Datasets:

timaeus

/

pile-pubmed_central

like 0

Follow

Timaeus 9

Introduction {#s1} ============ Rsp5, the sole member of the Nedd4 E3 ligase family in *S. cerevisiae*, has been implicated in various steps in intracellular trafficking, including endocytosis of plasma membrane proteins [@pone.0075372-Horak1] such as the Fur4 uracil permease [@pone.0075372-Galan1] and the Mat α receptor Ste2 [@pone.0075372-Dunn1]. Apart from a key initial role in substrate ubiquitination immediately preceding target internalization, Rsp5 is involved in other steps in vesicle trafficking, such as sorting at the multivesicular body (MVB) [@pone.0075372-Morvan1], [@pone.0075372-Macdonald1], [@pone.0075372-Erpapazoglou1] that precedes cargo delivery and degradation at the vacuole, and more recently, in mediating Golgi to ER trafficking [@pone.0075372-Jarmoszewicz1]. Similar to other Nedd4 ligases, the Rsp5 protein contains specific domains which are important for function: an N-terminal C2 lipid binding domain, important for subcellular localization to endosomal membranes [@pone.0075372-Dunn2], three WW protein binding domains in the central region important for the recruitment of substrates and binding to accessory cofactors, and a [H]{.ul}omologous to the [E]{.ul}6-AP [C]{.ul}arboxyl [T]{.ul}erminus (HECT) catalytic domain on the C-terminal end of the protein, responsible for ligase activity [@pone.0075372-Ingham1]. Regulation of Rsp5 activity is known to occur by various mechanisms, including binding to various adaptor proteins through specific protein-protein interactions. For example, the Rsp5 cofactors Bul1 and Bul2 have been implicated in dual roles in the Rsp5-mediated sorting of the amino acid permease Gap1 by facilitating ubiquitination at the plasma membrane [@pone.0075372-Soetens1], and by dictating trafficking of newly synthesized Gap1 from the Golgi to the vacuole by specifying poly-ubiquitin chain length, presumably through modulation of Rsp5 activity [@pone.0075372-Helliwell1]. In addition, recognition of substrates by Rsp5 can occur either by direct binding through its WW domains, or indirectly through additional adaptors [@pone.0075372-Leon1], [@pone.0075372-Merhi1], [@pone.0075372-Novoselova1], [@pone.0075372-MacDonald1], [@pone.0075372-Hatakeyama1], [@pone.0075372-Hettema1], [@pone.0075372-Lin1]. An additional potential route of ligase regulation may occur through E3 ubiquitination, either through an intra-molecular (*cis*) or an inter-molecular (*trans*) mechanism. In almost all cases studied to date, E3 auto-ubiquitination results in the down-regulation of ligase activity through proteasome-associated degradation, as in the case of the mammalian ligase Nedd4-2 [@pone.0075372-Bruce1], responsible for sorting of transmembrane proteins in a manner analogous to yeast Rsp5. However, non-proteolytic ubiquitination has also been reported, such as for the mammalian Itch ligase [@pone.0075372-Gallagher1] involved in immune signaling [@pone.0075372-Melino1], [@pone.0075372-Scialpi1]. Unlike most other E3 ubiquitin ligases, the ubiquitination of Rsp5 has not been well characterized, and a catalytically inactive mutant is comparably stable as wildtype Rsp5 [@pone.0075372-Wang1]. While Rsp5 has been reported in large scale studies to be ubiquitinated [@pone.0075372-Starita1], [@pone.0075372-Hitchcock1], [@pone.0075372-Peng1], [@pone.0075372-Ziv1] and to become auto-ubiquitinated [@pone.0075372-Huibregtse1], [@pone.0075372-Lee1], [@pone.0075372-French1], the physiological significance of this modification has not been demonstrated. E3 ligases can also be regulated through the intra-molecular folding of the enzyme. This has been reported for several mammalian cases, including Itch and Nedd4-2 [@pone.0075372-Bruce1]. In the case of Nedd4-2, ligase activity is catalytically inhibited in the absence of substrate by physical associations between its different domains [@pone.0075372-Bruce1]; the WW domains on Nedd4-2 binds to a PY motif located within the HECT catalytic domain of the protein. Whether this occurs through a *cis* or *trans* (i.e. intra- vs inter-molecular) mechanism is not yet known. Nevertheless, this binding inactivates the enzyme. Presumably, unfolding and release occurs upon the recognition and interaction with a substrate, resulting in 'on demand' re-activation [@pone.0075372-Bruce1], leading to the rapid ubiquitination and down-regulation of both the substrate and the ligase itself by 26S proteasome degradation. Association with a deubiquitinating enzyme (DUB) is another increasingly recognized mode of E3 regulation. Many E3 ligases, for example the mammalian RING ligase Mdm2 [@pone.0075372-Li1], as well as Itch [@pone.0075372-Mouchantaf1], are normally complexed with DUBs *in vivo*. In general, deubiquitinating enzymes have been reported to cleave off ubiquitin from auto-ubiquitinated ligases, resulting in protein stabilization. Previous studies in yeast by our group and others [@pone.0075372-Kee1], [@pone.0075372-Lam1] have shown that Rsp5 interacts both physically and functionally with the ubiquitin specific protease Ubp2. Ubp2 was reported to be important for proper trafficking at the MVB of Rsp5 membrane protein substrates, including the uracil permease Fur4 [@pone.0075372-Lam1]. The interaction of Ubp2 and Rsp5 is stabilized by the co-factor Rup1, which binds to both Ubp2 and Rsp5 but has an unclear physiological role [@pone.0075372-Kee1], [@pone.0075372-Lam1]. Although the exact role of Ubp2 and Rup1, is uncertain, they have been implicated as positive effectors of Rsp5-dependent ubiquitination [@pone.0075372-Lam1], [@pone.0075372-Ren1]. Given the accumulating evidence from studies showing auto-ubiquitination as an important mode of E3 regulation in mammalian systems, we examined if Rsp5 is auto-ubiquitinated under physiological circumstances in yeast and whether Ubp2 modulates this enzyme state. Here, we show for the first time that a significant fraction of Rsp5 is extensively poly- and auto-ubiquitinated *in vivo*, but this form of Rsp5 is both resistant to proteolysis and normally rapidly reversed by Ubp2. This effect is direct since Ubp2 is able to efficiently deubiquitinate recombinant auto-ubiquitinated Rsp5 *in vitro.* Moreover, similar to its mammalian homologues such as Nedd4-2, binding to a substrate or the Rup1 cofactor markedly stimulates Rsp5 auto-ubiquitination activity *in vitro*, implicating cofactor binding and intra-molecular unfolding in the Rsp5 activation cycle. These data indicate that Ubp2 and Rup1 enable a previously unappreciated regulatory feedback mechanism in Rsp5, and suggest that a tightly-coordinated pattern of auto-ubiquitination and deubiquitination forms a universal mechanism in the regulation of E3 ligase activity regardless of effects on enzyme stability. Results {#s2} ======= Rsp5 is Ubiquitinated and Deubiquitinated by Ubp2 *in vivo* {#s2a} ----------------------------------------------------------- As previous studies [@pone.0075372-Kee1], [@pone.0075372-Lam1] had shown that Ubp2 interacts both physically and functionally with Rsp5, we examined a possible role of Ubp2 in modulating the ubiquitination status of Rsp5. We hypothesized that Ub-Rsp5 conjugates are transient and hence might only accumulate in DUB deficient cells. To assess this, we looked for evidence of ubiquitination by Western blotting of N-terminally HA-tagged Rsp5, expressed under control of its endogenous promoter [@pone.0075372-Kaminska1], in soluble whole cell extracts prepared from both wildtype and an otherwise isogenic *ubp2* null mutant haploid yeast strains. The cell-free lysates were prepared in the presence of DUB and general protease inhibitors, and HA-Rsp5 was immunoprecipitated in native conditions prior to SDS-PAGE to enhance detection sensitivity. Whereas only a single dominant molecular weight species was apparent in WT cell derived extracts ([Fig. 1A](#pone-0075372-g001){ref-type="fig"}, centre and bottom panels), anti-ubiquitin immunoblotting revealed the presence of lower-mobility Rsp5 protein forms corresponding in size to mono and poly/multi-ubiquitinated in extracts isolated from *ubp2*Δ cells ([Fig. 1A](#pone-0075372-g001){ref-type="fig"}, top panel). These lower-mobility forms are not simply ubiquitinated proteins binding to Rsp5, as they were also detected with anti-HA antibody ([Fig. 1A](#pone-0075372-g001){ref-type="fig"}, bottom panel). Although only a small fraction of Rsp5 was ubiquitinated, these results nevertheless suggest that Rsp5 is indeed normally ubiquitinated transiently *in vivo* and that Ubp2 serves to deubiquitinate Rsp5. {#pone-0075372-g001} The Catalytic Activity of Ubp2 is Required for the Deubiquitination of Rsp5 {#s2b} --------------------------------------------------------------------------- One interpretation for the above data is that Ubp2 may be responsible for the removal of Ub chains from Rsp5. To test this, we examined whether the catalytic function of Ubp2 was required for this effect by monitoring the ubiquitination status of Rsp5 in a strain bearing a cysteine-to-serine point mutation at residue 745 of the genomic copy of Ubp2, a critical residue of the core conserved Cys box motif essential for catalytic activity [@pone.0075372-Baker1]. Indeed, mobility shifted Ub-Rsp5 species accumulated in the *ubp2-C745S* mutant strain to a level equivalent to that of an *ubp2* deletion strain ([Fig. 1B](#pone-0075372-g001){ref-type="fig"}), confirming that Ubp2 enzyme activity is required for the removal of Ub from Rsp5. Rsp5 is Auto-ubiquitinated {#s2c} -------------------------- Although the mammalian homologues of yeast E3 ligases have generally been reported to be auto-ubiquitinated [@pone.0075372-Scialpi1], [@pone.0075372-Fang1], examples of cross-ubiquitination by other E3 ligases have also been documented [@pone.0075372-Itahana1]. Hence, to determine if the ubiquitin modification on Rsp5 is the result of auto-ubiquitination or of the activity of another (unknown) E3 ligase, we examined the effects of inactivation of Rsp5 catalytic activity on the formation of ubiquitination species. To this end, we transformed a plasmid bearing an HA-tagged version of the conditional hypomorphic mutant *rsp5-1* allele (HA-*rsp5-1*) into yeast mutants deleted for *UBP2*. HA-tagged Rsp5-1 protein was then immunoprecipitated using anti-HA antibody and its ubiquitination status examined by Western blotting. Strikingly, cells expressing the Rsp5-1 mutant variant showed a drastic reduction in the level of mobility-shifted Ub-Rsp5 species compared with cells expressing wildtype HA-tagged Rsp5 ([Fig. 1C](#pone-0075372-g001){ref-type="fig"}). This result implies that, like most mammalian E3 ligases examined, Rsp5 is auto-ubiquitinated, possibly via an intra-molecular (*cis*) reaction. We reasoned that the low level of ubiquitination still detectable likely stems from residual *rsp5-1* enzyme activity. However, since Rsp5 has the ability to interact with itself (i.e. dimerize) [@pone.0075372-Dunn1], it could also reflect modification in *trans* via dimerization with endogenous Rsp5 present in these same cells, or possibly by an alternate ligase. To address this, we expressed and purified HA-Rsp5-1 from a yeast mutant strain wherein the native *RSP5* locus had likewise been converted to a conditional allele (*ubp2*Δ *rsp5-1*). In this context, ubiquitinated Rsp5 was virtually undetectable when the double mutant was grown at a non-permissive temperature, which was confirmed by the lack of cross-reactivity with anti-ubiquitin antibody, unlike the previous two lanes ([Fig. 1C](#pone-0075372-g001){ref-type="fig"}). Therefore, the minor residual ubiquitination seen in *ubp2*Δ cells is most readily explained as an inter-molecular (i.e. *trans*) ubiquitination reaction by endogenous (i.e. wildtype) Rsp5. Taken together, these data collectively indicate that the Rsp5 ubiquitinated forms detected *in vivo* are likely not due to the activity of another E3, and that Ubp2 inactivation is essential to reveal the innate intra-molecular (*cis*) ubiquitination process occurring on Rsp5. Substrate Recognition Stimulates Rsp5 Auto-ubiquitination {#s2d} --------------------------------------------------------- In the case of the mammalian HECT E3 Nedd4-2, the ligase is thought to remain in a folded, catalytically inactive conformation in the absence of a substrate, presumably through intra- or inter-molecular binding between the WW domain(s) and a PY motif in the HECT domain [@pone.0075372-Bruce1]. However, once a substrate binds to Nedd4-2, this inhibitory interaction is proposed to become disrupted, causing a change in the enzyme leading to an active conformation [@pone.0075372-Bruce1]. To explore whether this mechanism is likewise conserved in Rsp5, we performed an *in vitro* auto-ubiquitination reaction both in the absence and presence of substrate. For the reaction, we combined recombinant full-length Rsp5, E1 and E2 enzymes purified from *E. coli*, and ubiquitin, with or without a model substrate, the carboxy-terminal domain of RNA polymerase II (CTD) also purified from *E. coli* [@pone.0075372-Gupta1] ([Fig. 2](#pone-0075372-g002){ref-type="fig"}). RNAPII is ubiquitinated by Rsp5 *in vivo* in response to transcriptional arrest [@pone.0075372-Beaudenon1], [@pone.0075372-Somesh1], and the unphosphorylated form has been reported to be an excellent substrate *in vitro* [@pone.0075372-Kus1]. Ubiquitination was initiated by the addition of ATP and allowed to proceed at room temperature for various periods of time to monitor progression. The ubiquitination of both Rsp5 (GST-Rsp5) and the substrate (GST-CTD) were monitored by Western blotting using antibodies directed against the GST tag and Rsp5 itself. {#pone-0075372-g002} As expected, ubiquitinated CTD appeared upon addition of ATP, with higher molecular weight species indicating multi or poly-ubiquitination increasing proportional to reaction time ([Fig. 2A](#pone-0075372-g002){ref-type="fig"}, top panel). Auto-ubiquitination of Rsp5 also increased over time, with the non-modified form correspondingly decreasing in abundance. Strikingly, the substrate containing reactions ([Fig. 2A](#pone-0075372-g002){ref-type="fig"}, bottom panel) resulted in a markedly enhanced amount of shifted GST-Rsp5, corresponding to both mono-ubiquitinated and more extensively auto-ubiquitinated forms, as compared to the lanes in which no CTD was added ([Fig. 2B](#pone-0075372-g002){ref-type="fig"}, bottom panel). By 30 minutes, there is a notable accumulation of low-mobility forms, likely corresponding to Rsp5 conjugated to multiple ubiquitins representing either multi-ubiquitination or long poly-ubiquitin chains. In contrast, in the absence of substrate, only less extensively modified Rsp5 species were detectable, potentially corresponding to mono and short ubiquitin chains. Together, these results suggest that the presence of a substrate increases Rsp5 ubiquitination activity, consistent with a possible conformation change analogous to that reported for mammalian Nedd4-2. The Ubp2 Cofactor Rup1 Stimulates Rsp5 Auto-ubiquitination {#s2e} ---------------------------------------------------------- Since we and others have shown that Rup1 binds to Rsp5 directly [@pone.0075372-Kee1], [@pone.0075372-Lam1], we next examined if Rup1 exerts an analogous influence on the auto-ubiquitination of Rsp5 *in vitro*. To this end, Rup1 was affinity-purified from a TAP-tagged yeast strain lacking a functional copy of *UBP2* to remove the possible confounding effects of deubiquitination. Even in the absence of substrate, we found that addition of a sub-stoichiometric amount of Rup1-TAP to the Rsp5 *in vitro* ubiquitination mixture described above resulted in a significant increase in the rate of accumulation of band intensities corresponding to mono and multi/poly-Ub-Rsp5 species as compared to reactions in which Rup-TAP was omitted ([Fig. 3](#pone-0075372-g003){ref-type="fig"}). The effect of Rup1 on Rsp5 auto-ubiquitination was qualitatively similar, albeit less pronounced, to that seen following addition of a molar excess of CTD substrate. {#pone-0075372-g003} Taken together, these data show that, analogous to its mammalian homologue Nedd4-2 [@pone.0075372-Bruce1], Rsp5 auto-ubiquitination is stimulated both by the presence of substrates or cofactors. Hence, upon binding, Rsp5 may undergo an induced conformational change, causing an increase in its catalytic activity that leads in turn to both enhanced autocatalysis and substrate-directed ubiquitination. While future studies are needed to confirm this model, these results suggest that Rup1 exerts a regulatory effect independent of simply facilitating the recruitment or tethering of Ubp2 to Rsp5. Ubp2 Deubiquitinates Auto-ubiquitinated Rsp5 *in Vitro* {#s2f} ------------------------------------------------------- Our *in vivo* data ([Fig. 1](#pone-0075372-g001){ref-type="fig"}) indicate that auto-ubiquitinated Rsp5 is stabilized in the absence of Ubp2. This suggests that Ubp2 deubiquitinates Rsp5, thereby minimizing the accumulation of auto-ubiquitinated species. To test this directly, we examined the ability of purified TAP-tagged Ubp2 (Ubp2-TAP) isolated from an otherwise wild-type yeast strain (i.e. together in a complex with endogenous Rup1) to deubiquitinate Ub-GST-Rsp5 species *in vitro*. First, we assayed for deubiquitination activity *in vitro* by incubating various amounts of the purified Ubp2 preparation with K63-linked poly-ubiquitin chains, which Ubp2 has previously been reported to cleave [@pone.0075372-Kee2]. As predicted, Ubp2 was able to deubiquitinate poly-Ub chains down to mono-ubiquitin, and could be fully blocked by the addition of the DUB inhibitor ubiquitin aldehyde ([Fig. S1](#pone.0075372.s001){ref-type="supplementary-material"}). Next, we examined the ability of Ubp2 to deubiquitinate Ub-Rsp5 generated in an *in vitro* reaction in which substrate (CTD) was added to stimulate auto-ubiquitination (evident as an extensive smear by Western blot). Affinity-purified Ubp2-TAP, alone or with a DUB inhibitor (ubiquitin aldehyde) included as a control, or buffer alone, were then added, and the mixture incubated for an additional 2 hours. The ubiquitination status of Rsp5 was then monitored by Western blotting. As seen in [Figure 4](#pone-0075372-g004){ref-type="fig"}, the addition of Ubp2-TAP led to a dramatic reduction in the level of auto-ubiquitinated Rsp5, and a corresponding increase in both non-ubiquitinated and mono-ubiquitinated Rsp5 compared to control (i.e. buffer or inhibitor). These results establish Ubp2 as a direct regulator of Rsp5. While the Ubp2 preparation was, however, less effective at removing residual short ubiquitin chains and the terminal ubiquitin on Rsp5, our observations are in line with a previously reported preference of Ubp2 for processing longer ubiquitin chains [@pone.0075372-Kee2]. {#pone-0075372-g004} Discussion {#s3} ========== Detection of a Novel Ubiquitin Modification on Rsp5 {#s3a} --------------------------------------------------- It would appear that while most E3 ubiquitin ligases are auto-ubiquitinated, they also associate with deubiquitinating enzymes. Stably bound DUBs appear to reverse auto-ubiquitination [@pone.0075372-Li1], [@pone.0075372-Mouchantaf1], although they may also deubiquitinate E3 substrates as well [@pone.0075372-Li2], [@pone.0075372-Li3]. In this study, we show that, similar to its mammalian homologues, yeast Rsp5 is transiently auto-ubiquitinated. Although this modification has previously been detected, our work provides the first in-depth analysis of Rsp5 auto-ubiquitination *in vivo*, implying the modification may be biologically significant. Rsp5 auto-ubiquitination may have been difficult to detect previously in a physiological context because it is rapidly reversed by Ubp2. We found that a catalytically inactive Ubp2 point mutant, *ubp2C745S,* accumulates Ub-Rsp5 comparable to a full *ubp2* deletion. Moreover, affinity purified Ubp2 is able to deubiquitinate auto-ubiquitinated Rsp5 *in vitro*. Although a significant portion of Rsp5 remains mono-ubiquitinated, this is consistent with the previously identified catalytic activity of Ubp2 which has a preference for longer ubiquitin chains [@pone.0075372-Kee2]. Previous studies have shown differences in the function of a protein that was mono- versus poly-ubiquitinated (ex. RNA polymerase II ubiquitination [@pone.0075372-Harreman1] and PCNA ubiquitination as reviewed in [@pone.0075372-Bergink1]). Therefore, the removal of ubiquitin by Ubp2, even if not complete, still has the potential to modify Rsp5 function. In addition, it is also quite possible that a larger subset of cellular Rsp5 may be ubiquitinated *in vivo* that was not visible under the conditions of our experiments. For example, this could occur if ubiquitination is transient or if deubiquitinating enzymes other than Ubp2 (of which there are 15 other candidate Ubps in yeast) are likewise responsible for cleavage of ubiquitin from Rsp5 under certain circumstances. It is currently unknown whether Rsp5 is subject to long chain (poly-ubiquitination), multi-site ubiquitination, or if several short linked chains are attached. Since most blots of ubiquitinated Rsp5 show several discrete bands, it seems likely that poly- or multi-ubiquitination occur. It may be possible to determine the type of ubiquitin chain present on Rsp5, either by using chain type specific antibodies [@pone.0075372-Wang2], [@pone.0075372-Newton1], tandem mass spectrometry, or a panel of site--specific lysine mutant ubiquitin chains to prime the *in vitro* ubiquitination reaction. Relevant to this, Rsp5 and Ubp2 have been reported to catalyze the ubiquitination and deubiquitination [@pone.0075372-Kee2], respectively, of K63-linked chains, which suggests the presence of auto-ubiquitinated K63-linked chains on Rsp5. Rsp5 is Auto-ubiquitinated through an Intra-molecular Mechanism {#s3b} --------------------------------------------------------------- Another avenue to consider is the molecular mechanism by which the auto-ubiquitination reaction occurs. From previous studies on mammalian ubiquitin ligases [@pone.0075372-Scialpi1], [@pone.0075372-Mouchantaf1], [@pone.0075372-Fang1], [@pone.0075372-Luhtala1], the attachment of a ubiquitin chain to the ligase can either result from an intra-molecular reaction, in that the activated enzyme itself transfers ubiquitin to a residue on the same molecule, or as an inter-molecular reaction, wherein the ligase will ubiquitinate another enzyme molecule, such as another subunit of a dimer. From our results presented in this study, it seems that ubiquitination on Rsp5 happens preferentially as an intra-molecular reaction *in vivo*, although some low level of *trans* ubiquitination does occur as the remaining (a small percentage of the total) Ub-Rsp5-1 smear is eliminated when the genomic wildtype copy of Rsp5 is deleted. *Trans* ubiquitination of Rsp5 is, however, consistent with reports showing that Rsp5 has the ability to form dimers [@pone.0075372-Dunn1]. The disappearance of detectable Ub-Rsp5-1 species in the absence of functional endogenous Rsp5 also implies that no other ubiquitin ligases can stably ubiquitinate Rsp5 under standard yeast culture growth conditions, at least with a high enough efficiency to generate a band detectable by Western blot analysis. Binding to Substrates or Cofactors may Unfold and Activate Rsp5 {#s3c} --------------------------------------------------------------- Given that intra-molecular ubiquitin transfer seems to be the most plausible route by which Rsp5 becomes auto-ubiquitinated, we wanted to determine if an Rsp5 substrate, such as a recombinant form of the carboxy-terminal domain (CTD) of RNA polymerase II, influences the rate or degree of auto-ubiquitination of Rsp5. As evidenced by our *in vitro* data, the presence of either CTD or the cofactor Rup1 resulted in increased Rsp5 auto-ubiquitination over time, with the Ub-Rsp5 abundance elevated in terms of chain length and/or the number of attached ubiquitins. This effect can be compared to reactions without ectopic substrate, for which only mono-Ub or a short poly-Ub chain was detected. The increase in more extensively ubiquitinated forms, which is especially striking in reactions containing CTD, may indicate either increased processivity by Rsp5 or an increased preference for the formation of polyubiquitin chains under this condition. It is important to note that Rup1 has been reported to act as a cofactor for Ubp2, increasing the *in vitro* deubiquitinating activity of Ubp2 [@pone.0075372-Kee1]. It would be interesting to determine the overall impact of Rup1 on the Ubp2-Rup1-Rsp5 complex by examining the activity of Ubp2 on Rsp5 without Rup1 present. Since the addition of an Rsp5 substrate markedly increases Rsp5 auto-ubiquitination, it is tempting to speculate that, similar to the model proposed for its mammalian homologue Nedd4-2 [@pone.0075372-Bruce1], that Rsp5 is maintained in a "non-active" conformation or a poised folded state in the absence of suitable substrates or co-factors such as Rup1. This "inhibited" condition, as in the case with Nedd4-2, could be due to an intra- or inter-molecular interaction between one or more WW domains in the middle of Rsp5 with a sequence in the HECT domain. As previously reported [@pone.0075372-Gupta1], Rsp5 possesses a putative 'LPQY' motif in the HECT domain, similar to mammalian Nedd4 family ligases, which may serve as a candidate for binding to WW domain-containing cofactors. Upon recruitment of a binding partner or a substrate such as RNAPII CTD to the enzyme, Rsp5 may become unfolded as the WW domain shifts to bind the new factor, allowing the HECT-located PY motif on Rsp5 to become displaced such that the catalytic activity is engaged and the HECT domain is free to ubiquitinate both the substrate and Rsp5 itself. Evidence, from this study and others [@pone.0075372-Wang1], has established that proteasomal degradation is an unlikely fate for Ub-Rsp5, as catalytic mutants of Rsp5 are as stable as the wildtype forms. Given that ubiquitinated Rsp5 is seemingly not rapidly degraded by the proteasome, if poly-ubiquitination occurs it likely involves K63 chains consistent with the known catalytic activity of Rsp5. Recent studies, for example, on the ubiquitination of the mammalian Itch, has shown that ligase ubiquitination results in an as yet unknown non-degradative fate for the enzyme [@pone.0075372-Scialpi1]. The ubiquitination of Rsp5, then, may affect its function in a non-catalytic manner. For example, it could serve as a negative feedback mechanism, in which the presence of an Rsp5 substrate leads to the ubiquitination and inactivation of Rsp5 back to a resting ground state immediately after ubiquitination of the target. On the other hand, physical association of Rsp5 with Ubp2 would result in the deubiquitination and re-poising of Rsp5. Ubiquitination of Rsp5 may result in a decrease in enzymatic activity, possibly through a conformation change in the ligase. Auto-ubiquitination of Rsp5 may proceed until a suitable Ub chain length is reached for the inhibition of Rsp5 activity. Specifically, since the deletion of *UBP2* somewhat paradoxically causes a phenotype of Fur4 internalization similar to a substrate level ubiquitination defect [@pone.0075372-Lam1], the accumulation of ubiquitinated Rsp5 *in vivo* (albeit to substoichiometric levels) may result in a down-regulation of Rsp5 ligase activity at the MVB, leading to the protein trafficking defect observed in *ubp2* mutant cells. Along with the modification of Rsp5 activity, Ubp2, as recently reported [@pone.0075372-Erpapazoglou1], is also involved in the deubiquitination of ESCRT proteins. Ubp2-dependent deubiquitination was found to be required for proper MVB sorting [@pone.0075372-Erpapazoglou1]. The sum of these two roles of Ubp2 may result in the ultimate fate of an endocytosed protein at the MVB. Apart from a possible role in the direct modification of Rsp5 activity, it is also possible that auto-ubiquitination could result in the subcellular re-localization of Rsp5 protein to another cell compartment. Rsp5 has been reported to be present in various parts of the cell, most notably along the endocytic and vesicle trafficking pathways [@pone.0075372-Wang3]. Recent studies have implicated ubiquitination as an important signal specifying the intracellular localization of proteins [@pone.0075372-Marchenko1]. Therefore, it is possible that auto-ubiquitination of Rsp5 could represent one signal by which Rsp5 is re-directed to various compartments and new roles after ubiquitinating a substrate. "Activation" of Rsp5 by Ubp2 deubiquitination in this case then, may actually occur through a change in the localization of Rsp5, and therefore a change in possible Rsp5 substrates. Another possibility is that auto-ubiquitination prevents binding of Rsp5 to one or more of its co-factors, interacting proteins, or even certain substrates. In this case, auto-ubiquitination would lead to differential substrate recognition or result in a change in Rsp5 activity. This could be tested easily by monitoring the binding of various known co-factors and substrates of Rsp5 to both ubiquitinated and non-ubiquitinated forms of Rsp5. In this case, the ubiquitination of Rsp5, and the reverse, deubiquitination by Ubp2, could therefore lead to shuttling between pathways or processes due to differential protein binding. The exact ubiquitination site(s) on Rsp5 is currently unknown, although candidate lysines, specifically in the C2 and first WW domain, have been identified in at least one recent large scale study [@pone.0075372-Starita1]. Ubiquitination at these sites, therefore, may result in the translocation of Rsp5 from one cell compartment to another, or affect binding of Rsp5 to WW domain-binding cofactors. Future mapping experiments will be required to determine the effect the modification(s) have on Rsp5 folding, localization, and protein-protein interactions. Whichever case it may be, modification of Rsp5 by ubiquitin seemingly only affects a subset of the overall Rsp5 present in a cell. It is possible that only this subset is accessible to cofactors or substrates, resulting in Rsp5 activation and subsequent auto-ubiquitination, which, in turn, may lead to the fates discussed above. Taken together, the data presented in this study suggest that deubiquitinating enzymes in yeast, similar to well characterized DUBs in mammalian systems, have complex and important roles in regulating ubiquitination through direct modification of a ubiquitin ligase. Materials and Methods {#s4} ===================== Strains and Plasmids {#s4a} -------------------- A yeast expression plasmid encoding wild-type *RSP5* under control of its natural promoter pHA-Rsp5(p\[HA-*RSP5*, *CEN*, *LEU2*\]) [@pone.0075372-Kaminska1] and the catalytic mutant clone pHA-rsp5-1(pPC33) were obtained from Teresa Zoladek. Bacterial expression plasmids for generating recombinant proteins pET15b-UBC4, pET21a-GST-TEV-CTD, pGEX-6P2-RSP5 are described in [@pone.0075372-Gupta1]. A strain carrying a *UBP2* allele (*ubp2-C745S*) in the BY4741 strain background, encoding catalytically inactive Ubp2, was constructed in two steps: First, we disrupted the endogenous wildtype *UBP2* locus with a targeting cassette encoding *URA3* in a haploid uracil-auxotroph, creating an *ubp2*::*URA3* deletion allele. Then, sequences surrounding the *ubp2C745S* mutation in the yeast expression plasmid pF/H426-*ubp2*-*C745S* [@pone.0075372-Lam1] were PCR amplified and transformed into the *ubp2*::*URA3* strain. Integrants lacking a functional *URA3* marker were selected for on media containing 5-Fluoroorotic Acid (5-FOA). To eliminate possible secondary/unintended mutations, 5-FOA transformants were individually mated to haploid mat α *ubp2*Δ cells, diploids were selected, and haploid segregants scored after tetrad dissection. Clones carrying the *ubp2 C745S* allele showing a growth rate comparable to wildtype were identified by genomic amplification and sequencing of the modified *UBP2* locus. The TAP strains Rup1-TAP *ubp2*Δ and Ubp2-TAP are as described in [@pone.0075372-Lam1]. Cell Lysis, Immunoprecipitation, and Western Blots {#s4b} -------------------------------------------------- Yeast cells were cultured, harvested and lysed by bead beating essentially as previously described in [@pone.0075372-Lam1] in the presence of 10 mM of the DUB inhibitor N-ethylmaleimide (NEM; Sigma) and 0.1 nM of the proteasome inhibitor MG132 (Sigma). For anti-HA immunoprecipitations, approximately 6 mg of soluble cell-free protein lysate was first pre-cleared for 2 hrs at 4°C with protein-G agarose (Millipore), followed by immunoprecipitation with mouse monoclonal anti-HA antibody (generated from a cultured 12CA5 cell line; a kind gift from Mike Tyers, Samuel Lunenfeld Research Institute, Mount Sinai Hospital, Toronto, Ontario) and 20 µl of a 50% slurry of protein-G beads (Millipore) for 1hr. After extensive bead washing with lysis buffer, bound proteins were eluted using SDS sample loading buffer, subjected to PAGE, and transferred onto a nitrocellulose membrane as described in [@pone.0075372-Lam1]. Western blots were probed with either mouse monoclonal anti-GST antibody (B-14, Santa Cruz Biotechnology; 1∶2000 dilution), mouse monoclonal anti-ubiquitin antibody (clone 6C1; Sigma; 1∶5000 dilution), or rabbit anti-Rsp5 antiserum [@pone.0075372-Stamenova1]. Incubation with secondary antibody and detection of target proteins by ECL were then performed. Purification of Recombinant Ubc4, Rsp5, and CTD {#s4c} ----------------------------------------------- Inducible bacterial expression vectors bearing HIS-*UBC4*, GST-*CTD*, and GST-*RSP5* fusion cassettes were transformed into *E. coli* strain BL21(DE3) and recombinant proteins were expressed and affinity purified essentially as described in [@pone.0075372-Gupta1] with some modifications. Cells were grown in LB media at 37°C with vigorous shaking to an OD~590~ of 0.6, and 1 mM isopropyl-β-1-thio-D-galactopyranoside (IPTG) added to induce protein expression. Cells were incubated for a further 12 hrs at 16°C, then pelleted and lysed in sonication/binding buffer (20 mM HEPES, pH 8.0, 500 mM NaCl, 10% glycerol, and 0.5 mM tris(2-chloroethyl) phosphate (TCEP)) containing a protease inhibitor cocktail (Complete, Roche). For HIS-*UBC4*, 5 mM imidazole was added to the buffer to minimize non-specific binding during purification. After sonication, whole cell lysates were clarified by centrifugation at 80,000×g for 1 h at 4°C, and soluble protein incubated with either nickel-nitrilotriacetic acid agarose (Ni-NTA, Qiagen) to isolate the HIS-UBC4 fusion protein, or glutathione sepharose (Amersham) to recover the GST-tagged proteins. Beads were washed repeatedly with sonication/binding buffer (with 30 mM imidazole added to minimize non-specific background for the Ni-NTA purification). Bound HIS-*UBC4* was eluted with sonication/binding buffer containing 500 mM imidazole, while the GST-tagged proteins were eluted with sonication/binding buffer containing 15 mM glutathione. Protein eluates were concentrated to around 1 mg/ml by centrifugal filtration (MW cutoff 10 kDa; Microcon, Amicon-Millipore) as needed, and the purity and yield assessed by electrophoresis together with BSA control protein to estimate concentration on a 4--12% Bis-Tris polyacrylamide gel (Invitrogen) followed by Coomassie staining ([Fig. S2](#pone.0075372.s002){ref-type="supplementary-material"}). TAP Tag Purifications and Deubiquitinating Activity Assay {#s4d} --------------------------------------------------------- Tandem affinity purification of endogenous Ubp2-TAP and Rup1-TAP fusion proteins were performed on 4L yeast cultures essentially as described in [@pone.0075372-Krogan1], except the final elution buffer was DUB reaction buffer (10% glycerol, 50 mM Tris-Cl pH 7.9, 50 mM NaCl, 1 mM EDTA, 2 mM DTT, and 3 mM EGTA). Yield and purity of the purifications were estimated by SDS-PAGE followed by staining with silver. To confirm the deubiquitination activity of the Ubp2-TAP preparation, 1--30 µl of the TAP preparation (0.5 µg/ml) was incubated with 1.5 µl K63-linked ubiquitin chain (Boston Biochem) (2 mg/ml). The reaction was incubated for 1 hour at room temperature to allow for deubiquitination, and stopped by the addition of 3.6 µl of 100% trichloroacetic acid (TCA), followed by incubation on ice for 30 mins, and centrifugation to collect precipitated protein. Precipitates were resuspended in sample loading buffer, along with unbuffered Tris to neutralize the sample. The above was repeated but with 1--5 µl of ubiquitin aldehyde (1 mg/ml) to confirm that deubiquitination was Ubp dependent. *In vitro* Ubiquitination and Deubiquitination Assays {#s4e} ----------------------------------------------------- Ubiquitination assays in microtube format were done essentially as described in [@pone.0075372-Gupta1] with the following modifications. Each reaction (15 µl total volume) consisted of 5x assay buffer (250 mM HEPES, pH 7.4, 25 mM MgOAc, 2.5 mM TCEP, 500 mM NaCl, and 50% glycerol), 10 µg of purified yeast recombinant ubiquitin (Boston Biochem), 0.16 µg of recombinant commercial yeast E1 (Ube1; Boston Biochem), 3.8 µg of purified E2 (HIS-Ubc4), 1.2 µg of purified E3 (GST-Rsp5), 8 pmol of purified substrate GST-CTD, and 3.3 mM ATP (Sigma). In reactions supplemented with tandem affinity-purified yeast Rup1, 3.9 µl of a dilute Rup1-TAP preparation was added. Water was added to each reaction to bring the final volume of all reactions to 15 µl. Finally, ATP was added last to initiate enzyme activity, at time = 0, to all reactions rapidly in succession. Samples were incubated at room temperature and stopped at indicated times by boiling with SDS-PAGE sample buffer. For the deubiquitination experiments, samples were first subject to the ubiquitination conditions described above, except using 1 µg ubiquitin, and approximately 0.6 µg of a concentrated preparation of tandem-affinity purified yeast Ubp2-TAP, alone or together with 1 µg of the DUB inhibitor Ub-aldehyde (Affinity Research) as a negative control where indicated. Calmodulin elution buffer [@pone.0075372-Krogan1] was added to bring the final reaction volume up to a total of 30 µl. Reactions were then incubated for a further 2 hrs at room temperature to allow for deubiquitination to proceed, then stopped by boiling in sample loading buffer. The reaction products were analyzed by separation on a 4--12% SDS-PAGE gel and visualized by Western blotting using anti-GST antibody. Supporting Information {#s5} ====================== ###### The Ubp2-TAP preparation is catalytically active. (A) K63-linked ubiquitin chain was incubated *in vitro* with varying amounts of TAP purified Ubp2 to check for deubiquitination activity. The reaction was stopped and proteins precipitated by the addition of trichloroacetic acid (TCA), electrophoresed and visualized by staining with silver. The grey arrow bar indicates decreasing amounts of Ubp2 added, and the grey bar indicates an equivalent amount of Ub chain added in each reaction. Arrows point to the location of the various mono and polyubiquitin species. Mono ubiquitin (Ub) was located at the dye front. (B) To test for DUB specificity, the experiment was repeated, with an equivalent amount of Ubp2 and K63 chain in each reaction. Increasing amounts of ubiquitin aldehyde, a DUB inhibitor, was added to each reaction (grey arrow bar). The mobility of reference molecular weight markers is shown at the left. Mono and di-ubiquitin (Ub, Ub-Ub) were located at the dye front. (TIF) ###### Click here for additional data file. ###### Purification of recombinant Ubc4, Rsp5, and CTD. HIS-Ubc4, GST-Rsp5, and GST-CTD were purified as described in materials and methods. The indicated volumes of the purified proteins, and indicated amounts of bovine serum albumin (BSA) standards were electrophoresed on a SDS-PAGE gel, followed by Coomassie staining to visualize proteins. Protein concentrations of the purified preps were estimated by comparisons with the BSA standard. (TIF) ###### Click here for additional data file. We thank D. Rotin, T. Zoladek, and M. Tyers for strains, plasmids, and reagents. We also acknowledge A. Spence, T. Hughes, B. Lavoie, D. Goring, E. Conibear and members of the Emili Laboratory for constructive discussions. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: ML AE. Performed the experiments: ML. Analyzed the data: ML AE. Contributed reagents/materials/analysis tools: ML AE. Wrote the paper: ML AE. [^3]: Current address: Donnelly Centre for Cellular and Biomolecular Research, Department of Biochemistry and Department of Molecular Genetics, University of Toronto, Toronto, Ontario, Canada

