Datasets:

health360

/

Healix-Shot

like 5

Follow

Health 360 7

combined. Sigmoid Inhibition-An enzyme that displays sigmoidal kinetics will satisfy the Michaelis-Menten equation with the following modification. Koshland, Nemethy, and Filmer (11) have proposed a little used device that can discriminate between cooperative and noncooperative protein-ligand interaction. They suggested that by det.ermining the cooperativity index (R,), which is the ratio of substrate concentration necessary to give 90% saturation to that concentration which gives 10% saturation, any deviation from the Michaelis-Menten hyperbolic relationship can easily be determined. R, should be 81 for all cases that follow Michaelian kinetics. At (S)sO and (X),0, Equation 1 can be written in the form of two simultaneous equations as follows. and NADP+ against oleyl-CoA inhibition. I;nset, an enlargement of the curves at low oleyl-CoA concentrations. The concentrations of activators used are given in Fig. 2 legend. The expression "All Activators" represents the four activators given above. between ligand and protein and evaluate the interaction coefficient, n, which is only an approximate indication of the number of substrate-binding sites involved in the interaction. When a substrate acts as an allosteric inhibitor, theoretically, it should be possible to analyze the saturation curve on the same basis by using Equation 5. It must be emphasized that most substrate inhibition curves would give n. values greater than 1 when Equation 5 is used. Other diagnostic procedures would be required to confirm whether or not true cooperativity is involved. This approach has some value only when modulators activate an enzyme at high substrate concentration ranges. The computational procedure we followed was to select as zero inhibitor concentration the concentration of substrate at which there is neither an increase nor a decrease in the reaction rate when the substrate level was increased further by at least 10%. The concentration of substrate required to reduce this optimal rate by 10% was taken to represent (S&, and the concentration of substrate that reduced the optimal rate by 90% taken as (S1)90. (1)0.6 M 1.3 x 10-d 1.65 X lo+ 6.5 X 1O-4 3.3 x 10-d 3.0 x 10-d 4.5 x 10-S Extremely high 7.0 x 10-4 3.75 x 10-a 1.55 X 10-Z Extremely high 6.0 X 1O-6 9.1 x 10-e 6.1 X lo+ 6.1 X lo+ 6.2 X 1O-6 1.65 X lo-& 2.0 x 10-s Analysis of Fig. 3a in which ammonia acted as a substrate and an inhibitor of the reductive amination reaction of Pythium NAD-specific glutamic dehydrogenase, based on these suggestions, showed that the interaction of the inhibitor with the enzyme may be positive cooperative. (X1)90/(Sl)10 was estimated as 3, and n value computed as 4. A similar computation done for a-ketoglutarate as an inhibitor and substrate (Fig. 3b) gave an (XI),~/(SI)n, value of 3.72 and n of 3.3. ATP, as an activator, has been referred to above under adenylate control. Activation by all effecters was more pronounced on the Pythium catalyst than on AchZya glutamic dehydrogenase ( Table I). The activators were tested for their efficacy in the reductive amination and oxidative deamination reactions of both enzymes. Cyclic nucleotides, 3',5'-AMP and 3',5'-GMP, had slight stimulatory influence on the biosynthetic reaction but inhibited the catabolic process. Besides quantitative differences shown in the influence of the activators on the two enzymes, the effects of adenylates and P-enolpyruvate on these glutamic dehydrogenases were markedly different. Pytlzium glutamic dehydrogenase was inhibited by P-enolpyruvate when the catabolic reaction was catalyzed whereas Achlya glutamic dehydrogenase was activated. ATP inhibited the catabolic reaction of Achlyu glutamic dehydrogenase but not that of Pythium. Other than these subtle modifica- Allosteric Inhibiters-In addition to citrate and AMP, long chain fatty acid-CoA esters were found to be inhibitors of these glutamio dehydrogenases (Fig. 4). The two esters studied in some detail were palmitoyl-Coil and oleyl-CoA. (I)0.5 values' for these esters are given in Table II. At very low concentrations, the esters activated the enzyme slightly. This may be caused by the presence of traces of free CoA that are present in the commercial preparation of the esters. Alternatively, the ester may be acting as a competitive inhibitor with one of the substrates that inhibit, the enzyme. At higher concentrations, the ester would bind at its own inhibitor site as well. @I) 0.5~ and Activation-We have mentioned elsewhere (4) that the activation mechanism operative on the glutamic dehydro-1 Following a suggestion of Koshland et al., (II), discussed in detail by Atkinson (12), we will restrict our use of the term (X),,., to mean the concentration of ligand required to fill half of the sites of a given type on the enzyme. When the ligand is a substrate, we will use (S) in place of (X) ; when an inhibitor, genases is geared toward modulation of the enzymatic activity at high substrate concentrations where substrates appear to inhibit allosterically. By studying the influence of effectorson the enzyme at (X1)o.b of a substrate, a valuable estimate of the extent of ligand activation or inhibition can be made. This approach was taken to evaluate all activators of these glutamic dehydrogenases. The data given here are predominantly for the Pythium catalyst. Ammonia was held fixed at @I),., and the other substrates kept at optimal levels. c-r-Ketoglutarate, being also an allosteric substrate inhibitor, was not used at saturating concentrations. An activation curve for GTP is presented in Fig. 5 and the activation curves for four of the nine short chain acyl-CoA derivatives in Fig. 6. The results are presented as plots of (0, -a,) against the concentration of the activators. The notation ve is the reaction rate with activator and v0 is the rate without activator. The same data were analyzed in Line-weaverBurk double reciprocal form. This method permitted us to determine the K, (activation constant) values for the activators and also evaluate the nature of ligand binding. Although the activation of all of the short chain acyl-CoA derivatives were analyzed in this manner, the plots given are the curves for CoA, acetyl-CoA, butyryl-CoA, and succinyl-CoA. A curve for dephospho-CoA, which neither activated nor inhibited the enzyme, has been included but the double reciprocal form omitted. There were relatively little differences in the degree of activation caused by the acyl-CoA derivatives. The activation constants for GTP and the acyl-CoA derivatives were approximately the same at about 4 X lop5 M. Since acyl-CoA derivatives and free CoA functioned as activators, the sulfhydryl group of CoA does not appear to be essential for activation. The 3'-phosphate group of CoA, however, is essential (Table III). All of the activators displayed what appears to be a linear function in kinetic binding of the activators to the enzyme. But on replotting the data as log (v, -v,)/(V,,, -v,) against log of activator concentration, the log plots were biphasic with one slope of n value equaling 1 and a second slope with n values varying between 2 and 4. Only the data for GTP and butyryl-CoA are presented as they represent the extreme cases of the various biphasic log plots obtained. Presumably, some form of cooperative binding of the activators does occur. Although the activators of these glutamic dehydrogenases are phosphorylated compounds, their structures are so different that a clear picture cannot be obtained from the structures about important functional groups. Phosphate is unlikely to be the single important factor because AMP is an inhibitor and several other phosphorylated compounds tested have no influence on the enzyme. An alternative interpretation is that multiple sites may be involved in the activation process. Tests were done to see if there was any cumulative property of the activators in their interaction with the enzyme. Multivalency The ability of the activators to antagonize the inhibitory action of AMP, citrate, and long chain fatty acid-CoA esters was used as the test model. The reason for this approach rested on an early observation that all of the activators, except ATP, can easily reverse the allosteric inhibition by the substrates, ammonia, and cY-ketoglutarate. Nonsubstrate inhibitors were more toxic and resisted antagonism by single activators. AMP InhiKttin-When the activators were used singly and in combination to antagonize AMP effects on Pythium glutamic dehydrogenases, only NADP+ and P-enolpyruvate could, independently, overcome most of the inhibition (Table IV). GTP, CoA (or acetyl-CoA) did not antagonize AMP inhibition significantly. Combined, GTP and CoA were no better than CoA alone. On the contrary, NADPf and P-enolpyruvate acted cumulatively and antagonized AMP completely. Although P-enolpyruvate was cumulative with either GTP or CoA, NADP+ did not show any cumulative property with these two compounds. It would appear from these results that NADP+, GTP, and CoA were interacting at the same or closely related sites. P-Enolpyruvate, definitely, has a distinct site. However, because of the small relative differences in percentage deinhibition elicited when GTP and CoA were used, it became necessary to estimate the number of activator sites by using other inhibitors. Citrate Inhibition-All of the activators were tested for their ability to release the enzyme from citrate inhibition. A previous study had shown that NADP+ and P-enolpyruvate acted cumulatively against citrate (4). Similar studies were done here with GTP, ATP, and acyl-CoA derivatives and then compared against NADP+ and P-enolpyruvate antagonism. Although ATP is an activator (based on the adenylate control hypothesis) it failed to antagonize citrate (Fig. 7). GTP, COA, and P-enolpyruvate were weakly antagonistic to citrate. Only NADP+ showed a significant antagonistic property. GTP and CoA had some cumulative ability, but they were considerably less effective than NADP+ alone. When P-enolpyruvate and by guest on March 24, 2020 http://www.jbc.org/ Downloaded from either CoA or GTP were used, their cumulative effect did not quite match that of NADP+ as an antagonist (see Table II for a record of (Z)0.5 of citrate in the presence of various activators). P-Enolpyruvate, GTP, and CoA combined were cumulatively as effective as NADP+ alone. When all four activators (NADP+, P-enolpyruvate, GTP, and CoA) were used, complete antagonism of citrate inhibition occurred (Fig. 7). From these results, it appears that all of the activators have separate binding sites. Okyl-CoA Inhibition-Long chain fatty acid-CoA derivatives have been studied in detail by Taketa and Pogell (13) with regard to their inhibitory effect on enzymes not related to fatty acid biosynthesis. They concluded that palmitoyl-CoA may act as a detergent during inhibition. Normally, inhibition of this type would lead to an irreversible denaturation of the protein. Zahler,Barden,and Cleland (14) have estimated that the mixed micellar concentration of palmitoyl-Coil is 2 to 4 pM. Dorsey and Porter (15) showed that the critical mixed micellar concentration of palmitoyl-CoA and fatty acid synthetase is 5 PM. Above this concentration, the enzyme was inhibited irreversibly. We do not know what the critical mixed micellar level for Pylthium glutamic dehydrogenase and palmitoyl or oleyl-CoA may be. But Pythium glutamic dehydrogenase was markedly inhibited by oleyl-and palmitoyl-Coil at concentrations above 1.5 PM (Fig. 4) Activators used were at the concentrations indicated for Fig. 2. Reactants as outlined in Fig. 3b with 4 pg of enzyme. The values for ATP (3 mM) as antagonist were nearly coincident with the control plot and so have been omitted. PEP, phosphoenolpyruvate. protective effect. As shown in Table II and Fig. 4 inset, significant protection against oleyl-CoA inhibition occurred when all the activators were present. The enzyme was not inhibited by oleyl-CoA at concentrations below 6 PM. At 10 pM oleyl-CoA concentration, the enzyme was 80% inhibited in the absence of the activators, but only 20% inhibited when they were added. Significant inhibition always occurred at oleyl-CoA levels above 10 PM whether activators were present or not. It is possible that the coenzyme may inhibit in this range by detergent action. A more critical study of molar ratio of coenzyme and protein that leads to inhibition and how activators alter this ratio would be required to resolve this problem. Physiological Action of Modulators At this stage of our studies, it is not possible to give a complete synthesis of the physiological reasons that would account for the multivalent control of glutamic dehydrogenases from these fungi (Oomycetes). NADP+ and P-Enolpyruvate-Conclusions drawn from data presented previously (4) on the physiological basis of NADPi and P-enolpyruvate activation of Pythium, Achlya, and Sapro-Zegnia glutamic dehydrogenases are not contradicted. The NADP-specific isocitric and NAD-specific glutamic dehydrogenases act cooperatively to maintain a balance of pyridine nucleotides as follows. Isocitrate + NH,+ + NADP+ + NADH $ NAD+ + NADPH + glutamate + CO, P-Enolpyruvate may not be involved in this

reaction since the substrates can provide sufficient energy to effect the conversion. Under energy-rich conditions when P-enolpyruvate accumulates, GTP and P-enolpyruvate activation of the biosynthetic reaction of glutamic dehydrogenase is a reasonable effect because the citric acid cycle would be operating as a biosynthetic unit supplying its intermediates for amino acids and nucleotides. The questions, why do these enzymes display marked sensitivity to inhibition by their substrates, ammonia and cY-ketoglutarate; and why do effecters modulate the enzyme only at these toxic levels of substrate, remain unanswered. But two possible explanations can be offered. First, if cu-ketoglutarate is liable to accumulate during its production from carbohydrates and transamination reactions of glutamate, then this metabolite may act directly as a substrate and indirectly as an end product (transaminase-glutamic dehydrogenase couple) feedback effector of the glutamic dehydrogenase. One role of the activators would be to relieve the enzyme from allosteric inhibition by substrates and permit continued biosynthesis. The second and more attractive proposal is that substrate control is related to catabolism. The enzyme must have evolved as a catabolic catalyst, because it is subjected to catabolite repression. During deamination reactions, ar-ketoglutarate and ammonia may accumulate. This may have led to the development of a product inhibition control mechanism which is observed as substrate inhibition in these studies The multivalent control by activators may have evolved subsequently to overcome this inhibition when the enzyme has to function in a biosynthetic capacity. Acyl-CoA Derivatives-According to the transhydrogenase hypothesis, NADPH is produced at the expense of NADH. One use to which the NADPH may be put is to synthesize fatty acids. Activation of the enzyme by short chain intermediates of fatty acid synthesis in a manner that favors NADH utilization fits this concept very well. Palmitoyl-and oleyl-CoA, which may be considered "end products" of fatty acid biosynthesis feedback, inhibit the enzyme. This is a common method of control of the first enzyme involved in most biosynthetic sequences studied. Although citrate is an inhibitor, its effect is qualitatively unidirectional on the catabolic reaction of the enzyme (4). Therefore, the anomaly of citrate inhibition would not contradict acetyl-CoA activation of the enzyme. In fact, it does support the concept that the enzyme has catabolic function because citrate can be looked on as an end product. GTP-A reasonable explanation for the unidirectional stimulation of the enzyme by GTP is difficult because many interrelated reactions utilize GTP. For example, substrate level oxidative phosphorylation of succinyl-Coil produces GTP; a GTP-linked fatty acid activation enzyme (acyl-CoA synthetase) is present in mitochondria and closely linked to the oxidation of a-ketoglutarate; GTP is utilized during synthesis of P-enolpyruvate from oxalacetate via P-enolpyruvate carboxykinase action. It is unlikely that GTP has the same function as ATP because the latter compound responds very differently in multivalent studies. The only definite conclusion we can draw is that GTP, like the other activators, activates the enzyme for active biosynthesis. AdenyZutesThere are two suggestions that satisfactorily account for the effect of adenylates on the enzyme. (a) The adenylate control may reflect the physiological response of the enzyme to the energy state of the cell. Under energy-rich conditions, there would be an ample supply of ATP that could be utilized for biosynthesis. The biosynthetic reaction of the glutamic dehydrogenase, consequently, would be encouraged. (b) ATP may be involved in an NAD-kinase reaction of the type ATP + NAD+ ti NADP+ + ADP (7) which would assure a continued production of NADP+ for the transhydrogenase couple of isocitric and glutamic dehydrogenases. Table V summarizes those physiological reactions that may be directly influenced by the activity of the glutamic dehydrogenase. The list is not exhaustive. We have only presented those interrelations pertinent to the data presented in this communication. Envoy Two different mechanisms of allosteric control of NADspecific glutamic dehydrogenases have been observed among members of the simple fungi, Phycomycetes (4, 7). A broad sampling covering some 30 different species and all the major orders of this taxonomic group has been made.3 By a remarkable coincidence, the two complex enzyme control systems are distributed among the fungi in an identical fashion as the only two known pathways of lysine biosynthesis (16). Fungi are unique organisms in that they alone appear to possess these two biosynthetic pathways. A biochemical rationale for this correlation is not yet evident. It is hoped that more work on these and related enzymatic reactions of amino acid biosynthesis in these fungi may throw some light on the evolutionary aspect of our observations. Immunoinformatic approach for the construction of multi-epitopes vaccine against omicron COVID-19 variant The newly discovered SARS-CoV-2 Omicron variant B.1.1.529 is a Variant of Concern (VOC) announced by the World Health Organization (WHO). It's becoming increasingly difficult to keep these variants from spreading over the planet. The fifth wave has begun in several countries because of Omicron variant, and it is posing a threat to human civilization. As a result, we need effective vaccination that can tackle Omicron SARS-CoV-2 variants that are bound to emerge. Therefore, the current study is an initiative to design a peptide-based chimeric vaccine that may potentially battle SARS-CoV-2 Omicron variant. As a result, the most relevant epitopes present in the mutagenic areas of Omicron spike protein were identified using a set of computational tools and immunoinformatic techniques to uncover common MHC-1, MHC-II, and B cell epitopes that may have the ability to influence the host immune mechanism. A final of three epitopes from CD8+ T-cell, CD4+ T-cell epitopes, and B-cell were shortlisted from spike protein, and that are highly antigenic, IFN-γ inducer, as well as overlapping for the construction of twelve vaccine models. As a result, the antigenic epitopes were coupled with a flexible and stable peptide linker, and the adjuvant was added at the N-terminal end to create a unique vaccine candidate. The structure of a 3D vaccine candidate was refined, and its quality was assessed by using web servers. However, the applied immunoinformatic study along with the molecular docking and simulation of 12 modeled vaccines constructs against six distinct HLAs, and TLRs (TLR2, and TLR4) complexes revealed that the V1 construct was non-allergenic, non-toxic, highly immunogenic, antigenic, and most stable. The vaccine candidate's stability was confirmed by molecular dynamics investigations. Finally, we studied the expression of the suggested vaccination using codon optimization and in-silico cloning. The current study proposed V1 Multi-Epitope Vaccine (MEV) as a significant vaccine candidate that may help the scientific community to treat SARS-CoV-2 infections. Introduction In December 2019, a novel Beta coronavirus SARS-CoV-2 was identified as the etiological agent of COVID-19, which has been affecting more than 0.318 billion global populations and resulted in 5.5 million fatalities (https://www.worldometers.info/coronavirus/) (Poon and Peiris, 2020). On the other hand, several variants (i.e. Alpha, Beta, Gamma, Delta, Kappa, and so on) have been evolved as a result of mutations in the SARS-CoV-2 genome since the start of the COVID-19 pandemic. These mutations have a higher impact on the pace of transmission and the immunological escape mechanism (Tao et al., 2021;Thakur et al., 2021). The emergence of multiple variants leads to the emergence of several waves of devastating pandemics around the world Li et al., 2021). Recently, a novel mutant variant of SARS-CoV-2 was identified in Botswana, South Africa. The World Health Organization reported this on November 24, 2021. On November 26, 2021, WHO designated the novel variant Omicron as a Variant of Concern (VOC) (Poudel et al., 2022). The emergence of a highly mutated SARS-CoV-2 strain (B.1.1.529, Omicron) and its rapid spread across six continents within a week of its first detection has raised worldwide public health concern (Nishiura et al., 2022). Omicron's mutational profile is critical for understanding whether it shares or differs from other SARS-CoV-2 variants in terms of clinical manifestations. Various countries have taken strict actions to minimize the transmission of the Omicron variant of SARS-CoV-2 as a result of the heavily mutated variant emergence. That includes implementing measures, such as restrictions on international flights, strengthened genomic surveillance, and strain sequencing (Petersen et al., 2022). SARS-CoV-2 pathogenesis is initiated by viral particles adhering to host cell cellular surface receptors, where the Spike glycoprotein (S) of the SARS-CoV-2 virus identifies and binds to Human Angiotensin Converting Enzyme 2 (hACE2) present on the lungs cells (Harrison et al., 2020). Previous studies reported that all the recognized variants showed most of the mutations in their Spike glycoprotein (Harvey et al., 2021). As a result, developing an antiviral drug targeting Spike protein is an appealing strategy for preventing the transmission of COVID variants. The conventional approach of vaccine development, which involves whole organisms or proteins, results in excessive antigenic load and an increased risk of allergic reactions. This issue can be overcome by developing a Multi-Epitope Vaccine (MEV), which comprises of a short immunogenic peptide, adequate linkers, and an adjuvant capable of eliciting robust and targeted immune responses while reducing allergic reactions. The MEVs are more efficient than single-epitope vaccines because of their specificity towards target, stability of complex, timesaving features, and cost-effectiveness (Bahrami et al., 2019). Furthermore, they are likely to elicit strong humoral and cellular immunological responses at the same time because of the incorporation of T-cell and B-cell epitopes. The MEVs with adjuvants are thought to stimulate long-term immune responses and increased immunogenicity as adjuvants are crucial for reducing the amount of antigen used and the number of shots (Reed et al., 2013). In this continuing pandemic, in silico approaches are the scientifically reliable alternative options for specific vaccine development applicable to all SARS-CoV-2 Variant of Concern (VOC) and Variant of Interest (VOI). Thus, in this study, we hypothesized MEVs against spike protein of Omicron variant using immunoinformatic approaches. In brief, the SARS-CoV-2 Omicron Spike protein was searched for antigenic determinants and then tested to forecast B-cell and T-cell epitopes with Class-I and II MHC (Major Histocompatibility Complex) alleles. Antigenicity, conservancy, and global coverage of anticipated epitopes were investigated. Multiple in silico approaches were also employed to validate the structural stability, physiochemical properties, immunogenicity, toxicity, and antigenicity of the constructed MEVs. The MEV's binding affinity and stability with human pathogenic receptors were further investigated using molecular docking and MD simulations. Finally, the MEV codons were optimized for utilization in the E. coli system, and MEV expression profile was validated using in silico cloning. The findings provide the way for the construction of Omicron variant vaccines, however further experimental validation is needed to battle the pandemic. Data retrieval The Omicron spike genomic sequence was obtained from the Global Initiative for Sharing All Influenza Data (GISAID) under the accession number EPI ISL 8616776, which was reported by the National Health Laboratory South Africa, and then translated into amino acids sequence using the BLASTX tool. Multiple Sequence Alignment The spike proteins of Omicron variant and Wuhan (wild type) were aligned to study the variation in these strains of SARS-CoV-2. For this purpose, ClustalO tool was used that work on progressive alignment techniques. Reverse Vaccinology The Vaccinomics, biochemistry, immunology, proteomics, molecular biology, and genomics have all recently evolved, and transformed traditional vaccinology into Reverse Vaccinology (RV). The RV is an innovative and developing computational method that has been widely utilized to improve vaccine target and model prediction especially for those pathogens that are difficult to cultivate in the lab (Moxon et al., 2019). This methodology integrates immunogenomics, immunogenetics, and bioinformatics to scan the whole proteome of a pathogen to find a new vaccine candidate and analyze its ability to elicit a host-immune response. The complete applied reverse vaccinology methodology is highlighted in Fig. 1. Antigenicity identification Antigenicity refers to an antigen's capacity of pathogen to be identified by the immune system and elicit an immunological response. The VaxiJen v 2.0 was used to assess the antigenicity of spike protein of Omicron with 0.4 as a threshold value (Doytchinova and Flower, 2007). Prediction of MHC-I T-cell epitopes The immunomodulatory effects of different types of epitopes were studied using the NetCTL service to detect T-cell epitopes from spike protein (Doytchinova et al., 2006;Jalal et al., 2021) using a combined score of anticipated characteristics and a 0.75 threshold (Hasan et al., 2016). The T cell epitopes were also submitted to the Immune Epitope Database and Analysis Resource (IEDB AR) for MHC I binding prediction, which demonstrated that T cells recognized MHC-I antigens (Kim et al., 2012). The IEDB server has lots of information

on epitope immunogenicity for adaptive immunity. The default consensus settings of CombLib, ANN, NetMHCpan and SMM together with all HLA alleles and human MHC were used for epitope prediction in the current study (Kim et al., 2012). The HLA alleles analyzed in present study were HLA-A 0205, HLA_0201, HLA-A3, HLA-B 5401, HLA-A2, HLA-A 2.1, and HLA-B 5102 (Almofti et al., 2021). The threshold parameters on the basis of IC 50 < 100μM and <0.2 of percentile rank were used as cut-off values for narrowing down of MHC class-I epitopes (Rahman et al., 2020). MHC-I immunogenicity, antigenicity, toxicity and conservancy analysis The T-cell epitope must be immunogenic to trigger either CD4 + or CD8 + T cells. To determine their ability to elicit an immunological response, we employed the IEBD AR tool for MHC-I immunogenicity prediction. The default parameter settings were used for immunogenicity prediction, and positive value epitopes were chosen for future study (Dhanda et al., 2019). The conservancy, toxicity, and antigenic attributes of promiscuous MHC-I immunogenic epitopes were explored further. The epitope sequence should be conserved across all detected variant sequences in order to construct broad-spectrum peptide-based vaccines (Esmailnia et al., 2020). The IEBD server was utilized to perform the conversancy analysis (Angelo et al., 2017). The antigenicity of conserved epitopes was also predicted using the VaxiJen server, which has a 70-80% accuracy rate and a 0.5 probability threshold score (Doytchinova and Flower, 2007). The relative toxicity level of epitopes was determine through ToxinPred server with cut-off value set as 0.5 (Gupta et al., 2013). Identified MHC restricted alleles The identified MHCI -II epitopes were clustered to validate their respected MHC restricted alleles using the MHCcluster (Thomsen et al., 2013). The clustering performed for these epitopes is resulted in the plotting of heat map for the expression relation epitopes with corresponding alleles. It also generated phylogenetic tree that helped in the assessment of functional relation identification between HLAs and shortlisted epitopes (Ullah et al., 2020). B-cell epitope identification A perfect peptide vaccine may induce long-lasting humoral immunity, identical to the natural immunological response produced by pathogenic infections. The B-cell epitopes stimulate humoral immunity, which can kill infection by creating antibodies against exposed antigens to the human body. The ABCPred (Saha and Raghava, 2006), BCPred (Bhattacharya et al., 2019), and FCPred (EL-Manzalawy et al., 2008) servers are using sequence-based technique and a cut-off score of >0.8 was used for the identification of B-cell epitopes. Moreover, ElliPro server (Ponomarenko et al., 2008) was used for the assessment of hydrophobicity content of B-cell epitopes (EL-Manzalawy et al., 2008), Chou and Fashman tool (Chou, 1978) for beta turn prediction, Kaprplus and Schulz for flexibility scale (Karplus and Schulz, 1985), and Parker hydrophilicity scale for the identification of hydrophobic content (Parker et al., 1986), respectively. Epitope mapping (B-cell and T-cell) The development of epitope-based vaccines requires an epitope that can trigger immune cells (B and T cells) to react (Khalid et al., 2022). Consequently, the epitopes having binding affinity and similarity with each other from shortlisted MHC-I/II and B-cell epitopes of Omicron spike protein were determined. Manual comparison was carried out, and overlapping epitopes are classified as plausible peptide sequence. These peptides were compiled and used as final anticipated epitopes for vaccine modeling. Vaccines construction and structure modeling The shortlisted epitopes were conjugated sequentially with suitable adjuvant, PADRE (Pan HLA-DR reactive epitope) sequence, and linkers to create various vaccine constructs. We employed a variety of epitope sequences with four different adjuvants such as L7/L12 ribosomal protein, beta-defensin, HBHA protein, and HBHA conserved sequence to construct a vaccine model with minimal toxicity, allergenicity, and immunogenicity (Y. Yang et al., 2015). The addition of linkers boosts immunogenicity, while the PADRE peptide assists in the production of CD4 + T-cells, improving the efficacy and potency of peptide vaccines (Ghaffari-Nazari et al., 2015). HEYGAEALERAG and GGGS linkers were employed to combine HTL, CTL, and B-cell epitopes, while EAAAK linkers were utilized to link adjuvant sequences at both the N-and C-terminus. The TLR connections that have a profound immuno-stimulatory effect on polarized CTL response were further investigated in the fused vaccine models (Araújo et al., 2019). Antigenicity, allergenicity, and solubility assessment for vaccines constructs The AlgPred server for allergenicity analysis with a cut-off of − 0.4 and an accuracy of 85 percent was utilized to examine their allergenicity effects to overcome the allergic reaction characteristic of vaccine constructs. Scores less than a threshold are considered as non-allergenic vaccine (Yukeswaran et al., 2021). The antigenic potential of model vaccines was further predicted through VaxiJen and ANTIGENpro servers with the default threshold value of >0.5. Furthermore, using default settings of 74 percent accuracy and corresponding probability (i. e. 0.5), the SOLpro program was used to determine the solubility characteristic of the vaccine model that will aid in the successful expression of vaccine construct in E. coli plasmid (Dhanda et al., 2019). Physiochemical analysis of constructed vaccines The Expasy ProtParam (Garg et al., 2016) program was used to perform physicochemical analysis and functional characterization of vaccines based on parameters such as pK values, molecular weight, estimated half-life, GRAVY values, aliphatic index, instability index, isoelectric pH and half-life of generated vaccine model . Physiochemical properties must be assessed to determine the safety and effectiveness of vaccine candidates. Structure modeling and molecular dynamic simulation The interaction of the vaccine component with its receptors was studied to induce a consistent immune response in the vaccine model to target cells. The molecular docking method is a useful tool for estimating the binding energies and assessing interactions between epitopes and HLA molecules (Morris and Lim-Wilby, 2008). The final possible vaccine constructs that met all of the framework's criteria were then docked into the binding cavities of six HLA alleles commonly found in the human population i.e. 1H15 (HLA-DR B5*01:01), 2FSE (HLA-DR B1*01:01), 1A6A (HLA-DR B1*03:01), 3C5J (HLA-DR B3*02:02), 2Q6W (HLA-DR B3*01:01), and 2SEB(HLA-DRB1*04:01) retrieved from the protein database (Burley et al., 2021). The HLAs and vaccine interactions were estimated using the PatchDock server (Schneidman-Duhovny et al., 2005). Furthermore, The GRAMMX tool was also employed to validate the vaccine and TLR4/MD complex docking. The TLR4 is involved in the recognition of viral proteins, which results in the release of inflammatory cytokines. Hu et al. (2012) reported that TLR4 is required for an effective immune response against SARS-CoV-2 (Hu et al., 2012). Therefore, molecular docking of vaccine construct and TLR4 was performed through GRAMMX (Tovchigrechko and Vakser, 2006) and the interactions such as hydrogen bonding were validated and visualized by PDBsum (Laskowski, 2001) and UCSF Chimera (Pettersen et al., 2004). The MD simulation studies were carried out to evaluate the structural stability of the designed vaccine using GROMACS v. 2020 (Van Der Spoel et al., 2005). The topological parameters for the vaccine were processed by using GROMOS96 54A7 force field (Lin and van Gunsteren, 2013). The vaccine was put at 1.0 Å apart from the box edge in a cubic box that was constructed. Further, solvent molecules of SPC water model were added by using periodic boundary conditions. The addition of counter ions neutralized the solvated system. With a maximum force of <1000 kJ mol − 1 Å − 1 , the neutralized system was minimized using the steepest descent technique. The system was gradually heated at constant temperature and pressure in NVT and NPT ensemble for 100 ps using thermostat and Berendsen barostat algorithms. For constraining the bond length and calculation of electrostatic long-range interactions, LINCS algorithm, and PME method was applied, respectively. Finally, a 50 ns MD production was performed with coordinates and energies stored every 10 ps in the output trajectory file, according to established procedure (Jalal et al., 2021). Subsequently, the Root Mean Square Fluctuation (RMSF), Root Mean Square Deviation (RMSD), and Radius of gyration (Rg) were plotted to evaluate the stability of the system. Furthermore, molecular dynamic simulation of docked complex (vaccine with TLR4) was performed via iMODs server. The iMODs defines and estimates the flexibility of protein complex based on the direction and extent of the immanent motions of the complex in terms of deformability, covariance, B-factors, and eigenvalue. Immune simulation of final vaccine construct The immune response profile and vaccine immune simulation were anticipated using the C-ImmSim simulation server (Rahman et al., 2020). Three injections of modeled prophylactic Omicron variant vaccine were administered at 1, 82, and 126 h time periods and 12345 random seed for up to 4 weeks at three different intervals with default simulation parameters containing no LPS, volume, and simulation stages at 10, and 1000, respectively, with homozygous host haplotypes HLA-A*0101, HLA-A*0201, HLA-B*0702, HLA-DRB1*0101, and HLA-DRB1*0401 (Kaba et al., 2018). In-silico cloning and codon optimization of final vaccine construct The Java Codon Adaptation Tool (JCAT) was used to modify the vaccine's codon within the E. coli host strain k-12 (Grote et al., 2005). The amino acid sequence of the vaccine was back translated to DNA and then modified for codon use in E. coli. The adaptation of vaccine model was expressed in terms of Codon Adaptation Index (CAI) value and GC content estimated through the JCAT algorithm. In addition, the Snap-Gene tool (Hess et al., 1992) was used to clone the modified gene sequence of the final vaccine construct in E. coli pET28a vector to ensure vaccine construct expression. Antigenicity prediction for Omicron Spike protein The antigenicity of Omicron spike proteins analyzed through Vax-iJen server was identified as 0.4125 with a cut-off value of 0.4. The analysis indicates the spike protein as antigenic that can stimulate hostimmune response. Prediction of MHC Class-I T-cell epitopes The NetCTL server was used to forecast 1261 T-cell epitopes using a threshold value of 0.75 (Supplementary file 1), however, only 131 epitopes were found to exhibit optimum T-cell binding. The MHC-I binding on these 131 epitopes was evaluated using IEBD tool. It identified approximately 2657 MHC-I epitopes (Supplementary file 2). However, based on MHC-I and T-cell interaction, 163 epitopes were identified that evoked strong binding affinity using a cut-off criterion of IC 50 < 100 and percentile rank ≤0.2. All the shortlisted epitopes were predicted to be efficient T-cell binders and were evaluated further. Immunogenicity antigenicity, conservancy and toxicity analysis for shortlisted epitopes The immunogenicity of epitopes determines their ability to induce Tcell responses. The immunogenicity prediction was performed for the shortlisted epitopes. The higher the immunogenicity score, the better epitopes will be at stimulating cellular immunity and T-cells. From the 163 MHC-I shortlisted epitopes, 96 immunogenic epitopes with a positive score cut-off were identified as having significant immunogenic values using the IEBD service. These immunogenic selected epitopes were employed in the development of vaccines. The toxicity, antigenicity, and conservancy analyses were also performed on the 96 immunogenic epitopes that were shortlisted. The results of the ToxinPred and IEBD conservation tools showed that all 96 sequences were non-toxic and 100% conserved. However, antigenicity analysis using the VaxiJen server resulted in a total of 24 epitopes (Table 1) as antigenic (i.e., score between 0.5 and 1.0) and subjected to further evaluation whereas discarding the non-antigen one. MHC-II epitopes identification and conservancy analysis Furthermore, 916 MHC-II epitopes were identified employing the IEBD server in addition to MHC-I epitope prediction (Supplementary file 3). We evaluated epitopes based on their percentile rank with <0.2 values and a binding affinity with IC 50 < 100 nM. A total of 15 MHC-II epitopes were predicted by the server following the applied threshold and 100 percent conserved predicted by the IEBD conservancy analysis for MHC-II epitope (Table 2). MHC restriction cluster analysis of shortlisted epitopes The MHCclusters tool was used to evaluate the identified MHC-I/II epitopes in relation to MHC restricted allele and their suitable peptide validated the epitopes found in T-cells. In terms of annotation, the interaction between MHC-I/II and HLAs is displayed as a heat map and phylogenetic dynamic tree with red color representing stronger interaction and yellow color representing weak interaction (Fig. 3). Prediction of B-cell epitope Ideally, both humoral and cellular immunity are required to successfully eradicate the virus from the body. As a result, B-cell epitopes against Omicron spike protein were discovered using ABCPred, BCPred, and FBCPred. These methods identified 155, 22, and 42 B-cell epitopes using a threshold value of 0.51, and a specificity of 75 percent, (Supplementary File 4- Table S1). Chou-Fasman beta-turn prediction, Kolaskar Tongaonkar antigenicity, BepiPred linear epitope prediction,

Parker hydrophilicity prediction, Karplus-Schulz flexibility prediction, and Emini surface accessibility prediction were also used to further examine and select the resulting B-cell epitopes as highlighted in Fig. 4. Epitope mapping and prioritization The selected B-cell epitopes were used as a template and manually evaluated against MHC-I and MHC-II epitopes to screen out overlapping epitopes. The comparison analysis was resulted in the shortlisting of three epitopes that overlap MHC-I, MHC-II, and B-cell epitopes (Table 3) i.e., TESIVRFPNITNLCPFDEVFNATRFASVYAWNRKRISNCVADYSVLYN LAPFFTFKCYG, PQSAPHGVVFLHVTYVPAQEKNFTTAPAICHDGKAHF-PREGVFVSNGTHWF, and TQQLIRAAEIRASANLAATKMSECVLGQSKRV DFCGKG, respectively. Vaccine construction, allergenicity, antigenicity solubility and physicochemical properties prediction Four adjuvants, PADRE sequences, GGGS, HEYGAEALERAG, and EAAAK linkers were used to insert these four epitopes in a sequential way. The PADRE sequence helped to overcome the universal polymorphism impact of HLA-DR molecules in varied cultures, whereas this combination of adjuvants and linkers leads to activating a large immune response in the body against viruses. The selected epitopes were connected manually through bash shell script to each other using a specialized Glycine-Serine linker i.e., GGGS sequence, H-linker (HEY-GAEALERAG) and PADRE Sequence (AKFVAAWTLKAAA). A four molecular adjuvant was added towards the N-terminal direction of the epitopes. These adjuvants were attached to the epitopes by one EAAAK stiff linker. Eventually, twelve vaccine models were constructed using three shortlisted epitopes. Table 4 showed the details of the vaccine constructs. The antigenicity, solubility, and allergenicity of the twelve vaccine models were evaluated further. Eight vaccine models (i.e., V2, V3, V4, V7, V8, V10, V11, and V12) were identified as highly allergic in nature using the AlgPred program with scores ranging from 0.2 to 0.3 and were eliminated. However, the antigenicity predicted by the ANTIGENpro server, and the solubility predicted through SOLpro tool of the remaining four vaccine constructs for their successful expression in E. coli vector. It resulted in the high antigenicity and solubility for remaining four vaccine constructs (V1, V5, V6, and V9) with scores ranging from 0.4 to 0.8 and were thus subjected to further investigation. The ProtParam was used to predict physicochemical features for four shortlisted vaccine designs. The estimated molecular weight of the vaccine's models was ~41 KDa, with a pI score of ~5, an instability index score of 28-39, and a high aliphatic score of 83-86. On the other hand, the grand average of hydropathicity was estimated to be in the range of − 0.2. Nevertheless, among these four vaccines only V1 construct was identified to be stable with adjuvant HBHA. Table 5 lists the allergenicity, solubility, and antigenicity of all twelve vaccines and physico-chemical properties of four vaccine models. Structure prediction and validation of vaccine constructs The 3D structure of V1 vaccine construct was modeled using the SWISSMODEL tool. The template for V1 was PDB ID: 6NB3, a Spike glycoprotein from MERS-CoV complex with human neutralizing LCA60 antibody Fab fragment with 18% sequence identity, 0.25 GMQE score, and QMEANDisCo Global:0.42 ± 0.06 (Fig. 5). Furthermore, the stereochemical properties of the final selected vaccine construct were identified as 90.0 percent residues are in favorable region, 8.9 percent residues are in additionally allowed region, and 0.5 percent residues are in disallowed region based on the 3D structure evaluation through PROCHECK (Supplementary File 4-Fig. S1a). The ProSA program projected a Z-score of − 2.52, indicating that the model is like structures determined from NMR/X-ray crystallography (Supplementary File 4-Fig. S1b). Furthermore, as shown in Supplementary File 4-Fig. S1c, the PSIPRED program was utilized to validate the 2D (secondary) structure which revealed similar number of alpha helices, beta sheets, and beta turns. Molecular docking of vaccine construct (V1) Adaptive immunity is initiated by T cells that bind to HLA molecules. (Fig. 6a). Based on these global energies, HLA-DRB1*01:01, HLA-DR B1*03:01, and HLA-DR B5*01:01 showed more potency toward V1 construct. From docking results, it can be observed that V1 may act as a potential inhibitor against Omicron variant because of high binding affinities towards HLAs. To evaluate the immunological response, the GRAMMX tool was employed to perform a docking analysis to predict the interactions between V1 and the TLR 2 (2Z7X) and TLR 4 complex (PDB 3FXI). The adjuvant HBHA protein, which serves as TLR2 and 4 agonist and produces a variety of immune responses was used to build the V1 construct. The PatchDock docking revealed binding energies of − 12.12 and − 1.29 kcal/mol, showing that V1 and the TLR-2/4 complex have a significant interaction (Table 6). However, it can be clearly observed that V1 has showed more potency towards TLR2 complex. The Protein-Protein Interactions (PPIs) of the V1 model revealed that it mediates five hydrogen bonds with TLR2, mainly involve Glu177-Leu137, Tyr109-Thr149, Glu178-Arg150, Glu180-Arg150, and Leu57-Ser252 and three salt bridges between Lys208-Arg155, Glu180-Arg150, Glu178-Arg150. Whereas two hydrogen bonds with TLR4 were observed with Thr499-Pro254, Ser570-Tyr215 as shown in Fig. 6b. Molecular dynamics and immune response simulation study The MD simulation studies were conducted to determine the dynamics and structural stability of the designed vaccine in a timedependent manner. The simulated trajectories were analyzed to ensure structural stability during the simulation by plotting the RMSD graph of the backbone carbon atom. As evident from Fig. 7a, during the initial 15 ns of simulation, the system showed considerable fluctuations. However, after 15 ns the system gradually stabilized and remained stable till the end of the simulation with an average deviation of 1.23 Å that indicates the convergence of the simulated system. The predicted 3D structure of the vaccine is mostly consisted of helix and loop regions. The RMSF was plotted to further evaluate the intrinsic fluctuations of the amino acid residues. The plot showed variable fluctuations during simulation as depicted by Fig. 7b. An average RMSF was found to be 0.56 Å for the simulated system. The RMSF plot showed a few highly fluctuated regions; Ile162-Glu188, Val226-Gly247, and Val262-Gln315. All these regions were consisting of loops that are intrinsically flexible. Similarly, all the other amino acid residues of the simulated system showed significantly stable RMSF. Furthermore, the Radius of gyration was estimated to evaluate the compactness of the vaccine structure. As evident from Fig. 7c, during the initial 5 ns of simulation, the system showed high fluctuations. However, after 10 ns the system gradually compacts, contributing to the overall stability of the simulated vaccine construct. In addition, the C-Immune tool was used to predict human immune system response after vaccine injection at various time intervals. It confirmed that the immune response was consistent with the real immune reactions, such as the identification of T-cytotoxic cells, T-helper cells, and B-cells, Natural killer cells production, interleukins/interferons production, and antibody production (Fig. 7). Following the induction of vaccine injection, an increase in IgG1+IgG2, IgM, and IgG + IgM was seen, leading to a decrease in antigen concentration ( Fig. 8a and 8b). Upon vaccine construct injections, there was an increase in the production of NK cells, Th (helper), and Tc (cytotoxic) (Fig. 8c, d, and 8e). In addition, IFN-g production was increased after vaccination (Fig. 8f). V1 in silico cloning and codon optimization To optimize the codons and reverse translate the V1 for optimal production in E. coli (strain K12), the JCAT tool was employed. The average GC content and CAI value for V1 were estimated to be 41.9 percent and 0.61, respectively, resulting in successful vaccine construct expression. Finally, the SnapGene program was used to insert the optimized codon sequence (V1) into the pET30a (+) vector to create the recombinant plasmid (Fig. 9). Discussion The COVID-19 declared as pandemic and surged to cause increase in (Riley et al., 2021), Scotland (Angelo et al., 2017;Sheikh et al., 2021), Israel, Sydney, and South Asia) (https://www.bbc.com/news/world-asia -53420537). A successful vaccination campaign in the start of 2021 has substantially increased immunity of population whereas the advent Fig. 9. Codon optimization and in-silico cloning of vaccine model. In silico restriction cloning of the multi-epitope vaccine sequence into the pET30a (+) expression vector using SnapGene software, the red part represents the vaccine's gene coding, and the black circle represents the vector backbone. of the B.1.617.2 lineage of SARS-CoV-2 has put the delaying dosing tactics of immunization to a new challenge (first dose of vaccine followed by second dose given within 12 weeks) (W. Yang and Shaman, 2021). Additionally, lack of data regarding the spread of Omicron along with delta variant is currently unknown providing a huge challenge to overcome this pandemic. Nevertheless, some studies also suggested the use of multiple (three) dose regimes of vaccine to minimize such variant of concern effects and induce immunity (https://edition.cnn.com/202 1/07/08/health/us-coronavirus-thursday/index.html). Due to high transmission and immune evasion of ongoing Omicron-delta (Deltacron) variant, the development of new vaccine models is urgently needed to overcome the spread of fifth wave of pandemic. Traditional vaccine design methods use large proteins or whole organisms, which resulted in an excessive antigenic load and higher allergic responses (Chauhan et al., 2019;Sette and Fikes, 2003). Immunoinformatic techniques, on the other hand, are cost-effective and timesaving, and they may solve this problem by creating peptide-based vaccines that activate a powerful yet targeted immune response (He et al., 2018;Lu et al., 2017). As a result, the current study employed an in silico RV technique to create a multi-epitope vaccination against the Omicron spike protein (Fig. 1), which can activate immune cells (Abraham Peele et al., 2020). Multi-epitopes-based vaccine design is a new discipline that creates vaccine models that are not only protective in vivo but also effective (Cao et al., 2017;Guo et al., 2014;Zhou et al., 2009) (Jiang et al., 2017;Lennerz et al., 2014;Slingluff et al., 2013;Toledo et al., 2001). The current work identified immunogenic MHC-I, MHC-II, and B-cell epitopes that may be used to build a multi-epitope vaccine utilizing various filters such as: (i) The epitopes must be non-toxic, antigenic, non-allergenic, and highly conserved (Tables 1 and 2), (ii) have ability to bind to MHC-I/II alleles, and should be overlapping to CTL, HTL, and B-cell epitopes (Table 3). Bazhan et al. applied the similar approach to design T-cell multi epitope vaccine model against Ebola virus that was significantly immunogenic in mice (Bazhan et al., 2019). The current work utilized four adjuvants, including beta-defensin, and L7/L12 ribosomal protein, HBHA protein, HBHA conserved sequence, as well as GGGS, PADRE sequences, HEYGAEALERAG, and EAAAK linkers, to simulate twelve distinct vaccine constructs (Table 4). These twelve vaccination models, antigenicity, solubility, allergenicity and physiochemical properties were studied to narrow down the most promiscuous vaccine construct against the Omicron variant. The V1 was identified as the most powerful vaccine construct against omicron variant as non-allergenic, most antigenic, most soluble over expressing in E. coli, and having suitable physicochemical features because of the filtering (Table 5). Similar in silico strategies were also applied by Foroutan et al. against Toxoplasma gondii to evaluate the allergenicity and physiochemical properties of their model vaccine and through laboratory validation. It was validated that this vaccine design approach was able to trigger immune response in mice (Foroutan et al., 2020). As a result, the shortlisted V1 vaccine was modeled using the SWISSMODEL program (Fig. 5), and the modeled structure was validated using PRO-CHECK. The Ramachandran plot revealed that 90% of residues were classified in the desired area, indicating that the vaccine's tertiary structure was verified. The protein's Z-score of − 2.52 showed that it belongs to the experimentally validated structures of proteins solved by NMR and X-ray crystallographic methods, as assessed by the ProSA online site. The spike protein of the Omicron variant must interact with Toll-Like Receptor 4 (TLR4) produced in immune cells to elicit CTB (Boehme and Compton, 2004;Carty and Bowie, 2010;Xagorari and Chlichlia, 2008). It has been reported that the CTB lost ability to trigger inflammatory response in TLR4-deficient macrophages (Vaure and Liu, 2014). It was demonstrated through ELISA-based assays that the direct binding of CTB with TLR4 inflicts the activation of NF-κB (Phongsisay et al., 2015). Additionally, the molecular docking experiments were conducted to evaluate the interaction of the modeled vaccine design with Human Leukocyte Antigen (HLA) and TLR2 and 4. The TLR2-vaccine model mainly involves five hydrogen bonds and three salt bridges among Glu177-Leu137, Tyr109-Thr149, Glu178-Arg150, Glu180-Arg150, and Leu57-Ser252, Lys208-Arg155, Glu180-Arg150, Glu178-Arg150 residues. Whereas two hydrogen bonds with TLR4 were observed with Thr499-Pro254, Ser570-Tyr215 (Fig. 6). Several studies highlighted the importance of interaction of vaccines with TLR4 such as, Totura et al. demonstrated the susceptibility of mice to SARS-CoV infection is relatively high in TLR4 deficient mice

compared to wild type (Totura et al., 2015). Similarly Hu et al. observed that upregulation in expression of TLR4 when exposed to SARS-CoV infection, suggesting the importance of TLR in immune response stimulation (Hu et al., 2012). Importantly, the vaccine model was seen to be stable at 15 ns after a 50-ns molecular dynamics simulation (Fig. 7). The V1 model's codon optimization was reverse translated to its cDNA to enable effective expression in the E. coli pET-28a(+) expression vector. The expected GC and CAI values of V1 were 41.9 and 0.61%, respectively, indicating vaccine expression success (Fig. 9). Comparably, Foroutan et al. performed in silico codon optimization before expressing it in mice (Foroutan et al., 2020). Immune simulations of vaccine models revealed that the developed vaccine against the Omicron variant evoked a substantial immune response (Fig. 8). Correspondingly, the immune-simulation studies have been widely used for the construction of chimeric vaccine model against Klebsiella pneumoniae (Solanki et al., 2021), Mycobacterium tuberculosis (Bibi et al., 2021), Acinetobacter baumannii (Solanki and Tiwari, 2018), Ebola virus (Ullah et al., 2020) as well as against cancerous antigens (Zhang, 2018). Finally, using a reverse vaccinology technique, one vaccine construct against the Omicron version of SARS-CoV-2 was found and selected. The proposed vaccine candidate met all of the required criteria including physiochemical characteristics, antigenicity, and nonallergenicity. As a result of this finding, we determined that our vaccine construct (V1) is safe and could be administered to humans following the completion of a successful pre-clinical and clinical trial. This V1 construct is capable of inducing an innate, cellular and humoral immune response and displayed c8onsiderable expression when back translated to cDNA in the E. coli vector pET30a(+) plasmid. Immunoinformatic has proven that the vaccine construct can stimulate a reasonable immune response. However, to ensure vaccine design efficacy in animal models and human clinical trials, in vitro and in vivo evaluation of identified V1 model is crucial. This developed vaccine candidate has the potential to help manage the next COVID-19 wave and the global pandemic crisis. Ethical statement This article does not contain any studies with human participants performed by any of the authors. Funding This study was not funded by any funding agency. CRediT authorship contribution statement Kanwal Khan: performed the jobs equally, Writingoriginal draft, jointly wrote the article. Salman Ali Khan: performed the jobs equally, Writingoriginal draft, jointly wrote the article. Khurshid Jalal: performed the jobs equally, Writingoriginal draft, jointly wrote the article. Zaheer Ul-Haq: Formal analysis, Data curation, has contributed to the analysis of the data. Reaz Uddin: Supervision, has conceived the idea and supervised the work, Writingreview & editing, has reviewed and proof-read the manuscript. Declaration of competing interest All Authors declare that they don't have any conflict of interest. Data availability Data will be made available on request. Ethical Dilemma of Error Disclosure in Wrong-Site Craniotomy: A Systematic Review Introduction: Disclosure of errors to patients has long been fully endorsed by professional organizations and bioethicists because of its respect for patient autonomy, enhances informed decision-making and upholds the physician decision to tell the truth. Wrong site craniotomy have massive effects on both the patients and the surgeons. The purpose of this review was to apply moral philosophical analysis to this delicate and important issue that will for sure continually confront neurosurgeons in practice. Methods: A comprehensive selected literature search was performed to establish evidence for informed and evidenced based discussion including the current literature relating to wrong-site craniotomy error disclosure. A total of 232 articles were gathered, but after removing duplicates, editorial comments and correspondences, only two articles met the criteria. Results: In terms of the quality of the gathered evidence, only peer reviewed papers reached the limit. Generally, there is paucity of literature in the ethics of wrong site craniotomy. As a result, only two articles met the criteria (Cohen and Wu et al.). Conclusion: Deontology strongly believed that all major accidental errors which occur when facing patients should be disclosed. This is due to the fact that surgeons are duty-bound to do the right thing including not lying, respecting patients’ dignity, practicing beneficence, sympathy and acting with gratitude and conscience without arrogance. Introduction Medical errors are the eight leading causes of death in the United States accounting for annual patient fatalities that will equivalently result in three jumbo jets crashing every two days. 1 Error disclosure is very fundamental to error prevention. Clinicians face a dilemma when deciding on whether and how harmful medical or surgical errors were for patients. This is issue is because of their fear of lawsuits and self-perception of incompetence and unskillfulness which could be dispelled by organizational cultures emphasizing safety rather than blame. Harvard Medical Practice found out that more than 70% of errors resulting in harmful or adverse events were considered to be secondary to negligence and more than 90% were judged to be preventable. 2 As most errors are never voluntarily reported, there is always significant failure of systems and mechanism set ups in a complex health related environment to improve patient's safety. Healthcare providers are usually extremely embarrassed about their errors or mistakes. In most cases, this issue leads to making furious attempts. They actually immediately shift the blame to someone else or something else to conceal their mistakes and in other words to defend themselves. Disclosure of errors to patients has long been fully endorsed by professional organizations and bioethicists because it respects patient autonomy, enhances informed decision making and upholds the physician decision to tell the truth .Although the majority of clinicians endorsed the principle of medical error disclosure to patients, evidence reveals that such disclosure may be uncommon. Two national surveys highlighted how huge the gap between the principle and practice is and suggests that physicians may be disclosing as few as 30% of harmful errors to the patients. 2 Wrong site surgery may be the most regrettable and harmful error a surgeon can commit. It can actually lead to devastating outcomes, the consequences extend beyond the patient's imagination leading to serious legal and professional implications. Neurosurgery is the third most vulnerable specialty to wrong site surgery behind orthopedic and general surgery. 3 These specialties are consistent with the potential risk factors for wrong site surgery such as multiple surgeons may be involved in a case, and time pressures causing verification steps to be rushed or skipped. Wrong site craniotomy effects both patients and surgeons and will lead to definitely further exacerbating clinical outcomes. Cohen et al. 4 carried out an in-depth case series on wrong-site craniotomy: analysis of 35 cases and systems for prevention. Despite the implementation of preoperative checklists and surgical time-outs, wrong-site surgery still occurs today. The authors searched medical, legal and media databases including Medline, LexisNexis, Westlaw (from 1966 to July 1, 2009) and medial searches and also contacted medical licensing boards which yielded 34 cases of wrong-site craniotomy. They surveyed 50 medical licensing boards by using a brief survey instrument administered by email to the medical councils. Four categories of factors that contribute to wrong-site craniotomy were revealed after analyzing the cases. The categories are as follows: communication breakdowns, inadequate preoperative checks (technical factors and imaging) and human errors. This study is an examination and discussion of a case study in which an ethical dilemma presents the results of the wrong-site craniotomy error disclosure in a patient. This topic has been chosen as it is significant to the author, the occurrence of it is unique in neurosurgery and has caused a great deal of news and concern regarding its ethical and moral dilemma. The aim of this review is to evaluate and assess current evidence regarding the application of the Deontological theory in error disclosure of wrong-site craniotomy using a case study. Authors examined the duty of physicians and surgeons (with particular interest in neurosurgery) to disclose errors they had committed from a less paternalistic view using the moral philosophy of Immanuel Kant. Methods A comprehensive literature search was performed to establish evidence for an informed and evidenced based discussion. Current literature, relating to wrong-site craniotomy error disclosure or wrong site neurosurgery error disclosure were gathered by using databases including Medline and EMBASE following Preferred items for Reporting of Systematic Review and Metaanalysis (PRISMA). The keywords employed in this search strategy included deontology, wrong-site craniotomy, error disclosure, and ethical dilemma. The inclusion criteria included any type of study relevant to the review, studies on adult human patients, papers published in English, and seminal papers relevant to this study. Editorials, comments, and correspondences were excluded. Results In terms of the quality of the gathered evidence, only peer reviewed papers had the required quality. Generally, there is insufficiency in regards to the literature of the ethics of wrong site craniotomy. Actually, only two articles met the criteria (Cohen and Wu et al. 4,5 ). A total of 232 articles were gathered, but after removing duplicates, editorial comments and correspondences only two articles met the criteria (Fig 1). Figure 1. Search Flow Diagram Discussion The deontological theory implies that medical practitioners should adhere to their obligations and duties when analyzing ethical dilemma. This means that a person will perform his obligations as upholding one's duty is considered to be ethically correct. 6 Furthermore, the moral obligation for surgeons to openly recognize their fallibility and attempt to develop tools and strategies to minimize medical errors, may lead to the need for disclosure. The deontological theory was at first defined by a German philosopher named Immanuel Kant who spent most of his life in Konigsberg, Prussia non-relevant studies This theory greatly contributed to the development of the deontological theory. It has been well known that this theory is one of the main theories at the foundation and origin of medical bioethics. Two main types of normative ethical system exist which are: teleological or consequence-based, and deontological or duty-based. The main teleological is the Unitarianism theory which is based on the research and writing of two British philosophers, Jeremy Bentham and John Stuart Mills. 7 Also, the main deontological theory is based on what Immanuel Kant wrote and was latter translated into several volumes. 7 He believes that the moral worth of an action is not at all related to the outcome it brings. It is actually related to the action being done with a sense of duty or obligation. He states "the moral worth of an action does not lie in the effect expected from it and so too does not lie in any principle of action that need to borrow its motive from this expected effect. 8 What this really means is to behave not just with the sense of duty or obligation or any other motive, as duty or obligation is to actually do the right thing. In deontology, some duties are absolute. These are the obligations to do certain types of actions. Kant named this general type of obligation "categorical imperative". These actions are imperative because they fall within a certain category. Furthermore, Kant states that there is only a single categorical imperative or unconditional command governing the morality of human behavior. This command says "act only on the maxim through which you can at the same time will it to become a universal law". The Principle of the Doctrine of Virtue, is to "act on a maxim through which you can and at the same time will that it should become a universal law". 8 The medicine ethos is generally built on deontology principle: no one will deny that doctors have duties. In several ways deontology represents the voice of the patient. Healthcare professionals must always keep their promises and must never lie but sometimes this is not enough today's complicated world. Wrong-site craniotomy is a huge ethical issue characterized with some level of difficulty in decision making. Medical ethics in practice is simply the application of moral principles to specific situations requiring solutions. This discussion attempts to apply the deontology theory to ascertain how far error disclosure can go and also any lessons which can be learned for future reference and reflective practice. Four major principles have been derived from the moral theories including beneficence, non-maleficence, autonomy and justice which are very popular among health care practitioners and are accepted as moral ethical codes guiding morality in decision making 9 . "Beneficence" actually

guides the practitioner to help his patients and leads them not to harm them (non-maleficence), "Justice" is the principle of non-discrimination. Physicians must work for the welfare of the entire society. "Autonomy" refers to the respect for patients and their rights include informed consent, professional secrecy and respect for the independence of his colleagues' professional decisions. 10 There is still the question of whether medical errors should be disclosed to patients and to what extent irrespective of the outcomes, or whether they are obvious to the patients or not. In our case study of wrong-site craniotomy which is a serious and obvious error, to not disclose would mean telling a blatant and obvious lie which is unacceptable to deontology ethics. A lie is defined as merely an unintentional untrue declaration of another. It actually harms another most of the time, even if that other is not another individual, nevertheless humanity generally, inasmuch as it makes the source of right unusable. It is said that the fruitfulness has a specific duty which makes no distinction between those of whom has this duty and to whom one can exempt oneself from it. This is due to the fact that it is actually instead an unconditional duty which one holds in all relations. 8 There is nothing from the Deontological theory which allows neurosurgeons to hide this type of error from a patient. If the patient asks, "did everything go well?", not disclosing the error would be considered as an act of lie, deception and preservation of narrow professional interest over the well-being of patents, which is a breach of professional ethics. On the other hand, if the patient doesn't ask, it's the surgeon's duty to disclose major errors. Physicians may tend to lie for the fear of lawsuits and other professional consequences. However, the Deontological theory insists physicians to disclose medical errors no matter what the consequences or outcomes will turn out to be. The Kant deontology theory is not absolute and inflexible; it may allow room for times when conflicting roles arise. This theory describes two domains of obligation. When these two domains conflict with each other, the applied philosophy says that the stronger domain of obligation has priority. For example, when a physician's opinion about error disclosure in a certain condition where the patient is completely unaware of, and as a result if becoming aware of the error might do more harm than good by the anxiety and lack of evidence it evokes, the stronger ground of obligation will reside in not telling the error while the weaker ground of obligation will reside in telling the truth. Considering the ethical dilemma in wrong site craniotomy, there is a link between the decision-making process and management risk application while there may be other cognitive and psychological connections. Also, behavioral associations and verbal communication interactions occur between groups and individuals in the course of rationalization of problems. The theoretical concepts that effect and guide the process are really considered. 11 The deontology theory has been significantly influential in the development of ethics to guide the moral conductions of all health careers. 12 One important discussion which has been mentioned by Kant in the deontology theory is the kingdom of end. This discussion states that "Act in such a way you will always treat … another person, never simply as a means but always at the same time as an end". 8 What this actually means is that it is acceptable to benefit from an interaction with people as far as it's not the sole motivation of the specific interaction. Despite a global significant comprehensive generalizability and application of generic codes of professional, ethical and moral conduct, the wrong-site craniotomy still remains a controversial and abnormal cosmetic outlook condition. Hence the neurosurgical community must embrace a culture of openness and understanding towards errors to move forward and ultimately improve the quality of patient care. Wrong-site craniotomies are errors that are eradicable. 4 Wu et al. 5 carried out a survey on disclosing medical errors to patients, which is a reflection of the deontology theory. They found out that patients will favorably respond to physicians who apologies and accept responsibility for medical errors than those who do not apologies or give ambiguous responses. Results revealed that disclosure does not seem to increase the tendency to sue among those who had been aware of an error. Nevertheless, the ethical imperative should compel a full disclosure to patients who suffered preventable harms. In this case study, the surgeon displayed full and strict deontological theory in practice, took full responsibility despite the lawsuit and negative outcome, above all settled out of court. There are several assertions that support and refute the importance, usefulness and limitation of deontology in the consideration of ethical dilemma. Most ethicists feel that deontology is less useful in difficult situations and circumstances. For example, in triage deontology cannot tell what duties must be prioritized over others since some duties are absolute. However, the deontological theory was set up in order to establish a uniform medical practice at a behavioral level with the ability to generate the necessary skills of both doctors and patients. The professional and also human behavior of the physician who was considered to be worthy and optimal by his guild has been expressed in the medical deontological code. Discounting the outcomes as a valid factor in evaluating the morality of an action is considered to be a big criticism the of deontology ethics theory. Obviously it's not appropriate to completely ignore the outcomes, but at the same time it's not knowledgeable to rely on an outcome sole as in utilitarianism and consequentialism. Deontological theory is weaker when it comes to inform us about how to live or develop a character of virtue. Disclosure of large-scale adverse events is a challenging dilemma, not all the events were alike and this difference has implications for disclosure. Disclosure should be a norm, even with low probability of harm. Not much detail was said about the mechanism and steps for disclosure. Although the majority of the cases had disclosure to patients and authority, some other were not. Deontological ethical error has been extremely significant in the key development of bioethical theory to guide the moral conduct of surgeons and other members of the health care profession. Deontology strongly believes that all the major accidental errors that occur when facing patients should be disclosed. This is due to the fact that surgeons are duty-bound to do the right thing including not lying, respecting patients' dignity, practicing beneficence, sympathy and acting with gratitude and conscience without arrogance. Above all, never treating a patient as a means to an end but an end in itself. Three main ways surgeons can deal with errors are acknowledge the error, apologies to the patient and family as the case may be and acquire knowledge that will help to avoid same error in the near future. Conclusions Evidence based approach is of paramount importance in interpretation of ethical theory to modern day multidisciplinary practice. Deontology theory can initiate meaningful debate to facilitate robust decisionmaking. Regardless of which ethical theory or combination of ethical theories a neurosurgeon decides consciously or unconsciously to employ to facilitate decision-making, the key focus should always be the patients' best interests. Some interpret that deontology may allow withholding of information from patients especially if disclosing such information will do more harm than good, but purists will not agree with this. However not all answers are on this controversial issues hence further studies on deontology of error disclosure are needed. This study is characterized with the lack of evidenced based articles on deontology and error disclosure, which is concerned to be a significant limitation of this study. Ethical Approval Not applicable. Authors' Contributions TA and AZ contributed equally to this study. Immunity to adult Heligmosomoides polygyrus (Nematospiroides dubius): survival or rejection of adult worms following transplantation to mice refractory to larval challenge ABSTRACT Experiments were carried out to explore the survival of 14-day adult H. polygyrus following transplantation to mice of four strains, immunized by various protocols. Adult worm establishment and survival was unimpaired in CFLP mice which were totally refractory to larval challenge. Transplanted adult worms were also successful in NIH mice immunized by the 9-day abbreviated infection regime. However, NIH mice exposed to irradiated larvae or subjected to the divided primary infection, expelled transplanted adults. The 9-day abbreviated infection was further examined in SJL and (C57 Bl10 X NIH) F1 mice which expel adult worms during a primary infection and although this regime was unsuccessful in causing NIH mice to reject adult worms, expulsion of adult worms was accelerated in SJL and F1 mice. The survival of adult H. polygyrus was discussed in the context of stage-specific immunity and the delicate balance between the immunogenic stimuli from developing larvae, the immunomodulatory activities of adult stages and the host's genetically determined capacity to respond to these opposing signals. INTRODUCTION Heligmosomoides polygyrus is a long-lived parasite in most laboratory strains of mice and is known to survive for 8-10 months following a single oral administration of infective larvae (EHRENFORD, 1954;KEYMER & HIORNS, 1986;BEHNKE et al. 1987). The survival of the parasite is dependent on a delicate balance between the immunogenic stimuli from developing larvae, the immunomodulatory activities of adult worms (BEHNKE, 1987) and the host's genetically determined capacity to respond to these opposing signals (BEHNKE & ROBINSON, 1985). BARTLETT & BALL (1974) concluded that H. polygyrus differed from most other intestinal nematodes in that the fecundity and survival of adult worms seemed to be completely unaffected by acquired resistance, the latter being considered to act only against the larvae, leaving the adult worms totally unimpaired. This remarkable ability of adult H. polygyrus to live in mice totally refractory to larval infections was reinforced by JACOBSON et al. (1982) who demonstrated that immune LAF/J mice, totally resistant to larval challenge, tolerated transplanted adult worms for 2 weeks following implantation. Despite these studies, evidence is available that adult H. polygyrus are not totally resistant to immune responses in the intestine and that the host's genotype is an important factor in this equation. Some mouse strains have been shown to expel adult worms prematurely (LAF/J-CYPESS et al., 1977;SJL-MITCHELL et al., 1982). 'Present address: Department of Immunology, Mayo Clinic, Rochester, Minnesota 55905, USA. 2 Author to whom all correspondence should be addressed. Furthermore, experiments have been reported in which effective anti-adult worm responses have been initiated in mice by appropriate laboratory manipulation. Balb/c mice immunized by intraperitoneal injections of adult worms, when challenged orally by larvae, expelled the mature parasites within 20 days of infection (DAY et al. 7 1979). Even the weak responder strain C 57 B1 H ) was induced to reject adult worms by implantation of mature parasites i.p. and concurrent administration of the adjuvant, pertussigen (MITCHELL & MUNOZ, 1983;MITCHELL & CRUISE, 1984). These studies showed that although adult worms caused chronic primary infections in many widely used laboratory strains of mice and could survive in animals totally resistant to larval infection, the parasites were still susceptible to host effector mechanisms. In the experiments reported in this paper, we began by re-examining the effectiveness if i.p. implanted adult worms in eliciting responses against adult worm challenge. Adult worms were found to survive in CFLP mice, irrespective of the immunizing protocol used even when the mice were resistant to larvae. It was then necessary to compare the efficacy of other immunizing regimes, with well established properties of inducing strong acquired resistance to H. polygyrus in mice and to determine the influence of host genotype on the survival of transplanted adult worms in mice immunized by these procedures. Animals Four strains of female mice were used in the present study. Randomly bred CFLP, syngeneic NIH and (C 57 B1 H) X NIH)F, were all bred in the departmental animal house. SJL mice were purchased from Harlan Olac Ltd., Bicester, Oxon. All the animals were maintained under conventional animal house conditions with access to food and water ad libitum. Parasite The methods used to maintain H. polygyrus, infect mice and recover worms at autopsy have all been described previously (JENKINS & BEHNKE, 1977). Faecal egg counts were monitored as reported by BEHNKE & PARISH (1979). A detailed description of the

methods used to transplant adult worms was given by BEHNKE et al. (1983). Worms for transplantation were obtained from CFLP mice infected with 250-400 L 3 (third-stage larvae), 14 days earlier. Immunization of mice Several different methods were used to immunize mice to larval or adult stages of H. polygyrus. Most of these have been described previously but briefly the following procedures were used: Experiments 1-3. Adult worms were isolated, washed three times in warm sterile saline and injected into the peritoneal cavity of mice in 0-5 ml of sterile saline using a sharp 18 gauge hypodermic needle. Larvae were cleaned by repeated washing in saline and were injected in 0-5 ml of sterile saline through a 21-gauge needle. Both larvae and adult worms were killed prior to injection by freezing to -40°C and thawing, twice in rapid succession. Experiment 4. The divided primary infection protocol adopted was described by BEHNKE & WAKELIN (1977). Mice were given 100 L 3 larvae on days 0, 2, 4, 7, 9, 11 and were treated with pyrantel embonate (Strongid-P paste -Pfizer) on days 15, 22,29, 35 and 42. The animals were challenged on day 51. This protocol allows the overlapping development of larval stages, prolonging larval exposure in the intestine and limits adult worm populations by the anthelmintic treatments administered from day 15 onwards. It is extremely effective at inducing acquired resistance to larval challenge. Experiments 5, 8 and 9. The 9-day abbreviated infection regime used was described by BEHNKE & ROBINSON (1985). Mice were given a single dose of L 3 on day 0 and were treated with pyrantel on days 9, 10, 14 and 21. Challenge infections were administered a week later. This protocol allows the synchronized normal development of L 3 and L 4 but the anthelmintic treatment from day 9 onwards removes the adult worms from the gut lumen as they emerge from the submucosa without affecting parasites which are still developing in the tissues. This regime has been found to be very effective in discriminating between weak and strong responder mouse strains. Experiments 6, 7 and 9. Immunization by irradiated larvae was carried out as described by BEHNKE et al. (1980) and HAGAN et al. (1981). Infective larvae were exposed to 25 krad of gamma radiation from a Cobalt 60 source. Mice were given 250 L 3 on day 0 and were challenged on day 21 (Experiment 6), day 24 (Experiment 7) or day 28 (Experiment 9). This protocol does not permit the development of adult parasites but the L 3 and L 4 stages are believed to persist as arrested forms in the intestinal tissues (ALI & BEHNKE, 1985). Statistical analysis of results Faecal egg counts are presented as the mean value of 4 counts on each 1 sample of faeces. Worm counts are given as the mean worm burden ±S.E.M. When applicable the non-parametric Mann-Whitney U test was used to compare groups for significant difference (SOKAL & ROHLF, 1969) and a value of P<0-05 was considered to represent a significant difference. Survival ofH. polygyrus in CFLP mice immunized by 3 intraperitoneal injections of adult worms or 3 doses of L 3 stages Initially, several exploratory experiments were carried out in CFLP mice in order to determine whether i.p. injections of adult worms would sensitize mice and enable them to reject a challenge infection with transplanted adult worms. These preliminary studies confirmed that adult H. polygyrus could survive in mice resistant to larval challenge. Table I summarizes the results from three such experiments in which the treatments used to induce resistance were 50 adult worms injected i.p. on days -25, -23 and -21, 500 L 3 injected on the same days or 200 L 3 given orally, again on days -25, -23 and -21. All the mice, together with a naive control group, were treated with pyrantel 14 and 7 days before challenge to eliminate worms which had established in the intestine and the animals were challenged either with 50 adult worms by laparotomy or 200 L 3 orally. Adult worms survived without loss for 56 days in mice which showed varying levels of resistance to larval challenge (Table I). When 3 doses of 50 adult worms injected i.p. were used, up to 28-5% protection from larval challenge was observed (Experiment 3, groups BvD). Mice immunized by 3 doses of 500 L 3 given i.p. or 200 L 3 given orally showed respectively 47-7% (Experiment 3, groups BvF) and 62-5% (Experiment 3, groups BvH) reduction in parasite survival to day 56. Curiously, whilst these treatments failed to cause the rejection of transplanted adult worms, adult stages of H. polygyrus developing from the challenge inoculum were rejected more rapidly than in naive control mice. Thus in Experiment 3, mice immunized by 50 adult worms, 500 L 3 i.p. and 200 L 3 orally sustained respectively a further 45%, 44% and 46% reduction in worm burden between day 14 and 56. In contrast, naive control mice lost only 19-9% of their worm burden over the same period (group AvB). Survival and rejection of H. polygyrus in NIH and CFLP mice immunized by 3 different immunizing protocols The preceding experiments established that transplanted adult worms could indeed survive in CFLP mice immunized by treatments which elicited varying degrees of resistance to larval challenge. However, adult worms which survived the tissue phase of infection in immunized mice were less successful at avoiding expulsion since some were rejected within 8 weeks of infection. Thus there appeared to be an intrinsic difference in the capacity of these distinct adult worm populations to overcome the host's intestinal immune response. In an effort to explore further the resistance of transplanted adult worms to host effector mechanisms, a series of experiments was carried out using three further immunizing protocols, all known to stimulate strong acquired resistance to a challenge infection with H. polygyrus. Three of these experiments were carried out using NIH mice, a strain considered a strong responder to this parasite. In each experiment naive mice and mice immunized by the relevant immunizing protocol were challenged either by transplanted adult worms or by infective larvae given orally. The course of infection was then followed by faecal egg counts and parasite survival was assessed by worm recovery. The results are summarized in Table II. As can be seen, all three procedures were highly immunogenic in NIH mice and produced respectively 98-1, 93-8, and 98-9% protection against larval challenge. The protocols differed, however, with respect to their efficacy in mediating resistance to challenge by transplanted adult worms. Two protocols, irradiated larvae (Experiment 7) and the divided primary infection (Experiment 4) caused 80-2 and 97-4% loss of adult worms respectively. Faecal egg counts (Fig. 1) showed that in Experiment 7, adult worms established in mice immunized by irradiated larvae, initially producing high egg counts, but egg production dropped promptly and stayed depressed suggesting that many of the adult worms had been expelled within 2 weeks of transplantation. In control mice (Experiment 7, group A) faecal egg counts were more steady and the parasites survived for 6 weeks post transplantation. Faecal egg count data is not shown for Experiment 4, but no eggs were detected in the faeces of group B mice (immunized by the divided primary infection) on days 13, 20 and 27, whereas the naive mice receiving adult worms by laparotomy (group A) had counts of 800, 1200 and 1100 on these days. In contrast to NIH mice, CFLP mice were unable to reject transplanted adult worms when immunized by irradiated larvae (Experiment 6) despite the fact that the same protocol rendered CFLP mice almost totally refractory to larval challenge (95-9% protection). Furthermore, NIH mice were unable to reject adult worms when immunized by the 9-day abbreviated infection regime (Experiment 5). Again the latter protocol was shown to be highly immunogenic and elicited 93-8% protection against larval challenge. Indeed, in order to allow for a possible delayed response to transplanted adult worms, this experiment was continued for 10 weeks, but at autopsy worm counts were not significantly different in groups A and B. Nevertheless faecal egg counts in group B were depressed relative to group A throughout the period of observation and were very low in the 6-10 weeks of infection suggesting that despite the hosts failure to reject the worms, an anti-parasite response was initiated. Rejection of adult H. polygyrus from (C 57 Bl 10 X NIH) Fj hybrid and SJL mice. Recent studies have shown that three strains of mice expel adult H. polygyrus developing from orally administered L 3 within 10 weeks of primary infection (ROBINSON et al., in press). In (C 57 Bl 10 X NIH)Fi hybrids this is a dose-dependent phenomenon, low intensity infections being expelled earlier than high intensity infections. SJL mice, however, expel H. polygyrus in a dose-independent fashion usually within 7 weeks of larval administration. In the final section of this paper, we report the results of experiments carried out to determine whether the immunizing protocol, which failed to elicit resistance to transplanted adult worms in NIH, would accelerate worm rejection by these two strains of mice. Experiment 8 was designed along the lines of Experiment 5. Thus 27 Fj mice were arranged into 4 groups A(n=4), B(n=6), C(n=9) and D(n=8) which were treated identically to the mice in Experiment 5, except that groups C and D were challenged by 150 L3 larvae and not 50. This experiment was monitored by faecal egg counts as shown in Fig. 2 and the mice were finally killed for worm counts on day 56 post challenge. Naive Fi mice receiving transplanted adult worms (Group A), produced eggs in their faeces throughout the period of observation, but the egg counts declined towards week 5 and at autopsy 14-8±7-4 worms were recovered. Two of the mice still had high parasite numbers, the remainder had lost the majority of worms. In contrast immunized mice receiving transplanted adult worms (Group B) had rejected all the parasites by day 56 but the worms had established following transplantation as eggs were detected in the faeces from day 4 to 39. Thus ¥i mice immunized by the 9-day abbreviated infection protocol showed enhanced resistance to adult parasites which were rejected prematurely in relation to naive controls. No eggs were detected at any time in the faeces of immunized F, mice challenged with L 3 (group D) and no parasites were recovered at autopsy. Group C which comprised naive F, mice given L 3 orally produced eggs throughout the experiment and at autopsy 28-2±15-5 worms were recovered but the worm burdens were variable and 4 of the 9 mice had expelled the parasites before day 56. Thus in this group also, expulsion of adult worms (developing from L 3 larvae) had been initiated, some mice shedding their worm burden before the termination of the experiment. In the final experiment SJL mice were used. One group was designated the naive control group (A) and the other two groups were immunized, one by the abbreviated infection protocol (B) and the other by irradiated larvae (C). All three groups received transplanted adult worms and the course of infection was monitored by faecal egg counts. Mice in group C ceased producing eggs on day 29, those in group B on day 44, whilst eggs were detected in group A until day 57. At autopsy, the majority of mice were totally without worms. DISCUSSION H. polygyrus is clearly a remarkable parasite whose adult stages survive in situations where larvae cannot form patent infections. Our results have drawn attention again to this property of adult H. polygyrus and we have extended previous studies by exploring the survival of adult worms in mouse strains which have the capacity to expel primary infections with this organism. One interpretation of the results from our experiments could be that they reflect stage-specific resistance e.g. resistance induced by larvae only acting on larval and not adult parasites. Indeed, H. polygyrus expresses stage-specific antigens (PRITCHARD et al., 1984) and in Experiment 3, mice immunized by adult worms did not express anti-larval resistance, but subsequently rejected adult worms developing from the challenge inoculum. However, the same protocol did not accelerate the rejection of transplanted adult worms and although this may seem paradoxical, there are reasons for believing that the two adult

worm populations may be phenotypically distinct. Although no biochemical, enzymatic or antigenic differences have been reported, parasites developing in partially resistant hosts are stunted and less fecund than normal (BARTLETT & BALL, 1974;DOBSON, 1982;SITEPU & DOBSON, 1982) and by analogy with Nippostrongylus brasiliensis might be expected to show other differences (OGILIVIE & JONES, 1973). Such worms may already be partially damaged as a consequence of developing in immune animals and hence may be more susceptible to host effector mechanisms. However, stage-specific immunity is not an explanation which is entirely compatible with our data. Immunization with irradiated larvae induced strong resistance to larval challenge in both CFLP and NIH mice (Experiments 6 & 7, Table II) but allowed transplanted parasites to establish and survive in the former strain but not in the latter. Larvae subjected to 25 krad of irradiation do not develop into mature worms (BEHNKE et al., 1980) and in consequence the mice would have been exposed only to prolonged stimulation by L 3 and L 4 antigens. Both strains, therefore, had comparable experience of larval antigens, but only one acquired an anti-adult worm response. It is also pertinent that NIH mice were unable to expel transplanted adult worms when immunized by the 9-day abbreviated infection. Again, one explanation may be that the 9-day abbreviated protocol elicited anti-larval stage-specific immunity and for this reason we explored the effects of this immunizing protocol in two strains of mice which rank as stronger responders than NIH mice ((C 57 Bl 10 X NIH)F, and SJL strains; BEHNKE & ROBINSON, 1985). In both strains, the protocol accelerated the rejection of transplanted adult worms and we conclude that stage-specific immunity was unlikely to have been involved, unless there were genetically determined qualitative differences between the strains in recognizing and responding to the inducing signals from the inoculum. Mice exposed to the 9-day abbreviated infection would have experienced only larval stages of H. polygyrus, but would not have been subjected to such prolonged stimulation as mice exposed to the irradiated larvae (ALI & BEHNKE, 1985). Irradiated larvae stimulate strong resistance to challenge in both NIH and C 57 B 10 mice (the latter is a weak responder strain to H. polygyrus) but the 9-day abbreviated infection is only effective in the former (BEHNKE & ROBINSON, 1985). Therefore, the explanation for the different efficacies of the two immunizing protocols in inducing anti-adult worm responses in NIH mice may reside in the varying intensity of stimulation which they provide. An alternate explanation for our results may lie in considering the host's response to the immunomodulatory mechanisms used by adult worms for evading immunity (reviewed BEHNKE, 1987). CFLP mice may have been more severely compromised by the immunodulatory effects of adult parasites than NIH mice and hence worms were able to survive only in immunized individuals of the former strain. There is further evidence in support of this suggestion. Thus adult worms can readily be shown to depress homologous immunity in CFLP mice immunized by irradiated larvae (BEHNKE et al., 1983) and similar results have been found in C 57 B1 H1 mice, but we have been unable to repeat these observations in NIH mice. In one experiment, 104 adult H. polygyrus resulted in 12% reduction of immunity elicited by the irradiated vaccine in NIH mice, whereas in C 57 Bl ln only 13 adult worms caused 84% depression (Crawford, Behnke & Pritchard, unpublished observations). It is quite clear from our results and those of others (JACOBSON et al., 1982) that under a variety of circumstances adult H. polygyrus can evade host protective responses, but the parasites are not always successful. An important additional factor, not investigated previously in this context, appears to be the host genotype. Mouse strains such as SJL, (C 57 B1 U ) X NIH)F, and LAF,/J have the capacity to expel adult worms and if immunomodulation is an important component of the parasite's evasive strategies, it follows that these strains must benefit from a genetically determined insusceptibility to parasite mediated immunomodulation. The latter is an attractive, testable but as yet unproven hypothesis. It is interesting to note here that, the expulsion of primary infection adult worms from SJL mice is not accompanied by an intestinal mast cell response, such as that mounted by mice of this strain to Trichinella spiralis or secondary infections with H. polygyrus (DEHLAWI et al., 1987). SJL mice can therefore respond with a mastocytosis but fail to do so during a primary infection, presumably because the response is depressed by adult worms. In other words, H. polygyrus are partially successful in immunomodulating in SJL mice, in so far as the mast cell response is incapacitated, but the fact that the worms are lost implies that other component processes of the overall response in the intestine are less severely affected. These observations have not yet been extended to encompass other mouse strains which expel adult worms but such experiments are currently in progress. The key to understanding the survival strategy of H. polygyrus and hopefully other long-lived nematode parasites, will be to identify qualitative differences in response phenotype associated with survival or loss of adult parasites and to link these to genetic differences between the mouse strains. Personalized Three-Dimensional Printed Models in Congenital Heart Disease Patient-specific three-dimensional (3D) printed models have been increasingly used in cardiology and cardiac surgery, in particular, showing great value in the domain of congenital heart disease (CHD). CHD is characterized by complex cardiac anomalies with disease variations between individuals; thus, it is difficult to obtain comprehensive spatial conceptualization of the cardiac structures based on the current imaging visualizations. 3D printed models derived from patient’s cardiac imaging data overcome this limitation by creating personalized 3D heart models, which not only improve spatial visualization, but also assist preoperative planning and simulation of cardiac procedures, serve as a useful tool in medical education and training, and improve doctor–patient communication. This review article provides an overall view of the clinical applications and usefulness of 3D printed models in CHD. Current limitations and future research directions of 3D printed heart models are highlighted. Introduction Computed tomography (CT), magnetic resonance imaging (MRI), and echocardiography represent commonly used imaging modalities in the diagnostic assessment of congenital heart disease (CHD). These imaging techniques allow for generation of two-dimensional (2D) and three-dimensional (3D) visualizations, which play an important role in understanding the complexity of CHD and assisting pre-procedural planning of cardiac procedures. Despite useful information provided by these imaging modalities, the images are still limited to be viewed on 2D screens which is very different from the physical models that offer realistic visualization of 3D spatial relationship between normal and anomalous anatomy. 3D printing overcomes this limitation by creating patient-specific or personalized medical models [1][2][3]. The tactile experience offered by 3D printed models is another advantage over traditional image visualizations as the physical models enable comprehensive evaluation of anatomical and pathological structures which cannot be obtained by other methods [4]. 3D printing has been increasingly utilized in the medical field with studies confirming its clinical value and usefulness in many areas, ranging from medical education to pre-surgical planning and simulation of complex surgeries, and to patient-doctor communication [5][6][7][8][9][10]. In particular, personalized 3D printed models have been shown to offer valuable information for treating patients with CHD due to complexity and anatomic variation associated with this disease. Most of the current reports on 3D printing in CHD are dominated by isolated case reports or case series, with only a few single or multi-center studies and randomized controlled trials (RCT) available in the literature. This review aims to provide an in-depth overview of the current applications of 3D printed models in CHD, with limitations and future directions briefly highlighted. Image Post-Processing and Segmentation Process for Three-Dimensional (3D) Printing in Congenital Heart Disease (CHD) The first step to generate a 3D printed heart model is to undergo a series of image post-processing and segmentation of volumetric data, which are commonly acquired with cardiac CT or MRI imaging. While high-resolution original datasets are important for accurately isolating the desired anatomy of interest and pathology from surrounding structures, segmentation of cardiac structures remains challenging due to complexity of cardiac features, especially in the CHD cases. Different software is used for segmentation, with Mimics (Materialise HQ, Leuven, Belgium) being the most commonly used commercial software and 3D Slicer (Brigham and Women's Hospital, Boston, Mass) as the most common open-source tool. Several review articles have provided excellent description of details about the steps required from data acquisition to image post-processing and segmentation [11][12][13][14]. Figure 1 shows the steps to create 3D printed models from data acquisition to image post-processing and segmentation. Steps involved in fabrication of 3D printed heart models. CTA-computed tomography angiography; CMR-cardiac magnetic resonance; 3D-three-dimensional. Accuracy of 3D Printed Heart Models The most important part of creating 3D printed models is to ensure that 3D models accurately delineate anatomical structures and pathologies since the model accuracy is essential for treatment planning [15]. Current research evidence indicates that 3D printed heart models are generally accurate [16], and this has been validated by other studies, either based on case reports/series or singleor multi-center studies [17][18][19][20][21][22][23]. In most of the studies, model accuracy is determined by the degree of agreement between the measured dimensions of the 3D printed model and the dimensions of original source images, usually using cardiac CT, MRI, and sometimes using rotational angiography or echocardiography [16][17][18]22], or intraoperative findings [19]. Currently, there is no standardized method to measure the dimensions of the 3D printed heart models. Most of the studies carried out measurement using calipers on the physical 3D printed models [17,20,21]. Only a few studies conducted measurement on the standard tessellation language (STL) file [18] and conducted CT scan on the 3D printed model for measurement [16]. The authors claimed it is easier to replicate the exact plane for measurement comparison, hence improving the accuracy of the results [16,18]. Despite limited studies available in the literature regarding quantitative assessment of 3D printed heart models, the accuracy of 3D printed heart models is within 1 mm in terms of dimensional differences when compared to the original images. Lau et al. compared model accuracy between contrast-enhanced CT images of the 3D printed heart model (Figures 2 and 3) and original cardiac CT images in 10 anatomical locations including ventricular septal defect (VSD) [16]. High accuracy was achieved between these measurements by only 0.23 mm difference in average. Ma et al. in their study comprising 35 patients of Tetralogy of Fallot (ToF), compared measurements of VSD sizes between 3D printed models and intraoperative findings with no significant differences found (mean value ± standard deviation: 14.98 ± 1.91 vs. 15.11 ± 2.06 mm, p > 0.05) [19]. This is further confirmed by a multi-center study showing the model accuracy. Valverde et al. recruited 40 patients diagnosed with complex CHD from 10 international centers in their prospective study [21]. 3D printed models were created from CT or MRI images, and they were found to be highly accurate with mean differences of 0.27 ± 0.73 mm between measurements performed on the 3D printed models and CT/MRI images. Accuracy of 3D Printed Heart Models The most important part of creating 3D printed models is to ensure that 3D models accurately delineate anatomical structures and pathologies since the model accuracy is essential for treatment planning [15]. Current research evidence indicates that 3D printed heart models are generally accurate [16], and this has been validated by other studies, either based on case reports/series or single-or multi-center studies [17][18][19][20][21][22][23]. In most of the studies, model accuracy is determined by the degree of agreement between the measured dimensions of the 3D printed model and the dimensions of original source images, usually using cardiac CT, MRI, and sometimes using rotational angiography or echocardiography [16][17][18]22], or intraoperative findings [19]. Currently, there is no standardized method to measure the dimensions of the 3D printed heart models. Most of the studies carried out measurement using calipers on the physical 3D printed models [17,20,21]. Only a few studies conducted measurement on the standard tessellation language (STL) file [18] and conducted CT scan on the 3D printed model for measurement [16]. The authors claimed it is easier to replicate the exact plane for measurement comparison, hence improving the accuracy of the results

[16,18]. Despite limited studies available in the literature regarding quantitative assessment of 3D printed heart models, the accuracy of 3D printed heart models is within 1 mm in terms of dimensional differences when compared to the original images. Lau et al. compared model accuracy between contrast-enhanced CT images of the 3D printed heart model (Figures 2 and 3) and original cardiac CT images in 10 anatomical locations including ventricular septal defect (VSD) [16]. High accuracy was achieved between these measurements by only 0.23 mm difference in average. Ma et al., in their study comprising 35 patients of Tetralogy of Fallot (ToF), compared measurements of VSD sizes between 3D printed models and intraoperative findings with no significant differences found (mean value ± standard deviation: 14.98 ± 1.91 vs. 15.11 ± 2.06 mm, p > 0.05) [19]. This is further confirmed by a multi-center study showing the model accuracy. Valverde et al. recruited 40 patients diagnosed with complex CHD from 10 international centers in their prospective study [21]. 3D printed models were created from CT or MRI images, and they were found to be highly accurate with mean differences of 0.27 ± 0.73 mm between measurements performed on the 3D printed models and CT/MRI images. Currently, there is a lack of measurement comparison between 3D printed heart models and STL file, hence, it is unknown whether there is any dimensional error introduced during 3D printing process. This needs to be addressed by future studies. 3D Printed Models in Medical Education and Training 3D printed heart models have been shown to serve as a novel teaching tool in medical education and training and this is confirmed by RCT available in the literature [22][23][24][25]. Three of them reported the usefulness of 3D printed models of different types of CHD in medical education [22][23][24]. Table 1 shows details of these three studies and other single-and multi-center reports. Studies by Loke et al. and White et al. investigated how 3D printed models improved pediatric residents' knowledge and understanding of CHD, while the study by Su et al. focused on how 3D printed models improved medical students' knowledge in CHD. In these studies, 3D printed models of VSD and ToF representing simple and complex CHD were created and used for teaching in the test groups (Figure 4), while the control groups were only given the usual lectures with 2D images. Overall results showed significant improvements of residents and medical students' learning and confidence in managing CHD, especially in dealing with complex CHD situations such as ToF as confirmed by White et al. [24]. Furthermore, 3D printed models significantly improved residents' satisfaction and self-efficiency scores when compared to learning from 2D images [22] ( Figure 5). Currently, there is a lack of measurement comparison between 3D printed heart models and STL file, hence, it is unknown whether there is any dimensional error introduced during 3D printing process. This needs to be addressed by future studies. 3D Printed Models in Medical Education and Training 3D printed heart models have been shown to serve as a novel teaching tool in medical education and training and this is confirmed by RCT available in the literature [22][23][24][25]. Three of them reported the usefulness of 3D printed models of different types of CHD in medical education [22][23][24]. Table 1 shows details of these three studies and other single-and multi-center reports. Studies by Loke et al. and White et al. investigated how 3D printed models improved pediatric residents' knowledge and understanding of CHD, while the study by Su et al. focused on how 3D printed models improved medical students' knowledge in CHD. In these studies, 3D printed models of VSD and ToF representing simple and complex CHD were created and used for teaching in the test groups ( Figure 4), while the control groups were only given the usual lectures with 2D images. Overall results showed significant improvements of residents and medical students' learning and confidence in managing CHD, especially in dealing with complex CHD situations such as ToF as confirmed by White et al. [24]. Furthermore, 3D printed models significantly improved residents' satisfaction and self-efficiency scores when compared to learning from 2D images [22] ( Figure 5). 3D printed models resulted in significantly higher satisfaction scores than 2D images (p = 0.03). 3D printed models improved residents' self-efficacy scores in managing ToF, although this did not reach significant difference when compared to 2D images (p = 0.39). Su et al., 2018 [23] RCT: study group participated in teaching seminar including 3D printed models, while control group only attended teaching seminar without having 3D models 63 medical students: 32 in study group and 31 in control group Ventricular septal defect (VSD) Significant improvement in VSD learning and structure conceptualization in the study group compared to the control group (p < 0.05). White et al., 2018 [24] RCT: study group was given 3D printed models in addition to lectures, while control group received only the lectures 60 pediatric residents:31 in study group and 29 in control group VSD and ToF 3D printed models of CHD improved residents' knowledge and confidence in managing complex CHD such as ToF but did not seem to improve simple CHD such as VSD. Olivieri et al., 2016 [26] Single-center report of 3D printed models for training and simulation 10 3D printed models, 70 clinicians participated in the training sessions Cardiac and vascular anomalies 3D printed models can be used as a simulation training tool for multidisciplinary intensive care providers by enhancing their anatomic knowledge and clinical management of CHD patients. Hoashi et al., 2018 [27] Single-center experience 20 cases DORV and other cardiac anomalies 3D printed heart models improved understanding of the relationship between intraventricular communications and great vessels. Further, 3D printed models allowed simulation of cardiac surgeries by creating intracardiac pathways, thus providing benefits to inexperienced cardiac surgeons. Valverde et al., 2017 [21] Multi-center study consisting of 10 international centers 40 patients with complex CHD DORV (50%) and other cardiac anomalies 3D models were accurate in replicating anatomy. 3D models refined the surgical approach in nearly 50% cases. 3D models resulted in significant change in the surgical plan in 24% of cases. Zhao et al., 2018 [28] Single-center experience 25 patients with 8 in 3D printing group and 17 in control group DORV 3D printed models showed high accuracy in measurements of aortic diameters and the size of VSD when compared to original CT data. 3D printed models significantly reduced ICU time and mechanical ventilation time (p < 0.05). Results from cross-sectional studies are consistent with these findings from RCT [5,6,[30][31][32]. To date, there is sufficient evidence to prove that 3D printed models of CHD play a valuable role in education and training of medical students, pediatric residents, and healthcare professionals in improving their understanding of complex cardiac pathology and increasing their confidence in managing CHD patients. Figure 5. Impact of 3D printed heart models on medical education. Improvement was found in residents' knowledge on congenital heart disease with use of 3D printed models when compared to 2D images. A statistically significant difference was noticed in satisfaction ratings in the group having 3D printed heart models when compared to the control group (p = 0.03). While residents in the 3D printed model group had higher self-efficacy scores, this did not reach significant difference compared to the control group using 2D images/drawings (p = 0.39). Reprinted with permission under the open access from Loke et al. [22]. Results from cross-sectional studies are consistent with these findings from RCT [5,6,[30][31][32]. To date, there is sufficient evidence to prove that 3D printed models of CHD play a valuable role in education and training of medical students, pediatric residents, and healthcare professionals in improving their understanding of complex cardiac pathology and increasing their confidence in managing CHD patients. 3D Printed Models in Pre-Surgical Planning and Simulation Due to complexity of the cardiac conditions with wide variations between individuals with CHD, 3D printed models demonstrate great advantages over traditional image visualizations in presurgical planning and simulation of cardiac surgeries. A recent systematic review has summarized findings from a number of case reports and series with regard to the use of 3D printed models in facilitating preoperative planning and surgical decision-making in CHD cases [33]. Table 1 shows some results from single-and multi-center studies which involved more than 20 cases or participants about the value of 3D printed heart models in this aspect [20,[26][27][28][29]. These studies reported the usefulness of 3D printed heart models from different perspectives. Among all types of CHD, double outlet right ventricle (DORV) and ToF represent the most common types of CHD for fabrication of 3D printed models. This is reported in four out of the five studies mentioned above [21,[27][28][29]. Olivieri et al. created 3D printed models from 10 patients who underwent congenital cardiac surgery due to various cardiac and vascular anomalies [26]. They presented the 3D models to 70 clinicians including 22 physicians, 38 critical care nurses, and 10 ancillary providers. At completion of the cardiac surgeries, all participants underwent a training session of simulating intra-and postoperative care using 3D printed heart models. The use of 3D printed models was found to be more effective than standard verbal hand off with average score of 8.4 out of 10. In total, 90% of participants scored it very highly with regard to the efficacy of 3D printed models in improving cardiac anatomy understanding, surgical understanding, and ability to manage CHD clinically. Two other studies reported utilizing 3D printed heart models in the diagnostic management of patients with CHD [21,27]. Hoashi et al. created 20 3D printed heart models for the purpose of preoperative simulations of cardiac surgeries [27]. Despite realistic and expensive models being produced (each model costs between $2000 and $3000), this study mainly focused on findings related Figure 5. Impact of 3D printed heart models on medical education. Improvement was found in residents' knowledge on congenital heart disease with use of 3D printed models when compared to 2D images. A statistically significant difference was noticed in satisfaction ratings in the group having 3D printed heart models when compared to the control group (p = 0.03). While residents in the 3D printed model group had higher self-efficacy scores, this did not reach significant difference compared to the control group using 2D images/drawings (p = 0.39). Reprinted with permission under the open access from Loke et al. [22]. 3D Printed Models in Pre-Surgical Planning and Simulation Due to complexity of the cardiac conditions with wide variations between individuals with CHD, 3D printed models demonstrate great advantages over traditional image visualizations in pre-surgical planning and simulation of cardiac surgeries. A recent systematic review has summarized findings from a number of case reports and series with regard to the use of 3D printed models in facilitating preoperative planning and surgical decision-making in CHD cases [33]. Table 1 shows some results from single-and multi-center studies which involved more than 20 cases or participants about the value of 3D printed heart models in this aspect [20,[26][27][28][29]. These studies reported the usefulness of 3D printed heart models from different perspectives. Among all types of CHD, double outlet right ventricle (DORV) and ToF represent the most common types of CHD for fabrication of 3D printed models. This is reported in four out of the five studies mentioned above [21,[27][28][29]. Olivieri et al. created 3D printed models from 10 patients who underwent congenital cardiac surgery due to various cardiac and vascular anomalies [26]. They presented the 3D models to 70 clinicians including 22 physicians, 38 critical care nurses, and 10 ancillary providers. At completion of the cardiac surgeries, all participants underwent a training session of simulating intra-and post-operative care using 3D printed heart models. The use of 3D printed models was found to be more effective than standard verbal hand off with average score of 8.4 out of 10. In total, 90% of participants scored it very highly with regard to the efficacy of 3D printed models in improving cardiac anatomy understanding, surgical understanding, and ability to manage CHD clinically. Two other studies reported utilizing 3D printed heart models in the diagnostic management of patients with CHD [21,27]. Hoashi et al. created 20 3D printed heart models for the purpose of preoperative simulations of cardiac surgeries [27]. Despite realistic and expensive models being produced (each model

costs between $2000 and $3000), this study mainly focused on findings related to patient's cardiac surgery outcomes, while the value of 3D printed models was briefly mentioned in some sample cases. Specifically, authors concluded that 3D printed heart models did not reduce cardiopulmonary bypass time. In contrast, Valverde et al. conducted a multi-center study and performed both quantitative and qualitative assessments of the role of 3D printed models in clinical decision-making in patients with complex CHD [21]. Forty patients recruited from 10 international centers were included in this prospective study with 3D models fabricated using CT or MRI images. 3D printed models were assessed as to whether they changed the surgical decision (from conservative management to surgical intervention) and whether the surgical plan was modified. In more than half of the cases (52.5%), 3D printed models did not result in any change to the surgical decision. However, 3D printed models showed significant clinical impact on redefining the surgical approach in 47.5% cases. In 25% of cases, after inspection of 3D printed models, the surgical plan was modified with conservative management changed to surgery. As the only multi-center study available in the literature, this study shows the impact of 3D printed models on deciding the best surgical approach. However, more similar studies are desirable to validate this. The other two studies are based on single-center experience reporting the clinical impact of 3D printed models in CHD treatment outcomes [28,29]. Zhao et al. divided 25 patients with complex DORV into two groups, 8 in the 3D printing group and 17 in the control group, with all patients undergoing cardiac surgery [28]. The intensive care unit stay time and mechanical ventilation time in the 3D printing group was significantly shorter than in the control group (p < 0.05). Although the operative duration, cardiopulmonary bypass time, and aortic cross-clamping time in the 3D printing group was shorter than the control group, this did not reach statistical significance (p > 0.05). Similar findings are reported by Ryan and colleagues [29]. The authors presented their single-site three-year experience of using 3D printed models for managing CHD cases. Of 164 models fabricated for different purposes, 79 models covering a range of CHD complexities were selected for surgical planning. When compared to the standard care (without anatomical models) group, the 3D printed heart model group was found to have shorter mean duration in the operative room and lower 30-day readmission and mortality rates. However, it is worthwhile to note that it did not reach statistical significance, and it is likely due to limited study sizes for each CHD types. These reductions in durations could contribute to lower morbidity and mortality associated with management of CHD, although this needs to be validated by further studies. One example would be by investigating the impact of 3D printed models on 30-day post-operative outcome. 3D Printed Models in Doctor-Patient Communication Physician-patient relationship and working alliance plays a crucial role in improving patient adherence, level of satisfaction, and treatment outcomes [34]. Due to complexity and variations of cardiac anatomy in CHD, it is especially challenging in achieving good physician-patient communication (physicians specifically refer to cardiologists and cardiac surgeons in the situation of managing patients with CHD) [35]. Traditional approaches of using diagrams or image visualizations for explanation of complicated cardiac pathologies do not allow doctors to effectively communicate to patients or parents because of difficulty in interpreting 3D conceptualization of spatial relationship between cardiac structures. 3D printed models are able to eliminate this limitation as observers have no restriction in appreciating the spatial relationship between cardiac structures in all dimensions, thus improving doctor-patient communication. A study by Biglino et al. first attempted to quantify the benefit of 3D printed models in doctor-patient communication [36]. Ninety-two parents of patients with CHD were randomly allocated to two groups with 45 assigned to the model group using 3D printed heart models during their visit, and 52 to the control group with no models during consultation. Parents were asked to complete two questionnaires: A first brief questionnaire before their child's consultation and a second brief questionnaire after the consultation with regard to understanding of their child's heart condition, identification of cardiac defects, and clarity of planned intervention or procedure. Both cardiologists and parents rated the 3D printed models as very useful. Despite the improvement in doctor-parent communication, 3D printed models did not lead to improving parents' knowledge and understanding of their child's heart condition. Furthermore, consultations using the 3D printed models were found to be longer than those without the models (21 ± 10 vs. 16 ± 7 min, p = 0.02), although this did not show significant impact on overall duration of the visits. The same group conducted another study determining the impact of using 3D printed heart models on facilitating consultations between doctor and young people with CHD [37]. Twenty adolescent patients with CHD (age range 15-18 years) were included in this study with use of the same approach as stated in the previous study involving completion of two questionnaires, pre and post-consultations with their doctors. Positive responses were found in the study with use of 3D printed models with significant improvements in their knowledge of CHD (p < 0.05), confidence in explaining conditions to others (p < 0.001), and overall satisfaction (p < 0.05) (Figure 6). The majority of participants indicated that the 3D printed models improved their clinical visits, however, 30% of them expressed their concern of feeling more anxious about their heart condition with use of the 3D models (Figure 7). models were found to be longer than those without the models (21 ± 10 vs. 16 ± 7 min, p = 0.02), although this did not show significant impact on overall duration of the visits. The same group conducted another study determining the impact of using 3D printed heart models on facilitating consultations between doctor and young people with CHD [37]. Twenty adolescent patients with CHD (age range 15-18 years) were included in this study with use of the same approach as stated in the previous study involving completion of two questionnaires, pre and post-consultations with their doctors. Positive responses were found in the study with use of 3D printed models with significant improvements in their knowledge of CHD (p < 0.05), confidence in explaining conditions to others (p < 0.001), and overall satisfaction (p < 0.05) ( Figure 6). The majority of participants indicated that the 3D printed models improved their clinical visits, however, 30% of them expressed their concern of feeling more anxious about their heart condition with use of the 3D models (Figure 7). More research is needed to investigate the clinical translation of 3D printed heart models for doctor-patient/parent communication with involvement of different stakeholders including patients, parents, families, clinicians, and other healthcare professionals such as nurses and ancillary providers [28]. Further studies are also required to address the limitations of lack of evidence of clinical followup with regard to the impact of 3D printed models on patient's lifestyle and eventually patient outcomes. Summary and Future Research Directions Personalized 3D printed models of CHD are changing the current practice in the diagnostic management of patients with CHD. 3D printed models have demonstrated advantages over traditional image visualizations in the assessment of complex cardiac structures as observers are able to appreciate various CHD conditions with more confidence. Three main applications of 3D printed heart models have been discussed in this review, including medical education and training, presurgical planning and simulation, and doctor-patient communication. Despite attractiveness of 3D More research is needed to investigate the clinical translation of 3D printed heart models for doctor-patient/parent communication with involvement of different stakeholders including patients, parents, families, clinicians, and other healthcare professionals such as nurses and ancillary providers [28]. Further studies are also required to address the limitations of lack of evidence of clinical follow-up with regard to the impact of 3D printed models on patient's lifestyle and eventually patient outcomes. Summary and Future Research Directions Personalized 3D printed models of CHD are changing the current practice in the diagnostic management of patients with CHD. 3D printed models have demonstrated advantages over traditional image visualizations in the assessment of complex cardiac structures as observers are able to appreciate various CHD conditions with more confidence. Three main applications of 3D printed heart models have been discussed in this review, including medical education and training, pre-surgical planning and simulation, and doctor-patient communication. Despite attractiveness of 3D printed realistic models and promising results associated with their applications, more scientific evidence with high statistical power is needed before 3D printing is widely used in clinical practice. In addition to the lack of large-scaled studies (prospective and multi-center studies), some limitations should be addressed with future technical developments so that 3D printing will be more practicable in medical applications. One of the main limitations in generating 3D printed heart models lies in image post-processing and segmentation of cardiac imaging data, which is exceptionally time-consuming and requires expertise in image analysis. The duration needed to complete the segmentation and image post-processing is highly dependent on segmentation software tools (whether it is powerful enough for automatic segmentation), and researcher's familiarity with the software, as highlighted by two systematic reviews and other review articles (Table 2) [29,35,[38][39][40]. This operator-dependent process is inevitably associated with interobserver and intraobserver variability. This limitation could be resolved with the use of artificial intelligence or machine learning algorithms which are increasingly used in the domain of cardiovascular disease by providing automated image segmentation of coronary plaque lesions or coronary lumen [41][42][43][44], thus improving the workflow efficiency from image acquisition to 3D printing. Table 2. Summary of systematic reviews of 3D printed models in congenital heart disease. Authors Number High 3D-printing costs represent another obstacle, however this is being resolved with the use of low-cost 3D printing materials, provided that the accuracy of the 3D printed models is not affected. A recent study has demonstrated the feasibility of creating accurate 3D printed models using low-cost as opposed to high-cost material ($50 vs. $300) for delineation of cardiac anatomy and defects (Figures 8 and 9) [45]. With further cost reductions in 3D printers and printing materials in the near future, personalized 3D printed heart models will become more affordable to patients with CHD. Table 3 shows different types of 3D printers and printing materials that are available in the 3D printing of cardiac models and strengths and weaknesses of these models corresponding to each type of 3D printers [36,[46][47][48][49]. Batteux et al. 2019 [38] NR 3D printed models in surgical planning in complex CHD cardiac anatomy and disease and can be used to guide surgical planning. Lau and Sun 2018 [29] 28 Clinical value of 3D printed models in CHD 3D printed models accurately replicate cardiac anatomy and pathology and are shown to be valuable in preoperative planning and simulation of cardiac procedures. NR-Not reported. (A) (B) (C) Very few materials that are currently used for creating 3D printed heart models represent elastic properties similar to human tissue which allow for performance of realistic surgical simulation such as cutting and suturing of cardiac structures. Despite softness of the Tango Plus material as shown in Figure 9, the mechanical properties of these materials are still different from biologic heart tissues. Furthermore, current 3D printers generate a static heart model instead of a dynamic organ, therefore, allowing for assessment of morphological cardiac features rather than hemodynamics of the cardiovascular system. Future developments in printing technologies should aim to produce 3D printed dynamic heart models which enable detection of both anatomic and physiological changes during the cardiac cycle [11,39,50]. 3D bioprinting represents another major advancement with the capability of printing biomaterials, 3D printed tissue scaffolds, and 3D printed functional vascular networks [51][52][53]. Bioprinting of patient-specific heart tissues will broaden applications of 3D printing in CHD, although many challenges need to be overcome before it can be translated to clinical applications [54,55]. Currently, no guidelines or recommendations are available regarding the standardized use of 3D printed models in CHD patients. Use of 3D printed models is limited to complex CHD such as DORV ( Figure 10) and ToF as evidenced by anecdotal reports and case series [16,[21][22][23][31][32][33]56,57]. Other clinical benefits of using 3D printed heart models such as its impact on procedural safety and long-term outcomes are still yet

to be investigated. Future research should focus on these areas as they will contribute to the development of clinical recommendations of using 3D printed models routinely in medical practice, therefore having great impact on the treatment of congenital heart disease. Figure 11 presents a summary of the current applications and future directions of 3D printing in CHD. Conflicts of Interest: Authors declare no conflict of interest. The Management of Delayed Onset Cramps Related to Irinotecan during Oncological Treatment Background: Irinotecan-induced gastro-intestinal side effects are well described in the literature. The pathophysiology and treatment of diarrhea, especially acute onset, is the most understood. Yet delayed onset cramps and management have not been sufficiently documented. J Cancer Sci Clin Ther 2021; 5 (4): 448-458 DOI: 10.26502/jcsct.5079130 Journal of Cancer Science and Clinical Therapeutics 449 Materials: A case report of a nine-year-old girl diagnosed with rhabdomyosarcoma who was treated with irinotecan in the relapse setting, is presented. A limited literature review was done based on the challenges experienced during the management of a patient with delayed onset irinotecaninduced cramps. The pathophysiology of delayed onset irinotecan induced cramps is based on cholinergic, inflammatory and enzymatic factors as well as neural damage. Treatment was based on the etiology described in the literature with a positive result. Conclusion: Delayed onset irinotecan related cramps are part of a multifactorial etiology that should be approached systematically, especially when first line treatment does not yield an adequate response. Introduction Irinotecan, a topoisomerase I inhibitor, has become an effective chemotherapeutic in the first line treatment of colon cancer in adults and relapsed and resistant solid tumours in children [1,2]. Gastro-intestinal side effects and neutropenia are the most commonly reported toxicities in both adults and children [1,2]. Early and late onset diarrhea are well described in the literature [1,3]. Early diarrhea may be accompanied by cholinergic symptoms such as cramps [3]. Concomitant administration of atropine before irinotecan administration may ameliorate the short-term cholinergic symptoms [3]. Yet atropine has an elimination half-life of 2-5 hours [4]. The hepatic metabolisation of atropine is incomplete and is then mainly excreted via the urine. Approximately 50% of the dose is excreted within 4-5 hours with 90% excreted within 24 hours [4]. Late diarrhea is defined as diarrhea starting more than 24 hours after administration of irinotecan [5]. Fluid and electrolyte monitoring and treatment with loperamide are effective in the management [5]. Studies have shown that neuropathic changes in especially the colon may manifest as delayedonset diarrhea even after the irinotecan therapy has been completed [6]. What is not clear is whether this also causes late onset of cramps. Although early onset cramping may ease with atropine, the management of late onset cramps due to irinotecan has not been discussed in the literature. Methods and Materials This case report describes the clinical course of abdominal cramps after irinotecan administration, defining the causality and then presents a solution to improve the quality of life in a patient receiving chemotherapy in a tumour relapse setting. A Pubmed-and Google Scholar search was conducted in the English-language literature using the keywords 'irinotecan', 'cramps', 'late onset', 'delayed onset' and/or 'treatment'. Case Presentation A nine-year-old girl was diagnosed with metastatic alveolar rhabdomyosarcoma, involving a mass originating from the muscles of the right hand with metastatic disease involved her lungs, liver, T12 -L3 vertebrae, abdominal and thoracic lymph nodes. She was managed according to the very high risk (VHR) arm of the European Paediatric Soft The abdominal cramps reemerged as previously on day seven lasting for seven days (Table 2). Following the IT-P regimens diarrhea was present from and the analgesics were less effective than in the VIT regimen. The duration of pain relief decreased and the frequency of analgesic administration increased. Additional anti-spasmodic therapy with hyoscine butylbromide (Buscopan®) had no effect. Isolating irinotecan as causative agent: The patient received two independent courses that included irinotecan. During the first relapse vincristine, irinotecan, temozolomide (VIT) and during progression of disease irinotecan, temozolomide and palbociclib (IT-P). Vincristine can cause abdominal cramps in 1-10% of cases and may be independent or caused by constipation [8]. The cramps were present in both courses, but vincristine was only present in the first course. Abdominal cramps are not a reported side effect of palbociclib and temozolomide [9,10]. The Naranjo scores and pharmaceutical interactions: The Naranjo scores, a causality assessment method for determining the likelihood of whether an adverse drug reaction is actually due to the drug [10], of the irinotecan, vincristine, temozolomide and palbociclib were respectively 9, 6, 6, 6 ( Journal of Cancer Science and Clinical Therapeutics 451 regimen or IT-P regimen that could contribute to the abdominal cramps [8,9,11]. We cannot confirm whether this has been as specific study question in a clinical trial. The pathophysiology of irinotecan abdominal cramps: Where the pathophysiology of acute symptoms has been documented, the aetiology of late onset symptoms is more complex [3]. Early-onset diarrhoea is caused by irinotecan-induced cholinergic syndrome. This acute cholinergic reaction is associated with both acute and delayed diarrhoea, reduced capacity of the intestinal mucosa for absorption and intestinal hyper-peristalsis leading to abdominal cramps [3,12]. Although acute irinotecan-induced diarrhoea may be prevented by blocking anticholinergic receptors with atropine, no prophylactic treatment for delayed irinotecan-induced diarrhoea has been described [3]. A number of contributing factors have been described with late onset irinotecan-induced diarrhoea [3,13]. The pathophysiology has a neural, inflammatory and enzymatic component that leads to intestinal dysfunction and contributes to late onset cramps [3,5,6,12,13]. This set of etiological factors contributes to long-term gastrointestinal dysfunction and symptoms. Irinotecan is converted in the liver into an active metabolite SN-38 by enzymatic deesterification. This active substance is excreted via bile into the intestine [3,5]. A second parallel process occurs when the SN-38 glucuronide is converted by beta-glucuronidase from gut bacteria from a noncytotoxic metabolite to active SN-38 [3,5]. The cytotoxicity causes mucositis and the epithelial damage [13,14]. Studies evaluating beta-glucuronidase activity in intestinal fluids concluded that the highest activity was in the caecum and low in the duodenal and jejunum [15]. Thus, direct cytotoxicity to the caecum and colon leads to mucosal damage. The duodenum and jejunum are sites where carboxylesterase activity converts irinotecan to SN-38 and are thus spared from mucosal damage [15]. The mucosal damage is observed further down the intestinal tract [15]. The main process of damage is inflammation or intestinal mucositis [14,15]. in the ileum [15,16]. In the caecum goblet-cell hyperplasia occurs with excessive production of sulfomucin. The abnormal intestinal mucosa is therefore the site of abnormal ion transport, malabsorption and eventually increased electrolyte and water secretion [15,16]. Simultaneously the irinotecan-induced neuronal loss was associated with changes in colonic motility and gastrointestinal transit [6]. TLR4 signalling may be a contributing factor of both delayed diarrhoea and cramps [12]. The aetiology of irinotecan-induced mucositis is mediated through the interleukin-1/Toll-like receptor [3]. Mucosal damage and inflammation can be present following short and long-term treatment with irinotecan [13,14]. Therefore, irinotecan toxicity is not just dependent on dose but also on the duration of exposure as well. Discussion The pathology of delayed onset cramps is a multifactorial process involving mucosal inflammation and damage, neuronal dysfunction as well as enzymatic and electrolyte imbalances (Table 4) [3]. Since escalation of analgesia, which included opioid-based therapy, produced no effect, we postulated that various aetiological pathways of the cramps had to be addressed. Irinotecan-induced diarrhoea may be caused by intestinal bacterial deglucuronidation of SN-38G by beta-glucuronidase-producing flora [15]. Daily disease [17]. Budesonide has been proven to be locally effective in mild CD with distal inflammation (mainly in the ileum or caecum) or mild UC [17]. We hypothesised that therefore it could also be effective in stopping ileoceacal inflammation caused by irinotecan, causing fewer side-effects than systemic corticosteroids such as prednisone. During the third cycle budesonide was therefore administrated after completion of the irinotecan cycle on day six. This still resulted in severe (score 8/10) delayed onset cramps, but of a shorter duration (2 days) and responsive to paracetamol monotherapy. During the fourth cycle budesonide was started on the second day of the cycle which resulted in delayed onset cramps of mild intensity. Very limited literature exists on treatment for delayed neuronal dysfunction [6]. In the literature for inflammatory bowel disorders, peppermint oil has been studied for symptomatic management [18,19]. Through the direct enteric nervous system stimulation and calcium channel blockade, peppermint oil causes smooth muscle relaxation [18,19]. Visceral sensitivity modulation is achieved through transient receptor potential cation channels. This antispasmodic effect was present in both the small and large intestine. Independent from the neuronal effects, peppermint oil has anti-microbial and anti-inflammatory activity [18,19]. The diarrhoea was a new symptom during the IT-P regimen. The diarrhoea was of late onset which made cholinergic dysfunction less likely and prophylaxis for an enzymatic aetiology was already administered. We postulated that the late onset diarrhoea was due to inflammatory changes in the gut with subsequent neuronal and electrolyte imbalances. Despite not presenting with early onset symptoms, atropine was included before treatment. After the first cycle that included both atropine and steroid therapy the diarrhoea had resolved. In the event that the diarrhoea did not resolve loperamide and/or a peripheral opioid agent could have been introduced to control diarrhoea [3]. Yet these two options do not modulate intestinal mucositis-associated inflammation. Alternative treatment options Although our patient responded well to our management, other treatment options were considered, but never initiated. Upwards titration of budesonide The advised dose for children between six years and 18 years is 9 mg/day PO. A short-term increased dose can be given [21]. Anti-inflammatory treatment with an TRPA1 receptor agonist Anti-inflammatory potential of a TRPA1 agonist in reducing intestinal mucositis is achieved by reducing the production and/or release of pro-inflammatory cytokines Journal of Cancer Science and Clinical Therapeutics 456 TNF-a, IL-1b, and KC [14]. Inflammation can be reduced by limiting oxidative stress [14]. Reduction of intestinal motility or water and electrolyte loss Late onset symptoms have variable etiology and may not adequately respond to standard antispasmodic-and antidiarrheal medication [16]. This is due to mucosal damage of the intestine, loss of abnormal ion transportation and increased electrolyte and water losses [16]. This may be associated with cramps. Although the first line management may include loperamide and antispasmodics, with continued fluid and electrolyte losses other medications should be included in the management [16]. Racecadotril or acetorphan is a peripheral acting enkephalinase inhibitor that decreases intestinal secretion and promotes absorption. Octreotide is a peptide that reduces gastro-intestinal motility and secretion of fluids in the intestines and pancreas [22]. Octreotide is indicated in loperamide refractory diarrhea in cancer related diseases [23]. Conclusion Delayed onset irinotecan related cramps are part of a multifactorial etiology that should be approached systematically, especially when first line treatment does not yield an adequate response. Rotator cable and rotator interval: anatomy, biomechanics and clinical importance The rotator cable and rotator interval are among the most recent topics of interest in current shoulder literature. Most of the research has been published in the last two decades and our understanding about the importance of these anatomical structures has improved with biomechanical studies, which changed the pre- and intra-operative approaches of shoulder surgeons to rotator cuff tears in symptomatic patients. The rotator cable is a thick fibrous bundle that carries the applied forces to the rotator cuff like a ‘suspension bridge’. Tears including this weight-bearing bridge result in more symptoms. On the other hand, the rotator interval is more like a protective cover consisting of multiple layers of ligaments and the capsule rather than a single anatomical formation like the rotator cable. Advances in our knowledge about the rotator interval demonstrate that even basic anatomical structures often have greater importance than we may have understood. Misdiagnosis of these two important structures may lead to persistent symptoms. Furthermore, some distinct rotator cuff tear patterns can be associated with concomitant rotator interval injuries because of the anatomical proximity of these two anatomical regions. We summarize these two important structures from the aspect of anatomy, biomechanics, radiology and clinical importance in a review of the literature. Cite this article: EFORT Open Rev 2019;4:56-62. DOI: 10.1302/2058-5241.4.170071. Introduction Rotator cuff problems are a common cause of shoulder pain and represent a significant expense and burden on the healthcare system. Despite the high rate of radiological prevalence of rotator cuff tears in elderly

patients, many patients continue their lives with minimal symptoms. Several studies investigated asymptomatic shoulders in the general population and found that rotator cuff tears are relatively common, in the range of 6% to 34% in all age groups. 1,2 A recent study by Yamamoto et al reported that nearly 20% of the general population have rotator cuff tears, with an increasing prevalence in elderly patients. They found that 50% of patients who are in their 80s have rotator cuff tears regardless of whether they are symptomatic or asymptomatic. Even patients without any shoulder symptoms accounted for 13.5% of the rotator cuff tears in the general population. 3 Understanding the contribution of the rotator cable (RC) or rotator interval (RI) could be the answer to those. In his first publication about supraspinatus tears in 1934, Codman presented two cases of surgical repair and defined an avascular crescentic area at the insertion site of the supra and infraspinatus. 4 He identified this insertion site in rotator cuff tear patients. Nearly 60 years later, Clark and Harryman published a detailed anatomical study of the rotator cuff and described a fibrous tissue which originates from the deep layer of coracohumeral ligament and moves along the perpendicular axis of supraspinatus and infraspinatus tendon fibres. 5 Later, Burkhart et al named this anatomical structure the 'rotator cable' and described the biomechanical role of this ligament as a stress shielder of the rotator crescent, which is an avascular crescentic area at the insertion site of the supra and infraspinatus. 6 They compared the RC to a 'suspension bridge' that works to keep load stress out of the rotator crescent, which tends to be more vulnerable to tears. More studies of the gross anatomic and arthroscopic appearance of the RC were published after the abovementioned publications. 7,8 The RI is more like a protective cover consisting of multiple layers of ligaments and the capsule rather than a single anatomical formation like the RC. This close relationship with the joint capsule and nearby anatomical structures, such as the long head of biceps tendon and coracohumeral Rotator cable and rotator interval: anatomy, biomechanics and clinical importance Gazi Huri 1 Mehmet Kaymakoglu 1 Nickolas Garbis 2 3.1700EOR0010.1302/2058-5241.4.170071 research-article2018 ligament, etc., increases its importance. Ignoring the RI causes many misdiagnoses in clinical routine and can explain some 'refractory' shoulder pains. The RI's role in biceps tendinopathy, SLAp lesions and glenohumeral instability has been recently popularized in shoulder literature. 9 The aim of this paper is to recall the importance of the RC and RI and summarize their anatomical, biomechanical and radiological features. Awareness of these important structures should be raised among shoulder surgeons in the interest of proper diagnosis and treatment of shoulder pathologies. Anatomy and function Understanding the anatomy of the rotator cuff is essential for a shoulder surgeon in order to apply the appropriate surgical repair. Recent research has helped to advance our understanding of the complexity of the shoulder anatomy. The coracohumeral ligament (CHL) arises from the lateral side of the coracoid process and orientates obliquely to the humeral head. It divides into two layers (superficial and deep), surrounds the rotator cuff and inserts into lesser and greater tuberosities. The superficial layer of the CHL covers the articular surface of the supraspinatus and infraspinatus tendons. The deeper part of the CHL is thicker and inserts into the greater tuberosity. Here, the deep layer forms a crescent-shaped structure, running posteriorly from the anterior insertion site of the supraspinatus to the inferior border of the infraspinatus, perpendicular to the axis of tendon fibres (Fig. 1). Kolts et al contributed to the anatomy literature and named this structure the ligamentum semicirculare humeri. 10 This arch showed variable thickness in ultrasonographic and cadaveric studies. 11 It is approximately 2.59 times thicker than the rotator crescent. Burkhart et al found that younger cadaveric specimens have a thicker rotator crescent compared to the cable. They proposed that two categories can be defined. 6 One of these categories is cable-dominant, which is seen in older specimens, and the other is crescentdominant. Their data suggested that as a person gets older, his/her RC becomes thicker and the rotator crescent thinner. Therefore, it can be said that the cable/crescent ratio is bigger in older shoulders and the crescent area becomes thinner. This implies that the integrity of the superior rotator cuff is more dependent on RC integrity in older patients. Despite these findings, some researchers found that the RC is more detectable with ultrasound (US) in younger patients. 10,12 Understanding which muscles are suspended by the RC is crucial to making a proper diagnosis and surgical repair. The supraspinatus tendon has two parts: a superficial part which reaches more anteriorly and attaches to the lesser tuberosity, whereas the deeper and major part fuses into the RC in conjunction with the CHL. The anterior insertion area of the RC mainly consists of supraspinatus tendon fibres. The posterior insertion area is shared by the supraspinatus, infraspinatus and partly the teres minor tendons. 9 In addition to rotator cuff insertion, there has been some recent interest in studying the anatomy and function of the RI. The RI was presumed to be a semi-functional and relatively empty area in the rotator cuff. There are two different RIs in shoulder anatomy: the anterior and posterior RIs. 13 Here, we will discuss only the anterior RI. In 2002, Kolts et al identified the borders of the RI in a cadaver study on 19 shoulder joints. 7 It is described as a triangularshaped space instead of a two-dimensional 'area', whose borders are drawn superiorly by the supraspinatus, inferiorly by the subscapularis tendon and by the joint capsule as a ceiling. The floor of this space is the articular surface of the humeral head (Fig. 2). It is strengthened by the CHL laterally and by the superior glenohumeral ligament (SGHL) medially (Fig. 3a). 14 The long head of the biceps tendon lies on the base of the RI space and is covered by a fibrous sheet of capsule. 15 The coracoid process constitutes the medial edge of this triangular space (Fig. 3b). Microscopically, the RI consists of four layers, as described by Jost et al. 16 The first layer consists of superficial CHL fibres which originate from the coracoid process and insert into the greater and lesser tuberosities following the supraspinatus and subscapularis tendons, respectively. The second layer is a blend of the CHL and rotator cuff tendons. The third layer consists of deep CHL fibres and the fourth layer is a mesh of the SGHL and capsule. 17 Another important structure in the RI is the coracoglenoid ligament (CGL) which forms the anterosuperior ligament complex with the CHL and SGHL. It runs from the coracoid process to the supraglenoid tubercle, extra-articularly. Its function is still not fully understood, but a few studies suggest the contribution of stability to the shoulder. 18 Wilson et al described how the capsular anatomy in the RI can be variable. They stated that most of the patients (59%) have a RI capsular opening (RICO) that is superolateral to the middle glenohumeral ligament (MGHL). 19 A shoulder surgeon should also be aware of how close the RI and anterior portion of the RC (or supraspinatus tendon) are. This variability and proximity can give us a clue about why the RI and its associated structures (ligaments and biceps tendon, etc.) are involved in some cuff tears. 20 Radiological manifestation The RC and RI structures are not visible on plain radiographs. MRI or US is widely used in the screening of these anatomical structures. The RC can be observed in routine shoulder MRI screening in patients with rotator cuff symptoms in all planes. Some authors state that the abduction, external rotation (ABER) is the best position for viewing the rotator cuff and cable pathologies. This is attributed to the decreased tension of ruptured rotator cuff fibres and avoidance of adherence to adjacent intact cuff tissue. 21 However, evaluation of this imaging technique requires experience and can be complicated for some radiologists and orthopaedic surgeons. A routine oblique coronal plane MRI often offers a good view of the RC (Fig. 4). On the fat-suppressed T2 view, the RC can be visualized as a hypo-intense region 2 mm to 5 mm wide on the articular side of the supraspinatus tendon. Even in the best-case scenario, visualization of the RC is limited up to nearly 75% in asymptomatic patients on MRI due to its small diameter and the difficulty in distinguishing it in crescentdominant younger patients. 6,22 In addition, it is hard to distinguish the RC from a small anterior supraspinatus tear. Using axial images may help identify the RC for the clinician. On the axial plane, hypo-intensity of the RC runs from the greater tuberosity to its posterior attachment next to the infraspinatus tendon, the same as its anatomical shape. Coronal and axial images provide sufficient visibility in most cases, whereas ABER and sagittal planes may also be used to confirm the previous commonly used planes. 23 Ultrasound is another screening technique used to assess the RC. Some studies report improved identification of RI structures compared to MRI screening, in the range of 77% to 99%. 24,25 Sconfienza et al reported more frequent detection of the RC in elderly patients. This finding is consistent with the findings by Burkhart, in which elderly patients have cable-dominant anatomy. Another explanation for this relationship is the concomitant subclinical tendinosis in elderly patients, which makes the cable more visible. 26 Because the RI is an intra-articular structure, it can only be seen on MRI if synovial fluid is prominent (Fig. 5). 27 For this reason, a MR arthrogram is suggested in the presence of any injury of RI structures. The oblique sagittal MRI plane is the most useful sequence in evaluating the RI. On the sagittal plane, the RI capsule lies over the biceps pulley as a hypo-intense band. More medially, the CHL is visualized running from the coracoid process to the humeral head covering the long head of biceps tendon. Conversely, the SGHL is harder to visualize than the other structures of the RI. In an MRI study by Chung et al, the SGHL is not seen at all in all routine MRI views, and the CHL is observed only in 60% of cases. 28 The CGL can also be observed by experienced radiologists as a part of the anterosuperior capsuloligamentous complex. 29 Tamborrini et al reported that musculoskeletal US is another superior technique used to visualize the RI. Many pathologies of the anatomic structures within the RI (including tendinosis, tears and capsulitis) can be imaged using US with a better resolution than any other techniques. The only limiting factor is the decreased visualization while observing the soft tissue beneath any bony structure such as the acromion. 30 Also, US is dependent on the ability of radiologists' diagnostic experience; decisions on appropriate imaging techniques should be made taking this into consideration. Biomechanical and clinical importance of the rotator cable and rotator interval The relationship of cuff tear type and size with patients' clinical presentation has been discussed for a long time. Early studies suggested a correlation between the tear size and symptomatic disease with decreased muscle strength, pre-and post-operatively; 30 however, some researchers contradicted this idea, stating that post-operative outcomes and tear size have no relationship. 17 After the RC concept was introduced, there have been studies specifically evaluating the biomechanics. In 2013, Mesiha et al published the biomechanical properties of the RC in a cadaveric shoulder model. 31 They compared crescent-and cable-involving tears and proved that the cable-involved tears which are located at the anterior insertion site of the supraspinatus tendon are more relevant with bigger tear size, reduced tendon stiffness and increased tendon stress. These findings may explain increased fatty degeneration in anterior supraspinatus tendon ruptures and give us a clue about why some rotator cuff tears are more likely to be symptomatic. Using this concept, the treatment strategy for full thickness rotator cuff ruptures involving the anterior portion of the supraspinatus tendon is more often surgical. Our knowledge continued to evolve with the findings of pinkowski et al about partial thickness ruptures involving the RC. Using five frozen cadavers, they demonstrated partial thickness cable rupture and found out that even partial thickness rupture

would be enough to change the glenohumeral kinematics. The simulated tear in the anterior cable increases the anterior translation of humeral head by 38.6% in 30° external rotation. They concluded that these results can explain why orthopaedic surgeons sometimes fail to treat overhead athletes with partial thickness cuff rupture and their shoulder pain if they only perform debridement without repair in surgery. 32 Besides the more complex pattern and inherent risk for fatty degeneration and disrupted glenohumeral kinematics, re-tears occur more frequently in the setting of the rotator cuff ruptures involving the RC. 33 Additionally, Denard et al laid emphasis upon the relation of the cable integrity and pseudoparalysis. They compared the integrity of the RC in two different patient groups that underwent arthroscopic repair with massive rotator cuff tear, retrospectively. The first group, in which all patients had pseudoparalysis pre-operatively, had more disrupted RC attachment than the other group, in which patients had more active forward flexion > 90°. 34 This shows us the importance of the cable integrity in shoulder biomechanics. The RI plays an important role in stabilizing the glenohumeral joint. Harryman et al proved this relationship in a cadaveric study, in which they measured the motion of glenohumeral joint in three conditions: in the normal state; after capsule dissection; and imbrication. The capsular and ligamentous components of the RI limit the inferior translation of the glenohumeral joint in an adducted position. Sectioning increases the range of passive flexion, external rotation and adduction. Imbrication causes a decrease in these motions. 35 A second function of the RI is the stabilization of the long head of the biceps tendon via the CHL and SGHL. Both of these ligaments form a U-shaped cover around the biceps tendon, named the 'biceps pulley'. The superior part of the subscapularis tendon also contributes to this stabilization with a buttress effect. 36 Many studies reveal the biomechanical and clinical effects of RI closure. Daly et al suggested that an additional RI closure improves the biomechanical strength of the repaired subscapularis tendon. 37 Furthermore, Mologne et al put forward the idea that anterior shoulder stability increases with the same closure. 38 A study by Tsai et al showed good clinical results with arthroscopic release of the RI in adhesive capsulitis patients. They declared that the technique does not increase the risk of instability and avoids injury of the axillary nerve and fluid extravasation. 39 Adhesive capsulitis (frozen shoulder) is another pathologic condition that the RI plays a fundamental role in. It was first described by Nevasier et al in 1945 and is seen in patients aged > 40 years. 40 To put it all in simple terms, we can describe it as a thickening or rigidification of the glenohumeral capsule around the RI, as we mentioned above that the deepest layer of the RI consists of the capsule. Histological studies revealed that higher cytokine levels and lack of metalloproteinases could be the cause of the pathology. 41 Although CHL thickening and autoimmune response are also accused of being the cause, we still do not know the exact mechanism. Synovitis in adhesive capsulitis involves all of the anatomical parts of the RI, such as the sheath of long head of the biceps, the CHL and the capsule. Thus, these findings prove to us that whatever the exact mechanism adhesive capsulitis has, it can be said that adhesive capsulitis is a disease of the RI, specifically. The diagnosis is based mainly on clinical findings: pain and loss of range of the motion around the shoulder circle. MRI findings such as obliteration of the RI fat pad and increased signal density of the inferior glenohumeral ligament are diagnostic. 42 Imaging is helpful to exclude other pathologies such as rotator cuff tears, etc. Ultrasound can also be used for diagnosis but it is more difficult to view than in MRI. Treatment is mainly conservative, as most of the cases resolve by themselves. If physiotherapy fails to succeed, manipulation under general anaesthesia or arthroscopic release of the capsule is indicated. Anatomical proximity of the anterior RC and RI lead to a new definition: 'anterior superior lesions'. This distinct type of rotator cuff tear involves the subscapularis and anterior portion of the supraspinatus tendon and adjacent RI structures: SGHL, CHL and biceps tendon. 43 Case-control clinical studies and further investigation are needed to better define the clinical importance of this new concept. Conclusion The RC is one of the most important contributors to glenohumeral biomechanics. Even in cases of partial thickness rotator cuff tear involving the RC, there are significant changes in joint translation. This knowledge may help shoulder surgeons to clarify the indication for repairing versus debridement based on the involvement of the RC rather than the traditional percentage of medial to lateral thickness involvement. Biomechanical studies have been transforming into clinical prospective trials; the benefits of repairing the RC will be studied further as we develop longer-term results. 32 Our understanding of the RI can help us treat the shoulder during arthroscopic intervention in conditions such as adhesive capsulitis or shoulder instability. 39,44 The relatively new finding of an anatomical structure in the RI like the CGL may continue to change our knowledge about shoulder biomechanics and further clarify identification and clinical management of pathologic processes. Additional studies are needed to clarify the role of the RI and RC and its associated structures in the setting of shoulder instability to better define indications and improve surgical techniques. Funding statement The author or authors choose not to respond. Appraising psychotherapy case studies in practice-based evidence: introducing Case Study Evaluation-tool (CaSE) Systematic case studies are often placed at the low end of evidence-based practice (EBP) due to lack of critical appraisal. This paper seeks to attend to this research gap by introducing a novel Case Study Evaluation-tool (CaSE). First, issues around knowledge generation and validity are assessed in both EBP and practice-based evidence (PBE) paradigms. Although systematic case studies are more aligned with PBE paradigm, the paper argues for a complimentary, third way approach between the two paradigms and their ‘exemplary’ methodologies: case studies and randomised controlled trials (RCTs). Second, the paper argues that all forms of research can produce ‘valid evidence’ but the validity itself needs to be assessed against each specific research method and purpose. Existing appraisal tools for qualitative research (JBI, CASP, ETQS) are shown to have limited relevance for the appraisal of systematic case studies through a comparative tool assessment. Third, the paper develops purpose-oriented evaluation criteria for systematic case studies through CaSE Checklist for Essential Components in Systematic Case Studies and CaSE Purpose-based Evaluative Framework for Systematic Case Studies. The checklist approach aids reviewers in assessing the presence or absence of essential case study components (internal validity). The framework approach aims to assess the effectiveness of each case against its set out research objectives and aims (external validity), based on different systematic case study purposes in psychotherapy. Finally, the paper demonstrates the application of the tool with a case example and notes further research trajectories for the development of CaSE tool. Introduction Due to growing demands of evidence-based practice, standardised research assessment and appraisal tools have become common in healthcare and clinical treatment (Hannes, Lockwood, & Pearson, 2010;Hartling, Chisholm, Thomson, & Dryden, 2012;Katrak, Bialocerkowski, Massy-Westropp, Kumar, & Grimmer, 2004). This allows researchers to critically appraise research findings on the basis of their validity, results, and usefulness (Hill & Spittlehouse, 2003). Despite the upsurge of critical appraisal in qualitative research (Williams, Boylan, & Nunan, 2019), there are no assessment or appraisal tools designed for psychotherapy case studies. Although not without controversies (Michels, 2000), case studies remain central to the investigation of psychotherapy processes (Midgley, 2006;Willemsen, Della Rosa, & Kegerreis, 2017). This is particularly true of systematic case studies, the most common form of case study in contemporary psychotherapy research (Davison & Lazarus, 2007;McLeod & Elliott, 2011). Unlike the classic clinical case study, systematic cases usually involve a team of researchers, who gather data from multiple different sources (e.g., questionnaires, observations by the therapist, interviews, statistical findings, clinical assessment, etc.), and involve a rigorous data triangulation process to assess whether the data from different sources converge (McLeod, 2010). Since systematic case studies are methodologically pluralistic, they have a greater interest in situating patients within the study of a broader population than clinical case studies (Iwakabe & Gazzola, 2009). Systematic case studies are considered to be an accessible method for developing research evidencebase in psychotherapy (Widdowson, 2011), especially since they correct some of the methodological limitations (e.g. lack of 'third party' perspectives and bias in data analysis) inherent to classic clinical case studies (Iwakabe & Gazzola, 2009). They have been used for the purposes of clinical training (Tuckett, 2008), outcome assessment (Hilliard, 1993), development of clinical techniques (Almond, 2004) and meta-analysis of qualitative findings (Timulak, 2009). All these developments signal a revived interest in the case study method, but also point to the obvious lack of a research assessment tool suitable for case studies in psychotherapy (Table 1). To attend to this research gap, this paper first reviews issues around the conceptualisation of validity within the paradigms of evidence-based practice (EBP) and practice-based evidence (PBE). Although case studies are often positioned at the low end of EBP (Aveline, 2005), the paper suggests that systematic cases are a valuable form of evidence, capable of complimenting large-scale studies such as randomised controlled trials (RCTs). However, there remains a difficulty in assessing the quality and relevance of case study findings to broader psychotherapy research. As a way forward, the paper introduces a novel Case Study Evaluation-tool (CaSE) in the form of CaSE Purpose-based Evaluative Framework for Systematic Case Studies and CaSE Checklist for Essential Components in Systematic Case Studies. The long-term development of CaSE would contribute to psychotherapy research and practice in three ways. Given the significance of methodological pluralism and diverse research aims in systematic case studies, CaSE will not seek to prescribe explicit case study writing guidelines, which has already been done by numerous authors (McLeod, 2010;Meganck, Inslegers, Krivzov, & Notaerts, 2017;Willemsen et al., 2017). Instead, CaSE will enable the retrospective assessment of systematic case study findings and their relevance (or lack thereof) to broader psychotherapy research and practice. However, there is no reason to assume that CaSE cannot be used prospectively (i.e. producing systematic case studies in accordance to CaSE evaluative framework, as per point 3 in Table 2). The development of a research assessment or appraisal tool is a lengthy, ongoing process (Long & Godfrey, 2004). It is particularly challenging to develop a comprehensive purpose-oriented evaluative framework, suitable for the assessment of diverse methodologies, aims and outcomes. As such, this paper should be treated as an introduction to the broader development of CaSE tool. It will introduce the rationale behind CaSE and lay out its main approach to evidence and evaluation, with further development in mind. A case example from the Single Case Archive (SCA) (https://singlecasearchive.com) will be used to demonstrate the application of the tool 'in action'. The paper notes further research trajectories and discusses some of the limitations around the use of the tool. Separating the wheat from the chaff: what is and is not evidence in psychotherapy (and who gets to decide?) The common approach: evidence-based practice In the last two decades, psychotherapy has become increasingly centred around the idea of an evidence-based practice (EBP). Initially introduced in medicine, EBP has been defined as 'conscientious, explicit and judicious use of current best evidence in making decisions about the care of individual patients' (Sackett, Rosenberg, Gray, Haynes, & Richardson, 1996). EBP revolves around efficacy research: it seeks to examine whether a specific intervention has a causal (in this case, measurable) effect on clinical populations (Barkham & Mellor-Clark, 2003). From a conceptual standpoint, Sackett and colleagues defined EBP as a paradigm that is inclusive of many methodologies, so long as they contribute towards clinical decision-making process and accumulation of best currently available evidence in any given set of circumstances (Gabbay & le May, 2011). Similarly, the American Psychological Association (APA, 2010) has recently issued calls for evidence-based systematic case studies in Table 1 Key concept: systematic case study Systematic case study is a systematised alternative to the classical clinical case study. Systematic case studies generally involve a team of researchers, gather data from multiple