Introduction ============ Cell fate is known to be regulated by signals from the microenvironment,[@cit1] not only through soluble signals but also through biophysical cues.[@cit2]--[@cit5] Biophysical cues, including topographical features, geometries and substrate stiffness, control upstream cytoskeleton assembly and downstream gene expression *via* a mechanotransduction pathway,[@cit6] which can ultimately modulate cellular behaviors, such as cell spreading, migration, proliferation and differentiation.[@cit7]--[@cit9] Recent literature suggests that substrate stiffness can affect cell adhesion and consequently cell growth by changing the cell shape and actin cytoskeleton, and meanwhile demonstrates key roles in cell signaling and differentiation.[@cit10],[@cit11] A traditional cell-culture dish, such as polystyrene and glass, is used for unphysiologically stiff materials, and cells cultured on these substrates tend to display aberrant behaviors: anomalous polarization, flattened shapes and loss of differentiated phenotypes.[@cit12] Thus, it\'s crucial for a desirable biomaterial to own distinct properties including biocompatibility and the ability to support appropriate cellular growth and function, especially tunable stiffness similar to native tissues. Hydrogels have provided a useful platform to reveal fundamental phenomena, regulate cell behavior and direct stem cell differentiation in ways not possible with conventional culture substrates.[@cit13] However, the fabrication of hydrogel-based substrates is an extremely complicated, expensive and skilled process. Besides, it may produce secondary effects of hydrogels, that is to say, changes in hydrogel stiffness could lead to alteration of the density of cell-adhesive ligands or the porosity of the underlying scaffold. A lower collagen anchoring density and larger anchoring distance was demonstrated to result in increased differentiation.[@cit14] But Engler\'s group claimed that the stiffness of planar matrices regulated stem cell differentiation independent of protein tethering and porosity by modulating substrate porosity without altering stiffness in polyacrylamide (PAAm) hydrogels.[@cit15] Thus, there is still a debate whether stiffness or the secondary effects control cell function. To address this issue, a graphene oxide (GO) sheet, which is a single layer of sp^2^ hybridized carbon atoms tightly packed into a two-dimensional (2D) honeycomb lattice with oxygen-containing hydrophilic groups, was taken into consideration.[@cit16] We hypothesize that GO can act as a biocompatible coating material to generate homogeneous topographical features in the surface of GO-coated PAAm hydrogels with different stiffnesses. This GO/PAAm composite scaffold will have the following advantages: (i) no secondary effects as GO covered the surface of PAAm hydrogels, (ii) the tunable stiffness ability, (iii) biocompatibility (GO has been widely regarded as a very low cytotoxicity material).[@cit17],[@cit18] Herein, we report a simple, practicable and cost-effective strategy to regulate cell behaviors and functions *via* a tunable stiffness enabled GO/PAAm composite scaffold. Benefited from the biocompatibility and the capability for blocking the secondary effects of PAAm hydrogels, the GO/PAAm composite structure can mimic the stiffness of native tissues to investigate the influence of substrate stiffness on cell behaviors and gene expression during cell growth. It\'s found that the tunable stiffness of the GO/PAAm composite scaffold could affect cytoskeleton assembly, morphology and the expression of cellular signal regulation and cytoskeleton-related genes. Thus, this GO/PAAm composite scaffold can be regarded as an alternative biomaterial to investigate the deep molecular mechanism for the influence of mechanical cues on cell growth in their physical microenvironments. Results and discussion ====================== Overview of GO/PAAm composite scaffolds --------------------------------------- [Scheme 1](#sch1){ref-type="fig"} illustrates the processes employed for the fabrication of the stiffness-tunable GO/PAAm composite scaffolds and the effect of substrate stiffness on cell growth. As illustrated in [Scheme 1A](#sch1){ref-type="fig"}, to make the stiffness-tunable GO/PAAm composite scaffolds a substrate for cell growth, piranha-cleaned glass was modified with amine groups by using 3-aminopropyltriethoxysilane (APTES), and then treated with 0.5% glutaraldehyde/PBS to generate a reactive surface. By varying the amounts and ratios of monomers acrylamide and cross-linker bis-acrylamide, the stiffness-tunable PAAm hydrogels were synthesized. Considering the 2D flexible structure and the biocompatibility of GO sheets, we demonstrate the use of GO sheets as an effective coating material in combination with the stiffness-tunable PAAm hydrogels for investigating the influence of substrate stiffness on cell growth. {#sch1} A substrate stiffness-dependent cellular behavior is presented in [Scheme 1B](#sch1){ref-type="fig"}. Briefly, the cells cultured on the soft matrix present a round shape, while the cells on the stiff matrix show a spindle shape and obvious stress fibers. [Scheme 1C](#sch1){ref-type="fig"} illustrates the hypothetical gene pathways associated with cytoskeleton assembly. The cells transfer the extracellular stiffness information into different intracellular gene pathways related to cytoskeletal rearrangement *via* mechanotransduction, and then these signaling pathways induce a specific gene expression change, form actin filaments and promote cytoskeleton assembly and cell growth. Fabrication of GO/PAAm composite scaffolds with tunable stiffness ----------------------------------------------------------------- To fabricate the stiffness-tunable GO/PAAm composite scaffolds, we integrated the PAAm hydrogels with GO sheets *via* a spin-coating method. As shown in [Fig. 1A](#fig1){ref-type="fig"}, flexible and large-size GO sheets were synthesized by a modified Hummers\' method.[@cit19],[@cit20] The GO sheets displayed a large lateral dimension of over 30 μm with a typical wrinkled topography (Fig. S1A--C[†](#fn1){ref-type="fn"}) and had a typical thickness (∼1.58 nm) of sub-bilayer GO sheets (Fig. S1D and E[†](#fn1){ref-type="fn"}). In contrast, the surface of the freeze-dried PAAm hydrogel presented a multiporous network structure with uniformly distributed pore sizes of almost 5 μm ([Fig. 1B](#fig1){ref-type="fig"}). It is precisely the gel porosity and its variance in the hydrogels of differing stiffness that result in the secondary effects. Fortunately, this issue was simply and effectively addressed *via* a GO-coated PAAm hydrogel composite structure, in which the flexible and large-size GO sheets covered the porous structure on the PAAm hydrogel surface, formed a homogeneous GO film with a mildly wrinkled topography and eliminated the secondary effects that resulted from the gel porosity ([Fig. 1C](#fig1){ref-type="fig"}). To obtain a suitable GO concentration for an effective coating, the GO dispersions in a range of 0.005 to 0.5 mg ml^--1^ were dispersed in water and ultrasonically treated for an hour before coating. From the SEM images of the GO/PAAm composite structures (Fig. S2[†](#fn1){ref-type="fn"}), we could see that as the concentration of GO dispersions increases, the multiporous surface of the PAAm hydrogel was gradually covered with GO sheets. Finally, we chose 0.1 mg ml^--1^ as an ideal concentration; in this case, GO sheets not only were effectively coated onto the PAAm hydrogels, but also were tightly bound to the surface of the PAAm hydrogel and hard to exfoliate. The stability of GO/PAAm composite structures benefits from the electrostatic interaction between PAAm hydrogels (positive) and GO sheets (negative).[@cit21] As shown in [Fig. 1D](#fig1){ref-type="fig"}, the Raman spectra of the GO/PAAm composite scaffolds (red) showed the existence of the two characteristic peaks of the D band and G band of GO sheets (black), and at the same time, presented the characteristic peaks of the PAAm hydrogel (blue) at 1112 cm^--1^ and 2930 cm^--1^, indicating the formation of GO/PAAm composite scaffolds.[@cit22] Furthermore, the as-prepared GO/PAAm composite structure was also confirmed by using FT-IR spectra (Fig. S3[†](#fn1){ref-type="fn"}). {#fig1} To simulate the stiffness of different human organs, 4 formulations of monomer-to-crosslinker (acrylamide/bis-acrylamide) were polymerized to yield PAAm hydrogels of ∼2, ∼7, ∼16 and ∼32 kPa, which correspond to the stiffness of the brain tissue, adipose tissue, muscles and osteoid, respectively.[@cit23]--[@cit25] The compressive modulus (namely the stiffness) is a key physical property for the GO/PAAm composite scaffolds, which contribute to cell adhesion and growth.[@cit26],[@cit27] The hydrogel stiffness was controlled by the ratio of acrylamide and bis-acrylamide and the details are supplied in Table S1.[†](#fn1){ref-type="fn"} The different hydrogels revealed significant differences in mass swelling ratios, stiffness and surface texture (or porosity). When coated with an optimum concentration of GO sheets, the composite scaffolds showed a typical topography similar to that of GO thin films (Fig. S4[†](#fn1){ref-type="fn"}) and the stress--strain curves of PAAm hydrogels and GO/PAAm composite scaffolds are also analogical (Fig. S5[†](#fn1){ref-type="fn"}). As shown in [Fig. 1E](#fig1){ref-type="fig"}, the compressive modulus of the composite scaffold for a monomer-to-crosslinker ratio of 10/0.3 was significantly higher than that of 3/0.1, and as the ratio of acrylamide/bis-acrylamide increases, the compressive modulus increased from 2 kPa to 32 kPa. Together, the GO/PAAm composite scaffolds not only could simulate the stiffness of native tissues, but also owned a uniform surface appearance to eliminate the intrinsic secondary effects in hydrogels. Therefore, the GO/PAAm composite scaffolds will provide a powerful platform for investigating cell behaviors related to the substrate stiffness. The influence of ECM stiffness on cytoskeleton assembly and cell morphology --------------------------------------------------------------------------- The biocompatibility of scaffold materials is essential for cell survival and function in engineered tissues.[@cit28] To evaluate cellular proliferation cultured on the GO/PAAm composite scaffolds, a cell counting kit (CCK) assay was employed to measure the cell viability during the culture period ([Fig. 2B](#fig2){ref-type="fig"}). For the first 12 h, cells cultured on the GO/PAAm composite scaffolds (red columns) appeared to support cellular growth with a higher cell viability compared to bare PAAm hydrogels (light grey columns), as in the case of the cell cultured for 24 h. By contrast, for 12 and 24 h, the cell viability increased, independent of the substrate stiffness in all groups including the GO/PAAm composite scaffolds and the bare PAAm hydrogels. Besides, the composite scaffolds showed no significant difference in cell viability from 2 kPa to 32 kPa for cells incubated for 12 or 24 h. The abnormal and wrinkled state for cells on bare PAAm hydrogels implied high cytotoxicity of the PAAm hydrogel (Fig. S6[†](#fn1){ref-type="fn"}). Compared to the traditional PAAm hydrogel coated with collagen, cells cultured on the GO/PAAm composite scaffold presented a similar cell morphology and spreading area, indicating the excellent biocompatibility of the composite structure (Fig. S7[†](#fn1){ref-type="fn"}). Thus, the GO/PAAm composite scaffolds displayed better biocompatibility than bare PAAm hydrogels and the stiffness of the substrate had no significant influence on the cell viability. {#fig2} To investigate the influence of ECM stiffness on cytoskeleton assembly and cell morphology, HeLa cells were cultured on GO/PAAm composite scaffolds with varied stiffness. When incubated on the scaffolds for 24 h, the cells were fixed and stained with Alexa 488-conjugated phalloidin and DAPI, to reveal the actin filament network (green) and the nuclei (blue), respectively. The representative fluorescence images of HeLa cells are shown in [Fig. 2A](#fig2){ref-type="fig"}; obvious stress fibers (filamentous actin bundles) were seen in cells grown on the stiffer matrix (32 kPa), but not in cells grown on the softer matrix (2 kPa, 7 kPa or 16 kPa). Stress fibers, which play an important role in cellular cytoskeleton assembly, can provide force for cells to sense and transmit the signal of the ECM stiffness.[@cit29] As shown in Fig. S8,[†](#fn1){ref-type="fn"} it is confirmed that cells grown on stiffer GO/PAAm composite scaffolds (optical image) presented stress fibers (fluorescence images), and for the softer scaffolds, the cells showed a smaller shape and no obvious stress fibers were seen. Next, to further quantify the relationship between cell morphology and matrix stiffness, the cell spreading area was taken into statistical analysis. As presented in [Fig. 2C](#fig2){ref-type="fig"}, the cell spreading area demonstrated normal distribution characteristics, and as the matrix stiffness increases, the shift of distributions suggested increase of cell spreading area, and the distribution was getting wider. Typically, the cells on the soft substrate (2 kPa) presented a round shape and smaller spreading area (400 μm^2^), while the cells cultured on the stiff substrate (32 kPa) showed a spindle shape and larger spreading area (2000 μm^2^) ([Fig. 2C](#fig2){ref-type="fig"}). Therefore, the morphological analysis suggests that the stiffness of GO/PAAm composite scaffolds would significantly affect the cytoskeleton assembly, shape and spreading. The role of the PAAm hydrogel in the GO/PAAm composite structure ---------------------------------------------------------------- To further confirm that the PAAm hydrogel plays an important role in the GO/PAAm composite structure for guiding cell behaviors, we investigated the effect of cell growth in the case of only GO films. [Fig. 3A](#fig3){ref-type="fig"} shows the typical topographical features of GO films with concentrations in the 0.05--1.0 mg ml^--1^ range. The surface root-mean-square roughness (*R*~q~) of the GO films were 4.46 nm, 7.64 nm, 7.90 nm and 11.1 nm, respectively. For the GO film fabricated using 0.05 mg ml^--1^ GO solution, we observed a very thin, continuous and uniform GO film with an appearance of micrometer-long wrinkles. With the GO concentration increasing, we observed an increased disorder and roughness, and long, broad wrinkles were observed across the film surface. As shown in [Fig. 3B](#fig3){ref-type="fig"}, cells cultured on the different GO films showed no significant difference in cellular morphology. Besides, we found that it displays no remarkable difference of stress fibers and cell spreading area for cells under different GO films ([Fig. 3C](#fig3){ref-type="fig"}), though the roughness of these GO film surfaces showed differences. That is to say, there was no significant effect on cell morphology, despite some mild changes in the topographical features of the substrate surface.[@cit30],[@cit31] However, for the substrate stiffness, the case is completely different. The stiffness of substrates not only regulated the cellular morphology, but also affected cytoskeleton assembly (the formation of stress fibers). Therefore, the tunable stiffness of the PAAm hydrogel plays a central role in GO/PAAm composite scaffolds, providing the ability to regulate cell behaviors. Also GO sheets are very important, and enhance the biocompatibility and address the issue of secondary effects due to the different surface structures of the PAAm hydrogel. {#fig3} The role of gene expression programs in cytoskeleton assembly ------------------------------------------------------------- It is well known that gene expression programs and cell signaling pathways play important roles in regulating cell functions.[@cit32] For instance, Steven and Andre suggested that the effect of the matrix stiffness on cell directed migration depended on the balance of the ECM-triggered signaling pathways PI(3)K and ROCK.[@cit33] To investigate the relationship between the substrate stiffness and cell specific gene expressions, RT-qPCR was employed to analyze the expression of cytoskeleton-related genes (ACTB, PFN1 and CFL1) and cell signaling-related genes, such as the ROCK pathway (RhoA and ROCK) and PI3K pathway (PI3K, FAK and Rac1). These genes regulated or participated in cytoskeleton assembly, which is an important process for the formation of stress fibers and cell spreading.[@cit34]--[@cit37] As shown in [Fig. 4A](#fig4){ref-type="fig"}, the expression level of cytoskeleton-related genes (ACTB, PFN1 and CFL1) on the soft matrix presented a significant reduction compared to the stiff matrix, indicating a lower efficiency or depletion in number to produce stress fibers, because these genes participated in cytoskeleton assembly. Therefore, cells grown on the stiff matrix presented obvious stress fibers, but not for cells on the soft one (consistent with the experiment above). On the other hand, the expression level of RhoA and cell signaling-related genes decreased and Rac1 increased distinctly on the soft matrix compared to the stiff matrix, indicating that matrix stiffness could regulate gene expression programs, and besides RhoA and Rac1 may play different roles in cytoskeleton assembly and cell morphology. {#fig4} To further investigate how the matrix stiffness affects cytoskeleton assembly and cell morphology *via* regulating specific gene expression programs, HeLa cells were cultured on stiff and soft matrixes, and the cells on the stiff matrix were treated with a panel of pathway and protein inhibitors, including Rho-associated kinase (Y-27632) and PI3K (LY-294002) (RT-qPCR data are shown in Fig. S9[†](#fn1){ref-type="fn"}). As shown in [Fig. 4B](#fig4){ref-type="fig"}, the fold change in gene expression for the soft group is consistent with the group added PI3K inhibitor, and almost opposite with the group added ROCK inhibitor except for Rac1. This phenomenon may suggest that the PI3K pathway for cells on the soft matrix was restrained through matrix stiffness-related gene regulatory processes. To more rigorously verify the findings, cell morphology under different stimulations was taken into consideration ([Fig. 4B](#fig4){ref-type="fig"}), and the cell spreading area was significantly reduced in the presence of inhibitors of PI3K (LY-294002) and the cell presented a corrugated state as the stress fibers disappeared, which is consistent with the case for the cell on the soft matrix. In contrast, when inhibiting Rho-associated kinase with Y-27632, there was no distinct change in cell spreading area, but the cell also couldn\'t form stress fibers resulting in an unconsolidated cell morphology. Together, all of this confirms the role of the PI3K and the ROCK pathway in mediating the formation of stress fibers, and the cell morphology is also relative to the PI3K pathway. Mechanical cues from the microenvironment regulate cell behaviors during growth *via* gene-regulatory processes, in which specific gene expression programs are activated and induce cytoskeleton assembly, promoting the formation of stress fibers and cell spreading. The GO/PAAm composite scaffold with tunable stiffness has served as a powerful platform for investigating the underlying mechanisms of how substrate stiffness regulates cytoskeleton assembly under physical microenvironments. Conclusions =========== In summary, we have demonstrated the capability of a unique GO/PAAm composite scaffold to provide instructive physical cues that regulate specific gene expression programs and cell behaviors, such as cytoskeleton assembly and cell morphology. For the composite scaffold, the PAAm hydrogel plays an exceptional role in controlling the substrate stiffness; in contrast, GO sheets not only provide permissive surfaces for cell adhesion and growth, but also effectively reduce the secondary effects utilizing their uniform surface topographical features. It\'s found that cytoskeleton assembly and cell morphology can efficiently be regulated by the substrate stiffness. And cytoskeleton-related genes (ACTB, PFN1 and CFL1) participated in cytoskeleton assembly, and the PI3K and ROCK pathways play important roles in mediating the formation of stress fibers. We envision that these findings can help in understanding the molecular mechanism for the influence of ECM mechanical cues on cell growth and metastasis at a deeper level. Furthermore, such a GO/PAAm composite scaffold can serve as a powerful platform for developing future therapies for ECM defect-related diseases and injuries. Experimental section ==================== Preparation of the GO/PAAm hydrogel composite scaffold ------------------------------------------------------ Glass coverslips were cleaned with Piranha solution (H~2~SO~4~ : H~2~O~2~ = 3 : 1), and then modified using 3-(trimethoxysilyl)propyl methacrylate (APTES) and glutaraldehyde to facilitate covalent attachment of hydrogel substrates to the amino-silanated coverslips. The preparation of PAAm hydrogels was adapted from a previously described protocol with some minor modifications. Briefly, after the diluted suspension was bubbled with nitrogen gas for at least 15 min to remove oxygen, designated amounts of acrylamide monomers, cross-linker bis-acrylamide, tetramethylethylenediamine (TEMED) and ammonium persulphate were prepared in the PAAm hydrogel with varied stiffness. The ratio of acrylamide and bis-acrylamide and the final concentrations were varied to control the mechanical properties and porosity of the hydrogel (details in ESI Table S1[†](#fn1){ref-type="fn"}). The gel solution was sandwiched between the functionalized coverslip and a chloro-silanated glass slide to ensure easy detachment of hydrogels. GO was dispersed in deionized water at varying concentrations (0.005, 0.01, 0.05, 0.1, 0.5 and 1.0 mg ml^--1^). The substrates (cover glass or the PAAm hydrogel with different stiffnesses) were dipped with the GO dispersions directly on top of the substrate for 30 min, followed by spin-coating with 800 rpm for 30 seconds. Cell culture ------------ HeLa cells were cultured in standard Dulbecco\'s modified Eagle\'s medium (DMEM) with 10% fetal bovine serum, 1% penicillin/streptomycin at 37 °C, 5% CO~2~ and 95% air humidity. The GO/PAAm hydrogel substrates with varied stiffness were immersed in PBS and placed in the cell culture hood for 30 min under UV light for sterilization before cell seeding. For cell seeding, HeLa cells were seeded at a proper density on the substrates modified with GO, PAAm hydrogels or the GO/PAAm composite scaffold, so that they had enough space to spread and didn\'t contact other cells. Biocompatibility test --------------------- The biocompatibility of the various substrates with varied stiffness (the PAAm hydrogels and GO/PAAm composite scaffold substrate) was tested by examining the cell viability of HeLa cells using a cell counting kit-8 (CCK-8).[@cit38] HeLa cells were incubated on each substrate for 12 and 24 h in 96-well plates. Following incubation, 100 μL of the cell culture medium (containing 10 μL CCK-8 solution) was added to each well and incubated with cells for 120 min at 37 °C. The data are represented as the absorbance at 450 nm, considering the cells cultured in a well with CCK-8 as the control (*A*~c~), the cells cultured on the substrates with CCK-8 as the experiment group (*A*~s~), and the well with only CCK-8 added as the blank (*A*~b~). Finally, the cell viability is calculated as (*A*~s~ -- *A*~b~)/(*A*~c~ -- *A*~b~). Cell staining and image analysis -------------------------------- After incubation for 24 h and brief washing with sterilized PBS, HeLa cells were fixed in 4% (w/v) paraformaldehyde in phosphate buffered saline (PBS) for 15 min at room temperature (25 °C), permeabilized for 5 min with 0.5% v/v Triton-X100 in PBS at room temperature, and then blocked with 1% BSA for 1 h for actin filament staining. Actin staining was performed using FITC conjugated to phalloidin. After post-stain washing with PBS, the cells were mounted in 4′,6-diamidino-2-phenylindole (DAPI) for nuclear staining. For measurements of the cell-spreading area, the 2D images of phalloidin/DAPI-stained cells were taken using a Leica TCS SP5 inverted confocal microscope (Leica, Germany) with a 63× oil-immersion objective. Only those cells that did not exhibit any cell--cell contacts were considered in the analysis. The boundaries of all single cells were then outlined manually on the basis of the actin stain, and the cell spreading area was determined using Image J software. Real-time quantitative PCR (RT-qPCR) analysis --------------------------------------------- HeLa cells were harvested after counting of cells and the total RNA was extracted by using the TransZol reagent following the manufacturer\'s instructions. Total RNA samples were investigated on a NanoDrop spectrometer (ND_200, NanoDrop Technologies, USA). RT-qPCR analysis of mRNA was performed with SYBR Select Master Mix according to the manufacturer\'s instructions on a Bio-Rad C1000TM (Bio-Rad, USA). The primers used for RT-qPCR are listed in Table S2.[†](#fn1){ref-type="fn"} Briefly, for this PCR, the 20 μL reaction solution contained 2 μL of cDNA sample, 10 μL 2× SYBR Select Master Mix, 2 μL forward primer (5 μM), 2 μL reverse primer (5 μM), and 4 μL RNase-free water. The total PCR volume was 20 μL and the PCR was carried out for 2 min at 95 °C, followed by cycling 45 × (95 °C for 15 s and 60 °C for 1 min), and finished with 60 °C for 5 min. The experiment was repeated three times. The relative expression level of target mRNAs was evaluated by referring to the expression of GAPDH mRNA using the 2^--ΔΔ*C*~t~^ method.[@cit39] Inhibition experiments ---------------------- For drug inhibition experiments, drug inhibitors were added at 24 h of culture, and sequentially incubated for another 24 h for cell staining and the RT-qPCR experiment. The concentration of the inhibitor used was 20 μM for LY294002 (PI3K inhibitor) and 10 μM for Y27632 (ROCK inhibitor). Conflicts of interest ===================== There are no conflicts to declare. Supplementary Material ====================== Supplementary information ###### Click here for additional data file. This work was financially supported by the National Natural Science Foundation of China (No. 21621003, No. 21235004, No. 21327806), and Tsinghua University Initiative Scientific Research Program. [^1]: †Electronic supplementary information (ESI) available: Experimental procedures, characterization of GO sheets and RT-qPCR data. See DOI: [10.1039/c8sc02100g](10.1039/c8sc02100g)

All relevant data are within the paper files. Introduction {#sec001} ============ Osteoporosis is a global public health problem currently affecting nearly 50 million of individuals in the industrialized countries and representing a major public health burden nowadays and for the foreseeable future \[[@pone.0172100.ref001]\]. It is characterized by an imbalance in bone remodelling process that leads to progressive loss of bone mass and bone microarchitecture, thus increasing the fracture risk \[[@pone.0172100.ref002]\]. Osteoporotic fractures have serious direct consequences on the patient quality of life (e.g. decreased functional mobility) and indirect consequences on the whole society (e.g. increase request of professional home-care services). The resulting socio-economic cost is estimated at about 40 billion of euro per year in the European Union \[[@pone.0172100.ref003], [@pone.0172100.ref004]\]. Currently, together with the lifestyle recommendations for reducing osteoporotic fracture risk, the strategies to treat or limit the progress of the osteoporosis include the surgical management of osteoporotic fractures by minimally invasive techniques with injectable methyl methacrylate cements \[[@pone.0172100.ref005]\], pharmacological therapies with bisphosphonates, hormonal therapy, antagonists of the Wnt signaling pathways and anti-resorptive drugs and molecules acting on calcium-sensing receptors \[[@pone.0172100.ref006]--[@pone.0172100.ref008]\]. However, all these strategies confine the problem without resolving it. For example, the oral administration of bisphosphonates has several severe side effects and the current injectable biomaterials addressed to bone healing are still unable to exhibit adequate bioactivity and osteoconductivity, associated with the adequate mechanical performance to withstand the early biomechanical loads \[[@pone.0172100.ref009]--[@pone.0172100.ref012]\]. In this respect, the development of injectable biomaterials provided with bioactivity, bioresorbability as well as the ability to be implanted by minimal invasive surgery and self-harden *in vivo*, stabilizing even complex-shape bone fractures is highly demanded. Nowadays, a growing interest on the synthesis and development of calcium phosphate bone cements (CPCs) is demonstrated in literature, especially due to their excellent bioactivity deriving from the chemical similarity to the inorganic part of bone \[[@pone.0172100.ref013]--[@pone.0172100.ref015]\]. Among the various approaches reported for the synthesis of CPCs, the one based on the hydrolysis and transformation of α-Ca~3~(PO~4~)~2~ (αTCP) into nanostructured calcium-deficient hydroxyapatite (CD-HA) particles is particularly valued \[[@pone.0172100.ref009]\]. Moreover, the possibility to endow CD-HA structure of bioactive ions increased the possible application of this bone cement. In particular, the CD-HA doping with strontium ions is particularly effective against osteoporotic bone weakening \[[@pone.0172100.ref016], [@pone.0172100.ref017]\]. It was found that Sr^2+^ enhances the proliferation and differentiation of osteoprogenitor cells into bone-forming osteoblasts, with mechanisms probably involving membrane-bound calcium sensing receptor (CaSR) and Wnt/β-catenin signalling \[[@pone.0172100.ref018], [@pone.0172100.ref019]\]. Moreover, Sr^2+^ decreases osteoclastogenesis and the resorbing activity of mature osteoclasts \[[@pone.0172100.ref020], [@pone.0172100.ref021]\]. In the present work, Sr-doped hydroxyapatite (HA) bone cements were prepared by mixing Sr-substituted αTCP phases with setting solutions enriched with sodium alginate, as already described and characterized in our previous work \[[@pone.0172100.ref022]\], and for the first time were deeper biological evaluates *in vitro*, to investigate its biological effect on mouse mesenchymal stem cells (MSCs), pre-osteoblasts (OBs) and osteoclasts (OCLs). Materials and methods {#sec002} ===================== Sr-BCs synthesis {#sec003} ---------------- Sr-doped HA cements were prepared as reported in our previous work \[[@pone.0172100.ref022]\]. Briefly, Sr-substituted αTCP powders with different strontium content (i.e. Sr/(Ca+Sr) = 0, 2, 5 mol%, henceforth coded as bone cement (BC), Sr2%-BC and Sr5%-BC respectively), were synthesized by solid state reaction at 1400°C for 1 hour of mixtures of calcium carbonate (CaCO~3~, Carlo Erba, Italy), calcium hydrogen phosphate (CaHPO~4~, Sigma Aldrich) and strontium carbonate (SrCO~3~, Carlo Erba, Italy). After rapid quenching, the powders were milled by planetary mono mill (Pulverisette 6 classic line, Fritsch, Germany) at 400 rpm for 50 minutes with 5 mm diameter zirconia balls. Aqueous solutions containing 5 wt% of Na~2~HPO~4~ (Fluka) and 2 wt% of sodium alginate (Alginic Acid Sodium Salt from Brown Algae, Sigma Aldrich) were mixed with the powders by using a liquid-to-powder ratio equals to 0.48 ml/g. Solid and liquid components were treated with γ irradiation (25 kGy) and autoclaving (121°C for 20 minutes), respectively, and then mixed together by using a commercial Bone Cement Delivery System equipment (P-system, Medmix, Switzerland). Cylindrical cement specimens for *in vitro* tests (diameter = 10 mm; height = 2 mm) were obtained after 72 hours of dry incubation at 37°C and washing in ethanol 70% for 20 min followed by three washes in 1x PBS for 10 min each. The discs were then dried and sterilized by UV irradiation for 30 min per side under laminar flow hood and finally preconditioned for 72 hours in standard cell culture medium. Ion concentration measurements {#sec004} ------------------------------ The elemental composition of the cement precursors as well as the release of calcium and strontium ions in cell culture medium were analysed by Optical Emission Spectrometry (ICP-OES, Liberty 200, Varian, Clayton South, Australia). In particular, 20 mg of precursor powders were dissolved into 2 ml of nitric acid (HNO~3~, Sigma Aldrich), followed by dilution to 100 ml with bi-distilled water. Then, the release of calcium and strontium ions was evaluated by submerging disk-shaped samples (diameter = 10 mm; height = 2 mm) in Dulbecco\'s modified Eagle\'s medium (DMEM/F12 GlutaMAX (Gibco, Carlsbad, CA) supplemented with 10% Fetal Bovine Serum and 1% Penicillin-Streptomycin (PAA, Pasching, Austria) and aged at 37°C and 5% CO~2~. At each time point (after 1, 2, 3, 7 and 14 days) of incubation the supernatants were completely removed from the sample and collected for ion quantification (in mM): the supernatants were diluted in 0.5 ml HNO~3~ and 8.5 ml bidistilled water. Three specimens for each precursor powder and cement formulation were tested. The amount of Sr^2+^ ions (in weight) released at each time point was also referred to the correspondent amount in the precursor powder (in weight) ([Fig 1C](#pone.0172100.g001){ref-type="fig"}). {#pone.0172100.g001} Cell culture {#sec005} ------------ Mouse mesenchymal stem cells (C57BL/6) (MSCs), purchased from Invitrogen (Carlsbad, CA), were cultured in DMEM/F12 GlutaMAX (Gibco, Carlsbad, CA) supplemented with 10% Fetal Bovine Serum (FBS) and 1% Penicillin-Streptomycin (100 U/ml-100 μg/ml); PAA, Pasching, Austria). For the experiments, cells were plated at 1.5×10^4^/cement and cultured for up to 14 days in αMEM GlutaMAX (Gibco) supplemented with 10% FBS and 1% Penicillin-Streptomycin and osteogenic factors (10 mM β-glycerophaspate, 10^−7^ M dexamethasone, 50 μg/ml ascorbic acid). Mouse pre-osteoblast cell line MC3T3-E1 Subclone 14 (OBs), obtained from ATCC cell bank (Manassas, VA, USA), was used as a model of osteoblasts \[[@pone.0172100.ref023]\]. 1.5×10^4^/cement of OBs were cultured in αMEM containing ribonucleosides, deoxyribonucleosides (GIBCO), and L-glutamine, 10% FBS and 1% Penicillin-Streptomycin and osteogenic factors (10 mM β-glycerophaspate, 50 μg/ml ascorbic acid). The RAW 264.7 cell line (ATCC) (OCLs), an established mouse monocyte cell line used as a model of osteoclast formation *in vitro* \[[@pone.0172100.ref024]\], was cultured in DMEM/F12 GlutaMAX supplemented with 10% FBS and 1% Penicillin-Streptomycin. To initiate the cell differentiation, 25 ng/mL soluble murine Receptor Activator for Nuclear Factor kB Ligand (RANKL, Sigma-Aldrich, Dorset, UK) were added. For the experiments, cells were plated at 3.0×10^4^/cement and cultured for up to 14 days. Sr-BCs and BC discs, sterilized and preconditioned as above derived, were placed one per well in a 24-well plate and a drop of 50 μl, containing cell suspension, was seeded on the centre of the upper discs surface allowing cell attachment for 30 min in the incubator, before adding 1.5 ml of cell culture medium into each well. The cell medium was changed every 3 days. All cell-handling procedures were performed in a sterile laminar flow hood. All cell culture incubation steps were performed at 37°C with 5% CO~2~ and controlled humidity. Cell viability and proliferation {#sec006} -------------------------------- The viability and proliferation of the MSCs and OBs were measured by evaluating metabolically active cells. The MTT reagent (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide) is reduced to formazan dye in metabolically active cells. The formazan production can be observed at λ~max~ of 570 nm, using a Multiskan FC Microplate Photometer (Thermo Scientific), and the absorbance is directly proportional to the number of metabolically active cells. The reagent was prepared at 5 mg/mL in 1x PBS. Cell were incubated with the MTT reagent 1:10 for 2 h at 37°C. Medium was collected and cells incubated with dimethyl sulfoxide for 15 min. In this assay, the mean values of absorbance were determined. Tree biologically independent samples were analysed at day 1, 2, 3, 7, and 14 and the percentage of viability with respect to that of cells cultured on BCs, used as control group, was shown. Cell morphology evaluation {#sec007} -------------------------- ### Actin filaments staining {#sec008} After 3 days of culture, samples were washed in 1x PBS for 5 min and fixed in 4% (w/v) paraformaldehyde for 15 min. Permeabilization was performed with 1x PBS with 0.1% (v/v) Triton X-100 for 5 min. FITC conjugated phalloidin (Invitrogen) 38 nM in 1x PBS was added for 20 min at room temperature in the dark. For nuclear staining the cells were incubated with DAPI 300 nM (Invitrogen) for 5 min. Images were acquired by an Inverted Ti-E fluorescence microscope (Nikon). The analysis was performed for all the cements tested, but only one representative image for each cell type grown on BC samples has been reported. ### Scanning Electron Microscopy (SEM) analysis {#sec009} After 3 days of culture, Sr-BCs and BCs samples were washed with 0.1 M sodium cacodylate buffer pH 7.4 and fixed in 2.5% glutaraldehyde in 0.1 M sodium cacodylate buffer pH 7.4 for 2 h at 4°C, washed in 0.1 M sodium cacodylate buffer pH 7.4 and dehydrated in a graded series of ethanol (30%, 50%, 70%, 90% and 100% for 10 min/each). Dehydrated samples were sputter-coated with gold and observed using Stereoscan 360 SEM (Cambridge Instruments, UK). The analysis was performed for all the cements tested, but one representative image for each cell type grown on BC samples has been reported. Quantitative real-time polymerase chain reaction (qPCR) {#sec010} ------------------------------------------------------- After 7 and 14 days, cells grown on the cements were lysed and total RNA extraction was performed by use of the Tri Reagent, followed by the Direct-zol^™^ RNA MiniPrep kit (Euroclone, Milano, Italy) according to manufacturer\'s instructions. RNA integrity was analysed by native agarose gel electrophoresis and quantification performed by the Qubit^®^ 2.0 Fluorometer together with the Qubit^®^ RNA BR assay kit, following manufacturer\'s instructions (Invitrogen). Total RNA (500 ng) was reverse transcribed to cDNA using the High-Capacity cDNA Reverse Transcription Kit, according to manufacturer\'s instructions (Life Technologies, Carlsbad, CA). Quantification of the gene expression was performed by use the StepOne^™^ Real-Time PCR System (Applied Biosystems, Foster City, CA, USA). The target genes evaluated for MSCs were: runt-related transcription factor 2 (Runx2, Mm 00501580), alkaline phosphatase (ALP, Mm00475834); target genes for MC3T3-E1 evaluation were: Osterix (Mm 04209856), osteocalcin (Bglap, Mm 00649782) and integrin-binding sialoprotein (IBSP, Mm 00492555); for the quantification of the osteoclast activity the target genes evaluated were: osteoclast-associated immunoglobulin-like receptor (OSCAR, Mm00558665), integrin beta-3 (Itgβ3, Mm00443980) and cathepsin K (CtsK, Mm00484039). The housekeeping gene used for all the assays was glyceraldehyde 3-phosphate dehydrogenase (GAPDH, Mm99999915) (Life Technologies). Three biologically independent samples were analysed. BC's error bars reflect one standard deviation from the mean of 3 technical replicates as described \[[@pone.0172100.ref025], [@pone.0172100.ref026]\]. Data were collected using the StepOne Software (v.2.2.2) and relative quantification was performed using the comparative threshold (Ct) method (ΔΔCt), where relative gene expression level equals 2^-ΔΔCt^ \[[@pone.0172100.ref027]\]; BC sample was used as calibrator. Statistical analysis {#sec011} -------------------- Results were expressed as mean ± standard error of the mean (SEM) plotted on graph. Analysis of the effect of the cements on cell culture and of the ion release in culture medium was made by two-way analysis of variance (ANOVA), followed by Bonferroni's post hoc test for biological results and by Sidak\'s multiple comparisons test for chemical tests. Statistical analyses were performed by the GraphPad Prism software (version 6.0). Results {#sec012} ======= Strontium and calcium release {#sec013} ----------------------------- Upon immersion of the cement samples, a slight acidification of the culture medium was observed, as a colour change towards yellow occurred, especially for BC formulation. In this respect, if compared with the pH of physiological fluids \[[@pone.0172100.ref028]\], after 7 days, a ΔpH = -0.4, -0.36 and -0.13 was detected for BC, Sr2%-BC and Sr5%-BC, respectively. The concentrations of Ca^2+^ and Sr^2+^ ions, net of the amount of ions leached out during preconditioning, are shown in [Fig 1A](#pone.0172100.g001){ref-type="fig"}. Significant calcium depletion over time was registered from day 3 to day 14 (p\<0.05) ([Fig 1A](#pone.0172100.g001){ref-type="fig"}), without difference among the cements. Conversely, the Sr^2+^ release steadily was increased for all the samples up to 14 days; Sr5%-BC exhibited a higher strontium release, if compared with Sr2%-BC, with a decreased release rate after 1 day. After 14 days Sr5%-BC exhibited a Sr^2+^ concentration of 0.63 mM, two times higher than for Sr2%-BC (0.34 mM) ([Fig 1B](#pone.0172100.g001){ref-type="fig"}). Moreover, the overall amount of strontium released up to 14 days was very limited (in respect to the amount of ion detected in the precursor powders) and without plateaus ([Fig 1C](#pone.0172100.g001){ref-type="fig"}, 0.68 wt% for Sr2%-BC and 0.48 wt% for Sr5%-BC), suggesting a potential capability of the cement to further ion release. Biological effect of Sr-BCs on MSCs {#sec014} ----------------------------------- The results showed that the presence of the Sr^2+^ increased the bioactivity of this cement. Concerning the MSCs viability and proliferation, measured by evaluating metabolically active cells (MTT assay), no significant difference existed between the effect of Sr2%-BC and Sr5%-BC and no significant differences existed also in cell viability over the time of culture ([Fig 2](#pone.0172100.g002){ref-type="fig"}). However, it was possible to observe a weak decreasing at day 3 and day 7. That variability can be attributed to the imbalance of ions induces by the presence of the cements in a static culture condition that, however, do not compromise the MSCs viability. {#pone.0172100.g002} To evaluate the inductive effects on the gene profile related to the osteogenic differentiation, the expression of Runx2 and ALP, considered the principal markers of osteoblast commitment, has been quantified. MSCs, grown for 7 days on Sr2%-BC, displayed a significant increase in mRNA level of both the genes compared to the Sr5%-BC (Runx2 p≤0.05 and ALP p≤0.01) ([Fig 3](#pone.0172100.g003){ref-type="fig"}). {#pone.0172100.g003} The cell morphological analysis ([Fig 4A](#pone.0172100.g004){ref-type="fig"}) showed that MSCs were well spread and appeared to be homogenously distributed on the surface of all the tested cements, without any remarkable differences among them. After 3 days the cement surface of all the samples was nearly completely covered by the MSCs that exhibited their characteristic morphology and showed cytoplasmic extensions connecting them to each other and to the biomaterial surface, indicating the good cytocompatibility of the proposed cements ([Fig 4D](#pone.0172100.g004){ref-type="fig"}). Since there are no differences in cell morphology between the three different cements, only one representative image for each cell type was reported in [Fig 4](#pone.0172100.g004){ref-type="fig"}. {#pone.0172100.g004} Biological effect of Sr-BCs on OBs {#sec015} ---------------------------------- OBs viability, assessed by MTT analysis, showed an increase over the experimental time. Sr5%-BC induced a significant higher level of proliferation compared to Sr2%-BC, after 3 days, 7 days (p≤0.01) and 14 days (p≤0.0001) of culture. Moreover significant differences were observed between BC and both Sr substituted BC at all the time points (p≤0.0001) ([Fig 5](#pone.0172100.g005){ref-type="fig"}). Instead, the analysis of gene expression profile showed no significant difference directly ascribable to Sr^2+^ in the induction of osteoblast related genes (i.e. Osterix, BGlap and IBSP) among all the BCs tested ([Fig 6](#pone.0172100.g006){ref-type="fig"}). {#pone.0172100.g005} {#pone.0172100.g006} The study of cell morphology, through the cytoskeletal actin filaments analysis and the SEM evaluation, showed OBs exhibiting the typical morphology and strictly interacting with the BCs surface without any differences among all the tested cements. In [Fig 4B and 4E](#pone.0172100.g004){ref-type="fig"} it is possible to observe stretched cellular pseudopodia anchoring the cement surface or connecting the contiguous cells. Biological effect of Sr-BCs on OCLs {#sec016} ----------------------------------- Gene expression analysis ([Fig 7](#pone.0172100.g007){ref-type="fig"}) showed a significant decrease occurring over the time of Oscar and CtsK mRNA level, genes involved in the mechanism of OCLs differentiation and bioactivity. The Itgβ3 mRNA level showed a difference between Sr2%-BCs and Sr5%-BC after 7 days (p≤0.05), that became negligible after 14 days of culture. {#pone.0172100.g007} Actin staining showed several undifferentiated monocytes and few mature OCLs with the typical podosome belts, an actin-rich structure, which appeared as prominent rings encasing multiple nuclei. SEM images showed the details of one mature OCL grown on BC surface exhibiting the basoapically polarized morphology and the osteoclast/mineralized substratum interface typically of the resorption process ([Fig 4C and 4F](#pone.0172100.g004){ref-type="fig"}). Discussion {#sec017} ========== Biomaterials play a crucial role in personalized medicine due to their unique features in tuning cell behaviour. The injectable biomaterials represent an excellent strategy to improve the interaction of the material itself with the surrounding tissues in osteoporotic patients \[[@pone.0172100.ref029]--[@pone.0172100.ref031]\]. In the present study, it has been demonstrated that novel injectable strontium-doped apatitic bone cements \[[@pone.0172100.ref022]\], influenced the viability, morphology and gene expression profiling of bone cells. A detailed morphological analysis showed MSCs and OBs well attached and spread on all the cement surfaces with their characteristic morphology. It was possible to observe stretched cellular pseudopodia anchoring the cement surface or connecting the contiguous cells as index of high level of cell-cell and cell/material interaction. The qualitative OCLs morphological analysis confirmed the presence of few mature osteoclasts on all the BCs surface, surrounded by undifferentiated monocytes. This was considered as a marker of OCLs integration and a good indicator of the grade of the mimicry of the cement apatite structure \[[@pone.0172100.ref032], [@pone.0172100.ref033]\]. These findings suggested that the complex phenomena occurring during the cell/material interaction could be directly related to the biomimetic apatitic structure of the bone cements. Although the presence of Sr^2+^ seemed to not induce modification in cell morphology, Sr^2+^ conferred to the BCs biological activity able to modulate the bone cell behaviour due to its already demonstrated anabolic activity on MSCs and OBs, and inhibitory effect on OCLs \[[@pone.0172100.ref019], [@pone.0172100.ref034]--[@pone.0172100.ref036]\]. Currently, one major treatment of osteoporosis is the systemic administration of strontium ranelate. However, the use of this drug has been associated with adverse side effects such as venous thrombosis, cutaneous hypersensitivity, chronic renal failure, and other, and its bioavailability is very low by oral administration (\<20%) \[[@pone.0172100.ref037], [@pone.0172100.ref038]\]. Consequently, sustained local release of Sr^2+^ ions from biomaterials could be preferable to the systemic administration. In this work, the effect of Sr^2+^ substitution on the degradation properties of Sr-BCs, which consist in the releasing of Sr^2+^ and Ca^2+^ ions, up to 14 days of immersion in complex media able to replicate the *in vivo* fluids was studied. Generally, the introduction of strontium in the apatite structure is associated to an increased solubility of the cements, lead to an increase of ions released, due to its lattice expansion \[[@pone.0172100.ref039], [@pone.0172100.ref040]\]. It has been already proved that modification of ions concentration (e.g. Calcium) in the cell culture medium can even greatly affect both cell proliferation and differentiation \[[@pone.0172100.ref041], [@pone.0172100.ref042]\]. It was observed that αTCP-based cements generally induce a slight acidification and a depletion of Ca^2+^ concentration in the immersion medium possibly due to the progressive formation of apatite crystals \[[@pone.0172100.ref040]\]. In our system, a Ca^2+^ depletion in the culture medium overtime was also observed, while the introduction of strontium seems to attenuate the pH fluctuation; conversely, the Sr^2+^ release increased overtime, possibly attenuating the reported inhibiting effect of Ca^2+^ depletion on cell behaviour \[[@pone.0172100.ref043]\]. It should be noted that, after 1 day, the strontium release of Sr2%-BC (0.125 mM) was really close to the reported circulating strontium concentrations in osteoporotic patients treated with strontium ranelate orally administrated at a dose of 2 g daily (0.12 mM) \[[@pone.0172100.ref044]--[@pone.0172100.ref046]\]. This result makes this cement a good candidate for the treatment of osteoporotic bone fractures and defects, as a local reservoir of Sr^2+^ ions, to avoid the side effects related to the high oral dose treatments administered to compensate the relatively low bioavailability of the drugs. *In vitro* results suggested that the Sr^2+^ concentration released from the Sr-BCs is able to induce specific cellular response depending on the bone cell types, with totally absence of cytotoxicity also at the highest concentration detected (0.628 mM). Confirming the already known biological role of the Sr^2+^ \[[@pone.0172100.ref047]\], here it has been shown an effect of the released Sr^2+^ on the induction of MSCs osteogenesis-related gene (Runx2 and ALP). Interestingly, the activity of 0.24 mM of Sr^2+^, released from Sr2%-BC after 7 days, was higher compared to the effect of 0.54 mM of Sr^2+^ released from Sr5%-BC. The presence of Sr^2+^ had a different behaviour on precursor cells committed to the bone lineage (OBs). In fact, Sr^2+^ evoked a dose-dependent inductive effect on the proliferative process of OBs, but it seemed to be not directly involved in the expression of the mature osteoblast related genes. Several *in vitro* studies demonstrated a wide range of different responses induced by Sr^2+^ in term of differentiation and proliferation, depending on the cell type, time and dose of Sr^2+^ administration \[[@pone.0172100.ref047]--[@pone.0172100.ref050]\]. It seems reasonable to hypothesize that the variability of biological effects observed could be explained by the differences in the cell models, pre-osteoblasts and mesenchymal stem cells, as Sr^2+^ exerted a defined role that probably depended on the different commitment stage of the cells. The *in vitro* study confirmed also that the Sr^2+^ released from Sr-BCs exerts its inhibitory effect on the expression of the principal genes involved in OCLs activity (Oscar and CtsK) \[[@pone.0172100.ref051]--[@pone.0172100.ref053]\]. Although Itgβ3 mRNA level seems to follow the same trend of decreasing over the time of culture, no statistically difference was observed, probably due to the proved role of integrin β3 in the interaction with apatitic structure \[[@pone.0172100.ref054]\]. In fact, Itgβ3 is considered the principal receptor in osteoclasts mechanisms involved in adhesion to the extracellular matrix, motility and activity. Therefore, the observed absence of Itgβ3 down-regulation mediated by Sr^2+^ is in accordance with the high level of OCLs interaction with the apatitic structure of the BCs, as demonstrated by the morphological analysis. Conclusion {#sec018} ========== In this work, strontium-doped apatitic bone cements doped with tailored amounts of strontium and enriched with sodium alginate were biological characterized. The three different cell types studied exhibited different behaviour, depending on the Sr^2+^ doping amount. In this respect, it was possible to conclude that the Sr^2+^ is able to promote: i) an inductive effect on MCS osteogenic gene expression, especially at 2 mol% concentration; ii) a dose-dependent inductive effect on OBs proliferation and iii) an inhibitory effect on OCLs activity. Thanks to the combination of the biomimetic features of the apatitic bone cement structure with the biological effect of Sr^2+^, such cements were able to evoke a specific cellular response, thus exhibiting promising features to assist new bone formation, particularly relevant in the case of bone turnover impairment due to metabolic diseases, such as osteoporosis. The research leading to these results has received funding from the European Union\'s Seventh Framework Programme (FP7/2007-2013) under grant agreement n° 246373, OPHIS. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: **Conceptualization:** MM SP.**Data curation:** MM MD.**Formal analysis:** MM MD.**Funding acquisition:** AT SS.**Investigation:** MM MD.**Methodology:** MM MD.**Supervision:** MM SP SS.**Validation:** MM SS AT.**Writing -- original draft:** MM MD.**Writing -- review & editing:** MM MD SP SS AT.