different sources (questionnaires, observations by the therapist, interviews, statistical findings, etc.) and feature data triangulation processes in order to assess whether the data from different sources converge. 2. Using CaSE to evaluate the varying evidential quality of systematic case studies, which is particularly problematic for qualitative metaanalysis and meta-synthesis of published case studies in psychotherapy (Duncan & Sparks, 2020;Iwakabe & Gazzola, 2009;Thorne, Jensen, Kearney, Noblit, & Sandelowski, 2004); order to produce standardised measures for evaluating process and outcome data across different therapeutic modalities. However, given EBP's focus on establishing cause-andeffect relationships (Rosqvist, Thomas, & Truax, 2011), it is unsurprising that qualitative research is generally not considered to be 'gold standard' or 'efficacious' within this paradigm (Aveline, 2005;Cartwright & Hardie, 2012;Edwards, 2013;Edwards, Dattilio, & Bromley, 2004;Longhofer, Floersch, & Hartmann, 2017). Qualitative methods like systematic case studies maintain an appreciation for context, complexity and meaning making. Therefore, instead of measuring regularly occurring causal relations (as in quantitative studies), the focus is on studying complex social phenomena (e.g. relationships, events, experiences, feelings, etc.) (Erickson, 2012;Maxwell, 2004). Edwards (2013) points out that, although context-based research in systematic case studies is the bedrock of psychotherapy theory and practice, it has also become shrouded by an unfortunate ideological description: 'anecdotal' case studies (i.e. unscientific narratives lacking evidence, as opposed to 'gold standard' evidence, a term often used to describe the RCT method and the therapeutic modalities supported by it), leading to a further need for advocacy in and defence of the unique epistemic process involved in case study research (Fishman, Messer, Edwards, & Dattilio, 2017). The EBP paradigm prioritises the quantitative approach to causality, most notably through its focus on high generalisability and the ability to deal with bias through randomisation process. These conditions are associated with randomised controlled trials (RCTs) but are limited (or, as some argue, impossible) in qualitative research methods such as the case study (Margison et al., 2000) (Table 3). 'Evidence' from an EBP standpoint hovers over the epistemological assumption of procedural objectivity: knowledge can be generated in a standardised, nonerroneous way, thus producing objective (i.e. with minimised bias) data. This can be achieved by anyone, as long as they are able to perform the methodological procedure (e.g. RCT) appropriately, in a 'clearly defined and accepted process that assists with knowledge production' (Douglas, 2004, p. 131). If there is a well-outlined quantitative form for knowledge production, the same outcome should be achieved regardless of who processes or interprets the information. For example, researchers using Cochrane Review assess the strength of evidence using meticulously controlled and scrupulous techniques; in turn, this minimises individual judgment and creates unanimity of outcomes across different groups of people (Gabbay & le May, 2011). The typical process of knowledge generation (through employing RCTs and procedural objectivity) in EBP is demonstrated in Fig. 1. In EBP, the concept of validity remains somewhat controversial, with many critics stating that it limits rather than strengthens knowledge generation (Berg, 2019;Berg & Slaattelid, 2017;Lilienfeld, Ritschel, Lynn, Cautin, & Latzman, 2013). This is because efficacy research relies on internal validity. At a general level, this concept refers to the congruence between the research study and the research findings (i.e. the research findings were not influenced by anything external to the study, such as confounding variables, methodological errors and bias); at a more specific level, internal validity determines the extent to which a study establishes a reliable causal relationship between an independent variable (e.g. treatment) and independent variable (outcome or effect) (Margison et al., 2000). This approach to validity is demonstrated in Fig. 2. Social scientists have argued that there is a trade-off between research rigour and generalisability: the more specific the sample and the more rigorously defined the intervention, the outcome is likely to be less applicable to everyday, routine practice. As such, there remains a tension between employing procedural objectivity which increases the rigour of research outcomes and applying such outcomes to routine psychotherapy practice where scientific standards of evidence are not uniform. According to McLeod (2002), inability to address questions that are most relevant for practitioners contributed to a deepening research-practice divide in psychotherapy. Studies investigating how practitioners make clinical decisions and the kinds of evidence they refer to show that there is a strong preference for knowledge that is not generated procedurally, i.e. knowledge that encompasses concrete clinical situations, experiences and techniques. A study by Stewart and Chambless (2007) sought to assess how a larger population of clinicians (under APA, from varying clinical schools of thought and independent practices, sample size 591) make treatment decisions in private practice. The study found that large-scale statistical data was not the primary source of information sought by clinicians. The most important influences were identified as past clinical experiences and clinical expertise (M = 5.62). Treatment materials based on clinical case observations and theory (M = 4.72) were used almost as frequently as psychotherapy outcome research findings (M = 4.80) (i.e. evidence-based research). These numbers are likely to fluctuate across different forms of psychotherapy; however, they are indicative of the need for research about Table 3 Key concept: evidence-based practice (EBP) Evidence-based practice (EBP) was introduced in medicine as a conscientious use of current best evidence in clinical-decision making about individual patients. EBP revolves around efficacy research, which assesses whether specific interventions produce causal (measurable) effects on clinical populations. Internal validity and randomisation of samples are crucial to efficacy research. An example of such research is randomised controlled trials (RCTs). routine clinical settings that does not isolate or generalise the effect of an intervention but examines the variations in psychotherapy processes. The alternative approach: practice-based evidence In an attempt to dissolve or lessen the research-practice divide, an alternative paradigm of practice-based evidence (PBE) has been suggested (Barkham & Mellor-Clark, 2003;Fox, 2003;Green & Latchford, 2012;Iwakabe & Gazzola, 2009;Laska, Motulsky, Wertz, Morrow, & Ponterotto, 2014;Margison et al., 2000). PBE represents a shift in how we think about evidence and knowledge generation in psychotherapy. PBE treats research as a local and contingent process (at least initially), which means it focuses on variations (e.g. in patient symptoms) and complexities (e.g. of clinical setting) in the studied phenomena (Fox, 2003). Moreover, research and theory-building are seen as complementary rather than detached activities from clinical practice. That is to say, PBE seeks to examine how and which treatments can be improved in everyday clinical practice by flagging up clinically salient issues and developing clinical techniques (Barkham & Mellor-Clark, 2003). For this reason, PBE is concerned with the effectiveness of research findings: it evaluates how well interventions work in realworld settings (Rosqvist et al., 2011). Therefore, although it is not unlikely for RCTs to be used in order to generate practice-informed evidence (Horn & Gassaway, 2007), qualitative methods like the systematic case study are seen as ideal for demonstrating the effectiveness of therapeutic interventions with individual patients (van Hennik, 2020) ( Table 4). PBE's epistemological approach to 'evidence' may be understood through the process of concordant objectivity (Douglas, 2004): 'Instead of seeking to eliminate individual judgment, … [concordant objectivity] checks to see whether the individual judgments of people in fact do agree' (p. 462). This does not mean that anyone can contribute to the evaluation process like in procedural objectivity, where the main criterion is following a set quantitative protocol or knowing how to operate a specific research design. Concordant objectivity requires that there is a set of competent observers who are closely familiar with the studied phenomenon (e.g. researchers and practitioners who are familiar with depression from a variety of therapeutic approaches). Systematic case studies are a good example of PBE 'in action': they allow for the examination of detailed unfolding of events in psychotherapy practice, making it the most pragmatic and practice-oriented form of psychotherapy research (Fishman, 1999(Fishman, , 2005). Furthermore, systematic case studies approach evidence and results through concordant objectivity (Douglas, 2004) by involving a team of researchers and rigorous data triangulation processes (McLeod, 2010). This means that, although systematic case studies remain focused on particular clinical situations and detailed subjective experiences (similar to classic clinical case studies; see Iwakabe & Gazzola, 2009), they still involve a series of validity checks and considerations on how findings from a single systematic case pertain to broader psychotherapy research (Fishman, 2005). The typical process of knowledge generation (through employing systematic case studies and concordant objectivity) in PBE is demonstrated in Fig. 3. The figure exemplifies a bidirectional approach to research and practice, which includes the development of research-supported psychological treatments (through systematic reviews of existing evidence) as well as the perspectives of clinical practitioners in the research process (through the study of local and contingent patient and/or treatment processes) (Teachman et al., 2012;Westen, Novotny, & Thompson-Brenner, 2004). From a PBE standpoint, external validity is a desirable research condition: it measures extent to which the impact of interventions apply to real patients and therapists in everyday clinical settings. As such, external validity is not based on the strength of causal relationships between treatment interventions and outcomes (as in internal validity); instead, the use of specific therapeutic techniques and problem-solving decisions are considered to be important for generalising findings onto routine clinical practice (even if the findings are explicated from a single case study; see Aveline, 2005). This approach to validity is demonstrated in Fig. 4. Since effectiveness research is less focused on limiting the context of the studied phenomenon (indeed, explicating the context is often one of the research aims), there is more potential for confounding factors (e.g. bias and uncontrolled variables) which in turn can reduce the study's internal validity (Barkham & Mellor-Clark, 2003). This is also an important challenge for research appraisal. Douglas (2004) argues that appraising research in terms of its effectiveness may produce significant disagreements or group illusions, since what might work for some practitioners may not work for others: 'It cannot guarantee that values are not influencing or supplanting reasoning; the observers may have shared values that cause them to all disregard important aspects of an event' (Douglas, 2004, p. 462). Douglas further proposes that an interactive approach to objectivity may be Practice-based evidence (PBE) was introduced as an alternative paradigm to EBP. PBE focuses on assessing the variations and complexities of treatment in routine clinical practice. Research in PBE is concerned with the effectiveness of findings by examining how interventions work in real-world settings. External validity and contingency of findings is crucial to effectiveness research. An example of such research is the systematic case study. employed as a more complex process in debating the evidential quality of a research study: it requires a discussion among observers and evaluators in the form of peer-review, scientific discourse, as well as research appraisal tools and instruments. While these processes of rigour are also applied in EBP, there appears to be much more space for debate, disagreement and interpretation in PBE's approach to research evaluation, partly because the evaluation criteria themselves are subject of methodological debate and are often employed in different ways by researchers (Williams et al., 2019). This issue will be addressed more explicitly again in relation to CaSE development ('Developing purpose-oriented evaluation criteria for systematic case studies' section). A third way approach to validity and evidence The research-practice divide shows us that there may be something significant in establishing complementarity between EBP and PBE rather than treating them as mutually exclusive forms of research (Fishman et al., 2017). For one, EBP is not a sufficient condition for delivering research relevant to practice settings (Bower, 2003). While RCTs can demonstrate that an intervention works on average in a group, clinicians who are facing individual patients need to answer a different question: how can I make therapy work with this particular case? (Cartwright & Hardie, 2012). Systematic case studies are ideal for filling this gap: they contain descriptions of microprocesses (e.g. patient symptoms, therapeutic relationships, therapist attitudes) in psychotherapy practice that are often overlooked in large-scale RCTs (Iwakabe & Gazzola, 2009). In particular, systematic case studies describing the use of specific interventions with less researched psychological conditions (e.g. childhood depression or complex post-traumatic stress disorder) can deepen practitioners' understanding of effective clinical techniques before the results of large-scale outcome studies are disseminated. Secondly, establishing a working relationship between systematic case studies and RCTs will contribute towards a more pragmatic understanding of validity in psychotherapy research. Indeed, the very

tension and so-called tradeoff between internal and external validity is based on the assumption that research methods are designed on an either/or basis; either they provide a sufficiently rigorous study design or they produce findings that can be applied to real-life practice. Jimenez-Buedo and Miller (2010) call this assumption into question: in their view, if a study is not internally valid, then 'little, or rather nothing, can be said of the outside world' (p. 302). In this sense, internal validity may be seen as a pre-requisite for any form of applied research and its external validity, but it need not be constrained to the quantitative approach of causality. For example, Levitt, Motulsky, Wertz, Morrow, and Ponterotto (2017) argue that, what is typically conceptualised as internal validity, is, in fact, a much broader construct, involving the assessment of how the research method (whether qualitative or quantitative) is best suited for the research goal, and whether it obtains the relevant conclusions. Similarly, Truijens, Cornelis, Desmet, and De Smet (2019) suggest that we should think about validity in a broader epistemic sense-not just in terms of psychometric measures, but also in terms of the research design, procedure, goals (research questions), approaches to inquiry (paradigms, epistemological assumptions), etc. The overarching argument from research cited above is that all forms of research-qualitative and quantitative-can produce 'valid evidence' but the validity itself needs to be assessed against each specific research method and purpose. For example, RCTs are accompanied with a variety of clearly outlined appraisal tools and instruments such as CASP (Critical Appraisal Skills Programme) that are well suited for the assessment of RCT validity and their implications for EBP. Systematic case studies (or case studies more generally) currently have no appraisal tools in any discipline. The next section evaluates whether existing qualitative research appraisal tools are relevant for systematic case studies in psychotherapy and specifies the missing evaluative criteria. The relevance of existing appraisal tools for qualitative research to systematic case studies in psychotherapy What is a research tool? Currently, there are several research appraisal tools, checklists and frameworks for qualitative studies. It is important to note that tools, checklists and frameworks are not equivalent to one another but actually refer to different approaches to appraising the validity of a research study. As such, it is erroneous to assume that all forms of qualitative appraisal feature the same aims and methods (Hannes et al., 2010;Williams et al., 2019). Generally, research assessment falls into two categories: checklists and frameworks. Checklist approaches are often contrasted with quantitative research, since the focus is on assessing the internal validity of research (i.e. researcher's independence from the study). This involves the assessment of bias in sampling, participant recruitment, data collection and analysis. Framework approaches to research appraisal, on the other hand, revolve around traditional qualitative concepts such as transparency, reflexivity, dependability and transferability (Williams et al., 2019). Framework approaches to appraisal are often challenging to use because they depend on the reviewer's familiarisation and interpretation of the qualitative concepts. Because of these different approaches, there is some ambiguity in terminology, particularly between research appraisal instruments and research appraisal tools. These terms are often used interchangeably in appraisal literature (Williams et al., 2019). In this paper, research appraisal tool is defined as a method-specific (i.e. it identifies a specific research method or component) form of appraisal that draws from both checklist and framework approaches. Furthermore, a research appraisal tool seeks to inform decision making in EBP or PBE paradigms and provides explicit definitions of the tool's evaluative framework (thus minimising-but by no means eliminating-the reviewers' interpretation of the tool). This definition will be applied to CaSE (Table 5). In contrast, research appraisal instruments are generally seen as a broader form of appraisal in the sense that they may evaluate a variety of methods (i.e. they are non-method specific or they do not target a particular research component), and are aimed at checking whether the research findings and/or the study design contain specific elements (e.g. the aims of research, the rationale behind design methodology, participant recruitment strategies, etc.). There is often an implicit difference in audience between appraisal tools and instruments. Research appraisal instruments are often aimed at researchers who want to assess the strength of their study; however, the process of appraisal may not be made explicit in the study itself (besides mentioning that the tool was used to appraise the study). Research appraisal tools are aimed at researchers who wish to explicitly demonstrate the evidential quality of the study to the readers (which is particularly common in RCTs). All forms of appraisal used in the comparative exercise below are defined as 'tools', even though they have different appraisal approaches and aims. Comparing different qualitative tools Hannes et al. (2010) identified CASP (Critical Appraisal Skills Programme-tool), JBI (Joanna Briggs Institutetool) and ETQS (Evaluation Tool for Qualitative Studies) as the most frequently used critical appraisal tools by qualitative researchers. All three instruments are available online and are free of charge, which means that any researcher or reviewer can readily utilise CASP, JBI or ETQS evaluative frameworks to their research. Furthermore, all three instruments were developed within the context of organisational, institutional or consortium support (Tables 6, 7 and 8). It is important to note that neither of the three tools is specific to systematic case studies or psychotherapy case studies (which would include not only systematic but also experimental and clinical cases). This means that using CASP, JBI or ETQS for case study appraisal may come at a cost of overlooking elements and components specific to the systematic case study method. Based on Hannes et al. (2010) comparative study of qualitative appraisal tools as well as the different evaluation criteria explicated in CASP, JBI and ETQS evaluative frameworks, I assessed how well each of the three tools is attuned to the methodological, clinical and Table 5 Key concept: research appraisal tool Research appraisal tool is a method-specific (or a research componentspecific) form of appraisal that draws from both checklist and framework approaches. A research appraisal tool will usually provide explicit definitions for its evaluative framework and will be used by researchers who wish to demonstrate the evidential quality of their study to the readers. CASP is part of the Oxford Centre for Triple Value Healthcare enterprise, which seeks to support healthcare systems and achieve optimal outcomes for populations. CASP has a variety of checklists, many of which are aimed at RCTs (e.g. RCT checklist, systematic review checklist, cohort study checklist, etc.). theoretical aspects of systematic case studies in psychotherapy. The latter components were based on case study guidelines featured in the journal of Pragmatic Case Studies in Psychotherapy as well as components commonly used by published systematic case studies across a variety of other psychotherapy journals (e.g. Psychotherapy Research, Research In Psychotherapy: Psychopathology Process And Outcome, etc.) (see Table 9 for detailed descriptions of each component). The evaluation criteria for each tool in Table 9 follows Joanna Briggs Institute (JBI) (2017a, 2017b); Critical Appraisal Skills Programme (CASP) (2018); and ETQS Questionnaire (first published in 2004 but revised continuously since). Table 10 demonstrates how each tool should be used (i.e. recommended reviewer responses to checklists and questionnaires). Using CASP, JBI and ETQS for systematic case study appraisal Although JBI, CASP and ETQS were all developed to appraise qualitative research, it is evident from the above comparison that there are significant differences between the three tools. For example, JBI and ETQS are well suited to assess researcher's interpretations (Hannes et al. (2010) defined this as interpretive validity, a subcategory of internal validity): the researcher's ability to portray, understand and reflect on the research participants' experiences, thoughts, viewpoints and intentions. JBI has an explicit requirement for participant voices to be clearly represented, whereas ETQS involves a set of questions about key characteristics of events, persons, times and settings that are relevant to the study. Furthermore, both JBI and ETQS seek to assess the researcher's influence on the research, with ETQS particularly focusing on the evaluation of reflexivity (the researcher's personal influence on the interpretation and collection of data). These elements are absent or addressed to a lesser extent in the CASP tool. The appraisal of transferability of findings (what this paper previously referred to as external validity) is addressed only by ETQS and CASP. Both tools have detailed questions about the value of research to practice and policy as well as its transferability to other populations and settings. Methodological research aspects are also extensively addressed by CASP and ETQS, but less so by JBI (which relies predominantly on congruity between research methodology and objectives without any particular assessment criteria for other data sources and/ or data collection methods). Finally, the evaluation of theoretical aspects (referred to by Hannes et al. (2010) as theoretical validity) is addressed only by JBI and ETQS; there are no assessment criteria for theoretical framework in CASP. Given these differences, it is unsurprising that CASP, JBI and ETQS have limited relevance for systematic case studies in psychotherapy. First, it is evident that neither of the three tools has specific evaluative criteria for the clinical component of systematic case studies. Although JBI and ETQS feature some relevant questions about participants and their context, the conceptualisation of patients (and/or clients) in psychotherapy involves other kinds of data elements (e.g. diagnostic tools and questionnaires as well as therapist observations) that go beyond the usual participant data. Furthermore, much of the clinical data is intertwined with the therapist's clinical decision-making and thinking style (Kaluzeviciute & Willemsen, 2020). As such, there is a need to appraise patient data and therapist interpretations not only on a separate basis, but also as two forms of knowledge that are deeply intertwined in the case narrative. Secondly, since systematic case studies involve various forms of data, there is a need to appraise how these data converge (or how different methods complement one another in the case context) and how they can be transferred or applied in broader psychotherapy research and practice. These systematic case study components are attended to a degree by CASP (which is particularly attentive of methodological components) and ETQS (particularly specific criteria for research transferability onto policy and practice). These components are not addressed or less explicitly addressed by JBI. Overall, neither of the tools is attuned to all methodological, theoretical and clinical components of the systematic case study. Specifically, there are no clear evaluation criteria for the description of research teams (i.e. different data analysts and/or clinicians); the suitability of the systematic case study method; the description of patient's clinical assessment; the use of other methods or data sources; the general data about therapeutic progress. Finally, there is something to be said about the recommended reviewer responses (Table 10). Systematic case studies can vary significantly in their formulation and purpose. The methodological, theoretical and clinical components outlined in Table 9 follow guidelines made by case study journals; however, these are recommendations, not (1997)(1998)(1999). Out of the three tools, ETQS is most attuned to the qualitative research paradigm; it seeks to assess and enhance evidence that is 'different' from the common efficacy research in EBP (Long & Godfrey, 2004). Developing purpose-oriented evaluation criteria for systematic case studies The starting point in developing evaluation criteria for Case Study Evaluation-tool (CaSE) is addressing the significance of pluralism in systematic case studies. Unlike RCTs, systematic case studies are pluralistic in the sense that they employ divergent practices in methodological procedures (research process), and they may include significantly different research aims and purpose (the endgoal) (Kaluzeviciute & Willemsen, 2020). While some systematic case studies will have an explicit intention to conceptualise and situate a single patient's experiences and symptoms within a broader clinical population, others will focus on the exploration of phenomena as they emerge from the data. It is therefore important that CaSE is positioned within a purpose-oriented evaluative framework, suitable for the assessment of what each systematic case is good for (rather than determining an absolute measure of 'good' and 'bad' systematic case studies). This approach to evidence and appraisal is in line with the PBE paradigm. PBE emphasises the study of clinical complexities and variations through local and contingent settings (e.g. single case studies) and promotes methodological pluralism (Barkham & Mellor-Clark, 2003). CaSE checklist for essential components in systematic case studies In order to conceptualise purpose-oriented

appraisal questions, we must first look at what unites and differentiates systematic case studies in psychotherapy. The commonly used theoretical, clinical and methodological systematic case study components were identified earlier in Table 9. These components will be seen as essential and common to most systematic case studies in CaSE evaluative criteria. If these essential components are missing in a systematic case study, then it may be implied there is a lack of information, which in turn diminishes the evidential quality of the case. As such, the checklist serves as a tool for checking whether a case study is, indeed, systematic (as opposed to experimental or clinical; see Iwakabe & Gazzola, 2009 for further differentiation between methodologically distinct case study types) and should be used before CaSE Purposebased Evaluative Framework for Systematic Case Studies (which is designed for the appraisal of different purposes common to systematic case studies). As noted earlier in the paper, checklist approaches to appraisal are useful when evaluating the presence or absence of specific information in a research study. This approach can be used to appraise essential components in systematic case studies, as shown below. From a pragmatic point view (Levitt et al., 2017;Truijens et al., 2019), CaSE Checklist for Essential Components in Systematic Case Studies can be seen as a way to ensure the internal validity of systematic case study: the reviewer is assessing whether sufficient information is provided about the case design, procedure, approaches to inquiry, etc., and whether they are relevant to the researcher's objectives and conclusions (Table 11). CaSE purpose-based evaluative framework for systematic case studies Identifying differences between systematic case studies means identifying the different purposes systematic case studies have in psychotherapy. Based on the earlier work by social scientist Yin (1984Yin ( , 1993, we can differentiate between exploratory (hypothesis generating, indicating a beginning phase of research), descriptive (particularising case data as it emerges) and representative (a case that is typical of a broader clinical population, referred to as the 'explanatory case' by Yin) cases. Another increasingly significant strand of systematic case studies is transferable (aggregating and transferring case study findings) cases. These cases are based on the process of meta-synthesis (Iwakabe & Gazzola, 2009): by Comprehensive and detailed responses in relation to the study examining processes and outcomes in many different case studies dealing with similar clinical issues, researchers can identify common themes and inferences. In this way, single case studies that have relatively little impact on clinical practice, research or health care policy (in the sense that they capture psychotherapy processes rather than produce generalisable claims as in Yin's representative case studies) can contribute to the generation of a wider knowledge base in psychotherapy (Iwakabe, 2003(Iwakabe, , 2005. However, there is an ongoing issue of assessing the evidential quality of such transferable cases. According to Duncan and Sparks (2020), although metasynthesis and meta-analysis are considered to be 'gold Two other types of systematic case study research include: critical (testing and/or confirming existing theories) cases, which are described as an excellent method for falsifying existing theoretical concepts and testing whether therapeutic interventions work in practice with concrete patients (Kaluzeviciute, 2021), and unique (going beyond the 'typical' cases and demonstrating deviations) cases (Merriam, 1998). These two systematic case study types are often seen as less valuable for psychotherapy research given that unique/falsificatory findings are difficult to generalise. But it is clear that practitioners and researchers in our field seek out context-specific data, as well as detailed information on the effectiveness of therapeutic techniques in single cases (Stiles, 2007) (Table 12). Each purpose-based case study contributes to PBE in different ways. Representative cases provide qualitatively rich, in-depth data about a clinical phenomenon within its particular context. This offers other clinicians and researchers access to a 'closed world' (Mackrill & Iwakabe, 2013) containing a wide range of attributes about a conceptual type (e.g. clinical condition or therapeutic technique). Descriptive cases generally seek to demonstrate a realistic snapshot of therapeutic processes, including complex dynamics in therapeutic relationships, and instances of therapeutic failure (Maggio, Molgora, & Oasi, 2019). Data in descriptive cases should be presented in a transparent manner (e.g. if there are issues in standardising patient responses to a self-report questionnaire, this should be made explicit). Descriptive cases are commonly used in psychotherapy training and supervision. Unique cases are relevant for both clinicians and researchers: they often contain novel treatment approaches and/or introduce new diagnostic considerations about patients who deviate from the clinical population. Critical cases demonstrate the application of psychological theories 'in action' with particular patients; as such, they are relevant to clinicians, researchers and policymakers (Mackrill & Iwakabe, 2013). Exploratory cases bring new insight and observations into clinical practice and research. This is particularly useful when comparing (or introducing) different clinical approaches and techniques (Trad & Raine, 1994). Findings from exploratory cases often include future research suggestions. Finally, transferable cases provide one solution to the generalisation issue in psychotherapy research through the previously mentioned process of meta-synthesis. Grouped together, transferable cases can contribute to theory building and development, as well as higher levels of abstraction about a chosen area of psychotherapy research (Iwakabe & Gazzola, 2009). With this plurality in mind, it is evident that CaSE has a challenging task of appraising research components that are distinct across six different types of purposebased systematic case studies. The purpose-specific evaluative criteria in Table 13 was developed in close consultation with epistemological literature associated with each type of case study, including: Yin's (1984Yin's ( , 1993 work on establishing the typicality of representative cases; Duncan and Sparks' (2020) and Iwakabe and Gazzola's (2009) case selection criteria for metasynthesis and meta-analysis; Stake's (1995Stake's ( , 2010 research on particularising case narratives; Merriam's (1998) guidelines on distinctive attributes of unique case studies; Kennedy's (1979) epistemological rules for generalising from case studies; Mahrer's (1988) discovery oriented case study approach; and Edelson's (1986) guidelines for rigorous hypothesis generation in case studies. Research on epistemic issues in case writing (Kaluzeviciute, 2021) and different forms of scientific thinking in psychoanalytic case studies (Kaluzeviciute & Willemsen, 2020) was also utilised to identify case study components that would help improve therapist clinical decision-making and reflexivity. For the analysis of more complex research components (e.g. the degree of therapist reflexivity), the purpose-based evaluation will utilise a framework approach, in line with comprehensive and open-ended reviewer responses in ETQS (Evaluation Tool for Qualitative Studies) (Long & Godfrey, 2004) (Table 13). That is to say, the evaluation here is not so much about the presence or absence of information (as in the checklist approach) but the degree to which the information helps the case with its unique purpose, whether it is generalisability or typicality. Therefore, although the purpose-oriented evaluation criteria below encompasses comprehensive questions at a considerable level of generality (in the sense that not all components may be required or relevant for each case study), it nevertheless 2. Descriptive cases that capture specific psychotherapy processes as they emerge in treatment (particularity); 3. Unique cases due to unusual variations that go beyond the 'average' population (deviation); 4. Critical cases that test existing theories (faslficiation/confirmation); 5. Exploratory cases that indicate a beginning phase of a multiple case study research (hypothesis generation); 6. Transferable cases that seek to aggregate and transfer case study findings onto other cases (generalisability). The studied phenomenon • What is the studied phenomenon or 'conceptual type' (e.g, clinical condition, therapeutic technique, patient's symptoms)? There is generally one specific phenomenon. • Is the studied phenomenon sufficiently distinguished from other kinds of (potentially similar) phenomena? Patient data • Are patient characteristics relevant to the wider clinical population? (e.g. is there a good match between symptoms and experiences?) • What is the rationale for choosing this patient? • Does the patient present any unique or deviant characteristics? (e.g. symptoms that are not representative of the studied clinical condition) The clinical discourse • Is there a detailed clinical narrative in the form of therapist reflections and observations? • Does the case move from the particularity of the patient to a more general (theoretically abstract) claim about the studied phenomenon? Research • Is there a sufficient review of literature on the studied phenomenon? • Does the case refer to other cases and/or studies that replicate their findings? Case purpose • Does the case demonstrate the typical characteristics of the studied phenomenon? • Does the case provide findings relevant for the broader clinical population? • Can the case contribute to psychotherapy theory? 2 Descriptive cases (purpose: particularity) The studied phenomenon • What phenomena are studied in the case (e.g. clinical condition, therapeutic technique, patient's symptoms)? There can be multiple phenomena. • Does the case present events and processes common to clinical practice? (e.g. therapeutic relationship difficulties) Patient data • Are patient characteristics described in detail, with particular attention to uniqueness, subjectivity and meaning of "lived experiences"? • Is the patient clearly positioned within their cultural and psycho-social context? The clinical discourse • Does the case convey the process behind therapist's practical decisions in the consulting room? • Does the case provide 'know-how' knowledge on how practitioners can deal with clinically salient issues and situations? • Does the therapist provide a reflexive account on how their views and theoretical assumptions might impact the therapeutic relationship and clinical decision-making? Research • Does the case include patient's self-assessment? (e.g. through self-report measures and dialogic exchange) • Does the case include excerpts of dialogue between therapist and patient? Case purpose • Does the case provide a relational understanding (with which readers can empathise) of the studied phenomenon? • Does the case narrative sufficiently portray 'real analytic practice' rather than 'ideal models'? (e.g. by demonstrating disparity between clinical theory/research and practice) • Can the case contribute to psychotherapy training and practice? 3 Unique cases (purpose: deviation) The studied phenomenon • What phenomena are studied in the case (e.g. clinical condition, therapeutic technique, patient's symptoms)? There can be multiple phenomena. • Does the case explain how the studied phenomena are different or unique from the established theory/research? (e.g. the patient's experience of transference is different from the experiences of transference across a broader clinical population) • Are patient characteristics described in detail, with particular attention to uniqueness, subjectivity and meaning of "lived experiences"? • What is the rationale for choosing this patient? The clinical discourse • Does the case convey a detailed description of therapeutic interventions and their effectiveness? • Does the therapist provide a reflexive account on how their views and theoretical assumptions might impact clinical decision-making, particularly in terms of their understanding of the uniqueness/deviation in the case? • Does the case include sufficient considerations as to the cause of the deviation/uniqueness in patient's clinical condition or symptoms? Research • Does the case convey more than one theoretical and/or research perspective? (e.g. clinical assessment by multiple practitioners or data analysis by multiple researchers) • Are there considerations of alternative explanations to the observed deviation/uniqueness of the case? (e.g. by referring to other published case studies or research) Case purpose • Does the case provide insight into a novel phenomenon? (e.g. by describing unique patient symptoms or experiences) • Does the case provide novel theoretical knowledge in relation to unique/deviant phenomenon? (E.g., by developing a new therapeutic technique) • Can the case contribute to psychotherapy theory, training and/or practice? 4 Critical cases (purpose: falsification/confirmation) The studied phenomenon • What is the studied phenomenon in the case (e.g. clinical condition, therapeutic technique, patient's symptoms)? There is generally one specific phenomenon. • Does the case seek to test an existing theory/research about the studied phenomenon? (e.g. testing the effectiveness of a well-established therapeutic intervention) Patient data • Are patient characteristics described in detail? • Is the patient clearly outlined within their cultural and psycho-social context? • What is the rationale for choosing this patient? The clinical discourse • Does the case link therapist narrative and observations with the theoretical/research considerations? • Does the case convey a detailed description of therapeutic interventions and their effectiveness? Research • Does the case convey more than one theoretical and/or research perspective? (e.g. clinical assessment by multiple practitioners or data analysis by multiple researchers) • Does the case show how the theory/research that is being tested accounts for the clinical observations in the case? • Does the case provide a sufficient explanation on why their chosen theory/research is more

appropriate than another? • If the case falsifies an existing theory/research, are there sufficient sample considerations? (e.g. the case may be unique and therefore the original theory/research still stands) Case purpose • Does the case examine an existing theory/research successfully? (e.g. by showing whether a theory is effective with a specific patient) • If the case falsifies an existing theory/research, does it offer any novel suggestions or revisions to the falsified theory/research? • If the case confirms an existing theory/research, does it rule out alternative explanations for the tested hypothesis? (e.g. to show that a therapeutic intervention is effective, the positive effects of other variables like medication may need to be ruled out) Exploratory cases (purpose: hypothesis generation) The studied phenomenon • What phenomena are studied in the case (e.g. clinical condition, therapeutic technique, patient's symptoms)? There can be multiple phenomena. • Is the case discovery-led, in the sense that it explores data as it emerges? seeks to engage with each type of purpose-based systematic case study on an individual basis (attending to research or clinical components that are unique to each of type of case study). It is important to note that, as this is an introductory paper to CaSE, the evaluative framework is still preliminary: it involves some of the core questions that pertain to the nature of all six purpose-based systematic case studies. However, there is a need to develop a more comprehensive and detailed CaSE appraisal framework for each purpose-based systematic case study in the future. • Is the patient clearly outlined within their cultural and psycho-social context? The clinical discourse • Does the case link therapist narrative and observations with the theoretical/research considerations? • Does the case narrative explore the 'how' and 'what' questions in relation to patient experiences and treatment processes? • Does the case identify complex processes and mechanisms in the treatment and link them to theory? Research • Is there a sufficient review of literature of the studied phenomenon? • Does the case convey more than one theoretical and/or research perspective? (e.g. clinical assessment by multiple practitioners or data analysis by multiple researchers) • Does the data converge? Are different/conflicting findings reported? Case purpose • Does the case convey more than one set of outcomes? • Does the case indicate future research trajectories? • Can the case contribute to psychotherapy theory, training and/or practice? 6 Transferable cases (purpose: generalisability) The studied phenomenon • What is studied phenomenon (e.g. clinical condition, therapeutic technique, patient's symptoms)? There is generally one specific phenomenon. • Is the studied phenomenon explicitly defined and differentiated from other kinds of (potentially similar) phenomena? Patient data • Are patient characteristics described in detail? • Is the patient clearly outlined within their cultural and psycho-social context? • Does the patient present characteristics typical of the studied phenomenon? Is there sufficient information (clinical, theoretical) to link the patient with the studied phenomenon? The clinical discourse • Is there a detailed clinical narrative in the form of therapist reflections and observations? • Does the case shed light on specific characteristics of the therapeutic process? (e.g. the development of therapeutic alliance) • Is the case narrative theme-focused? (e.g. the case identifies specific treatment patterns across different sessions) • Is there a clear description of the therapeutic process, usually involving a session-by-session description? Research • Is there a sufficient review of literature on the studied phenomenon? • Does the case involve a specific therapeutic, theoretical and research framework, and is the framework made explicit by the researchers? • Is there a clear description of the research process? (e.g. step-by-step description of data analysis procedures) Case purpose • Does the case provide information about common or specific psychotherapy processes? • Can the case be compared to and aggregated with other psychotherapy case studies on the basis of its studied phenomenon and formulation? Using CaSE on published systematic case studies in psychotherapy: an example To illustrate the use of CaSE Purpose-based Evaluative Framework for Systematic Case Studies, a case study by Lunn, Daniel, and Poulsen (2016) titled 'Psychoanalytic Psychotherapy With a Client With Bulimia Nervosa' was selected from the Single Case Archive (SCA) and analysed in Table 14. Based on the core questions associated with the six purpose-based systematic case study types in Table 13(1 to 6), the purpose of Lunn et al.'s (2016) case was identified as critical (testing an existing theoretical suggestion). Sometimes, case study authors will explicitly define the purpose of their case in the form of research objectives (as was the case in Lunn et al.'s study); this helps identifying which purpose-based questions are most relevant for the evaluation of the case. However, some case studies will require comprehensive analysis in order to identify their purpose (or multiple purposes). As such, it is recommended that CaSE reviewers first assess the degree and manner in which information about the studied phenomenon, patient data, clinical discourse and research are presented before deciding on the case purpose. Although each purpose-based systematic case study will contribute to different strands of psychotherapy (theory, practice, training, etc.) and focus on different forms of data (e.g. theory testing vs extensive clinical descriptions), the overarching aim across all systematic case studies in psychotherapy is to study local and contingent processes, such as variations in patient symptoms and complexities of the clinical setting. The comprehensive framework approach will therefore allow reviewers to assess the degree of external validity in systematic case studies (Barkham & Mellor-Clark, 2003). Furthermore, assessing the case against its purpose will let reviewers determine whether the case achieves its set goals (research objectives and aims). The example below shows that Lunn et al.'s (2016) case is successful in functioning as a critical case as the authors provide relevant, high-quality information about their tested therapeutic conditions. Finally, it is also possible to use CaSE to gather specific type of systematic case studies for one's research, practice, training, etc. For example, a CaSE reviewer might want to identify as many descriptive case studies focusing on negative therapeutic relationships as possible for Several theoretical and research perspectives are explored in order to tailor the most suitable approach for the patient, including attachment styles, mentalization and integrative approaches. One of the authors acted as a therapist, while the two other authors were involved in data analysis; this improved the data triangulation process. Several hypothetical assumptions were made about therapeutic setting and relationship and their suitability for this patient; they are shown to be highly effective and helpful later in the case (e.g. nondirective PP therapy was experienced as more helpful by the patient than directive CBT therapy). The case demonstrates that insight-oriented, nondirective PP can yield significant successes for patients with bulimia nervosa who also display low reflective functioning and insecure attachments. This case is an important critical follow-up to larger RCT study, which by and large favoured CBT to PP for patients with eating disorders. their clinical supervision. The reviewer will therefore only need to refer to CaSE questions in Table 13(2) on descriptive cases. If the reviewed cases do not align with the questions in Table 13(2), then they are not suitable for the CaSE reviewer who is looking for "know-how" knowledge and detailed clinical narratives. Concluding comments This paper introduces a novel Case Study Evaluationtool (CaSE) for systematic case studies in psychotherapy. Unlike most appraisal tools in EBP, CaSE is positioned within purpose-oriented evaluation criteria, in line with the PBE paradigm. CaSE enables reviewers to assess what each systematic case is good for (rather than determining an absolute measure of 'good' and 'bad' systematic case studies). In order to explicate a purpose-based evaluative framework, six different systematic case study purposes in psychotherapy have been identified: representative cases (purpose: typicality), descriptive cases (purpose: particularity), unique cases (purpose: deviation), critical cases (purpose: falsification/confirmation), exploratory cases (purpose: hypothesis generation) and transferable cases (purpose: generalisability). Each case was linked with an existing epistemological network, such as Iwakabe and Gazzola's (2009) work on case selection criteria for meta-synthesis. The framework approach includes core questions specific to each purposebased case study (Table 13 (1-6)). The aim is to assess the external validity and effectiveness of each case study against its set out research objectives and aims. Reviewers are required to perform a comprehensive and open-ended data analysis, as shown in the example in Table 14. Along with CaSE Purpose-based Evaluative Framework (Table 13), the paper also developed CaSE Checklist for Essential Components in Systematic Case Studies (Table 12). The checklist approach is meant to aid reviewers in assessing the presence or absence of essential case study components, such as the rationale behind choosing the case study method and description of patient's history. If essential components are missing in a systematic case study, then it may be implied that there is a lack of information, which in turn diminishes the evidential quality of the case. Following broader definitions of validity set out by Levitt et al. (2017) and Truijens et al. (2019), it could be argued that the checklist approach allows for the assessment of (non-quantitative) internal validity in systematic case studies: does the researcher provide sufficient information about the case study design, rationale, research objectives, epistemological/philosophical paradigms, assessment procedures, data analysis, etc., to account for their research conclusions? It is important to note that this paper is set as an introduction to CaSE; by extension, it is also set as an introduction to research evaluation and appraisal processes for case study researchers in psychotherapy. As such, it was important to provide a step-by-step epistemological rationale and process behind the development of CaSE evaluative framework and checklist. However, this also means that further research needs to be conducted in order to develop the tool. While CaSE Purpose-based Evaluative Framework involves some of the core questions that pertain to the nature of all six purpose-based systematic case studies, there is a need to develop individual and comprehensive CaSE evaluative frameworks for each of the purpose-based systematic case studies in the future. This line of research is likely to enhance CaSE target audience: clinicians interested in reviewing highly particular clinical narratives will attend to descriptive case study appraisal frameworks; researchers working with qualitative meta-synthesis will find transferable case study appraisal frameworks most relevant to their work; while teachers on psychotherapy and counselling modules may seek out unique case study appraisal frameworks. Furthermore, although CaSE Checklist for Essential Components in Systematic Case Studies and CaSE Purpose-based Evaluative Framework for Systematic Case Studies are presented in a comprehensive, detailed manner, with definitions and examples that would enable reviewers to have a good grasp of the appraisal process, it is likely that different reviewers may have different interpretations or ideas of what might be 'substantial' case study data. This, in part, is due to the methodologically pluralistic nature of the case study genre itself; what is relevant for one case study may not be relevant for another, and vice-versa. To aid with the review process, future research on CaSE should include a comprehensive paper on using the tool. This paper should involve evaluation examples with all six purpose-based systematic case studies, as well as a 'search' exercise (using CaSE to assess the relevance of case studies for one's research, practice, training, etc.). Finally, further research needs to be developed on how (and, indeed, whether) systematic case studies should be reviewed with specific 'grades' or 'assessments' that go beyond the qualitative examination in Table 14. This would be particularly significant for the processes of qualitative meta-synthesis and metaanalysis. These research developments will further enhance CaSE tool, and, in turn, enable psychotherapy researchers to appraise their findings within clear, purpose-based evaluative criteria appropriate for systematic case studies. Polyomavirus Nephropathy: Quantitative Urinary Polyomavirus-Haufen Testing Accurately Predicts the Degree of Intrarenal Viral Disease Background A qualitative highly predictive urinary test for polyomavirus nephropathy (PVN) is the PV-Haufen test. This article evaluates whether a quantitative PV-Haufen analysis, that is, the number of PV-Haufen shed per milliliter urine, predicts PVN disease grades and the severity of intrarenal PV replication. Methods Polyomavirus-Haufen were counted in 40 urine samples from patients with biopsy-proven definitive PVN. The number of PV-Haufen was correlated with both histologic PVN disease grades 1 to 3 and the number of SV40-T–expressing cells as indicators of intrarenal PV replication in corresponding renal allograft biopsies (manual counts and automated morphometry). Findings from quantitative PV-Haufen analyses were

compared to conventional laboratory test results, that is, BK viremia (quantitative polymerase chain reaction [PCR]) and BK viruria (quantitative PCR and decoy cell counts). Results Polyomavirus-Haufen counts showed excellent correlation (α0.77–0.86) with the severity of intrarenal PV replication and disease grades. In particular, low PV-Haufen numbers strongly correlated with early PVN grade 1 and minimal intrarenal expression of SV40-T antigen (P < 0.001). In comparison, BK viremia and viruria levels by PCR showed only modest correlations with histologic SV40-T expression (α0.40–0.49) and no significant correlation with disease grades or minimal intrarenal PV replication. No correlations were seen with urinary decoy cell counts. In contrast to conventional quantitative PCR assays or decoy cell counts, quantitative urinary PV-Haufen testing accurately reflects the severity of PV replication, tissue injury, and PVN disease grades. Conclusions Quantitative PV-Haufen testing is a novel noninvasive approach to patient management for the diagnosis and prediction of PVN disease grades and monitoring of disease course during therapy. (Transplantation 2015;99: 609-615) P olyomavirus nephropathy (PVN) is the most important viral infection in renal allografts. The incidence of PVN in kidney transplant recipients ranges from 4% to 12%. [1][2][3][4] ABO-incompatible grafts seem to be at increased risk with a prevalence of 18%. 5 Graft failure because of PVN occurs in approximately 20% to 30% with a higher prevalence of 50% in cases of sclerosed PVN disease grade 3 (data collected by the Banff working group on polyomavirus nephropathy). Specific anti-PV therapy is not available, and outcome largely depends on an early diagnosis of PVN with limited renal injury that tends to respond favorably to therapeutic intervention (mainly based on the reduction of immunosuppression). Polyomavirus nephropathy is defined morphologically by the histologic demonstration of polyomavirus (PV) replication in the renal parenchyma, and a renal biopsy is required to render a definitive diagnosis. [6][7][8][9] However, establishing a diagnosis of PVN in a kidney biopsy can be challenging. Deterioration of allograft function and a rise in serum creatinine levels does not always accompany early PVN, resulting in a possible delay of a diagnostic renal biopsy. 6,8 Patient management after kidney transplantation includes risk assessment for potential PVN with quantitative polymerase chain reaction (PCR) assays for BK virus loads in plasma or urine or quantitation of urinary decoy cells in urine cytology specimens. These assays have been the mainstays for both clinical care and intervention. 2,[10][11][12][13][14][15][16][17][18][19] They are based on the observation that all patients with PVN show systemic signs of PV replication with viremia or viruria, and the assumption that plasma PCR levels for BK virus reflect intrarenal PV replication. However, only a minority of patients with PV viremia or viruria ultimately develops PVN, underscoring a well-established fact: latency establishing DNA viruses, such as PV, can be reactivated without causing manifest disease. 8,18,[20][21][22][23][24] Moreover, in immunocompromised patients, BK virus-associated disease with viremia can also be seen in the bladder (hemorrhagic cystitis) or in the salivary glands (human immunodeficiency virus-associated salivary gland sclerosis), typically lacking concurrent PVN. [25][26][27][28] Thus, positive PCR-based BK viremia or viruria test results or the presence of decoy cells in the urine indicate viral activation and an increased risk for PVN, but the tests cannot reliably predict the actual presence of intrarenal viral disease, that is, definitive PVN. In the setting of PVN, an ideal biomarker would not only indicate the presence or absence of intrarenal disease but also provide information on the severity of virally induced tissue injury. We recently described a noninvasive biomarker to accurately predict PVN in fixed voided urine samples, that is, the PV-Haufen test. We showed that a qualitative analysis, that is, PV-Haufen present or absent, marked intrarenal disease with positive and negative predictive values exceeding 95%. 20,29 The PV-Haufen test is based on a novel concept, that is, the detection of dense, cast-like PV aggregates in fixed voided urine samples. Polyomavirus-Haufen form in virally injured renal tubules subsequent to PV release from infected tubular epithelial cells. They are then flushed out of the diseased kidneys and can be detected in the urine. In our previous publication, we reported that PV-Haufen were exclusively found in patients with biopsy-proven definitive PVN and were never seen in 139 control patients with varying degrees of BK viremia and viruria and absence of PVN. In patients with a definitive biopsy-proven diagnosis of PVN who were followed longitudinally, PV-Haufen shedding went from negative to positive at the time of initial biopsy diagnosis. Polyomavirus-Haufen testing subsequently remained positive during persistent viral nephropathy. At the time of disease resolution, PV-Haufen testing turned from positive to negative and remained negative thereafter during long-term follow-up. Thus, the qualitative detection of urinary PV-Haufen, that is, present or absent, is disease-specific and represents a novel clinical biomarker for definitive PVN. 20,30 Currently, it is undetermined whether a quantitative urinary PV-Haufen analysis, that is, recording the number of PV-Haufen shed per milliliter urine, could provide additional clinical information on the severity of intrarenal disease. It is also undetermined how a quantitative urinary PV-Haufen assay might compare to conventional mainly PCR-based laboratory screening tests for PVN. We address these questions in our current study that expands on our previous qualitative PV-Haufen test findings by now focusing on quantitative testing. Because PV-Haufen shedding is only found in patients with biopsy-proven definitive PVN, we hypothesize that the degree of urinary PV-Haufen shedding reflects the severity of virally induced kidney injury. The aim of our current study is to correlate traditional screening assays and quantitative PV-Haufen test results with the severity of intrarenal PV replication or PV burden and disease grades in patients with biopsy-proven PVN. Which test is most accurate? Correlation of Quantitative Screening Test Results With the Severity of Intrarenal PV Replication The severity of intrarenal PV replication was determined by using an immunohistochemical stain for the SV40-T antigen and manually counting the number of positive staining nuclei, the number of positive staining tubular cross-sections, or alternatively performing (automated) morphometry. Irrespective of manual or automated analysis, the number of urinary PV-Haufen most accurately reflected the severity of intrarenal PV replication-SV40-T expression with excellent correlation coefficients between 0.77 and 0.86 (Table 1 and Figure 1). Of note, correlations were tightest for comparisons based on the manual counts rather than the (automated) morphometric deconvolution analyses using the Aperio system. Plasma and urine PCR readings (BK viremia and viruria) showed a modest correlation with intrarenal PV replication (correlation coefficients between 0.40 and 0.49). No correlation was seen between the number of decoy cells in the urine and the SV40-T expression in the corresponding biopsy samples. For correlation studies of a subgroup of PVN cases with minimal PV burden, six biopsies with 25 or less intrarenal SV40-T antigen-expressing cells were selected. Five of these six biopsies lacked intranuclear PV inclusion bodies by light microscopy further supporting an early and limited PVN disease stage in these cases. All six cases showed low urinary PV-Haufen numbers that tightly correlated with the low intrarenal PV burden (σ = 0.91; P = 0.01). Lowest PV-Haufen counts were found in one case with only two intrarenal SV40-T antigen-expressing cells ( Table 2). The PV-Haufen counts increased in increments in the remaining five cases parallel to increased intrarenal PV load levels. In contrast, in this cohort, no significant correlations were found between all other screening assays and minimal intrarenal PV replication: plasma PCR σ = 0.20, P = 0.7; urine PCR σ = −0.09, P = 0.9; and decoy cell shedding σ = 0.02, P = 1.0. Correlation of Screening Tests With PVN Disease Grades Low numbers of urinary PV-Haufen were statistically significantly associated with early or mild PVN disease grade 1 in 12 of the 12 cases with a median PV-Haufen count of 282/mL urine and a maximum count of 677 PV-Haufen/mL urine seen in one patient. Polyomavirus nephropathy disease grade 2 showed median PV-Haufen counts significantly higher at 1693/mL (range, 112-9977/mL urine; P =< 0.0001). On an individual case basis, PV-Haufen counts showed only little overlap between PVN disease grades 1 and 2 (Table 3 and Figure 1S, A, SDC, http://links.lww.com/TP/B46). Polyomavirus nephropathy disease grade 3 demonstrated median PV-Haufen counts of 2822/mL, which differed significantly from PVN disease grade 1 (1 vs. 3 P = 0.002). In comparison, the degree of BK viremia and BK viruria by PCR testing and the degree of decoy cell shedding by urine cytology were not significantly associated with PVN disease grades 1-3 (Table 3). DISCUSSION Urinary PV-Haufen are tight three-dimensional cast-like viral aggregates that are easily detected in fixed urine samples by negative staining electron microscopy (EM). 20,29 In a previous study, we showed that the qualitative detection of PV-Haufen, that is, present or absent, in a fixed urine specimen predicted PVN with positive and negative predictive values greater than 95%. 20 In the current study, we extended our analysis to quantitative urinary PV-Haufen testing. We correlated the number of PV-Haufen in voided urine samples with the severity of PVN and compared the data to standard PV screening assays. This comprehensive comparative analysis showed that quantitative urinary PV-Haufen testing most accurately reflected the severity of intrarenal PV replication and virally induced renal injury. Few PV-Haufen in urine samples were associated with limited intrarenal PV replication, including one case with only two intrarenal SV40-T-expressing cells. Counts increased in increments parallel to the increase of intrarenal PV replication and high numbers of urinary PV-Haufen reflected marked virally induced tubular injury, a high intrarenal PV burden (correlation coefficient, 0.86). Quantitative PV-Haufen testing also reflected PVN disease grades as proposed by the Banff Working Group on PVN. 31 Early mild PVN disease grade 1 was reflected by low numbers of urinary PV-Haufen that differed significantly from much higher counts in the florid PVN disease grade 2, and the late and sclerosed PVN disease grade 3. How can the striking correlation between the severity of virally induced tissue injury in PVN and the numbers of urinary PV-Haufen shed into the urine be explained? The formation of PV-Haufen appears to be similar to the formation of other intratubular casts such as red or white blood cell casts. Polyomavirus-Haufen form within virally injured renal tubules containing high concentrations of uromodulin (Tamm-Horsfall protein) that has previously been shown to facilitate aggregation of influenza, mumps, and Newcastle viruses. 32 High concentrations of Tamm-Horsfall protein facilitate the aggregation of PV in vitro (personal observation; manuscript in preparation). 30 Because PV-Haufen are flushed with the urine out of tubules and into the bladder, they can serve as specific urinary biomarkers for intrarenal PVN. In contrast, quantitative PCR assays to detect BK virus in plasma and urine only showed modest correlation with the severity of intrarenal PV replication or PV burden (correlation coefficients, 0.43 to 0.48). No significant correlations were found between quantitative PCR test results and PVN disease grades or limited PVN with only minimal intrarenal disease. No significant correlations were seen with urinary decoy cell counts. These results could be explained by potential PV activation in sites other than the kidneys, such as the urothelium or salivary glands. [25][26][27][28] In daily practice, we do not recommend using quantitative urinary PV-Haufen testing as a mass screening tool. Rather, it should be used as a targeted test to separate patients with asymptomatic transient BK viral activation from those with intrarenal disease, that is, definitive PVN. The test provides specific diagnostic evidence of intrarenal PVN and gives prognostic information on the severity of virally induced tissue injury. It is particularly suited for the diagnosis of PVN in early disease grade 1 with minimal intrarenal disease. Quantitative urinary PV-Haufen testing could be used to monitor PVN progression or regression during persistent disease. Urinary PV-Haufen testing is performed using negative staining EM, a well-established, easily performed technique specifically developed for the identification and quantitation of viruses in various body fluids. 33 It has been in routine use in virology since the early 1960s and is the method of choice for viral studies at major virology centers, such as The Centers for Disease Control in Atlanta, GA. The incorporation of urinary PV-Haufen testing only requires "thinking outside the box." The turnaround time from receipt of a voided urine sample to reporting of results is approximately 3 hr with costs lower or comparable to PCR-based assays. Although this single center study has limitations because of relatively few cases analyzed, we nevertheless think that quantitative urinary PV-Haufen testing

will help to better define, diagnose, and manage PVN. In conclusion, urinary PV-Haufen are accurate biomarkers of PVN and the number of PV-Haufen shed in fixed voided urine samples accurately reflects the severity of intrarenal disease. Quantitative PV-Haufen testing allows for an early diagnosis of PVN and monitoring of changing intrarenal viral load levels during therapeutic interventions such as in the setting of new clinical drug trials. MATERIALS AND METHODS Based on the availability of concomitant blood, urine, and renal biopsy samples, cases of definitive PVN were selected from the archives of The University of North Carolina at Chapel Hill (total n = 40 sample sets). The study protocol was approved by the institutional review board. Per definition, all diagnostic renal allograft biopsies showed evidence of PVN. [6][7][8] Histologic material was evaluated by two pathologists (V.N., H.K.S.) at a multiheaded scope and scoring results based on consensus (including the count of SV40-Tpositive cells and tubules) were used for statistical analysis. Urine samples collected on the day of biopsy were analyzed by conventional urine cytology (ThinPrep; Hologic Incorporated, Bedford, MA) for decoy cells, negative staining EM for the detection and quantitation of PV-Haufen, and quantitative PCR for BKV gene sequences. For negative staining EM, urine was fixed in a 1:1 ratio of 4% paraformaldehyde with subsequent EM grid preparation using previously published methods. This protocol does not require any embedding procedures and thin sectioning is not needed (SDC, Materials and Methods, http://links.lww.com/TP/B46). 20,29,33 Plasma samples collected on the day of biopsy were analyzed by quantitative PCR for BKV gene sequences. Biopsy Samples and Diagnosis of PVN All renal allograft biopsies fulfilled the Banff '97 criteria for tissue adequacy and were evaluated according to standard guidelines. 34 The PVN classification into disease grades 1 to 3 followed proposals made by the 2013 BANFF working group on the classification of Polyomavirus Nephropathy. 31 Polyomavirus nephropathy grades (1 = early/mild; 2 = florid; 3 = late/sclerosed) were defined based on the severity of PV replication or degree of staining for the SV40-T antigen in combination with the Banff "ci" interstitial fibrosis score. Twelve of 40 cases were diagnosed as early or mild PVN grade 1 ( Figure 2S, A and D, SDC, http://links.lww.com/TP/B46), 23/40 as florid PVN grade 2 ( Figure 2S, B and E, SDC, http://links.lww.com/TP/B46) and 5 of 40 as late sclerosed PVN grade 3 ( Figure 2S, C and F, SDC, http://links.lww.com/TP/B46). Quantitative Analysis of the SV40-T Antigen Immunostain in Renal Biopsies Intrarenal polyomavirus replication was assessed by immunohistochemistry with a monoclonal mouse antibody directed against the SV40-T antigen following previously published protocols (Clone 416, Calbiochem, San Diego, CA) ( Figure 2S, D, E, and F, SDC, http://links.lww.com/TP/B46). 35 Expression levels of the SV40-T antigen, that is, the number of epithelial cell nuclei expressing the T antigen, were quantitatively analyzed by manual count and also by (automated) morphometry using the Aperio system (Aperio, Vista, CA). Manual Counts (a) The total number of tubular epithelial cell nuclei expressing the SV40-T antigen and (b) the total number of stained tubular cross-sections lined by one or more epithelial cells expressing the SV-40 large T antigen were recorded. Morphometric Analysis The percentage area of a renal biopsy demonstrating positive expression with the SV40-T antigen was accomplished as follows: Aperio Color Deconvolution Software version 10.0 was applied to 40× virtual images in "svs" format scanned on the Aperio CS scanner (Aperio, Vista, CA). The total biopsy area was determined using the total stained area of all cellular components read using a channel optimized for blue nuclear counterstain. To evaluate the % of SV40-T-positive staining in the total biopsy area (TS 1 ), the total immunopositive area including both specific intranuclear staining and nonspecific background staining was determined using the DAB channel as specified by Aperio (% DAB ). To calculate the specific intranuclear SV-40 large T antigen signals, the weak positive threshold (TS DAB ) was set to eliminate presumed nonspecific background staining as, for example, seen in the cytoplasm. The final % of SV-40-positive staining (% IMMUNOPOSITIVE ) was calculated as: Negative Staining EM for the Detection and Quantitation of PV Haufen PV-Haufen Definition "Haufen" (after a German word for "cluster" or "stack") were defined as discrete three-dimensional, cast-like, dense polyomavirus aggregates in voided urine samples analyzed by negative staining EM (Figure 2). Polyomavirus casts containing six or more polyomavirus virions were classified as "PV-Haufen" following our previously published approach. 20,29 Negative Staining EM and PV-Haufen Quantitation Negative staining EM was performed using previously published standard procedures with a two-step methodology of urine clarification and urine concentration (SDC, Materials and Methods, http://links.lww.com/TP/B46 and Table 1S, SDC, http://links.lww.com/TP/B46). 20,29,33 One EM grid was prepared for each voided urine sample. Haufen quantitation was performed according to established previously published methodologies specifically designed for the quantitation of viruses in body fluids. 33,36 Briefly, PV-Haufen in the voided urine sample adhered to the EM grid by electrostatic binding properties ( Figure 3S, SDC, http://links.lww.com/TP/B46). Electron microscopy grids were examined on a transmission electron microscope (LEO EM-910, LEO Electron Microscopy, Thornwood, NY, accelerating voltage, 80-100 kV). A total of 25 randomly selected grid fields were viewed at 50,000× magnification and PV-Haufen were counted in these 25 grid fields at 80,000-100,000× magnification. The examination time was, on average, 20 to 30 min per grid. The total number of PV-Haufen per mL urine was then calculated using the following formula: Quantitative PCR Plasma and urine BK polyomavirus loads were determined by a real-time quantitative PCR assay using the ABI PRISM 7900HT Sequence Detection System with well characterized probes and primers specific for BK virus as previously published. 15,20 Real-time detection of PCR products was achieved with a fluorescent hydrolysis (Taqman) probe. Primers and probes were purchased from TIB Molbiol LLC, Adelphia, NJ. Quantitative linearity of the assay extended from 10 to 10 9 measured copies of BKV equivalents correlating with a dynamic linear range of 250 to 2.5 × 10 10 BKV copies/mL urine or plasma. Detectable BKV DNA below the lower limit of linearity (<250 BKV copies/mL) was classified as a "lowpositive" result. Urine Cytology for Decoy Cell Quantitation All urine samples were processed for cytology using standard liquid-based procedures (Hologic Incorporated, Bedford, MA) and stained by Papanicolaou method. Decoy cells were identified and quantified as reported previously. 11 Statistical Analysis Statistical analyses were performed with standard programs (Microsoft Excel and Stata 12.1; Stata Corp, College Station, TX). The number of urinary PV-Haufen, number of decoy cells, urine and plasma BKV DNA load levels were correlated with: (a) the severity of intrarenal PV replication/ SV40-T staining using Pearson's correlation coefficient and Spearman's correlation coefficient and, (b) PVN disease grades 1 to 3 using Wilcoxon rank-sum testing, and Kruskal-Wallis testing with ties. Wilcoxon rank-sum testing for pairwise comparisons with Bonferroni correction was performed only if overall group comparison by Kruskal-Wallis testing was significant. All testing was performed with two-sided α = 0.05. Adhesions Assemble!—Autoinhibition as a Major Regulatory Mechanism of Integrin-Mediated Adhesion The advent of cell-cell and cell-extracellular adhesion enabled cells to interact in a coherent manner, forming larger structures and giving rise to the development of tissues, organs and complex multicellular life forms. The development of such organisms required tight regulation of dynamic adhesive structures by signaling pathways that coordinate cell attachment. Integrin-mediated adhesion to the extracellular matrix provides cells with support, survival signals and context-dependent cues that enable cells to run different cellular programs. One mysterious aspect of the process is how hundreds of proteins assemble seemingly spontaneously onto the activated integrin. An emerging concept is that adhesion assembly is regulated by autoinhibition of key proteins, a highly dynamic event that is modulated by a variety of signaling events. By enabling precise control of the activation state of proteins, autoinhibition enables localization of inactive proteins and the formation of pre-complexes. In response to the correct signals, these proteins become active and interact with other proteins, ultimately leading to development of cell-matrix junctions. Autoinhibition of key components of such adhesion complexes—including core components integrin, talin, vinculin, and FAK and important peripheral regulators such as RIAM, Src, and DLC1—leads to a view that the majority of proteins involved in complex assembly might be regulated by intramolecular interactions. Autoinhibition is relieved via multiple different signals including post-translation modification and proteolysis. More recently, mechanical forces have been shown to stabilize and increase the lifetimes of active conformations, identifying autoinhibition as a means of encoding mechanosensitivity. The complexity and scope for nuanced adhesion dynamics facilitated via autoinhibition provides numerous points of regulation. In this review, we discuss what is known about this mode of regulation and how it leads to rapid and tightly controlled assembly and disassembly of cell-matrix adhesion. INTRODUCTION The Origin of Complex Cell Systems The emergence of life is an enigmatic question that humankind has pondered for millennia. At some point millions of years ago, the last universal common ancestor (LUCA) appeared and all organisms on earth descended from this initial cell (Yutin et al., 2008), or so the story goes. The steps to produce such an organism are controversial and mystifying. However, evolution from this single cell seems somewhat easier to imagine. A key step in the formation of multicellularity and more complex organisms was the development of cell adhesion molecules. The ancient origins of integrin-mediated adhesions has been traced back to the genesis of multicellularity (Sebé-Pedrós et al., 2010;Brunet and King, 2017). Initially, these cell adhesion molecules enabled cells to form interactions with other cells to form cell-cell junctions and, in animal cells, interactions with the newly-acquired extracellular matrix (ECM) to form cell-matrix junctions. The ability of cells to form sheets of cells and attach to an underlying matrix was central to the development of multicellular animal life. These attachment points also developed into sensitive sensory modules, able to feel the mechanics of the microenvironment and adopt the role of mechanotransduction centers-enabling cells to monitor and respond to mechanical cues and convert them into biological signals to elicit different cellular responses. Autoinhibition as a Regulator of Protein Activity Key to the development of complexity is the ability to regulate adhesion and control the proteins that assemble together to form adhesive structures. One way that proteins can be dynamically regulated is via formation of an intramolecular interaction that maintains the protein in an inactive state until adhesion assembly is required. The regulation of autoinhibition, and the factors that enable regulation of the activity of the protein, provides regulatable checkpoints in the system ( Figure 1A). The concept of autoinhibitory domains can be found throughout biology-these are regions of proteins which can form intramolecular interactions which regulate behavior and activity. We refer the reader to the review on autoinhibitory domains by Pufall and Graves as it provides an excellent, still highly relevant description of the fundamentals of autoinhibition (Pufall and Graves, 2002). CELL-MATRIX ADHESIONS: ATTACHMENT AND SIGNALING CENTERS Integrin Adhesion Complexes The affinity of integrins for the matrix is regulated by the process of integrin activation whereby the integrin that resides in a low-affinity, compact arrangement at rest is converted to, and stabilized in, an extended high-affinity conformation for ligand (Shattil et al., 2010;Sun et al., 2016). Integrin activation serves as a paradigm for how protein activity can be regulated by conformation ( Figure 1B). Following activation, huge protein complexes coalesce onto the short cytoplasmic integrin tails, ultimately giving rise to large signaling complexes. The order of this assembly is the focus of intense research. Autoinhibition at the Heart of Integrin Adhesion Complexes Many of the proteins that assemble to form integrin adhesion complexes are recruited to the adhesion site in an inactive state and, upon receiving the correct signals, are activated and set to work mediating adhesion and mechanotransduction. As we will see later, autoinhibition is not limited to the pre-adhesion phase of the protein's life; autoinhibition plays a central role in coordinating the dynamic assembly and disassembly required for cell migration-cells need to adhere and detach in perfect synchrony to enable progressive directional movement. Due to limitation in space, it is not possible to detail all of the adhesion proteins that are regulated by autoinhibition. Here, we will focus on some of the best characterized to serve as paradigms of autoinhibition regulating adhesion. Integrins Perhaps the best place to start is at the integrins themselves. The inside-out/outside-in activation

of integrins represents a textbook example of autoinhibition and the regulation of ligandbinding affinity via allosteric effectors. Whilst the concept of adhesions being formed between cells and the substratum had been appreciated since the work of Abercrombie (Abercrombie, 1961), integrins, as the receptors mediating these cell-ECM links, were first characterized in the 80s (Tamkun et al., 1986) and over the last 35 years have been the subject of extensive research effort (a PubMed search for the term "integrin" on 9th September 2019 yields 73,512 results). The integrin family of proteins consists of transmembrane, α/β-heterodimeric cell surface adhesion receptors that generally consist of a large ectodomain, a single transmembrane domain and a short cytoplasmic tail domain. This structure facilitates the characteristic bidirectional signaling ability of integrins wherein signals that regulate countless crucial cellular activities can be transmitted across the plasma membrane. In vertebrates, 18 α-and 8 β-subunits have been identified which form noncovalent links in different combinations to generate 24 integrin heterodimers (Hynes, 1992). Many great reviews have covered the complexity of bidirectional signaling through integrins (Qin et al., 2004;Luo and Springer, 2006;Shattil et al., 2010;Campbell and Humphries, 2011) but it is worth a quick summary here. The concept of integrin activation has been around for over 30 years, and the first paper showing that conformational changes regulate integrins was published in 1990 (Frelinger et al., 1990). The concept of "outside-in" signaling, whereby extracellular ligands can trigger large conformational changes to the integrin structure leading to increased integrin activation, quickly followed (Du et al., 1991). The use of monoclonal antibodies that recognized epitopes on the integrins that are only accessible in certain conformations led to the notion of allosteric behavior of integrins (Mould et al., 1996;Askari et al., 2009) mediated by autoinhibition. "Inside-out" signaling refers to integrins sensing intracellular signals, resulting in the binding of talin and kindlin to the FIGURE 1 | Autoinhibition as a means of regulating protein activity. (A) Simplified schematic of the general principle of autoinhibition. (B) Integrins exist in a closed, autoinhibited conformation when inactive (left). Upon activation, the headpiece extends and the cytoplasmic tails are held further apart. In this high-affinity conformation, the headpieces can interact with the extracellular matrix and an adhesion can assemble. (C) The structure of the complex formed between the transmembrane domains of integrin αIIb and β3 [PDB ID: 2K9J Lau et al., 2009] showing the salt bridge formed between R995 of αIIb and D723 of β3 which maintains the integrin in the autoinhibited state. cytoplasmic tail of the β-subunit leading to activation of the integrin. In reality, it is likely that both of these bidirectional signaling axes work in tandem, with extra-and intracellular cues contributing constantly to orchestrate the overall dynamics of the integrin. The structural changes in integrin conformation that occur upon activation, leading to the relief of autoinhibition and exposure of the ligand binding sites both extracellularly and intracellularly, are extensive and result in large-scale reorganization of the ectodomains and separation of the transmembrane and cytoplasmic tail domains (Kim, 2003) ( Figure 1B). Integrins are generally thought to have three major conformations: a bent, closed form, an extended form with the headpiece of the ectodomain still closed, and an extended form with the headpiece open and the α-and β-cytoplasmic tails a greater distance apart (Shimaoka et al., 2002;Li et al., 2017). The binding sites for many integrin ligands are cryptic-integrin activation causes a switch to the extended, open conformation and triggers exposure of these surfaces. Autoinhibition in the intracellular region is maintained via an electrostatic interaction between the two cytoplasmic tails: for example, in αIIbβ3 integrin, Asp723 in the β3 tail binds Arg995 in the αIIbβ3 tail (Anthis and Campbell, 2011) (Figure 1C). This salt bridge is crucial for maintaining the low-affinity state, holding the legs together (Hughes et al., 1996;Vinogradova et al., 2002;Kim et al., 2009;Lau et al., 2009) and preventing the separation of the legs that drives activation of the integrin. Protein-protein interactions between the β-tail and the actin-binding protein talin (see next section) can relieve this autoinhibitory interaction and drive tail separation, propagating a conformational rearrangement of the ectodomains and unfurling to reveal the ligand-binding motifs (Harburger and Calderwood, 2009). The structure of the talin2 F2F3 domains bound to the cytoplasmic tail of the β1d tail [PDB ID: 3G9W Anthis et al., 2009] revealed that part of the activation process mediated by the talin head binding to integrins (Calderwood et al., 1999) was to not just break this salt bridge but to form an alternate salt bridge between the Asp723 and a conserved basic residue in talin [Lys327 in mouse talin2 Anthis et al., 2009]. Active talin, with a little help from the FERM domain protein kindlin (Rogalski et al., 2000;Moser et al., 2008), can therefore convert integrin from an autoinhibited to an active conformation. Talin activates integrins by binding to the cytoplasmic tail of the β-integrin subunit via a phosphotyrosine-binding (PTB)like region in F3-this domain is termed integrin binding site 1 (IBS1). As mentioned in the previous section, the interaction between the talin head domain with integrin is sufficient to relieve autoinhibition of the integrin, exposing the previously cryptic binding sites for integrin ligands. However, talin itself is regulated by multiple layers of autoinhibition. The first evidence that talin was autoinhibited came from the initial structural and biochemical characterization of talin in 1987 (Molony et al., 1987). Electron microscopy analysis of purified talin from chicken gizzard smooth muscle revealed that talin was a flexible elongated molecule but could adopt a more globular compact form at low ionic strength. It has since been demonstrated that the autoinhibition of talin into this compact cytosolic conformation (Goult et al., 2013a;Dedden et al., 2019) occurs primarily via an interaction between the integrin binding site in F3 and the rod domain R9 (Goksoy et al., 2008;Goult et al., 2009;Song et al., 2012) with additional weaker interactions including that between F2F3 and R1R2 (Banno et al., 2012;Goult et al., 2013a). The compact autoinhibited structure (Molony et al., 1987;Goult et al., 2013a) is facilitated by the formation of a talin homodimer, formed via the dimerization domain (DD) at the very C-terminal helix . This stabilizes the autoinhibited conformation via the various inter-and intramolecular interactions taking place. In this configuration, the integrin binding site in the F3 domain and the actin-binding sites in the rod are masked which implies that, in order for integrin activation and subsequent signaling to occur, a conformational change must be induced in order to relieve talin autoinhibition. Layers of Talin Autoinhibition Many binding sites in talin are concealed as a result of these layers of autoinhibition (Gough and Goult, 2018)-most notably the sites for integrin, actin and vinculin-but also the sites in the rod with which the majority of talin binding partners interact. The folded conformation has the rod domains wrapped around the edge of the compact conformation with the two head domains of the dimer buried inside (Goult et al., 2013a). The rod domains have binding sites on their faces, some of which will be buried inside the closed conformation and inaccessible. However, some binding sites are still accessible-for example, Rap1-interacting adapter molecule (RIAM) is able to interact with autoinhibited talin, binding to a folded surface on R2R3 which is outwardfacing. RIAM-mediated coupling of talin to Rap1's membranetargeting motifs is a key event before integrin activation can occur (Lee et al., 2009(Lee et al., , 2013Shattil et al., 2010;Yao et al., 2014a;Lagarrigue et al., 2015). R9 and integrin bind to the same site on F3, indicating that the autoinhibitory interaction of talin and its integrin binding are mutually exclusive. The specific sites for talin autoinhibition have been investigated: on R9, these include a negatively charged surface comprised of residues Asp1676, Asp1763, Glu1770, Glu1798, and Glu1805 ) which bind to the positively charged integrin "activation" loop on F3 including Lys316, Lys318, Lys320, Lys322, and Lys324 (Goksoy et al., 2008;Goult et al., 2009;Song et al., 2012). The crystal structure of the complex between F2F3 and R9 [PDB ID: 4F7G Song et al., 2012] and the recent cryo-EM structure of the autoinhibited form of talin (Dedden et al., 2019) have provided atomic detail of this interface and confirmed the key role of Glu1770 in mediating autoinhibition via a buried salt bridge with Lys318. Mutations to Glu1770 in R9 have been shown to reduce the autoinhibitory interaction , and this mutant has allowed detailed analysis of talin uncoupled from the upstream signaling pathways (see later section "Manipulation of autoinhibition by mutation"). The layers of autoinhibition in talin extend beyond the headtail interaction. Once this "top layer" of autoinhibition is relieved, the binding sites on talin for integrin and actin are exposed and form the core of the adhesion complex. Talin has three actinbinding sites (Hemmings et al., 1996) and the regulation by autoinhibition of two actin-binding sites in the rod region, ABS2 and ABS3, demonstrates two further layers of talin regulation. The C-terminal ABS3 (McCann and Craig, 1997;Gingras et al., 2008) is comprised of the two R13 domains of the dimer linked together via the DD. ABS3 is inaccessible in the closed, compact form (Goult et al., 2013a) but, upon talin unfurling, is available to bind actin. The affinity of ABS3 for actin is further regulated via autoinhibition within the R13 domain, as the first helix (upstream helix, USH) of R13 limits actin binding (McCann and Craig, 1997). This domain-level autoinhibition requires force and/or changes in local pH (Srivastava et al., 2008) to allow maximal actin binding. Once ABS3 engages actin, it can capture the retrograde flow of actin that begins to exert forces onto the tethered talin molecule. These forces can cause conformational changes in talin and relieve autoinhibition or disrupt binding interactions of other domains. The second actin binding site in talin, ABS2, is in domains R4-R8 in the middle of the talin rod (Atherton et al., 2015). In the absence of force, ABS2 is cryptic and maintained in a low-affinity autoinhibited state via the adjacent domains R3 and R9 (Atherton et al., 2015). Forces exerted on talin via ABS3 relieve this autoinhibition and reveal ABS2 which can then form the high-affinity, tension-bearing cytoskeletal linkages with actin (Atherton et al., 2015;Kumar et al., 2016;Ringer et al., 2017). Here, mechanical force is the major driver relieving ABS2 autoinhibition, but this only occurs following relief of the layers of autoinhibition preceding it: i.e., talin head-tail autoinhibition and the changes in ABS3 that facilitate actin binding to talin. Many of the binding partners of talin require sites which are not constitutively accessible. Rod domain partners may bind when domains are folded, unfolded, or at intermediate levels of folding (Goult et al., 2013b. These sites can be exposed by various stimuli; for example, talin has 11 cryptic vinculinbinding sites (VBS) which are buried in the hydrophobic core of the folded rod helical bundles. These VBS are exposed when talin is under force, causing sequential rod domain unfolding and allowing the first subdomain of the vinculin head, Vd1, K318 and E1770 are required for this interaction to take place, forming a key buried salt bridge. Adapted from crystal structure PDB ID: 4F7G (Song et al., 2012). (B) Vinculin: the vinculin tail, Vt, interacts with both Vd1 and Vd4 domains of the head to form a strong autoinhibitory conformation. Mutating residues D974, K975, R976, R978 (Cohen et al., 2005) results in a constitutively active vinculin by disrupting the interaction between Vt and Vd4. Adapted from crystal structure PDB ID: 1TR2 . (C) Focal adhesion kinase: F2 of the FERM domain of FAK interacts with the C-lobe of the kinase domain to keep FAK in a closed conformation, rendering a key tyrosine in the linker between these domains (Y397) inaccessible to phosphorylation. Adapted from crystal structure PDB ID: 2J0J (Lietha et al., 2007). Images made using PyMOL. Frontiers in Molecular Biosciences | www.frontiersin.org to bind (Hytönen and Vogel, 2008;del Rio et al., 2009;Yao et al., 2016). Vinculin binding to these VBS subsequently allows for stabilization and maturation of the adhesion (Yao et al., 2014a). Each VBS-containing rod domain has a different force threshold at which it unfolds, leading to different forces required to relieve autoinhibition of each VBS. Further, each VBS helix has a different mechanical stability, and so the VBS-vinculin interactions at each site also have different strengths (Wang et al., 2019).

It remains to be determined exactly which binding sites are available in each conformation, and the catalog of talin ligands is expanding constantly . Furthermore, the lifetimes of each conformation are regulated by autoinhibition modulated by many different signaling cues including PTMs, force and calpain cleavage (see later section "Control of autoinhibition"). This leads to a domino effect of autoinhibitory relief steps downstream of Rap1 activation: (i) RIAM is activated, which translocates talin to the plasma membrane where (ii) the talin head-tail autoinhibition is relieved. (iii) Once active, talin relieves the autoinhibition of integrin. (iv) By connecting integrins to the actin retrograde flow, talin autoinhibition is further relieved, first with activation of enhanced actin binding to ABS3. (v) As force increases, mechanical activation of vinculin-binding sites in talin occurs, initially by unfolding of the R3 domain, the least stable of the rod domains (Goult et al., 2013b;Yao et al., 2014a). (vi) Mechanical exposure of high-affinity actin binding takes place via relief of autoinhibition of ABS2. This is just one example of a simplified, linear route through the autoinhibitory landscape regulating adhesion assembly; many other factors can feed into and modulate these steps and the order in which they occur. Vinculin Like talin, vinculin was first discovered as a component of adhesion plaques (Geiger, 1979). Vinculin is a ∼116 kDa actinbinding protein comprised of a large, globular head consisting of four α-helix-containing domains (Vd1-Vd4) linked to a tail domain (Vt) by a proline-rich hinge region ( Figure 2B). When talin is subjected to force, sequential unfolding of helical bundles of the rod domain occurs which exposes up to 11 cryptic vinculin binding sites (VBS) to which Vd1 can bind Papagrigoriou et al., 2004). Filamentous actin, on the other hand, interacts with Vt (Johnson and Craig, 1995). Autoinhibition of vinculin, like talin, is mediated by a headtail interaction that conceals the binding surfaces for most known ligands, including for talin and actin (Cohen et al., 2005(Cohen et al., , 2006. This interaction occurs via two interfaces: between Vd1 and Vt and between Vd4 and Vt. Although these are individually low-affinity, the combined interaction amounts to a strong autoinhibitory interaction (Cohen et al., 2005). The crystal structure of full-length vinculin in its autoinhibited form (Bakolitsa et al., 2004;Borgon et al., 2004) shows the two head domains interacting with Vt. A constitutively active vinculin termed vinculin T12 has been developed wherein a cluster of four charged residues in Vt are mutated: Asp974Ala, Lys975Ala, Arg976Ala, and Arg978Ala. This mutant has significant loss of affinity to the vinculin head compared to wildtype vinculin (Cohen et al., 2005). More recently, an improved constitutively active vinculin has been developed, the T12K mutant, with Asp974 mutated to a lysine (Asp974Lys) to further destabilize the interaction with the head (Chorev et al., 2018). Constitutively active vinculin markedly reduces adhesion turnover (Humphries et al., 2007;Carisey et al., 2013), and locks talin into an extended conformation (Yao et al., 2014a). Not surprisingly, this loss of adhesion dynamics is lethal in flies (Maartens et al., 2016). Focal Adhesion Kinase (FAK) The mechanisms underlying autoinhibition of focal adhesion kinase (FAK) are well-characterized and, being a kinase, a striking example of coupling regulation of protein activity to downstream signaling cascades. FAK is another linear multidomain protein with numerous binding sites for ligands. As the name suggests, FAK contains an enzymatically active kinase domain able to phosphorylate tyrosine residues in target proteins. FAK comprises an N-terminal FERM domain, a kinase domain, a ligand-binding region and a C-terminal 4-helix bundle termed the focal adhesion targeting (FAT) domain (Figure 2C). At rest, FAK adopts an autoinhibited, closed conformation where the FERM domain directly binds and occludes both the catalytic cleft and the activation loop of the kinase domain (Cooper et al., 2003;Lietha et al., 2007). The crystal structure of autoinhibited FAK allowed identification of the residues on each domain involved in the interaction (Lietha et al., 2007). In this closed conformation, the FAK activation loop is sequestered, preventing autophosphorylation of Tyr397. Interestingly, FERMmediated autoinhibition of kinase activity has also been observed in the Janus Kinase (JAK) family. JAK proteins have a similar domain structure to FAK and share this common mode of regulation (Zhou et al., 2001). Two mutants have been designed which significantly increase kinase activity of FAK: a double mutant in F2 of the FERM domain (Tyr180Ala and Met183Ala) and a mutation in the kinase domain (Phe596Asp) (Lietha et al., 2007). Activation of FAK requires relief of autoinhibition and this can occur through the FAK-FERM domain engaging the integrin tail, leading to release of the activation loop and autophosphorylation. As a result, binding between FAK and the Src tyrosine kinase is enhanced, leading to an active FAK-Src kinase complex able to activate and regulate many other proteins as a major driver of adhesion signaling (Schlaepfer et al., 1999;Zhao and Guan, 2011;Horton et al., 2016). Further, upon release of autoinhibition, FAK can become part of a tethered linkage in force-transmission pathways, tethering to integrins via its FERM domain while the Cterminal region couples to cytoskeletal proteins. This suggests that mechanical force may also contribute to the lifetime of the active form of the protein (see later section: Mechanical regulation of autoinhibition). Rap1-Interacting Adapter Molecule (RIAM) An important regulator of the integrin adhesion complex assembly is Rap1-interacting adapter molecule (RIAM). RIAM is a Rap1 effector that can interact directly with talin and translocate it to the plasma membrane, and thus into close proximity to the integrins (Han et al., 2006;Lee et al., 2013). The direct interaction of RIAM and talin is mediated via the N-terminus of RIAM which contains two talin binding sites (TBS) which interact with four of the talin rod domains (Goult et al., 2013b), with high-affinity binding occurring to the R2 and R3 talin rod domains. RIAM is another linear molecule, comprised of the two TBS and a long flexible linker connecting to Ras association (RA) and pleckstrin homology (PH) domains ( Figure 3A). The structural basis of RIAM autoinhibition was recently shown to be mediated by the region between the two TBS and overlapping with TBS2, termed the inhibitory region (IN) (Figure 3A). This IN region, mapped to residues 27-93, was shown to interact with, and occlude, the Rap1 binding site on . Autoinhibition is mediated by interaction between the SH2 domain and phosphorylated Tyr527 at the C-terminus (Shenoy et al., 1992). (C) Zyxin, comprised of an N-terminal region with Proline-rich "ActA" repeats, Nuclear Export Sequences (NES), and three LIM domains. Autoinhibition is mediated by interaction between the ActA and LIM domains (Nix et al., 2001;Call et al., 2011). (D) DLC1, comprised of a Sterile Alpha Motif (SAM), a Rho-GTPase Activating Protein domain (RhoGAP) and a STeroidogenic Acute Regulatory protein-related lipid-Transfer domain (START). Autoinhibition is mediated by interaction between the SAM and RhoGAP domains (Kim et al., 2008). (E) FHOD1, comprised of a GTPase-Binding Domain (GBD), a DAD Interacting Domain (DID), a Formin-Homology 2 (FH2) and a Diaphanous-Autoregulatory Domain (DAD). Autoinhibition is mediated by interaction between the DID and DAD domains (Takeya et al., 2008). (F) WASP, comprised of an Ena/VASP Homology-1 domain (EVH1), a GTPase-Binding Domain (GBD), and a Verprolin homology, Central hydrophobic and Acidic domain (VCA). Autoinhibition is mediated by interaction between the GBD and VCA domains (Kim et al., 2000). the RA domain (Zhang et al., 2014;Chang et al., 2019) via a switchblade-type autoinhibitory interaction. A tyrosine residue forming part of the binding site, Tyr45, was also shown to be a substrate for focal adhesion kinase (FAK). This suggests a novel regulatory axis whereby FAK-mediated phosphorylation of RIAM relieves RIAM autoinhibition, exposing both the RA and TBS domains, increasing the ability of RIAM to (i) co-localize with Rap1 at the leading edge of cells and (ii) to recruit talin to the same location. A Glu60Ala/Asp63Ala mutation was shown to render RIAM constitutively active (Chang et al., 2019). This Rap1:RIAM:talin nexus is thus tightly controlled in a myriad of ways, and the dynamic balance of activity status of each protein is implicitly entangled. RIAM is likely activated by active Rap1 (Bos, 2005;Stefanini and Bergmeier, 2016), and the interaction with talin is likely to tilt the equilibrium balance further. The interaction between the PH domain and membrane phospholipids will further tip the equilibrium. The lifetime of the interaction can then be controlled by phosphorylation via FAK and dephosphorylation by the relevant phosphatase. All of these factors will control the activation state of RIAM, highlighting the complex balance of factors that determine the lifetime of interactions. Src Relief of autoinhibition of FAK following activation at adhesion sites leads to FAK autophosphorylation of Tyr397 (Frisch, 1996) which provides one mechanism for the recruitment of Src to adhesion complexes. Src, like FAK, is a non-receptor tyrosine kinase that associates with sites of adhesion. The cellular Src protein was originally discovered from its homology to the Rous sarcoma virus oncogene protein product, v-Src (Bishop et al., 1978). Determining the domain structure of Src transformed the field of cell signaling as it is comprised of three modular domains: a kinase domain (Src Homology 1), a phosphotyrosine-binding SH2 domain (Src Homology 2) and a polyproline-recognizing SH3 domain (Src Homology 3). Identification of these modular binding domains in Src (Pawson and Gish, 1992) set a paradigm for cell signaling pathways. Interestingly, Src is maintained in an inactive cytosolic autoinhibited state as the C-terminus of Src contains a tyrosine, Tyr527, which when phosphorylated interacts with Src's own SH2 domain (Shenoy et al., 1992) ( Figure 3B). Interestingly, v-Src is constitutively active as it lacks this regulatory C-terminal autoinhibition motif. Phosphorylation of FAK Tyr397 can also enhance Src activation since the Src SH2 domain binds FAK, thus displacing the autoinhibitory tail. Relief of Src autoinhibition can be sustained by phosphatases that dephosphorylate Tyr527 and switch the molecule to its active form. Src recruitment to, and phosphorylation of, downstream molecules allows many pathways to be regulated through Src, and drives crosstalk between integrin, Src and Rho-family GTPases (Huveneers and Danen, 2009). This provides one route through the complex activation process where a sequence of autoinhibitory interactions are relieved: integrin activation, leading to increased FAK activation, which leads to increased Src activation. All of these enhanced activities can be further modulated by post-translational modification. And the Rest… Many other adhesome components have also been shown to autoinhibit. The core adhesion proteins are all long linear molecules so it is possible that they all contain autoinhibitory domains. Table 1 contains a non-exhaustive list of adhesion proteins regulated in this way, and constitutively activating mutations that can be introduced. The autoinhibited and active conformations of the proteins zyxin, WASP, FHOD1, and DLC1 are shown in Figure 3. These are just a small selection of the 200+ proteins associated with integrin adhesion complexes, but highlights the similarities between the regulatory mechanisms. In each case, the activity of the protein is controlled by a head-tail interaction which is mediated via an autoinhibitory domain binding to, and occluding, a major binding site. Very few proteins in cells are constitutively active. Zyxin ActA and LIM region S142D Nix et al., 2001;Call et al., 2011 Frontiers in Molecular Biosciences | www.frontiersin.org CONTROL OF AUTOINHIBITION Studies into the regulatory complexity of proteins' activation status is revealing adhesions to be highly dynamic. With recent advances in microscopy, it is now possible to observe constant switching between inactive and active conformations of proteins within an adhesion, as exemplified by the mounting evidence that integrins segregate into distinct nanoclusters within focal adhesions (Rossier et al., 2012;Spiess et al., 2018;Orré et al., 2019). Mutants that perturb talin autoinhibition [E1770A Ellis et al., 2013;Haage et al., 2018] or vinculin autoinhibition [T12 Cohen et al., 2005;Carisey et al., 2013] also result in strong attenuation of adhesion dynamics. Together, these results suggest autoinhibition of adhesion components is critical to the tight, dynamic regulation. Methods of Relief of Autoinhibition There are numerous ways that autoinhibition of proteins is controlled with multiple entry points for the regulation of protein activity, some of which are shown in Figure 4. The equilibrium between autoinhibited and active protein is exquisitely poised, such that a subtle shift toward a more active state of one protein can be sufficient to perturb the system and cause large-scale changes to cellular processes. Rather like the "butterfly effect" in chaos theory where a minor change in the

initial conditions can lead to large changes in the system overall (Lorenz, 1962), these chaotic events give rise to order in the cellular world. Upregulation of a protein in response to a signal of some kind, say activation of a GTPase or interaction of two proteins, can trigger rapid amplification and cascades of activation steps that result in large global changes in the adhesion and the cytoskeleton of a cell. One common mechanism for relieving autoinhibition is posttranslational modifications (PTMs) (Figure 4A). Here, a kinase or other enzyme chemically alters the target protein and relieves the autoinhibition, switching the protein "on" (or "off " in the case of proteins like Src). PTMs can also stabilize an active state once open and thus control the lifetime of the active state. The most well-studied PTM is phosphorylation but there are many others, including acetylation, methylation (Gunawan et al., 2015), sumoylation (Huang et al., 2018), etc. The converse of this is the removal of PTMs, where a second enzyme such as a phosphatase removes the PTM and reverses the switch. As well as temporary PTMs, there are also irreversible modifications that can occur, such as proteolytic cleavage ( Figure 4B). Here, the protein is cleaved, separating into two polypeptides. The best-characterized proteases linked to adhesion belong to the calpain family of calcium-dependent cysteine proteases which cleave many adhesion components. Talin has been shown to be a substrate for calpain (Franco et al., 2004), as have paxillin and FAK (Chan et al., 2010). Calpain cleaves talin in three sites: one site in the neck liberating the head from the rod (Franco et al., 2004), one site at the C-terminus immediately before the dimerization domain , FIGURE 4 | Potential ways of regulating autoinhibition. Center: A schematic of a protein regulated by autoinhibition, where an autoinhibitory domain (square) binds to, and inhibits, a ligand-binding domain (circle). There are many ways that autoinhibition can be relieved including: (A) Post-translational modification: a reversible way to disrupt the autoinhibitory interaction and activate the protein; (B) Proteolytic cleavage: by severing the link between the two domains, the protein can be constitutively activated; (C) Allosteric activation: another protein, small molecule, lipid surface, etc. binds to the protein and triggers conformational change which leads to activation; (D) Direct competition: a protein interacting with the target protein opens up the molecule; (E) Mutual relief: two autoinhibited proteins come together to relieve each other's autoinhibition, leading to simultaneous activation; (F) Mechanical relief: the two domains form part of a mechanical linkage where, by being tethered apart, they are maintained in an active conformation. and a cryptic third site in the R10 domain (Zhang et al., 2012). Each of these cleavage sites will result in constitutively active talin fragments. Interestingly, whilst the initially-identified role of calpain cleavage was primarily to regulate adhesion disassembly (Franco et al., 2004), recent work has shown that calpain cleavage activation of talin is required at adhesion genesis to trigger proper adhesion formation, recruitment of further noncleaved talin and adhesion maturation (Saxena et al., 2017), with the cleavage products playing key additional, presumably non-adhesive roles. Many proteins are activated by interaction with an activating molecule-this can be a protein, the membrane, etc. Here, the activating molecule can activate the target via allosteric activation ( Figure 4C) where binding causes conformational changes to the target leading to relief of autoinhibition. This can also be via a small molecule binding ( Figure 4C): for instance, calpain proteases are activated by influx of calcium and, once active, can activate other proteins by cleavage. Another scenario is that binding of the activating moiety can activate by direct competition (Figure 4D) with the autoinhibitory interaction. For instance, talin can be activated by the plasma membrane, where the interaction between the plasma membrane and the F3 domain of talin is stronger than the autoinhibitory interaction holding talin in a closed conformation (Saltel et al., 2009;Song et al., 2012), causing talin to undergo large-scale reorganization upon membrane engagement. A further scenario that is likely to be playing a major role in adhesion assembly is the mutual relief of autoinhibition ( Figure 4E). Here, two proteins in an "off " conformation come together and mutually activate each other. Alternatively, one protein becomes activated and, in doing so, is able to activate another, triggering a "domino rally"-style activation cascade. Mechanical Relief-Mechanotransduction A common theme of adhesion regulation by autoinhibition is that the proteins are regulated by head-tail autoinhibition. Here, the protein folds in a manner that is mediated by autoinhibitory domains which, when interacting, mask binding sites for ligands. Upon relief of autoinhibition, the protein unfurls to reveal the active form and expose additional binding sites. The protein changes dramatically from a more compact form to an open and extended conformation. The protein will be in equilibrium between the closed and the open states, and the lifetime of each state will be controlled by the factors described in the previous section. Many proteins in adhesion complexes form tethered linkages where the head and the tail are held apart (Figures 4F, 5B), for instance, vinculin when active forms a linkage between talin and actin. This dramatically reduces the affinity of autoinhibition as the two interacting domains are held apart and less able to interact. In these tethered systems, autoinhibition is a major mechanism enabling mechanotransduction as tethering alters the dynamics of the protein maintaining it for longer in the active state. This effect can be further enhanced if the tether is under mechanical force: if the protein is stretched, it is even harder for autoinhibition to occur (Wang et al., 2019). Many proteins can adopt multiple conformational states where binding sites are masked via intramolecular interactions and autoinhibition is where the lowest energy state results in suppression of the protein's functional role. As mechanical forces can lower the energy of active states, they can thus relieve autoinhibition. Autoinhibition in proteins that form part of mechanical linkages in the cell are particularly sensitive to mechanical forces. Whilst force might not be required to open up talin or vinculin from their head-tail autoinhibition, force will play a major role in coordinating the lifetime of the active state (Wang et al., 2019), in essence holding the molecule in an open, extended conformation. As such, the affinity of the protein with its ligands is directly correlated to the force on the system. MANIPULATION OF AUTOINHIBITION BY MUTATION Detailed biochemical characterization of the autoinhibitory domains in proteins enables the development of precise mutations that disrupt autoinhibition and render the protein constitutively active. Such mutations (examples of which are shown in Table 1) provide a way of uncoupling proteins from their activation processes and enable the detailed study of systems where the upstream signaling pathways are uncoupled. Such approaches complete a pipeline from biochemical to in cellulo to in vivo characterization and can enable detailed analysis of the protein in its active form, detached from the complex regulatory elements that control their activities. CASE STUDY: E1770A-CONSTITUTIVELY ACTIVE TALIN The autoinhibition of talin is mediated by intramolecular and intermolecular interactions, and the major autoinhibitory domain has been mapped to the R9 of the rod domain binding to the integrin-binding site in F3. Biochemical characterization of talin autoinhibition showed that the integrin-binding F3 domain interacted with the talin rod (Goksoy et al., 2008). Further refinement showed that the major autoinhibitory interaction was between F3 and the R9 rod domain , an interaction mediated by a negative surface on R9 which interacts with and occludes the basic activation loop on F3 (Wegener et al., 2007;Song et al., 2012). This structural information enabled identification of a R9 mutation (E1770A) which disrupted the interaction with F3. Analysis of this E1770A mutant in the context of full-length talin in HUVEC cells (Kopp et al., 2010) showed that a constitutively active talin leads to adhesions forming much more rapidly and in significantly larger numbers. Introduction of the E1770A mutant in Drosophila (E1777A in fly talin) led to insight into the loss of essential developmental phenotypes (Ellis et al., 2013). In a mouse model this mutation results in various defects as a result of more mature and stable focal adhesions, which ultimately results in impeded wound healing (Haage et al., 2018). This pipeline from in vitro to in cellulo to in vivo, encompassing structural and cellular methods, allows a broader understanding of autoinhibition and the consequences of disrupting such integral regulatory mechanisms. As our understanding of autoinhibition of integrin-mediated adhesion proteins develops, such a multi-disciplinary approach is likely to be an increasingly effective means of investigation. PRE-COMPLEXES The mechanically-sensitive interactions that occur following integrin activation have been the subject of intensive research and are reasonably well-understood. However, how all these proteins come to be in the right place and time prior to activation is much less clear. This coalescence of proteins together prior to the assembly of force-dependent linkages would suggest the formation of pre-complexes ( Figure 5A). It is likely that autoinhibition maintaining the proteins in an "off " state plays a major role in the formation of these pre-complexes. The conditions prior to formation of mechanical linkages will be very different, suggesting that some of these proteins interact in different force-independent ways. The interactions between the core proteins in the pre-complex state prior to force are likely to be fundamentally different to those that form following activation. The precise mechanism of how these proteins interact to give rise to pre-complexes is not well-understood. Using "knock-sideways" experiments where proteins are fused with mitochondrial-targeting motifs, pre-complexes between paxillin and autoinhibited talin and vinculin have been observed (Atherton et al., 2019). Furthermore, the use of constitutively active talin and vinculin mutants targeted to mitochondria enables the study of complexes following activation of one component. Fully autoinhibited talin and vinculin do not preassemble on the mitochondria but, by simulating the tethered state of either talin or vinculin using mutation, the release of the head-tail interaction of one was sufficient to trigger interaction of the two proteins. Such technologies are a powerful way to study the most basic of questions-in a cellular context, which proteins bind which, when, and where? The rapid development of new sophisticated microscopy techniques is allowing the observation of complexes within adhesions to be seen with unprecedented levels of detail (Kanchanawong et al., 2010), facilitating detection of multiple populations of a molecule within an adhesion. For instance, vinculin has at least three distinct states within an adhesion (Case et al., 2015) dictated by interactions with paxillin, talin and actin. Inactive vinculin is initially recruited by paxillin to the plasma membrane at the integrin adhesion "ground zero." Subsequent vinculin activation, promoted by talin, leads to vinculin moving away from the plasma membrane and to a "signaling layer" and ultimately the "force transmission" layer. Further, evidence for pre-complexation of inactive adhesion proteins comes from the use of fluorescent fluctuation analysis approaches (Bachir et al., 2014). Here, talin and vinculin are seen to associate before the formation of the integrin-talin complex in the nascent adhesion. More recently, new technologies utilizing machine learning trained on high-resolution traction force microscopy, coupled with single particle tracking and fluorescence fluctuation time-series analysis, is enabling the genesis of nascent adhesions to be visualized (Han et al., 2019). This visualization of the earliest stages of adhesion assembly is revealing that talin-vinculin-paxillin pre-complexes are a prerequisite for efficient and meaningful adhesion maturation. Talin and vinculin are required to be together at the moment of force generation to provide efficient maturation-if this is disrupted, maturation is limited. Nascent adhesions where talin and vinculin are recruited at different times fail to mature efficiently. Pre-complex formation appears to be required for efficient assembly of force-bearing linkages and for adhesions to mature following traction force. It has recently been shown that many of the talin molecules at adhesion sites are non-force bearing, suggesting they are targeted to adhesion sites, ready for action, but are not all simultaneously engaged (Lemke et al., 2019). It will be interesting to explore if this is common to other adhesion molecules and what complexes maintain them at the adhesion site. Altogether, this amounts to compelling evidence that adhesion proteins form pre-complexes prior to adhesion assembly. Having the key proteins localized together at time zero may be a requirement for rapid and productive adhesion assembly. For example, talin, vinculin and paxillin

form a pre-complex (Bachir et al., 2014;Han et al., 2019) prior to association with integrins and prior to mechanical forces, and a second pre-complex of kindlin and integrin has been identified (Rossier and Giannone, 2016). Could these two pre-complexes coming together be sufficient to lead to mutual relief of autoinhibition and trigger adhesion assembly? The use of FRET-based tension sensors (Austen et al., 2015;Kumar et al., 2016), traction force experiments, microscopy and the rapid advances in artificial intelligence, material science and microfluidics are enabling incredible insight into the interactions of proteins in mature adhesions, in 3D and nascent adhesions as the resolution improves. These technical advances are helping to answer the major open question in the field regarding how these proteins assemble and the interactions that mediate complex formation prior to activation of the integrin. CONCLUSION AND PHILOSOPHICAL REFLECTIONS Cells sense the chemical and mechanical properties of their environment through integrins, clustered in punctate adhesion complexes linked to the actin cytoskeleton. The precision and speed with which these adhesions assemble following activation of an integrin is remarkable and highlights the need for incredible robustness in the process of assembling such huge multiprotein complexes. The core of most integrin adhesions comprises talin and kindlin bound to and coordinating the activation state of integrin. Once engaged to integrin, the talin rod serves as a mechanosensitive signaling hub (MSH) and, onto this hub, hundreds of proteins are recruited . Each protein is regulated, many by autoinhibition, and the activation status of each protein provides a regulatable point in the process. A signal leading to activation of one protein (i.e., integrin-ligand interaction outside the cell, or Rap1 activation inside the cell, etc.) might trigger a cascade of activation where autoinhibition of many proteins is simultaneously or sequentially relieved. This enables the system to exact both subtle and sweeping changes to the adhesion and signaling outputs of the adhesion. Tight regulation and control of adhesion is crucial during development and adult tissue homeostasis. Defects in this tight regulation are implicated in many pathological conditions. Out of the 60 core consensus integrin adhesome proteins, 32 have been shown to be involved in cancer development and progression (Winograd-Katz et al., 2014). Inherited gene mutations in adhesome components are a significant source of disease and disability. Misregulation of the activity of any protein in the network will disrupt the fine balance needed to ensure the exquisitely tuned adhesive and signaling balance. Mutations in adhesion proteins and regulators give rise to diverse malfunctions and diseases as they perturb this balance in different ways which tilt the system in different tissues. Various intra-and extracellular changes in the environment of the cell trigger subtle readjustments and changes to the adhesive structures enabling appropriate responses. The whole adhesion machinery can be affected from afar by tweaks and changes via signaling pathways that alter the regulation of adhesion proteins. Likewise, external adjustments to the world outside the cell can be propagated into the cell through these adhesions, and subtle change in the lifetime of an interaction, or small reduction in the affinity of the autoinhibition of a protein (such as by increased force extending the lifetime of a linkage), will trigger changes to the system as it reorients to this new homeostasis, adopting an altered adhesion signaling complex which augments the programming of the cell. AUTHOR CONTRIBUTIONS BG and RK wrote the manuscript. A standardized definition of placental infection by SARS-CoV-2, a consensus statement from the National Institutes of Health/Eunice Kennedy Shriver National Institute of Child Health and Human Development SARS-CoV-2 Placental Infection Workshop Pregnant individuals infected with SARS-CoV-2 have higher rates of intensive care unit admission, oxygen requirement, need for mechanical ventilation, and death than nonpregnant individuals. Increased COVID-19 disease severity may be associated with an increased risk of viremia and placental infection. Maternal SARS-CoV-2 infection is also associated with pregnancy complications such as preeclampsia and preterm birth, which can be either placentally mediated or reflected in the placenta. Maternal viremia followed by placental infection may lead to maternal-fetal transmission (vertical), which affects 1% to 3% of exposed newborns. However, there is no agreed-upon or standard definition of placental infection. The National Institutes of Health/Eunice Kennedy Shriver National Institute of Child Health and Human Development convened a group of experts to propose a working definition of placental infection to inform ongoing studies of SARS-CoV-2 during pregnancy. Experts recommended that placental infection be defined using techniques that allow virus detection and localization in placental tissue by one or more of the following methods: in situ hybridization with antisense probe (detects replication) or a sense probe (detects viral messenger RNA) or immunohistochemistry to detect viral nucleocapsid or spike proteins. If the abovementioned methods are not possible, reverse transcription polymerase chain reaction detection or quantification of viral RNA in placental homogenates, or electron microscopy are alternative approaches. A graded classification for the likelihood of placental infection as definitive, probable, possible, and unlikely was proposed. Manuscripts reporting placental infection should describe the sampling method (location and number of samples collected), method of preservation of tissue, and detection technique. Recommendations were made for the handling of the placenta, examination, and sampling and the use of validated reagents and sample protocols (included as appendices). Introduction Pregnant individuals infected with SARS-CoV-2 have higher rates of admission to the intensive care unit, requirement for mechanical ventilation, extracorporeal membrane oxygenation, and death than nonpregnant individuals 1e3 but the risk of placental, fetal, and pregnancy complications have been less well described. Studies show that maternal SARS-CoV-2 infection is associated with an increased risk of preterm birth and preeclampsia, 4e6 both of which may be placentally mediated or placentallyreflected complications. Moreover, published reports 7e9 show that firstand second-trimester infections with SARS-CoV-2 are possible and can result in pregnancy loss. Recent reports from Ireland found second-trimester miscarriage and stillbirth associated with presumed placental SARS-CoV-2 infection, placentitis, and specifically the B.1.1.7 variant. 10,11 These reports suggest that placental infection does carry substantial perinatal morbidity and that clinical surveillance of exposed pregnancies and placental evaluation for infection are warranted. In addition, placental infection does not equate with vertical transmission but may cause placental damage, which leads to perinatal morbidity without infection. Many have noted the triad of histiocytic intervillositis, increased perivillous fibrin deposition, and trophoblast necrosis associated with placental infection. 10,12e19 Although these histologic findings are not specific for infection, they are characteristic enough to indicate a study to evaluate whether the placenta was infected. Although early reports were focused on SARS-CoV-2 placental infection and in utero transmission, 19e26 larger studies and meta-analyses indicate that placental infection and vertical transmission were rare, 4,9,27e32 particularly in comparison with other maternal viral infections in pregnancy (such as cytomegalovirus, Zika, rubella, and untreated HIV 33e36 ). Diagnostic criteria for confirming placental infection have been inconsistent. Early reports conflated placental Pregnant individuals infected with SARS-CoV-2 have higher rates of intensive care unit admission, oxygen requirement, need for mechanical ventilation, and death than nonpregnant individuals. Increased COVID-19 disease severity may be associated with an increased risk of viremia and placental infection. Maternal SARS-CoV-2 infection is also associated with pregnancy complications such as preeclampsia and preterm birth, which can be either placentally mediated or reflected in the placenta. Maternal viremia followed by placental infection may lead to maternal-fetal transmission (vertical), which affects 1% to 3% of exposed newborns. However, there is no agreed-upon or standard definition of placental infection. The National Institutes of Health/Eunice Kennedy Shriver National Institute of Child Health and Human Development convened a group of experts to propose a working definition of placental infection to inform ongoing studies of SARS-CoV-2 during pregnancy. Experts recommended that placental infection be defined using techniques that allow virus detection and localization in placental tissue by one or more of the following methods: in situ hybridization with antisense probe (detects replication) or a sense probe (detects viral messenger RNA) or immunohistochemistry to detect viral nucleocapsid or spike proteins. If the abovementioned methods are not possible, reverse transcription polymerase chain reaction detection or quantification of viral RNA in placental homogenates, or electron microscopy are alternative approaches. A graded classification for the likelihood of placental infection as definitive, probable, possible, and unlikely was proposed. Manuscripts reporting placental infection should describe the sampling method (location and number of samples collected), method of preservation of tissue, and detection technique. Recommendations were made for the handling of the placenta, examination, and sampling and the use of validated reagents and sample protocols (included as appendices). infection with vertical transmission 22 , highlighting the need for consensus definitions and nomenclature in this regard. One report found SARS-CoV-2 infection in 100% of placentas from mothers with COVID-19 based on positive immunohistochemistry studies, 37 whereas others have rarely reported placental infection, if at all. 17,29,38,39 Investigators have employed a variety of techniques to diagnose placental infection, including reverse transcription polymerase chain reaction (RT-PCR) on placental homogenates, 9,40,41 immunohistochemistry, 9,19,37 RNA in situ hybridization (RNA-ISH), 9,21,29,39 and electron microscopy, 21,42e44 making it difficult to compare across studies and establish definitive evidence regarding placental infection risks. The relative sensitivity and specificity of these techniques in the placenta have not been rigorously studied, but we have included what is known in the appendices. Given the patchy nature of infection of the placenta by SARS-CoV-2, 45 legitimate concerns are raised about false positive and false negative results by various detection methods. Misclassification might lead to erroneous diagnoses with potential clinical and research implications. Indeed, knowing the true prevalence of placental infection depends on accurate diagnoses, and thus, these techniques are critically important to perform with validation and proper controls. Despite these caveats, a systematic review and metaanalysis of these disparate case reports and case series reported a placental infection rate of 7% (2 of 26 cases). 46 Vertical transmission rates of 1.1% to 3% have been noted in cross-sectional studies ranging in size from 101 to 2399 mothers. 4,27,30,32,47 There is clearly a need for standardization of methods to ensure robust and reproducible results and to facilitate cross-study comparisons. Although a definition of transplacental infection has been proposed (World Health Organization reference number: WHO/2019-nCoV/mother-tochild transmission/2021.1 48 ), the criteria for placental infection have not. Although not all viruses infect the placenta before infecting the fetus, 49,50 placental infection seems to be a necessary intermediate step for in utero transmission of SARS-CoV-2 to the fetus. 15 Objective, robust, consensusdriven criteria for the documentation of placental infection are needed to provide guidelines for research and clinical care. Herein, we present consensus definitions, tiered by rigor, formulated by a team of experts on the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD)eassembled task force, for what constitutes placental infection by SARS-CoV-2 for clinical and research applications. We include, in appendices, protocols for implementation with specifics and references. Methods A group of active investigators with expertise in placental pathology, virology, obstetrics, infectious diseases, immunology, and molecular biology convened virtually to discuss and critique methods used to diagnose placental infection with SARS-CoV-2. In this multidisciplinary virtual workshop, experts in each field presented the following methods for documenting placental infection by SARS-CoV-2 (highlighting the strengths and limitations of each method): RNA RT-PCR or quantitative RT-PCR from placental homogenates, immunohistochemistry, RNA-ISH, electron microscopy, and histopathology. A round-table discussion followed the presentations and dialogue continued over several days. A 3person team (D.J.R., A.G.E., R.J.R.) then devised a ranked template of specific diagnostic techniques and procedures in descending order of rigor (Table 1). This was then distributed to the larger committee for further discussion until consensus was achieved on the recommendations that follow. Recommendations for Definition of Placental Infection Recognizing that there may be a difference between what represents the most stringent criteria for defining placental infection and what can be done in the greatest number of real-world settings, we seek to offer guidance on definitions of placental SARS-CoV-2 infection in clinical settings and in biomedical investigation. As such, we offer 5 levels of rigor for detecting placental infection with SARS-CoV-2 (both acute and current or persistent) (Table 1), followed by a pragmatic approach for defining placental infection. For each individual investigation, we suggest choosing 1 Note that an alternative approach is a 2-step approach, in which RT-PCR is used as a screen and then followed up with one of the methods recommended to confirm "definite" or "probable" infection. This

hybrid/2-step approach would be more rigorous than RT-PCR alone and potentially more sensitive than the "definite" and "probable" approaches. 2) Electron microscopic detection of viral-like particles in placental tissues. Unlikely: Negative results from any of the above tests Table 3. Recommended placental handling, processing, and examination For RT-PCR testing, we recommend at least 2 fresh placental samples (0.5 cm 3 ) taken within 20 minutes of placental delivery (and not >1 hour after placental delivery to avoid RNA degradation), best obtained in the delivery room by appropriately protected and trained individuals, from the fetal side of the placenta (Figures 1 and 2) rinsed in sterile normal saline or phosphate-buffered saline, and then either snap-frozen (in liquid nitrogen or on dry ice until placed in À80 C) or placed in 5 to 10 volumes of RNAlater (Thermo Fisher Scientific, Waltham, MA) for subsequent freezing and storage at À80 C. Biopsies should be taken from a midpoint between the chorionic and basal plates and midway between the cord insertion and the placental edge (at least 3 cm from the cord insertion and 3 cm from the placental edge) to ensure viable villous parenchyma is collected (Figures 1 and 2). The fetal membranes should be dissected off the fetal surface and not included in the biopsy. Notably, 2 samples are recommended for protection against failure of RNA extraction and for biological sampling diversity of different placental Recommended reporting guidelines for scientific manuscripts Scientific manuscripts that report on placental detection of SARS-CoV-2 should report on: (1) Sampling method (including location and number of sites sampled) (2) Time from delivery to sample preservation Gross photograph of the full-thickness slab section through the white line in Figure 1 The white brace highlights chorionic plate-fetal membranes should be excluded from the biopsy. The white bracket indicates fetal parenchyma as target for sampling; approximately 0.5 cm depth is recommended. The white arrows highlight the brace and the bracket. Special Report ajog.org regions, given that patchy infection is possible. 45 Snap-frozen biopsies may permit single-cell RNA-Seq/nuc-Seq approaches, other RNA analyses, and protein analyses, whereas biopsies preserved in RNAlater can be used for RNA/DNA analyses and protein analyses, but not for scRNA-Seq/nuc-Seq. RNA quality is likely better for longer periods of time with the use of RNAlater. For RNA-ISH and immunohistochemistry studies, formalin-fixed paraffin-embedded fullthickness placental parenchyma cut at 5 mm onto glass slides should be used. All studied placentas should be examined and reported promptly by pathology following the Amsterdam Criteria for sampling and histologic diagnoses. 52 Conclusion We submit this set of recommendations to standardize the definition of SARS-CoV-2 infection of the placenta. These recommendations are made by a consensus panel of experts in the fields of obstetrics, virology, placental pathology, infectious disease, immunology, and molecular biology. The definitions are tiered by rigor of the diagnostic technique. Investigators and clinical care providers are encouraged to use the most rigorous method available for documenting placental infection in studies and in clinical diagnosis. Comparisons across studies are more valuable when the studies use the same standardized methods. We understand that not all, or perhaps not any, of these methods will be available at all institutions, but we predict that most institutions will have at least one recommended method at their disposal, and collaboration or consultation is advised for those that do not have any recommended method available. We anticipate that utilization of these recommended definitions of placental infection by SARS-COV-2 will facilitate comparisons of results among studies and that improved interpretation will follow, allowing for maximum impact of research in this area and optimization of patient care. Note that the gold standard of SARS-CoV-2 titer demonstration via plaque assay or TCID50 require a bio-safety level 3 laboratory setting, which is not achievable in most real-world settings. Although the test characteristics for SARS-CoV-2 RNA-ISH in placental samples are not well defined given the limited data in these samples, the D.T.T. laboratory has performed a paired analysis of the Centers for Disease Control and Prevention (CDC) approved quantitative reverse transcription polymerase chain reaction (RT-qPCR) assay and the RNA-ISH in a variety of human autopsy materials from patients who died of COVID-19. 53 Using 90 human autopsy samples, the D.T.T. laboratory performed the RT-qPCR assay using a cycle threshold (Ct) cutoff of 42.5 for both N1 and N2 primer sets as the gold standard. RNA-ISH in these paired samples had a specificity of 96.7% and a sensitivity of 76.7%. A paired analysis of placenta will be needed to determine the work test characteristics of RNA-ISH in this setting. Appropriate positive control should be a FFPE specimen that is from an individual confirmed to be positive for SARS-CoV-2 by RT-PCR or RNA-Seq and show positive signal by RNA-ISH in the appropriate tissues. Appropriate negative control should be a similar FFPE specimen that is confirmed to be negative for SARS-CoV-2 by RT-qPCR. Appropriate positive controls include infected lung tissues from patients who died of COVID-19 (autopsy cases) and known infected placentas appropriately preserved; appropriate negative controls include tissues with similar pathologic processes (eg, diffuse alveolar damage in the lung and placental tissues of the same gestational age with similar pathologic features). Controls should be performed on each lot of RNA-ISH reagents and on each batch run of cases. An example image of RNA-ISH performed using the above technique is provided in Figure 3. A poor appetite or ability to eat and its association with physical function amongst community-dwelling older adults: age, gene/environment susceptibility-Reykjavik study A poor appetite or ability to eat and its association with physical function have not been explored considerably amongst community-dwelling older adults. The current study examined whether having an illness or physical condition affecting one’s appetite or ability to eat is associated with body composition, muscle strength, or physical function amongst community-dwelling older adults. This is a secondary analysis of cross-sectional data from the age, gene/environment susceptibility-Reykjavik study (n = 5764). Illnesses or physical conditions affecting one’s appetite or ability to eat, activities of daily living, current level of physical activity, and smoking habits were assessed with a questionnaire. Fat mass, fat-free mass, body mass index, knee extension strength, and grip strength were measured, and the 6-m walk test and timed up-and-go test were administered. Individuals who reported illnesses or physical conditions affecting their appetite or ability to eat were considered to have a poor appetite. The associations of appetite or the ability to eat with body composition and physical function were analysed with stepwise linear regression models. A total of 804 (14%) individuals reported having conditions affecting their appetite or ability to eat and had a significantly lower fat-free mass and body mass index, less grip strength, and poorer physical function than did those without any conditions affecting their appetite or ability to eat. Although the factors reported to affect one’s appetite or ability to eat are seldom considered severe, their strong associations with physical function suggest that any condition affecting one’s appetite or ability to eat requires attention. Introduction Ageing is associated with a decline in physical function, which is linked to serious health issues, including falls and related injuries, a loss of independence, institutionalization, 1 3 and mortality (Fried et al. 2001;Cesari et al. 2005). Previous research has shown that weight loss due to insufficient food intake and anorexia increases the risk of malnutrition amongst the older population (Buys et al. 2014;Landi et al. 2017). Diet plays an important role in susceptibility to chronic diseases during the ageing process (Mendoza et al. 2007). Moreover, there are many conditions that can contribute to a decline in appetite or interfere with the ability to eat, which subsequently increases the risk of malnutrition (Schilp et al. 2011;Pilgrim et al. 2016). Physical conditions that may contribute to increasing the risk of malnutrition include decreased saliva production; poor dentition, including difficulties in chewing and wearing dentures (Schilp et al. 2011;Pilgrim et al. 2016); and a reduced sense of taste and poor oral health (Solemdal et al. 2012). Since gastric emptying occurs more slowly in older people, food tends to remain longer in the stomach, prolonging satiation, which can reduce their appetite (de Boer et al. 2013). Few meta-analyses and reviews on appetite amongst community-dwelling older adults have been conducted (Giezenaar et al. 2016;O'Keeffe et al. 2019). One meta-analysis on appetite reported that older adults have lower levels of appetite and energy intake than do younger adults, suggesting that ageing itself affects food intake (Pilgrim et al. 2016;Giezenaar et al. 2016). It has been reported that older adults with a reduced appetite are more likely to have illnesses and physical conditions that affect appetite (Lee et al. 2006). Other studies have reported that community-dwelling older adults with a poor appetite and those with a good appetite have different levels of dietary intake (Vesnaver et al. 2012;Meij et al. 2017). In general, older adults experience a gradual decline in muscle mass, strength, and function with ageing (Granic et al. 2017). A balanced diet is essential for older adults to live an independent and healthy life (McLean et al. 2016;Rempe et al. 2019). Evidence shows that people with a poor appetite or ability to eat have a significantly lower protein consumption than do controls (Meij et al. 2017). Malnutrition as well as insufficient protein and energy intake are associated with an increased risk of functional decline and mobility limitations (van der Meij et al. 2017;Houston et al. 2017;Rempe et al. 2019). However, few studies have examined the associations of a decreased ability to eat or a poor appetite with physical performance, muscle mass, and muscle strength amongst community-dwelling older adults (Schilp et al. 2011;Chang and Lin 2016). The aim of the current study was to investigate whether having an illness or physical condition affecting one's appetite or ability to eat is associated with body composition, muscle strength, and physical function amongst community-dwelling older adults. Age gene/environment susceptibility (AGES)-Reykjavik study The current study has a cross-sectional design and includes a secondary analysis of data from a large community-based population residing in Reykjavik, Iceland (n = 5764). The AGES-Reykjavik Study, which is a part of the Reykjavik study, was established in 1967 to prospectively study cardiovascular disease in Iceland. The AGES-Reykjavik Study is an epidemiological study focused on four biologic systems: the vascular, neurocognitive (including sensory), and musculoskeletal systems and body composition/metabolism (Harris et al. 2007). In brief, a total of 30,795 men and women who were born in 1907-1935 and were living in Reykjavik, Iceland were followed as part of the Reykjavik Study (Sigurdsson et al. 1995). In 2002, the cohort members were re-invited to participate in the AGES-Reykjavik Study. At that time, 11,549 (38%) participants from the Reykjavik Study were still alive. From this group, participants were randomly selected (Saczynski et al. 2008). The AGES-Reykjavik examination for each participant was completed in three clinic visits that took place within four to six weeks. Details of the study design and the baseline AGES-Reykjavik Study assessments have been described elsewhere (Harris et al. 2007). Assessment of appetite or the ability to eat Illnesses or physical conditions affecting appetite or the ability to eat were assessed by a questionnaire during the second clinic visit for the AGES-Reykjavik Study. The first question related to this issue was as follows: "Do you have an illness or physical condition that interferes with your appetite or ability to eat?" (yes/no). For the participants who answered "yes", a second question was asked, with the following response options: (1) problems with your teeth, (2) swallowing problems, (3) pain when chewing, (4) poor sense of taste, (5) poor sense of smell, (6) stomach/abdominal pain, (7) gas/ bloating, (8) unspecific problems with digestion or heartburn, (9) diarrhoea, and (10) other types of illness, including chronic long-and short-term diseases. The status of having a poor appetite or ability to eat was defined on the basis of the first question. The participants who reported having illnesses or physical conditions interfering with their appetite or ability to eat were considered to have a poor appetite, and those who did not report these conditions were considered to have a normal appetite. Anthropometric measurements Body weight was measured whilst the participants were wearing only light undergarments and socks. Each participant stood quietly in the centre of the platform (Marel M1100, Iceland) with his/her weight equally distributed on both feet, without

touching any other form of support. Standing height was measured in centimetres (cm) when the participant stood with his/her back against the wall on a wall-mounted stadiometer (Seca stadiometer 242, Germany), with the heels together. Body mass index (BMI) was calculated as the body weight in kg divided by the height in metres squared (kg/m 2 ). According to the BMI classification system from the World Health Organization (WHO 2019), the following groups were identified based on BMI: underweight (BMI below 18.5 kg/m 2 ), normal (BMI 18.5 to 24.9 kg/m 2 ), overweight (BMI 25.0 to 29.9 kg/m 2 ) and obese (BMI ≥ 30 kg/m 2 ). Fat-free mass (FFM) and body fat mass (BFM) were assessed using bioelectrical impedance analysis (BIA) (Xitron 4200 Analyser with BIS4200 Utilities Software, San Diego, CA, USA). Whilst Dual-energy X-ray absorptiometry is considered the gold standard measurement tool for body composition in large epidemiological studies, BIA is frequently used to evaluate body composition for both epidemiological and clinical purposes (Buchholz et al. 2004;Marra et al. 2019). Evidence has shown that BIA results are highly correlated with DXA results (Ramel et al. 2011), and BIA is a relatively simple, low cost, quick, and non-invasive technique (Buchholz et al. 2004;Dehghan and Merchant 2008). FFM was calculated on the basis of the sex, height, weight, and resistance of the individuals using the following equation, FFM (kg) = 4.228 + (0.463 * H 2 /R50) + (0.293 * W) + ( 5.060 * sex), where H = body height (cm), R50 = resistance 50 kHz, W = body weight (kg), and sex = 0 for women and 1 for men. BFM was calculated as follows: BFM (kg) = total weight (kg) − FFM (kg). Knee extension strength The maximum isometric strength in knee extension on the dominant side was measured using an adjustable and computerized dynamometer on a fixed chair (Good Strength, Metitur, Palokka. Finland). Knee extension strength is considered a standard measure of leg muscle strength (Rantanen et al. 1994(Rantanen et al. , 1999. Knee extension strength was measured in Newtons at a fixed knee angle of 60 degrees from full extension using a strain-gauge transducer that was fastened to the ankle with a belt. Three 4-s trials were performed with 30 s of rest between trials. The highest performance value was used for the analysis (Heistaro and Kansanterveyslaitos 2008). Participants were excluded from the test if they reported having had surgery on their legs or experienced any ischaemic heart conditions within the past 2 months. Grip strength Grip strength is an indicator of mobility and a determinant of functional performance amongst community-living older adults (Sallinen et al. 2010); it was assessed using an adjustable and computerized dynamometer on a fixed chair, with the participant in a sitting position (Good Strength, Metitur, Palokka. Finland). The chair was equipped with armrests that were height-adjustable to ensure that the participant's shoulders were relaxed whilst they held the bar positioned directly above the elbow. Three trials were performed with the dominant hand to measure the highest grip strength using the same procedure used for knee extension strength. Each trial lasted 4 s, and after each trial, the participant was allowed to rest for half a minute. The participants were instructed to squeeze the bar using his/her maximum strength. The highest maximum force (N) was used for analysis in the current study (Heistaro and Kansanterveyslaitos 2008). Participants were excluded from the test if they reported having had surgery on their hand or experienced any ischaemic heart conditions within the past 2 months. Assessment of physical function Physical function was measured using the assessments that have been used in other large epidemiological and clinical studies, including the 6-m walk test and the timed up-andgo (TUG) test (Lord et al. 2003). Physical function amongst older adults is generally measured with tests of lower extremity physical performance, including the 6-m walk test and TUG test (Cesari et al. 2005). 6-m walk test The time it took for individuals to walk a distance of 6 m at their usual pace was measured in seconds. The short-distance walk test is reliable when it is performed with standardized methods, and it is easy for older people to complete (Harris et al. 2007). The total walk time was measured two times, and the average of the two trials was used in the study. Timed up-and-go test The TUG test is useful in monitoring clinical changes over time and in evaluating the outcomes of an intervention programme (Whitney et al. 1998). For the TUG test, the total time from when the participant stands up from a sitting position to when the participant has walked 3 m, turned around, walked back to the chair, and sat down again is measured in 1 3 seconds. The TUG test is well known to be a useful screening tool for older people with balance problems (Podsiadlo and Richardson 1991). The participants were required to use their own footwear and could use a cane or walker if necessary. Participants were excluded if they were not able to rise from the chair (height = 45.5 cm) or walk without assistance. The time for the first completed trial was used in the study. Activities of daily living (ADL) In assessing physical function amongst older adults, epidemiologic studies often count the ADL items that participants find difficult or cannot perform (Jette 1994;Fried et al. 1999;Ostir et al. 2001;Tinetti et al. 2011). In this study, ADL ability was measured by self-reported difficulty in performing five different activities: walking across the room, eating, dressing, bathing, and getting in and out of bed without assistance. In their answers to each ADL question, participants selected between four levels of difficulty (no difficulty, some difficulty, much difficulty, and unable to perform the activity). The response for each ADL item was converted into a dichotomous scale (Lawton and Brody 1969). The choices of some difficulty, much difficulty, and unable to perform the activity were coded as 1. The choice of no difficulty was coded as 0. The summary score of the 5-item ADL questionnaire ranged from 0 to 5, where a score of 0 indicated that the participant had no difficulty performing any of the 5 ADL items, and scores of 1-5 corresponded to the number of ADL items which participants had difficulty in performing or were unable to perform. The reliability and validity of the ADL assessment used in the current study and the data obtained have not been studied. Statistical analysis The first analysis compared the characteristics of two groups (those with a poor appetite and those with a normal appetite). Analysis of variance was used for continuous variables, and chi-square (χ 2 ) tests were used for categorical variables. Second, linear regression analysis was performed with various models adjusted for covariates to examine the association of appetite or the ability to eat with body composition, muscle strength, and physical function. Model 1 was adjusted for age and sex, model 2 was additionally adjusted for FFM (or BFM) and height, and model 3 was additionally adjusted for the level of physical activity within the past 12 months and smoking habits. Weekly physical activity (never, rarely, < 1 h, 1-3 h, 4-7 h, > 7 h) and smoking habits (never and previously/current) were assessed by a self-report screening questionnaire constructed for the AGES-Reykjavik Study (Harris et al. 2007;Mijnarends et al. 2016). Model 3 was repeated in the subgroup after excluding people with an underweight BMI. The analyses were separately conducted for the parameters of body composition and physical function. For the association between appetite or the ability to eat and physical function, model 2 was adjusted for BFM because it is a more precise measurement than is weight or BMI (Beaudart et al. 2015). A general linear model was used to examine the trend in the prevalence of a poor appetite in quartiles of the continuous outcome variables and ADL dependence. p < 0.05 was considered statistically significant in all analyses, and all statistical analyses were performed using STATA software, version 10 (Statacorp, Texas, USA). Results Amongst the AGES-Reykjavik Study population (n = 5764), 804 (14%) individuals reported having illnesses or physical conditions interfering with their appetite or ability to eat and were included in the poor appetite group. Compared with the normal appetite group, the poor appetite group had a lower body weight, height, and FFM; less muscle strength; and worse physical function (Table 1). Table 2 shows the association between appetite or the ability to eat and body composition. After the full adjustment in model 3 (age, sex, BFM/FFM, height, PA, and smoking status), compared with the individuals who had no illnesses or conditions interfering with their appetite or ability to eat, those with a poor appetite or ability to eat had a significantly lower BMI (Model 3: beta = − 0.25, 95% confidence interval − 0.40 to − 0.10, p = 0.001) and FFM (− 0.73, − 1.15 to − 0.32, p = 0.001), whilst the association of a poor appetite or ability to eat with BFM was not significant. After the participants with an underweight BMI were excluded from model 3 for both BMI and FFM, the association was still significant, with 12-16% attenuation from model 3. Table 3 shows the association between a poor appetite or ability to eat with physical performance and ADL dependence. For knee extension strength, individuals with a poor appetite or ability to eat had significantly lower knee extension strength after the adjustment of model 2. However, the association became non-significant (model 3, p = 0.053) after the model was adjusted for physical activity and smoking habits. Grip strength was significantly associated with appetite or the ability to eat in model 3 (p = 0.018), even after individuals with an underweight BMI were excluded (p = 0.011), but it was non-significant after the model was adjusted for FFM (p = 0.076). Compared with the people who had a normal appetite, the people who reported having illnesses or conditions interfering with appetite or the ability to eat performed significantly worse in the timed 6-m walk test and TUG test and had higher ADL dependence after the full adjustment. The association remained similar even after the model was additionally adjusted for FFM. Figure 1 shows the trend in the unadjusted prevalence of a poor appetite or ability to eat according to the quartile levels of body composition, muscle strength, physical function, and ADL dependence. The prevalence of a poor appetite or ability to eat was the highest in the lowest quartile of BMI, FFM, knee extension strength, and grip strength as well as in the highest quartile/level of the 6-m walk test, TUG test, and ADL dependence scale. Discussion In the current cross-sectional study, we found that having illness or any physical condition interfering with appetite or the ability to eat was associated with body composition and poor physical function in a large cohort of community-dwelling older adults in Iceland, regardless of the participants having a lower BMI or lower BFM. Previous studies have examined the association of malnutrition and poor appetite with various health characteristics (Lee et al. 2006;Reijnierse et al. 2015;Giezenaar et al. 2016). However, research amongst community-dwelling older adults is generally lacking (Beasley et al. 2013;Giezenaar et al. 2016;Houston et al. 2017). Having conditions that contribute to a decline in appetite or interfere with a person's ability to eat subsequently increases the risk of malnutrition (Schilp et al. 2011). Poor appetite is recognized as a risk factor for low quality of life and functional decline amongst hospitalized patients and nursing home residents (Wilson et al. 2005). For community-dwelling older adults, having a poor appetite can be a predictive factor for the unintentional decline of food intake, which can lead to weight loss (Meij et al. 2017). The current study has several additional contributions to the current knowledge on the association between appetite or the ability to eat and physical function. The question used to assess appetite in the AGES-Reykjavik Study focused on illnesses or conditions interfering with appetite or the ability to eat, which was different from those used in many previous studies (Wilson et al. 2005;Lee et al. 2006;van der Meij et al. 2017). The association of a poor appetite or ability to eat and muscle strength

in the current study provided results similar to those in previous studies (Malafarina et al. 2013;Reijnierse et al. 2015;Lardiés-Sánchez et al. 2017). In particular, grip strength was significantly lower amongst individuals with a poor appetite or ability to eat, whilst knee extension strength showed a trend in the same direction. The strong association between a poor appetite and lower grip strength has also been reported in a previous study (Reijnierse et al. 2015). Our findings suggest that a poor appetite or ability to eat may accelerate functional decline through less muscle strength. Previous research has shown that oral pain and chewing impairment in older adults are significantly related to frailty and its components, not only through a nutritional pathway of involuntary weight loss but possibly through a loss of strength and physical performance and reduced physical activity (Okuyama et al. 2011;Kamdem et al. 2017). Muscle mass explains most of the variance in muscle strength , indicating that both muscle mass and strength might contribute to a decline in physical function Beavers et al. 2013;Reinders et al. 2015). The significant association between appetite or the ability to eat with FFM and muscle strength observed in the current study indicates that individuals having problems interfering with their appetite or ability to eat have lower levels of FFM and muscle strength as well as poorer physical function and higher ADL dependence. Even after additional adjustments were made for BFM and FFM and people with underweight BMI values were excluded from the analysis, having a problem interfering with one's appetite or ability to eat was strongly associated with poor physical function in the current study. A previous study showed strong evidence that people with a poor appetite, regardless of the cause, are at a high risk of malnutrition (Lardiés-Sánchez et al. 2017). Diet quality may be one of the factors that explain this association because it may vary amongst persons with different appetite levels (van der Meij et al. 2017). The current study suggests there is a link between having problems affecting appetite or the ability to eat and poor physical function amongst community-dwelling older adults, regardless of one's BMI and fat-free mass. However, future research needs to examine longitudinal associations between appetite and physical function amongst community-dwelling older adults. Regarding limitations, the current study was a secondary analysis of cross-sectional data. The study used a questionnaire to determine whether the participants had illnesses or condition interfering with their appetite or ability to eat and their current level of physical activity. These questions were originally constructed for screening purposes of the AGES-Reykjavik Study, and the validity of the questions was not studied. Although the information on the validity and reliability of the ADL summary score is not available as part of the current study, it is common practice to present results of ADL assessment by the number of activities that participants reported having difficulty in performing or were unable to perform Ostir et al. 2001;Volpato et al. 2008;Tinetti et al. 2011), whilst other studies use scores with various levels of difficulty or inability to perform the activity (Volpato et al. 2007;Iwarsson et al. 2009;Bouwstra et al. 2019). It should be noted that the ADL summary score does not refer to the level of difficulty or disability because the ADL assessment in the current study was different from a conventional ADL assessment. Furthermore, the questionnaire used to assess appetite or the ability to eat was different from the scale used in previous studies, which directly asked participants about their level of appetite (van der Meij et al. 2017). In addition, it is possible that the participants who completed physical performance measurements may have been healthier than those who were excluded or did not complete the measurement. Our analyses did not account for the effect of recent weight loss, the presence of other diseases, or the use of medication, which is nevertheless a major limitation of the study. Finally, due to the cross-sectional study design, it is not possible to draw conclusions about cause and effect relationships because low physical functioning also negatively impacts appetite or the ability to eat amongst community-dwelling older adults (Lee et al. 2006). In conclusion, our study shows there is an association between a poor appetite or ability to eat and body composition, muscle strength, and physical function amongst community-dwelling older adults. Our findings have the potential to encourage both older individuals and the community to (N, < 215.4, 215.8-274.6, 274.7-367.4, 367.9 +), 6-m walk (s, < 5.59, 5.60-6.37, 6.38-7.47, 7.48 +), and timed up and go (s, < 10.24, 10.25-11.76, 11.77-13.90, 13.91 +). ADL dependence (0 = independent, 1-5 = number of ADL items with any difficulty or unable to perform). ADL activities of daily living be aware of conditions that can interfere with or decrease appetite or the ability to eat, which could prevent negative consequences, such as malnutrition. Although some of the illnesses or conditions reported to interfere with appetite or the ability to eat are seldom considered to be severe (such as heartburn, gas, or bloating), their strong associations with physical function suggest that any condition interfering with appetite or the ability to eat amongst community-dwelling older adults requires attention. Author contributions MC: study concept and design, analysis and interpretation of data, and preparation of manuscript. MV, IG, and OGG: study concept and design, interpretation of data, and critical review of manuscript. VG and LJL: acquisition of participants and data. Funding The current analysis, applying data from the AGES-Reykjavik study, was supported by the European Horizon 2020 PROMISS Project "PRevention Of Malnutrition In Senior Subjects in the EU", Grant Agreement 678732. Compliance with ethical standards Conflict of interest All authors declare no financial interest/conflict of interest associated with this manuscript. Ethics approval The AGES-Reykjavik Study was approved by the Icelandic National Bioethics Committee (VSN 00-063) and by the Institutional Review Board of the U.S. National Institute on Aging, National Institutes of Health. Informed consent was signed by all participants. The current study protocol was approved by the National Bioethics Committee in Iceland (VSN-16-012-V2). Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/. Effectiveness of a parenting programme in a public health setting: a randomised controlled trial of the positive parenting programme (Triple P) level 3 versus care as usual provided by the preventive child healthcare (PCH) Background Considering the high burden of disease of psychosocial problems in children and adolescents, early intervention regarding problem behaviour of young children is very important. The Preventive Child Healthcare (PCH) offers a good setting to detect such problem behaviour and to provide parenting support to the parents concerned. This paper aims to describe the design of an effectiveness study of a parenting programme for parents of children with mild psychosocial problems after an initial, evidence based screening in routine PCH. Methods/Design The effects of the intervention will be studied in a randomised controlled trial. Prior to a routine PCH health examination, parents complete a screening questionnaire on psychosocial problems. Parents of children with increased but still subclinical levels of psychosocial problems will be assigned at random to the experimental group (Triple P, level 3) or to the control group (care as usual). Outcome measures, such as problem behaviour in the child and parenting behaviour, will be assessed before, directly after and 6 and 12 months after the intervention. Discussion Parenting support may be an effective intervention to reduce psychosocial problems in children but evidence-based parenting programmes that fit the needs of the PCH are not available as yet. Although the Triple P programme seems promising and suitable for a universal population approach, evidence on its effectiveness in routine PCH still lacks. Trial registration NTR1338 Background Psychosocial problems (e.g. aggressive behaviour, fear, anxiety) frequently occur in children and may lead to serious restrictions in daily functioning currently and in later life, and are the major cause of long-term work disability in young adults [1][2][3]. Several population-based studies in the Netherlands show that about 20% of all children struggle with psychosocial problems [4,5]. A study in primary and secondary education showed that 13% of all pupils had internalising problems, 11% had externalising problems and 3% had other behavioural problems [6]. Parenting style and child well-being are closely connected which makes parenting support a suitable way to decrease psychosocial problems in young children. Parents can help to prevent these problems in children by teaching them social skills [7]. Therefore methods for early treatment of psychosocial problems in children by enhancing parenting skills become increasingly available [8], but no evidence-based programmes for parenting support are available that suit the child healthcare. Only a minority of the children (13%) with psychosocial problems is under treatment by youth care or youth mental care [4,5,9] whereas early detection and treatment of psychosocial problems in children can improve their prognosis substantially ('the earlier, the better') [4,10,11]. In the Netherlands, Preventive Child Healthcare (PCH) offers an ideal opportunity for the early detection of psychosocial problems among preschool children, comparable to community pediatrics in the USA. In this system, child health professionals (further: CHP), i.e. doctors and nurses, working in preventive child healthcare offer routine well-child care clinics, including the early detection and treatment of psychosocial problems to the entire Dutch population [12,13]. Access is free of charge. This offers an ideal setting to provide parenting support following an evidence-based method of early detection of psychosocial problems in children. For this aim, there is a need for standardised parenting support interventions that are short and suit the competences of CHPs. Triple P level 3, the so-called Primary Care Triple P, is such a short intervention but evidence for it after an initial screening on psychosocial problems in children in a preventive child healthcare setting still lacks. The intervention consists of practical advice and coaching on managing a specific behavioural problem during four short, individual consultations (20-30 minutes) with parents and their child by a trained child healthcare nurse. It is part of a multilevel system of early intervention for parents of children who have or are at risk of developing behavioural or emotional problems which aims at preventing and decreasing psychosocial problems in children by providing parenting support [14,15]. Research showed that Triple P, including level 3, seems promising when compared with a wait-list control group that receives no help [16]. A quasi-experimental study on the effects of Triple P level 3 in the Netherlands [17] showed significant decreases in the emotional and behavioural problems of children just as effects on parental satisfaction, parental efficacy and overall parental sense of competence. However, a randomised controlled trial investigating the effects of parenting support after an evidence-based, initial screening on psychosocial problems in children has never been done before and longterm follow-up data is currently not available. Objective This paper aims to describe the design of an effectiveness study of a parenting programme for parents of children with mild problem behaviour after an initial screening in routine PCH. Methods/Design Design The study is designed as a randomised controlled trial (RCT) with follow-up assessments directly after completion of the intervention, after six months and after twelve months. Eligible families will be randomly assigned to either the Triple P intervention or the care as usual (control) condition. The Medical Ethics Committee of the University Medical Center of Groningen approved the study design, protocols, procedures and informed consent. Participation is voluntary and all participants sign an informed consent form. In order to describe the design of this study, the CONSORT statement is followed [18], a checklist that intends to improve the quality of the reporting of randomised controlled trials. Participants

Eligible participants for the trial will be selected from a community sample of parents of 9-11 year old primary school children in the four northern provinces of the Netherlands, who will be examined during their routine PCH screening,. In the Netherlands, the PCH examines almost all children (>90%) at regular times. The four Northern provinces cover approximately 13% of all Dutch children of this age [19]. Recruitment of study population Prior to the contact with the PCH, all parents will be asked to complete a screening questionnaire on psychosocial problems in children. As part of the study, we added a baseline questionnaire about parenting behaviour. During routine examination by the PCH, the SDQ (Strengths and Difficulties Questionnaire) [20] total problems score of all children will be computed. Eligible participants, i.e. parents of children with a subclinical SDQ total problems score of 11-13, will be identified and invited by the CHPs to participate in the trial. After informed consent, they will be randomly assigned to the intervention or care as usual group ( Figure 1). Families included in the trial will be asked to nominate a primary participating partner who will attend the intervention and complete the accompanying questionnaires. However, both parents are welcome to attend the counseling sessions, which will be carried out at the clinic or at home. Study inclusion and exclusion criteria The inclusion criteria are: 1) Age 9 to 11 years, 2) An SDQ total problems score in the subclinical range, i.e. 11-13 [21], 3) Parents must acknowledge mild problem behaviour in their child, 4) Parents are willing to work on their child's psychosocial problems. The exclusion criteria are: 1) A diagnosis of developmental delay, developmental disorder (e.g. autism), conduct disorder or ADHD in the child, 2) Currently receiving treatment for behavioural problems, 3) A chronic disease for which three or more medical consultations in the past two months have been made, 4) Parental divorce, death or severe disease of someone to whom the child feels attached to (e.g. (grand)parent, sibling, friend, nanny) in the past two months, 5) Parents being in therapy for psychological or relationship problems, 6) Parents being unable to read or speak Dutch, 7) Severe and/or general behavioural or emotional problems beyond the scope of the Triple P level 3 intervention, 8) Suspected parental dysfunction, such as child maltreatment, psychiatric disease, alcohol or drug abuse. Randomisation procedure At the clinic, a research assistant ensures that participants enrolled by CHPs are eligible or have to be excluded from the trial. Randomisation, based on a computer-generated randomisation programme, occurs at the level of individual children in sequence of entrance. To prevent unequal randomisation, participants are pre-stratified and randomised by centre using block randomisation (blocks of six). Sample size The SDQ will be the primary outcome measure for the power calculation. For an average 3-point decrease in SDQ total problems score, i.e. a change from the subclinical to the normal range, SD = 6 at alpha = 0.05 (twosided) and beta = 0.20, 64 pairs of children need to be included with an initial SDQ total problems score between 11 and 13. As a loss to follow-up rate of 20% is expected, an initial study population of 80 pairs of children is needed in each treatment group (total: 160). This suffices to show an effect size of 0.5 of the Triple P programme regarding parent-reported child psychosocial problems and parenting problems. Blinding Participants do not know which research group they will be assigned to. Nurses carrying out the treatment cannot be blinded for the allocated treatment. Since all follow-up questionnaires will be sent by mail, no direct influence by the researchers or the CHPs is likely to occur. After randomisation, all participants receive a research code unknown by the researcher. Therefore, analysis of the data by the researcher will be blind. Intervention The intervention to be evaluated is Triple P level 3. Triple P is a multilevel system of family intervention which provides five levels of intervention of increasing strength [22]. Triple P intervention at level 3 (Primary Care Triple P) is a brief, narrow-focus parent programme that is aimed at parents with specific concerns about their child's behaviour or development [23]. It combines advice, rehearsal and self-evaluation to teach parents to manage a discrete child problem behaviour during four individual consultations of 20-30 minutes with the parents and their child (Table 1) [24]. To ensure the quality of delivery, intervention nurses completed a two-day training course delivered by the Netherlands Youth Institute and the developers of the programme and accreditation in levels 2 and 3 of the Triple P method by an accredited practitioner prior to the start of the project. The professionals conducting the interventions are all nurses employed in the PCH daily practice. Professional adherence to the Triple P method is warranted through several supervision sessions with an accredited practitioner. Control condition Families in the control condition will receive the usual care initiated by a CHP (i.e. a community nurse). Protocols on psychosocial problems of the PCH-organisations prescribe, in case of a deviant SDQ total problem score, that the CHPs verify the gravity of the situation with the parents and, sometimes, with the schoolteacher. If the parents acknowledge the behavioural problems in their child and experience parenting problems, the CHP will try to clarify the problem and provide parenting support (e.g. strategies, tips, tricks). Usually CHPs have three extra contacts at their disposal to provide parenting support. If the problem exceeds the expertise of the CHPs, the parents and child will be referred to a specialist (e.g. psychologist, psychiatrist, child welfare). All CHPs have at least a bachelor's degree in nursing, which means that they have completed four years of education. Most of them also have received a two year specialisation training in community nursing. Outcome measures The primary outcome of the study is problem behaviour of the child after intervention, measured by the (1) Strengths and Difficulties Questionnaire (SDQ) [20], 30 items divided into 5 subscales on prosocial behaviour, hyperactivity, emotional symptoms, conduct problems and peer problems and (2) the Eyberg Child Behavior Inventory (ECBI) [25], 36 items on parental perception of disruptive behaviour including intensity and problem score. The secondary outcome of the study is parenting behaviour since the intervention aims at parenting as mediator. Parental competence and parenting style will be measured by the (1) Problem Setting and Behaviour Checklist (PSBC) [26], a 28-item rating scale that assesses how confident parents are in dealing with child behaviour problems in various settings and (2) the Parenting Scale (PS) [27], 30 items on parenting behaviour (permissiveness, overreactivity, verbosity). Parenting stress will be measured by (1) the Dutch Parental Stress Index (PSI) [28], 11 items on parenting stress, and (2) the 21-item Depression, Anxiety and Stress Scale (DASS) [29]. At baseline family characteristics will be assessed, i.e.: family situation, age of the parents, parental education, employment and financial situation of the parents, and the ethnicity of the parents and the child. Data collection procedure Data will be obtained by questionnaires. Participants will be asked to return each of the questionnaires within one week. To minimise loss to follow-up, the parents will be called by phone if the questionnaire has not been returned within one week. The outcome assessments will take place, regard for equal time intervals, directly after treatment (T1), six months (T2) and twelve months (T3) after treatment. Parents who complete all questionnaires will receive a gift voucher. Analysis Statistical analysis will be performed according to the 'intention-to-treat' principle. With multiple measurements over time, data on subjects lost to follow-up will be handled by imputation techniques. Change in child behaviour and parenting behaviour will be expressed as standardised effect sizes. To analyse the development of the outcome measures in time, a longitudinal data analysis technique, i.e. random coefficient analysis, will be applied. Baseline characteristics of the parents in the two research groups will be compared using Chisquared tests for categorical variables, Wilcoxon's test for ordinal variables, and t-tests for continuous variables. Reporting will follow the CONSORT guidelines. Time frame of the study The preparatory period will take six months. In this period, nurses will be trained to carry out the intervention (Triple P, level 3). Furthermore, CHPs will be trained in recruiting potential participants in the trial. The inclusion phase will last two years and the follow-up phase will be twelve months. Analysing data and reporting the findings will last for six months. Therefore, the total duration of the study will be four years. Discussion This paper presents the design of a randomised controlled trial to investigate the effectiveness of a parenting programme (Triple P level 3) for parents of children with psychosocial problems after an initial, evidence based screening in routine PCH. Research on the effectiveness of interventions in routine PCH is very scarce [8] and no previous trial has been accomodated in a further evidence-based procedure. Studying the effects of this intervention is important as it aims (1) to reduce the burden of disease of psychosocial problems in children [4,5] and (2) to contribute to the use of evidence-based care for children with psychosocial problems. Furthermore, introducing an evidencebased programme may lead to a more recognisable and consistent approach to parenting support by the PCH. This short intervention can be easily embedded in the regular procedure. If CHPs suspect parenting problems or behavioural problems in children due to incompetent parenting, they can offer help to parents in three extra contacts as is customary. If Triple P level 3 proves to be effective, this will be conducive to further implementation. CHPs as well as parents of children with mild problem behaviour may benefit from a structured approach to working on and solving problem behaviour in young children. If proven effective, level 3 of the Triple P programme in combination with an available evidencebased detection of psychosocial problems in children may lead to a comprehensive evidence-based method to reduce the burden of disease due to psychosocial problems. Governmental organisations and policymakers may use the results of this study to develop future policy concerning parenting support provided by community healthcare workers. Strengths and limitations A randomisation procedure is applied to reduce the risk of selection and allocation bias. Since participants do not know each other, mutual influencing is considered unlikely. A broad array of outcome measurements gains insight into the treatment effect in many areas of parenting and child behaviour. Furthermore, this study is original in evaluating the effectiveness of an intervention on parenting support in a preventive healthcare organisation (i.e. the Dutch PCH) delivered with regular staff from multiple centres. While the intervention will be carried out in the daily PCH practice, conclusions can be generalised without reservations which makes the external validity of the trial strong. The intention-totreat analysis gives the trial high internal validity. Contrary to earlier studies, in this research the long-term effects will be assessed by measurements after twelve months after treatment. There are also some limitations. This study does not provide an independent, professional evaluation of psychosocial problems in children apart from parental judgement. However, parental concerns are a good indicator of problems indentified by professional [30] and observational research is difficult to objectify. Furthermore, the effects of the intervention may depend on the CHPs affinity for parenting support and treatment adherence. Assessment of the effects of parenting support at the level of independent health care workers within this study would be to laborious. The results of this study will become available in 2012. Ending life with grace and agreement Making the decision to withdraw or withhold life-sustaining care is difficult at the best of times, but when families and caregivers disagree about the course of action, the conflict disrupts patient care and consumes enormous personal resources for all parties. The case of Samuel Golubchuk, an ways emphasizes the best interest of the patient. In these circumstances, it is the disease that kills the patient, not the doctor or the treatment. Second, some families will overreach in their demands to preserve life, often because they use fundamentalist religious beliefs or cultural norms to narrowly define the best interest of the patient. In dying, just as in living, no right, including the constitutional freedom to one's religion, is absolute. Requiring caregivers to stand in the way of the natural process of dying by forcing them to

impose what, under the circumstances, are extraordinary measures with no foreseeable benefit, should not be accepted by the courts or society at large. Although religion and culture assist in better understanding the context of decisionmaking, it should not impinge on the rights of others. To make religion a factor in allocating a hospital bed is inequitable, for it privileges religious patients over others who may need care more urgently and whose chance of a favourable outcome may be better. Third, it is right that families who have strong beliefs, religious or otherwise, about the deaths of their loved ones be heard. In dying, just as in living, procedural fairness must be respected. The judge in Mr. Golubchuk's case noted that "communication broke down between the [family] and the physicians" and criticized the hospital for not following the recommendation of the Canadian Critical Care Society, which recommends "recourse to either mediation or adjudication" in cases of impasse. 2 By not providing impartial, arm's-length mediation as part of a well-established process including clear rules and laws, the hospital paid dearly: it was sued. Finally, it may be worthwhile to consider law reform. The State of Texas, for example, enacted the Texas Advance Directive Act, 1999, which provides a framework and a timeline for resolving disputes. 5 It outlines the process for families and physicians to talk and includes a disputeresolution mechanism as well as a time limit of 12 days. If the process fails, the hospital, in conjunction with the family, must try to arrange transfer of the patient to another physician or institution willing to give the treatment requested by the family. If no such provider can be found, the hospital and physician may unilaterally withhold or withdraw therapy that has been deemed futile, although the patient or surrogate can ask a state judge for an extension of time. In a recent study of the implementation of this directive over 2 years at Baylor University Medical Center in Dallas, Texas, the authors found that the data "suggest that the law represents a first step toward practical resolution of this controversial area of modern health care." 5 Although the whole law is not perfect, it offers an instructive example. And first steps are important. Ultimately, neither side in the Golubchuk case seems especially in the right. The hospital failed the family by not giving them a fair and impartial hearing, and the family failed society by using their religion to privilege their father over other needy patients. As the judge noted, the 2 sides "might have avoided litigation" if they had used "a knowledgeable, trained and objective mediator." 2 To avoid future failures, widespread adoption of formal mediation processes, or even better, laws like that in Texas, would provide much needed clarity. Acceptability, Precision and Accuracy of 3D Photonic Scanning for Measurement of Body Shape in a Multi-Ethnic Sample of Children Aged 5-11 Years: The SLIC Study Background Information on body size and shape is used to interpret many aspects of physiology, including nutritional status, cardio-metabolic risk and lung function. Such data have traditionally been obtained through manual anthropometry, which becomes time-consuming when many measurements are required. 3D photonic scanning (3D-PS) of body surface topography represents an alternative digital technique, previously applied successfully in large studies of adults. The acceptability, precision and accuracy of 3D-PS in young children have not been assessed. Methods We attempted to obtain data on girth, width and depth of the chest and waist, and girth of the knee and calf, manually and by 3D-PS in a multi-ethnic sample of 1484 children aged 5–11 years. The rate of 3D-PS success, and reasons for failure, were documented. Precision and accuracy of 3D-PS were assessed relative to manual measurements using the methods of Bland and Altman. Results Manual measurements were successful in all cases. Although 97.4% of children agreed to undergo 3D-PS, successful scans were only obtained in 70.7% of these. Unsuccessful scans were primarily due to body movement, or inability of the software to extract shape outputs. The odds of scan failure, and the underlying reason, differed by age, size and ethnicity. 3D-PS measurements tended to be greater than those obtained manually (p<0.05), however ranking consistency was high (r2>0.90 for most outcomes). Conclusions 3D-PS is acceptable in children aged ≥5 years, though with current hardware/software, and body movement artefacts, approximately one third of scans may be unsuccessful. The technique had poorer technical success than manual measurements, and had poorer precision when the measurements were viable. Compared to manual measurements, 3D-PS showed modest average biases but acceptable limits of agreement for large surveys, and little evidence that bias varied substantially with size. Most of the issues we identified could be addressed through further technological development. Introduction Information on body size and shape is used to interpret many aspects of physiology in children, including nutritional status, cardio-metabolic risk and lung function [1][2][3][4][5]. Measurements of regional body girths or diameters can provide detailed information on the size of specific body components such as the limbs, abdomen and chest. Such data have traditionally been obtained through manual anthropometry, and are considered relatively reliable, and acceptable to the subject, in most age groups. However, manual anthropometry becomes time-consuming if multiple outcomes are desired, and the measurement of multiple dimensions (e.g. girths and diameters) requires different equipment. A new alternative approach for the assessment of body size and shape comprises three dimensional photonic scanning (3D-PS) [6,7]. This technique projects lasers or light stripes on to the body surface, and cameras record the distortion of these light patterns. Computer algorithms then reconstruct the 3D body surface topography, allowing automatic landmarks to be located through customised software [8,9]. From these landmarks, a variety of girths, diameters and volumes can be derived using further algorithms [6,7,10]. 3D-PS offers a number of advantages over manual methods of shape assessment. First, raw data collection is extremely rapid, lasting only a few seconds. Second, a wide variety of digital shape outputs can be extracted through different software programs, and scans can be electronically archived for analysis with improved software in the future. Third, digital outputs can extend to 2D or 3D format, whereas manual measurements (e.g. girths) are only 1D. Fourth, composite models of shape can also be constructed, allowing the topographical location of shape change to be identified by accumulating scans over time [6,7], though the need to adjust for growth makes this more challenging in children. In recognition of these benefits, several large surveys have utilised 3D-PS to acquire 1D and 2D body shape outputs in adults, and have shown substantial shape variability in association with age, gender and ethnicity [9][10][11][12][13][14]. Since young children are unlikely to tolerate large numbers of manual measurements, 3D-PS could be especially valuable in this age group. Within any population, children vary substantially in size and shape, indicating variability in body proportions and the distribution of organ, muscle and fat tissue [15,16]. Such variability in regional shape may prove a better marker of traits such as abdominal obesity or lung function than simple size indices such as height or body mass index (BMI). Furthermore, body shape differences between ethnic groups are already established at young ages [17][18][19]. In adults, ethnic differences in body shape are an important factor contributing to variability in cardio-metabolic risk and physiological function [20][21][22]. Less is known about this scenario in children, but ethnic differences in body composition, even after standardising for BMI, have been reported [17,18]. Indeed, the limits of BMI as a proxy for body composition in children, particularly in multi-ethnic populations, are well established [15,17,21]. However, as yet, 3D-PS has received minimal application in younger age groups, and its acceptability and quality of performance have not been evaluated. Two studies have reported validation data for 3D-PS versus manual measurements in adults, showing a high level of ranking consistency between the methods, but with systematic bias, where 3D-PS provides slightly larger values than manual measurements [9,23]. Equivalent data for children are required. We therefore sought to evaluate the acceptability, technical success, precision and accuracy of 3D-PS in a multi-ethnic sample of children aged 5-11 years. This work was part of the Size and Lung function In Children (SLIC) study, which was conducted in London, UK to investigate sources of variability in children's lung function [24]. Methods The SLIC study is an epidemiological study conducted in London primary schools, aiming to explore ethnic differences in lung function in a young, multi-ethnic population of 5-11 year old children. The study was granted ethical approval by the London-Hampstead Research Ethics Committee (REC 10/H0720/53). All children with written informed parental consent were eligible to participate. Children were grouped into four ethnic groups, "White" which included European, Hispanic or Latino, and Middle Eastern children; "Black" which included African and Caribbean children; "South Asian" which included Indian, Pakistani, Sri Lankan and Bangladeshi children; and "Other" which included children of any other ethnicities such as Chinese or Filipino, as well as children of mixed ethnic ancestry (e.g. Mixed South Asian / White). Two indices on social deprivation were obtained. The Index of Multiple Deprivation 2010 (IMD2010) is a UK government-derived assessment assigned by post-code, based on a weighted amalgamation of income deprivation; employment deprivation; health deprivation and disability; education deprivation; crime deprivation; barriers to housing and services deprivation; and living environment deprivation. It is categorised in quintiles, with the first quintile being the least deprived. The Adapted Family Affluence Scale is a 0-6 point scale which is based on whether the child has their own bedroom, and how many vehicles and computers the child's family has. Differences in age and social deprivation were tested by ANOVA, with Bonferroni correction. Measurement of body size and proportions Attempts were made to measure body size and proportions using both manual measurements and 3D-PS in all children who were recruited during the first year of data collection for the SLIC study (October 2011-July 2012). Manual body size measurements included weight, height, girths of the chest, waist, knee and calf, and width and depth of the chest and waist. All these anthropometric measurements were performed using established protocols, as previously described [24]. Duplicate manual measures were obtained, and the average used in subsequent analyses. Duplicate 3D body scans were attempted using NX16 instrumentation ([TC] 2 , Cary, North Carolina), the same technology used in previous surveys of adult shape [9,10,12]. The scanner was installed in a customised trailer, which was driven to a series of schools in order to collect data during school hours. Parents were not present during the scanning procedure. The subjects wore form-fitting clothing and adopted a standardised position, with their feet located on landmarks on the scanner floor. The scanner projects strips of safe 'white light' (in the visible part of the electromagnetic spectrum) onto the body form and records the distortions induced by the body's surface topography using 6 non-moving cameras over a period of~8 seconds (Fig 1). The scanner obtains a 3D 'point cloud' of raw photonic data and then reconstructs the skin topography using computer algorithms. More detailed descriptions of this process have been published elsewhere [8,25]. The software used was Body Measurement System Version 5.3; [TC] 2 ). For our study, the software automatically extracted girths, widths and depths of the chest and waist, and girths of the knee and calf on both sides of the body. The left and right knee and calf data were averaged for use in subsequent statistical analyses. Either one or two scans were used to extract 3D outcomes, according to whether they were considered technically successful. Acceptability and feasibility Acceptability was assessed in terms of the numbers invited to undergo 3D scans, and their participation rates, while feasibility was assessed in terms of successful scan extraction rates. We calculated the frequency of those with parental consent, who refused to undergo the scan, and the frequency of technical failure in those who agreed to be scanned. We further categorised technical failure as follows: movement during the scan; interference by clothing (clothing artefact); failure of the software to extract a raw body image; and failure of the software to apply automatic landmarks and extract digital shape outputs. In order to stimulate further work to improve the technical performance of 3D-PS in children, examples of specific problems were selected for illustration. Between-group variability in scanning success rates,

and in the prevalence of specific causes of technical failure, was calculated using odds ratios obtained using logistic regression. In these models, we also took into account age, sex and the indices of social deprivation. Precision Precision was calculated from duplicate scans when available. In some instances, a third scan was obtained at the time of measurement due to concern over the quality of one of the two original scans. The mean of two or three scans was used when comparing methods. However, in calculating precision we discarded data from children with 3 scans, as this approach inflates precision. Using only duplicate scans, we calculated precision using the following equation, given by Bland and Altman [26]: Precision ¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi sum of squared differences between scans=n p Agreement between methods Agreement was assessed using the method of Bland and Altman [26], which calculates the mean bias between techniques along with its limits of agreement, which are obtained as twice the standard deviation of the bias. This bias was tested for significant difference from zero by paired t-test. We also used correlation analysis to establish whether the bias varied in relation to the mean value obtained using both methods, i.e. whether bias was associated with the magnitude of the outcome [26]. We further regressed 3D outputs on manual outputs, in order to derive the intercept and slope. Where the intercept differed from zero, outputs from one technique differed systematically in magnitude from the other. Where the slope differed significantly from 1, the two techniques showed inconsistent agreement across the range of size. All analyses were conducted in Excel or SPSS. A cut-off of p = 0.05 was used for statistical significance. Results Parental consent for their child to undergo 3D-PS was obtained for 94.8% (1496/1578) of those approached. Amongst these 1496 children, a scan was not attempted in 12 (0.8%), as the equipment was not ready, or the child was not available at the appropriate time. These 12 children were not included when calculating acceptability or feasibility. Of the remaining 1484 children, 38 (2.6%) refused, giving an acceptability rate of 97.4% (Table 1). Those refusing were younger (Δ = 1.0 (SE 0.3) y; p<0.0001), shorter (Δ = 6.8 (SE 1.9) cm; p<0.0001) and had lower BMI (Δ = 1.4 (SE 0.5) kg/m 2 ; p<0.0001) than those agreeing to be scanned ( Table 2), but the differences in body size were no longer significant after adjustment for age. There was no ethnic difference in the likelihood of refusal. The age-associated reduction in the percentage of children refusing is illustrated in Fig 2. There was no difference in age between the ethnic groups by ANOVA. Scores for the Index of Multiple Deprivation were significantly greater (p<0.0001) in the three 'non-White' groups relative to the 'White' children (3.27 SD 1.36). The 'Black' children (4.46, SD 0.85) further had scores significantly greater (p<0.0001) than the 'South Asian' (3.83, SD 0.99) and 'Other' (3.78, SD 1.32) groups. For the Family Affluence Scale, the 'White' (3.70, SD 1.42) and 'Other' (3.72, SD 1.40) children had significantly higher (p<0.0001) scores than the 'South Asian' children, 3.15, SD 1.33), and all three of these groups had significantly higher (p<0.005) scores than the 'Black' children (2.88, SD 1.25). On average, the duration for performing detailed manual anthropometry was~10 minutes per child. Although each 3D-PS measurement can be captured in 8 seconds, the preparation and processing time for this assessment, which includes undressing and dressing, encouraging the child to stand still and having to repeat the scan (if child willing) to achieve at least 2 acceptable scans also took an average of 10 minutes per child. Of the 1446 children who underwent a scan, the process was unsuccessful in 29.3%, due to problems with clothing, body movement, or failure to extract a viable body model. Thus, the technique proved technically feasible in 70.7%. Adjusting for age, the odds of technical failure were higher in 'Black' children (1.84; 95%CI 1.34, 2.51; p<0.0001) relative to 'White' children, however the other two groups showed no difference. The two measures of social deprivation were not significant in this model. Aside from ethnic variability, the risk of scan failure was associated with the child's age and body size ( Table 2), but not with sex. The mean age of those scanned successfully was 0.7 years (p<0.0001) greater than those whose scans were not successful. Adjusting for age, the mean weight and height of those not scanned successfully were 1.7 kg and 3.3 cm lower respectively (p<0.0001) than those scanned successfully, however there was no significant difference in BMI. The age-associated reduction in the percentage of children with unsuccessful scans is illustrated in Fig 2. Categorisation of the reasons for scan failure is given in Table 3. Among reasons for technical failure, the prevalence of excessive body movement was relatively low (2.8% of all those scanned, and 9.4% of all failed scans). Approximately equal proportions (8 to 9%) of scans overall failed because of clothing artefact, failure of the software to extract a body image, and failure of the software to locate landmarks for shape outputs. The challenge of extracting a body image was significantly more common amongst the 'Black' and 'South Asian' children, while the likelihood of clothing artefact was greater in the 'White' children. A small proportion of scans failed when the trailer was sensitive to vibrations from nearby busy roads. Examples of technical failures are illustrated, including inability to extract chest data due to atypical posture (Fig 3), failure to locate leg girths and body movement of the arms (Fig 4), and artefacts arising from inappropriate clothing and vibrations affecting the scanning truck ( Fig 5). Age-associated reductions in the percentages of children refusing to be scanned, or producing scans that were not technically successful. The rate of refusal fell steadily from 6% at 5 years to zero at 11 years, whereas the percentage of failed scans decreased with age, but remained~25% even in the older age groups. Precision of duplicate measurements by 3D-PS and by manual measurement is given in Table 4. Duplicate successful 3D scans were available for 589 children, 57.6% of those in whom 3D-PS was technically successful overall. The paired measurements by 3D-PS are illustrated for each outcome in supporting information file S1 Fig. The magnitude of precision in the manual measurements was better by at least a factor of 10 than that from 3D-PS. For example, 3D-PS precision of waist and chest girth was~1.5 cm, whereas manual precision was~0.1 cm. Height was not compared between methods as it is not extracted by 3D-PS. The scatter of 3D-PS values against manual values is shown for each outcome in Fig 6, while Table 5 shows correlations between 3D-PS and manual measurements in the whole sample, and by ethnicity. The majority of correlations were >0.90, and all were 0.84. This indicates relatively high consistency in ranking, with variability in 3D-PS outcomes explaining~75% to~85% of variability in torso depths and widths, and~92% to~97% of the variability in body girths. Table 6 shows Bland-Altman statistics for the assessment of agreement between techniques. There was a significant bias between 3D-PS and manual measurements for all outcomes, in all cases the 3D-PS outcomes being larger. These biases were furthermore significantly associated with the mean of the measurement by the two methods, indicating that bias varied with outcome size. There were negligible ethnic differences in these biases and their correlations with size: chest girth bias was significantly smaller in the 'White' than the 'South Asian' or 'Other' group; and the 'White' group had smaller bias in chest width, waist depth and knee and calf than the 'Black' group. In all cases, these ethnic differences were of small magnitude. Supporting information file S1 Table provides additional information on the level of agreement between techniques for the whole sample, by providing intercepts and slopes for the regression of 3D-PS measurements on manual measurements. These clarify that 3D-PS outcomes were systematically greater (intercept significantly >1) for all outcomes except girths of the chest, knee and calf. However, slopes differed significantly from 1 only for chest girth and depth and knee girth, indicating that the bias varied over the magnitude of the outcome. Discussion The speed of data collection per subject of 3D-PS potentially makes it a valuable resource for collecting multiple anthropometric outcomes in very large surveys, but these advantages are offset by the need for expensive equipment, and currently by lack of evidence regarding its viability and accuracy. In adults, 3D-PS measurements has shown high ranking consistency with manual anthropometry, although 3D-PS provides slightly but systematically larger values [9,23]. Acceptability of the technique has not been formally assessed, but large samples took part in national sizing surveys in several countries [9,10,12]. Our study represents the first effort to assess the acceptability, feasibility, precision and accuracy of 3D-PS in children, using a large multi-ethnic sample aged 5-11 years. This information will help develop this method for future application in pediatric research and clinical practice. Acceptability The acceptability proved high across the entire sample, with only 2.6% of those invited to participate declining. The rate of acceptance might have been even higher had parents of younger children been present. Those who declined, or refused, tended to be younger than those who accepted and were scanned successfully. Nevertheless, the vast majority of all age groups underwent a scan, and the main challenge that we encountered in this study was not with children's acceptance but with technical aspects of 3D-PS, with 29.3% failing to produce a viable scan. Table 5 and supporting information file S1 Table. doi:10.1371/journal.pone.0124193.g006 Technical problems Several generic reasons for technical failure were documented, some pertaining to the practical challenges of scanning young children, and others relating to technical limitations of the current hardware and software. Even after taking age into account, those in whom the scan was not successful tended to be slightly shorter and lighter, indicating that the software is marginally less successful in extracting data from smaller children. Nevertheless, the majority of scans at all ages were successful, as indicated in Fig 2. Problems with clothing affected 8.7% of the whole sample, and were more common amongst the 'White' children. This problem could be resolved by offering standardised formfitting clothing, although a minority of children may still be unwilling to comply due to sensitivity concerning body shape, or cultural variability regarding modesty. Some scans failed due to extreme clothing colour, an issue that again could be resolved by making available appropriate clothing, as was done in the adult surveys. Body movements must be anticipated in young children. Adults undergoing 3D-PS have the advantage of a stabilising handhold to aid maintenance of an ideal posture during assessments. However in this study, attempts to adapt the handhold for children were unsuccessful and will need manufacturer's input to customise the handhold for children to minimise body movements. The effect of such movements on scan quality could potentially also be reduced, by shortening the duration of the scan. This could be achieved in part through physical alterations to the scanner: since children are shorter than adults, the cameras do not need to descend from as great a height above the floor. Other options are also available, such as speeding up the capture of photonic data. Thus, technical development could potentially resolve this problem. Failure of the software to extract body shape or body measurements can be attributed to insufficient tailoring of the computer algorithms to the range of variability in children's shape. The software we used was originally developed for application with adults, and therefore lacks 'training' on younger age groups. The likelihood of failure to extract outcomes varied by ethnicity, suggesting that such training must be done on multi-ethnic datasets. There was no indication that ethnic differences in technical failure were associated with age or indices of social deprivation, suggesting that ethnic variability in posture or physique was the primary challenge. However, there were also minor differences between ethnic groups in the likelihood of body movement. Overall, this source of error could likewise be minimised through further development of 3D-PS software algorithms. When the trailer housing the scanner was parked near busy roads, vibrations could interfere with data capture. This problem could potentially be resolved by increasing the stability of the mobile unit. From the opposite perspective,

the mobile unit proved very valuable for introducing the technology in standardised format into a number of schools. Overall, the relatively high failure rate of 3D-PS in young children compared to adults can be attributed to issues that should be possible to resolve through technological development. For example, if we had resolved all technical problems with scan extraction, and encountered only problems of body movement and clothing artifact, the success rate would have been 87%. The feasibility problems we encountered relate primarily to a specific version of the technology we used, rather than universal limitations of 3D-PS. The time required to complete 3D-PS was similar to that required to collect equivalent manual data. If greater numbers of shape outcomes were required, 3D-PS would quickly become more time-effective, whilst technical changes could also speed up data collection. These issues are often very important in large surveys, and may have staff cost implications. Height and sitting height are not extracted by 3D-PS, and must be obtained manually, requiring additional time and equipment. However, height is routinely measured in clinical practice, hence manual measurement of this outcome is relatively undemanding. On the other hand, the initial cost of equipment is substantially greater for 3D-PS than manual approaches, hence any cost-benefits of 3D-PS are likely to restricted to large surveys. Precision and accuracy Precision of 3D-PS was poorer by a factor of 10 than that of manual measurements, which is likely to arise for several different reasons. Manual measurements are conducted with much greater control over the subject, as the investigator can ask the child to adopt and hold a given posture, potentially increasing the likelihood of making identifying the landmark correctly. However, this may also artificially inflate precision of manual anthropometry as it is human nature to return to the same landmarks when collecting duplicate data. 3D-PS must identify body landmarks automatically de novo each scan, and subtle changes in posture between repeated scans may confound this process. Despite this difference between techniques, 3D-PS achieves a level of precision that is adequate for large surveys, in which the aim is to rank individuals in terms of population variability in size and shape outcomes. Using manual measurements as the gold standard reference method, the accuracy of 3D-PS showed a relatively consistent bias to larger measurements, although the magnitude of this difference was always modest. This is consistent with previous validation studies, and can be attributed to the fact that manual measurements involve a slight tension, as the tape is held over the skin [23]. For example, disagreement between manual and 3D measurements decreased when a mannequin was measured instead of a human body [23]. The limits of agreement were of intermediate magnitude, as visually evident in Fig 6. While the correlations indicate relatively good ranking consistency between techniques, the level of agreement within individual children was wider than would be acceptable for studies where, for example, the aim was to identify children above or below specific cut-offs. These biases and limits of agreement are of sufficient magnitude for us to conclude that the techniques cannot be regarded as interchangeable. As yet, it is unclear what proportion of the bias can be attributed to different body locations being measured, as opposed to inaccuracy of the software algorithms, and further work is needed to address this. In any comparison using the Bland-Altman method, it must be remembered that both techniques contribute to the lack of agreement, however in this case the main source of imprecision was clearly 3D-PS. Overall, we suggest that 3D-PS shows great promise as a new digital tool for anthropometric surveys, but we have outlined a number of issues that will require attention in order to improve the quality of the data. Abdominal Aortic Aneurysms Targeted by Functionalized Polysaccharide Microparticles: a new Tool for SPECT Imaging Aneurysm diagnostic is nowadays limited by the lack of technology that enables early detection and rupture risk prediction. New non invasive tools for molecular imaging are still required. In the present study, we present an innovative SPECT diagnostic tool for abdominal aortic aneurysm (AAA) produced from injectable polysaccharide microparticles radiolabeled with technetium 99m (99mTc) and functionalized with fucoidan, a sulfated polysaccharide with the ability to target P-Selectin. P-Selectin is a cell adhesion molecule expressed on activated endothelial cells and platelets which can be found in the thrombus of aneurysms, as well as in other vascular pathologies. Microparticles with a maximum hydrodynamic diameter of 4 µm were obtained by crosslinking the polysaccharides dextran and pullulan. They were functionalized with fucoidan. In vitro interactions with human activated platelets were assessed by flow cytometry that demonstrated a specific affinity of fucoidan functionalized microparticles for P-Selectin expressed by activated platelets. For in vivo AAA imaging, microparticles were radiolabeled with 99mTc and intravenously injected into healthy and AAA rats obtained by elastase perfusion through the aorta wall. Animals were scanned by SPECT imaging. A strong contrast enhancement located in the abdominal aorta of AAA rats was obtained, while no signal was obtained in healthy rats or in AAA rats after injection of non-functionalized control microparticles. Histological studies revealed that functionalized radiolabeled polysaccharide microparticles were localized in the AAA wall, in the same location where P-Selectin was expressed. These microparticles therefore constitute a promising SPECT imaging tool for AAA and potentially for other vascular diseases characterized by P-Selectin expression. Future work will focus on validating the efficiency of the microparticles to diagnose these other pathologies and the different stages of AAA. Incorporation of a therapeutic molecule is also considered. Introduction Abdominal aortic aneurysm (AAA) is a pathological dilatation of the abdominal aorta with a mortality associated to its rupture of approximately 90% [1,2]. Nowadays, AAA is usually diagnosed by anatomical imaging techniques such as ultrasound, computed tomography and/or magnetic resonance Ivyspring International Publisher imaging [3]. Limitations of these methods are that i) they only give anatomical and morphological information and ii) the arterial wall dilatation must be relatively advanced to clearly identify the pathology. Furthermore, it has been reported that the size measurement of the AAA is not sufficient for predicting its rupture [4]. Consequently, new non-invasive techniques enabling early identification and evaluation of the AAAs rupture risk are needed [5]. A recent strategy to fulfill these requirements is to produce injectable diagnostic tools that are able to target key molecules involved in early arterial process [3]. For this purpose, several biological markers of AAA development have been identified as potential targets of the pathology [6,7]. Many molecular imaging probes of AAA are thus developed from various contrast agents targeted toward proteins such as elastin, collagen, matrix metalloproteinase or adhesion molecules like VCAM-1, ICAM and selectins [8][9][10][11]. In this work, we focused on adhesion molecules as they are expressed earlier in the inflammatory process than the other protein markers [12]. Among those, we selected P-Selectin which has the advantage to be present on platelets and endothelial cells on activation. The role of this key targeted molecule in AAA is not yet perfectly well established but it has been clearly identified to be involved in inflammatory cell recruitment in arterial diseases though the interaction with its counterreceptor P-Selectin Glycoprotein Ligand 1 [13,14]. For this reason, it is related with the renewal and growth of biologically active arterial thrombus and with the expansion of AAA [15,16]. Revealing the expression of P-Selectin with a molecular imaging tool is therefore a promising clinically relevant strategy for AAA early detection, growth prevision and rupture risk assessment. Besides, several spherical injectable polymeric particles have been developed to target P-Selectin with promising binding and/or diagnostic properties. PLGA nanoparticles conjugated with the external fraction of glycoprotein Ibα showed great affinity for P-Selectin coated surfaces and for activated endothelial cell layer [17]. Van Kasteren S. et al have imaged brain diseases with carbohydrate nanoparticles functionalized with natural complex glycan ligand of selectins [18]. McAteer et al. have developed microparticles of iron oxide conjugated with P-Selectin and VCAM-1 antibodies that revealed in vivo endothelial activation on MRI scans [19]. Molecular and functional imaging of AAA employs a wide variety of imaging modalities. A large amount of magnetic resonance imaging (MRI) dedicated functionalized contrast agent including ultrasmall superparamagnetic particle of iron oxide (USPIO) or gadolinium showed feasibility of AAA prone to rupture site identification [20]. All the other modalities are considered including ultrasound, computed tomography, optical imaging but the most promising seems to be nuclear imaging methods as they provides a highly sensitive detection of the injected radioactive imaging agent. 18F-fluorodeoxy-glucose (18F-FDG) which reveals the metabolic cells on positron emission tomography (PET) enabled identification of focal inflammatory sites in AAA which may be correlated with AAA progression and rupture risk [21,22]. However, this method as mean to predict AAA evolution is disputed [23]. Moreover, even if PET provides a better detection sensitivity than SPECT, 99m Tc is the most widely used radioisotope in nuclear medicine because of its physical characteristics: optimal gamma energy for SPECT imaging (140 keV) and short physical half-life (T = 6.01 h), allowing a low radiation burden to patients [24]. Several diagnostic tools of the AAA have thus been developed using 99m Tc detection by SPECT. In 1976, Ryo et al. published the use of 99m Tc labeled red blood cells for AAA detection [25]. Iwasaki et al. have developed in 2001 a diagnostic method for non-invasive imaging of aortic dissection using 99m Tc-labeled murine anti-smooth muscle myosin monoclonal antibody in rats [26]. In 2006, Sarda-Mantel et al. imaged luminal thrombi in murine AAA with radiolabeled annexin V that specifically bind to phosphatidylserine exposed to the surface of apoptic cells and activated platelets [27]. In our study, a particular interest was given to dextran and pullulan which are widely used in clinical applications [28]. An implantable biodegradable hydrogel for tissue engineering applications was obtained by crosslinking these polysaccharides [29,30]. Since this hydrogel was not injectable, we herein developed a novel water-in-oil emulsification process combined to the crosslinking of dextran and pullulan in order to obtain injectable particles. Fucoidan, a sulfated polysaccharide derived from seaweed that happens to be an occurring mimic of sialyl Lewis X, the natural ligand of P-Selectin [31], was used to functionalize the particles. Our group previously demonstrated its ability to bind P-Selectin and developed a radiotracer by combining 99m Tc to fucoidan [32,33]. The aim of this study was to produce an efficient AAA diagnostic tool from an injectable polymeric device able to be combined with a contrast agent and to target the P-Selectin which is expressed in acute AAA. For this purpose, we have developed a microparticle system which is functionalized with fucoidan and radiolabeled with 99m Tc. We demonstrate a strong SPECT contrast enhancement located in the ab-dominal aorta, revealing the presence of P-Selectin inside the aneurysm of an elastase induced AAA rat model. These microparticles could have clinical uses as a SPECT diagnostic tool for early detection and progression risk assessment of AAA and potentially other arterial diseases characterized by the expression of P-Selectin. Microparticles preparation Polysaccharide microparticles (MP) were obtained from a previously described crosslinking protocol [34] coupled to a water-in-oil emulsification process. The suspension was centrifuged (BR4i, JOUAN S.A., Saint Herblain, France) for 10 minutes at 3,000 g, then the supernatant was centrifuged for 10 minutes at 5,000 g. The resulting pellet was suspended at 10 mg/mL in saline buffer and stored at 4°C until use. Microparticles characterization The surface morphology of MP and MP-Fucoidan particles was imaged using scanning electron microscopy (SEM) (Philips XL 30 ESEM-FEG, Hilsboro, OR, USA) on dried samples coated with a thin gold layer. Mean diameter, size distribution and zeta potential were analyzed by dynamic light scattering method (NanoZS, Malvern Instruments S.A., Orsay, France). Surface sulfur presence was evidenced by energy dispersive X-ray spectroscopy (EDX) (Philips XL 30 ESEM-FEG, Hilsboro, OR, USA) and global sulfur content was quantified by UV fluo-rescence spectroscopy (THERMO TN-TS 3000, Thermo Fisher Scientific, Pittsburgh, PA, USA) on freeze-dried samples of MP, MP-Fucoidan and plain fucoidan. Fucoidan content in MP-Fucoidan was calculated from sulfur content in MP-Fucoidan and in fucoidan. Hemolytic Toxicity Assay Blood from healthy volunteers was collected in sodium citrate 3.8 % (w/v). Erythrocytes were separated from blood plasma by centrifugation (800 g, 5 min) and resuspended at 20 % (v/v) into distilled water, which was considered as producing 100 % hemolysis, and into normal saline producing no

hemolysis, considered as a blank. To reproduce the in vivo parameters, the same microparticle suspensions were assessed in a corresponding amount of blood (rats of 400 g average weight was considered to have a total blood volume of 24 mL). 5 µL of MP and 5 µL of MP-Fucoidan (5 mg/mL) were mixed with 500 µL of erythrocytes suspensions diluted in normal saline. All preparations were incubated for 1h at 37°C and centrifuged (3000 g, 5 min). Supernatants were taken and absorbance was measured at 540 nm. The percentage of hemolysis was determined for red blood cell samples incubated with MP, MP-Fucoidan and saline by comparing to water as 100% hemolytic sample. Results were expressed as mean values ± SEM (n=3). Microparticles radiolabeling Technetium-99m ( 99m Tc) labeling required the reduction of pertechnetate by a reducing agent. Labeling of MP or MP-Fucoidan was carried out by mixing 0.030 mL of a 5 mM stannous chloride solution (reducing agent), 0.5 mL of microparticle suspension (10 mg/mL), 0.2 mL of Na + , 99m Tc0 4corresponding to an activity of 370 MBq. After incubation for 1 h at room temperature, radiolabeled microparticles were separated from Na + , 99m TcO 4excess by centrifugation (5,000 g). In order to determine radiolabeling efficiency, microparticle pellet and supernatant activities were measured in an activimeter (Medi 40, Medisystem, Guyancourt, France) Labeling efficiency is expressed as percentage of the ratio between radioactivity associated with the microparticles and total radioactivity. To assess the labeling stability, the radiolabeled microparticles ( 99m Tc-MP or 99m Tc-MP-Fucoidan ) were resuspended in 1 mL of 0.9 % NaCl or rat plasma, and incubated at room temperature for 3 hours. Every 60 minutes, microparticles suspension were centrifuged and radioactivity associated with particles and in the supernatant was measured (n=3). Stability was expressed as a percentage of the initial labeling. For in vivo experiments, radiolabeled micropar-ticles were resuspended in saline (5 mg/mL) and 200 µL, corresponding to an activity of about 37 MBq, were administrated intravenously to rats. In Vitro Binding Assays Affinity of soluble fucoidan for P-Selectin was assessed with a BIAcore X100 (GE Healthcare, Freïburg Germany). A CM5 sensorchip was coupled with recombinant human P-Selectin and fucoidan or dextran solutions was successively injected at 0 M, 100 nM, 300 nM, 1 µM and 3 µM at a flow rate of 30 µL/min. The response in resonance units (RU) was recorded as a function of time. The apparent binding affinities of fucoidan for P-Selectin were determined by analysis of the kinetic of the association assuming a 1:1 Langmuir model using BIAcore evaluation software, following a previously described protocol [33]. Affinity of fucoidan functionalized microparticles for P-Selectin expressed on the surface of activated human platelets was assessed by flow cytometry. Five mL of blood from healthy adult volunteers was collected in sodium citrate 3.8 % (w/v). Platelet-rich plasma (PRP) was obtained by centrifugation at 120 g for 15 min and platelet concentration was adjusted at 2.10 8 /mL with autologous platelet-poor plasma (PPP). Activated PRP was obtained by stimulation of PRP with 20 µM of TRAP (thrombin receptor-activating-peptide). P-Selectin expression at the platelet surface was assessed using an anti-human CD62P-FITC (0.11 mg/mL, Ancell, Bayport, MN, USA) and its isotype-matched control. In some experiments, a non-labeled anti human CD62P (1 mg/mL, Ancell) was used to block P-Selectin in activated PRP. To evaluate the binding of microparticles to platelet P-Selectin, 5 µL of non-activated PRP, activated PRP or anti P-Selectin-treated activated PRP were incubated for 20 minutes with 5 µL of fluorescent (FITC) MP or MP-Fucoidan together with 5 µL of PE-Cy5 Mouse Anti-Human CD41a (BD Biosciences Pharmingen, Le Pont De Claix, France) to label platelets. In addition each PRP sample was incubated with an isotype-matched control antibody. Samples were analyzed on a LSRII flow cytometer (BD Bioscience Pharmingen), with 10,000 events collected per sample with area of double positivity reflecting the affinity of microparticles (FITC) for platelets (PE-Cy5). The data were processed with FACS DiVa software and the mean FITC fluorescence intensity (MFI) was measured in the area of double positivity. Results were presented as a ratio of MFI to the control MFI (MP incubated with PRP). In vivo arterial disease model The ability of the radiolabeled functionalized microparticles to target P-Selectin expression in vivo was assessed in an abdominal aortic aneurysm (AAA) experimental model in rats. All experimental procedures involving the use of rats were approved by the Animal Care and Use Committee of the Claude Bernard Institute (Paris, France). The elastase model was performed on 8 male adult Wistar rats (7 weeks, CEJ) [35]. Animals were anesthetized with pentobarbital (1 µL/g body weight). Porcine pancreatic elastase (2.7 mg/mL, Sigma Aldrich) was perfused into the lumen of an isolated segment of the infrarenal abdominal aorta for 15 minutes at a rate of 2.5 mL/h. Two weeks after the surgery, when the animals present biologically active abdominal aorta aneurysm, characterized by the presence of intraluminal thrombus [36] and the expression of P-Selectin [27,32], rats were injected with the microparticles and imaged. Injections of 200 µL of 99m Tc-MP or 99m Tc-MP-Fucoidan (5 mg/mL) were performed slowly, to avoid aggregation, into the penis vein. Four healthy rats served as control experiments. Immediately, after injection of 99m Tc-MP or 99m Tc-MP-Fucoidan, CT acquisition focused on the abdomen was started and 15 minutes after injection, SPECT imaging was performed in the same abdomen range. The SPECT acquisition was performed with the following parameters: helical scan with 28 projections per rotation plus circular scan at the beginning and at the end of the scan range, matrix size=256x256, zoom 1.14 (pixel size: 1 mm 2 ), correction for energy, linearity and uniformity CT data were reconstructed using filtered back projection algorithm with Ram-Lak filter in plane (voxel size 147 x 147 µm 2 ) and slice thickness equal to 147 µm 2 . SPECT data were reconstructed using HiSPECT iterative reconstruction software on PC workstation. Images were visualized using the Bioscan InVivoScope software with co-registration of SPECT and CT images. Reconstructed slices were visually assessed in 3 planes from same stereotaxic slices (sagittal slice, coronal slice 1 centered on abdominal aorta area and coronal slice 2 centered on AAA area) with and without CT coregistration to determine the presence of a focal uptake in the abdominal aorta according to the model. Quantification was performed on DICOM images with a DICOM processing software (OsiriX Imaging Software, Osirix, France) by calculating the ratio between the activity (mean counts) in the AAA area and in a normal region (background) on short-axis slices. The background activity was derived from a region of interest drawn over the renal aorta avoiding renal activity. Autoradiography, histology and immunohistochemistry Immediately after achievement of SPECT (60 minutes after injection) animals were sacrificed with pentobarbital overdose. Abdominal aorta aneurysms of AAA rats and healthy aorta of healthy control rats were removed, washed in 0.9 % saline and weighed. Radioactivity of AAA and healthy aorta was determined by gamma counting (COBRA II -Auto Gamma, Packard, Prospect, CT, USA) and the percentage of injected activity per gram was calculated. Then, aorta samples from AAA rats and control rats were frozen and cut into 20 µm thick frozen sections for autoradiography and 10 µm thick frozen sections for histology and immunochemistry studies. Autoradiographic images were obtained after 12 hours exposition of fifteen slides (corresponding to about 30 sections of AAA and 15 sections of healthy aorta) using a ß-imager (Beta Imager TM , Biospace Lab, Paris, France). Signal intensity values of regions of interest were assessed using a quantification sofware (M3 Vision, Biospace Lab, Paris, France). The ratio between activity (mean counts/mm 2 corrected for background) of the aneurysm sections and activity of the healthy abdominal aorta sections was calculated and compared statistically. These same sections were stained both with Masson trichrome to visualize cells, nuclei, and fibrin and with Alcian blue counter stained with nuclear red to reveal polysaccharide microparticles and cell nuclei. Immunohistochemistry studies were performed on 10 µm thick sections using goat mouse anti P-Selectin (4 µg/mL, Cruz Biotechnology Inc., Heidelberg, Germany) as a primary antibody and IgG rabbit anti goat as a secondary antibody (1 µg/mL, Dako, Les Ulis, France) revealed by DAB enhancer (Dako) and counterstained with hematoxylin. Control sections were obtained by omitting the primary antibody. Observations were performed using light and fluorescent microscopy (Nanozoomer, Hamamatsu Photonics France SARL, Massy, France). Micro-autoradiography Track of internal conversion electrons emitted during 99m Tc deexcitation was revealed by microautoradiography method [37]. Briefly, 10 µm frozen sections of AAA and control rats were deposited on 1 µm thick gelatin-coated glass slides. Five hundred µl of nuclear emulsion K5 (Ilford photo Harman technology Ltd, Ilford, United Kingdom) were poured onto section slides to obtain a 25 µm thick coating. After 24 hours of exposition at 4°C, slides were developed for 20 minutes at 15 ± 1°C by Brussels formula developer (1.8 % (w/v) sodium sulphite, 0.08 % (w/v) potassium bromide, 0.45 % (w/v) amidol and 3.5 % boric acid) [38], rinsed in stop bath, fixed for 40 minutes, rinsed in tap water and finally dryed dust free, overnight at room temperature. Then, microautoradiographied slides were stained with Alcian blue in order to compare localizations of electron tracks and polysaccharide microparticles on the same sections. Observation of electrons tracks and microparticles was performed using light microscopy (Nanozoomer, Hamamatsu Photonics France SARL, Massy, France). Statistical analysis Flow cytometry results were analyzed statically with a one-way ANOVA with Bonferroni post tests to compare data obtained with MP-Fucoidan and a two-way ANOVA with Bonferroni post tests to compare data obtained with MP-Fucoidan and MP, performed with Graphpad Prism (GraphPad Software, San Diego, CA, USA). Other results were analyzed with unpaired Student's t-tests to compare 2 groups and one-way ANOVA with Bonferroni post test to compare 3 groups. A difference of p<0.05 was considered significant. Characterization of polysaccharide microparticles Polysaccharide microparticles were prepared by a chemical crosslinking process with STMP combined with an emulsion step. Functionalized microparticles (MP-Fucoidan) were obtained by adding the sulfated polysaccharide fucoidan. MP (Fig 1a) and MP-Fucoidan (Fig 1b) both exhibited a spherical morphology and a smooth surface as shown on scanning electronic microscopy images. Dynamic light scattering measurement revealed that all microparticles had a hydrodynamic diameter smaller than 4 µm and a mean hydrodynamic diameter of 503 ± 110 nm and 358 ± 60 nm for MP and MP-Fucoidan, respectively ( Table 1). Zeta potential was measured and microparticles prepared with fucoidan had a significantly higher electronegativity than control microparticles (-16.2 ± 0.8 mV versus -9.1 ± 0.6 mV, respectively, p<0.05) ( Table 1). Global sulfur content was quantified by UV fluorescence spectroscopy, using MP and plain fucoidan (in powder) as negative and positive controls, respectively. MP-Fucoidan contained 1100 ± 400 ppm of sulfur. Since plain fucoidan contained 6.92 % (w/w) of sulfur, we calculated the fucoidan content in MP-Fucoidan to be 1.64 ± 0.57 % (w/w). Energy-dispersive X-ray spectroscopy analysis showed the absence of sulfur at the surface of control MP and the presence of sulfur at the surface of the MP-Fucoidan ( Table 1). The possible toxicity induced by microparticles on red blood cells was assessed in vitro in the same conditions than in vivo administration. To reproduce these conditions, micro particle suspensions (5 µL at 5 mg/mL) were mixed with 500 µL of erythrocytes. Hemolytic toxicity assay revealed that the MP and MP-Fucoidan suspensions lead to similarly low red blood cell lysis than normal saline (2.00 ± 0.31 % for MP and 2.33 ± 0.28 % for MP versus 2.34 ± 0.22 % for normal saline, n=3, no significant differences). In vitro affinity for P-Selectin Binding of free fucoidan to P-Selectin was analyzed by Surface Plasmon Resonance. 57 kD fucoidan and 40 kD dextran, used as a non-sulfated polysaccharide control, were flowed on a sensorchip coated with recombinant human P-Selectin. The interaction sensorgrams revealed that fucoidan binds to P-Selectin but dextran does not (Supplementary Material: Fig. S1). Association (Ka), dissociation (Kd) and affinity (KD) constants of fucoidan for P-Selectin, calculated using a 1:1 Langmuir binding model, were 2.5 10 3 M, 3.2 10 -5 M and 1.2 10 -8 M, respectively. Interaction of non-functionalized microparticles with non-activated platelets (PRP), activated platelets (PRP + TRAP) and P-Selectin blocked platelets (PRP + TRAP + CD62P) was assessed and compared

to the interaction of fucoidan functionalized microparticles (Fig 1c). First, a weak binding of non-functionalized microparticles for platelets was observed, whether they were incubated with activated, unactivated, or activated then blocked platelets, as shown by low values of mean fluorescence intensity (MFI). A weak binding was also noticed for fucoidan functionalized microparticles when incubated with unactivated platelets or with activated platelets blocked with anti P-Selectin, with ratios of MFI to the control MFI (MP incubated with PRP) of 1.0 ± 0.2 and 1.5 ± 0.2 respectively. In contrast, MP-Fucoidan exhibited a significantly higher MFI ratio with activated platelets, as compared to unactivated platelets or activated then blocked platelets (p<0.01 in both cases). Interaction with platelets activated with TRAP was 8 times higher with fucoidan functionalized microparticles, as compared to non-functionalized microparticles (MFI ratio of 8.5 ± 2.3 versus 1.1 ± 0.1, respectively, p<0.001). Radiolabeling of microparticles The 99m Tc radiolabeling yield was 90.2 ± 2.4 % on MP-Fucoidan and 91.7 ± 1.8 % on MP. It was measured after each radiolabeling and mean values were not significantly different (n=6). Concerning the labeling stability (Supplementary Material: Fig. S2), the measurements revealed that radioactivity was still associated to microparticles after three hours of incubation into saline (89 % of initial radioactivity for 99m Tc-MP and 92 % for 99m Tc-MP-Fucoidan). In plasma of rat, more than 75 % of radioactivity was still associated after 30 minutes of incubation and more than 50 % after 1 hour for both 99m Tc-MP and 99m Tc-MP-Fucoidan. In vivo SPECT imaging We evaluated the ability of the radiolabeled functionalized microparticles to target P-Selectin expression in vivo in an abdominal aortic aneurysm (AAA) experimental model in rats with SPECT-CT imaging (Fig 2a). While no enhancement was ob-served by SPECT 30 minutes after administration of 99m Tc-MP to AAA rats or of 99m Tc-MP-Fucoidan to healthy rats, administration of 99m Tc-MP-Fucoidan to AAA rats showed a strong contrast enhancement in the area of the abdominal aorta (Fig 2a, arrow). Uptake ratios were calculated from SPECT images by the ratio of the mean intensity value of the AAA region of interest (ROI) compared to the mean intensity value of the renal aorta ROI (Fig 2b). There was no uptake after administration of 99m Tc-MP-Fucoidan to healthy rats (ratio 1.03 ± 0.06) or administration of 99m Tc-MP to AAA rats (ratio 1.22 ± 0.10), while a significant increase of the ratio uptake with administration of 99m Tc-MP-Fucoidan to AAA (ratio 3.49 ± 0.35, p<0.01 in both cases) was obtained. Quantification of radioactivity was performed by gamma counting and results were expressed as a percentage of the injected activity per gram of tissue (Fig 2c). These results confirmed a significantly higher radioactivity accumulation after injection of 99m Tc-MP-Fucoidan into AAA (0.12 ± 0.01 %) as compared to 99m Tc-MP injected to AAA (0.04 ± 0.01 %, p<0.001) and 99m Tc-MP-Fucoidan injected to healthy rats (0.03 ± 0.02 %, p<0.01) Localization of 99m Tc-MP-Fucoidan within the AAA wall We performed cryosections of both AAA and healthy aorta samples to analyze the fate of the functionalized microparticles. Autoradiography analysis (Supplementary Material: Fig. S3) confirmed the presence of a radioactive signal 7 times stronger in the AAA wall as compared to healthy aorta wall 60 minutes after injection of 99m Tc-MP-Fucoidan (0.36 ± 0.14 cpm/mm 2 versus 0.05 ± 0.02 cpm/mm 2 , respectively, p<0.05, n=3 independent experiments). Histology staining was then performed on AAA cryosections. Masson trichrome staining (Fig 3a) confirmed the presence of a thrombus and a media layer of the degraded arterial wall. In inflamed areas where cells are activated and express P-Selectin (Fig 3b left image and negative control without primary antibody on the right), the presence of MP-Fucoidan was clearly identified in the arterial wall using Alcian blue staining (Fig 3c). This localization was confirmed by fluorescent microscopy, where green microparticles (prepared with FITC-dextran) were observed in the arterial wall, between the thrombus and the media layer (Fig 3d). Finally, micro-autoradiography coupled to Alcian blue and nuclear red staining (Fig 3e) showed electron tracks printed into the nuclear emulsion on the same blue polysaccharide spots, confirming colo-calization of MP-Fucoidan with electron emission, i.e. radioactivity. Discussion In this study, we described polysaccharides microparticles functionalized with fucoidan and 99m Tc radiolabeled. Our group previously developed tools for diagnostic of the abdominal aortic aneurysm (AAA) intraluminal thrombus with SPECT scan, using 99m Tc radiolabeled Annexin V and 99m Tc radiolabeled Fucoidan [27,32]. Besides their ability to image the pathology, both of these systems used small molecules and it has been reported that injectable systems that present a bigger hydrodynamic diameter have many advantages as targeting devices. Having a longer circulation time [39,40] and presenting many ligands [41,42] at the surface of each single object, the binding at the pathological site would improve. One other advantage working with a microparticle system instead of small molecule is that the purification steps are easy to perform by a density gradient separation method. We could thus easily remove several forms of colloid technetium and free 99m Tc0 4that tend to distort the biodistribution of the radiopharmaceuticals [43,44]. For all these reasons we have decided to develop a microparticle type diagnostic tool. A functionalized polymeric injectable vehicle The water-in-oil emulsion crosslinking method produced a reproducible distribution of microparticle size. SEM images showed that the functionalization of the microparticles did not have significant effect on surface morphology since both type of microparticles exhibit smooth surfaces. All microparticles were below 4 µm, though, microparticles prepared with fucoidan present a smaller hydrodynamic diameter (358 ± 60 nm versus 503 ± 110 nm, p<0.05). We attribute this difference to the modification in the water-in-oil emulsion parameters due to the addition of fucoidan chains in the dispersed phase. First, it altered the total polysaccharide concentration in the water phase. Secondly, the fucoidan chains have a molecular weight of 57,000 g/mol which is small compared to dextran and pullulan chains that present a molecular weight of 500,000 g/mol and 200,000 g/mol respectively. It has been reported that decreasing the molecular weight of the polymers results in smaller particles for a water-in-oil emulsion process [45]. Finally, the fucoidan is a negatively charged polysaccharide and may modify the ionic strength of the system, known to have an influence on emulsions stability [46,47]. The literature describes many other polymeric carriers also developed to target vascular diseases with sizes, from 100 nm to 5 µm of mean diameter [17,[48][49][50][51]. On the one hand, nanoscale vehicles offer lower risk of vascular occlusion and avoid phagocytosis [52,53]. On the other hand, bigger vehicles over 500 nm would have a more adapted hemodynamic behaviour and have been reported as better candidates for targeting the wall in medium to large vessels relevant in several cardiovascular diseases [54][55][56]. Along these lines, the size distribution of our microparticles is well chosen as a compromise. They are not exclusively nanoscaled in order to improve the binding capacity and every single particle has a mean hydrodynamic diameter smaller than 4 µm to avoid vessel occlusion (the smallest human capillaries are about 5 microns). Moreover, hemolytic assay demonstrated that the developed polysaccharide microparticles are not toxic, at the described concentration, for red blood cells. For targeting the AAA, we functionalized the microparticles with fucoidan polysaccharide chains. Before adding the ligand to the system, the binding capacity of the fucoidan used in this work was studied. The SPR analyses confirmed that plain fucoidan with a molecular weight of 57,000 g/mol, has a strong affinity for P-Selectin (KD = 12 nM). It exhibited a better affinity for P-Selectin than its natural ligand does (KD = 320 nM) [57]. Addition of fucoidan resulted in a significant decrease of the zeta potential (-9.1 ± 0.6 mV versus -16.2 ± 0.8 mV, p<0.01) and an increase of the global amount of sulfur (1.5 ± 0.5 10 -3 % wt versus 113 ± 39 10 -3 % wt, p<0.05). Furthermore, EDX analysis revealed sulfur presence at the surface of the microparticles. These results confirmed that fucoidan, which is an anionic sulfated polysaccharide [58], is present at the surface of the functionalized microparticles. We quantified the presence of fucoidan at 1.64 ± 0.57 % wt. We then demonstrated that the functionalized microparticles have a strong affinity for human activated platelets. We measured by flow cytometry with functionalized microparticles a binding signal 7.7 higher than with non-functionalized control microparticles (8.5 ± 2.3 MFI ratio versus 1.1 ± 0.1 MFI ratio, p<0.001). For comparison, Modery et al. assessed by receptor-specific binding studies the affinity of liposomal nanoconstructs functionalized with EWVDV peptide for activated platelets and demonstrated a 5 times increase of binding magnitude versus control [59]. We also evidenced that this affinity is specific to the P-Selectin since we measured the same level of interaction with activated-then-P-Selectin-blocked platelets than with non-activated platelets. An efficient diagnostic tool of abdominal aortic aneurysm The radiolabeled functionalized microparticles enabled a strong signal uptake in the aneurysm area of the AAA rats on SPECT 30 minutes after intravenous injections. The mean SPECT signal measured in the AAA areas was more than three times higher than the mean signal measured on the renal aorta whereas no enhancement was observed on SPECT of healthy rats 30 minutes after injection of the same radiolabeled functionalized microparticles (3.49 ± 0.35 versus 1.03 ± 0.06, p<0.01). Rouzet et al. measured a median 99m Tc-fucoidan SPECT uptake ratio of 3.6, which means that fucoidan functionalized microparticle systems show similar diagnostic efficiency to the free radiolabeled fucoidan [32]. However, it appears that the areas of contrast uptake obtained in the AAA are larger with 99m Tc-MP-Fucoidan than with 99m Tc-fucoidan. In fact, when we compared the autoradiography analysis of cryosections of the harvested AAAs obtained in both cases, we evidenced that 99m Tc-fucoidan brought a radioactive signal limited to the intraluminal thrombus of the AAA whereas 99m Tc-MP-Fucoidan brought a signal to the whole aneurysm. Indeed, Rouzet et al. have reported that 99m Tc-fucoidan uptake and retention are more localized in the thrombus itself. The size of the diagnostic device could explain this difference. These results emphasized the reported advantages of micro and nanoscale carriers for targeting vessel walls. While free fucoidan molecules could not detect the injured endothelial cells which were not in the thrombus, this new SPECT diagnostic microtool would present an improved binding capacity, as compared to free fucoidan molecules, and an improved hemodynamic behavior that favors a contrast to the entire AAA injured wall. This strategy to develop a diagnostic tool of a higher hydrodynamic diameter seems therefore highly relevant. Actually, this research seems relevant regarding the other molecular imaging tools of AAA. Indeed, our radiolabeled microparticles resulted in a higher signal uptake of AAA wall, in acute stages AAA rat models, than the 18 F-FDG tool in symptomatic AAA patients; uptake ratios of 3.5 was obtained with 99m Tc-MP-Fucoidan versus 2.5 with 18 F-FDG [21]. Our technology therefore exhibited better displays than a molecular imaging method validated by clinical trials. Moreover, certain techniques requires a long time before reaching their targets. For instance, in magnetic resonance molecular imaging, macrophages accumulation sites imaging with USPIO and collagen imaging with targeted micelles incorporating gadolinium provides excellent contrast uptake in aneurysms, but only more than 30 hours after administration [10,60]. Our method with a diagnostic in the 30 minutes following the tracer injection is obviously more convenient for clinical application. We herein validated a molecular imaging tool of P-Selectin highly effective on acute AAA, and we will, in future studies, assess its ability to detect different stages and to predict the pathology growth. A biomimetic targeting device Histological analysis revealed that fluorescent 99m Tc-MP-Fucoidan localized inside the AAA wall. Light microscopy observation of Alcian blue stained sections and fluorescent microscopy observation of adjacent sections show microparticles in the media layer. Micro-autoradiography analysis coupled with Alcian blue staining demonstrated that the emission of electron was correlated with the presence of polysaccharide. Immunostaining revealed expression of P-Selectin in the same AAA area where 99m Tc-MP-Fucoidan accumulated. One hour after injection, the functionalized microparticles developed in this study seem to be able to penetrate inside the wall of the AAA. Accumulation in the inner wall of the AAA is similar to monocyte infiltration in the inflammatory cell recruitment involved in the pathology development [61,62]. Indeed the

microparticles accumulate in the same area where macrophages have been observed on rats elastase induced AAA [63] or on mice angiotensin II induced AAA [64]. It should be noticed that leukocytes express PSGL-1 [65,66] and our microparticles are functionalized with fucoidan chains that present sulfated sites that mimic the PSGL-1 anchor sites. Hence this technology could be considered as leukocyte mimetic microparticles. The property to migrate inside the arterial wall of the AAA could be a drawback for this technology and this should be taken into consideration for future work. Nevertheless, the number of microparticles seen on tissue sections is limited as compared to the quantity of leukocytes involved in the physiopathology [62]. Furthermore, this localization of the microparticles could be useful for future applications which will consist in associating a therapeutic molecule to this targeting device. Indeed, it is well known that the arterial wall degradation in aneurysm is due to a pathological proteolytic activity that occurs within the media layer and a promising strategy to treat AAA is aiming to inhibit this proteolytic activity [6]. For instance, Yoshimura et al. obtained regression of AAA with c-Jun N-terminal kinase inhibitors which reduce MMP proteolytic activity and enhance tissue repair [67]. Liao et al. demonstrated that angiotensin-converting enzyme inhibitors prohibited the degradation of medial elastin in AAA development [68]. The herein developed microparticles would act as a carrier that would bring the proteolytic inhibitor drug directly inside the AAA wall, where the therapeutic activity is needed. Besides, this targeted proteolytic action has been seriously considered for theranostic applications [69,70]. Moreover, this functionalized system targeted toward P-Selectin will certainly exhibit a binding avidity reflecting the evolution of the pathology and therefore provide a mean to self-adjust the amount of therapeutic molecules according to the inflammation degree. Future works will then consist in selecting the most adapted proteolytic inhibitor(s) and develop a method for its incorporation into the polysaccharide microparticles and its controlled release. As the microparticles developed are composed by reticulated polysaccharide chains, the biodegradability of a similar hydrogel being observed in a month on injured sites [71], we will propose a technology based on biodegradability for the release of the drug. Since many peptide of interest are hydrophilic and have amino groups, we can easily trap them into the microparticles or graft them with a chemical conjugation though amine bonds [72] in order to combine the therapeutic molecule of interest to the polysaccharide microparticles. Conclusion We hereby developed an efficient polysaccharide based SPECT diagnostic tool for the abdominal aorta aneurysm (AAA). In this work, we firstly described its fabrication with an emulsion-crosslinking process, confirmed its hydrodynamic diameter distribution and the presence of fucoidan ligand at the surface and characterized the radiolabeled product. We then demonstrated in vitro by flow cytometry that this fucoidan-functionalized system specifically binds to the P-Selectin expressed by human activated platelets. Finally, we showed in vivo on rats its ability to reveal AAA by SPECT imaging and we studied its fate in the AAA wall by histology experiments. Future works will consist in testing the ability of this diagnostic tool to detect the earliest stages of the AAA and in the incorporation of a therapeutic molecule. Association between common cardiovascular drugs and depression Abstract Objective: Cardiovascular diseases are associated with an increased risk of depression, but it remains unclear whether treatment with cardiovascular agents decreases or increases this risk. The effects of drugs on individual usage are also often unknown. This review aimed to examine the correlation between depression and common cardiovascular drugs, develop more potent interventions for depression in cardiovascular patients, and further research on the bio-behavioural mechanisms linking cardiovascular drugs to depression. Data sources: The data in this review were obtained from articles included in PubMed, EMBASE, and Web of Science. Study selection: Clinical trials, observational studies, review literature, and guidelines about depression and cardiovascular drugs were selected for the article. Results: We systematically investigated whether the seven most used cardiovascular drugs were associated with altered risk of incident depression in this literature review. Statins have been proven to have antidepressant effects. Some studies believe angiotensin-converting enzyme inhibitors (ACEIs)/angiotensin receptor blocker (ARB) can exert an antidepressant influence by acting on the renin-angiotensin system, but further clinical trials are needed to confirm this. Beta-blockers have previously been associated with depression, but the current study found no significant association between beta blockers and the risk of depression. Aspirin may have antidepressant effects by suppressing the immune response, but its role as an antidepressant remains controversial. calcium channel blockers (CCBs) can regulate nerve signal transduction by adjusting calcium channels, but whether this effect is beneficial or harmful to depression remains unclear. Finally, some cases have reported that nitrates and diuretics are associated with depression, but the current clinical evidence is insufficient. Conclusions: Statins have been proven to have antidepressant effect, and the antidepressant effects of ACEIs/ARB and aspirin are still controversial. CCBs are associated with depression, but it is unclear whether it is beneficial or harmful. No association has been found with β-blockers, diuretics, and nitrates. Introduction Depression, one of the most common mental health disorders, is among the leading causes of health-related disability worldwide. Studies have confirmed that depression occurs with even greater frequency in populations suffering from a wide range of cardiovascular conditions, such as coronary heart disease (CHD), hypertension, atrial fibrillation (AF), and heart failure (HF). Depression is two to three times more common in patients with CHD than in the general community, [1] and the research has demonstrated that depression is a risk factor for incident CHD or cardiovascular morbidity and mortality in patients with established CHD. [2] As a multifactorial pathology that affects between 30% and 40% of the general population, hypertension is also an essential risk factor in cardiovas-cular disease. A cross-sectional comparison study in Ghana and Nigeria suggest the prevalence of depression is high among patients with hypertension in Ghana and Nigeria. [3] Depression affects approximately one-third of hypertensive patients, affecting their quality of life and their prognosis. Thus, effective patient-centered interventions for depression are needed. AF is a clinically common arrhythmia, mostly occurring in patients with other underlying diseases, and mainly caused by ectopic pacemaker activities generated in the atrium during atrial conduction and reentry. Studies have shown that AF patients are more prone to psychological problems than the general healthy population. About 33% of AF patients have associated depression, which can not only increase the recurrence of AF but also seriously affects the quality of life and mortality of patients [4] ; among patients with HF, depression and anxiety disorders are common, with prevalence markedly higher than that in the general population. A meta-analysis of 36 studies found that clinically significant depressive symptoms affect 21.5% of HF patients, with one-third of HF patients reporting more depressive symptoms on questionnaires. [5] Cardiovascular diseases are associated with an increased risk of depression, but it remains unclear whether treatment with cardiovascular agents decreases or increases this risk. Accumulated studies have confirmed that some cardiovascular drugs have antidepressant effects. It is accepted that some cardiovascular drugs have anti-inflammatory or anti-oxidant actions, show protective effects toward the vascular endothelium, or alter regulation of neurotransmitters, [6] effects which may be related to their antidepressant mechanisms. At the same time, some studies have demonstrated that certain cardiovascular drugs positively correlate with the development of depression. For instance, reserpine can induce an imbalance in the regulation of monoamine neurotransmitters in the brain, leading to depression and is widely used to induce depression in disease models. [7] By expatiating upon the antidepressant effects of some common cardiovascular drugs, we aim to provide an up-to-date overview of the latest evidence regarding interventions used to treat depression among patients with cardiac disease. Data synthesis We conducted a literature search of PubMed, EMBASE, and Web of Science from January 1990 through May 2021, including clinical trials, observational studies, review literature, and guidelines limited to studies published in English. To explore the following topics: (1) pathophysiological mechanisms of cardiovascular drugs and depression, (2) the role of statins in depression, (3) the role of b-blockers in depression, (4) the role of antiplatelet medications in depression, (5) the role of calcium channel blocker (CCBs) in depression, (6) the role of angiotensin-converting enzyme inhibitors (ACEI)/angiotensin receptor blockers (ARB) in depression, (7) the role of diuretics in depression, and (8) the role of nitrate drugs in depression. After excluding duplicate studies, other publication types, those not relevant to the topic, or having insufficient sample size for significant conclusions, a total of 623 records were eligible for final reviewing. We picked out the controlled clinical trials as listed in Table 1. The screening process used was as illustrated in Figure 1. The representative articles selected are summarized as follows: Pathophysiological mechanisms of cardiovascular drugs and depression As we know, the link between depression and cardiovascular disease is bidirectional. Both have common pathophysiology and behavior as their basis [ Figure 2]. Cardiovascular drugs can improve the emotional state of cardiovascular patients, which indicates a close relationship to the regulation of a common pathogenic cause between depression and cardiovascular disease. Within each of the seven main classes of medication for cardiovascular disease, the active compounds show different overall pharmacological characteristics. Each drug is characterized by specific pharmacological properties, including selectivity of action dependent on receptor subtypes, intrinsic sympathomimetic activity, lipid solubility, and pharmacokinetic profile, as well as potential antiinflammatory properties. Thus, by reducing blood lipids and affecting cholesterol synthesis, statins can regulate metabolism in the nervous system. [8] Concurrently, studies have confirmed that statins can reduce inflammatory factors, suggesting that statins can play an antidepressant role by reducing the inflammatory response. [9] Previous studies had linked b-blockers use with an increased risk of depression, but more recent studies have found no significant association between the two. [10] Antiplatelet drugs, such as aspirin, may reduce the risk of depression by reducing inflammation, [11] but this remains controversial. CCBs modulate nerve cell signaling by regulating calcium channels, but whether this effect is beneficial or harmful in depression remains unclear. ACEI/ARB targeting-drugs may regulate the body's response to stress by regulating the renin-angiotensin system (RAS), thus reducing the risk of depression, [12] but there is still no solid clinical evidence to support this. Some cases have reported that nitrates and diuretics are associated with depression, but the current clinical evidence for this is insufficient. Common cardiovascular drugs in depression Currently, the main cardiovascular drugs used in clinical practice include lipid-lowering drugs such as statins, antiplatelet drugs such as aspirin and clopidogrel, ACEIs and ARBs, b-blockers, CCBs, diuretics, and nitrate ester drugs. We will now explore the correlation between the above medications and depression. The role of statins in depression Statins are commonly prescribed lipid-lowering drugs used primarily to treat high cholesterol and cardiovascular disease. In 1992, Lechleitner et al [13] reported four cases with primary hypercholesterolemia who developed severe depressive symptoms during treatment with pravastatin. This association between depression and statins has subsequently drawn much attention from researchers. Initially, the main viewpoint was that statins were associated with increased risk of depression. This perspective may be based on reports of a reduction of cholesterol synthesis in the brain on statin exposure. [14] Cholesterol is a membrane component that may bind to and alter the behavior of specific G-protein-coupled receptors, including a1A-adrenergic, b2-adrenergic, adenosine A2A, dopamine D1, and 5-HT1A receptors; and reports suggested that decreasing cholesterol can induce psychological disease. [15] However, subsequent controlled clinical studies did not find statins to be associated with an increased risk of symptoms of depression. Results from 14 controlled trials conducted by Bristol Myers Squibb do not support an association between pravastatin therapy and depressive symptoms. There was no difference between patients treated with pravastatin or placebo for up to 2 years. [16] A prospective cohort study that included 1631 subjects discovered that regular aspirin or statin at outset Chinese Medical Journal 2021;134 (22) www.cmj.org did not reduce the incidence of major depressive disorder (MDD) during a 3-year follow-up. [17] Another 12-week triple-blind, randomized, controlled trial enrolled 130 participants with moderate to severe MDD (aged 15-25 years old); participants were randomized to receive aspirin (n = 40), rosuvastatin (n = 48), or placebo, and the results indicated that neither aspirin nor rosuvastatin conferred any beneficial effect over and above routine treatment for depression in young

people. [18] Soh et al [19] conducted a randomized, double-blind clinical trial that enrolled 60 bipolar disorder and MDD patients using lithium. Their experimental data show that depression relapse during a 12-week follow-up was not significantly different between groups. A double-blind investigation carried by Glaus et al [17] also found that there was no significant effect of lovastatin in treatment for depression (P > 0.2). Similar results were obtained by Agustini et al [20] , Stewart et al, [21] and Muldoon et al [22] There are even some studies that have found that statins increase the risk of depression. For example, Chan et al [23] found that after 24 months of statin treatment, there was an increase of 2.8 points (1.5-4.0) on the Hamilton Depression Rating scale (P < 0.0001), although 64 (68%) of 94 patients' scores at 24 months still suggested either no depression or mild depression. The differences in these results may be related to the criteria for depression and the dosages of statins. HADS Metoprolol (+) Duch et al 1992 [43] b-blockers Spain/25 RCT BDI and Zung self-rating depression b-blockers (+) Hu et al 2020 [46] Aspirin Sweden/ 316,904 Cohort study ICD Aspirin (À) Weir et al 1996 [59] Amlodipine, Bisoprolol, Enalapril [60] CCB, b-blocker, and ACE inhibitor Kessing et al 2020 [70] Antihypertensive drugs Denmark/3,747,190 Cohort study Comprehensive diagnostic methods Angiotensin agents (À) Calcium antagonists (À) b-blockers (À) Diuretic ( * ) Potempa et al 1993 [71] Pindolol, propranolol, and hydrochlorothiazide CHD. Recent studies show that statins are related to lower risk in patients with other systemic diseases. A trial enrolling 465 stroke patients supports the conclusion that statin use was not associated with poststroke depression (PSD) status at outset but was significantly associated with reductions in PSD and significant PSD risk, specifically at follow-up. By enrolling two asthma-chronic obstructive pulmonary disease overlap syndrome (ACOS) cohorts, one taking statins (n = 1252) and one comprised of nonstatin users, matched by age, sex, and index date (n = 7887), Yeh et al [27] found that the ACOS cohort using statins had lower risks of anxiety and depression. The incidence of anxiety and depression was relatively low among users of statins with inhaled corticosteroids or oral steroids in the ACOS cohort. However, the association between statin use and depression is complex as research findings have been mixed. The possible mechanism underlying the antidepressant effect of statins may predominantly include the anti-inflammatory, anti-oxidant, and lipid-lowering properties of this drug class. An animal study has found that treating mice orally with vehicle (saline, 0.9%, control group) compared to 1 or 10 mg · kg À1 day À1 of atorvastatin or fluoxetine for seven days, the mice treated with atorvastatin presented lower depressive status than the placebo group; this was associated with the prevention of lipopolysaccharide (LPS)-induced depressive-like state and of an LPS-induced increase in tumor necrosis factor-a level and reduction in brain derived nerve factor level in the hippocampus and prefrontal cortex. [28] A clinical trial enrolling 217 coronary artery disease (CAD) patients reported that the patients treated with statin showed reduced Interleukin-1b (IL-1b) and Nuclear factor kappa B (NF-kB) levels. By comparing the correlation between simvastatin and pravastatin and depression, Harrison and Ashton [29] found that both kinds of statins could relieve the symptoms of depression, and the effect of simvastatin was greater than that of pravastatin. In a randomized, double-blind clinical trial, Haghighi et al [30] also found that statins had an antidepressant effect. Statins may function as antiinflammatory agents in therapy for depression in patients with CAD via downregulation of IL-1b and NF-kB. [31] The role of b blockers in depression b-blockers are a family of agents widely used to treat hypertension, angina pectoris, and cardiac arrhythmias. [32] They are widely used in patients with cardiovascular disease. The first b-blocker to be implicated in depression was propranolol, whose lipophilic structure and penetration of the blood-brain barrier (BBB) could explain its role in depression. [33] b-blockers can be divided into two types, fat-soluble and water-soluble. Compared with water-soluble b-blockers, fat-soluble blockers can more easily permeate into the central nervous system (CNS) through the BBB to reach high concentrations. Studies suggested that b-blockers can be used in the treatment of essential tremors and have unique therapeutic significance, and previous work supports that they can induce depression, insomnia, dreaming, and other side effects. [34] In a 1967 article published in the British Medical Journal, Fitzgerald et al [35] reported on a case group of 89 hypertensive patients who were being treated for arrhythmia. Results of subsequent research on the link between b-blockers and depression have been mixed, prompting more research on the subject. We feel that the association between depression and b-blockers may need to be further explored due to the possibilities of publication bias and use of different criteria for depression assessment. Since Stoudemire et al [36] posed, in 1984, that there is very little evidence to link propranolol with mood disturbance, subsequent studies have consistently challenged the dogma that b-blockers cause depression. Recent studies in patients with heart disease have tended to sever the link between b-blockers and depression, treatment with b-blockers even being associated with a lower risk of depression in some cases. A retrospective cohort study of 6915 patients, in a multicenter integrated healthcare system, diagnosed with HF between 2008 and 2014, found that among them, 1252 had a diagnosis of depression. During a mean follow-up of 2.6 years, the authors reported that depressed patients not treated with a b-blocker had higher mortality than non-depressed patients (adjusted hazard ratio [aHR], 1.4, 95% CI 1.09-1.70, P < 0.05). When treated with b-blockers, their risk of mortality was attenuated (hazard ratio [HR] 1.1, 95% CI 0.97-1.20, P = 0.14). [37] Randomized trials of b-blockers used in the treatment of myocardial infarction (MI), HF, or hypertension were identified by searching the MEDLINE database for English-language articles . A paper published in JAMA also supports the idea that there is no significant increased risk of depressive symptoms and only small increased risks of fatigue and sexual dysfunction. [38] In a multicenter study, 381 MI patients were assessed for depressive symptoms using the Beck Depression Inventory at outset and at t = 3 months, 6 months, and 12 months post-MI; patients were matched using the frequency matching procedure according to age, gender, hospital of admission, presence of baseline depressive symptoms, and left ventricular function. No significant differences were found between non-b-blocker users and b-blocker users in the presence of depressive symptoms. [10] Additionally, the results of Sorgi et al, [39] Rosenberg et al, [40] and Pérez-Stable et al [41] also support this view. There are also some studies suggesting that the use of b-blockers increased the risk of depression. For example, Liu et al [42] analyzed the correlation between metoprolol use and depression in patients with chronic HF. It was found that their Hospital Anxiety and Depression Scale and the Copenhagen Burnout Inventory (CBI) scale scores significantly increased from baseline throughout the study's timeframe. Similar results were found by Duch et al's [43] study. The differences in these results may be related to the quantity of b-blocker crossing the BBB. The amount of a fat-soluble b-blocker crossing the BBB is determined by plasma protein-binding (low uptake) and fat-solubility. The potency of action is determined by the local concentration of the drug, the number of receptors, and the affinity for the relevant receptors. [44] The role of antiplatelet medications in depression Aspirin is a commonly used antiplatelet drug in the clinic. In addition to its inhibitory effect on platelet aggregation, it can also be used as an anti-inflammatory and analgesic agent. [45] Numerous clinical and animal studies have confirmed that aspirin treatment has a specific correlation with reduction in signs of depression. A cohort study involving all patients in Sweden diagnosed with a first primary malignancy between July 2006 and December 2013, found that among 316,904 patients identified, 5613 patients received a diagnosis of depression, anxiety, or stress-related disorder within 1 year after a cancer diagnosis. Compared with those with no use of nonsteroidal anti-inflammatory drugs, the use of aspirin alone was associated with a lower rate of depression, anxiety, and stress-related disorders (HR, 0.88; 95% CI 0.81-0.97). [46] Bhatt et al [47] demonstrated in animal studies that aspirin improved depression in sprague-dawleyrats both in combination with dexamethasone and alone. The animals treated with aspirin showed increased sucrose preference, decreased immobility-time in the forced swim test, decreased serum cortisol, and increased brain serotonin levels signifying antidepressant action. Studies have shown that aspirin's antidepressant mechanisms are closely related to its anti-inflammatory effects. Many clinical trials and observational studies have been conducted based on the premise that the anti-inflammatory effect of aspirin may prevent depression. Inflammatory processes associated with persistent infection have long been discussed as etiological factors in psychiatric disorders. Studies have found that people with major depression have higher levels of pro-inflammatory cytokines, such as IL-1, IL-6, tumor necrosis factor-alpha, and C-reactive protein. After antidepressant treatment for depression, a decrease in IL-4, IL-6, and IL-10 cytokine levels was observed in serum. [48][49][50] These observations raise the possibility that the occurrence and development of depression are associated with the levels of inflammatory factors. Indeed, preclinical, pharmacoepidemiologic, and pilot clinical trial data suggest that aspirin may have clinical potential in psychiatric use based on its anti-inflammatory properties. [51,52] However, whether aspirin has an antidepressant effect is still controversial. Some studies show no significant association between aspirin and depression. A prospective cohort study including 1631 subjects (43.6% women, mean age 51.7 years), randomly selected from the general population of an urban area reported that, after a mean duration of 5.2 years follow-up, and adjusted for a wide array of potential confounders, data did not support largescale preventive treatment of depression using aspirin or statins in subjects aged between 35 and 66. [16] To determine whether low-dose aspirin reduces the risk of depression in healthy older adults, Berk et al [53] observed 199,114 participants in whom 9525 received aspirin and 9589 received a placebo with a median follow-up of 4.7 (3.5-5.6) years, and a rate of depression of 70.4 events per 1000 person-years in the aspirin group and 69.1 in the placebo group; there were no significant differences at annual visits in the proportion of depression reports between the two groups. The result may be connected to an apparent increase in the risk of bleeding events on aspirin treatment. These negative results may be related to the dose of aspirin used in the trial. It is not clear whether low-dose aspirin reduces inflammatory cytokines. Studies found that aspirin reduced the risk of schizophrenia at doses ranging from 1 g/day to 45 g/day. [11] At the same time, aspirin can increase the risk of bleeding, which increases the economic and mental burden on patients to a certain extent and may also affect the experimental results. Clopidogrel is another widely used antiplatelet drug in cardiovascular disease. It can significantly improve primary cardiovascular outcomes in patients following percutaneous coronary intervention. An antidepressant effect of clopidogrel has been reported in a few cases, but there is still a lack of robust clinical and animal experiments to confirm this. The role of CCBs in depression CCBs are a heterogeneous group of structurally unrelated compounds that are widely used today in the treatment of cardiovascular disease. CNS effects of verapamil and diltiazem have been described previously. Studies have confirmed that drugs with antagonistic effects on cellular calcium also possess psychotropic properties, an action which is associated with CCBs' has an influence on intracellular calcium homeostasis. [54] Cellular calcium homeostasis not only plays a crucial role in neuronal cell signal processing but also has an impact on both synthesis and release of neurotransmitters. In a rat model, CCBs can competitively combine with neurotransmitter receptors and induce psychotic symptoms. [55] It has been suggested that receptors interact with a neurotransmitter or neuromodulator in somewhat the same way that receptors for gamma-aminobutyric acid and benzodiazepines interact in the regulation of chloride-ion channels. Following observations that verapamil could improve response to antidepressants, an animal study sought to investigate whether this was the result of verapamil effects on the Pglycoprotein transporter in the BBB. Administered by mouth or intravenously, verapamil crossed the BBB, and verapamil pre-treatment significantly elevated concentrations of antidepressant imipramine in all brain

regions studied; the effect was most pronounced in the brainstem and frontal cortex with about a doubling of brain region to serum concentration ratios being observed by Clarke et al [56] Further studies have confirmed that CCBs could be used alongside antidepressants to treat models of depression: Aburawi et al [57] found that combining nifedipine with alprazolam produced additional antidepressant effects in animal trials, indicating that they exert antidepressant effects through different mechanisms. However, some clinical trials have shown no correlation between depression and CCBs. In a multicenter randomized, double-blind, two-way, crossover study design, used to compare the antidepressant effect of bisoprolol and nifedipine in hypertension, there was no noticeable change in the depression scores observed in the 4 to 6 weeks follow-up. [58] Weir et al [59] also reported that there was no clear evidence of a link between depression and CCBs. Rathmann et al [60] reported that the use of CCBs is associated with increased risk of depression. A trial that enrolled 972 diabetic patients indicated that, among diabetic patients, new prescriptions of calcium channel and b-blockers were associated with an estimated two-to three-fold increased risk of subsequently diagnosed depression. Although there are shreds of evidence that CCBs can interfere with nervous system activity, a qualitative association between depression and CCBs still lacks evidence derived from clinical trials; this may relate to the type and dosage of CCBs used. It has been suggested that fat-soluble CCBs are more strongly associated with depression than non-fat-soluble CCBs, which may reflect the ease of entry of fat-soluble drugs into the and across the BBB. [61] The role of ACEIs/ARBs in depression ACEIs/ARBs play a pivotal role in regulating the RAS by adjusting the synthesis and release of angiotensins in order to regulate blood pressure. In recent years, multiple studies found that ACEIs/ARBs can also delay and reverse ventricular remodeling, prevent further development of myocardial hypertrophy, improve endothelial function and heart function, and reduce the incidence of arrhythmia, and effectively improve healing in cardiovascular diseases and improve survival rate. [62] In addition to the regulation of blood pressure, the RAS is also a vital regulator of the inflammatory state of the nervous system. Several studies have shown that depression is closely related to the occurrence of inflammation in the brain. [63][64][65] Therefore, in recent years, many studies have proposed that ACEI/ARB drugs can reduce the risk of depression. Several observations have linked angiotensinconverting enzyme polymorphisms with depression and the serotonin and dopamine neurotransmitter systems. In a clinical trial that enrolled 625 Caucasian men with mild hypertension, the results suggested use of ACE inhibitors was associated with a reduced likelihood of risk of depression. [66] Prospective trials were initiated to compare the relative efficacy and influence on the quality of life of angiotensin converting enzyme inhibitors, and the results confirmed that ACEI captopril was associated with a more significant reduction in complaint rate as an index of life quality (P < 0.05) compared with methyldopa and showed a tendency to produce fewer symptoms of depression compared with other cardiovascular drugs. [67] However, there are few randomized clinical trials on ACEI and ARB targeting drugs and depression, which indicates the need to undertake future clinical studies. The role of diuretics in depression Diuretics are the most commonly chosen treatment for mild to moderate hypertension and are among the most commonly prescribed medications. [68] There is little evidence that these medications have effects on the CNS. Okada et al [69] reported eight cases (five men, three women; ages ranging from 44 years to 69 years) of depression induced by antihypertensive diuretics, but there is still no substantial further evidence. It is unclear whether the depression was caused by the drug directly or by lowered blood pressure. Some clinical studies have looked at the relationship between diuretics and depression, but the results have not shown any significant correlation. A nationwide population-based study investigated whether four antihypertensive drugs, angiotensin agents, calcium antagonists, b-blockers, and diuretics can decrease the risk of depression. They observed participants in the Danish population beginning January 2005 with follow up until December 2015. The results showed "Continued use of classes of angiotensin agents, calcium antagonists, and b-blockers was associated with significantly decreased rates of depression." [70] In a case-control study which enrolled 41 patients, Potempa et al [71] found there were no significant within or between subject effects for drug group or drug treatment effects with hydrochlorothiazide. Another multicenter randomized placebo-controlled trial of the treatment of isolated systolic hypertension in elderly subjects used chlorthalidone as the step one treatment. At the 5-year follow-up, no meaningful changes in mood were found in this population. [72] Although no meaningful results were observed, this study may serve as substantial evidence of no association between depression and diuretics due to the study's large sample size. Up to now, the majority view has been that diuretic's effects on depression or CNS behavior occur only when electrolyte disturbances develop. Otherwise, it appears that these medications have little or no effect on mood. The role of nitrate drugs in depression Nitrate drugs are one of the oldest cardiovascular drugs in clinical practice. They were initially used to prevent and treat angina pectoris and gradually applied to the treatment of HF and hypertension. [73] The primary mechanism of action is that nitrate esters can be converted into NO by the action of a series of enzymes; this then activates intracellular signal transduction pathways, reducing the concentration of calcium ions in smooth muscle cells, and relaxing smooth muscle. [74] There are few studies on the correlation between nitrates and depression. However, some studies suggest that plasma NO concentration can be related to the functioning of nerve cells, and the increase of NO concentration can change the functional activities of nerve cells and induce depression, anxiety, and other mental-health conditions. [75,76] A casecontrol study in 50 treatment-naïve young adults with a first episode of major depression and 50 healthy control subjects was conducted. The authors reported that decreased plasma concentrations of nitric oxide metabolites was not associated with vascular endothelial dysfunction in these young subjects. [77] Reduced nitrate levels could reflect decreased nitric oxide production in the CNS of depressed subjects. Further studies are needed to confirm this hypothesis. Discussion There is a bidirectional association between depression and cardiovascular disease. Patients with depression are more likely to develop cardiovascular disease than healthy individuals and vice versa. Studies have confirmed that depression is an independent risk factor for predicting the prognosis of patients with cardiovascular disease. [78,79] Clinical trials have found that the depressive symptoms of This article describes the relationship between seven common cardiovascular drugs and depression and their possible mechanisms. Statins can reduce the risk of depression by reducing inflammatory factors, while the antidepressant effect of aspirin is still controversial. There have been case reports on ACEI and ARB drugs' abilities to reduce the risk of depression, but there is less compelling data on their potential physiological benefits. b-blockers were previously thought to reduce the risk of depression, but recent studies have found no significant association between the two. CCBs affect nervous system activity by affecting calcium channels, but their role in depression remains undetermined. Some cases of depression induced by diuretic drugs have been reported, but it is believed that depression may be related to electrolyte disturbance. Finally, nitrate drugs were not significantly associated with depression. The present study suggests the correlations between cardiovascular drugs and depression remain unclear; the reason for differences between research results, in addition to the human factor, may arise mainly from the following: First, the drug dose differs; in studies on statins, Berk et al [18] found that the use of statins can relieve the symptoms of depression. At the same time, Jun-Jun et al [27] believe that low doses of statins and depression show no significant correlation. Therefore, the relationship between a high dose and a low dose of drug administered and depression needs to be further studied. Moreover, the effect of different doses on the treatment outcome should not be ignored. Therefore, the relationship between high dose or low dose therapy and depression needs to be further studied. Second, there are different criteria for depression evaluation. At present, the main diagnostic methods for depression include Patient Health Questionnaire (PHQ-9), Beck Depression Questionnaire (BDI), Hamilton Depression Scale (HAMD), etc. [80] To eliminate this difference, large-scale clinical trials need to be implemented. Third, we may mention the influence of other drugs. Because clinical patients may have various diseases, it is inevitable that they may take different drugs, a variable which may not be controlled for. Although the association between cardiovascular medications, such as nitroglycerin and diuretics and depression, is unclear, several studies have found that cardiovascular medications can enhance the effectiveness of antidepressants. Conclusion In this literature review examining PubMed, EMBASE, and Web of Science from January 1990 through May 2021, we systematically investigated whether the seven most used antihypertensive drugs were associated with altered risk of incident depression. Statins have been proven to have antidepressant effects. Some studies believe ACEIs/ARBs can exert an antidepressant influence by acting on the RAS, but further clinical trials are needed to confirm this. b-blockers have previously been associated with depression, but the current survey found no significant association between b-blockers and the risk of depression. Aspirin may have antidepressant effects by suppressing the inflammatory response, but its role as an antidepressant remains controversial. CCBs can regulate nerve signal transduction by modulating calcium channels, but whether this effect is beneficial or harmful to depression remains unclear. Finally, some cases have reported that nitrates and diuretics are associated with depression, but the current clinical evidence is insufficient. However, the findings should be replicated in welldesigned, more extensive randomized controlled trials using appropriate designs and statistical analyses to address selection and confounding factors. Further research is needed to determine whether common cardiovascular drugs reduce the risk of depression and to identify the factors that are associated with such a reduction. Funding The study was granted by the grants from the National Natural Science Foundation of China (No. 81970447) and China Women's Development Foundation (No. 2019300). Temporal Dynamics of Salmonella enterica subsp. enterica Serovar Agona Isolates From a Recurrent Multistate Outbreak The largest outbreak of Salmonella Agona in the United States occurred in 1998. It affected more than 400 patients and was linked to toasted oat cereal. Ten years later, a similar outbreak occurred with the same outbreak strain linked to the same production facility. In this study, whole-genome sequence (WGS) data from a set of 46 Salmonella Agona including five isolates associated with the 1998 outbreak and 25 isolates associated with the 2008 outbreak were analyzed. From each outbreak one isolate was sequenced on the Pacific Biosciences RS II Sequencer to determine the complete genome sequence. We reconstructed a phylogenetic hypothesis of the samples using a reference-based method for identifying variable sites. Using Single Nucleotide Polymorphism (SNP) analyses, we were able to distinguish and separate Salmonella Agona isolates from both outbreaks with only a mean of eight SNP differences between them. The phylogeny illustrates that the 2008 outbreak involves direct descendants from the 1998 outbreak rather than a second independent contamination event. Based on these results, there is evidence supporting the persistence of Salmonella over time in food processing facilities and highlights the need for consistent monitoring and control of organisms in the supply chain to minimize the risk of successive outbreaks. INTRODUCTION Salmonella enterica subsp. enterica is comprised of more than 1,500 serovars, including Salmonella enterica subsp. enterica serovar Agona (Salmonella Agona). Salmonella Agona was first isolated from cattle in Ghana in 1952 (Guinee et al., 1964). Within the last several years, Salmonella Agona has been one of the top 20 most commonly reported serotypes causing human infections (CDC, 2014). It has been responsible for numerous human outbreaks associated with various dry food products such as dried milk, machacado, dried snacks, and cereal (Clark et al., 1973;Killalea et al., 1996;Taylor et al., 1998). In the United States, the largest known outbreak of Salmonella Agona infections occurred between April and May in 1998 (Russo et al., 2013). During that time, the Centers for Disease Control and Prevention (CDC) investigated a multistate (11 states) outbreak of Salmonella Agona infections comprised of 209 confirmed cases (Centers for Disease Control and Prevention, 1998).

Among the 162 case patients for which there was available information, 46 (28%) had been hospitalized (Centers for Disease Control and Prevention, 1998). Collaborative investigative efforts by state and federal officials implicated toasted oat cereal as the source of this Agona outbreak. The pulsed-field gel electrophoresis (PFGE) Xbal pattern (JABX01.0001) of the Agona isolate from the cereal production facility was indistinguishable from the predominant PFGE pattern among outbreak-associated clinical isolates. After the investigation, the implicated production lines from the company's plant were sealed off, all equipment was removed, all surfaces were stripped to bare concrete, decontaminated, refinished and new production lines were installed (Russo et al., 2013). A decade later, in April 2008, the CDC announced a 15state outbreak associated with Salmonella Agona resulting in 28 identified cases, with eight individuals being hospitalized. The federal investigation suggested that unsweetened puffedrice cereals and unsweetened puffed-wheat cereals from the same company that was associated with the 1998 outbreak were the likely sources of contamination (CDC, 2008). The PFGE pattern of the 2008 outbreak strain had the same Xbal pattern (JABX01.0001) as of Salmonella Agona identified during the 1998 outbreak associated with toasted oat cereal by company X. On April 5, 2008 the same company X voluntarily recalled its unsweetened puffed-rice cereals and unsweetened puffed-wheat cereals after finding Salmonella Agona contamination during routine testing from the same plant (Russo et al., 2013). In this study, whole genome sequence (WGS) data from a group of closely related Salmonella Agona isolates were analyzed. The majority of the isolates were investigated as a part of the two outbreaks of salmonellosis associated with cereals that were separated by 10 years and our objective was to determine the evolutionary relationships among these outbreak isolates as a means to differentiate among two competing hypotheses: (1) persistence of Salmonella within the facility and (2) reintroduction of the same strain. We also demonstrate how the use of WGS can identify stable SNP targets that can be utilized for differentiating closely related Salmonella Agona isolates and clustering closely related isolates together, such as those of the 1998 and 2008 outbreak. Further, we show that the 2008 outbreak involves direct descendants from the 1998 outbreak rather than a second independent event. Bacterial Isolates We analyzed 46 Salmonella Agona whole genome sequences, 31 of which are first reported here while the remainder were publicly available from GenBank at NCBI (Table 1). Of these 46 isolates, 30 were linked with the original investigation into the outbreak of salmonellosis associated with breakfast cereal including one clinical isolate and 4 food isolates obtained in 1998 and 5 clinical isolates, 8 food isolates, and 12 environmental isolates acquired in 2008. All 30 outbreak associated isolates exhibited indistinguishable Xbal PFGE patterns (JABX01.0001). We also included one clinical Salmonella Agona isolate SL483 containing the same Xbal PFGE pattern from a different event isolated in Wisconsin 2003. The remaining Salmonella Agona isolates were chosen to provide a measure of the genetic diversity that is found among other isolates within the serovar. DNA Preparation, PFGE, and Genome Sequencing Pulsed-field gel electrophoresis was performed according to the PulseNet protocol of the Center for Disease Control and Prevention 1 (CDC). Genomic DNA of each strain was isolated from overnight cultures using DNeasy Blood and Tissue kit (Qiagen, CA, United States). Twenty-eight isolates were shotgun sequenced with the Roche Genome Sequencer FLX (454 Life Sciences, Roche, Branford, CT, United States) using the GS FLX Titanium Sequencing Kit XLR70 according to the manufacturer's protocol. One isolate was sequenced using Illumina's MiSeq platform (Illumina, Inc., CA, United States). Sample preparation and the sequencing library was prepared using the Nextera XT Sample Preparation Kit and then sequenced for 2 × 151 cycles. Two outbreak associated isolates (1998 and 2008) were sequenced on the Pacific Biosciences (PacBio) RS II Sequencer and assembled as previously described (Hoffmann et al., 2015). Specifically, a single SMRTbell 10-kb library was prepared according to the 10-kb PacBio sample preparation protocol and sequenced using C2 chemistry on two singlemolecule real-time (SMRT) cells with a 180-min collection protocol. The 10-kb continuous long read (CLR) data were de novo assembled using the PacBio hierarchical genome assembly process (HGAP3)/Quiver software package, followed by Minimus2, and polished by Quiver (Chin et al., 2013). The assembled sequences were annotated using the NCBI Prokaryotic Genomes Annotation Pipeline (PGAP) and have been deposited at DDBJ/EMBL/GenBank. Construction of SNP Matrices and Phylogenetic Inference We used the CFSAN SNP Pipeline 2 (Davis et al., 2015) to construct a SNP matrix among the 46 individuals. This pipeline is a reference-based approach within which Bowtie2 (Langmead and Salzberg, 2012) is used to align reads to the reference and VarScan (Koboldt et al., 2012) is used to detect variants. Within the pipeline, we used Salmonella Agona SL483 (CP001138) as the reference; using NCBI's methodology (Ciufo et al., 2018) CP001138 has a type strain ANI of 98.6994 confirming taxonomic accuracy. We applied a SNP density filter (i.e., three or more SNPs in a 1000 bp window were removed) to remove variants that are likely the result of recombination or mobile elements (e.g., Petkau et al., 2017). With the resulting SNP matrix from the CFSAN SNP Pipeline, we constructed a phylogenetic hypothesis of the evolutionary relationships among the 46 isolates investigated using the Genetic Zwickl, 2006). We used the default settings within GARLI and the best tree was chosen among 10 replicate runs using the observed SNP matrix; topological support was estimated by conducting 1,000 bootstrap replicates. Trees were rooted based on preliminary analyses including the serovar Soerenga as an outgroup. We also applied the ACCTRAN method within the R package phangorn (Schliep, 2011) to provide the number of changes along branches of the best tree from GARLI. We used PopART to construct a minimum spanning network (Bandelt et al., 1999) using the SNP matrix. Functional and Genomic Differences A custom python script was written to determine the position of each SNP in the reference genome, the gene within which it was found, and the type of mutation (synonymous or nonsynonymous). A custom script was also used to identify fixed SNPs between the outbreak clade and all other isolates. Both scripts are available at https://github.com/CFSAN-Biostatistics/ snps_in_genes. Using the COG information in the feature table for the reference found in the Genome database at NCBI, we then assigned each gene to one of the higher functional categories. We compared the abundance of each category between the isolates found in the outbreak and the complete dataset, which indicates what functions might be more pronounced in the former. To further explore genomic differences among the isolates, the complete chromosomes were aligned with MAUVE aligner version 20150226 using progressive algorithm with default settings (Darling et al., 2004). We constructed a circle plot of genomic differences between the publicly available Agona strain SL483 and the two completely closed outbreak associated isolates CFSAN000477 and CFSAN000471using BRIG (Alikhan et al., 2011). RESULTS AND DISCUSSION Public health authorities were informed in 1998 and 2008 of outbreaks involving Salmonella Agona that resulted from consumption of contaminated cereal from the same company X. The PFGE pattern of these highly clonal Agona isolates from 1998 and 2008 were indistinguishable. At that time PFGE was the method used for traceback during an outbreak investigation. However, PFGE does not have the resolution to distinguish a transient strain from a resident strain (Pightling et al., 2018) and, therefore, at that time using this method officials from the state health department were able to correctly identify the source of an outbreak but were not able to determine if the 2008 Agona isolate is the same strain as the 1998 Agona isolate. Since 2012, several studies have demonstrated the higher resolution and increased discriminatory power of WGS relative to that of PFGE where the former can differentiate highly clonal isolates. This increased resolution of WGS is critical in a successful investigation of foodborne outbreaks to determine the source(s) of contamination (Hoffmann et al., 2014(Hoffmann et al., , 2016Allard et al., 2018). Thus, WGS was invoked retrospectively to the cereal outbreaks of 1998 and 2008 in an attempt to resolve and ameliorate the shortcomings of the other more traditional molecular strategies used for such purposes, as well as to characterize the dynamics of the two outbreak scenarios. First, using long read sequencing the complete genome sequences of Salmonella Agona isolate CFSAN000471 obtained from cereal during the outbreak in 1998 and of Salmonella Agona isolate CFSAN000477 obtained from unsweetened puffed-rice cereal during the outbreak in 2008 (Hoffmann et al., 2015) were sequenced with the Single Molecule, Real-Time (SMRT) sequencing technology. Isolate CFSAN000471 was sequenced with 110× coverage; its genome size was 4,798,531 bp, and the G + C content was 52.1% and Isolate CFSAN000477 was sequenced with 130× coverage; its genome size was 4,797,172 bp, and the G + C content was also 52.1%. At the time Salmonella Agona SL483 was the only complete genome from Salmonella Agona available in Genbank; it had a genome size of 4,798,660 bp, and a G + C content of 52.1%. The strain was isolated from a clinical sample and is associated with a sporadic case in 2003. Figure 1 shows the pronounced similarity between these three closed Salmonella Agona isolates suggesting that they are highly related with each other. To further characterize them we used the MAUVE aligner to Frontiers in Microbiology | www.frontiersin.org align them and checked manually for differences. The different genome sizes can be explained by the insertion/deletion of transposases. CFSAN000477 and CFSAN000471 carry the same transposase but at different locations within the chromosome (CFSAN000471 transposase insertion: 710,499;CFSAN000477 (transposase insertion: 987,229) while SL483 is missing the specific transposase but instead carries an insertion sequence element, IS10, at three different loci. Further, very few SNPS and deletions and insertions were found among the three isolates listed in Supplementary Table 1. Explicitly, CFSAN000471 isolated in 1998 showed only one unique informative SNP in gene int (locus_tag = SEEA8692_015865) not found in CFSAN000477 and SL483. SL483 has one unique informative SNP in gene hslO (locus_tag = SeAg_B3697), two unique SNPs, two unique deletions and two unique insertions in non-coding region, and is missing a hypothetical protein. CFSAN000477 has six unique including five informative SNPs [fusA (locus_tag = SEEA0421_002795), GDPmannose pyrophosphatase (locus_tag = SEEA0421_021175), transporter gene (locus_tag = SEEA0421_019215), protease (locus_tag = SEEA0421_011315), hypothetical protein (locus_tag = SEEA0421_010320), and fructose-6-phospharte aldolase A (locus_tag = SEEA0421_006170)], four SNPs, one insertion and one deletion in non-coding region. The genomic size (including extrab. -chromosomal DNA) of the 46 Salmonella Agona isolates varied from 4.6 to 5.1 Mb. FIGURE 2 | Phylogenetic reconstruction of the 46 Salmonella Agona isolates investigated herein. The tree was constructed from a SNP matrix obtained with the CFSAN SNP Pipeline within which the isolate Salmonella Agona SL483 was used as the reference. The program GARLI v2.01.1067 was used to infer the tree. The heatmap to the right is ordered according to the tips of the tree and shows the pairwise SNP distances among isolates (e.g., the first two columns and last two rows represent the two most basal isolates in the tree that are very different from all other isolates but similar to one another). Primarily, differences in genomic size was due to the presence or absence of mobile genetic elements, such as phages and genetic islands. Using the complete genome of Salmonella Agona (SL483), we were able to more accurately map the raw reads to the reference genome to build a SNP matrix and subsequently construct a phylogenetic tree (Figure 2). Overall, we identified 441 variable SNPs with 112 not located on a gene. Supplementary Table 2 lists all 441 SNPs including SNP position in the genome, description of the gene where the SNP is located, SNP position in the gene, amino acid change and the substitution type; Supplementary Figure S1 shows the number of changes along branches of the phylogenetic tree inferred with GARLI. The data derived from the phylogenetic analysis of 46 Salmonella Agona isolates (Table 1), collected from various sources between 1987 and 2009 including 5 isolates associated with the 1998 outbreak, and 25 isolates associated with the 2008 outbreak provided critically relevant information. First and most importantly, the 1998 and 2008 outbreak associated isolates clustered together to the exclusion of other isolates with strong support (i.e., 19 fixed SNPs and 100% bootstrap support).

Second, the tree showed that all of the isolates associated with the 2008 outbreak, including 6 clinical, 7 cereal, and 12 environmental isolates from the implicated food production facility clustered together with 92% bootstrap support (Figure 2) and three fixed SNPs. Third, the two clinical and three food isolates associated with the 1998 outbreak clustered together with 96% bootstrap support and four fixed SNPs. Interesting, the reference isolate SL483 grouped together with the 1998 outbreak isolates. SL483 is a clinical isolate (sporadic case) obtained in 2003. The analyses suggest that most likely the patient was infected with the same Salmonella Agona strain. Interestingly, there are also two strains from 1998 and two strains from 2008 that were part of a previous publication Zhou et al., 2013 that also cluster with the sample associated with the particular cereal outbreak. The metadata for those samples indicates that they indeed are part of the outbreak event ["A food isolate, isolated at the time of the (1998| 2008) outbreak in the United States"]. Fourth, there is a tendency for isolates to cluster based on geographical location such as United States, Europe and China and proximity. The most distinct isolate in this study was collected from herbal tea in Mexico in 2007. Given the importance of the tree topology to the accurate interpretation of this study results, we constructed using the same SNP matrix a minimum spanning tree (Figure 3) to further illustrate the relationships among the isolates and to provide additional insight into whether the observed phylogenetic relationships may be influenced by recombination and/or mobile elements. With the phylogenetic analyses linking the 1998 and 2008 strains, the network analysis illustrates that the outbreak isolates form a star-like clade (Figure 3). Star-like clusters are often attributed to rapid diversification and/or a clonal swarm. In this situation, it reveals additional evidence that indeed the 2008 outbreak are direct descendants from the 1998 outbreak rather than a second independent contamination event. Therefore, our analyses suggest that company X had a resident strain surviving inside their facility for a minimum of a decade. Given that the facility was thought to have taken appropriate measures to eliminate the presence of Salmonella, the question arose as to how a strain found 10 years earlier could have persisted in the plant and caused the subsequent outbreak in 2008. One possible explanation is that the bacteria persisted in the cement walls and there is evidence for Salmonella forming dense biofilms on such a surface (Joseph et al., 2001). Based on our SNP analyses where variable positions were mapped to an annotated reference genome, there does appear to be fewer number of SNP differences within genes linked to cell mobility, intracellular transport and transcription within the cereal clade (Figure 4). Such evolutionary changes may be driven by selection pressures from bacterial controls instituted by the food industry or by production processes, resulting in enhanced survival, persistence and even growth within food matrices and in the production environment. These increase the likelihood of foodborne outbreaks with morbidity and mortality that threaten public health. To prevent such outbreaks, we must identify the genetic factors responsible for pathogen persistence across the farm to fork supply chain. CONCLUSION The subtyping of isolates associated with a foodborne outbreak event is essential for successful investigation and eventual traceback to a specific food or environmental source (Allard et al., 2012). In this regard, PFGE has been thought of as the "gold standard" by augmenting public health investigations with information regarding genetic similarity of isolation for nearly two decades (Bergholz et al., 2014). However, numerous clonal strains, a particularly common phenomenon within Salmonella, confound epidemiological investigations because PFGE and other traditional molecular typing tools cannot separate these clonally related strains (CDC, 2010). The population and evolutionary dynamics associated with clonal bacterial populations can now be investigated with much greater resolution afforded by next generation sequencing (NGS) data. We are now at a stage that WGS can identify if a company's product is contaminated with a new strain of a pathogen from the environment or if they have a resident strain within their facility that has persisted over a period of time (e.g., multiple years). In this study, WGS data from a set of Salmonella Agona isolates were analyzed to provide insight into the evolutionary relationships among strains linked to two outbreaks of salmonellosis separated by ten years. All of the Salmonella Agona strains associated with the 1998 and 2008 outbreaks could be traced to the same cereal facility in the United States and the analyses of genomic data suggests that the same strain from 1998 was responsible for the 2008 outbreak and it was able to survive in the facility. This highlights the persistence of Salmonella and that it can survive in dry food production environments for years. DATA AVAILABILITY STATEMENT GenBank accession numbers for all new sequences are listed in Table 1. AUTHOR CONTRIBUTIONS All authors played an integral part of project conception. Each author has read and approved the final version of the manuscript. Specifically, MH, MA, EB, and JP conceived and designed the experiments. MH performed the experiments. JP, MH, JM, and DM analyzed the data. MH and JP wrote the manuscript. FUNDING This project was supported by the internal FDA/CFSAN research funding. The effects of maturity matched and un-matched opposition on physical performance and spatial exploration behavior during youth basketball matches The aim of this study was analyze the effect of playing against biological matched and un-matched opposition, on physical performance and spatial exploration behavior of youth basketball players. Thirty under-14 to 16 basketball players were assigned to different teams according to maturity status (Pre-, Mid-, and Post-Peak Height Velocity [PHV]), and participated in basketball matches against matched (same maturity status), and un-matched (different maturity status) opposition. Maturity status was estimated considering the percentage of predicted adult height. Workload data was collected via inertial devices (IMUs) and Ultra-Wide Band (UWB)-based system. Heart rate was recorded with individual HR monitors. The Pre-PHV performed significantly more accelerations and decelerations and explored more space against matched opposition. Against un-matched opposition, the Pre-PHV presented higher average speed, body impacts, and Player Load. Both Mid- and Post-PHV covered more distance against matched opposition than against Pre-PHV. Games against Pre-PHV involved lower distance covered, average speed, Player Load, and higher accelerations and decelerations, than against Mid- and Post-PHV. The Pre-PHV athletes performed a higher number of accelerations and decelerations comparing to the Mid and Post-PHV players. Also, a significant interaction effect (group x time) was found in distance covered, average speed, body impacts, and Player Load. The type of opposition influenced physical performance and spatial exploration behavior during basketball matches, particularly of less-mature players. Based on present findings, practitioners can select the most suitable game format, considering the physical, technical, tactical, and psychological development needs, individualizing training stimulus. Introduction Biological maturation refers to progress toward the adult or mature state and can be defined in terms of status, timing, and tempo [1]. Maturational status refers to the stage of maturation significantly higher number of lower body impacts (> 5g) and Player load [13] were observed in the bio-banded competition [13]. Nevertheless, further research is needed to extrapolate which variables, including external, and internal load, and individual exploration behavior, are interchangeable in the bio-banding approach. Considering the previous statements and the need to provide some recommendations to support coaches how to recognize differences in maturation level, and use the bio-banding approach to individualize training, this research study aimed to analyze the effect of playing against biological matched and un-matched opposition, on physical performance and spatial exploration behavior of youth basketball players. More specifically, we sought to examine how parameters of physical performances varied when athletes competed in game formats when they were matched or not matched by maturational status. Experimental design Using a descriptive design, all data were gathered over 2 sessions with a 48-h interval, involving matched and un-matched game formats. Due to equipment availability, only eighteen players (six per maturity band were randomly selected) were monitored. Players' movement demands were assessed using the Ultra-Wide Band (UWB)-based system (Realtrack Systems, Almeria, Spain). Physical performance tests were conducted one week before the commencement of the study, as described elsewhere [5,14] for better understanding of the influence of somatic maturation on physical parameters among sample subjects. Participants Eighteen under-14 and twelve under-16 national level basketball players (age = 13.45±1.22 years; height = 164.52±11.85 cm; body mass = 55.90±12.84 kg; percentage of their predicted adult height [% PAH] = 90.75±5.40%) were recruited from Portuguese Basketball Academy to participate in this study. All players averaged a total of six hours of basketball training (4 teamsessions/week, 90 minutes/session), and 1 competitive match (national level) per week. Written informed consent was obtained from all participants and their parents before this investigation. The present study was approved by the Ethics Committee of the University of Trás-os-Montes and Alto Douro and conformed to the recommendations of the Declaration of Helsinki. Somatic maturation Height was recorded using a commercial portable stadiometer (Tanita BF-522W, Japan, nearest 0.1 cm). Body mass was estimated using the scales (Tanita BF-522W, Japan, nearest 0.1 kg). All measurements were taken following the guidelines outlined by the International Society for the Advancement of Kinanthropometry (ISAK) by the same researcher, who holds an ISAK Level 1 accreditation. Players' height, weight, chronological age, and mid-parent height were used to predict the adult height of each player [15]. The height of the biological parents of each player were self-reported and adjusted for over-estimation using the previously established equations [16]. The current height of each player was then expressed as a percentage of their predicted adult height (% PAH), which can then be used as an index of somatic maturation [17]. Players were grouped into three maturity bands based on the percentage of predicted adult height attained at the time of the tournament [2]: <88% (Pre-PHV), 88-95% (Mid-PHV) and >95% (Post-PHV) of predicted adult stature (Table 1). Only for descriptive reasons, maturity timing was estimated for each player based on z-scores: average or on-time (z-score between +0.5 and -0.5), early (z-score >+0.5), and late (z-score <-0.5) [9]. Game formats Subjects were included in one of two teams per maturity band. Participants competed in maturity matched (i.e., bio-banded) and non-matched (i.e., different bio-bands) formats against external opposition that had also been grouped in three equivalent maturity bands (Fig 1). Each team competed twice against each maturity band. In total, each participant completed a total of 6 games (Fig 1). Games were played in 5 vs. 5 formats, with 8-min duration, and conformed to standard officiating and rule procedures. Matches were played on a full-sized standard basketball court (28 × 15 m), with full-sized baskets (3.05 m), with a size 7-ball. Games were played without substitutions, verbal encouragement and technical or tactical instructions (including time-outs). Subjects were instructed to stay at the basketball field during the game [18]. Moreover, the main researcher provided new balls when necessary to ensure the quick restart of game situations [19], and called any foul or rules violation. Physical and tactical data The physical and tactical data was collected using inertial devices (IMUs) and Ultra-Wide Band (UWB)-based system placed in indoor basketball court (WIMU PRO 1 ; Realtrack Systems, Almeria, Spain). IMUs integrates multiple sensors registering at different sample frequencies [20]. For example, sampling frequency for a 3-axis accelerometer, gyroscope, and magnetometer was 100 Hz, 120 kPa for the barometer, and 18 Hz for the positioning system [20]. [22]. The UWB system was composed of 6 antennae placed in a hexagon around the indoor basketball court [23,24]. The UWB system uses triangulations between the antennas and the inertial units, six antennas send a signal to the units every 50 milliseconds [24]. Then, the device calculates the time required to receive the signal and derives the unit position (coordinates X and Y), using one of the antennas as a reference [24]. The use of inertial devices and calibration process occurred as described elsewhere [23]. . Also, average and peak speed (km�h -1 ) peak acceleration (PAcc; m�s -2 ) and deceleration (PDec; m�s -2 ) were calculated [20]. Speed zones during the basketball game were adapted from a previous research [24]. The > 2 m�s -2 and > -2 m�s -2 was the criteria to detect high-intensity accelerations and

decelerations, respectively [25]. Body impacts correspond to the number of jumps and impacts that exceed 5G forces, measured with the inertial accelerometer in the z, x, and y axes [26]. The instantaneous Player load (PLn) was computed using the following formula: PLn ¼ ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi . The accumulated Player Load (PL) ðPL ¼ P m n¼0 ðPLnÞ x 0; 01Þ was computed and calculated per minute for further analysis. The heart rate (HR) data were recorded continuously with individual HR monitors (Garmin, Soft Strap Premium, USA). Average and participants' peak HR (HR peak), taken as the highest HR recorded throughout the testing period [27] were recorded. The HR peak zones were defined as: zone 1 (50-60%); zone 2 (60-70%); zone 3 (70-80%); zone 4 (80-90%), and zone 5 (90-100%). The Edward's training impulses (TRIMP) was calculated based on the following formula and presented in arbitrary unit (a.u.) [28]: TRIMP = (time spent in zone 1 � 1) + (time spent in zone 2 � 2) + (time spent in zone 3 � 3) + (time spent in zone 4 � 4) + (time spent in zone 5 � 5). Sample Entropy (SampEn) was also used to assess each player's HR regularity during the games. SampEn (m,r,n) is defined as the negative natural logarithm of the conditional probability that two sequences, similar form points (length of the vector to be compared), remain similar at the next point m + 1 [29] The values used to calculate SampEn were 2 to vector length (m) and 0.2 ±SD to the tolerance (r) [29]. Values of SampEn range from zero towards infinity, where values close to zero were indicative of higher regularity in HR, while the higher the SampEn, the more unpredictable the HR. All data were analyzed using the commercially available software (WIMU SPRO Software; Realtrack Systems SL). Spatial exploration index (SEI) was obtained for each player by calculating his mean pitch position, computing the distance from each positioning time-series to the mean position and, finally, computing the mean value from all the obtained distances [30]. Statistical analyses Data are presented as mean ± SD. All data was found to be normally distributed using the Shapiro-Wilk test. Univariate analysis of variance (ANOVA) test and Tukey's post-hoc test were used in conjunction to examine the differences between groups (Pre-, Mid-, and Post-PHV), in all parameters. Effect sizes (ES) of the differences between and within-groups were evaluated using the Cohen's "d". Also, repeated-measures ANOVA was used to analyze within-group changes. A 3x3 repeated-measures ANOVA was performed on the absolute values of all parameters to determine the main effects between groups (Pre-, Mid-, and Post-PHV) and game formats (Pre-, Mid-, and Post-PHV). Sphericity of the data was checked by Mauchly's statistic, and where violated, Greenhouse-Geiser adjustment was applied. The level of statistical significance was set at p � 0.05. All the statistical analyses were performed using SPSS software (version 24 for Windows; SPSS Inc., Chicago, IL, USA). All graphs were constructed in computing environment R (Version 1.2.1335, RStudio, 2019), using ggplot2 package [31]. Inferential analysis also confirmed a significant main effect of time in DC, DC in SZ2, Acc, Dec, AS, PS, PL, and SEI (all p < 0.01). Games involving Pre-PHV involved lower DC (Mid and Post-PHV), DC in SZ2 (Mid and Post-PHV), AS (Mid and Post-PHV), BI (Post-PHV), and PL (Mid and Post-PHV), but also higher Acc and Dec (Mid and Post-PHV), PS (Post-PHV), and SEI (Mid-PHV). Games involving Mid-PHV involved lower DC in SZ2, but also higher PS comparing to those involving Post-PHV. A significant effect of group was found in Acc, Dec, BI, and PL (all p < 0.05). The Pre-PHV athletes performed more Acc and Dec comparing to the Mid and Post-PHV players. Moreover, performed higher values of HIDec, BI, PL than Mid-PHV. Also, a significant interaction effect (group x time) was found in DC, DC in SZ2, AS, BI, and PL (all p < 0.05). Discussion The purpose of this investigation was to examine the impact of maturity matching upon physical performance and spatial exploration behavior in youth basketball players. As expected, the degree to which players were matched or not matched, influenced the tactical complexity and physical demands of the games. The impact of maturity matching was most pronounced in less mature (i.e., Pre-PHV) players who appeared to benefit the most from competing with maturity matched peers. In the present study, the internal load remained constant irrespective of game formats. Pre-, Mid-, and Post-PHV players present the physical effort equal regardless of whether they were competing against maturity matched or unmatched opposition. These findings are consistent with previous study which revealed similar internal load (maximum and HR av ), irrespective of the competition format (chronological vs bio-banding) [13]. The absence of differences in the internal load may reflect limited impact of maturation status upon aerobic capacity. For example, laboratory studies demonstrated no significant main effect of PLOS ONE Biological maturation and youth basketball matches maturation status (Pre-, Mid-, and Post-Pubertal) in HR response during submaximal cycling exercise 30 to 70% of peak oxygen uptake [32]. The games involving the Pre-PHV teams (Pre-vs Pre-PHV; Pre-vs Mid-PHV; and Pre-PHV vs. Post-PHV) were characterized by a decrease in total distance covered. That is, all players covered less distances when competing in games that involved Pre-PHV players. The lower distances covered by Mid-and Post-PHV players against Pre-PHV players may result from greater full-court pressure which generated more opportunities to tackle and steal the ball, and perform more effective shots closer to the basket [33,34]. Furthermore, in maturity mismatched competition, steals are extensively performed by heavier, taller, and more mature players [3,4]. Thus, when competing against their less mature peers, the more mature players (i.e., Post-PHV) in the current study, may have used their physical advantage to score more points, without having to run such long distances. Afterwards, it is difficult to explain lower distance in Pre-PHV matched game based on previous arguments, and more studies are need for better understanding. The results of analysis pertaining spatial exploration were of particular interest. More specifically, Pre-PHV players presented higher mean values for individual spatial exploration behavior (i.e., SEI) when competing in maturity matched games. This indicates that Pre-PHV players adjusted their tactical behaviors according to the maturational status of the opponents. The SEI has been shown to be highly dependent on environmental constraints such as the amount space available to play [30]. The physical and functional advantages associated with variance in maturity may equally serve as important task-based constraint, impacting spatial exploration in youth basketball [35]. According to previous findings, players tend to explore more space when they play without restriction in space occupation [30], but also against lower anthropometrical disadvantage [35]. When Pre-PHV players played against matched opposition probably feel freer to move through the space [10]. It seems likely that anthropometrical differences could generate a distinct perception of free space available, underpinning different individual exploration behavior. The concern about the anthropometric aspects, the pitch size, and the spatial awareness has been addressed in previous research exploring bio-banding in youth football [9]. In all game formats, from 6-a-side to 11-a-side, less mature players (80-85% PAH) play in smaller pitch sizes comparing to more mature players (91-95% PAH) [9]. This adjustment reflects one advance comparing with previous bio-banding experience where players played in standard pitch irrespective of their biological maturation [10]. Thus, future research could examine how the variation of available space and anthropometric aspects may affect spatiotemporal interpretation in youth athletes. Moreover, the game involving matched opposition imposes longer offensive sequences, a greater number of passes [12], and more displacements in lateral and longitudinal directions [36] to find a solution to complete the offensive phase. In contrast, Pre-PHV against non-matched opposition significantly explored less space, in line with previous findings [35]. Against un-matched maturity opposition, players need to execute faster than they normally would [10], and they had less time for better interpretation of spatiotemporal information [35]. However, against matched opposition, there is greater physical equity [1,9,10,12] and therefore the differences in size and strength are likely to have a greater impact on performance, and less mature players seem less reluctant to perform their ball or non-ball skills. Player Load and the body impacts are frequently used to quantify the external training load during basketball training and matches [13,20,28]. According to the present findings, the games involving the Pre-PHV resulted in lower values of Player Load and body impacts. Although in youth basketball more mature players are frequently involved in actions that involve physicality [14], when competing against their less mature peers, they play less physical and competitive intensity. Matched groups increase the opportunities to work with right balance between competence and challenge (i.e. zone of proximal development), which is beneficial for psychological development [37]. In this regard, playing against matched opposition resulting in higher level of enjoyment and pleasantly, but also higher physical and technical challenge [9]. However, if more mature players are given a decreased challenge (e.g., players with decreased physical attributes) then they will be less likely to try hard, which may result in lower values of Player Load and body impacts. Considering the lower values of Player Load and body impacts, playing against un-matched opposition may be especially useful when players are going through developmental phases in which they may be more susceptible to injury risk (i.e., the pubertal growth spurt) [11], because it imposes lower external demands. Nevertheless, Pre-PHV faced higher inter-game variability and absolute values (against Post-PHV) in terms of PL and BI, resulting in a higher neuromuscular load experience [38]. Thus, play against matched opposition may reduce match physicality, and to be less likely to induce contact injuries [9]. Notwithstanding, previous studies examining the process of development of expert basketball players revealed competing against older players as a key aspect to be more likely to become expert [39,40]. That said, playing against older or more mature players facing bigger challenges can also be a development opportunity to consider if players are robust enough to tolerate physical differences, including reduced risk of injury. Based on these findings, practitioners should select the most suitable game format, considering the individual physical, technical, tactical, and psychological development needs. Limitations of the present study must be acknowledged. First, the participants in the current sample are slightly shorter and lighter when compared to youth basketball players of an equivalent age in other studies [3,4,6]. The degree to which maturity matching may impact physical and technical performance may vary relative to the nature of the sample. Maturity selection biases, for example, tend to be greater in more elite samples [3,4]. Second, it is important to recognize that the sample size in the current study was small and that future studies should look to consider the impact of maturity matching on a large sample [9,10,12]. Similarly, it is important to note that each athlete only completed short-duration bio-banded game formats. The impact of maturity matching may vary relative to game duration and future studies could consider playing longer matches. That said, given the current lack of knowledge regarding the effects of bio-banding in youth basketball, this study provides important starting point from which to discuss the matter and consider future research studies. Further research with a larger sample, longer game durations, players of different anthropometric attributes (height, body weight, and wingspan) should be conducted before the applicability of the current results can be generalized. Finally, it would be interesting to analyze other performance variables, such as collective tactical behavior (e.g., team synchronization), biochemical markers (e.g., creatine kinase, and testosterone), and coaches and players perceptions of unmatched competition. Present strategy offers different challenges, and changes on environment, provide distinct physical demands and individual spatial exploration behavior according to the game format (i.e., maturity matched and un-matched). These findings may be especially useful when youth players are competing through chronological age-based competitions in which they may be more susceptible to maturation bias, particularly the less mature. Thus, our findings may help practitioners in understanding

the real and potential values of talented young players during basketball training and competitions, improving the accuracy of the talent identification and recruitment processes. Furthermore, manipulate the type of opposition could be applied as a suitable strategy to individualize the prescription of training and competitive situations, considering physical and tactical development needs, and consequently encouraging the longterm developmental perspective. Moreover, this practice may be especially useful when players are going through developmental phases in which they may be more susceptible to injury risk (i.e., the pubertal growth spurt). Finally, this approach is a grouping strategy that has proposed to help counter some of the challenges presented by individual differences in growth and maturation, which not replace chronological-age competition. The synergistic compatibility mechanisms of fuzi against chronic heart failure in animals: A systematic review and meta-analysis Background: Fuzi’s compatibilities with other medicines are effective treatments for chronic heart failure. Pre-clinical animal experiments have indicated many possible synergistic compatibility mechanisms of it, but the results were not reliable and reproducible enough. Therefore, we performed this systematic review and meta-analysis of pre-clinical animal studies to integrate evidence, conducted both qualitative and quantitative evaluations of the compatibility and summarized potential synergistic mechanisms. Method: An exhaustive search was conducted for potentially relevant studies in nine online databases. The selection criteria were based on the Participants, Interventions, Control, Outcomes, and Study designs strategy. The SYRCLE risk of bias tool for animal trials was used to perform the methodological quality assessment. RevMan V.5.3 and STATA/SE 15.1 were used to perform the meta-analysis following the Cochrane Handbook for Systematic Reviews of Interventions. Result: 24 studies were included in the systematic review and meta-analysis. 12 outcomes were evaluated in the meta-analysis, including BNP, HR, HWI, ALD, LVEDP, LVSP, EF, FS, +dP/dtmax, −dP/dtmax, TNF-α and the activity of Na + -K + -ATPase. Subgroup analyses were performed depending on the modeling methods and duration. Conclusion: The synergistic Fuzi compatibility therapeutic effects against CHF animals were superior to those of Fuzi alone, as shown by improvements in cardiac function, resistance to ventricular remodeling and cardiac damage, regulation of myocardial energy metabolism disorder and RAAS, alleviation of inflammation, the metabolic process in vivo, and inhibition of cardiomyocyte apoptosis. Variations in CHF modeling methods and medication duration brought out possible model–effect and time-effect relationships. Introduction Heart failure (HF) is a global condition that affects approximately 38 million patients worldwide with increasing prevalence and incidence as a population ages (Braunwald, 2015). HF is not caused by a single factor but by a complex terminal-stage syndrome induced by various cardiovascular diseases and characterized by typical symptoms (e.g., breathlessness, ankle swelling, and fatigue) that may be accompanied by signs (e.g., elevated jugular venous pressure, pulmonary crackles, and peripheral edema) caused by a structural or functional cardiac abnormality, resulting in reduced cardiac output and elevated intracardiac pressures (Ponikowski et al., 2016). Depending on the time of onset, it is classified as acute and chronic HF (Pagliaro et al., 2020). Chronic heart failure (CHF) is one of the most common chronic conditions worldwide (Ewen et al., 2016), with an estimated prevalence of 26 million patients worldwide (Ponikowski et al., 2014). Pathogenic causes of CHF mainly include single or combined factors, such as reduced left ventricular myocardial function or dysfunction of the pericardium, endocardium, heart valves, great vessels, and the right ventricular myocardium. Related pathogenic mechanisms leading to CHF involve ventricular remodeling, increased hemodynamic overload, abnormal cardiac myocyte calcium circulation, ischemia-related dysfunction, excessive neurohumoral stimulation, accelerated apoptosis, and genetic mutations (Dassanayaka and Jones, 2015). The health expenditures of CHF are considerable and will rise dramatically with an ageing population. Although significant advances and progress have been made in therapies and prevention, the 5-years mortality rate of CHF remains at 50%, which is the same as that of a malignant tumor (O'Gara et al., 2013), and the quality of life and prognosis remains poor (Savarese and Lund, 2017), which impose a huge burden on socioeconomics and medical resources (Lloyd-Jones et al., 2009). The development of more effective and promising treatment is of great importance and urgency. In recent years, traditional Chinese medicine (TCM) has made great progress in treating CHF and plays a dominant role in numerous treatments, showing superior and unique efficacy in controlling symptoms, enhancing therapeutic effects, and improving prognosis, which has made TCM a popular treatment of CHF. According to the basic theory of Chinese medicine, CHF belongs to the categories of palpitation, chest discomfort, edema, and heart obstruction, and the causes and underlying mechanisms are defined as qi deficiency and yang deficiency, whose corresponding treatment is tonifying qi and warming yang. Statistically, the use of this treatment approach ranked first at 64.3% among all reported treatments of CHF (Xu, 2016). Aconiti Lateralis Radix Praeparata (Fuzi in Chinese), the lateral root of Aconitum carmichaelii Debx., is a time-honored interior-warming TCM used in CHF that is usually processed into salty Fuzi, hei-shun-pian, or bai-fu-pian. The main bioactive components of Fuzi are diester-diterpene alkaloids (DDAs), which mainly include aconitine (AC), mesaconitine, and hypaconitine (HA). Fuzi is praised as one of the most effective TCMs for revitalizing yang and rescuing patients from collapse, indicating its pharmacological cardiotonic efficacy by elevating blood pressure, causing vasodilation, increasing blood flow volume, enhancing hypoxia tolerance and cold resistance, and providing anti-shock, antiarrhythmic, and anti-myocardial ischemic activities. Powerful as Fuzi presents, attention should also be paid to its toxicity. Aconitine, the main bioactive part of Fuzi, serves as dual effects of toxicity and efficacy. For humans, the estimated lethal dose is 2 mg of pure aconitine, 5 ml of aconite tincture, and 1 g of the wild plant (Singh et al., 1986). In addition, it is especially known for its cardiotoxicity, with electrocardiograms presenting a transient decrease in heart rate, atrial and ventricular extrasystole, and tachycardia as well as non-paroxysmal ventricular tachycardia and ventricular fibrillation. Therefore, Fuzi is only used clinically after proper processing and always combined with other TCMs to reduce its toxic alkaloid content, thus enhancing the curative effects and alleviating adverse reactions. The compatibility of TCMs is the fundamental principle of clinical Chinese medical administration, presenting the characteristics and edges of TCM. With proper compatibility, the therapeutic effects can be enhanced and the toxic and side effects decreased, widening clinical applications, regulating interactions, and sometimes resulting in new therapeutic effects. Compatibility refers to purposeful, organized, and well-founded combinations between TCMs based on the rule of Qi Qing. Literally, Qi means the number seven, and Qing represents emotions. Chinese ancient medical scholars summarized that there were seven kinds of compatibility relationships between TCMs: single, mutual reinforcement, mutual assistance, mutual restraint, mutual suppression, mutual inhibition, and antagonism. A TCM prescription is based on the compatibility of the ingredients. According to the contribution of each medicine to the final effect of a prescription, different roles are determined as sovereign, minister, assistant, and courier. Clinically, Fuzi is the most frequently used yang-warming medicine among TCMs at 80.3% (Chan, 2009), and it is most compatible with dried ginger (Zingiber officinale Rosc.) according to the rule of mutual reinforcement and is usually used in the treatment of yang collapse (e.g., Sini Decoction (SND)). The main bioactive component of dried ginger is gingerol, which is capable of resisting and reducing the toxicity of aconitine. Peng (2013) found that the compatibility of aconite and ginger can promote the dissolution of aconitine alkaloids, inhibit the transport of DDAs to the heart and their absorption in the body while promoting the absorption of monoester alkaloids, including benzoylaconine (BAC), benzoylhypaconine (BHA), and benzoylmesaconine (BMA), delaying their elimination in vivo. Ye et al. (2013) compared the content of curative alkaloids of dried ginger, fresh ginger, and sand-fried ginger after compatibility with Fuzi by referencing a high-performance liquid chromatography atlas and found that the content of curative alkaloids in the dried ginger's compatibility group was higher than that of a simple Fuzi decoction, proving that dried ginger, when used with Fuzi, can increase the therapeutic effect. In addition, when Fuzi is combined with ginseng (Panax ginseng C.A. Mey.), mutual assistance makes it effective compatibility for both yang and qi collapse (e.g., Shenfu Decoction (SFD)). Pharmacology research (Cui et al., 2013) has found that ginsenoside, the main active ingredient of ginseng, can improve myocardial metabolism, protect the myocardial structure, and enhance cardiac contractility. Li et al. (2011) found that the water decoction of Fuzi in combination with ginseng at 1:2 enhanced the heartstrengthening effect by significantly improving the rats' hemodynamic indexes relative to the effect of Fuzi water decoction alone. Another study (Li et al., 2013) found that the compatibility of Fuzi and ginseng also improved the myocardial diastolic and contraction function and changed the levels of N-terminal pro-brain natriuretic peptide, angiotensin II (Ang II), and atrial natriuretic peptide in the plasma of rats with acute heart failure, achieving the effects of alleviating HF. In recent years, the clinical application of Fuzi against CHF is becoming more popular, along with which, however, are increasing aconite poisoning and inefficacious reports due to inappropriate processing, compatibility, or usage (Chan, 2011;Chen et al., 2012). Currently, prominent progress has been made in Fuzi's chemical components, pharmacological effects, attenuated processing, and attenuated compatibility (Wei et al., 2019;Yang et al., 2019). However, a lack of further research on the mechanisms of synergistic compatibility of Fuzi against CHF remains. Although multiple possible mechanisms have been demonstrated through animal experiments, small sample size effects and individual differences still exist, making it hard to draw a specific and reliable conclusion. For a long time, quality requirements in clinical research are normally higher than in animal experiments, hence the derivative is a relatively comprehensive way of quality control and assessment of clinical interventions, resulting in, however, a poor human clinical and toxicological utility of animal experiments (Knight, 2007). A systematic review is conducive to assembling overall evidence, enlarging sample size, and lowering various risks of bias, thus systematic evaluations of high-qualified randomized controlled trials (RCTs) are acknowledged to be one the most reliable pieces of evidence when the efficacy of interventions needs proving. For the past few years, facing the fact that new drug test results on animals often failed to apply to clinical research (Van Norman, 2019;Van Norman, 2020), reminds people of the necessity of quality assessment of animal experiments. Ferreira et al. (Ferreira et al., 2020) addressed the importance of levelling the translational gap for the animal to human efficacy data and pointed out that selecting optimal animal models of disease may prevent the conducting of clinical trials, based on unreliable preclinical data. Without a doubt, a systematic review of animal experiments is of importance and beneficial to avoiding animal resource wasting, lowering experimental costs and preventing patients from unknown risks (van Meer et al., 2015). All things considered, our study was designed to conduct a systematic review and meta-analysis of pre-clinical animal studies to integrate evidence from all relevant studies, conduct both qualitative and quantitative evaluations of the compatibility and summarize all potential synergistic mechanisms of compatibility, through which the reliability and accuracy can be increased. The purpose of this study was to 1) assess the efficacy of Fuzi compatibility with any other TCMs or their bioactive constituents against CHF by reviewing all related animal studies, 2) identify provable evidence for synergistic compatibility mechanisms of Fuzi for CHF, 3) evaluate the impact of possible publication bias and small-study effect, 4) explore the factors that may affect the efficacy of compatibility, and 5) hopefully serve as a reference for further clinical trials and applications. Method This study was performed following the Cochrane Handbook for Systematic Reviews of Interventions (Higgins et al., 2019). The protocol for this meta-analysis is available in PROSPERO (CRD42021290955). databases whereas the corresponding terms were used in Chinese databases. In addition, possibly related studies (e.g., conference literature) were also carefully identified through manual searching. The full retrieval details are attached in Supplementary Material, Supplementary Appendix S1. Selection criteria The selection criteria were based on the Participants, Interventions, Control, Outcomes, and Study designs (PICOS) strategy. Exclusion criteria 1) Participants: models other than CHF, in vitro studies, or clinical trials in humans. 2) Interventions: administration combined with other non-TCM treatments (e.g., electroacupuncture) or without compatibility. 3) Control: administration combined with other TCM or treatments. 4) Study type: case reports, reviews, conference proceedings, catalogue, and technical achievement reports. 5) Duplicate publication. 6)