Introduction ============ Protein--protein interactions (PPIs) are critical to almost all biological process, and a good knowledge of the network of interacting proteins is crucial to understanding cellular mechanisms ([@ref-81]). Recent advances in biotechnology, such as high-throughput yeast two-hybrid screening, have allowed scientists to build maps of proteome-wide PPI, or interactome. Conventionally, a PPI map is a static network, in which each node represents a protein and an edge connecting two proteins indicates that there is experimental evidence showing that, under certain circumstances, the two proteins would interact. In reality, a PPI network (PPIN) should be viewed as a dynamic entity: it is an interaction network that is intrinsically controlled by regulatory mechanisms and changes with time and space ([@ref-59]), as determined by the physiological condition of the cell in which the proteins reside. If there is a PPIN that includes all possible PPIs, then, under a specific physiological condition only a specific sub-network of the PPIN is realized. MicroRNAs (miRNAs) are small (∼22 nucleotides) noncoding regulatory RNA molecules in plants, animals, and some viruses. In a process known as RNA interference, a miRNA regulates gene expression by destabilizing and/or disrupting the translation of fully or partially sequenced mRNA ([@ref-13]; [@ref-53]). In this way a miRNA regulates the formation of all PPINs to which its target is connected, and by extension all biological processes (BP) with which those PPINs are involved. As well as acting as a tumor suppressor gene (TSG), a miRNA may also act as an oncogene, say, by targeting a TSG ([@ref-104]). The function of a specific biological process, or its malfunction, such as associated with a disease, typically involves a complex composed of a set of miRNA-regulated proteins, together with their interacting protein partners. The study of such miRNA-protein complexes should be an integral part of understanding BP ([@ref-43]) as well as diseases. An understanding of the molecular and physio-pathological mechanisms of diseases is crucial for the design of disease preventive and therapeutic strategies. The combination of experimental and computational methods has led to the discovery of disease-related genes ([@ref-19]; [@ref-49]). An example is the causal relation connecting the malfunction causing mutations in the enzyme phenylalanine hydroxylase to the metabolic disorder Phenylketonuria ([@ref-85]). Many human diseases cannot be attributed to single-gene malfunctions but arise from complex interactions among multiple genetic variants ([@ref-41]). How a disease is caused and how it can be treated can be better studied on the basis of a body of knowledge including all associated genes and biological pathways involving those genes. Diseases are usually defined by a set of phenotypes that are associated with various pathological processes and their mutual interactions. Some relations between phenotypes of different diseases may be understood on the basis of common underlying molecular processes ([@ref-12]), such as when there are genes associated with both diseases. It has been shown that genes associated with the same disorder encode proteins that have a strong tendency to interact with each other ([@ref-31]). More specifically, one may consider two diseases to be related if their metabolic reactions within a cell share common enzymes ([@ref-54]). Networks of PPIs have also been studied in the context of disease interactions ([@ref-44]; [@ref-60]). Here, we report on a web service platform, miRNA disease regulatory network (miRDRN) (<http://mirdrn.ncu.edu.tw/mirdrn/>). The platform contains two parts, a database that stores a set of newly constructed set of 6,973,875 *p*-valued target-specific regulatory sub-pathways (RSPs) associated with 116 diseases in 78 tissue types built from 207 diseases-associated miRNA regulating 389 genes; and a novel web-based tool that, using the RSPs stored in miRDRN and information from miRNA-related databases, facilitates the construction and visualization of disease and tissue-specific miRNA-protein regulatory networks for user specified single diseases and, for comorbidity studies, disease-pairs. We demonstrate three applications of miRDRN: to explore the molecular and network properties of the single disease colorectal neoplasm; to study the comorbidity of the disease-pair Alzheimer's disease-Type 2 diabetes (AD-T2D); and, by using miRDRN to construct a miRNA regulatory sub-network centered on the gene *BACE1*, to look for insights that may explain why several anti-AD, *BACE1* inhibiting drugs that failed recent late-phase trials worsened conditions of treatment groups. We believe findings from miRDRN, even exploratory in nature, may potentially lead to the identification of new drug targets and new understanding in modes of drug action. Materials and Methods ===================== Data integration ---------------- miRDRM integrated data from several existing database on disease-miRNA association, miRNA-target gene association, gene ontology, biological pathway, and PPI. For disease-specific cases disease-associated miRNAs and targets were obtained from human microRNA and disease associations database (HMDD) ([@ref-58]) (v2.0, <http://www.cuilab.cn/hmdd>). For non-disease specific cases, miRNA/siRNA and targets were obtained from TarBase ([@ref-95]) (v7.0, <http://carolina.imis.athena-innovation.gr/diana_tools/web/index.php?r=tarbasev8%2Findex/>). In disease-specific cases, the optional filter requiring miRNA-target association be assay validated used TarBase data; the filter excludes miRNA-target pairs appearing in HMDD but not in TarBase (if and when this happens). Gene-tissue associations were taken from NCBI-Entrez ([@ref-66]); RSP associations with known pathways from the Kyoto Encyclopedia of Genes and Genomes (KEGG) ([@ref-47]) (<http://www.genome.jp/kegg>); gene-tumor associations from TAG ([@ref-24]) (<http://www.binfo.ncku.edu.tw/TAG>); OMIM IDs of target genes and genes on regulatory networks from NCBI-OMIM ([@ref-66]) (<https://www.ncbi.nlm.nih.gov/omim>); gene IDs and transcription factor and/or receptor from NCBI-GeneBank ([@ref-66]) (<https://www.ncbi.nlm.nih.gov/genbank>); data on human PPI from biological general repository for interaction datasets (BioGRID) ([@ref-23]) (<https://thebiogrid.org/>); and gene ontology, gene annotation and gene product information from GO ([@ref-8]) (<http://geneontology.org/>). ([Table 1](#table-1){ref-type="table"}). 10.7717/peerj.7309/table-1 ###### Databases used in the construction of miRDRN and data usage.  ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Database Information used Where used Reference --------------- ---------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------- ----------- HMDD 2.0 Disease-miRNA association,\ Query interface (disease selection) [@ref-58] miRNA-target association TarBase 7.0 miRNA-target association\ Query interface (miRNA/siRNA selection)\ [@ref-95] miRNA-target validation Query interface (filter---miRNA-target validation) NCBI-Entrez Gene-tissue association Query interface (filter---tissue) [@ref-66] KEGG Biological pathways Query interface (filter---pathway) [@ref-47] TAG Gene-tumor association Query interface (filter---tumor genes) [@ref-24] NCBI-GeneBank Gene ID, TF, and/or RC Query interface (filter---TF and/or RC) [@ref-66] NCBI-OMIM OMIM ID Target genes; genes on regulatory network (activated by mouse-click) [@ref-66] BioGRID Protein--protein interaction Construction of regulatory sub-pathways [@ref-23] GO Biological processes, molecular functions; gene ontology, gene annotation, and product Computation of Jaccard index for regulatory sub-pathways; annotation of genes on regulatory network (activated by mouse-click) [@ref-8] ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Construction of miRNA-associated target-specific regulatory sub-pathways ------------------------------------------------------------------------ Consider a linked sequence (*M*, *T*, *G*~1~, *G*~2~) ([Fig. 1](#fig-1){ref-type="fig"}), where *M* is a miRNA, *T* is its regulatory target gene, *G*~1~ is a gene whose encoded protein (*p*~1~) interacts (according to PPI data) with the protein (*p*~*T*~) encoded by *T*, and *G*~2~ is a gene whose encoded protein (*p*~2~) interacts with *p*~1~. In what follows, when there is little risk of misunderstanding, the same symbol will be used to represent a gene or the protein it encodes. We call the sequence (*T*, *G*~1~, *G*~2~) a target-specific RSP, or simply a RSP, and (*M*, *T*, *G*~1~, *G*~2~) a miRNA-specific RSP (MRSP). Given a target gene *T*, we use PPI data from BioGRID to collect all RSPs by extending from *T* two levels of interaction. {#fig-1} Jaccard score of a regulatory sub-pathway ----------------------------------------- Jaccard similarity coefficients ([@ref-68]) were used to score the RSPs, based on the assumption that there is a tendency for two directly interacting proteins to participate in the same set of BP or share the same set of molecular functions (MF). Given two sets *S*1 and *S*2 (in the current application, a set will be either a list of BP or a list of MF, both according to GO), the Jaccard (similarity) coefficient (JC) of *S*1 and *S*2 is defined as,$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$${{\rm{JC }}(S1,S2)} = \frac{| S1 \cap S2|}{| S1 \cup S2 |}$$\end{document}$$ Where ∪ is the union (of two sets), ∩ is the intersection, and \|*Z*\| is the cardinality of *Z*. JC, which ranges from zero to one, is a quantitative measure of the similarity between two sets. For example, when *S1* = {*a*, *b*, *c*} and *S2* = {*b*, *c*, *d*}, JC (*S1, S2*) = 2/4 = 0.5. Let (*T*, *G*~1~, *G*~2~) be an RSP as defined in the previous section and denote by \[*G*\] the set of BP (or pathways) ([@ref-47]; [@ref-8]) that involve the gene *G*. We define the Jaccard score, or JS, of RSP as,$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\[{\rm{J}}{{\rm{S}}_{\rm{X}}}(T,{G_1},{G_2}) = \frac{1}{2}\left( {{\rm{JC}}\left( {{{\left[ T \right]}_{\rm{X}}},{{\left[ {{G_1}} \right]}_{\rm{X}}}} \right) + {\rm{JC}}\left( {{{\left[ {{G_1}} \right]}_{\rm{X}}},{{\left[ {{G_2}} \right]}_{\rm{X}}}} \right)} \right)\]\end{document}$$ Where X may be BP or MF. If the pair \[*T*\] and \[*G*~1~\] do not share a common term, then the corresponding JC has a zero value; similarly for the pair \[*G*~1~\] and \[*G*~2~\]. In either case the RSP is considered to be not viable and discarded. In other words, miRDRN excludes any RSP with zero JC score. *p*-Value of a regulatory sub-pathway ------------------------------------- A *p*-value for an RSP (*T*, *G*~1~, *G*~2~) was assigned as follows. Let the total number of BP (or MF, as the case may be) terms be *N*, and the number of terms in \[*T*\], \[*G*~1~\], \[*G*~2~\], \[*T*\] ∉ \[*G*~1~\], \[*G*~1~\] ∉ \[*G*~2~\] be *x*, *y*, *z*, *n*~1~, and *n*~2~, respectively, then the *p*-values, *P*~1~ and *P*~2~, for (*T*, *G*~1~) and (*G*~1~, *G*~2~) are, respectively$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$${P_1} = \>{{C_{{n_1}}^NC_{x - {n_1}}^{N - {n_1}}C_{y - {n_1}}^{N - x}} \over {C_x^NC_y^N}}$$\end{document}$$and$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$${P_2} = \>{{C_{{n_2}}^NC_{y - {n_2}}^{N - {n_2}}C_{z - {n_2}}^{N - y}} \over {C_y^NC_z^N}}$$\end{document}$$ The *p*-value for the RSP was set to be the greater of *P*~1~ and *P*~2~. Assembly and storage of target-specific regulatory sub-pathways --------------------------------------------------------------- A union set of miRNA-associated target genes were collected from HMDD and TarBase and for every target a complete set of RSPs, with BP- and MF-type JC scores and *p*-values assigned, was assembled. The entire set of RSPs for all targets was stored in miRDRN. Construction of disease-specific miRNA regulatory network --------------------------------------------------------- A user-initiated construction of a disease-specific miRNA regulatory network (RRN) proceeds as follows. Step 1. Select a disease. Step 2. Collect from HMDD all miRNAs (*M*'s) and target genes associated with the disease. Step 3. For each *M* and each of its targets retrieve from miRDRN storage all target-specific RSPs, thus forming a set of MRSPs. The union of the sets of MRSPs over all *M*'s is the set of disease-specific MRSPs for the selected disease. Step 4. Construct the disease-specific RRN from the set of disease-specific MRSPs by linking all unlinked pairs of genes/proteins if they have interaction according to BioGRID ([Fig. 2](#fig-2){ref-type="fig"}). {#fig-2} Results ======= miRNA disease regulatory network (miRDRN)---A database and web service platform ------------------------------------------------------------------------------- We built miRDRN (<http://mirdrn.ncu.edu.tw/mirdrn/>), a web-based service that allows the user to construct a disease and tissue-specific, *p*-valued, miRNA-protein regulatory network, or RRN. The current version of miRDRN contains 6,973,875 *p*-valued target-specific RSPs constructed through 389 miRNA-regulated genes from 207 diseases-associated miRNAs associated with 116 diseases ([Table 2](#table-2){ref-type="table"}). 10.7717/peerj.7309/table-2 ###### Data contained in current version miRDRN.  Type of data Disease miRNA miRNA regulated gene Target-specific RSP -------------- --------- ------- ---------------------- --------------------- Number 116 207 389 6,973,875 Comparison of miRDRN with other miRNA-related databases ------------------------------------------------------- A number of databases and/or web service platforms on miRNA-related topics are publicly available ([Table 3](#table-3){ref-type="table"}). Aside from HMDD and TarBase on which miRDRN was built ([Table 1](#table-1){ref-type="table"}), PhenomiR ([@ref-82]) is a database on disease-miRNA association, miRwayDB ([@ref-28]) is a database on disease-miRNA-target and target-KEGG term association, and miRPathDB ([@ref-10]) is a database on miRNA-pathway association. New and unique as a database, miRDRN stores the 6,973,875 *p*-valued target-specific RSPs it has assembled ([Table 2](#table-2){ref-type="table"}). As a web service platform miRDRN is a tool that facilitates the construction and visualization of disease-specific RRNs using these RSPs in combination with resources from HMDD, TarBase, and several other databases ([Table 1](#table-1){ref-type="table"}). 10.7717/peerj.7309/table-3 ###### Comparison of miRDRN with other miRNA-related databases.  Information Database ------------------------------------------------------------------- ---------- ----- ----- ----- ----- ----- Disease-associated miRNA Yes Yes -- Yes Yes -- miRNA-associated target gene Yes Yes Yes -- Yes -- Target gene-KEGG association Yes -- -- -- Yes -- Pathway-associated miRNA Yes -- -- -- -- Yes Target-specific regulatory sub-pathway (RSP) Yes -- -- -- -- -- Disease-specific miRNA-protein regulatory network Yes -- -- -- -- -- Gene annotation (GeneBank, TAG, KEGG, GO) on all genes on network Yes -- -- -- -- -- Comorbidity of disease-pair Yes -- -- -- -- -- Brief description of usage of miRDRN ------------------------------------ miRNA disease regulatory network is reasonably user friendly; its many features are easily discovered by user exploration. Here, we give a brief description of its main features. User may use miRDRN to explore a single disease, or the comorbidity of a disease-pair. In the course of either type of study, all relevant miRNAs, genes, and RSPs are made accessible to the user in tabulated form, and RRNs in the form of interactive maps, both of which may be downloaded by the user. Often a map is too large for practical visualization, and in such a case the user may use options such as setting a *p*-value cut-off, or requiring a specific gene to be present in the map, or both, to obtain a partial RRN. The entrance interface of miRDRN (<http://mirdrn.ncu.edu.tw/mirdrn/>) asks the user to select "Single Search" to explore a single disease (or miRNA/siRNA) or "Comorbidity Search" to explore the comorbidity of a disease-pair ([Fig. 3](#fig-3){ref-type="fig"}). The user is then asked to specify the disease or disease-pair to be explored and tissue/tumor types, and *p*-value threshold for RSP evaluation, and to click on (or not) several optional filters, respectively, on targets and on RSPs. The filter on miRNA targets allows the user to admit only targets positively validated by the seven direct experimental methods: HITS-CLIP, PAR-CLIP, IMPACT-Seq, CLASH, Luciferase Reporter Assay, 3LIFE, and Genetic Testing ([@ref-95]); filters on RSP allow the user to select only those RSPs with some or all of the proteins to be cancer related ([Fig. 4](#fig-4){ref-type="fig"}). The user may then click on "Query" to start the computation. Tabulated results of disease-associated miRNAs and their target genes ([Fig. 5](#fig-5){ref-type="fig"}), a multi-page list of all RSPs ([Fig. 6](#fig-6){ref-type="fig"}) and, in the case of Comorbidity Search, a list of all comorbid genes ([Fig. 7](#fig-7){ref-type="fig"}) will then automatically appear. After the first, automatic iteration, the user may reduce the size of the RSP-list by using the "Gene filter" and "Show top ... sub-pathways" options ([Fig. 6](#fig-6){ref-type="fig"}). The next interface ([Fig. 8](#fig-8){ref-type="fig"}), in ready mode on first appearance, waits for the user to select one of three network layouts: "Tree," "Circle," or "Radial." After "Go" is clicked on, the platform displays an interactive map showing the RRN built from RSPs selected by user-specified options ([Fig. 8](#fig-8){ref-type="fig"}). When the mouse is placed on a node (a miRNA or a gene) on the map a small pop-up window opens to show the name of the node/gene and the number of other nodes it is linked to, and annotation on the node from GO, OMIM, KEGG, and GeneBank databases. {#fig-3} {#fig-4} {ref-type="fig"}, on miRNAs and literature source (blue area) and target genes (green). For each gene the gene symbol and its OMIM id are given, as well as information on whether the protein it encodes has a cancerous protein tag: CRG, cancer related gene; OCG, oncogene; TSG, tumor suppressor gene.](peerj-07-7309-g005){#fig-5} {#fig-6} {#fig-7} {ref-type="fig"}.\ The option "Show top-70" RSPs (by *p*-value) was used. When the mouse is placed on a node (in this case the gene IRS1) in the displayed network, a small pop-up window opens to show the name of the node/gene and the number of other nodes it is linked to, and annotation on the node from GO, OMIM, KEGG, and GeneBank databases.](peerj-07-7309-g008){#fig-8} Discussion ========== Here, we demonstrate the utility of miRDRN by presenting three applications. Case 1. A single disease study of colorectal neoplasm ----------------------------------------------------- Here, we demonstrate a single disease application of miRDRN. On the query interface select "colorectal neoplasms" (or colorectal cancer (CRC)), tissue type "colorectal tumor," pathway ranking by "Jaccard index (MF)," and *p*-value \< "0.001," and no option on target genes or RSP. The result yielded 33 miRNAs and 37 target genes (if the option "miRNA target experimental validation is positive and direct\" on the query interface was selected then there would be 20 miRNAs and 23 target genes ([Fig. 5](#fig-5){ref-type="fig"})), and 45,565 RSPs involving 3,079 genes (reduced to 2,111 RSPs and 1,650 genes when target is restricted to being "positive and direct" and cancer related ([Fig. 6](#fig-6){ref-type="fig"})) ([Table 4](#table-4){ref-type="table"}). 10.7717/peerj.7309/table-4 ###### Result for sample Single Search: disease, colorectal neoplasm; tissue type, colorectal.  ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Disease ------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Disease name Colorectal neoplasms Tissue filter Colorectal tumor Associated miRNAs\ hsa-mir-491, hsa-mir-185, hsa-mir-20a, hsa-mir-221, hsa-mir-199a, hsa-mir-34a, hsa-mir-199b, hsa-mir-34c, hsa-mir-34b, hsa-mir-148a, hsa-mir-342, hsa-mir-21, hsa-mir-499a, hsa-let-7c, hsa-mir-148b, hsa-mir-1915, hsa-mir-17, hsa-mir-320a, hsa-mir-200c, hsa-mir-143, hsa-mir-139, hsa-mir-103a, hsa-mir-103b, hsa-mir-107, hsa-mir-497, hsa-mir-106a, hsa-mir-429, hsa-mir-7, hsa-mir-362, hsa-mir-330, hsa-mir-367, hsa-mir-339, hsa-mir-133a (total 33) Targeted genes\ BCL2L1, RHOA, CDC42, BNIP2, CDKN1C, AXL, MYC, BCL2, DNMT1, RHOB, FOXO4, PDCD4, MMP11, PBX3, CCKBR, CCL20, RND3, NRP1, ZEB1, CTNNB1, MACC1, IGF1R, DAPK1, KLF4, RAP1B, TGFBR2, SOX2, YY1, RBL2, E2F1, USF2, PTPN1, RYR3, PLRG1, RFFL, DNMT3A, KRAS (total 37) Regulatory sub-pathways 45,565 Distinct genes 3,079 ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ By default, the interface "target-specific RSPs" ([Fig. 6](#fig-6){ref-type="fig"}) lists all the constructed RSPs, namely all 45,565 of them in the present case and, if requested, would present a map including all the RSPs which, however, would be impractical to visualize, not to say interact with. On the same interface are two options for displaying/using a smaller RSP set: "Gene filter," where the user can restrict the set to only those RSPs containing a specified gene; and "Show top ... sub-pathways," where the user can ask for only the top-*N* RSPs having the smallest *p*-values be listed and used for network construction. The interface "Disease specific RRN" then allows the user to choose one among the layouts "Tree," "Circle," and "Radial." Here, a tree-map, with several disconnected parts, built from the top-70 RSPs is shown ([Fig. 9](#fig-9){ref-type="fig"}). {#fig-9} The largest connected sub-RRN, or "Network-1" ([Fig. 10](#fig-10){ref-type="fig"}), is composed of six miRNAs targeting four genes connected to 52 other genes ([Table 5](#table-5){ref-type="table"}). Of the 56 genes in Network-1, 22 have known CRC connections (CORECG database, <http://lms.snu.edu.in/corecg>) ([@ref-2]), and 26 others have references linking them either directly or indirectly to CRC ([@ref-48]; [@ref-87]; [@ref-45]; [@ref-100]; [@ref-98]; [@ref-99]; [@ref-64]; [@ref-5]; [@ref-103]; [@ref-90]; [@ref-105]; [@ref-26]; [@ref-77]; [@ref-34]; [@ref-83]; [@ref-17]; [@ref-39]; [@ref-74]; [@ref-106]; [@ref-32]; [@ref-102]; [@ref-50]; [@ref-56]; [@ref-38]; [@ref-70]; [@ref-6]; [@ref-72]; [@ref-46]) ([Table 6](#table-6){ref-type="table"}). Among these, *TNIK* ([@ref-64]) and *TNK2* ([@ref-72]) have been used as drug targets for CRC treatment. We consider the remaining eight genes---*PRKACA, MAP3K12, LRRK1, RIOK2, OXSR1, CDK17, EIF2AK1, TSSK4*---to be potential novel CRC-related genes. Noticeably, Network-1 has two parts, one 28 nodes (five miRNAs targeting three genes) and the other 34 nodes (one miRNA targeting one gene), connected by a single link, or PPI. The three types of genes, known CRC-related, reference-supported, and potential CRC-related, are more or less proportionately distributed in these two parts. {#fig-10} 10.7717/peerj.7309/table-5 ###### Statistics and gene information in the network-1, the largest connected sub-network of the CRC-specific miRNA regulatory network.  Number Item set ----------------------------------- -------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------- Network-1 miRNAs 6 hsa-mir-199a, hsa-mir-34a, hsa-mir-199b, hsa-mir-139, hsa-mir-497, hsa-mir-106a Target genes 4 AXL, IGF1R, RAP1B, TGFBR2 Gene set (including target genes) 56 AXL, CSK, TNK2, LCK, PRKACA, FGR, MAPK15, IGF1R, MERTK, ERBB2, PTK2, EGFR, JAK2, JAK1, PRKCD, TEC, EPHB2, PHKG2, ROR1, FES, MAP3K12, RAP1B, MST4, PAK1, LRRK1, MAP2K3, CDK11B, ACVR1, TGFBR2, RIOK2, TGFBR1, MAP3K7, NEK8, NUAK2, OXSR1, CDK1, ACVRL1, MKNK2, STK35, CDK17, EIF2AK4, DAPK2, EIF2AK1, TSSK4, ZAK, MAP2K6, SIK3, VRK2, PINK1, TAOK2, TNIK, MAPK6, PRKACB, WNK1, PAK6, PKMYT1 10.7717/peerj.7309/table-6 ###### Known, literature supported, and potential novel CRC-related genes.  Number Item set --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------- Gene set (Network-1) Known CRC genes 22 AXL, LCK, FGR, IGF1R, MERTK, ERBB2, PTK2, EGFR, JAK2, JAK1, EPHB2, FES, PAK1, MAP2K3, ACVR1, TGFBR2, TGFBR1, CDK1, EIF2AK4, DAPK2, MAP2K6, PAK6 Reference supported ([@ref-48]; [@ref-87]; [@ref-45]; [@ref-100]; [@ref-98]; [@ref-99]; [@ref-64]; [@ref-5]; [@ref-103]; [@ref-90]; [@ref-105]; [@ref-26]; [@ref-77]; [@ref-34]; [@ref-83]; [@ref-17]; [@ref-39]; [@ref-74]; [@ref-106]; [@ref-32]; [@ref-102]; [@ref-50]; [@ref-56]; [@ref-38]; [@ref-70]; [@ref-6]; [@ref-72]; [@ref-46]) 26 CSK, TNK2[\*](#table-6fn1){ref-type="fn"}, MAPK15, PRKCD, TEC, PHKG2, ROR1, RAP1B, MST4, CDK11B, MAP3K7, NEK8, NUAK2, ACVRL1, MKNK2, STK35, ZAK, SIK3, VRK2, PINK1, TAOK2, TNIK[\*](#table-6fn1){ref-type="fn"}, MAPK6, PRKACB, PKMYT1, WNK1 Potential novel cancer-related gene 8 PRKACA, MAP3K12, LRRK1, RIOK2, OXSR1, CDK17, EIF2AK1, TSSK4 **Note:** Known target genes used in treatment of CRC. The "Gene filter" option ([Fig. 6](#fig-6){ref-type="fig"}) allows the user to focus on a specific gene in RRN construction. As example, *TNK2*, a key drug target for the treatment of metastatic CRC ([@ref-72]), was selected as the filter, together with the "Show top 70 RSPs" option. The result was a nine-node sub-RRN: the target gene *AXL* regulated by three miRNAs---hsa-mir-199b, hsa-mir-34a, hsa-mir-199a---and linked (by PPI) to *TNK2*, itself linked to four other genes *AXL*(OCG), *MAGI3, HSP90AB2P, MERTK*(OCG), *KAT8* ([Fig. 11](#fig-11){ref-type="fig"}). {#fig-11} Case 2. A comorbidity study of the disease-pair Alzheimer's disease-Type 2 diabetes ----------------------------------------------------------------------------------- Recent studies suggest a possible AD-T2D comorbidity. Known pathophysiological factors shared by AD and T2D include insulin, cholesterol, β-amyloid aggregation and tau ([@ref-4]). High cholesterol level impacts β-amyloid formation in the brain ([@ref-76]); abnormal insulin function, a key factor of T2D and related disorders ([@ref-14]), increases the risk for AD ([@ref-80]); cardiovascular risk factors such as high cholesterol and hypertension are common to T2D and AD ([@ref-52]). Evidences connecting AD to impaired function of insulin/IGF suggested AD might be viewed as a new type, "type 3," of diabetes ([@ref-55]). However, another study claims T2D to be associated with cerebrovascular but not Alzheimer neuropathology ([@ref-1]). Here, we demonstrate a two-disease application of miRDRN. After logging onto miRDRN's entry interface ([Fig. 3](#fig-3){ref-type="fig"}), click on "Comorbidity Search" to see a new interface urging the user to select two diseases; for "Disease 1," "Alzheimer disease" (AD) and tissue type "brain" were selected and for "Disease 2," "Type 2" (which stands for T2D) and tissue type "pancreas." Pathway ranking by "Jaccard index (MF)," and *p*-value \< "0.005" for both diseases were selected. Both AD and T2D are complex diseases and share aging for a risk factor; accumulated evidence indicates a connection between these two diseases at the molecular level ([@ref-3]). For this case miRDRN yielded, for AD (T2D), three (one) associated-miRNAs, three (one) targeted genes, 644 (3,908) RSPs, involving 633 (2,187) genes ([Table 7](#table-7){ref-type="table"}). Because AD and T2D did not have any common associated-miRNA target gene, they had distinct sets of RSPs. The 500 genes common to the two sets of RSPs (25 of which are shown in [Fig. 7](#fig-7){ref-type="fig"}) are significantly enriched in three KEGG terms: hsa03040:Spliceosome (*p*-value = 0.00549), hsa03018:RNA degradation (*p*-value = 0.00802), and hsa03022:Basal transcription factors (*p*-value = 0.00415). Abnormality of spliceosome has been reported in both AD ([@ref-62]) and T2D ([@ref-29]) patients. Among the comorbid genes, 8---*ALOX5, APP, BIN1, CHGB, VWF, NEFL, LETMD1, CELF1*- were identified as known AD target genes ([@ref-16]; [@ref-88]; [@ref-11]) and 14*---TCF7L2, APOA1, VWF, CDKN2B, CAT, ITGB2, ISL1, POLD3, APP, NFKBIB, GNA12, DEDD, LDLR, PRKAB1*- as known T2D target genes ([@ref-27]), *APP* and *VWF* are known targets of both diseases ([Table 8](#table-8){ref-type="table"}). With the exception of three---*LEMD1*, *POLD3, GNA12*, the comorbidity of all the others have literature support ([Table 8](#table-8){ref-type="table"}). 10.7717/peerj.7309/table-7 ###### Results for the AD-T2D comorbidity study.  ----------------------------------------------------------------------------------- Disease 1 Disease 2 Comorbidity ------------------------- ------------------ ------------- ------------------------ Disease name AD T2D AD/T2D Tissue filter Brain Pancreas Brain/pancreas Associated-miRNA hsa-mir-29a,\ hsa-mir-144 hsa-mir-29a,\ hsa-mir-195,\ hsa-mir-195,\ hsa-mir-146a hsa-mir-146a,\ hsa-mir-144 Targeted gene NAV3, BACE1, CFH IRS1 NAV3, BACE1, CFH, IRS1 Regulatory sub-pathways 644 3,908 4,552 Total no. of genes 633 2,187 2,320 No. of common genes -- -- 500 ----------------------------------------------------------------------------------- 10.7717/peerj.7309/table-8 ###### Known, literature supported, and potential novel AD-T2D comorbid genes.  ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- No. of targets in comorbidity gene set (500) Comorbid genes (references) ------------------- ---------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Known data Known AD target\ 8 ALOX5 ([@ref-71]; [@ref-67]; [@ref-40]), APP[\*](#table-8fn2){ref-type="fn"}, BIN1 ([@ref-36]), CHGB ([@ref-42]), VWF[\*](#table-8fn2){ref-type="fn"}, NEFL ([@ref-97]; [@ref-21]), LETMD1[^\#^](#table-8fn3){ref-type="fn"}, CELF1 ([@ref-92]; [@ref-15]) (210) Known T2D target\ 14 TCF7L2 ([@ref-18]; [@ref-91]; [@ref-7]; [@ref-78]), APOA1 ([@ref-96]; [@ref-73]; [@ref-61]), VWF[\*](#table-8fn2){ref-type="fn"}, CDKN2B ([@ref-69]), CAT ([@ref-37]; [@ref-63]; [@ref-51]), ITGB2 ([@ref-65]; [@ref-25]), ISL1 ([@ref-101]), POLD3[^\#^](#table-8fn3){ref-type="fn"}, APP[\*](#table-8fn2){ref-type="fn"}, NFKBIB ([@ref-57]), GNA12[^\#^](#table-8fn3){ref-type="fn"}, DEDD ([@ref-30]; [@ref-75]), LDLR ([@ref-33]; [@ref-20]), PRKAB1 ([@ref-94]; [@ref-35]; [@ref-84]) (497) ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **Notes:** The 210 known AD target genes were built by integrating gene lists from AlzGene ([@ref-16]), AlzBIG ([@ref-88]) and AlzBase ([@ref-11]); the 497 known T2D targets were from T-HOD ([@ref-27]). Known AD and T2D target. No literature support. Case 3. A sub-RRN centered on the AD-associated gene *BACE1* ------------------------------------------------------------ In recent years a number of anti-AD drugs designed on the basis of the amyloid-beta (Aβ) hypothesis of AD, which holds that Aβ aggregate in the brain is the main causative factor of AD, failed late-phase trials. These include the γ-secretase inhibitor Semagacestat ([@ref-86]) and two *BACE1* inhibitors, Verubecestat ([@ref-93]) and Atabecestat ([@ref-9]). In all three cases treatment groups scored worse than the control group on the Alzheimer's disease cooperative study activities of daily living inventory (ADCS-ADL) functional measure and reported more anxiety, depression, and sleep problems than controls. In a "Single Search" application on AD (tissue, brain; *p*-value threshold, 0.005), we had miRDRN construct a partial RRN (Gene filter, *BACE1*; Show top 70 sub-pathways; Network layout, Radial) centered on *BACE1*, which is a regulatory target of hsa-mir-195. The result shows the genes *PSEN1*, *NCSTN*, *RANBP9*, *PLSCR1*, *MMP2*, and *FURIN* to be immediately downstream to *BACE1* in the RRN ([Fig. 12](#fig-12){ref-type="fig"}). *PSEN1* and *NCSTN* encode proteins that are, respectively, catalytic and essential subunits of the γ-secretase complex; suppression of these genes are presumably the purpose of *BACE1* inhibition. On the other hand, *RANBP9* encodes a protein that facilitates the progression of mitosis in developing neuroepithelial cells ([@ref-22]); *PLSCR1* encodes a protein that acts in the control of intracellular calcium homeostasis and has a central role in signal transduction ([@ref-89]); *MMP2* encodes a protein that promotes neural progenitor cell migration ([@ref-79]). Suppression of these genes (by *BACE1* inhibition) may therefore adversely affect signal transduction and the nerve system, and could be part of the reason why Semagacestat, Verubecestat, and Atabecestat worsened the ADCS-ADL functional measure of treatment groups. {#fig-12} Conclusion ========== This work describes miRDRN (<http://mirdrn.ncu.edu.tw/mirdrn/>), composed of a new database on target-specific RSPs and a web service platform that allows the user to use the stored RSPs to construct disease and tissue-specific RRNs, which may aid the user to explore disease related molecular and pathway associations, or find new ones. As demonstration, miRDRN was applied to study the single disease CRC, where 34 potential target genes were identified, 26 of which have literature support; to study the comorbidity of the disease-pair AD-T2D, where 20 potential novel AD-T2D comorbid genes were identified, 17 of which have literature support; and to construct a partial miRNA regulatory sub-network centered on the AD-associated gene *BACE1*, which in turn suggests a possible explanation why, in late-phase trials that ended in failure, several γ/β-secretase inhibiting anti-AD drugs worsened the functional measure of treatment groups. We believe that findings from miRDRN, even exploratory in nature, may potentially lead to the identification of new drug targets and new understanding in modes of drug action. AD : Alzheimer's disease BioGrid : biological general repository for interaction datasets CRG : cancer related gene GO : gene ontology database HMDD : human microRNA and disease associations database KEGG : Kyoto encyclopedia of genes and genomes miRDRN : miRNA disease regulatory network database and web service platform MRSP : miRNA-specific regulatory sub-pathway OCG : oncogene PPI : protein--protein interaction PPIN : PPI network RSP : target-specific regulatory sub-pathway RRN : disease-specific miRNA regulatory network T2D : Type 2 diabetes TarBase : database on miRNA:mRNA interactions TSG : tumor suppressor gene. Additional Information and Declarations ======================================= The authors declare that they have no competing interests. [Hsueh-Chuan Liu](#author-1){ref-type="contrib"} conceived and designed the experiments, performed the experiments, analyzed the data, contributed reagents/materials/analysis tools, prepared figures and/or tables, authored or reviewed drafts of the paper, approved the final draft. [Yi-Shian Peng](#author-2){ref-type="contrib"} authored or reviewed drafts of the paper, approved the final draft, provided material and did literature search for discussion in the manuscript related to AD, T2D, and anti-AD drugs. [Hoong-Chien Lee](#author-3){ref-type="contrib"} conceived and designed the experiments, analyzed the data, authored or reviewed drafts of the paper, approved the final draft. The following information was supplied regarding data availability: The source code and data of miRDRN is available at and freely downloadable from the public repository GitHub: <https://github.com/o2snow/2019MIRDRN>.

Introduction ============ Physical inactivity is related to all-cause mortality and implicated in 6% of total deaths globally ([@B71], [@B71]). Physical inactivity is also associated with increased risk from multiple chronic illnesses and conditions including cardiovascular disease, diabetes, some cancers, and obesity ([@B8]; [@B38]; [@B6]; [@B9]). Numerous studies have demonstrated the benefits of participating in regular physical activity. Population studies have shown that participation in regular physical activity is likely to reduced all-cause mortality, notably deaths caused by cardiovascular disease ([@B8]). This has led to the development of national guidelines for the type, frequency, intensity and duration of physical activity thought to confer health benefits. While there are idiosyncratic differences in guideline content, most advocate at least 30 min of moderate-to-vigorous physical activity on at least 5 days of the week ([@B49]; [@B72]). As a consequence many governments have aimed to develop national programs to intervene and promote participation in physical activity at the population level ([@B16]). The development of such interventions, however, needs to be based on persuasive techniques and strategies to encourage individuals to take up physical activity and to provide them with the skills and personal resources to do so. Research on such techniques and strategies is based on formative psychological research that identifies the factors underpinning health behavior. At the forefront of this research are social cognitive theories derived from social psychology, which have provided insight into the factors that impact on decision making and have demonstrated efficacy in understanding the processes involved ([@B34]; [@B13]; [@B31]). These factors can then be targeted in behavioral interventions using specific techniques to manipulate or change the influential factors ([@B7]). Research, however, is needed to target and manipulate each specific component to fully understand the precise mechanisms underpinning behavior change. This can be done in factorial designs in which one or more intervention techniques can be targeted and their individual and interactive effects on a behavioral outcome evaluated ([@B53]). This is also important if the effectiveness of theories of behavior change in guiding behavior change interventions is to be evaluated. Many social cognitive theories applied to predict and understand health behavior have intention as a focal variable and the most proximal predictor of behavior ([@B7]; [@B47]; [@B14]; [@B59]). However, research has consistently noted the generally weak link between intentions and behavior, known as the intention-behavior 'gap' ([@B51]; [@B58]; [@B67]; [@B64]). According to [@B63], intention measures account only for 28% of the behavior. Furthermore, interventions focusing on changing interventions and its antecedents have been shown to be successful in changing intentions but are much less effective in changing behavior ([@B70]). Research has shown that some individuals can have strong intentions to engage in health behavior but experience difficulties in converting those intentions into action. These individuals have been identified as 'inclined abstainers' ([@B51]) or 'unsuccessful intenders' ([@B58]). Researchers have therefore sought to examine the process by which intentions are converted into action. One approach has been to focus on volitional processes, that is, the processes involved in the enactment of intentions after they have been formed ([@B34]; [@B62]). Prominent among these is the model of action phases ([@B34]; [@B22]), which makes the distinction between volitional and motivational phases prior to action. The motivational phase makes reference to the formation of actual intentions in which individuals form intentions based on their beliefs, similar to the process outline in the theory of planned behavior. The volitional phase occurs after the individual has formed and intention and made the decision to act. This phase outlines the process by which an individual enacts their intentions and is dependent on the individual developing a sufficient future strategy to enact the planned behavior. Whether individuals are successful in enacting a plan is, therefore, dependent on whether they have an intention to act and whether they have a sufficient strategy to carry out their intention. Accordingly, the model of action phases implies that the motivation is a necessary but not sufficient condition for behavioral enactment, which is where the intention-behavior gap arises. Intentions must, therefore, be augmented with sufficient strategies in the volitional phase that enables individuals to carry out their plans ([@B43], [@B42], [@B44]; [@B65]; [@B11]). A prominent strategy that has been purported as a means to assist individuals in enacting their intentions and countering the gap between intention and behavior is implementation intentions ([@B64]). An implementation intention is a self-regulation strategy by which an individual augments their intention with a cue-initiated plan to increase the likelihood that they will carry out their intention. An implementation intention is a mental act linking an anticipated critical situation or 'cue' with the appropriate intended response ([@B22]). By identifying a cue present in the environment and pairing it with the intended action, the individual will increase the likelihood of the intention being enacted. The cue should be an appropriate condition, time and place in which the action should be performed. [@B22] suggests that the implementation intention leads to the establishment of a strong link between the cue and the goal-directed response. The mechanism by which implementation intentions exerts its effects is through the formation of effective plans at the post-intentional stage, i.e., rather than being mediated by intentions, implementation intentions should affect behavior through planning ([@B46]). Meta-analyses of implementation intention interventions have shown a medium-to-large effect size in promoting health-related behaviors generally ([@B25]) and a small-to-medium effect size in promoting physical activity ([@B5]). However, it is important to note that these meta-analyses included few unpublished studies and there are some indications of null or detrimental effects for implementation intention interventions in the literature ([@B36], [@B37]; [@B66]; [@B35]; [@B61]; [@B39]). While implementation intentions have been found to be effective across studies, meta analyses have also identified considerable heterogeneity in the effects of implementation intention across the literature. [@B32] indicated that the way implementation intention manipulations are operationalized could potentially moderate their effect. The format and presentation of plans feature prominently as a potential moderator of planning interventions. Research has suggested that implementation intention effects are often facilitated by prompting individuals to form plans using and 'if-then' format, that is, prompting individuals to identify a critical situation and pair it with their intended behavior ([@B24]), as opposed to more 'global' or open-ended plans without an 'if-then' prompt ([@B10]). Other critical moderators that have been identified include intrapersonal factors such as individual differences in planning capacity and time delay between the administration of the planning technique and the enactment of the target behavior. There have, however, been few systematic tests of these moderators using factorial designs in the literature ([@B32]). The effectiveness of implementation intentions is also dependent on individuals having sufficient motivation to engage in the behavior. Maximizing intentions, a construct that reflects an individual's level of motivation toward a given behavior and the amount of effort they are prepared to invest in pursuing it, is an important prerequisite for the effectiveness of any planning intervention ([@B65]; [@B27]). This is based on the assumption in the model of action phases that volitional components like planning affect behavior in a post-decisional manner. Means to bolster motivation to participate in the behavior alongside a volitional strategy like implementation intentions, therefore, will ensure that an individual is an optimal motivational and volitional state to enact their intentions and engage in the behavior. An effective means to bolster motivation is through mental simulation. Mental simulation is a strategy in which individuals visualize or imagine engaging in the intended action ([@B54]) and is supposed to reinforce individuals' intention to act by increasing their propensity and the readiness to act ([@B54]). As a consequence, the effects of mental simulation occur through changes in self-efficacy and it is considered to affect behavior as part of the motivational phase in [@B34] model. Specifically, mental simulations boost self-efficacy as the imagined action serves as a 'self-model' or subjective 'vicarious experience' ([@B4]). The effects are therefore expected to be mediated by self-efficacy perceptions. A sequential multiple mediation model can therefore be envisaged in which effects of mental simulations on action occur through changes in self-efficacy followed by changes in motivation or intention. There are two types of mental simulations. Outcome mental simulation requires individuals to imagine the achievement of the goal (e.g., simulating the situation once the goal is reached alongside the emotions and cognitions the individual is likely to have). Process mental simulation requires individuals to imagine the means required to attain the behavioral goal (e.g., simulating how the individual is enacting the behavior focusing on the details, like body sensations, emotions, details of the environment). Research on mental simulations has shown them to be effective in boosting behavioral engagement and goal-progress in many domains including health ([@B54]; [@B69]; [@B29]). A further proposed mechanism for the effects of mental simulation on behavior change is through their capacity to 'activate' or make salient certain action-oriented mindsets or behavioral tendencies. Mental simulations may work by providing a bridge between thoughts and goal-directed behavior ([@B54]). Mental representations of the goal-directed behaviors may involve fantasizing about past events that have already been experienced or future events that have never been realized. For example, studies have suggested that mental simulation activates an open-minded mindset while implementation intention activates an implemental mindset by focusing on the relevant components of if-then plans ([@B19]). But when mental simulation is realized before implementation intention, it would enable the individual to reflect on the different paths to set up a goal in the motivational phase of decision making and then set up strategies to achieve the goal in the volitional phase ([@B34]). Studies have investigated the effectiveness of combined mental simulation and implementation intentions interventions to promote reductions in alcohol consumption. The interventions were based on the hypothesis that combining a mental simulation intervention that promotes motivation with an implementation intention that promotes volition will optimize participation in health behavior ([@B40]; [@B29],[@B30]). This follows research that suggests that creating conditions that maximize intentions to engage in the health behavior and augmenting the intentions with means to implement the intentions should lead to optimal conditions for behavioral enactment ([@B56]). A review of nine studies testing effects increasing intention strength and implementation intentions using full-factorial designs revealed that six showed significant interaction effect such that behavior participation was optimal when intentions were strengthened alongside implementation intentions. [@B29] adopted a similar full-factorial design to examine the independent and interactive effects of mental simulations and implementation intentions on alcohol consumption. Results revealed that the combination of mental simulation and implementation intention interventions resulted in a significant decrease in the number of alcohol units consumed relative to participants receiving each of the intervention conditions alone or neither, but only among participants with high baseline alcohol consumption. Research adopting the same design has found that only implementation intention intervention was effective in evoking change ([@B30]). While these findings provide some indication that the combination of mental simulation and implementation intention strategies can lead to optimal engagement in health behavior, results do not provide unequivocal support and [@B29] conditional findings for high alcohol consumers suggest that effects are most substantive when motivation is low and individuals are resistant to change and have high levels of an undesired behavior. The model of action phases clearly dictates that motivation is a necessary but insufficient condition for behavioral enactment, and that behavioral enactment is also dependent on conditions in the volitional phase such as planning that lead to better behavioral enactment. Research has provided some support for the interactive effect of motivational and volitional interventions in promoting health behavior, including mental simulations with implementation intentions ([@B48]; [@B29]), but some research has not supported the interactive effect. These may be findings that are specific to a particular behavior or particular context, or, as in the [@B30] study, among participants with high levels of the undesirable behavior to be changed. In the present study we aimed to conduct a series of interventions using mental simulations and implementation intentions to promote physical activity behavior. In Study 1, we examined the effect of an intervention combining implementation intentions and mental simulations on physical activity participation relative to an implementation intention intervention alone and a no intervention control condition in a student sample. We expected that the combined intervention condition would yield greater participation in physical activity relative to the implementation intention only condition and the control condition after controlling for baseline physical activity. We also expected participants in the implementation intention condition to report greater physical activity participation than those in control condition. In Study 2 we examined the unique and interactive effects of these manipulations on physical activity using a full-factorial design. In addition, we aimed to provide a robust test of the interactive effects of these interventions by examining their effects in a participants including students and members of the general public joining a fitness center with little or no previous experience of physical activity and using center attendance as an objective proxy measure of physical activity participation. We expected to find main effects of the mental simulation and implementation intention interventions on physical activity participation post-intervention controlling for baseline values on physical activity. We also expected to find an interaction effect of the mental simulation and implementation intention interventions such that participants receiving both intervention would exhibit the highest levels of physical activity participation relative to each condition alone and a control condition receiving neither. Study 1 ======= Adopting a three-group cluster-randomized design, we aimed to examine the effects of a mental simulation and implementation intention intervention on physical activity behavior in a sample of students relative to an implementation intention only intervention and a measurement-only control group (Supplementary File). The intervention conditions and measures of behavioral and psychological variables were administered at baseline (T1) with 1-week (T2)^[1](#fn01){ref-type="fn"}^ and 4-week (T3) behavioral follow-up occasions. We expected that the participants who formed a mental simulation alongside an implementation intention (mental simulation plus implementation intention group) would report higher levels of physical activity at 1 and 4 weeks post-intervention than participants who formed an implementation intention (implementation intention group) alone and participants who received neither of the intervention components (control group). Materials and Methods {#s1} --------------------- ### Participants Participants were undergraduate students from Grenoble Alpes University (*N* = 92, females = 79, males = 13; *M*~age~ = 24.4 years, *SD* = 6.44) recruited on a voluntary basis by advertising the study to students eligible to participate before their classes. Students were considered ineligible to participate if they were aged less than 18 years or had reduced mobility or a disability which prevented them from participating in physical activity. ### Design and Procedure At baseline, all participants completed consent forms followed by a baseline questionnaire containing self-report measures of demographic details (age, BMI, smoking status, number of alcoholic drinks consumed per occasion and frequency of alcohol consumption) ([@B17]; [@B33]), theory of planned behavior variables (attitude, subjective norms, perceived behavioral control and intention) ([@B3]) and past physical activity behavior ([@B12]). Participants allocated to the implementation intention group were then prompted to form an implementation intention with respect to physical activity presented as a pen-and-paper exercise. Participants allocated to the mental simulation plus implementation intention group formed a mental simulation, again as a pen-and-paper exercise, prior to forming an implementation intention. Participants from the two experimental groups then completed manipulation check measures of mental simulation and/or implementation intentions according to their allocated group. Participants allocated to the control group completed the measures but did not receive any intervention. One and 4 weeks after baseline, participants completed follow-up self-report measures of theory of planned behavior variables, and physical activity behavior. Three teachers from the Grenoble Alpes University were contacted to give them information about the study and ask for their authorisation to recruit students from their classes. We chose to randomize the sample by clusters determined by university class using a random numbers table in order to reduce the likelihood of data contamination through participants allocated to different conditions conferring. Regarding the sequence generation, no allocation concealment was made. Patients were blinded to group allocation, but the experimenters administering the study materials were not. Data were collected from March to May 2015 and we stopped collecting the data when sufficient numbers of participants were included to achieve adequate statistical power. ### Informed Consent At baseline, participants read an information sheet describing the study and outlining their rights and benefits, and the potential risks of participation. They then signed a written informed consent form which was detached from the questionnaire thereafter to maintain participant anonymity. At the three data collection occasions, participants formed a unique identifier comprising the first two letters of their mother's name, father's name, and their birth month and day. Questionnaires were matched across data collection points using participants' unique identifier. Prior to data collection ethical approval was obtained from the institutional review board of the University of Grenoble. ### Implementation Intention Intervention Participants allocated to the implementation intention condition and to the mental simulation plus implementation intention condition were asked to form implementation intentions with respect to regular physical activity for the forthcoming month. An 'if-then' format was used to link the situation (preceded by the prompt "if") with the goal-oriented behavior (preceded by the prompt "then") ([@B22]). Participants were presented with the following script, which was followed by a series of prompts including the day and hour the participant planned to exercise, the type of exercise, and the place where they planned to exercise accompanied by a series of blank lines so that they could write down their responses: > "A lot of people have the intention to be more physically active, but they forget or do not find the time to do it. Some studies have shown that if you form a plan specifying when, where and how you will do your physical activity, you will be more likely to remember to do it and find the time to do it. Thus, it would be useful to spend a few minutes to think about the times when you will participate to a physical activity in the next month. Try to think about the best time for you to exercise, and do not share your answers with your colleague." ### Mental Simulation Intervention Participants in the mental simulation plus implementation intention condition were prompted to carry out a mental simulation exercise prior to the formation of an implementation intention. Our mental simulation manipulation was identical to that used by [@B54] adapted for physical activity and focused on process mental simulation. Participants were presented with the following written message, which was followed by a series of blank lines on the page for participants to write down their response: > "In this exercise, you will be asked to visualize yourself doing physical activity to increase your regular participation in physical activity. As of today, and for the forthcoming month, imagine how you would do physical exercise regularly. It is very important that you actually 'see' yourself doing physical activity and have that picture in your mind. Please write on the lines below how you imagine you will achieve your goal of doing regular physical activity, in the following month." Measures -------- ### Theory of Planned Behavior Variables We measured the variables related to the theory of planned behavior ([@B3]) adapted to refer to regular participation in physical activity. Intention to engage in physical activity was measured using four items (e.g., "I intend to do physical activity regularly in the forthcoming month"). Attitudes toward physical activity were measured using five items (e.g., "For me doing physical activity regularly in the forthcoming month is harmful/beneficial"). Subjective norms (e.g., "Most people who are important to me think that I should/I should not do physical activity regularly") and the probability the participant would enact the behavior (e.g., "It is expected of me that I do physical activity regularly in the forthcoming month") were measured using four items each. Perceived behavioral control as function of the perceived difficulty (e.g., "For me doing physical activity regularly in the forthcoming month would be impossible/possible") and ability (e.g., "If I wanted to, I could practice physical activity regularly in the forthcoming month") to engage in the behavior were measured using two items each, the total score for each was computed with the mean of the two items. All the responses were made on 7-point Likert scales, with the exception of the attitudes scale which was measured on 7-point bi-polar semantic differential scales. ### Physical Activity Behavior The International Physical Activity Questionnaire -- Short Form (IPAQ-7; [@B15]) was used to measure self-reported physical activity. Participants indicated the number of minutes, hours, or days during which they did physical activity over the previous seven last days at three levels of intensity: vigorous activity, moderate physical activity, and walking. Each of level of intensity was accompanied by a detailed description so that participants were familiar with the types of activity that fell into each category. The measure yields a total physical activity score expressed in Metabolic Equivalent Task per minute (METs/min), an indicator of energy expenditure as a function of physical activity intensity. ### Manipulation Checks Participants in the intervention groups also completed manipulation check measures ([@B29]) to measure the extent to which they adhered to the instructions for the mental simulation (e.g., "To what extent have you figured out exactly how you might do physical activity regularly over the next month") and implementation intention (e.g., "To what extent do you have a plan for when, where, and how you might do regular physical activity in the following month") exercises on 7-point Likert scales (from 1 "I have no idea" to 7 "I have figured out exactly"). Results ------- ### Participants Ninety-two students agreed to participate and completed study measures questionnaires at T1 (females = 79, males = 13; *M*~age~ = 24.4 years, *SD* = 6.44). Seven participants (five in the control group and two in the mental simulation plus implementation intention group; 7.61% of the initial sample) dropped out of the study at T2 and 11 participants (two in the control group, four in the implementation intention group and five in the mental simulation plus implementation intention group; 9.47% of the initial sample) dropped out of the study at T3 (**Figure [1](#F1){ref-type="fig"}**). Of the recruited participants, five dropped out of the study at T2 only (three in the control group and two in the mental simulation plus implementation intention group), 15 dropped out at T3 only (five in the control group, three in the implementation intention group and seven in the mental simulation plus implementation intention group) and two dropped out at both T2 and T3 (all in the control group). Thus 64 participants (58.88% of retention rate) completed the study at the 2- and 4-week follow-up occasions (females = 59, males = 10; *M*~age~ = 25.19 years, *SD* = 5.44). {#F1} ### Missing Data Preliminary analyses of the raw data from the questionnaire measures at follow-up revealed that 7.12% of the data values were missing due to failures to respond. Little's MCAR test revealed that values were missing completely at random (χ^2^ = 163.352, df = 175, *p* = 0.726). Missing data were imputed using multiple imputation in SPSS v. 20 using the fully conditional specification imputation method with 10 iterations and a linear regression model for scaled variables (SPSS Inc.). ### Attrition Checks Chi-square tests and MANOVAs were used to test for differences on study variables due to attrition across the time points. Differences on categorical variable (i.e., gender) were tested using a χ^2^ analysis with the level of the variable of interest (male vs. female) cross-tabulated with attrition (dropped out of study after T1 vs. remained in study at T3). Differences on continuous demographic (age, BMI, alcohol consumption) and psychological (theory of planned behavior, self-efficacy) and behavioral (physical activity) variables were tested using MANOVAs with the study variables as multiple dependent variables and attrition as the independent variable. Participants who completed the study did not differ significantly from the participants who dropped out on gender \[χ^2^(1, *N* = 92) = 0.03, *p* = 0.863, $\eta_{p}^{2}$ = -0.018\] and on demographic, psychological and behavioral variables \[Wilks' Λ = 0.87, *F*(9,75) = 1.25, *p* = 0.277, $\eta_{p}^{2}$ = 0.131\]. ### Randomisation Checks Sample demographic and psychological data are presented in **Table [1](#T1){ref-type="table"}**. Randomisation checks were conducted using χ^2^ tests and a MANOVA. There was no significant multivariate effect for intervention condition on the demographic, psychological and behavioral variables at baseline, but the effect only narrowly fell short of conventional levels of statistical significance \[Wilks' Λ = 0.69, *F*(18,148) = 1.66, *p* = 0.052, $\eta_{p}^{2}$ = 0.168\]. When we looked at the between-participants effects, the three groups differed significantly on the physical activity measure at baseline \[*F*(2,82) = 3.80, *p* = 0.026, $\eta_{p}^{2}$ = 0.085\]. In addition, there were fewer smokers in the mental simulation plus implementation intention group (3 smokers) than in the implementation intention group (15 smokers) and in the control group \[14 smokers; χ^2^(2, *N* = 91) = 10.94, *p* = 0.027, $\eta_{p}^{2}$ = 0.245\]. There were also more men in the mental simulation plus implementation intention group (eight males) compared to the implementation intention (three males) and control (two males) groups \[χ^2^ (2, *N* = 92) = 6.98, *p* = 0.031, $\eta_{p}^{2}$ = 0.275\]. ###### Self-reported sample characteristics of Study 1 (*n* = 92). Control group (*n* = 30) Implementation intention group (*n* = 34) Mental simulation plus implementation intention group (*n* = 28) ----------------------------------------- -------------------------- ------------------------------------------- ------------------------------------------------------------------ **Age (years)** Mean (*SD*) 24.75 (6.04) 23.21 (5.03) 25.87 (8.63) **Gender** Women 28 (35.4%) 31 (39.2%) 20 (25.3%) Men 2 (15.4%) 3 (23.1%) 8 (61.5%) **BMI** Mean (SD) 21.49 (2.30) 21.94 (2.65) 21.33 (3.54) **Smoking status** Smoker 14 (43.8%) 15 (46.9%) 3 (9.4%) Non-smoker 15 (28.8%) 17 (32.7%) 20 (38.5%) Former smoker 1 (14.3%) 2 (28.6%) 4 (57.1%) **Number of alcoholic drinks per week** Mean (SD) 2.68 (1.22) 3.09 (1.40) 2.26 (1.54) **Frequency of alcohol consumption** Mean (SD) 2.14 (1.46) 1.74 (1.08) 1.78 (1.68) **Attitude** Mean (SD) at T1 5.90 (0.79) 5.58 (1.53) 5.81 (1.17) Mean (SD) at T2 5.87 (0.85) 5.79 (1.14) 6.03 (1.01) Mean (SD) at T3 5.99 (0.95) 5.80 (1.24) 5.90 (1.47) **Subjective norms** Mean (SD) at T1 4.85 (0.81) 4.38 (0.96) 4.78 (1.10) Mean (SD) at T2 4.59 (0.89) 4.16 (1.18) 4.43 (1.32) Mean (SD) at T3 4.61 (0.94) 4.14 (1.27) 4.24 (0.95) **Intention** Mean (SD) at T1 4.92 (2.01) 5.35 (1.79) 4.70 (2.17) Mean (SD) at T2 5.00 (1.77) 5.36 (1.66) 4.92 (1.93) Mean (SD) at T3 5.04 (1.78) 5.00 (1.88) 4.91 (1.97) **Perceived behavioral control** Mean (SD) at T1 4.88 (1.46) 4.63 (1.57) 4.52 (1.51) Mean (SD) at T2 4.63 (1.40) 4.47 (1.68) 4.43 (1.63) Mean (SD) at T3 4.58 (1.56) 4.20 (1.83) 4.80 (1.61) **MET-min per week** Mean (SD) at T1 7.38 (1.18) 7.57 (0.85) 6.90 (1.52) Mean (SD) at T2 7.49 (0.87) 7.37 (1.49) 7.21 (1.30) Mean (SD) at T3 7.68 (0.86) 7.64 (0.98) 7.35 (1.17) ### Intervention Effects The primary physical activity outcome variable expressed in MET-min did not follow a normal distribution, so we computed a logarithmic transformation. We opted for a natural logarithmic transformation, because the data were calculated on a proportional basis and were easily interpretable ([@B20]). Physical activity data for the intervention conditions are presented in **Table [2](#T2){ref-type="table"}**. ###### Results of linear mixed effect model of physical activity measured at T2 and T3 on interventions (Study 1). Mean between-group difference^a^ Standard deviations^b^ Cohen's *d*^c^ *t*-value *p*-value ----------------------------------------------------------------------------------------------- ---------------------------------- ------------------------ ---------------- ----------- ----------- **Fixed effects** Implementation intention only group vs. control group -0.28 0.26 -0.38 -1.10 0.27 Mental simulation plus implementation intention group vs. control group 0.07 0.27 0.10 0.25 0.79 Mental simulation plus implementation intention group vs. Implementation intention only group 0.36 0.23 0.49 1.53 0.13 Residual 0.73 AIC 537.69 \- \- \- Log-likelihood -260.85 \- \- \- Linear mixed effect model based on 92 participants and 276 observations. Dependent variable in all cases in metabolic equivalent (MET) per minute (MET.min -1 ). AIC, Akaike Information Criterion. a Difference in mean MET.min -1 across intervention groups at T2 and T3 combined. b Standard deviations of effect sizes estimates. c Cohen's d effect size coefficients were computed using residual standard deviations. Given the risk of contamination related to the cluster design and the three-wave design, we tested our hypotheses using a regression model with random effects using recursive maximum-likelihood estimates and using R version 3.1.1 with the "nlme" package ([@B45]). We analyzed the effect of the intervention by conducting a linear mixed effect model regression of the intervention conditions on physical activity measured at T2 and T3 controlling for physical activity measured a T1. A dataset was created for the mixed model with the three groups (control group, implementation intention group and mental simulation and implementation intention group) as between-participants effects and time as a within-participants effect on the physical activity scores at each time-point. Model selection based on Akaike's information criterion selected a model with a random slope. The resulting effects indicate between-group comparisons Cohen's *d* effect size coefficients were computed using residual standard deviations ([@B60]). Results are presented in **Table [2](#T2){ref-type="table"}**. Results revealed no significant between-group effects on aggregate values for on physical activity measured in MET-min per week at T2 and T3, controlling for gender (as the groups were not equivalent on the males percentage) and physical activity at T1^[2](#fn02){ref-type="fn"}^. However, effect size statistics suggested medium effect sizes for the between-group comparisons for the combined group, which implies that we had insufficient statistical power to detect these effects. Discussion ---------- The purpose of the current study was to examine the effects of an intervention combining mental simulation and implementation intention strategies to promote physical activity at 1- and 4-week post-intervention follow-up. We conducted the intervention in a sample of undergraduate students adopting a three-group cluster-randomized design with classes allocated to an implementation intention group, a mental simulation plus implementation intention group, or a control group. Findings revealed no differences in physical activity participation at follow up between the implementation intention, implementation intention plus mental simulation, and control groups. While these data indicated that these intervention techniques were not effective in promoting physical activity participation in this study and were inconsistent with the bulk of the research on implementation intention and mental simulation in research on health behavior, the effect size data indicated that possible differences were present but the study was underpowered. It is, therefore, important to interpret current findings in the context of its limitations and acknowledge the importance of replication in larger, representative samples. The current study had some limitations. The lack of provision of a goal intention in advance of completing the implementation intention exercise meant that the implementation intention exercise would have had very little relevance to those who did not have a goal of increasing their physical activity participation. The adoption of a cluster randomisation design may not have led to homogenous groups on key demographic and behavioral variables, although we controlled for potential variations on gender distribution, smoking status, and the number of alcoholic drinks per week to counteract differences across clusters. The reliance on a self-report measure of physical activity despite acceptable psychometric properties of the IPAQ-7 measure and good correlations with objectives measures in previous research ([@B15]) was a potential source of error variance. However, the most prominent limitation was the small sample size and the medium-sized effect sizes seemed to indicate this despite the non-significant findings. As a consequence, the current effect deserves further replication in a sufficiently powered trial. A final limitation was the lack of a full-factorial randomized controlled design enabling us to study the main effects of each intervention component, mental simulation and implementation intentions, independent of the interaction effect. This is important if researchers are to decode the precise components responsible for affecting a change in physical activity behavior, and whether the interactive effects are truly separate from the main effects of each component. Study 2 ======= Study 2 added to and extended the findings of Study 1 by adopting a full-factorial, fully randomized between-participants design with a larger sample. Study 2 also had the advantage of recruiting participants who had signed up to attend a fitness center and had not attended the center previously, who were low active, and had no recent experience of physical activity behavior (Supplementary File). Two weeks after their enrolment to the fitness center (Time 1, T1), they were randomly allocated to mental simulation, implementation intention, mental simulation plus implementation intention group, or control groups. In addition, our target dependent variable was fitness center attendance verified through participants' fitness center records and used as a proxy measure of physical activity participation, with 4-week (Time 2, T2) and 19-week (Time 3, T3, only for a subsample) behavioral follow-up occasions. We expected that participants who received the mental simulation and implementation intention manipulations together would exhibit higher fitness center attendance at follow-up than participants receiving the mental simulation and implementation intention interventions alone. We also expected participants receiving either the mental simulation manipulation, implementation intention manipulation, or both, to have greater attendance at the center than participants in the control group that received neither of the manipulations. Materials and Methods {#s2} --------------------- ### Participants Participants (*N* = 184, females = 97, males = 87; *M*~age~ = 26.95 years, *SD* = 9.59) were recruited from two different locations and randomized to each intervention conditions within each location, on a voluntary basis. In the first sample, participants were recruited from the University of Essex fitness center, which had both student and non-student attendees. In the second sample, participants were recruited from a commercial fitness center in the city of Stockport, UK, which comprised individuals from the general population. As the two studies adopted the same design and used the same materials and protocol, we opted to combine the samples. Although this decision violated randomisation for the overall sample, randomisation was retained within the samples (see Results section), and our decision to pool the samples was based on strict criteria to ensure there were no differences in the samples on key demographic factors. Inclusion criteria were identical to those of Study 1 except that participants aged under 18 years were also eligible for inclusion in the study as they had passed the fitness center induction criteria. ### Design and Procedure A randomized controlled design was adopted with four groups: implementation intention only, mental simulation only, mental simulation plus implementation intention, and control groups. The design was identical in both samples. Statistical power analysis revealed that a sample size of 144 participants was required (power = 0.95, alpha = 0.05, *d* = 0.65, *f* = 0.325; [@B25]) for a two-way mixed-model ANOVA with time \[with two levels: baseline (T1), 4-week follow-up (T2) or 19-week follow-up (T3)\] as the within-participants factor and experimental condition (condition: implementation intention only, mental simulation only, mental simulation plus implementation intention group, control group) as the between-participants factor. The participants were asked to participate at T1 in the 2nd week after their enrolment to the centers. They were randomly allocated to the mental simulation plus implementation intention group, the implementation intention group, the mental simulation group, or the control group within each sample. At baseline, all participants completed measures of socio-demographic, past behavior and theory of planned behavior variables (i.e., attitude, intention, subjective norms, and perceived behavioral control). Consistent with the findings of Study 1, BMI, smoking status, alcohol consumption and self-control were excluded as these measures had no relation to the behavioral outcome and we wanted to reduce response burden on participants. At T1, participants were allocated to the mental simulation plus implementation intention group, the implementation intention group, the mental simulation group or the control group using a random numbers table generated by the experimenter. No allocation concealment was made regarding the sequence generation. Patients were blinded to group allocation, but the experimenter administering the study materials was not. The participants were allocated to the implementation intention group were prompted to form implementation intentions and participants allocated to the mental simulation group were required to undertake a mental simulation exercise. Participants from the mental simulation and implementation intention group were required to undertake both exercises. Participants allocated to the control group completed the demographic, psychological, and physical activity measures only. At T2, all participants completed follow-up measures of theory of planned behavior variables. Participants from University of Essex campus fitness center also completed these measures at T3 because we were unable to obtain further access to the participants from Stockport fitness center at T3. Fitness center attendance was measured at the three time points, 2 weeks after initial enrolment to the fitness center (T1), at a 6-week follow-up occasion (T2) for both samples, and at a 19-week follow-up (T3) for participants from University of Essex campus fitness center. Participants were contacted via email with their email address and participant code stored in a database separate to the main data with data subsequently matched by a researcher who did not have access to the participants' data to retain participant anonymity. Data were collected between September and April, 2008. We stopped collected the data when sufficient numbers of participants were recruited according to our power analysis. ### Informed Consent and Anonymity Participants read an information sheet describing the study and outlining the rights, benefits, and potential risks of the study to participants before being enrolled in the study. They then signed a written informed consent form which was detached from the questionnaire thereafter in order to maintain participant anonymity. A unique identifier comprising the first two letters of participants' home post code, the first three letters of their mother's maiden name and their landline phone number was formed at each time point of the study. This study obtained the ethical approval from the University of Essex Department of Psychology Research Ethics Committee prior to data collection. ### Intervention Manipulations Before the mental simulation and implementation intention statements, the intervention groups read the following introductory passage: > "Please complete the following mental and/or visualization exercises, they will only take about 5 min of your time. Please take care to complete each exercise in full and follow the instructions carefully, they will ultimately benefit you in future." In addition, participants were provided with WHO guidelines for physical activity at the beginning of the questionnaire using the following passage: "3--4 periods of exercise per week lasting no less than 30 min." The intervention was a pen-and-paper exercise and participants were not assisted in performing the mental simulation and implementation intention exercises. ### Implementation Intention Manipulation Participants allocated to the implementation intention condition or mental simulation plus implementation intention condition were prompted to form an implementation intention with respect to their fitness center (gym) attendance. They were presented the following script which was followed by a series of blank lines to write down their plan: > "You are more likely to carry out your intention to use the gym regularly if you make a decision about where and when you will do so. Decide now when and how you will use the gym over the following weeks. You may find it useful go to the gym just before or just after something else that you do regularly, such as lectures. Please write in below when and how you will use the gym in the next 6 weeks (e.g., 'I will do the 'Fat Burning' class on every evening after my lecture')." ### Mental Simulation Intervention Participants in the mental simulation condition and mental simulation plus implementation intention condition performed the mental simulation exercise. We used the same manipulation as [@B54] adapted to physical activity with some minor modification to encourage individuals to engage in the visualization exercises and focusing on the outcomes. They were presented with the following text introducing the exercise: > "Think about the exercise related goal that you have just decided on. Visualize yourself having achieved that goal. You have put a lot of effort into the achievement of your task and you have finally accomplished it. Imagine the effort you have made. See yourself standing at the point of success from where you look back on the work you did to get there. Imagine how your life is different since you started exercising regularly. Visualize the changes that resulted from the accomplishment of this goal. How does it feel to have started a new lifestyle that is good for you? Picture your life as it is now. Concentrate the on the feelings that you have when you do something that is really good for you. Visualize the satisfaction you feel at having achieved your goal. Picture the pride you feel, the confidence you feel in yourself, knowing you were successful. Try to feel the satisfaction you have with this accomplishment. Feel how proud and confident you are. Think about you daily routine. What does your day look like, now that exercising is a firm part of it? Imagine a typical day and see yourself engaging in your new exercise routine. See yourself standing at the point of success. Picture yourself thinking back to when you started working on your goal. How do you feel having successfully accomplished what you wanted? Concentrate on the energy that your healthier lifestyle contributes to your life. How does it feel to have more energy, more confidence, and to know that you successfully started a new healthier lifestyle?" Measures -------- ### Theory of Planned Behavior Variables Measures of the theory of planned behavior constructs were similar to those used in Study 1 with the exception that they made reference to fitness center (gym) attendance. Intention to engage in physical activity was measured using three items (e.g., "I intend to to use the gym 3--4 days a week over the next 6 weeks"). Attitudes toward physical activity were measured using six items (e.g., "For me, using the gym 3--4 days a week over the next 6 weeks is harmful/beneficial"). Subjective norms (e.g., "Most people who are important to me would want me to use the gym 3--4 days a week over the next 6 weeks") and the likelihood the participant would enact the behavior (e.g., "Most people close to me expect me to use the gym 3--4 days a week over the next 6 weeks") were measured using four items. Our measure of perceived behavioral control comprised six items relating to perceived control (e.g., "How much personal control do you have in using the gym 3--4 days a week over the next 6 weeks?") and ability (e.g., "I believe I have the ability to use the gym 3--4 days a week over the next 6 weeks") with respect to engaging in the behavior. All the responses were made on 6-point Likert scales, with the exception of the attitude scale which was measured on 6-point bi-polar semantic differential scales. ### Past Behavior Self-reported past physical activity behavior was measured using a single item ("How often in the past have you used a gym?") with responses provided on a 6-point Likert scale ranging from 1 (never) to 6 (always). This single item measure is derived from [@B21] leisure-time exercise questionnaire and has been used in previous studies ([@B55]). ### Fitness Center Attendance We measured participants' frequency of attendance at the fitness center at a 6-week follow-up time point post-intervention (T2) as well as a 19-week follow-up time point (T3) for the University of Essex participants. Attendees were issued with a unique coded pass card that they were required to swipe on an electronic card reader on each occasion they attended the fitness center. Attendance data for each participant was automatically logged and stored electronically. Participating fitness centers provided the researchers with an anonymised list of participants' attendance records using the separate list of codes and participant names provided by researchers. This ensured that study data was never matched with participant names to retain anonymity. Participants consented to having their fitness center attendance data provided to the research team prior to signing their consent form. Results ------- ### Participants One hundred and eighty-four participants completed the questionnaires at T1 (females = 87, males = 97; *M*~age~ = 26.95 years, *SD* = 9.59). Eight participants (two participants in each group; 7.61% of the initial sample) dropped out of the study at T2 leaving 178 participants (94.57% of response rate) remaining in the study at 6-week follow-up (females = 93, males = 83; *M*~age~ = 27.15 years, *SD* = 9.72). At T3, seventy-eight participants (42.39% of the initial sample) from the sample from the University of Essex campus fitness center were contacted to complete the 19-week follow-up measures (females = 48, males = 20; *M*~age~ = 21.81 years, *SD* = 3.93; **Figure [2](#F2){ref-type="fig"}**). {#F2} . ### Preliminary Analyses #### Missing data Preliminary analyses revealed that 5.44% of the data values were missing at follow-up due to response failures and Little's MCAR test revealed that values were not missing completely at random (χ^2^ = 113.68, df = 21, *p* \< 0.001). Missing data were imputed using multiple imputation using the same parameters as Study 1. #### Attrition checks A χ^2^ analysis revealed that participants who completed the study did not differ significantly from participants who dropped out on gender \[χ^2^(1, *N* = 184) = 0.03, *p* = 0.875, $\eta_{p}^{2}$ = 0.012\]. A MANOVA was used to test the differences on continuous demographic (i.e., age), psychological (theory of planned behavior), and behavioral (physical activity) variables by attrition rate. The MANOVA revealed an overall statistically significant multivariate effect for attrition \[Wilks' Λ = 0.929, *F*(6,177) = 2.24, *p* = 0.042, $\eta_{p}^{2}$ = 0.071\]. Univariate tests showed that participants who completed the study had a higher fitness center attendance at baseline \[*F*(1,182) = 8.14, *p* = 0.005, $\eta_{p}^{2}$ = 0.043\] and a higher perceived behavioral control \[*F*(1,182) = 10.14, *p* = 0.002, $\eta_{p}^{2}$ = 0.053\] compared to participants who dropped out. As a consequence, we controlled for these variables in the main analyses. #### Randomisation checks Sample demographic and psychological data are presented in **Table [3](#T3){ref-type="table"}**. Randomisation checks, conducted using χ^2^ tests and a MANOVA, showed no differences across intervention conditions on gender distribution \[χ^2^(3, *N* = 184) = 3.92, *p* = 0.270, $\eta_{p}^{2}$ = 0.146\] and demographic, psychological and behavioral variables \[Wilks' Λ = 0.92, *F*(18,495) = 0.79, *p* = 0.719, $\eta_{p}^{2}$ = 0.026; **Table [3](#T3){ref-type="table"}**\], for the overall sample. With reference to the first sample, there were also no differences in conditions on gender \[χ^2^(3, *N* = 76) = 1.72, *p* = 0.633, $\eta_{p}^{2}$ = 0.150\] and the demographic, psychological and behavioral variables \[Wilks' Λ = 0.71, *F*(21,190) = 1.15, *p* = 0.300, $\eta_{p}^{2}$ = 0.108\]. In the second sample, no differences in condition were found on the demographic, psychological and behavioral variables \[Wilks' Λ = 0.75, *F*(21,281) = 1.39, *p* = 0.119, $\eta_{p}^{2}$ = 0.090\]. However, there were significantly fewer males in the mental simulation group (10 males) than in the control (15 males), implementation intention (20 males), and mental simulation plus implementation intention groups \[18 males; χ^2^(3, *N* = 108) = 8.37, *p* = 0.039, $\eta_{p}^{2}$ = 0.278\]. These data provided good indication that there were no variations across samples from the two locations due to randomisation, and given the identical study design, we pooled data from both locations. ###### Sample characteristics in Study 2 (*n* = 184). Control group (*n* = 43) Implementation intention group (*n* = 48) Mental simulation group (*n* = 47) Mental simulation and implementation intention group (*n* = 46) ---------------------------------- -------------------------- ------------------------------------------- ------------------------------------ ----------------------------------------------------------------- **Age (years)** Mean (SD) 27.74 (7.88) 27.13 (10.66) 26.26 (9.93) 26.72 (9.77) **Gender** Women 22 (22.7%) 21 (21.6%) 30 (30.9%) 24 (24.7%) Men 21 (24.1%) 27 (31%) 17 (19.5%) 22 (25.3%) **Attitude** Mean (SD) at T1 4.74 (0.74) 4.69 (0.66) 4.68 (0.66) 4.71 (0.81) Mean (SD) at T2 4.78 (0.69) 4.81 (0.52) 4.74 (0.66) 4.77 (0.75) **Subjective norms** Mean (SD) at T1 4.50 (0.82) 4.25 (1.01) 4.32 (0.88) 4.42 (0.79) Mean (SD) at T2 4.35 (0.94) 4.24 (0.87) 4.53 (0.84) 4.45 (0.71) **Intention** Mean (SD) at T1 4.44 (1.12) 4.16 (1.01) 4.34 (0.74) 4.41 (0.95) Mean (SD) at T2 4.58 (1.14) 4.21 (0.98) 4.58 (1.08) 4.34 (0.92) **Perceived behavioral control** Mean (SD) at T1 4.46 (0.79) 4.46 (0.76) 4.41 (0.63) 4.45 (0.72) Mean (SD) at T2 4.51 (0.68) 4.49 (0.77) 4.49 (0.71) 4.52 (0.69) #### Intervention Effects^[3](#fn03){ref-type="fn"}^ A 2 (mental simulation: present vs. absent) × 2 (implementation intention: present vs. absent) ANCOVA was conducted on the dependant variable of fitness center attendance at T2 controlling for fitness center attendance at T1. Estimated marginal means for gym attendance at T2 for the mental simulation and implementation intention (*M* = 10.15, *SE* = 0.44, 95% CI = 9.29, 11.02), mental simulation (*M* = 10.63, *SE* = 0.44, 95% CI = 9.77, 11.49), implementation intention (*M* = 10.49, *SE* = 0.43, 95% CI = 9.63, 11.34), and the control (*M* = 10.44, *SE* = 0.46, 95% CI = 9.54, 11.34) intervention groups suggested few differences (**Table [4](#T4){ref-type="table"}**). This was corroborated by our analyses. We found no statistically significant main effects of mental simulation \[*F*(3,180) = 0.02, *p* = 0.877, $\eta_{p}^{2}$ \< 0.001\] and implementation intention \[*F*(3,180) = 0.24, *p* = 0.626, $\eta_{p}^{2}$ = 0.001\], or a significant time × group interaction effect \[*F*(3,180) = 0.36, *p* = 0.549, $\eta_{p}^{2}$ = 0.002\], on fitness center attendance. We also conducted an analysis for the subsample that completed the 19-week follow-up measure at T3. A 2 × 2 ANCOVA with fitness center attendance at T1 as the covariate revealed no significant main effects of mental simulation \[*F*(3,72) = 0.41, *p* = 0.524, $\eta_{p}^{2}$ = 0.006\] and implementation intention \[*F*(3,72) = 0.03, *p* = 0.868, $\eta_{p}^{2}$ = 0.000\], or a significant time × group interaction effect, on fitness center attendance \[*F*(3,72) = 0.06, *p* = 0.816, $\eta_{p}^{2}$ = 0.001\]. ###### Results of the linear mixedmodel of the effects of the intervention on physical activity in Study 2. Outcome measures Control group (*n* = 43) Mean (*SD*) Implementation intention group (*n* = 48) Mean (*SD*) Mental simulation group (*n* = 47) Mean (*SD*) Mental simulation plus implementation intention group (*n* = 46) Mean (*SD*) *F*~mental\ simulation\ ×\ implementation\ intention~ Effect size (main effect of implementation intention)^c^ Effect size (main effect of mental simulation)^d^ Effect size (interaction effect of mental simulation and implementation intention)^e^ ----------------------------------- -------------------------------------- ------------------------------------------------------- ------------------------------------------------ ------------------------------------------------------------------------------ ------------------------------------------------------- ---------------------------------------------------------- --------------------------------------------------- --------------------------------------------------------------------------------------- ------------- ------ ------------- Fitness center attendance (T1) 7.33 (4.93) 6.23 (4.37) 7.47 (5.01) 7.30 (5.62) Fitness center attendance (T2) 10.77 (7.04) 9.40 (5.91) 11.15 (7.48) 10.46 (7.96) 0.36 0.01 -0.36; 0.39 0.09 -0.29; 0.46 0.08 -0.30; 0.46 Fitness center attendance (T3)^a^ 19.42 (12.18) 16.22 (11.16) 19.45 (11.89) 16.92 (10.70) 0.06 0.10 -0.35; 0.55 0.05 -0.40; 0.50 0.14 -0.38; 0.65 T1, Time 1; T2, Time 2; T3, Time 3. a Sample size for all groups, n = 20. b 95%CI = 95% confidence interval of the effect size. c Effect sizes for the mental simulation and implementation intention group and implementation intention group vs. Mental simulation group and control group comparisons at T2 and T3 controlling for T1. d Effect sizes for the mental simulation and implementation intention group and mental simulation group vs. Implementation intention group and control group comparisons at T2 and T3 controlling for T1. e Effect sizes for the mental simulation and implementation intention group vs. Mental simulation group and implementation intention group and control group comparisons at T2 and T3 controlling for T1. Discussion ---------- The aim of this study was to test the unique and interactive effects of an intervention comprising motivational and volitional components on fitness center attendance. In contrast to Study 1, the intervention adopted mental simulation to target motivation and implementation intention to target volitional components in an appropriately powered, full-factorial, randomized-controlled design. Results revealed no main effects for either of the implementation intention or mental simulation intervention, or their interaction, on fitness center attendance participation at 6 weeks post-intervention. We observed identical results for the 19-week follow-up for the University of Essex fitness center participants. These results provide further evidence against the efficacy of mental simulations and implementation intentions, and a combination of the two, in changing physical activity behavior. Current data were based on a much larger sample size relative to Study 1 and used appropriate means to manipulate the intervention components consistent with the underlying theory and methods adopted in the original mental simulation ([@B54]) and implementation intention ([@B51]; [@B23]) studies. The study also adopted fitness center attendance an objective indicator of the physical activity behavior. In addition, we adopted a factorial design to compare the efficacy of each intervention component separately. General Discussion ================== The aim of the present research was to test the unique and combined effects of two theory-based motivational and volitional components to promote physical activity behavior: mental simulations and implementation intentions. The studies adopted experimental designs to examine the unique and interactive effects of these intervention components on self-reported physical activity (Study 1) and attendance at a fitness center (Study 2). Results of both studies indicated no statistically significant effects of the mental simulation and implementation intention interventions on physical activity behavior. As a consequence, we had to reject our primary hypothesis that the motivational (mental simulation) and volitional (implementation intention) components would lead to increased physical activity behavior relative to each of the components alone. We also rejected our hypothesis that each of the intervention components would have main effects on behavior with greater physical activity participation for participants receiving either of the intervention conditions relative to participants in the control group. We were not able to conduct mediation analyses with the self-efficacy and planning measures as mediating variables as there were no effects on the main outcomes in both studies, contrary to expectations. However, it is important to put these findings into perspective with respect to the limitations of the studies. Study 1, in particular, was conducted on a small sample and effect size statistics indicated medium-sized effects of the interventions, but the study was not sufficiently powered to detect effects of this size. Current findings are inconsistent with trends in the literature if we compare our results with findings from studies that have investigated the unique and combined effects of mental simulation and implementation intention on health behavior. Meta-analyses of interventions examining effects of implementation intention on health behaviors have shown small-to-large effect sizes ([@B25]; [@B2]; [@B5]), but with substantive unresolved heterogeneity and a number of individual studies with similar designs have found null effects ([@B36], [@B37]; [@B66]; [@B35]; [@B61]; [@B39]). Similarly, research examining the effects of mental simulations has largely supported effects of the intervention, again, some studies report null findings. The combination of motivational and volitional components was supposed to facilitate the enactment of the behavior through maximizing motivation and facilitating implementation ([@B48]; [@B28]). In Study 1, it is possible that the intervention effects may have been effective and that the study had been inadequately powered. However, this possibility was not corroborated by the null findings and weak effects in Study 2. It is important to note that intervention studies combining both techniques have not always supported this pattern of effects. For example, [@B30] found a significant effect of implementation intention on alcohol consumption but no effect for mental simulation or interaction effect of mental simulation and implementation intention, while [@B29] demonstrated that mental simulation was significantly more effective for the same behavior with no effect for implementation intentions or the interaction. It is important to note that the methods adopted in both of the current studies included variations to the conventional operationalization of the intervention manipulations. For example, we adopted slightly different versions of the mental simulation and implementation intention exercises across the studies. An 'if-then' format was used in Study 1 while a global format (a 'what, when, and where' plan) was adopted in Study 2. Some researchers ([@B10]; [@B31]) have advocated the use of the 'if-then' format in implementation intention research to make the link between encountered situational cue and the behavior ([@B25]). However, there is, as yet, no definitive consensus on the appropriate format to adopt ([@B32]) and few researchers have compared the moderating effect of plan format (e.g., 'if-then' format vs. the global format) on the effect of implementation intentions on health-related behaviors ([@B10]; [@B52]). Despite these variations, there was no indication of a moderating effect of format in the current studies, as we found consistently null effects regardless of implementation intention format. As a consequence, we do not think that our findings can be dismissed on the basis of methodological variation. Another factor that may have influenced current findings was the operationalization of the mental simulation manipulation. Two different mental simulation manipulations were used across the studies: a process mental simulation manipulation was used in Study 1 while an outcome mental simulation was used in Study 2. One possible explanation of the current findings is that neither of the simulation techniques was effective in promoting greater motivation. In previous studies, process simulations have been generally found to be more effective than the outcome simulations ([@B54]). The greater efficacy of process simulations may have been due to the promoting greater self-efficacy through self-modeling and more positive attitudes toward the behavior through increasing knowledge of the regulation skills required. Outcome mental simulations, on the other hand may reflect 'idealized' outcome states which are 'fantasy'-based and may be disruptive when presented alongside planning techniques like implementation intentions. Outcome simulations, for example, may distract from participants' attention from the goal rather than to the behavior and the appropriate cues. However, while this may explain why the outcome simulations were not effective in the current study, it does not provide an explanation of the poor effectiveness of the process simulation alongside the implementation intentions in Study 1. One possibility is that because the mental simulation techniques, both process and outcome, were self-directed, the imagery did not focus on behaviourally relevant actions and cues that would lead to more effective behavioral engagement. There are other imagery-based techniques such as mental contrasting ([@B68]; [@B1]; [@B50]), which may help to implement the goal by bolstering motivation without distracting the attention to other goals ([@B41]). Mental contrasting focuses on the reinforcement of the volitional phase of the Model of Action Phases and tap into personally relevant barriers rather than focusing on outcomes which may not be realistic or efficacious. Certainly further interventions may consider including 'personally tailored visualization' manipulations that focus the actor's attention on the specific action (i.e., doing physical activity) and relevant cues to action. Limitations and Suggested Solutions ----------------------------------- A key limitation of the current study was the relatively small sample sizes, particularly in Study 1. This would found to be particularly pertinent given that the effect size seemed to indicate possible medium-sized effects. Study 2 adopted a larger sample size and used similar, but not identical, methods but did not find any effects. It is important to note that in Study 2, we based our *a priori* sample size calculation on the implementation intention effect size only ([@B25]), rather than the interaction effect for the mental simulation component or the interaction of this component with implementation intentions. This means that the study was not powered to find the interaction effect. In addition, powering Study 2 to detect a small effect size, such as that found in [@B5] research, would have required a much larger sample. But, powering the study to find such a small effect size would mean to focus on an effect that would be of little practical significance. Interestingly, adhering to such an estimate would mean identification of a sample size that would exceed the largest sample size in the sample of studies included in [@B5] meta-analysis. Meaning that all studies testing for planning intervention effects in physical activity are underpowered. In reality, it seems that the true effect size derived from the current study was even smaller than this estimate. This means that the likelihood of finding a statistically significant effect size would require a very large sample size. This example, therefore, illustrates the imperative of an appropriate sample size, but also illustrates that, in some cases, a very small effect size would make it unfeasible to attain the required sample size. However, such studies would perhaps have limited value because they would be attempting to find an effect which has little or no practical significance. The variability in effect sizes for these interventions is also an issue noted in other studies on implementation intentions. Focusing on implementation intentions, previous meta-analyses in physical activity, and across health behaviors more generally, have reported highly variable effect sizes ranging from small to large ([@B25]; [@B2]; [@B5]). Such variability points to the likely presence of moderator variables. This presents a problem when deciding which effect size should be used in statistical power analyses for new interventions and replications of implementation intention studies. There is, therefore, a real need to adopt standardized designs for the manipulation of intervention techniques like implementation intentions and mental simulations and conduct a systematic evaluation of the effect as a function of candidate moderators ([@B57]; [@B32]). This is consistent with research in other areas of psychology advocating the adoption of standardized designs and large-scale, highly powered replications of intervention effects to allay problems associated with small samples and lack of statistical power ([@B18]; [@B26]). A further limitation of the current research which may have influenced the findings is the definition of the physical activity provided for the participants and failure to prompt a goal intention. No formal definition of the regular physical activity was provided to participants. In Study 1, our self-report measure of physical activity, the IPAQ, provided a description of each level of intensity of physical activity but it was not defined prior to completing the manipulations. Further, participants in Study 1 were not provided with a specific goal intention to participate in more physical activity than they did before the trial. In Study 2, the WHO physical activity recommendations were introduced alongside mental simulation and implementation intention manipulations. Providing a specific definition may have served as a prompt for participants to form a goal to reduce physical activity when responding to the intervention manipulations. Nevertheless, a specific goal intention was not prompted, which may have stymied the effectiveness of the intervention. Although we used a self-report measure of physical activity with good psychometric properties in Study 1 ([@B15]) and fitness center attendance as an objective proxy measure of physical activity in Study 2, both measures have limitations. Even though self-report measures have good concurrent validity with objective measures of physical activity, there is still the possibility of response bias. Fitness center attendance is also problematic as it only accounts for frequency of attendance and not the type, duration or intensity of the activities performed once there. These measures may, therefore, have led to bias in the reporting of behavior. The inclusion of a self-report measure of physical activity in Study 2 alongside the fitness center attendance measures would have provided additional indication of the types of physical activity behavior participants engaged in during their fitness center attendances. Replication of the current paradigm with 'gold standard' objective measures of physical activity like heart rate monitors and accelerometers would be required to unequivocally support the current null findings. Conclusion ========== Current findings do not support the effectiveness of the individual or combined use of mental simulations and implementation intention techniques to promote physical activity behavior. The studies had certain strengths, particularly Study 2 which included a factorial design and efforts to match of manipulations and stimuli with those adopted in previous studies. Even if we tried to closely mirror the suggested features of each intervention component in our manipulations, there are some salient deviations from other implementation intention and mental simulation studies in terms of the format adopted. The relatively small sample sizes, particularly that of Study 1, is a significant limitation. Notwithstanding these limitations, the current findings provide indication that these intervention techniques may not be as effective as expected or reported in previous meta-analyses. However, we cannot, of course, unequivocally rule out the possibility that changes in sample size and format may overturn the null findings, and we therefore look to further replications that refine our methods to provide further replication tests of these effects to confirm or reject our null findings. Author Contributions ==================== CM, AG, BA, OF, and MSH were equally involved in the conception of the studies, the acquisition, analysis and interpretation of data, in drafting the manuscript and the final approval of the version to be published. Conflict of Interest Statement ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. **Funding.** This work was supported by Grenoble Innovation Recherche \[grant number AGI13SHS04\]. MSH's contribution was supported by a Finnish Distinguished Professor Programme (FiDiPro) award from Tekes, the Finnish funding agency for innovation. We would like to thank Caroline Poulet and Mélanie Cerles for having accepted to recruit the students in their classes for Study 1, as well as Rebecca Moult, Jake Brazier, and Paul Farrow who collected the data for Study 2. We adopted a 1-week follow-up in order to conduct a test of the concurrent validity of our self-report measure of physical activity using accelerometers worn by a subsample of the participants. However, we were not able to retrieve the data from the accelerometers due to equipment failure. The majority of participants scored at least 3 (on a 6-point scale) on the manipulation check variables for mental simulation and implementation intention manipulation checks suggesting that the majority had sufficiently engaged in the intervention. However, a small minority reported the lowest score (1) on the mental simulation (*n* = 2) and implementation intention (*n* = 4) manipulation check variables suggested that they had not engaged. However, excluding these participants from the main analyses did not alter the effects with no main effect of groups on physical activity in MET-min per week at T2 and T3, taking into account physical activity at T1. Similar to Study 1, we tested whether the mental simulation manipulation had an effect on the motivation by examining the effects of the intervention effects on intention. A univariate ANCOVA with mental simulation as fixed factor (present vs. absent) was conducted on the dependant variable of intention at T2 controlling for intention at T1. Yet, the analyses revealed no significant time × group interaction effect \[*F*(1,182) = 0.02, *p* = 0.902, $\eta_{p}^{2}$ = 0.000\] on intention. Supplementary Material ====================== The Supplementary Material for this article can be found online at: <http://journal.frontiersin.org/article/10.3389/fpsyg.2016.01789/full#supplementary-material> ###### Click here for additional data file. [^1]: Edited by: *Thomas L. Webb, University of Sheffield, UK* [^2]: Reviewed by: *Aukje Verhoeven, University of Amsterdam, Netherlands; Pier-Eric Chamberland, Université du Québec à Trois-Rivières, Canada; Andrew Prestwich, University of Leeds, UK* [^3]: This article was submitted to Personality and Social Psychology, a section of the journal Frontiers in Psychology