Studies without full text. Study screening All retrieved articles were imported into EndNote 20, and duplicate studies were removed first. Then, two reviewers independently performed primary screening based on titles and abstracts, and studies inconsistent with the selection criteria were excluded directly. Next, a full-text screening for eligible articles was performed, and two reviewers recorded specific exclusion reasons independently. Finally, a cross-check of results between the two reviewers was performed to ensure the consistency of the screening. During this procedure, any disagreements between the two researchers were discussed with a third reviewer, and the final judgment was made. Data extraction A form for data collection was designed and established in Microsoft Excel 2021, and two reviewers extracted the following items from the included papers independently: 1) Study ID: first author's name and publication year. 2) Participant's information: number, sex, species, weight, and CHF model. All outcome measures were analyzed as continuous variables, estimated mean and standard deviation values were extracted, and each variable was extracted from the last time point. For trials with more than one experimental group but sharing a common control group, the control group was divided into multiple groups with its number in accordance with experimental groups, and these added comparisons were also incorporated into the meta-analysis (Higgins et al., 2019). An email of the request was sent to authors of related studies for more information when the details of studies or data about outcomes were missing or only presented in graphs. If relevant data were not available or insufficient for quantitative analysis, qualitative analysis was performed. Quality assessment Two researchers conducted quality assessments independently, and the SYRCLE risk of bias tool for animal trials was adopted to perform the methodological quality assessment (Hooijmans et al., 2014), which consisted of the following aspects: 1) Selection bias: sequence generation, baseline characteristics, and allocation concealment. 2) Performance bias: random housing and blinding of caregivers and/or investigators. 3) Detection bias: random outcome assessment and blinding of the outcome assessor. 4) Attribution bias: incomplete outcome data. 5) Reporting bias: selective outcome reporting. 6) Other sources of bias. Frontiers in Pharmacology frontiersin.org 04 The bias grade of each field was classified as high, low, or unclear risk. Any disagreements were discussed and resolved with a third assessor. Statistical analysis Following the Cochrane Handbook for Systematic Reviews of Interventions (Higgins et al., 2019), the overall effects of all continuous variables were reported by weighted mean difference (MD) with the 95% confidence interval (CI), and p values <0.05 were defined as indicative of statistical significance. The heterogeneity of multiple results was tested by performing the Chi-square (Chi 2 ) and I-square (I 2 ) tests. For the Chi 2 test, the level of α = 0.1 was considered to be indicative of statistical significance, and for the I 2 test, the level was I 2 = 50%. More specifically, a fixed-effects model was used for I 2 < 50%; otherwise, a random-effects model was applied. To explore potential sources of heterogeneity and how they affected the pooled effects, subgroup analyses were performed depending on . For studies with medium or high heterogeneity, sensitivity analyses were organized to investigate the influence of each study by omitting studies one by one when there were adequate studies. Publication bias and small-study effects were assessed by funnel plots and Egger's test on the condition that ≥10 trials were included for an outcome. The results of Egger's test were considered to be associated with considerable publication bias or small-study effects when p values were <0.05 (Egger et al., 1997). This meta-analysis was performed in RevMan V.5.3 software, whereas subgroup analysis, sensitivity analysis, and the evaluation of publication bias were performed in STATA/SE 15.1 software. Study characteristics Twenty-three studies involving 686 animals, with 445 in the intervention group and 241 in the control group, were included in the systematic review and meta-analysis, and 1 study did not report the number of participating animals. Twenty-two trials chose rats as the experimental subjects, whereas two chose mice, mostly males. The weights of the rats ranged from 160 to 300 g, but 1 study used mice weighing from 20 to 24 g. For modeling methods, drug inducement and surgery were the major choices, and DOX-inducement was the most frequently used (13 of 24 studies, 54.2%), with LAD and TAC used less frequently (4 of 24 studies, 16.7% each). The use of AAC was mentioned in only three (12.4%) studies, which was the least. Interventions involved in all studies were mainly Fuzi or its bioactive constituents, including AC, HA, higenamine (HG), hei-shunpian, and total alkaloids (TA) that were compatible with ginseng, red ginseng, shan-zhu-yu (Cornus officinalis Sieb. et Zucc.), banxia (Pinellia ternate (Thunb.) Breit.), fuling (Poria cocos), gancao (Glycyrrhizae Radix et Rhizoma) or its bioactive constituents, including glycyrrhetinic acid (GA) and liquiritin, dried ginger (Zingiberis Rhizoma) or its bioactive constituents, including [6]-gingerol ([6]-GR) and SND or its components, including total alkaloids and total gingerols (TAG), total alkaloids, flavones, and saponins (TAFS), and total alkaloids, total gingerols, total flavones, and saponins (TAGFS). In contrast, the administrations in the control groups were only of Fuzi or its bioactive constituents. Moreover, the duration of each study was noted and divided into short-or long-duration groups based on the average duration (21 days), with 10 studies in the short group, 13 in the long, and 1 without duration reported. The outcome indexes included in each study and all of the above characteristics of the studies are presented in Supplementary Material, Supplementary Table S1. Study quality The results of the quality assessment are presented in Supplementary Material, Supplementary Table S2. 1) Selection bias: random allocation to the control group and intervention group were mentioned in all included studies, of which three studies described the detailed method of random sequence generation, referring to a random number table. The animals' baseline characteristics (sex, age, and weight) of each trial were similar in both the control and intervention groups, and the timing of CHF induction was also adequate. None of the studies mentioned allocation concealment. As a result, three in the 24 studies were identified as having low risks of bias in sequence generation, and all studies were assessed to have low risks of bias in the baseline characteristics and unclear risks of bias in allocation concealment. 2) Performance bias: none of the studies reported if the animals were randomly housed during the experiment or not or if the caregivers and investigators were blinded, so the assessment of random housing and blinding of this item were identified as unclear risks of bias in all studies. 3) Detection bias: none of the studies described randomly picking an animal for outcome assessment or using a random component in the sequence generation for outcome assessment as well as the blinding of outcome assessors, which indicated unclear risks of bias in all studies in random outcome assessments and blinding of this aspect. 4) Attribution bias: all animals were included in the analysis, and no outcome data were missing from any of the studies, so for Frontiers in Pharmacology frontiersin.org incomplete outcome data, low risks of biased data were judged for all studies. 5) Reporting bias: it was clear that all published reports included all expected outcomes and comparative methods, and results sections, suggesting low risks of bias in selective outcome reporting. 6) Other sources of bias: for this aspect, funding was the main factor considered, and 19 studies reported that their studies were free of the inappropriate influence of funders, whereas 5 did not, resulting in 19 studies at low risks of bias in funding and 5 at unclear risks of bias. Effectiveness A total of 12 outcome indexes were identified as sufficient for meta-analysis, including BNP, HR, HWI, ALD, LVEDP, LVSP, EF, FS, +dP/dt max , −dP/dt max , TNF-α, and the activity of Na + -K + -ATPase. Meta-analysis was not performed for the other outcomes due to insufficient data. BNP As shown in Figure 2, a total of 14 studies were included to assess the general compatibility effect of Fuzi on BNP. Statistical tests revealed a significant difference between the two groups (MD = −42.96, 95%CI [−69.45, −16.48], p = 0.001), indicating a significant effect on decreasing the level of BNP in the compatible group compared with the control group, but with a high heterogeneity (Chi 2 = 196.38, p < 0.00001; I 2 = 93%). Figure 3 shows the results obtained from the preliminary analysis of HR. Fourteen studies reported the compatibility effect of Fuzi on HR, and a significant difference was found in the compatibility group when compared with the control group (MD = 22.76,95%CI [4.16,41.37], p = 0.02), with a medium heterogeneity (Chi 2 = 35.75, p = 0.0006; I 2 = 64%), suggesting that the compatibility group increased the HR of the CHF patients. Figure 4 shows that nine studies were included in the analysis of the compatibility effect of Fuzi with other treatments on HWI altogether. A more valid correlation was found between compatibility and HWI drop compared with the control group (MD = −0.12, 95%CI [−0.14, −0.10], p < 0.00001), with a low heterogeneity (Chi 2 = 7.42, p = 0.49; I 2 = 0%). ALD The pooled results of the compatibility effect of Fuzi with other treatments on ALD are shown in Figure 5. Eight trials were included in this comparison. The results showed a slight decrease in ALD in the compatibility group but the difference between two groups was not significant (MD = −1.44, 95%CI [−5.97, 3.10], p = 0.53), with medium heterogeneity (Chi 2 = 14.75, p = 0.04; I 2 = 53%). Figures 6A,B shows that there were 16 and 17 studies included in the meta-analysis of the effect of Fuzi compatibility with other treatments on LVEDP and LVSP, respectively. Although the pooled result showed slight decreases in LVEDP, the difference between two groups was not significant (MD = −0.03, 95%CI [−3.44, 3.38], p = 0.99) and with high heterogeneity (Chi 2 = 236.74, p < 0.00001; I 2 = 94%). For LVSP, the overall effect test showed the significant superiority of the compatibility group in increasing LVSP versus the control group (MD = 8.54,95%CI [2.71,14.37], p = 0.004) with medium heterogeneity (Chi 2 = 27.31, p = 0.04; I 2 = 41%). EF and FS Figures 7A,B shows the summary statistics of EF and FS from five trials for each. It was apparent that the compatibility Frontiers in Pharmacology frontiersin.org FIGURE 3 Forest plot of HR. FIGURE 4 Forest plot of HWI. 3.4.7 + dP/dt max and −dP/dt max The overall effects of Fuzi compatibility on + dP/dt max and −dP/dt max are shown in Figures 8A,B, with 17 studies included for each. The pooled effect of Fuzi increased + dP/ dt max but decreased −dP/dt max relative to the control group. The compatibility group was able to increase both the maximum increase rate and decrease rate of left ventricular pressure significantly (+dP/dt max : MD = 769.81, 95%CI [144.08, 1395.53], p = 0.02; −dP/dt max : MD = −1130.54, 95%CI [−1705.66, −555.42], p = 0.0001). Both + dP/dt max and −dP/ dt max showed high heterogeneity (+dP/dt max : Chi 2 = 130.91, p < 0.00001; I 2 = 88%; −dP/dt max : Chi 2 = 101.25, p < 0.00001; I 2 = 84%). Figure 9 shows the results of the pooled Fuzi compatibility effect on TNF-α. Only two studies reported on this index with analyzable data, and the compatibility group showed a decrease in the level of TNF-α versus that in the control group, but the difference was not 3.4.9 Na + -K+-ATPase Figure 10 shows that three studies were included in the analysis of the Fuzi compatibility effect on the activity of Na + -K + -ATPase. The compatibility group improved the activity of Na + -K + -ATPase significantly versus the control group (MD = 0.19, 95%CI [0.06, 0.33], p = 0.006), with high heterogeneity (Chi 2 = 12.11, p = 0.002; I 2 = 83%). Subgroup analysis Subgroup analyses were performed depending on two factors: the modeling methods of CHF and duration. For CHF modeling methods, four were mentioned, including DOX, TAC, AAC, and LAD, in keeping with the four kinds of subgroups. Duration, the average of which (21 days) was used to classify the subgroups

into short-(<21 days) and long-(≥21 days) duration subgroups. The outcome indexes of BNP, HR, HWI, ALD, LVEDP, LVSP, +dP/ dt max , and −dP/dt max were selected for subgroup analyses whereas the other outcomes were excluded due to insufficient studies. The results of subgroup analysis can be found in Supplementary Material, Supplementary Tables 3-10. BNP The subgroup analysis results for BNP are presented in Supplementary Table S3. Among the modeling methods, there were significant differences in the efficacy of Fuzi compatibility in the AAC (p = 0.000), TAC (p = 0.013), and LAD (p = 0.000) subgroups, but not in the DOX subgroup (p = 0.523), indicating that the outcome in DOX group was not credible though it presented to be positive. The efficacy of Fuzi compatibility was significant in the two duration subgroups (short: p = 0.013; long: p = 0.002). However, the two factors were not the source of its high heterogeneity. HR As shown in Supplementary Table S4, the subgroup analyses for HR found that there was a significant difference in the efficacy of Fuzi compatibility in the DOX group (p = 0.015) but not in the AAC group (p = 0.388) versus the control group, suggesting that the results of the experiment on HR in the AAC group were invalid. In addition, significant compatible efficacy was shown in the short-duration subgroup (p = 0.015) but not in the longduration subgroup (p = 0.388), which was proof of the suspect outcome of the experiment on HR in the long-duration group. Neither of the two factors was the source of heterogeneity. HWI The results of subgroup analyses on HWI are presented in Supplementary Table S5. For modeling methods, significant efficacy of compatibility was only shown in the TAC group (p = 0.000) and not in the DOX (p = 0.323) and AAC (p = Frontiers in Pharmacology frontiersin.org 0.185) groups, indicating that the results for HWI were not valid in the DOX and AAC groups. Moreover, duration did not have a significant effect on the pooled compatibility effect (short: p = 0.000; long: p = 0.045). A heterogeneity test was not performed for this outcome index due to its low heterogeneity. ALD For ALD, subgroup analysis was performed only for the modeling methods with sufficient studies on the duration (Supplementary Table S6). There was a significant difference in the Fuzi compatibility efficacy between the DOX group (p = 0.000) but not between the AAC group (p = 0.440) and the FIGURE 8 Forest plot of +dP/dt max and −dP/dt max . FIGURE 9 Forest plot of TNF-α Frontiers in Pharmacology frontiersin.org 11 control group, which indicated that the modeling method was a key factor affecting the overall results. Further, the heterogeneity results of ALD suggested that the modeling method was the source of its heterogeneity (). LVEDP and LVSP For LVEDP, no significant associations between the efficacy of compatibility and LVEDP in the modeling method subgroups (DOX: p = 0.332; AAC: p = 0.543) and durations (short: p = 0.312; long: p = 0.998) were found (Supplementary Table S7). The results of subgroup analyses on LVSP are shown in Supplementary Table S8. Depending on the modeling method, the efficacy of compatibility was significant in the DOX group (p = 0.000) but not in the AAC group (p = 0.599). The shortduration subgroup showed significant efficacy of compatibility (p = 0.001), which was not observed in the long-duration subgroup (p = 0.400). Interestingly, the pooled effects of the DOX and AAC subgroups in the subgroup analyses of both LVEDP and LVSP were opposite from each other, which was also observed for the short-and long-duration subgroups. Neither of the grouping bases led to LVEDP's high heterogeneity, whereas the modeling method was shown to be the source of LVSP heterogeneity (). +dP/dt max Supplementary Table S9 shows the results of subgroup analyses on the efficacy of compatibility on +dP/dt max . A significant was found for the DOX subgroup (p = 0.001) but not in the AAC subgroup (p = 0.868). Significant efficacy of compatibility was observed for the short-duration subgroup (p = 0.002) but not in the long-duration subgroup (p = 0.559). Although the efficacy results were opposite in the DOX and AAC subgroups, the two factors were not the source of high heterogeneity for +dP/dt max . 3.5.7 −dP/dt max Supplementary Table S10 shows the results of subgroup analyses on the efficacy of compatibility on −dP/dt max . There was a significant effect on efficacy in the DOX subgroup (p = 0.000) but not in the AAC subgroup (p = 0.289). Significant efficacy of compatibility was observed for the short-duration subgroup (p = 0.000), but not in the long-duration subgroup (p = 0.121). The test of heterogeneity showed that the two factors were not the source of high heterogeneity for −dP/dt max . Sensitivity analysis Sensitivity analyses were performed on BNP, HR, ALD, EF, FS, LVEDP, LVSP, +dP/dt max , −dP/dt max , and the activity of Na + -K + -ATPase by omitting studies one by one to investigate how each study influenced the pooled effects. No sensitivity analysis was performed on the other outcome indexes due to insufficient studies. After successively excluding each of the studies, the combined results of BNP, HR, ALD, LVEDP, LVSP, +dP/ dt max , and −dP/dt max did not change substantially, indicating that although their heterogeneity was high, the results were stable and credible. However, the sensitivity analyses on EF, FS, and Na + -K + -ATPase showed changes when certain studies were omitted, and the pooled effects varied from those for the primary analysis and were no longer statistically significant, which indicated that some of the included studies influenced the overall evaluation. Publication bias and small-study effect Publication bias and the small-study effect were evaluated through funnel plots and Egger's test on BNP, HR, LVEDP, LVSP, +dP/dt max , and −dP/dt max but not on the other outcome indexes because they had <10 included studies. Supplementary Figures F1-F6 Hemodynamics analyses are considered the most accurate among a series of cardiac function tests and an essential assessment of heart condition in rats (Khorrami et al., 2014). HR represents the number of times the heart ventricles contract per unit of time. For CHF rats, HR will rise due to conditioned reflexes to make up for heart-induced insufficient blood supply (Huff et al., 1999). LVSP and LVEDP mainly relate to myocardial contraction and diastole. The ±dP/dt max parameter evaluates the rate of left ventricular pressure and degree of myocardial diastole, which is susceptible to cardiac load, and also represents the ability of myocardial contraction and diastole. Hence, the above five outcome indexes are classic parameters for hemodynamic analyses, directly reflecting an animal's blood flow state. For animals in a CHF model, their LVSP and +dP/dt max will drop dramatically, whereas LVEDP and −dP/dt max will rise significantly, indicating a prominent decrease in myocardial contractility, left ventricular overfilling, and impaired diastolic function (Thibault et al., 2013). The present meta-analysis results showed that the synergistic effect of Fuzi compatibility with other reviewed study treatments increased HR, LVSP, and +dP/dt max as well as decreased −dP/dt max relative to the effect of Fuzi used alone. Although there was no significant difference in the pooled effect of LVEDP, some individual studies showed decreases. All things considered, the mechanism of Fuzi synergistic compatibility on hemodynamics is likely to be enhancement of myocardial contraction and diastolic ability, increased coronary blood flow, reduction of myocardial oxygen consumption, improvement of myocardial cellular energy metabolism, and alleviation of heart load, all of which would inhibit further deterioration of HF. Heart pumping function and cardiac contractility Echocardiography is one of the important techniques commonly used to detect heart function and can be used clinically to evaluate the severity of HF. Common indicators of cardiac structural and functional changes are EF and FS (Wan et al., 2017). FS refers to the left ventricular (LV) short-axis shortening rate, and its normal reference range is 25%-45% (average: 30%). The EF is more commonly used clinically. Left ventricular EF (LVEF) refers to stroke volume as a percentage of ventricular end-diastolic volume. When the ventricles contract, all of the blood in the ventricles cannot be injected into the arteries. In a normal adult resting state, the diastolic volume of the left ventricle is about 125 ml, and the stroke volume is 60-80 ml. Generally, 55%-75% is considered the normal range. When the human body is at rest, the EF is about 55%-65% and is related to myocardium contractility; specifically, stroke volume and EF increase with stronger myocardial contractility. Under normal circumstances, LVEF is ≥50%, and <50% is regarded as insufficiency of LV contractility. Clinically, CHF patients are classified as HF with reduced EF (HFrEF: EF < 40%), HF with mid-range EF (HFmrEF: EF 40%-49%), and HF with preserved EF (HFpEF: EF ≥ 50%) (Iwano and Little, 2013;Tschöpe et al., 2019). In this paper, the levels of both EF and FS decreased and returned to the relatively normal range in the Fuzi compatibility group but not in the Fuzi alone control group, suggesting that there were improvements in the cardiac pumping function and enhancement of cardiac contractility with Fuzi compatibility. Reverse cardiac remodeling to some extent was also implied. Improvements in ventricular remodeling and cardiac damage HF mainly damages the myocardium, including ventricular overload during diastole and systole, which changes the number of myocardial cells (myocardial infarction, diffuse myocarditis) that causes atrial and ventricular hypertrophy. Pathological cardiac hypertrophy is associated with cardiac insufficiency and is usually induced by many factors, such as prolonged and abnormal hemodynamic stress, activation of proinflammatory cytokines, and cellular dysfunction, and all these complicated responses cause cardiac remodeling changes and potentially CHF Shimizu and Minamino, 2016). According to the results of hematoxylin and eosin staining on the cardiac tissue of CHF rats, pathological changes were found that compared with the normal group, the HWI of the CHF group increased, with obvious myocardial hypertrophy, myocardial tissue infarction, myocardial tissue cell death area, myocardial layer thickening, and myocardial fibrous space widening (Xie et al., 2021). HWI refers to the relative value of heart weight and body weight, which reflects the degree of cardiac remodeling and hypertrophy. As an evaluation of heart function, its increase is often accompanied by obvious myocardial hypertrophy or myocardial tissue infarction (Xu, 2015). LDH is an important enzyme in the glycolysis pathway, widely presented in various tissues of the human body. The content of LDH in red blood cells is about 100 times that of normal serum and is abundant in heart and lung tissues. LDH reversibly catalyzes the oxidation reaction of lactic acid into pyruvic acid and catalyzes the oxidation of lactic acid to propionic acid while simultaneously transferring hydrogen to the coenzyme to become NADH. Due to LDH's wide distribution, its specificity for disease diagnosis is poor. However, because of the large difference in enzyme content between tissue and plasma, it has high sensitivity. When HF occurs, the activity Frontiers in Pharmacology frontiersin.org of LDH is hundreds of times higher in myocardial cells than in serum, so even the slightest damage to the myocardium causes a significant increase in serum LDH activity. CK is an important enzyme in human energy metabolism and is mainly distributed in skeletal muscle and myocardium, followed by brain tissue. The activity of CK increases within 3-8 h after myocardial injury, peaks at 12-36 h, and returns to the level in healthy people in 3-5 days (Li et al., 2001). Continuous increase in CK activity indicates more severe myocardial damage. LDH and CK are used as myocardial injury markers whose activity changes can be dynamically monitored to determine the degree of HF and observe treatment efficacy and predict prognosis. found that the levels of LDH and CK in mice were elevated after DOX inducement. However, when administered with SND and preparations containing TA, both the levels of LDH and CK were improved to normal ranges, which was not observed with TA alone, suggesting the compatibility synergistic effect. Wen et al. (Wen et al., 2019) measured the activities of myocardial damage markers, including LDH and CK, to determine whether heart damage had occurred as reflected in serum biomarkers, and the results showed that the marker levels increased significantly in the DOX-inducement model group but were reduced in

the compatibility group, which was the combination of Fuzi and dried ginger. This meta-analysis showed a significant decrease in the level of HWI in the compatibility group relative to that in the control group as reflected by the LDH and CK levels, suggesting that the addition of Fuzi improved the efficacy via synergistic mechanisms involved in reversing ventricular remodeling, resisting cardiac hypertrophy, and alleviating cardiac damage. Increase in the activity of Na + -K + -ATPase and Ca 2+ overload suppression The heart is one of the main energy-consuming organs in the human body, and energy metabolism is the material basis of myocardial cell activity, of which the tricarboxylic acid cycle is the main process. Under normal circumstances, mitochondria are the primary sites for myocardial energy production and storage. ATP is an important guarantee for maintaining cardiac contractile function and energy metabolism, 60%-90% of which comes from the oxidation of fatty acids, and the rest comes from carbohydrates (sugar and lactic acid) and a small amount of ketone oxidation (Stanley et al., 1997). The continuous and rhythmic contraction and relaxation of normal cardiomyocytes require a steady stream of ATP to provide energy for their activities. Disorders in the metabolism of fatty acids, glucose, lactic acid, and amino acids in myocardial cells during HF can cause changes in cardiac energy metabolism, leading to myocardial contraction-coupling disorder and progression of the ventricular remodeling (van Bilsen et al., 2004). Current research suggests (Chen Y. et al., 2014) that when HF occurs, the balance between oxygen supply and oxygen consumption is broken, and mitochondrial function is changed from the use of fatty acids to greater use of glucose, phosphorylation potential, as well as ATP production and utilization, are reduced, and the AMP and ADP contents increase. Consequently, the combination of ventricular remodeling accompanied by energy metabolic disorders is considered to be the main pathophysiological mechanism of CHF, which has been confirmed by the detection of highenergy phosphoric acid compounds in the CHF myocardium (Neubauer, 2007). Na + -K + -ATPase, also known as the sodium pump or sodium-potassium pump, is mainly used in energy supply, elimination of intracellular waste, and maintaining cell-pressure balance. Currently, studies have found that when the activity of the sodium-potassium pump on myocardial cells continues or severely decreases, it will cause severe dysfunction of the sodium pump, leading to abnormal myocardial energy metabolism, heart damage, or cardiac insufficiency (Huang, 2013). Na + -K + -ATPase is widely distributed on the cell membrane. In the pathophysiological process of CHF, the change in Ca 2+ regulation is the central link between various mechanisms. The protein level and activation degree of Na + -K + -ATPase in the myocardial cell membrane are the key factors of intracellular Ca 2+ homeostasis. Studies have shown that when CHF occurs, Na + -K + -ATPase activity decreases, intracellular Na + increases, and Na + -K + exchange increases, leading to mitochondrial dysfunction and reduced ATP production. Plasma and sarcoplasmic membranes that rely on energy Ca 2+ -ATPase cannot pump out or take in the excess Ca 2+ into the sarcoplasmic reticulum due to insufficient energy, causing calcium overload in myocardial cells (Zhu, 2012). Miao et al. (Miao et al., 2016) found that in the state of CHF, the content of ATP in rat myocardial cells was significantly reduced, whereas that of AMP and ADP was increased. After intervention with SND, a significant increase in the content of ATP was found in the myocardium of rats with CHF, indicating that improving myocardial cell energy metabolism may be one of the important curative mechanisms of SND and its different compatible prescriptions for myocardial protection. In addition, our meta-analysis results on the activity of Na + -K + -ATPase showed that the compatibility of Fuzi improved its activity relative to that of Fuzi alone, providing evidence for Fuzi's synergistic compatibility mechanisms on energy metabolism, which include increasing the activity of Na + -K + -ATPase on myocardial cells, regulating myocardial energy metabolic disorders, and ultimately enhancing myocardial cell activity and heart pump function. Regulation of the renin-angiotensin-aldosterone system The renin-angiotensin-aldosterone system (RAAS) is a vital human body system because it maintains plasma sodium concentration, arterial blood pressure, and extracellular volume. Kidney-secreted renin enzyme acts on its substrate to form Ang II, a versatile effector peptide hormone (Patel et al., 2017), which is the main medium for RAAS to achieve its biological effects. ALD is a hormone that regulates blood Frontiers in Pharmacology frontiersin.org volume in the human body, and the level is positively correlated with the severity of HF. Many experiments have confirmed that the RAAS and renin-angiotensin system (RAS) are activated during HF, in which ALD is particularly important, directly reflecting the effects of drugs in the treatment of HF. Elevating ALD in a short period in HF can increase cardiac output and thus have a compensatory effect, but if ALD continues to increase long term, it can cause water and sodium retention, increase blood volume of the circulatory system, and promote ventricular preload and myocardial cell fibrosis (McMahon, 2001), causing malignant arrhythmia or sudden death. Studies have confirmed that the reduction in ALD can reduce ventricular load and reduce the responsiveness of blood vessels to sympathetic nerves, thereby directly dilating blood vessels and improving heart function. Normally, Ang II can maintain vascular tension, excite sympathetic nerves, and regulate ALD secretion. However, under pathological conditions, Ang II becomes the main factor leading to ventricular remodeling and stimulates the increase in ALD secretion. Simultaneously, the increased release of ALD causes myocardial cell apoptosis, myocardial ischemia, and arrhythmia promoting ventricular remodeling and inducing and aggravating HF (Xiong and Wang, 2013), forming a vicious cycle. ET-1, a strong vasoconstrictor in the body, is secreted by vascular endothelial cells and cardiomyocytes. ET-1 can promote the synthesis and release of Ang II in cardiomyocytes and vascular smooth muscle cells as well as promote the proliferation of cardiac fibroblasts and synthesis of collagen. Studies have found that CHF can be accompanied by increased ET-1 levels (Xie and Peng, 2013). BNP is a peptide hormone of the natriuretic peptide class. Studies have found that the content of BNP is much higher in human heart tissue than in brain tissue. The BNP content is highest in the right and left atria of the human body. Ventricular volume load and ventricular wall tension affect the synthesis and secretion of BNP, so the concentration of BNP in serum can be used to reflect ventricular function (Xu, 2015). When HF occurs, the increase in LV filling pressure and ventricular wall tension are the key reasons for the increase in BNP secretion (Miao et al., 2015). Studies (Wang et al., 2012) have confirmed that with the gradual aggravation of HF, the level of BNP also increases, and when improvements in heart function are made, the level of BNP will significantly decrease, suggesting that BNP can be used as a marker to observe the severity of HF and assess the risk stratification and prognosis of HF. In addition, BNP, as a cardiac neuroendocrine hormone secreted by ventricular cardiomyocytes, is capable of dilating blood vessels, causing diuresis, and inhibiting RAAS and sympathetic nervous system activity (Zhang et al., 2012). Due to the diuretic and vasodilatory effects of BNP, it can resist the effect of RAAS in narrowing blood vessels and effectively prevent myocardial hypertrophy and fibrosis. Therefore, when HF occurs, the BNP plasma concentration is significantly increased; thus, by checking the blood BNP level, the activation condition of the RAAS and the severity of HF can be monitored. In this systematic review, a meta-analysis was conducted on the level of BNP and ALD in CHF rats, and the results showed significant decreases in BNP levels in the compatibility groups versus those in the single utilization groups, whereas for ALD, although there were no significant differences in the overall effect of ALD, decreases could be seen in individual studies. Moreover, it can be concluded from qualitative analyses that the Fuzi compatibility can reduce the levels of ET-1 and AngⅡ in plasma more effectively than using Fuzi alone and achieve a regulatory effect on the RAAS (Fu et al., 2018), proving that the synergistic compatibility mechanism of Fuzi is related to the regulation of RAAS and inhibition of the excessive activation of neuroendocrine factors, thereby enhancing the effect of anti-CHF. Inhibition of pro-inflammatory cytokines and alleviation on inflammation Interleukin (IL)-10 is a cytokine that mediates the interaction between leukocytes, and its main role is to inhibit the synthesis and release of inflammatory factors while simultaneously directly inhibiting the activation of inflammatory cells (Iyer and Cheng, 2012), thereby exerting a protective effect on the myocardium (Zhao et al., 2013). IL-18 is mainly produced by activated mononuclear macrophages, which is both a pre-and proinflammatory cytokine closely related to various cardiorenal and vascular complications (Bouquegneau et al., 2015), the levels of which are closely related to angina pectoris, myocardial infarction, and fatal coronary events (Schöttker et al., 2013). TNF-α is a multi-dominant cytokine produced by monocytes and macrophages that has a variety of physiological functions, such as regulating the immune response and promoting cell growth and differentiation. The biological activity of TNF-α is transmitted through specific receptors on the cell surface. As an important inflammatory mediator, TNF-α is secreted in large quantities induced by inflammation. Highly expressed TNF-α promotes the differentiation of B cells and improves the activity of B cells, which then promote the synthesis of excess IgE and induce a series of inflammatory reactions. Studies (Wang and Liu, 2014;Ghigo et al., 2014) have shown that increased levels of TNF-α are closely related to myocardial remodeling and decreased ventricular pumping function. CHF is often accompanied by severe inflammation, and myocardial cells can induce a variety of pro-inflammatory factors under chronic stress, including TNF-α, IL-10, and IL-18. Pro-inflammatory factors can activate macrophages through a cascade effect, producing more proinflammatory factors and inducing a stronger inflammatory response. Excessive pro-inflammatory factors can both directly inhibit heart function and accelerate the development of HF by inducing myocardial fibrosis and remodeling (Pinchuk et al., 2014). The meta-analysis results for TNF-α provided evidence of the superior therapeutic effects of Fuzi compatibility for decreasing the level of TNF-α in CHF animals relative to the Frontiers in Pharmacology frontiersin.org level in single utilization, suggesting that the synergistic compatibility mechanisms of Fuzi involve inhibition of proinflammatory cytokines and alleviation of inflammation. Effect on drug metabolism process in vivo AC and HA are the main active ingredients of Fuzi, and their pharmacokinetics in the body reflect the therapeutic effect of aconite on HF to a certain extent. Through pharmacokinetic experiments, Xie et al. (Xie et al., 2021) found that the absorption of aconite in a CHF rat model was less than normal, elimination was quick, and the retention time in the body was short. After continuous administration of red ginseng, compared with the normal group, the model group showed that AC and HA were mostly eliminated within 24 h, whereas the group administered red ginseng showed almost complete elimination within 48 h, indicating that the absorption of AC in the HF rats increased after compatibility. As a result, the elimination rate was reduced, the working time in the body was prolonged, and the curative effect was enhanced. In addition, preliminary laboratory research (Yang, 2019) found that in normal animals, the compatibility of drug-metabolizing enzymes could produce drug-drug interactions. Among them, the main metabolic CYP450 enzyme subtype of AC and HA was CYP3A4, whose corresponding type in rats was CYP3A2. Further, continuous administration of red ginseng inhibited the activity of CYP3A2 in rats. Further mechanistic studies have found that red ginseng can affect the metabolism of DDAs in vivo by regulating the PXR-CYP3A4 pathway. In conclusion, the potential synergistic mechanisms of Fuzi compatibility led to increased absorption of bioactive constituents, a reduced elimination rate so that the in vivo duration is prolonged, and the activity of drug-metabolizing enzymes, which affect metabolic behaviours, are inhibited, thus therapeutic effects enhanced. Inhibition of cardiomyocyte apoptosis The basic pathogenesis of CHF is ventricular remodeling, which is manifested as heart enlargement and/or myocardial hypertrophy (Cohn et al., 2000). At present, the factors affecting ventricular remodeling mainly include BNP and other neuroendocrine factors as well as the participation in cell apoptosis (Hirota et al., 1999). Among them, cardiomyocyte apoptosis is an

important process in the progression of CHF from the compensatory phase to the decompensated phase. Cardiomyocyte apoptosis can cause the hypertrophy of other cardiomyocytes and eventually lead to HF. Apoptosis is the process of autonomous cell death controlled by genes and is an important mechanism for maintaining the normal development and tissue state of the body. Its hyperfunction is closely related to the occurrence of some diseases. Wan et al. (2017) found that the compatibility of GA and liquiritin with HA significantly reduced the expression of B-cell lymphoma-2 (Bcl-2) associated X protein (Bax), caspase-3, fatty acid synthetase (Fas), and Fas-L proteins, and increased expression of Bcl-2 protein. The combination of HA and liquiritin with HA improved more significantly and has a prominent effect on cardiomyocyte apoptosis. Among these effects, the changes in Bcl-2 and Bax were equivalent to the efficacy of the positive control drug digoxin, but the difference was not significant. Simultaneously, GA combined with HA and liquiritin combined with HA significantly reduced the expression of Bax, caspase-3, Fas, and Fas-L protein and increased the expression of Bcl-2 protein relative to those in the HA-single group, which showed an inhibitory effect on cardiomyocyte apoptosis, providing evidence for the potential synergistic mechanisms of Fuzi compatibility providing enhanced inhibition of cardiomyocyte apoptosis, thereby preventing the progression of ventricular remodeling and exerting curative effects in CHF patients. Implication for further studies Systematic evaluation of animal experiments is an inevitable trend in the development of evidence-based medicine in human research and animal experiments as well as is a useful tool to improve the quality of animal experiments. Additionally, systematic evaluation is a potential solution to the common problem that current animal experimental results often fail to transform into clinical research. Systematic evaluation of animal experiments can provide a comprehensive and objective understanding of the research status of existing animal experiments and a more reliable basis for screening interventions that can enter clinical trials. Therefore, it is necessary to conduct a rigorous systematic review of animal experiments before clinical trials, which helps avoid waste of both animal and human research resources and prevents putting patients at unknown treatment risks. The establishment of a CHF model enables further study of the progression of HF and testing of new treatments against it (Sun and Ma, 2016). Although many reliable HF models, such as the use of surgery (e.g., TAC, LAD, AAC) or drug inducements (e.g., DOX) have emerged in recent years with the rapid development of experimental zoology and have made great contributions to clinical trials, drawbacks remain. CHF is a multifactorial systemic disease. After cardiac damage, the structure, neurohumoral, cellular, and molecular mechanisms are activated and serve as a network for maintaining physiological functions. All of these coordinated and complex processes can cause excessive capacity overload, increased sympathetic nerve activities, circulatory redistribution, and lead to different and parallel clinical symptoms (Tanai and Frantz, 2011), resulting in its multifactorial etiology and complicated pathogenesis. Moreover, the performances of hemodynamics, myocardial fibrosis, and cardiac hypertrophy at each stage vary, and the levels involved are diverse, ranging from cellular to molecular to genetic. Consequently, the Frontiers in Pharmacology frontiersin.org pathogenesis of CHF can be elucidated from multiple directions. Therefore, it is necessary to use models with different characteristics for more comprehensive analyses. Additionally, the process of human CHF is often accompanied by other diseases, such as hypertension or diabetes, but this is rare in animal and cell models, so a combination of multiple technologies is required (Geng and Liu, 2014), which should be the focus of future research. In our evaluation of experiments in CHF animals, the modeling method has a substantial role in the overall effect. According to our subgroup analysis, variability in modeling methods was associated with the levels of BNP, HR, HWI, ALD, LVSP, +dP/dt max , and −dP/dt max . The significance of the results of the above outcome indexes varied when subgroups were generated based on the modeling methods, sometimes from valid to invalid or the reverse, indicating a significant influence of the choice of modeling methods on the pooled results. More interestingly, we found that for the hemodynamics indicators (HR, LVSP, +dP/dt max , −dt/dt max ), the DOX-inducement modeling method was more likely to achieve a valid positive result on the pooled effects of compatibility than AAC, indicating that research on the hemodynamic process of CHF should use the DOX-inducement model. For research on the ventricular remodeling index (e.g., HWI), TAC showed a significant overall effect of compatibility, whereas DOXinducement and AAC did not, indicating that the TAC modeling method should be chosen. Furthermore, for research on the appropriate level of ALD to investigate how RAAS affects CHF, according to our subgroup analysis on ALD, the DOXinducement method is recommended for establishing a CHF model if a positive outcome is expected. However, if the research focuses on the level of BNP, surgery methods would be more suitable. With the continuous development of experimental zoology, some models will more accurately reflect future clinical results and be more stable, easier to use, economic, and reproducible, all of which should promote a more comprehensive understanding of CHF in humans. The focus of the present research was mainly on how to choose a more reasonable and more appropriate model for a better exploration of the different aspects of CHF, which depends on the model-effect relationship, for eventual use in the design of pre-clinical trials. Moreover, we investigated the time-effect relationship, which also has a critical role in clinical medicinal treatments. Subgroup analysis revealed that the difference in treatment duration was associated with the overall effects of HR, LVSP, +dP/dt max , and −dP/dt max , suggesting that duration had an influence on the ultimate results and that potential time-effect relationships should be explored. All things considered, these findings indicated that in pre-clinical studies of Fuzi compatibility effects in CHF animals, more attention should be given to the importance of establishing appropriate animal models and medication durations. Limitations The following limitations in this systematic review and meta-analysis were inevitable and should be considered carefully: 1) Neutral and negative results are normally less likely to be published in clinical trials, which increases the publication bias, thus exaggerating the effects of interventions. In our publication bias and small-study effect tests, the different degrees of asymmetry in the funnel plots and the statistical analysis results of Egger's test both indicated that publication bias and small-study effects should not be ignored and that the efficacies of compatible interventions were probably exaggerated, so the positive results of the Fuzi compatibility effects in CHF animals should be cautiously interpreted. 2) In the quality assessment of the included studies, some were evaluated as of poor methodological quality for various reasons, such as the unclear illustration of random allocation, which may have led to selection bias. 3) In the meta-analysis, a few studies were not included for certain outcome measures because of insufficient or unavailable data even though positive results were mentioned, but those studies were still included in the systematic review. Conclusion This systematic review and meta-analysis aimed to evaluate the efficacy of Fuzi compatibility on CHF animals and identify potential synergistic compatibility mechanisms. The results showed that Fuzi compatibility has therapeutic effects against CHF animals superior to those of using Fuzi alone, and the potential synergistic compatibility mechanisms were associated with improvements in the hemodynamic process, increased activity of Na + -K + -ATPase, suppression of Ca 2+ overload, regulation of the RAAS, inhibition of proinflammatory cytokines, alleviation of inflammation, improvements in ventricular remodeling and cardiac damage, influence on the metabolic process in vivo, and inhibition of cardiomyocyte apoptosis. We also found that the variations in modeling methods of CHF and medication duration influenced the pooled effects and thereby affected the stability of positive results, through which we concluded that there were possible model-effect and time-effect relationships. It should be clarified that factors, such as publication bias or other bias, the methodological quality of the included studies, and the small sample size effect can limit the accuracy and credibility of the positive results, which reminded us to dialectically treat the outcome of this paper. We hope more reasonably designed, fully reported, and high-quality preclinical animal experiments will be carried out so that more credible evidence can be provided for future clinical trials. Frontiers in Pharmacology frontiersin.org Integrity of Databases for Literature Searches in Nursing The quality of literature used as the foundation to any research or scholarly project is critical. The purpose of this study was to analyze the extent to which predatory nursing journals were included in credible databases, MEDLINE, Cumulative Index to Nursing and Allied Health Literature (CINAHL), and Scopus, commonly used by nurse scholars when searching for information. Findings indicated that no predatory nursing journals were currently indexed in MEDLINE or CINAHL, and only one journal was in Scopus. Citations to articles published in predatory nursing journals are not likely found in a search using these curated databases but rather through Google or Google Scholar search engines. R esearch, evidence-based practice, quality improvement studies, and other scholarly projects typically begin with a literature review. In research, the review of the literature describes existing knowledge about the topic, reveals gaps and further research questions to be answered, and provides a rationale for engaging in a new study. In evidence-based practice, the literature review provides evidence to answer clinical questions and make informed decisions. Quality improvement studies also begin with a search of the literature to gather available knowledge about a problem and explore interventions used in other settings. The appearance of journals that are published by predatory publishers has introduced the danger that reviews of the literature include Statement of Significance What is known or assumed to be true about this topic? The quality of nursing literature used is vital for the development of research studies, application of evidence in clinical settings, and other scholarly projects. Nurse scholars need to be confident as they search the literature that they are accessing sound information sources and not articles from predatory nursing journals, which do not adhere to quality and ethical publishing standards. Citations of articles in predatory nursing journals may be found when searching Google and Google Scholar, making these citations easy to access but potentially resulting in the integration of poor quality research into the nursing literature. On the other hand, searches through credible databases-MEDLINE, CINAHL, and Scopus-are less likely to yield citations from predatory publications. What this article adds: This study helps validate the trustworthiness of these databases for conducting searches in nursing. inadequate, poorly designed, and low-quality information being used as "evidence"raising the possibility of risky and harmful practice. Researchers and authors should be confident in the literature they cite; readers should have assurance that the literature review is based on sound, authoritative sources. When predatory journals are cited, that trust is eroded. No matter what type of study or project is being done, the quality of literature is critical for the development of nursing knowledge and for providing up-to-date information, concepts, theories, and approaches to care. 1 An effective literature review requires searching various reliable and credible databases such as MEDLINE (through PubMed or Ovid) and the Cumulative Index to Nursing and Allied Health Literature (CINAHL), among others that are relevant to the topic. The ease of searching using a web browser (now commonly referred to as "googling") has increased the risk of finding sources published in predatory and low-quality journals that have not met the standards of research and scholarship that can be trusted as credible and reliable evidence. The purpose of this article is to present an analysis of the extent to which predatory nursing journals are included in MEDLINE, CINAHL, and Scopus databases, used by nurse researchers and other nurses when searching for information, and in the Directory of Open Access Journals. This directory indexes "high-quality, open access, peerreviewed journals" and should not include any predatory journals. 2 PREDATORY JOURNALS Many studies have documented the problem of predatory journals. These journals do not adhere to quality and ethical publishing standards, often use deceptive language in emails to encourage authors to submit their manuscripts to them, are open access but may not be transparent with the article processing charge, may have quick but questionable peer review, and may publish inaccurate information on their Web

sites such as impact factor and indexing. [3][4][5][6] Predatory publishing is an issue in many fields including nursing. In a recent study, 127 predatory journals were identified in nursing. 7 Citations acknowledge the ideas of others and give credit to the authors of the original work. When articles are cited in a subsequent publication, those citations disseminate the information beyond the original source, and the article in which it is cited might in turn be referenced again, transferring knowledge from one source to yet another. When articles in predatory journals are cited, the same process occurs. Those citations transfer knowledge from the predatory publication beyond that source. Studies have found that authors are citing articles published in predatory journals in nursing as well as other 104 ADVANCES IN NURSING SCIENCE/APRIL-JUNE 2021 fields. [7][8][9][10] Nurse scholars need to be confident as they search the literature that they are accessing sound information sources and not articles from predatory journals. NATIONAL LIBRARY OF MEDICINE INFORMATION RESOURCES The National Library of Medicine (NLM) supports researchers and clinicians through its multiple health information resources including PubMed, MEDLINE, and PubMed Central (PMC). PubMed serves as the search engine to access the MEDLINE database, PMC, and books, chapters, and other documents that are indexed by the NLM. PubMed is free and publicly available: by using PubMed, researchers can search more than 30 million citations to the biomedical literature. 11 The majority of records in PubMed are from MEDLINE, which has citations from more than 5200 scholarly journals. For inclusion in MEDLINE, journals are assessed for their quality by the Literature Selection Technical Review Committee. 12 Five areas are included in this assessment: scope of the journal (ie, in a biomedical subject); quality of the content (validity, importance of the content, originality, and contribution of the journal to the coverage of the field); editorial standards and practices; production quality (eg, layout and graphics); and audience (content addresses health care professionals). PMC includes journal citations and full-text articles that are selected by the NLM for digital archiving. To be included in PMC, journals are evaluated for their scope and scientific, editorial, and technical quality. 13 Journals considered for inclusion are evaluated by independent individuals both inside and outside PMC. 14 PMC serves as the repository for articles to meet the compliance requirements of the National Institutes of Health (NIH) and other funding agencies for public access to funded research. About 12% of the articles in PMC are deposited by individual authors to be in compliance with funders and 64% by publishers, scholarly societies, and other groups. 15 Beginning in June 2020, as a pilot program, preprints reporting research funded by the NIH also can be deposited in PMC. 16 CINAHL AND SCOPUS The journal assessment and indexing processes for CINAHL and Scopus are similar to those used by the NLM. However, as private corporations, EBSCO (CINAHL) and Elsevier (Scopus) are not required to make journal selection processes publicly available or explicit. CINAHL has an advisory board for journal selection. A CINAHL representative provided the following criteria for indexing of journals in CINAHL: high impact factor; usage in reputable subject indexes (eg, the NLM catalog); peer-reviewed journals covered by other databases (eg, Web of Science and Scopus); top-ranked journals by industry studies; and article quality (avoiding low-quality journals) (personal communication, October 19, 2020). Elsevier's Scopus provides a webpage referring to the journal selection and assessment processes. Journals being considered for indexing in Scopus are evaluated by the Content Selection and Advisory Board and must meet the following criteria: peerreviewed with a publicly available description of the peer review process; published on a regular basis; has a registered International Standard Serial Number (ISSN); includes references in Roman (Latin) script; has English language titles and abstracts; and has publicly available publication ethics and publication malpractice statements. 17 LITERATURE REVIEW Studies have shown that in health care fields, researchers, clinicians, faculty, and students regularly search MEDLINE for their research and other scholarly and clinical information. [18][19][20][21] De Groote et al 18 found that 81% of health science faculty used MEDLINE to locate articles for their research. MEDLINE was used by the majority of faculty in each individual health care field including nursing (75%) and medicine (87.5%) for searching the literature and finding articles. In another study of 15 different resources, medical faculty and residents reported that PubMed was used most frequently for searching the databases of the NLM, primarily MEDLINE. 20 Few studies have focused on the search practices of nurses. In a review of the literature, Alving et al 22 found that hospital nurses primarily searched Google for information on evidence-based nursing. They used Google more than bibliographic databases. The quality of content that is retrieved when using PubMed as a search engine is important considering its widespread use for accessing scholarly and clinical information in nursing and other fields. Manca et al 23 reported that articles published in predatory journals were being retrieved when conducting searches using PubMed and were a concern for researchers. Based on their studies of predatory journals in neurology 24 and rehabilitation, 25 they concluded that predatory journals "leaked into PubMed" through PMC because of less stringent criteria for inclusion of journals. 23 Citations to articles from predatory journals then could be found using the PubMed search engine. However, in a letter to the editor, Topper et al 26 from the NLM clarified that individual articles published in predatory journals might be deposited in PMC to meet the requirements of research funding and be searchable in PubMed. Topper and colleagues make a clear distinction between journals indexed in MEDLINE or PMC and citations of individual articles that were deposited in PMC to meet funder requirements. Purposes The aim of this study was to determine whether predatory nursing journals were included in databases used by nurse researchers and other nurses when searching for information. These databases included MEDLINE (searched via PubMed), CINAHL (EBSCO), and Scopus (Elsevier) and in the Directory of Open Access Journals. METHODS In an earlier study, 127 predatory nursing journals were identified and assessed for characteristics of predatory publications. That dataset was used for the current study. For each predatory nursing journal, information was retrieved from the NLM Catalog, Ulrichsweb, and journal and publisher Web sites. Ulrichsweb 27 provides bibliographic and publisher information on academic and scholarly journals, open access journals, peer-reviewed titles, magazines, newspapers, and other publications. Journal titles of the predatory journals were often similar to nonpredatory journals and could be easily mistaken. To ensure accuracy, the information for each journal was checked for consistency between these sources using the ISSN, exact journal title, and publisher name. The purpose of an ISSN is to identify a publication and distinguish it from other publications with similar names. An ISSN is mandatory for all publications in many countries and having one assigned is considered a journal best practice. 28 For each predatory journal, the following data were collected if available: complete journal title; abbreviated journal title; acronym; ISSN (electronic and/or print); DOI prefix; publisher name and Web site URL; NLM index status; number of predatory journal articles cited in MEDLINE and PMC (when searching using PubMed), in CINAHL, and in Scopus; if the journal was indexed in the Directory of Open Access Journals; status in Ulrichsweb; and Google Scholar profile URL. Counts of articles cited were checked individually by journal title, publisher, and/or ISSN. Once ISSNs (both electronic and print where available) were assembled, a search algorithm was created, which included all retrieved journal ISSNs. MEDLINE was searched via PubMed using a combination of NLM journal title abbreviations and ISSNs. CINAHL, Scopus, and the Directory of Open Access Journals were searched using a combination of ISSN, journal title abbreviation, full title, and publisher. Results were visually inspected for accuracy and alignment with dataset fields. Data analysis Data were collected between January and April 2020. Data were entered into an Excel spreadsheet and organized by predatory journal name; abbreviated journal title; acronym; ISSN (electronic, print); DOI prefix; Web site URL; entry in NLM Catalog (yes/no); index status; number of articles cited in PubMed, CINAHL, and Scopus; Directory of Open Access Journals (included/not included); Ulrichsweb status (active/ceased); publisher; and Google Scholar profile URL. Frequencies and medians are reported. RESULTS Of the 127 predatory nursing journals in the dataset, only 102 had ISSNs to use for the search. Eighteen of the journals had records in the NLM Catalog, but only 2 of those had ever been indexed in MEDLINE, and neither are currently indexed. These 2 journals had been published earlier by a reputable publisher but then were sold to one of the large predatory publishers. The NLM Catalog record for these journals indicates that citations of articles from them appeared in MEDLINE through 2014 for one of the journals and 2018 for other, but following their transition to the new publisher are no longer included. Consistent with the MEDLINE results, these same 2 journals had been indexed in Scopus as well. Citations of articles from one of these journals were added to Scopus up to 2014, with no articles cited thereafter. Articles from the second journal continue to be added through 2020. One additional journal from the predatory journal dataset is currently in Scopus, however, only through 2014. None of the predatory nursing journals were indexed in CINAHL based on full journal title, title abbreviation, ISSN, or publisher. Two journals in the dataset were found in the Directory of Open Access Journals. When searching PubMed, we found citations of articles from 16 predatory nursing journals. The number of citations ranged from 1 to 372 citations (from one of the journals indexed earlier in MEDLINE but sold to a predatory publisher). The second highest number of citations (n = 168) was of articles from a predatory nursing journal that had been depositing articles in PMC (and thus were retrievable when searching PubMed) but is no longer adding new material to PMC. The other citations were of articles deposited in PMC to meet requirements of NIH and other research funding. The predatory journals in which these articles were published, however, are not indexed in MEDLINE or PMC. There were no articles from predatory nursing journals cited in CINAHL. Scopus has citations from the 2 predatory nursing journals that are no longer indexed there: 616 that were published in one of the journals and 120 from the other. Articles from a third predatory nursing journal in the study dataset, which is currently indexed in Scopus, totaled 173 (see Table). This analysis documented that none of the predatory nursing journals in the study dataset were currently indexed in MEDLINE or CINAHL, and only one journal is still in Scopus. Most of the citations of articles from predatory journals found in a search of these databases are from earlier years before the journals were sold to one of the large predatory publishers. Other citations are to articles deposited in PMC in compliance with research funder requirements. DISCUSSION By using PubMed as a search engine and entry point to the databases of the NLM, researchers can search millions of records included in MEDLINE, or in process for inclusion, and articles from PMC deposited by publishers or authors for compliance with funders. Six million records, and about 5500 journals, can be searched in CINAHL Complete, 29 and Scopus, the largest of the proprietary databases, provides access to 24000 journals and 60 million records. 30 Results from this study show that very few articles published in predatory nursing journals find their way into a search done using PubMed and Scopus and none into CINAHL. In a prior study, 814 citations of articles in predatory nursing journals were found in articles published in nonpredatory nursing journals. 7 Based on this current study, the conclusion can be made that these citations are not coming from searches in MEDLINE/PubMed, CINAHL, or Scopus and are likely from searches done using Google or Google Scholar as the search engine. The databases examined in this study are curated by organizations with a vested interest in maintaining and improving the quality of the research literature in those databases. Searching multiple databases using different search engines can be frustrating and time consuming. There is overlap among MED-LINE, CINAHL, and Scopus. However, these are curated databases and, as this study found, are

unlikely to return many, if any, predatory citations as part of the search results. Still, it falls on the searcher to eliminate duplicates and redundant citations. Further, certain types of literature, such as theses, dissertations, and fugitive (or "gray" literature), 31 are unlikely to be found in any of these databases, even though those citations may be important or relevant sources. Given this, it is easy to understand the intuitive appeal of Google Scholar, which provides "one stop shopping": "From one place, you can search across many disciplines and sources: articles, theses, books, abstracts and court opinions, from academic publishers, professional societies, online repositories, universities and other web sites. Google Scholar helps you find relevant work across the world of scholarly research." 32 Google and Google Scholar were founded with a mission to become the most comprehensive search engines in the world. While this allows someone to scour the World Wide Web and Internet for some of the most obscure facts available, at the same time, little is done to verify or validate the results that are returned. Thus, it falls on the searcher to be diligent and evaluate the results of a Google or Google Scholar search, which will include citations of articles in predatory journals. This is easily confirmed by the fact that many predatory journal Web sites promote the Google Scholar logo as a sign of indexing or a badge of legitimacy. Another vexing issue that was revealed in this study is that of reputable journals that have been bought by predatory publishers. This study found 2 journals in this category. Brown 33 reported on 16 medical specialty journals that were purchased from 2 Canadian commercial publishers by a predatory publisher. In all these cases, it is the same predatory publisher, although some of the purchases were made under a different business imprint, adding further confusion to an already muddied situation. Jeffrey Beall, who coined the term "predatory publisher" and maintained the blog "Scholarly Open Access" for almost a decade, was quoted by Brown 33 : "[The company] is not only buying journals, it is buying metrics and indexing, such as the journals' impact factors and listing in Scopus and PubMed, in order to look legitimate." One positive finding from this study was that the 2 purchased journals that were identified were quickly de-accessioned by the NLM and are no longer indexed in MEDLINE, although citations from their pre-predatory era remain intact. Recommendations All of this presents a confusing picture, but it is possible to make some specific recommendations to aid researchers, clinicians, faculty, and students in their literature searches. First, become familiar with the journals and publications in your field. This is a basic foundation of scholarship. As you read articles, remember where they were published, learn journal titles, and focus on sources as well as the content. As you come across predatory journals in nursing and health care, make note of them and learn their titles too. Remember that many predatory journals adopt names that are intended to be confusing and may differ from a legitimate journal by only one letter, such as "Africa" and "African." Second, consider carefully how to approach your search from the outset. If you choose to start with MEDLINE (searched via PubMed), CINAHL, or Scopus, then you can have some assurance that the results will not return citations from predatory journalsalthough you should still verify every citation that you receive. On the other hand, Google and Google Scholar can be a "quick and easy" way to get started but will require that you carefully review and evaluate the results. If you need to venture to other more specialized databases, such as PsycInfo or ERIC (Education Resources Information Center), it is important to carefully inspect the results that you receive. To reduce the risk of including a predatory journal article in research, nursing scholars should use reputable bibliographic databases, which have clear criteria for journal indexing, for their searches. Third, when you come across a journal title that is not familiar, take time to research it, visit the journal Web site and evaluate the information at the Web site, and determine whether it is a credible source to include in your results. If something seems irregular, then it is worth your time to do more investigating-either on your own or by enlisting the help of a knowledgeable colleague or librarian. Journals change publishers all the time, and while most of these business transfers are benign and probably will not impact you as an end consumer of the literature, that is not always the case. Likewise, the major publishers in the world today are large, multinational conglomerates that regularly spin off or purchase other companies. While this probably will not impact you on a day-to-day basis, it is important to investigate any irregularities when conducting a search of the literature. Last, because these issues are complex and multifaceted, it is always wise to consult with a librarian who can assist you in every step of the search process. Their knowledge and expertise in information literacy, data sources, and searching techniques can help to ensure that you find the information you need from sources that are reliable and credible. SUMMARY Researchers, clinicians, faculty, and students need to be careful not to include citations from predatory sources in their literature searches and articles. Predatory journals publish low-quality studies and citing this work erodes the scholarly literature in nursing. The findings of this study offer some reassurance to those who search the professional nursing literature: if you begin a search in a database such as MEDLINE, CINAHL, or Scopus, then the results will probably not include citations to predatory publications. Google and Google Scholar searches, however, may very well include predatory citations, and in that case, it is the searcher's responsibility to carefully evaluate the output and discard findings from nonlegitimate sources. Enlisting the help of a librarian is always beneficial and highly recommended. Metabolic characterization of triple negative breast cancer Background The aims of this study were to characterize the metabolite profiles of triple negative breast cancer (TNBC) and to investigate the metabolite profiles associated with human epidermal growth factor receptor-2/neu (HER-2) overexpression using ex vivo high resolution magic angle spinning magnetic resonance spectroscopy (HR MAS MRS). Metabolic alterations caused by the different estrogen receptor (ER), progesterone receptor (PgR) and HER-2 receptor statuses were also examined. To investigate the metabolic differences between two distinct receptor groups, TNBC tumors were compared to tumors with ERpos/PgRpos/HER-2pos status which for the sake of simplicity is called triple positive breast cancer (TPBC). Methods The study included 75 breast cancer patients without known distant metastases. HR MAS MRS was performed for identification and quantification of the metabolite content in the tumors. Multivariate partial least squares discriminant analysis (PLS-DA) modeling and relative metabolite quantification were used to analyze the MR data. Results Choline levels were found to be higher in TNBC compared to TPBC tumors, possibly related to cell proliferation and oncogenic signaling. In addition, TNBC tumors contain a lower level of Glutamine and a higher level of Glutamate compared to TPBC tumors, which indicate an increase in glutaminolysis metabolism. The development of glutamine dependent cell growth or “Glutamine addiction” has been suggested as a new therapeutic target in cancer. Our results show that the metabolite profiles associated with HER-2 overexpression may affect the metabolic characterization of TNBC. High Glycine levels were found in HER-2pos tumors, which support Glycine as potential marker for tumor aggressiveness. Conclusions Metabolic alterations caused by the individual and combined receptors involved in breast cancer progression can provide a better understanding of the biochemical changes underlying the different breast cancer subtypes. Studies are needed to validate the potential of metabolic markers as targets for personalized treatment of breast cancer subtypes. Background Triple negative breast cancer (TNBC) is a heterogeneous subgroup of breast cancer characterized by the absence of expression of estrogen receptor (ER), progesterone receptor (PgR) and human epidermal growth factor receptor-2/neu (HER-2). TNBC represents approximately 15-20% of all breast cancer cases and is generally considered as the most severe subgroup of breast cancer. Patients diagnosed with TNBC are largely unresponsive to currently available targeted therapies, such as Tamoxifen and Trastuzumab, in addition to having a higher risk of relapse and a higher mortality rate compared to other breast cancer subtypes [1]. Treatment with protein inhibitors against PI3KCA and HSP90 have shown to be efficient in only a subset of TNBC [2]. Therefore, there is an urgent need to identify new molecular targets for treatment of TNBC to improve treatment care and survival of this breast cancer subgroup. Classification of breast cancer according to molecular subtypes is highly relevant and may provide significant prognostic information related to patient outcome. Several studies have investigated the underlying genomic and transcriptomic characteristics of TNBC [3][4][5]. The results suggest the existence of a variety of TNBC subtypes including basal and non-basal, p53 mutated and high genomic instability, among others [3]. For example, five distinct subtypes of TNBC have been suggested based on gene expression profiles [5]. In a recent study, TNBC was subdivided into basal or 5-negative phenotype dependent on the expressions of assorted basal markers, including cytokeratin 5 (CK5) and epithelial growth factor receptor (EGFR) using immunohistochemistry (IHC) and in situ hybridization [6]. The validation of reliable markers for breast cancer sub-classification is still ongoing. Altered energy metabolism is a new emerging hallmark of cancer [7]. Increasing evidence suggests that alterations in cancer metabolism, especially choline phospholipid and amino acid metabolism may provide potential targets for treatment of breast cancer. To our knowledge, the metabolite profiles of TNBC and the metabolic influences of HER-2 overexpression have not yet been investigated in detail. Metabolomics, defined as a systematic study of the metabolism, has proven to be an important tool for the identification of new biomarkers for targeted treatment, treatment evaluation and prediction of cancer survival [8][9][10][11]. Previous studies have shown the potential and benefit of combining the different OMICS approaches, e.g. transcriptomics and metabolomics, for better molecular characterization and stratification of breast cancer [12][13][14][15]. Ex vivo high resolution magic angle spinning magnetic resonance spectroscopy (HR MAS MRS) can be used for the identification and quantification of the metabolite content in a biological tissue sample. HR MAS MRS is a non-destructive technique meaning that the tissue remains intact after examination and can be used for other OMICS approaches, thus allowing for a comprehensive and detailed study of the molecular composition of the tissue. By using HR MAS MRS, more than 30 metabolites can be detected and assigned simultaneously in breast cancer tissue [16]. HR MAS MRS has been widely used to study cancer related pathways, including choline phospholipid metabolism, glycolysis (the Warburg effect), amino acids, lipids and polyamines, among others [17][18][19]. The metabolite profiles acquired by HR MAS MRS have shown to correlate to hormone receptor status, treatment response and survival in breast cancer [20][21][22][23][24]. Analysis of HR MAS MRS spectra can be challenging due to the high number of collinear variables (exceeding tens of thousands of data points per sample). Multivariate data analysis is a suitable method for analyzing the complex and high dimensional MRS data. Partial least squares discriminant analysis (PLS-DA) can be used to identify metabolic differences between distinct classes by finding linear relationships between the spectral data and class variables, e.g. receptor status [25]. In addition to multivariate modeling, quantification of the individual metabolites can be achieved by calculating the area under the peak signal. Most studies have compared TNBC with non-triple negative breast cancer, most commonly ER pos /PgR pos breast cancer subtype, in those studies the effects of HER-2 overexpression were not considered. In this study, we have investigated the metabolic differences between TNBC tumors and tumors with ER pos /PgR pos / HER-2 pos status, which for the sake of simplicity is called triple positive breast cancer (TPBC). We have also examined the influences of ER, PgR and HER-2 receptors status individually on breast cancer metabolism and explored the metabolite profiles associated with HER-2 overexpression. Metabolic alterations caused by the individual and combined hormone and growth receptors may help identify potential targets for treatment of breast cancer subtypes. Patients and tumor receptor status Included in this study were patients (n = 75) aged 34 to 90 diagnosed with breast cancer without known

distant metastasis. The patients did not receive any pre-surgical therapy for their cancer disease. The biopsies were extracted immediately after surgical removal of the tumor. Parts of the tumor were used for routine analyses, including tumor grade, ER, PgR and HER-2 status (Table 1). Tumors were considered positive for ER and PgR when more than 10% of tumor cells showed positive staining by IHC. The samples were tested for HER-2 gene expression using a validated dual probe fluorescence in situ hybridization (FISH) assay (HER-2 IQFISH pharmDx/HER-2FISHpharm Dx) or for protein overexpression using a validated IHC assay (HercepTest, DAKO). The HER-2 gene was considered amplified if the gene to chromosome 17 ratio was larger than 2.0 analyzed by FISH or evidence of protein overexpression by IHC score 3+. Another part of the tumor was snap frozen immediately during surgery and stored in liquid nitrogen for MRS analysis. All patients have signed a written informed consent, and the study was approved by the Regional Ethics Committee, Central Norway. Imprint cytology Cytological imprint was performed to confirm the presence of tumor cells in the sample before HR MAS MRS and was used as an inclusion criterion and not as a quantitative measurement [26]. This technique is fast and requires minimal preparation. In brief, the tissue was gently pressed on a glass slide and air-dried for approximately 10 minutes. The imprints were fixed in ethanol and stained with May-Grünwald-Giemsa stain (Color-Rapid, Med-Kjemi, Norway). All imprints were reviewed by a well-trained pathologist. Samples with absence of tumor cells were excluded from further analysis. High resolution magic angle spinning To minimize the effect of tissue degradation on the metabolite profiles, the samples were prepared on ice block and within a short period (5 ± 1 min). The biopsies (13 ± 3 mg) were cut to fit 30 μl disposable inserts filled with 3 μl phosphate buffered saline (PBS) in D 2 O containing 1.0 mM TSP for chemical shift referencing and 1.0 mM Format for shimming. The HR MAS spectra were acquired on a Bruker Avance DRX600 spectrometer equipped with a 1 H/ 13 C MAS probe with gradient (Bruker Biospin GmbH, Germany) using the following parameters; 5 kHz spin rate, 4°C probe temperature, cpmgpr1D sequence (Bruker Biospin GmbH, Germany) with 273.5 ms total echo time, a spectral width of 20 ppm (−5 to 15 ppm) and 256 scans (NS). For some patients, more than one biopsy (taken from different places in the tumor) were prepared and analyzed by HR MAS MRS. Data analysis Following acquisition, the spectra were Fourier transformed into 65.5 k after 0.3 Hz line broadening and TSP was calibrated to 0.00 ppm (Topspin 3.1, Bruker Biospin GmbH, Germany). The following spectral preprocessing steps were carried out using Matlab R2009a (The Mathworks, Inc., USA). Spectral regions containing signals from chemical contaminations (e.g. ethanol), water, and lipids were removed before multivariate data analysis. Baseline offset was corrected by setting the lowest point of each spectrum to zero. The spectra were normalized to equal total area to account for differences in sample size. Furthermore, the spectra were peak aligned using icoshift [27]. The spectral region between 1.5 -4.7 ppm, containing the majority of low-molecular weight metabolites, was used as the final input for the multivariate models. PLS-DA and metabolite relative quantification were performed to evaluate the metabolic differences between the tested groups using Matlab and PLS_Toolbox 6.2.1 (Eigenvector Research, USA). The spectra were meancentered before the PLS-DA modeling. The classification results were calculated using random cross validation (20% for testing and 80% for training, repeated 20 times). In cases where there were multiple spectra from the same patient, all of these spectra were either used for training or testing. The number of latent variables (LVs) used for all repetitions was chosen by leave one patient out cross-validation of the whole data set. Permutation testing, carried out by randomly assigning the class labels, was performed to evaluate the statistical significance of the classification results [25]. The permuted classification result was calculated as described for the PLS-DA models and repeated 1000 times. Metabolites importance in the PLS-DA loading were identified by variable importance in the projection (VIP) scores [28]. Relative metabolite quantification was performed by peak integration using mean normalized spectra after removal of water, lipids and contaminations. Statistical differences between the groups were tested by Wilcoxon testing with Benjamini Hochberg correction for multiple testing. P-values ≤ 0.05 were considered significant. The p-values adjusted for multiple testing are given as qvalues. While P-values are used as an indicator of the false positives in all tested values in the dataset, the qvalues are used to interpret the false discovery rate (FDR) among significant p-values. To give a more accurate indication of the FDR both p-and q-values are listed in the results. The quantification results are illustrated by heat maps (Matlab R2009a). Results Spectra from biopsies with absence of tumor cells and low spectral quality with high noise and severe chemical contamination were excluded from further analysis (n = 4). In total, 106 biopsies from 73 patients were included in the data analyses. A representative metabolite spectrum of breast cancer tissue obtained by HR MAS MRS is shown in Figure 1. The metabolite data shows no significant association with tumor grade and lymph node status by PCA and PLS-DA modeling (data not shown). The PLS-DA classification results of TNBC, ER, PgR and HER-2 are summarized in Table 2. TNBC versus TPBC The PLS-DA shows the highest CV accuracy for separating TNBC and TPBC (77.7%, p = 0.001). The corresponding score and loading plots show a clear separation between the two groups. TNBC is characterized with higher levels of Choline and Glycerophosphocholine (GPC), and a lower level of Creatine compared to TPBC (Figure 2A). Based on the loadings, high levels of PC and Glycine were observed in some tumors, but their influence in the classification model are unclear. Relative quantification shows consistently higher levels of Choline (p = 0.008, q = 0.041) in TNBC tumors. Lower levels of Glutamine (p < 0.001, q = 0.001) and higher levels of Glutamate (p = 0.002, q = 0.015) were also observed in TNBC compared to TPBC tumors ( Figure 3A). Creatine appears to be important for separating TNBC and TPBC in the multivariate analysis identified by a high value of VIP score. Lower levels of Creatine were also found in TNBC compared to TPBC tumors by relative quantification, however, the q-value was not significant (p-value = 0.031, q-value = 0.109). Hormone receptor status PLS-DA models show clear separations between ER neg and ER pos (72.2%, p < 0.001), and PgR neg and PgR pos (67.8%, p < 0.001) tumors. ER neg tumors show higher levels of Glycine, Choline, and Lactate compared to ER pos tumors, as shown in the score and loading plots ( Figure 2B). According to the VIP scores, Glycine appears to be most important for the discrimination between ER neg and ER pos . Higher levels of Glycine (p = 0.002, q = 0.010), Choline (p = 0.021, q = 0.067), Lactate (p < 0.000, q = 0.001), and Glutamate (p <0.001, q <0.001) and lower level of Glutamine (p <0.001, q <0.001) were observed in ER neg compared to ER pos tumors by relative quantification ( Figure 3B). PC levels appear to be high in some tumors from both groups, and could not be used to discriminate between ER neg and ER pos tumors. PLS-DA classification and relative quantification of PgR neg and PgR pos tumors show similar metabolite profiles as ER neg and ER pos tumors (data not shown). HER-2 status HER-2 neg and HER-2 pos tumors were discriminated by PLS-DA with 69.1% CV accuracy (p < 0.001). Contrary to ER neg and PgR neg , HER-2 neg tumors have a lower level of Glycine (p = 0.002, q = 0.012) compared to HER-2 pos tumors ( Figures 2C and 3C). Similar to what was observed for ER neg and PgR neg tumors, lower levels of Glutamine (p = 0.003, q = 0.017) were observed in HER-2 neg compared to HER-2 pos tumors detected by relative quantification. In addition to the changes in Glycine and Glutamine, HER-2 neg tumors also display higher levels of Alanine (p = 0.010, q = 0.039), and lower levels of Succinate (p = 0.001, q = 0.012) and Creatine (p = 0.024, q = 0.075) compared to HER-2 pos tumors by relative quantification. In the loading plot, PC levels appear to be higher in some HER-2 neg compared to HER-2 pos tumors. However, the relative quantification result shows no significant difference in PC levels between the two groups. HER-2 metabolite profiles in tumors with different ER and PgR status To investigate the metabolic influences of HER-2 status independently of the hormone receptors status, the metabolite profiles associated with HER-2 status were examined within ER neg , ER pos , PgR neg , and PgR pos tumors separately. The PLS-DA results are shown in Table 3. The scores and loadings of PLS-DA models show higher levels of Glycine in HER-2 pos compared to HER-2 neg tumors irrespective of ER and PgR status ( Figure 4A-D). Glycine levels determined by relative quantification showed a trend of higher levels in HER-2 pos compared to HER-2 neg tumors in the different ER and PgR status groups (p < 0.021 and q < 0.133). Glutamine also showed a trend of higher level in HER-2 pos compared to HER-2 neg tumors (p < 0.034 and q < 0.179). In the PLS-DA models, PC appears to be high in some HER-2 neg tumors. However, PC level was not significantly different between HER-2 pos and HER-2 neg by relative quantification. Discussion Triple negative breast cancer is characterized as being estrogen receptor, progesterone receptor and HER-2/neu receptor negative; it is a heterogeneous breast cancer subtype that is difficult to treat and is associated with high recurrence and poor outcome [1]. Several studies have investigated the underlying genomic and gene expression patterns of TNBC [2][3][4][5] while the metabolite profiles of TNBC have not yet been investigated in detail. Most studies have compared TNBC with ER pos / PgR pos breast cancer subtype, which does not take into consideration the influence of HER-2 status on breast cancer molecular profiles. In this study we investigated the metabolite profiles of patients with TNBC compared to TPBC and showed that these two groups could be successfully separated based on the metabolite profiles of tissue biopsies. In accordance with previous studies, altered metabolite profiles were observed in tumors with different expression of ER and PgR [21]. Furthermore, our results show that overexpression of HER-2 might cause alterations to the metabolite profiles of breast cancer independent of hormone receptor status, thus affecting the differentiation between TNBC and TPBC. The basal-like breast cancer subtype is defined through gene expression profiling and is considered to be a more aggressive breast cancer subtype compared to luminallike and HER-2 enriched gene expression subtypes. The majority of basal-like tumors are TNBC, but not all TNBC are defined as basal-like by gene expression. As previously published, the discrepancy rate is approximately 20-30 % [29]. Furthermore, there exists a significant overlap between TNBC, basal-like and BRCA-1 breast cancer [30]. In our study, TNBC has significantly higher Choline levels compared to TPBC, and this is in accordance with previous findings where higher Choline levels were detected in the more aggressive basal-like xenografts and TNBC patients as compared to the less aggressive luminal-like xenografts and ER pos /PgR pos breast cancer patients [12]. In another study, a significantly higher total Choline (tCho = PC + GPC + Choline) signal to noise ratio (tCho/SNR) was detected in TNBC when compared to non-triple negative tumors using in vivo MRS [31]. Choline-containing metabolites are involved in cell signaling, lipid metabolism, and cell membrane synthesis and degradation. The tCho level detected by in vivo and ex vivo MRS has been suggested as a biomarker for breast cancer diagnosis and response to chemotherapy [19]. Patients with basal-like breast cancer have been shown to be more sensitive to anthracycline-based neoadjuvant chemotherapy than the luminal subtype and a higher percentage of patients with a pathological complete response (pCR) to the treatment was achieved in the basal-like compared to luminal subtypes [32]. However, for patients with residual disease after chemotherapy, the basal-like subtypes

showed worse overall survival better survival rate [22,23]. Targeting the genes and enzymes involved in the choline phospholipid metabolism is currently under investigation, and so far the results have been promising. Down-regulation of choline kinase alpha (CHKA), the gene regulating the conversion of Choline to PC, has been shown to decrease cell proliferation, and to increase the effect of chemotherapy in ovarian [33] and breast cancers [34], whereas CHKA overexpression increases drug resistance in breast cancer cells [35]. The CHKA inhibitor is currently under phase I clinical trial. Our results suggest that targeting the genes/enzymes responsible for the choline phospholipid metabolism may provide new molecular targets for treatment of TNBC. Alterations in ER, PgR and HER-2 expression have proven to play a major role in breast cancer progression, with ER pos and PgR pos tumors having better prognosis, while HER-2 overexpression is associated with a worse prognosis. Thus metabolic alterations caused by these hormone and growth receptors are highly relevant, especially because the molecular reasons behind their overexpression/amplification remain largely unknown. Similar metabolite profiles were observed in tumors with ER neg and PgR neg , and ER pos and PgR pos status. In accordance with previous findings, we found a higher level of Glycine in patients with ER neg and PgR neg tumors compared to ER pos and PgR pos , respectively [21]. Higher levels of Choline and Lactate were also observed in ER neg and PgR neg tumors, which suggest enhanced glycolytic activity and tumor aggressiveness. ER status is generally accepted as an independent prognostic and predictive factor, while the significance of PgR status is less clear [36]. Although TNBC is considered to be a more aggressive breast cancer subgroup due to low response to available treatment, the overexpression of HER-2 itself is associated with poorer prognosis compared to HER-2 negativity [37]. Patients identified with HER-2 pos tumors are often treated with Trastuzumab. Noticeably, it has been reported that about 20-30% of HER-2 pos patients fail to respond to first time treatment with Trastuzumab and about 15% of patients will develop resistance to this drug [38,39]. Therefore, there is also a need to identify new molecular targets for treatment of this breast cancer subtype. In our study, we found high levels of Glycine and Alanine to be associated with HER-2 pos breast tumors. Alanine is involved in the synthesis of Glycine from Pyruvate and Serine. High levels of Glycine have previously been shown to correlate with poor prognosis in breast cancer [23,40,41]. Glycine is an amino acid involved in the synthesis of proteins, nucleotides and glutathione. The potential role of Glycine as a tumor biomarker has also been studied in human brain tumors, where it was found to positively correlate with tumor grade [42,43]. Moreover, the synthesis of Glycine from Glucose has been shown to correlate with rapid cancer cell proliferation [44]. Interestingly, we found higher levels of Glycine to be associated with HER-2 overexpression in the ER neg , ER pos , PgR neg , and PgR pos tumors separately, which suggest Glycine to be a specific marker for HER-2 amplification regardless of the ER and PgR status. ER neg and PgR neg tumors and ER pos and PgR pos tumors show comparable results. Tumors overexpressing HER-2 have shown to acquire resistance to estrogen therapy which suggests that there exists a crosstalk between ER and HER-2 status [45]. The p-values for differences in Glycine relative concentration between HER-2 pos and HER-2 neg were significant before multiple corrections, while the adjusted p-values showed trends towards significance. However, the chance of false positive result as described by the false discovery rate was low. In our cohort, we could not detect any differences in Glycine between TNBC and TPBC, possibly due to the high level of Glycine in ER neg and PgR neg and low level of Glycine in HER-2 neg tumors, which may cancel out the differences in Glycine between TNBC and TPBC. Based on our results we suggest Glycine to be associated with tumor aggressiveness in HER-2 pos breast cancer. Recently, there has been an increasing interest in detecting the circulating HER-2 protein in serum samples for use as a complementary assay to IHC and FISH analysis for diagnosis, but also for use as a prognostic marker for breast cancer recurrence [46,47]. High throughput screening of serum metabolites, including Glycine, is feasible using MRS or other laboratory assays and should be investigated further as a breast cancer prognostic marker. Furthermore, significant changes in other amino acids were also observed between the different breast cancer subtypes. Tumors negative for ER and PgR, and TNBC tumors contain lower levels of Glutamine and higher levels of Glutamate compared to tumors with positive receptor statuses which might result from increased glutaminolysis metabolism. Glutamine plays an important role in nucleotide and protein synthesis and in mitochondrial energy metabolism. Increased uptake and metabolism of Glutamine through glutaminolysis can provide a proliferating cell with significant amount of NADPH requirement [48]. Some cancer cells develop addiction to Glutamine and become dependent on Glutamine to support cell growth and activation of signaling molecules, e.g. mTOR kinase [49]. Recent studies have explored the potential of targeting amino-oxyacetic acid (AOA) for inhibition of cell proliferation in breast cancer xenograft models [50]. In a recent study, the expression of glutamine-related proteins was found to be highest in HER-2 subtypes compared to other breast cancer subtypes [51]. In our study, we found a higher level of Glutamine in HER-2 pos compared to HER-2 neg tumors. The role of Glutamine metabolism in breast cancer prognosis and treatment is still under investigation. However, increasing evidence suggests that alterations in cancer metabolism, especially the choline phospholipid and amino acid metabolisms may provide potential targets for treatment of breast cancer. In 2010, the American Society of Clinical Oncology (ASCO) and College of American Pathologists (CAP) issued guidelines that recommended the threshold for determining ER and PgR positivity to be decreased from 10% to 1%, in order to standardize the determination of hormone receptor status by IHC and also to increase the number of patients eligible for hormone therapy. In this study, the 10% cutoff value was used according to The Norwegian Breast Cancer Group recommendation at the time of inclusion. There is an ongoing debate about whether the decrease in ER and PgR threshold has led to a group of false ER pos tumors. Studies have shown that the majority of low ER pos tumors (≥1 < 10%) were identified as basal-like or HER-2 enriched tumors with pathological features more similar to ER neg than ER pos tumors, while only a minority of low ER neg tumors was classified as luminal A subtype [52][53][54]. In a large breast cancer study by Engstrøm et al., only 24 out of 909 cases (2.7%) showed positive staining for ER in ≥1 < 10% of the tumor cells and were classified as ER pos according to the new guidelines [6]. The authors found little or no change in the Kaplan-Meier and Cox results when comparing the new 1% cut-off with the previously 10% cut-off in their study. In our cohort, we found no correlation between the metabolite profiles and tumor grade and also no correlation between the metabolite profiles and lymph node status. We have previously investigated nodal metastasis using metabolite data from the primary tumor using PLS-DA, and the results showed only a weak correlation with nodal spread [21]. Based on our results, the differences in the metabolite profiles observed are indeed resulting from the hormone and growth receptor status and not dependent on tumor grade or nodal metastasis. This study is restricted by some limitations including the small number of samples in each subgroup and the lack of normal control tissues. Most of the patients in this study were recruited less than 5 years ago, long term follow-up is thus yet not available, however this aspect will be important in future studies. In addition, it would be interesting to investigate if Ki67 overexpression is associated with adverse metabolic profiles; Ki67 was not included, however, as part of the standard histochemical staining at the time of patient recruitment in our study. Statins are a class of drugs that reduce the production of cholesterol by inhibiting the enzyme HMG-CoA reductase. In a recent study, treatment with Lovastatin was shown to decrease choline-containing phospholipids and inhibit the proliferation in breast cancer cells in vitro [55]. The effects of statins on the metabolite profiles should also be investigated in more details. Breast cancer tumor heterogeneity is a common challenge. To minimize the effect of heterogeneity, we have chosen to include only tumors with T1/T2 stages (<5 cm in diameter) without distant metastasis. In this study, the effect of heterogeneity on the metabolic profile when sampling multiple biopsies was tested by comparing the average correlation of 35 random pairs repeated 1000 times versus 35 pairs of sample from the same patient. Our results show that the variation between patients is significantly higher than the variation within a patient (p-value < 0.001). Conclusion The classification of TNBC and TPBC tumors were successfully separated based on the metabolite profiles. Choline levels were found to be higher in TNBC compared to TPBC, possibly related to tumor proliferation and oncogenic signaling. TNBC tumors had a lower level of Glutamine and a higher level of Glutamate compared to TPBC tumors, which indicates an increase in glutaminolysis metabolism and suggests the development of glutamine dependent cell growth. The classification of ER, PgR and HER-2 status were also successful. We found significantly higher levels of Glycine in HER-2 pos breast cancer, which supports the potential of Glycine as a marker for tumor aggressiveness. Further studies are needed to validate the potential of metabolic markers as targets for personalized treatment of breast cancer subtypes. Clustering technique to determinate signal-to-noise ratio of Rhizophora spp. binderless and araldite resin particleboard as phantom material on computed tomography images The signal-to-noise ratio (SNR) is an important measure of the quality of computed tomography (CT) images. In this study, a new clustering method is proposed to calculate the SNR ratio of CT image. Multi- Objective Simulated Annealing clustering is used for the comparison based on segmentation parameters such as SNR ratio. Two samples are used in this study as phantom materials, namely, Rhizophora Spp. binderless and araldite resin particleboard, with dimension of 20 cm x 20 cm. For each scanned datum, ImageJ software is utilised as the combination method to analyse CT images. Results shows that the automatic clustering algorithm improves the SNR results of the sample images. In addition, the SNR value of images using MOSA clustering is higher than that of normal CT images. Introduction Computed tomography scan (CT) is a cross-sectional medical imaging technology that uses the computer to scan the inner parts of the body. Diagnostic and classification of the disease by CT imaging depends on morphological parameters, such as tissue, tissue attenuation and size [1]. Therefore, to better understand and use any image, the performance of the imaging system must be determined in terms of image quality. The image quality of the system is usually described as the image noise in terms of signalto-noise ratio (SNR). Phantoms are important tools for the quality assurance of diagnostic radiology and radiotherapy equipment. Previous studies assumed that Rhizophora spp. wood has properties comparable with those obtained from other standard phantom materials used for radiation dosimetry [2][3][4]. Marashdeh et al. [5] proposed that Rhizophora spp. wood be shredded into particles and compressed into binderless particleboards. Moreover, Marashdeh et al. [6] showed that the binderless particleboard phantom can be used for dosimetric measurements in kilovoltage X-ray beam dosimetry. In addition, the epoxy resin is often used in phantom fabrication with a mixture of compounds. A series of phantom materials that are equivalent to tissue based on epoxy resin is modified to mimic biological tissues [7]. To understand the imaging system for both phantom materials, the clustering method was used. Clustering is a data mining technique in the field of unsupervised datasets; this technique is used to explore and understand large collections of data [8]. The clustering technique distributes the dataset into clusters of similar features. The clustering has widespread applications in many fields, such as gene expression data [9][10],marketing [11] [12] and image processing [13] [14]. Image segmentation is one of the most important

techniques in image processing [15]. Clustering an image is a good technique used for the segmentation of images. The goal of clustering in image segmentation is to subdivide an image into different regions of certain properties and extract the desired parts [16]. Clustering has different types: hierarchical clustering [17], fuzzy C-means clustering [18] and K-means clustering [19]. In this study, automatic clustering algorithm called Multi-Objective Particle Swarm Optimization with Simulated Annealing (MOPSOSA) [20] was used for noisy image segmentation to obtain clearer images. This technique was applied on Rhizophora Spp. binderless and araldite resin particleboards (ARRP) as two phantom materials. Samples preparation Rhizophora spp. binderless particleboards were prepared in accordance with the method described by Marashdeh et al., [6]. Binderless particleboard samples with a particle size of < 50 μm were fabricated as described by Marashdeh et al. [21] at a moisture content of 5%-6%. The binderless particleboards had dimensions of 20 cm x 20 cm. Epoxy resin (Araldite MY 6010) and hardener (Jeffamine D-230) ratio of 3:1 were thoroughly mixed by a hand mixer for 30 min to ensure good homogeneity. The mixture was decanted into a 20 cm x 20 cm plastic Teflon-coated mould. The mixture underwent curing for 48 h. CT number measurements The phantom was scanned using a DECT scanner (SOMATOM Definition AS, Siemens 2014, Germany) to investigate the CT number inside the Rhizophora spp. binderless particleboard and ARRP samples. Nine square zones over CT image were considered as regions of interest (ROI) for each sample. ImageJ was used to calculate the mean and standard deviation of the scanned images at ROI. The coefficient of variation (COV) of the SNR were computed using the following equations: Image segmentation The CT images were segmented using automatic clustering algorithm MOPSOSA (figure 1). Matlab was used in coding the algorithm. The MOPSOSA algorithm is a hybrid K-means technique, multiobjective PSO, multi-objective SA, and sharing fitness. The main strategy of MOPSOSA algorithm starts with the K-means technique [19], which is used to improve the selection of the initial particle position (swarm) because of its significance in the overall performance of the search process. The Multi-Objective Particle Swarm Optimization (MPSO) is implemented, where all particles in the swarm are launched through the search space by following the current optimum particles to search for the best solution. The MPSO simultaneously deals with three different cluster validity indices, namely, DB-index [22], Sym-index [23] and Conn-index [24]. During the search, the Multi-Objective Simulated Annealing (MOSA) was used if no change occurred in the position of a particle, or if the particle moved to a bad position. The MOSA algorithm was merged to improve the performance of particles in the search process, as well as escape the stagnation and entrapment in local solutions. Through trade-off between the three different validity indices in the MOPSOSA algorithm, many Pareto optimal solutions were created. Therefore, the idea of sharing fitness [25] was incorporated to maintain diversity in the repository that contained optimal Pareto solutions. Optimal Pareto solutions are important for decision makers to choose from. Results and discussion Automatic clustering algorithm improved the SNR results of the sample images with different ratios based on variation in CT number of binderless particleboard and ARRP samples (figures 2 and 3). The results show the difference in noise levels by comparing standard deviation (SD) values in binderless particleboard and AARP samples images. The ImageJ calculated the image noise in CT images by measuring the mean SD values at nine ROIs. The range of calculated SD values were 19.62-31.59 and 19.61-29.79 for binderless particleboard and AARP sample images, respectively, before MOPSOSA clustering. By contrast, the range of calculated SD values were 17.65-20.71 and 16.07-20.78 for binderless particleboard and AARP sample images, respectively, after using MOPSOSA clustering. The results showed no significant variation between SD values and a slight decrease in image noise level after using MOPSOSA clustering. The CT images did not include different materials (lesions) such that no significant difference was observed between the pixel values when using MOPSOSA. Thus, MOPSOSA clustering can be useful and effective for images containing different lesion areas. Table 1 shows the results of the COV of the SNR results in the nine ROIs of binderless particleboard and ARRP sample were 8.18 and 8.98, respectively, after using MOPSOSA compared with 20.88 and 21.77 before clustering, respectively. In addition, ARRP sample showed slightly higher COV value compared with binderless particleboard sample. Therefore, the results indicated that binderless sample had a more homogenous profile density than the ARRP sample. Conclusion The automatic clustering algorithm improved the image quality. In addition, the binderless particleboard showed the least variation in CT number profile compared with ARRP samples. Thus, it has potential use as a tissue-equivalent phantom material in diagnostic imaging application. The role of functional data in interpreting the effects of genetic variation Progress in DNA-sequencing technologies has provided a catalogue of millions of DNA variants in the human population, but characterization of the functional effects of these variants has lagged far behind. For example, sequencing of tumor samples is driving an urgent need to classify whether or not mutations seen in cancers affect disease progression or treatment effectiveness or instead are benign. Furthermore, mutations can interact with genetic background and with environmental effects. A new approach, termed deep mutational scanning, has enabled the quantitative assessment of the effects of thousands of mutations in a protein. However, this type of experiment is carried out in model organisms, tissue culture, or in vitro; typically addresses only a single biochemical function of a protein; and is generally performed under a single condition. The current challenge lies in using these functional data to generate useful models for the phenotypic consequences of genetic variation in humans. ABSTRACT Progress in DNA-sequencing technologies has provided a catalogue of millions of DNA variants in the human population, but characterization of the functional effects of these variants has lagged far behind. For example, sequencing of tumor samples is driving an urgent need to classify whether or not mutations seen in cancers affect disease progression or treatment effectiveness or instead are benign. Furthermore, mutations can interact with genetic background and with environmental effects. A new approach, termed deep mutational scanning, has enabled the quantitative assessment of the effects of thousands of mutations in a protein. However, this type of experiment is carried out in model organisms, tissue culture, or in vitro; typically addresses only a single biochemical function of a protein; and is generally performed under a single condition. The current challenge lies in using these functional data to generate useful models for the phenotypic consequences of genetic variation in humans. Genome sequences are available for many individual humans, but these constitute only one level of data. The blueprints for cellular structure and behavior are not stored solely in the linear sequence of the genome, but embedded within the network of interacting molecular components. Via current sequencing technologies, we have become accomplished in determining genotype. One consequence is that, at the protein level, we can readily establish the differences in protein sequences present in any person or during any disease process. Such a difference may lead to a molecular phenotype, whereby a protein is changed in its activity, stability, localization, or other property. This molecular change can lead, in turn, to a cellular phenotype, whereby the cell may alter how it divides, responds to signals, repairs its DNA, or carries out some other process. Finally, the cellular change can lead to an organismal phenotype, such as cancer. However, the complex hierarchical network of interacting components makes it exceedingly difficult to predict with confidence the effect of a mutation on an organism's phenotype, even though a small change in the genome can have profound effects on development, morphology, behavior, and disease. The many attempts to computationally predict phenotype from genotype have led to remarkable improvements but have not achieved sufficient accuracy for use in a clinical setting. "Prediction" here is meant to include both the ability to accurately label individual variants as either deleterious or benign and the ability to combine individual variant effects into larger models of phenotypic effect, though most efforts to date have focused on the former type of prediction. These approaches use evolutionary conservation-based metrics (Grantham, 1974;Ng and Henikoff, 2001) or apply methods such as machine learning to combine diverse features of a variant, including conservation, amino acid chemical similarity, and surrounding sequence features into statistical models of variant effect (Adzhubei et al., 2010). These models do not predict risk for a specific disease but instead are trained to label variants as either deleterious or benign by using data either from large databases of individually annotated disease genes (e.g., ClinVar; Landrum et al., 2014), which are expected to be deleterious, or from lists of common variants or lists of fixed but recently derived human alleles, which both are expected to be benign (Kircher et al., 2014). The models are therefore limited by the quality and quantity of the labeled variants used during training and will perform poorly for diseases or for types of detrimental variants that are not well represented in the training data (Li et al., 2014). Most of these models ant's phenotypic effect size and its effect on reproductive fitness (Henn et al., 2015). When a disease strongly affects reproductive success, rare, large-effect variants are more likely to be the major cause of the disease (Agarwala et al., 2013). Even among variants identified by their association with disease, our understanding is poor. The Human Gene Mutation Database, which contains only such variants, is rapidly increasing beyond its current 140,000 mutations, and the vast majority have not been functionally characterized (Stenson et al., 2014). Determining the phenotypic consequences of all these variants one at a time using site-directed mutagenesis is obviously infeasible. However, there is a strong incentive to functionally characterize variants that are inaccessible to association studies, as many of them likely affect disease risk. The direct functional characterization of genetic variation is achieved through mutagenesis, but mutagenesis experiments have typically been low throughput, sometimes analyzing only a few mutations. These studies seek to uncover more about a protein's function or to characterize mutations observed in the context of human disease. However, for both of these goals, the utility of mutagenesis has been limited by its throughput. It is often unknown which substitutions will be most informative, and there are too many potential mutations in every gene and regulatory element to perform another experiment each time a new mutation is observed in humans. Deep mutational scanning (reviewed in Fowler and Fields, 2014) increases the throughput of mutagenesis studies such that tens of thousands of genotype-phenotype associations can be assayed in a single experiment. For most proteins, this approach can analyze all singleamino acid substitutions and a large number of double substitutions at once. Similar strategies have been developed for DNA regulatory elements and for RNA (Patwardhan et al., 2009) A deep mutational scan begins with the synthesis of a pool of mutants. The mutations can be introduced by "doped" oligonucleotides, in which random mutations are introduced during gene synthesis, by error-prone PCR, or by synthesis of designed oligonucleotide primers that introduce defined mutations (Fowler and Fields, 2014). The pool of mutations must be introduced into an expression system in which each mutant gene and its encoded protein are physically linked. Both cellular and phage-based expression can meet this requirement. Finally, a selective pressure is applied, unique to the protein function being assayed, such that the activity of the mutant proteins can be differentiated. For example, cells expressing the protein can be dependent on it for normal growth. Cells harboring detrimental mutants thus grow more slowly. This difference in growth rate can be measured by high-throughput sequencing. The steps involved include first sequencing the pool of mutants at some initial time point and determining the frequency of each mutant in the pool (number of reads for mutant X/total number of reads). Second, the pool of mutants is sequenced after a period of selective growth, with the frequency of each mutant again determined by sequencing. Third, the change in frequency of each mutant from the initial to the final time point is calculated. Mutants with higher growth rates will increase in frequency, and those with lower growth rates will decrease in frequency. The data

can be analyzed statistically by programs like Enrich (Fowler et al., 2011). Other methods for differentiating between mutants in the pool include linking protein function to the expression of an essential gene or to a fluorescent readout, in which case flow cytometry can be used to separate functional and nonfunctional mutants. Regardless of the details of the selective pressure, the approach leverages the ability of high-throughput sequencing to provide the frequency of each mutant in a large pool, thus greatly increasing the throughput of mutagenesis studies. also rely heavily on measures of evolutionary constraint that might not be constant at a genetic locus across the related sequences used in the alignment. For example, sites that have only recently in evolutionary history either become important or lost their importance for organismal function will lead to false negatives and false positives, respectively, when using distantly related species or sequences to estimate evolutionary constraint. In addition, the relationship between phenotypic effect and evolutionary constraint is complex, as many phenotypes, including some common diseases, might not have a large effect on reproductive success if they cause morbidity and mortality only later in life. Partly for these reasons, false-positive and false-negative rates are too high to use these predictions in the context of human disease (Dong et al., 2015). Even though the high standard of clinical utility has not been met, these models still perform well, hinting that a more accurate prediction of phenotype from genotype might be possible. To achieve this goal, we will need to experimentally measure the phenotypic effects of genetic variation in a diversity of cellular and environmental contexts. The resulting large data sets of functional measurements will serve as both input variables and as more dense sets of output variables in the training of statistical models, allowing us to predict more accurately the effect of genetic variation on a wide range of both molecular and organismal phenotypes. The accuracy of genotype-based predictions will always be constrained by the heritability of the predicted trait-the amount of phenotypic variation that is attributable to genotypic variation. For this reason, functional data on genetic variation will be most informative for highly heritable traits, and these are the traits that we should focus on first. CONSTRUCTING A GENOTYPE-PHENOTYPE MAP BY DEEP MUTATIONAL SCANNING Genotype-phenotype maps were classically derived from forward genetic screens, in which random mutations introduced into a genome yielded interesting phenotypes. The mutations were localized within the genome, and then the corresponding genes were identified, a process that could take months to years. Reverse genetics, in which defined genes are first mutated and the effects of the mutations are then tested, sped up the process of mapping genotype to phenotype. However, we now know that that there are hundreds to thousands of variants in each gene in humans alive today, with more than 97 million single-nucleotide variants catalogued to date (dbSNP build 144, www.ncbi.nlm.nih.gov/news/06-09-2015-dbsnp -build-144). There is no indication yet of a plateau in the discovery of new variants. Most human variation arose recently and is rare, with 99% of variants present in <1% of individuals (Exome Aggregation Consortium, 2015; Fu et al., 2013). Each person has, on average, around 40-90 de novo mutations (Kong et al., 2012;Besenbacher et al., 2015); thus every possible point mutation in the genome-or at least those compatible with life-may have occurred among the seven billion humans on the planet. Establishing the contribution of rare variation to disease is difficult with most existing methodologies. An examination of the power of whole-genome resequencing-association studies to identify rare variants of modest effect (1% of phenotypic variance explained) showed that both gene-based and single variant-based methods are significantly underpowered even with tens of thousands of individuals (Moutsianas et al., 2015). Worse, purifying selection keeps frequencies of most detrimental variants low, so these rare variants may be enriched for detrimental effects (Gibson, 2012;Tennessen et al., 2012). Whether rare, large-effect variants or common, moderate-effect variants are more likely to explain the heritability of common diseases depends strongly on the relationship between a vari-effect prediction algorithms. Thus useful, gene-specific information from assaying variants directly is not captured by models based largely on evolutionary data. Many potential variants have the capacity to be damaging in the context of cancer, as the model predicts that roughly 20% of the assayed mutations in the RING domain would show impaired HDR activity. As more of the variants seen in large-scale cancer sequencing projects are subjected to direct functional assessment, models that integrate these results should provide better predictions and highlight areas of incomplete understanding. Statistical models could be used to integrate these direct measurements with other evolutionary and functional data to predict deleteriousness. Alternatively, for molecular phenotypes that have strong correlations with disease, such as protein instability, functional scores from these assays could be used as the target of prediction. The validated disease variants currently used to train some variant effect prediction models are spread sparsely throughout the genome, and their collection is subject to ascertainment biases, resulting in models that are not likely generalizable. Direct functional measurements would serve as much more dense gold standards for training. The eventual objective is to advance beyond predicting only the molecular effects of single variants and to attempt to predict individuals' phenotypes from their genome sequences. To this end, direct functional measurements could be used alongside data from genotype-phenotype association studies to predict disease risk, prognosis, or treatment effectiveness. Several groups have recently used mixed linear models on genotype-phenotype association data from large association studies to estimate the total contribution of genetic variation across a large set of genetic variants to phenotypic variation (Yang et al., 2011). The resulting model, which takes into account the marginal effects of all variants simultaneously, explains more of the heritability of complex traits than the much smaller set of individually significantly associated variants and can be used to predict phenotype from the identity of genotyped single-nucleotide polymorphisms (SNPs). However, the prediction accuracy is still low, and the phenotypic variance that is accounted for is still much lower than the actual heritability, as there is sampling error when estimating the effect of each SNP, and this problem will likely require large increases in sample size to reduce, especially when many contributing variants are rare (Chatterjee et al., 2013). Deep mutational scanning data could potentially improve SNPbased prediction models by providing information as to how each SNP marginally adds to phenotypic variation. An additional set of parameters, one for each assay, could be added, with each parameter representing the contribution of the functional score to the phenotypic variation. These parameters would replace the corresponding parameters for the variance explained by SNP presence. In cancer-resequencing studies, the resulting model could make use of the presence of rare SNPs that would otherwise not contribute to the prediction. Other types of models, like random forests and neural networks, which can implicitly take into account interactions between the molecular function results, might yield even better predictions, providing that the prediction target population is sufficiently similar to the training population (Quang et al., 2015;Stephan et al., 2015). A major challenge in collecting the data will be the choice of phenotypes to assay. It will be essential to determine the types of data (e.g., stability-based, enzymatic, or expression-based) that are most predictive, so as to prioritize experimental efforts and avoid overfitting by adding unhelpful data. This goal could be facilitated by using methods that rank model inputs by their contribution to the model's accuracy (Saeys et al., 2007). GENOTYPE-PHENOTYPE MAPS IN DISEASE A compelling reason to create genotype-phenotype maps is that these data will be valuable in diagnosing, treating, and understanding disease. These maps seem especially critical in cancer, a disease characterized by the accumulation of novel changes in genotype. Large projects have coordinated the extensive sequencing and transcriptional analysis of thousands of samples from many of the most common and deadly types of cancer (McLendon et al., 2008;Cancer Genome Atlas Research Network, 2013). The rationale for these studies, which require the coordinated efforts of clinicians, pathologists, sequencing centers, and bioinformaticians, is threefold. First, finding genes frequently mutated in cancers sheds light on the mechanisms of tumorigenesis, with frequently mutated genes often clustering in a small number of pathways (Vogelstein et al., 2013). Second, the molecular profiling of many samples from one type of cancer, previously typed solely by histology, may reveal multiple subtypes with different clinical properties. For example, gene expression profiles are predictive of prognosis in breast cancer (Millar et al., 2009). Finally, molecular characterization may lead to the discovery of predictive biomarkers that can help guide treatment decisions. For example, cancers that express certain proteins respond better to treatment with drugs that inhibit these proteins (Weigel and Dowsett, 2010). In addition, mutations in drug targets may predict response to the drug (Traina et al., 2014). Despite these few successes, the search for new prognostic and predictive biomarkers, especially those personalized for individual patients, is complicated by the numbers of rare mutations in cancer and the difficulties in interpreting their effects. All cancers are heterogeneous mixes of clones, each with up to hundreds of de novo nonsynonymous mutations (Vogelstein et al., 2013). We currently try to assign causative mutations by finding genes or networks of genes associated with cancer progression or risk and using predictive tools to stratify mutations occurring in those genes. This approach can miss rare variants of large effect as well as variants that are pathogenic only when combined. Even a recurrent mutation may not be clinically important. For example, large numbers of mutations in a panel of genes recurrently mutated in skin cancers were found in histologically normal skin tissue, suggesting that a cancer-inducing environment or some combination of these mutations is necessary for complete transformation (Martincorena et al., 2015). Estimates that it may take more than a decade for tumor expansion to occur after the initiating driver mutation indicate that the multiple changes required for malignancy accumulate slowly (Yachida et al., 2010). However, tumor expansion likely occurs in a quick, punctuated manner without obvious intermediates between normal and bulk tumor cells (Navin et al., 2011), complicating efforts to determine which variants contribute to the tumor phenotype. It would clearly be useful in prioritizing drug targets and assessing risk to know the marginal fitness advantage conferred to a cancer by each genomic change. In this vein, deep mutational scanning experiments have advantages over association tests. First and foremost, the functional effect of all single-nucleotide variants can be assayed directly without regard to their frequency in the population. An analysis of the breast cancer susceptibility gene, BRCA1, measured the effect of thousands of mutations in the BRCA1 RING domain on two biochemical functions: binding to the BARD1 protein and E3 ubiquitin ligase activity (Starita et al., 2015). These functional scores were used as the inputs to model a third function, homology-directed DNA repair (HDR), which is difficult to measure in bulk but correlates best with cancer susceptibility. The model captures roughly 70% of the variation in HDR scores and greatly outperforms commonly used biological Finally, changes in cell type, temperature, nutrients, chemical messengers, and drug treatments, as well as the presence of other organisms, can also mask or unmask a phenotypic change caused by a mutation. Deep mutational scanning experiments performed under more than one condition have shown that these gene-by-environment interactions are common (Guy et al., 2014;Stiffler et al., 2015). It is still unclear how much gene-by-gene and gene-by-environment interactions contribute to phenotype. For some phenotypes, like the probability that a given splicing event occurs, epistasis between sequence features and cell type dependence are both important enough that accuracy suffers greatly when these interactions are not included in phenotype predictions (Xiong et al., 2015). In other cases, for example, in single-step enzymatic processes, we might expect less dependence on cellular context and less intergenic epistasis, though intragenic epistasis might still be significant. Deep mutational scanning experiments can functionally characterize all single mutants of a given target but only a subset of double and triple mutants, so errors in predictive models due to epistasis will likely not be overcome by brute force. However, it remains to be seen whether the effects of genetic and environmental interactions will have an appreciable impact on accuracy. CONCLUSIONS The genomic diversity of humanity is staggering, and the genomic diversity

within cancers greater still. Although only a small fraction of this variation may be relevant to any given disease phenotype, identifying these relevant variants will require functional data for all rare variations. High-throughput phenotyping experiments can score many thousands of genetic variants for their effects on a diversity of molecular properties, but it is unclear which of these properties and targets will be most informative when going from molecular to organismal phenotype. Perhaps the most daunting obstacle is the potential contribution of environmental and genetic interaction terms, especially higher-order interaction terms, to phenotype. By combining diverse types of functional data for genetic variants within a structured mathematical model, we can probe the contribution of molecular phenotypes to disease phenotypes directly, for both rare and common variants. Obtaining high-quality, reproducible functional measurements throughout the genome will require a large, coordinated effort across many laboratories, but the effort will be well placed if the result is a more interpretable genome. CONTEXT DEPENDENCE Mutations do not exist in a vacuum, making the prediction of phenotype from genotype even more difficult. Mutations interact with other mutations, and they interact with features of the environment, such that their effects often depend on context. Consider two mutations in the same gene. If the gene is implicated in a phenotype such as growth rate, then we might expect that the combined effect of the two mutations on growth rate could be found by adding their individual effects. When the actual growth rate differs from this expectation, then the two mutations display epistasis, described quantitatively as how much the double mutant's growth rate differs from the expectation. Several deep mutational scanning experiments have demonstrated such intragenic epistasis, both positive and negative (Araya et al., 2012;McLaughlin et al., 2012). However, these same experiments have shown that the general rule for most mutations in a protein is that they do not exhibit epistasis, with the additive model providing a good estimate of log-transformed double-mutant function. Epistasis can also occur between two separate genes that both affect the same phenotype; for example, the encoded proteins may interact with each other or take part in the same cellular process. Gene-by-gene epistasis is usually studied through gene deletion, as in synthetic lethality screens, but has not been systematically measured at the level of individual single-nucleotide variants. However, a huge number of such interacting variants might exist, given that a screen of 75% of yeast gene pairs found 170,000 genetic interactions affecting cell growth (Costanzo et al., 2010). There remain substantial technical hurdles to analyzing gene-by-gene interactions by deep mutational scanning, but these types of interactions have been probed via short hairpin RNA screens and have proven to be useful for predicting survival and drug response in cancer patients based on the genetic architecture of their tumors (Jerby-Arnon et al., 2014). This result supports the idea that the specific genetic variants unique to a given tumor might render it vulnerable to inhibition of other genes that interact with those variants, providing a first glimpse of the potential utility of predictive models that make use of a mix of functional and association data. Epistasis is not limited to deviations from the expected interactions of just two genetic changes. Higher-order epistasis can occur after taking into account epistasis at lower levels (Weinreich et al., 2013). The fitness of a triple mutant may deviate from expectation even if all pairwise epistasis scores are considered, and higher-order epistasis coefficients are in many cases as large or larger than pairwise epistatic coefficients (Weinreich et al., 2013). Instances of higher-order intergenic (as opposed to intragenic) epistasis have also been identified (Taylor and Ehrenreich, 2014). With more interacting components, such as those involved in common disease processes, there will be an intractable number of even third-order epistasis terms that contribute significantly to phenotype. In Drosophila, whole genetic interaction networks can change substantially based on genetic background (Chari and Dworkin, 2013). This complexity might seem discouraging, but it is unknown to what extent higherorder interactions play a role in observed human phenotypes. The extent to which epistasis affects phenotypic variation among humans or within tumors depends on the relative frequencies of the interacting sites, and there is some evidence that the contribution of epistasis, while significant when assayed directly, may play a smaller role in human phenotypic variation in complex disease (Hill et al., 2008). One estimate from the analysis of expression quantitative trait loci is that pairwise epistasis explains approximately one-tenth the amount of phenotypic variance that additive effects do (Hemani et al., 2014). Novel Clostridium thermocellum Type I Cohesin-Dockerin Complexes Reveal a Single Binding Mode* Background: In general, dockerins present two homologous cohesin-binding interfaces, which confer increased flexibility into cellulosomes. Results: The structure of two novel Coh-Doc complexes reveals a dockerin single-binding mode. Conclusion: Single-binding mode dockerins bind, preferentially, to cell surface cohesins. Significance: The dual binding mode is a property of cellulosomal dockerins. Protein-protein interactions play a pivotal role in a large number of biological processes exemplified by the assembly of the cellulosome. Integration of cellulosomal components occurs through the binding of type I cohesin modules located in a non-catalytic molecular scaffold to type I dockerin modules located at the C terminus of cellulosomal enzymes. The majority of type I dockerins display internal symmetry reflected by the presence of two essentially identical cohesin-binding surfaces. Here we report the crystal structures of two novel Clostridium thermocellum type I cohesin-dockerin complexes (CohOlpC-Doc124A and CohOlpA-Doc918). The data revealed that the two dockerins, Doc918 and Doc124A, are unusual because they lack the structural symmetry required to support a dual binding mode. Thus, in both cases, cohesin recognition is dominated by residues located at positions 11, 12, and 19 of one of the dockerin binding surfaces. The alternative binding mode is not possible (Doc918) or highly limited (Doc124A) because residues that assume the critical interacting positions, when dockerins are reoriented by 180°, make steric clashes with the cohesin. In common with a third dockerin (Doc258) that also presents a single binding mode, Doc124A directs the appended cellulase, Cel124A, to the surface of C. thermocellum and not to cellulosomes because it binds preferentially to type I cohesins located at the cell envelope. Although there are a few exceptions, such as Doc918 described here, these data suggest that there is considerable selective pressure for the evolution of a dual binding mode in type I dockerins that direct enzymes into cellulosomes. Protein-protein interactions play a pivotal role in a large number of biological processes exemplified by the assembly of the cellulosome. Integration of cellulosomal components occurs through the binding of type I cohesin modules located in a non-catalytic molecular scaffold to type I dockerin modules located at the C terminus of cellulosomal enzymes. The majority of type I dockerins display internal symmetry reflected by the presence of two essentially identical cohesinbinding surfaces. Here we report the crystal structures of two novel Clostridium thermocellum type I cohesin-dockerin complexes (CohOlpC-Doc124A and CohOlpA-Doc918). The data revealed that the two dockerins, Doc918 and Doc124A, are unusual because they lack the structural symmetry required to support a dual binding mode. Thus, in both cases, cohesin recognition is dominated by residues located at positions 11, 12, and 19 of one of the dockerin binding surfaces. The alternative binding mode is not possible (Doc918) or highly limited (Doc124A) because residues that assume the critical interacting positions, when dockerins are reoriented by 180°, make steric clashes with the cohesin. In common with a third dockerin (Doc258) that also presents a single binding mode, Doc124A directs the appended cellulase, Cel124A, to the surface of C. thermocellum and not to cellulosomes because it binds preferentially to type I cohesins located at the cell envelope. Although there are a few exceptions, such as Doc918 described here, these data suggest that there is considerable selective pressure for the evolution of a dual binding mode in type I dockerins that direct enzymes into cellulosomes. Biological nanomachines combining a range of complimentary enzyme activities are critical to cellular function. Cellulosomes are one of nature's most elaborate and highly efficient multienzyme complexes that deconstruct cellulose and hemicellulose, two of the most abundant polymers on Earth (1)(2)(3). Thus, cellulosomes play a major role in carbon recycling and provide an opportunity to explore the largely untapped energy provided by plant biomass, by the bioenergy and bioprocessing sectors. It is now well established that the complex physical and chemical structure of plant cell walls restrict their access to hydrolytic enzymes. Aerobic microorganisms that utilize plant biomass as a significant nutrient express extensive repertoires of degradative enzymes, primarily glycoside hydrolases but also lyases and esterases, which attack the structural polysaccharides of the plant cell wall. In contrast, microbial anaerobes, due to environmental selective pressures, have a lower protein producing capacity and organize enzymes into cellulosomes, which enhance enzyme synergy and substrate targeting (see Refs. 1 and 2 for review). The cellulosome of the thermophilic bacterium Clostridium thermocellum has been extensively explored (4,5). It consists of a large non-catalytic multimodular protein, termed CipA, that contains nine tandemly repeated type I cohesins that recognize type I dockerins located in the cellulosomal enzymes (6,7). Type I cohesins of CipA display a very high level of sequence identity. It was thus suggested that there is little discrimination by the dockerins and their protein receptors presented by the cellulosome scaffold (8). Primary scaffoldins, such as CipA, may also contain a C-terminal divergent type II dockerin that specifically recognizes type II cohesins located on the bacterium's envelope, thereby providing a mechanism for the cell surface attachment of cellulosomes (9). Thus, different cohesin-dock-erin (Coh-Doc) 5 specificities (in C. thermocellum type I and type II) are responsible for the correct assembly of the multienzyme complex (type I) and its direct attachment to the organism (type II), respectively. Structural studies on type I Coh-Doc complexes of C. thermocellum (10,11) and Clostridium cellulolyticum (12), a mesophilic bacterium that produces a cellulosome analogous to the former microorganism, provided insights into the molecular determinants of protein-protein recognition that mediate the assembly of these protein complexes. Dockerins fold into two ␣-helices and EF-hand calcium-binding loop motifs, each corresponding to one of the two duplicated segments (10,12). Thus, the structure of the N-terminal ␣-helix and EF-hand calcium-binding loop can be precisely superimposed over the equivalent structures at the C-terminal end, leading to an internal 2-fold symmetry in the dockerin molecule (11). The implications of this internal symmetry were realized when it was observed that type I dockerins present two cohesin binding surfaces because they can bind their cognate protein module either through the analogous N-or C-terminal ␣-helices (11). In C. thermocellum type I dockerins, residues that dominate the hydrogen bond network with cohesins are located at positions 11 and 12 of the calcium binding loop and are usually a Ser-Thr pair (10). When the dockerin is 180°reverse oriented, the equivalent residues (Ser 45 and Thr 46 ) in the C-terminal dockerin helix participate in cohesin recognition (11). The Ser-Thr dyad symmetry observed in C. thermocellum dockerins is replaced, in C. cellulolyticum, by hydrophobic residues, which accounts for the lack of affinity between protein partners from different species. The dockerin dual binding mode may reduce the steric constraints that are likely to be imposed by assembling a large number of different catalytic modules into a single cellulosome. In addition, the switching of the binding mode between two conformations may also introduce quaternary flexibility into multienzyme complexes, thus enhancing substrate targeting and the synergistic interactions between some enzymes, particularly exo-and endo-acting cellulases. Currently, it is unclear whether the dual-binding mode displayed by C. thermocellum and C. cellulolyticum dockerins is universal to all cellulosomal enzymes. The genome sequence of C. thermocellum ATCC 27405 encodes 72 polypeptides containing type I dockerin sequences. Alignment of the 72 dockerin sequences at the two ligand binding sites revealed a strong conservation of the amino acids that mediate cohesin recognition (particularly Ser 11 , Thr 12 , and a Lys-Arg motif at positions 18 and 19). Recently, we described the identification of four dockerins, of proteins Cthe_0435 (Cel124A), Cthe_0918, Cthe_0258, and Cthe_0624 (Cel9D-Cel44A), which deviate from the canonical C. thermocellum motifs at least in one of the cohesin binding interfaces (13). Here we describe the structure of two complexes in which two different type I cohesins are bound to these unusual dockerin modules. The data indicate

that a cohort of C. thermocellum type I dockerins display a single binding mode. The possible biological significance for the single binding mode displayed by these dockerins is discussed. Cloning and Expression DNA encoding type I dockerins of Cthe_0435 (Cel124A, residues 31-112) and Cthe_0918 (residues 1146 -1209) and type I cohesins of Cthe_0452 (OlpC, residues 108 -258) and Cthe_3080 (OlpA, residues 30 -177) were amplified by PCR from C. thermocellum genomic DNA using the thermostable DNA polymerase NZYDNAChange (NZYTech Ltd.) and primers described in Table 1. Genes encoding the type I dockerin modules of Cthe_0435 and Cthe_0918, here termed Doc124A and Doc918, respectively, were ligated into NdeI_BamHI-digested pET3a (Novagen). Genes encoding cohesin modules termed CohOlpC and CohOlpA, which derive from proteins Cthe_0452 and Cthe_3080, respectively, were ligated into NheI_XhoI-restricted pET21a (Novagen). Recombinant cohesins contained a C-terminal His 6 tag. To express the dockerin and the cohesin genes in the same plasmid, the recombinant pET3a derivative was digested with BglII and BamHI, to excise the dockerin gene under the control of the T7 promoter, which was subcloned into the BglII site of recombinant pET21a so that both genes were organized in tandem. Through this approach, it was possible to express both Doc124A and Coh452 and also Doc918 and Coh3080 in the same cells. Doc124A and Doc918 were also subcloned into pET32a vector (Merck) restricted with EcoRI and XhoI. Recombinant dockerins were expressed in fusion with thioredoxin to improve dockerin solubility and stability. OlpA and OlpC cohesins were also cloned into BglII-and EcoRI-digested pRSETa (Invitrogen). Mutant derivatives of both dockerins were synthesized (NZYTech Ltd.) with codon usage optimized for expression in Escherichia coli ( Table 2). The synthesized genes contained engineered EcoRI and XhoI recognition sequences at the 5Ј-and 3Ј-ends, respectively, which were used for subsequent subcloning into pET-32a (Merck), as described above. Protein Purification Cohesin-Dockerin Complexes-The Coh-Doc complexes CohOlpC-Doc124A and CohOlpA-Doc918 were expressed in E. coli Tuner cells, grown at 37°C to A 600 of 0.5. Recombinant protein expression was induced by adding isopropyl-␤-D-thiogalactopyranoside to a final concentration of 0.2 mM and incubation for 16 h at 19°C. The recombinant proteins were purified by immobilized metal ion affinity chromatography using Sepharose columns charged with nickel (HisTrap TM ). Fractions containing the purified Coh-Doc complexes were bufferexchanged, using PD-10 Sephadex G-25 M gel filtration col- Unbound Cohesins and Dockerins-Dockerins Doc124A, Doc918, and the respective mutant derivatives cloned in pET32a were expressed in E. coli Origami cells. CohOlpC and CohOlpA cloned in pRSETa vector were expressed in E. coli Tuner cells. Growth was performed at 37°C to midexponential phase (A 600 ϭ 0.5) in Luria broth. Recombinant protein expression was induced with 1 mM (Origami) or 0.2 mM (Tuner) isopropyl-␤-D-thiogalactopyranoside and incubation for 16 h at 19°C. The recombinant proteins were purified by immobilized metal ion affinity chromatography as described above and buffer-exchanged into 50 mM Na-Hepes buffer, pH 7.5, containing 2 mM CaCl 2 and then subjected to gel filtration using a HiLoad 16/60 Superdex 75 column (GE Healthcare) at a flow rate of 1 ml/min. Isothermal Titration Calorimetry (ITC) ITC experiments were carried out essentially as described previously (11,12), except that the titrations were at 55°C, and proteins were in 50 mM Na-HEPES buffer, pH 7.5, containing 2 mM CaCl 2 . During titration, the dockerin (40 M) was stirred at 300 rpm in the reaction cell, which was injected with 28 successive 10-l aliquots of ligand comprising cohesin (180 M) at 200-s intervals. Integrated heat effects, after correction for heats of dilution, were analyzed by non-linear regression using a single site-binding model (Microcal ORIGIN, version 5.0, Microcal Software). The fitted data yielded the association constant (K a ) and the enthalpy of binding (⌬H). Other thermodynamic parameters were calculated by using the standard thermodynamic equation, ⌬RTlnK a ϭ ⌬G ϭ ⌬H Ϫ T⌬S. Crystallization and Data Collection Protein crystals were obtained using the hanging drop, vapor diffusion method. CohOlpC-Doc124A complex crystals grew in 2 M ammonium sulfate, pH 4.6 (condition 32 of Crystal Screen HR2-110 from Hampton Research) in drops with 7 g/liter protein and were harvested after 5-7 weeks at 19°C. CohOlpA-Doc918 complex crystals grew in 0.2 M lithium sulfate, 10% (w/v) PEG 8000ϩ, 10% (w/v) PEG 1000, pH 7.5 (condition 14 of Clear Strategy Screen I MD1-14 from Molecular Dimensions) in drops with 16 g/liter protein and were harvested after 3-5 weeks at 19°C. Crystals were cryocooled with paratone in liquid nitrogen prior to data collection at beamline ID14 -2 at the European Synchrotron Radiation Facility (ESRF, Grenoble, France) at 100 K using an ADSC QUANTUM 4R CCD detector and at a wavelength of 0.9330 Å. Structure Determination and Refinement Structure determination of CohOlpC-Doc124A was based on two data sets that were processed in MOSFLM (14) merged and combined with SORTMTZ from the CCP4 suite (16) and scaled in SCALA (15). Phasing was performed by molecular replacement with the program BALBES (18) using a search model based on the Protein Data Bank structures 2ccl, 1aoh, 2vn6, 1nv8, and 1ixh, mainly related to cohesin and dockerin modules from C. thermocellum and C. cellulolyticum (11,12). Density modification, together with non-crystallographic symmetry averaging, was done with the DM program from the Novel Type I Coh-Doc Complexes from C. thermocellum CCP4 suite (16). ARP/wARP (19) was used to automatically build the protein model. Model completion, editing, and initial validation were carried out in COOT (21). Initial restrained refinement of the molecular model was done using REFMAC version 5.5 (22), and water molecules were added/validated according to the following criteria: a compatible water-shaped , whose center was within acceptable hydrogen bond distance to the closest protein atoms or other waters (ϳ2.4 -3.2 Å), and B-factors similar to neighboring atoms but less than 80 Å 2 . The final cycles of refinement were done with the program PHENIX.REFINE from the PHENIX suite (23). The two molecules in the asymmetric unit are arranged as a dimer of heterodimers, the later composed of chains A/B or C/D. All atoms in the protein could be properly assigned and refined, apart from a few initial (first 6 and 7 residues from chains A and C, respectively, and the first 6 residues from chains B and D) and final residues (last 6 and 7 residues from chains A and C, respectively, and the last 6 and 8 residues from chains B and D, respectively) in the polypeptide chains. The final model also includes 461 water molecules and four calcium ions. R-work and R-free converged to 18.1 and 21.5%, respectively. Model assessment and validation were carried out by PHENIX.POLYGON (24) and MOLPROBITY (25) from the PHENIX suite and PROCHECK (26). According to these programs, the final model contains 99.5% of the residues in most favored and allowed regions of the Ramachandran plot and 0.5% of the residues in generously allowed regions of the plot. Structure determination of CohOlpA-Doc918 was similarly done by molecular replacement with BALBES using structures with Protein Data Bank entries 2ccl and 2vn6 as models (11,12). Density modification with non-crystallographic symmetry was done with the PARROT (27) program from the CCP4 suite. ARP/wARP was used to automatically build the protein model. Model completion, editing, and initial validation were also carried out in COOT. The dockerin start model had to be manually rebuilt due to a mistracing error by ARP/wARP, originating from the presence of dockerin's two duplicated segments that share a striking sequence similarity and strong structural conservation. Refinement procedures were done as described for CohOlpC-Doc124A. R-work and R-free converged to 17.5 and 20.6%, respectively. Two chains were found in the asymmetric unit, arranged as a dimer (A/B). Protein residues could be properly assigned and refined, apart from the first 5 and last 10 residues from chain A and the first 2 residues from chain B. The final model includes 224 water molecules and two calcium ions. Model assessment and validation using the above-mentioned tools produced a final model with 100% of the residues in the most favored and allowed regions of the Ramachandran plot. Data collection and refinement details data for the two complete structures are summarized in Table 3. Expression and Crystallization of Novel Coh-Doc Complexes In a previous study (13), C. thermocellum type I dockerins were shown to bind to the nine type I cohesins of CipA and the single cohesin modules of the cell surface protein OlpA or OlpC (Fig. 1). The majority of C. thermocellum dockerins (68 of 72), exemplified by the well characterized dockerin of Xyn10B (10, Novel Type I Coh-Doc Complexes from C. thermocellum DECEMBER 28, 2012 • VOLUME 287 • NUMBER 53 11), display a distinctive internal symmetry that is compatible with a dual binding mode. These dockerins display preferential recognition for OlpA and CipA cohesins. In contrast, two C. thermocellum dockerins, from the protein of unknown function, Cthe_0258, and the recently described cellulase, Cel124A (28), display a 2-and 10-fold preferential binding, respectively, to the cell envelope cohesin of OlpC (13). A third dockerin, of the bifunctional cellulase Cel9D-Cel44A, displays two cohesinbinding interfaces with different specificities; the dockerin can interact with C. cellulolyticum cohesins through the N-terminal interface and with C. thermocellum counterparts through the C-terminal binding site (13). A fourth dockerin, from the protein of unknown function Cthe_0918, binds equally well to CipA and OlpA and displays a lower affinity for OlpC. The primary sequences of these four dockerins lack the distinctive symmetry at the binding interfaces, which may explain, at least for dockerins of Cthe_0258, Cel124A and Cel9D-Cel44A, the observed differences in ligand specificity (13). The dockerin dual binding mode does not favor crystallization of protein complexes, and the usual approach used to study the Coh-Doc interaction involves the inactivation of one of the dockerin cohesin-binding interfaces through site-directed mutagenesis (11,12). Because the four unusual dockerins described above seem to present a single binding interface, wild-type proteins were used for these structural studies. Here, we have used established strategies for the production and purification of Coh-Doc complexes, which involve the co-expression of both proteins in E. coli cells (29). Structure of Type I Coh-Doc Complexes The structures of OlpA type I cohesin bound to the dockerin of the protein Cthe_0918 (CohOlpA-Doc918) and of the OlpC type I cohesin in complex with the dockerin of Cel124A (CohOlpC-Doc124A) were solved to 1.95 and 1.75 Å resolution, respectively (Fig. 2). In C. thermocellum, OlpA and OlpC cohesins are the only two type I cohesins that do not belong to CipA and show significant deviations in the putative residues that participate in dockerin recognition, when compared with the nine highly homologous cohesins of CipA (13) (for details, see supplemental Fig. 1S). Both type I cohesin structures in complex with their respective protein partners reveal a striking similarity (supplemental Fig. 2SA). In comparison with the second cohesin of CipA (CohCipA2), a structure superposition with CohOlpA has an r.m.s. deviation of 1.04 Å (between 136 C ␣ pairs with a sequence identity of 35.3%) and 1.18 Å with CohOlpC (133 C ␣ pairs, 38.3% sequence identity). The two novel cohesins superpose with each other with an r.m.s. deviation of 1.14 Å (139 C ␣ pairs, 34.5% sequence identity). Noteworthy structural divergences occur between ␤-strands 4 and 5 (which include a small ␣-helix), where both CohOlpA and CohOlpC have a shorter loop than CohCipA2, and on the loop between ␤-strands 7 and 8 that, compared with CohCipA2, is slightly longer in CohOlpA and considerably larger in CohOlpC, increasing the main longitudinal axis length of these two proteins by around 2 and 10 Å, respectively (supplemental Fig. 2SA). The ␤-sheet B interface area, evaluated on the basis of its solvent-accessible area when in complex with its cognate dockerin (PDBePISA) (20), was 686 Å 3 for CohCipA2, 803 Å 3 for CohOlpA, and 729 Å 3 for CohOlpC. Structure of Type I Dockerins-The structures of dockerins of Cthe_0918 and Cel124A, here termed Doc918 and Doc124A, respectively, are organized in two ␣-helices, arranged in an antiparallel orientation (N-terminal or helix-1 and C-terminal or helix-3) connected through an extended loop displaying a small helix (helix-2) (Fig. 3). In Doc918, helix-1 is composed of residues 15-27, helix-2 extends between residues 35 and 39, and helix-3 extends from residue 48 to 60. In Doc124A, the respective residues are as follows: helix-1 (residues 17-29), helix-2 (residues 39 -45), and helix-3 (residues 53-65). Helix-2 connects the other two helices, both