1. Background {#sec1} ============= Volatile oils (VOs) have antioxidant, anti-inflammatory, anticancer, anthelmintic, antimalarial, antiviral, antibacterial, cholesterol inhibition, and insecticide activities \[[@B1], [@B2]\]. The chemical structures of VOs determine their therapeutic activities \[[@B3]\]. Recently, there is great attention towards natural antioxidants from plants. Antioxidants can act as a radical scavenger, promote health, and produce anticancer activity \[[@B4]\]. There is a growing concern about antimicrobial resistance issue \[[@B5]\]. The complications of multidrug resistance enforced scientists to search for new antimicrobial agents from various sources such as medicinal plants \[[@B6]\]. In recent time, there has been considerable interest in VOs and extracts of medicinal plants for the development of alternatives to prevent or to delay the growth of pathogens \[[@B7]\]. Many scientific investigations reported that the chemical composition, total yield, and the aroma of VOs may be different due to growing conditions (climate, type of soil and composition, and altitude), plant age, geoclimatic location, and environmental conditions of collection time and site \[[@B8]\]. *Micromeria fruticosa* subspecies *serpyllifolia* (Lamiaceae), also known as White *Micromeria*, is an aromatic herb \[[@B9]\], dominant in the eastern Mediterranean regions including Palestine, which has a pleasant minty fragrance that in hot summer provides a sensation of coolness \[[@B10], [@B11]\]. In Palestinian society known as Duqat 'Adas, 'Ishbit esh-shai, Qurnya and as Thyme-leave savory in English, the aerial parts of the plant (flower, leaves, and stalk) are used in folk medicine \[[@B12]\]. It has been widely used in traditional medicine for the treatment of hypertension, heart disorders, diarrhea, abdominal pains, colds, headache, wounds, and infections such as skin and eye infections and as an anti-inflammatory agent \[[@B9], [@B11], [@B13]--[@B17]\]. *M. fruticosa serpyllifolia* VOs also exhibit antibacterial, antifungal, antioxidant, insecticide, analgesic, anticonvulsant, and CNS depressant activity \[[@B11], [@B15]\]. The authors of the current study conducted several tests on wild-growing *M. fruticosa serpyllifolia* in three regions in the West Bank area in Palestine. Part of these tests has been published previously \[[@B18]\]. Specifically, structural elucidation and identification of the chemical composition using the GC-MS analysis section represent shared results with this study, therefore elaborating and discussing some aspects linked to the findings of the published manuscript. As a continued step, the aims of this study were to screen antioxidant and antimicrobial activities of *M. fruticosa serpyllifolia* collected from three different regions of West Bank, Palestine. 2. Methods {#sec2} ========== 2.1. Chemicals and Reagents {#sec2.1} --------------------------- Calcium chloride, 2,2-diphenyl-1-picrylhydrazyl (DPPH), and Trolox were purchased from Sigma-Aldrich, Germany. Dimethyl sulfoxide (DMSO) was purchased from CARLO ERBA, France; Nutrient Agar, Mannitol, MacConKey Agar, and Mueller Hinton Broth were purchased from HiMedia Laboratories, Mumbai, India; Sabouraud Dextrose Agar was purchased from Oxoid, UK; 3-\[N-morpholino\] propanesulfonic acid (MOPS) buffer and RPMI-1640 medium (with L-glutamine, without sodium bicarbonate) (developed at Roswell Park Memorial Institute) were purchased from Sigma-Aldrich, UK. Tween 80 (0.05%) was purchased from ACROS Organics, Belgium. Sodium hydroxide and ethyl alcohol 99.9% were purchased from Sun Pharm. Drug Stars, Nablus, Palestine; cefuroxime 250 mg (as axetil) tablet and doxycycline 100 mg (hyclate) tablet (Jerusalem Pharmaceutical Company, Albiereh, Ramallah, Palestine), levofloxacin 500 mg tablet (Birzeit Pharmaceutical Company, Birzeit, Ramallah, Palestine), azithromycin 250 mg capsules (Pharmacare Company, Birzeit, Ramallah, Palestine), tinidazole 500 mg tablet (Jerusalem Pharmaceutical Company, Albiereh, Ramallah, Palestine), and terbinafine hydrochloride 250 mg tablet (Birzeit Pharmaceutical Company, Birzeit, Ramallah, Palestine), all these pharmaceuticals were donations from Military Medical Services Ramallah Palestine. 2.2. Instrumentation {#sec2.2} -------------------- Grinder (Moulinex model, Uno, China) was used to fracture the dried herbs. A balance (Radway ag, AS 220/c/2, Poland) was used to weigh the plant material; ultrasonic-microwave cooperative extractor/reactor (CW-2000, China) was employed for the extraction of volatile oil. A balance (Boeco, 4500 g, Germany), a UV-visible spectrophotometer (Jenway 7315, UK) for the assessment of the antioxidant activity of VOs, a water bath (Memmert, Germany), micropipettes (Finnpipette, Finland), a heater (Lab-Tech, Korea), and a balance (Sartorius AY 303, Canada) were used. 2.3. Plant Material Collection and Extraction Procedures {#sec2.3} -------------------------------------------------------- The aerial parts of *M. fruticosa serpyllifolia* were collected in April of 2017 from three cities in the West Bank (WB) in Palestine: Nablus, Ramallah, and Hebron representing north, middle, and south of the WB in Palestine, respectively. The samples were botanically identified and coded by Dr. Nidal Jaradat, the pharmacognosist at the Department of Pharmacy in Pharmacognosy and Herbal Products Laboratory, An-Najah National University, and the voucher specimen code was Pharm-PCT-1575. The extraction of VOs was performed following the procedure in accordance to \[[@B19], [@B20]\]. 2.4. Antioxidant Activity {#sec2.4} ------------------------- The scavenging activity of *M. fruticosa serpyllifolia* VOs of the three samples from three regions in WB in Palestine was assessed using the methods described in \[[@B20], [@B21]\]. Stock solutions at a concentration of 1 mg/mL in methanol and Trolox were prepared from *M. fruticosa serpyllifolia* VOs that were collected from three Palestinian regions. Each one of these stock solutions was diluted in methanol to prepare 12 of the working solutions with the following concentrations: 1, 2, 3, 5, 7, 10, 20, 30, 40, 50, 80, and 100 *μ*g/mL. A freshly prepared DPPH solution (0.002% *w*/*v*) was mixed with both methanol and with each of the abovementioned working solutions at 1 : 1 : 1 ratio. In addition, a negative control solution was prepared by mixing the mentioned DPPH solution with methanol in a 1 : 1 ratio. All of these solutions were incubated at room temperature in a dark cabinet for 30 min. By the end of the incubation period, the optical density of these solutions was determined using spectrophotometric absorbances at a wavelength of 517 nm using methanol as the blank solution. Antioxidant activity was monitored by measuring the absorbance at 517 nm wavelength. The antioxidant activities of *M. fruticosa serpyllifolia* VOs and Trolox were assessed by their ability to donate a hydrogen atom or electron and were identified from converting the deep violet color of a methanol solution of DPPH to colorless or pale yellow; for that, the inhibition percentage of DPPH activity was used to determine the antioxidant activity of *M. fruticosa serpyllifolia* VOs and Trolox using the following equation (Inhibition% of antioxidant activity \[[@B6]\]): $$\begin{matrix} {\text{In}\% = \frac{A_{\text{blank}} - A_{\text{sample}}}{A_{\text{blank}}} \times 100,} \\ \end{matrix}$$where *A*~blank~ represented the absorption of the control reaction (all reagent without the sample) and *A*~sample~ represented the absorbance of the sample. 2.5. Antimicrobial Screening {#sec2.5} ---------------------------- The antibacterial activities of *M. fruticosa serpyllifolia* VOs were investigated against the growth of nine reference bacterial strains obtained from the American Type Culture Collection (ATCC): *Escherichia coli* (ATCC 25922), *Enterococcus faecium* (ATCC 700221, USA), *Klebsiella pneumoniae* (ATCC 13883, UK), *Pseudomonas aeruginosa* (ATCC 27853, USA), *Shigella sonnei* (ATCC 25931, USA), and *Staphylococcus aureus* (ATCC 25923, USA). In addition, diagnostically proven clinical isolates *Proteus mirabilis*, *Staphylococcus aureus*, and methicillin-resistant *Staphylococcus aureus* (MRSA) were tested. Four antibiotics were used to test the sensitivity of bacterial strains: azithromycin, levofloxacin, cefuroxime, and doxycycline; the antibiotics were dissolved in a suitable solvent according to solubility test to obtain stock solution \[[@B22]--[@B24]\]. The antifungal activity of VOs was examined against the growth of two fungal strains acquired, from the American Type Culture Collection (ATCC), *Candida albicans* (ATCC 90028, USA) and *Epidermophyton floccosum* (ATCC 52066, UK). The antifungal agents (terbinafine and tinidazole) were used for susceptibility tests to *Candida albicans*, used with *M. fruticosa serpyllifolia* VOs \[[@B25]\]. The antibacterial activity of three VO samples was determined using broth microdilution method described in Jaradat et al. \[[@B20]\]. Each one of the isolated *M. fruticosa serpyllifolia* VOs was dissolved in DMSO (100%) at a concentration of 50 mg/mL. The prepared *M. fruticosa serpyllifolia* VO solutions were filter sterilized and then were serially microdiluted (2-fold) eleven times in sterile nutrient broth. In 96-well plates, the dilution processes were carried out under aseptic conditions. In the microwells that were assigned to evaluate the antibacterial activities of the extracted *M. fruticosa serpyllifolia* VOs, the concentration of these oils ranged from 0.024 to 25 mg/mL. On the other hand, the concentrations of these essential oils in the microwells assigned to evaluate their antifungal activities ranged from 8.467∗10^−3^ to 16.666 mg/mL. In these plates, microwell number 11 contained essential oil-free nutrient broth, which was used as a positive control for microbial growth. In addition, the microwell number 12 contained essential oil-free nutrient broth that was left uninoculated with any of the test microbes. This well was used as a negative control for microbial growth. Microwell numbers 1 to 11 were inoculated aseptically with the test microbes. At the time of inoculation, the final concentrations of microbial cells were about 2.5 × 10^5^ and 0.333--1.666 × 10^3^ colony-forming unit (CFU)/mL for the tested bacterial pathogens and *Candida albicans*, respectively. Each of the included microbes in this study was examined in duplicate for being inhibited by the *M. fruticosa serpyllifolia* essential oils. At 35°C, all the inoculated plates were incubated and the incubation period lasted for about 16-20 hours for the plates inoculated with the test bacterial strains and for about 48 hours for the plates inoculated with *Candida albicans*. The lowest concentration of *M. fruticosa serpyllifolia* essential oils, at which there was no visible microbial growth in that microwell, was observed. 2.6. Statistical Analysis {#sec2.6} ------------------------- IC~50~ values of antioxidant activity and I% of DPPH free radical were determined in triplicate for *M. fruticosa serpyllifolia* VOs obtained from three different regions in Palestine. The results were expressed as mean ± standard deviation (SD), and the obtained data were compared using one-way ANOVA with post hoc Tukey-Kramer HSD multiple comparison calculation; *p* values of 0.05 or less were considered statistically significant \[[@B26]\]. 3. Results {#sec3} ========== 3.1. Yields and Chemical Composition {#sec3.1} ------------------------------------ Volatile oils of the three samples of *M. fruticosa serpyllifolia* were extracted using microwave-ultrasonic apparatus; still, the results of this section can be obtained from the previously published work by the same research group \[[@B18]\]. However, these results are shown in [Figure 1](#fig1){ref-type="fig"} modified from \[[@B18]\]. Briefly, [Figure 1](#fig1){ref-type="fig"} represents the chemical composition obtained using GC-MS analysis in which the most abundant components in all three samples were pulegone and isomenthone. The total identified components in the three samples were almost consistent in which 90.48, 94.44, and 93.55% of the constituents were identified in Nablus, Ramallah, and Hebron districts, respectively. 3.2. Antioxidant Activity {#sec3.2} ------------------------- DPPH assay was used as in vitro approach to determine the free radical-scavenging activity and to screen for the possible antioxidant activity of the *M. fruticosa serpyllifolia* VOs from different regions in Palestine. IC~50~ values were used to assess the ability of the examined samples to inhibit DPPH. The assay revealed that the VO samples exhibited higher antioxidant potency compared to Trolox but they showed lower efficacy (maximum inhibition) ([Table 1](#tab1){ref-type="table"} and [Figure 2](#fig2){ref-type="fig"}). Statistical analysis using one-way ANOVA was performed to compare the antioxidant potency (IC~50~) and efficacy among samples. There were significant differences in antioxidant potency and efficacy of VOs compared to Trolox (*p* \< 0.05 or \<0.01). There were significant differences in antioxidant efficacy of VOs compared to each other (*p* \< 0.05 or \<0.01), but there were no significant differences in antioxidant potency of VOs compared to each other (*p* \> 0.05). 3.3. Antimicrobial Activity {#sec3.3} --------------------------- The minimum inhibitory concentrations (MICs) of *M. fruticosa serpyllifolia* VOs from different regions of Palestine were reported in [Table 2](#tab2){ref-type="table"}. The majority of Gram (+) and Gram (-) bacterial strains were sensitive to *M. fruticosa serpyllifolia* VOs at MIC of 3.13 mg/mL. There were no statistically significant differences in activity against nine bacterial strains between *M. fruticosa serpyllifolia* VOs from the three regions in Palestine. There were significant differences of Hebron VO sample compared to Nablus and Ramallah VO samples against the American Type Culture Collection *Shigella sonnei* (ATCC 25931), *p* \< 0.01, and the VO sample from Hebron had the highest potency at a MIC value of 1.56 mg/mL. There were significant differences in Hebron and Ramallah VO samples compared to Nablus VO samples against two clinical isolate of Gram (+) bacterial strains, *Staphylococcus aureus* and MRSA, *p* \< 0.01, the VO sample from Nablus provided the lowest potency at a MIC value of 6.250 mg/mL. To evaluate the sensitivity of bacterial strains, four antibacterial drugs were used: azithromycin 250 mg, levofloxacin 500 mg, doxycycline 100 mg, and cefuroxime 250 mg. The MIC values of the drugs were in the range 1.28∗10^−6^ mg/mL--22.5∗10^−3^ mg/mL; [Table 3](#tab3){ref-type="table"} lists the MICs for drugs. In addition, the antifungal activity against the fungal strains was tested for sensitivity to *M. fruticosa serpyllifolia* VOs; the yeast was the most sensitive followed by the fungus; the American Type Culture Collection *C. albicans* (ATCC 90028) yeast was found to be the most sensitive to *M. fruticosa serpyllifolia* VO samples at a MIC value of 0.206 mg/mL followed by the fungus *Epidermophyton floccosum* (ATCC 52066) at a VO MIC value of 0.78 mg/mL ([Table 2](#tab2){ref-type="table"}). To evaluate the sensitivity of fungal strains, two antifungal dugs were used: terbinafine 250 mg and tinidazole 500 mg, and the MIC value of antifungal drugs was 18.52 *μ*g/mL. 4. Discussion {#sec4} ============= Natural antioxidants have been widely investigated to find protective compounds against damages and diseases developed from free radicals and oxidative stress. *Micromeria* species were identified as a rich source of antioxidant agents \[[@B27]\]. Different results were reported by Güllüce and coauthors in Turkey in which the antioxidant activity of the VOs of *M. fruticosa serpyllifolia* was observed with an IC~50~ value of 98.2 *μ*g/mL \[[@B28]\]. The study showed that VOs abundant with oxygenated monoterpene such as pulegone have antioxidant activity \[[@B29], [@B30]\]. However, this result may support the antioxidant potency of Ramallah sample VOs, which contained the highest amount of total oxygenated compounds and pulegone (89.88 and 86.04%, respectively) among the three samples of VOs. Interestingly, multidrug resistance of bacterial species causes health difficulties. Extracts of volatile oils have been investigated as new potential antimicrobial agents, biopreservative products, and promising antiseptic enhancers for topical uses \[[@B31]\]. *Micromeria* species VOs were considered to have strong broad-spectrum antimicrobial activity \[[@B32]\]. The results listed in [Table 2](#tab2){ref-type="table"} showed that the VOs of the three samples exhibited considerable antifungal potency but little antibacterial potency. The results of the antimicrobial activity of the VOs of three samples revealed that this activity was specific against *Shigella sonnei*, *Staphylococcus aureus* (CI), and MRSA and nonspecific against the rest of the microbial organisms. Screening the potential antimicrobial activity of *M. fruticosa serpyllifolia* VOs and methanolic extract, growing in Turkey conducted by Güllüce et al. \[[@B28]\], concluded that the VO provided stronger antimicrobial properties than methanolic extract (methanolic extract did not show any antimicrobial activity); the MIC values of 0.5 mg/mL volatile oil stock solution for bacterial species which were susceptible to the oil ranged from 31.25 to 125 *μ*g/mL and for fungi which were susceptible to VO ranged from 31.25 to 62.50 *μ*g/mL. A study conducted by Omari et al. \[[@B33]\] evaluating the antifungal activity of *M. barbata* growing in Lebanon, using different fungal strains and yeasts, including *Epidermophyton floccosum* and *Candida albicans*, concluded that the *M. barbata* VOs showed a high fungistatic activity. Investigating the antimicrobial activity of *Micromeria cilicica* VOs growing in Turkey resulted in the finding that the *Micromeria cilicica* VOs and pulegone crude compound (the main component) showed a significant antifungal and antibacterial activity; the activities increased relying on the amount of pulegone and VOs and *Candida albicans* was the most sensitive to pulegone \[[@B34]\]. *Micromeria congesta* VOs were considered as a significant antibacterial due to abundant components such as pulegone and isomenthone \[[@B35]\]. Studying the chemical ingredients and antibacterial and antifungal activity of the volatile oils of four plants including *Mentha spicata* growing in Iran by Kazemi et al. \[[@B36]\], and crude menthone (the dominant component) for antimicrobial activity, reported that VOs showed very strong antimicrobial properties against *Staphylococcus aureus*, all of *Shigella species*, *Escherichia coli*, *Klebsiella sp.*, *Pseudomonas aeruginosa*, *Proteus sp.*, *Candida albicans*, and other strains and concluded that menthone (isomenthone) exhibited strong antibacterial properties with MIC 1.5-3.5 *μ*g/mL. These findings could be linked with our results obtained from Hebron in which the VOs of *M. fruticosa serpyllifolia* owned the highest quantity of isomenthone (14.41%) which is thought to be effective against *Shigella sonnei*. On the contrary, isomenthone represented the lowest amount of constituents (3.16%) in Nablus and therefore exhibited the lowest potency against *Staphylococcus aureus* (CI) and MRSA. 5. Conclusions {#sec5} ============== *M. fruticosa serpyllifolia* VOs from different regions in Palestine represented by three cities showed variable antioxidant and antimicrobial activities depending on the phytochemical constituents of the volatile oils. The sample from middle Palestine (Ramallah) showed the most potent antioxidant properties. The plant extract exhibited strong antifungal activities and minimal antibacterial activities. The sample of the south region showed higher potency against *Shigella sonnei* while the sample of the northern region showed lower potency against *Staphylococcus aureus* (CI) and MRSA. These findings enable *M. fruticosa serpyllifolia* VOs to be good agents in curing or preventing oxidative stress and healing wounds and skin dermatitis and a good food preservative agent. Further *in vivo* studies are needed to evaluate the potential pharmacological activities, to isolate the basic components responsible for potential pharmacological activities, and to evaluate the safety and toxicity of plant extract. The authors would like to thank the Faculty of Graduate Studies-An-Najah National University for donating the chemical agents and helping in providing necessary instruments. CNS: : Central nervous system DMSO: : Dimethyl sulfoxide DPPH: : 2,2-Diphenyl-1-picrylhydrazyl GC-MS: : Gas chromatography mass spectrometry MIC: : Minimum inhibitory concentration MRSA: : Methicillin-resistant *Staphylococcus aureus* VOs: : Volatile oils. Data Availability ================= All raw data are available upon request from the corresponding author. Conflicts of Interest ===================== The authors declare that they have no competing interest. Authors\' Contributions ======================= NS performed the experimental part and drafted the manuscript; NYS designed, supervised, coordinated the study, and drafted the manuscript; NJ designed and conceptualized the study; ME and RA conducted the antimicrobial tests; SK, MA, and LA conducted the GC analysis; AR and MA drafted the manuscript. All authors read and approved the final manuscript. Supplementary Materials {#supplementary-material-1} ======================= ###### Supplementary data 1: GC-MS chromatograms of volatile oil samples of *M. fruticosa serpyllifolia* collected from different regions. Supplementary data 2: GC-MS analysis and component identification and quantification of volatile oil samples. Supplementary data 3: detailed procedures for antibacterial agent preparation. Supplementary data 4: detailed protocol for antifungal agent preparation. Supplementary data 5: antimicrobial effect of *M. fruticosa serpyllifolia* volatile oils collected from different regions using the agar dilution method. ###### Click here for additional data file. {#fig1} {#fig2} ###### IC~50~ of DPPH radical scavenging activity of *M. fruticosa serpyllifolia* VOs from different regions of Palestine and Trolox. Trolox Nablus Ramallah Hebron ------------------------------------------ -------------- ----------------- ------------------ ------------------- IC~50~ (*μ*g/mL) 0.64 ± 0.12 0.47 ± 0.02^a^ 0.45 ± 0.01^a^ 0.47 ± 0.01^d^ Max. I% DPPH radical scavenging activity 96.80 ± 2.83 49.25 ± 0.33^d^ 50.19 ± 0.65^bd^ 45.01 ± 0.86^cde^ ^a^ *p* \< 0.05 compared to Trolox, ^b^*p* \< 0.05 compared to Nablus, ^c^*p* \< 0.01 compared to Ramallah, ^d^*p* \< 0.01 compared to Trolox, and ^e^*p* \< 0.01 compared to Nablus; ^∗^mean ± SD, *n* = 3. ###### Antimicrobial activity (MIC in mg/mL) of *M. fruticosa serpyllifolia* VOs from different regions of Palestine based on the broth microdilution method and agar dilution method. -------------------------------------------------------------------------------------- MIC\ MIC\ MIC\ DMSO 100% Nablus Ramallah Hebron ----------------------------------------- -------- ---------- ------------ ----------- Yeast *C. albicans* (ATCC 90028) 0.206 0.206 0.206 3.70% Fungus *Epidermophyton floccosum* (ATCC 52066) 0.781 0.781 0.781 6.25% Bacterial strains *Staphylococcus aureus* (ATCC 25923) 3.125 3.125 3.125 12.50% *Staphylococcus aureus* (CI) 6.250 3.125^a^ 3.125^a^ 12.50% MRSA (CI) 6.250 3.125^a^ 3.125^a^ 12.50% *Enterococcus faecium* (ATCC 700221) 3.125 3.125 3.125 6.25% *Escherichia coli* (ATCC 25922) 3.125 3.125 3.125 12.50% *Pseudomonas aeruginosa* (ATCC 27853) 3.125 3.125 3.125 12.50% *Shigella sonnei* (ATCC 25931) 3.125 3.125 1.5625^ab^ 12.50% *Proteus mirabilis* (CI) 3.125 3.125 3.125 12.50% *Klebsiella pneumoniae* (ATCC 13883) 3.125 3.125 3.125 12.50% -------------------------------------------------------------------------------------- ^a^ *p* \< 0.01 compared to Nablus, ^b^*p* \< 0.01 compared to Ramallah. ###### Minimum inhibitory concentration (*μ*g/mL) of some antimicrobial drugs. Azithromycin Levofloxacin Doxycycline Cefuroxime --------------------------------------- -------------- -------------- ------------- ------------ Bacterial strains *Staphylococcus aureus* (ATCC 25923) 0.352 5.125∗10^−3^ 0.012 2.356 *Staphylococcus aureus* (CI) 0.352 6.4∗10^−3^ 0.097 4.713 MRSA (CI) 0.176 6.4∗10^−3^ 0.097 4.713 *Proteus mirabilis* (CI) 5.625 1.28∗10^−3^ 0.387 4.713 *Pseudomonas aeruginosa* (ATCC 27853) 0.703 1.28∗10^−3^ 0.387 2.356 *Escherichia coli* (ATCC 25922) 0.703 1.28∗10^−3^ 0.012 2.356 *Klebsiella pneumoniae* (ATCC 13883) 1.406 0.012 0.387 4.713 *Shigella sonnei* (ATCC 25931) 0.703 --- 0.387 2.356 *Enterococcus faecium* (ATCC 700221) 22.5 1.64 0.0.097 4.713 Terbinafine Tinidazole *Candida albicans* (ATCC 90028) 18.5185 --- [^1]: Guest Editor: Marta Laranjo

INTRODUCTION {#s1} ============ The presentation of two simultaneous primary tumours is rare \[[@RJU066C1]\]. In the paper we report on an 82-year-old female who was diagnosed with primary breast and bladder cancer. We review the literature on multiple primary cancers (MPC) and discuss the diagnostic and treatment challenges of metachronous and synchronous breast and bladder cancer. CASE REPORT {#s2} =========== An 82-year-old female with known metastatic lobular breast cancer presented to a urology outpatient clinic with an incidental finding of a bladder mass at the left vesicoureteric junction (VUJ, Fig. [1](#RJU066F1){ref-type="fig"}). The bladder lesion had originally been identified on a surveillance computed tomography scan and was associated with left-sided hydronephrosis (Fig. [1](#RJU066F1){ref-type="fig"}). Figure 1:TCC at base of bladder. Symptomatically the patient described a recent history of dysuria which had been treated unsuccessfully by the General Practitioner with a 3-day course of nitrofurantoin. There was no history of macroscopic haematuria. Urine dip at the time of clinic was positive for leucocytes and nitrates and further culture identified a significant growth of *Staphylococcus aureus* sensitive to penicillin. Renal function confirmed previous chronic kidney disease attributable to hypertension. A previous history of smoking that ceased 40 years ago was the only environmental risk factor identified. Following discussion at a urological multi-disciplinary team (MDT) meeting, flexible cystoscopy was performed. This confirmed a 4 cm solid lesion on the left side of the bladder base with surface necrosis. Neither ureteric orifice could be identified. Two sub-centimetre satellite lesions were also noted on the right lateral wall of the bladder. Suspecting either primary urothelial cancer or adenocarcinomatous metastatic breast deposits a rigid cystoscopy, transurethral resection of bladder tumour and biopsies were scheduled. Rigid cystoscopy identified bullous oedema surrounding the 4--5 cm solid tumour that covered the entire bladder base. It had typical endoscopic appearance of a transitional cell carcinoma (TCC). Only after complete resection were the ureteric orifices identified with both showing good signs of peristalsis and efflux negating the need for stenting. Loop biopsies were taken to completely resect the satellite lesions on the right wall which had appearances atypical of a TCC and tissue was sent separately for histological diagnosis (Fig. [2](#RJU066F2){ref-type="fig"}). A 22 Fr catheter remained *in situ* until post-operative renal function stabilized and the patient was discharged awaiting histological diagnosis. Figure 2:Metastatic lobular carcinoma of the breast identified on right wall of bladder. The histopathological analysis identified two distinct malignant entities. The solid tumour at the base of bladder was confirmed as muscle invasive bladder cancer (G3pT2) whilst the satellite lesion showed features suggestive of metastatic lobular carcinoma of the breast. Following discussion at both urological and breast MDTs palliative chemotherapy was administered to address the metastatic breast cancer whilst palliative radiotherapy was used to contain the primary bladder cancer. DISCUSSION {#s3} ========== Whilst the incidence of patients developing MPC is increasing, the presentation of synchronous primary tumours remains rare \[[@RJU066C1], [@RJU066C2]\]. We have found only one other study that has documented a simultaneous presentation of primary urological and breast cancer \[[@RJU066C1]\] whilst a few others have described a metachronous association \[[@RJU066C2], [@RJU066C3]\]. Genetic pre-disposition through the identification of genes common to different organ tumours has been postulated as a cause for MPC \[[@RJU066C1], [@RJU066C4]\]. *UROC28* is a gene that has been found in bladder, breast and prostate tumours leading to the theory that having a primary tumour might be a risk factor for developing another one, either synchronously or metachronously \[[@RJU066C1], [@RJU066C4]\]. Other theories include accumulation of free radicals and subsequent mistakes in DNA replication and reduced function of lipid laden macrophages leading to impaired host immune surveillance \[[@RJU066C1]\]. Environmental and demographic risk factors must also be considered \[[@RJU066C1]\]. Patients suffering specifically from primary breast cancer are at an increased risk of a secondary tumour in the female genitals and endocrine systems \[[@RJU066C2]\]. Obesity, tobacco and young age were all postulated as potential shared risk factors for this association \[[@RJU066C2]\]. Similarly, primary bladder cancer is associated with suffering subsequent malignancy in the genitals and respiratory system \[[@RJU066C2]\]. Tobacco is a shared risk factor for both bladder and lung cancer whilst topographic proximity could potentially explain the relationship between bladder cancer and subsequent genital malignancy \[[@RJU066C2]\]. There was no familial history of breast or urothelial cancer and our patient was not clinically obese. Often secondary and potentially simultaneous malignancies can be attributed to the first tumour treatment \[[@RJU066C1], [@RJU066C2]\]; however, there is no reported evidence to suggest that primary endocrine therapy for lobular breast cancer increases the risk of TCC of the bladder. The increased detection of MPC and potentially of synchronous tumours can be attributed to increased use of more advanced diagnostic technique, in particular radiology, the introduction of screening practices and more rigorous oncological surveillance and peri-operative evaluation \[[@RJU066C1], [@RJU066C2]\]. There is no consensus on treatment recommendations for synchronous tumours. The opinion of most urologists in a questionnaire-based survey was that they would prefer to perform the tumour resections simultaneously if logistically and topographically feasible \[[@RJU066C1]\]. The only caveat was if one tumour was more aggressive than the other resection of the more aggressive tumour was prioritized \[[@RJU066C1]\]. Modification of behavioural risk factors and necessary lifestyle changes as a primary prevention to MPC has also been supported \[[@RJU066C2]\]. The index of suspicion for potential simultaneous primary malignancies must be heightened in clinicians. This is displayed in our case report where lesions on the base of the bladder and right lateral wall were not assumed to be of the same histological subtype and separate biopsies were taken and submitted for histopathological diagnosis. Smoking was the only causative risk factor we could identify to explain the synchronous bladder and breast tumours. Other potential explanations for the development and identification of synchronous tumours include increased life expectancy, improved screening practices, increased exposure to environmental causative agents and genetic pre-disposition. Treatment of synchronous tumours is dependent on the tumour\'s histological sub-type, location and progression. The surgical preference is for resection of both tumours simultaneously whilst modification of behavioural risk factors may prevent the occurrence of MPC.

INTRODUCTION {#sec1-1} ============ Topical timolol maleate is a nonselective beta-adrenergic receptor blocker that is commonly used for treatment of open-angle glaucoma. It has been frequently used as first-line therapy for reduction of associated ocular hypertension. While systemic concentration after administration of topical beta-blockers is low in comparison to that achieved with oral beta-blockers, it is well known that topical therapy can induce cardiovascular, respiratory, central nervous system, and metabolic side effects.\[[@ref1]\] We report an interesting case of a sinus bradycardia with intermittent atrioventricular block induced by topical timolol maleate. CASE REPORT {#sec1-2} =========== A 70-year-old black woman with a history of dyslipidemia, diabetes, hypertension, obesity, depression, and primary open-angle glaucoma presented to her cardiology appointment complaining of intermittent palpitations and dyspnea. At the time of examination at 9:00 AM, she was found to be bradycardic, with a normal sinus rhythm of 44 beats per minute; her cardiopulmonary examination was otherwise unremarkable. She had no history of structural or acquired cardiac disease. Her medications included hydrochlorothiazide, lisinopril, simvastatin, bupoprion, topical timolol, and topical brimonidine. She was admitted to the telemetry unit for further monitoring and evaluation of bradycardia. A serum electrolyte panel, complete blood count, and cardiac marker panel were drawn and all were within normal limits. Initial management consisted of inpatient telemetry monitoring with collection of serial cardiac enzymes and potential cardiac catheterization the following day. While on telemetry, she returned to normal sinus rate and rhythm until 9:45 PM, when her heart rate dropped to 41 beats per minute. An EKG at that time revealed atrioventricular block \[[Figure 1](#F1){ref-type="fig"}\]. She was asymptomatic and after approximately 14 minutes spontaneously converted back to a regular rate and rhythm. The next morning at 8:50 AM, the same phenomenon occurred, where she transiently went into an episode of asymptomatic atrioventricular heart block with a rate of 44 beats per minute \[[Figure 2](#F2){ref-type="fig"}\]. Upon questioning we discovered that she had instilled her topical timolol maleate approximately 30 minutes prior to each of these episodes. Topical timolol was discontinued and the conduction abnormality resolved. She was diagnosed as having intermittent sinus bradycardia with intermittent atrioventricular block, likely induced by topical beta-blocker therapy. Subsequently, topical timolol was substituted with topical dorzolamide and a permanent pacemaker was placed. {#F1} {#F2} DISCUSSION {#sec1-3} ========== Topical timolol maleate has long been known to be effective in the treatment of ocular hypertension.\[[@ref2]\] It typically lacks direct myocardial depressant activity.\[[@ref3]\] Timolol maleate like most topical ophthalmic agents is absorbed into the conjunctival, nasal, oropharyngeal, and gastrointestinal mucosal capillaries.\[[@ref1]\] After administration, onset of action is in 20 minutes, with peak effect in 4 hours; the total effect usually lasts 24 hours.\[[@ref3]\] Peak plasma levels following topical administration have been shown to vary from undetectable to 9.6 ng/ml, but on average is accepted to be approximately 1 ng/ml. Although this is significantly lower than plasma concentrations following oral administration, it is sufficient to induce some degree of systemic beta-adrenergic blockade.\[[@ref1]\] With regard to cardiac side effects, topical timolol has been shown to induce a decrease in heart rate and diastolic blood pressure.\[[@ref4]\] This effect has been shown to be synergistic with oral beta-blockers. A recent randomized controlled trial showed that in the subset of glaucoma patients not using either topical or oral beta-blockers, the mean resting pulse was 76 beats per minute (bpm); in comparison, patients using topical beta-blockers had a resting pulse of 70.3 bpm, patients on oral beta-blockers had a resting pulse of 64.7 bpm, and patients using both topical and oral beta-blockers had a resting pulse rate of 58 bpm.\[[@ref5]\] Our patient was not using oral beta-blockers. However, this information is important in the management of patients with both ocular and systemic hypertension. Numerous reports exist of various cardiac and respiratory disturbances occurring as a result of topical timolol. Third-degree atrioventricular block has been reported as a consequence of topical administration.\[[@ref6]\] However, our case report is the first to describe sinus bradycardia with intermittent atrioventricular block. CONCLUSIONS {#sec1-4} =========== Topical timolol maleate is an effective treatment for ocular hypertension, acting by reducing aqueous fluid production. However, it can induce systemic side effects and should be used with caution in patients with, or predisposed to, cardiac or respiratory depression. More recently, topical prostaglandin E2 has emerged as a first-line therapy in glaucoma owing to its effect of increasing aqueous outflow while avoiding the systemic effects of topical beta-blockade.\[[@ref2]\] Additionally, a variety of medications and therapeutic options exist for both systemic hypertension and ocular hypertension associated with open-angle glaucoma. The clinician treating patients with coexisting ocular and systemic hypertension should be aware of these options, especially since topical beta-blockers can induce severe systemic effects in patients without known cardiac or pulmonary dysfunction. **Source of Support:** Nil **Conflict of Interest:** None declared.

Small organic and metal-containing molecules (MW \< 1000) can catalyse synthetically useful reactions with the high levels of stereo-selectivity typically associated with macromolecular enzymatic catalysts. Whereas enzymes are generally understood to accelerate reactions and impart selectivity by *stabilising* specific transition structures through networks of cooperative interactions, enantioselectivity with chiral, small molecule catalysts is typically rationalised by the steric *destabilisation* of all but one dominant pathway. However, it is increasingly apparent that stabilising effects play an important role in small-molecule catalysis as well, although the mechanistic characterisation of such systems is rare. Here it is shown that arylpyrrolidino amido thiourea catalysts catalyse the enantioselective nucleophilic ring opening of episulfonium ions by indoles. Evidence is provided for selective transition state stabilisation of the major pathway by the thiourea catalyst in the rate- and selectivity-determining step. Enantioselectivity is achieved through a network of attractive anion binding, cation-π, and hydrogen bonding interactions between the catalyst and the reacting components in the transition structure assembly. Multi-functional urea and thiourea derivatives have been shown to promote enantioselective reactions of cationic species in non-polar media by binding the corresponding counteranion through hydrogen bonding,^[@R1],[@R2],[@R3],[@R4]^ with selectivity imparted through a combination of electrostatic association and additional noncovalent interactions that differentiate the diastereomeric transition structures leading to the product enantiomers.^[@R5],[@R6],[@R7]^ We sought to extend the anion-binding catalysis concept to episulfonium ions, highly reactive electrophilic species that readily undergo diastereospecific bond-forming reactions with nucleophiles.^[@R8],[@R9],[@R10],[@R11]^ Recently, Toste and coworkers demonstrated enantioselective catalysis of the addition of alcohols to episulfonium ions using a chiral phosphoric acid catalyst.^[@R12]^ On the basis of our previous work in anion-binding catalysis, we envisioned that a (thio)urea could serve as a suitable host for an in situ formed episulfonium ion through interactions with the counteranion ([Figure 1](#F1){ref-type="fig"}). High enantioselectivity might be achieved if additional interactions between the catalyst and episulfonium intermediate could be incorporated to differentially stabilise the diastereomeric pathways. This hypothesis was investigated in the context of a Friedel--Crafts-type indole alkylation reaction.^[@R13]^ Results and Discussion {#S1} ====================== A. Reaction Methodology Development {#S2} ----------------------------------- Preliminary efforts to identify a suitable episulfonium ion precursor revealed that a relatively non-nucleophilic leaving group was required in order to achieve the desired reactivity. Ultimately, we found that stable trichloroacetimidates (TCA) of type **1a**^[@R12]^ were particularly useful substrates (eq. 1), undergoing protonolysis and substitution with a variety of strong Brønsted acids to form a *meso*--episulfonium ion with a counteranion that is readily varied based on the identity of the acid employed. A wide variety of chiral urea and thiourea derivatives was evaluated in the model reaction ([Table 1](#T1){ref-type="table"}). Only arylpyrrolidine-derived thioureas of type **3** were found to induce reactivity above the background rate of **1a** and acid alone. A broad screen of Brønsted acids revealed a pronounced counterion effect. In conjunction with thiourea **3b**, mineral acids with a nucleophilic counter-anion, such as HCl, produced only trace amount of the desired indole addition product **2a** (entry 1), with the corresponding chloride addition product predominating. In contrast, sulfonic acid co-catalysts afforded **2a** in useful yields and varying levels of enantioselectivity, with 4-nitrobenzenesulfonic acid (4-NBSA) providing the most promising result (entry 5). The identity of the aromatic substituent on the pyrrolidino amide portion of the catalyst was also found to exert a profound effect on reaction enantioselectivity (entries 5, 7-12). Catalyst **3a**, lacking an aryl group, induced little rate acceleration above the background reaction (entry 6) and afforded nearly racemic product. In contrast, catalyst **3c-3g** bearing more extended aromatic substituents proved more enantioselective than catalyst **3b**. Correlation between ee and either the electronic properties of the aryl substituents or the polarisability of the aromatic ring has been noted in other reactions using this family of catalysts.^[@R2],[@R6],[@R14]^ However, in the present case no such straightforward relationship was observed. Instead, ee was observed to improve upon expanding the aryl group from phenyl to phenanthryl (entries 7-10), and then to descrease slightly with more expansive aryl substituents (entries 11-12). Urea catalyst **4e** induced only marginally lower ee than its thiourea counterpart (entries 10 vs 13), indicating that any mechanism wherein the thiourea sulfur is engaged productively as a Lewis base catalyst is not operative.^[@R15],[@R16],[@R17],[@R18]^ A variety of substrate combinations was evaluated in order to define the scope as well as gain insight into the mechanism of the reaction ([Table 2](#T2){ref-type="table"}). Substrates bearing electronically and sterically diverse sulfur substituents (R^2^) underwent enantioselective reactions (entries 1-9), with *S*-benzyl-substituted derivatives affording highest ee's (entries 1 & 7). Electron potential maps calculated using DFT showed that the benzylic protons in the *S*-benzyl episulfonium ions bear a substantial amount of partial positive charge, and this may serve to enhance attractive interactions between the cationic intermediate and an electronegative functionality on the catalyst (*vide infra*). Various indole derivatives bearing electron-donating and -withdrawing substituents at the 2-, 4-, 5- or 6-positions all underwent the addition reaction with high levels of enantioselectivity (entries 10-16). In sharp contrast, *N*-methyl indole provided the desired product **2p** with moderate yield and in almost racemic form (entry 17). This suggested that the indole N-H motif may be involved in a key interaction during the ee-determining transition state (*vide infra*). Benzotriazole also underwent the addition reaction, forming a C-N bond with synthetically useful ee (entry 18). Less nucleophilic heterocycles (i.e., π-nucleophiles with Mayr nucleophilicity parameters *N* \< 4)^[@R19]^ proved unreactive. Variation of the substituents on the carbon backbone of the electrophile revealed that aryl groups with *meta*- and *ortho*- functionalities were compatible (entries 19-22); substrates bearing a *para*-substituent, regardless of its steric and electronic properties, resulted in substantially lower enantioselectivity (entries 23-26). Ongoing computational studies suggested that in the transition state leading to the major enantiomeric product, one of the *para*-C--H bonds is engaged in an attractive, electrostatic interaction with the thiourea-bound sulfonate. We reasoned that the lack of this interaction in the cases with *para*-substituted substrates might result in a less well-organized transition structure and reduced selectivity. Finally, three different acetimidate leaving groups displayed essentially the same reactivity and enantioselectivity, suggesting that the leaving group is not directly involved in the ee-determining step (entries 28-30). B. Determination of the Rate- and Enantio- Determining Step {#S3} ----------------------------------------------------------- Thiourea derivatives of type **3** bearing specific arylpyrrolidino residues have been identified as highly enantioselective catalysts for nucleophilic additions to a remarkable variety of cationic electrophilic intermediates, including oxocarbenium ions,^[@R2]^ acyliminium ions,^[@R6]^ acylpyridinium ions,^[@R20]^ and now episulfonium ions. Elucidation of the mechanisms that underlie catalytic activity and stereoinduction by these thioureas may potentially serve as a foundation for new reaction discovery and provide broader insights into cooperative non-covalent pathways. We therefore undertook a detailed experimental and computational investigation of the indole addition to episulfonium ions promoted by **3e** and related catalysts. On the basis of the qualitative observations described above and previous studies involving thiourea anion-binding pathways, a simple catalytic cycle for the reaction of indoles with **1** mediated by **3e** can be advanced ([Figure 2](#F2){ref-type="fig"}). The absence of dependence of enantioselectivity on the identity of the acetimidate group suggests that the reaction begins with protonation of the trichloroacetimidate substrate, followed by episulfonium ion formation in a step that may or may not be under the influence of the thiourea catalyst. A series of ^1^H NMR studies of pre-formed episulfonium ion derivatives^[@R17]^ revealed that the episulfonium sulfonate most likely exists as a covalent adduct both in the presence or absence of the thiourea catalyst, so an endothermic ionization to the episulfonium ion complex is presumably taking place before addition of the indole nucleophile. Finally re-aromatisation to form the product and regenerate the acid closes the catalytic cycle. In the thiourea-promoted pathway, the chiral catalyst must therefore induce enantioselectivity through association with the charged intermediates and transition structures, indicated in [Figure 2](#F2){ref-type="fig"} as taking place via direct binding to the sulfonate counterion. In order to identify the rate- and enantioselectivity-determining step in the catalytic mechanism, reaction progress kinetic analysis^[@R21]^ of the reaction employing substrate **1a**, indole, thiourea **3e**, and 4-NBSA in toluene at 0 °C was conducted with in situ IR spectroscopy.^[@R22],[@R23]^ Using reaction rates measured over 10% to 60% conversion with "different excess" experiments,^[@R22]^ the empirical rate equations were determined for both the racemic reaction catalysed by only 4-NBSA ([equation 3](#FD1){ref-type="disp-formula"}) and the asymmetric reaction co-catalysed by 4-NBSA and thiourea ([equation 4](#FD2){ref-type="disp-formula"}). $$r_{rac} = d\lbrack 1\mathbf{a}\rbrack/dt = k_{rac}{\lbrack 4­{NBSA}\rbrack}_{T}\lbrack{indole}\rbrack$$ $$r_{asym} = d\lbrack 1\mathbf{a}\rbrack/dt = k_{asym}{\lbrack 4­{NBSA}\rbrack}_{T}{\lbrack 3\mathbf{e}\rbrack}_{T}\lbrack{indole}\rbrack$$ Here, *k*~rac~ is the second-order rate constant for the racemic reaction, and *k*~asym~ is the third-order rate constant for the asymmetric reaction, and \[4-NBSA\]~T~ and \[**3e**\]~T~ are the total initial concentrations of acid and thiourea, respectively. The rate of the asymmetric reaction was accelerated by the chiral thiourea catalyst relative to the reaction catalysed by 4-NBSA alone. For instance, at 273 K, \[**3e**\]~T~ = 50 mmol/L, *r*~asym~/*r*~rac~ at 10% conversion of **1a** is 43.4 ± 5.0. This rate acceleration corresponds to a lowering of the free energy of activation of the reaction by **3e** by 2.0 ± 0.1 kcal/mol. The 0^th^-order rate dependence on **1a** and first-order rate dependence on 4-NBSA indicate that quantitative protonation of the substrate occurs under the reaction conditions prior to the rate-determining step (p*K*~a\ (4-NBSA)~ \~ −7, p*K*~a\ (**1a**)~ \~ 2).^[@R24]^ In support of this conclusion, treatment of substrate **1a** with 1 equiv of 4-NBSA resulted in instantaneous, complete consumption of **1a** and concomitant, quantitative formation of trichloroacetamide (the by-product generated during episulfonium ion formation) as determined by in-situ IR spectroscopy. The first-order dependence on indole in both the reaction catalysed by 4-NBSA alone or by 4-NBSA/**3e** reveals that indole is present in the rate-determining transition structure and that episulfonium•4-nitrobenzenesulfonate (existing predominantly as the covalent adduct) is the resting state of the substrate in the reaction. The kinetic data are consistent with the reaction with indole involving either addition or reversible addition followed by slow re-aromatisation as the rate-determining step ([Figure 2](#F2){ref-type="fig"}). To distinguish between these two possibilities, linear free energy relationship and kinetic isotope effect studies were carried out. In the racemic reaction catalysed by 4-NBSA alone, a linear correlation was observed between reaction rate and Mayr's nucleophilicity parameter *N* for five different 5-substituted indoles (log(*k*~rac~) vs. *s*~N~*N*, R^2^ = 0.997), consistent with the indole addition step resulting in nucleophilic ring opening of episulfonium ion being the rate-limiting step.^[@R25]^ A correlation between indole nucleophilicity and rate was also obtained in the thiourea-catalysed reaction (log(*k*~asym~) vs. *s*~N~*N*, R^2^ = 0.757). As discussed below, a strict linear correlation is not observed in this case because the Brønsted acidity of the indole N-H group also influences reaction rate in the thiourea-catalysed reaction (see Section C2). Evaluation of 3-deuterioindole in the thiourea-catalysed addition reaction revealed a very small effect of isotopic substitution (*k*~H~/*k*~D~ = 0.93 ± 0.12). This rules out re-aromatisation as the rate-determining step, which would be expected to display a significant primary isotope effect (*k*~H~/*k*~D~ \> 2.5),^[@R26]^ and is fully consistent with rate-determining indole addition. It can be concluded that indole addition is enantio-determining as well, since the product's stereogenic centers are generated in this rate-determining step.^[@R27]^ C. Elucidation of the Catalyst--Substrate Interactions in the Enantio-Determining Step {#S4} -------------------------------------------------------------------------------------- While the kinetic studies served to define the stoichiometry of the transition structure in the rate- and ee-determining addition of indoles to **1**, they do not provide any direct insight into the specific manner by which the thiourea catalyst induces rate acceleration and enantioselectivity. This proved attainable, however, through the series of structure-reactivity and structure-enantioselectivity studies detailed below. ### C.1 Thiourea Dual Hydrogen Bond Donation to the Sulfonate Anion {#S5} Anion-binding catalysis has been recognised as the primary mode of substrate activation in a variety of asymmetric reactions involving H-bond donors.^[@R1]^ In studies relevant to the system described here, sulfonate ion association to urea or thiourea derivatives via dual hydrogen bond donation interactions has been identified in both binding and reactivity studies.^[@R28],[@R29]^ To test whether such sulfonate binding is operative in the current reaction, model binding studies and catalyst structure--activity studies were conducted. Titration of solutions of thiourea **3e** in *d*~8~-toluene with dibenzylmethylsulfonium triflate \[(Bn~2~MeS)^+^(OTF)^−^\] (**5**) revealed formation of a 1:1 complex as determined by ^1^H NMR. Resonances assigned to each of the thiourea N-H protons sharpened and shifted downfield upon complexation, consistent with a dual hydrogen bonding interaction.^[@R29],[@R30]^ Perturbations to the chiral catalyst that diminished its ability to act as an H-bond donor, either by reduction of its acidity through substituent effects or by excision of one of the donor groups, led to strong or complete decreases in reactivity and enantioselectivity in the indole addition reaction (see [Supplementary Information](#SD2){ref-type="supplementary-material"}, Section 4). These data, taken together with the strong ee-dependence on the sulfonate counterion of the Brønsted acid ([Table 1](#T1){ref-type="table"}) suggests strongly that the thiourea--sulfonate interaction is a key element in the mechanism for catalysis and stereoinduction. ### C.2 Hydrogen Bonding Interaction with the Indole N-H {#S6} As noted above, a striking difference in enantioselectivity was observed between *N*-H indole and *N*-methyl indole analogs in the addition reaction (93% vs 3% ee using catalyst **3e** in additions to **1a**). This suggests an important organisation role of the indole N-H in the stereoinduction mechanism, likely through hydrogen bonding to a Lewis basic functionality on the catalyst. To define the catalyst-indole interactions, a structure-reactivity and structure-enantioselectivity relationship study was undertaken with a series of π-nucleophiles ([Figure 3a](#F3){ref-type="fig"}). In general, the absence of an N-H motif in a (1,3) relationship with the reactive nucleophilic site leads to very low levels of rate acceleration and enantioselectivity (entries 1-2 vs. entries 3-5). To further probe the role of H-bonding interactions between indole nucleophile and catalyst, various 5-substituted indoles were evaluated in competition experiments under both racemic and thiourea-catalysed reaction conditions. The degree of rate acceleration induced by thiourea **3e** was found to be linked directly to the acidity of the indoles, and a linear correlation between log(*k*~asym~/*k*~rac~) and the p*K*~a~ of indoles was observed ([Figure 3b](#F3){ref-type="fig"}). Therefore, in the presence of thiourea, the rate of the reaction is correlated not only to the intrinsic nucleophilicity of the indole, but also to its H-bond donor properties. No correlation was observed between reaction enantioselectivity and the acidity of the indoles, however, indicating that this interaction is likely to exist to a similar degree in both the major and the minor pathways. The kinetic data are thus consistent with general base activation of indole through an indole N-H--catalyst hydrogen bonding interaction in the rate-determining addition to episulfonium ions.^[@R31],[@R32],[@R33],[@R34]^ Catalyst **3e** possesses few Lewis basic functionalities -- namely the thiourea, the amide, and perhaps the extended arene substituent -- and therefore the possible catalyst H-bond acceptor sites are limited in number. The similar reactivity and enantioselectivity displayed with urea **4e** and thiourea **3e** would appear to rule out a direct role of the thiourea sulfur atom. As discussed below, the extended aromatic plays a key role that can be tied to interactions with the episulfonium ion. By this simple process of elimination, we therefore propose that the amide oxygen is the most likely H-bond acceptor site for activation of the indole.^[@R35],[@R36]^ Consistent with this hypothesis, catalyst **3a** was found to be more reactive than the corresponding thioamide analog in the model reaction (eq 2). In addition, the reaction catalyzed by **3a** is ca. 4 times as fast as the reaction with Schreiner's thiourea \[1,3-bis(3,5-bis(trifluoromethyl)phenyl)thiourea\], which is a stronger H-bond donor but lacks an amide appendage. ### C.3 Stabilisation of the Cationic Transition State Through Cation-π Interactions {#S7} The strong correlation between reaction enantioselectivity and the identity of the arene on the catalyst suggests a direct role of the extended π-system in the mechanism of stereoinduction. In principle, this arene effect may be due primarily either to acceleration of the major pathway through transition state stabilisation, or to inhibition of pathways leading to the minor enantiomer through destabilising interactions. This question was addressed through kinetic analysis of the reaction, taking advantage of the fact that the indole addition step is both rate- and ee- determining. The rate constant corresponding to the major pathway (*k*~asym,major~) could be deduced for catalysts **3a-3g** from in situ IR-based kinetic measurements combined with er (enantiomeric ratio) determinations. A strong correlation between this rate and reaction enantioselectivity was observed, with plots of ln(*k*~asym,major~) vs ln(er) providing a good linear fit ([Figure 4a](#F4){ref-type="fig"}). This provides unambiguous evidence that enantioselectivity increases due to variation of the aryl component of the catalyst **3** are indeed tied to stabilisation of the major transition structure.^[@R14]^ The rate of the pathway leading to the minor enantiomer also displays a linear, positive correlation with reaction er, indicating that the minor transition structure is also stabilised selectively by the more enantioselective catalysts, albeit to a substantially lesser extent. Insight into the nature of the transition state stabilising interactions that may be at play was provided through spectroscopic analysis of thiourea derivatives **3a** and **3e** complexed to a sulfonium ion model system. The dibenzylmethylsulfonium ion triflate **5** was selected for these studies because episulfonium sulfonate could not be examined directly (see section B), and the cation in **5** was found by computational methods to have a very similar charge distribution to the episulfonium ion ([Figure 4 c](#F4){ref-type="fig"}). With thiourea **3e**, the resonances of the benzylic and the methyl protons of **5** underwent a significant upfield shift (0.6-0.8 ppm) upon formation of the 1:1 complex ([Figure 4b](#F4){ref-type="fig"}).^[@R37]^ In contrast, the chemical shift of those protons was bearly perturbed to any measurable extent (Δδ \< 0.05 ppm) upon complexation of **5** with the thiourea derivative **3a**, which lacks an aryl substituent. This points to an attractive cation-π interaction between the π-face of the arene in catalyst **3e** and the sulfonium ion in the model system.^[@R38],[@R39],[@R40],[@R41],[@R42]^ Such interaction in the transition structure of the indole addition to episulfonium ions could underlie the observed enantioselectivity effects, as the cation-π interaction would be expected to increase in magnitude with more extended aromatic substituents.^[@R43],[@R44]^ Based on the kinetic and enantioselectivity data, we therefore propose that the difference in the strength of the cation-π interaction between the major and minor pathways lies at the origin of the observed high reaction enantioselectivity.^[@R6],[@R14],[@R45]^ It should be noted that on the basis of polarisability effects alone, which are known to correlate directly with cation-π binding ability,^[@R46]^ the most expansive aryl-substituted catalysts **3f** and **3g** might be expected to be most enantioselective. However, as noted above ([Table 1](#T1){ref-type="table"}) these two thiourea derivatives afford slightly lower ee's in the model reaction than the smaller, phenanthryl catalyst **3e**. The reason for the lack of an exact correlation between the polarisability of the aromatic substituent and reaction enantioselectivity has not yet been established; however, it seems reasonable to expect that any number of minor steric or conformational factors could attenuate the ability of the largest substituents to engage fully in the stabilising cation-π interaction. Taken together, the data presented above allow construction of a detailed mechanistic model for the enantioselective reaction, wherein rate-acceleration and enantioselectivity are induced by the thiourea catalyst through a network of attractive non-covalent interactions. In particular, we propose that the transition structure for the rate-determining addition of indole to the episulfonium ion is stabilised by a combination of anion binding of the thiourea to the sulfonate, general base activation of the indole via a catalyst amide--indole N-H interaction, and a cation-π interaction between the arene of the catalyst and the benzylic protons of the episulfonium ion ([Figure 5](#F5){ref-type="fig"}). We anticipate that characterisation of these enzyme-like non-covalent stabilising elements with small molecule catalysts such as **3e** may enable the future design and application of such biomimetic strategies in organic asymmetric synthesis. Methods {#S8} ======= General procedure for thiourea 3e-catalysed nucleophilic ring opening of episulfonium ions with indole derivatives {#S9} ------------------------------------------------------------------------------------------------------------------ An oven-dried 1.5 dram vial was charged with substrate **1** (0.05 mmol, 1.0 equiv), **3e** (3.2 mg, 0.0050 mmol, 0.10 equiv), indole (11.7 mg, 0.10 mmol, 2.0 equiv) and 4Å molecular sieves (25 mg, powder, activated) under an atmosphere of N~2~. The vial was cooled to −30 °C, and toluene (1 mL) was added with stirring. Once the reactants and catalyst were fully dissolved, the mixture was cooled to −78 °C, and solid 4-NBSA (0.7 mg, 0.0035 mmol, 0.07 equiv) was added at once against a counterflow of N~2~. The resulting solution was stirred at −30 °C and the progress of the reaction was monitored by thin layer chromatography (TLC) (see [Supplementary Information](#SD2){ref-type="supplementary-material"}, Section 3). When the progress of the reaction was determined to be complete, triethylamine (\~10 μL) was added at −30 °C. The resulting mixture was applied directly to a pipette column containing 4-5 cm of silica gel, and product was isolated by eluting hexanes/ethyl acetate (20:1 to 10:1) and solvent removal. **The Supplementary Information contains:** detailed experimental procedures, synthesis of substrates and catalysts, characterisation data for all new compounds, procedures and data for mechanistic investigations (including reaction progress kinetic analysis, linear free energy relationship studies with Mayr's reactivity parameters, kinetic isotope effect studies, model binding studies by ^1^H NMR, and other experimental kinetic studies). Crystallographic information for compounds **2b, 2g** and **2q** have been deposited at the Cambridge Crystallographic Data Centre and allocated the deposition numbers (CCDC 862750, 862751 and 862752, respectively). Supplementary Material {#S10} ====================== This work was supported by the NIGMS (P50 GM-69721 and RO1 GM-43214) and by a predoctoral fellowship to S.L. from Eli Lilly. We thank Dr. Robert Knowles, Dr. Katrien Brak and Prof. Dr. Herbert Mayr for helpful discussions, Adam Brown, Dr. Dan Lehnherr and Dr. Alan Hyde for the use of catalysts, and Dr. Shao-Liang Zheng for crystal structure determinations. Author Contributions S.L. conducted the experiments; S.L. and E.N.J. wrote the manuscript; E.N.J. guided the research. Supplementary information and chemical compound information accompany this paper at [www.nature.com/naturechemistry](www.nature.com/naturechemistry). The authors declare no competing financial interests. {#F1} {#F2} {ref-type="supplementary-material"}, Section 14 for detailed experimental procedure and data analysis). Error bars reflect the range of experimental data from 2-3 individual measurements, and the line represents the least-squares fit.](nihms399652f3){#F3} {ref-type="disp-formula"} & [4](#FD2){ref-type="disp-formula"}) and the following equations: (a) *r*~asym~ = *r*~asym,major~ + *r*~asym,minor~, (b) er = *r*~asym,major~/*r*~asym,minor~. Lines represent least-squares fits. **b**. ^1^H NMR binding study of thiourea and **5** in *d*~8~-toluene, showing attractive interactions between the aromatic group in **3e** and the a-protons in **5**. The resonances of the benzylic protons and the methyl protons in **5** are labled with blue and green dots, respectively. In the two bottom spectra, the methyl resonance overlaps with the solvent peak, but can be identified when zoomed-in **c**. Electrostatic potential maps for fully optimized structures (B3LYP/6-31G(d)) of the episulfonium ion derived from **1a** and the sulfonium ion in **5**, revealing a similar distribution of positive charge over the benzylic protons. Negative potentials are shown in red and positive potentials in blue.](nihms399652f4){#F4} {#F5} ###### Reaction optimisation.[a](#TFN1){ref-type="table-fn"} ---- -------- ------------ ---- ----- 1 **3b** HCl 10 5 2 **3b** HOTf 73 32 3 **3b** FSO~3~H 78 19 4 **3b** 2,4-diNBSA 79 63 5 **3b** 4-NBSA 72 73 6 -- 4-NBSA 7 n/a 7 **3a** 4-NBSA 16 12 8 **3c** 4-NBSA 84 84 9 **3d** 4-NBSA 80 85 10 **3e** 4-NBSA 93 93 11 **3f** 4-NBSA 91 91 12 **3g** 4-NBSA 97 88 13 **4e** 4-NBSA 98 92 ---- -------- ------------ ---- ----- Optimisation reactions were performed on 0.05 mmol scale. Isolated yields of material purified chromatographically. Enantiomeric excesses (ee's) determined by HPLC analysis. MS = molecular sieves; 2,4-diNBSA = 2,4-dinitrobenzenesulfonic acid; 4-NBSA = 4-nitrobenzenesulfonic acid. ###### Substrate scope of nucleophilic ring opening of episulfonium ions with indole derivatives.[a](#TFN5){ref-type="table-fn"} ----------------------------------- ------------------ --------------- ----- ----------------------------------------------- -------- ------ ---- 1 Bn **2a** 99 93 2 Ph **2b** 83 85 3[b](#TFN6){ref-type="table-fn"} Ph **2b** 84 80 4 4-F-C~6~H~4~ **2c** 73 81 5 Ph 4-Me-C~6~H~4~ H **2d** 76 87 6 2-naphthyl **2e** 90 88 7 PMB **2f** \>99 94 8 Me **2g** 72 84 9 *t*-Bu **2h** 89 87 10 5-Me **2i** 97 91 11 5-MeO **2j** 93 93 12 5-Br **2k** 83 92 13 5-F **2l** 88 95 14 Ph Bn TCA 6-F **2m** 92 85 15 4-MeO **2n** 83 91 16 2-Me **2o** 95 79 17 *N*-Me **2p** 54 3 18 benzotriazole[f](#TFN10){ref-type="table-fn"} **2q** 92 80 19[c](#TFN7){ref-type="table-fn"} 3-MeO-C~6~H~4~ **2r** 85 93 20 3-F-C~6~H~4~ **2s** 97 95 21 3-Me-C~6~H~4~ **2t** 95 93 22 2-Me-C~6~H~4~ **2u** \>99 79 23 4-F-C~6~H~4~ Bn TCA H **2v** 91 45 24 4-Me-C~6~H~4~ **2w** 89 60 25[c](#TFN7){ref-type="table-fn"} 4-MeO-C~6~H~4~ **2x** 67 6 26 4-CF3-C~6~H~4~ **2y** 18 5 27[d](#TFN8){ref-type="table-fn"} ---(CH~2~)~4~--- **2z** 16 9 28[e](#TFN9){ref-type="table-fn"} TCA **2b** 87 87 29[e](#TFN9){ref-type="table-fn"} Ph Ph H **2b** 91 85 30[e](#TFN9){ref-type="table-fn"} **2b** 85 85 ----------------------------------- ------------------ --------------- ----- ----------------------------------------------- -------- ------ ---- Isolated yields of material purified chromatographically are reported. With urea **4e**. Yields determined by ^1^H NMR are reported. Product isolated via flash column chromatography usually contains a small amount of **3e**. R^2^ = Ph, with HOTf. With 20 mol% **3e** and 10 mol% 4-NBSA. Benzotriazole as nucleophile instead of indole.

Introduction ============ Neurotrophic keratitis (NK) is a rare degenerative corneal disease caused by damage of trigeminal innervation. Corneal nerves play an essential role in tear production and preservation of the normal metabolism and function of the ocular surface. Loss of sensitivity impairs corneal wound healing, leading to epithelial changes including punctate epithelial keratopathy, persistent epithelial defects, and corneal ulcer.[@b1-eb-10-037] Various ocular and systemic diseases can cause damage to the fifth cranial nerve at different levels, from the trigeminal nucleus to the corneal nerve endings. Common causes are herpetic keratitis, diabetes, chemical or surgical damage, and neurosurgical procedures.[@b2-eb-10-037],[@b3-eb-10-037] Medical treatment consists mainly of supportive measures that do not address the underlying cause of the disease. Recently, novel promising treatments have been proposed and are currently under evaluation. These include mainly topical treatment with nerve growth factor (NGF) and surgical corneal neurotization. Such approaches target the pathogenic elements of NK, carrying the potential of restoring normal corneal innervation and sensitivity. Pathogenesis ------------ The cornea is innervated by the ophthalmic branch of the trigeminal nerve as well as by autonomic nerves. Nerve bundles enter the cornea at the limbus, move toward the center below the anterior third of the stroma, penetrate Bowman's layer, and create a dense network of nerve fibers between Bowman's layer and basal epithelial cells (i.e., sub-basal nerve plexus).[@b4-eb-10-037] The great number of sensory nerve endings makes the cornea the most densely innervated tissue in the human body.[@b5-eb-10-037] Nerves have a key role in maintaining a healthy ocular surface, both by triggering protective reflexes after injury, and by providing trophic factors to the corneal cells. Corneal denervation causes decreased vitality, metabolism, and mitosis of epithelial cells, with subsequent epithelial changes including intracellular edema, loss of microvilli, and abnormal development of the basal lamina.[@b6-eb-10-037],[@b7-eb-10-037] Corneal sensory innervation reacts to mechanical, chemical, and thermal stimuli with two reflex arcs: a motor arc eliciting blinking, and an autonomous arc stimulating tear secretion. Therefore, the reduction of corneal sensation impairs these two reflexes with an alteration of production and stability of the tear film.[@b8-eb-10-037] Corneal nerves and epithelial cells support each other mutually through the release of trophic factors promoting epithelial cell proliferation, migration, and differentiation as well as nerve development and survival. Corneal nerves express many epitheliotrophic neuromediators, such as substance P, calcitonin gene-related peptide, acetylcholine, noradrenaline, serotonin, neuropeptide Y, and vasointestinal peptide. On the other side, corneal epithelial cells release various neurotrophic growth factors, including NGF, ciliary neurotrophic factor, and glial-cell-derived neurotrophic factor. These factors are fundamental in ocular surface homeostasis and wound healing.[@b4-eb-10-037] In particular, NGF is a neurotrophin discovered in the early 1950s by R. Levi-Montalcini that promotes neuronal sprouting by intact and injured neurons. It provides trophic support to neurons after injury and reverses pathologic changes induced by peripheral nerve damage.[@b9-eb-10-037] NGF exerts its biologic functions by binding to low-affinity p75NTR (neurotrophin receptor, a member of the tumor necrosis factor receptor superfamily) and high-affinity TrkA receptor (a tyrosine kinase receptor family). The human cornea produces NGF and expresses high-affinity NGF receptors.[@b10-eb-10-037] NGF demonstrated an important role in the integrity and function of the ocular surface, stimulating both epithelial and nerve fiber proliferation and survival.[@b11-eb-10-037] In vivo studies proved that NGF concentration increases after injury, and that administration of NGF accelerates corneal healing.[@b10-eb-10-037] These complex reciprocal relations between epithelial cells and nerves are crucial for the corneal physiologic renewal and wound healing. An impairment of corneal sensitivity interrupts these trophic relations and may produce the pathologic changes typical of NK. Causes ------ All conditions that impair trigeminal innervation at any level can cause NK, such as ocular surface diseases, systemic diseases, and central or peripheral nervous damages ([Figure 1](#f1-eb-10-037){ref-type="fig"}).[@b12-eb-10-037],[@b13-eb-10-037] Herpes simplex and herpes zoster eye infections are among the most common conditions leading to corneal anesthesia. In a large case series from Bonini et al, herpetic keratitis was the most common cause of NK (27% of the patients).[@b2-eb-10-037] A severe reduction of sub-basal nerve plexus density was demonstrated in herpetic keratitis.[@b14-eb-10-037] Corneal surgeries like laser-assisted in situ keratomileusis, keratoplasty, and corneal incisions for cataract may disrupt corneal innervation and cause NK.[@b15-eb-10-037],[@b16-eb-10-037] After nerve transection at the peripheral cornea, the distal nerves undergo Wallerian degeneration with immediate loss of sensation to the corresponding region. Nerve regeneration occurs gradually over months with variable return of corneal sensitivity.[@b15-eb-10-037] Other ocular surface conditions associated with NK are corneal chemical and traumatic injuries, corneal dystrophies, contact lens wearing, and chronic use of topical medications (such as anesthetics, beta-blockers, antivirals, glaucoma medications, antibiotics, and nonsteroidal anti-inflammatory drugs).[@b12-eb-10-037],[@b13-eb-10-037] Surgical procedures for the treatment of trigeminal neuralgia may induce permanent damage to the trigeminal nerve with subsequent corneal anesthesia. The main surgical options include microvascular decompression, balloon compression, radiofrequency thermocoagulation, and gamma knife radiosurgery. The intraoperative damage occurs at the preganglionic or ganglionic region of the nerve and is not associated with a reduction of sub-basal nerve plexus density.[@b17-eb-10-037] Other less common causes of trigeminal palsy are intracranial and orbital tumors, facial trauma, aneurysm, and stroke.[@b1-eb-10-037],[@b12-eb-10-037] Systemic conditions like diabetes, multiple sclerosis, and leprosy may impair corneal sensitivity and cause NK. In particular, neuropathy is one of the most common complications of diabetes mellitus and results from microvascular damage of myelinated nerve fibers. The extent of nerve damage has been demonstrated to be associated with the duration of hyperglycemia.[@b18-eb-10-037] Corneal nerve evaluation in patients with diabetes has shown that the decreased corneal sensitivity was related to a reduction of sub-basal nerve density, with abnormally tortuous nerve fiber bundles. Additionally, a correlation between the decrease of corneal sensitivity and the degree of somatic polyneuropathy was found.[@b18-eb-10-037] Recent studies using in vivo confocal microscopy (IVCM) have reported an association between levels of glycated hemoglobin and the density of corneal innervation.[@b19-eb-10-037] Long-term systemic therapy with antipsychotics and antihistamines may also cause NK. Very rare congenital causes of NK include Riley--Day syndrome, Moebius syndrome, Goldenhar--Gorlin syndrome, and congenital corneal anesthesia.[@b12-eb-10-037] Clinical presentation --------------------- The clinical presentation of NK is characterized by well-defined and progressive ocular surface changes, which are caused by the loss of sensitivity, regardless the cause of trigeminal damage. A three-stage classification of NK based on severity of corneal damage was introduced by Mackie.[@b1-eb-10-037] Stage 1 is characterized by corneal abnormalities such as punctate keratopathy, corneal edema, corneal epithelial hyperplasia, and irregularity ([Figure 2A](#f2-eb-10-037){ref-type="fig"}). Rose Bengal staining of the conjunctiva, increased tear viscosity, and decreased breakup time are often associated. In long-standing cases, stromal scarring and superficial corneal neovascularization may develop. Stage 2 is characterized by a recurrent or persistent epithelial defect, commonly located paracentrally in the superior half of the cornea. The epithelial defect is often delimited by smooth, rolled, and loose edges of an opaque and edematous epithelium ([Figure 2B](#f2-eb-10-037){ref-type="fig"}). Stromal edema with Descemet's membrane folds may be observed and sometimes anterior chamber inflammation is associated. Stage 3 is characterized by corneal ulcer ([Figure 2C](#f2-eb-10-037){ref-type="fig"}). The stromal melting may progress to corneal thinning and eventually perforation. This may occur without significant ocular symptoms because of impaired corneal sensitivity; however, patients may complain of blurred vision in case of corneal ulcer, edema, or scarring.[@b1-eb-10-037],[@b12-eb-10-037] Diagnosis --------- Patient's clinical history should be accurately collected and reviewed to identify any ocular or systemic disease potentially associated with the disorder, such as ocular surface conditions (e.g., previous herpetic infection, ocular surgery, chemical burns, topical drug use, contact lens wearing), systemic diseases (e.g., diabetes, multiple sclerosis), and neurologic conditions (e.g., brain tumors, neurosurgery, stroke, trauma). A careful ocular surface examination should be performed. In particular, eyelids should be examined to rule out the presence of lagophthalmos or blepharitis. Slit-lamp examination is fundamental to detect sings of NK and to classify NK severity, ranging from punctate corneal keratopathy to stromal melting and corneal perforation. Corneal anesthesia causes lack of ocular discomfort symptoms; thus, NK should be always suspected in case of significant discrepancy between ocular signs and symptoms. Ocular examination helps to identify also possible causes of decreased corneal sensitivity: corneal scarring may indicate previous injuries or infections; patchy iris atrophy may be a sign of previous herpetic infection. Fundus examination may show signs of diabetic retinopathy or optic nerve abnormalities due to intracranial pathology.[@b12-eb-10-037],[@b13-eb-10-037] Vital staining with fluorescein or lissamine green is useful to assess corneal and conjunctival epithelial integrity. Tear film production and stability should be evaluated with lacrimal function tests. It is known that neurotropic keratitis is associated with an increase of tear osmolarity and a decrease of both Schirmer test and tear breakup time values.[@b8-eb-10-037],[@b20-eb-10-037] The decrease or absence of corneal sensitivity is the diagnostic hallmark of NK. Corneal sensitivity can be quantitatively measured by Cochet--Bonnet esthesiometer, which employs a nylon monofilament of adjustable length touching the cornea and recording the patient's response. Each quadrant of the cornea may be tested. As the length of the filament is decreased, the pressure transmitted is increased.[@b21-eb-10-037] The noncontact Belmonte esthesiometer is another device allowing to investigate also chemical and thermal corneal sensitivity.[@b22-eb-10-037] However, it is typically reserved for research purposes, and not routinely used in clinical practice to date. IVCM is a noninvasive imaging technique that allows the in vivo study of the different layers of the ocular surface at a cellular level. IVCM has been used to evaluate corneal nerve morphology in healthy and diseased corneas. A significant reduction of corneal nerve density strongly correlated with the decrease of corneal sensation was shown in patients with NK.[@b14-eb-10-037]--[@b18-eb-10-037] In addition, IVCM showed also a higher number of hyper-reflective keratocytes and a lower epithelial and endothelial cell density in patients with NK.[@b16-eb-10-037] These findings suggest that corneal sensory nerves play a role also in the function and survival of the corneal endothelium. Treatment --------- Treatment approach of NK should be prompt and based on the stage and severity of the disease. Despite various medical and surgical therapies having been proposed, NK remains yet difficult and challenging to treat, and the lack of positive response is commonly observed in the clinical practice. In stage 1, the therapy aims at preventing an epithelial breakdown and improving epithelial quality. All topical and systemic medications associated with ocular surface toxicity should be discontinued. Preservative-free tear substitutes and lubricant ointments are useful to improve lubrication and ocular surface health. Other associated ocular surface diseases, such as dry eye, exposure keratitis, and limbal stem cell deficiency, should be properly treated.[@b12-eb-10-037] In stage 2, therapy is addressed to promote healing of the epithelial defect, and to prevent the progression to corneal ulcer. Patient should be monitored frequently, because the progression to stromal melting and perforation may occur without any symptom. Topical antibiotics are recommended in this stage to prevent infections. On the contrary, topical steroids should be used with caution, because they may inhibit the healing process and increase the risk of corneal melting. Therapeutic contact lenses are useful to promote corneal healing, maintaining a fluid layer in stable contact with the cornea, and protecting it from the rubbing of the eyelids. However, attention should be paid because of the increased risk of infections.[@b23-eb-10-037] Autologous serum eye drops have become increasingly popular for treating ocular surface disorders, including NK.[@b24-eb-10-037] Autologous serum contains several natural components of the normal tears, such as growths factors, neuromediators, cytokines, and vitamins.[@b25-eb-10-037] These substances are known to promote proliferation, migration, and differentiation of the ocular surface epithelium, and are essential in corneal homeostasis and wound healing.[@b26-eb-10-037] Matsumoto et al reported the complete healing of all the 14 eyes with NK treated with autologous serum drops and an increase in corneal sensitivity in 64.2% of cases.[@b26-eb-10-037] The study demonstrated that serum harbors neurotrophins and growth factors to the ocular surface. Other more recent studies confirmed that autologous serum eye drops allowed high rates of corneal healing, and also the improvement of corneal nerve morphology with increased number, length, width, and density detected by IVCM.[@b27-eb-10-037]--[@b29-eb-10-037] The use of autologous serum may be inconvenient or contraindicated in patients with coexisting general conditions, such as anemia and blood discrasia. Therefore, the use of allogeneic sera obtained from healthy donors has been proposed as a viable alternative. Among these, umbilical cord blood serum is rich of epitheliotrophic growth factors, with higher levels of epidermal growth factor, insulin-like growth factor, transforming growth factor-beta, and vascular endothelial growth factors compared to peripheral blood serum.[@b30-eb-10-037],[@b31-eb-10-037] It has been shown that umbilical cord blood serum eye drops is effective in treating NK, achieving a complete corneal healing in all the patients under study.[@b32-eb-10-037] A recent IVCM study showed an improvement in corneal nerve morphology in patients treated with umbilical cord serum eye drops, with an increase of total nerve number and a decrease of nerve tortuosity.[@b33-eb-10-037] Two randomized clinical trials are currently ongoing to evaluate safety and efficacy of topical umbilical cord blood serum and plasma rich in growth factors in patients with NK (NCT02707120, NCT03084861). Surgical treatment is usually limited to corneal ulcers (stage 3) not responding to medical treatment, and/or to the related complications. Several approaches are available, including tarsorraphy, conjunctival flap, and amniotic membrane transplantation. Tarsorraphy is the most commonly used procedure to promote corneal healing in NK. It reduces the width of the palpebral fissure, protecting the cornea from the traumatic rubbing of eyelids and decreasing tear evaporation rate.[@b34-eb-10-037] However, the poor cosmetic outcome is often a major concern for patients. The injection of Botulin A toxin into the superior elevator palpebrae superioris muscle is an effective alternative, allowing also an easier and better examination of the ocular surface compared to tarsorraphy.[@b35-eb-10-037] The use of conjunctival flaps was first proposed by Gundersen in 1958 and became a standard surgical procedure to treat corneal ulcers with or without perforation.[@b36-eb-10-037] After a 360° peritomy, a pedunculated conjunctival flap is mobilized from the upper bulbar conjunctiva and sutured to the conjunctiva in the lower limbus to cover the affected cornea. The conjunctival flap provides metabolic and mechanical support for corneal healing, thanks to the conjunctival blood vessels, which transport nutrients and growth factors to the corneal surface.[@b37-eb-10-037] The procedure restores the anatomical integrity of the ocular surface but sacrifices visual function. Amniotic membrane transplantation has been successfully used to treat corneal ulcers and persistent epithelial defects from different causes, including NK.[@b38-eb-10-037],[@b39-eb-10-037] The amniotic membrane provides mechanical protection, releases growth factors, and supports epithelial cell adhesion and proliferation.[@b38-eb-10-037] Evidence from a randomized controlled trial showed that both amniotic membrane transplantation and conventional management (tarsorraphy or therapeutic contact lenses) are effective in treating refractive NK.[@b40-eb-10-037] Small perforations can be effectively treated with cyanoacrylate gluing followed by application of a bandage contact lens, whereas lamellar or penetrating keratoplasty is required for larger defects.[@b41-eb-10-037] However, persistent anesthesia causes impaired wound healing and a high risk of recurrence of corneal ulceration in the graft. This explains the low success rate of corneal transplantation in patients with NK. Three cases from our cohort patients, which may be representative of different history and course of the disease, are shown in [Figure 3](#f3-eb-10-037){ref-type="fig"}. Explanation of each case is given in the legend. Future prospects ---------------- NK was traditionally considered an orphan disease with no effective treatment. In recent years, novel promising medical and surgical treatments have been developed and are currently under investigation. Recent articles of novel medical treatments are summarized in [Table 1](#t1-eb-10-037){ref-type="table"}. Regenerating agent (RGTA, Cacicol20^®^; OTR3, Paris, France) is a new matrix agent containing large polymers mimicking heparan sulfates. RGTA creates a microenvironment that induces cell migration and adhesion and promotes epithelial healing. A Phase III clinical trial is currently ongoing (NCT01794312). In a recent study, topical drops containing RGTA achieved complete healing of corneal ulcer in 73% of 11 patients with NK.[@b42-eb-10-037] Another work showed less favorable results, with a failure of corneal healing in 67% of treated patients.[@b43-eb-10-037] The combined therapy of RGTA and umbilical cord blood serum eye drops has been recently proposed, thanks to their synergistic effect.[@b44-eb-10-037] RGTA could replace destroyed heparan sulfates, protecting the bioavailability of growth factors and providing a matrix in which cells can migrate and multiply.[@b45-eb-10-037] Thymosin beta-4 is an intracellular protein with various functions in different cellular processes, including wound healing, suppression of inflammation and apoptosis, synaptogenesis, and axon growth.[@b46-eb-10-037] The efficacy of topical treatment with thymosin beta-4 in nine patients with NK was described in an open study that reported a complete corneal healing in six patients with corneal ulcers, while the remaining three patients with punctuate epithelial defects did not show significant clinical changes.[@b46-eb-10-037] A randomized-controlled trial is currently ongoing to evaluate safety and efficacy of topical treatment with thymosin beta-4 in patients with NK (NCT02600429). Topical substance P (SP) and insulin-like growth factor 1 (IGF-1) stimulate synergistically the proliferation and migration of corneal epithelial cells both in vitro and in vivo.[@b47-eb-10-037],[@b48-eb-10-037] In two recent studies, patients with epithelial defects associated to NK were treated with eye drops containing a combination of an SP-derived peptide and IGF-1-derived peptide.[@b49-eb-10-037],[@b50-eb-10-037] The treatment induced the healing of the epithelial defects in 89%[@b49-eb-10-037] and 73% of cases[@b50-eb-10-037] without any adverse side effect. Topical treatment with NGF represents one of the most promising therapies for NK. NGF is a neurotrophin that promotes growth and survival of sensory and sympathetic neurons and can restore the function of injured neurons.[@b9-eb-10-037] NGF demonstrated a key role in the immune modulation, trophism, and healing of the ocular surface.[@b11-eb-10-037] Two open-label studies tested murine-derived NGF in patients with moderate (stage 2) and severe (stage 3) NK. The treatment induced rapid corneal healing in all the patients, with improved corneal sensitivity, tear production, and visual acuity without any adverse event.[@b2-eb-10-037],[@b51-eb-10-037] A recombinant human NGF eye drop formulation has been subsequently developed, and Phase I studies showed a good safety profile.[@b52-eb-10-037] Phase II trials are currently ongoing to evaluate the efficacy of two different formulations of this treatment in patients with NK stage 2 and 3 (NCT02227147, NCT01756456). Nicergoline is an ergoline derivative used to treat degenerative and vascular dementia. In animal rat models, it was shown to promote corneal healing and to increase NGF levels in the cornea.[@b53-eb-10-037] A recent study evaluated treatment with oral nicergoline in patients with NK, demonstrating a healing of the epithelial defect in 23 out of 27 eyes, with an improvement of NGF levels and corneal sensitivity.[@b54-eb-10-037] Corneal neurotization is a novel and promising surgical procedure for NK, which can be performed through two different techniques, based on the transposition of the supraorbital and/or supratrochlear nerves (direct neurotization),[@b55-eb-10-037],[@b56-eb-10-037] or on the sural nerve graft (indirect neurotization).[@b57-eb-10-037] In the first technique, contralateral healthy supraorbital and supratrochlear nerves are harvested through a large coronal incision, tunneled across the bridge of the nose and inserted around the limbus of the anesthetic eye. Six patients with NK were treated with this technique, and all the patients showed the improvement of corneal health and sensitivity.[@b55-eb-10-037] Allevi et al performed corneal neurotization according to the same technique, combined with upper eyelid neurotized platysma graft, and followed by penetrating keratoplasty in a patient with associated seventh and fifth cranial nerve palsy. The overall procedure was successful allowing patient to regain corneal sensitivity, eyelid movement, and visual acuity.[@b56-eb-10-037] As the ipsilateral frontal nerve is still intact in case of isolated corneal anesthesia from local damage to the long ciliary nerves, ipsilateral supraorbital nerve can be used to directly neurotize the cornea, allowing the decrease of the size of the scalp incision and the extent of the dissection needed.[@b58-eb-10-037] In the indirect technique, corneal neurotization was performed using a sural nerve autograft anastomosed to the supratrochlear nerve and tunneled through the upper eyelid incision. Unlike the direct technique, this approach enables the management of bilateral NK and avoids the large bicoronal incision. Four eyes with NK were included in the study: three of them had noticeably improved corneal sensitivity 6 months after surgery.[@b57-eb-10-037] A recent cadaver feasibility study described a new minimally invasive technique for corneal neurotization using an endoscopic approach for direct nerve supraorbital nerve transfer.[@b59-eb-10-037] The potential advantages of the endoscopic corneal neurotization include smaller incisions, decreased operative and healing time, decreased forehead edema, less risk for facial nerve injury, and minimal blood loss. A pilot study is currently ongoing for NK stages 2 and 3 (NCT03037450). The patients gave their consent for the photographs to be used in medical publication. **Disclosure** The authors report no conflicts of interest in this work. {#f1-eb-10-037} {#f2-eb-10-037} {#f3-eb-10-037} ###### Novel medical treatments for neurotrophic keratitis Study Eyes (no.) Treatment Posology Onset-treatment interval (days) Complete healing Healing time (days) -------------------------------- ------------ -------------------- ----------------------------------- --------------------------------- ------------------ ---------------------------------------------------- Aifa et al[@b42-eb-10-037] 11 RGTA 1x/on alternate days \>15 72.7% 60.9 Arvola et al[@b43-eb-10-037] 6 RGTA 1x/on alternate days 45 33% 56 Dunn et al[@b46-eb-10-037] 9 Thymosin beta-4 4x/day \>42 67% 45 Nishida et al[@b49-eb-10-037] 9 SP and IGF1 4x/day 141 89% 13.3 Yamada et al[@b50-eb-10-037] 26 SP and IGF1 4x/day 96 73% 10.5 Lambiase et al[@b51-eb-10-037] 14 NGF Every 2 h for 2 days, then 6x/day 45 100% 21 Bonini et al[@b2-eb-10-037] 45 NGF Every 2 h for 2 days, then 6x/day 38 100% 22.8/26.6[a](#tfn1-eb-10-037){ref-type="table-fn"} Lee et al[@b54-eb-10-037] 27 Nicergoline (oral) 10 mg 2x/day \>60 85% 15.6 **Note:** Range (minimum/maximum). **Abbreviations:** RGTA, regenerating agent; SP, substance P; IGF, insulin-like growth factor; NGF nerve growth factor; h, hours.