of which provide the two putative cohesin binding interfaces. In fact, this region, limited by the distal end of helix-1 and the C terminus of helix-2, contains a large amount of the structural variability found among the core C␣ trace of these dockerins. In Doc918, the region connecting the two helices is less structured than in DocXyn10B, presenting a single turn on its ␣-helix, similar to a type I C. cellulolyticum dockerin (12). The internal sequence duplication and nearly perfect 2-fold symmetry was quantified by an internal superposition between helix-1 and -3 within each structure. Doc918 shows an r.m.s. deviation of 0.57 Å for 23 C ␣ pairs, and in Doc124A, both segments overlap almost as well, with an r.m.s. deviation of 0.66 Å for 26 C ␣ pairs. Lack of conservation in the key contacting residues when the two putative binding surfaces are compared should prevent a dual binding mode, which is explored below. Both dockerins contain two Ca 2ϩ ions coordinated by several residues in the canonical EFhand calcium-binding loop. The coordination of the two cal- Novel Type I Coh-Doc Complexes from C. thermocellum DECEMBER 28, 2012 • VOLUME 287 • NUMBER 53 JOURNAL OF BIOLOGICAL CHEMISTRY 44399 cium ions is similar to the metal ions observed in the type I dockerins of C. thermocellum and C. cellulolyticum in complex with their cognate protein partners (supplemental Fig. 3SA). Novel Type I Coh-Doc Complex Interfaces In contrast with what was previously observed for other type I complexes, the dockerins described here in complex with their protein partners seem to present a single binding mode. Thus, in the CohOlpA-Doc918 complex, binding is dominated by the Doc918 C-terminal helix. In contrast, in the CohOlpC-Doc124A complex, binding is orchestrated by the dockerin N-terminal helix (Fig. 2). In these two novel Coh-Doc structures, the complex interface has a significant hydrophobic nature. Using the solvation free energy gain at complexation, calculated by PDBePISA (⌬ i G in kcal/mol (20)), the CohOlpA-Doc918 interaction is more hydrophobic (Ϫ10.6 kcal/mol) than that of CohOlpC-Doc124A (Ϫ7.7 kcal/mol), which in turn exceeds the CohCipA-DocXyn10B value of Ϫ6.4 kcal/mol. However, the negative values upon binding are less significant than those of the highly hydrophobic C. cellulolyticum type I complex (Protein Data Bank code 2vn6) with Ϫ14.9 kcal/mol. These differences reflect the numerous hydrophobic residues, involved in the Coh-Doc complex interface, enumerated in detail in supplemental Table 1S and highlighted in supplemental Figs. 1S and 3S. Thus, the numbers of cohesin and dockerin hydrophobic residues implicated in the interface of the CohOlpA-Doc918 are greater than in the CohCipA2-DocXyn10B complex. Although the hydrophobic contact network of CohOlpC-Doc124A is also extensive, the hydrophobic residues that contribute to the heterodimer interface are contributed primarily by Doc124A. The major hydrophobic contact residues located at the surface of cohesins CipA2, OlpA, and OlpC include a completely conserved leucine (Leu 83 , Leu 83 , and Leu 92 , respectively), which is assisted by upstream hydrophobic residues Val 81 , Ala 81 , and Val 90 and downstream by Ala 85 , Leu 85 , and a divergent Asp 94 in OlpC, respectively. Other important contributors correspond to Leu 129 , Met 132 , and Leu 146 , respectively. With respect to the dockerins Xyn10B-␣3/Xyn10B-␣1, 918, and 124A, the major hydrophobic contact residues are Leu 22 /Leu 56 , Leu 27 , and Leu 65 , respectively, at position 22 of the less interacting binding interface. In addition, in position 15 of the dominating interface, residues Leu 49 /Thr 15 , Leu 53 , and Val 22 make a significant contribution to cohesin recognition. The above mentioned conserved leucine located at the surface of the three cohesins is part of an important hydrophobic pocket formed in CohCipA2 by Ala 72 , Tyr 74 , Val 81 , and Leu 83 , which is occupied by Leu 22 or Leu 56 from DocXyn10B in the two possible binding modes, respectively. Using the same relative structural positioning order, for CohOlpA, we find Asn 72 , Ala 81 , and Leu 83 , which accommodate Leu 27 from Doc918. As for CohOlpC, residues Asn 81 , Val 90 , and Leu 92 form a hydrophobic pocket that is occupied by the equivalent Doc124A residue, Leu 65 , found in the opposite C-terminal interface. The heterodimer interfaces are assisted by a network of direct and bridged hydrogen bonds and salt bridge interactions (described in detail in Table 4). Compared with DocXyn10B(␣3) in a similar C-terminal binding conformation (10), Doc918 reveals a more imbalanced distribution of polar bonds, favoring helix-3 residues. Although the Ser/Thr dyads of both complexes share an equivalent contribution, the main difference occurs at the Lys 56 /Lys 57 pair of Doc918 that contribute with one salt bridge and two direct H-bonds, whereas in DocXyn10B(␣3), the equivalent Ser 52 makes no polar bonds, and Arg 53 establishes a single salt bridge. Again, in comparison with the N-terminal bound DocXyn10B(␣1), Doc124A reveals some striking differences with respect to the relevant Ser-Thr pair, which is replaced by a divergent Ile 18 -Ser 19 motif. In Doc124A the N-terminal binding face interacts with CohOlpC, through significant hydrophobic contacts. The only direct polar interactions mediated by helix-1 occur via positions 18 and 19 (Lys 25 -Arg 26 ), through six direct bonds (two salt bridges from Lys 25 and four H-bonds from Arg 26 ) and a couple of water bridged H-bonds involving Ile 18 . In contrast, the N-terminal bound configuration of DocXyn10B reveals a hydrogen bond network around, and dominated by, the conserved Ser-Thr pair and also some involvement of residues 18 and 19. Also in contrast to DocXyn10B(␣1), Doc124A presents in the opposite interface (␣3) a stronger polar contribution participated in by four residues, Lys 61 (one salt bridge), Leu 64 (one H-bond), Leu 65 (one H-bond), and His 66 (one salt bridge), whereas in DocXyn10B, Leu 56 and mainly Arg 57 make polar contacts with the CipA cohesin (Fig. 3, A and C; detailed contacts in supplemental Figs. 1S and 3S). Extending the comparison of Doc124A to C. cellulolyticum type I complex (2vn5/2vn6) in an analogous Network of polar interactions in novel type I Coh-Doc complex interfaces binding conformation, the major difference consists of a much subdued polar interaction network found in the latter, especially at the ␣3 interface, where only positions 22 and 23 reveal direct contacts (12). The cohesin-interacting residues can be grouped into three regions corresponding to ␤-strands ␤3, ␤-strands ␤5/␤6, and the loop between ␤-strands ␤8 and ␤9 ( Fig. 3B; detailed contacts in supplemental Fig. 1S). Around the ␤3 region, the important interactions are quite similar among CohCipA2/ CohOlpA, because equivalent residues Asn 37 /Ser 39 and Asp 39 / Asp 41 , respectively, establish relevant polar contacts with the dockerin Ser/Thr pair. Conversely, the equivalent CohOlpC residue Ser 48 does not display any polar contacts, and Asn 50 , equivalent to CohCipA2 Asp 39 , establishes a single H-bond with the dockerin. In the ␤5/␤6 cohesin region, notable differences between CohCipA2, CohOlpA, and CohOlpC occur, respectively, at Arg 77 , Asp 77 , and Asp 86 residues; Arg 77 makes an H-bond with its target dockerin, whereas the equivalent acidic residues of the other two cohesins are not implicated on the interface. In the ␤8-loop-␤9 region, the corresponding residues Asn 127 , Asn 130 , and Phe 144 in CohCipA2, OlpA, and OlpC, respectively, reveal some differences in their capacity to recognize the dockerin protein partner. In a helix-3-dominated binding, CohCipA2-Asn 127 does not exhibit any contacts with its dockerin, whereas CohOlpA-Asn 130 makes two bridged H-bonds. However, in a helix-1-dominated binding, CohCipA2 uses its Asn 127 to make two H-bonds with Doc-Arg 19 , whereas in CohOlpC, the backbone of Phe 144 establishes two H-bonds with Doc-Arg 26 . In addition, in the Glu 131 /Glu 134 /Pro 148 position of the cohesins, both acidic residues from CohCipA2 and CohOlpA form an H-bond with the critical threonine found at position 12 of the dockerin, whereas CohOlpC-Pro 148 does not contribute to dockerin recognition. Further analysis of the differences between the canonical type I cohesin and this work on cell-bound cellulosomal cohesins was based on the predicted negative hydrogen bondaccepting regions in an electrostatic surface potential evaluation using the Poisson-Boltzmann electrostatics calculation on the PDB2PQR server (30) and visualization of the results in UCSF Chimera (31) (Fig. 4). As reported previously (32), cohesins are strikingly negatively charged in the binding interface plateau, whereas dockerins present a suitable complementary positive-to-neutral surface. Compared with CohOlpC and CohCipA2, CohOlpA shows an elongated polar region that extends beyond the binding interface. As described for the type II cohesin of SdbA (32), the opposite cohesin surfaces in CohOlpC and CohOlpA are more positively/neutrally charged, which was suggested to be important to promote a tighter interaction of cell surface cohesins to the negatively charged peptidoglycan layer. Analysis of the type I Coh-Doc interfaces provides significant insights into the previously described tight binding of Doc124A to CohOlpC, in comparison with the lower affinity displayed by this dockerin toward the cohesins of CipA and OlpA (13). The hydrophobic nature of Ile 18 at the critical position 11 of the Doc124A interface, establishes a strong network of apolar contacts with CohOlpC, namely with Asn 50 , Asn 140 , and Cys 142 . These CohOlpC pivotal residues are replaced by an aspartate (position 50) and by small residues, namely glycine or alanine, at the other two positions in CohCipA and CohOlpA cohesins. The aspartate residue, equivalent to CohOlpC-Asn 50 , found in both OlpA and CipA cohesins is also highly relevant for the recognition of typical type I dockerins because, together with Asn 37 , it establishes conserved hydrogen bonds with the canonical serine residues usually found at position 11. In CohOlpC, the latter residues are replaced with Ser 48 and Asn 50 , respectively, whose side chains were found more than 4 Å apart and thus presumably unavailable for H-bond formation. In addition, dockerin position 12 of the binding interface makes some relevant polar contacts with the mentioned residue of CohCipA2-Asn 37 and also Glu 131 . Again, in CohOlpC, the equivalent Ser 48 side chain orientation is unsuitable for those contacts, and the Pro 148 , which substitutes Glu 131 , is manifestly non-reactive. Probing the Importance of Contact Residues in Dockerins To identify the dockerin residues that are involved in cohesin recognition, a mutagenesis study informed by the previously described type I complex structures was implemented. Previous data suggest that the implications of single changes in dockerin activity may be relatively modest, so the strategy used here involved the change of particular groups of residues that are believed to play a cooperative role in cohesin recognition (10 -13). CohOlpA-Doc918 Complex-Site-directed mutagenesis and ITC data of the CohOlpA-Doc918 complex ( Fig. 5 and Table 5), show that the replication of the relevant residue environment from dockerin helix-1 into the C-terminal helix-3, which dominates ligand recognition (mutant Doc918_m1: S49D, T50E, and K57N), precluded any binding, which reinforces a vital role for the Ser-Thr motif, similar to the canonical type I Coh-Doc interaction (10). The drastic decrease in affinity obtained with mutant Doc918_m3 (S49Q and T50Q) also supports a major role for the C-terminal dockerin Ser-Thr motif. The K57N mutation also emphasizes the relative importance of a basic residue, such as lysine or arginine (equivalent to Arg 53 in DocXyn10B), in this position for efficient binding. The Doc918_m2 mutant design provided additional insights into the pivotal residues mediating cohesin recognition. Essentially, using the inactive dockerin, Doc918_m1, an attempt was made to force the alternate helix-1 binding mode. Thus, the non-functional N-terminal helix of Doc918_m1 was engineered to restore an N-terminal cohesin-binding interface by introducing the three pivotal residues (mutant Doc918_m2: D16S, E17T, and N24K) identified in helix-3 in the corresponding positions in helix-1. ITC data showed that, although with a 10-fold reduction in affinity, this strategy indeed allowed binding through helix-1. In CohOlpC-Doc124A and CohCipA2-DocXyn10B (i.e. the S45A/T46A mutant of DocXyn10B) (11), where helix-1 dominates the binding interface, there is a bulky positively charged residue in helix-3 (His 66 and Arg 57 , respectively), which provides polar and hydrophobic interactions to the interface but which was replaced by a Gly 61 in Doc918 when binding was engineered at the N-terminal

face (Fig. 3). This divergent substitution could thus contribute to the reduced affinity displayed by the Doc918_m2 mutant. Overall, the data presented here confirm that Doc918 presents a single proteinbinding interface that is dominated by the C-terminal helix and where Ser 49 , Thr 50 , and Lys 57 dominate cohesin recognition. CohOlpC-Doc124A Complex-As described above, the Doc124A Lys 25 -Arg 26 pair dominates the polar binding network with OlpC cohesin, whereas Lys 61 makes an important salt bridge with Asp 79 present at the surface of the cohesin. Thus, Doc124A mutants m1 and m2 were used to explore the importance of helix-1 Lys 25 -Arg 26 and helix-3 Lys 61 -Arg 62 pairs, by mutating them separately (m1) or simultaneously (m2) to Ala ( Fig. 6 and Table 5). As expected, based on these multiple polar contacts, the lesion in helix-1 (m1) caused an ϳ400-fold decrease in affinity. In addition, the additive effect of mutating the two basic pairs at helix-1 and helix-3 simultaneously (m2) led to complete loss in cohesin recognition, confirming the importance of Lys 61 in heterodimer formation. Thus, the basic pair at helix-1 plays a key role in cohesin recognition, and the massive reduction in affinity suggests a single binding mode for Doc124A. However, because the helix-3 Lys 61 -Arg 62 pair is in a position symmetry-related to that of Lys 25 -Arg 26 in helix-1, it is also possible that, following a 180°rotation of the dockerin, these latter residues could participate in a lower affinity cohesin recognition mediated by helix-3. Under these circumstances, the lower affinity of m1 would result from substitution of the critical Ile by an Asp at position 11 and by the loss of a putative Lys 25 -mediated salt bridge at the other helix. Data presented above suggest that Doc124A could eventually present two cohesin binding interfaces expressing different affinities. To explore this possibility, Doc124A Ile 18 , Val 22 , and Leu 23 , which are part of the hydrophobic platform of the helix-1 binding interface, were mutated to replicate their symmetryrelated counterparts in helix-3 (m3) (Fig. 6 and Table 5). The data revealed that these mutations lead to a reduction in the capacity of Doc124A to bind its cohesin partner. The Doc124A_m4 mutant introduces into the m3 background, in which helix-1 binding is reduced, the mutations D54I, N58V, and Y59L, with the intention of promoting a reversal in binding through the C-terminal helix ( Fig. 6 and Table 5). ITC results show an 8-fold increase in affinity over the m3 mutant, similar to the wild type dockerin, suggesting that although a dual binding mode is not feasible in the native form of Doc124A, in the m4 mutant, binding is probably dominated by the C-terminal interface. Thus, overall, the data suggest that Doc124A presents a single binding mode driven by helix-1. The importance of the hydrophobic network established between Doc124A and OlpC was further explored in the mutant m5, which investigated the role of a second residue pair, Leu 64 -Leu 65 , in the interactions established with the cohesin (Fig. 6 and Table 5). As described above, the Doc124A dockerin presents a symmetry-related pair at helix-1, Ile 28 -Leu 29 , which could be involved in a similar interaction if binding was mediated by the helix-3 lower affinity interface. The importance of this pair was explored in m6. The knock-out of the Leu 64 -Leu 65 helix-3 pair (m5) induced a 10-fold decrease in affinity, confirming the relevance of these residues in binding the cohesin when helix-1 is the dominant binding face. Indeed, it is reasonable to assume that the loss of Leu 64 and Leu 65 in m5 reorientates the major binding face to helix-3. Consistent with this view is the further reduction in affinity by the concurrent mutation of the proximal helix-1 residue pair (m6). CONCLUSIONS The structure of two type I Coh-Doc complexes presented here revealed that unlike the large majority of C. thermocellum dockerins, the dockerins of cellulase Cel124A and of Cthe_0918 protein, presently of unknown function, display a single cohesin-binding surface. The structures of the two dockerins were solved in complex with the two unique cell surface type I cohesins of C. thermocellum, OlpA and OlpC, which direct plant cell wall hydrolytic enzymes directly to the cell surface. A recent study (33) revealed that cellulosomes act in synergy with enzymes located at the bacterium cell envelope, which include the abundant Cel124A endocellulase that targets cellulose crystalline-amorphous junctions. The fact that high quality crystals for both complexes were obtained using wild type dockerins was an initial good indication that these dockerins present essentially a single interacting surface. The structures of the two complexes revealed that the critical positions 11 and 12 of the dockerin non-interacting interface are occupied predominantly with acidic residues (Glu and Asp). Acid residues are not suitable for interacting with the highly negatively charged cohesin platform. Site-directed mutagenesis data demonstrate the importance of the Ser/Ile-Thr motif at positions 11 and 12 and the Lys/Lys-Arg pair at positions 18 and 19 in cohesin recognition. Inspection of the primary sequences of dockerins of Cthe_0258, which recognizes OlpC with higher affinity, and cellulase Cel9D-Cel44A, which binds both C. thermocellum and C. cellulolyticum cohesins, also revealed unsuitable substitutions at one of the dockerin binding faces, which should result in only one binding face capable of recognizing C. thermocellum cohesins. It is presently unclear why a subset of four dockerins, the two described here and those from Cthe_0258 and Cel9D-Cel44A, have not evolved the dual binding mode characteristic of the other 68 C. thermocellum cellulosomal enzymes and extensively described for Xyn10B dockerin. Whereas the Cel124A dockerin directs the appended enzyme to the cell surface, because it binds predominantly to the OlpC cohesin, the dockerin of Cel9D-Cel44A is believed to present two cohesin binding interfaces with different cohesin specificities (the N-terminal face binds C. cellulolyticum-like cohesins, and the C-terminal interface binds their C. thermocellum counterparts). Thus, together, these data suggest that a dual binding mode is of primary importance for enzymes binding CipA, the multimodular cohesin scaffolding responsible for cellulosome assembly in C. thermocellum. An exception to this general rule is the dockerin of Cthe_0918, which recognizes CipA cohesins with higher affinity. The elucidation of the functional role of the protein domain appended to Cthe_0918 dockerin would help to clarify this issue. Nevertheless, the presence of two cohesin binding interfaces in dockerins integrated in multienzyme complexes may contribute to the capacity of the cellulosome to adjust its catalytic machinery to a highly insoluble and recalcitrant substrate. How Therapists Use Visualizations of Upper Limb Movement Information From Stroke Patients: A Qualitative Study With Simulated Information Background Stroke is a leading cause of disability worldwide, with upper limb deficits affecting an estimated 30% to 60% of survivors. The effectiveness of upper limb rehabilitation relies on numerous factors, particularly patient compliance to home programs and exercises set by therapists. However, therapists lack objective information about their patients’ adherence to rehabilitation exercises as well as other uses of the affected arm and hand in everyday life outside the clinic. We developed a system that consists of wearable sensor technology to monitor a patient’s arm movement and a Web-based dashboard to visualize this information for therapists. Objective The aim of our study was to evaluate how therapists use upper limb movement information visualized on a dashboard to support the rehabilitation process. Methods An interactive dashboard prototype with simulated movement information was created and evaluated through a user-centered design process with therapists (N=8) at a rehabilitation clinic. Data were collected through observations of therapists interacting with an interactive dashboard prototype, think-aloud data, and interviews. Data were analyzed qualitatively through thematic analysis. Results Therapists use visualizations of upper limb information in the following ways: (1) to obtain objective data of patients’ activity levels, exercise, and neglect outside the clinic, (2) to engage patients in the rehabilitation process through education, motivation, and discussion of experiences with activities of daily living, and (3) to engage with other clinicians and researchers based on objective data. A major limitation is the lack of contextual data, which is needed by therapists to discern how movement data visualized on the dashboard relate to activities of daily living. Conclusions Upper limb information captured through wearable devices provides novel insights for therapists and helps to engage patients and other clinicians in therapy. Consideration needs to be given to the collection and visualization of contextual information to provide meaningful insights into patient engagement in activities of daily living. These findings open the door for further work to develop a fully functioning system and to trial it with patients and clinicians during therapy. Introduction Stroke is the leading cause of acquired adult disability in high-income countries [1], with upper limb deficits affecting an estimated 30% to 60% of survivors [2,3]. Stroke causes damage within the brain that, when affecting somatosensory circuitry, lead to difficulties sensing and controlling movement of the body's contralateral side. Due to these limitations, stroke patients tend to reduce the utilization of the affected limb, which may cause muscle shortening and weakness, thus further compromising arm functionality [4]. As a result, performance in basic activities of daily living (ADL) such as eating, bathing, and dressing can be heavily affected, impacting on a patient's independence, social engagement, quality of life, and well-being [5]. Therapists (occupational therapists and physiotherapists) deliver effective upper limb rehabilitation interventions in hospitals. Interventions generally start by setting goals that target meaningful activities (eg, use of cutlery), functional movements (eg, grasp and retrieve objects), or specific impairments (eg, muscle weakness). Training is often task-specific and involves practicing tasks relevant to daily life. Along with this training, therapists employ a variety of techniques to support rehabilitation, such as mirror therapy, muscle electrical stimulation, strength training, stretching and positioning, mental practice, robotics, and virtual reality applications [4,[6][7][8]. Since therapy time is limited, the use of the affected arm in between sessions is crucial for enhancing functional outcomes. Therapists generally prepare daily exercise routines considering a patient's personal goals, or they utilize constraint-induced movement therapy to encourage patients' use of the affected arm in daily life [4]. Although the use of activity diaries such as the Motor Activity Log (MAL) allow determining compliance with therapy when not in the clinic, these are subject to various biases including the ability and motivation of patients and caregivers to provide accurate information [9]. The lack of objective information is particularly concerning because adherence to rehabilitation programs at home is often low due to lack of motivation, musculoskeletal issues, and fatigue [10]. Wearable sensor technology offers potential to provide therapists with objective information about a patient's arm movement in everyday life. Specifically, inertial measurement units (IMUs) appear promising, because these sensors can be embedded in wristbands, gloves, or garments, and thereby track changes in the acceleration and orientation of the affected arm. Various studies in controlled settings show that IMUs can track arm, hand, and finger movements [11][12][13][14]. This line of research is typically focused on technical challenges (ie, the accuracy of motion tracking [12,15]), reliability of tracking over long periods of time [16], wearability for patients [17], and the processing of metrics from sensor data [18]. While all of these issues are important to realize the potential of wearable sensor technology, to date there has been little consideration for the needs of therapists and whether this information is useful for the rehabilitation process. The aim of this research is to explore the information needs of therapists in order to help them understand how patients use their arm in everyday life in between rehabilitation sessions. In particular, this research seeks to address how therapists use visualizations of upper limb information presented on a dashboard to support therapy. A dashboard in this sense refers to a visual display of information on a computer screen. Similar to a car dashboard, the information on a digital dashboard needs to be compact to be monitored at a glance, to help people achieve one or more objectives [19]. Since neither wearable sensors nor dashboards are readily available, we conducted a design-driven investigation where we built a dashboard prototype that visualizes arm movement information, and we evaluated this Web-based prototype in a qualitative study with therapists. Based on a qualitative analysis we discuss the potential uses of these visualizations and

identify areas for improvement. Dashboard Design Process The dashboard design process is part of a larger research project into the development of a system to monitor upper limb movement of stroke patients in everyday life. The envisioned system consists of (1) wearable sensor technology that patients wear on their arm over several weeks to monitor upper limb data in everyday life; and (2) a dashboard to present the sensor data to therapists for use in consultations with patients. A wearable sensor prototype has been evaluated in a movement laboratory to establish the feasibility of this approach [20]. The prototype captures motion of the arm through IMUs placed at the wrist, above the elbow, and at the shoulder. From these sensors, motions in three degrees of freedom in the shoulder (adduction/adduction, flexion/extension, internal/external rotation), one in the elbow (flexion/extension), and one in the wrist (pronation/supination) can be calculated. The current system is not capable of capturing wrist extension or finger movements. The project team is now working on a sensor prototype that is comfortable to wear and robust enough for use in everyday life. We designed a dashboard prototype that visualizes sensor data to support therapists in their consultations with patients. The prototype was created through a user-centered design process, a standard approach in the field of human-computer interaction, to ensure that the dashboard that is being developed meets the needs of users [19,21]. The design process started with informal interviews with 3 occupational therapists (OTs) to understand the problems faced by therapists and the need for objective information. Based on these insights, 3 rounds of design workshops were conducted to generate and review ideas for information and visualizations that could be useful to support the work of therapists. These workshops involved 2 OTs, 1 physiotherapist, 2 mechanical engineers, 2 experts on wearable technology, and 2 interaction design researchers. As is common in a user-centered design process [22], ideas were initially sketched on paper for review and discussion. For the second and third workshops these sketches were refined as paper prototypes and digital prototypes. The final dashboard prototype was built with the prototyping software Axure, which supports the implementation of interactive Web-based prototypes without requiring software development skills. The strengths of such a prototyping approach are that they capture the key ideas of the entire team, allow quick evaluation and iteration, and facilitate discussion about relevant information and visualizations before effort is spent on developing the actual software [22,23]. Dashboard Prototype We developed an interactive dashboard prototype to gather feedback from therapists on the usefulness of various upper limb visualizations before a fully functioning system is implemented. As illustrated in the following figures, the prototype was designed in a sketchy manner to invite feedback, and to avoid giving the impression that this was a fully functioning website. The dashboard prototype evaluated in this study contained upper limb movement information for each patient (Textbox 1). This information was based on interviews and design workshops with therapists, as well as related work on kinematic measures for upper limb movements [18]. Related work shows that inertial sensors can provide information on the amount of arm movement and time spent using the arm in daily life [24]. Quality of movement and range of motion (ROM) are typically generated through robotic technologies or opto-electronic systems [18]. These systems can provide more precise measurements than inertial sensors, but they rely on a controlled environment and hence are not readily available for daily life use. Part of the information displayed on the website was based on sensor data collected in a movement laboratory [20]. We created additional fictional information in consultation with therapists to ensure that the information presented on the dashboard is complete and realistic for a stroke patient. The following figures show how this information was presented on the dashboard through 5 screens, which support different views and analysis of the various data. Overview Page The first page provides an overview of a patient's upper limb information ( Figure 1). It includes a brief patient profile, showing age, affected arm, dominant arm, and date of incident. An overview is provided of key movement information, including a tabular summary of number of movements overall, quality of movement, and time active. The therapists in the design workshops wanted both information about averages and for particular time periods. Furthermore, a timeline shows the number of movements over the last week, and the quality of movement on a scale from 1 (low quality) to 10 (high quality). The visualizations here were inspired by related work [19] and commercial dashboards of activity trackers (eg, Fitbit, Jawbone Up). Therapists can add notes. This is important as patients are usually seen by multiple therapists in the course of their therapy. Timeline Page The timeline page, which provides detailed movement information at two different time scales is shown in Figure 2. The timeline on the top presents movement patterns over long periods of time, from several hours to several days. The data presented here shows the level of activity, for example, 50% means that the arm is moved for 5 minutes during a 10-minute window. This information was included to provide therapists with a quick snapshot of how active patients are throughout a day. Therapists can annotate this data by dragging and dropping tags like "exercising" and "eating" to the activity timeline. The timeline on the bottom of the page presents movement for each degree of freedom over several seconds. The red progress bar connects the two time lines. This information was included so that therapists can explore movement in more detail and obtain insights into the quality of movement. For example, they can select a data point in the activity timeline (on top of the page) from a period of exercising, and on the bottom of the page they can see how the exercise was performed (eg, whether the movement was initiated by abducting from the shoulder which would indicate a compensatory movement). A media player (bottom right) shows arm position and movement corresponding to the progress bar on the time line to visualize how the arm moves to aid with this analysis. Joints Page The joint-based visualization illustrated in Figure 3 structures movement information around the entire arm. Therapists can click on a particular plane of movement in each joint (eg, shoulder abduction/adduction) to access a summary of a number of movements, quality, time active, and active ROM for the selected movement. Inspired by related work [25], the ROM is further illustrated for the selected joint through an avatar that visualizes the ROM achieved by the patient in daily life compared with the maximum ROM possible for this type of movement. This page was developed during the design workshops to show patients how the information collected through sensors relates to the different types of upper limb movement. Figure 5 shows the spreadsheet, which allows therapists to inspect all movements captured by the sensor and to sort them by time, quality, duration, and range of motion. A media player can be used to illustrate the arm movement selected in the spreadsheet. The data can be exported for further analysis (eg, for research into the effectiveness of interventions). This page was included during the design workshops to provide support detailed analysis of movements for therapists engaged in research activities. Study Participants We recruited 8 therapists (all female) to evaluate the dashboard prototype. Participants were recruited through the Royal Melbourne Hospital, Australia. All therapists were actively engaged in upper limb therapy with patients with neurological conditions including stroke, multiple sclerosis, traumatic brain injuries, and Parkinson's disease. Their clinical experience ranged from 3 months to 12 years. Five therapists worked predominantly with acute patients (within the first few weeks after presenting to hospital) and 3 therapists worked with chronic patients (ranging from several weeks to several years after a stroke). These 8 therapists had not been involved in the design process. They were recruited for the evaluation to provide unbiased feedback on the dashboard. Book vouchers were offered to participants for their time and involvement in the dashboard evaluation. Dashboard Evaluation A qualitative evaluation was conducted to explore how therapists would use the information presented and visualized on the dashboard. The evaluations took place in a meeting room at the hospital and lasted 60 minutes per therapist. Ethics approval was obtained through the University of Melbourne (#1545866). The evaluation followed a standard procedure. First, a background interview was conducted to learn about upper limb rehabilitation practices and the information therapists desire about their patients. Second, we conducted observations of therapists exploring each of the 5 dashboard pages. The therapists were instructed to think aloud in order to get a better understanding about their impressions of each visualization on the website and any questions or expectations that they may have. Finally, through a semi-structured interview, the therapists were asked to compare and rate the 5 visualizations in terms of usefulness for their work with stroke patients. These ratings were used as prompts to discuss how the dashboard could be integrated with their current work practices and the potential impact on improving rehabilitation outcomes. Each evaluation was audio-recorded and transcribed for later analysis. The examination of the dashboard was also screen-recorded with input from a webcam to capture facial expression of participants as they interacted with the website. The data were analyzed qualitatively, following a thematic analysis approach [26]. The authors read through all transcripts and coded the data to identify the various uses for each visualization as well as areas for improvement. Data were coded by the authors (BP, JF, SN) through SaturateApp, a Web-based tool for collaborative qualitative analysis. In total, 249 codes were generated about the uses for the 5 dashboard pages, 35 codes about ranking the different visualizations according to their potential usefulness, and 55 codes about the usefulness of the dashboard as a whole. In consultation with the research team these codes were collated into 3 themes that describe the uses of the dashboard and 1 theme about a major limitation in using the system, which are presented next. Theme 1: Objective Data About Activity Levels, Exercise, and Neglect The main use of the dashboard is to obtain objective patient data. Therapists can glance at the dashboard before or during consultations to assess how patients engage their upper limb outside the clinic including how actively they engage the affected limb, their adherence to exercise regimens, and possible neglect of the affected limb. The overview page was preferred by 63% (5/8) of therapists to assess the activity levels of patients outside the clinic. The overview page provides a quick snapshot of the patient's activity levels through visualizations of the number of movements performed over a week, the average quality of these movements, and the time spent active for each day. A simple timeline showing movements performed over a week offers therapists a quick glance of days when their patients performed well and when their patients did not reach their target levels. A lot of patients will try really hard today, and then tomorrow they really suffer, and then the next day they will probably do somewhere in between, and then two days later they will be like "oh I haven't done my exercises very much." And educating a patient around that when you've got hard data spike is really valuable. [OT8] The timeline page was preferred to assess whether patients adhered to the prescribed exercise regimens. The first visualization on this page shows the times and the intensity of arm activities over several days. Therapists used this information to infer activities based on time (eg, eating), duration (eg, exercise), or through conversation with patients. Some patients keep exercise diaries that therapists can use to compare with the timeline data. The timeline supports tagging, meaning that therapists can manually annotate events on the timeline with labels such as exercising and eating. It is important to note that the second timeline on the bottom of this page was not considered useful. This timeline would support analysis of movements for each degree of freedom over several seconds, for example, to inspect how patients perform an exercise. However, therapists commented that they would not have the time to analyze the data in this way. Finally, therapists found the heatmaps useful to assess patients with very low

levels of mobility and patients with hemispatial neglect, who have difficulty attending to one side of space. The heatmaps indicate where the hand is resting, and can be used to identify whether the hand is resting in a "natural" position. The heatmaps also show whether the hand of the patient crosses the midline of their body. This indicates attendance to the neglected side in neglect patients, and it shows an increased range of activities of daily living that a patient is able to perform. Theme 2: Engage Patients to Learn About Therapy, Provide Motivation, and Reflect on Progress A second area of use for the dashboard is to engage patients in a dialogue about the data to become more actively involved in the rehabilitation process. Therapists and patients can collaboratively examine the data presented on the dashboard to foster motivation and to inquire how patients cope in their everyday life. Particularly the timeline data and the tagging feature invited opportunities for therapists to engage their patients to learn more about exercise and other activities. Therapists can use the data to inquire about how well patients cope with the exercise programs that they have been given. Therapists may also use peaks and troughs in the timeline data to ask more broadly about the well-being of their patients in daily life. Theme 3: Engage With Other Clinicians and Researchers Based on Objective Data The information presented on the dashboard can also be useful beyond the interactions between a therapist and a patient during therapy. Finally, the information available through the dashboard provides opportunities for research into the effectiveness of rehabilitation services provided at the clinic. The spreadsheet page allows therapists to sort data by time, duration, and quality to support detailed analysis of the motions performed by individual patients. While the spreadsheet page was not considered useful for therapy, being able to export this data was seen as useful for further therapists engaged in research activities in order to assess the effectiveness of interventions across different patients. Your spreadsheet is only helpful for data analysis and research, which I think is a great thing to have incorporated but there's only going to be a small group of people that would utilize that. [OT8] Theme 4: Contextual Information is Critical to Analyze Movement Data A major limitation is the lack of contextual information presented across the different dashboard pages. The different dashboard pages presented various movement data (number, range, duration, quality of movement). However, a recurring discussion point with therapists was the lack of contextual information to understand the significance of these movements in daily life. First, the lack of contextual information was evident in discussions of the quality ratings. The quality rating was displayed on the overview page as an average value between 1 and 10 for all the movements performed over the course of a day, thus allowing the therapists to see trends in the data over several days and weeks. The therapists confirmed the findings from study 1 that information about the quality of the movements outside the clinic is critical, for some even more so than the number of movements. However, while the therapists desired a quality score, they also felt that in order to truly judge the quality of a movement they would have to see their patient making the movement. This is because the quality of a movement is dependent on its purpose in a particular context. For example, lifting the shoulder and shoulder abduction are often used as indicators for low quality movements, because many stroke patients use these movements to compensate for difficulties in reaching forward, or involuntarily abduct the shoulder when intending to reach forward. However, in certain contexts lifting the shoulder and abduction can be desirable and indicative of a normal, high quality movement, which cannot be distinguished by the system. Second, the lack of contextual information was evident in discussions about the timeline page. Based on the dashboard alone therapists cannot know if a movement constitutes an exercise activity, if the patient is engaging in an activity of daily living like eating, if the arm is swinging while walking, or if the arm is moved by a caretaker who helps the patient get dressed. The timeline presents some contextual information through the time of the day when movements are performed, which can indicate that a patient is eating or washing. However, the precise nature of the activity needs to be confirmed in conversation with a patient. Principal Findings This research identified core principles for the visualization of information collected through wearable sensor technologies for use by occupational therapists. Dashboards provide objective data for therapists about the activities of patients outside the clinic. This is important because prior work shows that the quality of subjective data through retrospective recall and exercise diaries is limited, and it relies on patients who are motivated and have adequate cognition [9]. Hence, data from wearable devices presented on the dashboard can verify subjective accounts from patients through objective data about activity levels in between therapy sessions, exercises performed at home, and attendance to the neglected side of the body. In accessing objective data, therapists emphasized the importance of getting an overview, over being able to see details. In line with the principal idea of a dashboard [19], the overview needs to provide a quick glance of the patient data. This overview needs to support comparison between different timescales, from several hours to several weeks, and between different joint movements (eg, to compare shoulder abduction with shoulder flexion). Unlike in other domains [27], the therapists expressed that they would not have time to inspect details of individual movements or outliers in the data, because it would take time away from working hands-on with patients. Hence the spreadsheet and the detailed timeline to analyze movements over several seconds were seen as superfluous. Visualizations need to engage patients in the therapy process. In particular, visualizations play an important role in discussing progress, motivating patients, and prompting reflection about exercises and activities of daily living performed in their own homes. Timeline visualizations were useful to discuss progress with patients. Heatmaps were useful to present spatial information about common positions and postures of the arm for reflection with patients. This is important to foster patient participation and motivation to achieve positive rehabilitation outcomes [28]. Visualizations and objective data are important to help therapists advocate on behalf of their patients in discussions with other clinicians. The work of therapists depends to a large extent on subjective judgments about a patient's ability to engage in activities of daily living. Hence, having objective movement data captured in daily life provides an objective indicator of a patient's capabilities that therapists can use in discussions with other clinicians. Contextual information is critical to analyze the information visualized on the dashboard. The lack of contextual information was raised as a key limitation because the therapists wanted to understand how much patients use their affected upper limb in daily life outside therapy (eg, to exercise, eat, or dress themselves). There was a disparity between the generally hands-on work of therapists, where they can touch and observe patients and understand the intentions of their actions, and the visualizations generated from sensor data that were disembodied and lacked references to the settings in which movements occur. Prior work on clinicians interpreting sensor data from patients with Parkinson's disease [29] and multiple sclerosis [30] highlights similar challenges in interpreting sensor data where therapists find it difficult to interpret sensor data in the absence of the patient, even though these studies [29,30] used sensors for short assessments in clinical settings, rather than to collect data over days and weeks in real-life. Health data are often not self-evident, and additional work is required to make sense of the data and to apply it in practice [29,31]. However contextual information is particularly important for therapists to interpret body movement, including understanding how movements relate to activities of daily living ranging from personal and domestic tasks, to community, employment, leisure, and recreational activities [32]. Hence, subsequent phases of this project will explore how contextual information can be gathered, such as through sensors embedded in objects and places that indicate activities (like sensors embedded in cutlery to indicate eating), or through mobile apps that allow patients or their caretakers to annotate movement information with pictures or personal notes about daily life activities. Furthermore, we seek to investigate to what extent the revised dashboard can elicit contextual information through dialogue between patients and therapists. Figure 6 summarizes the findings through a revised dashboard design. Based on the results presented above we combined the most useful elements of the 5 original dashboard pages into a design that fits on a single page to support meaningful comparison and minimize time spent navigating the dashboard. The annotations to Figure 6 summarize the key findings about the uses of the dashboard (obtain objective data, and to engage patients and clinicians) and the areas identified for improvement (capture contextual information, changes to enhance the clarity of the information presented, and content omitted due to lack of use). Figure 6. Revised dashboard design based on the findings from this study. The annotations on the left side show how the new design maintains the key features that the therapists found useful. The annotations on the right side highlight changes to the design. Limitations The main limitation of this study lies in the ecological validity. The findings of this study provide rich insights into the potential uses of a dashboard to support upper limb therapy. However, evaluations in a laboratory or simulated setting do not allow for evaluation of how a system would be used in a real-world setting and how it fits into the work practices of therapists. Furthermore, the prototype relied on mock data because real-life data about upper limb movement over extended periods of time is currently not available. If real-life sensor data were available, it is likely that the data would contain a degree of inaccuracy due to movement of the sensors on the patient's body and due to sensor drift, which would affect measures of quality and range of motion. Finally, the therapists in this study spoke about the potential uses of the dashboard to engage patients, yet these claims have not been verified with patients. A deployment study of a functioning dashboard and wearable technology with patients engaged in upper limb therapy and their therapists will be conducted in the next phase of this project to address these limitations. A further limitation of the dashboard and wearable technology developed in this project is the lack of data on wrist and finger extension. The current system focusses on the movement of the arm (shoulder, elbow, and wrist supination/pronation), which is critical for many stroke patients with low levels of mobility. However, activities of daily living like eating, dressing, and washing rely to a great extent on our ability to move the wrist and the fingers, which are not captured in the current design. Related work shows the potential of capturing finger and wrist movements through sensors captured through gloves [33,34] or rings worn on the finger [12,16], which we aim to explore in subsequent phases of this research project. Conclusions Upper limb information from wearable technology provides hitherto unavailable insights into the activities of stroke patients outside the clinic. Visualization of this information provides therapists with objective data, engages patients and supports discussion with other clinicians. Consideration needs to be given to contextual information, such as how to collect this information and how to integrate it with existing visualizations to provide meaningful insights into activities of daily living performed by patients. These findings open the door for further work to develop wearable technology for patients to collect upper limb data in real life, and to develop visualizations that present this information to therapists and patients to support rehabilitation. Lymphoma-Like T Cell Infiltration in Liver Is Associated with Increased Copy Number of Dominant Negative Form of TGFβ Receptor II Hepatosplenic T cell lymphoma (HSTCL) is a distinct and lethal subtype of peripheral T cell lymphoma with an aggressive course and poor outcome despite multiagent chemotherapy. Contradictory literature, an unknown etiology, and poor response to treatment highlight the need to define the malignant process and identify molecular targets with potential for

successful therapeutic interventions. Herein, we report that mice homozygously expressing a dominant negative TGFβRII (dnTGFβRII) under the control of the CD4 promoter spontaneously develop lymphoma-like T cell infiltration involving both spleen and liver. Splenomegaly, hepatomegaly and liver dysfunction were observed in homozygous dnTGFβRII mice between 10 weeks and 10 months of age associated with a predominant infiltration of CD4−CD8−TCRβ+NK1.1+ or CD8+TCRβ+NK1.1− T cell subsets. Notch 1 and c-Myc expression at the mRNA levels were significantly increased and positively correlated with the cell number of lymphoid infiltrates in the liver of dnTGFβRII homozygous compared to hemizygous mice. Further, 2×104 isolated lymphoma-like cells transplant disease by adoptive cell transfers. Collectively, our data demonstrate that increased copy number of dnTGFβRII is critical for development of lymphoma-like T cell infiltration. Introduction Transforming growth factor beta (TGFb) is a multifunctional protein that acts as an important regulator of cell growth, proliferation, differentiation, morphogenesis and inflammation. TGFb exerts biological effects by ligation of its cognate cell surface TGFb receptors with activation of downstream effectors including the TGFbR Smad family [1,2,3]. Alterations of specific components involved along the TGFb signaling pathway results in loss of TGFb receptor function and disruption of the intracellular TGFb signaling cascade. Such loss of TGFbR function is implicated in the pathogenesis of aortic pathology, various cancers and fibrotic and inflammatory disease [1,2,4]. There is a reduction of TGFbRII expression in Burkitt's lymphoma [5] and advanced cutaneous T cell lymphomas [6,7]. Similarly, loss of surface TGFbRII or mutations in the TGFbRII gene have been reported in human T cell malignancies [8] and colorectal cancer [2,9,10], suggesting that abnormal expression of TGFbRII is associated with malignant progression. Thus, abnormalities in the TGFb signaling pathway may be involved in the molecular pathogenesis of lymphoid malignancy. In the study herein, we generated homozygous dominant negative TGFbRII mice in which a 2-fold increase in expression of dnTGFbRII transgene was detected under control of the CD4 promoter. We demonstrated that mice homozygous for dnTGFbRII spontaneously developed lymphoma-like T cell infiltration involving both the spleen and liver with a significantly elevated pro-oncogene expression of Notch 1 and c-Myc. Further, we demonstrated that 2610 4 lymphoma-like cells were able to transplant disease by adoptive cell transfer. For intracellular cytokine analysis, mononuclear cells isolated from splenic tissues were cultured in media containing Leukocyte Activation Cocktail, with BD GolgiPlug TM (BD Pharmingen, San Diego, CA) for 4 hours. Cells were stained for cell surface markers with PerCP anti-CD8a (clone 53-6.7, Biolegend), APC-conjugated anti-TCR-b (clone H57-597, eBiosciences), and APC-eFluorH 780-conjugated anti-NK1.1 (clone PK136, eBiosciences). After staining, the cells were fixed with BD Cytofix/Cytoperm TM solution and permeabilized with BD Perm/Wash TM buffer (BD Pharmingen, San Diego, CA). Aliquots of these cells were stained with FITC-or PEconjugated antibodies against IFN-c, IL-2 as well as their respective isotype control antibodies. Stained cells were analyzed using a FACScan flow cytometer (BD Bioscience) that was upgraded by Cytec Development (Fremont, CA), which allows for five-color analysis. Data were analyzed utilizing CELL-QUEST software (BD Bioscience). Appropriate known positive and negative controls were used throughout. Histopathology Immediately after sacrifice, the lung, spleen, liver and colon were harvested, fixed in 4% paraformaldehyde (PFA) at room temperature for 2 days, embedded in paraffin, and cut into 4micrometer sections. The liver sections were de-paraffinized, stained with hematoxylin and eosin (H&E), and evaluated using light microscopy. Real time RT-PCR Analysis To determine the relative mRNA levels of the genes including Notch-1, the Notch-1 ligands DLL1/DLL4, the Hes-1 gene which involved in Notch signaling, and for purposes of control the gene PTEN that is associated with DNA repair were quantitated in RNA extracted from liver tissues of the appropriate strains of mice. Total RNA was extracted from individual liver tissues using the QIAGEN RNeasy Mini Kit (Qiagen, Valencia, CA). For real time PCR analysis, 1 mg of total RNA was reverse transcribed and then quantified on an ABI ViiA TM 7 Real-Time PCR System (Applied Biosystems, Foster City, CA). Amplification was performed for 40 cycles in a total volume of 20 ml and products were detected using SYBR Green (Applied Biosystems, Foster City, CA). The relative level of expression of each target gene was determined by normalizing its mRNA level to an internal control gene GAPDH. Clonality Analysis The clonal T cell expansions were identified by CDR3-length analysis of TCRVb gene segments as described [13]. Relative Quantification of Transgene Copy Number Relative quantification of transgene copy number was established as previously described [14]. Briefly, genomic DNA was obtained from mouse ear using the QIAamp DNA Mini Kit (QIAGEN, Valencia, CA), and then diluted to 2 ng/ml. Amplification was performed for 40 cycles in a total volume of 10 ml and products were detected using SYBR Green (Applied Biosystems, Foster City, CA) and quantified on an ABI ViiA TM 7 Real-Time PCR System. The primer sequences for dominant negative TGFbRII were as follows. Forward: GCTGCA-CATCGTCCTGTG, Reverse: ACTTGACTG-CACCGTTGTTG. A single copy gene within the mouse genome, Survival Motor Neuron (SMN) gene [15,16] was used as a reference gene. The primer sequences for SMN were as follows. Forward: TGGGAGTCCATCCATCCTAA, Reverse: CGACTGGGTAGACTGCCTTC. Relative quantification methods (2 2DDCt methods) were used for relative quantification of transgene in mouse genome by normalizing target gene to the reference gene SMN. Statistical Analysis The data are presented as the mean 6 SEM. Two-sample comparisons were analyzed using the two-tailed unpaired t-test. The correlation between two parameters was analyzed using Spearman Correlation Method. A value of p,0.05 was considered statistically significant. Lymphoma-like T Cell Infiltration in Intercrossed dnTGFbRII IL-6 2/2 Littermates Our lab has previously documented that deletion of the IL-6 gene from the hemizygous dnTGFbRII mice significantly improved colitis as indicated by substantially reduced intestinal lymphocytic infiltration, reduced diarrhea and increased body weight, while maintaining the autoimmune cholangitis. In the follow-up study of cholangitis in this mouse model, we expanded this colony by intercrossing hemizygous TGFbRII IL-6 2/2 litters. In this process hemizygous dnTGFbRII +/2 IL-6 2/2 mice were generated along with homozygous TGFbRII +/+ IL-6 2/2 and TGFbRII 2/2 IL-6 2/2 littermates, in a ratio that follows Mendel's law of segregation. When individual animals generated by intercrossing were examined for the liver infiltrating mononuclear cells (MNCs), two distinct subsets of littermates with TGFbRII transgene were found to have dramatically different (6-to 12-fold) numbers of liver infiltrating MNC ( Figure 1A). One of the subsets, approximately one third of the TGFbRII littermates, had 152.7622.0610 6 hepatic mononuclear cells (HMNCs), while the rest of the animals only had 20.961.8610 6 HMNCs. Such massive HMNC increase suggested a lymphoma-like disease. Therefore we examined the mRNA levels of the lymphomarelated proto-oncogenes involved in the Notch-1 signal pathway, including Notch-1, DLL1/4, Hes-1, PTEN and c-Myc in the liver tissues of individual animals. Among these genes, the relative mRNA levels of Notch-1 and c-Myc were the most pronounced and significantly increased in the littermates with massive HMNCs when compared to those with fewer HMNC infiltration (c-Myc: 9.261.7 vs. 1.760.6, n = 6, p = 0.0016) ( Figure 1B), or compared with the hemizygous dnTGFbRII mice (c-Myc: 9.261.7 vs. 2.460.6, n = 6, p = 0.0032) ( Figure 1B). A significant positive correlation was detected between the number of HMNCs and Notch-1 (r = 0.8502, p = 0.0005, Spearsman correlation), and between HMNCs and c-Myc in interbred mice (r = 0.9347, p,0.0001, Figure 1C). These results indicate that lymphoma-like T cell infiltration occurred in the interbred dnTGFbRII + IL-6 2/2 mice. Since the lymphoma-like HMNCs are seen in approximately one third of the dnTGFbRII + IL-6 2/2 mice generated by intercross breeding that resulted in a mixture of littermates with hemizygous (+/2) and homozygous (+/+) dnTGFbRII gene, but not seen in pure hemizygous dnTGFbRII +/2 IL-6 2/2 litters generated without intercrossing, the results also suggest that the lymphoma-like T cell infiltration occurred in mice with the homozygous dnTGFbRII gene. Characterization of the Lymphoma-like HMNCs Flow cytometric analysis was performed to determine the phenotype of the lymphoma-like HMNCs. As shown in Fig. 2A, in mice with greatly elevated HMNC infiltration (HMNC high), or hepatic lymphoma, the HMNCs were comprised of two major phenotypes. Seven out of 12 lymphoma-like mice were characterized by the predominance of CD4 2 CD8 2 TCRb + NK1.1 + cells (termed NK1.1), whereas the other 5 lymphoma-like mice had a predominant CD8 + TCRb + NK1.1 2 T cell subset (termed NK1.1 2 ) in the liver tissues. The percentage of HMNCs with these phenotypes ranged from 70-98%, indicating that these mice developed lymphoma-like T cell infiltration. Mice with HMNCs in these two phenotypes had a significant increase in splenic weight ( Figure 2B) and hepatic MNC count ( Figure 2C) compared to mice without such predominant HMNC phenotypes. Histological examinations were performed on lymphoid ( Figure 2D, a-h) and non-lymphoid ( Figure 2D, i-s) organs. Massive atypical lymphoid infiltration was observed in the grossly enlarged spleen and liver, but not in the small intestine or colon, of the mice with these predominant HMNC phenotypes ( Figure 2D). Marked lymphoid aggregation was only found in the lung of 1/7 mouse with predominant CD4 + TCRb + NK1.1 + infiltrates. In liver sections, a diffuse infiltration with atypical lymphocytes was observed in mice with CD4 2 CD8 2 TCRb + NK1.1 + HMNC phenotype, whereas the CD8 + TCRb + NK1.1 2 phenotype was associated with focal lymphoid aggregates ( Figure 2D, d:6200, t:640). These histological changes were not present in the inbred dnTGFbRII IL-6 2/2 littermates without massive HMNC infiltration. NK1.1 + lymphoma-like T cells isolated from the liver had a markedly reduced ability to produce IFN-c and IL-2 compared to NK1.1 2 lymphoma-like cells and HMNCs of non-lymphoma dnTGFbRII IL-6 2/2 mice ( Figure 3A), while such difference was not seen in the spleen ( Figure 3B). We next determined the clonality of the T cells in HMNC by examining T cell receptor Vb repertoire (Vb1-20) CDR3 length and joining beta (Jb1.1-1.6 and Jb 2.1-2.7). The results demonstrated that both lymphoma-like and non-lymphoma dnTGFbRII IL-6 2/2 mice displayed restricted Vb repertoires. However, Vb2/cp Jb2.7, Vb4 Jb 2.3, Vb7 Jb1.5 and Vb9 Jb2.3 were highly restricted in lymphoma-like mice compared to the non-lymphoma littermates, indicating that expanded lymphoid populations are clonally heterogeneous ( Figure 4). The Development of Lymphoma-like T Cell Infiltration was Associated with Homozygosity of dnTGFbRII Gene It is critical to exclude the possibility that the observed severe HMNC infiltration in dnTGFbRII IL-6 2/2 littermates actually reflects severe autoimmune liver inflammation rather than hepatic lymphoma. We have previously demonstrated that the autoimmune cholangitis in dnTGFbRII mice is mediated by Th1 cells, and that abrogation of the Th1 pathway by depleting IL-12 p40 gene completely protect dnTGFbRII mice from both cholangitis and colitis [12]. In order to further confirm the relationship between dnTGFbRII homozygosity and lymphoma-like T cell infiltration and to differentiate lymphoma from severe autoimmune cholangitis, we intercrossed our previously reported hemizygous dnTGFbRII p40 2/2 mice, which were cholangitisfree, to generate a mixture of hemizygous and homozygous dnTGFbRII p40 2/2 littermates with a ratio that follows Mendel's law of segregation. As expected, substantially increased HMNCs (9.460.12610 7 ) were observed in the liver tissues of two out of seven, similar to the expected one-third frequency, intercrossed dnTGFbRII p40 2/2 mice at the age of 10 weeks. Liver pathology examination demonstrated massive atypical lymphoid hepatic infiltration in these mice, indicating that introduction of a homozygous dnTGFbRII gene into the cholangitis-free dnTGFbRII p40 2/2 mice resulted in development of lymphomalike T cell infiltration ( Figure 5A). Finally we generated intercrossed dnTGFbRII mice to determine if homozygous dnTGFbRII gene alone is sufficient to cause lymphoma-like T cell infiltration. Severe IBD in female hemizygous dnTGFbRII gene mice reduced the rate of fertility. Therefore only 3 interbred littermates were obtained. In one out of these 3 littermates a predominant CD4 2 CD8 2 TCRb + NK1.1 + cell subset was found in the liver, which also demonstrated massive atypical lymphoid hepatic infiltration at 12 weeks ( Figure 5B). Taken together, the findings in the intercrossed dnTGFbRII IL-6 2/2 , dnTGFbRII p40 2/2 and dnTGFbRII mice indicate that dnTGFbRII homozygosity is associated with occurrence of lymphoma-like T cell infiltration. Increased Copy Number of dnTGFbRII is Associated with the Development of Lymphoma-like T Cell Infiltration To directly confirm that lymphoma-like disease occurs with a higher copy number of

dnTGFbRII gene in homozygous dnTGFbRII mice, we determined the relative dnTGFbRII transgene copy number in intercrossed dnTGFbRII littermates by quantitative real-time PCR. Given the fact that the parental generations are hemizygous, Mendel's law of segregation predicts that the entire litter would be comprised of 50% of hemizygous (dnTGFbRII +/2 ), 25% homozygous (dnTGFbRII +/+ ) and 25 percent negative (dnTGFbRII 2/2 ) in the dnTGFbRII transgene. We determined the relative levels of the transgene in genomic DNA in comparison to a reference single copy gene SMN in 42 mice derived from 7 litters generated by intercrossing mice that carried hemizygous dnTGFbRII gene. These included 5 litters of intercrossed dnTGFbRII IL-6 2/2 mice, 1 litter of dnTGFbRII p40 2/2 mice and 1 litter of dnTGFbRII mice. In 22 mice (52%), the relative transgene level was 1.2960.03; in 8 mice (19%) the relative transgene levels was 2.8360.11, while in 12 mice (29%) the transgene was not detected ( Figure 5C). Among the 8 mice with the higher level of dnTGFbRII transgene, 4 mice died around 12 weeks, two mice had splenomegaly, hepatomegaly and jaundice with an 18-fold increase of total HMNCs at age 12 weeks compared to the lower transgene level controls; the last two mice had an approximate 3-fold higher number of HMNCs at age of 15 weeks. Importantly, all mice with lymphomatous lesions, including dnTGFbRII and dnTGFbRII p40 2/2 mice shown in Figure 5A and 5B, had higher levels of TGFbRII transgene than those without lymphomatous lesions (2.4160.22, n = 8 vs. 1.2760.03, n = 15, p,0.0001). There was a highly significant positive correlation between hepatic cellular infiltrates and the copy number of dnTGFbRII transgene (r = 0.9342, p,0.0001), indicating that abrogation of TGFb signaling by higher copy number of dnTGFbRII contributes to the emergence of T cell lymphomalike T cell infiltration. Adoptive Cell Transfer to Rag1 2/2 Mice To evaluate whether the lymphoma-like T cell infiltration in dnTGFbRII mice are transferable, we carried out standard adoptive transfer studies. We reported previously that transferring CD8 + T cells from hemizygous dnTGFbRII mice induced changes consistent with autoimmune cholangitis in Rag1 2/2 mice only when the number of transferred cells reached one million [17]. We thus isolated HMNCs from hemizygous and homozygous dnTGFbRII mice and transferred 5-to 50-fold fewer cells (2610 4 or 2610 5 ) than the previous transfer studies [18,19] into Rag1 2/2 mice. The frequency of T cells expressing NK1.1 phenotype was approximately 95% in the liver of homozygous TGFbRII donors, 95% of TCRb + NK1.1 + cells were CD4 and CD8 double negative ( Figure 6A). Six weeks after intravenous injection, pathological and phenotypic changes identical to that seen in the donor homozygous mice were observed in spleen and liver in 8/8 recipient mice ( Figure 6B and 6D). Histopathological studies revealed massive atypical lymphoid cell infiltrates and hepatocellular damage in recipient mice, even in mice that received as few as 2610 4 HMNCs from TGFbRII homozygous mice ( Figure 6D). Consistent with the loss of cytokine function noted in the donor lymphoma-like cells, T cells from the recipient mice showed reduced cytokine production ( Figure 6C). In contrast, no obvious lymphoid cell infiltrates were found in Rag1 2/2 recipients of donor cells from hemizygous TGFbRII mouse ( Figure 6D Adoptive Transfer of Hepatic CD8ab Cells from TGFbRII Hemizygous Mice Resulted in Lymphoma-like Infiltration Since two distinct phenotypic T cell expansions were found in the TGFbRII homozygous mice, we addressed whether they were derived from the same precursors and transited through a TCRb + NK1.1 2 stage before they became NK1.1 + T cells during the terminal stage [20]. Flow cytometry sorted populations of hepatic CD8ab cells from TGFbRII hemizygous mice were adoptively transferred to Rag1 2/2 mice (10 6 sorted CD8 T cells per mouse, n = 3). The purity of the sorted CD8ab cells was .95% as assessed by flow cytometric analysis ( Figure 7A). Recipients were euthanized six weeks after adoptive transfer. Lung, heart, kidney, intestine, colon, spleen and liver were removed and examined histologically. The wet weight of spleen and liver were determined as shown in Figure 7C. Two of the three recipients had increased liver mass and cell count of mononuclear cells in both spleen and liver. Only 53% of MNCs from recipient 2 retained the TCRb + NK1.1 2 phenotype, while 45% of donor cells became CD4 CD8 double negative TCRb + NK1.1 + in liver ( Figure 7B). The frequency of TCRb + NK1.1 + was higher in liver compared to that in spleen ( Figure 7B). Histology revealed that bridging necrosis with severe to massive atypical lymphoid infiltration and hepatocellular damage were found in recipient 1 and recipient 3. On the other hand, only mild focal necrosis with mild lymphoid infiltration was observed in recipient 2. Our data collectively suggest that CD8 T cells from dnTGFbRII mice possess a capability to develop an NK1.1 + phenotype (Figure 7) with high pathogenic potential. In contrast, no NK1.1 positive T cells turned into NK1.1 negative T cells in immunodeficient Rag1 2/2 mice ( Figure 6). Discussion The TGFb signaling pathway plays an important role in T cell development and proliferation, naïve T cell homeostasis, peripheral T cell tolerance and effector T cell differentiation [20,21]. To date, different strategies have been employed to generate mice with a T cell-targeted disruption of TGFb signaling. In these models, consequences of TGFb defects are limited to T cells. Mice that have a CD4-Cre-mediated deletion of a TGFbRII allele [20] develop a progressive lymphoid infiltration into multiple organs before 5 weeks of age [22]. Inserting a truncated TGFbRII under a CD2 promoter/enhancer in mice results in a CD8 T cell lymphoproliferative disorder with small lymphocyte infiltration [23]. Expression of a dominant-negative form of TGFbRII (dnTGFbRII) under a CD4 promoter without a CD8 silencer leads to spontaneous activation and differentiation of both CD4 and CD8 T cells and development of autoimmune diseases at 4-6 months of age [24]; the different pathological outcomes demonstrated that the TGFb signaling pathway is not completely abrogated by the expression of the dominant negative form of TGFbRII. We previously reported that hemizygous dnTGFbRII mice manifest autoimmune cholangitis with elevated Th1 cytokines in serum and liver [25]. In the present study, we unexpectedly found shown in the middle and right panels were gated on TCRb + NK1.1 2 or TCRb + NK1.1 + populations as indicated in the left panels. B. The spleen weight of dnTGFbRII, dnTGFbRII IL-6 2/2 and dnTGFbRII IL-6 2/2 mice with a predominant NK1.1 positive or negative phenotype at age of 24-40 weeks. C. The total HMNC counts of dnTGFbRII, dnTGFbRII IL-6 2/2 and dnTGFbRII IL-6 2/2 mice with a predominant NK1.1 positive or negative phenotype at age of 24-40 weeks. D. Representative H&E stained sections of tissue sample including liver (a-d), spleen (e-h), small intestine (i-l), colon (m-p) and lung (qs) were prepared from dnTGFbRII and dnTGFbRII IL-6 2/2 mice at age of 24-40 weeks (a-s,6200; t,640). Typical diffuse lymphomatous lesions were found in liver (c) and spleen (g) of dnTGFbRII IL-6 2/2 mice with a predominant NK1.1 + phenotype, while large focal lymphomatous lesions were found in liver (d,6200)(t,640) and spleen (h) of dnTGFbRII IL-6 2/2 mice with a predominant TCRb + NK1.1 2 phenotype. No obvious lymphomatous lesions were found in lung, small intestine and colon in dnTGFbRII and dnTGFbRII IL-6 2/2 mice. doi:10.1371/journal.pone.0049413.g002 MNCs were CD4 2 CD8 2 TCRb + NK1.1 + or CD8 + TCRb + NK1.1 + cells. Second, a significantly increased expression of the proto-oncogenes Notch-1 and c-Myc, which correlates significantly with the number of HMNCs in mice homozygous for dnTGFbRII, compared to hemizygous dnTGFbRII mice. This observation has to be taken with caution partly because Myc and Notch can be induced upon normal T cell activation [26,27]. Third, histological analysis demonstrated massive atypical lymphoid cell infiltration in the grossly enlarged IL-2-producing T cells is lower in liver of dnTGFbRII IL-6 2/2 mice with a predominant NK1.1 + phenotype than dnTGFbRII IL-6 2/2 mice with a predominant NK1.1 2 phenotype (IFN-c + T: 6.13% vs. 50.13%; IL-2 + T: 7.45% vs. 29.10%). The data are representative of three independent experiments. doi:10.1371/journal.pone.0049413.g003 spleen and liver from the homozygous dnTGFbRII, but not hemizygous dnTGFbRII mice. Fourth, adoptive transfer of very small numbers of infiltrating cells (2610 4 ) isolated from the liver of homozygous dnTGFbRII, but not from hemizygous dnTGFbRII mice, resulted in massive atypical lymphoid cell infiltration with CD4 2 CD8 2 TCRb + NK1.1 + phenotype in Rag1 2/2 recipients, which were identical to that of the donor. However, our present data here do not clearly discriminate lymphoma from inflammatory T-cell infiltration. NKT cells are most abundant in the liver. TGFb signaling is critical for the differentiation of NKT subsets [28], but how TGFb signaling involved in the regulation and differentiation of NKT in our mice model with abrogated TGFbRII needs to be further elucidated. Oligoclonal expansion of T cells was detected in the liver of homozygous and hemizygous dnTGFbRII mice by clonality analysis, indicating that the expanded T cells are heterogeneous. Of note, clonality is not equivalent to malignancy, since benign and inflammatory conditions may show monoclonal rearrangement [29,30]. Our finding of heterogeneous clonal expansion of lymphoma-like T cell in homozygous dnTGFbRII mice is consistent with previously documented heterogeneous clonal restrictions of T cell populations in patients with angioimmunoblastic T-cell [31] and cutaneous T-cell lymphoma [32]. Moreover, our clonality results support a previous study in AKR/J mice demonstrating that restricted TCR Vb repertoire and lack of CDR3 conservation displayed thymic lymphomas [33]. In addition, lymphoma-like changes develop in dnTGFbRII mice with a skewed but polyclonal TCR repertoire, which is in agreement with a recently published study in which TCR-diversity suppressed development of mature T-cell lymphoma [34]. The clonal competition hypothesis might be a possible explanation for outgrowth of atypical lymphocytes in such a skewed TCR repertoire situation. Another possible explanation for local atypical infiltrates in our experimental system could be outgrowth of activated T cells accumulated in liver and mutations in certain oncogenes such as p53 [35]. The liver is a ''graveyard'' that actively sequesters activated and eventually apoptotic T cells [36]. However, activated T cells without the regulation of TGFb signaling might undergo cell divisions rather than apoptosis due to mutations in oncogenes, resulting in outgrowth of atypical massive T cells. Further studies will be required to characterize the outgrowth of atypical lymphocytes in detail, including selections of malignant clones, deep sequencing and potential gene mutations to elucidate the underlying mechanism. Patients with various autoimmune diseases have demonstrated an increased risk of developing non-Hodgkin lymphoma and multiple myeloma. Previous studies have also shown that loss of response to TGFb is associated with the progression of different types of malignancies including T-cell lymphomas [7,8]. Since it has been previously implicated that the expression of a single copy of dnTGFbRII transgene does not completely block TGFb signaling in T cells, we reasoned that a second copy of dnTGFbRII transgene in homozygous mice further suppresses the downstream signaling, resulting in the development of T cell lymphoma. The pathological and immunological presentation in these lymphoma mice resembles the main clinical features in patients with HSTCL, although to date no cd + T-cell lymphomas have been found in homozygous dnTGFbRII mice. Strikingly, although elevated circulating IL-6 was reported to correlate with adverse clinical features and survival in non-Hodgkin lymphoma [37,38], while clinical trials have showed anticancer effects of IL-12 on cutaneous T cell lymphoma [39], genetic depletion of IL-6 or IL-12p40 did not rescue outgrowth of lymphoma-like T cell in homozygous dnTGFbRII mice indicating that dnTGFbRII homozygosity is critical for the outgrowth of lymphoma-like T cells. Peripheral T-cell lymphomas (PTCL) are rare and aggressive malignancies that are distinct from the more common cutaneous T-cell lymphomas. Hepatosplenic T cell lymphoma (HSTCL) is a distinct and lethal subtype of peripheral T-cell lymphoma with an aggressive clinical course and a dismal outcome despite multiagent chemotherapy. HSTCL likely arises from cytotoxic T-cells that express the cd T-cell receptor type. However, it is important to note that an ab T-cell phenotype has been described increasingly in HSTCL [40,41,42,43]. Lymphoma cells usually have the following phenotype: CD2 + , CD3 + , CD4 2 , CD5 2 , CD7 + and CD8 2 . The World Health Organization

has updated the classification of lymphomas, which has led to the application of a more stringent criteria for the diagnosis of enteropathyassociated T-cell lymphoma (EATL) [44]. However, the challenges in understanding and treating PTCLs remain, since published literature consists mostly of case reports. A wide range of pathologic subdivisions with varied clinical features also impedes systematic study on PTCL. Our findings highlight a potential role for TGFb signaling in the development of HSTCL. Consistent with the previous work by Marie and colleagues [20], we found that in the hemizygous TGFbRII mice an expanded subset of T cells expressed the NK1.1 marker, although lymphoma was not found in these mice. However, when we adoptively transfer the dysfunctionally activated TGFbRII-CD8 NK1.1 2 T cells from these mice into the immunocompromised micro-environment in the Rag1 2/2 recipients without regulatory T and B cells, expression of NK1.1 and disease pathology were observed (Figure 7), suggesting that expression of NK1.1 marker is Figure 7. CD4 and CD8 double negative T cells were detected in Ly5.1Rag1 2/2 mice six weeks after adoptively transferred with one million hepatic CD8ab T cells from hemizygous dnTGFbRII mice (lymphomatous lesion-free). A. Flow cytometry analysis demonstrated the purity of hepatic CD8ab T cells from hemizygous dnTGFbRII mice. The numbers in the plots indicate the percentage of cells. B. Flow cytometric analysis of splenic and hepatic mononuclear cells of recipient mice at 6 weeks post-transfer. Three Ly5.1Rag1 2/2 mice were adoptively transferred with 1610 6 hepatic CD8ab T cells from hemizygous dnTGFbRII mice. TCRb staining was gated on CD45.2 + cells. The numbers in the plots indicate the percentage of cells. C. Weight and total MNC counts of spleen and liver from Ly5.1Rag1 2/2 recipients 6 weeks post-transfer. D. H&E staining sections of liver tissues from Ly5.1Rag1 2/2 recipients six weeks post-transfer. R1, Recipient 1; R2, Recipient 2; R3, Recipient 3; doi:10.1371/journal.pone.0049413.g007 associated with enhanced pathogenic potential. Although some lymphomas are phenotypically and genotypically of T cell origin, there are also lymphomas that are positive for the CD56 marker and are of NK cell origin [45]. NK cell markers are frequently expressed in HSTCL and other types of T cell lymphomas except for nasal and extranodal NK/T-cell lymphomas [46]. In comparison to CD56 2 T-cell lymphoma, the CD56 + NK-like T cell lymphomas demonstrated an aggressive clinical course [45,47] associated with a poor prognosis [48]. Our present study suggests that the acquisition of the NK1.1 cell surface marker by dnTGFbRII-CD8 T cells resulted in a highly pathogenic population that leads to development of T cell lymphoma with an aggressive clinical course. We speculate that blockage of the transition from NK1.1 2 to NK1.1 + T cells could be a potential strategy for the management of lymphoma disease. In summary, our data demonstrate that several features of human HSTCL are manifested in homozygous dnTGFbRII mice, suggesting that selective CD4 targeted functional abrogation of TGFbRII by increased copy number of dominant negative form of TGFbRII in mice can serve as a model of HSTCL for studying the disease mechanism and therapeutic strategies. Factors Predicting Mortality in Perforation Peritonitis at a Tertiary Care Center in Nepal Introduction Perforation peritonitis is a common surgical emergency. Despite advances in surgical techniques, antimicrobial therapy and intensive care support, mortality due to this illness still remains high. To analyze various demographic and clinical parameters associated with mortality in patients undergoing surgery for hollow viscus perforation peritonitis. Methods A prospective observational study was conducted at the Surgical Gastroenterology Unit, Department of Surgery, Tribhuvan University Teaching Hospital, in patients undergoing surgery for perforation peritonitis over a period of 18 months. Demographic characteristics, physiological variables and laboratory values were obtained. End point of the study was patients’ condition at discharge or in-hospital mortality. Univariate and multivariate analysis of various demographic, clinical and laboratory parameters were performed. Results Among 121 patients, mean age was 41.5±18.9 years (range of 16 to 87 years). There was a male preponderance (74.1%). In-hospital mortality was seen in 19 patients (17.0%). Age, female sex, pulse rate, blood pressure, white blood cell count, urea and creatinine were significant parameters associated with mortality. Multivariate logistic regression analysis of the variables found to be of significance from univariate analysis showed that only pulse, systolic blood pressure and creatinine were the independent variables associated with mortality. Conclusion Higher mortality was seen in elderly patients. Despite of male preponderance, larger proportion of females succumbed to their disease. Derrangement in vital parameters like pulse and blood pressure and renal function test had negative impact in survival. INTRODUCTION M ortality associated with perforation peritonitis remains relatively high. 1 It is difficult to predict mortality at the time of presentation. However, there are certain demographic, clinical and laboratory parameters regarded as high risk. Presence of these high risk factors may guide clinicians to be more watchful in such subset of patients. 2 Objective of this study was to identify various demographic, clinical and laboratory parameters associated with mortality in patients undergoing surgery for perforation peritonitis. METHODS A prospective observational study was conducted at the Surgical Gastroenterology Units of Department of Surgery, Tribhuvan University Teaching Hospital, Kathmandu, Nepal, over a period of 18 months (July 1, 2011 to November 30, 2012). Ethical approval was obtained from the Institutional Review Board of Institute of Medicine. Informed consent was taken from all the patients before enrolling them into the study. All patients undergoing surgery for suspected perforation peritonitis were included in the study. Patients aged less than 16 years, discharged on request and patients in whom no perforation found during surgery were excluded from the study. Statistical analysis performed with SPSS 16.0 (SPSS, Chicago, Ill). Results expressed as mean ±standard deviation.T-test to compare difference between two groups. All statistical tests with P<0.05 considered statistically significant. RESULTS A total of 121 patients underwent surgery for perforation peritonitis during the study period. Mean age of the patients was 41.5 ± 18.9 years with range of 16 -87 years. There was a male preponderance with male to female ratio of 2.9 : 1. 19 patients (17.0%) of the patient died. Comparison of demographic and clinical parameters between survivors and non survivors are depicted in Table 1. Non-survivors were significantly older in comparison to survivors. (53.2 ± 15.8 versus 39.2 ± 18.7 years). Though male patients outnumbered female patients, there were a larger proportion of females among non survivors. However, on analyzing other confounding factors such as age (p = 0.295) and duration of symptoms (p= 0.364), the difference in survival was not significant. Similarly, the median duration of symptoms likewise did not vary statistically (p = 0.241) Pulse rate in non survivors was 100.9 ± 19.2 as compared to 119.6 ± 24.9 in non survivors (p < 0.001). Systolic blood pressure in survivors and non-survivors was 116.3 ± 19.9 mmHg and 90.7 ± 16.8 mmHg respectively (p < 0.001). Hence, variation in vital parameters, such as pulse and systolic blood pressure was found to be significantly different between survivors and non survivors. Table 2. Levels of hemoglobin, sodium and potassium were not found to be significantly different between survivors and non-survivors. White blood cell count was lower in the non-survivors (p = 0.048). Levels of urea and creatinine were significantly elevated in the non survivors. Comparison of laboratory parameters between survivors and non survivors is given in Non traumatic bowel perforation accounted for 63.4% of all patients. Mode of bowel perforation, whether non traumatic, penetrating or blunt injuries were not found to be associated with increased mortality (p = 0.068). Duodenum was the commonest site of perforation which accounted VOLUME 41 | NUMBER 3 | DECEMBER 2019 www.jiom.com.np Kansakar et al. to nearly one third of the cases, followed by ileal and appendicular perforations. One third of the patients with ileal perforation succumbed to their illness. Similarly, the site of perforation likewise did not differ significantly in survivors and non survivors (Table 3). Intraoperative blood loss did not show any correration with survival. Multivariate logistic regression analysis done for the variables found to be significant from univariate analysis showed that only pulse, systolic blood pressure and creatinine were the independent variables associated with mortality. (Table 4) DISCUSSION Overall mortality in hollow viscus perforation in various series ranged from 10.6 % to 36 %. [3][4][5][6] The results in the present series were comparable with most series with exception to Mulari, et al, who reported mortality of 36 %. 4 The reason for high mortality rate in Mulari series may be inclusion of only ICU patients. As with present study, many other studies have found correlation between increasing age and mortality in perforation peritonitis. 2,7,8 Increasing age seems to be an important prognostic factor that may probably be mediated indirectly through latent or more severe co-morbidity, poor nutritional status, general fragilty, and changed immune function. 9 Because of the fact that increasing age relates to high mortality, it has been given a high score in both the scoring systems. Similarly, male preponderance in perforation peritonitis was also comparable with other series. 5,7 Higher proportion of male smokers which is a risk factor for peptic ulcer perforation maybe the reason for higher rate of perforation. Traumatic perforations resulting from physical assaults are also seen more in the male population. Several studies showed duration of symptoms to be a predictor of mortality. 5,7 Delay in presentation has been attributed to many factors. Low socioeconomic status, level of literacy and long distance needed for travel to reach a referral centre are the important factors causing delay in presentation. Svanes et al, 10 reported that delay in presentation more than 24 hours increased the mortality rate by seven to eight fold. However, in the present study, duration of symptoms did not differ significantly between survivors and non survivors. There is a three-fold increased risk of mortality when patients are in shock on admission and the mortality approaches 50% in presence of septic shock. 11 Present study showed significant association between hypotension and increased mortality in perforation peritonitis similar to other studies. 2,7,8 Regardless of the cause, hypotension leads to diminished tissue perfusion and increased incidence of dehiscence at the repair site. Laboratory parameters such as hemoglobin, sodium and potassium did not differ between survivors and non-survivors which correlates with published series. 7,12 WBC count was not associated with mortality in the study by Horiuchi,et al. 12 In the present series, though WBC count was a significant factor associated with mortality, multivariate analysis failed to establish it as an independent risk factor. Levels of urea and creatinine depicting the renal function had a strong association with mortality in our study as well as several other studies. 2,7,8 Proximal site of bowel perforation was seen in a studies from South Asia. 5,8 Duodenal ulcer perforation was the commonest site in the current study. Smoking, especially among young males, has been hypothesized for higher proportion of peptic ulcer perforation. 13 Distal bowel perforation has been seen to be predominant in the western countries mainly due to higher prevalance of diverticular disease. 14 On the other hand, a Nigerian study has shown 58% typhoid perforations. 15 Hence, the site of perforation has been shown to vary in different geographical regions. Overall, age, sex, site of perforation, preoperative shock, VOLUME 41 | NUMBER 3 | DECEMBER 2019 JIOM Nepal Mortality in perforation peritonitis hypoglycemia, renal dysfunction and duration of symptoms have been reported as the determinants of mortality in patients with perforation peritonitis. 7 However, from the present study, only pulse, blood pressure and creatinine were found to be significant factors from multivariate logistic regression analysis associated with mortality. CONCLUSION Approximately one fifth of the patients undergoing surgery for perforation peritonitis succumbed to their disease. Higher mortality was seen in elderly patients. Despite of male preponderance, larger proportion of females succumbed to their disease. Vital parameters like pulse and blood pressure were independent variables associated with mortality. Renal function test had a negative impact in survival. Symptom relief with MVP (mitomycin C, vinblastine and cisplatin) chemotherapy in advanced non-small-cell lung cancer. The role of chemotherapy in the palliation of patients with advanced stage (IIIB and IV non-small-cell lung cancer (NSCLC) remains controversial. We have carried out a chemotherapy study emphasising symptom relief, a topic not normally discussed in previous similar studies. A total of 120 patients with locally advanced or metastatic

non-small-cell lung cancer (NSCLC) were treated with a moderate-dose palliative chemotherapy regimen consisting of mitomycin C 8 mg m-2 i.v. on day 1 (alternate courses), vinblastine 6 mg m-2 i.v. on day 1 and cisplatin 50 mg m-2 i.v. on day 1 (MVP), repeating every 21 days for a maximum of six courses. Thirty-eight of 118 assessable patients (32%) achieved an objective response. Patients with locally advanced disease (stage IIIB) had a significantly better response rate (52%) than those with metastatic disease (25%) (P < 0.01). In 76 out of 110 (69%) patients, with tumour-related symptoms including 24 out of 31 patients (78%) with locally advanced disease, symptoms completely disappeared or substantially improved. In only 15 patients (14%) did symptoms progress during treatment. Symptomatic improvement was achieved after one course of chemotherapy in 61% and after two courses in 96% of responding patients. The schedule was well tolerated. Only 19% developed WHO grade 3/4 nausea/vomiting, and only 3% developed significant alopecia. Other toxicities were minimal. MVP is a pragmatic inexpensive chemotherapy regimen that offers useful symptom palliation in patients with advanced NSCLC and merits a 1-2 course therapeutic trial in such patients. The schedule should also be assessed as primary (neoadjuvant) chemotherapy before radical radiotherapy for locally advanced NSCLC in a randomised trial. The role of chemotherapy in the treatment of advanced non-small-cell lung cancer (NSCLC) remains controversial. Some, but not all, recent trials have shown a survival advantage for combination chemotherapy over best supportive care in this condition (Cormier et al., 1982;Rapp et al., 1988;Cartei et al., 1993), and one has shown chemotherapy to be more cost-effective than supportive care (Jaakkimairn et al., 1990). Very recently, an overview analysis of seven trials has confirmed a statistically significant survival benefit, but a modest one (Souquet et al., 1993), and chemotherapy remains very much a palliative approach. No single chemotherapy regimen has been shown superior to others in the treatment of NSCLC, but best response rates have consistently been achieved with cisplatin-based regimens (Veronesi et al., 1988;Luedke et al., 1990). The combination of isplatin with mitomycin C and vinblasti has been shown in randomised trials to be one of the most effective regimens (Ruckdeschel et al., 1986), and this has the advantage over most other combinations of very rarely causing signiicant alopecia (an important consideration for palliative treatment). There is a tendency to use cisplatin in high dosage (100-120mgm-) in many of these regimens, but this is associated with significnt toxicity, including emesis, peripheral neuropathy, nephrotoxicity and high-frequency hearing loss. These problems disappear or are at least very markedly reduced with moderate-dose cisplatin (50mgm-2) and probably without significant reduction in efficacy (Hardy et al., 1989). We have therefore developed a moderate-dose cisplatin (P) regimen in combination with mitomycin C (M) and vinblastine (V) (MVP) which is well tolerated with few side-effects (Hardy et al., 1989). We have previously shown that this schedule achieves good symptom relief and therefore useful palliation in a small series of patients. Symptom relief has not been adequately addressed in most other studies of chemotherapy for NSCLC. Here we update our results in a much larger series of patients with emphasis on symptom relief as well as response rate and toxicity. patients, PS 3. Fifty-two patients had received prior treatment. Five had recurrent disease following surgery. Thirty-two had received previous radiotherapy (ten radical and 22 palliative). One patient had received prior conventional combination chemotherapy for coexistent hairy cell leukaemia. Fifteen patients had been previously treated with expermental single-agent therapy. Experimental agents included: CL 286,558 (Zeniplatin), eight patients; lonidamine, four patients; mitozolomide, one patient; ifosfamide, one patient; carboplatin, one patient; CB 10-277 (analogue of dacarbazine), one patient. Patient characteristics are summarised in Table I. Treatment regimen All patients received the following regimen: mitomycin C 8 mg m 2 i.v. on day 1 (given on alternate courses), vinblastine 6mg mr-2 (maximum 1Omg) i.v. on day I and cisplatin 50mgm-2 i.v. on day 1, repeated every 21 days. Standard intravenous pre-and post-treatment hydration was given with cisplatin. This consisted of 1.51 of normal saline + 40 mmol of potassium chloride + 40 mg frusemide over 2 h pre-cisplatin, followed by 2 1 of normal saline + 40 mmol of potassium chloride over 12 h post-cisplatin. Patients received prophylactic antiemetic therapy with high-dose metoclopramide and dexamethasone or a 5-HT3 antagonist and dexa-s_. rdg a or inW PA ENs etd X methasone, often as part of separate antiemetic trials. Renal flmction was checked with 5"Cr-EDTA clearance before alternate courses and the dose of cisplatin reduced as follows: EDTA >60mlmin-', full dose; 40-60 ml min', 25% dose reduction; <40 ml min-, no treatment with cisplatin. Treatment was continued until the development of progressive disease, unacceptable toxicity or to a minimum of six cycles in patients achieving objective response and/or symptomatic relief. Pretreatment investigations and assessment of response and toxicity All patients underwent pretreatment physical examination, including penpheral full blood count (FBC), plasma electrolytes, urea and creatinine, serum liver function tests, 5"Cr-EDTA clarance and chest radiography. Other more elaborate radiological investigations were only carried out if chnically indcated. FBC was repeated in patients at the time of their predicted nadir following the first cycle of treatment. Following this patients were ased before each treatment for symptomatic response (see below) and for objective response and toxicity according to standard WHO criteria (Miler et al., 1981). Complete response (CR) was defined as the disappearance of all known disease for at least 4 weeks. Partial response (PR) was defined as > 50% decrease in the sum of the products of the tumours' longest dimension and its widest perpendiclar measurement for at least 4 weeks, without the appearance of new lesions or progression of any one lesion. Stable disea (SD/NC) was defined as <50% derease or <25% increase in the size of the measurable disease, without the appearance of new lesions or progression of any lesion >25%. Progressive disease (PD) was defined >25% increase in one or more of the measurable lesions or the appearance of a new lesion(s). Response duration and survival were calulated from the date of first treatment using the standard life-table method of Kaplan and Meier (1958). Assessment ofsymptomatic response Tumour-related symptoms were recorded at the start of treatment under the following general headings: malaise, pain, cough, dyspnoea or 'other', which was then specified. Symptoms were then reassessed following each course of treatment with patients asked to grade change in symptoms using simple descriptive criteria as follows: (i) complete disappearance of symptoms (CR); (ii) good improvement of symptoms (PR); (iii) minor or no change of symptoms (NC); (iv) worse (PD). Result Of the 120 patients entered into this study, two patients were not evaluable for objective and symptomatic response, one being lost to follow-up before completion of the first treatment cycle, and the other dying of sepsis during the first cycle. Nine other patients with progressive disease before completion of the first treatment cycle have been assssed as PD for both objective and symptomatic response. Seven patients were excluded from toxicity analysis (six because of early progression; plus the patient lost to follow-up). Eight patients were asymptomatic at presentation and were therefore not assessable for symptomatic response. All patients were included in the survival analysis. The median number of treatment courses given was 3 (range 1-8 Table II. Median response duration for responding patients was 6 months. Median survival for all patients was 5 months (6 months for locally advanced patients and 4 months for metastatic disease patients). Overall survival is given in Figure 1. Median survival for responding patients was 9 months compared with 4 months for non-responders (P<0.005). Response related to previous chemotherapy Previous treatment with experimental chemotherapy did not influence response in this study. The response rate for 103 patients who had received no previous chemotherapy was 32% compared with 29% for the 15 patients who had received previous experimental chemotherapy. Total (%) 17 (15) 59 (54) 19 (17) 15 (14) 110 sion of symptoms during chemotherapy. Details of symptomatic response are given in Table III. Forty-six of the 76 responding patients (61%) had a symptomatic response after only one course of chemotherapy and 73 (96%) after two courses. Thirty of these had a further improvement of symptoms with more treatment. Only three patients who failed to achieve symptomatic relief with two courses of treatment (4%) gained a symptomatic response with further treatment. Thirty-three of 36 patients (92%) with an objective response to treatment gained symptomatic relief. However, 43 patients (54%) with no change or progressive disease on objective response assessment also gained symptomatic relief. Those patients who had a symptomatic response to therapy had a significant survival advantage over those who did not (median survival 6 months vs 3 months respectively, P<0.005) (Figure 1). Median symptomatic response duration was 15 weeks from start of treatment (range 4-65 weeks). Toxicity Overall, treatment with this moderate-dose cisplatin regimen was well tolerated. Haematological toxicty was minimal (Table IV). Only seven patients (6%) developed WHO grade 3/4 neutropenia, and only four patients (3%) developed grade 3/4 thrombocytopenia. Two patients developed neutropenic fever, but only one of these developed a significnt neutropenic infection (1%). One other patient died in his local hospital of presumed neutropenic sepsis. The most significant adverse effect was emesis, with 22 patients (19%) developing grade 3/4 nausea and vomiting on at least one treatment cycle. Alopecia was minimal, with only three patients (3%) developing signifcant hair loss. Other signicant side-effects including neuropathy and nephrotoxicity were rare (Table V). Dose reductions and delays Eleven patients (9%) required a 25% dose reduction during chemotherapy. In five cases this was due to a reduction in EDTA clearance below 60 ml min-1, in three patients because of grade 3/4 leucopenia and in three because of grade 3/4 thrombocytopenia. No patient stopped treatment as a result of treatment-induced toxicity. Chemotherapy costs The cost of treatment at May 1993 prices for an average patient of surface area. 1.75 m2 is shown in Table VI. Chemotherapy does not yet have an established role in the treatment of non-small-cell lung cancer (NSCLC). In some respects, this is surpnsing. At least three randomised trials have shown a statistically significnt survival benefit over best supportive care (Cormier et al., 1982;Rapp et al., 1988;Cartei et al., 1993). In one of these trials, the total costs of supportive care alone exceeded those of chemotherapy because more in-patient time was required for symptom control (Jaakkimainen et al., 1990). On the other hand, not all trials have shown a statistically significnt survival benefit (Ganz et al., 1989;Woods et al., 1990;Kaasa et al., 1991). Very recently, an overview analysis of seven such trials has confirmed an overall reduction in mortality with chemotherapy (Souquet et al., 1993), but even in the positive trials the survival benefit has been small at around 4-5 months. In addition, chemotherapy is frequently perceived as being expensive and toxic. For these reasons, its role as palliative treatment for NSCLC has remained controversial. Our results from this study belie some of these critiisms. First and most important, this MVP schedule achieves useful symptomatic benefit in around two-thirds of patients. An interim analysis of another large trial has also suggested improvement in tumour-related symptoms following chemotherapy (Cullen, 1993). This in its own right is a major issue in palliative therapy, irrespective of survival benefit. Second, this benefit is achieved with a low incidence of serious sideeffects, particularly in comparison with some other commonly used schedules for NSCLC. For example, by careful choice of drugs we have dramatically reduced the risk of 4 fu% l 2 severe alopecia (3%); for most other schedules this problem is almost universal. In addition, by using cisplatin in moderate dosage, we have abolished the risk of severe druginduced nephroand neurotoxicity reported in other studies (Ruckdeschel et al., 1986;Rapp et al., 1988). Third, this is a simple and inexpensive regimen; the drug cost of around £78 per course is up to 10-fold cheaper than some other commonly used chemotherapy schedules in cancer medicine. Most of our patients were treated on an overnight in-patient basis, but the schedule can readily be adapted for day care use. Have these benefits been gained at the expense of decreased efficacy compared with other more intensive and expensive regimens? Higher dose cisplatin regimens have been advocated for NSCLC (Donnadieu et al., 1991) and

at least one early trial showed a survival benefit in responding patients for high vs moderate-dose cisplatin (Gralla et al., 1981). Subsequent trials, however, have in general failed to confirm this benefit (Klastersky et al., 1986;Ruckdeschel et al., 1986;Shinkai et al., 1986), and we believe that current evidence does not justify a high dose of cisplatin when balanced against significantly increased toxicity. Our own overall objective response rate and survival results are within the range reported in most other studies. An exception however is the so-called MIC (mitomyci-C, ifosfamide and cisplatin) schedule, one of the most active regimens so far reported with a response rate of 56% (Cullen et al., 1988). It is unlikely that selection bias can entirely explain this difference, and a comparative trial would be useful in assessing the extent to which putative increased efficacy might outweigh the disadvantages of increased cost and toxicity of MIC in a palliative setting. Meanwhile, we believe that MVP, a pragmatic and inexpensive chemotherapy regimen, should now be considered as an established therapy for symptom palliation in patients with metastatic NSCLC. An important practical point for clinical management is that symptom relief was usually achieved after one course of chemotherapy and nearly always after two. Palliative MVP chemotherapy should therefore be given as a therapeutic trial of two courses, and in general stopped after this if useful symptomatic relief has not been obtained. This prevents cumulative toxicity in patients failing to respond and targets more prolonged treatment at patients receiving real benefit. Furthermore, since objective response is nearly always associated with symptomatic relief, the latter can be used in the great majority of patients as evidence of clinical benefit without the need for elaborate and expensive restaging investigations including CT scans to confirm objective response. Where does chemotherapy go next in NSCLC? An important area for study is primary (neoadjuvant) chemotherapy before surgery or radiotherapy for locally advanced (usually IIIB) disease. Several trials have already been reported, some showing significant survival benefit over radiotherapy alone (Dillman et al., 1990), some showing benefit just below significance (LeChevalier et al., 1991) and some showing no benefit (Morton et al., 1991). Further trials are under way, including an important one using the MIC schedule (Cullen et al., 1988), but results are not yet available. If survival improvement is established for primary chemotherapy in this context it is likely to be modest. Real clinical benefit will therefore also depend on cost-effectiveness and low treatment-associated morbidity. In this context, we believe that this pragmatic low-risk MVP schedule, with established efficacy and low toxicity, should now be assessed as primary (neoadjuvant) chemotherapy before radical radiotherapy for locally advanced NSCLC in a randomised trial. An Improved Composite Multiscale Fuzzy Entropy for Feature Extraction of MI-EEG Motor Imagery Electroencephalography (MI-EEG) has shown good prospects in neurorehabilitation, and the entropy-based nonlinear dynamic methods have been successfully applied to feature extraction of MI-EEG. Especially based on Multiscale Fuzzy Entropy (MFE), the fuzzy entropies of the τ coarse-grained sequences in τ scale are calculated and averaged to develop the Composite MFE (CMFE) with more feature information. However, the coarse-grained process fails to match the nonstationary characteristic of MI-EEG by a mean filtering algorithm. In this paper, CMFE is improved by assigning the different weight factors to the different sample points in the coarse-grained process, i.e., using the weighted mean filters instead of the original mean filters, which is conductive to signal filtering and feature extraction, and the resulting personalized Weighted CMFE (WCMFE) is more suitable to represent the nonstationary MI-EEG for different subjects. All the WCMFEs of multi-channel MI-EEG are fused in serial to construct the feature vector, which is evaluated by a back-propagation neural network. Based on a public dataset, extensive experiments are conducted, yielding a relatively higher classification accuracy by WCMFE, and the statistical significance is examined by two-sample t-test. The results suggest that WCMFE is superior to the other entropy-based and traditional feature extraction methods. Introduction In response to imaginary movements, the brain cortex produces a corresponding Motor Imagery Electroencephalography (MI-EEG) with a rhythmic activity. The MI-EEG-based Brain-Computer Interface (BCI) technology appeals to patients with neurological disabilities, such as stroke; it can help them with neurorehabilitation to restore more effective motion control [1,2]. Due to the sensitivity to noise, time-varying and fuzziness of MI-EEG, its feature extraction has become an important issue in BCI-based rehabilitation engineering. At present, many feature extraction methods have been developed in time, frequency, time-frequency and spatial domains. The Autoregressive (AR) model is a classical feature extraction method in time domain. EEG signals are modeled by AR model and AR coefficients act as the features of EEG. This method can reflect the time-varying property of EEG, but it is sensitive to the data length [3]. The conventional time-frequency methods include Hilbert Huang Transform (HHT) [4,5], Empirical Mode Decomposition (EMD), Multivariate Empirical Mode Decomposition (MEMD) and Short Time Fourier Transform (STFT) [6][7][8], Discrete Wavelet Transform (DWT) based on Wavelet Transform (WT) [9][10][11] and Wavelet Packet Transform (WPT) [12], etc., in which the power Feature Extraction Based on WCMFE CMFE has better stability and consistency than MFE and other entropy-based methods. It benefits from the multiple information of different coarse-grained sequences at the same scale factor [34]. Each coarse-grained time series of CMFE is approximately regarded as the arithmetic mean filter [35], which is helpful to eliminate general random disturbance and make the signal smoother. Nevertheless, this method fails to match the time-varying MI-EEG signals. It is necessary to change the weight factors in the coarse-grained procedure. In fact, the arithmetic mean filters will be replaced with weighted mean filters from the perspective of signal processing [36,37]. In the following, a Weighted CMFE (WCMFE) is developed and used for extracting the nonlinear features of MI-EEG. The detained process is described as follows. Preprocessing of MI-EEG Times Series Suppose that X M 0 = [x M (1), x M (2), · · · , x M ( j), · · · x M (N)] T ∈ R N×1 with length N is the M-th channel MI-EEG signal of a trial, where M = 1, 2, · · · , n C , n C is the total number of channels. By analyzing the impact of scale factor τ on CMFE of MI-EEG, the suitable value of τ can be selected. Then, the coarse-grained sequences and the corresponding CMFE in the scale factor are calculated, and the optimal time interval is determined according to the maximum difference of CMFE between two motor imagery tasks. So, the MI-EEG signal is rewritten as where H = d − b + 1, b and d represent the serial numbers of the start point and the end point in the optimal interval, respectively. Hence, the WCMFE of the M-th channel MI-EEG can be calculated by averaging MFEs in τ scale: Construction of Feature Vector For τ scale, the WCMFEs of all the relevant channels can be expressed as where 1 ≤ τ ≤ τ max . The fusion feature vector of MI-EEG is defined as Data Source In this paper, the Data set III from BCI competition II is used to evaluate the superior performance of the proposed methods. This dataset was recorded from a normal subject (female, 25y) during a feedback session, and it was made up of the EEG signals about left-right-hand motor imagery. In the quiet and relaxed state, the corresponding imaginary tasks can be completed according to the screen prompt. Figure 1 displays the electrode positions. EEG signals were recorded on channels C3, Cz and C4 with sampling rate 128 Hz, and the signals were filtered from 0.5 to 30 Hz. There are 280 trials in the dataset. Similar to the training set, the test set was 140 trials in total, in which left and right hand motor imagery tasks were both performed 70 times. All trials were completed on the same day [38]. Data Source In this paper, the Data set III from BCI competition II is used to evaluate the superior performance of the proposed methods. This dataset was recorded from a normal subject (female, 25y) during a feedback session, and it was made up of the EEG signals about left-right-hand motor imagery. In the quiet and relaxed state, the corresponding imaginary tasks can be completed according to the screen prompt. Figure 1 displays the electrode positions. EEG signals were recorded on channels C3, Cz and C4 with sampling rate 128 Hz, and the signals were filtered from 0.5 to 30 Hz. There are 280 trials in the dataset. Similar to the training set, the test set was 140 trials in total, in which left and right hand motor imagery tasks were both performed 70 times. All trials were completed on the same day [38]. As shown in Figure 2, a single test was 9s. In the first two seconds, the subject kept rest. When t = 2s, the screen displayed a "+" cursor and began with a trial sound signal. Between 3s and 9s, the subject proceeded with the corresponding imaginary task according to the direction of the arrow on the screen. As shown in Figure 2, a single test was 9s. In the first two seconds, the subject kept rest. When t = 2s, the screen displayed a "+" cursor and began with a trial sound signal. Between 3s and 9s, the subject proceeded with the corresponding imaginary task according to the direction of the arrow on the screen. Data Source In this paper, the Data set III from BCI competition II is used to evaluate the superior performance of the proposed methods. This dataset was recorded from a normal subject (female, 25y) during a feedback session, and it was made up of the EEG signals about left-right-hand motor imagery. In the quiet and relaxed state, the corresponding imaginary tasks can be completed according to the screen prompt. Figure 1 displays the electrode positions. EEG signals were recorded on channels C3, Cz and C4 with sampling rate 128 Hz, and the signals were filtered from 0.5 to 30 Hz. There are 280 trials in the dataset. Similar to the training set, the test set was 140 trials in total, in which left and right hand motor imagery tasks were both performed 70 times. All trials were completed on the same day [38]. As shown in Figure 2, a single test was 9s. In the first two seconds, the subject kept rest. When t = 2s, the screen displayed a "+" cursor and began with a trial sound signal. Between 3s and 9s, the subject proceeded with the corresponding imaginary task according to the direction of the arrow on the screen. Interval Selection of MI-EEG The data of the training set on channels C3 and C4 were applied to select the optimal interval. For left-right-hand imaginary movement, CMFE sequences of each trail in the training set on channels C3 and C4 were calculated, respectively. For the same task, CMFE sequences of 70 trials were superimposed and averaged by a sliding window with window length of 1s and step size of 1 sampling point to obtain the mean CMFE, where two channels of MI-EEG signals of 9s were considered in one trial, and the MI task was left or right hand motor imagery. Furthermore, the related parameters were selected as: m = 2, n = 2, r = 0.15SD and τ = 2. The mean CMFE time series curves are displayed in Figure 3. It can be seen that CMFE values on channel C3 increase gradually and on channel C4 decrease with the left-hand motor imagery. However, it is opposite for the right-hand motor imagery. It is concerned with the Event-Related Desynchronization (ERD)/Event-Related Synchronization (ERS) phenomenon. Moreover, within the sampling interval of [450, 900], the changes of mean CMFE on channels C3 and C4 are prominent and the difference is the most obvious, so the sampling interval in [450, 900] is chosen for subsequent experimental study. Interval Selection of MI-EEG The data of the training set on channels C3 and C4 were applied to select the optimal interval. For left-right-hand imaginary movement, CMFE sequences of each trail in the training set on channels C3 and C4 were calculated, respectively. For the same task, CMFE sequences of 70 trials were superimposed and averaged by a sliding

window with window length of 1s and step size of 1 sampling point to obtain the mean CMFE, where two channels of MI-EEG signals of 9s were considered in one trial, and the MI task was left or right hand motor imagery. Furthermore, the related parameters were selected as: The mean CMFE time series curves are displayed in Figure 3. It can be seen that CMFE values on channel C3 increase gradually and on channel C4 decrease with the left-hand motor imagery. However, it is opposite for the right-hand motor imagery. It is concerned with the Event-Related Desynchronization (ERD)/Event-Related Synchronization (ERS) phenomenon. Moreover, within the sampling interval of [450, 900], the changes of mean CMFE on channels C3 and C4 are prominent and the difference is the most obvious, so the sampling interval in [450, 900] is chosen for subsequent experimental study. Selection of Weight Factors It is necessary to select the appropriate weight factors for estimating the WCMFE of MI-EEG. Different weight factors of coarse-grained series on WCMFE will change the performance of the filter and affect the classification results successively. . From the spectrums, the weighted mean filter has a low-pass characteristic and it can restrain high frequency components of the original signal. Meanwhile, the selection of weight factors will change the cut-off frequency and the spectrums as well. Selection of Weight Factors It is necessary to select the appropriate weight factors for estimating the WCMFE of MI-EEG. Different weight factors of coarse-grained series on WCMFE will change the performance of the filter and affect the classification results successively. The three-point weighted mean filters with different weight factors (a) Figure 4. It shows the variation of amplitude with normalized frequency ω π . From the spectrums, the weighted mean filter has a low-pass characteristic and it can restrain high frequency components of the original signal. Meanwhile, the selection of weight factors will change the cut-off frequency and the spectrums as well. Linear phase Finite Impulse Response (FIR) filter can process data without phase distortion and it has been widely used in speech signal processing, adaptive processing and other aspects [39]. Its unit impulse response has symmetry property. According to the basis, the weight factors are only selected by the coarse-grained procedure and the rules are as follows: Define the weight factors as Specifically, The coarse-grained process of CMFE can be regarded as the arithmetic mean filter, which can eliminate the noise to a certain extent and the calculation is simple. In order to detect the filter effect, a comparative experiment of the coarse-grained sequences using WCMFE, CMFE and original MI-EEG on channel C3 with motor imaginary tasks was carried out. Moving average was realized by a sliding window with sample interval of one. In Figure 5, there is no doubt that all of the coarsegrained sequence curves follow the trend of the original MI-EEG and weaken the influence due to noise or exceptional circumstances of MI-EEG to a certain degree. Further, they play the role of smoothing. The coarse-grained curves by CMFE have larger fluctuation and more dispersed points than WCMFE, which will yield misclassification in recognition and the poor classification accuracy. In contrast, the filtered MI-EEG by WCMFE is smoother and has lesser short-term variations. It is advantageous to correctly distinguish the different motor imaginary tasks. Linear phase Finite Impulse Response (FIR) filter can process data without phase distortion and it has been widely used in speech signal processing, adaptive processing and other aspects [39]. Its unit impulse response has symmetry property. According to the basis, the weight factors are only selected by the coarse-grained procedure and the rules are as follows: Define the weight factors as that the first and the last points remain equal and are set as 0.1, A τ,2 means the start and the end points are both set as 0.2 and so on. In addition, the middle τ − 2 points of A τ,h remain equal and the sum of the weight factors is 1. When the other parameters are fixed, select different weight factors to calculate WCMFEs in multiscale and input to BP neural network to classification. The appropriate weight factors can be selected according to the recognition rate. The coarse-grained process of CMFE can be regarded as the arithmetic mean filter, which can eliminate the noise to a certain extent and the calculation is simple. In order to detect the filter effect, a comparative experiment of the coarse-grained sequences using WCMFE, CMFE and original MI-EEG on channel C3 with motor imaginary tasks was carried out. Moving average was realized by a sliding window with sample interval of one. In Figure 5, there is no doubt that all of the coarse-grained sequence curves follow the trend of the original MI-EEG and weaken the influence due to noise or exceptional circumstances of MI-EEG to a certain degree. Further, they play the role of smoothing. The coarse-grained curves by CMFE have larger fluctuation and more dispersed points than WCMFE, which will yield misclassification in recognition and the poor classification accuracy. In contrast, the filtered MI-EEG by WCMFE is smoother and has lesser short-term variations. It is advantageous to correctly distinguish the different motor imaginary tasks. To further explain the effectiveness of weight factors in WCMFE, a comparison of the coarsegrained sequences on channel C3 was performed with a sliding window. In this case, the sampling interval was intercepted into [500, 800] to better display the difference between two motor imagery tasks, and the weight factors of selected. In Figure 6, differences of the coarse-grained sequences after these two weighted methods are displayed. In general, the variation trend of coarse-grained sequences produced by the two types of weighting factors is consistent, and the stability of the coarse-grained sequences with A5,4 is better than that with A5,1. To further explain the effectiveness of weight factors in WCMFE, a comparison of the coarse-grained sequences on channel C3 was performed with a sliding window. In this case, the sampling interval was intercepted into [500, 800] to better display the difference between two motor imagery tasks, and the weight factors of A 5 2 5 ] were selected. In Figure 6, differences of the coarse-grained sequences after these two weighted methods are displayed. In general, the variation trend of coarse-grained sequences produced by the two types of weighting factors is consistent, and the stability of the coarse-grained sequences with A 5,4 is better than that with A 5,1 . To further explain the effectiveness of weight factors in WCMFE, a comparison of the coarsegrained sequences on channel C3 was performed with a sliding window. In this case, the sampling interval was intercepted into [500, 800] to better display the difference between two motor imagery tasks, and the weight factors of selected. In Figure 6, differences of the coarse-grained sequences after these two weighted methods are displayed. In general, the variation trend of coarse-grained sequences produced by the two types of weighting factors is consistent, and the stability of the coarse-grained sequences with A5,4 is better than that with A5,1. To acquire better classification results, more appropriate weight factors were selected through experiments. For τ = 3 ∼ 7, m = 2, n = 2, r = 0.15SD, the weight factors were set as A τ,1 , A τ,2 , A τ,3 , A τ,4 to compare. The experimental results are shown in Figure 7. It illustrates that multiple weight factors yield different recognition results. For τ = 4, the most obvious difference in classification results is between A 4,1 and A 4,4 , and it is almost up to 2%. It implies that the appropriate weight factors are very important to obtain a better classification result; the four curves of the recognition results are basically Gaussian distribution with the increase in the scale factors. Further, from τ = 5 to 7, A τ,3 ≥ A τ,4 ≥ A τ,2 ≥ A τ,1 are displayed in order of classification results. When τ = 5, all of the four groups' classification results are the best and A 5,3 has the highest recognition rate, suggesting that A τ,3 is more suitable for MI-EEG in this study. When τ=5, all of the four groups' classification results are the best and Selection of Scale Factor Selection of scale factor needs to be taken seriously. Different scale factors will change the information we obtain and affect the recognition results of MI-EEG in turn. Note that we only do the coarse-grained procedure to select the scale factor τ and the detailed rules can be summarized as follows: There is no coarse-grained operation when τ=1, which means the original MI-EEG. At the sampling interval [450, 900], two trials on channel C3 were selected randomly for imaging left-right hands movement. For the weight factor ,3 A τ , the MI-EEG signals were filtered by the coarse-grained process of WCMFE on different scales. The filtered results are displayed in Figure 8. At multiscale, the coarse-grained sequences fluctuate with the trend of the original MI-EEG, but the fluctuation is smaller. Additionally, with the increase in the scale factor, the smoothness of the filtered MI-EEG is better. At the expense of it, the differences of the curves after filtered become smaller and smaller. Therefore, it has the maximum scale factor of 7 in this study. Selection of Scale Factor Selection of scale factor needs to be taken seriously. Different scale factors will change the information we obtain and affect the recognition results of MI-EEG in turn. Note that we only do the coarse-grained procedure to select the scale factor τ and the detailed rules can be summarized as follows: There is no coarse-grained operation when τ=1, which means the original MI-EEG. At the sampling interval [450, 900], two trials on channel C3 were selected randomly for imaging left-right hands movement. For the weight factor A τ,3 , the MI-EEG signals were filtered by the coarse-grained process of WCMFE on different scales. The filtered results are displayed in Figure 8. At multiscale, the coarse-grained sequences fluctuate with the trend of the original MI-EEG, but the fluctuation is smaller. Additionally, with the increase in the scale factor, the smoothness of the filtered MI-EEG is better. At the expense of it, the differences of the curves after filtered become smaller and smaller. Therefore, it has the maximum scale factor of 7 in this study. Selection of Parameters in FE It can be concluded from Equation (5) that, except for the original sequence length N, the weight factors and the scale factor τ, the calculation of WCMFE is also related to the embedding dimension m, the boundary width r and the boundary gradient n. In this section, we will determine the selection of m, n, and r through experiments to optimize the recognition performance of MI-EEG. The detailed selection rules are as follows: The parameters m and n are fixed to calculate the mean and standard deviation of WCMFE with parameter r for two-class imaginary tasks, respectively. Analogously, we can obtain the mean and standard deviation of WCMFE with parameter m or n, respectively. At the sampling interval of [450,900], training samples of 140 trials were selected to calculate the mean WCMFE on channels C3 and C4, respectively. Then, the definition is as follows: Selection of Parameters in FE It can be concluded from Equation (5) that, except for the original sequence length N, the weight factors and the scale factor τ, the calculation of WCMFE is also related to the embedding dimension m, the boundary width r and the boundary gradient n. In this section, we will determine the selection of m, n, and r through experiments to optimize the recognition performance of MI-EEG. The detailed selection rules are as follows: The parameters m and n are fixed to calculate the mean and standard deviation of WCMFE with parameter r for two-class imaginary tasks, respectively. Analogously, we can obtain the mean and standard deviation of WCMFE with parameter m or n, respectively. At the sampling interval of [450, 900], training samples of 140 trials were selected to calculate the mean WCMFE on channels C3 and C4, respectively. Then, the definition is as follows: D WC = WCMFE C3 − WCMFE C4 , where WCMFE C3 means the WCMFE of MI-EEG on channel C3, and WCMFE C4 represents the WCMFE of MI-EEG

on channel C4. M WC and SD WC are the mean and the standard deviation of D WC , respectively. For τ = 3, they were drawn with the parameters m, n, and r in Figure 9. The bigger difference of M WC as possible and the smaller values of SD WC as possible are the statistical basis of selecting the parameters of m, n and r for better discrimination of two motor imagery tasks. In Figure 9a, it illustrates the changes of M WC and SD WC with the embedding dimension m when τ = 3, n = 2 and r = 0.15SD. The difference of M WC between the two motor imagery tasks is the most obvious at m = 2 and it is the most beneficial to classification. Therefore, m = 2 is taken into consideration. The parameter n determines the boundary gradient of the similar tolerance in the process of fuzzy calculation. The variations of M WC and SD WC with n are shown in Figure 9b when τ = 3, m = 2 and r = 0.15SD. The larger n is, the larger the gradient will be, and the more information will be lost [40]. On the contrary, SD WC is the largest when n is set as 1. The effects of D WC and SD WC on classification are synthesized, a smaller n is selected to 2 in this paper. Similarity tolerance r mainly controls the similarity of template matching [37]. Figure 9c displays the changes of M WC and SD WC with the boundary width r when τ = 3, m = 2 and n = 2. Seen from Figure 9c, with the increase in r, the distinction between the left-right hand motor imagery is going down as well as the values of D WC . In general, more difficult the matching of templates will be with the increase in r. Nevertheless, SD WC increases with the decrease in r; it is also harmful to classify and leads to increased sensitivity to noise. Therefore, we will select r = 0.15 in this paper. Comparison of WCMFE and CMFE Traditional classification algorithms mainly include linear discriminant analysis, SVM, logistic regression and so on. As a complex nonlinear problem, EEG should be considered from the nonlinear perspective in feature extraction and classification algorithms. The Back-Propagation (BP) neural network is a multi-layer feedforward network trained according to the error back-propagation algorithm. Without limiting the number of hidden layer nodes, a Back-Propagation (BP) neural network with only one hidden layer can achieve arbitrary nonlinear mapping [41]. Therefore, the BP network can be used to learn the complex nonlinear problem of MI-EEG recognition, and it does not have any requirements or restrictions on the distribution of training sample data. To verify the effectiveness and separability of WCMFE for two kinds of MI tasks, the comparison between WCMFE and CMFE was conducted and BP neural network was utilized for classification. The BP neural network consists of an input layer, a hidden layer and an output layer. The neuron number of input layer equals to the dimension of feature vector, which is set to 14. The hidden layer is the encoders with six neurons while the output layer has two neurons. The activation functions of neurons are sigmoid functions for input layer and hidden layer, and they are pure linear functions for output layer. The mean squared error is used as the loss function to evaluate the performance during the training process. The structure diagram of the BP neural network is shown in Figure 10. classify and leads to increased sensitivity to noise. Therefore, we will select r = 0.15 in this paper. Comparison of WCMFE and CMFE Traditional classification algorithms mainly include linear discriminant analysis, SVM, logistic regression and so on. As a complex nonlinear problem, EEG should be considered from the nonlinear perspective in feature extraction and classification algorithms. The Back-Propagation (BP) neural network is a multi-layer feedforward network trained according to the error back-propagation algorithm. Without limiting the number of hidden layer nodes, a Back-Propagation (BP) neural network with only one hidden layer can achieve arbitrary nonlinear mapping [41]. Therefore, the BP network can be used to learn the complex nonlinear problem of MI-EEG recognition, and it does not have any requirements or restrictions on the distribution of training sample data. To verify the effectiveness and separability of WCMFE for two kinds of MI tasks, the comparison between WCMFE and CMFE was conducted and BP neural network was utilized for classification. The BP neural network consists of an input layer, a hidden layer and an output layer. The neuron number of input layer equals to the dimension of feature vector, which is set to 14. The hidden layer is the encoders with six neurons while the output layer has two neurons. The activation functions of neurons are sigmoid functions for input layer and hidden layer, and they are pure linear functions for output layer. The mean squared error is used as the loss function to evaluate the performance during the training process. The structure diagram of the BP neural network is shown in Figure 10. For the purpose of eliminating the contingency of the feature extraction, the average classification result can be taken as the average of 10 × 10-fold Cross Validation (CV) based on all 280 trails from the Data set III of BCI Competition II, which contains the training set and the test set; one half is left-hand motor imagery and the other half is right hand motor imagery. In addition, the For the purpose of eliminating the contingency of the feature extraction, the average classification result can be taken as the average of 10 × 10-fold Cross Validation (CV) based on all 280 trails from the Data set III of BCI Competition II, which contains the training set and the test set; one half is left-hand motor imagery and the other half is right hand motor imagery. In addition, the experimental parameters were set as: τ max = 7, m = 2, n = 2, r = 0.15SD and A τ,h = A τ,3 . The classification results are displayed in Table 1. It is clear that although the top classification results of CMFE and WCMFE are equal, the average classification accuracy is slightly better than CMFE. Table 1. Comparison of WCMFE and CMFE in the case of τ max = 7, m = 2, n = 2, r = 0.15SD and A τ,h = A τ,3 . In addition, computation cost is another important index of algorithm. CMFE and WCMFE are compared in the same software (Matlab 2015, Windows10) and hardware (a Hewlett-Packard computer, which is equipped with an Inter(R) Core (TM) i7-9700 CPU @ 3.00GHz, a NVIDIA GeForce RTX 2070 GPU) environment. The computation time of one trail is 6.454 ms for CMFE and 6.477 ms for WCMFE. The small difference is resulted from the coarse-grained process, which is a mean filtering in CMFE and a weighted mean filtering in WCMFE, namely, the weight factor is A τ,h = [1, 1, · · · , 1, 1] ∈ R 1×τ in CMFE and A τ,h = [ h 10 , 5−h 5×(τ−2) , · · · , 5−h 5×(τ−2) , h 10 ] ∈ R 1×τ in WCMFE for each scale 1 ≤ τ ≤ τ max ,τ max = 7 and h = 1. So, τ multiplications are added in calculation of WCMFE for each scale factor τ. Even so, the computation costs of CMFE and WCMFE are very close because of the excellent performance of computer. Statistical Analysis To further analyze the resulted classification difference from WCMFE and CMFE statistically, a two-sample t-test is devoted to detecting whether there is a significant difference when they are applied to extract features of MI-EEG. First, the Lilliefors test (lillietest) is used to verify whether the classification results produced by WCMFE and CMFE conform to the normal distribution. In our experiment, Population 1 and Population 2 represent the classification results of 10 × 10-fold CV corresponding to WCMFE and CMFE, respectively, and they all consist of 100 individuals. The results of the test are displayed in Table 2 where the output results include the Hypothesis test result h, which returned as a logical value, and the p-value, which returned as a scalar value in the range (0, 1). From Table 2, it is obtained that the output results of Population 1 are h = 0 and p = 0.50 > 0.05, which means that the hypothesis that Population 1 is a normal distribution is accepted, and the output results of Population 2 are h = 0 and p = 0.27 > 0.05, which means that the hypothesis that Population 2 is a normal distribution is accepted. So, the two populations all fit the normally distributed. Then, we test the homogeneity of equal pooled variance of populations by Test Grouped Data for Equal Variances and the null hypothesis of the test is that the variances of populations are equal. The results of the test are also shown in Table 2. Where the p-value of the Homogeneity test of variance is 0.09, which is greater than 0.05, it indicates that the null hypothesis that the variances of populations are equal is not rejected. Therefore, the results show that two populations are consistent with normal distribution with equal variance. After the normal distribution and homogeneity of variance of two populations were examined, we would perform the two-sample t-test. Assume that two samples were chosen independently and randomly from the above-mentioned two normal populations with equal variances (namely, Population 1 and Population 2), and they had the same sample size, then the test statistic t could be calculated by Equation (8). Where M WCMFE and M CMFE are the mean values of the two samples, n WCMFE and n CMFE denote the sample size, and S 2 WCMFE and S 2 CMFE stand for the variance, respectively. Especially, n WCMFE = n CMFE =15. Defined, the null hypothesis is H 0 : the classification results of WCMFE and CMFE are derived from independent random samples from normal distributions with equal means; the alternative hypothesis is H 1 : the results of WCMFE and CMFE are derived from populations with unequal means. The significance level can be set as α = 0.05. The decision rule is to reject H 0 , if: p = P t > t α (n − 1) ≤ 0.05 (9) It can be obtained that the value of t is 3.01, and the corresponding value of p is 0.0055, which is less than 0.05. Hence, the null hypothesis H 0 is rejected at the 0.05 significance level. This indicates WCMFE outperforms CMFE in feature extraction of MI-EEG. Comparison with Multiple Traditional Feature Extraction Methods Based on BP Neural Network To illustrate the feasibility of WCMFE in extracting features of MI-EEG, the comparison experiment with DWT, WT+ICA, HHT and WPE in references [5,9,12,15] was carried out. It was executed based on the same dataset (Data set III from BCI Competition II) and BP neural network was the classifier. The classification results are shown in Table 3. It indicates that the classification result of integrating WCMFE with BP neural network is better than the referenced methods. WCMFE has the advantage to quantify the complexity and the irregularity of sequences than the traditional feature extraction methods. Additionally, weight factors of WCMFE reflect the important degree of different sample points and have better adaptability to nonlinear non-stationary signals. "+" means the combination of feature extraction methods or optimization of classifiers; "-" represents that the average recognition rate of ten times repetition of a 10-fold CV is not given in the reference. Comparison of Multiple Entropy-Based Feature Extraction Methods To verify the validity of WCMFE in extracting features of MI-EEG, some comparative experiments were performed based on various nonlinear dynamics methods in the same dataset. The average classification results are displayed in Figure 11. this analysis ignores deeper feature information consistent with ApEn and SampEn. Thanks to the abundant characteristic information from multiple scales, we get better classification results when MSE, MPE, MFE and IMFE are designed to extract features of MI-EEG. Moreover, the CMFE method improves the performance of the coarse-grained sequences to overcome the

drawbacks of the previous entropy methods, which has better stability for short time series. Due to the improved filter method of coarse-grained process, WCMFE enhances the recognition result. In addition, the standard deviation of 10 × 10-fold CV is smaller than the above entropy-based methods; it shows that WCMFE has better stability. Further, a two-sample t-test is designed to detect whether there is a significant difference between MFE and WCMFE or IMFE and WCMFE. The similar experiments were finished as in Section 3.3.5, and the values of p are both less than 0.01. It illustrates the superiority of WCMFE compared with MFE and IMFE in feature extraction of MI-EEG. Figure 11. The average classification accuracies and standard deviations of 10 × 10-fold CV by using WCMFE and multiple nonlinear dynamic methods. Comparison with Multiple Traditional Recognition Methods In this section, the comparison experiments with multiple recognition methods were carried out, including the other traditional recognition methods [5][6][7][8][10][11][12]15] and the top three recognition It is easy to see that ApEn and SampEn have poor classification performance. Due to the usage of Heaviside Function in the similarity measurement, it yields the mutation of entropy value. FE uses fuzzy function instead of Heaviside Function, which has better stability and consistency. However, this analysis ignores deeper feature information consistent with ApEn and SampEn. Thanks to the abundant characteristic information from multiple scales, we get better classification results when MSE, MPE, MFE and IMFE are designed to extract features of MI-EEG. Moreover, the CMFE method improves the performance of the coarse-grained sequences to overcome the drawbacks of the previous entropy methods, which has better stability for short time series. Due to the improved filter method of coarse-grained process, WCMFE enhances the recognition result. In addition, the standard deviation of 10 × 10-fold CV is smaller than the above entropy-based methods; it shows that WCMFE has better stability. Further, a two-sample t-test is designed to detect whether there is a significant difference between MFE and WCMFE or IMFE and WCMFE. The similar experiments were finished as in Section 3.3.5, and the values of p are both less than 0.01. It illustrates the superiority of WCMFE compared with MFE and IMFE in feature extraction of MI-EEG. Comparison with Multiple Traditional Recognition Methods In this section, the comparison experiments with multiple recognition methods were carried out, including the other traditional recognition methods [5][6][7][8][10][11][12]15] and the top three recognition methods [42][43][44] based on the Data set III from BCI Competition II. Table 4 displays the detailed information. The combination of WCMF and BP achieves the highest recognition rate of 100%, and the average result of 10 × 10-fold CV is better than the best one in Data set III from BCI Competition II and the traditional recognition methods in references. It illustrates that WCMFE has better applicability to extract MI-EEG-related features, and it matches the BP neural network as well, which provides a new idea to extract features of MI-EEG signals. Discussion Entropy, as a measure of complexity, has received much attention and been developed well. Especially in consideration of the fuzzy, multiscale, nonstationary and individual difference characteristics of MI-EEG, a personalized WCMFE is proposed to explore its feature extraction problems. As an improved method of CMFE, the weight factors of the coarse-grained process in WCMFE were introduced to change the parameters and performance of filters, yielding the smoother, less overlapping and less fluctuation of filtered MI-EEG signals for left-right hand motor imagery tasks. It is helpful for signal filtering and feature extraction simultaneously, while the pure denoising technology cannot give consideration to feature extraction. Concerning this topic, successive studies were carried out. Based on the ERD/ERS phenomenon, the mean CMFE time series curves of MI-EEG on channels C3 and C4 were drawn in Figure 3 under different motor imagery tasks to determine the sampling interval for showing the best obvious difference, which helps find the time period that has class separability for a subject and will be used in the following experiments. Then, take the scale factor τ = 3 as an example, the impact of weight factors on linear phase FIR filter performance was displayed in Figure 4, the symmetrical form of weight factors A τ,h was determined, and the resulting coarse-grained sequences were different from that of CMFE, as in Figure 5, which indicates that WCMFE can weaken the influence due to noise or exceptional circumstances of MI-EEG to a certain degree. To further demonstrate the effectiveness of different types of weight factors in WCMFE, the effects of weight factors on coarse-grained sequences are shown in Figure 6, which means that under the conditions of the same scale factor τ = 5 and the different parameters h = {1, 4}, the changing trends of coarse-grained sequences are consistent, however, the vibration strengths are each different. Therefore, the multi types of weight factors A τ,h (τ = {3, 4, 5, 6, 7}, h = {1, 2, 3, 4}) were compared and the classification accuracies are shown in Figure 7, in which A 5,3 obtains the highest recognition rate and is more suitable for MI-EEG. Next, the parameters' selection of WCMFE was discussed. When τ varied from 1 to 7 and the weight factor A τ,3 was employed, the coarse-grained sequences over the best time period were demonstrated in Figure 8, where it can be seen that the change of coarse-grained sequences is getting weaker and weaker with the increasing of τ, and the difference between τ = 6 and τ = 7 is very small. Therefore, the maximum of scale factor τ max is set to 7. Furthermore, the embedding dimension m, boundary gradient n and boundary width r were studied through experiments in order to optimize the recognition performance of MI-EEG, and the results are shown in Figure 9 in the case of τ = 3. In the following, a BP network, as a nonlinear classifier, was designed (as in Figure 10) to compare WCMFE and CMFE (see Table 1). After the Lilliefors test, to verify that the feature sample conforms to the normal distribution, shown in Table 2, a two-sample t-test was employed to detect whether there is a significant difference when WCMFE and CMFE are utilized to extract the nonlinear features of MI-EEG, WCMFE demonstrates the superiority of classification rate as well as the comparable computation cost. It is worth noting that WCMFE achieved a minor improvement of 0.68% in classification accuracy compared with CMFE; this might be because only the symmetrical form of weight factors A τ,h and the simple assignment of parameters were implemented. Further, the comparison experiments with traditional feature extraction methods, entropy-based nonlinear dynamic methods and multiple recognition methods in the references were carried out; for the details see Tables 3 and 4, and Figure 11. It suggests that the results by WCMFE and BP neural network are better than the other methods, which indicates that the nonlinear features extracted by WCMFE is matched well with the nonlinear BP neural network classifier, and it is feasible and effective for the feature extraction of MI-EEG using WCMFE. It is also noticeable that the 10-fold cross validation is used for eliminating the contingency of feature extraction. However, we have to point out that the specific form and parameter values of weight factors in WCMFE are artificially set, and we have not a general method to set and obtain their optimal values; further research will continue with regard to the design and optimization of weight factors. In addition, we have only finished some research on a publicly available dataset, and two classes of MI-EEG were classified by using WCMFE. In the future, we will focus on the performance evaluation of WCMFE for multi-class motor imagery tasks and more subjects. Conclusions Aiming at the non-stationary, multi-scale and individual difference characteristics of complex MI-EEG signals, a personalized WCMFE is developed by introducing weight factors in CMFE. Instead of the mean filters in CMFE, a weighted mean filter is applied to the original MI-EEG signal in each scale. This makes the filtered MI-EEG signal, namely, the coarse-grained sequences be coincident with the time-varying characteristic of MI-EEG and have less fluctuation than the original MI-EEG and the coarse-grained sequences by CMFE as well. It is helpful to objectively measure the complexity and represent the deeper nonlinear dynamic features in multiscale. The selection and optimization of the parameters in MFE are analyzed, and several setting modes of weight factors are given and discussed from the perspective of signal processing. The extensive comparative experiments are carried out on a publicly available dataset, and the relatively higher classification accuracy and the comparative computation cost show the effectiveness of WCMFE. The proposed WCMFE is an important supplement of Entropy theory, and it will promote the application of CMFE in time-varying signals, especially the biological signals such as Electrocardiographic (ECG) and Electromyographic (EMG). However, only the linear phase FIR filters were considered in our study and the parameters of weight factors were artificially set, this simplifies the design steps but limits the performance of WCMFE. How to design more reasonable filters to improve WCMFE is a potential problem. Intravenous Paracetamol Versus Patient-Controlled Analgesia With Morphine for the Pain Management Following Diagnostic Knee Arthroscopy in Trauma Patients: A Randomized Clinical Trial Background: Most patients undergoing outpatient surgeries have the unpleasant experience of high level pain after surgery. Compared with open surgeries, arthroscopic procedures are less painful; however, inadequate pain management could be associated with significant concerns. Opioids alone or in combination with local anesthetics are frequently used for diminishing postoperative pain using intravenous or epidural infusion pumps. Despite morphine various disadvantages, it is commonly used for controlling pain after surgery. Objectives: The aim of this study was to compare intravenous paracetamol and patient-controlled analgesia (PCA) with morphine for the pain management following diagnostic knee arthroscopy in trauma patients. Patients and Methods: Sixty trauma patients who were scheduled to undergo knee arthroscopy were randomly divided into two groups. Patients immediately received intravenous infusion of 1 g paracetamol within 15 minutes after surgery and every 6 hours to 24 hours in the paracetamol group. The patient-controlled analgesia group received morphine through PCA infusion pump at 2 mL/h base rate and 1mL bolus every 15 minutes. Pain level, nausea and vomiting, and sedation were measured and recorded during entering the recovery, 15 and 30 minutes after entering the recovery, 2, 6, and 24 hours after starting morphine pump infusion in the morphine and paracetamol in the paracetamol groups. Results: There was no significant difference regarding the pain level at different times after entering the recovery between the two groups. No one from the paracetamol group developed drug complications. However, 22.3% in the PCA morphine suffered from postoperative nausea; there was a statistically significant difference regarding the sedation level, nausea, and vomiting at various times between the two groups. Conclusions: Intravenous administration of paracetamol immediately after knee arthroscopy improved postoperative pain, decreased analgesic administration, maintained stable hemodynamic parameters, had no complications related to opiates, no nausea and vomiting, and increased patient satisfaction and comfort in comparison to PCA with morphine. Background Patient safety has always been a major concern for the physician in all eras (1) and even ancient physicians used their own methods to reduce pain of the patients (2). Recent advances in surgical techniques and anesthesia have increased the number of surgeries in outpatient setting worldwide. Consequently, more than 60% of patients in the West are outpatient surgery cases (3). Diagnostic and therapeutic arthroscopies are common procedures for outpatient surgeries under general anesthesia or regional anesthesia. Compared with open surgeries, arthroscopic procedures are less painful; however, inadequate pain management could be associated with signifi-cant concerns. Cost-effectiveness of outpatient surgeries is well-known. Effective postoperative pain controlling is necessary and important for optimizing patients comfort and satisfaction. Furthermore, providing a rapid recovery and reducing pain and discomfort for controlling pain after outpatient surgery is necessary to improve the patient outcome. Different analgesic techniques from multimodal approaches (4,5) to blocks (6,7), corticosteroids (7,8), and local anesthetics (9) have been introduced. Analgesics administration is growing up in outpatient surgeries in order to prevent postoperative pain. Inad-equate control of pain after surgery is one of the key factors for failing to discharge patients (3,10,11). Postoperative pain control improves the patient's ability and increases their performance in their daily

life activities (12). Most patients undergoing outpatient surgeries have the unpleasant experience of high level pain after surgery (3,10,11). About 30% -40% of the outpatient surgery candidates experience moderate to severe pain during first 24 -48 hours after surgery (13)(14)(15). In some cases, the pain may become more severe over time and affect sleep and daily activities (16,17). Moreover, the duration and type of surgery have a significant effect on pain severity which may ultimately lead to further analgesic requirements (13,15,18). Different techniques of anesthesia and analgesia have their own advantages and disadvantages. Patient-controlled analgesia (PCA) pump is one of the pain management methods in which the patient manages the pain using programmed infusion and further analgesic boluses. Drug infusion pump is set according to patient's needs to maintain adequate analgesia. Different groups of analgesics exert their effects through different mechanisms. Opioids are gold-standard analgesics for postoperative analgesia; however, due to their undesirable side effects, such as apnea, urinary retention, nausea, and vomiting can lead to patient discomfort. Opioids alone or in combination with local anesthetics are frequently used for diminishing postoperative pain using Intravenous or epidural infusion pumps. Despite morphine various disadvantages, it is commonly used for controlling pain after surgery. Onset time for intravenous morphine is 1 -2 minutes after injection; its peak effect is 3 -5 minutes after injection and the duration of its analgesic effect is 1-2 hours (18). Opioids and none-steroids anti-inflammatory drugs (NSAIDs) are the main analgesic drugs that are commonly used to control pain after surgery. Other strategies are recently implemented to improve postoperative pain control and decline the side effects associated with opioids. Intravenous paracetamol (acetaminophen) is one of the new strategies used for improving mild to moderate pain intensity (19). With an onset of action of 5 -10 minutes after the injection, acetaminophen displays a peak effect one hour after injection and its analgesic effects endure for 4 -6 hours (18)(19)(20). Paracetamol is used for managing mild to mediocre pains (20). Objectives The present study was conducted to compare Intravenous paracetamol and patient controlled analgesia with morphine for the pain management following diagnostic knee arthroscopy in Trauma patients. Patients and Methods This randomized clinical trial was conducted on 60 eligible trauma patients who were scheduled for under-going diagnostic knee arthroscopy at Akhtar hospital in Tehran city, Iran, during October 2013 to October 2014. Sample size was acquired using the two means comparison formula. The power value was determined to be 80%, with an assumed dropout rate of 20%. Considering d = 4, sample size of 30 for each group was obtained. Before recruitment of first subject, study protocol was approved by local ethics committee of Shahid Beheshti University of medical sciences. The study has been performed in accordance with the ethical standards of the 1964 Declaration of Helsinki. All patients signed the informed consent forms prior to recruitment in the study. The sample consisted of sequential selection of 60 patients who were divided consecutively and randomly into two groups (intravenous paracetamol or morphine PCA pumps). Inclusion criteria were: trauma patients undergoing diagnostic knee arthroscopy under spinal anesthesia, patients between 15 -60 years of age, American society of anesthesiology classification (ASA) I-II, lack of other coexisting diseases and lack of history of neurologic drugs use. Exclusion criteria were: contraindications of spinal anesthesia, history of analgesic medication 24 hours before surgery, inability to use the PCA, history of bleeding or coagulopathies within the previous month, liver or kidney failure, severe cardiopulmonary, obesity (body mass index > 30 kg/m 2 ), history of postoperative nausea and vomiting, history of migraine, history of complications during or after surgery, postoperative complications during 24 hours after surgery requiring intervention, known allergy, allergy or contraindications to opioid analgesics or nonopioid drugs, pregnancy or lactation, and history of alcohol or drug abuse. All the patients were operated with the simultaneous presence of the same surgeon and anesthetist and the same surgical routine. The duration of the procedure was considered to be 30 to 40 minutes and in case of any complication which would make it longer, the patients were excluded from the study. All the patients first received preanesthesia consisting of 50 mg of fentanyl and 1 mg of midazolam. Blood pressure and heart rate were measured and recorded by Non invasion Blood Pressure (NIBP) monitoring and a pulse oximetry. Spinal anesthesia was made by a puncture in the L4-L5 intervertebral space and infusion of 10 mg of 0.5% isobaric Bupivacaine®. Systolic pressure decline of more than 30% of base level or less than 100 mmHg was considered as hypotension which was treated by infusion of intravenous fluids and 6 mg of ephedrine every 2 minutes until reaching normal blood pressure levels. Decrease of heart rate to less than 60 beats per minute was considered a major change, which was treated by the administration of 0.5 mg of intravenous atropine. After the procedure, patients of the first group received intravenous infusion of 1g paracetamol (Apotel; Uni-Pharma. SA, India) within 15 minutes after surgery and every 6 hours to 24 hours in the paracetamol group. At the same time, The PCA group received morphine through PCA infusion pump at 2 mL/h base rate and 1 mL bolus every 15 minutes. Pain level, nausea and vomiting and sedation were measured and recorded during entering the recovery, 15 and 30 minutes after entering the recovery, 2, 6, and 24 hours after starting morphine pump infusion in the morphine and paracetamol in the paracetamol groups. Side effects of morphine were assessed every 4 hours in the first 24 hours after surgery. The drugs used in the first 24 hours after surgery, possible side effects of paracetamol (rash, thrombocytopenia, leukopenia, hypotension, and renal or liver failure), and patient analgesia satisfaction quality after 24 hours were evaluated based on VAS (zero = dissatisfaction, 1 -3 = medium satisfaction, 4 -6 = good satisfaction, and 7 -10 = high satisfaction). The assessment was performed in recovery and in the ward by trained nursing staff who were not part of the team for the present study. Data were analyzed using SPSS 19 (SPSS Inc, Chicago, IL, USA). Comparisons between the groups were made using Chi-square or Fisherʼ s exact test when appropriate and P value less than 0.05 was considered statistically significant. Results In the present study, 30 patients (50%) were in the paracetamol and 30 patients (50%) in the PCA morphine groups. Comparison of demographic data between the groups showed no significant statistical difference (Table 1). Changes in pain score between the two groups at different times are demonstrated in Figure 1. There was no significant difference regarding the pain level at different times after entering recovery between two groups (P = 0.076). Recovery time in the paracetamol and PCA morphine, administered further analgesic (morphine), the time for receiving first dose of opioids were compared and analyzed. None of the mentioned outcome measures showed a statistically significant difference between the two groups (P > 0.05) ( Table 2). No one from the paracetamol group developed drug complication. Nevertheless, 7 patients (22.3%) in the PCA morphine suffered from postoperative nausea; there was a statistically significant difference between the two groups in this regard (P = 0.005). There was a statistically significant difference regarding the sedation level at various times between the two groups. Furthermore, there was a statistically significant difference regarding nausea and vomiting at the various times between the two groups. Moreover, there was a statistically significant difference regarding patient satisfaction between the two groups (P = 0.0001). Discussion Effective control of postoperative pain is a major concern for anesthesiologists. Despite the development of improved methods of pain control, patients still suffer from inadequate postoperative pain management. Effective control of postoperative pain by opioids has several side-effects such as nausea, vomiting, pruritus, urinary retention and respiratory depression (23). Therefore, the use of non-opioids has become more popular to avoid opioid-related side-effects (24). In this study, changes in pain scores between the two groups depicted no significant difference; during the 24 hours after surgery, paracetamol and morphine were similarly reduced the severity of postoperative pain. The pain severity 24 hours after operation was mild and none of the patients had pain in both paracetamol and morphine groups. Intravenous paracetamol has been reported to provide effective analgesia and pain control and consequently reduced opioid consumption after breast surgery outpatients. Reducing opiates consumption is very important because it leads to a decrease in side-effects and life-threatening complications of opioids. Admittedly, any therapy that reduces total opioid administration dose is considered an advantage. The results of this study are consistent with previous studies in which paracetamol declines the administration of opiates (25)(26)(27). In the clinical studies, one gram of intravenous paracetamol has been reported to be as effective as Ketorolac 30 mg, 75 mg diclofenac, and morphine 10 mg. It is beneficial for the treatment of moderate to severe pain after surgery (28,29). Sinatra et al. study showed that one gram of intravenous paracetamol in patients with moderate to severe pain resulted in rapid and effective analgesia in the first 24 hours after orthopedic surgeries (30). Another study showed that nonopioid analgesics are beneficial as well as opioids in controlling pain after endonasal surgery (31). The exact mechanism of action of paracetamol has not been determined. Paracetamol may lead to the activation of serotonin receptor II and inhibition of cyclooxygenase pathway in the synthesis of prostaglandins with doublebarreled effect (32)(33)(34). Paracetamol may be associated with hemorrhagic side-effects by inhibiting the synthe-sis of prostaglandins and related compounds which has only been reported for NSAIDs (32). There was no bleeding case in the present study. All of the paracetamol patients were calm and alert at discharge while 7 patients of the PCA morphine were anxious, agitated, and restless 15 and 30 minutes after entering the recovery. This may be due to insufficient control of pain in the paracetamol group. In the present study, no significant changes in systolic blood pressure, diastolic blood pressure, means arterial blood pressure, and heart rate were observed in the paracetamol group. Paracetamol may decrease cardiac output and peripheral vascular resistance (35). In our study, systolic and diastolic blood pressure, and mean arterial blood pressure decline was 5 mmHg and blood pressure was maintained within the normal range; none of patients developed hypotension. Also, in the morphine pump group, imbalance in hemodynamic was observed frequently; while, the paracetamol group patients experienced a further stable hemodynamics. The results obtained from our study presented that intravenous administration of one gram of paracetamol in diagnostic arthroscopic knee surgery immediately after the surgery has clinical benefits such as improved postoperative pain, decreasing of analgesics administration, stable hemodynamic parameters, reducing length of hospitalization in the recovery room, lack of complications related to opiates, no nausea and vomiting, and increased patient satisfaction and comfort. Conducting studies with different doses of paracetamol, different surgical processes and larger sample sizes are recommended for the future studies. Footnote Authors' Contribution:Seyed Masoud Hashemi had primary responsiblity for protocol development, patient screening, enrollment, outcome assessment, preliminary data analysis and writing the manuscript. Aliakbar Esmaeelijah and Samad Golzari participated in the development of the protocol and analytical framework for the study and contributed to the writing of the manu-script. Azita Maserrat contributed in the development of the protocol and analytical framework for the study and contributed to the writing of the manuscript and was responsible for patient screening. Gholamreza Mohseni supervised the design and execution of the study, performed the final data analyses and contributed to the writing of the manuscript. Evaluation of the immune response in conventionally weaned pigs infected with porcine deltacoronavirus Although porcine deltacoronavirus (PDCoV) is a significant pandemic threat in the swine population and has caused significant economic losses, information regarding the immune response in conventionally weaned pigs infected with PDCoV is scarce. Hence, the immune response in conventionally weaned pigs infected with PDCoV was assessed after challenge and rechallenge. After the first challenge, obvious diarrhea and viral shedding developed successively in all pigs in the four inoculation dose groups from 3 to 14 days postinfection (dpi), and all pigs recovered (no clinical symptoms or viral shedding) by 21 dpi. All pigs in the four groups exhibited significantly increased PDCoV-specific IgG, IgA and virus-neutralizing (VN) antibody (Ab) titers and IFN-γ levels in the serum