Introduction {#s0005} ============ Cholangiocarcinoma (CCA) is a malignant cancer with an unknown etiology and an unfavorable prognosis. The morbidity and mortality of CCA have increased year by year worldwide [@bb0005]. Because of limited diagnostic methodologies, the majority of patients are diagnosed at advanced stages and are not eligible for surgery, and many of these patients are further chemo-resistant to conventional chemotherapy [@bb0010]. Therefore, for the past three decades, the 5-year survival rates of CCA has remained only 10% [@bb0010]. Consequently, it is essential to find novel curative targets and therapeutic strategies for CCA. To meet the high demand for rapid proliferation under different stressed conditions, cancer cells reprogram their metabolisms [@bb0015]. Known as the Warburg effect, this metabolic reprograming is marked by a conversion from oxidative phosphorylation to aerobic glycolysis [@bb0020], [@bb0025]. This metabolic reprogramming produces anti-oxidant substances, reduces tumor oxidative stress, thereby finally promoting the proliferation of tumors. These metabolic changes facilitate the accumulation of lactate and glycolytic intermediates to promote the growth and invasion of tumor. Hence, reversing the Warburg effect can be a potential mechanism for cancer therapy [@bb0030]. Changes in the expression of key enzymes that catalyze these biosynthesis pathways have been found in multiple tumor metabolic reprogramming [@bb0035]. Pyruvate dehydrogenase (PDH) complex is a multi-enzyme complex that regulates carbohydrate and fat metabolism, of which PDHA1 is the major component [@bb0040]. PDH can catalyze the irreversible decarboxylation of pyruvate into acetyl-CoA, which plays a crucial role in many biological reactions [@bb0045]. Abnormal expression of PDH has been found in a variety of tumors. Several studies have found that PDH activates the metabolism of cancer cells from glycolysis to glucose oxidation, which reduces the lactate production and increases reactive oxygen species (ROS), thus inducing the apoptosis and proliferation of tumor cells [@bb0050]. In the process of metabolic reprogramming, another important factor for cancer cell proliferation is the activation of the serine synthesis pathway (SSP) [@bb0055]. Serine is a non-essential amino acid in mammals, but plays an essential role in cancer progression [@bb0060]. Glycolytic metabolisms of glucose can be catalyzed by 3-phosphoglycerate dehydrogenase (PHGDH), phosphoserine aminotransferase (PSAT1), and phosphoserine phosphatase (PSPH) to produce serine. Therefore, PHGDH, PSAT1, and PSPH are considered important components of SSP. As a scavenger of ROS, serine contributes to the redox balance to promote the proliferation of cancer cells [@bb0065]. However, the specific role of SSP in cancer remains largely unexplored. Post-translational modifications in metabolism regulation have attracted attention because of their ability to respond to changes in cellular metabolic status and regulate through upstream signal transduction pathways. Acetylation is a crucial post-translational modification found in the metabolic enzymes of cell regulation [@bb0070]. Deacetylases are classified into four groups (I-IV), wherein sirtuin (SIRTs) are NAD^+^-dependent class III histone deacetylases (HDACs). Among the seven SIRT homologues in mammals [@bb0075], SIRT2 catalyzes a variety of biological processes such as metabolism and gene expressions [@bb0080]. There is growing evidence that SIRT2 promotes proliferation in malignant tumors [@bb0080], [@bb0085]. Studies have demonstrated that SIRT2 inhibitors have anti-tumor, anti-diabetic, and anti-inflammatory effects [@bb0090]. Recent studies have shown that SIRT2 can inhibit the ubiquitin degradation of cMYC, thus contributing to the stability of cMYC and the proliferation of cancer. However, a SIRT2 inhibitor, thiomyristoyl (Tm), can inhibit the growth of cancer cells by degrading cMYC [@bb0095]. cMYC is an important transcription factor, which is up-regulated in numerous human tumors [@bb0100]. A recent research has further found that cMYC can stimulate SSP activation by transcriptionally up-regulating the expressions of SSP-related enzymes [@bb0105]. Together, these findings suggest that targeting the inhibition of SIRT2/cMYC may play a crucial role in cancer prevention and treatment. However, it remains unclear regarding whether the SIRT2/cMYC pathway has a synergistic effect on the metabolic reprogramming of CCA. In the present study, we found that SIRT2/cMYC promoted metabolic reprogramming in CCA through two signaling pathways: 1) promoting PDHA1 phosphorylation (p-PDHA1) to inhibit oxidative phosphorylation; and 2) promoting serine synthesis by activating the SSP pathway to reduce apoptosis. The SIRT2/cMYC pathway plays a crucial role in transforming glucose oxidative metabolism to serine anabolic metabolism and thus provides antioxidants for stress resistance. SIRT2/cMYC-induced metabolic reprogramming may thus represent a new therapeutic target for treating CCA. Materials and Methods {#s0010} ===================== Ethics Approval and Consent to Participate {#s0015} ------------------------------------------ All experiments utilizing human samples and involving animals were reviewed and approved by the Ethical Committee and Animal Welfare Committee of Drum Tower Hospital, Nanjing University. Cell Culture {#s0020} ------------ The cells were maintained at 37°C with 5% CO~2~. Three human CCA cell lines were used: HIBEpic (ScienCell, Carlsbad, CA, USA), HuCCT1 (JCRB, Osaka, Japan) and RBE (The Institute of Biochemistry and Cell Biology, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, Shanghai, China). The cells were maintained in RPMI-1640 medium (Invitrogen, Waltham, MA, USA) containing 10% fetal bovine serum (Biological Industries, Cromwell, CT, USA), penicillin (Invitrogen) (100 U/ml), and streptomycin (Invitrogen) (100 U/ml). N-acetyl-L-cysteine (NAC, A7250) and CPI-613 (S2776) were purchased from Sigma-Aldrich (St. Louis, MO, USA) and Selleck (Houston, TX, USA), respectively. Tm (MCE, Monmouth Junction, NJ, USA) was commercially purchased. Cell Transfection {#s0025} ----------------- SiRNAs targeting cMYC (5′-GUCAAGAGGCGAACACACA-3′), PDHA1 (5′-CCGAATGGAGTTGAAAGCAGAT-3′), and negative control siRNA were purchased from RiboBio (Guangzhou, China). The full-length cMYC and PDHA1 plasmids were gifts from the Zhao lab of Fudan University (Shanghai, China). HuCCT1 and RBE cells were transfected using Lipofectamine RNAiMax reagent (Invitrogen) according to the manufacturer's instructions. Cell Viability Assay {#s0030} -------------------- Cell viability was investigated by CCK-8 assay in 96-well plates (5×10^3^ cells/well). Different concentrations of Tm or 0.1% dimethyl sulfoxide (DMSO) were added at indicated times. A total of 10 μl CCK-8 solutions (Dojindo, Minato-ku, Tokyo, Japan) was added to each well and they were incubated at 37 °C for 1 h. Absorbance was recorded at 450 nm. For each experimental trial, wells were conducted in triplicate and each well was assayed in triplicate. Migration and Invasion Assays {#s0035} ----------------------------- Cellular motility and invasive abilities were determined using Transwell (Corning Life Sciences, Bedford, MA, USA) and Matrigel invasion (BD Biosciences, San Jose, CA, USA), respectively. For the transwell migration and invasion assay, 5 × 10^4^ cells and 2 × 10^5^ cells were seeded, respectively. The cells that migrated to the underside of the membrane were fixed and stained with 0.1% crystal violet, and 10 microscopic fields were counted. The average values of the migrating or invading cells were expressed as percentages. Each experiment was performed in replicate inserts, and the average value was calculated from three independent experiments. Colony Formation Assay {#s0040} ---------------------- CCA cells were counted and seeded in 6-well plates with 500 cells/well. After treatment with 25 μM Tm for 24 hours, the Tm was removed and normal medium was added for 14 days. The cells were fixed in methanol and colonies were counted after staining with 0.1% crystal violet in 25% methanol for 20 mins. The results were presented as the average number of counted colonies per well under each condition. All the experiments were performed in triplicate and repeated three times independently. Apoptosis Assay {#s0045} --------------- Apoptosis was detected by flow cytometry using the Annexin V-fluorescein isothiocyanate (FITC) Apoptosis Detection Kit (556, 547, BD Biosciences), following the manufacturer's instructions. Western Blotting {#s0050} ---------------- The cells were lysed using RIPA buffer (150 mM NaCl, 50 mM Tris-HCl at pH 7.4, 1 mM EDTA, 0.1% SDS, 1% Triton X-100, 1% sodium deoxycholate and 1% NP-40) mixed with a protease and phosphatase inhibitor cocktail (Roche Diagnostics GmbH, Mannheim, Germany) and phenylmethylsulfonyl fluoride (PMSF) (Biosharp, Hefei, China) for 15 min on ice. The proteins were subjected to western blotting according to standard protocols. Antibodies were as follows: SIRT2 (HPA011165,Sigma), SIRT2 (s8447, Sigma), cMYC (ab32072, Abcam, Cambridge, UK), p293-PDHA1 (ab92696, Abcam), GAPDH (ab128915, Abcam), cleaved-caspase3 (9664s, CST, Danvers, MA, USA), cleaved-PARP (5625s, CST), PDHA1 (ab110334, Abcam), PHGDH (14719-1-AP, Proteintech, Chicago, IL, USA), PSAT1 (10501-1-AP, Proteintech), PSPH (14513-1-AP, Proteintech), HRP-conjugated anti-rabbit (7074, CST), and anti-mouse antibodies (7076, CST). Metabolite Analysis {#s0055} ------------------- Metabolism experiments were carried out as described previously [@bb0110]. Metabolite extracts were collected. A total of 2 μl of metabolite extracts were injected for gas chromatography-mass spectrometer (GC-MS) analysis using an Agilent 6980 GC coupled to an Agilent 5973 MS system. Relative metabolite abundances were investigated by comparing the abundance of each metabolite with internal standards and cell protein standards. Mitochondrial Oxidative Phosphorylation Analysis {#s0060} ------------------------------------------------ The real-time oxygen consumption rate (OCR) was detected using an XF96 Extracellular Flux Analyzer (Seahorse Bioscience, North Billerica, MA, USA). According to the manufacturer's instructions, the cells were seeded at 1x10^4^ cells/well in 96-well XF cell culture microplates and incubated at 37 °C for 24 h. A total of 25 μM Tm was added to each well of each plate. Before measurement, the medium was replaced with 175 μl/well XF-96 running medium (supplemented with RPMI-1640 without serum) and pre-incubated at 37°C without CO~2~ for 20 min. For each analysis, different compounds that modulate mitochondrial respiration were injected into each well, according to standard protocols: for the OCR -- oligomycin, carbonylcyanide p-trifluoromethoxy-phenylhydrazone, rotenone and antimycin A; for the extracellular acidification rate (ECAR) -- glucose, oligomycin and 2-deoxy-D-glucose. Cell number was used for data normalization. The OCR was expressed as picomoles/minute (pmol/min). The ECAR was expressed as milli-pH units per minute (mpH/min). ROS and Glutathione Measurement. {#s0065} -------------------------------- For cellular ROS the analyses, 3 × 10^5^ cells/well were seeded into a six-well plate and different concentrations of Tm were added at the indicated times. After treatment, the redox-sensitive probe DCF (5 μM) was incubated for 30 min. Fluorescence intensity was measured immediately by flow cytometry analysis in the six-well plate-treated cells. The levels of GSH and GSSG were measured in the cells or tumor lysates according to the GSH and GSSG Assay Kit instructions (S0053, Beyotime Biotechnology, Jiangsu, China). The experiments were conducted in triplicate and repeated three times independently. Differentially expressed genes (DEGs) of Paired-CCA from the Cancer Genome Atlas (TCGA) Data {#s0070} -------------------------------------------------------------------------------------------- The CCA RNA-Seq data was downloaded from the TCGA database using the GDC Data Portal at <https://gdc-portal.nci.nih.gov>. The mRNA expression data included 18 samples consisting of 9 normal samples and 9 matched CCA samples. The sequencing data were public, and ethical issues were not relevant. The edgeR package in Bioconductor was used to screen the DEGs in the CCA and normal tissue samples. The edgeR package is based on a negative binomial (NB) distribution, which corrects the problem of overdispersion in RNA-Seq data using the Poisson model and the Bayes procedure. Data with a value of zero were deleted. If \|FoldChange\| \>2, the genes were considered to be DEGs with a *P* value \<.01 and a false discovery rate (FDR) \<0.05. Functional Annotation {#s0075} --------------------- The Database for Annotation Visualization and Integrated Discovery (DAVID) online tool (<https://david.ncifcrf.gov/>) was used to perform functional and pathway enrichment analyses in our study. GO and KEGG pathway enrichment assays were performed to detect the potential biological functions and pathways of the high- and low-expression genes in CCA. Immunohistochemistry {#s0080} -------------------- Two to five human CCA specimens from one patient were used in the immunohistochemistry (IHC) study. Tumor specimens from humans and mice were fixed in 4% paraformaldehyde and embedded in paraffin. Except for target detection, the IHC followed standard procedures. The intensity of staining in each tissue was evaluated by multiplying the staining's range and intensity [@bb0115]. CCA Cancer Xenograft Model {#s0085} -------------------------- Female nude mice were subcutaneously injected in the flanks with 100 μL PBS or 2 × 10^6^ HuCCT1 cells suspended in 100 μL Matrigel. The mice were allowed to recover after the injections and monitored for three weeks. Tumors were measured using a caliper and once the majority of tumors reached a threshold size of 200 mm^3^, intraperitoneal (IP) injections of DMSO or Tm were initiated. IP injections of 1.5 mg Tm in 50 μL DMSO were administered once a day. After one month of treatment, or if the mice met the humane endpoint criteria, they were euthanized via CO~2~ asphyxiation. Tissues were collected, fixed with 10% neutral formaldehyde, embedded in paraffin, sectioned, and stained with hematoxylin and eosin (HE). Serum, organs, and tumor tissues were frozen in liquid nitrogen and then stored at -80°C for subsequent analyses. Statistics {#s0090} ---------- All the data were presented as means ± SEM. The data were analyzed using one-way ANOVA followed by post hoc Duncan tests (GraphPad Prism version 5.0, GraphPad Software, La Jolla, CA, USA). *P* \< .05 was considered significant. Results {#s0095} ======= SIRT2, cMYC, and p-PDHA1 contributed to CCA metabolic reprogramming and were associated with poor prognosis {#s0100} ----------------------------------------------------------------------------------------------------------- Metabolic reprogramming can be regulated through the expression of multiple genes to accelerate the malignant behavior of tumor cells [@bb0120]. Therefore, differentially expressed genes (DEGs) in CCA were analyzed using data from the Cancer Genome Atlas (TCGA) and the most relevant DEGs associated with metabolism and apoptosis were confirmed ([sFigure 1](#f0040){ref-type="graphic"}*A*). As a transcription factor and downstream protein of SIRT2, cMYC has attracted extensive interest for its potential role in tumorigenesis in many different human cancers -- especially for its role in regulating the expression of metabolic genes such as lactate dehydrogenase (LDH) and muscle-specific pyruvate kinase 2 (PKM2) [@bb0125], [@bb0130]. Furthermore, PDH is a gatekeeper of tricarboxylic acid cycle (TCA) and is always highly phosphorylated, thereby leading to an inactive state and contributing to the Warburg effect [@bb0135]. In order to identify the relevant protein targets of metabolic reprogramming, we first screened the cMYC-related targets in a normal bile duct epithelial cell line (Hibepicc) and two human CCA cell lines (HuCCT1 and RBE). We confirmed high levels of p-PDHA1, SIRT2, and its upstream target cMYC in CCA cells ([Figure 1](#f0005){ref-type="fig"}*A*). Because the Warburg effect is the key feature of tumor metabolism, the metabolic reprogramming changes were then evaluated by over-expressing SIRT2 in CCA cells. We found that while SIRT2 decreased the levels of TCA-related metabolites, it increased the level of lactate, thereby promoting the Warburg effect ([Figure 1](#f0005){ref-type="fig"}*A* and [sFigure 1](#f0040){ref-type="graphic"}*B*). Overall, 48 CCA samples were used and the clinical characteristics were summarized in [Table 1](#t0005){ref-type="table"}. We found that the expressions levels of SIRT2, cMYC, and p-PDHA1 were elevated in CCA tissues compared to those in adjacent tissues ([Figure 1](#f0005){ref-type="fig"}, *C* and *D*). In addition, these elevated expressions were associated with lower survival rates ([Figure 1](#f0005){ref-type="fig"}*E*). Collectively, these results revealed that high levels of SIRT2, cMYC, and p-PDHA1 contributed to CCA metabolic reprogramming and were associated with a poor prognosis. PDHA1 is the Downstream Target of SIRT2/cMYC, But Is Unnecessary for CCA Apoptosis {#s0105} ---------------------------------------------------------------------------------- PDHA1 is the downstream target of SIRT2/cMYC, but is unnecessary for CCA apoptosis. As a well-known metabolic regulator, cMYC can be stabilized by the deacetylation activity of SIRT2 [@bb0095]. We confirmed that SIRT2 stabilized cMYC protein through post-translational modification in CCA ([sFigure 1](#f0040){ref-type="graphic"}*B* and 1C). To further investigate how SIRT2 regulates metabolic reprogramming, Tm (a SIRT2 inhibitor) was added to CCA cells. After the inhibition of SIRT2, cMYC and p-PDHA1 decreased while apoptosis-related proteins (c-Caspase 3 and c-PARP) increased ([Figure 2](#f0010){ref-type="fig"}*A*). Furthermore, after knocking-down cMYC in CCA cells, we found that p-PDHA1 and apoptosis-related proteins decreased ([Figure 2](#f0010){ref-type="fig"}*B*). Also, rescue experiments showed over-expressing cMYC help CCA cells resist apoptosis in the presence of TM ([Figure 2](#f0010){ref-type="fig"}*C*). This suggests that PDHA1 is the downstream of the SIRT2/cMYC pathway. Conversely, neither CPI613 (a PDHA1 inhibitor) nor PDHA1 knockdown could significantly change the levels of apoptotic proteins in CCA cells ([Figure 2](#f0010){ref-type="fig"}, *D* and *E*). Additionally, the inhibition of PDHA1 led to a reduction of the OCR, which suggests a decrease in TCA ([Figure 2](#f0010){ref-type="fig"}*F*). The aforementioned results together suggest that PDHA1 is the downstream target of SIRT2/cMYC by regulating glucose metabolism in CCA. The inhibition of PDHA1 alone was insufficient to promote apoptosis in CCA. SIRT2 Inhibition Reduces CCA Biological Activity {#s0110} ------------------------------------------------ To observe the effects of SIRT2 inhibitor on the biological activity of CCA, HuCCT1 and RBE cells were treated with indicated Tm. After the inhibition of SIRT2, CCA cells showed marginal differences in phenotypes ([Figure 3](#f0015){ref-type="fig"}*A*). The cell growth was consistently affected after Tm treatment ([Figure 3](#f0015){ref-type="fig"}*B*). In accordance with these results, colony formation in CCA cells decreased when treated with Tm ([Figure 3](#f0015){ref-type="fig"}*C*). Additional transwell assays were then performed, and when SIRT2 was inhibited, a reduction in cell migration and invasion was observed ([Figure 3](#f0015){ref-type="fig"}*D*). Next, flow cytometry was used to study the mechanism by which Tm inhibited CCA cell proliferation. It was found that Tm enhanced apoptosis in CCA cells ([Figure 3](#f0015){ref-type="fig"}, *E* and *F*). This indicates that Tm, the SIRT2 inhibitor, can reduce the biological activity of CCA cells.Figure 1**High expression levels of SIRT2, cMYC, and p-PDHA1 were found in CCA and are associated with poor prognosis.**(A) The expression levels of SIRT2, cMYC, and p-PDHA1 in a normal bile duct epithelial cell line (HIBEpic) and two human CCA cell lines (HuCCT1 and RBE) were determined by western blotting. (B) CCA cells were overexpressed with either SIRT2 plasmids or control vectors; the cells were then collected and lysed and the concentrations of metabolites were measured by MS. (C & D) The SIRT2, cMYC, and p-PDHA1 protein levels in the tumors and adjacent tissues from 60 CCA patients were detected by IHC and quantified. The magnification is ×200. Scale bars, 100 μm. (E) The 3-year survival was evaluated in CCA patients with SIRT2, cMYC, and p-PDHA1 protein expressions, respectively. Data represent mean ± SEM, n ≥ 3. \*\**P* \< .01.Figure 1Figure 2**PDHA1 is the downstream target of SIRT2/cMYC.**(A) HuCCT1 and RBE cells were treated with 0, 25, or 50 μm Tm and subjected to western blotting. (B & C & D & E) HuCCT1 and RBE cells were transfected with siRNA of cMYC, plasmids of cMYC, CPI-613, Tm, or siRNA of PDHA1, respectively, and subjected to western blotting. (F) The OCRs were detected at different time points. The OCR was under oligomycin, carbonyl cyanide-m-chlorophenylhydrazone (FCCP), and antimycin A/rotenone treatments, respectively. Data represent mean ± SEM, n ≥ 3.Figure 2Table 1Clinical characteristics of patients with cholangiocarcinomaTable 1nLow (%)High (%)X^2^*P* valuecMYC expression, n (%)Gender0.333.564Male2411 (45.8)13 (54.2)Female2413 (54.2)11 (45.8)Age1.371.242≤60208 (40.0)12 (60.0)\>602816(57.1)12 (42.9)Location0.785.675Intrahepatic198 (42.1)11 (57.9)Hepatic portal95 (55.6)4 (44.4)Extrahepatic2011 (55.0)9 (45.0)Size (mm)0.403.525≥40146 (42.9)8 (57.1)\<403418 (52.9)16 (47.1)Differentiation6.155.046Well164 (25.0)12 (75.0)Medium2818 (64.3)10 (35.7)Poor42 (50.0)2 (50.0)T Stage0.949.330T1\~T23519 (54.3)16 (45.7)T3\~T4135 (38.5)8 (61.5)Lymph node metastasis0.403.525Negative3418 (52.9)16 (47.1)Positive146 (42.9)8 (57.1)Distant metastasis1.091.296Negative4423 (50.0)21 (50.0)Positive41 (50.0)3 (50.0)Venous invasion0.085.771Negative2714 (51.9)13 (48.1)Positive2110 (47.6)11 (52.4)Nerve invasion1.613.204Negative145 (35.7)9 (64.3)Positive3419 (55.9)15 (44.1)*SIRT2 expression, n (%)*Gender0.784.376Male2413 (54.2)11 (45.8)Female2416 (66.7)8 (33.3)Age0.002.960≤602012 (60.0)8 (40.0)\>602817 (60.7)11 (39.3)Location0.843.656Intrahepatic1913 (68.4)6 (31.6)Hepatic portal95 (55.6)4 (44.4)Extrahepatic2011 (63.6)9 (36.4)Size (mm)0.089.766≥40148 (57.1)6 (42.9)\<403421 (61.8)13 (38.2)Differentiation0.463.793Well169 (56.3)7 (43.8)Medium2818 (64.3)10 (35.7)Poor42 (50.0)2 (50.0)T Stage6.553.010T1\~T23525(71.4)10 (28.6)T3\~T4134 (30.8)9 (69.2)Lymph node metastasis0.089.766Negative3421 (61.8)13 (38.2)Positive148 (57.1)6 (42.9)Distant metastasis6.660.010Negative4429 (65.9)15 (34.1)Positive40 (0.0)4 (100.0)Venous invasion0.167.683Negative2717 (63.0)10 (37.0)Positive2112 (57.1)9 (42.9)Nerve invasion1.002.317Negative1410 (71.4)4 (28.6)Positive3419 (55.9)15 (44.1)Gender4.269.039*SIRT2 expression, n (%)*Male2411 (45.8)13 (54.2)Female2418 (75.0)6 (25)Age3.407.065≤60209 (45.0)11 (55.0)\>602820 (71.4)8 (28.6)Location3.220.200Intrahepatic1910 (52.6)9 (47.4)Hepatic portal94 (44.4)5 (55.6)Extrahepatic2015 (75.0)5 (25.0)Size (mm)0.897.344≥40147 (50.0)7 (50.0)\<403422 (64.7)12 (35.3)Differentiation3.388.184Well167 (43.8)9 (56.3)Medium2820 (71.4)8 (28.6)Poor411(50.0)10 (50.0)T Stage4.365.037T1\~T23518 (48.6)17 (51.4)T3\~T41311 (84.6)2 (15.4)Lymph node metastasis2.859.091Negative4425 (56.8)19 (43.2)Positive44 (100.0)0 (0.0)Distant metastasis01Negative5628 (50.0)28 (50.0)Positive42 (50.0)2 (50.0)Venous invasion0.167.683Negative2717 (63.0)10 (37.0)Positive2112 (57.1)9 (42.9)Nerve invasion5.043.025Negative145 (35.7)9 (64.3)Positive3424 (70.6)10 (29.4)Figure 3**SIRT2 inhibitor reduces CCA biological activity.**(A) HuCCT1 and RBE cells were treated with or without 25 μM Tm for 24 h or 72 h and the morphological changes were observed. Scale bars, 100 μm. (B) CCA cells were treated with 0-100 μM Tm for 0-72 h, and the cell proliferation was then quantified via CCK-8 assays. (C) CCA cells were treated with or without 25 μM Tm; colonies were stained with crystal violet (left) and quantified (right). Scale bars, 1 cm. (D) CCA cells were treated with or without 25 μM Tm; cell invasion was quantified. (E & F) CCA cells were treated with 0, 25, or 50 μM Tm; apoptotic cells were measured and quantified using flow cytometry. Data represent mean ± SEM, n ≥ 3. \**P* \< .05, \*\**P* \< .01, and \*\*\**P* \< .001.Figure 3 The SIRT2/cMYC Pathway has Antioxidant Effects on CCA by Promoting the Downstream SSP Pathway {#s0115} --------------------------------------------------------------------------------------------- The imbalance between the antioxidants and pro-oxidants caused by superfluous reactive oxygen species (ROS) can promote apoptosis [@bb0140]. On the other hand, SSP plays a key role in antioxidant (T-GSH/GSSG/GSH) production and anti-apoptotic processes [@bb0145]. Consequently, we investigated whether Tm-induced apoptosis is related to ROS production. We found that Tm increased ROS level in a concentration-dependent manner ([Figure 4](#f0020){ref-type="fig"}*A*). More importantly, Tm decreased antioxidant (T-GSH/GSSG/GSH) levels, which can be reversed by NAC (acetylcysteine, a known antioxidant) ([Figure 4](#f0020){ref-type="fig"}*B*). We also found that Tm treatment decreased serine synthesis and the SSP pathway ([Figure 4](#f0020){ref-type="fig"}, *C* and *D*). Our results further revealed that Tm treatment inhibited the SSP pathway in both mRNA and protein levels ([Figure 4](#f0020){ref-type="fig"}, D--*F*, *H* and *I*). Similar to Tm treatment, cMYC knockdown also significantly decreased the levels of SSP pathway proteins in CCA cells ([Figure 4](#f0020){ref-type="fig"}, *G* and *J*). These results suggest that the SIRT2/cMYC pathway inhibited oxidative stress by promoting the SSP pathway and antioxidant production.Figure 4**SIRT2/cMYC has antioxidant effects on CCA by promoting the downstream SSP pathway.**(A) HuCCT1 and RBE cells were treated with 0, 25, or 50 μM Tm and the levels of ROS were detected. (B) CCA cells were treated with NAC, Tm, or a combination of both. The levels of T-GSH, GSSG, and GSH were detected. (C & D) CCA cells were treated with or without indicated Tm. The levels of pyruvate and serine were detected by MS. The relative mRNA levels of PHGDH, PSAT1, and PSPH were detected by qPCR. (E) The expressions of PHGDH, PSAT1, and PSPH in the HIBEpic and CCA cells were determined by western blotting. (F) CCA cells were treated with 0, 25, or 50 μM Tm and the levels of PHGDH, PSAT1, and PSPH were detected by Western blotting. (G) CCA cells were transfected with siRNA of cMYC and subjected to western blotting. Data represent mean ± SEM, n ≥ 3. \**P* \< .05 and \*\**P* \< .01Figure 4 SIRT2/cMYC Pathway Inhibits Oxidative Stress-Related Apoptosis by Promoting Antioxidants Production {#s0120} --------------------------------------------------------------------------------------------------- Studies have found that both lowering glycolytic metabolism and promoting antioxidant production can lead to tumor apoptosis [@bb0140], [@bb0150]. As previously demonstrated, the SIRT2/cMYC pathway promotes metabolic reprogramming in CCA through two signaling pathways: 1) promoting p-PDHA1 to inhibit oxidative phosphorylation and 2) promoting serine synthesis by activating the SSP pathway to increase antioxidant production. To evaluate whether the SSP pathway plays a key role in CCA cell apoptosis, NAC (an antioxidant) was used in combination with Tm. Our results revealed that NAC treatment alone did not affect apoptosis in CCA cells, though Tm treatment alone was able to promote apoptosis. The combination of Tm and NAC, however, led to a significant decrease in apoptosis as compared with Tm used alone ([Figure 5](#f0025){ref-type="fig"}, *A* and *B*). Accordingly, ROS levels did not change in the NAC-treated cells and increased in the Tm-treated cells, and the combination of Tm and NAC led to a decrease in ROS compared to Tm used alone ([Figure 5](#f0025){ref-type="fig"}*C*). Detection of apoptosis proteins showed similar results ([Figure 5](#f0025){ref-type="fig"}*D*), suggesting that the SIRT2/cMYC pathway reduced oxidative stress-induced apoptosis by promoting the downstream SSP pathway.Figure 5**The SIRT2/cMYC pathway inhibits oxidative stress-related apoptosis by promoting antioxidant production.**(A & B) After treatment with Tm, NAC, or a combination of both, CCA cells were analyzed using Annexin V/PI staining and apoptosis was measured by flow cytometry. Percentages of early and late apoptotic cells were quantified. (C & D) After treatment with Tm, NAC, or a combination of both, the levels of ROS and proteins were detected. Data represent mean ± SEM, n ≥ 3. \**P* \< .05 and \*\**P* \< .01.Figure 5 SIRT2 Inhibitor has an Antitumor Effect on CCA *In Vivo* {#s0125} -------------------------------------------------------- Finally, the expression levels of SIRT2, cMYC, and p-PDHA1 were assessed in fresh human CCA tissues. We found that these proteins significantly increased in tumor tissues as compared to that in adjacent tissues ([Figure 6](#f0030){ref-type="fig"}*A*). To further validate our findings, the anti-tumor effects of Tm were evaluated *in vivo* using a CCA cell tumor xenograft model. We observed that the Tm treatment group significantly inhibited tumor growth ([Figure 6](#f0030){ref-type="fig"}, *B* and *C*) without affecting mouse body weight ([Figure 6](#f0030){ref-type="fig"}*D*), suggesting that Tm treatment was safe *in vivo*. Lastly, the expression levels of these proteins were assessed in fresh xenograft tissues by both staining and western blotting, and we found that cMYC, p-PDHA1, PHGDH, PSAT1, and PSPH were markedly down-regulated following Tm treatment ([Figure 6](#f0030){ref-type="fig"}, *E* and *F*). These findings indicated that by inhibiting SIRT2 activity, Tm can degrade cMYC to decrease p-PDHA1 and SSP, thus exerting an anti-tumor effect *in vivo*.Figure 6**SIRT2 inhibitor has an anti-tumor effect on CCA *in vivo*.**(A) The levels of cMYC, p-PDHA1, and SIRT2 proteins in tumors and adjacent normal tissues from 6 donors were detected by western blotting. (B) HuCCT1 cells were injected subcutaneously into the flanks of nude mice. When tumors were palpable, DMSO (control) or Tm were administered. Tumors were collected and photographed. Scale bars, 1 cm. (C) Tumor volumes of mice were measured. (D) The body weights of mice were compared among the groups. (E & F) cMYC, p-PDHA1, PHGDH, PSAT1 and PSPH protein levels in xenograft tissues from control and Tm group were detected by staining and Western blotting. Data represent mean ± SEM, n ≥ 3. \**P* \< .05 and \*\**P* \< .01.Figure 6 Discussion {#s0130} ========== Metabolic reprogramming of tumor cells includes abnormal metabolism in glucose and other substances such as amino acids and fats. Among these, the metabolic reprogramming of glucose is called the Warburg effect. Nonetheless, the question of how metabolites in metabolic reprogramming help induce tumorigenesis and which proteins are key targets for this metabolism remains unanswered. In this study, we identified the promoting role of the SIRT2/cMYC pathway in CCA, identified its downstream targets, and evaluated its therapeutic effect ([Figure 7](#f0035){ref-type="fig"}).Figure 7**Schematic model of the effects of SIRT2/cMYC pathway in CCA cells.**SIRT2 and cMYC reprogrammed the metabolism of CCA cells by up-regulating downstream p-PHDA1 and PHGDH/PSAT1/PSPH. The metabolic reprogramming increased SSP and decreased mitochondrial oxidative phosphorylation, thus contributing to redox homeostasis, consequently protecting CCA cells from oxidative stress-induced apoptosis.Figure 7 SIRT2 is a highly conserved enzyme that widely exists across species, from bacteria to humans. The removal of the acetyl group from lysine residues is coupled with the hydrolysis of NAD to generate nicotinamide, lysine, and O-acetyl-ADP-ribose. SIRT2 is responsible for catalyzing many biological processes, including genetic control, development and metabolism. Many previous studies have shown that SIRT2 increases the migration and invasion of cancer cells, demonstrating that it may play a role in promoting cancer proliferation and metastasis [@bb0155], [@bb0160]. However, some studies have shown that SIRT2 may be a tumor-suppressor gene in various other cancers [@bb0165]. Therefore, SIRT2 might promote tumor progression in certain cases, such as human gastric cancer and pancreatic cancer, while under other conditions it may inhibit tumor progression. It is worth noting that LDH-A [@bb0170], tubulin [@bb0175], and phosphoenolpyruvate carboxykinase (PEPCK) [@bb0180] are substrates of this deacetylase, suggesting that SIRT2 affects different metabolism in different tumors to promote the biological behaviors of various tumors. In fact, after inhibiting SIRT2 in CCA cells, two significant changes in different metabolic pathways are observed: one is glycolysis reversal by promoting PDHA1 phosphorylation (p-PDHA1) to inhibit oxidative phosphorylation and the other is promotion of the SSP pathway via aiding serine synthesis activity. Interestingly, we found that the reversal of PDHA1 phosphorylation alone was insufficient to induce significant apoptosis. This indicates that although glucose metabolism changes induced by SIRT2 are observed in CCA apoptosis, a high degree of serine anabolic metabolism provides an antioxidant effect against stress, which plays a more important role. There are many kinds of SIRT2 inhibitors, mainly including nicotinamide, which inhibits NAD^+^-dependent reactions [@bb0185], and competitive inhibitors such as sirtinol, spitomicin, and cambinol [@bb0190]. Researchers have found that SIRT2 inhibitors have anti-tumor [@bb0125], anti-inflammatory [@bb0195], and anti-diabetic properties. SIRT2 shares a common structure with other members of the SIRT family, such as SIRT1, which is one of the most widely studied SIRTs and regulates cell migration, apoptosis, and proliferation [@bb0200]. Therefore, the off-target effects of inhibitors should not be ignored. Consequently, a specific inhibitor (Tm) was used. It inhibited SIRT2 with an IC50 value of 0.028 mM and SIRT1 with an IC50 value of 98 mM, but did not inhibit SIRT3, even at 200 mM [@bb0125]. Hence, the inhibition of SIRT2, rather than other SIRTs, may play a leading and more efficient role in reversing metabolic reprogramming. In addition, in the CCA tumorigenesis mouse model in which SIRT2 was inhibited by Tm, tumor growth was delayed without significant side effects. As an oncogene, *cMYC* has received much attention as a result of its potential role in promoting tumorigenesis. There are many mechanisms by which cMYC induces tumorigenesis, and the enhancement of the Warburg effect is one method. This study found that cMYC is a crucial downstream target of SIRT2. Many studies have also elucidated that cMYC can elevate metabolic proteins such as LDHA and PKM2 [@bb0130], and is therefore considered a promising anti-cancer target. The present study demonstrates that cMYC has an elevated expression in CCA and predicts an unfavorable prognosis. In addition, cMYC is positively associated with p-PDHA1 and SSP-related enzymes and contributes to a high serine level. Our work has determined that the inhibition of cMYC can be used to reduce serine, thereby effectively promoting apoptosis in CCA cells. Glucose is the main source for serine biosynthesis [@bb0205]. However, little is known about how glycolysis intertwines with SSP activation to coordinate cell survival and proliferation. Previous studies have reported that increased SSP activation was observed in breast cancer during serine starvation to maintain cell survival [@bb0210]. Herein we confirmed the importance of SSP activation in CCA proliferation. We also observed that by inhibiting the SIRT2/cMYC pathway, the reduced glycolytic metabolism changed to an elevated serine synthesis metabolism. These results suggest that high serine anabolic metabolism, which provides antioxidants for stress resistance, is essential for preventing apoptosis in CCA cells. We further observed that SIRT2/cMYC and its downstream targets may represent the intersecting points of metabolic reprogramming between glycolysis and SSP, highlighting the importance of SIRT2 for cell survival under redox imbalance conditions. Our data show that the SIRT2/cMYC pathway plays a crucial role in the conversion of glucose oxidative metabolism to serine anabolic metabolism, thus providing antioxidants for stress resistance. SIRT2/cMYC-induced metabolic reprogramming may thus represent a new target for the treatment of CCA. The following are the supplementary data related to this article.Supplementary Figure 1(A) Differentially expressed genes (DEGs) in CCA were analyzed using data from The Cancer Genome Atlas (TCGA). (B & C) CCA cells were treated with or without indicated Tm, MG132 or CHX for 24 hours. Cells were harvested and lysed, and endogenous cMYC was visualized by Western blotting. Data represent mean ± SEM, n ≥ 3.Supplementary Figure 1 Author contributions {#s0135} ==================== Mingming Zhang, Lei Wang, and Xiaoping Zou designed the study. Lei Xu, Dehua Tang, and Yida Pan conducted the cell experiments. Yuming Wang and Lei Xu collected the tissue samples. Lei Xu and Mingming Zhang performed the protein analysis. Lixing Zhou, Yuming Wang, Mingming Zhang, and Dehua Tang drafted the manuscript and conducted the immunohistochemistry experiments. Lei Xu performed the metabolite analysis. Robert G. Dorfman and Mingming Zhang wrote the manuscript. Mingming Zhang and Xiaoping Zou supported the study. All of the authors read and approved the final manuscript. Acknowledgments {#s0140} =============== We thank the Zhao lab for their assistance. Grant support {#s0145} ============= This work was supported by grants from the National Natural Science Foundation of China (Nos. 81602076, 81672935, and 81472756), the Outstanding Youth Project of Nanjing City (No. JQX17002), the Jiangsu Clinical Medical Center of Digestive Disease (BL2012001), the Natural Science Foundation from the Department of Science & Technology of Jiangsu Province (BK20160113), the Fund of Jiangsu Provincial Commission of Health and Family Planning (No. Q201611) and the Fundamental Research Funds for the Central Universities (No. 021414380244). Conflicts of interest {#s0150} ===================== The authors declare no conflicts of interest. [^1]: These authors contribute equally to this work.