after the first challenge. All pigs were completely protected against rechallenge at 21 dpi. The serum levels of PDCoV-specific IgG, IgA, and VN Abs increased further after rechallenge. Notably, the IFN-γ level declined continuously after 7 dpi. In addition, the levels of PDCoV-specific IgG, IgA and VN Abs in saliva increased significantly after rechallenge and correlated well with the serum Ab titers. Furthermore, the appearance of clinical symptoms of PDCoV infection in conventionally weaned pigs was delayed with reduced inoculation doses. In summary, the data presented here offer important reference information for future PDCoV animal infection and vaccine-induced immunoprotection experiments. PDCoV is an enveloped, single-stranded, positivesense RNA virus [7,12]. The PDCoV genome contains a 5' untranslated region (UTR) followed by open reading frame 1a/1b (ORF 1a/1b); sequences encoding the spike (S), envelope (E), membrane (M), accessory protein 6 (NS6), nucleocapsid (N), accessory protein 7 (NS7), and accessory protein 7a (NS7a) proteins; and a 3' UTR [13,14]. The E protein is a small envelope protein that plays an important role in viral assembly and infection [15]. The M protein plays an important role in inducing protection and mediating the course of disease [16,17]. The N protein is principally produced in infected cells and plays multiple roles in viral replication and pathogenesis [13,18]. The S protein contains three receptor-binding S1 subunits tightly packed into a crown-like structure and three membrane fusion S2 subunits forming a stalk. This protein has several structural features that may facilitate viral immune evasion, including compactness, concealed receptor-binding sites, and shielded critical epitopes [19]. Because the S protein is the major surface protein, it is also the main target of neutralizing antibodies (Abs) during infection and a focus in vaccine design [20]. PDCoV can infect pigs regardless of age; unweaned piglets infected with PDCoV display clinical symptoms similar to those seen in pigs infected with porcine epidemic diarrhea virus (PEDV) and transmissible gastroenteritis virus (TGEV), including watery diarrhea, vomiting, dehydration and death [7,11,21]. Although previous studies have confirmed the postinfection effect of PDCoV in newborn and suckling piglets, few reports have addressed the pathogenicity of and immune response to PDCoV in conventionally weaned pigs. In addition, although a correlation of resistance to disease induced by PEDV infection with age has been described, little is known about the effect of PDCoV infection and re-challenge on weaned pigs. Therefore, 45-day-old weaned pigs were challenged and rechallenged with different inoculation doses of sixth-passage (P6) culture-adapted PDCoV strain CH/XJYN/2016 (GenBank accession number MN064712), and the immune response in the PDCoVinfected, conventionally weaned pigs was evaluated. Twenty-five 45-day-old PDCoV-naïve conventionally weaned pigs were obtained from a commercial pig farm with no previous herd history of a PDCoV outbreak and confirmed to be seronegative for PDCoV by indirect ELISA (developed in our laboratory). In addition, these pigs were confirmed to be negative for PEDV, TGEV and porcine kobuvirus (PKoV) by reverse transcription PCR (RT-PCR) and real-time PCR (developed in our laboratory). The pigs were randomly divided into four experimental groups (G1-G4) and one mock-infected control group, each containing five pigs. Each group of pigs was housed in a different room. Pigs in groups G1 to G4 were separately challenged orally with 3 mL of tenfold serial dilutions (from 10 0 to 10 −3 ) of P6 cell-culture-adapted CH/XJYN/2016 (original titer: 4.0 log 10 TCID 50 /mL). The mock-infected group received MEM. At 21 days post-inoculation (dpi), pigs in G1 to G4 were rechallenged with 1000 TCID 50 of P6 cell-culture-adapted CH/XJYN/2016. After challenge, clinical symptoms were observed and the fecal consistency was scored daily as described previously (scores: 0 = normal, 1 = pasty, 2 = semiliquid, and 3 = liquid) [22]. Serum samples and rectal swabs were collected from all pigs at 0, 7, 14, 21, 28 and 35 dpi. Saliva specimens were collected at 21, 28 and 35 dpi. An indirect ELISA was developed by our laboratory to measure the serum Ab titers in the pigs. In brief, 96-well plates (Costar, USA) were coated with eukaryotically expressed N protein. Each well was then blocked with 100 μl of a solution containing 1.5% bovine serum albumin (Solarbio, Beijing) and incubated at 37 °C for 1 h. After three washes with PBS (1×, pH 7.2-7.4), the plates were loaded with 100 μl of serum sample (diluted 1:100) or original saliva sample per well. Horseradish peroxidase (HRP)conjugated goat anti-pig IgG (Abcam, Cambridge, UK) diluted 1:30,000 (serum) or 1:8,000 (saliva) was added, and the reaction was visualized using 3,3´,5,5´-tetramethylbenzidine (TMB) (Solarbio ® , Beijing). The titers were determined by measuring the optical density (OD) at 450 nm using an ELISA plate reader (Thermo Scientific, Waltham, USA) operated with professional software (SkanIt RE 4.1). A virus neutralization assay was performed as in our previous study with some modifications [22]. Twofold serial dilutions of serum starting at 1:2 were coincubated at 37 °C for 1 h with equal volumes of viral stock containing 200 TCID 50 of PDCoV strain CH/XJYN/2016 in 96-well plates (Corning, USA). Virus-neutralizing (VN) Ab titers were calculated as the reciprocal of the highest serum dilution that prevented a cytopathic effect (CPE). The levels of IFN-γ in serum samples from all groups of piglets were determined using a commercial porcine IFN-γ ELISA kit (Solarbio ® , China) according to the manufacturer's instructions. A real-time PCR method based on the PDCoV N gene was developed in our laboratory. The primer and probe sequences were as follows: sense, 5'-ACG TCG TAA GAC CCA GCA TC-3'; antisense, 5'-CCC ACC TGA AAG TTG CTC TC-3'; probe B, FAM-GTA TGG CTG ATC CTC GCA TCA TGG C-BHQ1. The PCR conditions were as follows: 42 °C for 5 min and 95 °C for 10 s, followed by 39 cycles of 95 °C for 10 s and 57 °C for 20 s. Cycle threshold (Ct) values greater than 30 were considered negative; Ct values less than 30 were considered positive. Statistical analysis was performed using SPSS 16 software. Statistical significance among the different experimental groups was determined using one-way ANOVA with Tukey's test for multiple comparisons. Differences for which the p-value was less than 0.05 were considered significant. All piglets were monitored closely before and after the experiment. Prior to the first challenge, all pigs in the five groups were healthy and tested negative for PDCoV-specific IgG Ab and DNA by real-time PCR. After the first challenge with the original concentration and tenfold serial dilutions of P6 cell-culture-adapted CH/XJYN/2016, the clinical symptoms in each group were observed, and fecal viral shedding was detected (Table 1). No obvious clinical symptoms were observed in any pigs in the four experimental groups, and viral shedding in all fecal samples was negative at 1 and 2 dpi. All pigs in the original virus challenge group (G1, n = 5) and the 10 −1 -diluted virus challenge (G2, n = 5) group exhibited obvious diarrhea and fecal viral shedding at 4 and 6 dpi, respectively (Table 1). In the groups challenged with 10 −2 -and 10 −3 -fold dilutions of virus, 3 of 5 and 4 of 5 pigs, respectively, displayed diarrhea and viral shedding at 14 dpi (Table 1). All pigs in groups G1-G4 recovered (no clinical symptoms or viral shedding) by 21 dpi. In addition, no pigs in groups G1-G4 exhibited any obvious clinical symptoms or viral shedding in feces after rechallenge with 3.0 log 10 TCID 50 /ml of virus at 21 dpi (Table 1). The mock-infected control group exhibited no clinical symptoms or viral shedding from 0 to 35 dpi ( Table 1). The postinoculation titers of serum PDCoV-specific IgG Abs among G1-G4 pigs showed a sustained increase from 7 to 14 dpi (Fig. 1a). However, the PDCoV-specific IgG Ab titers in G1 and G2 pigs decreased at 21 dpi, whereas those in G3 and G4 pigs continued to increase at 21 dpi (Fig. 1a). This difference may be due to the differences in the onset time of the clinical signs. After rechallenge with a 10 3 TCID 50 dose of P6 cell-culture-adapted CH/XJYN/2016 at 21 dpi, the PDCoV-specific IgG Ab titers in G1-G4 pigs increased further from 28 to 35 dpi (Fig. 1a). The PDCoVspecific IgG Ab titers in the mock-infected control group did not change significantly throughout the experiment (Fig. 1a). The PDCoV-specific IgG Ab titers in G1-G4 pigs were significantly elevated (p < 0.05) compared with those in the mock-infected control group from 7 to 35 dpi (Fig. 1a). Similar to the serum PDCoV-specific IgG Ab titers, the serum PDCoV-specific IgA Ab titers increased from 7 to 14 dpi in G1-G4 pigs and declined slightly at 21 dpi, except in G4 (Fig. 1b). After rechallenge with a 10 3 TCID 50 dose of P6 cell-culture-adapted CH/XJYN/2016 at 21 dpi, the serum PDCoV-specific IgA Ab titers increased significantly in G1-G4 pigs from 28 to 35 dpi (Fig. 1b). The PDCoVspecific IgA Ab titers in the mock-infected control group did not change significantly throughout the experiment (Fig. 1b). The PDCoV-specific IgA Ab titers in G1-G4 pigs were significantly elevated (p < 0.05) compared with those in the mock-infected control group from 7 to 35 dpi. In addition, in all G1-G3 pigs, the VN Ab titers increased from 7 to 14 dpi but declined slightly at 21 dpi (Fig. 1c). Similarly, after rechallenge with a 10 3 TCID 50 dose of P6 cell-culture-adapted CH/XJYN/2016 at 21 dpi, the VN Ab titers increased markedly from 28 to 35 dpi (p < 0.05; Fig. 1c). No VN Abs were detected in the mock-infected control group. Almost all previously conducted animal studies with PDCoV have used newborn and suckling piglets as animal models [23] and have indicated that the pathogenicity and clinical symptoms differ in pigs of different ages, a concept called the "age-dependent axis" [24]. Therefore, to better understand the pathogenicity and immune response, this PDCoV infection study used conventionally weaned pigs as an animal infection model. Unlike piglets, which typically develop diarrhea at 1-2 dpi after PDCoV challenge, the conventionally weaned pigs in groups G1 and G2 in this study developed diarrhea and viral shedding at 3-7 dpi. Furthermore, G3 and G4 pigs exhibited no clinical symptoms or viral shedding up to 7 dpi. These results Blood samples were collected from each pig at 0, 7, 14, 21, 28, and 35 days after the first challenge, and the PDCoV-specific IgG and IgA Ab titers were measured using an indirect ELISA developed in our laboratory. The average OD value in each group (n = 5) was measured, and the standard deviation (SD) was plotted at each point. Significant differences among the five groups at 7-35 dpi are indicated by asterisks (p < 0.05). In VN tests, serum samples were heatinactivated for 30 min at 56 °C before being serially diluted twofold (50 µL) and mixed with a volume equal to 200 TCID 50 of the CH/ XJYN03/2016 strain. The samples were examined using an inverted microscope on day 5. The VN Ab titer in each serum sample was measured three times, and the average was calculated. Significant differences are indicated by asterisks (p < 0.05) Ab titers in saliva after rechallenge. Saliva was extracted by soaking in 1× PBS followed by centrifugation. The average OD value in each group (n = 5) was measured, and the SD was plotted on each bar graph. Significant differences are indicated by asterisks (p < 0.05). In the VN tests, saliva samples were heat-inactivated for 30 min at 56 °C, serially diluted twofold (50 µL) and mixed with a volume equal to 200 TCID 50 of the CH/XJYN03/2016 strain. The samples were exam-ined using an inverted microscope on day 5. The VN Ab titer in each saliva sample was measured three times, and the average VN Ab titer was calculated. Significant differences are indicated by asterisks (p < 0.05). (d) The concentration of serum IFN-γ in weaned pigs after two challenges. Serum IFN-γ levels were measured using a commercial ELISA kit. The concentration of serum IFN-γ was obtained from a standard curve by multiplying by the dilution factor. The average serum IFN-γ levels were calculated, and the SD was plotted at each point. Significant differences are indicated by asterisks (p < 0.05) indicate that the appearance of clinical symptoms induced by PDCoV infection is delayed in conventionally weaned pigs with reduced inoculation doses. A similar pattern has

been observed in previous PEDV infection studies [25,26]. Interestingly, we observed that the levels of IFN-γ were negatively correlated with the challenge dose and that viral rechallenge had no effect on the level of serum IFN-γ. A previous study showed that PDCoV significantly induced expression of IFN mRNA in infected Peyer's patches at 3 dpi [27]. In this study, however, the levels of serum IFN-γ peaked 7 dpi. The difference may be due to the longer sampling interval. Therefore, a shorter sampling interval will help to obtain more-detailed information about the levels of serum IFN-γ after PDCoV-infection. Generally, swine enteric coronavirus challenge studies use the TCID 50 as the infectious titer [11,21]. Practically, however, the infectious titer of PDCoV strains differs between pigs and cells, but the median pig diarrhea dose (PDD 50 ) of PDCoV in pigs of different ages is unknown. Therefore, before starting this study, a separate experiment was performed, and the results showed that 1000 TCID 50 is the lowest dose that can still reliably produce clinical symptoms in weaned pigs. Therefore, in this study, the infection dose used for rechallenge was 1000 TCID 50 . In this study, the dynamics of the PDCoV-specific IgG, IgA and VN Ab titers in serum and saliva were determined after PDCoV challenge and rechallenge, and the serum IFN-γ levels were measured. The results indicated that the serum titers of PDCoV-specific IgG, IgA, and VN Abs correlated well with those in saliva and revealed that the appearance of PDCoVinduced clinical symptoms in conventionally weaned pigs is delayed with reduced inoculation doses. Finally, all pigs were completely protected against rechallenge after infection with different PDCoV inoculation doses. The Modulation of Phosphatase Expression Impacts the Proliferation Efficiency of HSV-1 in Infected Astrocytes Herpes Simplex Virus 1 (HSV-1) is a major pathogen that causes human neurological diseases, including herpes simplex encephalitis (HSE). Previous studies have shown that astrocytes are involved in HSV-1 systemic pathogenesis in the central nervous system (CNS), although the mechanism remains unclear. In this study, a high-throughput RNAi library screening method was used to analyze the effect of host phosphatase gene regulation on HSV-1 replication using Macaca mulatta primary astrocytes in an in vitro culture system. The results showed that the downregulation of five phosphatase genes (PNKP, SNAP23, PTPRU, LOC714621 and PPM1M) significantly inhibited HSV-1 infection, suggesting that these phosphatases were needed in HSV-1 replication in rhesus astrocytes. Although statistically significant, the effect of downregulation of these phosphatases on HSV-1 replication in a human astrocytoma cell line appears to be more limited. Our results suggest that the phosphatase genes in astrocytes may regulate the immunological and pathological reactions caused by HSV-1 CNS infection through the regulation of HSV-1 replication or of multiple signal transduction pathways. Introduction Herpes Simplex Virus 1 (HSV-1), a DNA virus with a complex gene expression program [1], can not only manipulate transcription regulation leading to latent infection [2], but also cause comprehensive pathological injury, such as herpes simplex encephalitis (HSE) with poor prognosis, in hosts during its infection process [3,4]. The studies of the mechanism of HSV-1 virus genetic transcription regulation and HSV-1 pathogenesis have shown that the interaction between HSV-1 and the host largely determines the specific virus infection process [5,6]. Therefore, studies of the molecular events of interactions between viruses and host cells may provide direct evidence for understanding the mechanism of HSV-1 pathogenesis. Previous studies have shown that astrocytes in the central nervous system (CNS) are involved in the pathogenesis of HSV-1 encephalitis [7]. This involvement occurs because astrocytes, with a fast response to invasive pathogens [8], are important immune cells in the CNS, and because of the capability of HSV-1 to infect astrocytes and promote systemic pathogenesis in the CNS [9]. Studies have suggested that production of certain inflammatory factors, such as MCP-1 and TNF, were increased following HSV-1 infection in astrocytes [10], which was observed to be mediated in a TLR-2-dependent manner [10]. Although the mechanism of interaction between host and HSV-1 in astrocytes, especially the impact of cytokine production on the viral replication, is not clear, the phosphatases involved in the TLR2 response signaling pathway were confirmed to be very important components [11,12]. Thus, to explore the role of certain phosphatases in HSV-1 replication might be helpful to yield insight into the cellular response to viral infection and into their further significance in the pathogenesis of HSV-1 infection in astrocytes, which is a part of the pathological process of encephalitis induced by this virus [13]. Therefore, our work in this study was focused on the interaction of phosphatases with HSV-1 in astrocytes to better understand the molelular mechanism HSV-1 pathogenesis in infectious encephalitis. Studies of the biological activities of astrocytes have shown that as a major component of the CNS, astrocytes play important roles in neuron regulation and immune responses [14,15]. In general, the neuron-astrocyte communication is very important for the systemic stability of the CNS [16], which includes both the activation of astrocytes by synaptically released neurotransmitters and their signaling back to neurons [16]. During HSV-1 infection, upregulation of cytokine release by infected astrocytes may alter the normal neuron-astrocyte communication and lead to local instability of the CNS. [17]. This study investigated the mechanism involved in HSV-1 infection of astrocytes. We focused on using an RNAi library screening method to evaluate the role of phosphatases on HSV-1 infection of astrocytes. We targeted phosphatases directly involved in cell regulation to understand the molecular mechanism of the change in the biological activities of astrocytes during HSV-1 infection. Ethics statement All animal experimental procedures were carried out in strict accordance with the National Institute of Health Guidelines for the Care and Use of Laboratory Animals. The experimental animal procedure was approved by the Office of Laboratory Animal Management of Yunnan Province and the Ethics Committee of the Institute of Medical Biology. The monkey was kept and bred according to the guidelines of the committee of experimental animals at the Institute of Medical Biology, Chinese Academy of Medical Sciences (CAMS) [18]. We used a single monkey, with a weight of 3 kg and an age of 3.5 years, for each primary culture of astrocytes; astrocyte collection was performed three times in this study. Before the initiation of the study, the monkey was kept in isolation for 1 day from a group that is provided with toys and facilities to allow swinging, jumping and engaging in activities from floor to ceiling. The monkey was housed in a pen with a height of 1.8 m and an area of 2.5 m 2 in a light-dark (LD) 12:12 cycle (12 hours light, 12 hours dark). This monkey has constant access to water which had been purified with filters, sterilized or additionally chlorinated or acidified in order to eliminate biological pollution and was fed every day with sterilized foods and fresh apples. This monkey was euthanized with sodium pentobarbital (10 mg/kg). The cerebral material was obtained 15 minutes after the sacrifice of the animal. Preparation of monkey cerebrocortical astrocytes Astrocytes were prepared from the cortex of adult monkeys using a protocol derived from previously described methods [19][20][21]. Cortices were washed several times with PBS, cut into small cubes (,1 mm 3 ) and digested with 0.25% trypsin for 25 min at 37uC. Trypsinization was terminated by the addition of Dulbecco's modified Eagle's medium (DMEM) that contained 10% fetal bovine serum (FBS) (GIBCO). Cell suspensions were sieved through a 40 mm cell strainer. The filtrate was seeded at a density of 1610 6 cells/cm 2 in 100 mm dishes (Corning, USA) after being pre-adhered for 30 min to remove fibroblasts. The plated cells were cultured in a 5% CO 2 incubator at 37uC for 21 days and the culture medium was changed at 5-day intervals as described [22,23]; After confluence, the primary cell culture was washed 3 times with PBS and digested with 0.25% trypsin at room temperature for 5 min. Trypsinization was also terminated by the addition of DMEM containing 10% fetal bovine serum. After centrifugation, the cells were subsequently cultured in DMEM containing 10% fetal calf serum, 100 U/ml penicillin and 20 mg/ ml streptomycin in a 5% CO 2 incubator at 37uC. Other glia, such as oligodendroglia and microglia were removed during this procedure. The purity of astrocytes was assessed by the percentage of cells showing fluorescence for the glial fibrillary acidic protein (GFAP) marker. SiRNA transfection Both the siRNA used for knockdown and the siRNA used as a control were designed and synthesized by Guangzhou RiboBio Co., Ltd. One day before transfection, the cells were plated in growth medium without antibiotics. The cell monolayer was then transfected with siRNA at a final concentration of 50 nM with the Lipofectamine TM 2000 transfection reagent (Invitrogen) according to the manufacturer's instructions. Cell lines and Viruses HEK293 and Vero cells were grown in Modified Eagle Medium (MEM) supplemented with 10% newborn bovine serum (GIBCO). U-87MG cells were grown in Dulbecco's Modified Eagle's Medium (DMEM) supplemented with 10% fatal bovine serum (GIBCO). Stocks of HSV-1(strain 17) [5] were prepared in Vero cells and gradient purified. Growth curve analysis The cell monolayers were then infected with HSV-1 at a MOI of 0.01 PFU/cell. The cell culture supernatants were harvested at 0, 12, 24, 36, 48, 60, 72, 84 and 96 hr post-infection., clarified and stored at -80uC. HSV-1 titers were determined by a plaque assay in Vero cells. Monolayers were infected with HSV-1 and overlaid with 700 mL essential medium containing 2% FBS in 6-well plates. After virus adsorption, monolayers were overlaid with 2.5 mL essential medium containing 2% FBS and 0.9% agarose. After fixation of the cells with 4% formaldehyde (Solarbio) and removal of the agarose overlay at 4-5 days post-infection, the plaques were visualized by staining with 1% crystal violet (Solarbio) in distilled water. Immunofluorescence Primary cell cultures on coverslips were washed with PBS, fixed in 4% paraformaldehyde and permeabilized with 0.2% Triton X-100. The cells were incubated with a rabbit anti-GFAP antibody (1:8) for 1 hr and sebsequently incubated with a FITC-conjugated goat anti-rabbit antibody (1:250) for 1 hr. The slides were examined using a Nikon E600 fluorescence microscope. SiRNA library screening All 212 monkey phosphatase genes and putative phosphatase genes were chosen from the NCBI. The siRNAs (3 per gene) targeting the phosphatases were designed and purchased from Guangzhou RiboBio Co., Ltd. Three siRNAs for each gene target were cotransfected into monkey astrocytes for screening in 96-well plates. Forty-eight hours after transfection (described above), the cells were infected with HSV-1 (MOI = 0.01). Forty-eight hours after infection, the titer of each well was tested by microtitration assay. Briefly, viral stocks were serially diluted 10 times, and 100 mL of the diluted stocks were added to Vero cells with 80% confluence in each of the 96-well plates. Overall, 8 parallel wells for each dilution stock were used. The 96-well plates were incubated at 37uC in a 5% CO 2 incubator for 5-7 days to observe the cytopathic effects. The 50% cell culture infective dose (CCID50) was determined by Reed-Muench assay. The following protocol was used for these PCR assays: 3 min at 95uC, followed by 40 cycles of 95uC for 10 s, 56uC for 10 s and 68uC for 40 s. All experiments were repeated three times. Electron microscopy analysis The tissue was prepared and fixed in a solution containing 0.5% glutaraldehyde, 4% paraformaldehyde and 0.1 M of sodium phosphate buffer (pH 7.6) on ice for 1 hr. After washes with 4% sucrose in 0.1 M phosphate buffer, the tissue was decalcified with a 2.5% EDTA solution for 5 days. After three washes, the samples were fixed with 2% osmium tetraoxide on ice for 1 hr and dehydrated with a series of ethanol gradients. For transmission electron microscopy, the tissues were embedded in Epon 812 resin mixture, and ultrathin (70-nm) sections were cut with Leica EM UC6 ultramicrotome (Leica Co., Austria). The ultrathin sections were stained with 2% uranyl acetate in 70% ethanol for 5 min at room temperature and then with Reynold's lead for 5 min at room temperature. Sections were analyzed with an H-7650 transmission electron microscope (HITACHI, Japan). Statistical analysis The statistical analysis was performed using SPSS 13.0 software. Data obtained from all experiments performed in triplicate were processed by variance analysis, and P,0.05 was considered to be statistically significant. Replication

of HSV-1 in Macaca mulatta astrocytes The pathological study of HSE caused by HSV-1 suggested that replication in the CNS was one of the major events in HSV-1 pathogenesis [4]. Although studies have confirmed that HSV-1 can cause latent infections in neurons [24], the HSV-1 replication mechanism in neurons is still not clear due to the special biological environment of neurons. Therefore, astrocytes, which are important cell components in the CNS, are one of the major target cells study for a better understanding of HSV-1 in encephalitis pathogenesis. An astrocyte tissue culture system (purity . 90%) was obtained using a differential adsorption method of passaging of astrocytes from a M. mulatta CNS. The purity of the cell culture was verified using fluorescence detection with a specific antibody against the glial fibrillary acidic protein (GFAP), which is expressed in astrocytes [ Fig.S1]. HSV-1 (MOI = 0.01) was used to infect astrocytes in the astrocyte culture system, and virus replication was observed. As the control, human astrocytoma cells U-87MG strain, and HEK293 cells were also used. The results showed that after HSV-1 infection, astrocytes showed obvious cytopathic effects (CPE), similar to the effects of HSV-1 infection of U-87MG and HEK293 cells [ Fig.1a]. In addition, transmission electron microscope (TEM) results confirmed the microscopic cellular structure change and the virus replication in rhesus astrocytes [ Fig.1b]. A subsequent kinetics study involving HSV-1infected astrocytes at different time points post-infection showed that the infection had typical viral infection kinetics [25] [Fig.1c]. The HSV-1 replication kinetics in astrocytes were similar to the replication kinetics the U-87MG and HEK293 cell lines. Astrocytes have multiple regulatory functions in the CNS [17]. We designed the siRNA library for phosphatase systems [26] involved in cellular activity in astrocytes. This library included siRNAs for 212 phosphatase genes [ Table S1] and was used to screen phosphatases involved in HSV-1 infection as well as to further understand the molecular mechanism of the interaction between HSV-1 and astrocytes. Our study first investigated the efficiency of the siRNA transfection in astrocytes. Cyanine 3 (Cy3)labeled siRNA transfection results showed that the transfection method was successful [ Fig.S2]. Six phosphatase genes (ACP6, ALPL, DUSP11, PNKP, PPP1CC and PTPN11) were further analyzed using qRT-PCR, and the results showed that siRNA transfection effectively downregulated corresponding phosphatase RNA expression [Fig. 2a]. The downregulation showed a three-to fivefold times reduction compared with the control RNA expression. Two phosphatase genes (ALPL and PNKP) were used to study the inhibitory kinetics, and the results showed that the inhibition was effective from 24 to 72 hr post-transfection [Fig. 2b,c]. All of these results confirmed the effectiveness of the phosphatase siRNA transfection-inhibition system. The effect of the downregulation of 212 phosphatase genes in astrocytes on HSV-1 replication The siRNA for each phosphatase was used to transfect astrocytes (3 siRNAs per gene, 3 replicate wells per gene). HSV-1 (MOI = 0.01) was used to infect astrocytes at 48 hr post- Figure 3. siRNA knockdown of different phosphatases led to different responses to HSV-1 infections. Three siRNAs per gene were transfected in astrocytes followed by infection with HSV-1 (at an MOI of 0.01) in three independent experiments. Virus solutions were harvested at 48 h post infection, and the virus titer was determined by the CPE method and analyzed by calculating the normalized fold of CCID50/mL. The color scale reflects the fold of reduction or increase level of viral replication. These values (0.003-80) represent ratios of CCID50/mL (target/NC). NC refers to the negative control (value considered as 1). doi:10.1371/journal.pone.0079648.g003 transfection. Viruses were harvested at 48 hr post-infection, and the titer was determined. Non-functional siRNA (NC-siRNA) transfection of astrocytes was used as the negative control. The results showed that with siRNA knockdown of different phosphatases led to different cellular responses to HSV-1 infections [Fig.3]. However, siRNA knockdown of most phosphatases did not affect virus replications significantly; only the knockdown of five phosphatases led to significant decreases (.100x reduction of HSV-1 replication) in virus infection efficiency [ Fig.S3 and Table S2]. The effect of five phosphatases on HSV-1 replication efficiency in astrocytes The study of HSV-1 infection efficiency with the downregulation of 212 phosphatase genes showed that the downregulation of five phosphatases (PNKP, SNAP23, PTPRU, LOC714621 and PPM1M) by siRNA transfection led to significant decreases in the HSV-1 titer [Fig. 4a]. The downregulation of these phosphatases can cause a 100-to 1000-fold reduction in HSV-1 infection. This result suggested the possibility that these five phosphatases were needed for the replication of HSV-1 in astrocytes and that these phosphatases, which are involved in cellular signal transduction, might contribute to the pathogenesis of HSV-1 encephalitis by mediating cytokine upregulation. Although the complete data regarding these phosphatases are not yet available, the fact that these five phosphatases were observed to be required in human astrocytes [ Fig. 4b] but not HEK293 cells [ Fig. 4c] seems supportive of this possibility. The downregulation of these five phosphatases can cause a 5-to 20-fold reduction in HSV-1 infection in U-87MG [ Fig. 4b]. Thease results indicated that the effect of downregulation of these phosphatases on HSV-1 replication in a human astrocytoma cell line appears to be more limited. Additionally, further study of the functions of these five phosphatases suggested that the phosphatases were involved in the phosphorylation regulation of several functional molecules and structural proteins in cells [ Table 1]. Discussion HSV-1, which is a common virus found in human populations [27], has been proven to cause numerous diseases through complicated pathogenesis pathways. Among all of these diseases, HSE has a low occurrence but a poor prognosis [28], as well as important clinical significance. The mechanism by which HSV-1 causes HSE remains unclear. However, a previous study of HSV-1 infection in astrocytes indicated that the infected astrocytes experienced an upregulation of various cytokines which caused instability of neuron-astrocyte communication and subsequently the dysfunction of the CNS [8]. Therefore, analysis of the interaction between HSV-1 and astrocytes, which are important functional cells in the CNS, may provide important evidence for understanding the clinical pathogenesis of HSE. This study investigated the interaction between HSV-1 and cellular phosphatases involved in the regulation of HSV-1 infection using siRNA interference of 212 phosphatases in astrocytes to observe the titer changes in HSV-1. The results showed that the downregulation of five phosphatases had significant inhibitory effects on HSV-1 infection, suggesting that these phosphatases were needed for HSV-1 replication in astrocytes. This observation was extended to human astrocytoma cells, in which the downregulation of these five phosphatases inhibited the proliferation of HSV-1 and implied a role of these five phosphatases in the replication of HSV-1 in astrocytes. Studies have shown that among the five phosphatases (PNKP, SNAP23, PTPRU, LOC714621 and PPM1M ), the PNKP-relevant molecules XRCC4 and LigIV have been demonstrated to play a role in the circularization of HSV-1 linear double-stranded genomes [29]. In addition, PNKP has been shown to be involved in doublestrand break (DSB) repair [30] with its 59-DNA kinase and 39-DNA phosphatase activities [31]. These functions which are directly involved in HSV-1 genome replication indicated that these phosphatases have important roles in HSV-1 replication in astrocytes. The regulatory molecules (SNAP23, PTPRU and LOC714621) involved in different cellular signal transduction pathways showed an inhibitory effect on HSV-1 replication efficiency, suggesting that these signaling pathways can directly or indirectly affect the HSV-1 replication mechanism [32][33][34]. While there is data to suggest that SNAP23 is the components of the TLR signaling pathway [35], our findings suggested the immunological response of astrocytes may influence HSV-1 replication. Clearly, while more exploration is needed to understand the details of the mechanism of this process, this study has provided insight for further studies regarding not only the interaction of HSV-1 and astrocytes but also the pathogenesis of encephalitis induced by this virus. were harvested at 48 h post infection, and the virus titer was determined by the CPE method and analyzed by calculating the lgCCID50/mL. Cut off = 2.7, and it means 100x reduction of HSV-1 replication. NC refers to the negative control. Interleukin-33/ Suppression of Tumorigenicity 2 in Renal Fibrosis: Emerging Roles in Prognosis and Treatment Chronic kidney disease (CKD) is a major public health problem that affects more than 10% of the population worldwide and has a high mortality rate. Therefore, it is necessary to identify novel treatment strategies for CKD. Incidentally, renal fibrosis plays a central role in the progression of CKD to end-stage renal disease (ESRD). The activation of inflammatory pathways leads to the development of renal fibrosis. In fact, interleukin-33 (IL-33), a newly discovered member of the interleukin 1 (IL-1) cytokine family, is a crucial regulator of the inflammatory process. It exerts pro-inflammatory and pro-fibrotic effects via the suppression of tumorigenicity 2 (ST2) receptor, which, in turn, activates other inflammatory pathways. Although the role of this pathway in cardiac, pulmonary, and hepatic fibrotic diseases has been extensively studied, its precise role in renal fibrosis has not yet been completely elucidated. Recent studies have shown that a sustained activation of IL-33/ST2 pathway promotes the development of renal fibrosis. However, with prolonged research in this field, it is expected that the IL-33/ST2 pathway will be used as a diagnostic and prognostic tool for renal diseases. In addition, the IL-33/ST2 pathway seems to be a new target for the future treatment of CKD. Here, we review the mechanisms and potential applications of the IL-33/ST2 pathway in renal fibrosis; such that it can help clinicians and researchers to explore effective treatment options and develop novel medicines for CKD patients. INTRODUCTION Chronic kidney disease (CKD) is a clinical syndrome with definite changes in kidney structure and function that last for more than 3 months (Ammirati, 2020). Incidentally, CKD is the third leading cause of premature death in patients, following acquired immunodeficiency syndrome and diabetes. In fact, more than 10% of the world's population is affected by CKD Djudjaj and Boor, 2019), thereby making it a major public health problem issue. Interestingly, renal fibrosis occurs in severe cases of CKD, and it is considered as an underlying pathological process of this disease; moreover, renal fibrosis ultimately leads to the progression of CKD to end-stage renal disease (ESRD; Berchtold et al., 2017;Chen et al., 2018a;Martínez-Klimova et al., 2019). Therefore, the underlying mechanisms of renal fibrosis should be extensively studied to ensure a better understanding of CKD development because the inhibition of renal fibrosis may help to decrease the rate of CKD progression. Fibrosis is a highly coordinated process that occurs as a result of the tissue repair response. Incidentally, the tissue repair response may become dysregulated if the healing response of tissues continues beyond normal wound healing, for instance during chronic inflammation, trauma, infection, metabolic disorders, inflammation, and autoimmunity. This results in the production of cytokines and chemokines. The release of these mediators leads to the local activation of collagen-producing mesenchymal cells, to a transition of various cell types into myofibroblasts as well as to the recruitment of fibroblast precursors (Weiskirchen et al., 2019;Henderson et al., 2020;. The main pathological characteristics of renal fibrosis are renal interstitial fibroblast hyperplasia and the aberrant and excessive deposition of extracellular matrix (ECM). These changes destroy the normal tubular and interstitial structures of the kidneys (Humphreys, 2018;. Interestingly, inflammatory responses are central to the progression of renal fibrosis (Zhao et al., 2015;Chen et al., 2019). They involve various cytokine-mediated multi-signaling pathways, including transforming growth factor-beta (TGF-β), tumor necrosis factor-alpha (TNF-α), and leukocyte mediators Chen et al., 2018a;Prakoura et al., 2019). Interleukin 33 (IL-33), the gene for which is located on chromosome 9, is a newly discovered member of the IL-1 cytokine family and a pivotal regulator of inflammatory and immune responses (Molofsky et al., 2015;Di Salvo et al., 2018;Chen et al., 2020). It functions in association with the suppression of tumorigenicity 2 (ST2) receptor. In fact, the IL-33/ST2 pathway is involved in causing an imbalance between widespread inflammation and tissue regeneration in several organs, such as the lungs, liver, skin, and gastrointestinal system, ultimately leading to fibrosis (Rankin et al., 2010;Tan et al., 2018;Wu et al., 2018;Imai et al., 2019). Due to the potential role of the IL-33/ST2 pathway in fibrotic diseases, it has received increasing attention. In fact, this review discusses the current progress in research related to the IL-33/ST2 pathway in renal fibrosis to provide new insights into developing treatment options for CKD patients. IL-33 In 1999, IL-33

was first identified in a canine model of arachnoid hemorrhage, which expressed the protein DV27 (Onda et al., 1999). Subsequently, in 2003, endogenous IL-33 and its mRNA were discovered in human tissues, and initially, defined as a nuclear factor from high endothelial venules (NF-HEV; Baekkevold et al., 2003). In 2005, Schmitz et al. (2005 discovered that the hydroxyl-terminal portion of NF-HEV resembles the three-dimensional folded structure of the IL-1 family of cytokines. Moreover, as a cytokine, this factor induces a type 2 immune response by binding to the ST2 receptor. Based on these observations, NF-HEV was classified as a new member of the IL-1 cytokine family, and it was named IL-33 (Schmitz et al., 2005). The IL-33 is a regulator of inflammation, and can induce T-helper 2 (Th2)-mediated innate and adaptive immune responses (Kakkar and Lee, 2008;Kotsiou et al., 2018). Additionally, IL-33 participates in the pathogenesis of renal, neurological, hepatic, pulmonary, and eye diseases. It is a cell cytokine that promotes inflammatory responses and has a characteristic of alarm (Moussion et al., 2008;Li et al., 2019b;Wang et al., 2020). Since the expression of IL-33 has been reported to increase in tissues during inflammation (Kotsiou et al., 2018), IL-33 is a potential mediator of various inflammatory diseases (Dinarello, 2005). Furthermore, IL-33 plays an invaluable role in various biological responses, such as regulation of immune responses, maintenance of tissue homeostasis, and tissue repair and remodeling (Drake and Kita, 2017). IL-33 Receptor The IL-33 receptor complex is a heterodimeric complex consisting of IL-1 receptor accessory protein (IL-1RACP) and ST2 encoded by IL-1receptor type 1(IL1R1; Cayrol and Girard, 2014). In fact, the IL-1RACP functions as a co-receptor for IL-33 signaling . The ST2, a member of the Toll-like receptor/ IL-1 receptor superfamily, has four subtypes, including two variant forms (ST2LV and ST2V), a soluble form (sST2), and a membrane-bound form (ST2L; Bergers et al., 1994;Iwahana et al., 2004;Schmitz et al., 2005;Griesenauer and Paczesny, 2017). The ST2LV is a form of ST2L that is produced by selective splicing (Iwahana et al., 2004). The ST2V is mainly expressed in organs of the digestive system. Its overexpression in cell lines leads to restricted membrane localization (De la Fuente et al., 2015). The most significant subtypes are ST2L and sST2, and they are highly expressed in the kidneys, lungs, placenta, and stomach (Li et al., 2000;Kaur et al., 2020). The ST2L is a functional component of IL-33 signaling, and it promotes inflammation and conducts a Th2-type immune response by activating downstream molecules, such as myeloid differentiation primary response 88 (MyD88) and tumor necrosis factor receptor (TNFR)-associated factor 6 (TRAF6) proteins (Milovanovic et al., 2012;Takeda et al., 2018). The sST2 lacks transmembrane and intracellular domains and can act as a decoy receptor for IL-33 that can help to block IL-33 signaling Antunes et al., 2018). It is expressed in mast cells and fibroblasts and its activity is induced by cytokines such as TNF-α. THE IL-33/ST2 PATHWAY AND RENAL FIBROSIS Distribution of IL-33 and ST2 in the Kidney Incidentally, IL-33 is widely expressed in various organs, including the heart, brain, kidneys, liver, spleen, and lungs. At the cellular level, IL-33 is mainly expressed by non-immune cells, including epithelial cells, endothelial cells, and fibroblasts (Schmitz et al., 2005;Martin and Martin, 2016;Chen et al., 2017;Drake and Kita, 2017). In addition, immune cells such as activated macrophages and dendritic cells are also sources of IL-33, but the expression level of IL-33 is low (Li et al., 2014a;Yang et al., 2016). In kidney, immunohistochemical analysis has revealed the presence of IL-33 + cells in the renal tubules of the patients with chronic allograft dysfunction . Similarly, in patients with renal allograft rejection, IL-33 has been detected in the renal tubules and interstitium (Yu et al., 2021). Furthermore, IL-33 is constitutively expressed in endothelial cells of large and small renal vessels (Akcay et al., 2011). Chen et al. (2016) have also reported the presence of a major population of IL-33 + cells among α-smooth muscle actin (α-SMA) + interstitial myofibroblasts and a minor population among CD31 + peritubular vascular endothelial cells in the kidneys of unilateral ureteral obstruction (UUO) mice. In addition, IL-33 was constitutively expressed throughout the kidney in peritubular and periglomerular spaces of mice kidneys (Ferhat et al., 2018). With respect to the IL-33 receptors, ST2L is selectively expressed by several immune cells involved in type 2 immune responses, such as macrophages, invariant natural killer T (iNKT) cells, mast cells, and lymphocytes (Kaur et al., 2015;Lu et al., 2015;Larsen et al., 2018;Kaur et al., 2020). The sST2 is mainly expressed in fibroblasts and epithelial cells (Akimoto and Takenaga, 2019). In the kidney, ST2 is strongly expressed in the glomeruli, renal tubules, and peritubular capillaries in patients with diffuse cutaneous systemic sclerosis (Manetti et al., 2010). Additionally, Yu et al. (2021) also reported ST2 may be localized to renal tubules and interstitial spaces. Another study reported that ST2 is expressed in macrophages, FIGURE 1 | The interleukin-33 (IL-33)/suppression of tumorigenicity 2 (ST2) pathway. Activation of the IL-33/ST2 pathway: IL-33 binds to a receptor complex composed of ST2L and interleukin 1 receptor accessory protein (IL-1RACP) on the cell membrane. The interaction between the C-terminal domain of this receptor complex promotes the recruitment of myeloid differentiation primary response 88 (MyD88), IL-1 receptor-associated kinase 1 (IRAK1), IRAK4, and tumor necrosis factor receptor (TNFR)-associated factor 6 (TRAF6), thus leading to the activation of nuclear factor kappa light chain enhancer of activated B cells (NF-κB), c-Jun N-terminal kinase (JNK), extracellular signal-regulated kinase (ERK), and P38 pathways. The activation of these pathways facilitates the gene expression of chemokines and growth factors. The inhibition of the IL-33/ST2 pathway: soluble ST2 (sST2) acts as a decoy receptor for IL-33 to block the binding of IL-33 to the receptor. single immunoglobulin IL1-related receptor (SIGIRR) can destroy ST2L/IL1RAcP heterodimers. Frontiers in Physiology | www.frontiersin.org 4 January 2022 | Volume 12 | Article 792897 bipotent T and innate lymphoid cells (T/ILC), neutrophils, and monocytes in kidneys of mice after obstructive injury (Chen et al., 2018b). However, the localization of ST2 to other non-immune cells in murine kidneys has not been completely elucidated, and it requires further investigation. The Role of IL-33/ST2 Pathway in Renal Fibrosis The IL-33/ST2 pathway is widely involved in the fibrosis of various organs, including the airway, lungs, heart, intestine, liver, and skin ( Table 1). As shown in Table 1, the role of this pathway in cardiac fibrosis is controversial. Sanada et al. (2007) observed that the IL-33/ST2 pathway limits overall ventricular fibrosis, whereas Ghali et al. (2020) reported IL-33 aggravated the deterioration of cardiac function and cardiac remodeling. In fact, it is associated with activation of myofibroblasts and an increase in the pro-fibrotic markers, such as connective tissue growth factor (CTGF) and TGF-β (Ghali et al., 2020). Overall, in majority of fibrotic diseases, this pathway mainly promotes inflammatory responses and facilitates organ fibrosis, ultimately leading to organ dysfunction. Diabetes (Abd Rachman Isnadi et al., 2018), obstructive nephropathy (Martínez-Klimova et al., 2020), ischemiareperfusion injury (IRI; Rossi et al., 2021), and lupus nephritis (Davidson, 2016) promote the development of CKD. In fact, there are many publications showing that the IL-33/ST2 pathway exacerbates the structural and functional damage of the kidneys in CKD (Table 2). Specifically, in studies of UUO model that is classical renal fibrosis model (Chevalier et al., 2009;Martínez-Klimova et al., 2019), IL-33 inhibited fibrosis in the early stages (Gatti et al., 2021). However, continuous stimulation resulted in significant upregulation of the IL-33/ST2 pathway (Chen et al., 2016;Li et al., 2019b;Gatti et al., 2021). In a word, the progression of renal disease is associated with the sustained activation of IL-33/ST2. Therefore, the IL-33/ST2 pathway appears to be a significant mechanism underlying renal fibrotic disease. Mechanisms of IL33/ST2 Involvement in Renal Fibrosis During chronic kidney injury and inflammation, endothelial cells, epithelial cells, and fibroblasts release IL-33 (Tonacci et al., 2019). This IL-33 activates ST2-expressing cells, such as polarized M2 macrophages, innate lymphoid type 2cells (ILC2s), CD4 + T cells, and iNKTs, thereby triggering inflammation and tissue repair responses. It has been reported that macrophages play an important role in obstructive kidney injury (Meng et al., 2014). Interestingly, IL-33, along with other cytokines, promotes the transformation of M0 macrophages into M2 macrophages (Liew et al., 2016), and persistence of these M2 macrophages contributes to fibrosis (Mosser and Edwards, 2008;Kim et al., 2015). Moreover, Li et al. (2019b) discovered that during UUO-induced renal fibrosis, IL-33 exacerbates renal injury and promotes renal fibrosis via the activity of M2 macrophages and an increased secretion of IL-13 and TGF-β1. Incidentally, ILC2s are activated in damaged kidneys; for instance increased amounts of ILC2s have been reported in the kidneys of UUO mice models (Chen et al., 2018b). Additionally, ILC2s promote lung and liver fibrosis, and they can secrete Th2-specific cytokines, such as IL-4 and IL-13 (Liu et al., 2019a), which, in turn, exhibit significant pro-fibrotic activity (Fertin et al., 1991;Wynn, 2003). However, one study reported that IL-33 for a short period of time may attenuate the IRI-induced renal damage via ILC2 activity (Riedel et al., 2017;Cao et al., 2018). This may have occurred due to the short duration and low dose of treatment with recombinant IL-33 (rIL-33). Akcay et al. (2011) demonstrated that a sustained high level of rIL-33 can exacerbate renal damage, but they did not specify the role of ILC2s in this process. Therefore, the role of ILC2s, stimulated by sustained high levels of rIL-33, in cases of renal fibrosis requires further investigation. Furthermore, IL-33 can also drive the activation of iNKTs and the infiltration of neutrophils into inflamed tissues (Bourgeois et al., 2011;Thierry et al., 2014;Ferhat et al., 2018). In fact, Ferhat et al. observed that IL-33 promotes IFN-γ/IL-17A production by iNKT cells in IRI, further amplifying the renal tissue injury (Ferhat et al., 2018). The CD4 + T cells are also vital for the progression of renal fibrosis . A previous study has reported that the infiltration of renal CD4 + T cells was reduced in acute kidney injury (AKI) mice model injected with sST2. Moreover, in wild-type mice, rIL-33 exacerbated AKI and raised levels of C-X-C motif ligand 1 (CXCL1), but similar results were not observed in CD4 deficient mice (Akcay et al., 2011). These results suggest that the deleterious effects of IL-33 are probably mediated via CD4 + T cells. In addition to effects of the IL-33/ST2 pathway on the abovementioned immune cells, a few studies also reported its effects on non-immune cells. IL-33 is involved in post-transplant interstitial fibrosis by activating the p38 MAPK signaling pathway and promoting epithelial-mesenchymal transition (EMT) in HK-2 cells . In fact, Chen et al. found elevated expression of IL-33 in mesenchymal myofibroblasts and perivascular endothelial cells of the renal tubules. They also confirmed that IL-33 deletion via gene knockdown reduced UUO-induced renal fibrosis (Chen et al., 2016). These observations suggest that the upregulation of the IL-33/ST2 pathway in case of obstructed kidney disease may promote tubular cell injury and interstitial fibrosis. Another study reported that upregulation of the IL-33/ST2 signaling pathway in case of systemic lupus erythematosus (SLE) promoted renal tubular cell injury and fibrosis primarily through renal EMT (Chen et al., 2017). Moreover, Liang et al. (2017) found that IL-33 worsened renal fibrosis after IRI. On the contrary, sST2 significantly inhibited collagen deposition in the kidneys of IRI-stressed mice by inhibiting IL-33. It is possible that IL-33 recruits bone marrow-derived fibroblasts and inflammatory cells into the kidney, thereby producing proinflammatory factors and profibrotic molecules (Liang et al., 2017). Similarly, Zhu et al. found that IL-33 induced the phenotypic transformation of bone marrow-derived monocytes into fibroblasts in a dosedependent manner, thereby leading to a marked increase in the expression of α-SMA and fibronectin . Generally, the mechanisms of the IL-33/ST2 pathway in the progression of renal fibrosis can be summarized in two aspects (Figure 2). On the one hand, this pathway promotes the secretion of pro-inflammatory and pro-fibrotic factors, such as IL-4, IL-13, and TGF-β by some immune cells. These cytokines promote inflammation. Notably, TGF-β is an important central mediator of renal fibrosis (Meng et al., 2015). Some previous studies found that IL-33 expression was positively correlated with TGF-β (Li et al., 2014a;Kotsiou et al., 2018;Elsherbiny et al., 2020). This may be related to macrophages (Li et al., 2019b) or ILC2 (Nagashima and

Iyoda, 2021). The specific link and mechanism between IL-33/ST2 and TFG-β need to be investigated further. On the other hand, the IL-33/ ST2 pathway can promote fibrogenesis by triggering EMT of renal tubular epithelial cells and activating myeloid fibroblasts. Ultimately, these changes promote the secretion of collagen and fibronectin to facilitate fibrogenesis. However, the effect of this pathway on other non-immune cells of the kidney, such as collecting duct cells and podocytes, is unknown and requires further research. THE POTENTIAL VALUE OF THE IL-33/ ST2 PATHWAY FOR RENAL FIBROSIS The IL-33/ST2 Pathway: A Promising Biological Marker of Kidney Disease Patients with kidney disease lack clinical symptoms and have low sensitivity to serum creatinine (Cr) tests in the early stages. Moreover, a decreased glomerular filtration rate is usually detected in the moderate or severe stages of CKD; this means that patients miss the optimal period for treatment. Early detection and correction of modifiable risk factors can decrease the rate of decline in kidney function (Ruggenenti et al., 1999;McClellan and Flanders, 2003;de Zeeuw et al., 2006;Ho et al., 2013). Therefore, researchers aim to identify novel biological markers in patients with kidney disease that will allow early detection and treatment of the condition. Currently, IL-33 and ST2 are used as biological markers mainly in patients with heart disease. In fact, sST2 has been shown to independently predict heart failure (HF), major cardiovascular events, and mortality of the patients (Seo et al., 2018;Alam et al., 2019;Homsak and Gruson, 2020). Incidentally, levels of cardiac ST2 mRNA and serum ST2 protein are known to increase in patients with myocardial infarction (Weinberg et al., 2002). Additionally, IL-33 has been associated with vascular disease dysfunction (Gungor et al., 2017). It has been reported that IL-33 effectively activates the type 2 cytokine milieu in the damaged heart and is associated with the deterioration of cardiac function as well as cardiac remodeling (Ghali et al., 2020). Furthermore, serum IL-33 and sST2 levels were elevated in patients with chronic HF , and they continue to increase with the degradation of cardiac function (Xiang et al., 2021). What is more, the expression of IL-33 and ST2 in myocardial tissue from patients with end-stage HF was reported to be elevated. In addition, their levels were significantly correlated with the degree of cardiac fibrosis and the expression level of TGFβ1 (Tseng et al., 2018). It is evident that the IL-33/ST2 pathway plays a regulatory role in the progression of cardiac fibrosis, thereby making it a useful tool for the prognosis and diagnosis of heart failure. It is well-known that the heart and kidneys are closely linked organs under physiological and pathophysiological conditions (Palazzuoli et al., 2015). Acute or chronic heart disease can directly lead to the deterioration of acute or chronic renal function and vice versa. This phenomenon referred to as the cardiorenal syndrome. Incidentally, the coexistence of cardiac and renal dysfunction can significantly increase the mortality and morbidity of the patients, as well as cost of patient care (Ronco et al., 2010;Cruz et al., 2011;Palazzuoli et al., 2015). HF is the main cardiovascular complication and the leading cause of mortality in patients with renal disease, especially the ones with ESRD (Bansal et al., 2017). It is clear that kidney disease and heart disease are closely related. Therefore, given that serum IL-33 and sST2 levels are known to reflect heart function, some researchers have investigated whether they might also reflect kidney function. One study reported that circulating levels of sST2 are independently and negatively associated with a poor diuretic response in patients with acute HF and renal dysfunction . Another study revealed that serum IL-33 is a sensitive marker of renal function in patients with gout . Gungor et al. (2017) reported that IL-33 and ST2 levels increase with the gradual progression of the CKD stages. In fact, in a cohort study of critically ill patients, the ones with elevated sST2 expression had the worst prognosis (Dieplinger et al., 2016). However, some studies indicate that sST2 is a better indicator of risk for end-stage dialysis than IL-33 in patients with kidney disease (Bao et al., 2012;de Boer et al., 2015;Tuegel et al., 2018;Lukic et al., 2020;Yan et al., 2020). For instance, Bao et al. (2012) did not observe any difference in the serum IL-33 concentrations of patients with CKD and that of healthy individuals. However, sST2 serum level was elevated in patients. Furthermore, a significant correlation between the sST2 levels and the disease severity was found by correlation analysis (Bao et al., 2012). Similarly, Lukic et al. (2020) reported that sST2, but not IL-33, is associated with ESRD parameters, and the serum sST2 concentration is positively correlated with that of serum urea and creatinine. Hence, it is unclear whether IL-33 can be used as a reference indicator in patients with kidney disease. However, sST2 appears to be a promising biological marker for kidney diseases, and deserves further study and wide application in clinical practice. The IL-33/ST2 Pathway as a New Therapeutic Target for CKD Patients with ESRD require dialysis or kidney transplantation (Hewitson, 2009). However, this treatment modality is limited in the developing and underdeveloped countries due to the high costs (Nogueira et al., 2017). At present, glucocorticoids are the mainstay of treatment for kidney disease. However, their widespread clinical use is largely limited because of their association with a variety of serious adverse effects, such as infections, hypertension, and osteoporosis (Ponticelli and Locatelli, 2018;Vandewalle et al., 2018). Therefore, it is necessary to identify new therapeutic targets and medicines to prevent disease progression and ultimately death. From the above discussion, it is evident that the IL-33/ST2 pathway plays a key role in renal fibrosis. In particular, in vivo studies in IL-33-deficient transgenic mice further supported the importance of the IL-33/ST2 pathway in the progression of renal fibrosis. For instance, the damage to the proximal tubules was reduced in the kidneys of the IL33−/− UUO model mice (Chen et al., 2016). Similarly, compared with 2 | Advancements in the study of IL-33/ST2 in chronic kidney disease (CKD). Elsherbiny et al., 2020 Protein↑ IL-33 level is elevated in DN rats with contrast-induced nephropathy. Inhibition of IL-33 provided functional and histological protection. Onk et al., 2016 Non-diabetic nephropathy Chronic obstructive nephropathy mRNA↑ mRNA↑ Elevated IL-33 is devastating renal injury and limit proliferation of tubular epithelial cells. Chen et al., 2016 mRNA↑ Nuclear IL-33 in fibroblasts inhibits the initial pro-fibrotic response, but continued stimulation by UUO and secretion of IL-33 exerts a pro-fibrotic effect via activated fibroblasts. Gatti et al., 2021 mRNA↑ mRNA↑ IL-33 promotes renal fibrosis through macrophages and increases secretion of IL-13 and TGF-β1. Li et al., 2019b Acute kidney injury (AKI) Endogenous IL-33 contributes to kidney IRI by promoting iNKT cell recruitment and cytokine production. Chronic allograft injury after kidney transplantation Protein↑ IL-33 may contribute to the development of kidney interstitial fibrosis via the p38 MAPK signaling pathway. Xu et al., 2018 Lupus nephritis Protein↑ (serum) IL-33 blockade may have a therapeutic effect on SLE by inhibiting the production of inflammatory cytokines, such as IL-1β, IL-6, and IL-17. Li et al., 2014b wild-type mice, IRI mice lacking IL-33 showed reduced levels of tubular cell injury (Ferhat et al., 2018). Moreover, the SLE mice treated with anti-IL-33Ab, which has a protective effect on SLE, showed lower serum and renal levels of IL-1β, IL-6, and IL-17 than that in the untreated mice (Li et al., 2014b). Therefore, the IL-33/ST2 pathway may form a new therapeutic target for CKD. In this regard, IL-33 blockers may be a novel treatment modality for CKD patients; for instance, the monoclonal antibody blockade of IL-33 is a novel immunotherapy agent. Incidentally, AstraZeneca is conducting the trial of a monoclonal antibody called MEDI3506 (anti-IL-33) in patients with diabetic nephropathy. However, this trial has not yet been completed (Jiang et al., 2021). It is known that the main side effect of monoclonal antibody administration is the risk of an adverse immune response (Hansel et al., 2010;Liu and Li, 2014). Therefore, this method requires extensive research. In addition, IL-2 and IL-33, especially as a hybrid cytokine (IL233-bearing IL-2 and IL-33 activities in one molecule), potentiate Tregs and ILC2s to prevent renal injury (Sabapathy et al., 2019). However, this method may be suitable only on a short-term basis as IL-33 becomes a deleterious factor with prolonged stimulation (Akcay et al., 2011). Traditional Chinese Medicine (TCM) and other natural plant compounds offer certain therapeutic advantages over standard medicine, such as potentially lower prices and ease of access. Therefore, these form a good choice for the development of new medicines. Incidentally, the use of TCM has been extensively studied in cases of renal fibrosis (Table 3). However, there is limited information about TCM targeting the IL-33/ST2 pathway in the treatment of kidney diseases. Interestingly, Calycosin, an isoflavone, is the predominant component of Radix Astragali, which has been widely administered to ameliorate the symptoms of diabetes and diabetic nephropathy . Calycosin treatment significantly reduced mRNA expression and protein levels of IL-33 and ST2 when compared to diabetic rats. Thereby, it affected the downstream pathway, p65 NF-κB. Finally, it delayed progression of renal fibrotic events (Elsherbiny et al., 2020). In addition, astragaloside IV (AS-IV; Bao et al., 2016), Cnidium Cnidii (Yang et al., 2020), and salidroside (Cai et al., 2020) have also been shown to reduce IL-33 to improve the disease condition. However, these studies were conducted in non-renal diseases, such as allergic inflammation. AS-IV, Cnidium cnidii, and salidroside have therapeutic effects on renal fibrosis (see Table 3). Thus, the role of these TCMs on IL-33/ST2 signaling in the kidney remains to be investigated. Other natural plant compounds can also attenuate IL-33 expression. Chrysin (CR) is one of the flavonoids that are commonly used as a traditional medicine and found in many plant extracts. CR has an ameliorative effect on CKD (Ali et al., 2015). Moreover, CR reduced PbAc-induced renal inflammation and significantly reduced renal IL-33 levels (Kucukler et al., 2021). Zingerone (ZO), a component of dry ginger root, has several pharmacological activities owing to its antioxidant, anti-inflammatory and anti-apoptotic properties. Kandemir et al. (2018) demonstrated that ZO remarkably reduced the levels of IL-33 in VCMinduced nephrotoxicity. FIGURE 2 | Mechanisms of IL-33/ST2 involvement in renal fibrosis. When the kidney is injured, endothelial cells, epithelial cells, and fibroblasts release IL-33. IL-33 binds to ST2L on the cell membrane and activates downstream pathways. For immune cells, such as M2 macrophages, ILC2s, iNKTs, CD4 + T cells, and neutrophils, IL-33 activates these cells and promotes the secretion of some cytokines that further aggravate inflammation. For other non-immune cells in the kidney, IL-33 can lead to EMT of renal tubular epithelial cells and activation of myofibroblasts. Eventually, COL1, α-SMA etc. are produced. Continued inflammation, EMT, and activation of myofibroblasts finally lead to renal fibrosis. In conclusion, blocking the IL-33/ST2 pathway may be a novel therapeutic strategy for treating renal diseases in the future. Calycosin, CR, and ZO can attenuate IL-33 to protect renal function, thereby making them potential candidates for developing new medicines. Other effective natural compounds targeting IL-33/ ST2 still need to be discovered for the treatment of kidney diseases. CONCLUSION Kidney disease is a major public health problem worldwide. Although kidney transplantation and dialysis treatments are available, there is a significant lack of effective treatments. Therefore, in recent years, researchers have focused on the identification of new treatment targets for patients suffering from kidney diseases. Gradually, the role of the IL-33/ST2 pathway in fibrotic diseases has been established. Although its effects have been extensively studied in cardiac, pulmonary, and liver fibrosis (Willems et al., 2012;Sun et al., 2017;Drake and Prakash, 2020), research related to renal fibrosis still in the nascent stages. In this review, we have focused on discussing the role of the IL-33/ST2 pathway in the development and progression of renal fibrosis as well as the underlying mechanisms of this process. Incidentally, IL-33 activates the immune cells and promotes the secretion of certain cytokines, further aggravating inflammation. However, in other non-immune cells of the kidney, IL-33 can lead to EMT of renal tubular epithelial cells and activation of myofibroblasts. Additionally, many clinical studies have revealed that the IL-33/ST2 pathway may be an effective biomarker of kidney disease, and this is important for the early diagnosis of CKD and assessment of patient prognosis. However, further detailed and comprehensive studies

are required for addressing the existing gaps in knowledge. Meanwhile, the IL-33/ST2 pathway appears to be a promising therapeutic target for renal fibrosis. TGF-β, a strong pro-fibrotic factor, and its relevant pathway have been extensively studied in cases of renal fibrosis (Loboda et al., 2016). However, IL-33 is a newly discovered pro-fibrotic factor. Hence, information related to the IL-33/ST2 pathway is still limited, and medicine targeting the IL-33/ST2 signaling is yet to be developed. We hope this review will contribute to the comprehensive studies of the IL-33/ST2 pathway in renal fibrosis and help in the development of novel therapeutic agents. AUTHOR CONTRIBUTIONS X-YT designed the ideas and wrote the first version of the manuscript. H-YJ contributed to resources. X-YT and H-YJ have contributed equally to this work. Y-RM reviewed the draft of the manuscript. All authors contributed to the article and approved the submitted version. FUNDING The study was sustained by National Natural Science Foundation of China (81973732). Tumor necrosis factor receptor (TNFR)-associated factor 6 UUO Unilateral ureteral obstruction ZO Zingerone Decreased gastrointestinal toxicity associated with a novel capecitabine schedule (7 days on and 7 days off): a systematic review Capecitabine is widely used in the management of metastatic breast cancer; however, drug delivery is limited by gastrointestinal and other toxicity. We employed mathematical modeling to rationally design an optimized dose and schedule for capecitabine of 2,000 mg twice daily, flat dosing, 7 days on, 7 days off. Preclinical data suggested increased efficacy and tolerability with this novel dosing, and three early-phase clinical trials have suggested a favorable toxicity profile. To further define the tolerability of this regimen, we conducted a systematic review of the gastrointestinal adverse events of patients on these studies. This review demonstrated a favorable gastrointestinal toxicity profile with capecitabine in this novel schedule when given as single agent or in combination therapy with either bevacizumab or lapatinib. No patients discontinued therapy for gastrointestinal toxicity, and there were no grade 4 or 5 gastrointestinal toxicities reported. Grade 3 or greater diarrhea occurred in two (2%); grade 2 or greater mucositis, constipation, and vomiting were reported in three (4%) patients. We conclude that capecitabine administered on a 7 days on, 7 days off schedule has limited gastrointestinal toxicity. Our methodology was based on an analysis of individual patient toxicity data from one phase I single-agent capecitabine and two phase II capecitabine combination studies (with bevacizumab and lapatinib, respectively), focusing specifically on gastrointestinal toxicity. INTRODUCTION In spite of significant advances over the last decades in the management of metastatic breast cancer, cure remains an elusive goal. Quality of life on treatment is of major importance and underscores the need to minimize toxicity while preserving or, ideally, improving the efficacy of drugs with known anticancer activity. Models that explore optimum drug delivery are a critical component of rational clinical trial design. 1 Capecitabine is an orally administered fluoropyrimidine that is absorbed intact through the intestinal wall and converted via three enzymatic reactions to 5-fluorouracil. 2 The rate-limiting step in this process is the final reaction, catalyzed by thymidine phosphorylase. 2 In many cancers, including breast and gastrointestinal, thymidine phosphorylase is found in higher levels in tumor than in surrounding normal tissue, allowing relatively selective tumor activity. 2 Several studies have demonstrated the activity of capecitabine in patients with metastatic breast cancer, [3][4][5] and it is widely used in the management of this and gastrointestinal malignancies. [6][7][8] The United States Food and Drug Administration (FDA) approved capecitabine at 1,250 mg/m 2 twice daily, with 14 days continuous dosing followed by a 7-day break . However, toxicity is a significant issue at this dose and schedule, with the most frequent adverse events being palmar-plantar erythrodysesthesia (PPE), diarrhea, and stomatitis. Diarrhea occurs in 28-54% of patients receiving capecitabine, with grade 3/4 diarrhea experienced by 7-19% of patients. 9 Dose reductions are required in 41-65% of patients receiving capecitabine, 9 with a median time to dose reduction of o 2 months. 10 Therefore, in clinical practice an empiric schedule of 1,000 mg/m 2 twice daily, 14 days on, 7 days off, is frequently employed to improve tolerability. 9,10 Toxicity is still a significant issue; dose modification is frequently required, and the efficacy of these alternative schedules has not been prospectively determined. 9,11 Of note, a phase II prospective study failed to demonstrate non-inferiority for an empirically selected lower dose capecitabine regimen. 12 Norton-Simon modeling can accurately predict future behavior of tumor systems on the basis of early growth measurements. 13 In addition, it has determined that the rate of tumor regression is proportional to the rate of tumor growth. 1,14 This model suggests that chemotherapy delivery at increased dose density can increase cancer cell kill by minimizing the opportunity for regrowth between cycles of therapy. 1 This concept has been borne out in the improved outcomes demonstrated with 'dose dense' adjuvant chemotherapy in breast cancer. 15 Our group applied Norton-Simon modeling to capecitabine dosing and observed that the maximum effect of capecitabine is reached at~8 days. 16 Therefore, further dosing through day 14 contributes to toxicity without enhancing efficacy. 16 This provided a rationale for an alternative, optimized dosing schedule that has been rationally rather than empirically designed. Preclinical xenograft models testing a 7 day on, 7 day off (7-7) schedule allowed for higher dosing and increased efficacy compared with conventional (14-7) scheduling. 16 A phase I study of single-agent capecitabine established that the maximumtolerated dose of capecitabine in this novel schedule is 2,000 mg flat dose twice daily. 17 Two phase II studies confirmed the feasibility of this capecitabine schedule in combination with the biologic agents lapatinib and bevacizumab for patients with metastatic HER2-positive and HER2-negative breast cancer, respectively. 18,19 Despite the potentially higher daily dose administered with this regimen, capecitabine was well tolerated in the 7-7 schedule. In particular, we observed improved gastrointestinal tolerability with 4grade 2 diarrhea of just 5-26% reported on the individual studies of the 7-7 schedule. [17][18][19] To further define this tolerability, we conducted a systematic review of the gastrointestinal toxicity observed in patients on these early-phase studies that explored the capecitabine (7-7) dosing schedule. Patient population This analysis includes 81 patients evaluable for toxicity; 18 received single-agent capecitabine, 40 received capecitabine and bevacizumab, and 23 received capecitabine and lapatinib. The median age of patients was 52 years (range, 29-73 years) and the median ECOG performance status was 0. Sixty-two patients (76%) had visceral tumor involvement. The median number of prior chemotherapies for metastatic breast cancer was 0 (0-2). Therapy delivery The median number of treatment cycles delivered was 6 (range, 1-40). Seventeen patients (21%) received ⩾ 12 months of therapy. No patients discontinued therapy for gastrointestinal toxicity. Treatment discontinuation was due to progressive disease in 54 patients (67%), toxicity in 15 (19%), withdrawal of consent in 4 (5%), and other reasons in 8 (10%). Of the patients who discontinued therapy for toxicity, one was receiving single-agent capecitabine, 10 were receiving capecitabine with bevacizumab, and 4 were receiving capecitabine with lapatinib. Treatment discontinuations for toxicity deemed possibly related to capecitabine included palmar-plantar erythrodysesthesia in six patients and laboratory abnormalities in four. Other reasons for patients to discontinue therapy for toxicity included atrial fibrillation, pulmonary embolus, and headache attributed to bevacizumab; rash attributed to lapatinib; and gout. Dose reduction or delay due to gastrointestinal toxicity occurred in five (6%) and four (5%) patients, respectively. Gastrointestinal toxicities These regimens were well tolerated with minimal gastrointestinal toxicity. Table 1 lists the treatment-related gastrointestinal toxicities of all grades. There were no grade 4 or 5 gastrointestinal toxicities reported. Grade 3 or greater diarrhea occurred in two (2%) patients; grade 2 or greater mucositis, constipation, and vomiting were reported in three (4%) patients. DISCUSSION Capecitabine is widely used in the management of metastatic breast and gastrointestinal cancers. Given the palliative goal of therapy for advanced disease in these malignancies, therapeutic tolerance of active agents is of critical importance. Unfortunately, gastrointestinal toxicity is a limiting issue with this drug at the conventional dose and schedule. 9 Despite the widespread use of this drug, the optimal capecitabine dose for patients with metastatic breast cancer has not been defined. 20 A number of empirically designed capecitabine dosing schedules have been employed in an effort to increase tolerance and facilitate continued drug delivery. 9,12 However, the efficacy of these regimens has not been prospectively demonstrated, 9,11 and it is not clear that this empiric dosing has validity. In fact, as noted above, a randomized phase II study failed to demonstrate the non-inferiority of lower dose capecitabine in combination with docetaxel. 12 In addition, a two-stage study exploring the feasibility of fixed-dose single-agent capecitabine 3,000 mg per day, in divided doses, failed to meet its first-stage response end point and was closed. 21 We employed Norton-Simon mathematical modeling to rationally design a novel and optimal dose and schedule of capecitabine. 16 Modeling suggested an optimized regimen of capecitabine twice daily 7-7. In preclinical models, this improved survival over conventional scheduling. Early-phase clinical studies have established the recommended dose and have demonstrated the feasibility of this approach. [16][17][18][19] Data pertaining to the efficacy of this approach is outstanding and is the subject of an ongoing randomized phase III study led by the Latin American and Caribbean Society of Medical Oncology (SLACOM). This trial is comparing conventional capecitabine scheduling with the 7-7 schedule (NCT02028494) and is powered to detect a progression-free survival benefit with the novel regimen. In this systematic review, we present individual patient gastrointestinal toxicity from one phase I study of single-agent capecitabine and two phase II studies combined with the biologic agents bevacizumab and lapatinib. [17][18][19] These data represent a heterogenous patient population receiving these three therapeutic regimens, each with the novel capecitabine backbone schedule as described. There were no grade 4 gastrointestinal toxicities observed with rare occurrences of grade 3 diarrhea and constipation. In addition, the rates of grade ⩾ 2 gastrointestinal toxicity were minimal, with only diarrhea reaching above 5% incidence. This appears to compare favorably with historical data from previous randomized studies of capecitabine with bevacizumab and lapatinib 22,23 (Table 2). Miller et al. noted ⩾ grade 2 diarrhea in 26% and 28% of patients receiving capecitabine alone and capecitabine with bevacizumab, respectively. 22 Similarly, Geyer et al. reported ⩾ grade 2 diarrhea in 25% and 33% of patients receiving capecitabine versus capecitabine and lapatinib, respectively. 23 In that study, 140 (65%) of patients in the capecitabine group and 181 (79%) of patients in the combination group required a dose reduction for toxicity; however, the toxicities necessitating these reductions were not specified. Notably, no patients in the current systematic review came off study for gastrointestinal toxicity, and 17 patients (21%) were on this cytotoxic therapy for over a year. Hand-foot syndrome, 9 or PPE, also significantly limits capecitabine delivery. Exploration of this symptom was not the primary end point of this review; however, given the significance of this toxicity clinically, it has also been highlighted. Six of the patients who discontinued therapy for toxicity that was possibly related to While the efficacy of this approach has to be determined, the improved gastrointestinal tolerability of this novel regimen does not appear to be at the cost of disease control. The median progression-free survival of the phase I study was not reported; however, in the phase II studies it was 8.0 and 9.4 months when combined with bevacizumab and lapatinib, respectively. 18,19 Although it is not possible to make cross-trial comparisons, this is reasonable in the context of previous data. The median progression-free survival of the phase III studies of capecitabine with bevacizumab and lapatinib was 4.87 months 22 and 8.4 months, 23 respectively. Acknowledging the limitations of this retrospective systematic review of prospectively collected data, our study provides an encouraging signal about the tolerability of this novel dose and schedule. The favorable gastrointestinal toxicity profile demonstrated may facilitate prolonged drug delivery with reasonable quality of life. We propose that this novel dose and schedule of capecitabine optimizes the therapeutic index. As noted, the efficacy of this novel approach is the subject of an ongoing phase III clinical trial. The potential for this dose and schedule to decrease toxicity and increase drug compliance has wide-reaching public health implications given the broad use of this drug worldwide in breast and gastrointestinal cancers. In addition, it

has the potential to reduce drug cost: in this novel schedule, patients receive capecitabine for 14 days out of every 28, as opposed to 14 out of every 21 in conventional scheduling. Mathematical modeling employed in the design of this regimen allows for dose and schedule determination based on efficacy by defining the point of maximum tumor perturbation. 16 This moves away from the traditional drug development focus of maximumtolerated toxicity rather than therapeutic index. The proposed novel capecitabine regimen also moves away from conventional dosing based on body surface area, a widely used practice in drug development that has not been well validated. 24 This regimen has significant implications for future drug development and will potentially provide proof of principal that rational dose and schedule design, determined by mathematical models, may efficiently optimize activity and reduce toxicity. We have demonstrated that this novel dose and schedule of capecitabine causes minimal and tolerable gastrointestinal toxicity. We eagerly await the outcome of the ongoing phase III study that will determine the relative efficacy of this approach. MATERIALS AND METHODS In reaching our conclusions from this review, our methodology, based on an analysis of individual patient toxicity data from one phase I single-agent capecitabine and two phase II capecitabine combination studies (with bevacizumab and lapatinib, respectively), focused specifically on gastrointestinal toxicity. This analysis combines the primary toxicity data from three prospective single-institution studies exploring the feasibility of a novel capecitabine dosing schedule as previously described. [17][18][19] These studies included patients with metastatic breast cancer and measurable disease per RECIST criteria. Patients with ECOG performance status ⩽ 2 and adequate hematologic, hepatic, and renal function were eligible. Each treatment cycle was 28 days, and disease evaluation occurred every 12 weeks with scans assessed by RECIST. For selected eligibility criteria and the design of each study, see the study characteristics table (Table 3). All studies were approved by the Memorial Sloan Kettering Cancer Center Institutional Review Board, and patients were provided written informed consent. Individual patient toxicity data from these studies were prospectively collected and graded using NCI CTCAE version 3 terminology, and Table 2. Toxicity of two phase III studies of capecitabine combined with biologic agents bevacizumab and lapatinib, respectively (14) 22 (14) 17 (11) -44 (27) 33 (20) 19 (12) 2 (1) Mucositis/stomatitis (14) 11 (7) 3 (2) -30 (18) 10 (6) 3 (2) -Abbreviation: NR, not reported. attribution assigned. These data were retrospectively extracted by two independent reviewers for the purposes of this systematic review. The primary end point of interest for this review is the incidence and severity of individual gastrointestinal toxicities (nausea, diarrhea, mucositis, constipation, vomiting) with this novel capecitabine schedule. The maximum grade per patient is used as a summary measure. Given the potential for PPE to limit therapy with capecitabine, the incidence and severity of this toxicity has also been highlighted. ACKNOWLEDGMENTS We thank the patients, and their families, who participated in the original studies included in this systematic review. This study was funded in part through NIH/NCI Cancer Center Support Grant P30 CA008748. We previously designed an optimized dose and schedule for capecitabine of 2,000 mg twice daily, flat dosing, 7 days on, 7 days off. Our systematic review of three early-phase studies utilizing this novel dose and schedule has demonstrated minimal and tolerable gastrointestinal toxicity. Characterization of the multigene family TaHKT 2;1 in bread wheat and the role of gene members in plant Na+ and K+ status Background A member of the TaHKT2;1 multigene family was previously identified as a Na+ transporter with a possible role in root Na+ uptake. In the present study, the existing full-length cDNA of this member was used as a basis to query the International Wheat Genome Survey Sequence to identify all members of the TaHKT2;1 family. Individual TaHKT2;1 genes were subsequently studied for gene and predicted protein structures, promoter variability, tissue expression and their role in Na+ and K+ status of wheat. Results Six TaHKT2;1 genes were characterized which included four functional genes (TaHKT2;1 7AL-1, TaHKT2;1 7BL-1, TaHKT2;1 7BL-2 and TaHKT2;1 7DL-1) and two pseudogenes (TaHKT2;1 7AL-2 and TaHKT2;1 7AL-3), on chromosomes 7A, 7B and 7D of hexaploid wheat. Variability in protein domains for cation specificity and in cis-regulatory elements for salt response in gene promoters, were identified amongst the functional TaHKT2;1 members. The functional genes were expressed under low and high NaCl conditions in roots and leaf sheaths, but were down regulated in leaf blades. Alternative splicing events were evident in TaHKT2;1 7AL-1. Aneuploid lines null for each functional gene were grown in high NaCl nutrient solution culture to identify potential role of each TaHKT2;1 member. Aneuploid lines null for TaHKT2;1 7AL-1, TaHKT2;1 7BL-1 and TaHKT2;1 7BL-2 showed no difference in Na+ concentration between Chinese Spring except for higher Na+ in sheaths. The same aneuploid lines had lower K+ in roots, sheath and youngest fully expanded leaf but only under high (200 mM) NaCl in the external solution. There was no difference in Na+ or K+ concentration for any treatment between aneuploid line null for the TaHKT2;1 7DL-1 gene and Chinese Spring. Conclusions TaHKT2;1 is a complex family consisting of pseudogenes and functional members. TaHKT2;1 genes do not have an apparent role in controlling root Na+ uptake in bread wheat seedlings under experimental conditions in this study, contrary to existing hypotheses. However, TaHKT2;1 genes or, indeed other genes in the same chromosome region on 7AL, are candidates that may control Na+ transport from root to sheath and regulate K+ levels in different plant tissues. Background Saline soils are a challenge to cereal production in many regions of the world. Osmotic and ion-specific damage that significantly reduces crop growth in saline soils is largely due to excess of Na + and Cl − ions [1,2]. In bread wheat, a balance between restricting Na + net uptake at roots and subsequent transport into shoots and subcellular compartmentation to cope with unfavourable levels of Na + is needed to maintain ion homeostasis for plant growth and development under saline conditions [3,4]. The regulation of Na + transport and its genetic basis is, therefore, of significant interest in order to assist in development of targeted strategies for enhanced salinity tolerance in crops. A major portion of Na + exclusion (>98%) in bread wheat is accomplished by restricting net Na + uptake at the soil-root interface and net xylem loading in roots [2], yet a number of channels and transporters facilitate Na + influx when Na + concentration is high in the soil solution [5][6][7][8]. Channel and transporter proteins are potential targets to reduce Na + net uptake by roots and have been a major focus to investigate their role in crop tolerance to salinity. Electrophysiological observations and pharmacology of 22 Na + tracer influxes in cereal roots support the hypothesis that the bulk of Na + influx is mediated by voltage-independent, non-selective cation channels (VICs/NSCCs) [9][10][11]. In addition, two transporters LCT1 [12,13] and HKT2;1 [14] can mediate high and/or low-affinity Na + transport in wheat. While both VICs/NSCCs [15] and LCT1 [13] facilitate Ca 2+ dependant unidirectional influx of various cations, TaHKT2;1 mediates Ca 2+ independent influx of Na + and K + [16,17]. Although, TaHKT2;1 is a Na + /K + co-transporter, the protein has a minor significance in K + nutrition as Na +coupled K + uptake has limited physiological relevance in terrestrial species [18]. Therefore, among the genes encoding candidate proteins involved in root Na + influx, TaHKT2;1 is of particular interest in crop breeding because of its high specificity for Na + and the potential to manipulate TaHKT2;1 genes without largely interfering with the homeostasis of nutritionally important ions (e.g., K + , Ca 2+ ) in plants. TaHKT2;1 has a Gly-Gly-Gly-Gly type K + /Na + selectivity pore [19,20] and when expressed in Xenopus oocytes and yeast heterologous systems, preferentially mediates high affinity K + uptake when external Na + is at sub-millimolar levels, but facilitates low-affinity Na + influx when millimolar levels of external Na + are in excess of K + [16,21,22]. The ion transport properties of TaHKT2;1 were determined when point mutations within the TaHKT2;1 gene improved K + specificity and reduced Na + influx in Xenopus oocytes and yeast [21,23]. Down regulation of TaHKT2;1 in transgenic bread wheat by expressing anti-sense or truncated sense constructs reduced Na + levels in K + starved seedling roots [24]. Since the gene is expressed in the cortical cells of roots, alongside in xylem parenchyma in leaves, it is possible that the protein encoded by TaHKT2;1 is associated with root Na + uptake and translocation [24]. Therefore, it appears that under saline conditions TaHKT2;1 is responsible for substantial Na + influx into roots of bread wheat and transport to other parts of the plant. TaHKT2;1 has been identified as a multi-gene family consisting of five individual members distributed across homeologous chromosome group 7 in bread wheat [25]. Although previous studies identified the function of Na + transport in one member of the TaHKT2;1 gene family in heterologous systems [21,23] and in transgenic plants by down-regulation of one or more genes [24], the contribution of each member of the gene family and their interactions to control ion accumulation in wheat remains unknown. This study aimed to understand the organization and variability of each member of the TaHKT2;1 gene family, transcriptional regulation, their encoded proteins, and potential contributions to controlling Na + and K + status in bread wheat. The study used an existing full length cDNA (FL-cDNA) sequence (U16709) as a basis to select and characterize all members of the TaHKT2;1 gene family using the International Wheat Genome Survey Sequence (IWGSS) (www.wheatgenome.org). Results Organization and structure of TaHKT2;1 multi-gene family The FL-cDNA of TaHKT2;1, U16709, was used as query sequence in BLASTN analysis of IWGSS to identify members of the TaHKT2;1 gene family in hexaploid wheat. Six significant hits having greater than 90% nucleotide identity with the FL-cDNA were retrieved that identified three TaHKT2;1 genes on the long arm of chromosome 7A (TaHKT2;1 7AL-1, TaHKT2;1 7AL-2, TaHKT2;1 7AL-3), two genes on the long arm of 7B (TaHKT2;1 7BL-1, TaHKT2;1 7BL-2), and one on the long arm of 7D (TaHKT2;1 7DL-1). The six genes were further studied for gene structure and sequence variability. Gene structure was analysed by splice junction prediction software, and subsequently validated in DNA sequence alignments with the FL-cDNA, U16709, revealed three exons interrupted by two introns for each gene ( Figure 1) with intron splice junction sites having the conserved motif, 5′ GT-AG 3′. TaHKT2;1 7AL-2 and TaHKT2;1 7AL-3 were identical in their introns and exons and each had a single nucleotide deletion at +191 bp in exon 1 relative to the other genes on chromosome 7A, 7B and 7D whereby the translated sequences predicted an in-frame stop codon causing pre-mature termination and a significantly truncated protein of 65 amino acid residues. Therefore, it appeared that TaHKT2;1 7AL-2 and TaHKT2;1 7AL-3 were duplicated pseudogenes. Accordingly, TaHKT2;1 represented a multi-gene family of six individual genes in bread wheat of which four (TaHKT2;1 7AL-1, TaHKT2;1 7BL-1, TaHKT2;1 7BL-2 and TaHKT2;1 7DL-1) were identified as functional members. Analysis of predicted proteins of TaHKT2;1 multi-gene family Functional members of TaHKT2;1 gene family on 7AL and 7BL encoded proteins of 533 amino acids whereas the gene on 7DL encoded a protein of 531 amino acids ( Table 2) having 91% to 93% protein identity (Table 1) with TaHKT2;1 7BL-1/-2. Sequence alignment ( Figure 2) coupled with analysis of hydrophobicity (Additional file 1: Figure S1) and 3D structures of the predicted proteins revealed the topology of four sequentially arranged membrane-pore-membrane (MPM) domains in each protein ( Figure 2). Glycine molecules at Gly91, Gly246, Gly370 and Gly473 that form the selectivity pore filters [19,20] were conserved in each protein ( Figure 2) predicting that each functional member has potential to mediate both K + and Na + transport. Interestingly, several amino acid substitutions were identified in the close proximity of filter glycine residues on first, second and fourth MPM domains of TaHKT2;1 7AL-1 and TaHKT2;1 7DL-1 compared to TaHKT2;1 7BL-1/-2 ( Figure 2). In the first M1 a P a M2 a , Ile95 in TaHKT2;1 7BL-1/-2 was substituted by Val, an amino acid having similar physical properties, in both TaHKT2;1 7AL-1 and TaHKT2;1 7DL-1. The substitution Thr96 for another

hydrophilic but larger amino acid Lys was shared in both TaHKT2;1 7AL-1 and TaHKT2;1 7DL-1. The second M1 b P b M2 b identified two substitutions, one at Ala240 for Ser, a more hydrophilic amino acid in the same size range and Cys242 for a large hydrophobic amino acid Phe in TaHKT2;1 7DL-1. The fourth M1 d P d M2 d identified Ala472 in both TaHKT2;1 7AL-1 and TaHKT2;1 7DL-1, substituted for Val, an amino acid carrying large aliphatic hydrophilic side chains. Altered physical properties and steric hindrance caused by the amino acid substitutions could, therefore, potentially change structural properties of the filters and manifest modified protein functions for ion selectivity and transport. Promoter analysis of functional TaHKT2;1 members Promoter diversity was investigated in functional members of the TaHKT2;1 gene family by analysis of 2000 bp region up-stream of the predicted translation start site. The core promoter element, TATAA box (−69 bp) was conserved in all functional members of the TaHKT2;1 family. Although TaHKT2;1 7BL-1/-2 were identical in exon and intron sequences, they shared only 42.7% DNA sequence identity between promoters. The promoter region of TaHKT2;1 7AL-1, and TaHKT2;1 7DL-1 genes showed 65.8% and 43.5% DNA sequence identity, respectively, compared with TaHKT2;1 7BL-1/-2. Table 3 summarizes the number and type of cis-acting regulatory elements (CREs) identified in promoters of functional members of the TaHKT2;1 family using the PLANTcare and PLACE databases. Promoters of all functional TaHKT2;1 genes contained a number of CREs known to be involved in salt activated response in plants and each gene shared at least five CREs of different categories. Therefore, it appeared that salt stress-activated transcription factors (TFs) interact with CREs to regulate TaHKT2;1 genes. WRKY transcription factors [31] The number of each element in the promoter represented in each gene is shown in parenthesis. amplify the region encoding the premature stop codon in exon 1 was confirmed by the absence of PCR amplicons from nullisomic-tetrasomic lines N7AT7B and N7AT7D ( Figure 3B). The single base deletion at +191 bp in TaHKT2;1 7AL-2/-3 relative to TaHKT2;1 7AL-1 ( Figure 3C) responsible for the in-frame stop codon was conserved in all diploid, tetraploid and hexaploid Triticum accessions analysed ( Figure 3D) indicating that the deletion event was an ancient evolutionary sequence variant and the resulting pseudogenes were retained in modern durum and bread wheat accessions. Selection constraints on the genes were estimated by replacement (Ka) to silent (Ks) ratio [32]. The ratio between the functional genes TaHKT2;1 7AL-1 and TaHKT2;1 7BL-1/-2 was 0.20. Higher ratio value of 0.56 between TaHKT2;1 7AL-2/ -3 and the functional counterpart TaHKT2;1 7AL-1 indicated relaxed selection pressure on the pseudogenes. Expression of functional members of TaHKT2;1 gene family The expression of functional members of the TaHKT2;1 gene family were investigated by imposing NaCl treatments on Chinese Spring seedlings and detection of transcripts specific for each gene in roots, sheaths and leaf blades. After three days exposure at 200 mM NaCl treatment in the external solution, Na + concentration was 4-fold higher in roots, 3-fold in sheath and 2-fold in leaves, compared to the control plants with no added NaCl in the external solution ( Figure 4A) and, therefore, showed significant (P < 0.01) phenotypic differences for transcript analysis of the different TaHKT2;1 genes. Primers were designed to amplify FL-cDNA of each functional member of the TaHKT2;1 gene family and gene specificity of primers was validated using nullisomic-tetrasomic lines ( Figure 4B). TaHKT2;1 7AL-1 and TaHKT2;1 7BL-1/-2 were expressed in roots, sheaths and leaf blades under both low and high NaCl conditions whilst TaHKT2;1 7DL-1 was identified only in roots and sheaths ( Figure 4C). Each functional member of the TaHKT2;1 gene family was down regulated in the leaf blades ( Figure 4C). A unique feature of the TaHKT2;1 7AL-1 gene was an additional smaller transcript of approximately 600 bp in length identified only in root and sheath tissues under both NaCl treated and non-treated conditions ( Figure 4C), indicating that alternative splicing was evident for this gene in these tissues. No RT-PCRamplification of TaHKT2;1 7AL-2 or TaHKT2;1 7AL-3 was detected in any of the tissues analysed (data not shown), providing further evidence that these were, indeed, unprocessed pseudogenes. FL-cDNA transcripts of functional TaHKT2;1 genes expressed in root, sheath and leaf under low and high NaCl conditions were sequenced. The cDNA sequence for TaHKT2;1 7AL-1 [Genbank: KJ540616]; TaHKT2;1 7BL-1/-2 [Genbank: U16709] and TaHKT2;1 7DL-1 [Genbank: KJ540618] were identical from each tissue and were in perfect agreement with the predictions of gene structure in Figure 1 when aligned against the cognate genomic DNA sequence. Sequence analysis identified a 571 bp alternatively spliced transcript [Genbank: KJ540617] from TaHKT2;1 7AL-1 encoding a predicted protein of 133 amino acids. Alignment with genomic DNA identified a splice site within exon 1 resulting in the deleting of exon 2 and exon 3 and intervening introns relative to its larger cDNA counterpart [Genbank: KJ540616]. The alternatively spliced TaHKT2;1 7AL-1 sequence had the dinucleotide motif 5′-AT and AC-3′ at the exon-intron splice junction different to the conserved 5′GT and AG-3′ splice junction motifs in all other transcripts. Tissue ion analyses of wheat aneuploid lines null for TaHKT2;1 genes An investigation of possible phenotypic effects of TaHKT2;1 genes was conducted by evaluating tissue Na + and K + concentrations in wheat aneuploid lines null for each member of the TaHKT2;1 gene family, when exposed to NaCl and different K + supply in the external solution. The absence of DNA amplicons from aneuploid lines using gene specific PCR primers ( Figure 5A) confirmed that TaHKT2;1 7AL-1, and TaHKT2;1 7DL-1 were located in the distal region of 7AL and 7DL respectively, whereas TaHKT2;1 7BL-1/-2 were located within the proximal region of 7BL ( Figure 5B). Therefore, the aneuploid lines 7AL-1 and Dt7BS were null for the functional genes TaHKT2;1 7AL-1 and TaHKT2;1 7BL-1/-2 respectively, whereas 7DL-2 was null for TaHKT2;1 7DL-1 and suitable for comparing ion concentrations with Chinese Spring under NaCl and K + supply. Analysis of tissue Na + concentration showed no difference between aneuploid lines and Chinese Spring in roots, bulked leaf blades or youngest fully expanded leaf blade regardless of low (0.2 mM) or high (4 mM) K + status when genotypes were treated with an external solution containing 200 mM NaCl ( Figure 6A). An increase of 24% in sheath Na + concentration was observed in the aneuploids, 7AL-1 ( Figure 6A) compared to Chinese Spring under both low and high K + conditions. Similarly, Dt7BS showed 25% increase in sheath Na + but only under low K + conditions. Therefore, the chromosomal regions containing TaHKT2;1 7AL-1 and TaHKT2;1 7BL-1/-2 may have a significant effect on Na + concentration in the sheath but not in the roots, bulk leaves nor the youngest fully expanded leaf blade. Interestingly, 7DL-2 null for TaHKT2;1 7DL-1 showed no significant difference between Chinese Spring ( Figure 6A) indicating this gene does not have a role in controlling Na + concentration in any tissue or treatment. The same aneuploid lines when exposed to an external solution containing 200 mM NaCl and 4 mM K + showed no differences in K + concentration in bulked leaf blades regardless of the genotype or the Na + status ( Figure 6B). However, K + concentration was reduced by 23% in the roots and 19% in the youngest fully expanded leaf blade for Dt7BS compared to Chinese Spring, but only under 200 mM NaCl treatment ( Figure 6B). Similarly, sheath K + was reduced by 12% in the aneuploid line 7AL-1 in the 200 mM NaCl treatment ( Figure 6B). Therefore, it appears that the chromosomal regions containing TaHKT2;1 7AL-1 and TaHKT2;1 7BL-1/-2 influence K + status under high NaCl conditions in a tissue-specific manner. There was no significant difference in K + between 7DL-2 and Chinese Spring ( Figure 6B) indicating that TaHKT2;1 7DL-1 does not have a role in regulating K + status in any tissue and treatments. Discussion This study provides a detailed characterization of the TaHKT2;1 multigene family in bread wheat which included four functional members and two pseudogenes. Despite presence of salt induced stress responsive elements in the promoter region it was found that the functional members of the TaHKT2;1 family were expressed under low and high NaCl levels in roots and sheaths, but were down regulated in leaf blades. Wheat aneuploid lines were used to provide evidence for a potential role of individual TaHKT2;1 gene family members in Na + and K + homeostasis in seedlings under low and high NaCl hydroponic conditions. Although all members of the TaHKT2;1 gene family contained three exons and two introns sharing the same gene structure as Group II HKT genes from other monocots [33], DNA sequence differences resulted in a degree of variability at the protein level. Variation amongst the TaHKT2;1 genes and their encoded proteins could be of functional relevance. TaHKT2;1 proteins potentially acting as K + /Na + cotransporters have a conserved Gly-Gly-Gly-Gly signature in the selectivity filter, yet the filter signature alone does not necessarily predict ion selective properties and other domains responsible for ion specificity need to be considered [34,35]. In this regard, induced mutations within the cytoplasmic (L149, E180) or P-loop (A240, L247, F463 and E464) domains affected selectivity and the rate of uptake of K + and Na + by the TaHKT2;1 7BL-1/-2 proteins in yeast and/or Xenopus oocytes heterologous systems [17,19,21,23,36,37]. Interestingly, the majority of the amino acids that were mutated in previous experiments were conserved between the members of TaHKT2;1 family except for a A240S substitution in TaHKT2;1 7DL-1. In addition to A240S, five amino acid differences were identified within six base pair proximity to the filter glycine residues in the P-loops of TaHKT2;1 7AL-1, TaHKT2;1 7DL-1 compared to TaHKT2;1 7BL-1/-2. Therefore, it is reasonable to hypothesize that the amino acid differences between members of the TaHKT2;1 gene family in these regions may affect ion uptake properties (e.g. specificity and rate of ion transport). Since this study cloned FL-cDNAs of each member of the gene family, further investigations can be pursued using heterologous systems to evaluate the functional significance of the protein sequence variability for ion transport in the TaHKT2;1 gene family. TaHKT2;1 7AL-2 and TaHKT2;1 7AL-3 shared typical characteristics of unprocessed pseudogenes, such as inframe stop codons, location within close proximity to the functional genes on the same chromosome, and the absence of transcription [38]. Pseudogenes have previously been reported in the HKT super-family but their functional relevance is unknown [39]. Pseudogenes are thought to be an integral feature in evolution of polyploidy genomes, representing 12% of the genes in bread wheat [40] and may be of importance in gene or genome function. Pseudogenes in eukaryotes have been reported to have a regulatory role for genes from which they were derived, have non-coding functions or act as a source of genetic variability and were subjected to natural selection [38]. Conservation across progenitor genomes indicated that the single base deletion in exon 1 that caused a pre-mature stop codon in TaHKT2;1 7AL-2 and TaHKT2;1 7AL-3 was an ancient event created during evolution of the A genome progenitor species prior to, and maintained during, polyploidization of the tetraploid and hexaploid genomes. The Ka/Ks between the functional genes was close to the average Ka/Ks ratio of 0.2 previously estimated for functional wheat genes [41], providing evidence that the functional members of the TaHKT2;1 gene family was subjected to resilient purifying selection. Although TaHKT2;1 7AL-2 and TaHKT2;1 7AL-3 had characteristics of pseudogenes the Ka/Ks value between these genes and TaHKT2;1 7AL-1 was less than the theoretical value of 1.0 for a bona-fide pseudogene [38], indicating a relaxed selective constraint during genome evolution. Furthermore, the fact that these pseudogenes were retained during evolution of the hexaploid wheat genome with gene structure conservation could indicate some functional importance. Each member of the TaHKT2;1 gene family was variable for presence and frequency of salt induced CREs. Plant response to salinity is mediated by immediate and extensive reprogramming of the spatial and temporal gene transcription, instigated by the interaction of transcription factors with four major groups of CREs including ABA-independent CREs (CBF/DREB; NAC and ZF-HD) and ABA-dependent CREs (AREB/ABF; MYC/ MYB) [42,43], all of which were represented in TaHKT2;1 gene family. Interestingly, promoter diversity in AtHKT1;1 genes resulted in variable salt response in

natural populations of Arabidopsis [44]. Similarly, variability in CREs contributed to diverse expression patterns of TaHKT1;5 in bread wheat and wild relatives under salt stress [45], supporting the hypothesis that HKT genes are regulated by CREs mediated mechanisms. The presence of these elements with varying copies and combinations predict salt response with variable expression patterns of individual TaHKT2;1 members. Given the CREs identified in this study, it was expected that each TaHKT2;1 gene would show diverse transcriptional regulation, yet, no obvious transcript differences were detected between low and high NaCl treated plants. Similar observations were reported [24] where saline treatments imposed with K + deficient conditions did not induce differences in TaHKT2;1 gene expression in two week old wheat seedlings. However, expression of TaHKT2;1 7BL-1/-2 was up regulated by imposing K + starved conditions in six day old wheat [46] indicating a rapid response of genes to external K + in younger seedlings. In the present study, there was an obvious down regulation of TaHKT2;1 transcripts in leaf blades regardless of the NaCl concentration of the external solution, confirming adjustment of gene expression in this tissue. The extent of gene expression was further explored by mining wheat EST databases including NCBI, Transcriptome Shotgun Assembly and the TIGR Plant Transcript Assemblies, identifying a number of ESTs derived from pre-anthesis spikes and kernel tissue with 98-100% identity to members of the TaHKT2;1 gene family (data not shown). Further studies are needed to understand the expression of TaHKT2;1 gene family and how the various members influence ion homeostasis during different stages of plant development, various environmental conditions and variable salt levels. An interesting feature of TaHKT2;1 gene expression was the level of complexity caused by alternative splicing of a 571 bp transcript unique to TaHKT2;1 7AL-1. Alternative splicing is evident in 25 to 42% of bread wheat genes whereby some forms are thought to play a role in regulating functional transcript levels [47]. Alternative splicing at non-conventional-splice junctions has been reported in rice OsHKT1;5 resulting in a premature stop codon and non-functional transcripts as a possible means of regulating mRNA degradation [34]. Similarly, alternative splicing event in TaHKT2;1 7AL-1 indicates a rare unstable splice-junction motif that may act as an embedded molecular switch for post-transcriptional regulation and mRNA turn-over processes, which presumably may then influence levels of protein production. The use of wheat aneuploid lines null for specific members of the TaHKT2;1 family provided a system to analyse phenotypic effects of the genes on tissue ion concentrations. This approach provided evidence that a gene does not have a specific role when there was no significant difference between its corresponding aneuploid line and Chinese Spring (containing the full chromosome complement) whereas differences indicated that the gene would be a candidate controlling tissue Na + and/or K + status. The results, therefore, corroborate that TaHKT2;1 genes neither regulate root Na + uptake nor Na + accumulation in bulk leaf blades and youngest fully expanded leaf blade. Nevertheless, presence of transcripts for all TaHKT2;1 gene family members in these tissues under low and high NaCl conditions indicate alternative roles if they are translated into proteins. The higher sheath Na + concentration in 7AL-1 and Dt7BS is likely from increased Na + concentration in the xylem flow and being preferentially stored in sheaths. Wheat and other grasses do have mechanisms for withdrawal of Na + from xylem in roots and sheath to reduce delivery of Na + into the leaf blade [48]. Therefore, TaHKT2;1 7AL-1 and TaHKT2;1 7BL-1/-2 genes are candidates regulating Na + accumulation in sheaths, perhaps by modifying Na + in the root xylem that affects the flow of Na + into sheaths. The possibility of other genes in the same chromosomal region contributing to this phenotypic effect cannot be excluded and further studies are needed to confirm the function of TaHKT2;1 genes on 7AL and 7BL in controlling Na + accumulation in the sheaths. The TaHKT2;1 7AL-1 gene may also regulate K + accumulation in the sheath but only under high NaCl conditions, whereas TaHKT2;1 7BL-1/-2 may serve a similar role in roots and the youngest fully expanded leaf blade, evident by the lower K + in these tissues in the Dt7BS line. Since TaHKT2;1 genes are not involved in K + uptake by roots under saline conditions [18,49], the lower root K + concentration observed for Dt7BS in this study would be due to a greater net K + translocation from roots to shoots. Therefore, it is reasonable to assume that TaHKT2;1 7BL-1/ TaHKT2;1 7BL-2 genes control K + translocation in specific tissues of salt stressed wheat. Extensive functional redundancy of K + transport systems [50] can be a challenge in isolated studies on individual transporter proteins, yet evidence in the present study identified the importance of further investigating TaHKT2;1 7BL-1/TaHKT2;1 7BL-2 and TaHKT2;1 7AL-1 genes in controlling K + homeostasis. Conclusions TaHKT2;1 genes has been of interest because the encoded proteins implement pathways for Na + entry into wheat in saline conditions [24]. However, previous studies do not recognize a role of individual members of the TaHKT2;1 gene family in controlling Na + or indeed, of K + , during plant growth and development. The present study has highlighted diversity amongst the TaHKT2;1 gene family and predicted proteins, as well as promoter variability in salt signalling and presence of TaHKT2;1 transcripts in different tissues, under low and high levels of external Na + and K + , gaining knowledge that TaHKT2;1 genes could assume multiple roles in bread wheat. The phenotypic effects when individual genes were deleted in bread wheat showed that TaHKT2;1 gene family are not involved in regulating root or leaf blade Na + concentration but TaHKT2;1 7AL-1 and TaHKT2;1 7BL-1/-2 may be candidate genes controlling sheath Na + concentration and K + status in different tissues and experimental conditions. TaHKT2;1 7DL-1 had no apparent role in regulating Na + and K + status in different tissue and treatments in this study. In summary, the present findings from combined molecular and physiological studies provided a broader perspective of the role of the TaHKT2;1 family of genes in bread wheat. In-silico analysis of TaHKT2;1 gene family The TaHKT2;1 FL-cDNA, U16709, was used as the query sequence in BLASTN to search IWGSS database (www.wheatgenome.org) and retrieve coding and promoter regions of the individual members of TaHKT2;1 gene family. In-silico gene structure predictions for individual members of the TaHKT2;1 family were made using Spidey (http://www.ncbi.nlm.nih.gov/spidey/), GeneSeqer (http://www.plantgdb.org/cgi-bin/GeneSeqer/index.cg) and SIM4 (http://pbil.univ-lyon1.fr/sim4.php) software. Intron-exon structures for each member were confirmed by aligning retrieved genomic sequences with the FL-cDNA, U16709. DNA and protein sequence analysis and structure predictions were made using GENEIOUS 6.0.3 [51] and phylogenetic analysis was performed in MEGA 5.05 [52]. DNA sequence 2000 bp upstream of predicted initiation site were analysed for cis-acting regulatory elements (CREs) associated with salt stress response in PlantCARE [54] and PLACE [55] databases. DNA extraction and PCR amplification Plant genomic DNA was extracted from leaves using a phenol-chloroform-based method as described in [56] and was used as a template for PCR to amplify individual members of the TaHKT2;1 family. Gene specific primer pairs were designed by targeting insertions and/or deletions (INDELs) and single nucleotide polymorphisms (SNPs) based on 3′ terminus mismatch described in [57]. NetPrimer (http://www.premierbiosoft.com/ netprimer/netprlaunch/netprlaunch.html) was used to confirm primer compatibility. Primer information is summarized in Table 1 Fe-EDTA, 0.05; MES, 1.0; and micronutrients of onequarter-concentration in Hoagland solution (pH was adjusted to 6.5 using Ca(OH) 2 ). Four day old seedlings were then transferred to full strength nutrient solution in aerated, foil-covered, 4.5 L pots for 12 days. Each pot carried one individual of each of all six genotypes arranged randomly in each pot. There were four replicates of each genotype for each treatment in the experiment and the pots were arranged in a completely randomised design. Seedlings were held upright by a foam holder at the stem base inserted into individual holes in the pot lids. The experiment was carried out in a temperature controlled phytotron (20 ± 3°C/15 ± 2°C day/night) and the photosynthetically active radiation (PAR) recorded at midday was at ca. 1,400 -1,500 μmol m −2 s −1 . Solution levels in pots were maintained by topping up with deionized water. In order to induce expression and study phenotypic effects of the TaHKT2;1 gene family, the plants were subjected to low K + conditions prior to imposing salt treatments [24,46]. As known TaHKT2;1 transporters might be involved with either Na + and/or K + transport [16,21,22] NaCl treatments were also imposed at the two K + levels: low (0.2 mM) K + , where high affinity transport systems for K + dominate and high (4 mM) K + , where K + uptake kinetics was determined by low affinity systems [58]. 12 day old seedlings were subjected to low K + conditions for four days in a modified nutrient solution without added K + . The media contained (mM) Ca 2+ , 4; Mg 2+ , 0.4; NH 4 + , 0.825; NO 3 − , 4.375; SO 4 2 -, 0.4; HPO 4 2 -, 0.2; Fe-EDTA, 0.05; MES, 1.0; and micronutrients at one-quarter-concentrations in Hoagland solution (pH = 6.5, adjusted with Ca(OH) 2 ) at a background Na + and K + concentrations of 0. 3 mM and 0.1 mM, respectively. After 4 days of low K + level pre-treatment, 200 mM NaCl was applied to designated pots, so as to obtain four different Na + /K + combinations: no added Na + /0.2 mM K + , no added Na + /4 mM K + , 200 mM Na + /0.2 mM K + and 200 mM Na + /4 mM K + . The NaCl treatment was applied in 50 mM increments at 12 hour intervals and plants were maintained at the final 200 mM concentration for a full 3 days. The youngest fully expanded leaf blade, bulk leaf blades (all others), sheaths and roots of each plant were sampled. Prior to excision, roots were washed three times for 10 secs each in a solution containing 4 mM CaSO 4 and 368 mM mannitol (200 mM treated plants) or in 4 mM CaSO 4 (plants subjected to no added NaCl). Leaves and sheath samples were rinsed in deionized water. All tissue samples for ion analyses were oven-dried at 70°C, weighed and then ground to a powder. Samples of dried powdered tissue (100 mg) were extracted in 5 ml of 0.5 M HNO 3 for 3 days on a mechanical shaker. Na + and K + were measured in technical triplicate using a flame photometer (Sherwood 410, Cambridge) and mean values represented each treatment replicate tissue sample. The reliability of the methods was confirmed by analyses of a reference tissue (broccoli, ASPAC Plant number 85) taken through the same procedures. Mean values from 4 treatment replicates of each genotype were analysed by ANOVA using SAS 9.4 (SAS Inc., USA) software. RNA extraction and RT-PCR T. aestivum var. Chinese Spring was grown in hydroponics and treated with NaCl as described above. Leaf blades (including youngest fully expanded leaf blade), sheaths and roots of plants treated with 4 mM K + with no added Na + and 4 mM K + with 200 mM NaCl were harvested after treating the plants for three days, snap frozen in liquid nitrogen for RNA extraction and RT-PCR. Duplicate tissue samples were harvested for Na + analysis (described above). Total RNA was isolated using RNeasy Plant Mini Kit (Qiagen, Hilden, Germany). First strand cDNA was synthesised using RT Omniscript Kit (Qiagen, Hilden, Germany). Integrity of synthesised cDNA was verified by RT-PCR using primers (forward 5′-CGCC AGGGTTTTC CCAGTCAC GAC −3′, reverse 5′-TCAC ACAGGAAAC AGCTATGAC −3′) designed for glyceraldehyde-3-phosphate dehydrogenase (GAPDH) gene [59]. FL-cDNA of individual members of the TaHKT2;1 gene family were amplified from cDNA preparations using primer pairs FL:A1 F and FL:A1 R, FL:A2/3 F and FL:A2/3R, FL:B1/2 F and FL:B1/2 R and FL:D1 F and FL:D1 R (Table 1). Early Postnatal Cardiac Changes and Premature Death in Transgenic Mice Overexpressing a Mutant Form of Serum Response Factor* Serum response factor (SRF) is a key regulator of a number of extracellular signal-regulated genes important for cell growth and differentiation. A form of the SRF gene with a double mutation (dmSRF) was generated.

This mutation reduced the binding activity of SRF protein to the serum response element and reduced the capability of SRF to activate the atrial natriuretic factor promoter that contains the serum response element. Cardiac-specific overexpression of dmSRF attenuated the total SRF binding activity and resulted in remarkable morphologic changes in the heart of the transgenic mice. These mice had dilated atrial and ventricular chambers, and their ventricular wall thicknesses were only 1 ⁄ 2 to 1 ⁄ 3 the thickness of that of nontransgenic mice. Also these mice had smaller cardiac myocytes and had less myofibrils in their myocytes relative to nontransgenic mice. Altered gene expression and slight interstitial fibrosis were observed in the myocardium of the transgenic mice. All the transgenic mice died within the first 12 days after birth, because of the early onset of severe, dilated cardiomyopathy. These results indicate that dmSRF overexpression in the heart apparently al-ters cardiac gene expression and blocks normal postnatal cardiac growth and development. serum and incubated at 37 °C overnight. The cells were then washed with 2 ml of serum-free DMEM and cultured in DMEM containing 0.1% serum for another 48 h. The DMEM containing 0.1% newborn bovine serum was then replaced by DMEM containing 10% newborn bovine serum, and the cells were cultured for an additional 3.5 h. Firefly luciferase activity was meas- ured as relative light units using a dual luciferase reporter assay system (Promega). The number of relative light units from individual transfection experiments was normalized by measurement of Renilla luciferase activity expressed from a cytomegalovirus promoter-driven vector in the same samples. Individual transfection experiments were carried out in triplicate, and the results were reported as mean firefly luciferase/ Renilla luciferase activity (mean (cid:1) S.D.) from at least three separate experiments. Generation of (cid:1) -MHC-dmSRF Transgenic Mouse Lines— The DNA fragment corresponding to the full-length coding region of dmSRF was subcloned into the blunted Sal I site of a pBluescript II KS( (cid:2) ) plasmid containing the (cid:1) -MHC promoter (a generous gift from Dr. J. Robbins, The Children’s Hospital and Research Foundation, Cincinnati, OH). A Not I fragment containing the transgenic construct and the pronuclear stage zygotes of FVB/N mouse were used for microinjection. At 2–3 weeks of age, a 1-cm portion of tail was removed from each mouse for DNA analysis. The potential transgenic mice were screened twice by the polymerase chain reaction using two different forward primers (primer 1, 5 (cid:3) -ACAGGTGGTGAACCTGGACAC-3 (cid:3) , and primer 2, 5 (cid:3) -CCATTCAAGTGCACCAGGC-3 (cid:3) ) and one reverse primer (5 (cid:3) -CACTG- GAGTGGCAACTTCCAG-3 (cid:3) ). Southern blot analysis, using a [ (cid:1) - 32 Serum response factor (SRF) 1 is a key regulator of many extracellular signal-regulated genes important for cell growth and differentiation (1)(2)(3). SRF was first identified as a critical transcription factor involved in mediating serum-induced transcriptional activation of the c-fos gene (4,5). Attention has been given to the role of SRF in the activation of c-fos gene in the G 0 -to-G 1 transition in the cell cycle and in many other cell activation responses (6,7). The importance of SRF for growth factor-regulated transcription is further suggested by the iden-tification of SRF-binding sites (serum response elements (SREs)) within the regulatory region of many other seruminducible genes, including members of the immediate-early genes and muscle-specific genes (8 -11). SRF is a member of the MADS (MCM-1, Agamous and Deficiens, SRF) box family of transcription factors (5). The MADS box motif is a 56-amino acid region that comprises a highly conserved basic N-terminal region followed by a less well conserved, relatively hydrophobic C-terminal segment. The SRFbinding domain is located at the N terminus of the protein in which the MADS box (amino acids 142-198) forms the center of the binding domain that is important for DNA binding specificity, dimerization, and recruitment of p62TCF proteins (12)(13)(14). Mutations in this domain attenuate SRF binding specificity (12,15). Recently, several studies have suggested that SRF is directly involved in the regulation of expression of a number of musclespecific genes, including skeletal ␣-actin (SKA), ␣-myosin heavy chain (␣-MHC), ␤-MHC, and cardiac ␣-actin (16), and in the regulation of mesoderm formation (17,18). SRF also plays an essential role in shaping the normal morphology and size of the embryonic body during embryonic cell differentiation and aggregation (19). To define the role of SRF in the regulation of function and morphology of the postnatal heart and to test the hypothesis that functional SRF is required for postnatal cardiac growth and development, we overexpressed a double mutant form of SRF (termed dmSRF), which has impaired binding ability to SRE, in the mouse heart. Our results revealed that dmSRF overexpression in the heart elevated the levels of SRF mRNA and protein but attenuated the total SRF binding activity to the SRE. It resulted in remarkable morphological changes in the transgenic mice, such that all the transgenic mice died within 12 days after birth, because of the early onset of severe, dilated cardiomyopathy. EXPERIMENTAL PROCEDURES SRF Mutant-A functional double mutant form of human SRF gene, termed dmSRF, was generated by site-directed mutagenesis. It contains a double mutation at amino acid positions 159 and 163 of the SRF protein (Thr 159 3 Ser 159 ; Lys 163 3 Glu 163 , based on the GeneBank TM sequence accession number J03161). When compared with wild type SRF, this mutant is severely compromised in its ability to bind to the c-fos SRE in in vitro binding assays. In Vitro Translation and Cotranslation of wtSRF and dmSRF Protein-The DNA fragments corresponding to the full-length coding region of dmSRF and wild type SRF (wtSRF) were subcloned into plasmid pBluescript SK(-). The transcription of both the mutant and wild type SRF genes was under the control of T7 promoter. The dmSRF and wtSRF proteins were in vitro translated by using a TNT-coupled transcription/translation system (Promega). Equal amounts of the in vitro * This work was supported in part by National Institutes of Health Grants AG00294, AG08812, AG13314, AG18388, and AG00251. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. translated dmSRF and wtSRF protein were used for electrophoretic mobility shift assays (EMSAs). In the cotranslation experiment, 1 l of in vitro translated wtSRF RNA was added to tubes containing 0, 1, 2, 4, or 6 l of dmSRF RNA, respectively. The wtSRF and dmSRF proteins were synthesized in the same cotranslation reaction in the presence of [ 35 S]methionine. The lysate of this cotranslation reaction and the SRE probe were utilized for EMSA. Transfection and Cotransfection Assays-The plasmids for the transfection assay were pGL3-ANF promoter (a generous gift from Dr. A. Usheva, Harvard Medical School, Boston, MA), pAdTrack-CMV (a generous gift from Dr. T. C. He, John Hopkins Oncology Center, Baltimore, MD), pAd-CMV-dmSRF, and pAd-CMV-wtSRF. In the transfection assays, NIH3T3 cells were transfected with complexes of 200 ng of reporter plasmid (pGL3-ANF-luciferase) and either 200 ng of pAd-CMV-dmSRF or 200 ng of pAd-CMV-wtSRF plasmid. In the cotransfection assays, the cells were transfected with 200 ng of pGL3-ANF-luciferase plasmid, 200 ng of pAd-CMV-wtSRF plasmid, and an excess amount of dmSRF plasmid relative to wtSRF plasmid (400 ng (2-fold), 800 ng (4-fold), and 1000 ng (5-fold) of dmSRF, respectively). Plasmid pRL-CMV (Promega; 10 ng/well) was transfected at the same time as an internal control. Transient transfections were carried out with a Lipo-fectAMINE transfection kit (Life Technologies, Inc.). Approximately 4 h after the transfection was initiated, NIH3T3 cells were placed in DMEM containing 10% newborn bovine serum and incubated at 37°C overnight. The cells were then washed with 2 ml of serum-free DMEM and cultured in DMEM containing 0.1% serum for another 48 h. The DMEM containing 0.1% newborn bovine serum was then replaced by DMEM containing 10% newborn bovine serum, and the cells were cultured for an additional 3.5 h. Firefly luciferase activity was measured as relative light units using a dual luciferase reporter assay system (Promega). The number of relative light units from individual transfection experiments was normalized by measurement of Renilla luciferase activity expressed from a cytomegalovirus promoter-driven vector in the same samples. Individual transfection experiments were carried out in triplicate, and the results were reported as mean firefly luciferase/Renilla luciferase activity (mean Ϯ S.D.) from at least three separate experiments. Generation of ␣-MHC-dmSRF Transgenic Mouse Lines-The DNA fragment corresponding to the full-length coding region of dmSRF was subcloned into the blunted SalI site of a pBluescript II KS(ϩ) plasmid containing the ␣-MHC promoter (a generous gift from Dr. J. Robbins, The Children's Hospital and Research Foundation, Cincinnati, OH). A NotI fragment containing the transgenic construct and the pronuclear stage zygotes of FVB/N mouse were used for microinjection. At 2-3 weeks of age, a 1-cm portion of tail was removed from each mouse for DNA analysis. The potential transgenic mice were screened twice by the polymerase chain reaction using two different forward primers (primer 1, 5Ј-ACAGGTGGTGAACCTGGACAC-3Ј, and primer 2, 5Ј-CCATTCAAGTGCACCAGGC-3Ј) and one reverse primer (5Ј-CACTG-GAGTGGCAACTTCCAG-3Ј). Southern blot analysis, using a [␣-32 P]dCTP-labeled dmSRF cDNA fragment from plasmid pCGN-dm-SRF, was employed to confirm the identification of transgenic founder mice and to determine the transgene copy number in different transgenic lines. In all experiments that were performed in this study, ageand sex-matched nontransgenic littermates were used for comparison with the dmSRF transgenic mice. The studies were conducted with approval of the Institutional Review Board of Beth Israel Deaconess Medical Center. Electrophoretic Mobility Shift Assays-The in vitro translated protein of wtSRF and dmSRF and the whole cell lysate of mouse ventricular tissue were employed for EMSAs. Whole cell extracts of ventricular tissue were prepared by a modification of the method described previously (20). Briefly, cells or tissue were washed with cold phosphatebuffered saline and then suspended in buffer C containing 20 mM Hepes (pH 8.0), 1.5 mM MgCl 2 , 25% (v/v) glycerol, 420 mM NaCl, 0.2 mM EDTA (pH 8.0), 1 mM dithiothreitol, and 1ϫ protease inhibitor mixture (Roche Molecular Biochemicals). They were homogenized and then incubated on ice for 30 min before centrifuging at 10,000 rpm for 15 min. The SRE consensus oligonucleotide was of the sequence 5Ј-GGATGTCCATATT-AGGACATCT-3Ј, which is derived from the c-fos promoter. The SRE mutant oligonucleotide was of the sequence 5Ј-GGATGTCCATATTAT-TACATCT-3Ј, to which SRF is unable to bind. Oligonucleotides were labeled with [␥-32 P]ATP using T4 polynucleotide kinase. Binding reaction mixtures were incubated at room temperature for 20 min and contained 0.5 ng of DNA probe and 5 g of protein in the binding buffer with 10 mM Tris (pH 7.5), 50 mM NaCl, 1 mM dithiothreitol, 1 mM EDTA, 5% glycerol, and 1 g of poly(dI-dC) to inhibit nonspecific binding of the labeled probe to the ventricular tissue protein. DNA-protein complexes were resolved by electrophoresis through 4% native polyacrylamide gels containing 50 mM Tris, 45 mM boric acid, 0.5 mM EDTA. The gels were subsequently dried and exposed to Kodak X-Omat film. Gel supershift assays were performed as described above with the exception that subsequent to incubation of oligonucleotide probes with the whole cell extract, 1 l of of anti-SRF antibody (1 g/l; Santa Cruz Biotechnology) was added to the reaction mixture and incubated at room temperature for 30 min. Measurement of SRF Protein Expression-50 g of protein prepared as described in the electrophoretic mobility shift assays section above was separated by SDS-polyacrylamide gel electrophoresis on a 10% polyacrylamide gel and transferred to nitrocellulose. The membrane was blocked for 2 h at room temperature in 5% nonfat milk in TBS-T (20 mM Tris, 137 mM sodium chloride, 0.1% Tween 20, pH 7.6) and then incubated for 2 h at room temperature with SRF antibody followed by incubation for 1 h with horseradish peroxidase-conjugated secondary antibody (both from Santa Cruz Biotechnology). Immunoreactive bands were visualized by chemiluminescence (ECL; Amersham Pharmacia Biotech). Northern Blot Analysis-Total RNA was isolated from mouse ventricular tissue using the ULTRASPEC RNA isolation reagent (Biotecx Laboratories, Houston, TX). Approximately 10 g of total RNA was then fractionated on a 1% formaldehyde-agarose gel and transferred to a Hybond nylon membrane (Amersham Pharmacia Biotech) by capillary action in high salt solution (10ϫ SSC, 1 mM EDTA). The blots were prehybridized in a hybridization solution containing 7% SDS, 0.5 M Na 2

HPO 4 (pH 7.2), and 250 g/ml salmon sperm DNA for 5 h at 65°C, followed by overnight hybridization with [␥-32 P]ATP-labeled oligonucleotide probes or a [␣-32 P]dCTP-labeled SRF cDNA probe. The blots were washed three times in 2ϫ SSC, 0.2% SDS at room temperature for 30 min and then in 0.5ϫ SSC, 0.2% SDS at 65°C for 15-30 min before exposure to x-ray film. Morphologic Analysis-Fresh ventricular tissues were placed in 10% neutral buffered formalin overnight. After fixing, the samples were subjected to a dehydration series and embedded in paraffin. The sections (3-4 M) were stained using standard hematoxylin and eosin or Masson trichrome staining protocols (Poly Scientific, Bayshore, NY). The ImmunoCruz System kit and SRF polyclonal antibody (Santa Cruz Biotechnology) were used for immunohistochemistry. Photomicrographs were obtained using a Nikon ES400 microscope. Measurement of Wall Thickness and Cell Size-Slides of cross-sections of the heart taken at the level of the left ventricular papillary muscles were stained with hematoxylin and eosin. Ventricular wall thicknesses were measured under the microscope using an objective micrometer with 0.01-mm ruler markings. The septum and the left and right ventricular walls were each divided into six subregions, and the wall thicknesses of the six subregions from the septum and the left and right ventricular walls of the transgenic mice and their respective counterparts from nontransgenic animals were measured. The relative cell size of the cardiac myocytes was determined in arbitrary units using Image-Pro Plus software. The results were reported as the means Ϯ S.D. Electron Microscopy-The tissue samples of the left ventricular free wall were preserved immediately and then processed following standard procedures in the histology core facility of Beth Israel Deaconess Medical Center. The intracellular volume ratios (%) of mitochondria, myofibrils, and remaining organellae were measured by point counting using a line grid with intercepts every 5 mm (21,22). The measurements from six electron micrographs from each specimen were averaged for each LV region in each group. Data Analysis-The values are expressed as the means Ϯ S.D. The data were analyzed by two independent observers, blind to the transgenic status of the mice. Normality testing was performed on all data, and the nonparametric Mann Whitney U Test was used to determine the differences between the two groups. A p value of less than 0.05 was considered to be statistically significant. Effect of the Double Mutation on SRF Protein Function- The amino acids Thr 159 and Lys 163 are located within an important domain of the SRF protein that is capable of dimerizing, binding to DNA, and recruiting p62TCF proteins (14,15 163 ) on the binding ability of SRF to its cognate binding site, in vitro translated dmSRF and wtSRF proteins were utilized in the gel shift assays. The results showed that the binding activity of the dmSRF protein was much lower than that of the wtSRF protein (Fig. 1A). To demonstrate that dmSRF can inhibit the binding of wtSRF to SRE in vitro, wtSRF protein was cotranslated with dmSRF. EMSA using lysate from the cotranslation reactions revealed that increasing the amount of dmSRF RNA in the cotranslation reaction resulted in progressive reduction of SRF binding activity (Fig. 1B). The effect of the double mutation on the function of SRF protein was further examined in ex vivo experiments. The plasmid containing the promoter of one of the cardiac marker genes, ANF, which contains an SRE site, was used to test this effect. NIH3T3 cells were cotransfected with plasmids containing the ANF promoter and with wtSRF and dmSRF plasmids. The results showed that wtSRF could significantly activate the ANF promoter activity relative to control (p Ͻ 0.01), whereas dmSRF did not significantly increase ANF promoter activity compared with control. To elucidate the mechanism by which dmSRF inhibits SRF function, we sought to test whether dmSRF acts to inhibit the endogenous SRF or increase its activity. We cotransfected dm-SRF with wtSRF and determined the ANF promoter activity. In the presence of the 2-, 4-, and 5-fold amounts of dmSRF relative to the amount of wtSRF, the ability of wtSRF to activate the ANF promoter was significantly and progressively attenuated (p Ͻ 0.01; Fig. 1C). Generation of dmSRF Transgenic Mice-To examine the effect of the dmSRF protein on cardiac function in the postnatal period, the ␣-MHC promoter was employed to generate dmSRF transgenic mice (27). A plasmid construct containing dmSRF cDNA under the transcriptional control of the ␣-MHC promoter was injected into pronuclear stage zygotes using a standard microinjection procedure. Polymerase chain reaction was employed to identify the transgenic founder mice, and only two transgenic founder mice were obtained from a total of 159 mice that were screened (1.2% yield, which is much lower than the normal yield of between 20 -30% at the transgenic facility of this institution). Southern blot analysis revealed that both founder mice A and B had only one single copy of the dmSRF transgene each (Table I). Furthermore, the transgenic mice were found to have elevated levels of SRF mRNA and protein in the heart. However, no overexpression of SRF was found in other muscle, such as skeletal muscle (see Fig. 3C). Early Onset of Cardiomyopathy and Early Mortality in the dmSRF Transgenic Mice-The two male transgenic founder mice were bred with nontransgenic female mice, but the pregnancy rate among the nontransgenic female mice was low. Female mice that were mated with the transgenic founder mouse A gave birth to 8 litters of live birth progeny with litter sizes of 7-10. Only one or two neonates in each litter were found to be transgenic. The transgenic neonatal mice of line A apparently looked normal at birth and during the first few days gained body weight similar to the nontransgenic littermates. They were initially not distinguishable from the nontransgenic littermates. By ϳ7-8 days post-birth, however, their growth appeared to be blunted ( Fig. 2A). All these transgenic mice died between the ages of 9 and 12 days (Fig. 2B). Autopsy revealed decreased body weight, increased heart weight, increased heart/body weight ratio, and four-chamber cardiac dilatation. The average body weight was 3.99 Ϯ 0.15 g in transgenic mice, FIG. 2. Early postnatal cardiac changes and premature death in dmSRF transgenic mice. A, early postnatal growth of the transgenic (Tg) and nontransgenic (NTg) mice. By ϳ7-8 days post-birth, growth of Tg mice was blunted. B, survival of the Tg mice. The F1 generation of Tg mice died between 9 and 12 days after birth. None of them survived. C, nontransgenic and transgenic mice at the age of 12 days. The body weight of the nontransgenic mouse was 5.11 g, and the body weight of the transgenic mouse was 3.94 g. The difference in body weight underlies an apparent difference in body size. D, hearts of the nontransgenic and transgenic mice at the age of 12 days. The heart weight was 28 mg in the nontransgenic mouse and 57 mg in the transgenic mouse. Note the severe dilatation of the transgenic heart. E, heart weight (mg) to body weight (g) ratio in nontransgenic and transgenic littermate mice. Note the increased heart weight/body weight ratio of transgenic compared with nontransgenic mice (15.75 Ϯ 2.10 and 6.39 Ϯ 0.81, respectively). The results are expressed as the means Ϯ S.D. (n ϭ 7, *, p Ͻ 0.001). The transgenic founder mouse of line B was bred with 25 young adult female nontransgenic mice, but only six female mice became pregnant. They delivered 7-11 neonates/litter. However, only one of the F1 neonates was ever found to be transgenic. This transgenic neonatal mouse appeared to be extremely dyspneic and in respiratory distress shortly after birth. It became moribund within the first 24 h. Autopsy revealed that the body weight of this 1-day-old transgenic neonate was 1.36 g, which was similar to that of nontransgenic littermates (1.36 Ϯ 0.051 g, n ϭ 10), whereas the heart weight was 22.9 mg, which was 1.7-fold heavier than that of nontransgenic mice (13.39 Ϯ 0.48 mg, n ϭ 10). The heart weight to body weight ratio in the transgenic neonate was 16.8, which was higher than that of nontransgenic mice (9.74 Ϯ 0.29, n ϭ 10). In an effort to better understand the low birth rate of the live birth transgenic offspring, three sets of mouse embryos at the embryonic development days 12-17 were obtained from line B to examine the presence of the dmSRF transgene. Only one of the 28 mouse embryos in these three litters was positive for the dmSRF transgene. Elevation of dmSRF and Alteration of Gene Expression in the Heart-Northern blotting revealed that the total SRF mRNA level was increased by roughly 4-fold in the ventricle of the transgenic relative to nontransgenic animals. The total SRF protein in the ventricle of the transgenic mice was also higher than that in nontransgenic animals (Fig. 3, A and B). To test whether overexpression of dmSRF could reduce the overall SRF binding activity to SRE in the heart, whole cell lysate from mouse ventricular tissue was employed in EMSAs. The whole cell lysate from transgenic and nontransgenic mice and a [␥-32 P]ATP-labeled SRE oligonucleotide corresponding to the c-fos promoter region were used in the binding reactions. EMSAs revealed that the SRF binding activity was indeed decreased in the ventricle of the transgenic relative to nontransgenic mice (Fig. 3D). Several cardiac genes that contain the SRF-binding site were examined in the dilated hearts of the transgenic mice. Elevated mRNA levels of ANF, SKA, ␤-MHC, and c-fos but decreased mRNA levels of ␣-MHC, MLC2v, cardiac ␣-actin, and SERCA2 were observed (Fig. 3E). Because the ␣-isoform of myosin heavy chain gene itself has a low level of expression in the ventricles during embryogenesis, the possibility that the ␣-MHC promoter drove dmSRF transgene expression during embryogenesis was examined. Northern blotting revealed elevated SRF mRNA level in the heart of the transgenic embryo at embryonic development day 17. In addition, the mRNA levels of ␤-MHC, MLC2v, and cardiac ␣-actin were notably increased in the transgenic compared with nontransgenic embryos. However, the SKA mRNA was hardly detectable in the transgenic embryos (Fig. 3E). The mRNA of the endogenous ␣-MHC isoform was not detected in the hearts of either nontransgenic or transgenic embryos (data not shown). Morphologic Changes-Both atria and ventricles of the transgenic mice displayed substantial dilatation. Cross-sectional views of the ventricles revealed severe dilatation of both ventricular chambers in the transgenic relative to nontransgenic littermate mice (Fig. 4A). In addition, the ventricular free wall and septum of the transgenic mice were thinner than those of nontransgenic littermates. In the 12-day-old transgenic mouse, the left ventricle (LV) wall was 0.070 Ϯ 0.0137 mm thick, which was 2 ⁄3 the thickness of that of nontransgenic littermates (0.102 Ϯ 0.00619, n ϭ 6, p Ͻ 0.001). The septum was 0.053 Ϯ 0.0069 mm thick, which was 1 ⁄2 the thickness of that of nontransgenic littermates (0.103 Ϯ 0.0096, n ϭ 6, p Ͻ 0.001). The right ventricle (RV) wall was 0.0295 Ϯ 0.0072 mm thick, which was 2 ⁄3 the thickness of that of nontransgenic littermates (0.0436 Ϯ 0.0072, n ϭ 6, p Ͻ 0.001) (Fig. 4B). FIG. 3. Total SRF expression and SRF binding activity in nontransgenic (NTg) and transgenic (Tg) mice. A, the level of total SRF mRNA in the ventricles of nontransgenic and transgenic mice. B, the SRF protein level in the ventricles of nontransgenic and transgenic mice. C, the SRF protein level in the skeletal muscle of nontransgenic and transgenic mice. Note that there is no overexpression of SRF in the skeletal muscle of Tg. D, dmSRF overexpression reduced the SRF binding activity in the heart of nontransgenic mice (see lanes C and E). E, the mRNA levels of the cardiac genes were altered in transgenic mouse and transgenic embryo compared with those of nontransgenic mice. In the 1-day-old F1 neonate of the transgenic mouse from line B, the ventricular free wall and septum were much thinner than those of nontransgenic mice. The LV wall was 0.018 Ϯ 0.0037 mm thick, which was 1 ⁄2 the thickness of that of nontransgenic littermates (0.0365 Ϯ 0.0031, n ϭ 6, p Ͻ 0.001). The septum was 0.0334 Ϯ 0.00412 mm thick, which was 2 ⁄3 the thickness of that of nontransgenic littermates (0.0540 Ϯ 0.0028, n ϭ 6, p Ͻ 0.001), and the RV wall was 0.0083 Ϯ

0.0067 mm thick, which was only 1 ⁄3 the thickness of that of nontransgenic littermates (0.0235 Ϯ 0.0067 mm, n ϭ 6, p Ͻ 0.001). Part of the RV wall of the 1-day-old transgenic mouse appeared to be comprised of only a few layers of cardiac muscle cells and was only 0.002 mm thick, or only 1 ⁄10 of the average RV wall thickness of the nontransgenic littermate (0.0235 Ϯ 0.0067 mm) (data not shown). Mitotic cells were visible in the ventricular sections of 1-dayold mice. However, no significant difference of mitotic index was found between the transgenic (0.030 Ϯ 0.018%) and nontransgenic (0.029 Ϯ 0.005%; not significant) mice. No mitotic cells were found in the hearts of 12-day-old mice. Immunohistochemistry confirmed elevated levels of SRF protein in the myocardium of the transgenic relative to nontransgenic mice (data not shown). Masson-Trichome staining revealed minimal interstitial fibrosis in the myocardium of the transgenic mice at the age of 12 days (Fig. 4C), but no fibrosis was visible in the myocardium of the transgenic mouse at the age of 1 day (data not shown). No cardiac myocyte hypertrophy was observed in the ventricles of either of the transgenic lines. Instead, the myocytes in the transgenic mice (24886 Ϯ 8992, arbitrary units) were smaller than those in nontransgenic (33113 Ϯ 8675, arbitrary units; p Ͻ 0.0001) mice (Fig. 4D). Electronic microscopy of the heart of transgenic mouse revealed substantial myofibrillar degeneration as well as mitochondrial swelling and deterioration (Fig. 4E). Furthermore, the intracellular volumes of mitochondria and myofibrils were decreased, whereas the volume of remaining organellae was elevated. DISCUSSION The complex of SRF and ternary complex factor is a downstream target of both the Ras/RhoA and mitogen-activated protein kinase pathways, which regulate cell growth and differentiation in response to a number of extracellular signals (23,24). To test whether functional SRF is required for postnatal myocardial growth and development, a procedure such as targeted disruption of the SRF gene would be the preferred method. However, SRF null allelic (Ϫ/Ϫ) embryos die during embryonic development because of a gastrulation defect and the absence of mesodermal cells that are required for further embryonic development (17). In addition, in these SRF (Ϫ/Ϫ) embryos, expression levels of the SRF target genes, such as c-fos, egr-1, and the ␣-actin genes are either very low or not FIG. 4. Morphological changes in hearts of the transgenic (Tg) mice compared with nontransgenic (NTg) mice. A, cross-section of the ventricles of 12-day-old mice at the level of papillary muscle stained with hematoxylin and eosin. The hearts of transgenic mice were dilated in both RV and LV compared with that of nontransgenic mice. B, ventricular wall thickness. The LV wall of transgenic mice was 0.070 Ϯ 0.0137 mm thick, which was 2 ⁄3 as thick as of that of nontransgenic mice (0.102 Ϯ 0.00619, n ϭ 6, p Ͻ 0.001). The septum of transgenic mice was 0.053 Ϯ 0.0069 mm thick, which was 1 ⁄2 the thickness of that of nontransgenic mice (0.103 Ϯ 0.0096, n ϭ 6, p Ͻ 0.001). The RV wall of transgenic mice was 0.0295 Ϯ 0.0072 mm, which was 2 ⁄3 as thick as of that of nontransgenic mice (0.0436 Ϯ 0.0072, n ϭ 6, p Ͻ 0.001). The data were obtained through direct measurement under the microscope, using an objective micrometer with 0.01-mm ruler markings. C, fibrosis was present in the left ventricle of transgenic mice mouse (see arrows on the right). Masson trichrome staining is shown (magnification of ϫ200). D, the cell size of the cardiac myocytes was determined in arbitrary units using Image-Pro Plus software. The myocytes in transgenic (24886 Ϯ 8992, arbitrary units) were smaller than those in nontransgenic (33113 Ϯ 8675, arbitrary units) mice (n ϭ 100, p Ͻ 0.0001). The results are reported as the means Ϯ S.D. Hematoxylin and eosin staining is shown (magnification of ϫ400). E, electron microscopic view of LV of nontransgenic mice and transgenic mice. There is myofibrillar degeneration and mitochondrial swelling and deterioration in the LV of transgenic mice. In addition, intracellular volumes of mitochondria and myofibrils are decreased in the LV of transgenic mice, and the intracellular volume of the remaining organellae is increased in the LV of transgenic mice. detectable by the sensitive reverse transcriptase-polymerase chain reaction measurement, suggesting that it would be extremely difficult if not impossible to study the function of SRF in the postnatal heart using conventional gene targeting techniques. We therefore chose to utilize a somewhat different approach, by overexpressing a mutant form of SRF driven by the ␣-MHC promoter to determine the role of SRF in the regulation of genes associated with postnatal development of cardiac structure and function. The double mutation of amino acids 159 (Thr 3 Ser) and 163 (Lys 3 Glu) is located within the area of the SRF protein that is important for dimerizing, binding DNA, and recruiting other cofactors such as p62TCF (15). The mutation at these positions apparently reduced the binding ability of dmSRF protein to SRE and made it incapable of activating the ANF promoter, which contains the SRE site. In addition, 2-5-fold excess amount of dmSRF relative to the amount of wtSRF was able to significantly attenuate the ability of wtSRF to activate the ANF promoter activity. Furthermore, increasing the dmSRF/ wtSRF protein ratio in cotranslation experiments progressively attenuated the SRF binding activity. Overexpression of the dmSRF transgene elevated the levels of total SRF mRNA and protein but reduced the total SRF binding activity in the heart of the transgenic animals. These findings demonstrate that, as is the case in vitro, dmSRF was able to compete effectively with mouse endogenous SRF for SRE in vivo and thereby partially blocked the signals regulating cell growth and differentiation that are mediated through SRE via various signaling cascades, such as the Ras/RhoA and mitogen-activated protein kinase pathways (25,26). To study the effect of overexpression of dmSRF on postnatal cardiac function, the ␣-MHC promoter was selected for the transgenic procedure (27). We were able to obtain a few transgenic mice. However, the rate of live births to the transgenic founder mice was extremely low. Only two founder mice were identified of 159 mice that were screened, and each founder mouse had only one single copy of the transgene. Although a few live birth transgenic F1 progeny were eventually obtained, they all developed severe cardiomyopathy soon after birth and died during early postnatal development. The F1 transgenic mice displayed severe dilatation of the atrial and ventricular chambers very early during the postnatal period, indicating that the heart of the dmSRF transgenic mouse was apparently unable to grow normally or maintain normal wall thicknesses in response to increased workload after birth. The blood volume in the circulation increases substantially around the perinatal period because of a rapid increase in body size (28 -30). The normal perinatal heart is able to accommodate the increased workload by rapidly increasing cardiomyocyte size and myofibrillar protein content, thereby enhancing cardiac function (31,32). The hearts of the dmSRF transgenic mice obviously failed to meet this challenge. The ventricular wall of the transgenic mice was much thinner than that of age-matched nontransgenic mice, and the cardiac myocytes in the transgenic mice were significantly smaller than those in nontransgenic mice. Furthermore, intracellular changes that could reduce the energy supply such as extensive mitochondrial swelling and deterioration were noted. Less myofibril volume and myofibril degeneration in the cardiac myocytes might have severely affected the cardiac performance of these animals. The morphological abnormalities in the heart of the transgenic mouse likely made it incapable of pumping blood with sufficient efficiency for survival. It has been reported that SRF is required for a cell to establish proper cell-cell interactions and to maintain normal morphology. However, heterozygous SRF (Ϫ/ϩ) embryonic stem cells apparently form embryonic bodies similar to wild type embryonic bodies. Only SRF (Ϫ/Ϫ) null allelic embryonic stem cells form smaller embryonic bodies with 40 -70% of the size of corresponding SRF (Ϫ/ϩ) and SRF (ϩ/ϩ) embryonic bodies, and with abnormal morphology, which is presumably due to impaired interactions of SRF(Ϫ/Ϫ)embryonic stem cells with each other during differentiation and aggregation (19). It is possible that dmSRF overexpression might make the heart more vulnerable to changes in cardiac morphology under increased workload during the early postnatal period, in part because of possibly altered cell-cell interactions. The remarkable morphologic changes observed in the heart of the transgenic mice, especially in the 1-day-old transgenic mouse, could be a result of dmSRF overexpression before as well as after birth. During embryonic development, ␤-MHC is the predominant isoform of myosin heavy chain that is expressed in the heart, whereas the mRNA level of ␣-MHC is very low. The ␣-MHC promoter that was used in this study contains a small portion of the ␤-MHC gene and the entire portion of 5Ј-end of the ␣-MHC gene and is 5441 base pairs in length. Immediately behind the multiple cloning site of the promoter is a 622-base pair DNA fragment of human growth hormone intron (Hgh) sequence that could enhance the transgene expression (27,33). This promoter is one of the most widely used promoters for transgenic mouse models of postnatal cardiac diseases. In the present study, the transcriptional unit ␣-MHC promoter-dmSRF transgene-Hgh was apparently activated at embryonic development day 17, as the mRNA level of the total SRF was moderately elevated in the heart of transgenic relative to nontransgenic embryos. Consequently, certain SRFregulated genes, such as ␤-MHC, MLC2v, cardiac ␣-actin, and SKA were apparently dysregulated. These alterations could have changed the ratio and components of the cardiac structural proteins during embryogenesis, such as at day embryonic development day 17, and could have made the heart of the transgenic mouse more susceptible to the increased workload that occurred soon after birth. The expression of the MLC2v, cardiac ␣-actin, and SKA genes at end stage cardiomyopathy (day 12) in the transgenic mouse was different from that of the dmSRF transgenic embryo at embryonic development day 17, whereas the expression of ␤-MHC remained the same at both periods. It is possible that different transcriptional regulators are active in the regulation of SRF target genes at these two periods. More work needs to be done to define the promoter region of the SRF target genes and to elucidate the potential role of other transcriptional modulators in the regulation of cardiac development and cardiac gene expression in response to hemodynamic stress. Recently, Shioi et al. (34) reported that cardiac-specific overexpression of constitutively active PI3K results in transgenic mice with larger hearts and hypertrophied myocytes, whereas overexpressing dominant-negative PI3K results in mice with smaller hearts with comparable decreases in myocyte size. This suggests that the PI3K pathway is necessary and sufficient to promote organ growth in mammals (34). PI3K activates gene expression by transactivating SRF-dependent transcription independently of the Rho and ETS ternary complex factor pathways. Also, PI3K-stimulated cell cycle progression requires transactivation of SRF, and dominant-interfering SRF mutants could potentially attenuate mitogen-stimulated cell cycle progression (35). Therefore SRF might be one of the downstream regulators that mediates events or factors that contribute to the determination of cell size. We previously reported that overexpression of the wild type SRF transgene results in cardiac myocyte hypertrophy (16), and in the present study we found that overexpression of the dmSRF resulted in dilated hearts with smaller myocytes. The lack of hypertrophy in the cardiac myocytes of these dmSRF mice raises the possibility that SRF may be important for mediating cardiomyocyte hypertrophy. Taken together, the results of the present study indicate that overexpression of dmSRF in the heart could alter the mRNA levels of SRF-regulated genes and could thereby impair early postnatal cardiac growth and development. These changes result in the early onset of severe, dilated cardiomyopathy, and premature death. It appears that SRF may be important in the regulation of genes that mediate cardiac myocyte hypertrophy. Statistical methods for comparing test positivity rates between countries: Which method should be used and why? The test positivity (TP) rate has emerged as an important metric for gauging the illness burden due to COVID-19. Given the importance of COVID-19 TP rates for understanding COVID-related morbidity, researchers and clinicians have become increasingly interested in comparing TP rates across countries. The statistical methods for performing such comparisons fall into two general categories: frequentist tests and Bayesian methods. Using

data from Our World in Data (ourworldindata.org), we performed comparisons for two prototypical yet disparate pairs of countries: Bolivia versus the United States (large vs. small-to-moderate TP rates), and South Korea vs. Uruguay (two very small TP rates of similar magnitude). Three different statistical procedures were used: two frequentist tests (an asymptotic z-test and the ‘N-1’ chi-square test), and a Bayesian method for comparing two proportions (TP rates are proportions). Results indicated that for the case of large vs. small-to-moderate TP rates (Bolivia versus the United States), the frequentist and Bayesian approaches both indicated that the two rates were substantially different. When the TP rates were very small and of similar magnitude (values of 0.009 and 0.007 for South Korea and Uruguay, respectively), the frequentist tests indicated a highly significant contrast, despite the apparent trivial amount by which the two rates differ. The Bayesian method, in comparison, suggested that the TP rates were practically equivalent—a finding that seems more consistent with the observed data. When TP rates are highly similar in magnitude, frequentist tests can lead to erroneous interpretations. A Bayesian approach, on the other hand, can help ensure more accurate inferences and thereby avoid potential decision errors that could lead to costly public health and policy-related consequences. Introduction The test positivity (TP) rate is defined as the proportion of all tested individuals who test positive for a particular illness or disease. In March of 2020, the World Health Organization [1] emphasized the importance of assessing SARS-CoV-2 test positivity. From that point forward the TP rate has emerged as a critical metric for gauging the illness burden due to COVID-19. TP rates are routinely reported by news media outlets and by many online data repositories, such as Our World in Data (ourworldindata.org). Given the importance of COVID-19 TP rates for understanding COVID-related morbidity between nations, researchers and clinicians have become increasingly interested in comparing TP rates across countries. If country A has a lower TP rate than country B, it would appear that country A has a lower disease burden. Perhaps country A mobilized a more effective public health campaign that emphasized the importance of consistent social distancing and mask use. But how do we determine whether country A truly has a lower TP rate than country B? The question is an important one that has both public health and policy-related ramifications. For example, if decision makers in country B mistakenly conclude that country A has a substantially, or significantly, lower TP rate, then country B might devote considerable resources and enact policy-related changes in order to mimic the apparent success of country A. Such efforts, however, will be in vain because in this example the TP rates for country A and country B are not meaningfully different. How then should a researcher or public health professional decide whether two TP rates are substantially different? One approach that is sometimes used, but is fraught with error and thus not recommended, is the eyeball approach (i.e., unaided human judgment). In a seminal article on this topic, Dawes and colleagues [2] discussed the issue in terms of clinical versus actuarial judgment-with actuarial methods (statistical and mathematical models) consistently outperforming human clinical judgments. Regarding statistical and mathematical approaches, there are several methods that could be used to compare TP rates (which, statistically speaking, are proportions bounded between 0 and 1). These methods can be divided into two groups: classical or frequentist methods and Bayesian methods. Frequentist methods for comparing two proportions, such as TP rates, include the asymptotic z-test [3] and the 'N-1' chisquared test [4]. One limitation of these frequentist tests is that with very large sample sizes, such as those forming the basis of TP rate calculations at the country-wide level, standard error estimates become very small which results in high statistical power and a high probability of rejecting the null hypothesis. When comparing two proportions, the null hypothesis states that there is no difference between the two values. However, when the sample size is very large, rejecting the null means that the researcher concludes there is a substantial/significant difference between the two proportions, even though the actual difference between the two values might be quite small. Related to the above point, a second limitation of frequentist methods is that results generated by these approaches are almost always interpreted from the perspective of Null Hypothesis Significance Testing (NHST). A key limitation of NHST is that the probability value (p-value) resulting from the test statistic measures the probability of the result (or one more impressive) occurring, assuming that the null hypothesis is true. The p-value does not provide direct evidence for or against the alternative hypothesis; that is, whether the two proportions (TP rates) truly differ from each other under the assumption that the null hypothesis is false. Frequentist confidence intervals don't eliminate this concern because they, too, rely on a p-value based interpretation (e.g., if the interval excludes zero, the two proportions are statistically significantly different). A third limitation of frequentist methods concerns their emphasis on dichotomous decision making: either the null hypothesis is supported or it is not supported. In contrast to frequentist-based NHST with its focus on rejecting or retaining the null hypothesis, it is the alternative hypothesis that is almost always of interest to the researcher. Unlike frequentist methods, Bayesian approaches directly evaluate the alternative hypothesis. Before outlining the merits of Bayesian analysis, let's first briefly define the goal of statistical inference: the results obtained in a sample (or samples) are used to make inferences about the population (or populations) from which the sample(s) are drawn. All else being equal, the sample-based statistics are interpreted as the best single estimates of the underlying true population values (i.e., the population parameters). In the present study, the parameter of interest is the true difference between the TP rates of two different countries. In a Bayesian analysis, a key assumption is that because parameters are estimated with error (sampling error, measurement error, etc.), some parameter estimates are better, or more credible, than others. In Bayesian inference, the goal is to ascertain which parameter value, or range of values, is most credible. When performing Bayesian estimation, greater credibility is allocated toward parameter values that are consistent with the data, and less credibility is given to parameter values that are inconsistent with the data. By "data", what is technically meant are the observed sample data combined with, or informed by, previous theory and/or research. In Bayesian estimation the previous theory and/or knowledge is codified in the form of a 'prior distribution'. The prior distribution, which is a type of probability distribution selected by the researcher (see the Method section below for more details), combines with/informs the observed data to give rise to a posterior distribution of parameter estimates. The process is an iterative one: thousands of samples of parameter estimates are drawn and examined, using a procedure called Markov Chain Monte Carlo (MCMC) sampling, to identify the most credible parameters of interest (i.e., the values that are most likely to occur in the population of interest). The collection of credible parameter estimates constitutes the 'posterior distribution'. The purpose of this study was to conduct pairwise, between-country comparisons of COVID-19 TP rates using three different methods: the asymptotic z-test and the 'N-1' chi-squared test (both are widely used frequentist methods), and a Bayesian procedure for comparing proportions. All three methods are designed to compare statistically independent proportions. In this study, because the TP rates are derived from different countries, and because different countries contain nonoverlapping populations, any pair of proportions (TP rates) are considered to be statistically independent. There were two important aims of the analyses: (1) to examine whether the results of the two methodological approaches (frequentist and Bayesian) lead to similar or different inferences, and (2) if the two approaches lead to different inferences, then which approach (frequentist or Bayesian) appears to be more accurate? By "accurate" we mean which approach seems more consistent with the observed between-country data. Method The TP rate data used in this study were obtained from Our World in Data (ourworldindata.org), which is a freely available online data repository. The database is updated frequently and the analyses performed in this study used the version of the database updated on September 18, 2020. All data were processed and analyzed using IBM SPSS and R. The z-score calculator available at www.socscistatistics.com/tests/ztest/ default2.aspx was used to conduct the asymptotic z-test. To perform the 'N-1' chi-squared test, the online MedCalc calculator was used (available at www.medcalc.org/calc/comparison_of_proportions.php). To perform the Bayesian analysis for comparing proportions, the prop.diff.eq R function written by Reza Norouzian was used (available at raw.githubusercontent.com/izeh/i/master/i.r). For the Bayesian analysis, the prior distribution that we used was a Beta (1.2, 1.2) probability distribution. The two values of 1.2 are hyperparameters for the two probabilities being compared (i.e., the two TP rates). The specifics about hyperparameters are not important for our purposes. What is important is that this prior distribution is commonly used when comparing proportions; it is a conservative distribution that depicts most of the proportions between 0 and 1 as being fairly equally likely to occur. The exceptions are the extreme proportions near 0 and 1, which are depicted as occurring notably less frequently. A picture of the Beta (1.2, 1.2) distribution is presented in Fig. 1. In a Bayesian analysis, it is important to select an appropriate prior distribution. As noted above, when comparing two independent proportions, the Beta (1.2, 1.2) distribution is an appropriate choice because this distribution (a) is not biased toward (does not favor) any particular proportion value, and (b) assumes that in most real-world applications extreme proportions are, probabilistically speaking, less likely to occur. When researchers are conducting a study in which prior research has been performed and/or there is strong theory, then the past research or theory can inform the selection of a particular type of prior distribution. In contrast, for scenarios in which past research and compelling theory are lacking, the researcher can select what is known as a non-informative prior distribution. A commonly selected non- Fig. 1. A Beta (1.2, 1.2) probability distribution. informative prior is the Uniform distribution, in which all empirical values are considered to be equally likely to occur. The Beta family of distributions is a popular choice for generating prior distributions because, depending on the particular hyperparameters selected, Beta distributions can assume the characteristics and shapes of many different distributions. To give some examples, Beta (1, 1) is the Uniform distribution, Beta (5, 5) is the Normal distribution, Beta (2, 8) is a Positively Skewed distribution, and Beta (8, 2) is a Negatively Skewed distribution. Graphs of these particular Beta distributions are presented in Fig. 2. Results and discussion Given the large number of countries in the world, there are, obviously, a multitude of pairwise comparisons that could be conducted. Rather than performing a large number of comparisons, it is perhaps instructive to examine a few prototypical yet disparate cases. One such case involves Bolivia and the United States; these two countries have large and small-to-moderate TP rates, respectively. In the 9-18-20 update of the Our World in Data database (ourworldindata.org), the TP values for Bolivia and the United States were 0.337 and 0.045. The discrepancy between these two proportions is quite large. In such cases, where there are large between-country differences, both frequentist and Bayesian analyses will converge on the same conclusion: that there are significant/substantial differences between the two TP rates. A second instructive case involves the scenario in which the TP rates for the two countries are quite similar. An example of this scenario involves the TP rates for South Korea and Uruguay: the values from the 9-18-20 update of the Our World in Data database were 0.009 and 0.007, respectively. These TP rates are very close together and would appear to not meaningfully differ from each other. However, despite this intuition, the p-values from both frequentist tests were highly statistically significant (z = 9.32, p < .00001; χ 2 = 86.94, p < .0001). Due to the extremely large sample sizes (close to 20,000 citizens in South Korea tested positive), both frequentist tests had very high statistical power and easily rejected the null hypothesis of equal TP rates. In contrast, the Bayesian procedure, which generates a distribution (a posterior distribution) of

credible parameter estimates by analyzing the observed data as informed by the prior distribution, indicated that the two TP rates were practically equivalent. To be more precise, the Bayesian results showed that although the difference between the two proportions (TP rates) is not exactly equivalent to zero, the difference can be regarded as being practically equivalent to zero. This result is diametrically opposed to the frequentist findings and aligns more closely with the observation of a 0.002 difference between the two TP rates-an amount that appears to be quite trivial. A graph of the Bayesian results depicting the posterior distribution of credible parameter estimates is presented in Fig. 3. In actuality, this graph contains several quantities of interest that warrant discussion. First, the 95% credible interval is indicated by the solid black line on the x-axis that is situated beneath the posterior distribution curve. As can be seen in the graph, the interval ranged from − 0.16% to − 0.24% (when converted to proportions, these values are − 0.0016 and − 0.0024). This credible interval can be interpreted as follows: there is a 95% probability (or, equivalently, we are 95% certain) that the true difference between the two proportions (the two TP rates) ranges between − 0.0016 and − 0.0024. The posterior parameter estimate, which in this case is the difference between the two TP rates, is labelled in Fig. 3 as Δ (p2-p1) . As indicated in the graph, this estimate was − 0.20% (which, when converted to a proportion, equals − 0.002). This estimate can be interpreted as follows: there is a 95% probability (or, equivalently, we are 95% certain) that the true difference between the two TP rates is − 0.002. Note how straightforward it is to interpret the posterior parameter estimate and accompanying 95% credible interval. In contrast, if one performed a frequentist analysis and thus calculated a 95% confidence interval, the interpretation of the confidence interval would be as follows: if the researcher drew a very large number of random samples (with replacement) from South Korea and Uruguay (samples of the same size as those examined in the actual study), and for each pair of random samples the researcher calculated the difference between the two TP rates and the accompanying 95% confidence interval, then 95% of the 95% confidence intervals would contain the true difference between the two TP rates. We strongly suspect that most researchers would find the interpretation of frequentist confidence intervals to be substantially more convoluted and less illuminating than the interpretation of Bayesian credible intervals. By the way, for all of the Bayesian results discussed above, the signs of the values are irrelevant-whether they are positive or negative is merely a function of which country was labelled group 1 versus group 2 in the Bayesian analysis. There are several reasons as to why a Bayesian analysis provides richer information than frequentist tests. First, unlike NHST procedures, Bayesian approaches directly evaluate the alternative hypothesis, which in the present study is that the two TP rates are truly different. Second, a Bayesian posterior parameter estimate is accompanied by a credible interval, which can be interpreted as a range of values that contain, with a specified degree of probability, the true parameter estimate/true population value [5]. As mentioned in the previous paragraph, we believe that credible intervals are easier to interpret than frequentist confidence intervals. The major take home message from this study is that when TP rates are very similar, performing a Bayesian, rather than frequentist, analysis can avoid a potentially costly false positive decision error. Specifically, the Bayesian approach will, in all likelihood, prevent researchers and policy makers from mistakenly concluding that two TP rates of similar magnitude differ significantly from each other. Our emphasis throughout this article has been to compare and contrast Bayesian and frequentist methods for analyzing TP rates. We did not discuss factors that could influence a test's actual positivity rate. Although a number of factors could be relevant, we believe that two in particular deserve mentionthe sensitivity of the test and the prevalence of the disease in the communities where testing is administered. A test's sensitivity is its ability to correctly identify those individuals infected or with the disease. If a test is highly sensitive, it will have a high accuracy rate when it comes to correctly identifying those infected or with the disease. Recall that a test's positivity rate represents the proportion of all tested individuals who test positive for a particular disease. In any group of individuals who are tested, there will be a certain number/proportion of people who have the disease. A highly sensitive test will be effective at correctly identifying such diseasepositive individuals which, relative to a less sensitive test, will result in a larger proportion of the individuals testing "positive" for the disease in question. In other words, all else being equal, a test with a higher (versus lower) level of sensitivity will result in a higher test positivity rate. Another characteristic of a test is its level of specificity, which is defined as a test's ability to correctly identify those individuals without the infection or disease. Because a test's positivity rate is concerned solely with identifying individuals who have the disease, the concept of specificity is less relevant for TP rates. An interesting issue to consider regarding the SARS-CoV-2 virus concerns the recent variants that have been identified. To the extent that the structures and/or biomolecular properties of the variants affect the sensitivities of SARS-CoV-2 tests, then the test positivity rates of those tests could likewise be affected when the variant viruses are driving infection rates. Regarding the second influential factor that we believe deserves mention (i.e., the prevalence of disease in communities being tested), Usher-Smith and colleagues [6] found that tests developed and evaluated in communities/settings with high disease prevalence may have lower sensitivity when used in lower disease prevalence settings. The lower sensitivity in lower disease prevalence settings implies, by extension, that test positivity rates could also be affected by cross-setting differences in disease prevalence. Finally, regarding the issue of statistical software, the Bayesian R function that we used in this study was easy to implement. However, there are other Bayesian programs for comparing two independent proportions, including the Fully Bayesian Evidence Synthesis online application (see bre-chryst.shinyapps.io/BayesApp/) and the Bayesian First Aid package for R (available at www.sumsar.net/blog/2014/01/ bayesian-first-aid/). There also are other frequentist tests (see [3] for alternatives), but the two that we selected are among the most commonly used. In conclusion, although frequentist hypothesis tests for comparing proportions are widely implemented, their use for comparing between-country TP rates, when those rates are similar in magnitude, can result in erroneous interpretations which could then lead to costly public health and policy-related consequences. Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. CDP-Diacylglycerol Synthetase Coordinates Cell Growth and Fat Storage through Phosphatidylinositol Metabolism and the Insulin Pathway During development, animals usually undergo a rapid growth phase followed by a homeostatic stage when growth has ceased. The increase in cell size and number during the growth phase requires a large amount of lipids; while in the static state, excess lipids are usually stored in adipose tissues in preparation for nutrient-limited conditions. How cells coordinate growth and fat storage is not fully understood. Through a genetic screen we identified Drosophila melanogaster CDP-diacylglycerol synthetase (CDS/CdsA), which diverts phosphatidic acid from triacylglycerol synthesis to phosphatidylinositol (PI) synthesis and coordinates cell growth and fat storage. Loss of CdsA function causes significant accumulation of neutral lipids in many tissues along with reduced cell/organ size. These phenotypes can be traced back to reduced PI levels and, subsequently, low insulin pathway activity. Overexpressing CdsA rescues the fat storage and cell growth phenotypes of insulin pathway mutants, suggesting that CdsA coordinates cell/tissue growth and lipid storage through the insulin pathway. We also revealed that a DAG-to-PE route mediated by the choline/ethanolamine phosphotransferase Bbc may contribute to the growth of fat cells in CdsA RNAi. Introduction During development, animals grow rapidly by both cell proliferation and cell growth. Cell growth is a heavily energy-dependent process and requires large amounts of phospholipids for expansion of cellular membranes and other cellular needs. Fat storage on the other hand is an energy-saving process which stores neutral lipids in the form of triacylglycerol (TAG) for utilization under nutrientlimited conditions such as starvation. In the de novo biosynthetic pathway, fatty acyl-CoA is utilized either for TAG synthesis or for phospholipid production ( Figure 1A), so it is reasonable to propose that growth and fat storage might be balanced during normal development [1]. Indeed, numerous observations support such a balance. For example, Caenorhabditis elegans Tor mutants grow slowly and exhibit excess fat storage [2]. Drosophila melanogaster insulin pathway chico mutants are less than half the size of wild type, but show an almost 2-fold increase in lipid levels [3]. The mechanisms that coordinate cell growth and neutral lipid storage are largely unknown and many intriguing questions remain to be addressed. During development, animals usually have an initial phase of rapid growth, which involves expansion of cell number and size, followed by a homeostatic state, when growth ceases and excess fat is stored in adipose tissue. How is the balance between cell growth and neutral lipid storage regulated in these two different developmental stages? How is the balance regulated in different tissues or cells? For example, how is the balance regulated in the adipocyte, since it both grows large and stores fat? In some disease states, ectopic lipid accumulation in non-adipose tissues such as muscle, pancreas, and liver is often observed [4]. How is the balance regulated in non-adipose tissues? Answering these questions would definitely lead to a significant advance of our knowledge in the fields of both fat storage and cell growth. It is well known that the insulin pathway regulates lipid homeostasis and couples nutritional conditions with systemic growth and metabolism. Besides inhibiting lipolysis, it also promotes fatty acid synthesis through activating the expression of different target genes, including acetyl-CoA carboxylase (ACC), fatty acid synthase (FAS), and sterol regulatory element binding protein (SREBP) [17][18][19]. Recently acyl-CoA synthetase (ACS)/Pudgy has been identified as a direct target of FOXO [20]. Since many of these regulators have been studied during the homeostatic stage, it is not known whether they contribute to the balance of cell growth and neutral lipid storage during the developmental stage. Furthermore, additional regulators or targets of the insulin pathway in balancing cell growth and fat storage remain to be identified. With its exquisite genetic tools and evolutionarily conserved metabolic pathways, Drosophila melanogaster has been accepted as a model for studying lipid metabolism [21][22][23][24]. Through continuous feeding, Drosophila larvae grow rapidly with a nearly 200-fold increase in body mass during the 4-day larval stage. During this time most tissues, such as brain, salivary gland, and imaginal discs, store very little neutral lipid, while the fat body accumulates large amounts of fat [23]. The fat body is a specialized energy storage organ, equivalent to the vertebrate adipose tissue, and most of the dietary fats and de novo-synthesized fats from the intestine are transported to the fat body via lipoproteins in the hemolymph [25]. As in mammals, many questions remain to be answered in Drosophila. How do Drosophila larvae coordinate cell growth and neutral lipid storage in non-adipose tissues? What mechanisms are used in the fat body to balance the storage of fat along with cell growth? In this study, we identified that Drosophila CDP-diacylglycerol synthetase, CdsA, coordinates cell growth and neutral lipid storage through phosphatidylinositol (PI) metabolism and the insulin pathway. We also found that when CdsA is defective, a DAG-to-PE route mediated by the choline/ethanolamine phosphotransferase Bbc may contribute to the growth of fat cells. CdsA broadly affects fat storage We previously reported that mutants of dSeipin, the Drosophila homolog of human lipodystrophy gene Seipin which is important for fat storage and lipid droplet size [26,27], exhibit ectopic lipid droplets in salivary glands, which can be stained by the neutral lipid dyes Nile red or Bodipy [28]. The Drosophila salivary gland is large and easy to manipulate and has been previously used as an in vivo system to study cell death

and autophagy, cell growth and proliferation, and signal transduction, for example in Hedgehog signaling [29]. Hence, we investigated neutral lipid storage regulation in salivary glands by screening for genes which, when knocked down or overexpressed, caused an ectopic lipid droplet phenotype in salivary gland. We used ppl-Gal4, which drives strong gene expression in both larval salivary gland and fat body. The strongest ectopic lipid storage phenotype was caused by CdsA RNAi ( Figure 1B). The phenotype was also verified with an independent CdsA RNAi line ( Figure S1A). In addition, CdsA 1 , a weak allele of CdsA, was previously found to have a dSeipin-like ectopic lipid storage phenotype in the salivary gland [28], demonstrating that CdsA affects fat storage. CdsA is the sole Drosophila CDP diglyceride synthetase (Cds), and diverts phosphatidic acid (PA) from TAG synthesis to the synthesis of cytidine diphosphate diacylglycerol (CDP-DAG), the precursor of two phospholipids, PI and phosphatidylglycerol (PG) ( Figure 1A). According to results from the Drosophila gene expression database FlyAtlas (http://www.flyatlas.org), CdsA is widely expressed in different larval tissues. In adults, CdsA is highly enriched in the eye and is important for the phototransduction pathway [30]. We confirmed the larval expression profile with Q-RT-PCR ( Figure S1B) and a Lac-Z enhancer trap line. The Lac-Z signal was detected in many places, including salivary gland, fat body, proventriculus, hind gut, brain, muscle, Malpighian tubules, and tracheal tubes ( Figure S1C). The broad expression of CdsA and the strong salivary gland fat storage phenotype of CdsA RNAi provide a unique opportunity to address whether other non-adipose tissues have the same capacity as salivary glands to store excess fat. When we drove UAS-CdsA RNAi transgene expression using the ubiquitous tub-Gal4 driver, a strong, fully penetrant phenotype was observed. The level of CdsA transcripts was reduced to about 40% with RNAi. The tub.CdsA RNAi larvae accumulated excess fat in many non-adipose tissues, including salivary gland, proventriculus, hind gut, Malpighian tubules, and trachea ( Figure 1C). Among the tissues examined, the most prominent lipid storage phenotype was found in salivary gland and prothoracic gland, both of which have polyploid cells. The excess lipid accumulation phenotype of tub. CdsA RNAi was reproduced using various tissue-specific Gal4 lines, suggesting that CdsA acts tissue-autonomously (data not shown). To direct measure the accumulation of TAG in non-adipose tissues, we measured the level of TAG from the fat body-removed whole larval samples. Compared to control, tub.CdsA RNAi animals have much higher level of TAG, consistent with the Nile red staining results ( Figure 1B). Together, these results indicate that although the fat body is the main energy reservoir for larvae, neutral lipids can still be stored in significant amounts in many non-adipose tissues when CdsA function is lost. However, it remains to be determined whether excess lipid accumulation in these non-adipose tissues can fulfill the energy reservoir function of the fat body, or causes deterioration of these tissues. CdsA regulates salivary gland fat storage and cell size Beside the fat storage phenotype, we noticed a significantly reduced organ size phenotype in tub.CdsA RNAi larvae. The imaginal discs, salivary gland, and brain are all smaller in tub. CdsA RNAi larvae than controls ( Figure 1D). This phenotype could be due to a decrease in cell size, or cell number, or both. We quantified the cell size and cell number in the salivary gland and found that while the cell number is not changed, the cell size is greatly reduced in tub.CdsA RNAi larvae (data not shown). Author Summary During development, animals undergo a rapid increase in cell size and number, which requires large amounts of lipids, in the form of phospholipids, for the expansion of cell membranes. Once the growth phase ends, excess lipids are usually stored as body fat, in the form of triacylglycerol (TAG), for use when nutrients are limited. How cells coordinate growth and fat storage is not fully understood. By screening for genes that affect lipid storage in the fruitfly Drosophila we discovered that the enzyme CDP-diacylglycerol synthetase (CdsA) coordinates cell growth and fat storage. Phospholipids and TAG have a common precursor, phosphatidic acid, which is diverted by CdsA from TAG synthesis to synthesis of the phospholipid phosphatidylinositol (PI). We also uncovered a link between CdsA and the insulin signaling pathway, which plays a major role in regulating cell and tissue growth. CdsA regulates the level of PI, which modulates insulin pathway activity; insulin pathway activity, in turn, influences the level of CdsA. The lipid metabolism pathways and the insulin signaling pathway are conserved in other animals including humans. Our findings may therefore provide further insights into clinically important imbalances in fat storage such as obesity. Therefore, reduced cell size is the likely cause of the small organ size in CdsA RNAi larvae. To rule out a possible off-target effect of RNAi, we turned to CdsA mutants. CdsA 1 is a weak and viable allele of CdsA and does not cause an obvious small cell size phenotype (data not shown). In the CdsA GS8005 allele, a transposon element inserted into the second exon of CdsA disrupts transcription (Figure 2A). CdsA GS8005 animals died at late embryonic to early larval stages. By RT-PCR, CdsA GS8005 is likely a strong loss-of-function or null allele ( Figure 2B). We generated CdsA GS8005 mutant clones in the salivary gland. Consistent with the RNAi result, CdsA GS8005 mutant salivary gland cells are smaller in size and accumulate large CdsA adds CTP to PA and generates CDP-DAG; Pis catalyzes the donation of the phosphatidyl group from CDP-DAG to inositol and produces PI, which is the precursor of all PI derivatives, such as PIP2 and PIP3; Bbc synthesizes PC and PE from DAG, which can also be converted to TAG by DGAT. (B) (left) RNAi knockdown of CdsA in salivary gland causes massive lipid accumulation. pplG4/+: the ppl-Gal4 driver only; ppl.CdsA RNAi: UAS-CdsA RNAi driven by ppl-Gal4. Blue: DAPI staining for nuclei; red: Nile red staining for neutral lipids. (right) TAG levels of fat body-removed whole larval samples measured by mass spectrometry. The level of TAG is normalized to total phospholipids. (C) CdsA affects fat storage in many tissues. In wandering 3 rd instar tub.CdsA RNAi larvae, massive fat storage in many non-adipose tissues is detected by Nile red staining. Blue: DAPI staining for nuclei; red: Nile red staining for neutral lipids. (D) CdsA RNAi reduces salivary gland, brain, and wing disc tissue size. CdsA was knocked down with ppl-Gal4 in salivary gland and tub-Gal4 in brain and wing disc. Different tissues dissected from wandering 3 rd instar larvae of Gal4 controls and CdsA RNAi were stained by Nile red (red) or DAPI (blue). Relative tissue sizes were quantified in multiple samples (salivary gland: n = 8; brain and wing disc: n = 10) based on the area occupied. Scale bar (B, C, D): 50 mm. doi:10.1371/journal.pgen.1004172.g001 amounts of fat compared to neighboring control cells ( Figure 2C), further demonstrating that CdsA functions cell-autonomously. On average, the size of CdsA GS8005 mutant salivary gland cells is only about 20% of wild-type cells. Together, these results suggest that CdsA plays a key role in balancing fat storage and cell growth. CdsA regulates salivary gland cell growth by affecting PI metabolism and insulin pathway activity Cell growth depends on nutrient conditions and is well-known to be regulated by the insulin signaling pathway. The activity of the insulin pathway is regulated by the PI-derived lipid PIP3. Since CdsA is a key enzyme involved in the production of PI from PA through CDP-DAG ( Figure 1A), and PI is the precursor of PIP, PIP2, and PIP3, we speculated that in CdsA RNAi or mutant animals, the total level of PI is reduced, resulting in a reduction in PIP3, which would lead to low insulin pathway activity and defective cell growth. To test this hypothesis, we first examined the level of PIP3 with a PIP3-specific GFP reporter, tGPH (tubulin-GFP-Pleckstrin homology), in the salivary gland [10]. GFP is recruited to the plasma membrane by binding to PIP3; therefore, the ratio of plasma membrane to cytosol GFP intensity reflects the relative level of PIP3. In controls, the GFP signal is strong in both plasma membranes and nuclei. In CdsA RNAi larvae, the overall GFP intensity is much lower due to unknown reasons and the plasma membrane to cytosol GFP intensity ratio is only 30% of wild type, indicating reduced levels of PIP3 ( Figure 3A). To further probe the activity of the insulin pathway in CdsA RNAi, we also determined the level of phosphorylation of residue S505 of the protein kinase Akt (P-Akt S505 , equivalent to S473 in mammalian Akt), which has often been used as an indicator of insulin pathway activity. The P-Akt S505 level is significantly decreased in CdsA RNAi whole larvae and salivary glands ( Figure 3B). To obtain direct evidence of reduced PI levels, we quantified PI lipids in whole larval and salivary gland samples using mass spectrometry. We found that in both whole larvae and salivary glands, the total level of PI is greatly reduced in CdsA RNAi samples compared to controls and the level of PI in salivary glands is reduced more than ten-fold ( Figure 3C). CdsA RNAi also causes a reduction in the total level of PG, accompanied by an increase in PA ( Figure 3C). In addition, ubiquitous overexpression of CdsA slightly increases the levels of PI and PG in whole larval samples ( Figure 3C). These data are consistent with the role of CdsA in catalyzing the conversion of PA to CDP-DAG, the precursor of PI and PG. Besides, compared to measurable amount of PIP2 in control, the level of PIP2 in CdsA RNAi salivary gland samples is below the threshold level detected by mass spectrometry ( Figure 3D). Moreover, likely due to the small size of the salivary glands, we were unable to measure the abundance of PIP3 in both control and CdsA RNAi samples with mass spectrometry. Instead, we measured the level of PIP3 in whole larval samples using an ELISA method. In CdsA knockdown larvae, the PIP3 level is reduced to less than half of the control ( Figure 3D). Taken together, our results indicate that CdsA influences insulin pathway activity and PI metabolism in the salivary gland. These results predict that blocking de novo PI synthesis in the salivary gland should mimic the CdsA mutant clone phenotype in affecting salivary gland cell growth. Phosphatidylinositol synthase (Pis) is the only enzyme which synthesizes PI from CDP-DAG and inositol ( Figure 1C). Similar to CdsA RNAi salivary glands, Pis RNAi salivary glands are small and accumulate lipid droplets ( Figure 3E). Similar to CdsA RNAi, the level of PI in salivary glands is greatly reduced in Pis RNAi samples compared to controls ( Figure 3F). On the other hand, the level of PG is significantly increased in Pis RNAi ( Figure 3F). We further examined Pis 1 , a likely null mutant of Pis [31]. The Pis 1 mutation is lethal, like CdsA GS8005 , so we generated Pis 1 mutant clones. Interestingly, Pis 1 mutant salivary gland cells do not accumulate large numbers of lipid droplets ( Figure 3G), suggesting that Pis mutations likely increase the level of CDP-DAG, which is converted to PG, without significantly elevating the level of PA. Similar to CdsA GS8005 mutant salivary gland cells, Pis 1 mutant salivary gland cells are significantly smaller than wild-type cells ( Figure 3G). On average, the size of Pis 1 mutant salivary gland cells is only about 25% of wild-type cells. The reduction of cell size in CdsA and Pis mutants is not as strong as in insulin pathway mutants, where salivary gland cell size can be reduced by over 90% [10,32], suggesting that PI from other resources, such as uptaken externally or inherited from mother cells, may partly contribute to salivary gland cell growth. Taking these results together, we conclude that CdsA regulates salivary gland cell growth by affecting PI metabolism and insulin pathway activity. The insulin pathway genetically interacts with CdsA and affects CdsA transcription levels To further reveal the connections between CdsA, PI metabolism, and the insulin pathway, we turned to genetic interaction assays. In the salivary gland genetic screen, we identified two insulin pathway components, PI3K and Akt, in addition to CdsA. Knockdown

of PI3K by dominant-negative PI3K (PI3K DN) or of Akt by RNAi leads to small cell size and ectopic lipid storage phenotypes in the salivary gland, although the lipid storage phenotype is weaker than that of CdsA RNAi ( Figure 4A and S2A). Since both CdsA RNAi and reduced insulin pathway activity have the same phenotypes in the salivary gland, we examined their genetic interaction by gain of insulin pathway activity in CdsA RNAi larvae and vice versa. We found that overexpressing wild-type Akt or constitutively active PI3K (PI3K CA) fully suppressed the small cell size phenotype of CdsA RNAi ( Figure 4B and 4C). Although we can't rule out the possibility that CdsA and insulin pathway act redundantly in a common pathway, these results are consistent with the notion that CdsA acts upstream of PI3K in positively regulating insulin pathway activity. Interestingly, the ectopic lipid accumulation phenotype of CdsA RNAi is partially rescued by overexpression of PI3K CA, but not Akt ( Figure 4B and S2B). We then tested whether overexpression of CdsA can rescue the cell size and ectopic lipid storage phenotypes caused by impaired insulin pathway activity. On its own, CdsA overexpression slightly increases the size of the salivary gland ( Figure 4D and 4E). Interestingly, CdsA overexpression partially, but significantly, rescued the small cell size phenotype of PI3K DN ( Figure 4D and 4E). Moreover, the salivary gland ectopic lipid storage phenotype of PI3K DN can be fully suppressed by CdsA overexpression (Figure 4D and S2B). These results raise the possibility that insulin pathway activity may influence the CdsA level. Since insulin signaling regulates lipid homeostasis mainly at the level of gene transcription and the forkhead box protein dFOXO can transcriptionally activate the upstream target InR via a feedback regulatory loop [33], we asked whether the transcription of CdsA is affected by the insulin pathway. We examined the level of CdsA under conditions which perturb insulin pathway activity. Compared to controls, CdsA transcription levels in PI3K DN or Akt RNAi salivary glands were reduced significantly to 40-50% as assayed by quantitative RT-PCR ( Figure 4F). Conversely, CdsA transcription levels slightly increased when Akt was overexpressed ( Figure 4F). Likewise, loss of FOXO slightly increased the level of CdsA ( Figure 4F). Taken together, these results pinpoint a positive feedback loop between insulin signaling and CdsA in modulating the balance of cell growth and lipid storage. In this positive feedback loop ( Figure 4G), CdsA positively regulates the activity of the insulin pathway through PI, and the insulin pathway affects the transcription of CdsA. Fat body cell growth and neutral lipid storage are not affected by CdsA Besides salivary gland, we also examined the role of CdsA in fat body by both loss-of-function and gain-of-function analyses. To our surprise, fat body-specific CdsA RNAi or CdsA overexpression using ppl-Gal4 or cg-Gal4 did not result in a significant lipid storage phenotype in the fat body ( Figure 5A and data not shown). Consistent with this, the total levels of glycerides are not significantly different in control, CdsA RNAi, and CdsA-overexpressing animals ( Figure 5B). The fat cell size is also unchanged in CdsA RNAi and CdsA-overexpressing animals ( Figure 5A). Although the RNAi efficiency was verified by Q-RT-PCR (data not shown), it is still possible that knockdown by CdsA RNAi is not efficient enough to cause a cell growth or fat storage phenotype in the fat body. To rule out this possibility and to obtain more definitive answers, we analyzed CdsA mutants. In contrast to the small size and massive neutral lipid storage of CdsA GS8005 mutant salivary gland cells ( Figure 2C), CdsA GS8005 mutant fat body cells are normal in size and fat storage ( Figure 5C), which is consistent with the RNAi result. Therefore, we concluded that fat body cell growth and neutral lipid storage are not affected by CdsA under normal conditions. The lack of a neutral lipid storage phenotype in CdsA RNAi fat bodies suggests that PA and probably de novo lipogenesis from fatty acyl-CoA contribute little to final TAG content in the fat body under normal conditions. Indeed, it was reported that DAG is the major form in which lipids are transferred, via lipoproteins, from the intestine to the fat body for storage [25]. Our results are also consistent with a previous finding that unknown mechanism(s) maintain the level of fat body lipid storage within a narrow range [25]. Next, we asked why CdsA does not affect fat body cell growth. As we showed before, CdsA affects PI levels in the salivary gland and subsequently the activity of the insulin pathway ( Figure 3). Is it possible that PI or insulin signaling is not important for fat body cell growth? Several previous reports involving manipulations of insulin pathway activity in fat cells clearly rule out this possibility [10,16,34]. Alternatively, it is possible that the PI level is not reduced in CdsA RNAi or mutant fat cells compared to salivary gland. We therefore examined the levels of PI and PIP3 in CdsA RNAi fat body cells. Similar to salivary gland, the overall tGPH reporter intensity is much lower in CdsA RNAi fat cell compare to control. The ratio of plasma membrane to cytosol GFP signal intensity from the tGPH reporter is reduced to around 50% of wild type in CdsA RNAi fat body cells ( Figure 5D). By lipid content analysis, the level of PA is slightly increased in CdsA RNAi fat bodies ( Figure 5E). However, compared to the .10fold reduction of PI levels in the CdsA RNAi salivary gland, the level of PI is only reduced to 35% of wild type in the CdsA RNAi fat body ( Figure 5E). Interestingly, the P-Akt S505 level is not changed in the CdsA RNAi fat body ( Figure 5F), consistent with the cell size phenotype. These results raise the possibility that although the level of PI, and probably PIP3, is reduced by CdsA RNAi, the reduction is not sufficient to affect Akt phosphorylation or other compensatory mechanism(s) that exist to support fat cell growth. A DAG-to-PE route mediated by the choline/ ethanolamine phosphotransferase Bbc may contribute to fat cell growth in CdsA RNAi larvae To further explore why the fat cell growth is not affected in CdsA mutant, we turned to Pis null mutant and Pis RNAi again. Pis 1 mutant fat body cells are about 30% smaller on average than control cells ( Figure 6A) and in Pis RNAi fat body, the PI level reduced to less than half of the control ( Figure 6B). Expression of dominant negative PI3K or Akt RNAi reduces the size of fat body cells by over 50% ( Figure 6C). Together, these results suggest that de novo-synthesized PI contributes partly to fat cell growth and PI from other origins (i.e. transported from other cells/tissues or inherited from mother cells) may contribute significantly to the growth of fat cells. If so, why is a cell growth phenotype not exhibited by CdsA mutant clones? To examine the CdsA mutant phenotype in a sensitive genetic background, we crossed CdsA RNAi to PI3K or Akt mutants. We found that reducing PI3K or Akt to one copy in the CdsA RNAi background results in a reduction in fat cell size of around 50% ( Figure 6D). These dosage-sensitive interactions suggest that CdsA RNAi mildly impairs fat cell growth. In addition, because CdsA acts one step earlier than Pis in PI biosynthesis ( Figure 1A), it is possible that accumulation of or lack of certain metabolites compensates for the mild growth defect in CdsA mutants. Through lipid measurements, we found that the levels of DAG and PE are significantly increased in CdsA RNAi fat body samples. In fat body, the DAG level is doubled and the PE level is increased by about 50%, while in salivary gland, the PE level is unchanged ( Figure 6E and 6F). In contrast, the levels of DAG and PE are only slightly increased in Pis RNAi fat body ( Figure 6E and 6F). PEBP, a PE binding protein, has been reported to affect Akt phosphorylation, suggesting that PE may be linked to insulin signaling [35]. If the growth compensation is through the DAG-to-PE route, blocking the synthesis of PE from DAG in CdsA RNAi animals might lead to a more obvious fat cell growth phenotype. Choline/ethanolamine phosphotransferase (Cept) Bbc is a key enzyme in Drosophila that converts DAG to PE ( Figure 1A) [36]. While bbc RNAi did not affect the P-Akt S505 level ( Figure 6G) and did not obviously affect the fat cell size ( Figure 6C), CdsA and bbc double RNAi exhibits a synergistic effect: the fat cell size is drastically decreased to a level comparable to Akt or PI3K RNAi ( Figure 6C). Moreover, the fat body P-Akt S505 level is greatly reduced by CdsA and bbc double RNAi ( Figure 6G). In contrast, the salivary gland cell size of CdsA and bbc double RNAi is comparable to CdsA single RNAi ( Figure S3). Together, these results suggest that the elevated conversion of DAG to PE mediated by the choline/ethanolamine phosphotransferase Bbc may contribute to fat cell growth in CdsA RNAi larvae. Discussion The regulation of cell growth in terms of both cell size and cell number has been studied intensively, and is known to involve many key core signaling pathways such as the insulin pathway, the Hippo pathway, and the mTOR pathway, as well as cross-talk between these pathways [5,7,37]. The downstream targets that execute and coordinate various specific aspects of cell growth are just starting to be elucidated. In this study we have revealed the intrinsic connections between CDP-diacylglycerol synthetase, PI metabolism, and insulin pathway activity in coordinating cell growth and neutral lipid storage. Lipid storage is a basic function of a cell. Within an organ, glycerol and fatty acyl-CoA is the precursor of all phospholipids, which are mainly utilized within the cell, and glycerolipids, which are largely destined for neutral lipid storage. Therefore, under normal developmental conditions when resources are not unlimited, a balance must be achieved between cell growth and storage (fast growth and less storage, or slow growth and increased storage). Moreover, at the systematic level, the tissue specificity of insulin action and the lipid flow between different tissues make this balance more complicated [38]. For non-adipose tissues, fast cell growth and low levels of storage are beneficial for maximizing growth potential. However, the balance encounters a problem in adipose tissue because both growth and storage are required. Several predictions can be made here. First, the key enzymes that act at the branch points of phospholipid and TAG synthesis are good candidates for controlling the balance between growth and storage. Second, the factors that control the difference in balance regulation between adipose tissue and non-adipose tissue likely sit after the regulatory points. Alternatively, adipose tissue may have two developmental phases, a growing phase and a fat storage phase. The timing of the switch between these two phases would determine the final size and storage capacity of adipose tissue. Third, excess nutrients may saturate the balance mechanism, leading to overload in both directions. Indeed, our results depict a scenario in which CDP-diacylglycerol synthetase coordinates cell growth and fat storage by partitioning the flow of fatty acyl-CoA between PI synthesis and TAG synthesis. Moreover, our results reveal different contributions of de novo-synthesized PI and PI derived from external sources to the growth of fat cells and salivary gland cells. De novo-synthesized PI contributes most to the growth of salivary gland cells, while fat body cell growth is mainly stimulated by externally-derived PI ( Figure 6H). Therefore, the role of CdsA in fat cell growth is only revealed under conditions in which growth is mildly compromised. Besides that, an external supply of DAG, most likely from the intestine, may bypass the CdsA-gated coordination, allowing adipose tissue (fat body) cells to both grow large and store fat ( Figure 6H). Consistent with this, it was previously reported that intestine-derived lipids, including DAG, are mainly transported to the fat body via lipoproteins, while relatively low levels of lipids are transported to non-adipose tissues either directly from the intestine or indirectly via the fat body [25,39]. By responding to environmental and nutritional cues, the insulin pathway regulates cell growth [5,9]. Through genetic

analyses, GFP reporter assays, and lipidomics, we have presented evidence for a positive feedback loop between the insulin pathway and CdsA levels ( Figure 4G). In this loop, lowering insulin pathway activity reduces CdsA transcription, leading to less PI production, which further decreases insulin pathway activity. There are numerous previous findings of FOXO as a transcriptional activator [20,33,40], but dFOXO appears to be a repressor of CdsA transcription. Microarray and ChIP studies have identified [41]. At this point, we do not know whether dFOXO represses CdsA directly or indirectly. The widespread cell growth and fat storage phenotype caused by CdsA RNAi suggests that the insulin pathway-CdsA feedback loop operates widely in various tissues. In addition, it shows that the fat storage capacity of most non-adipose tissues is enormous. The insulin pathway-CdsA feedback regulation allows cells to take full advantage of the environmental conditions to grow fast and stop growth quickly when necessary. For example, in a nutrient-rich environment, animals that can grow quickly have a tremendous competitive advantage over slow-growing ones. In nutrient-poor conditions, the ability to quickly convert growth to fat storage is beneficial for animals to survive through harsh times. This is not the first time that feedback regulation of the insulin pathway has been reported [33], suggesting that it might be a common mechanism to augment pathway sensitivity and ensure a rapid response to changes in nutrient conditions. Regulation of the insulin pathway by lipids is already known to occur at the points of PIP to PIP2 and PIP2 to PIP3 conversion. Compared to regulation by PI3K and PTEN, CdsA provides a new layer of control at the level of available lipids. Together with other pathways regulated by available nutrients/small molecules, such as the mTOR pathway, which responds to amino acid levels, and the AMPK pathway, which responds to ATP levels [42,43], this new regulatory mechanism helps cells to cope with diverse environmental changes. The insulin-CdsA feedback loop is important during cell development. When cells reach the homeostasis phase, the growth-favored loop must be switched towards fat storage. In the future, it will be interesting to determine what triggers the switch. In addition, the existence of the positive feedback loop suggests that excess nutrients will overload the fatty acyl-CoA partitioning mechanism and disrupt the balance. Therefore, impairing the insulin-CdsA feedback loop by genetic mutations or environmental conditions may cause an imbalance in fat storage, resulting in metabolic diseases such as obesity and related disorders. Re-establishing the feedback loop could be a way of treating these disorders. Drosophila stocks and husbandry Drosophila stocks were maintained in standard cornmeal food. w 1118 was used as wild type. All RNAi stocks were obtained from the VDRC or NIG RNAi stock centers. ppl-Gal4 was generously provided by Dr. Pierre Leopold. The expression pattern of ppl-Gal4 was confirmed with UAS-GFP. Other stocks used were: tub- RNAi screen and genetic analysis ppl-Gal4 virgin females were crossed to males harboring UAS-RNAi inserts to knock down target gene expression in the larval salivary gland. Wandering 3 rd instar larvae harboring both the Gal4 and the UAS-RNAi transgene were dissected and mounted on glass slides with PBS. For visualization, specimens were examined with a Zeiss DIC microscope. For GAL4/UAS experiments, flies were grown at 29uC to allow for maximal GAL4 activity. For tissue or clonal analysis, unless stated otherwise, all larvae were dissected at the wandering 3 rd instar stage. Salivary gland clones were generated by heat shock at 37uC for 1 hr immediately after collection of eggs for 4 hr. Fat body clones were induced 8 hr after egg laying for 1 hr at 37uC. Staining and microscopy After crossing ppl-Gal4 with RNAi lines, wandering 3 rd instar larval progeny were dissected and stained with Nile red or Bodipy, which are fluorescent dyes that specifically mark neutral lipids. For lipid droplet staining, larvae were dissected in PBS and fixed in 4% paraformaldehyde for 30 min at room temperature. Tissues were then rinsed twice with PBS, incubated for 30 min in a 1:2500 dilution with PBS of 0.5 mg/ml Nile red (Sigma), and then rinsed twice with distilled water. Stained samples were mounted in 80% glycerol for photo-taking. For b-galactosidase staining, dissected tissues were fixed for 2 min in 0.5% glutaraldehyde on ice. The tissues were washed three times for 10 min each in PBS containing 0.3% Triton X-100 (PBT) and then placed in X-gal staining solution (0.2% X-gal, 1 mM MgCl 2 , 5 mM K 4 [Fe(CN) 6 ], 5 mM K 3 [Fe(CN) 6 ]) overnight at 37uC. All images were taken using a Nikon confocal scope or Zeiss fluorescent scope. Quantification of mutant clone areas was performed by calculating the size of the area occupied by mutant clones versus wild-type cells. Quantification of the size of whole salivary gland, brain, and wing disc or the size of fat body cells was performed by measuring the area occupied by these organs/cells. To quantify lipid storage in salivary glands, the Nile red positive areas of salivary glands per staining of fat bodies in larvae with loss of function of insulin pathway components or silencing of CdsA and bbc. Silencing CdsA or bbc alone does not affect fat body cell size, whereas reduced insulin pathway activity dramatically decreases the cell size. Double silencing of CdsA and bbc also significantly reduces fat body cell size. Histogram: n$94 for each genotype. Blue: DAPI staining for nuclei; red: Nile red staining for neutral lipids. (D) Removing one copy of PI3K or Akt in a ppl.CdsA RNAi background is sufficient to reduce fat body cell size. Histogram: n$80 for each genotype. Blue: DAPI staining for nuclei; red: Nile red staining for neutral lipids. (E, F) DAG and PE levels obtained by lipid measurements of whole larva, salivary gland or fat body of wandering 3 rd instar larvae of Gal4 control, CdsA RNAi, CdsA overexpression and Pis RNAi. CdsA or Pis was knocked down with ppl-Gal4 in salivary gland and fat body and tub-Gal4 in whole larva. Assays were done in triplicate. Note that the PE level in the CdsA RNAi fat body sample was 47.6% higher than the Gal4 control. The levels of DAG and PE are normalized to total phospholipids. (G) Double silencing of CdsA and bbc decreases larval fat body Akt phosphorylation at Ser505. Total Akt and phosphorylated Akt (Ser505) levels were detected by western blotting. a-tubulin (Tub) was used as a loading control. Average and standard deviation of relative band intensity ratio of p-Akt versus total Akt from three replicates is indicated at the top after normalization. The Western blot result from one experiment is shown here. (H) Schematic model depicting the underlying mechanisms of different phenotypes observed in CdsA and Pis mutants. In wild type, de novo-synthesized PI contributes most to the growth of salivary gland cells, while PI from an external source along with DAG, probably from the intestine, contributes most to the growth and fat storage of fat body cells. Both CdsA and Pis mutations lead to ,80% growth reduction in salivary gland due to the loss of de novo-synthesized PI. In fat body, the loss of de novo-synthesized PI in Pis mutants results in ,30% reduction of cell growth, while the increase in PE (marked in green) in CdsA mutants compensates for the loss of de novo-synthesized PI. genotype were measured and normalized to the whole cell area as previously described [25]. The average lipid storage in salivary glands of CdsA RNAi was set as 1. For tGPH signal quantification, membrane-to-cytoplasmic GFP ratios were calculated from measurements of mean pixel intensities within equal areas of membrane versus cytoplasm. All measurements were done using Nikon Br analysis software. Quantitative RT-PCR Total RNA was isolated using an RNeasy kit (Qiagen) according to the manufacturer's protocol. To quantify CdsA transcript levels in tub.CdsA RNAi and tub/+ controls, three groups of 5 larvae per genotype were collected at the wandering 3 rd instar larval stage. To quantify CdsA transcript levels in different tissues such as salivary gland, fat body, brain, and gut, three groups of 30 tissues per genotype were dissected and collected from wandering 3 rd instar larvae. To quantify CdsA transcript levels in larval salivary gland with loss of function or overexpression of insulin pathway components, salivary glands were dissected and collected in three groups of 50 pairs per genotype from wandering 3 rd instar larvae. For each sample, 3 mg total RNA was used to synthesize single-stranded cDNA using a SuperScript II reverse transcriptase kit (Invitrogen). Quantitative RT-PCR (qRT-PCR) experiments (Agilent Stratagene Mx3000P system, SYBR Green PCR mix) were performed using specific primer pairs (59 GCAATGATGT-CATGGCGTAC 39 and 59 AAATCCAGAGGCAAAGAAGC 39). The level of each transcript was normalized to rp49 in the same sample. All qRT-PCR experiments were repeated at least three times. PIP3 ELISA assay The extraction and measurement of PIP3 were performed using the PIP3 Mass ELISA Kit (Echelon), according to the manufacturer's instructions. For extraction of PIP3, four groups of 20 larvae per genotype were collected at the wandering 3 rd instar larval stage. The level of PIP3 was normalized to larval dry mass in each group. The PIP3 ELISA assay was repeated four times. Total body glycerides were measured by homogenizing six groups of 5 male larvae per genotype in 100 ml PBS containing 0.05% Tween 20 (PBST). The homogenate was immediately heattreated at 70uC for 10 min. Then, 20 ml samples were incubated with 200 ml Triglyceride Reagent (Triglycerides Kit, Zhong-ShengBeiKong) for 10 min at 37uC and the absorbance at 540 nm was measured using a spectrophotometer. The total protein content of the same samples was measured by Bradford assay (Sigma). Glyceride levels were normalized to total protein amounts in each sample. Characterization and Differentiation of Iron Status in Anemic Very Low Birth Weight Infants Using a Diagnostic Nomogram Background In the early weeks of life, very low birth weight (VLBW) infants experience intense laboratory blood sampling leading to clinically significant anemia and the need for red blood cell transfusion. Although controversial, treatment with recombinant human erythropoietin (EPO) and iron has been recommended to stimulate erythropoiesis; optimal dosing of EPO and iron is still uncertain. Objectives: To assess the validity of a four-quadrant diagnostic plot of iron availability (ferritin index) versus iron demand for erythropoiesis (reticulocyte hemoglobin content, CHr) for differentiating iron status in anemic VLBW infants. Methods Study subjects were enrolled in a previously reported randomized controlled trial of clinically stable VLBW infants <31 weeks’ gestation and <1,300 g at birth to receive 18 days of treatment with: group 1: oral iron; group 2: EPO + oral iron, and group 3: EPO + intravenous + oral iron. Results At the end of treatment the ferritin index was significantly higher in both EPO groups compared to the control group. By day 18, CHr of the control group declined into the quadrant of the diagnostic plot characteristic of functional iron deficiency and anemia of chronic disease. Both EPO groups ended in the quadrants that are characteristic for latent iron deficiency and iron deficiency anemia, respectively. Conclusions The diagnostic plot for differentiating anemia in VLBW infants may be an informative, clinically useful tool for iron status assessment under different physiologic and therapeutic erythropoietic states. Larger additional studies in difficult patient populations are needed before the clinical utility of this diagnostic procedure can be unequivocally confirmed. controversial [6][7][8] . This controversy has in part focused on conflicting strategies concerning what constitutes optimal iron supplementation during r-HuEPO treatment of anemic infants, i.e. identification of an r-HuEPO and iron-dosing schedule that provides adequate available iron to meet the needs of accelerated erythropoiesis while avoiding the risk of iron toxicity. Unfortunately, characterization of iron-deficient states in infants is challenging. In addition to traditional parameters such as complete blood count and biochemical markers of iron metabolism, e.g., serum transferrin, transferrin saturation and serum ferritin, more recent parameters used in defining iron status and iron-deficient states include zinc erythrocyte protoporphyrin, soluble transferrin receptor (sTfR), and the ferritin index (FI) [9] . Zinc erythrocyte protoporphyrin is non-specific since its levels increase either when erythropoiesis is stimulated or when early functional iron deficiency (FID) is present. The interpretation of both sTfR and the FI is useful in the evaluation of iron status and erythropoietic activity as shown by Dimitriou et al. [9] in children and neonates.

High sTfR levels indicate increased iron demand when iron stores are being depleted. Consequently the measurement of sTfR is helpful for investigating the pathophysiology of anemia by allowing the evaluation of the adequacy of marrow proliferative capacity and the erythropoietic response to various therapies [10] . Recently, the combination of hemoglobin content of reticulocytes (CHr), an early indicator of iron demand, and FI have been investigated by Thomas et al. [11,12] as promising diagnostic tools for classifying and monitoring of iron deficiency in the anemia of chronic disease (ACD) in adult cancer patients. Thomas et al. [11] speculated that the diagnostic plot might also be a useful tool in assessing iron status in a broad spectrum of other anemias, including classical iron deficiency, end-stage renal failure and anemia of infection and inflammation. These investigators have devised a promising method for describing the relationship of FID (represented by CHr) to the availability of iron for erythropoiesis (represented by FI). These authors used a plot of these two variables to describe this relationship. Doing so offers the advantage of classifying iron deficiency into four discrete states ( fig. 1 ): quadrant 1 (Q1) indicates full or replete iron stores, quadrant 2 (Q2) indicates unsatisfactory iron stores prior to iron-deficient erythropoiesis -latent iron deficiency, quadrant 3 (Q3) indicates depletion of iron storages and decreased hemoglobinization of red cellsiron deficiency anemia, and quadrant 4 (Q4) indicates full or replete iron stores but inappropriate iron metabolism -FID. This plot offers the potential for monitoring erythropoietic activity, FID and the adequacy of iron availability for EPO-stimulated RBC production. [11] . a Therapeutic implications for the treatment of different phases of iron deficiency according to Thomas et al. [11] . The aim of the study was to apply the diagnostic plot to the results of a previous study of r-HuEPO and irontreated VLBW infants to assess whether this diagnostic tool is suitable for differentiating iron-deficient stages in AOP [3] . We hypothesized that the information derived has potential for being useful in therapeutic decisionmaking regarding the need for iron replacement and/or r-HuEPO therapy in anemic infants. Participants and Study Design Using the diagnostic plot to assess iron status relative to r-HuEPO and iron therapy, we retrospectively analyzed previously described data from a randomized controlled trial from clinically stable VLBW infants ! 31 weeks of gestation and ! 1,300 g birth weight [3] . In brief, inclusion criteria postnatal age 1 1 week, hematocrit 1 30%, phlebotomy loss averaging ! 1.0 ml/day over the previous 3 days, enteral caloric intake 1 20 kcal/kg/day and stable respiratory status defined as mean oxygen saturation 1 90% in a fraction of inspired oxygen ^ 0.40 for 1 48 h regardless of ventilation. Infants excluded were those who received RBC transfusions 3 days before study entry or during the study, those with an active, culture-proven infection, a major malformation, electroencephalographic-proven seizures, hypertension, intraventricular hemorrhage 6 grade III, immune hemolytic disease, necrotizing enterocolitis, or surgery. Among the study participants who began the study, 3 infants with sepsis or sepsis-like episodes, 2 with necrotizing enterocolitis, 2 who received RBC transfusions were excluded. There were no statistical differences among the three study groups for common neonatal morbidities including respiratory distress syndrome (p = 0.65), bronchopulmonary dysplasia (oxygen dependency at 36 weeks of postconceptional age; p = 0.31), patent ductus arteriosus (p = 0.60) and retinopathy of prematurity (p = 0.32). There were also no significant differences identified among the three study groups for birth weight or in age, body weight, or phlebotomy loss at the prior to and during the treatment period ( table 1 ). The 21-day study consisted of a 3-day run-in baseline period during which all subjects received 9 mg/kg/day of iron polymaltose complex (IPC; Ferrum Hausmann Syrup, Vifor International, St. Gallen, Switzerland). To remain in the study, subjects were not allowed to receive treatment with RBC transfusion. Study participants were randomized to receive 18 days of oral iron alone (group 1: oral iron; control group), r-HuEPO + oral iron (group 2: EPO + oral iron), or r-HuEPO + intravenous (IV) iron (group 3: EPO + IV iron) treatment beginning after a 3-day run-in period ( table 2 ). Analytical Procedures Hematological parameters and the iron status (sTfR, ferritin) were determined on peripheral venous or arterial blood drawn on treatment days -3, 0, 4, 11, and 18. Hematological parameters were analyzed immediately. Serum and plasma were aliquoted and stored at -80 ° C for later analysis. Blood counts, reticulocyte counts, RBC and CHr indices were performed on fresh whole blood samples by flow cytometry (Technicon H.3 Autoanalyzer, Statistical Analyses Within-group comparisons were performed by paired-samples t test (two-tailed). Between-group comparisons were performed by independent-samples t test and ANOVA with Bonferroni posthoc multiple comparisons test for normal distributed data or Mann-Whitney U test for non-parametric sample distribution. Results are expressed as mean 8 SEM. p ! 0.05 was considered statistically significant. Calculations were done with SPSS 14.0 (SPSS, Inc., Chicago, Ill., USA). Results The age of subjects at study entry was 23 8 2.9 days. Demographic and clinical features of the three study groups including gestational age and birth weight have been described previously and did not differ among the three groups at treatment day -3 or baseline. On day 0, none of the hematological or serological parameters demonstrated significance between group differences ( table 3 ) [3] . As anticipated, subsequent hematological and serological parameters demonstrated significant differences among the groups according to their therapeutic assignment ( fig. 2 a-c). Posttreatment changes in sTfR concentration were significantly higher in group 2 (EPO + oral iron) and group 3 (EPO + IV iron) compared to group 1 (oral iron; control group) (p ! 0.01; fig. 2 a). The sTfR values increased significantly from baseline in both EPO groups on days 11 and 18 compared to the control group. Similarly, there was a posttreatment increase in the FI for both EPO groups starting at day 11 (p ! 0.05) compared to the control group ( fig. 2 b). A significant increase in CHr occurred at day 18 (p ! 0.05) in group 2, while in group 3 only a trend toward higher CHr values was observed which did not quite achieve statistical significance (p = 0.059) ( fig. 2 c). Beginning on treatment day -3 and continuing to day 18 at the end of treatment, serial iron status laboratory data for each group were plotted (CHr vs. FI; fig. 3 ) according to the method of Thomas et al. [11] . As the study period progressed, the pattern of laboratory findings of group 1 infants differed significantly in the course of CHr and FI in comparison to both r-HuEPO-treated groups. For group 1 infants, CHr by the end of study declined into Q4, the quadrant is indicative of functional iron-deficient anemia, e.g., for ACD. In contrast, the two r-HuEPO-treated groups moved to the right, i.e. an increase in FI was observed. Although the courses of group 2 and group 3 paralleled one another over time beginning with the initiation of r-HuEPO treatment, on day 18, group 2 infants (EPO and oral iron) ended up in Q2 (latent iron deficiency) while group 3 IV EPO + IV irontreated infants ended in Q3 (iron-deficient anemia). Discussion The aim of the present study was to examine the clinical utility of using a diagnostic plot of Thomas et al. [11] to assess iron availability for erythropoiesis in VLBW infants with AOP as an informative tool for clinicians. The data analyzed were derived from a previous study of VLBW premature infants treated with two different r-HuEPO and iron-dosing regimens used in the treatment of AOP. This diagnostic plot utilizes two sensitive and specific parameters of the adequacy of iron for erythropoiesis for the x-and y-axes: FI and CHr. The CHr and the FI were used as the key hematological indicators for the intensity of the demand for iron and the adequacy of iron supply, respectively. The plot was used to identify the initial and subsequent erythropoietic and iron status of the three study groups that differed in the r-HuEPO and oral and parenteral iron supplementation each received. CHr has been used in the diagnosis of different iron-deficient states and in the response to iron and/or r-HuEPO treatment [11,[13][14][15][16][17] . Importantly, CHr provides a direct assessment of the incorporation of iron into erythrocyte hemoglobin and thus serves as a direct indicator of the recent functional availability of iron for the erythron [13,18] . sTfR and zinc erythrocyte protoporphyrin were proposed as useful indicators for iron deficiency [19] , because both parameters are directly associated with erythropoietic activity [10,20,21] . Both have therefore been used to help identifying iron deficiency states in children [9,22,23] . The acute-phase reactant, ferritin, has primarily been used as an indicator of iron storage [24][25][26] . In the absence of inflammatory disease, the sTfR to ferritin ratio (FI) is inversely correlated with iron stores in various anemic states, including the ACD and FID [11] . In the previous study from which the present data were derived, we investigated the efficacy and safety of high doses of EPO/iron therapy on the AOP [3] . In our application of the diagnostic plot of Thomas et al. [11] to the present data beginning on day -3, all three groups started in the upper left quadrant (Q1) of the plot ( fig. 3 ), i.e. indicating that baseline iron status of all three groups were adequate relative to their erythropoietic demand. We speculate that the control group became increasingly anemic as a result of multiple factors impacting iron status including: (1) low endogenous EPO production, (2) diminished iron stores as a result of rapid growth, and (3) on going phlebotomy loss. We further speculated that the later two of these contributed to the decrease of CHr ob-served in the control group despite receiving oral iron supplementation. Without administration of EPO, the course of the control group fell into the lower left quadrant (Q4) of the diagnostic plot and remained there until the end of the treatment period on day 18. Q4 is characteristic of anemic conditions associated with FID/ACD. These results suggest that the group 1 control subjects receiving oral iron supplementation alone experienced both r-HuEPO and iron deficiency by the end of the study. This unanticipated finding that control group infants moved from their initial position in iron sufficient Q1 to their final position at the end of the study in Q4, characteristic of anemic conditions associated with FID/ACD, is consistent with the findings of Thomas et al. [11] . On this point, we speculate that because the control group did not receive r-HuEPO treatment, iron was not incorporated in adequate amounts by reticulocytes. As a result, mean CHr values for control subjects fell below the 28 pg cutoff value for Q4. Thus, although iron stores were replete based on ferritin at treatment day 18 1 200 l/l and a FI ! 1.5, complete hemoglobinization did not occur. As such, treatment with r-Hu-EPO and additional iron supplementation might be considered as possible therapeutic considerations. In contrast to the control group, both r-HuEPO-treated groups had moved into quadrant Q2 by treatment day 11. This can be interpreted as a successful response to the r-HuEPO therapy with a pronounced stimulation of erythropoiesis by the second week of treatment. No significant increase in CHr was observed in the EPO + IV iron group (p = 0.059) by day 18, while in the EPO + oral iron group a slight, but significant, increase in CHr was observed compared to the control group (p ! 0.05). Based on our diagnostic plot data and the recent review of iron supplementation in preterm infants by of Ridley et al. [27] , it can be hypothesized that oral iron supplementation together with EPO therapy is adequate clinical therapy for preterm infants during their period of AOP which is in accordance with the considerations of Ridley et al. [27] . By day 18, the course of group 2 ended in Q2, reflecting a latent iron deficiency. In contrast, the course of group 3 ended in Q3, indicative of classic iron deficiency anemia. Additional trials are needed to assess whether r-HuEPO administration might be stopped at this point and if iron therapy (oral and/or IV iron) alone

is sufficient as a therapeutic intervention for patients entering Q2 or Q4 of the diagnostic plot. The diagnostic plot has served as a clinically useful tool in evaluating iron status of patients with renal anemia in which CHr provides helpful information regarding FID in dialysis patients receiving erythropoiesis-stimulating agents [28] . The diagnostic plot also appears helpful in monitoring iron therapy in cancer-related anemia [11] . Treatment of preterm infants with EPO and iron is one therapeutic approach for AOP exacerbated by phlebotomy (and therefore iron) loss. The diagnostic plot might also be helpful as a supporting diagnostic tool for clinicians seeking to identify the optimal r-HuEPO and iron-dosing regimens. The influence of supplementation with vitamin B 12 and folic acid with or without r-HuEPO treatment and with IV iron to reduce the need for RBC transfusion [29] is another example of a clinical situation that would be of interest to apply the diagnostic plot to determine if a positive effect in the stimulation of erythropoiesis is observed. A clinical study by Worthington-White et al. [30] indicated that premature infants require additionally vitamin B 12 and folate, and it could be shown that the severity of anemia was reduced. In their study, infants treated with vitamin B 12 alone or combined with folate had higher hemoglobin values than the untreated (p ! 0.0005) or solely folate-treated (p ! 0.01) groups. In future studies, it will be important to include untreated infants starting at birth when both ferritin and sTfR levels are normally (i.e., in the absence of r-HuEPO and IV iron treatment and before receiving RBC transfusions) at their highest. Conclusion Our results indicate that the diagnostic plot developed by Thomas et al. [11] offers promise as a useful tool for assessing iron status in anemic preterm infants. In its reliance on laboratory determinations performed in most hospital laboratories, the diagnostic plot has the potential for providing a readily available, real-time indicator of iron status. This has particular relevance for anemic, rapidly growing low birth weight infants during treatment with erythropoiesis-stimulating factors which result in different degrees of iron demand for hemoglobin synthesis. Additional studies carried out in a robust spectrum of clinical circumstances and other neonatal patient groups are needed before the clinical utility of this new diagnostic procedure can be unequivocally relied on. Ataxin-7 and Non-stop coordinate SCAR protein levels, subcellular localization, and actin cytoskeleton organization Atxn7, a subunit of SAGA chromatin remodeling complex, is subject to polyglutamine expansion at the amino terminus, causing spinocerebellar ataxia type 7 (SCA7), a progressive retinal and neurodegenerative disease. Within SAGA, the Atxn7 amino terminus anchors Non-stop, a deubiquitinase, to the complex. To understand the scope of Atxn7-dependent regulation of Non-stop, substrates of the deubiquitinase were sought. This revealed Non-stop, dissociated from Atxn7, interacts with Arp2/3 and WAVE regulatory complexes (WRC), which control actin cytoskeleton assembly. There, Non-stop countered polyubiquitination and proteasomal degradation of WRC subunit SCAR. Dependent on conserved WRC interacting receptor sequences (WIRS), Non-stop augmentation increased protein levels, and directed subcellular localization, of SCAR, decreasing cell area and number of protrusions. In vivo, heterozygous mutation of SCAR did not significantly rescue knockdown of Atxn7, but heterozygous mutation of Atxn7 rescued haploinsufficiency of SCAR. The progressive retinal and neurodegenerative disease Spinocerebellar Ataxia type 7 (SCA7) is caused by CAG trinucleotide repeat expansion of the ATXN7 gene, resulting in polyglutamine (polyQ) expansion at the amino terminus of the Atxn7 protein (Giunti et al., 1999;David et al., 1997). SCA7 disease is characterized by progressive cone-rod dystrophy leading to blindness, and progressive degeneration of the spine and cerebellum (Martin, 2012;Garden, 1993). In Drosophila, loss of Atxn7 leads to a phenotype similar to overexpression of the polyQexpanded amino terminal truncation of human Atxn7 (Latouche et al., 2007) -reduced life span, reduced mobility, and retinal degeneration (Mohan et al., 2014a). Without Atxn7, the Drosophila DUBm is released, enzymatically active, from SAGA. Once released, the module acts as a gain-offunction, leading to reduced ubiquitination of H2B. Similarly, the mammalian DUBm binds and deubiquitinates substrates without Atxn7 in vitro, albeit at a lower rate (Lan et al., 2015;Yang et al., 2015). Consistent with Non-stop release and over activity mediating the Drosophila Atxn7 loss-offunction phenotype, reducing non-stop copy number alleviates lethality associated with loss of Atxn7 (Mohan et al., 2014a). Non-stop is a critical mediator of retinal axon guidance and important for glial cell survival (Weake et al., 2008;Martin et al., 1995). Interestingly, abnormal ubiquitin signaling in the nervous system contributes to a number of genetic and spontaneous neurological and retinal diseases, including SCA3, SCAR16, Alzheimer's, Parkinson's, ALS, and Huntington's disease (Petrucelli and Dawson, 2004;Campello et al., 2013;Mohan et al., 2014b). Few functions are known for the DUBm beyond deubiquitination of H2Bub, H2Aub, shelterin, and FBP1 (Atanassov and Dent, 2011;Atanassov et al., 2009). In polyQ-Atxn7 overexpression models, sequestration of the DUBm may result in reduced deubiquitinase activity on critical substrates. Interestingly, both increased and decreased H2Bub have been shown to hinder gene expression, and imbalances (whether up or down) in ubiquitination are observed in retinal and neurological diseases (Ristic et al., 2014). Together, these data suggest more needs to be understood about Atxn7-mediated regulation of DUBm function. To better understand the Non-stop-Atxn7 regulatory axis, we set out to identify substrates for Non-stop and determine the consequence of Atxn7 loss on Non-stop/substrate interactions. To this end, we purified Non-stop-containing complexes, fractionated them by size, and tested enzymatic activity to identify novel complexes bearing enzymatically active Non-stop. This revealed the functionally active DUBm associates with protein complexes distinct from SAGA. Mass spectrometry revealed Arp2/3 complex and Wiskott-Aldrich syndrome protein (WASP)-family verprolin homologous protein (WAVE) Regulatory Complex (WRC) members suppressor of extracellular cAMP receptor (cAR) (SCAR), HEM protein (Hem), and specifically Rac1-associated protein 1 (Sra-1), as interaction partners of the independent DUBm. These complexes function together to initiate actin branching (Kurisu and Takenawa, 2009). Pull-down and immunofluorescence verified interaction and revealed extensive colocalization between Non-stop and WRC subunit SCAR. In flies, loss of Atxn7 led to increased interaction between Non-stop and SCAR. In cells and in flies, loss of Atxn7 resulted in a 2.5-fold increase in SCAR protein levels, while loss of Non-stop led to 70% decrease in SCAR protein levels. Spatiotemporal regulation of SCAR is essential for maintaining cytoskeletal organization. A constant ubiquitination-proteasomal degradation mechanism contributes to establishing and maintaining SCAR protein amount. Mutating the Non-stop enzymatic pocket to increase affinity for ubiquitin led to increased interaction with SCAR. An alternative enzymatic pocket mutation decreasing affinity for ubiquitin reduced interaction with SCAR. When Nonstop was knocked-down in cells, the amount of ubiquitinated SCAR increased. Loss of SCAR in Nonstop mutants was rescued by proteasome inhibition, suggesting Non-stop counters SCAR polyubiquitination and proteasomal degradation. An emerging class of proteins, important for neural function and stability bear WRC interacting receptor sequences (WIRS) which contribute to direct interaction and spatiotemporal regulation of SCAR (Chen et al., 2014). Close examination of Non-stop amino acid sequence revealed conserved WIRS motifs in Drosophila and human Non-stop orthologues. Mutating these decreased Non-stop-SCAR interaction, but allowed Non-stop to assemble into SAGA. Overexpression of Non-stop increased total SCAR protein levels approximately 5-fold. Interestingly, SCAR protein levels increased in cellular compartments where Non-stop increased, including in the nucleus. When SCAR protein levels were increased and relocalized by augmenting Non-stop, local F-actin amounts increased too, leading to cell rounding and fewer cell protrusions. Mutation of Non-stop WIRS motifs reduced Non-stop ability to increase SCAR protein, F-actin, and to reduce cellular protrusions. Although SCAR heterozygosity was not sufficient to rescue defective retinal axon targeting upon Atxn7 knockdown, loss of Atxn7 was sufficient to rescue loss of F-actin in SCAR heterozygotes. Non-stop-containing protein complexes were purified from nuclear extracts via tandem affinity purification of their epitope tags: FLAG, then HA (Suganuma et al., 2010). The complexity of these samples were assessed by silver stain ( Figure 1D). Isolated complexes from SAGA subunits Atxn7 and Will Decrease Acetylation (WDA) are shown for comparison (Mohan et al., 2014a). Although similar in profile to Atxn7 and WDA purifications, Non-stop-containing complexes were unique, indicating Non-stop has novel interaction partners. To separate SAGA from other Non-stop-interacting protein complexes, purified complexes were separated by size using Superose 6 gel filtration chromatography ( Figure 1E, upper) (Kusch et al., 2003). For comparison purposes, data for Atxn7 and Ada2B are reproduced here (Mohan et al., 2014a). Atxn7 and Ada2b are SAGA-specific subunits. Their elution from the gel-filtration column indicates those fractions contain SAGA (Mohan et al., 2014a;Kusch et al., 2003). Non-stop eluted in additional fractions ( Figure 1E, upper). To determine which fractions contained enzymatically active Non-stop, the ubiquitin-AMC deubiquitinase activity assay was performed, in which the fluorescent probe AMC is initially quenched by conjugation to ubiquitin but fluoresces when released by deubiquitination, permitting an indirect measurement of the relative amount of deubiquitination occurring in each sample ( Figure 1E, lower) (Kö hler et al., 2010;Samara et al., 2010). In complexes purified through Atxn7, peak deubiquitinase activity was detected around 1.8MDa, consistent with SAGA complex and previous purification, fractionation, and mass spectrometry of Atxn7, which is reproduced here for purposes of comparison (Mohan et al., 2014a;Kusch et al., 2003;Lee et al., 2011). In Non-stop purifications, however, deubiquitinase activity resolved into three major peaks. The major activity peak resolved at about 1.8 MDa, along with SAGA, and was labeled 'Group 1'. A secondary peak, 'Group 2', was resolved centering approximately 669 kDa, which was interpreted to be non-SAGA large multi-protein complexes. A final peak with the lowest enzymatic activity, centering about 75 kDa, was labeled 'Group 3.' The DUBm, consisting of Non-stop, Sgf11, and E(y)2 are predicted to have a molecular mass of 88 kDa, meaning this last peak was the DUBm interacting with very small proteins, if any. Silver staining of a single fraction at the center of Group 2 revealed the fractionation procedure resolved complexes into a pattern distinct from the total input ( Figure 1E, right). The three fractions comprising the center of each peak were combined, and MudPIT mass spectrometry was performed to identify their protein constituents (Washburn et al., 2001). MudPit proteomics provides a distributed normalized spectral abundance factor (dNSAF) which acts as a reporter for relative protein amount within a sample (Washburn et al., 2001). Consistent with previous characterizations of SAGA, Group 1 contained SAGA complex subunits from each of the major (HAT/TAF/SPT/DUB) modules (Mohan et al., 2014a;Kusch et al., 2003;;Lee et al., 2011;Setiaputra et al., 2015). Group 2 was selected for further interrogation because it was predicted to contain non-SAGA large multi-protein complexes. Group 2 contained the DUBm: Non-stop, Sgf11, and E(y)2 -indicating the DUBm remained intact during purification and fractionation (Kö hler et al., 2010;Samara et al., 2010;Lee et al., 2011). Histone H2A and H2B were also detected in Group 2, suggesting the DUBm co-purified with known substrates, further confirming conditions were suitable for preservation and identification of endogenous interactions ( Figure 2A) (Zhao et al., 2008;Henry et al., 2003). Furthermore, Non-stop dNSAF was not over-represented relative to Sgf11 and E(y)2, further confirming Non-stop bait protein was not expressed beyond physiological levels ( Figure 2A) . In Drosophila, SCAR is particularly important for cytoplasmic organization in the blastoderm, for egg chamber structure during oogenesis, axon development in the central nervous system, and adult eye morphology (Zallen et al., 2002). To further characterize the relationship between Nonstop, Atxn7, and WRC, we turned to the BG3 cell line (ML-DmBG3-c2, RRID:CVCL_Z728). These cells were derived from the Drosophila 3 rd instar larval central nervous system. Biochemical and RNAseq profiling of these cells indicate they express genes associated with the central nervous system. Their morphology is similar to that of Drosophila CNS primary cultured cells (Cherbas et al., 2011;Ui et al., 1994). To verify interaction between Non-stop and WRC, BG3 cells were transiently transfected with either Non-stop-FH or SCAR-FLAG-HA (SCAR-FH) and immunoprecipitated using anti-FLAG resin. Anti-HA antibody was used to verify pull down of the tagged protein. A previously verified anti-SCAR antibody confirmed the presence of SCAR in Non-stop-FH immunoprecipitate ( Figure 2B) Rodriguez-Mesa et al., 2012;Cetera et al., 2014;Tran et al., 2015;. Reciprocal immunoprecipitation of SCAR-FH utilizing an anti-FLAG resin followed by immunoblotting for Non-stop using an anti-Non-stop antibody confirmed SCAR-FH associated with endogenous Non-stop ( Figure 2C) (Mohan et al., 2014a). To visualize the scope of Non-stop and WRC colocalization, indirect immunofluorescence was used to

localize endogenous SCAR and Non-stop in BG3 cells. BG3 cells were plated on concanavalin A coated coverslips giving them an appearance similar to S2 cells plated on concanavalin A (Rogers et al., 2003) ( Figure 2D). SCAR localized in a characteristic pattern with high concentrations Figure 1 continued transfected cell lines. Parental S2 cells containing no plasmid compared to S2 cells stably transfected with Non-stop plasmid. Quantitation of anti-Nonstop immunoblot signal intensities normalized to Ponceau total protein loading control. Error bars represent standard error. (D) Analysis of purified complexes. Non-stop-containing complexes were analyzed by silver staining. Atxn7 and Will Decrease Acetylation (WDA)-containing complexes are shown for comparison. (E) Protein complexes, purified and fractionated as described in B were analyzed by immunoblotting to observe relative elution by size (top, Atxn7 and Ada2B recreated from Mohan et al., 2014b) followed by measure of deubiquitinase activity contained in each fraction as assayed by ubiquitin-AMC assay in which increased fluorescence correlates directly with deubiquitination activity. The complexity of eluted fractions was analyzed and Group 2 peak fraction, number 14, is shown (right). DOI: https://doi.org/10.7554/eLife.49677.002 The following figure supplement is available for figure 1: around the nucleus and at the cell periphery (Rogers et al., 2003). Non-stop localized similarly to SCAR, but also further within nuclei. Examination of the spatial relationship between Non-stop and F-actin, using phalloidin, revealed Non-stop overlapping with portions of F-actin/phalloidin ( Figure 2E) (Cooper, 1987). Atxn7 and Non-stop coordinate SCAR protein levels Atxn7 loss releases Non-stop from SAGA and is lethal, although, lethality is rescued by reducing Non-stop gene copy number. This implies a class of Non-stop target(s), which associate more with Non-stop in the absence of Atxn7. Furthermore, it raises the possibility that a subset of these may contribute to toxicity resulting from Atxn7 loss (Mohan et al., 2014a). To determine if interaction between Non-stop and SCAR is regulated by Atxn7, endogenous Non-stop was immunoprecipitated from either wild-type or Atxn7-mutant 3 rd instar larval whole cell extracts. Immunoprecipitates were immunoblotted to detect endogenous SCAR protein. In Atxn7 mutant extracts, approximately 1.4 times more SCAR was present per Non-stop captured ( Figure 3A, right). This suggests Atxn7 retains a pool of Non-stop that can be freed to interact with SCAR. If Non-stop was acting as a SCAR deubiquitinase to counter proteolytic degradation, then loss of Atxn7 would lead to increased SCAR protein. Conversely, reducing Non-stop would lead to less SCAR protein. To test these hypotheses, RNAi was used to knock down Non-stop or Atxn7 transcripts in cell culture and SCAR protein levels were observed by immunoblotting. BG3 cells were soaked in dsRNA targeting LacZ (negative control), Non-stop, or Atxn7. After six days, denaturing whole cell protein extracts were prepared from these cells and immunoblotted to detect SCAR. Ponceau S total protein stain was used as loading control to verify analysis of equal amounts of protein (Romero-Calvo et al., 2010;Lee et al., 2016). Protein extracts from Non-stop knockdown cells consistently showed approximately 70% decrease in SCAR protein levels relative to LacZ control knockdown. Consistent with a model in which Atxn7 restrains Non-stop deubiquitinase activity, cells treated with Atxn7-targeting RNAs showed 2.5-fold increase in SCAR protein ( Figure 3B). We verified these results by immunoblotting whole cell extracts prepared from the brains of 3 rd instar larvae bearing homozygous mutations for Atxn7 (Ataxin7 [KG02020] ), a loss-of-function allele (Mohan et al., 2014a) or non-stop (not 02069 ), also a loss of function allele (Poeck et al., 2001). Larval brain extracts from Atxn7 mutants showed about 2.5-fold increase in SCAR protein levels ( Figure 3C). In non-stop mutant extracts SCAR was decreased approximately 75% ( Figure 3C). SAGA was first identified as a chromatin modifying complex and predominantly characterized as a transcriptional coactivator. Therefore, we examined whether Non-stop influences SCAR gene expression. In genome wide analysis of glial nuclei isolated from mutant larvae, SCAR transcript was significantly down regulated by one fold in Non-stop deficient glia, but not in Sgf11 deficient glia (Ma et al., 2016). Since Sgf11 is required for Non-stop enzymatic activity, it is not clear what mechanism links Non-stop loss to reduction in SCAR gene expression. To confirm the relationship between Non-stop and SCAR gene expression, we used qRT-PCR to measure SCAR transcript levels in BG3 cells knocked-down for non-stop. We found a similar one half decrease in SCAR transcript levels ( Figure 3D). However, examination of Atxn7 knockdown showed no decrease in SCAR transcript levels ( Figure 3D). Therefore, while non-stop loss leads to modest decreases in SCAR transcripts, SCAR gene expression is independent of Atxn7 or Sgf11, suggesting this gene is not SAGA-dependent or dependent on Non-stop enzymatic activity. To determine the consequences of reducing SCAR transcripts by half on SCAR protein levels, we examined SCAR heterozygous mutant larval brains immunoprecipitated using anti-FLAG affinity resin and purified interactors analyzed by immunoblotting for the presence of endogenous SCAR. Anti-HA immunoblots verified the presence of Non-stop-FH bait. NS marks a non-specific band. (C) Pull-down was performed as in B, but with SCAR-FLAG-HA (SCAR-FH) expression vector. Immunoprecipitated proteins were probed by immunoblotting to detect endogenous Non-stop. Anti-HA immunoblots verified the presence of SCAR-FH bait. (D) Endogenous Non-stop and SCAR occupy similar subcellular regions. Immunofluorescence of SCAR and Nonstop in BG3 cells. Cells were immunostained with anti-Non-stop (magenta), anti-SCAR (yellow), and DAPI (white). Scale bar is 10 mM. Inset is enlarged to the right. (E) Endogenous Non-stop and F-actin are found in similar subcellular regions. Representative images of BG3 cells immunostained with anti-Non-stop (magenta), phalloidin (F-actin)(yellow), and DAPI (DNA)(white). Scale bar is 10 mM. Inset is enlarged to the right. Images in D and E were adjusted with contrast limited adaptive histogram equalization using the ImageJ CLAHE algorithm (Zuiderveld, 1994). DOI: https://doi.org/10.7554/eLife.49677.004 Figure 3. Non-stop and Atxn7 regulate SCAR protein levels. (A) Endogenous pull-down reveals increased interaction between Non-stop and SCAR in the absence of Atxn7. Whole cell extracts prepared from either OregonR (WT) or homozygous mutant Ataxin7 [KG02020] (Atxn7) third instar larvae were subject to immunoprecipitation with pre-immune guinea pig serum or anti-Non-stop antibody as indicated. Immunoprecipitates were analyzed by immunoblotting with anti-Non-stop to verify capture, and with anti-SCAR antibody to assess interaction. To compare the relative proportion of SCAR Figure 3 continued on next page (SCAR [Delta37] /+). qRT-PCR analysis showed heterozygous SCAR brains produced 40% less SCAR transcript than wild-type brains ( Figure 3E). However, heterozygous SCAR mutant and wild-type brains showed similar amounts of SCAR protein, demonstrating this amount of transcript is sufficient for full protein production ( Figure 3F). Since non-stop mutants showed a similarly modest decrease in SCAR gene expression but a drastic decrease in protein levels, we tested the possibility that Nonstop counters SCAR protein degradation post-translationally. Purification and mass spectrometry identified WRC members SCAR, HEM, and SRA-1, as interaction partners of the independent DUBm ( Figure 2A). In Drosophila, HEM, SRA-1, and Abi recruit an unidentified deubiquitinase to balance SCAR regulation through a constant ubiquitination-proteasomal degradation mechanism facilitating rapid changes in SCAR protein amount and localization (Chen et al., 2014;Kunda et al., 2003). In mammals, the USP7 deubiquitinase acts as a molecular rheostat controlling proteasomal degradation of WASH (Hao et al., 2015). To probe whether ubiquitinated SCAR is a substrate for Non-stop, the substrate binding pocket of Non-stop was modified to either augment (Non-stop(C406A)), or reduce (Non-stop(C406S)) substrate binding (Morrow et al., 2018). These mutants were immunoprecipitated from whole cell extracts prepared from transiently transfected BG3 cells. Immunoblotting revealed that Non-stop (C406A) bound to SCAR approximately 4-fold more than wild-type Non-stop while Non-stop(C406S) bound only about 1/3 rd as well ( Figure 4A and B, upper panels). To verify both mutants incorporated into SAGA, interaction with Ada2b was confirmed ( Figure 4A and B, lower panels). Entry into the proteasome for destruction can be triggered by polyubiquitination (Grice and Nathan, 2016;Meulmeester et al., 2005). To determine whether Non-stop counters polyubiquitination of SCAR, HA-ubiquitin and SCAR-FLAG were expressed in BG3 cells and either LacZ (negative control) or non-stop were knocked down followed by a brief period of proteasomal inhibition to preserve polyubiquitinated SCAR which would otherwise be rapidly destroyed by the proteasome (Xia et al., 2012). Immunoprecipitation of SCAR followed by immunoblotting with a ubiquitin-specific antibody revealed that knock-down of Non-stop nearly tripled levels of SCAR polyubiquitination ( Figure 4C). To determine whether Non-stop regulates SCAR entry into a proteasomal degradation pathway 3 rd instar larval brains were cultured ex vivo to permit pharmacological inhibition of the proteasome in defined genetic backgrounds (Rabinovich et al., 2015;Prithviraj et al., 2012). Wandering 3 rd instar larval brains were isolated from non-stop homozygous mutant larvae and cultured for 24 hr in the presence of MG132 proteasome inhibitor. Treating non-stop brains with MG132 rescued SCAR protein amounts to levels found in wild type brains ( Figure 4D). Together, these data show Nonstop counters polyubiquitination and entry of SCAR into the proteasomal degradation pathway. Non-stop regulates SCAR protein levels and localization To examine whether increasing Non-stop led to increased SCAR, we expressed recombinant Nonstop in BG3 cells and utilized indirect immunofluorescence to observe the amount and location of endogenous SCAR protein. Upon expression of Non-stop-FH in BG3 cells, endogenous SCAR protein levels increased five-fold compared with mock transfected cells ( Figure 5A). Utilizing anti-HA interacting with Non-stop, immunoblotting signals were quantified by densitometry and the amount of SCAR interacting per given amount of Non-stop was calculated and expressed relative to WT control. Error bars represent standard error. (B and C) Loss of Atxn7 increases, and loss of Non-stop decreases SCAR protein levels. (B) BG3 cells were treated with dsRNA to knock down either Atxn7, LacZ (control), or non-stop (not). Denaturing whole cell extracts were analyzed by immunoblotting to determine SCAR protein levels. Total protein (Ponceau, lower panel) was used as loading control. SCAR intensity was quantified and values were normalized to Ponceau and expressed relative to LacZ control (right panel). Error bars represent standard error. (C) Brain and central nervous systems from WT, non-stop 02069 (not), or Ataxin7 [KG02020] (Atxn7) third instar larvae were isolated and denaturing whole cell extracts prepared. SCAR protein levels were determined by immunoblotting and quantified as in B. (D, E, F) Atxn7-or Non-stopmediated gene regulation are not sufficient to change SCAR protein levels. (D) Relative quantification of SCAR transcripts in BG3 cells treated with dsRNA targeting either Atxn7, LacZ (control), or not. Transcripts from not and Atxn7 treated cells are expressed relative to LacZ. (E) Relative quantification of SCAR transcript levels in larval brains dissected from WT or SCAR [Delta37] /cyo-gfp (SCAR/+). SCAR transcript levels are set relative to WT. (F) Immunoblot for SCAR protein in larval brains dissected from OregonR (WT) or SCAR [Delta37] /cyo-gfp (SCAR/+). SCAR protein levels were quantified as in B. DOI: https://doi.org/10.7554/eLife.49677.005 immunofluorescence to observe recombinant Non-stop, three distinct subcellular localization patterns were apparent: nuclear, cytoplasmic, and evenly distributed between nucleus and cytoplasm ( Figure 5A). Increased SCAR protein co-compartmentalized with Non-stop, even within the nucleus, where SCAR function is more enigmatic ( Figure 5A). To quantify co-compartmentalization of SCAR and Non-stop, the nuclear to cytoplasmic ratios of Non-stop-FH and endogenous SCAR were calculated by dividing immunofluorescence intensity within each nucleus by that of the corresponding cytoplasm. This was plotted as the ratio nuclear:cytoplasmic SCAR on the Y axis and the ratio nuclear:cytoplasmic Non-stop on the X axis. As the amount of Non-stop in the nucleus increased, so did the amount of SCAR in the nucleus, showing Non-stop was driving SCAR localization (R 2 = 0.8) ( Figure 5A). SCAR functions, within WRC, to activate Arp2/3 and facilitate branching of F-actin filaments (Kurisu and Takenawa, 2009). With correct spatiotemporal regulation of these actin-branching complexes, cells are able to establish appropriate shape and size through extension of actin-based protrusions (Blanchoin et al., 2014). BG3 cells have a characteristic bimodal appearance when plated in the absence of concanavalin A (Liu et al., 2009). If Non-stop overexpression were disrupting SCAR function through augmenting and mislocalizing SCAR, changes in cell shape and ability to form protrusions were predicted (Machesky and Insall, 1998). Using phalloidin to observe F-actin in cells overexpressing Non-stop, a substantial change in cell morphology was observed. Cells displayed fewer projections and covered less total area ( Figure 5B). This was specific for cells expressing Nonstop in the cytoplasm or both nucleus and cytoplasm. Increased Non-stop signal associated closely with increased

phalloidin signal. Overexpression of SCAR alone similarly led to reductions in cell area and number of projections observed ( Figure 5B). Non-stop bears multiple, conserved, WRC interacting receptor sequence (WIRS) motifs The basis for interaction between the SAGA DUBm, Arp2/3, and WRC was sought. WRC subunits HEM, SRA-1, and Abi mediate interaction with an unidentified deubiquitinase to counteract proteasomal degradation of SCAR (Kunda et al., 2003). WRC subunits Abi and SRA-1 mediate interaction with WRC interacting receptor sequences (WIRS)-bearing proteins which contribute to spatiotemporal regulation of SCAR (Chen et al., 2014). Recall, Hem and Sra-1 were present in mass spectrometry analysis of Group 2 proteins, suggesting Non-stop was in vicinity of this surface of WRC. Primary sequence analysis of DUBm subunits revealed four WIRS-conforming sequences on Nonstop and three on mammalian Non-stop orthologue USP22. These were numbered from 0 to 3, with WIRS1 and 3 conserved in location between Drosophila and mammalian Non-stop. ( Figure 6A). None of the other DUBm components bore conserved WIRS motifs. Although analysis of the yeast orthologue of Non-stop (UBP8) did not reveal a WIRS-like motif, the Atxn7 orthologue Sgf73 did. In Figure 4 continued immunoblotted for SCAR. Immunoblots for Ada2b control for incorporation into the SAGA complex. Immunoblots for HA verify the presence of the Non-stop-FH constructs. Ada2b and SCAR intensity were quantified and normalized to HA. Values are shown as relative to WT. Error bars represent standard error. (B) Non-stop catalytic mutation which decreases substrate binding also decreases interaction with SCAR. The Non-stop catalytic cysteine was mutated to serine (C406S). BG3 cells were mock transfected (-), transfected with Non-stop-FH, or with Non-stop C406S-FH (C406S) and whole cell extracts were prepared. Flag resin was used to immunoprecipitate the Non-stop-FH constructs. Pull-downs were immunoblotted for SCAR. Incorporation into SAGA was verified by Ada2b immunoblot. HA immunoblots verify the presence of the Non-stop-FH constructs. Ada2b and SCAR intensity was quantified and normalized to HA. Values are shown as relative to wild type. Error bars are standard error. (C) Non-stop counters polyubiquitination of SCAR. BG3 cells were treated with dsRNA targeting either non-stop or LacZ and transfected with HA-ubiquitin (HA-Ub) and SCAR-FLAG (SCAR-F). Control cells were treated with LacZ dsRNA and a mock transfection was performed. After six days, cells were treated with MG132 protease inhibitor for 6 hr and denaturing whole cell extracts were made. Anti-flag resin was used to capture SCAR-FLAG. Immunoblots were performed for ubiquitin (VU-1) and Flag to verify presence of SCAR. Protein intensity for VU-1 was measured and normalized to Flag protein intensity. Values are shown as relative to LacZ. Error bars represent standard error. (D) Non-stop counters proteasomal degradation of SCAR. WT or not brains were dissected from 3 rd instar larva and cultured ex vivo. Half of the brains from each genotype were treated for 24 hr with protease inhibitor [50 mm MG132 (MG132+)] while the remaining brains served as untreated control (MG132-). Whole cell extracts were immunoblotted for SCAR. SCAR protein intensity was measured and normalized to Ponceau total protein control. Numbers are expressed as relative to WT untreated control. Error bars represent standard error. DOI: https://doi.org/10.7554/eLife.49677.006 yeast, the DUBm can also be found separate from SAGA, although not without Sgf73 (Lim et al., 2013). Therefore, we focused on Non-stop WIRS motifs. To verify Non-stop WIRS motifs are important for Non-stop/SCAR interaction, putative WIRS motifs were inactivated using a phenylalanine to alanine substitution, previously demonstrated to be effective at inactivating these motifs (Chen et al., 2014). This mutant displayed an approximate 80% reduction in SCAR binding, but still integrated into SAGA as observed by maintenance of Atxn7 binding ( Figure 6B). To determine whether Non-stop WIRS sequences are necessary for Non-stop-mediated modulation of SCAR function, a series of phenylalanine to alanine point mutants of individual WIRS domains (WIRS F-A) were generated and expressed in BG3 cells. As above (Figure 5), the ability of Non-stop WIRS mutants to increase the levels of endogenous SCAR protein, to change cell area, and to change the number of protrusions per cell was measured. Immunofluorescence was used to detect the HA epitope on exogenously expressed Non-stop and an antibody toward endogenous SCAR was used to observe SCAR protein levels. Cells expressing moderate amounts of Non-stop as judged by the spectrum of fluorescence intensity were analyzed to avoid assaying cells with either too little or too much exogenous Non-stop relative to endogenous protein. The ratio of HA fluorescence to SCAR fluorescence in each mutant was calculated and compared to wild-type Non-stop ( Figure 6C). In each case, the Non-stop mutant produced less SCAR protein than wild type, suggesting that these motifs are indeed important for Non-stop mediated increases in SCAR protein levels ( Figure 6C). Mutant Non-stop was also less potent for decreasing cell area and decreasing the number of cell protrusions ( Figure 6D). The effect of mutating each WIRS was not equivalent, suggesting a hierarchy for which is most important for WRC stabilization. In Drosophila, specific cell-type and subcellular location of actin branching enzymes facilitate fine tuning of actin remodeling in space and time (Zallen et al., 2002). SCAR and the single Drosophila WASp homolog act non-redundantly to orchestrate actin branching in different tissues and to regulate different developmental decisions. SCAR is particularly important in axon development in the central nervous system and adult eye morphology, but also essential for blastoderm organization and egg chamber structure during development. Within these tissues, cells display both cytoplasmic and nuclear localization of SCAR (Zallen et al., 2002). Considering the demonstrated importance of Atxn7, Non-stop, and SCAR in the nervous system and eye, the interplay between these genes was examined in 3 rd instar larval brains. To observe an output of SCAR function, phalloidin was used to visualize F-actin. In Non-stop mutants, F-actin intensity was reduced by approximately 70% ( Figure 7A). In Atxn7 mutants, F-actin levels increased approximately 1.7 times ( Figure 7B). To test for genetic interaction between the SAGA DUBm and SCAR, heterozygous mutations of Atxn7 and SCAR were combined and F-actin levels determined ( Figure 7C). In Atxn7 heterozygotes, F-actin levels were not significantly different than wild-type. In SCAR heterozygotes, however, F-actin levels were decreased. Combining Atxn7 and SCAR heterozygous mutants rescued F-actin levels. Together, these results support a model where Atxn7 restrains Non-stop from stabilizing and increasing levels of SCAR, which then increases F-actin. Figure 5 continued Scale bar is 10 mM. Hashed lines outline DAPI (inner circle) to approximate nuclear location and the outermost SCAR signal to approximate the cell edge (outer circles). Three categories of HA (Non-stop-FH) expression are shown: Non-stop 'nuclear', Non-stop 'cytoplasmic', and Non-stop 'nuclear and cytoplasmic.' Total SCAR signal was measured in HA positive cells (Non-stop-FH) and mock transfected cells (Control). The average increase in endogenous SCAR immunofluorescence intensity was determined relative to control (top, right panel). Error bars are standard error. To examine the relationship between Non-stop localization and increased SCAR protein, SCAR immunofluorescence intensity was measured in the nucleus (as defined by DAPI) and in the cytoplasm. A ratio of nuclear SCAR intensity divided by cytoplasmic SCAR intensity was calculated. HA intensity was similarly measured in the nucleus and cytoplasm. A ratio of nuclear HA intensity to cytoplasmic HA intensity was calculated. The HA ratio was plotted on the X axis and the SCAR ratio was plotted on the Y axis. A trend line was calculated and the R 2 was determined to be 0.80. (B) Increasing Non-stop alters F-actin organization similarly to increasing SCAR -reducing cell area and number of cell protrusions. BG3 cells were transfected with no-plasmid (Control), Non-stop-FH, or SCAR-FH. The location of exogenously expressed proteins were determined as above and F-actin was visualized using phalloidin. Cell area and total number of cellular projections were counted by quantifying the phalloidin signal distribution. HA-positive cells (Non-stop-FH or SCAR-FH) were compared to mock transfected (Control) cells (bottom panels). Cells expressing Non-stop-FH were split into three categories of HA (Non-stop-FH) expression as above. T-tests were used to compare samples and p-values are shown. Error bars are standard error. DOI: https://doi.org/10.7554/eLife.49677.007 To test whether SCAR rescues an Atxn7 phenotype, retinal axon targeting in 3 rd instar larval brains was observed using an anti-chaoptin antibody, a retinal axon-specific protein (Figure 8). Neural-specific expression of Atxn7 RNAi (Dietzl et al., 2007), driven by elav-Gal4, results in defective axon targeting. Similarly, knockdown of either Non-stop (Perkins et al., 2015) or SCAR (Perkins et al., 2015) led to defective axon targeting. Gcn5 knock-down (Dietzl et al., 2007), however, resulted in wild-type axon targeting. To test whether defective axon targeting upon Atxn7 loss, could be rescued by reducing SCAR (Perkins et al., 2015), Atxn7 was knocked-down in a SCAR heterozygous background. Although there were fewer defects in axon targeting (62% normal vs 52% normal), this difference was not sufficient to achieve statistical significance (p=0.73). Discussion Purification of Atxn7-containing complexes indicated that Atxn7 functions predominantly as a member of SAGA (Mohan et al., 2014a). In yeast, the Atxn7 orthologue, Sgf73, can be separated from SAGA along with the deubiquitinase module by the proteasome regulatory particle (Lim et al., 2013). Without Sgf73, the yeast deubiquitinase module is inactive (Samara et al., 2010). In higher eukaryotes, Atxn7 increases, but is not necessary for Non-stop/USP22 enzymatic activity in vitro (Mohan et al., 2014a;Lan et al., 2015). In Drosophila, loss of Atxn7 leads to a Non-stop over activity phenotype, with reduced levels of ubiquitinated H2B observed (Mohan et al., 2014a). Here, purification of Non-stop revealed the active SAGA DUBm associates with multi-protein complexes including WRC and Arp2/3 complexes separate from SAGA. SCAR was previously described to be regulated by a constant ubiquitination/deubiquitination mechanism (Kunda et al., 2003). SCAR protein levels increased upon knockdown of Atxn7 and decreased upon knockdown of non-stop. Decreases in SCAR protein levels in the absence of non-stop required a functional proteasome. Conversely, overexpression of Non-stop in cells led to increased SCAR protein levels and this increased SCAR protein colocalized to subcellular compartments where Non-stop was found. Nuclear Arp2/3 and WRC have been linked to nuclear reprogramming during early development, immune system function, and general regulation of gene expression (Yoo et al., 2007;Miyamoto et al., 2013;Taylor et al., 2010). Distortions of nuclear shape alter chromatin domain location within the nucleus, resulting in changes in gene expression (Navarro-Lérida et al., 2015;Rodriguez-Mesa et al., 2012). Nuclear pore stability is compromised in SAGA DUBm mutants, resulting in deficient mRNA export (Kö hler et al., 2008). Similarly, mutants of F-actin regulatory proteins, such as Wash, show nuclear pore loss . Non-stop may contribute to nuclear pore stability and mRNA export through multiple mechanisms. When we examined the basis for this unexpected regulatory mechanism, we uncovered a series of WIRS motifs (Chen et al., 2014) conserved in number and distribution between flies and mammals. These sequences functionally modulate Non-stop ability to increase SCAR protein levels. Point mutants of each WIRS resulted in less SCAR protein per increase in Non-stop protein. WIRS mutant Non-stop retained the ability to incorporate into SAGA, indicating these are separation of function mutants. Figure 6 continued immunoprecipitated using anti-FLAG resin to capture exogenous Non-stop and immunoblotted for SCAR. Immunoblot for Atxn7 verified incorporation into the SAGA complex. Immunoblot for HA verified capture of Non-stop-FH. Protein intensity of Atxn7 and SCAR were normalized to HA intensity. Intensities are shown relative to WT Non-stop. Error bars are standard error. (C) BG3 cells were transfected with either wild type (WT) or Non-stop-FH harboring a point mutation in one of the four putative WIRS motifs (W0-W3) as indicated. Exogenously expressed Non-stop and endogenous SCAR were detected by indirect immunofluorescence. Scale bar is 10 mm. Total fluorescence intensity of SCAR and HA were measured in HA containing cells. Box plots show the SCAR intensity divided by the intensity of HA. T-tests were used to compare samples and p-values are shown. P-values marked with an asterisk are significant. (D) BG3 cells transfected with either wild-type or Non-stop-FH harboring a point mutation in one of the four putative WIRS motifs (W0-W3) were immunostained for HA (magenta) and phalloidin (yellow). Scale bar is 10 mm. Arrows point to localized accumulation of Non-stop WIRS mutant protein, which coincide with aberrant actin protrusions commonly observed upon WRC or Arp2/3 loss of function. Cell area and number of projections protruding from cells was determined for HA positive cells. T-tests were used to compare

the samples and p-values are shown. P-values marked with an asterisk are significant. Error bars are standard error. DOI: https://doi.org/10.7554/eLife.49677.008 Microscope acquisition settings were identical to allow comparison. Scale bar is 50 mM. Charts show the averaged phalloidin fluorescence intensity measurements for individual brain lobes. Wild type average intensity was set to one and mutants were normalized to wild -type. A t-test was used to Figure 7 continued on next page Overall, these findings suggest that the cell maintains a pool of Non-stop that can be made available to act distally from the larger SAGA complex to modulate SCAR protein levels ( Figure 9). In yeast, the proteasome regulatory particle removes the DUBm from SAGA (Lim et al., 2013). In higher eukaryotes, caspase-7 cleaves Atxn7 at residues which would be expected to release the DUBm, although this remains to be shown explicitly (Young et al., 2007). The mechanisms orchestrating entry and exit of the DUBm from SAGA remain to be explored. Cell culture ML-DmBG3-c2 (Drosophila Genomics Resource Center #68, RRID:CVCL_Z728, FBtc0000068) cells were grown in Schneider's media supplemented with 10% fetal bovine serum and 1:10,000 insulin. S2_DRSC cells (Drosophila Genomics Resource Center #181, RRID:CVCL_Z992) were maintained in Schneider's media supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin (Ther-moFisher, Catalog number: 15070063, 5000 U/ml). BG3 and S2 cell lines were ordered fresh from the Drosophila Genomics Resource Center and verified free from mycoplasma by PCR test or by microscopy. They were further verified by characteristics including survival in growth medium (these cell lines grow in distinct media -one cannot survive in the wrong media), their distinct morphology, and cell doubling time. Transfection BG3 cells were plated at 1 Â 10 6 cells/ml, 24 hr later they were transfected using Lipofectamine 3000 (ThermoFisher catalog number: L3000015) following manufacturer's directions. The Lipofectamine 3000 was used at 1 ml per mg of DNA and p3000 reagent was used at 1 ml per mg of DNA. To induce expression of proteins from plasmids containing a metallothionein promoter 250 mm of CuSO 4 was added to the media. For transfections taking place in six-well dishes 1 mg of DNA was used per well, 4 mg of DNA was used per 10 cm dish, and 15 mg of DNA was used per 15 cm dish. Proteins were given 2 to 3 days of expression before the samples were used. Primer list for quickchange mutagenesis: RNAi treatment of BG3 cells Short dsRNA was produced by amplifying cDNA with primers that contained T7 promoter sequences on the ends. These PCR products were then used as a template for in vitro transcription using the Ampliscribe T7 high yield transcription kit (epicentre catalog number: AS3107) following the manufacturer's instructions. Transcription was allowed to continue for 4 hr. To make the RNA double stranded it was then heated to 65˚C for 30 min and cooled to room temperature. The RNA was then ethanol precipitated in order to clean it. BG3 cells were plated at 2 Â 10 6 cells/ml in six well dishes. Cells were allowed to adhere (approximately 4 hr) then treated with 10 mg of dsRNA. The cells were harvested for western blot after being soaked for 6 days and after 5 days for RNA analysis. For 10 cm dishes cells were plated at 16 Â 10 6 cells in 4 ml of serum-free media 120 mg of dsRNA was immediately added. Cells were incubated for 1 hr at 25˚C. After 1 hr 9 ml of complete media was added. The cells were harvested for immunoprecipitation after being soaked for 6 days in dsRNA. dsRNA-LacZ-R: GCTAATACGACTCACTATAGGCCAAACATGACCARGATTACGCCAAGCT dsRNA-LacZ-F: GCTAATACGACTCACTATAGGCCAAACGTCCCATTCGCCATTCAGGC Immunofluorescence Cells were plated directly onto coverslips in six-well dishes and allowed to adhere. BG3 cells were fixed in 4% methanol-free formaldehyde after 2 days of growth. Cells were washed in 1xPBS and then either stored in 1xPBS for up to 1 week or immediately stained. Coverslips were washed with 1xPBS containing 0.5% Triton X-100 four times for 10 min. Coverslips were blocked in 5% BSA and 0.2% TWEEN 20 diluted in 1xPBS for 1 hr at room temperature. They were then incubated in primary antibody diluted in blocking buffer overnight at 4˚C. The following day they were washed with 1xPBS containing 0.5% Triton X-100 four times for 10 min. They were then incubated in secondary antibody diluted 1:1000 in blocking buffer for 4hr at room temperature. Cells were then washed four times 10 min at room temperature in 1xPBS containing 5% Triton X-100. Coverslips were dried and mounted in Vectashield containing DAPI (Vector Labs catalog number: H-1200). Larval brains were dissected into 4% methanol free formaldehyde/1xPBS and fixed at 4˚C for 1 hr. Fixative was removed and the brains were washed with 1xPBS and then either stored at 4˚C in Figure 8. Chaoptin staining in the optic lobe of third instar larval brains reveals similar defects in Atxn7, SCAR, and not knockdowns in vivo. The elav-Gal4 driver was used to drive UAS-RNAi lines in larvae. Third instar larval brains were dissected and immunostained with a Chaoptin antibody. Images were adjusted with contrast limited adaptive histogram equalization using the ImageJ CLAHE algorithm (Zuiderveld, 1994). Optic lobes were analyzed for the presence of bundles, gaps, or weak axons. Examples of these are indicated in the figure. A brain was determined to be normal if it had three or fewer of these defects. The percentage of normal brains are shown for each genotype. A fisher's exact test was used to compare samples to the driver alone (elav-gal4/+). Significant p-values are marked with an asterisk. Error bars shows standard error. DOI: https://doi.org/10.7554/eLife.49677.010 1XPBS for up to 1 week or immediately stained. Brains were washed three times in 1xPBS containing 5% Triton X-100. They were then incubated in 1xPBS containing 5% Triton for 20 min at room temperature. Cells were blocked for 1 hr at room temperature in blocking buffer containing 0.2% TWEEN 20% and 5% BSA diluted in 1xPBS. Brains were then incubated in Alexafluor conjugated phalloidin diluted 1:20 in blocking buffer overnight at 4˚C. They were then wash four times for 10 min at room temperature in 1XPBS containing 5% Triton X-100. Larval brains were mounted in Vectashield containing DAPI (Vector Labs catalog number: H-1200). For Chaoptin staining, brains were incubated in primary antibody diluted in blocking buffer at 4˚C overnight following blocking. They were washed four times 10 min at room temperature in 1xPBS containing 5% Triton X-100. They were then incubated in secondary antibody diluted 1:1000 in blocking buffer overnight at 4˚C. They were then washed four times 10 min in 1xPBS containing 5% Triton. Larval brains were mounted in Vectashield containing DAPI (Vector Labs catalog number: H-1200). A Zeiss LSM-5 Pascal upright microscope and LSM software was used for imaging. Cells were imaged with a 40X objective aperture number 1.30 oil or a 63X objective aperture number 1.25 Oil and Z-stacks were acquired using LSM software with a slice every 1 mm. Brains were imaged with a 20X objective aperture number 0.45 air and Z stacks were taken with slices every 4 mm. All images were taken at room temperature. All settings remained the same within an experiment. Image analysis Stacks were exported as TIFFs from the LSM software and analyzed in FIJI (Schindelin et al., 2012) RRID:SCR_002285. For analysis z projections were created by sum or max. To analyze cell shape and phalloidin structures in BG3 cells the phalloidin images were thresholded in ImageJ and used to create ROIs. The number of projections was counted by hand. A t-test was used to test the significance between the mock transfected control and the Non-stop-2xFlag-2 transfected cell lines in cellular shape and the number of projections. To measure SCAR levels in BG3 cells projections were made by sum and ROIs were drawn by hand around the entirety of the scar staining and integrated density was measured for both HA and SCAR. For BG3 cells, a second ROI was drawn around the nuclei by thresholding the DAPI image and the SCAR intensity and HA intensity in the nucleus was measured. A t-test was used to compare between treated and untreated cells. To measure phalloidin intensity in larval brains, max projections were created and ROIs were drawn by hand. Brains were only measured if they were not mounted at an angle that excluded part of the VNC and the VNC was intact. Thresholding and image calculator were used to subtract background. A t-test was used to compare differences between genotypes. To measure integrity of the optic lobe max projections were created and images were thresholded identically using contrast limited adaptive histogram equalization using the ImageJ CLAHE algorithm (Zuiderveld, 1994) and the number of gaps and bundles were counted by hand. Fisher's exact test was used to compare differences between genotypes. Ex vivo culturing of third instar brains Ex vivo culturing of larval brains was performed as described (Rabinovich et al., 2015;Prithviraj et al., 2012). Third instar brains were dissected in 1xPBS and quickly placed into culture media (Schneider's insect media supplemented with 10% FBS, 1% pen/strp, 1:10,000 insulin, and 2 mg/ml ecdysone) in 24-well dish. The dish was wrapped in parafilm and incubated at 25˚C. After 6 hr of culturing, the media was changed. The wells receiving MG132 then had 50 mM of MG132 added. Brains were collected for protein samples after 24 hr. Protein sample preparation All western blots using tissue culture cells were done from cells collected from a six-well dish. The media was removed and 200 ml of 2xUREA sample buffer (50 mM Tris HCl pH 6.8, 1.6% SDS, 7% glycerol, 8M Urea, 4% b mercaptoethanol, 0.016% bromophenol blue) and 2 ml of 10 mg/ml benzonase was added to the dish. After the mixture was lysed they were boiled at 80˚C for 10 min. Larval brains were dissected into 1xPBS. The PBS was removed and 2xUREA sample buffer was added in a 1:1 ratio and 10 units of Benzonase was added. The brains were homogenized by pipetting and boiled at 80˚C for 10 min. Non-denaturing extract Transfected cells were resuspended in Extraction Buffer (20 mM HEPES (pH7.5), 25% Glycerol, 420 mM NaCl, 1.5 mM MgCl 2 , 0.2 mM EDTA, 1:100 ethidium bromide with protease inhibitors added. 1% NP-40 was added and the cells were pipetted up and down until the solution was homogenous. They were placed on ice for 1 hr with agitation every 10-15 min. They were then centrifuged for 30 min at 4˚C at 20,000 x g. The supernatant was placed into a new tube and stored at -80˚C for up to 1 week. Lysates were thawed and 36 mg of protein was removed for input and mixed with 2 x Urea sample buffer (50 mM Tris-HCl (pH 6.8), 1% SDS, 7% Glycerol, 8 M Urea, 4% beta-mercapto-ethanol, 0.016% Bromophenol blue) and boiled for ten minutes at 80˚C. An equal volume of Dignum A buffer (10 mM HEPES (pH 7.5), 1.5 mM MgCl 2 , 10 mM KCl) was added to the lysates in order to adjust the salt concentration to 210 mM NaCl. The entire volume was added to 15 ml of Mouse IGG agarose (Sigma Catalog number: A0919) that had been equilibrated in one volume extraction buffer plus one volume dignum A. They were then rotated at 4˚C for 1 hr. After 1 hr, they were centrifuged for 1 min at 4˚C at 4500 x g. The supernatant was then added to anti-FLAG M2 Affinity Gel (Sigma-Aldrich Catalog number: A2220, RRID:AB_10063035) that had been equilibrated in one volume extraction buffer plus one volume Dignum. They were then rotated for 1 hr at 4˚C. After 1 hr, they were centrifuged at 4500 x g at 4˚C for 30 s. The supernatant was removed and the gel was washed five times in extraction buffer plus dignum A. The resin was then resuspended in 40 ml 2XUREA sample buffer and boiled for 10 min at 80˚C. In vitro ubiquitination assays Immunoprecipitations with the intention to immunoblot for ubiquitin were prepared in denaturing extracts as described (Zhang et al., 2006;Wang and Li, 2006). The day of harvesting cells were treated with 50 mM of MG132 for 6 hr. After 6 hr, cells were scraped and resuspended in Denaturing Buffer (1% SDS/50 mM Tris [pH 7.5], 0.5 mM EDTA/1 mM DTT) plus the protease inhibitors PMSF, sodium orthovanadate,

Aprotinin, Leupeptin, Pepstatin A, Sodium butyrate, and Sodium fluoride. They were boiled at 100˚C for 5 min. They were then diluted 10 fold in dilution buffer (10 mM Tris-HCl, pH 8.0, 150 mM NaCl, 2 mM EDTA, 1% Triton), mixed well, and then centrifuged for 30 min at 4˚C at 20,000 x g. The supernatant was then added to 15 ml anti-FLAG M2 Affinity Gel (Sigma-Aldrich Catalog number: A2220, RRID:AB_10063035) that had been equilibrated dilution buffer. They were then rotated for 1 hr at 4˚C. After 1 hr, they were centrifuged at 4000 x g at 4˚C for 30 s. The supernatant was removed and the gel was washed four times with washing buffer (10 mM Tris-HCl, pH 8.0, 1 M NaCl, 1 mM EDTA, 1% NP-40). The resin was then resuspended in 33 ml 2XUREA sample buffer and boiled for 10 min at 80˚C. Endogenous Non-stop immunopreciptiation Larva were placed into Extraction Buffer (20 mM HEPES (pH7.5), 25% Glycerol, 420 mM NaCl, 1.5 mM MgCl 2 , 0.2 mM EDTA, and 1:100 ethidium bromide with protease inhibitors added. 1% NP-40 was added and the cells were pipetted up and down until the solution was homogenous. Samples were spun for 30 min at 20,000 xg at 4˚C. 36 mg of protein was removed for input and mixed with 2 x Urea sample buffer (50 mM Tris-HCl (pH 6.8), 1% SDS, 7% Glycerol, 8 M Urea, 4% beta-mercaptoethanol, 0.016% Bromophenol blue), then boiled for 10 min at 80˚C. Input was run on a gel and a western blot for Non-stop was performed to determine relative Non-stop concentrations and used to normalize the amount used in pull down. An equal volume of dignum A buffer (10 mM HEPES (pH 7.5), 1.5 mM MgCl 2 , 10 mM KCl) was added to the lysates in order to adjust the salt concentration to 210 mM NaCl. Extracts were centrifuged for 30 min at 4˚C at 20,000 x g. The Non-stop antibody was added at a concentration of 1:50 and extracts were rotated at 4˚C overnight. Extracts were added to 50 ml of protein A dynabeads (Life technologies: 10002D) equilibrated in Extraction Buffer and rotated at 4˚C for 2 hr. The extract was removed and the dynabeads were washed four times in Extraction Buffer with 1% NP-40. Bead were resuspended in 30 ml of 2XUREA sample buffer as above and boiled for 10 min at 80˚C. Western blot Protein samples were run on either 6% polyacrylamide gel (Immunoprecipitation) or on 8% polyacrylamide gels (all other samples) in 1X Laemli electrode buffer. Proteins were transferred to Amersham Hybond P 0.45 mM PVDF membrane (catalog number: 10600023) using the Trans-blot turbo transfer system from BioRad. The semi-dry transfer took place in Bjerrum Schafer-Nielsen Buffer with SDS (48 mM Tris, 39 mM Glycine, 20% Methanol, 0.1% SDS) at 1 Amp 25 V for 20 min. Membranes were stained in Ponceau (0.2% Ponceau S, 2% Acetic Acid) for 20 min at room temperature as a loading control. The membrane was then briefly washed in 1XPBS and imaged on a Chemidoc MP Imaging system. Membranes were then washed three times for 5 min to remove excess ponceau. Membranes were then blocked for 1 hr in 5% non-fat dried milk diluted in 0.05% TWEEN-20 in 1XPBS. The membrane was then briefly rinsed and incubated in primary antibody diluted in 1% non-fat dried milk diluted in 0.05% TWEEN-20 for either 1 hr at room temperature or overnight at 4˚C. The membrane was then washed four times for 5 min in 1% TWEEN-20 diluted in 1XPBS. The membrane was then incubated for 1 hr at room temperature in secondary antibody diluted 1:5000 in 1% non-fat dried milk diluted in 0.05% Triton-X 100. The membrane was then washed four times for 5 min in 1% TWEEN-20 diluted in 1xPBS. Bioworld ICL (catalog number: 20810000-1) was used to visualize the proteins following the manufacturer's instruction. The chemiluminescence was imaged on a Chemidoc MP Imaging system. Band intensities were measured on ImageLab software. QPCR RNA was extracted using the Invitrogen PureLink RNA mini kit (Fisher catalog number: 12 183 018A) following the manufacturer's instructions. The optional on column DNA digest was performed using the PureLink DNase set (Fisher Catalog number: 12185010). The high-capacity cDNA reverse transcription kit (ThermoFisher catalog number: 4374966) was used to create cDNA using the manufacturer's protocol. 3 ml of cDNA created from 1 mg of RNA was used in downstream qPCR analysis. TaqMan gene expression assays were run following the manufacturer's directions. PCR was performed using the TaqMan Universal PCR Master Mix (ThermoFisher catalog number: 4364340). The reactions were run on an ABI 7500 Real-time PCR machine. Fly strains Multidimensional protein identification technology TCA-precipitated protein pellets were solubilized using Tris-HCl pH 8.5 and 8 M urea, followed by addition of TCEP (Tris(2-carboxyethyl)phosphine hydrochloride; Pierce) and CAM (chloroacetamide; Sigma) to a final concentration of 5 mM and 10 mM, respectively. Proteins were digested using Endoproteinase Lys-C at 1:100 w/w (Roche) at 37˚C overnight. The samples were brought to a final concentration of 2 M urea and 2 mM CaCl 2 and a second digestion was performed overnight at 37C using trypsin (Roche) at 1:100 w/w. The reactions were stopped using formic acid (5% final). The digested size exclusion eluates were loaded on a split-triple-phase fused-silica micro-capillary column and placed in-line with a linear ion trap mass spectrometer (LTQ, Thermo Scientific), coupled with a Quaternary Agilent 1100 Series HPLC system. The digested Not and control FLAG-IP eluates were analyzed on an LTQ-Orbitrap (Thermo) coupled to an Eksigent NanoLC-2D. In both cases, a fully automated 10-step chromatography run was carried out. Each full MS scan (400-1600 m/z) was followed by five data-dependent MS/MS scans. The number of the micro scans was set to one both for MS and MS/MS. The settings were as follows: repeat count 2; repeat duration 30 s; exclusion list size 500 and exclusion duration 120 s, while the minimum signal threshold was set to 100. Mass spectrometry data processing The MS/MS data set was searched using ProLuCID (v. 1.3.3) against a database consisting of the long (703 amino acids) isoform of non-stop, 22,006 non-redundant Drosphila melanogaster proteins (merged and deduplicated entries from GenBank release 6, FlyBase release 6.2.2 and NCI RefSeq release 88), 225 usual contaminants, and, to estimate false discovery rates (FDRs), 22,007 randomized amino acid sequences derived from each NR protein entry. To account for alkylation by CAM, 57 Da were added statically to the cysteine residues. To account for the oxidation of methionine to methionine sulfoxide, 16 Da were added as a differential modification to the methionine residue. Peptide/spectrum matches were sorted and selected to an FDR less than 5% at the peptide and protein levels, using DTASelect in combination with swallow, an in-house software. The permanent URL to the dataset is: ftp://massive.ucsd.edu/MSV000082625. The data is also accessible from: Proteo-meXChange accession: PXD010462 http://proteomecentral.proteomexchange.org/cgi/GetDataset? ID=PXD010462. MassIVE | Accession ID: MSV000082625 -ProteomeXchange | Accession ID: PXD010462. Original mass spectrometry data underlying this manuscript can also be accessed from the Stowers Original Data Repository at http://www.stowers.org/research/publications/libpb-1242. Acknowledgements Jerry L Workman and Susan M Abmayr for advice and guidance. Jakob Waterborg for contributing plasticware reagents and small equipment. Boyko Atanassov and Stone (Bayou) Chen for critical reading of early draft manuscripts. University of Missouri Faculty Scholars training program. Funding agencies -School of Biological Science, UMKC; University of Missouri Research Board; UMKC Students Engaged in the Arts and Research (SEARCH); UMKC Summer Undergraduate Research Opportunity (SUROP) scholars programs; funds provided by the University of Missouri -Kansas City Provost Strategic Initiatives Fund through Project ADVANCER -Academic Development Via Applied and Cutting-Edge Research for UMKC undergraduate and professional students from underrepresented minorities Program; and NIGMS grant 5R35GM118068 to JL Workman. The Drosophila Genomics Resource Center (NIH Grant 2P40OD010949) for reagents. The mouse anti SCAR (deposited by Susan Parkhurst and mouse anti Chaoptin antibody (deposited by Benzer, S and Colley, N) were obtained from the Developmental Studies Hybridoma Bank, created by the NICHD of the NIH and maintained at The University of Iowa, Department of Biology, Iowa City, IA 52242. Transgenic fly stocks and/or plasmids were obtained from the Vienna Drosophila Resource Center (VDRC, www. vdrc.at). Stocks obtained from the Bloomington Drosophila Stock Center (NIH P40OD018537) were used in this study. The following dataset was generated: Uptake and fate of Di-2-ethylhexyl phthalate in aquatic organisms and in a model ecosystem. Di-2-ethylhexyl phthalate (DEHP), often referred to as dioctyl phthalate or DOP, is the most widely used plasticizer for vinyl plastics, and approximately 350 million pounds were produced in the United States during 1970. Altogether the production of phthalate ester plasticizers was 855 million pounds (1). The cumulative production of DEHP and related phthalate plasticizers in the United States since 1943 is in excess of 8000 million pounds (2). DEPH is an external plasticizer which softens resins without reacting with them chemically, and it may be present in concentrations up to 40% of the weight of the plastic as in the familiar laboratory tubing. As a result of the large production and wide distribution and destruction of plastics, DEHP has become uniquitous and has been found in milk (3), deep frying fat (4), and human blood plasma (5). DEHP has also begun to appear as a micropollutant in the tissues of a variety of organisms. Taborsky (6) isolated it from bovine pineal glands, and Nazir et al. (7) found it in mitochondria from the hearts of cattle, dogs, rabbit, and rat. DEHP has been Di-2-ethylhexyl phthalate (DEHP), often referred to as dioctyl phthalate or DOP, is the most widely used plasticizer for vinyl plastics, and approximately 350 million pounds were produced in the United States during 1970. Altogether the production of phthalate ester plasticizers was 855 million pounds (1). The cumulative production of DEHP and related phthalate plasticizers in the United States since 1943 is in excess of 8000 million pounds (2). DEPH is an external plasticizer which softens resins without reacting with them chemically, and it may be present in concentrations up to 40% of the weight of the plastic as in the familiar laboratory tubing. As a result of the large production and wide distribution and destruction of plastics, DEHP has become uniquitous and has been found in milk (3), deep frying fat (4), and human blood plasma (5). DEHP has also begun to appear as a micropollutant in the tissues of a variety of organisms. Taborsky (6) isolated it from bovine pineal glands, and Nazir et al. (7) found it in mitochondria from the hearts of cattle, dogs, rabbit, and rat (8). Therefore DEHP has all the requisite properties to be classified as an environmental micropollutant (9). Our investigation was undertaken to study its metabolism and possible biomagnification in a variety of aquatic organisms and its ecological behavior in food chains of a laboratory model ecosystem (10). Expeimentl Procedures 14C Carbonyl-labeled DEHP was prepared from 10 mg phthalic 7-1 4C-anhydride, 9.2 mCi/mmole (New England Nuclear), by heating it in a sealed tube for 25 hr at 1300C with 100 Ml of 2-ethylhexyl alcohol and a trace of phthalic acid. The product consisted of 78% DEHP and was purified by column chromatography on silica gel and by subse-quent preparative thin-layer chromatography (TLC) on silicic acid with a solvent system of benzene, Skellysolve B, acetone, and acetic acid in proportions of 65:25:25:5 by volume. The final product used in these experiments had a radiopurity of more than 99% and was diluted to a specific activity of 5.56 mCi/mmole or 31,400 dpm/pg. An additional sample of 14C-labeled DEHP was purchased (from New England Nuclear) with a specific activity of 1.64 mCi/mmole. The degradation products of 14C-DEHP in the various animal and plant tissues were separated by TLC and autoradiographs were made on Eastman no-screen x-ray film. The identities of the various metabolites were determined by cochromatography with known standards; radioactive spots were removed from the TLC plates by elution in scintillation fluid and determined quantitatively by liquid scintillation counting. Uptake Studies These were carried out to detennine the uptake of DEHP directly from water by a variety of aquatic organisms and to study the metabolic transformations occurring over a short period. The organisms chosen were the water flea Daphnia magna, the mosquito larva Culex pipiens quinquefasciatus, the "fingernail" clam Sphaerium striatinum, the guppy Lebistes reticulatus,

and the aquatic plant Elodea canadensis. Standard reference water (11) was made from glass-distilled water, and 2.5-liter portions were placed in 5-liter battery jars. The 14C-labeled DEHP was dissolved in 1 ml of acetone and added quantitatively to the battery jars at concentrations of 0.1 and 10 ppm. Groups of organisms were exposed to the DEHP for various intervals as indicated in the tables. Model Ecosystem Study A laboratory model ecosystem with a terrestrial-aquatic interface and a sevenelement food chain has been found very useful in estimating the potential environmental effects of DDT and other pesticides (10), particularly in regard to ecological magnification and biodegradability. 14C-DEHP was evaluated in this system by applying 5 mg of the carbonyl-labeled compound to Sorghum plants on the terrestrial end of the system. After 33 days, the various organisms in the system were homogenized in water, extracted with diethyl ether, and the concentrated extracts resolved by TLC on silicic acid by use of benzene, Skellysolve B, acetone, and acetic acid in the proportions 65:25:25:5 by volume. The radioactive spots were located by autoradiography and the parent compound and metabolites determined quantitatively by liquid scintillation counting. PRsults and Discussion Uptake Studies Autoradiograms of the homogenized extracts of the organisms exposed to 1 4C-DEHP at 10 ppm for 1, 2, and 7 days are shown in Figures 1 and 2; a quantitative estimation of the distribution of the radioactivity in the various spots is given in Table 1. The data demonstrate the steady decrease of DEHP in the guppy from 88.5% of total radioactivity after 1 day, to 37.1% after 2 days and 16.8% after 7 days. There was a commensurate increase in the polar metabolites at the bottom of the chromatograms, from 11.5% after 1 day to 34.2% after 2 days and to 80.6% after 7 days. One of the identified metabolites was phthalic acid, which comprised 23.8% of the radioactivity after 2 days and then decreased to 4.8% after 7 days. Small amounts of a metabolite which is apparently phthalic anhydride appeared after 2 and 7 days. Uncontaminated guppies fed Daphnia with a 1-day uptake of 14C0 labeled DEHP showed the same degradation pattem ( A detailed study was made of the rate of uptake of 14C-DEHP by Culex larvae, Daphnia, snail, Elodea, and guppy exposed to 0.1 ppm and 10 ppm over intervals from 1 hr to 48 hr. The distribution of 1 4C in the organisms at the various intervals is shown in Table 2. From these uptake studies, the following conclusions can be drawn. Model Ecosystem Study During the 33-day period of the model ecosystem study, the concentration of 14 C in the aquatic phase reached a peak of 0.031 ppm at the fifth day after treatment and declined to 0.0077 ppm at the end of the experiment. This decline was clearly the result of the uptake of DEHP and its degradation products by the organisms of the model ecosystem as shown in Table 3. The distribu- products found in the model ecosystem, the major degradative pathways seem to be through hydrolysis of the ester groups to gX produce MEHP, then phthalic acid, and then phthalic anhydride. tion of radioactivity was clearly indicated by the autoradiograph in Figure 5. At the conclusion of the experiment, the water contained 0.00034 ppm DEHP, the algae 18.32 ppm (53,890X ) the snails 7.30 ppm (21,480X ), the mosquito larvae 36.61 ppm (107,670X), and the fish 0.044 ppm (130X). Thus the model ecosystem experiment was in good agreement with the uptake studies in showing that the mosquito larvae had the highest bioconcentration factor, and the fish the least. The autoradiographs (Fig. 5) showed the presence of almost nothing but DEHP in algae and mosquito larvae, indicating that these organisms have little degradative capacity. The snail produced substantial amounts of mono-2-ethylhexyl phthalate (MEHP), phthalic anhydride, and phthalic acid. The fish were more active in metabolism, and about half the total amount of 14C found in their bodies was a compound, Rf 0.65, which cochromatographed with phthalic anhydride and is presumed to be that compound reformed from phthalic acid. It is interesting to compare the model ecosystem studies of DEHP with that of DDT conducted under identical conditions (10). The experiments reported above demonstrate that DEHP is a microchemical environmental pollutant which is rapidly biomagnified by a variety of plants and animals in an aquatic system. DEHP is biodegraded very slowly in algae, Daphnia, mosquito larvae, snails, and clams and more rapidly in fish by hydrolysis at the ester bonds to form monoethylhexyl phthalate, phthalic acid, phthalic anhydride, and a variety of polar metabolites and conjugates. However, DEHP closely resembles DDT in rate of uptake and storage, and it obviously partitions strongly in the lipids of plants and animals and is concentrated through food chains. The biomagnification of DEHP together with its teratogenic properties and its enormous rate of production and ubiquitous use indicate the need for much further study of its environmental distribution and fate. Present data suggest the need for restrictions on the use and waste disposal of DEHP. Robust and Gradient Thickness Porous Membranes for In Vitro Modeling of Physiological Barriers Porous membranes are fundamental elements for tissue-chip barrier and co-culture models. However, the exaggerated thickness of commonly available membranes impedes an accurate in vitro reproduction of the biological multi-cellular continuum as it occurs in vivo. Existing techniques to fabricate membranes such as solvent cast, spin-coating, sputtering and PE-CVD result in uniform thickness films. To understand critical separation distances for various barrier and co-culture models, a gradient thickness membrane is needed. Here, we developed a robust method to generate ultrathin porous parylene C (UPP) membranes not just with precise thicknesses down to 300 nm, but with variable gradients in thicknesses, while at the same time having porosities up to 25%. We also show surface etching and increased roughness lead to improved cell attachment. Next, we examined the mechanical properties of UPP membranes with varying porosity and thickness and fit our data to previously published models, which can help determine practical upper limits of porosity and lower limits of thickness. Lastly, we validate a straightforward approach allowing the successful integration of the UPP membranes into a prototyped 3D-printed scaffold enabling in vitro barrier modeling and investigation of cell-cell interplay over variable distances using thickness gradients. Introduction Modeling, in vitro, the structure of specific cell complexes and their environments is of utmost importance to examine the molecular mechanisms governing cell-to-cell interplay and to perform robust pharmacological studies. [1][2][3] In particular, modeling physiological vascular barriers is of fundamental translational value, and novel solutions for in vitro study are continuously emerging with the goal of an effective integration to, or even to by-pass, in vivo approaches. When modeling vascular barriers, selecting the appropriate model cells is crucial as demonstrated by testing primary or immortalized endothelial cells and, more recently, humanderived pluripotent stem cells. [4,5] However, this cellular-level advancement is not being matched with a research effort examining equally important aspects of in vitro modeling, specifically the development of porous membranes to mimic the nano-to-microscale level cell-to-cell interplay. [6] Literature indicates that porous membranes are an essential component for compartmentalized cell co-cultures and to support a tissue barrier. [6][7][8][9][10][11][12][13][14][15] However, most barrier models incorporate commercially available track-etched membranes such as those found in transwell, fiber-based, or micro-fluidic systems with a high thickness compared to the basal laminae in vivo (>10 µm vs. ~300 nm). By using available materials, the porosity cannot be adequately engineered, leading to low porosity, random pore distribution, and limited size control, potentially biasing physiological multi-cellular interplay, especially during co-culture conditions. [16,17] The ideal membrane should allow necessary cell-cell communication and transport while providing the opportunity to modulate cell-substrate interactions through a change in porosity and pore geometry. [18][19][20] In successful efforts to provide these properties, inorganic or polymeric ultrathin films have been developed and tested, providing physiologically relevant thickness, optical transparency, and controlled pore size. [6,9,13,15,[21][22][23][24] However, these solutions present significant fabrication or technical stumbling blocks that impede their integration into the in vitro modeling realm. Limited active area for cell culture, time-consuming and costly process of backside etching of silicon wafers, and the use of toxic or corrosive solvents are among issues that need to be resolved. Importantly, existing techniques to fabricate membranes such as solvent casting, spin-coating, sputtering, and plasma-enhanced chemical vapor deposition (PE-CVD) result in uniform thickness films. [25][26][27][28] In order to understand critical separation distances for various barrier and co-culture models, a gradient thickness membrane is needed. However, to the best of our knowledge, a continuous thickness gradient membrane that can be used for quantifying distance-dependent cell-cell communication in various tissue-on-a-chip and human barrier models have not been developed nor implemented. Parylene C has been proposed as a material solution compatible with in vitro cell attachment and biological response, with a relatively low-cost fabrication over large surface areas with tunable thicknesses. [29][30][31][32][33][34][35] Current protocols for parylene etching needs significant optimization, as they result in loss of pattern fidelity by changing membrane thickness. [36][37][38] Therefore, precise pore geometry requires addition and subsequent removal of a hard mask which can be a time and resource expensive process. [29,35] Parylene also tends to curl as a thin film as a result of thickness-dependent stress variation, which requires the addition of an undesirable and interfering chip or grid-based support. In the present study, we developed facile methods to generate ultrathin porous parylene C (UPP) membranes not just of precise thicknesses down to 300 nm, but with variable gradients in thicknesses, all with relatively high porosities. We examined the mechanical properties of UPP membranes with varying porosity and thickness and fit our data to available models. We developed a straightforward approach to reinforce the ultrathin membranes while maintaining a large uninterrupted central region and integrated the UPP membrane into a prototyped 3D printed scaffold for tissue-chip devices. Membrane Fabrication and Scanning Electron Microscopy Characterizations In order to maximize reproducibility, we sought to develop a simplified and reliable process for UPP membrane fabrication (Figure 1). Of the tested candidates for sacrificial layers, Micro-90 was chosen due to a straightforward lift-off process that can be accomplished within seconds by adding DI water. Parylene C deposition at a base pressure of 10 mTorr and deposition pressure of 25 mTorr resulted in conformal coating thickness across 6" wafers, and measured thicknesses were obtained by spectrophotometry and confirmed by profilometry. A combination of a top anti-reflective coating (TARC) and an optimized etching recipe was used to obtain a significantly lower pore expansion across samples, which were verified using scanning electron microscopy ( Figure S1 and S2). Our results are analogous to previous works that included the use of hard masks that are costly, time-consuming, and require a chemically harsh removal step, leaving metallic residuals on the membranes and negatively impact subsequent use for cell culture. The lifted-off membranes are supported by SU-8 frames that provide large active areas (>1 cm 2 ) and are easily transferable using tweezers (Figure 2A). SU-8 supports are used for easier handling and minimizing the deflection of membranes during transfer and use in cell culture. Figure 1. Microporous UPP membrane fabrication and introduction of a simple SU-8 frame to support robust handling and transfer. A step-by-step process for UPP membrane fabrication to achieve reliable pore geometries matching the mask design. Pore size fidelity relative to the mask design across three pore sizes (1, 3, and 8 µm) and two membrane thicknesses (300 nm and 1500 nm) revealed minimal pore size expansion under an optimized fabrication process ( Figure 2B). Excellent pore size fidelity was observed in 300 and 1500 nm thick membranes with 3 and 8 µm pore sizes, and also in 300 nm thick parylene C membranes with 1 µm pore size. The 1500 nm thick parylene C membranes with 1 µm pore size showed an average pore expansion of 25%, which can be corrected by a change in mask design to account for the effect of higher thickness. SEM imaging shows uniform patterning across large areas with no noticeable defects ( Figure 2C). The imaging also demonstrates the successful transfer of the pattern from the photoresist layer to the membrane layer with sharp edges ( Figure 2D). Thickness Gradient Membranes A simple and inexpensive, custom-made polymethyl methacrylate (PMMA) chamber provides a reproducible

system for creating a thickness gradient in the deposited membranes across the silicon wafer substrate, and the gradient can be easily tuned by changing the chamber width ( Figure 3A). By modifying the opening distance between 1 and 4 mm, a thickness gradient of more than 200 nm/cm is achieved across a 2 cm vertical distance of the wafer ( Figure 3B). An increase in the opening distance has a nearly linear correlation with the decrease in thickness gradient. A major challenge with fabricating porous membranes with a thickness gradient is that during the etching process the time required to create pores in the thicker regions will result in over-etching the thinner regions, leading to the expansion of the pores in the thinner regions. However, desirable pore size consistency was observed across the thickness gradient membrane, likely attributable to the combination of anisotropic etching and the use of TARC to minimize UV reflectance effect ( Figure 3C). . Thickness gradient membrane fabrication using a custom-made PMMA device. A) Illustration of the PMMA device designed with an adjustable top opening as the only path for parylene deposition on the enclosed silicon wafer. The restricted diffusion path for parylene monomers before polymerization at the wafer surface leads to a linear decrease in thickness from the top of the wafer down. B) Increases in the PMMA device opening results in a nearly linear decrease in thickness gradient across the membrane. C) Thickness difference across a frame-supported UPP membrane with a length of 2 cm. SEM images of different sites of frame supported membrane reveal a consistent pore size and distribution across different thicknesses (scale bar=1 μm) Mechanical Testing We next investigated the mechanical properties of UPP membranes at varying thickness and porosity. Previous studies that modeled mechanical properties of porous materials progressed from the simplistic assumption that no stress-concentration arises from pores (i.e. ultimate strength is correlated with volume fraction) to empirical models for material properties such as ultimate strength (σu), such as the following equation: where !" and are the ultimate strength and porosity of the porous material, respectively. [39][40][41][42] Recent studies have performed various simulations and modeling and combined these to obtain a more accurate representation of material properties. [43][44][45][46][47] They adopted the following equation: where ' is obtained through estimation and can be approximated as 2 based on theoretical and experimental data. [44,45] Simplifying the equation yields: Although these models have been tested experimentally for characterizing the strength of a variety of materials with different shapes, such experimental validation has not been tested on ultrathin porous membranes. While there have been several notable studies that have investigated the mechanical characterization and tensile properties of non-porous thin films, [48][49][50][51] we believe this is the first examination of the implications of introducing porosity in parylene nanomembranes. Previous studies on the mechanical properties of porous nanomembranes primarily used nanoindentation and bulge testing due to the challenge of conducting traditional tensile testing on ultrathin materials. [52][53][54] To evaluate the mechanical properties of our UPP membranes, we applied a SU-8 grip support layer to both ends of membrane test strips, which enables evaluation of the membranes with a commercial tensile tester ( Figure S3). A summary of the results from the tensile tests is presented in Figure 4. Both the ultimate and yield strength follow a second-order dependence on the solid fraction of the membranes (1 − ), showing good agreement with the model described in Equation 3. Results for the yield strength are included here because this property is relevant to the membrane application of tissue culture. The intended purpose of UPP membranes is to act as a nominally two-dimensional substrate to enable facile imaging of cells using optical microscopy. If the membranes undergo significant plastic deformation from the stresses of handling and device integration and cell culture processes, then the membrane would no longer be flat and would diminish the ease of planar cell imaging. Considering the apparent second-order dependence of the membrane strength on the solid fraction (i.e., (1 − )), it is clear that the membranes should quickly become too weak at higher porosities. However, both the 5% and 25% porous membranes display a strength that is robust with respect to the stresses experienced in lift-off, mounting into tissue culture devices and cell culturing activities. Our experience even showed that membranes with porosities as high as 60% could be successfully lifted-off and transferred to cell culture devices, but subsequent handling to perform mechanical testing was not feasible due to their fragility. We targeted up to 25% porosity because it is more than double what is commonly available in track-etched membranes. Furthermore, we previously demonstrated that porous membranes with 25% porosity modulate cell-substrate interactions and cell behavior like softer, non-porous materials with Young's moduli similar to in vivo tissue values. [18] Theoretically, maximum elongation for all conditions should remain the same under ideal conditions. However, various defects can be introduced by increasing porosity and decreasing thickness that may not be accurately predicted nor avoided. This is even more pronounced when dealing with porous membranes of submicron thickness. Test results of the UPP membranes show that elongation at failure is essentially independent of the membrane thickness at all porosities. In terms of the effects of thickness on mechanical properties, no significant difference is observed between 300 nm and 1500 nm thicknesses for ultimate stress and elongation at failure. Additionally, the force required for 5% elongation is seen to scale directly with membrane thickness as expected. In summary, mechanical characterization reveals that the strength characteristics of the UPP membranes follow an apparent second-order dependence on the solid fraction, which agrees with previous modeling and experimental work. The ultimate and yield strength exhibited by the 25% porous membrane is still at 50% of that of the non-porous one, and this is sufficient to ensure that the porous membrane is robust enough to handle the stresses of fabrication and cell culture. C) Elongation of the membrane at failure shows remarkable retention of resistance against mechanical strain after introducing 25% porosity while yielding no significant difference among different thicknesses. D) Force at a given elongation (5%) reveals the expected outcome that force is directly proportional to the thickness, and force decreases with increasing porosity, similarly to ultimate stress results. Surface Roughness and Improved Cell Adhesion Culture substrates are routinely modified to improve cell adhesion. Some approaches rely upon adsorption or attachment of extracellular matrix proteins or peptides, which can be expensive and have limited shelflife. [55,56] Alternatively, tissue culture polystyrene (TCPS) substrates are routinely treated with plasma. [67] In order to improve cell adhesion to the UPP membranes, we explored treating the surface with oxygen plasma. We exposed membranes to two reactive ion etching (RIE) oxygen plasma powers (100 W and 200 W) and measured endothelial cell attachment after storing the membranes in atmospheric conditions for 1, 7, or 14 days. Both 100 W and 200 W resulted in significant improvements in cell adhesion over untreated membranes, increasing adhesion to levels similar to TCPS, which is consistent with previous studies involving surface modification of parylene substrates for improved cell adhesion (Figure 5). [57,58] Another important aspect of these surface modifications is persistence over time. The advantages of the oxygen plasma treatment on some surfaces can be transient with hydrophobic recovery occurring within 24 hours. [59] In our studies, we found treated UPP membranes maintained high levels of cell adhesion even after 14 days of storage in atmospheric conditions at room temperature. Due to the persistent improvement in cell adhesion, we hypothesized that our plasma treatments may have induced a permanent physical change to the membrane surface. We used atomic force microscopy (AFM) to examine surface roughness of the UPP membranes before and after RIE plasma treatment. Our data show RIE treatment results in a rougher surface that may explain the improved and persistent cell adhesion (Figure 6). The changes are on a length scale that likely provides a greater surface area for protein adsorption, and may inhibit collagen motility, resulting in aggregation of collagen fibrils which can significantly increase cell gripping and adhesion as opposed to smooth collagen layer on smooth surfaces. [57,[60][61][62] Potential applicability of UPPM to flow-based cell culture devices Here, we aimed to show technical feasibility for UPP membrane integration into a prototype 3D-printed scaffold system for flow-based cell culture (Figure 7). To this end, frame-supported membranes were successfully seated on the top surface of the 3D-printed chamber ( Figure 7B). Feasibility tests showed that the UPP membrane remained stable in place and did not tear when the flow was applied at a range of 1 to 20 ml/min using a pulsatile pump. When the pulsatile flow is applied, the circulating media did not diffuse underneath the UPP membrane. Next, we tested whether a viable endothelial cell culture is attainable when UPP membranes are integrated into the 3D-printed chamber. Under these conditions, endothelial cells attached and grown to confluence within 3 days of seeding ( Figure 7C-D). Phalloidin and nuclei DAPI staining showed physiological cytoskeletal organization and cell morphology on the UPP membranes ( Figure 7E). Tight junction protein staining ( Figure 7F) indicates the development of continuous endothelial intercellular contacts supporting the potential applicability of UPP membranes to cell transmigration or drug transport studies. Taken together, these results advocate for the full development and testing of UPP membrane-based in vitro culture or coculture devices, including applicability to flow-based systems. Conclusion In summary, we developed methods to efficiently fabricate robust UPP membranes that can be used for viable cell culture and integrated into prototyped in vitro devices. UPP membranes can be produced with variable and even gradient thicknesses to investigate cell-cell interactions. Our study on membrane thickness, porosity, and mechanical properties demonstrated agreement with the existing empirical models used to predict practical upper limits of porosity and lower limits of thickness. Lastly, we showed that oxygen plasma etching improves cell adhesion on UPP membranes. Further studies are now needed to prove the biological advantages and the in vivo mimicry associated with the integration of UPP membranes into in vitro devices or chips. Experimental Section Sacrificial layer deposition and parylene coating: Micro-90 was used as a water-soluble sacrificial layer. This was achieved by spin-coating the solution on top of a 6-inch (100) silicon wafer (500 rpm for 3 s, followed by 3000 rpm for 45 s). The last step in the fabrication process requires simply exposing the attached membrane to Deionized (DI) Water, which dissolves Micro-90. A solution of Micro-90 diluted to 5% in DI water was identified as the optimal concentration. Although higher concentrations of Micro-90 led to much easier detachment, they often resulted in delamination during the fabrication process. Parylene-C coating was done using DPX-C dimer (Specialty Coating Systems, USA) in an SCS Labcoter® 2 parylene deposition system (PDS 2010, Specialty Coating Systems, USA). Briefly, silicon wafers are placed in the deposition chamber with an optimized separation distance (minimum of 10 cm) to ensure thickness uniformity across the wafer, and dimer is loaded in the vaporizer (nominally 800 nm final parylene C thickness per 1 gram of dimer). The process begins at a base chamber pressure of 10 mTorr, and the dimer-cracking furnace is heated to 690 °C. Then, the vaporizer is ramped to a final temperature of 175 °C, causing the sublimation of dimer. The temperature ramp rate of the vaporizer is controlled to maintain a chamber pressure of 25 mTorr. Thickness measurements were performed by both Tencor P2 long scan profilometer (KLA-Tencor, USA) and NanoSpec Spectrophotometer (Nanometrics Incorporated, USA). Membranes with a thickness gradient across the diameter of the wafer were obtained by placing the wafer in a custom-made polymethyl methacrylate (PMMA) fixture ( Figure 3A) for the deposition. The PMMA fixture creates a narrow slot that restricts the diffusion of the parylene monomers and thereby reduces the likelihood of the monomers reaching the wafer surface toward the bottom of the fixture. The gap between the wafer and the fixture wall can be adjusted allowing for a range of thickness variations. I-line photolithography: HDMS prime (Microchemicals, Germany) was used as an adhesion promoter between parylene and photoresist, and was applied by spin-coating at 3000 rpm for 45 s followed by baking at 140 °C for 1 min. AZ MiR 701 positive photoresist (14 cPs) (Microchemicals, Germany) was applied by

spin-coating at 1500 rpm for 45 s followed by soft-baking at 90 °C for 1 min to obtain a thickness of 1.5 µm. To prevent UV light reflection which leads to undesired pore size change, AZ Aquatar (Microchemicals, Germany) was used as a top anti-reflective coating (TARC) with a manufacturer-recommended spin-coating recipe of 500 rpm for 3 s followed by 2200 rpm for 30 s. The advantage of using TARC over bottom antireflective coating (BARC) such as AZ Barli II is the removal during the developing step without the necessity of separate etching steps while obtaining comparable results. Photoresist exposure was performed on an ASML PAS 5000 i-line 5X Stepper (ASML, Netherlands). In order to save time and efforts and minimize associated costs, multiple designs with different final pore sizes and pore spacings (1 µm, 1 µm, 3 µm, and 8 µm pore size with 2 µm, 8 µm, 6 µm, and 16 µm center to center spacings, respectively) were embedded in a single mask. Multiple parameters including exposure dose and focus offset were adjusted and tested for each patterned image. The wafers then went through a post-exposure bake at 110 °C for 1 min. The exposed photoresist was developed using Microposit MF CD-26 developer (Microchemicals, Germany) followed by hard-baking at 140 °C for 1 min. All photolithography steps were performed while the ambient cleanroom humidity was within a 38-42% range. Reactive ion etching: Previous works involving etching parylene mainly used Reactive Ion Etching (RIE) with Oxygen to etch parylene. However, creating high-resolution patterns with oxygen etching proved to be challenging due to a generally higher etching rate of the photoresist versus the parylene and isotropic etching of both the parylene and photoresist layers. A variety of etching recipes were evaluated using a Trion Phantom 3 system (Trion Technology, Inc., USA), and a combination of 50 sccm of Oxygen (O2), 5 sccm of Sulfur hexafluoride (SF6), and 5 sccm of Argon (Ar) proved to be an optimal etching recipe ( Figure S1). The chamber pressure was kept at 50 mTorr and power was set at 175 W with a tolerance of 2 mTorr and 5 W, respectively. The residual photoresist was removed by immersing that wafer in the PRS-2000 solution (JT Baker, Inc. USA) at 75 °C for 10 min. Membrane frame design and embedment: To enable easier transfer of the UPP membranes after lift-off, a "scaffold structure" was patterned on top of the UPP membrane through contact lithography of SU-8 to obtain frames with different designs. Wafers were dehydrated at 150 °C for 10 min. Then, SU-8 3025 was deposited with a spin-coating recipe of 500 rpm for 10 s followed by 3000 rpm for 45 s, soft-baked at 95 °C for 10 min, and transferred to a cool plate, which resulted in a SU-8 layer with 20 µm thickness. The wafers were exposed in contact lithography to broadband spectrum UV light for exposure of 250 mJ/cm 2 . Post-exposure bake consisted of 65 °C for 1 min, cool plate for 10 min, and 95 °C for 5 min. Wafers were developed using SU-8 developer (Microchemicals, Germany) for 5 min with mild agitation. The wafers were then rinsed with IPA for 10 sec, dried, and given a final hard bake at 120 °C for 150 s. Membrane lift-off and characterization: The wafers were cleaved into nominal 2 cm by 3 cm samples before membrane lift-off. The cleaved wafers were simply soaked in DI water for 1-2 minutes which dissolved the sacrificial layer and facilitated membrane lift-off. The frame-supported membranes were lifted with tweezers to transfer to other containers for subsequent characterization or incorporation into cell culture devices. The physical properties of the membranes were characterized by scanning electron microscopy (SEM) analysis. Mechanical characterization: In order to quantify the impact of porosity, pore size, and thickness on the mechanical properties of the porous membranes, 1 cm ⨉ 3 cm rectangles of UPP membranes were fabricated with thick SU-8 support regions covering 5-mm strips on either end to serve as gripping area supports. A CellScale® UniVert Biomaterial Tester (CellScale, Canada) equipped with a 10 N load cell was used for mechanical tensile testing. The SU-8 regions were held between Polyoxymethylene (POM) clamps specifically designed for soft materials. CellScale analysis software was used to extract ultimate stress, elongation at failure, and force at 5% elongation for three different porosities (0%, 5%, and 25% porosities) and two thicknesses (300 nm and 1500 nm). A minimum of five samples was used for each condition. RIE oxygen plasma treatment was used to modify the surface for better cell attachment. After completion of fabrication steps and before lift-off, membranes were exposed to RIE oxygen plasma in the Trion phantom III system (40 sccm, 50 mTorr) with 100 and 200 W power. Membranes were washed in 70% ethanol for 30 min and subsequently in ultrapure water for 30 min, all done in a laminar-flow biosafety cabinet after lift-off. Cells were seeded on the membranes, and tissue culture polystyrene (TCPS) as control with a density of 5000 cells/cm 2 . TCPS served as a typical standard cell culture substrate with adequate cell adhesion. The total number of attached cells was counted on membrane and control samples were counted after 6 hr. Although oxygen plasma treatment has been previously used for cell culture membranes for enhancing cell attachment, its effects and stability over time have not been quantified. To quantify its effects on UPP membranes, cell seeding was performed after 1, 7, and 14 days after oxygen plasma. Device fabrication and assembly: Cell culture devices were designed using SolidWorks® software and printed using Form 2 Stereolithography (SLA) printer (Formlabs Inc., USA) with a layer precision of 25 μm. This was followed by thorough washing with isopropyl alcohol (IPA) and curing at 60 ℃ for 45 min. After membrane lift-off using DI water, they were transferred using a tweezer to the 3D printed device. PBS was added on top of the membrane and left to dry to promote adhesion between the membrane and the device. Cell culture and immunofluorescence: Bovine aortic endothelial cells (BAECs) were cultured in Dulbecco's Modified Eagle Media (DMEM) supplemented with 10% FBS, 1% L-glutamine, 100 μg/ml penicillin, and 100 μg/ml streptomycin. Cells were detached and sub-cultured per manufacturer's instructions using TrypLE. BAEC media was exchanged every 2-3 days and cells were passaged at 80% confluence. BAECs were used between passages 3-4. Cells were seeded on the samples with a seeding density of 10,000 cells/cm2 in a 200 μl cell solution and left for 4 hr before flooding with media. After 3 days, cells were fixed with 3.7% formaldehyde for 15 minutes, washed three times with PBS, and then permeabilized with 0.1% Triton X-100 for 3 minutes. Cells were blocked with 2% BSA for 15 minutes and again washed with PBS. For visualization of nuclei and stress fibers, cells were stained with DAPI (300 nM), and 1:400 AlexaFluor 488 conjugated phalloidin. For visualization of tight junctional protein ZO-1, the cells were stained with 1:100 AlexaFluor488 conjugated anti-ZO-1/TJP1, Clone ZO1-1A12 (Affymetrix eBioscience, San Diego, CA). CD95 Structure, Aggregation and Cell Signaling CD95 is a pre-ligand-associated transmembrane (TM) receptor. The interaction with its ligand CD95L brings to a next level its aggregation and triggers different signaling pathways, leading to cell motility, differentiation or cell death. This diversity of biological responses associated with a unique receptor devoid of enzymatic property raises the question of whether different ligands exist, or whether the fine-tuned control of CD95 aggregation and conformation, its distribution within certain plasma membrane sub-domains or the pattern of post-translational modifications account for this such broad-range of cell signaling. Herein, we review how the different domains of CD95 and their post-translational modifications or the different forms of CD95L can participate in the receptor aggregation and induction of cell signaling. Understanding how CD95 response goes from cell death to cell proliferation, differentiation and motility is a prerequisite to reveal novel therapeutic options to treat chronic inflammatory disorders and cancers. INTRODUCTION Many Tumor necrosis factor (TNF) receptor superfamily members display significant roles in the progression of human diseases, such as the death domain (DD)-containing receptors including CD95, TNF-related apoptosis-inducing ligand receptor (DR4 and DR5), TNFR1, DR3, DR6, nerve growth factor receptor (NGFR), and ectodysplasin receptor (EDAR, Figure 1A). These receptors are characterized by the presence of an intracellular DD, which is required for their apoptosisinducing activity (Dostert et al., 2019). Several of them, including CD95 and TNFR1, are known to form multimers not only in the presence but also in the absence of their cognate trimeric ligands Siegel et al., 2000), rendering complex to determine the nature and role of their aggregation in the cell signaling process. This review discusses how the CD95 stoichiometry is controlled by receptor-dependent and independent processes, and how stoichiometry can affect the implementation of apoptotic or non-apoptotic signals. A UNIQUE CD95 RECEPTOR BUT AT LEAST TWO FORMS OF THE LIGAND CD95 CD95 (also known as Fas, Apo-1, TNFRSF6) is a 319 amino acid type I transmembrane glycoprotein (Itoh et al., 1991; Figure 1B). In the presence of its ligand CD95L, the receptor interacts with the adaptor protein Fas-associated protein with death domain (FADD) through homotypic DD-mediated interactions. FADD in turn recruits the protease caspase-8 and the long form of the regulator of apoptosis cellular FADD-like interleukin-1-β-converting enzyme-inhibitory protein (cFLIP L ) via death effector domain (DED) homotypic binding. Together, these proteins form a complex designated DISK for death-inducing signaling complex (Kischkel et al., 1995). The initial steps of CD95-DISK formation are quite well defined and some of them are shared with other death receptors of the TNFR superfamily. Although initially described as a pure death receptor, CD95 undergoes a paradigm change, which might lead to a therapeutic revolution. Indeed, cumulative evidence support that CD95 is not only able to trigger a cell death signal but can also promote inflammation and normal and tumor cell growth and migration through the implementation of non-apoptotic cellular functions including PI3K, NFkB, and JNK MAPKs (Desbarats and Newell, 2000;Desbarats et al., 2003;Kleber et al., 2008;O' Reilly et al., 2009;Hoogwater et al., 2010;Tauzin et al., 2011;Gao et al., 2016;Poissonnier et al., 2016). Members of DISK including FADD and caspase-8 could also participate in the induction of these non-apoptotic cell signaling pathways (Barnhart et al., 2004;Kreuz et al., 2004). Notably, caspase-8 acts through its scaffolding function to drive cytokines production in various cancer cell lines upon CD95L stimulation (Henry and Martin, 2017). Production of pro-inflammatory chemokines in dying cells results in the recruitment of monocytes and neutrophils that engulf the dying cells expressing the "find me" signal (Cullen et al., 2013). How CD95L triggers these apoptotic and non-apoptotic signaling pathways and their respective biologic functions remain to be better understood. CD95L CD95 ligand also known as CD95L (FasL, TNFSF6 or CD178) is a type II transmembrane protein with a long cytoplasmic domain, a transmembrane (TM) domain, a stalk region, a TNF homology domain (THD) that mediates homotrimerization and a C-terminal region involved in CD95 binding ( Figure 1C). The TM CD95L (membrane-CD95L or m-CD95L) can be cleaved in its stalk region by several matrix metalloproteases (MMPs) including MMP3, MMP7, MMP9, a disintegrin and metalloprotease-domain-containing protein (ADAM)-10 ( Guegan and Legembre, 2018). The resulting soluble form of CD95L (s-CD95L) is a homotrimer ) whose binding to CD95 fails to induce cell death (Suda et al., 1997;Schneider et al., 1998). Although the pathophysiological roles of s-CD95L remain to be elucidated, it accumulates in the bloodstream of patients suffering from a variety of diseases, including certain cancers such as NK cell lymphomas (Tanaka et al., 1996), ovarian cancers (De La Motte Rouge et al., 2019), and triple-negative breast cancer (TNBC) (Malleter et al., 2013). In TNBC women, high concentrations of s-CD95L are associated with the risk of relapse and metastatic dissemination (Malleter et al., 2013). s-CD95L levels are also elevated in inflammatory and autoimmune disorders such as systemic lupus erythematosus (SLE) (Tauzin et al., 2011;Poissonnier et al., 2016), rheumatoid arthritis (RA) (Hashimoto et al., 1998), and acute lung injury (ALI) (Herrero et al., 2011). CD95 STRUCTURE CD95 is detected homotrimerized independently of the presence of its ligand (Papoff et al., 1996;Siegel et al., 2000). Different domains in the death receptor seem to contribute to its aggregation, including the cytoplasmic DD (Ashkenazi and Dixit, 1998), TM

and extracellular regions. Due to the TM nature, aggregation propensity and domain flexibility, the whole CD95 structure has not been solved yet. Nevertheless, 3D structures of some parts of the receptor have been deciphered by electron microscopy, X-ray crystallography or NMR spectroscopy (Figure 2A). Although CD95 structure has been extensively studied by these biophysical methods, the conformation of some important domains within the receptor, including a part of the preligand assembly domain (PLAD) and the calcium-inducing domain (CID) (Figures 1B, 2A,B) are absent from these pictures, precluding a comprehensive understanding of the CD95-mediated cell signaling. Extracellular Region The extracellular region of TNF receptors is characterized by the presence of cysteine-rich domains (CRDs), which contain six cysteine residues engaged in the formation of three disulfide bridges (Bodmer et al., 2002). The number of CRDs in a given receptor varies from one to four, and CD95 encompasses 3 CRDs (Figures 1B, 2C; Bodmer et al., 2002). The repeated and regular arrangement of CRDs confers an elongated shape to the receptors. In the absence of stimulation, CD95 is found at the plasma membrane as monomers or homodimers and homotrimers associated through their respective extracellular N-terminal PLAD, encompassing the amino acid residues 17-82 (or amino acid residues 1-66 according to the mature protein) (Papoff et al., 1999;Siegel et al., 2000). Accordingly, the elimination of PLAD in DDmutated CD95 constructs abrogates their dominant-negative inhibitory effect, while expression of PLAD alone exerts a dominant negative action on the CD95-mediated apoptotic signal (Papoff et al., 1999;Siegel et al., 2000). More precisely, the minimal domain required for CD95 homotypic interaction contains amino acids 59-82 (43-66 without the peptide signal) (Edmond et al., 2012). The structure of CD95 extracellular domain (ECD) has been determined after complexation with a Fab fragment of agonistic (i.e., apoptotic) anti-CD95 antibodies, bound to the CRDs 1 and 2 (Chodorge et al., 2012). The CD95/Fab complex is monomeric (Figure 2C), and although the antibody is an agonist, it shares only a short region with the CD95/CD95L interface, mainly the arginine at position 102 (based on the human CD95 precursor sequence with 16 amino acids subtracted to obtain the position on the mature sequence corresponding to R86, see Figure 1B). CD95 ECD exhibits a linear organization and its putative orientation to the membrane is predicted based on the solved Holo trimeric structure of other members of the TNFR family, including DcR3/CD95L (PDB:4MSV), DR5/TRAIL (1D0G), TNFR2/TNF (3ALQ), and DcR3/TL1A (3MI8). The PLAD residues consist of amino acids 17-82 (1-66 without the peptide signal) (Papoff et al., 1996;Siegel et al., 2000) and part of this region is missing from the X-ray data, i.e., amino acids arginine 17 (first amino acid following the peptide signal) to histidine 54 ( Figure 1B). Also, the CD95 domains encompassing amino acid residues K148 to E156 and K164 to S170 are absent from X-ray and NMR analyses (Figure 2A). Threading approaches (Dunbrack, 2006) previously allowed us to build some plausible models of a completed ECD (including PLAD) in a trimeric organization (Levoin, 2017). However, in the present work focused on understanding of the trimeric assembly, we preferred not modeling a region whose structure is not strongly supported by experimental evidence. Therefore, to fill the two main gaps within the CD95 extracellular structure, we interrogated the PDB using PISA (Krissinel and Henrick, 2007) to find 3D homologs to the CD95 crystal structure (PDB : 3TJE). This method seems more appropriate than classical sequence-based screen, because sequence homology between DRs is rather low, and it is accepted that structure is more conserved than sequence (Murzin, 1998). Using this approach, we found that the structure of CD40 ECD (PDB : 3QD6) was close to that of CD95, with good geometric superposition and extended sequence solved (PISA Qscore = 0.55, with RMSD = 1.4 Å for 87 amino acids). Therefore, we completed the structure of CD95 protomer (residue N48 to E167, Figure 2B) using CD40 as a structural template. Afterward, each protomer served to assemble homotrimers through protein-protein docking, using SymmDock software (Schneidman-Duhovny et al., 2005b). We imposed a symmetric nature of the complex as a constraint, and the predicted four best solutions are presented Figure 3. These associations can be described as close Nt/remote Ct (model 1), close Nt/close Ct (model 2), remote Nt/close Ct (model 3), and remote Nt/remote Ct (model 4). The second solution is definitely the most compatible with biochemical data, because PLAD is known to be necessary and sufficient for receptor aggregation (Papoff et al., 1996;Siegel et al., 2000), and model 2 is the only one that orientates the amino terminal region in a way that can draw a large interface between PLADs. According to this computer-driven model, the interface between protomers occurred mostly between amino acid residues N48 to L52, K61 to P65 in CRD1, E114 to N118 in CRD2, R128 to V139 and C146 to E167 in CRD3. A noticeable structural feature of all trimer models is that protomers are tilted to the membrane, with an angle of about 45 • for model 1 to 9 • for model 2. It is noteworthy that the homotypic PLAD affinity is rather low, almost in the mM range (Cao et al., 2011) suggesting that other domains in CD95 could exert a complementary role for the receptor homotrimerization, in agreement with the proposed trimeric model. Domains of a monomeric CD95 whose structure has been experimentally solved. The plasma membrane is symbolized by two parallel planes, with the outer leaflet in purple and the cytosolic couleur in green. Note that the orientation toward membrane is a hypothesis. (C) Structure of the extracellular domain of CD95. Crystal structure of CD95 ECD domain (PDB:3TJE), colored according to the sequence order (blue to red, from Nt to Ct extremities). The yellow structure (amino acid residues N31 to D55) represents the gap in the crystal structure, which has been completed using CD40 homology. Nt: Amino-terminal region; Ct: COOH-terminal region. Transmembrane Domain (TM) Recent studies highlight that for several death receptors, the TM domain is involved in their aggregation. For example, DR5 TM helix promotes the assembly of high-order complexes responsible for cell death induction, independently of the ectodomain. Nuclear magnetic resonance (NMR) analysis of this TM region in bicelles shows different trimerization and dimerization interfaces responsible for a supramolecular dimer-trimer network (Pan et al., 2019). Surprisingly, elimination of the DR5 ECD triggers cell death in a TM-dependent and ligand (TRAIL)-independent manner, suggesting that the extracellular region of DR5 exerts an inhibitory action on the receptor activation; TRAIL binding overcoming this auto-inhibitory process (Pan et al., 2019). The CD95 TM domains have also been investigated by NMR in lipid/detergent bicelles (Fu et al., 2016) and these structures are found associated as stable trimers ( Figure 4A). While the ends of the three helices display a certain flexibility, their core was more rigid (Figure 4B). The amino terminal portions of the helices (extracellular side) are closer and less flexible than their C-terminal counterparts (mean d1 = 9.7 ± 0.5 Å vs. d2 = 18.2 ± 2.6 Å, respectively, between L174 or V195 Cα of each protomer, for the 15 NMR structures). Unlike DR5, a proline motif is present in CD95 TM and in many members of the TNFR superfamily, including TNFR1, DR3, DR4, and CD40. This proline-rich sequence (P183 and P185 residues in the human sequence) within the CD95 TM helix favors packing of CD95 protomers through van der Waals interactions (Fu et al., 2016). Transmembrane mutants affect the CD95L-mediated cell death program to a lesser extent than DD mutants (Fu et al., 2016). Indeed, co-expression of wild type and TM mutants does not disturb the formation of CD95 homotrimer in the absence of CD95L, indicating that the TM region of CD95 does not participate in its pre-ligand association (Fu et al., 2016). Nonetheless, CD95 TM domain probably stabilizes CD95 aggregation and/or conformation in the presence of CD95L because mutants within this domain impinge on the induction of apoptosis in cells exposed to CD95L (Fu et al., 2016). Super-resolution microscopy data points out that monomers, dimers, and trimers of receptors co-exist on the plasma membrane before ligand binding, supporting that CD95 stoichiometry results from a dynamic equilibrium among oligomeric states (Fu et al., 2016), which could differ according to the expression level of the receptor itself and other factors that remain to be identified. Interestingly, somatic mutations exist in the human CD95 TM domain associated with malignancy, such as P183L associated with lymphoma or C178R mutation with squamous cell carcinoma (Tauzin et al., 2012) and these mutations abrogate the trimerization of TM domains in bicelles (Fu et al., 2016). Because TM mutants disrupting the CD95 homotrimerization impede the CD95L-induced apoptotic program, we can envision that this stoichiometry corresponds to its minimal arrangement required for induction of cell signaling. Intracellular Domain (ICD) Like other death receptors, CD95 does not possess any intrinsic enzymatic activity and thereby initiates signaling cascades by recruiting proteins through protein-protein interactions (PPIs) in a dynamic manner. Most of the intracellular domain (ICD) is constituted by a DD, a scaffolding unit recruiting FADD through homotypic interactions. FADD in turn is a hub that binds caspase-8 and c-FLIP (Ferrao and Wu, 2012), and this complex cooperatively activates the apoptotic program (Hughes et al., 2016). The 3D structure of CD95 DD has been solved in complex with FADD by different teams (Figures 5A,B; Scott et al., 2009;Wang et al., 2010). Due to the different experimental conditions used in these studies, including low pH/high salt concentration vs. neutral pH and low salt concentration, and different methods (i.e., X-ray and NMR), the CD95 structures obtained are not completely superimposable ( Figure 5A) and probably represent different conformational states of the domain. The X-ray structure of CD95 performed at pH 4 (Scott et al., 2009) reveals a dramatic shift in the carboxy-terminal region of the DD encompassing helices 5 and 6, resulting in the opening of the globular structure to render amino acids of the interface accessible to the FADD DD. This modification of the DD conformation was not detected in other X-ray studies of the CD95/FADD complex (Wang et al., 2010) or NMR analyses of CD95 alone (Huang et al., 1996) or combined with FADD (Esposito et al., 2010). Interestingly, mutations of residues within DD, which favor the opening of helix 6, enhance CD95induced cell apoptosis, presumably because of an improved DISK formation. Unexpectedly, Driscoll's team showed that shifting pH from 6.2 to 4 causes the loss of CD95/FADD interaction (Esposito et al., 2010) weakening the conclusions drawn at low pH or suggesting that acidic conditions could affect the way CD95 implements cell signaling (Monet et al., 2016). The first crystallized CD95-DD/FADD complex showed a tetrameric arrangement (4:4) mostly mediated by CD95 domains (Figure 5C; Scott et al., 2009). However, the predicted orientation toward the membrane renders this model hardly compatible with the full assembly of the receptors. Indeed, Figure 5Cα illustrates that two chains of the tetramer are too far from the membrane. An alternative perpendicular orientation (Figure 5Cβ) seems also improbable for the same reason. Therefore, we suppose that this assembly results from crystal packing, and that a relevant biologic dimer is close to Figure 5Cγ. The second crystal structure of CD95 DD showed an asymmetric oligomeric complex composed of 5 CD95 DDs and 5 FADDs ( Figure 5D; Wang et al., 2010). This latter study revealed that half of the residues involved in CD95/CD95, CD95/FADD or FADD/FADD interfaces are positively or negatively charged, suggesting again a sensitivity to salts or pH for the formation of the aggregated complex and thereby signal induction. Although the structure of this complex matches with the data obtained using electron microscopy, it remains questionable because rebuilt from the supposed orientation shown in Figure 5A, the structure is asymmetric, and one DD penetrates the membrane (Figure 5Dα). Optimization of the pentameric complex shows again a questionable asymmetry (Figure 5Dβ), even if the juxtamembrane region that we designated CID for Calcium-Inducing Domain (Figure 1B) is long and flexible enough to accommodate such a variability. Calcium-inducing domain encompasses a 36 amino-acids sequence (amino acids K191 to D226), which is predicted to be disordered, explaining why it has never been solved by structural studies. Molecular modeling can, however, illuminate this structure at a single molecule level, showing that CID presented sparsely and transiently folded small α

helix (Poissonnier et al., 2016). The role of this peptide in the DD conformation and orientation to the plasma membrane and thereby in the recruitment of FADD is difficult to predict. While the DD (amino acid residues 210-303) is involved in cell death, the biological roles of the last 15 residues of The helix bundle is virtually inserted in a membrane, whose thickness is a hypothesis. Right panel: the Nt interdistance is less wide than its Ct counterpart (d1 = 9.7 ± 0.5 Å vs. d2 = 18.2 ± 2.6 Å between L174 or V195 Cα, for the 15 NMR structures). (B) Superposition of the 15 NMR structures, showing that the core of the bundle is quite rigid, while both ends are more flexible. . Note that there is still a conformational rearrangement of helices 5 and 6, but with a limited amplitude. (C) Different X-ray structures of ICD and their orientation toward the plasma membrane. Panels α to δ: proposed orientations of the tetrameric crystal structure of CD95:FADD complex (only CD95 is depicted). The N-terminal region of the death domain starting at N223 is the closest residue to TM and is labeled with black spheres. Chains in red seem correctly oriented regarding the plasma membrane, but the orientation of chains in yellow renders the position of the tetramer improbable. Panels γ to δ: only the closest dimers to the membrane are considered. Drawing in γ represents the most probable orientation toward the membrane. (D) Orientation of the pentameric CD95-DD, taking as reference the protomer showed in Figure 2B. Note that using this model, one DD is inserted into the plasma membrane. β. Optimized orientation of the pentameric CD95-DD regarding the position of the amino terminal residues K231 and Y232 (black balls and sticks) to the plasma membrane. Note the asymmetry of the structure, particularly for chain (A) (arrow). In conclusion, 3D structures of CD95 combined with biochemical and cellular data suggest the existence of different conformations for CD95-DD but their roles in the recruitment of FADD or other partners and the implementation of cell signaling remain to be understood. Apo CD95 Superposition of the trimeric model 2 shown in Figure 3 to the experimental TM bundle showed a near perfect alignment ( Figure 6A). The only 3 residues lacking in the ECD (i.e., E 168 GS) near the outer leaflet of the plasma membrane could form a short loop with a flexible glycine. This loop can easily fill the gap between ECD and TM, supporting our trimeric model 2. Indeed, the estimated distance between α carbon of each protomer of CD95 ECD at position E167 corresponds to 51 Å for model 1, 18 Å for model 2, 14 Å for model 3, and 64 Å for model 4, while trimeric NMR-based TM showed an average distance of 21 Å between α carbon at position R171 (Fu et al., 2016). Holo CD95 The structure of the CD95L complexed with the decoy receptor DcR3 in a trimeric complex has been solved (Liu et al., 2016). Based on structural similarities, we superimposed our previously rebuilt CD95 ECD to DcR3 receptor, resulting in a trimeric CD95L/CD95 complex ( Figure 6B). The homotrimeric Apo CD95 structure exhibits a packed conformation (Figure 6A), so a large opening of this quaternary structure is necessary to allow the insertion of the CD95L homotrimer ( Figure 6B). In this Holo conformation, the structural organization of the CD95L/CD95 trimer reveals that the missing three amino acids of CD95 ECD cannot fill anymore the gap between ECD and TM, with a distance between α carbon of each ECD CD95 at position E167 of 39 Å ( Figure 6B). This observation raises two hypotheses: either the distance of the TM bundle changes between Apo and Holo CD95 trimers, or CD95 CRD3 is very flexible and naturally pivots under CD95L to cover partially its bottom side, probably around the hinge formed by N132 ( Figure 6C). The second scenario seems the most plausible, because, first, TM domain has to be trimeric to implement the apoptotic signaling pathway in the presence of CD95L (Fu et al., 2016), and second, the lack of electron density in the crystal structure of CD95 CRD3 suggests a flexible domain. Moreover, the slope of the CD95 ECD protomers inside the trimer is reminiscent of an inverted iris-like mechanism observed for certain channels and transporters (Yoder et al., 2018;McCarthy et al., 2019), in which the inducer engenders a small conformation change, echoed into a huge amplitude modification at the opposite end of the structure. Marchesi et al. (2018) recently theorized the mechanical lever effect of the iris-like motion, and concluded that this mechanism reduces by 3 the force required to open channels. If this iris-like mechanism permits an amplification of motion from ECD to TM in response to small extracellular ligands (i.e., cyclic nucleotide and ATP), an inverted physical principle might here switch an important extracellular movement (i.e., insertion of a large ligand) into a minimal TM perturbation. Based on this mechanism, the trimeric TM would not need to dissociate when CD95 ECD widely opens to accept the homotrimeric CD95L. Moreover, a small motion in the juxtamembrane hinge region (for example, E 168 GS) should counterbalance the large shift of the PLAD domain ( Figure 6C). In agreement with a movement of the CD95 juxtamembrane domain, we estimated in our model a distance of 7 Å between the CD95 residue S137 and its partner on CD95L (P206), which was shown critical for the interaction (Schneider et al., 1997). Because this distance is too important for an implication of P206 in CD95/CD95L interaction, the receptor requires to approach the ligand, either through CRD3 flexibility or by a rotation/rocking of the receptor protomer following the iris-like hypothesis. These conformational rearrangements will require further investigations using normal mode analysis (Wako and Endo, 2017) and molecular dynamics (Arroyo-Manez et al., 2011;Wang et al., 2018). EXTERNAL FACTORS MEDIATED STOICHIOMETRY CD95 Post-translational Modifications (PTMs) CD95 can be glycosylated and different reports indicate that sialylation of asparagines 118 and 136 (corresponding to N102 and N120 in the human mature CD95 protein), improves the induction of the cell death program (Peter et al., 1995;Keppler et al., 1999). However, more recent data challenged the involvement of these glycosylations in the induction of cell death (Shatnyeva et al., 2011). Because the elimination of these glycosylations do not affect the stability or the plasma membrane expression of CD95 (Shatnyeva et al., 2011), it could be interesting in the future to explore the effect of these PTMs on the induction of the CD95-mediated non-apoptotic signaling pathways. Several other PTMs affect the extent of oligomerization of CD95 prior to or following its interaction with CD95L. These include S-palmitoylation of the juxtamembrane cysteine at position 199 (Chakrabandhu et al., 2007;Feig et al., 2007) and S-nitrosylations on both C199 and C304 (Leon-Bollotte et al., 2011). S-glutathionylation of CD95 at cysteine 294 (mouse amino acid sequence) promotes its aggregation and subsequent caspase activation and apoptosis (Anathy et al., 2009). Glutaredoxins (Grxs) reverse this process. Therefore, reactive oxygen species (ROS) can enhance CD95-mediated caspase-8 activation, which in turn cleaves and inactivates Grx1, generating a positive feedback loop sealing the cell fate. Also, CD95 S-glutathionylation promotes its distribution into lipid rafts and its avidity for CD95L (Anathy et al., 2009). These results highlight that CD95 aggregation and signaling can be modulated by a redox-based mechanism. Phosphorylation of CD95 on different serine/threonine and tyrosine (Y232 and Y291) within its intracellular region can modulate its signaling pathways (Chakrabandhu and Hueber, 2016). The replacement of Y291 by phenylalanine prevents recruitment of the AP-2 adaptor complex and the subsequent clathrin-mediated CD95 internalization but does not affect FADD binding and cell death induction (Lee et al., 2006). Interestingly, although this Y291 mutation inhibits the induction FIGURE 6 | CD95/CD95L complex and its association to the homotrimeric CD95 TM. (A) Manual alignment of trimeric Apo CD95 ECD model and experimental trimeric TM. Only residues E 168 GS are missing, and ECD E167 and TM R171 appeared close. (B) CD95-ECD was rebuilt using CD40 structure as a template. Next, this protomer was geometrically superimposed to the DcR3 homotrimer structure interacting with homotrimeric CD95L (yellow) with Maestro, Schrödinger Inc. (optimization of the α carbons superposition, with a final RMSD of 7.5 Å between DcR3 and CD95). CD95 TM is depicted in red. (C) Unlike the Apo CD95-ECD, the Holo Ct ends of CD95-ECD are too remote (each Ct end is distant from 40 Å) to be connected to the Nt ends of CD95-TM (in red) by the only 3 missing amino acids. The most probable rearrangement following ligand binding is a rotation/rocking of the whole CRD3 depicted in navy blue around N132 (the resulting modeled position is in navy blue ribbons, marked by blue arrows). of the apoptotic signaling pathway (Lee et al., 2006), it fails to alter the induction of non-apoptotic signals such as NF-KB and MAPK (Lee et al., 2006) suggesting that similarly to TNF-R1 signaling (Micheau and Tschopp, 2003), apoptotic and non-apoptotic machinery are assembled within different subcellular localizations. In addition to receptor tyrosine kinases (RTKs) described below (see paragraph III-2), src-family kinases (SFKs) can phosphorylate tyrosines in CD95 leading to the inhibition of the apoptotic program and these phosphorylation marks might serve as poor prognostic markers in several types of cancer, including breast, ovarian, and colon cancers . Of note, this Y291 phosphorylation can also recruit some phosphatases including SHP-1 and SHP-2 and SH2-containing inositol phosphatase (SHIP), whose activities counteract the granulocyte-macrophage colony-stimulating factor (GM-CSF)mediated pro-survival signal in neutrophils (Daigle et al., 2002). In conclusion, phosphorylation of Y291 within DD of CD95 might participate in the inhibition of the CD95mediated apoptotic pathway and at least in certain cells including neutrophils, might terminate the cytokine-mediated pro-survival signaling pathways rendering difficult to predict the role of this PTM in the cell fate. CD95 ECD Partners The tyrosine-protein kinase c-Met, also known as hepatocyte growth factor receptor (HGFR), can be associated with CD95, via a YLGA amino-acids sequence located in the N-terminal region of the c-Met α-chain (Wang et al., 2002), and CD95L also bears a 244 YLGA 247 sequence. Nonetheless, the observed competition between c-Met and CD95L for CD95 interaction raises some questions because the YLGA-containing c-Met sequence (i) competes with CD95L for CD95 binding, despite the fact that the CD95/CD95L interface involves amino acid residues different from the CD95L YLGA sequence (Schneider et al., 1997) and (ii) seems to disrupt CD95 oligomerization even if the CD95/CD95 aggregation requires CRD1 (PLAD) and TM domains different from the CRD2 and CRD3 regions involved in CD95/CD95L interface. Therefore, it remains to better understand how c-Met and CD95 interact to elucidate how this receptor affects the CD95 signaling pathway. Of note, an additional RTK, namely epidermal growth factor receptor (EGFR) has been linked to the modulation of the CD95-mediated signaling pathway. Accordingly, Haussinger's team reported that the hydrophobic bile salts can trigger cell death in hepatocytes through activation of EGFR, which induces CD95 tyrosine phosphorylation and implementation of cell death (Reinehr et al., 2003a,b). By contrast, other groups established that the EGFR-induced MAPK pathway counteracts CD95mediated apoptosis in hepatocyte cells exposed to bile salts (Qiao et al., 2001) and this RTK also inhibits the CD95mediated apoptotic signaling pathway in glioma cells (Steinbach et al., 2002) rendering difficult to conclude on the role of EGFR in the modulation of the CD95-mediated cell death program. On the other hand, the presence of EGFR exerts a pivotal role in the induction of the CD95-mediated nonapoptotic signaling pathways. In the presence of CD95L, CD95 recruits EGFR to implement the PI3K signaling pathway in TNBC cells (Malleter et al., 2013) or the MAPK pathway (i.e., extracellular signal-regulated kinase) in hepatic stellate cells (HSCs) (Reinehr et al., 2008) and thereby promotes cell migration or proliferation, respectively. Interestingly, a recent study highlighted the role of CD95 in sensing the cell survival of epithelial cells and thereby the maintain of tissue integrity (Gagnoux-Palacios et al., 2018). In adherens junction, the proximity of E-cadherin and α-catenin to CD95 favors the recruitment of Dlg1 to the C-terminal region of CD95 (Gagnoux-Palacios et al., 2018). Dlg1impinges on the DISK formation in cells exposed to m-CD95L and the loss of adherens junction will favor the release of this antiapoptotic factor to promote cell death, a mechanism

that could prevent metastatic dissemination of pre-tumor cells. Another method for adhesion molecules to control cell death has been also established for ICAM-2 (Perez et al., 2002). ICAM-2 overexpression or its interaction with leukocyte function-antigen-1 (LFA-1) induces ezrin phosphorylation by src tyrosine kinase and PI3K/AKT activation (Perez et al., 2002), which in turn impairs the induction of the CD95-mediated apoptotic program in leukocytes. This study points out that the PI3K activation by adhesion molecules can protect cells from apoptotic signal induced by death receptors. Ion-Driven CD95 Stoichiometry As aforementioned, upon addition of CD95L, CD95 undergoes conformational modification of its DD, inducing a shift of helix 6 and fusion with helix 5, promoting both oligomerization of the receptor and recruitment of the adaptor protein FADD (Scott et al., 2009). However, the idea of an elongated C-terminal α-helix favoring the cis-dimerization of CD95-DD in the acidic conditions (pH 4) was challenged by Driscoll and colleagues (Esposito et al., 2010) who did not observe the fusion of the last two helices at a more neutral pH (pH 6.2). These findings raise the question of whether a local decrease in intracellular pH might affect the CD95 conformation by promoting the opening of the CD95-DD and eventually by contributing to the formation of a complex that elicits a sequence of events distinct from what occurs at physiologic pH. Accordingly, we recently observed that CD95 activates the Na + /H + exchanger 1 (NHE1) in the presence of s-CD95L (Monet et al., 2016). NHE1 catalyzes an electroneutral exchange of extracellular Na + for intracellular H + , and its activity is necessary for cell migration (Putney et al., 2002;Frantz et al., 2007). While the presence of s-CD95L activates NHE1, no such modulation is observed in cells stimulated with a cytotoxic, multi-aggregated CD95L, suggesting that an acidic pH may surround the intracellular region of CD95 in cells stimulated with cytotoxic CD95L as compared to that in cells exposed to s-CD95L (Monet et al., 2016). This observation might explain how a drop of pH close to CD95 could promote a receptor conformation recruiting FADD and thereby, unleash the apoptotic signaling pathway. By contrast, NHE1 activation by CD95 (Monet et al., 2016) alkalinizes the intracellular region and could prevent modification of DD helix 5 and 6 fusion (Scott et al., 2009). Of note, acidification can affect the protein conformation through the modulation of histidines as demonstrated for the phosphoinositide binding of cofilin, which is pH-dependent and decreases at high pH (Frantz et al., 2008). Interestingly, the intracellular region of CD95 encompasses four histidines and two of them (H282 and H285, i.e., H266 and H269 in the mature protein) are localized upstream helix 5 (Tauzin et al., 2012). Lipid-Driven CD95 Stoichiometry CD95 aggregation relies on the plasma membrane composition in lipids. Indeed, CD95 aggregation is slower in 1,2dipalmitoyl-sn-glycero-3-phosphocholine (DPPC) than in 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC), thereby the apoptotic program is faster in the latter lipid environment (Gulculer Balta et al., 2019). CD95 engagement triggers the accumulation of ceramide in a caspase-8-dependent manner, which in turn contributes to its aggregation and thereby favors the induction of cell death (Grassme et al., 2003). The initial stage of the CD95 response could be described as a two-step process, first requiring a certain degree of CD95 aggregation to secondarily promoting a caspase-8-driven intracellular signaling pathway that results in the aggregation and distribution of unstimulated CD95 into lipid rafts (Siegmund et al., 2017) which seems to favor the apoptotic response (Gajate et al., 2004). S-palmitoylation of CD95 (Chakrabandhu et al., 2007;Feig et al., 2007) appears to promote CD95 redistribution into lipid rafts (Muppidi and Siegel, 2004;Chakrabandhu et al., 2007). Two Ligands Contrary to its receptor, CD95L is a type II transmembrane protein whose N-term extremity is in the cytoplasm (Figure 1C). The membrane-bound native CD95L (m-CD95L) can be processed by several proteases, including MMP3, MMP7, MMP9, and ADAM-10 (Tauzin et al., 2012), to release a soluble form of the ligand (s-CD95L) ( Figure 1C). m-CD95L is responsible for the DISK assembly, whereas s-CD95L can trigger the formation of a different complex designated MISC (Malleter et al., 2013;Poissonnier et al., 2016). While most of the studies on s-CD95L report that this ligand possesses an homotrimeric stoichiometry, its membrane-bound counterpart shows a higher degree of aggregation such as large synapse of CD95L are observed between CD95L-expressing T-cells or NK cells and their cellular targets. Human CD95L selfassociation domain spans between amino acid residues 137-183 (Voss et al., 2008) and the last three amino acids are important for CD95 interaction (Figure 1C; Orlinick et al., 1997). CD95L ectodomain contains three putative sites for N-linked glycosylation (N184, N250, and N260) and a lack of glycosylation alters the expression level of this ligand, probably by acting on its stability and/or intracellular trafficking (Schneider et al., 1997;Voss et al., 2008). Although CD95L/CD95 structure and surface plasmon resonance analyses reveal that CD95L glycosylation is not interfering with CD95 binding, the glycosylated ligand triggers a stronger cell death signal as compared to its sugar-free counterpart (Liu et al., 2016). While this difference of function has been associated with the fact that CD95L glycosylation might reduce the magnitude of its aggregation level (Liu et al., 2016), other studies showed no effect on the induction of cell death signal (Orlinick et al., 1997;Schneider et al., 1997;Voss et al., 2008), rendering difficult to conclude on the exact biological role played by the N-glycosylation of CD95L. Although O-glycosylation for DR5 (Wagner et al., 2007) and N-glycosylation for DR4 do not alter the intensity of their interaction with TRAIL (Dufour et al., 2017), these PTMs enhance the apoptotic signal through molecular mechanisms that remain to be elucidated (Micheau, 2018). A possible explanation could come from galectins, which are small proteins capable to bind to the β-galactoside sugars present in the extracellular region of TNF receptors family members. Of note, different galectins can bind and aggregate DR4, DR5, and CD95 and thereby, stimulate or inhibit cell death (Micheau, 2018) suggesting the subtle role played by glycosylation in the implementation of cell signaling by death receptors. CD95 Ligands and Aggregation The role of CD95L in the extend of CD95 aggregation is not clearly understood. Although most studies report that the metalloprotease-cleaved CD95L engenders homotrimer unable to induce cell death (Tanaka et al., 1996Tauzin et al., 2011;Suda et al., 1997;Schneider et al., 1998) but instead triggers pro-inflammatory signaling pathways, in certain pathologies, a soluble CD95L is accumulated and reaches an aggregation level of CD95 allowing the implementation of the cell death program (Bajou et al., 2008;Herrero et al., 2011). Moreover, although the homotrimeric s-CD95L does not induce cell death, a recombinant and hexameric form does (Holler et al., 2003), supporting that the extent to which CD95L is multimerized is a pivotal step in determining whether non-apoptotic signaling or cell death is induced. Notably, some pathophysiological conditions could favor s-CD95L oligomerization, thereby promoting its cytotoxic activity. CD95L in the bronchoalveolar lavage (BAL) fluid of patients suffering from acute respiratory distress syndrome (ARDS) undergoes oxidation at methionines 224 and 225, promoting its aggregation and thereby rendering it cytotoxic (Herrero et al., 2011). In addition, in ARDS BAL fluid, another methionine oxidation occurs at position 121 within CD95 and prevents its cleavage by MMP7 which can explain why this cytotoxic ligand retains its stalk region (Herrero et al., 2011). Nonetheless, whether this corresponds to an alternatively cleaved form of s-CD95L with higher-level stoichiometry or a full-length exosome-bound m-CD95L remains to be elucidated. The stoichiometry of CD95L can also be increased by external elements including fibronectin in the extracellular matrix (Aoki et al., 2001), rendering the inactive molecule apoptotic and raising the question of which domain within the soluble ligand interacts with fibronectin. The different responses obtained with soluble and membranebound CD95L are a common feature among the TNF superfamily. For instance, while soluble TNF binds efficiently both TNFR1 and TNFR2, it stimulates TNFR1 signaling and induces cell death, but fails to trigger any response with TNFR2 (Grell et al., 1995). More importantly, the artificial oligomerization of soluble ligands restores the implementation of a classical response (Wajant, 2015) indicating that the ligand stoichiometry modulates the cell signaling in this superfamily. However, the difference between soluble and membrane-bound ligand signaling can be not so tremendous that what is observed for TNR2 or CD95. For instance, membrane TWEAK (TNF-like weak inducer of apoptosis) induces both alternative and classical NFκB pathways while soluble TWEAK only triggers the classical NFκB pathway (Wajant, 2015). Agonistic Antibodies An interesting study using a set of agonistic anti-CD95 antibodies revealed an inverse correlation between antibody affinity and cell death (Chodorge et al., 2012). A structure-function analysis disclosed that dissociation rate (Koff) of anti-CD95 antibodies is crucial for receptor activation because beyond affinity, dissociation of one antibody arm allows antibodies to bring together more CD95 monomers, forming a receptor cluster required to trigger cell death (Chodorge et al., 2012). These observations strengthen that the level of CD95 aggregation is important to induce the cell death process. However, the role of aggregation in the induction of non-apoptotic signaling pathways has not been investigated in this study and could be interesting to address. DISCUSSION Unsurprisingly, initial therapeutic solutions involving CD95 focused on the apoptotic pathway. Most of research efforts have dealt on deciphering the molecular basis of apoptosis induction by CD95 and considering the biological functions of CD95 in light of this role. Although non-apoptotic functions of CD95 (Alderson et al., 1993) have been reported soon after CD95 cloning (Itoh et al., 1991;Alderson et al., 1993), these have been largely neglected over the years. As a consequence, no CD95 agonists have become a standard of care in inflammatory disorders or cancers. It is now clear that CD95 can contribute to multiple biological functions, including inflammation and tumorigenesis through the induction of non-apoptotic signaling pathways. Accordingly, a CD95 decoy receptor blocking both the apoptotic and non-apoptotic signaling pathways, Asunercept (APG101), has nevertheless entered clinical trials for glioma and myelodysplastic syndrome (Wick et al., 2014;Boch et al., 2018). Overall, the evidence that homotrimeric ligand can activate certain receptor-associated signaling pathways favors the concept of a two-step model of TNFRSF activation. In a first step, there is ligand induced formation of homotrimeric TNFSF/TNFRSF complex, triggering some signaling pathways (mainly non-apoptotic signaling pathways). In a second step, there is multimerization of the homotrimeric complex through different mechanisms including oligomerization, transactivation, plasma membrane or microdomain redistribution/exclusion inducing different signaling pathways. Each of these steps constitute possible targets for therapeutic agents and should be scrutinized in future studies. AUTHOR CONTRIBUTIONS NL and PL designed the experiments (computer modeling) and wrote the manuscript. MJ wrote the manuscript. FUNDING This work was supported by the INCa PLBIO, Ligue Contre le Cancer, Fondation ARC, Fondation de France (Price Jean Valade), and ANR PRICE. Progression of Danon disease with medical imaging: two case reports Danon disease is a rare X-linked dominant genetic disorder caused by loss-of-function mutations in the lysosome-associated membrane protein 2 gene. Progression of Danon disease is unknown because of its rare incidence in a diverse ethnic population. We report longitudinal data from two patients who were diagnosed with Danon disease by a genetic test. The evaluation protocol included electrocardiographic monitoring, echocardiography, and magnetic resonance imaging. Progression of hypertrophic cardiomyopathy to dilated cardiomyopathy was observed in the first patient. He died from sudden cardiac arrest. The second patient is currently suffering from hypertrophic cardiomyopathy. Development of the hypertrophic phase progressing into the dilated phase in Danon disease may provide useful information for early identification and clinical decisions in patients with this disease. Introduction Danon disease is a type of lysosomal glycogen storage disease with normal acid maltase. 1 Danon disease is a rare X-linked dominant genetic disorder caused by lossof-function mutations in the lysosomeassociated membrane protein 2 (LAMP-2) gene. This disease can be distinguished from other vacuolar myopathies by the presence of glycogen particles in cell debris, and acetylcholine and nonspecific esterase activity in small basophilic vacuolar membranes. 2 The triad of hypertrophic cardiomyopathy, myopathy, and intellectual disability are some of the classical clinical features of Danon disease in boys. [1][2][3][4] Cardiomyopathy in Danon disease classically presents as hypertrophic cardiomyopathy (HCM) in most male patients and progresses as dilated cardiomyopathy (DCM) in some of the patients. [5][6][7][8] In contrast, in female patients, DCM and HCM have equal prevalence. 6 Other conduction defects are common

in Danon disease, including atrial fibrillation, complete heart block, supraventricular tachycardia, and sinus node dysfunction. 9 Late gadolinium enhancement (LGE) in cardiomyocytes shows a marked progression of fibrosis, which is useful for distinguishing ischemic cardiomyopathies from nonischemic cardiomyopathies. 10,11 We report here longitudinal data from two cases of Danon disease. Clinical and biochemical data, electrocardiograms (ECGs), echocardiography, and magnetic resonance imaging (MRI) were performed to show the features of Danon disease. One of our patients who had Danon disease with a 10-year follow-up showed progress of HCM to DCM. Case 1 A 23-year-old man had Danon disease from the age of 12 years, when symptoms began, until 23 years old when he succumbed to his disease. He was first admitted to the Affiliated Hospital of Jining Medical University at 12 years old because of transient loss of consciousness with palpitation. He had suffered from a poor ability of physical activity, mild mental retardation, and learning difficulty. Pansystolic murmur along the left lower sternal border was not observed. Echocardiography showed HCM with a maximum left ventricular end-diastolic diameter (LVEDd) of 45 mm, left ventricular posterior wall (LVPW) thickness of 12 mm, maximum interventricular septal (IVS) thickness of 11 mm, and left ventricular ejection fraction (LVEF) of 60%. Electromyography showed normal muscle strength and deep tendon reflexes. He was diagnosed with Danon disease at 13 years of age in Beijing Union Hospital by genetic testing, which showed that he carried a deletion mutation (c.257_258delCC) in exon 3 of the LAMP-2 gene. An endomyocardial biopsy and electron microscopic examination showed autophagic vacuoles. ECG showed type B Wolff-Parkinson-White (WPW) syndrome, a high voltage in the left ventricle, and an inverted T wave (Figure 1a). He then underwent successful ablation in Beijing Union Hospital. 12 He was asymptomatic for the next 3 years. At the age of 17 years, he had persistent palpitations and visited the hospital. ECG (Figure 1b) showed an ectopic rhythm, paroxysmal supraventricular tachycardia, complete left bundle branch block, and high voltage in the left ventricle. He switched to sinus rhythm following treatment with propafenone. However, an ECG ( Figure 1c) showed ventricular premature contraction. At a physical examination, a heaving apical impulse and a grade 4/6 pansystolic murmur along the left lower sternal border were observed. Follow-up echocardiography indicated that left ventricular hypertrophy of the patient progressively worsened (Figure 2a, b). MRI was performed and showed LGE in the muscular interventricular septum and the front wall of the left ventricle. However, the liver and bilateral gastrocnemius skeletal muscles were negative for LGE ( Figure 3a). This process of cardiac hypertrophy was rapidly advancing at this stage. In the next 4 years, HCM progressed to the dilated phase. The LVEDd increased from 44 to 58 mm, while the IVS decreased from 24 to 19 mm and LVPW decreased from 31 to 17 mm. The LVEF was reduced from 60% to 22% (Figure 2c, d). At the age of 23 years, an ECG ( Figure 1d) showed atrial fibrillation. An MRI ( Figure 3b) showed that gadolinium contrast enhancement was visible in the liver, psoas muscle, and bilateral gastrocnemius muscles. These abnormal findings manifested with the classic triad of cardiomyopathy, skeletal myopathy, and intellectual disability in this patient. He received an automatic implantable cardioverterdefibrillator, even after considering a heart transplant, which he refused, and he finally succumbed to his disease with sudden cardiac arrest. Unfortunately, the patient did not undergo an investigation of his device to identify the trigger for cardiac death. Case 2 A Chinese boy has been followed up from 10 years old. Genetic analysis at 10 years old showed that he had Danon disease and was (Figure 4b). Findings at this time were similar to those in case 1 at 13 years old. Changes in the LVEDd (55 to 51 mm), IVS (15 to 21 mm), LVPW (15 to 24 mm), and LVEF (63% to 61%) were recorded in the next 2 years (Figure 5a, b), as well as changes in MRI ( Figure 6). However, he was too overweight to ensure good image quality. MRI showed an expanded scope of gadolinium contrast enhancement, which was visible in the myocardium. However, the liver, psoas muscle, and bilateral gastrocnemius muscles were normal. The boy was deprived of education owing to his intellectual disability. Based on these data, the boy will be followed up in the hypertrophic stage to monitor progression of HCM to DCM. Discussion Follow-up of case 1 showed two phases of rapidly progressing hypertrophy and rapidly progressing dilation of hypertrophy. In these two phases, typical and atypical arrhythmias occurred, and clinical Expanded scope of a late gadolinium enhancement pattern in the myocardium, but the liver, psoas muscle, and bilateral gastrocnemius muscles are negative for this enhancement. evaluation, echocardiograms, and MRI were performed. In Danon disease, LAMP-2 mutations cause a significant variety of clinical manifestations. This disease is a multisystem disorder that predominantly affects cardiac and skeletal muscles, and its morbidity is unknown. Malignant arrhythmias and heart failure are the leading causes of death in patients with this disease. 13 The average age of the first symptoms, diagnosis, and death in male patients with Danon disease is younger than that in female patients. 14 Electrical conduction abnormalities are equally common and present in almost all affected men. 14,15 WPW syndrome is a common ECG abnormality in this disease. 14 Atrial and ventricular arrhythmias are also present in Danon disease. The majority of male patients with Danon disease are mainly accompanied by HCM, but 10% show the DCM phenotype. 15 Congestive heart failure partly complicates progression of HCM to the dilated phenotype in later stages of this disease in young men. 15,16 Two phenotypes were noted in one of our patients. Followup of case 1 suggests that HCM is a characteristic feature of the early stage of Danon disease, but DCM may occur in the late stage. Because there is phenotypic variation of Danon disease and progression to DCM is not common in these patients, we are unsure if case 2 will develop DCM. MRI has recently been used for cardiomyopathies. MRI enables accurate assessment of left ventricular wall thickness, size, and function because of its high spatial resolution. Furthermore, LGE detects the presence of myocardial fibrosis. 17 In our patients, extensive LGE was consistently found in the anterior wall of the left ventricle and was normal in the liver and skeletal muscles in the HCM phase, but it showed myocardial fibrosis, liver damage, and skeletal muscles in the DCM phase. This finding indicates a possible limitation of LGE in the early stage of Danon disease and diffusion in the late stage. Unfortunately, patients with Danon disease succumb to lethal ventricular tachyarrhythmia. 16 A heart transplant is the most effective form of treatment for this condition. 18 Danon disease should be taken into consideration for adolescents with cardiomyopathy. A cardiac physical examination, ECG, echocardiography, and biochemical examination are the most common examinations for Danon disease. Especially for patients with HCM, risk stratification for sudden cardiac death needs to be identified, with performance of programed ventricular stimulation and techniques, such as MRI, for an individualized prevention strategy. 19,20 Genetic counseling should be provided to families with a family history of Danon disease. Because of the rapidly progressive nature of Danon disease, genetic testing should be performed as soon as possible in HCM. 9 With the latest developments in gene therapy and cell transplantation techniques, genetic manipulations and pharmaceutical approaches may provide a novel therapy for patients with genetic diseases, including Danon disease. In conclusion, we obtained longitudinal clinical data from disease onset to sudden cardiac death in a patient with classical Danon disease. In our other case, 10-year follow-up showed progression of the hypertrophic phase to the dilated phase. This development may provide useful information for early identification and clinical decisions in patients with Danon disease. Ethics statement The study was approved by the Ethics Committee of the Affiliated Hospital of Jining Medical University (approval no.: 2017-Article-001). The patients or parents provided informed consent for publication of the cases. Effect of Cisplatin on the Frequency and Immuno-inhibitory Function of Myeloid-derived Suppressor Cells in A375 Melanoma Model Based Abstract Background: To investigate the change of frequency and immuno-inhibitory function of myeloid-derived suppressor cells (MDSCs) after treatment of cisplatin (DDP) in A375 human melanoma model. Materials and Methods: BALB/c nude mice were inoculated with A375 cells to establish the human melanoma model and randomly divided into control group given normal saline (NS) and experimental group treated with DDP (5 mg/ kg). The percentages of MDSCs in the tumor tissue and peripheral blood after DDP treatment were detected by flow cytometry. The proliferation and interferon-γ (IFN-γ) secretion of T cells co-cultured with MDSCs were analyzed through carboxyfluorescein succinimidyl ester (CFSE) labeling assay and enzyme-linked immunospot (ELISPOT) assay, respectively. Results: In A375 human melanoma model, DDP treatment could significantly decrease the percentage of MDSCs in the tumor tissue, but exerted no effect on the level of MDSCs in peripheral blood. Moreover, DDP treatment could attenuate the immuno-inhibitory function of MDSCs. T cells co-cultured with DDP-treated MDSCs could dramatically elevate the proliferation and production of INF-γ. Conclusions: DDP can decrease the frequency and attenuate immuno-inhibitory function of MDSCs in A375 melanoma model, suggesting a potential strategy to augment the efficacy of combined immunotherapy. Introduction Myeloid-derived suppressor cells (MDSCs) are a heterogeneous population of cells deriving from the myeloid cell lineage and abnormally accumulating in the patients with various tumors (Crook et al., 2014). Gr-1 and CD11b, two primary cell surface markers, are usually used to identify the MDSCs (Medina-Echeverz et al., 2015). There is strong evidence that MDSCs could inhibit both adaptive and innate immunity mainly through their capacity to produce nitric oxide and radical oxygen species, thus leading to the inhibition of activation, proliferation and function of effector T cells (Fernandez et al., 2014). Given this immuno-inhibitory effect of MDSCs, it has been proposed that down-regulation of MDSCs may significantly augment the anti-tumor immune activity. Previous studies displayed that except for direct cytotoxic effects on tumor cells, some chemotherapeutic drugs could also eliminate MDSCs, resulting in enhanced anti-tumor response (Suzuki et al., 2005;Vincent et al., 2010;Chen et al., 2012;Alizadeh et al., 2014a;2014b;Zheng et al., 2015). Based on the previous research results that cisplatin (DDP) could decrease the percentage of MDSCs in immuno-competent B16 melanoma model (Chen et al., 2012), the down-regulatory effect of DDP General data Animals: SPF female BALB/c nude mice and BALB/c mice (6~8 weeks old) were purchased from Academy of Military Medical Science (Beijing, China) and housed in a pathogen-free animal facility with moderate temperature and humidity. The animal experiments were in accordance with the guide for care and use of laboratory animals as approved by Nanjing Medical University. Cells lines: Human melanoma cell line A375 was obtained from Shanghai Institute of Biochemistry and Cell Biology (Shanghai, China) and cultured in complete RPMI 1640 medium (Hyclone, Waltham, MA, USA) supplemented with 10% fetal bovine serum. All cells were cultured in a humidified atmosphere with 5% CO 2 at 37°C. Methods Tumor model: A375 cells (10^6/100 μL PBS) were subcutaneously inoculated in the right flank of BALB/c nude mice. When tumors were detectable (about 5 mm in diameter), the mice were administrated with the single dose of 5 mg/kg DDP (Qilu Pharmaceutical Co., Ltd, China) as experimental group approximately 7 days after tumor challenge, whereas the mice received normal saline (NS) alone as control group. Besides, both experimental and control groups included 3 mice at least. Flow cytometry: On the selected days after treatment, the single cell suspensions of peripheral blood and tumor tissue were prepared and stained with the indicated monoclonal antibodies. FITC-conjugated anti-CD11b, PE-conjugated anti-Gr-1, PE-Cy5-conjugated anti-CD3 and FITC-conjugated anti-CD8 antibodies were purchased from eBioscience (San Diego, CA, USA). To determine the percentage of MDSCs, the single cell suspension was subjected to flow cytometry and gated by plotting forward vs side scatter followed by gating on Gr-1 + /CD11b + cells. For the suspension from peripheral blood, 10,000 events were collected while 100,000 for the tumor tissue. All analyses were conducted using Accuri C6 software. Co-culture system: CD3 + /CD8 + T cells were isolated from the spleen of naive BALB/c mice using MACS system. Then CD11b + MDSCs were also purified from the spleen of mice bearing A375 tumors

using MACS system. These sorted T cells were mixed with enriched MDSCs that had been treated with either NS or single injection of 5 mg/kg DDP 48 h before isolation. The cell ratios were one MDSC for one T cell in triplicate. Soluble anti-CD3 (2 μg/mL), anti-CD28 (2 μg/mL) and interleukin-2 (IL-2, 1 000 IU/mL) obtained from PEPROTECH (Rocky Hill, NJ, USA) were added to the co-culture system. Stimulation was maintained for 24h. Carboxyfluorescein succinimidyl ester (CFSE) labeling assay: The purified splenic CD3 + /CD8 + T cells were incubated with 5 μmol/L CFSE obtained from Dojindo Laboratories (Tokyo, Japan) for 10 min and then mixed with MDSCs in the co-culture system. After incubation for 24 h, the cells were harvested and subjected to the flow cytometry for the determination of CFSE + cells. Enzyme-linked immunospot (ELISPOT) assay: The secretion of interferon-γ (IFN-γ) by T cells was detected by ELISPOT using Mouse ELISPOT kit (DAKEWEI, China) according to the manufacturer protocol. In brief, T cells (1×10 5 cells/well) co-cultured with MDSCs were added into anti-IFN-γ coated 96-well plates in triplicate, stimulated with anti-CD3 (2 μg/mL), anti-CD28 (2 μg/ mL) and IL-2 (1 000 IU/mL) for 24h. After washing steps, the secondary biotin-conjugated anti-IFN-γ detection antibody and diluted streptavidin-HRP conjugate solution were added and incubated for 1h successively. Finally, the plates were treated with AEC substrate solution and incubated at room temperature for 25 min in the dark. After developing, the spots were counted using an ELISPOT reader (Bioreader 4000; Bio-sys, Germany). Statistical data analysis Statistical analysis was conducted using SPSS 13.0 statistical software (Chicago, Illinois, USA). Measurement data were expressed as the mean±standard deviation (x±s) and compared with t test. Significant differences between groups were compared using ANOVA, and LSD was applied for multiple comparisons. Differences were considered to be significant at the time of P<0.05. DDP reduced the frequency of Gr-1 + /CD11b + MDSCs in the tumor tissue In preliminary experiments, we found that DDP could eliminate Gr-1 + /CD11b + MDSCs in B16 melanoma model (Chen et al., 2012). To investigate whether DDP had a similar effect on the frequency of MDSCs induced by other cancer cell lines, A375 human melanoma model was established and treated with the single dose of DDP (5 mg/ kg). In consistent with the data obtained in B16 melanoma model, the percentage of MDSCs elevated, which was associated with the increased tumor mass without DDP treatment (Figure 1). Administration of DDP was able to dramatically decrease these Gr-1 + /CD11b + MDSCs populations in mice bearing A375 tumor. Three days after DDP treatment, a typical flow cytometry trace showed a marked decrease in the percentage of MDSCs (25.77±5.55 vs 15.10±0.53, P=0.03). However, DDP-induced depletion seemed transient. The percentage of MDSCs rapidly rose to the level comparative to that in control group 6 and 9 d after DDP treatment. DDP exerted no effect on the frequency of Gr-1 + /CD11b + MDSCs in peripheral blood. To determine whether circulating MDSCs were affected by DDP treatment, we also examined the percentage of MDSCs in peripheral blood in A375 melanoma model. As shown in Figure 2, the frequency of Gr-1 + /CD11b + MDSCs in peripheral blood of tumor-bearing hosts increased with the growth of tumor mass without DDP intervention. However, DDP treatment did not affect the frequency of circulating MDSCs in mice bearing A375 DOI:http://dx.doi.org/10.7314/APJCP.2015.16.10.4329 Effect of Cisplatin on the Frequency and Immuno-inhibitory Function of Myeloid-derived Suppressor Cells in A375 Melanoma Model melanoma. Although there was a mild decline regarding the percentage of MDSCs compared with control group, significant difference was not presented. DDP enhanced the proliferation of T cells co-cultured with MDSCs Previous studies have reported that MDSCs could suppress the anti-tumor activity of CD8 + T cells (Umansky et al., 2013;Ostrand-Rosenberg et al., 2010). We speculated that except for the down-regulatory effect on the frequency of MDSCs, DDP treatment could also attenuate the immuno-inhibitory effect of MDSCs on T cells. CD3 + /CD8 + T cells were purified from BALB/c mice and stained with CFSE. To determine the suppressive effect of MDSCs on the proliferation of T cells, we mixed the sorted CD3 + /CD8 + T cells with CD11b + MDSCs from the mice bearing A375 melanoma that had been treated 3 d previously with DDP or NS. As shown in Figure 3, untreated MDSCs could significantly reduce the proliferation of T cells (4.06±0.94 vs 9.87±1.60, P=0.001), whereas DDP treatment partially abrogated this immuno-inhibitory effect of MDSCs. The proliferation of co-cultured T cells profoundly elevated (6.52±0.79 vs 4.96±0.94, P=0.025), but was still markedly inhibited by comparison to that without MDSCs (6.52±0.0.79 vs 9.87±1.60, P=0.031). DDP augmented the secretion of IFN-γ by T cells cocultured with MDSCs. Whether DDP treatment could alter the inhibitory effect of MDSCs on IFN-γ production of T cells was also investigated. After co-culture with untreated MDSCs in A375 melanoma-bearing mice, the secretion of IFN-γ by A375 melanoma were treated with 5 mg/kg of DDP (i.p., D0) or left untreated. Peripheral blood was isolated and prepared for the single cell suspension on D0, 3, 6, 9 after treatment. The percentage of Gr-1 + /CD11b + MDSCs in the peripheral blood was determined through cytometry (A) and the typical data that the experiment was shown (B). Columns: the mean of triplicate measurements; Bars: SE Figure 4. Augmented Secretion of IFN-γ by T Cells Co-Cultured with MDSCs after DDP Treatment. A375 melanoma-bearing mice were administered with DDP 5 mg/kg (i.p., D0) or left untreated. 3 d after DDP treatment, the spleen tissues were isolated from the mice and CD11b + MDSCs were sorted by flow cytometry. The obtained MDSCs were then cocultured with T cells for 24 h. The IFN-γ production of T cells was detected by ELISPOT (A) and the typical data that the experiment was shown (B). Columns: the mean of triplicate measurements; Bars: SE; *: P<0.05 Figure 3. Enhanced Proliferation of T Cells Co-Cultured with MDSCs after DDP Treatment. A375 melanoma-bearing mice were administered with DDP 5 mg/kg (i.p., D0) or left untreated. 3 d after DDP treatment, the spleen tissues were isolated from the mice and CD11b + MDSCs were sorted by flow cytometry. The obtained MDSCs were then cocultured with CFSE + T cells for 24 h. The proliferation of T cells was subjected to flow cytometry (A) and the typical data that the experiment was shown (B). Columns: the mean of triplicate measurements; Bars: SE; *P<0.05 T cells was obviously inhibited (3.80±1.22 vs 38.40±6.20, P=0.0007), whereas DDP treatment partially reversed this immuno-suppressive effect of MDSCs, resulting in significantly increased IFN-γ production by co-cultured T cells (24.06±6.10 vs 3.80±1.22, P=0.005). However, the secretion of IFN-γ by T cells remained in a decreased status compared with that without MDSCs (24.06±6.10 vs 38.40±6.20, P=0.046). Discussion In recent years, immunotherapy has displayed a great promise in the treatment of various tumors (Su et al., 2013;Zhang et al., 2014a;Wang et al., 2014b;Anagnostou et al., 2015;Curti et al., 2015;Swaika et al., 2015). However, the therapeutic activity of immunotherapy is limited by the immunosuppressive factors active in the tumor hosts. It is increasingly recognized that manipulation of the host immune system is essential to guarantee the anti-tumor effect of immunotherapy (Shekarian et al., 2015). As a population of immature myeloid cells, MDSCs possess the ability to inhibit T cell antigen-specific response mainly through the production of nitric oxide and radical oxygen species, consequently leading to the failure of anti-tumor activity (Fernandez et al., 2014). It has been proposed that elimination of MDSCs will be an effective approach to abrogate the tumor-induced immune-suppression and enhance the activity of immunotherapy. Accordingly, a number of strategies have been evaluated. Among which, chemotherapeutic drugs are demonstrated to have the ability to selectively eliminate MDSCs. It was reported that both gemcitabine and 5-fluorouracil could reduce the number of splenic Gr-1 + /CD11b + MDSCs while preserving the number of T cells, NK cells and B cells, accompanied by an increased anti-tumor activity of CD8 + T cells and NK cells (Suzuki et al., 2005;Vincent et al., 2010;Chen et al., 2012;Alizadeh et al., 2014a;Alizadeh et al., 2014b;Zheng et al., 2015). Besides, DDP could deplete the splenic Gr-1 + /CD11b + MDSCs in murine B16 melanoma (Chen et al., 2012). To date, there are few studies that delineate the function of MDSCs after treatment of chemotherapeutic drugs, even fewer studies on the change of MDSCs induced by chemotherapeutic drugs in immune-deficient mice bearing human tumors. Therefore, in this study, BALB/c nude mice bearing human melanoma were established, and the frequency and function of MDSCs after DDP treatment were also analyzed. The research results displayed that DDP (5 mg/ kg) could significantly reduce the frequency of MDSCs in the tumor tissue in mice bearing A375 human melanoma. However, this depletion was transient and only lasted for 3 d after DDP treatment, indicating that there may be an optimal time point when the combined regimen exerts the maximal synergistic anti-tumor activity. Moreover, DDPinduced depletion of MDSCs frequency was not detected in peripheral blood. We speculated that the elimination of MDSCs induced by DDP was tumor cells-dependent. Some cytokines released by tumor cells, such as IL-6, IL-10 and VEGF, could recruit MDSCs to migrate and accumulate into the tumor tissue (Gabrilovich et al., 2009). DDP was presumed to inhibit the release of these soluble factors by tumor cells, thus inducing the decline of MDSCs in tumor tissue rather than in peripheral blood. In addition to showing a decreased frequency of MDSCs in the tumor tissue of A375 melanoma-bearing mice after DDP treatment, it was also found a concomitant decrease in the immuno-inhibitory activity of MDSCs. DDP treatment attenuated the immunosuppressive effect of MDSCs on the proliferation and IFN-γ production of T cells. Notably, DDP treatment only partially reversed the immuno-inhibitory effect of MDSCs, the proliferation and IFN-γ production by T cells still went down profoundly by comparison to that without MDSCs. The mechanisms on whether DDP can hinder the immuno-inhibitory effect of MDSCs remain elusive. One speculation may be that MDSCs can differentiate into dendritic cells after DDP treatment. Cyclophosphamide is reported to induce a marked expansion of immature dendritic cells which share the same progenitor with MDSCs (Salem et al., 2009). However, further studies are still needed to elucidate the underlying mechanisms. In this study, the down-regulatory effect of DDP on MDSCs was observed in nude mice bearing human A375 melanoma. Nevertheless, the murine immune environment was not able to mimic the human immune background. Although chemotherapeutic drugs were capable of decreasing MDSCs in mice, an interesting question that remains to be solved is that whether the immune-modulating effect of chemotherapeutic drugs could be translated into clinical practice. In summary, DDP can decrease the frequency and attenuate immuno-inhibitory function of MDSCs in A375 melanoma model, suggesting a potential strategy to augment the efficacy of combined immunotherapy. The dark side of the force: Multiplicity issues in network meta‐analysis and how to address them Standard models for network meta‐analysis simultaneously estimate multiple relative treatment effects. In practice, after estimation, these multiple estimates usually pass through a formal or informal selection procedure, eg, when researchers draw conclusions about the effects of the best performing treatment in the network. In this paper, we present theoretical arguments as well as results from simulations to illustrate how such practices might lead to exaggerated and overconfident statements regarding relative treatment effects. We discuss how the issue can be addressed via multilevel Bayesian modelling, where treatment effects are modelled exchangeably, and hence estimates are shrunk away from large values. We present a set of alternative models for network meta‐analysis, and we show in simulations that in several scenarios, such models perform better than the usual network meta‐analysis model. Standard models for network meta-analysis simultaneously estimate multiple relative treatment effects. In practice, after estimation, these multiple estimates usually pass through a formal or informal selection procedure, eg, when researchers draw conclusions about the effects of the best performing treatment in the network. In this paper, we present theoretical arguments as well as results from simulations to illustrate how such practices might lead to exaggerated and overconfident statements regarding relative treatment effects. We discuss how the issue can be addressed via multilevel Bayesian modelling, where treatment effects are modelled exchangeably, and hence estimates are shrunk away from large

values. We present a set of alternative models for network meta-analysis, and we show in simulations that in several scenarios, such models perform better than the usual network metaanalysis model. | INTRODUCTION Network meta-analysis (NMA) is a statistical tool for synthesizing evidence from multiple studies comparing a range of alternative treatment options for the same disease. [1][2][3][4] NMA offers several distinct advantages over a series of standard (pairwise) meta-analyses, such as an increase in precision and power, the opportunity to compare interventions that have not been compared directly in any studies, and the capacity to provide a ranking of all competing treatments. NMA has been increasingly popular, with hundreds of application being published every year. 5,6 Despite the popularity of NMA, there have been some concerns regarding the validity of NMA findings. For example, Del Re et al 7 claimed that considerations related to multiple testing shed doubts on the validity of results of a previously published NMA by Cipriani et al. 8 This was a NMA on the efficacy of antidepressant drugs, which found several important differences between the drugs. Del Re et al claimed that such findings might be due to multiple testing. They performed simulations where they replicated the network of antidepressants by Cipriani et al, assuming, however, no treatment effects between the drugs. Their simulations showed that 72% of the simulated datasets found at least one statistically significant treatment comparison between the treatments, even though in truth there was none. Thus, they concluded that there is a high chance that the findings of Cipriani et al are false positives and that there might be no actual differences between the drugs. Faltinsen and colleagues 9 argue that a consensus is needed on how to address multiple testing issues in NMA, and they call for more methodological research regarding the optimal strategy for addressing this issue in future reviews. The arguments by Del Re et al are based on the concept of statistical significance. It should come as no surprise that using the null hypothesis testing framework in NMA leads to problems, given that the NMA model simultaneously estimates multiple relative treatment effects, ie, comparing all treatments in the network. For example, in a network of 20 treatments, the model estimates 190 different treatment comparisons. The concept of statistical significance, however, has attracted a lot of criticism lately. 10,11 For example, as discussed by Gelman and colleagues, 12,13 it is rather unlikely that any relative treatment effect between different interventions would be exactly zero in the real world (although they might be practically zero). In that sense, testing the null hypothesis of zero treatment effects using a fixed threshold for the p-value actually answers a question for which we already know the answer. Thus, retiring statistical significance 11 would solve the multiple testing issue in NMA, since we would not care anymore about Type I errors. On the other hand, moving away from a hypothesis testing framework would not fix all issues related to having multiple comparisons in NMA. Problems are expected to arise whenever there is some form of selection procedure performed upon the many estimated treatment effects. This is the case when researchers highlight the most extreme results from a NMA, or when they try to identify the best performing treatment in the network and quantify its effects versus a reference treatment. The latter is to the best of our knowledge common practice, ie, most NMAs are performed with exactly this aim in mind. To illustrate why this might be problematic, let us forget about NMA, and imagine a setting where someone simultaneously estimates 100 different parameters (eg, about the effects of different nutrients to some health outcome) and uses these estimates to draw some conclusion, eg, by highlighting in a guideline the nutrient with the largest effect. Irrespective of whether statistical significance is used to dichotomize results, just by chance some of these 100 estimates will be exaggerated, as compared with their true values. The selection procedure (picking up the largest effect and including it in the guideline) is expected to introduce bias and lead to overstated claims. This is expected to happen even if each specific estimate on each own is unbiassed. To draw parallels, when researchers use the results of a NMA to identify the best performing treatment, they run the exact same risk as in the nutrients case. In other words, and irrespective of whether we choose a p-value threshold to call our findings "statistical significant," the fact remains: when we independently estimate a large number of parameters (as the standard NMA model does), we increase the probability of getting some extreme results that do not reflect reality. Such extreme results can then easily be overemphasized in publications and subsequently impact clinical practice, leading to possibly worse patient outcomes and/or unnecessary increase in costs. This raises serious concerns regarding the possible clinical implications. For example, the NMA by Cipriani et al is one of the most highly cited NMAs ever published, and the findings of this paper might have guided clinical practice that affected the lives of millions of patients with depression around the world. Leucht et al placed NMA on the top of the evidence hierarchy when developing treatment guidelines, 14 and influential organizations such as the WHO have recently adopted NMA when drafting guidelines. 15 A clear and definite answer as to whether the scientific community can trust findings from NMAs is urgently needed. In this work, we try to answer this question. We start by discussing the problem theoretically in Section 3. We then What is new? • We argue that the standard NMA model might lead to exaggerated treatment effects regarding the best ranking treatment. It might also be associated with large type I error rates, ie, it can identify differences between treatments which are in truth equally effective or safe. • In this paper, we present a range of Bayesian, NMA models that account for possible similarities between the treatments by modelling exchangeable treatment effects. • We show in theory and in simulations that in several scenarios our models can have a much better performance than the usual NMA model. Potential impact for RSM readers outside the authors' field • Null hypothesis significance testing should be avoided in NMA. • When using NMA to synthesize the evidence regarding a clinical question, researchers can use our Bayesian models to take into account similarities between the treatments and possibly obtain better estimates of relative treatment effects. review some of the most widely used methods for addressing multiple testing. Then, we propose a series of new NMA models by slightly modifying the standard NMA model. Our models are Bayesian models that incorporate prior information about similarities between the treatments, without requiring strongly informative priors. Such models have previously appeared in the literature, [16][17][18][19] but have not been discussed in the context of multiplicity. We argue that such models can naturally address the multiplicity issues of the standard NMA model. These models do not employ any form of correction for multiple testing, eg, like Bonferroni. Instead they rely on shrinking effect sizes towards a common mean. In Section 4, we describe a series of simulations that we performed in order to illustrate the multiplicity issues of the standard NMA model, and to compare its performance to our models. In Section 5, we present the application of this new model in two real clinical examples. | A network of antidepressants Our first example comes from a published NMA by Cipriani et al, 8 on antidepressant drugs for treating unipolar major depression in adults. The network comprised 111 studies comparing 12 different active drugs (new-generation antipsychotics). The primary outcome was response to the treatment. This was defined as the proportion of patients who had a reduction of at least 50% from their baselines score on the Hamilton depression rating scale (HDRS), or Montgomery-Åsberg depression rating scale (MADRS), or who scored much improved or very much improved on the clinical global impression rating scale (CGI) at 8 weeks. The original NMA found important differences between several of the drugs in the network. The reason we chose this particular example is the already discussed concerns raised by Del Re et al . 7 Here, we re-analysed this network using the conventional NMA model as well as our models, aiming to highlight the differences between the two approaches, but also to help settle the difference between Cipriani et al and Del Re et al. The network is depicted in the left panel of Figure 1. | A network of neuroleptic drugs for schizophrenia Our second example comes from a published network comparing five neuroleptic drugs and placebo for patients with schizophrenia. 20 The outcome we will use is binary efficacy, defined by the authors as "experiencing ≥20%-30% reduction from baseline score on the Positive and Negative Syndrome Scale/Brief Psychiatric Rating Scale or a Clinical Global Impression of much improved." Thirty studies were included in the dataset. The network is depicted in the right panel of Figure 1. | METHODS In this section, we begin by describing the standard NMA model and discuss the multiplicity issue. We then present some approaches that have been used in the past for addressing such issues in other contexts. Subsequently, we propose a way to address the multiple testing issue in NMA, via modelling exchangeable treatment effects. | Multiplicity issues in the standard NMA model Usual approaches to fitting a NMA 1,21,22 are based on estimating a set of parameters (the "basic parameters" of the model). These basic parameters are usually set to be the relative treatment effects of each treatment vs a reference treatment. The reference treatment can be chosen arbitrarily, and F I G U R E 1 Network of antidepressants (left panel) and neuroleptic drugs (right panel) the other relative effects in the network (ie, between nonreference treatments) can be expressed as linear combinations of the basic parameters. If the network contains T treatments, there will be T − 1 basic parameters. In total, T(T − 1)/2 different treatment effects can be estimated from the network. Let us assume that the dataset comprises S two-armed studies, where each study i compares treatments t 1i and t 2i . Let us also assume that each study reports the observed relative treatment effect y i and the corresponding standard error s i . The standard common-effect ("fixed-effect") NMA model can be written as . d 1 is set to zero without loss of generality, in which case d t is the effect of treatment t vs treatment 1. The T − 1 different d t are the basic parameters of the model. In a Bayesian setting, we usually assign "vague" (flat) prior distributions to the basic parameters, eg, d t N(0,100 2 ), for t = 2,3,…,T. A random effects NMA model which assumes consistency can be written as follows: Model I: Standard NMA model Note that including in model I studies that compare more than two treatments requires changing the corresponding univariate normal distributions to multivariate ones. The model above applies to any type of outcome (continuous, binary, and time-to-event) and follows a two-stage, contrastbased approach. In this approach, at the first stage we obtain study-specific estimates of relative treatment effects, and at the second stage we perform the NMA. The model can be modified to accommodate a one-stage analysis, using armbased likelihoods; eg, for binary outcomes we could use a binomial likelihood for each arm in each study. Here, τ denotes the heterogeneity of treatment effects. Model I assumes a common τ for all comparisons in the network. 23 This is a very usual assumption in NMA, but there are more general formulations. 24 Setting τ = 0 leads to a commoneffect NMA model. If we use model I to make inference about a prespecified treatment comparison of interest, each estimated treatment comparison is unbiassed when seen in isolation, and there are no multiplicity issues. The problems arise when someone uses a collection of estimates to perform some form of selection. To explore multiplicity, we consider the case of a star network (ie, when all studies are two-armed and compare a treatment to the reference) where all relative treatment effects are zero. If we focus on the case where τ is known (eg, the common-effect model

τ = 0), the T − 1 estimates of the basic parameters are independent. Thus, if we use the concept of statistical significance to categorize results, we expect that on average a fraction 1 − (1 − 0.05) T − 1 of such NMAs will find at least one basic parameter to be statistically significant at the usual 5% level. In networks with many treatments, this false positive rate becomes very large. Eg, if we have a network of T = 20 equally effective treatments, there is a 62% chance of at least one statistically significant finding for the basic parameters. The expected fraction of NMAs with at least one statistically significant finding across all treatment comparison in the network (ie, not limiting to basic parameters) will of course be even larger. In summary, if the null hypothesis of no treatment effects in the network holds, a NMA is expected to have a very large false positive rate-the more the treatments in the network, the larger this rate. This implies that if model I is used in conjunction with a statistical significance threshold to answer the question "is any of the treatments in the network effective/safe?", or the question "are there any differences between the treatments in the network?", there is a high probability of spurious findings, if treatments have the same effects. Moving away from the case of star networks, for the case of networks where nonreference treatments are also directly compared, the basic parameters are no longer independently estimated. Eg, a BC study will jointly inform the basic parameters AB and AC. Thus, the false positive rate for the basic parameters is no longer straightforward to calculate, and it will depend on the geometry of the network. Similarly, extending the model to a random-effects setting (where τ is estimated) makes the theoretical calculation of the expected false positive rate nontrivial, because it will depend on the assumed variance-covariance structure of the random effects, the number of studies per comparison, and the structure of the network. Even if we abstain from categorizing findings according to their p-values, multiplicity can still be an issue. If T is large, we expect that by chance alone some of the basic parameters will receive large positive values, and some will receive large negative values. The larger the T and the smaller the precision of the studies, the more extreme such results are expected to be. Thus, when identifying the best and worst performing treatment in the network we run the risk of overestimating the corresponding treatment effects and ending up with an exaggerated estimate regarding how well/badly the best/worst treatment performs. This is expected to be the case regardless of what measure we use to identify the best treatment (eg, SUCRAs 25 or any other measure). The criticism of Del Re et al 7 is closely related to what we have discussed so far in this section. One should note, however, that the probability P 1 = P NMA shows treatment effects are not zero ð Þ treatment effects are zeroÞ is not the same as the probability P 2 = P treatment effects are zero ð j NMA shows treatment effects are not zeroÞ: The simulations of Del Re et al and the calculations presented in this section correspond to P 1 , but among the two, the clinically interesting quantity is P 2 . An attempt to distinguish between P 1 and P 2 naturally leads into a Bayesian way of thinking, since P 2 is a probability only in the Bayesian sense. Following this simple argument, in the next paragraphs we discuss how multiplicity can be naturally addressed within a Bayesian framework. | Currently available methods for correcting for multiple testing One popular way to fix the multiple testing issue in medical research is to use some form of correction method. Probably, the most popular correction method is the so-called Bonferroni correction (but there are other similar approaches, eg, see 26 ). Let us assume that we want to test n independent hypotheses at a desired significance level a (where usually a = 0.05). Using the Bonferroni method, we instead test each individual hypothesis on a "corrected" level equal to a/n. This very simple method comes however with several drawbacks. Perhaps, the most important one is that such corrections mitigate the false positive rate at the expense of statistical power. In other words, the Bonferroni as well as other alternative methods focus on holding the rate of type I error (incorrect rejection of the null hypothesis) at the nominal level, but by doing so they increase the rate of type II error (failure to reject an incorrect null hypothesis). Depending on the problem, this increase might be quite important. For these reasons (among others), the use of correction methods to attack the multiple testing issue has been strongly criticized. 27,28 For the case of NMA where we estimate T − 1 basic parameters, following the Bonferroni correction, we would call a basic parameter statistically significant only if the corresponding P value was lower than a/(T − 1). This, however, would lead to a dramatic loss of our power to detect any treatment effects, especially for larger networks. Moreover, it would lead to the counterintuitive case where the interpretation of a result (eg, regarding the comparison of treatments A and B) would depend on seemingly irrelevant pieces of information, such as the number of total comparisons in the network. Eg, the interpretation of results for A vs B would depend on whether A was also compared with D in other, independent trials (while it would not depend on the actual results of these A vs D trials). For these reasons, we believe that the use of Bonferroni (or any other similar) adjustment in NMA would create more problems than it would potentially solve. Another approach to handle the multiple testing issue is to instead focus on the "false discovery rate". 29 This approach does not focus on reducing the familywise error rate (ie, the probability of having at least one false positive) but controls the proportion of false positives. This is a less conservative method than the Bonferroni correction, but it has a larger power. As Gelman et al note, 12 such methods are generally useful in fields like genetics, but they are expected to be less useful in areas such as NMA, where we usually do not test for thousands of different hypotheses and where true effects are less likely to be exactly zero. In the next paragraph we describe how the introduction of a small modification to the usual NMA model can mitigate the multiplicity issues while bypassing some of the problems of the aforementioned correction methods. This approach closely follows a Bayesian multilevel modelling approach, discussed by Gelman et al. 12 In what follows, we start by outlining this approach and adapt it to the case of NMA. Note that multilevel hierarchical models that model exchangeable treatment effects have previously appeared in the NMA literature, eg, by Dakin et al, 16 DelGiovane et al, 17 Warren et al, 18 Owen et al, 19 and Senn et al 30 but have not been discussed in relation to multiplicity issues. | Network meta-analysis models with exchangeable treatment effects In their paper, Gelman and colleagues 12 employ data from the "Infant Health and Development Program," an intervention targeted at premature and low-birthweight infants. The intervention was administered at eight different sites, and the aim of the analysis was to estimate treatment effects for each site individually. As the authors discuss, the analysis can be performed in two extreme ways. One extreme is what they term "extreme pooling," which assumes equal (common) treatment effects across sites. This corresponds to the standard, "identical parameters" analysis, eg, as described by Spiegelhalter et al, 31 page 92. Extreme pooling avoids the multiple testing issue but of course fails to answer the question at hand, ie, it cannot estimate the treatment effect for each individual setting. The other extreme, which Gelman et al term "no pooling," is to estimate treatment effects in each site separately (ie, independently). Spiegelhalter et al term this "independent parameter" analysis. 31 This analysis can answer the question of interest, but might be problematic due to multiplicity issues. Gelman et al then describe a middle ground approach which they term "partial pooling," and which involves random effects modelling. Following this method, the true treatment effects in the sites are assumed to be exchangeable, ie, to be realizations of a common underlying distribution of treatment effects. Instead of inflating the uncertainty of each estimate, this approach shifts the point estimates towards the common mean, effects are shrunk away from extreme values, and the multiple comparisons problem gets resolved without completely diminishing the statistical power. This is the "exchangeable parameters" analysis in Spiegelhalter et al. 31 We now discuss how these ideas can be applied in a NMA framework. Following the identical treatment effects approach (extreme pooling), we can assume that the effects of all treatments in the network vs the reference are the same, and we can perform a simple pairwise meta-analysis. This meta-analysis does not suffer from issues related to multiplicity and will answer the question "are treatments on average better than the reference?". But, it cannot answer the question "how do different treatments compare to each other?". On the other extreme (ie, no pooling), we have the usual NMA model. This analysis tries to estimate relative effects between all treatments, but, as we discussed, it involves estimating a (possibly large) number of parameters, with all the (previously described) problems that this might entail. In order to follow a partial pooling (exchangeable parameters) approach, we propose a set of alternative NMA models. More specifically, we describe a set of Bayesian models that utilize prior information regarding possible similarities between the treatments. Let us first focus on the usual scenario of a network comprising several active treatments as well as a clearly defined control treatment, such as placebo, no treatment, waiting list, treatment as usual, etc. Moreover, let us assume that the active treatments are somewhat "similar" to each other (but not to control), eg, that they are drugs whose molecules have comparable structure; or that they are treatments known to work through similar biological pathways; or that they are similar surgical operations, etc. In such cases, it might be reasonable to think that the true effects (with respect to a specific effect measure) of these similar treatments form a common underlying distribution of effects. A Bayesian random-effects NMA model with exchangeable treatment effects for the case when only two-arm studies are present in the dataset can be written as follows: Model II: NMA model with exchangeable treatment effects Weakly informative priors can be employed for μ d and τ 2 d . Here we have assumed treatment 1 to be the control treatment, ie, to be dissimilar to the rest of the treatments. The basic parameters of the model (ie, the relative treatment effects of all treatments vs treatment 1) are no longer assigned independent prior distributions as in model I, but are modelled exchangeably. This exchangeability encodes our prior information about the similarity of treatments. Model II "pushes" the estimated relative treatment effects of the different active treatments towards their common mean, to an extent determined by parameter τ d , where τ d = 0 corresponds to extreme pooling, while τ d ! ∞ corresponds to no pooling. As discussed by Gelman et al, 12 this model will tend to correct the multiple comparison issue. Moreover, this model makes intuitive sense: when treatments are similar, it is reasonable to assume that their relative effects vs the control follow from a common underlying (normal) distribution, whose parameters we are estimating from the data. When there is large uncertainty regarding the effects of a treatment (eg, when there are only a few small trials that have studied it), then the corresponding estimates of all treatments vs the reference will be strongly pulled towards the overall mean. When uncertainty is small, the corresponding estimate will be shrunk only a little. In general, when we have a densely connected network with many studies per treatment comparison, we expect to find small differences between models I and II. Conversely, in sparse networks with

few studies per comparison, differences might be more pronounced. The only difference between model II and the standard Bayesian NMA model I regards the priors for d k . The standard model assigns independent, flat prior distributions to the effects of the various treatments. Thus, for example, in the standard model, an odds ratio (OR) equal to 0.2 regarding treatment X vs control, has the same prior likelihood as an OR equal to 5. In contrast, in Model II, the prior distributions of the effect sizes of the similar treatments vs the control are exchangeable, not independent. This implies, for example, that if the ORs vs the control for all treatments except X are found to be centred on 0.2, then we expect the OR of X vs the control to be also around 0.2. Thus, we would assign a much higher prior probability to this OR being around 0.2, rather than it being around 5. Note however that the marginal (unconditional) prior for this OR remains flat. When there are not enough treatments in the network to allow for a precise estimation of the hyperparameters μ d and τ d , the estimated treatment effects using model II will tend to become similar to the ones estimated using model I. For the extreme case of T = 2 treatments, models I and II will become equivalent when it comes to estimating treatment effects, as long as flat prior distributions are used for all parameters in both models. In that case they both correspond to a simple, pairwise meta-analysis model. Thus, metaanalysts can safely use model II even for the case of small networks. Let us now also consider the scenario where there is no clear control treatment in the network. This is the case for the network of antidepressants, which we described in Section 2.1. In such cases, model II is not well suited for the analysis, because it assumes different priors between comparisons vs the reference, and comparisons which are not vs the reference. This is problematic when there is no clear choice for the reference treatment. In such scenarios, we can implement a symmetric version of model II as follows: Model III: Symmetric NMA model with exchangeable treatment effects Model III can be more compactly written as d = (d 1 , ' is a vector with T identical elements, and Σ d is a T × T matrix with τ 2 d in the diagonal and 0:5τ 2 d in all nondiagonal elements. Note that in this model μ d does not contribute to the likelihood and can be set to zero. Also note that, as one would intuitively expect, the joint likelihood of treatment effects in model III is identical to the likelihood of the effects between similar treatments in model II (ie, after removing the reference treatment 1), with the only difference being a rescaling of τ 2 d . Model II can be extended to the case when active treatments (ie, not the control) belong to multiple classes, 16,18,19 where C 1 is the first class of similar treatments, C 2 the second class, and so on. For example, in a NMA of interventions for pain relief, one could include placebo, several nonopioid analgesics (class C 1 ), and several opioid analgesics (class C 2 ). Similarly, a NMA could include behavioural interventions (C 1 ) as well as pharmacological interventions (C 2 ) for the same condition. In such cases, model II can be modified to: Model IV: NMA model with exchangeable treatment effects for multiple classes In this model, we group the basic parameters that correspond to the treatments of each class, and we assign them a common prior distribution. Thus, each class C x is characterized by a class effect vs the reference (μ d,x ) and a classspecific variance (τ 2 d, x ), which measures the variability of treatment effects within this class. An additional assumption that can be used to simplify model IV is to set τ d,x to be common across all classes, ie, τ d,x = τ d 8 x. [16][17][18][19] Also, μ d,x could be modelled to be exchangeable across (some of the) classes. Note that model IV can be used if a class includes a few, or even a single treatment. Of course, in this case, the corresponding μ d,x and τ d,x will not be easy to estimate with precision, unless we employ some modelling assumption like the ones discussed above. If all classes in model IV include a single treatment and μ d,x are given independent, flat prior distributions, model IV is equivalent to the standard NMA model, when it comes to estimating d k and τ 2 . Finally, we can extend our symmetric model III for the case when treatments belong to multiple classes and there is no clear control treatment. Let us choose, without loss of generality, treatment 1 belonging to class C 1 to be the reference treatment of the network. Our symmetric multiclass model can be written as Model V: Symmetric NMA model with exchangeable treatment effects for multiple classes where again we could simplify by assuming τ d,x to be common, or by setting μ d,x to be exchangeable among (some of the) classes. This model could be used for example in a NMA of several opioid and nonopioid analgesics, where no placebo-controlled studies are included. | Considerations regarding statistical power Because models II-V incorporate prior beliefs about similarities between some of the treatments, they are expected to have less power to detect differences between them. In contrast, they are expected to have more power to detect differences between treatments not assumed to be similar. Eg, using model II instead of I, we may have less power to detect differences between drugs, but more power to detect differences between drugs and placebo. Using model III, we expect to have less power in detecting treatment effects between any of the treatments in the network. Using models IV and V, we expect to have less power to detect differences between treatments within each class, but more power to detect differences between treatments belonging to different classes. In Section 4, we use simulations to explore how models II and III compares to the standard NMA model I, with respect to the trade-off between the probability of spurious findings and power. | Overview In this section, we describe the simulations we performed to illustrate the multiple testing issue in the standard NMA model (model I), and to compare this model with our proposed models II, III, IV, and V. The primary focus was in comparing the models under different sets of treatment effects, but we also varied other factors, such as network geometry and the existence of random effects. We performed two simulation studies, each one including several scenarios. In simulation study 1, we explored the situation when all treatments were equally effective. The initial scenarios had one trial per comparison, and hence heterogeneity was ignored in the analysis. To explore the impact of heterogeneity, we also generated scenarios with multiple trials per comparison. For each scenario except scenario D (where we had three treatments in the network), we explored networks of two contrasting geometries: star networks and fully connected networks (Figure 2). Especially for scenario D, we only generated fully connected networks. We compared models II and III with three variants of model I, using two different priors on the treatment contrasts (flat and informative priors). In simulation study 2, we explored the situation where some treatments had different effectiveness. We explored several scenarios for the treatment effects, for both star and fully connected networks of 10 treatments. In this study, we simulated only one trial per comparison, ie, heterogeneity was ignored. For the analysis with model I, we used only flat priors. In this simulation study, we also used models IV and V. For model IV, we assumed two classes of treatments: treatments 2 to 5 comprised the first class of similar treatments, and treatments 6 to 10 the second; treatment 1 was the reference. For model V, we assumed two classes, treatments 1 to 5 and 6 to 10. | Data generating mechanisms for each scenario For each scenario, we simulated 1000 independent datasets. We only simulated two-armed studies. For each study i, we simulated an observed relative treatment effect (y i ) and a corresponding standard error (s i ). The study-specific observed effects (y i ) were generated in different ways depending on the scenario. For each study, s i was generated using a chi-square distribution with one degree of freedom, ie, s i 0:5 + 0:2 χ 2 1 . For scenarios B, we simulated random effects. There we assumed a common variance of the random effects (τ 2 ) for all treatment comparisons in the network, and we generated it by drawing from a log-normal distribution τ 2 LN(−2.56,1.74 2 ) for each dataset. This was based on the empirical distributions of heterogeneity proposed by Turner et al. 32 Note that we did not use this distribution for the F I G U R E 2 Network structures explored in simulation studies. Star network (left) and fully connected network (right) analyses (see next paragraph). All data were simulated in R. 33 Codes are available online in https://github.com/esmispm-unibe-ch-REPRODUCIBLE/the_dark_side_of_the_ force. We investigated eight scenarios (noted as A-H in Table 1). For each scenario X (where X denotes all scenarios A to H, except C), we simulated data for a star network (termed scenario X.1 in Table 1) and a fully connected network (X.2). Simulation study 1 assumed zero treatment effects in the network and included scenarios A, B, C and D. In scenarios A, we simulated a network of 10 treatments, one study per comparison, giving a total of nine studies for scenario A.1 and 45 studies for scenario A.2. In scenarios B, we assumed 10 treatments, and we simulated three studies per comparison, assuming heterogeneity. In scenarios C, we only had two competing treatments, and each dataset included five studies. In scenarios D, we assumed three treatments in the network, one study per comparison. In simulation study 2 (scenarios E-H), we assumed nonzero treatment effects in the network. We only used networks of 10 treatments. In scenarios E, we assumed one of the treatments to be the control, and the relative effects of all other treatments vs the control were assumed equal. In scenarios F, we split the treatments in two groups (treatments 1-5 and 6-10). Treatments in the same group were assumed to be equally effective, and we assumed different effectiveness between groups. In scenarios G, all treatments had different effectiveness. In scenarios H, we assumed two groups of treatments (treatments 2-5 and 6-10) and a control treatment (treatment 1). Treatments in the same group had equal efficacies vs the control. The details of the data generating mechanisms in each scenario are presented in Table 1. | Methods of analysis H. 2 Fully connected repeated the analysis with the same model, using informative priors d k N(0,1). This was to check whether reasonably informative priors can mitigate issues related to multiplicity without having to resort to more complicated modelling. Next, we analysed each dataset with models II and III, using vague priors for the extra parameters, ie, μ d N(0,10) and τ d U(0,2). For scenarios B and C, where we simulated multiple studies per comparison, we used the random-effects version of these models, with a vague prior for the standard deviation of random effects, τU(0,5). For the other scenarios, we used the common-effect version of all models. For simulation study 2, in each simulated dataset, we fitted model I with flat priors, d k N(0,100 2 ). We then repeated the analyses with the common-effect versions of models II and III, assuming vague priors, μ d N(0,10) and τ d U(0,2). | Fitting details We fitted all models in R using the package rjags. For each model, we fitted two independent chains, 15 000 iterations per chain, with a burn-in period 5000 iterations. In a sensitivity analysis, to ensure our results were robust, we also used the netmeta 34 command in R 33 to fit the standard NMA model in a frequentist setting. The code we used for model

fitting is also available online in https://github. com/esm-ispm-unibe-ch-REPRODUCIBLE/the_dark_side_ of_the_force. | Measures of model performance After performing the described analyses in each of the 1000 datasets, we extracted the posterior median estimate, the corresponding standard deviation, and the 95% credible intervals (CrI) for all treatment effects. In each analysis, we also identified the best and worst treatment based on the SUCRA values for ranking treatments. 25 Based on the extracted information, for each scenario of simulation study 1 (where treatments were equally effective) and for each method of analysis, we calculated (a) the mean absolute error for the basic parameters (treatment effects vs treatment 1); (b) the mean bias of the basic parameters; (c) the mean estimate and mean standard deviation for the best vs the worst treatment in the network; (d) the mean estimate and mean standard deviation for the best treatment in the network vs the reference; (e) the percent of datasets that showed with confidence that there were nonzero effects in the network, where in truth there were none. For (e), we checked whether the 95% CrI of the estimated relative treatment effects in the network (among any two treatments) included zero. For scenarios of simulation study 2 (where we assumed nonzero treatment effects in the network) and for each method of analysis, we calculated (e) as the percent of datasets that showed with confidence that there were nonzero effects among treatments that were in truth equally effective, and we additionally calculated (f) power, ie, the percentage of treatment effects that were shown with confidence to be nonzero (95% CrI excluding zero), when in truth they were nonzero. 4.6 | Results from the simulations 4.6.1 | Main findings A summary of the findings from simulation 1 is shown in Table 2. In scenarios A and B, where we assumed the 10 treatments in the network to be equally effective, models II and III clearly outperformed the standard NMA model I. Model I was associated with higher mean absolute errors; it gave more exaggerated treatment effects regarding the best vs worst treatment, and best vs reference; it also had the largest percentage of datasets showing with confidence that there are nonzero treatment effects among some of the treatments in the network. The results from scenario C showed the equivalence of model I with flat priors and model II, when T = 2. In scenario D, where we had a network of only three equally effective treatments, model III clearly outperformed model I. All models were unbiassed in all scenarios A to D. A summary of the findings from simulation 1 is shown in Table 3. In scenarios E, where we assumed all treatment effects vs the reference to be equal, model II performed overall the best. The standard NMA model (model I) gave the most exaggerated estimates regarding the best treatment and had the highest rate of false positive findings. In scenarios F, where treatments were split in two groups, models V performed best. Model I had the worst performance in terms of exaggerating the effects of the best treatment, and also false positive findings. Models II and III had the smallest power. In scenario G.1, where all treatment effects in the network were nonzero and the network was star-like, model I was the worst in terms of mean absolute error and exaggeration of treatment effects, and second worst in power. Model II had the best performance overall, followed by IV. In scenario G.2, all models performed comparably, but model II had the lowest, and model IV the largest power. In scenarios H, model IV had overall the best performance. Model I had the largest mean absolute error, it showed the most exaggerated results regarding the effects of the best treatment and had the largest false positive rate. In summary, in almost all cases, the standard NMA model (model I) led to the most exaggerated claims regarding treatment effects of the best treatment in the network. This exaggeration was especially pronounced for star networks, rather than fully connected ones. In scenarios, where no treatment effects were simulated, model I gave the largest T A B L E 2 Overview of results for simulation study 1 (zero true treatment effects in the network). For all scenarios except B. proportion of networks that found nonzero relative treatment effects. In the next sections, we provide some additional results for some of the scenarios. | Reversal of effects For scenarios G, where all treatments were assumed to be differently effective, we calculated the percent of estimates for which the 95% CrI did not include zero, but showing a wrong direction of effects (eg, when treatment 2 was found to be worse than treatment 1, although in truth it was better). For scenario G.1, using model I, we found that 1.1% of all estimated treatment effects across all datasets showed a reversed effects. For all other models, this was below 0.00%. Similarly, for G.2, the percentages were 0.03%, 0.00%, 0.02%, 0.00%, and 0.06% for models I to IV, respectively. These results highlight another possible advantage of modelling treatment effects exchangeably. | Treatment rankings In all scenarios of simulation study 1, due to the way we generated the data in this scenario treatments were identical. Thus, we would ideally like to see on average small differences between the maximum and minimum SUCRA value in each NMA. This, however, was not the case for model I. | Further findings Aiming to show the equivalence between model I with flat priors and model II in scenario C, we calculated the mean absolute difference of the estimated treatment effects across all datasets between these two models, and we found it to be 0.002. The mean absolute difference of the corresponding standard deviations was 0.001. These results confirm what we theoretically anticipate, ie, models I and II are equivalent when T = 2. | Antidepressants We used the data from Cipriani et al 8 to fit the usual random effects NMA model I. Following the original publication, we chose fluoxetine to be the reference treatment for the model. We also fitted the symmetric random-effects NMA model III. We opted for the symmetric model because there was no placebo or other obvious control treatment in the network. We fitted the models in R, using the rjags package. For both models, we used τU(0,3) as a prior for the standard deviation of random effects. For model III, we assumed μ d N(0,1) and τ d N(0,1)I(0, ), ie, the positive part of N(0,1). We used two independent chains per analysis, and we performed 50 000 iterations per chain, discarding the first 15 000. In both models, the posterior median for τ was 0. Table 5. Based on the results of Table 4, we conclude that there is a strong evidence of difference in efficacy between some of the antidepressants. These results are unlikely to be false findings due to multiplicity. In a series of sensitivity analysis, we (a) fitted the standard NMA model I after choosing a different treatment to be Venlafaxine the reference, and with τU(0,5); (b) fitted model III using less informative prior distributions: τU(0,5), μ d N(0,10), and τ d U(0,2); and (c) fitted the random-effects version of model II. These sensitivity analyses did not give substantially different results for most comparisons. In practice, one might want to perform additional analyses using models IV and/or V, after splitting the drugs into classes. However, given that the main focus of this paper is on the methods, we did not pursue this any further. We provide the code and data that we used for model fitting both for R and for OpenBUGS 35 in https://github. com/esm-ispm-unibe-ch-REPRODUCIBLE/the_dark_side_ of_the_force. | Neuroleptic drugs We used the data by Klemp et al 20 described in Section 2.2 to first fit the usual random effects NMA model I. We also fitted our NMA model II. We chose model II over the symmetric model III, because in this example there was an obvious control treatment in the network, ie, placebo. We fitted the models the same way as in Section 5.1. Table 6. In both models, all drugs were shown to be more efficacious than placebo. However, results from model II pointed to smaller differences between some of the active treatments. For example, the OR for aripiprazole vs clozapine was 0.63 [0.40; 1.00] using the standard model, while using model II it was estimated to be 0.73 [0.47; 1.09]. This highlights that model II might be more conservative in identifying differences between active treatments. The ranking of the treatments was unchanged using the two models and is shown in Table 7. | DISCUSSION In this paper, we highlighted the possible implications of the multiplicity issue in NMA. We presented the problems associated with multiplicity using both theoretical arguments as well as simulations. We showed that the model commonly used for NMA may give exaggerated estimates of the effects of the treatment identified to be the best in the network. It may also be associated with a high probability of (falsely) showing differences between treatments, when actually there is none. The problems stem from the fact that the NMA model simultaneously estimates tens (or even hundreds, depending on the network size) of relative treatment effects. By chance alone, some of these estimates may be very large. Any selection procedure following the estimation (eg, identifying the best ranking treatment, or focusing on "statistically significant" results) will lead to exaggerated findings. These issues will be more pronounced when the number of treatments in the network is large. In this paper, we discussed how a set of alternative Bayesian NMA models can be used to address problems associated with multiplicity. The key difference between these models and the standard NMA approach is that the former model treatment effects exchangeably, ie, assuming that the true effects of the treatments are realizations of one or more common underlying normal distributions. This small modification to the standard NMA model is enough to mitigate the multiplicity issue. The price to pay is a possible decrease in the statistical power to detect differences between treatments assumed to be similar (although there might be an increase in the power to detect differences between nonsimilar treatments). Simulations showed that in a range of different scenarios, our models had a better performance than the standard model. The latter gave in almost all cases the larger mean absolute error regarding the estimates of the basic parameters; it gave the most exaggerated estimates regarding the relative effects of the best treatment vs the reference, and best vs worst treatment. Also, it was worse in terms of the trade-off between false positives (falsely identifying differences among equally effective treatments) and power. Eg, in scenario E.1 (star network, all treatments had equal effects vs the reference), when we switched from model I to model II, we saw that the effects of the best treatment vs reference were estimated with a much smaller error, power almost doubled (from 37% to 69%), while false findings reduced from 55% to 3%. Likewise, in scenario G.1, where all treatments were generated to have different effectiveness, exaggeration of effects was much more pronounced in model I, while power was more or less similar across all models. One limitation of our simulations is that we did not include scenarios with multi-arm studies. Note however that the problems with multiplicity arise in NMA when some form of selection among the estimated effects is performed, regardless of whether these estimates used data from multiarm studies. At the same time, as we saw in simulations, when the network includes a lot of information (eg, when it is more densely connected), the exaggeration of the effect of the best treatment is expected to be smaller. In that sense, keeping the number of studies and number of patients per study-arm constant, networks with multi-arm trials are expected to give less exaggerated effects than networks with two-arm studies only. Also, let us note that the power rates we found for some of the scenarios are unrealistically large. This was an artefact of the (arbitrary) choices for the data generating mechanisms, and the fact that for some scenarios we considered fully connected networks. In practice, it is rather unlikely that a network of 10 treatments will ever be fully connected, and the high

power we showed in some scenarios should not be interpreted as a general feature of NMA. In addition, readers should keep in mind that the aim of our simulations was to compare the relative performance of the models, not to assess the absolute power of NMA in usual conditions of data availability. The practice of dichotomizing evidence according to "statistical significance" has been heavily criticized lately. We also think that researchers should abstain from characterizing evidence as statistically significant or not, and we showed in our simulations that this practice can be especially problematic in NMA, for all models we used. However, as we presented in this article, the multiplicity issue in NMA is not limited to multiple testing. Thus, although dropping the notion of statistical significance is a step in the right direction, it is by itself not enough to safeguard us against all multiplicity issues in NMA. In this paper, we showed how utilizing prior information regarding similarities among treatments can help address the issue and possibly lead to more trustworthy NMAs. One important aspect of our models is that they require some subjective decisions to be made. The decision of which treatments to include in the network (or how to group treatments in models IV and V) may be more important in our models as compared with standard NMA. Eg, with model II, including in the network an older drug that is less efficacious might lead to more conservative estimates about the newer drugs. In this scenario, however, if the older drug is substantially different from the newer drugs (eg, in terms of the molecule structure), it might make more sense to include it in a separate class, ie, using model IV. Also, the assumption of exchangeability of treatment effects might be subject to criticism, eg, in cases where drugs use different biological pathways. Furthermore, exchangeability might be a difficult concept to convey to clinicians. One approach might be to ask questions such as "You are trying to assess treatment X, and you are told that treatment Y is effective/ineffective/safe/unsafe. Would this information about Y impact your judgement about X?" "You are trying to assess treatment X, and you are told that one of the other treatments in the network is effective/ineffective/safe/unsafe. Would you need to know which other treatment it was?" Answers "yes" and "no," respectively, would support an exchangeability assumption. Decisions about exchangeability among treatments need to be taken a priori, by content experts, taking into account existing knowledge about the nature of treatments (and not the results of the studies in the network). Moreover, these decisions should be clearly reported and justified, preferably in the protocol stage of the review. Of course a Bayesian analysis will always include some level of subjectivity, but to our view this is not a disadvantage, as long as all decisions are transparent. An obvious question at this point is whether one should prefer one of our NMA models for the primary analysis, over the standard NMA model. We think that whenever there is prior knowledge regarding possible similarities among some of the treatments in the network, our models should be considered. The answer might also depend on the nature of the research question. For example, for the case of a NMA of a serious safety outcome (eg, mortality), where several drugs are compared with placebo, researchers might be more interested in increasing the power to detect differences vs placebo, rather than between drugs. In such case, our models might be more appropriate to use. Another case where our models might be very useful is when for some treatments there is limited and/or biassed evidence. Imagine that there is available only a single study, or a few small studies, of a new drug vs placebo. Assume that in these studies, due to publication bias (or due to other biases, or due to chance alone), the effect of the new drug is grossly exaggerated. This may be a relatively common situation; Fanelli et al showed that earlier studies often show inflated results. 36 In a case like this, the standard NMA model may identify this drug as being the best in the network and may estimate large treatment effects vs all other drugs. Conversely, our models will give less biassed estimates in this scenario, by pushing the effectiveness of this drug towards the mean effectiveness of the rest of the drugs. The models we have described could be further extended for the case when we have information about predictors at the level of the treatments. Eg, let us assume that we have a network of similar drugs being compared with placebo, and that there is available information regarding a treatment characteristic that is expected to affect the outcome, such as the duration of the treatment. In that case, instead of pulling the treatment effects towards the overall mean, we could use a multilevel regression model to pull treatment effects towards the regression line. In cases where treatments in the network cannot be assumed similar (so that our hierarchical models cannot be used), we think that the considerations presented in this paper can still be useful in highlighting the potential issues with performing a NMA to identify the best performing treatment and also in using the concept of statistical significance in NMA. To summarize, we think that the models we proposed are a useful addition to a network meta-analyst's arsenal of statistical methods, and that in several scenarios they should be preferred over the standard NMA model. DATA AVAILABILITY STATEMENT We provide the code and data that we used for model fitting both for R and for OpenBUGS 35 in https://github.com/esmispm-unibe-ch-REPRODUCIBLE/the_dark_side_of_the_ force. Prevalence and risk factors for nasal carriage of Staphylococcus aureus in children attending anganwaries (preschools) in Ujjain, India Background Children with nasal carriage of S. aureus play an important role in community spread of S. aureus and methicillin-resistant S. aureus (MRSA). Screening the nasal carriage isolates of S. aureus for antibiotic resistance patterns will provide guidelines for empiric therapy of community-acquired infections. The aim of the present study was to determine the prevalence of S. aureus and MRSA and it’s in vitro antibiotic susceptibility pattern among children in anganwaries (preschools) of Ujjain city India. This work is an extension to our previous publication in BMC Pediatrics (http://www.biomedcentral.com/1471-2431/10/100). Methods A prospective study was done among children aged 1 to 6 years of age attending 100 anganwaries chosen purposely for the study to evenly cover the city. From each anganwari 10 children were randomly selected for nasal swabbing. Children having pyoderma were not included. Information on risk factors for nasal colonization was collected using a pre-tested questionnaire. Swabs from anterior nares were plated on 5% sheep blood agar. Antibiotic susceptibility tests were performed using Kirby-Bauer’s disc diffusion method according to performance standards of Clinical and Laboratory Standard Institute guidelines. Results A total of 1002 children were included in the study. The prevalence of S. aureus nasal carriage was 35% (95% confidence interval CI 32.07 to 37.98) and that of MRSA nasal carriage was 29% (95% CI 24.28 to 33.88). The factors that were independently associated with nasal carriage of S. aureus were: “age-group” i.e. as the age increased beyond the age of 2 years the OR of nasal carriage decreased, “family size of more than 10 members” OR 2.59 (95% CI 1.53-4.37; P < 0.001), and protein energy malnutrition Grade 3 or 4 (OR 1.40, 95% CI 1.04-1.90; P = 0.026). The resistance pattern of S. aureus and MRSA showed resistance not only to single antibiotic class but co-resistance and multi-drug resistance was also common. Conclusions The high rates of nasal carriage of S. aureus and MRSA and presence of resistance to commonly used antibiotics are disturbing. Antibiotic stewardship programmes that promote judicious use of antibiotic along with strategies to prevent community spread of S. aureus are urgently needed. Background Worldwide there is increasing trend of community spread of methicillin-resistant S. aureus (MRSA) [1,2]. The dissemination of plasmid-borne mec A gene in the MRSA isolates makes them resistant to multiple antibiotics [3]. Preventing further spread of MRSA has become a major global public health problem [3,4]. Antibiotic use is the main determinant of extent of antibiotic resistance in a given geographical area [5,6]. The spread of multi-resistant strains of bacteria in the community is compounded with a paucity of new classes of antibiotics in the pipeline [7]. S. aureus is the most common bacterial cause for diverse range of infections, from folliculitis and furunculosis to life-threatening infections, including sepsis, deep abscesses, pneumonia, osteomyelitis, and infective endocarditis [2]. S. aureus colonises the skin and mucosae of human beings and several animal species [8]. Although multiple body sites can be colonised in human beings, the anterior nares of the nose is the most frequent carriage site for S. aureus [8]. The hand carriage and nasal carriage of S. aureus are strongly correlated [9] suggesting that contaminated hands most commonly cause the colonization of the nares. Nasal carriers can act as "cloud" individual during rhinitis, dispersing S. aureus into the environment [10]. Also, causal association between S. aureus nasal carriage and staphylococcal disease has been confirmed by many studies [2]. Therefore, it is important to study the prevalence of nasal carriage of S. aureus and factors associated with such carriage to prevent spread of S. aureus in the community. Screening the nasal carriage isolates of S. aureus for antibiotic resistance patterns will provide guidelines for empiric therapy of community-acquired infections. In a study from same geographical area in India the main risk factors for nasal carriage of S. aureus were related to overcrowding i.e. children going to school and preschool [11]. Therefore, the aim of the present study was to determine the prevalence of S. aureus and MRSA and its in vitro antibiotic susceptibility pattern among children in anganwaries (preschools) of Ujjain city, Madhya Pradesh, India. Methods This was a prospective study conducted during a 28-month period from January 2008 to April 2010. Study area and setting The study was conducted among children attending anganwaries (preschools) of Ujjain city. The anganwaris are focal point for delivering the services under the Integrated Child Development Scheme (ICDS) run by the Ministry of Women and Child Development, Government of India. The services include a package of nutritional supplementation, immunisation, health check-up, and nonformal pre-school education to children of the age group 3-6 years and health and nutrition education to women in the age group 15-45 years. Each anganwarie caters to about 1000 community members and is served by a female health care worker. There were 302 anganwaries in urban Ujjain at the time of start of the study. We purposely choose 100 anganwaries to evenly cover the city. Pilot study A one-month pilot study was conducted in December 2007 on 45 children from five anganwaries to estimate the nasal carriage rate of children attending anganwaries to facilitate sample size calculation and to study the feasibility of data collection. The nasal carriage rate for S. aureus in the pilot study was 11% (5/45). Sample size Sample size calculation was done based on the pilot study and a previous study done in the same geographical area, which reported a nasal carriage rate between 5-8% [11]. Thus, assuming 5% as the basic percentage of nasal carriage in children up-to five in general population, 11% as nasal carriage in children attending anganwaries, requesting a 95% confidence interval for the proportion with width no higher than 15% and power of 90% the minimum sample size needed is 191. A conservative estimate of design effect of 4 was considered appropriate [12]; this gave a minimum sample size needed of 764 (191 × 4) children. Study population The caregivers accompanying the children were verbally informed about the study one day prior to sample collection. The caregivers of the children who consented to participate were asked to accompany the child to anganwaries the next day. Ten children one to six years of age were randomly selected for nasal swabbing after obtaining written consent from their caregivers. Children having pyoderma (reported by caretakers) were not included. The caregiver accompanying the child was interviewed and a questionnaire filled-in. The questionnaire contained children demographic characteristics, information on any acute illness in the past two weeks like acute watery diarrhoea (defined as passage of 3 or more