Introduction: tracking energy expenditure in athletes {#s1} ===================================================== One of the unique characteristics of athletes is that energy requirements of training and competition increase their total daily energy expenditure (TDEE) beyond those of the general population (Westerterp, [@B52]). Energy requirements can vary considerably depending on exercise type, intensity, and duration, but sustained levels of energy expenditure (EE) can be in the range of 5,000--8,000 kcal/day (Westerterp et al., [@B53]; Westerterp, [@B51]). This high energy turnover has implications not only for weight gain and weight loss practices, which are prominent in sports with weight classes, anti-gravitational sports, or aesthetic sports; it also necessitates a sufficient dietary energy intake, as sustained energy deficiency can result in long-term detriments including impaired bone health and infertility (Loucks et al., [@B30]). In addition, recent data suggest that athletic performance may also be impaired in energy-deprived athletes (Vanheest et al., [@B44]). Because of the high energy demands and the consequences of energy deficiency, tracking EE is paramount for many athletes and their support staff. Considering that athletes expend up to 75% of their TDEE during exercise (Westerterp, [@B52]), quantifying energy needs during training and competition requires particular attention. The current gold-standard method for the assessment of TDEE in free-living situations is the doubly labeled water (DLW) method, which has been used in numerous athletic settings (Westerterp et al., [@B53]; Sjödin et al., [@B40]; Trappe et al., [@B43]; Hill and Davies, [@B20], [@B21]; Ebine et al., [@B15]; Ekelund et al., [@B16]; Koehler et al., [@B26]). However, the time resolution is limited and the method does not differentiate between various components contributing to TDEE, such as exercise energy expenditure (ExEE) (Westerterp et al., [@B53]). Improved resolution is provided by indirect calorimetry (IC), the reference method for EE quantification in controlled laboratory settings (Haugen et al., [@B19]). However, despite recent methodological advances, the method remains mostly limited to research and exercise testing. Further, the requirement of a face mask hinders natural training behaviors such as fluid or food intake. Therefore, other approaches that do not interfere with training and competition practices are needed to reliably quantify EE, and particularly ExEE, in athletes. Available methods include accelerometry, pedometry, heart-rate monitors, and self-report methods (Ndahimana and Kim, [@B35]). With the exception of self-report methods, which only provide subjective information and show low accuracy and reliability (Ndahimana and Kim, [@B35]), all of these approaches have been incorporated in activity monitors. These devices are less cost-prohibitive than DLW or IC, can be used during a wide range of activities and numerous settings, and allow for data collection over prolonged time intervals in large cohorts (Düking et al., [@B12]). Several such wearable devices, including the ActiGraph, Actical, RT3, ActivePAL, or GeneActiv, have been developed for research purposes, and various companies have introduced commercial physical activity trackers (e.g., Fitbit, Garmin, Jawbone, Nike). However, as these devices typically rely only on accelerometry, they provide mixed accuracy with regard to its ability to predict EE or time spent in different activities (Welk et al., [@B49]) and the ability to detect when devices are worn may be limited (Jaeschke et al., [@B22]). Technology of the SenseWear armband: features, functions, and modifications {#s2} =========================================================================== The SenseWear armband (SWA) developed by BodyMedia Inc. (Pittsburgh, PA, USA) combines accelerometry with additional biological variables, such as heat flux, skin temperature, near-body ambient temperature, and galvanic skin response. The device only collects data when it is in direct contact with the skin and its pattern-recognition algorithm has been shown to provide more accurate results for estimating EE and time spent in various activities when compared to the ActiGraph (Welk et al., [@B49]). Given these benefits, the SWA became a promising tool to objectively monitor EE in various exercise and non-exercise settings (Fruin and Rankin, [@B17]). Most basic principles and functions have remained the same since the initial introduction of the first prototypes in the late 1990s, but there have been several upgrades, the most notable modification being the addition of a third dimension accelerometer axis (Riou et al., [@B39]) along with increased data transfer and storage capacity. Per manufacturer instructions, the SWA is worn on the upper left arm, and can be used to record data continuously for up to 3--4 weeks (Koehler et al., [@B28]). Data can be downloaded, viewed, and exported for subsequent data processing using manufacturer software (InnerView, BodyMedia, Pittsburgh, PA). A proprietary algorithm converts raw data into estimates of EE, which are expressed both in kcal/min and metabolic equivalents (METs). In efforts to improve the validity of the SWA, this algorithm has been modified several times (Jakicic et al., [@B23]; Van Hoye et al., [@B45]). Although the technology was purchased by a competitor in 2013 and has since been discontinued (Welk et al., [@B50]), the SWA continues to be used extensively in research and clinical settings (Figure [1](#F1){ref-type="fig"}). Considering the continued popularity and the current lack of alternatives on the market, it was our goal to provide a critical review of the applicability of the SWA to measure EE specifically in athletes. As such, we provide a general overview of the strength and limitations of the SWA in the general population (section Validity of the SenseWear Armband in the General Population: Energy Expenditure, Physical Activity, and Exercise), followed by a review of the validity of the SWA in athletes and during various types of high-intensity exercise (section Validity of the SenseWear Armband during High-Intensity Exercise). We further discuss possible reasons for limitations (section Limitations of the SenseWear Armband: Algorithm vs. Methodology) and non-traditional applications of the SWA in athletic settings (section Application of the SenseWear Armband in Athletic Populations). To identify appropriate literature, a quasi-systematic PUBMED search (<https://www.ncbi.nlm.nih.gov/pubmed/>) was conducted in June 2017 independently by both authors, using "SenseWear" in combination with "exercise," "activity," or "athletes" as search terms. In addition, we included literature cited. Final inclusion was decided on by a joint decision from both authors based on each paper\'s relevance to the review\'s target group. {#F1} Validity of the SenseWear armband in the general population: energy expenditure, physical activity, and exercise {#s3} ================================================================================================================ In the general population, the SWA has been validated extensively and has been shown to provide accurate estimates of TDEE as well as EE at rest and during activities of light to moderate intensities when compared to DLW or IC (Cole et al., [@B7]; Fruin and Rankin, [@B17]; Jakicic et al., [@B23]; King et al., [@B25]; Mignault et al., [@B34]; Papazoglou et al., [@B36]; Malavolti et al., [@B32]; Patel et al., [@B37]; St-Onge et al., [@B41]; Johannsen et al., [@B24]; Casiraghi et al., [@B6]; Brazeau et al., [@B5]). When specific time periods of varying activity intensities were examined, however, the SWA generally overestimated EE at lower intensities, while EE was underestimated at higher intensities (Cole et al., [@B7]; Fruin and Rankin, [@B17]; Jakicic et al., [@B23]; Patel et al., [@B37]; Dwyer et al., [@B13]; Berntsen et al., [@B4]; Benito et al., [@B3]; Gastin et al., [@B18]). Accordingly, TDEE was overestimated in participants with low levels of TDEE and underestimated in participants with high TDEE (St-Onge et al., [@B41]; Johannsen et al., [@B24]). It should further be considered that the accuracy of the SWA is impacted by external factors such as treadmill incline, exercise mode (e.g., running vs. bicycling), or the use of the upper vs. lower body exercise (Fruin and Rankin, [@B17]; Jakicic et al., [@B23]; Berntsen et al., [@B4]; Vernillo et al., [@B47]; Brazeau et al., [@B5]; Gastin et al., [@B18]). Specifically, underestimation of EE during uphill walking has been reported in several studies, with increasing measurement errors at steeper inclines (Fruin and Rankin, [@B17]; Jakicic et al., [@B23]; Vernillo et al., [@B47]). Downhill walking, on the other hand, was associated with an overestimation of EE, and---although less pronounced---measurement errors increased as declines became steeper (Vernillo et al., [@B47]). During stationary cycling, total EE did not differ between the SWA and IC, but individual time point data were poorly correlated: At the beginning of the cycling trial, EE was underestimated, but EE estimates by the SWA increased gradually over time even though IC values remained stable (Fruin and Rankin, [@B17]; Brazeau et al., [@B5]). Further, Gastin et al. ([@B18]) reported an underestimation of EE during resistance type circuit exercise, most likely due to inaccuracies at higher intensities. In addition to problems related to activity type and intensity, body weight has been shown to affect measurement accuracy. Even though no particular bias toward over- or underestimation of EE was observed, measurement error increased with increasing BMI (Dwyer et al., [@B13]; Malavolti et al., [@B31]). Considering that athletes typically are on the extreme ends of the body composition spectrum (Meyer et al., [@B33]), it is unclear to which degree body weight or composition contribute to measurement errors in athletes. Differences in body weight or composition may also contribute to the considerable variability of measurement accuracy at the individual level (Fruin and Rankin, [@B17]; Brazeau et al., [@B5]). Nevertheless, a recent study reported accurate measurements of TDEE with a mean difference of 2.8 kcal/day and narrow 95% confidence intervals (−34.8 to 40.3 kcal/day) and a correlation coefficient of *r* = 0.88 when comparing SWA values to DLW in 191 generally healthy adults with diverse body weight and physical activity levels (Drenowatz et al., [@B11]). Overall, the SWA provides valid estimates of TDEE and ExEE with a measurement error of typically \<10% in a recreationally active population. Validity of the SenseWear armband during high-intensity exercise {#s4} ================================================================ To our knowledge, only one study has assessed the validity of SWA-measured TDEE specifically in athletes. Koehler et al. ([@B27]) reported an average difference of 65 kcal/day (\<2% of TDEE) between TDEE measured by SWA and DLW in 14 endurance trained athletes and a moderate to strong correlation (*r* = 0.73) However, higher levels of TDEE tended to be underestimated by the SWA, and the level of underestimation was related to the participant\'s exercise capacity, whereby EE was underestimated to a greater degree in better trained athletes (Koehler et al., [@B27]). Validity during high-intensity aerobic exercise ----------------------------------------------- Several studies have tested the validity of the SWA during high-intensity, continuous aerobic exercise. In two independent studies in trained male athletes, the SWA underestimated ExEE during treadmill running at speeds of \~10.1 km/h (6.3 miles/h) and greater (Koehler et al., [@B27], [@B28]). These findings were replicated by Drenowatz and Eisenmann ([@B9]), who demonstrated that ExEE was consistently underestimated in endurance-trained athletes running at 65, 75, and 85% of their aerobic capacity, corresponding to a similar speed range (9.9--14.6 km/h; 6.2--9.1 miles/h). In another study, the SWA underestimated ExEE even at speeds from 6.0 to 7.2 km/h (3.7--4.5 miles/h) (van Hoye et al., [@B46]). Similar findings were also reported during stationary bicycling, whereby the SWA underestimated ExEE at workloads between 140 and 380 W (Koehler et al., [@B27]). In all cases, the level of underestimation increased with increasing exercise intensity (Drenowatz and Eisenmann, [@B9]; Koehler et al., [@B27], [@B28]; van Hoye et al., [@B46]). However, visual inspection of the combined data from all five studies (Figure [2](#F2){ref-type="fig"}) suggests that differences between SWA and IC are rather modest at low-to-moderate exercise intensities. At exercise intensities above 35 mL/kg/min (10 METs) SWA-measured ExEE, however, tends to plateau whereas IC-measured ExEE increases continuously, resulting in a stark increase in the level of underestimation. It is noteworthy that all studies employed an incremental exercise test to assess the validity of the SWA at multiple exercise intensities. To our knowledge, only one study separately used a 30 min exercise bout at a self-selected intensity, resulting in a similar level of underestimation of 27% (Drenowatz and Eisenmann, [@B9]). ![Previously published data reporting the discrepancy between energy expenditure measured with the SenseWear armband (black symbols) in comparison to the reference method (indirect calorimetry; open symbols) and the difference between SenseWear and indirect calorimetry (gray symbols). The dotted line depicts an exercise intensity of 35 mL/kg/min (10 METs). Data published by Drenowatz and Eisenmann ([@B9]) stem from 20 male and female runners (VO~2peak~: 57 mL/kg/min); Data published by Koehler et al. ([@B27]) stem from 14 triathletes (VO~2peak~: 58 mL/kg/min) who were assessed while running and biking; Data published by Koehler et al. ([@B28]) stem from 19 endurance and strength trained men (VO~2peak~: 55 mL/kg/min) who were assessed while running; Data from van Hoye et al. ([@B46]) stem from 23 male kinesiology students (VO~2peak~: 69 mL/kg/min) and 20 female kinesiology students (VO~2peak~: 53 mL/kg/min) who were assessed while walking and running; Data published by Van Hoye et al. ([@B45]) stem from 39 male and female kinesiology students (VO~2peak~: 58 mL/kg/min) who were assessed while walking and running.](fphys-08-00983-g0002){#F2} Validity during resistance exercise ----------------------------------- Only few studies have examined the accuracy of the SWA during resistance-type exercise. Benito et al. ([@B3]) reported an underestimation of ExEE during circuit-type resistance training at 30, 50, and 70% of the 15RMmax in a mixed sample of 29 recreationally active participants. Compared to IC, SWA-estimated ExEE was 32% lower in men, corresponding to a difference of 2.3 METs, and 21% lower in women (1.1 METs). Furthermore, the degree of underestimation increased with increasing exercise intensity, although this effect was only significant in men (Benito et al., [@B3]). On the other hand, the SWA slightly overestimated exercise EE by an average 35 kcal per session during self-selected resistance exercise in a mixed sample of 52 participants of varying age and fitness level (Bai et al., [@B2]). The measurement error at the individual level was reported at 15%. However, the average exercise intensity was rather low during these sessions (3.2 METs) and may not resemble a typical resistance exercise session in athletic populations. Using a more traditional resistance training protocol of 9 exercises covering all major muscle groups with 3 sets of 10 repetitions at 70% of the 1-reptition maximum, the SWA provided accurate estimates of ExEE with an error of less than 5% and a strong correlation for ExEE (*r* = 0.77) and TDEE (*r* = 0.97) (Reeve et al., [@B38]). Measurement errors also remained constant across the ExEE spectrum with an almost perfect reliability of the SWA (test-retest *r* = 0.96). It should, however, be considered that ExEE was integrated over the course of the exercise bout; no information was provided on the measurement accuracy for specific exercise types (Reeve et al., [@B38]). Validity during mixed exercise forms ------------------------------------ Similar to studies addressing resistance-type exercise, there has been only limited research examining the accuracy of the SWA during mixed exercise forms, particularly in athletic populations. Zanetti et al. ([@B55]) assessed the accuracy of the SWA during a 42-min sport-specific intermittent exercise trial in 14 male rugby players. While there was no clear trend toward over- or underestimation of ExEE with a mean bias of −0.2 kcal/min (−1.9%), results revealed only a moderate correlation between the SWA and IC (*r* = 0.55). During a 30-min basketball-specific skill session, the SWA, however, was shown to underestimate ExEE by 1.1 kcal/min (15%) (Taylor, [@B42]). EE during recovery period following intermittent exercise training, on the other hand, was overestimated by 17% by the SWA when compared to IC (Zanetti et al., [@B55]). Limitations of the SenseWear armband: algorithm vs. methodology {#s5} =============================================================== Despite the tendency to underestimate ExEE during high-intensity exercise, available data suggest that the SWA can reliably detect activity patterns, rest periods, and varying levels of exercise intensity within individuals. For example, significant intra-individual correlations between IC and SWA was reported in 90% of endurance athletes who ran at exercise intensities between 65 and 85% VO~2max~ (Drenowatz and Eisenmann, [@B9]). In another study involving incremental treadmill running at speeds between 10.8 and 17.3 km/h, raw data including acceleration counts, and particularly counts in the longitudinal plane, increased continuously as workload increased (Koehler et al., [@B28]), demonstrating that the technology is suited to detect movement patterns even at higher exercise intensities. Consequently, limitations to the proprietary algorithm are a candidate source for the underestimation of ExEE during high-intensity exercise. Several studies have tested whether algorithm adjustments could improve the validity of the SWA during exercise. In one of the first published validation studies, Jakicic et al. ([@B23]) reported that the accuracy of the SWA improved after algorithm revisions. After the initial algorithm underestimated ExEE during walking, stepping, and cycling by 7--29% and overestimated ExEE during arm ergometry by 29%, the researchers provided a subset of their data to develop exercise-specific proprietary equations, which reduced errors in ExEE measured by the SWA to a non-significant level. However, ExEE values, which peaked during stair stepping at 5.3--9.2 kcal/min, did not exceed the 10 MET-threshold. More recently, Van Hoye et al. ([@B45]) compared two different algorithms during low- and moderate-intensity treadmill running in well-trained students, reasoning that a newer algorithm would provide more accurate estimates of EE as the manufacturer updates proprietary algorithms on a regular basis. When compared to the initially used algorithm (version 2.2.), data processed using a newer algorithm (version 5.2) reduced the measurement error from 18--24 to 5--17%, although ExEE remained underestimated. Application of the SenseWear armband in athletic populations {#s6} ============================================================ Despite the previously mentioned limitations, several groups have used the SWA to track EE in athletes. In adolescent sprinters undergoing high-intensity exercise training, Aerenhouts et al. ([@B1]) measured TDEE, ExEE, and activity patterns using the SWA. When compared to self-report, the SWA registered less time spent in high-intensity activity, although this difference did not result in differences in TDEE, which was within 6% of the TDEE derived from activity diaries. The authors also highlighted the need for additional information when athletes fail to wear the SWA for 24 h. The SWA was also used to record ExEE during the competitive season in volleyball players (Woodruff and Meloche, [@B54]). SWA-recorded ExEE was found to be higher during games when compared to practice and warm-up sessions. Combining SWA data with diet logs and body composition assessment, the authors further concluded that the majority of the athletes were in an energy-balanced state. Using the SWA to quantify non-exercise activity thermogenesis (NEAT) among endurance athletes undergoing periods of high and low training volume, Drenowatz et al. ([@B10]) demonstrated that the high training volume did not result in a compensatory reduction in NEAT; instead, athletes reduced their sedentary activities to allow for more training time. In professional Australian Football players, the SWA was used to document the contribution of NEAT to TDEE, which was greater on training days (85%) when compared to match days (69%) (Walker et al., [@B48]). Because the SWA can be worn continuously for several days, it has also been used for the assessment of sleep quantity and quality. In male elite rugby union players, SWA-derived sleep duration was shown to be lower during game nights when compared to non-game nights, although sleep efficiency was not different (Eagles and Lovell, [@B14]). In another trial comparing high-intensity interval training to strength training, SWA-derived sleep efficiency was lower in the high-intensity interval condition (Kölling et al., [@B29]). These applications demonstrate that the SWA is well-suited to capture other biological factors, such as characteristics of sleep and NEAT, that may have important implications for athletic performance. Conclusion and summary {#s7} ====================== Considering that the SWA has been designed for a broad market, it is not surprising that the device tends to underestimate ExEE for periods of high-intensity exercise. Although most data has been established for aerobic exercise, the SWA seems to equally underestimate ExEE during other exercise forms. When energy expenditure is integrated over longer time periods, including rest and recovery, the measurement error becomes less pronounced and estimations of TDEE tend to be more accurate, even in athletic populations. Adjustments to the proprietary algorithm that is used to derive EE may further help to improve the validity of the SWA. Unfortunately the sale of the SWA has been terminated. Recently, a new disposable device with similar functionality has been introduced but is not available for commercial application at this time (Welk et al., [@B50]). Another viable option is the combination of GPS data with accelerometry and heart rate to assess EE in outdoor sports (Costa et al., [@B8]), although the accuracy of such devices remains to be explored. Given the current lack of alternatives, the SWA continues to be used in research and practice, emphasizing the need for the continued development of wearable devices that reliably measure EE and related variables in athletic settings. Author contributions {#s8} ==================== All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication. Conflict of interest statement ------------------------------ The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. [^1]: Edited by: Billy Sperlich, Universität Würzburg, Germany [^2]: Reviewed by: Salvatore Tedesco, University College Cork, Ireland; Beat Knechtle, University of Zurich, Switzerland [^3]: This article was submitted to Exercise Physiology, a section of the journal Frontiers in Physiology

{#sp1 .21}

Indoleamine 2,3--dioxygenase 1 (IDO1, also known as tryptophan pyrrolase) is a heme-containing enzyme that catalyzes the oxidative cleavage of L-tryptophan to N-formyl-L-kynurenine in the first and rate-limiting step of tryptophan metabolism[@b1]. The great majority of tryptophan is known to be metabolized by the kynurenine pathway in which tryptophan is degraded through a series of metabolic reactions[@b2][@b3]. It has been reported that IDO1 is ubiquitously expressed in the colon[@b4], small intestine[@b5], brain[@b6], spleen[@b7], lung[@b8] and epididymis[@b9] as well as in myeloid cells including macrophages and dendritic cells[@b10]. Although the highest levels of expression were found in the colonic intestinal tissues[@b4], the specific function of IDO1 in this location has not been fully studied. IDO1 has attracted considerable attention in recent years because of its major roles in non-metabolic functions. The first discovery of the biologic relevance of IDO1 occurred in the placenta, where maternal T cells regulate immunity during gestation. These studies used 1-methyl-tryptophan (1-MT), an inhibitor of IDO1 activity[@b11], and showed that IDO1 regulated T cell activity by suppressing proliferation. In addition, several other studies have shown an immunosuppressive role of IDO1 in conditions involving T cell activation such as graft-versus-host disease (GVHD)[@b12][@b13], HIV infection[@b14] and immunodeficiency diseases[@b15]. It has been shown that enhanced expression of IDO1 increased the production of tryptophan metabolites, which promotes cell-cycle arrest of and apoptosis of T cells and induce the differentiation of T regulatory cells (T~regs~)[@b16]. Despite knowledge of the immune-modulatory role of IDO1, the mechanism by which IDO1 mediates inflammatory reactions is a field of active investigation and remains controversial. A contradictory effect of IDO1 on inflammatory responses during the development of colitis was recently described. Mechanistic studies using trinitrobenzene sulfonic acid (TNBS)-induced colitis suggested that inhibition of IDO1 leads to increased severity of colitis due to down-regulation of T~reg~ cell responses within the intestinal tract[@b17][@b18]. Conversely, another study reported that local increases in IDO1 production during active inflammatory responses resulted in more severe colitis promoted by key mediators of pro-inflammatory signaling[@b19]. Most previous results were drawn from studies of T cell-associated functions of IDO1. For example, TNBS is a chemical that induces colitis in a T cell-dependent manner[@b17] as a result of delayed-type hypersensitivity reaction to haptenized proteins; however, because other studies raised the points that IDO1 may play inflammatory roles in a T cell-independent manner, different approaches are needed to elucidate the role of the enzyme in the development of colitis. One of the potential approaches is to study the roles of IDO1 using dextran sulfate sodium (DSS) which induces colitis by disrupting epithelial barrier function of colon tissues. In this study, we performed for the first time an array-based transcriptome analysis to identify differentially-expressed genes targeted by IDO1 using *Ido1* knock-out (*Ido1*^−/−^) mice. We identified new molecular targets of IDO1 and described functionally distinct molecular mechanisms regulated by IDO1. In addition, we further examined the pathophysiological roles played by IDO1 in colitis development in studies using DSS-induced colitis model. We determined the effects of *Ido1* expression on colitis-related clinical parameters and histopathological damage. We also assessed changes in inflammatory cell recruitment using flow cytometry, and performed gene expression profiling analyses. To our knowledge, this is the first report of a comprehensive gene expression profile analysis of *Ido1*^−/−^ mice in either non- stimulated or DSS-stimulated conditions. This large-scale profiling may enhance our understandings of the contributions of IDO1 to colitis development, and provide novel target genes regulated by IDO1. Results ======= Identification of differentially expressed genes in *Ido1* knock-out mice ------------------------------------------------------------------------- As a first attempt to identify the targets of IDO1, which catalyzes the first and rate-limiting step of tryptophan degradation, transcriptome analysis was performed using *Ido1*^−/−^ and *Ido1*^+/+^ mice. Total RNAs from the colon tissue of each mouse was respectively applied to Illumina Mouse WG-6 v.2 BeadChip arrays containing a total of 45,281 transcripts. Principal component analyses (PCA) confirmed that the gene expression profiles of *Ido1*^−/−^ mice were readily distinguishable from those of *Ido1*^+/+^ mice as shown in [Fig. 1a](#f1){ref-type="fig"}. Differentially expressed genes in *Ido1*^−/−^ and *Ido1*^+/+^ mice were identified by unpaired t-test with thresholds of 1% false discovery rate \[FDR\] and 2-fold change restriction. A total of 102 significant genes were identified with 37 up- and 65 down-regulated genes. We confirmed that transcripts of *Ido1* (IDO1, marked by an asterisk in [Fig. 1b](#f1){ref-type="fig"}) were completely absent in *Ido1*^−/−^ mice. To gain a broader understanding of the biological processes regulated by IDO1, we performed a series of bioinformatic analyses. Hierarchical clustering analysis showed three distinct gene clusters ([Fig. 1b](#f1){ref-type="fig"}). Cluster 1, composed of up-regulated genes in *Ido1*^−/−^ mice was classified further to explore biological implications. Functional classification and Fisher's exact test identified Gene Expression, Embryonic Development, Nervous System Development and Function, Organ Development, and Organismal Development as the top 5 significant biological processes. Accordingly, the expression of many genes in the gene expression-related transcriptional network was significantly increased including *Hoxd10*, *Barx2*, *Msx1*, and *Pbx1* ([Fig. 1c](#f1){ref-type="fig"}). Likewise, the top five molecular and cellular functions of genes in cluster 2 were Inflammatory Response, Dermatological Disease and Conditions, Immunological Disease, Infectious Disease, and Cancer. The most significant category modulated by the absence of IDO1 was Inflammatory Response (P = 5.39E-05). These included *Cd6*, *Scrib*, *Il1rap*, *Cd80*, and *Trem3* in *Ido1*^−/−^ mice, suggesting that IDO1 might be an important mediator in inflammatory response ([Fig. 1d](#f1){ref-type="fig"}). In addition, gene set enrichment analysis (GSEA) confirmed that genes associated with the inflammatory response were significantly enriched in the set of differentially expressed genes between *Ido1*^−/−^ and *Ido1*^+/+^ mice (FDR q \< 0.01, [Fig. 1e](#f1){ref-type="fig"}). These findings prompted us to explore the roles of IDO1 in colonic inflammatory responses. Contribution of IDO1 to development of DSS-induced colitis ---------------------------------------------------------- Based on the array analyses showing that the inflammatory response was the most significant biological process affected by the absence of IDO1 in untreated mice, we next sought to study the effects of IDO1 deficiency on colitis development. We used an established murine model of colitis induced by oral administration of DSS, a reagent that disrupts the barrier function of mucosal epithelial cells[@b20]. *Ido1*^−/−^ and *Ido1*^+/+^ mice were given either 1% or 2% DSS in their drinking water for 7 days and daily water intakes were measured. There was no difference in DSS-containing water consumption among all groups (see [Supplementary Fig. S1](#S1){ref-type="supplementary-material"}). While treatment with 2% DSS resulted in a marked inflammatory reaction, 1% DSS treatment did not produce significant inflammatory reaction compared to the 0% control in *Ido1*^+/+^ mice. In treatment with 2% DSS, progressive weight loss, increased disease activity index (DAI) including severe diarrhea and intestinal bleeding, and greater shortening of colon length were detected in *Ido1*^+/+^ compared to *Ido1*^−/−^ mice ([Fig. 2a--c](#f2){ref-type="fig"}). These results suggest that IDO1 may play roles in promoting the inflammatory response in DSS-challenged mice. In addition, although treatment with 1% DSS did not induce significant differences in body weight and DAIs between *Ido1*^−/−^ and *Ido1*^+/+^ mice, it caused histological changes showing that the extent of tissue damage in *Ido1*^+/+^ mice was much greater than those for *Ido1*^−/−^ mice ([Fig. 2d](#f2){ref-type="fig"}). Inflammatory cell infiltration and crypt damage was more apparent in these mice as evidenced on H&E staining. Furthermore, to explore whether pharmacological inhibition of IDO1 reproduces the results observed following *Ido1* gene deletion, we administered *Ido1*^+/+^ mice with L-1MT, a specific IDO1 inhibitor, then followed by colitis induction with 2% DSS treatment. IDO1 blockade by L-1MT ameliorated severe diarrhea and intestinal bleeding, resulting in significant reduction in the DAI. Shortening of colon length was notably attenuated in mice administered with L-1MT compared to the placebo ([Fig. 2e,f](#f2){ref-type="fig"}). Taken together, our data indicate that the severity of DSS-induced colitis development was significantly reduced in *Ido1*^−/−^ mice and L-1MT administered mice, suggesting that IDO1 deficiency might protect against pro-inflammatory signals. We next carried out transcriptome analysis to elucidate the underlying molecular mechanisms of this effect. Gene expression profiling analysis of inflamed colon tissues in *Ido1* ^−/−^ mice --------------------------------------------------------------------------------- We analyzed the gene expression profiles of three to five *Ido1*^−/−^ or *Ido1*^+/+^ mice from the controls and from cohorts treated with 1% or 2% DSS ([Fig. 3a](#f3){ref-type="fig"}). As expected, the transcription levels of *Ido1* were significantly increased in the inflamed colon tissues of *Ido1*^+/+^ mice, suggesting that the inflammatory response induced *Ido1* transcription (data not shown). PCA analysis showed that *Ido1*^−/−^ mice had a distinct pattern compared with *Ido1*^+/+^ mice only in the 2% DSS treatment group ([Fig. 3a](#f3){ref-type="fig"}). We performed a one-way ANOVA analysis to identify differentially-expressed genes with statistical thresholds of 5% FDR and 2-fold change restriction among the six groups. A total of 6,421 genes were identified as significant and were classified further based on their biological functions. As expected, the Inflammatory Response category was the most significant key function (p = 1.23E-35 to 3.89E-08) ([Fig. 3b](#f3){ref-type="fig"}). In comparisons of *Ido1*^−/−^ and *Ido1*^+/+^ mice, much higher numbers of genes were differentially expressed under DSS-stimulated conditions than in the basal non-stimulated state. Cytokines including interleukins and interferons are known to induce a broad inflammatory reaction in response to infection or injury[@b21][@b22][@b23]. This prompted us to further investigate cytokine and chemokine gene expression signatures. Strikingly, the expression levels of pro-inflammatory cytokines and chemokines, such as *Il-1β*, *TNF, Cxcl1,* and *Ccl7*, which are central mediators of inflammation were significantly elevated in a dose-dependent manner by DSS treatment in *Ido1*^+/+^ mice, however, the elevation was substantially attenuated in *Ido1*^−/−^ mice ([Fig. 3c](#f3){ref-type="fig"}). In contrast, IL-7 and IL-18 are anti-inflammatory cytokines and their expression was much lower in DSS-treated *Ido1*^+/+^ compared with DSS-treated *Ido1*^−/−^ mice. These gene expression signatures suggested that *Ido1*^−/−^ mice would be less likely to develop severe colitis than *Ido1*^+/+^ mice. The pro-inflammatory cytokines and chemokines noted above were previously reported to be induced in inflammatory cells, including monocytes, as part of the inflammatory reaction[@b24][@b25]. These findings led us to assess CD11b^+^Gr-1^+^ cells in peripheral blood (PBL) and colon using FACS. The frequencies of CD11b^+^Gr-1^+^ (Ly6G) cells were higher in the PBL and colon of *Ido1*^+/+^ mice than in *Ido1*^−/−^ mice following treatment with 2% DSS, although the difference in PBL was not significant ([Fig. 4a](#f4){ref-type="fig"}). Interestingly, in the CD11b^+^Gr-1^+^ cells, the difference in the proportion of the Ly6G^hi^Ly6C^low^ subpopulation was significant between the two genotypes of mice with 2% DSS-treatment: the proportion of Ly6G^hi^Ly6C^low^ was higher in the *Ido1*^+/+^ mice than in the *Ido1*^−/−^ mice ([Fig. 4b](#f4){ref-type="fig"}). To verify the role of IDO1 in the expansion of CD11^+^Gr-1^+^ cells in the DSS-induced colitis model, we examined whether treatment with the IDO1 inhibitor, L-1MT, would reproduce the results from *Ido1*^−/−^ mice. Consistently, the frequency of CD11^+^Gr-1^+^ cells and the proportion of the Ly6G^hi^Ly6C^low^ subpopulation in CD11^+^Gr-1^+^ cells were lower in the L-1MT-treated mice with DSS-induced colitis, compared with their vehicle-treated control counterparts ([Fig. 4c,d](#f4){ref-type="fig"}). Moreover, the proportion of colonic granulocytes (CD11b^+^Ly6G^+^Ly6C^low^F4/80^-^CD11c^−^)[@b26] was significantly lower in both *Ido1*^−/−^ and L-1MT-treated *Ido1*^+/+^ mice compared to the *Ido1*^+/+^ control mice after DSS treatment (see [Supplementary Fig. S2a](#S1){ref-type="supplementary-material"}). However, no significant difference was observed in monocyte populations (CD11b^+^Ly6C^+^Ly6G^-^CD11c^−^)[@b26] among the three groups ([Supplementary Fig. S2b](#S1){ref-type="supplementary-material"}). This indicates that loss of the functional activity of IDO1 may contribute to reduced expansion of CD11b^+^Gr-1^+^ cells including granulocytes in the DSS-induced colitis model. These results suggest that IDO1 induced a systemic inflammatory response including a gut inflammatory response associated with increased expression of pro-inflammatory cytokines and chemokines. IDO1 deficiency results in down regulation of TLR and NF-kB signaling pathways ------------------------------------------------------------------------------ We next conducted an upstream regulator analysis to identify major upstream molecules of the differentially expressed genes identified in our data set[@b27]. It would be critical to identify the upstream regulatory molecules and their associated mechanisms of action to provide biological insights into the observed expression changes. In addition to transcription factors, upstream regulators can include any gene or small molecule that has been observed experimentally to affect gene expression directly or indirectly. The top 9 predicted upstream regulators ranked by z-score are lipopolysaccharide (p = 2.66E-170), TNF (p = 2.79E-166), IFNG (p = 3.00E-140), IL1B (p = 2.66E-114), NF-kB complex (p = 1.25E-83), STAT3 (p = 3.65E-69), MYD88 (p = 2.39E-54), p38 MAPK (p = 5.47E-47), and TLR (p = 1.51E-45) ([Fig. 5a](#f5){ref-type="fig"}). Surprisingly, all regulators are well known contributors to the development of inflammatory responses. As a point of interest, the greatest overlap with the regulators was in the pathways related to the Toll-like receptor and NF-kB signaling. Given these findings, we focused on genes belonging to those pathways in more detail. The most striking result was that expression of the majority of genes involved in TLR signaling was down-regulated in DSS-treated *Ido1*^−/−^ mice, suggesting that TLR signaling is essential for the contributions of IDO1 to the development of DSS-induced colitis ([Fig. 5b](#f5){ref-type="fig"}). In the basal state (non-stimulated), expressions of *Tlr2*, *Tlr6*, and *Myd88* remained low and the levels were similar in *Ido1*^−/−^ and *Ido1*^+/+^ mice. However, treatment with DSS resulted in dramatic increases in the expressions of those genes in *Ido1*^+/+^ mice in a dose-dependent manner. The expressions of these genes were not significantly induced by DSS treatment in IDO1-deficient mice ([Fig. 5d](#f5){ref-type="fig"}). Importantly, MyD88 is a key adaptor protein in the signal transduction cascades shared by most TLRs and MyD88-dependent TLR signaling pathways were found to be down-regulated in our study ([Fig. 5b](#f5){ref-type="fig"}). This suggests that the differences in colitis development between *Ido1*^−/−^ and *Ido1*^+/+^ mice result from dysregulation of the TLR-MyD88 signaling pathway. We also ascertained that TLR-triggered cascades downstream of NF-kB signaling molecules like NF-kB (nuclear factor kB, p = 1.54E-4), JNKs (JUN N-terminal kinases, p = 6.62E-3), and IRF5 (interferon regulatory factor 5, p = 1.20E-3), which were all down-regulated in DSS-treated *Ido1*^−/−^ mice ([Fig. 5c](#f5){ref-type="fig"}). These results implied that in mice with DSS-induced colitis, TLR signaling was inhibited by IDO1-deficiency, which suppressed pro-inflammatory cytokine and chemokine production through the regulation of a multitude of transcription factors such as NF-kB. Discussion ========== IDO1-mediated tryptophan metabolism in various tissues is linked to numerous biological and physiological functions. Heightened expression of IDO1 in colonic intestinal tissues led to the hypothesis that IDO1 plays critical roles in gut homeostasis. Various microorganisms and food antigens exist in the lumen and challenge the intestinal immune system. As the luminal surface of the gastrointestinal tract continually interacts with foreign antigens such as pathogenic bacteria, the mucosal immune system maintains immune tolerance to limit inflammatory responses elicited by these antigens. In the present study, we showed that in the basal state in which there was no chemical induction of inflammation, IDO1 deficiency did not lead to pathophysiologic changes in the gut. However, gene expression profiling analysis revealed that a transcriptional network of inflammatory responses was significantly down-regulate in the absence of IDO1. When the mucosal immune system fails to maintain tolerance, pathogenic microbes are able to infect the host. Impaired immune tolerance in the intestine can lead to inflammatory bowel diseases such as Crohn's disease and ulcerative colitis[@b28][@b29][@b30]. Treatment with DSS treatment mimics the condition of impaired immune tolerance because it disrupts the mucosal barrier of epithelial cells on the luminal surface. Studies of the DSS-induced colitis model system enabled us to characterize the role of IDO1 in driving inflammatory response, which is not apparent in the non-stimulated state. We showed that DSS-treated IDO1-deficient mice did not develop colitis of the same severity as normal control mice by assessing the loss in body weight, intestinal bleeding, diarrhea, shortening of colon length and histological lesions. It should be noted that there are conflicting results regarding the role of IDO1 in different mouse models of colitis. In studies that utilized TNBS to induce colitis, inhibition of IDO1 was found to exacerbate colitis[@b17][@b31]. This discrepancy can probably be explained, at least in part, by the fact that TNBS induces a T cell-dependent disease while the DSS induction is in a different manner. Although both DSS and TNBS colitis are used for inflammatory bowel disease animal model, DSS colitis has been distinguished from TNBS models by many others in the several points[@b20][@b32]. For instance, while TNBS-induced colitis develops as a result of delayed-type hypersensitivity reaction to haptenized proteins, DSS-induced colitis is the result of a change in epithelial barrier function[@b32]. Additionally, it is known that while TNBS induces Crohn's disease-like colitis, DSS induces ulcerative colitis[@b20]. Therefore, disease exacerbation in IDO1-deficient TNBS-treated mice was attributed to the inhibition of T~reg~ activity[@b17][@b31], which is a distinct mechanism from what we found in IDO1-deficient DSS-treated mice. It is well established that TLRs and NF-kB signaling pathways make important contributions to inflammatory responses. Strikingly, our transcriptome analyses showed that the expression of members of the TLR-MyD88-NF-kB signaling pathways was significantly decreased in the absence of IDO1. Consequently, down-regulation of those signaling pathways resulted in dramatic changes in the expression of cytokines and chemokines. We showed that in IDO1 deficient mice, the frequencies of circulating inflammatory cells in peripheral blood and in the gut were decreased as well. These findings suggest that IDO1 may require TLR-MyD88-NF-kB signaling to promote the development of colitis. It is notable that compared to normal mice, IDO1-deficient mice had higher expression of *Muc1,* which encodes mucin protein, the first line of host defense against invading bacteria ([Fig. 5d](#f5){ref-type="fig"}, p = 3.20E-3). Reduction in the expression of *Muc1* severely affects epithelial barrier function[@b33][@b34]. Several studies reported that mucin may be a negative regulator of TLR signaling[@b35][@b36]. This suggests that the absence of IDO1 may lead to increased mucin expression and subsequently down-regulate TLR-MyD88-NF-kB signaling. IDO1 has become an emerging target for the treatment of cancer, infection, autoimmunity, and other diseases associated with inflammatory responses and immunosuppression[@b37][@b38][@b39][@b40]. The present report utilizing the *Ido1*^−/−^ mouse model provides the first comprehensive analysis of IDO1 targets at the transcriptome level and broadens our understanding of the diverse functions of IDO1 during inflammatory responses. This study also suggest that IDO1 is likely to be a promising target of therapeutic intervention in colitic diseases. Methods ======= Mice ---- *Ido1*^−/−^ mice of the C57BL/6 (B6) genetic background were obtained from The Jackson Laboratory (Bar Harbor, ME, USA). *Ido1*^−/−^ mice were crossed with C57BL/6 wild-type (The Jackson Laboratory) to generate the *Ido1*^−/−^ and *Ido1*^+/+^ offspring used in this study. Genotypes of knockout mice were verified via PCR typing. The mice used were 10--12 weeks old and weighed 18--23 g. Age- and weight-matched female littermates were used as controls. C57BL/6 mice and *Ido1*^−/−^ mice were maintained under specific pathogen free (SPF) condition at the Center of Animal Resource Development, Seoul National University College of Medicine. The mice were maintained based on the guidelines of Seoul National University Animal Experiment Ethics Committee. All animal experimental protocols were approved by the Committee on the Ethics of animal experiments of Seoul National University (Institutional Animal Care and Use Committee permit number: SNU-150119-5). All experiments were carried out in accordance with the guidelines and regulations. Induction of colitis and evaluation of colitis severity ------------------------------------------------------- To generate an acute colitis experimental model, dextran sulfate sodium (DSS) (molecular mass 36--50 kDa; MP Biomedicals, Illkirch, France) was added to the drinking water at concentrations of 1% or 2% (w/v) given *ad libitum* for 7 days. Control mice received drinking water without DSS. The subsequent course of colitis development was evaluated by monitoring daily weight changes. Colitis severity also was scored by evaluating clinical disease activity through daily observation of the following parameters: weight loss (0 points = No weight loss or weight gain, 1 points = 5--10% weight loss, 2 points = 11--15% weight loss, 3 points = 16--20% weight loss, 4 points = \>21% weight loss); stool consistency (0 points = normal and well formed, 2 points = very soft and unformed, 4 points = watery stool); and bleeding stool score (0 points = normal color stool, 2 points = reddish color stool, 4 points = bloody stool). The disease activity index (DAI) was calculated based on the combined scores of weight loss, stool consistency, and bleeding ranging from 0 to 12. All parameters were scored from day 0 to day 7. At the 7^th^ day after DSS-colitis induction, mice were sacrificed and the entire colon was quickly removed. After colon length was determined as a marker of inflammation, the entire colon was cut open lengthwise and gently flushed with sterile phosphate-buffered saline (PBS) to remove any traces of feces. Colon segments were immediately frozen in liquid nitrogen and stored at −80 °C for subsequent extraction of total RNA. For histological analysis, colon segments were fixed in 10% neutral buffered formalin phosphate and stored at room temperature until study for evidence of inflammation. Administration of L-1MT ----------------------- 9--11 weeks old *Ido1*^+/+^ mice were administered L-1MT. To prepare L-1MT for oral gavage, 1g of L-1MT (purchased from Sigma-Aldrich) was added to a 15 ml conical tube with 10 ml Methocel/Tween \[0.5% Tween 80/0.5% Methylcellulose (v/v in water; both from Sigma-Aldrich)\]. The mixture was bead milled overnight by adding 2--3 mm glass beads and mixing inversion. The next day, the L-1MT concentration was adjusted to 80 mg/ml by adding an additional 2.5 ml Methocel/Tween and mixing again. The L-1MT slurry was administered by oral gavage at 400 mg/kg/dose (100 μl of total volume) using a curved feeding needle (20-gquge 1^1/2^ in; Fisher) as previously described[@b41]. For twice a day dosing, L-1MT was administered once in the morning and once in the evening. On day 5 of the experiment, all mice received 2% DSS treatment and mice were sacrificed on day 13. Histological analysis of colitis -------------------------------- Routinely processed, 4--6 μm paraffin-embedded sections of colon samples were prepared and stained with hematoxylin and eosin (H&E) for histological grading. Histological scores, including severity of colitis, were evaluated in a blinded manner as previously described by Laroui *et al.*[@b42]. Grades were evaluated from 0--4 for the following three criteria: severity of inflammation (0, rare inflammatory cell in the lamina propria; 1, increased inflammatory cells in the lamina propria; 2, confluent inflammatory cells extending into the submucosa; and 3, transmural extension of the inflammatory cell infiltrate); damage (0, none; 1, loss of the basal 1/3 of the crypt; 2, loss of the basal 2/3 of the crypt; 3, loss of the entire crypt but intact epithelial cells; and 4, loss of the entire crypt and of the surface epithelial cells); extension (0, none; 1, focal; 2, lesion involving 1/3 of the intestine; 3, lesion involving 2/3 of the intestine; and 4, lesion involving the entire intestine). Scores for each criterion were added to give an overall inflammation score for each sample with a range of 0--11. The histological grades were determined for each section, and the sum of the grades was reported as the histological score for each mouse. The level of colitis was blindly assessed by two histopathologists. Flow cytometry (FACS) analyses ------------------------------ Fresh peripheral blood lymphocytes (PBLs) were prepared by incubating blood with ACK (ammonium-chloride-potassium) buffer to lyse red blood cells at room temperature for 3--5 minutes, and stained using FACS buffer (1X phosphate-buffered saline \[PBS\] with 0.1% bovine calf serum and 0.05% sodium azide). Colonic cells in the lamina propria of mice were isolated according to the previously described protocol[@b43]. Briefly, colon pieces (1 cm pieces) were treated with 5 ml of predigestion solution (1× HBSS containing 2 mM EDTA and 1 mM DTT) for 20 min at 37 °C. After incubation, cells epithelial cells were decanted and again incubated the pieces for 20 min at 37 °C. Intestinal pieces were washed with 1× PBS to remove remaining EDTA. And collected tissues were incubated with digestion solution (1.5 mg/ml of collagenase D \[Roche\], 0.1 mg/ml of DNase I \[Sigma-Aldrich\] and 5% of fetal bovine serum in 100 ml of 1× PBS) for 20 min at 37 °C and repeated. Cells were harvested and centrifuge for 10 min at 1,500 rpm, and resuspended with 5 ml of 40% percoll solution and overlayed to 80% percoll solution. Cells were centrifuged for 20 min at 1,000g and resuspended with FACS buffer. And cells were stained FITC- and eFluor® 450-conjugated anti-Ly6G (Gr-1; RB6-8C5, eBioscience, San Diego, CA, USA), PE- and PE-Cy7-conjugated anti-CD11b (M1/70, eBioscience), APC-conjugated anti-Ly6C (HK1.4, eBioscience), PE-Cy5-conjugated anti-F4/80 (BM8, eBioscince), APC-Cy7-conjugated anti-CD11c (N418, Biolegend, San Diego, CA, USA) antibodies at 4 °C for 30 min. After washing with FACS buffer, the cells were analyzed by a FACS Calibur (BD Bioscience, Franklin Lakes, NJ, USA) and FACS LSRII (BD Bioscience) and Flowjo software (Tree star, Ashland, OR, USA). Microarray hybridization ------------------------ The gene expression profile was determined using the MouseWG-6 v.2 Expression BeadChips (Illumina®). For microarray hybridization, total RNA was isolated by homogenizing colon tissue samples and was purified using a DNA-free RNA isolation kit (RNAqueous-4PCR kit; Ambion, Austin, TX, USA) in accordance with the manufacturer's instructions. Total RNA integrity and quantity were assessed with a Nanodrop-2000 Spectrophotometer (Thermo Fisher Scientific, Wilmington, DE). Only total RNA with an OD 260/280 ratio \>2.0 was used for microarray hybridization. RNA samples were first amplified for array analyses using the Illumina Total Prep RNA Amplification Kit (Ambion, Austin, TX, USA) according to the manufacturer's instructions. Briefly, 500ng of total RNA, isolated from colon tissue, was used to prepare labelled cRNA with overnight incubation according to the manufacturer's protocol. The quality and quantity of the labelled cRNA were monitored using a Nanodrop-2000 Spectrophotometer. Amplified cRNA (1.5 μg) was hybridized on MouseWG-6 Expression BeadChip arrays, containing more than 45,281 well-annotated Ref transcripts, according to the manufacturer's standard protocol. The arrays were then scanned on a BeadArray Reader (BeadStation 500G Instrument, Illumina Inc.), and Spot images identification and quantification were obtained by the Genome Studio software v1.0.2. (Illumina Inc.). Identification of significant genes ----------------------------------- The raw data were pre-processed through three steps: background correction was performed, the data were then log-transformed to log 2 scale, and normalized by quantile normalization method implemented in the Genome Studio software (Illumina Inc.). Significant difference between two genotypes in each dose (0% DSS \[baseline\], 1% DSS treatment, 2% DSS treatment), differences between dose response effect in each genotype, and difference between genotype x dose interaction were identified using ANOVA test (*p* \< 0.05) on log 2-transformed normalized intensities using by Partek® Genomics Suite software v6.3 (Partek, St Louis, MI) (<http://www.partek.com/partekgs>). Transcripts with more than 2-fold differential expression and a false discovery rate (FDR) \< 0.01 were selected for each specific comparison analyzed. Functional enrichment and clustering analysis --------------------------------------------- Functional categorization analysis was performed based upon gene ontology consortium (GO). Broad Gene Set Enrichment Analysis (GSEA) was done to examine the significance of each functional category classified by GO[@b44] (<http://www.broadinstitute.org/gsea/index.jsp>). Hierarchical clustering analysis was carried out with Genesis software v1.7.5[@b45] using the Pearson correlation distance matrix with average linkage algorithm. Statistical Analysis -------------------- Data were expressed as the mean ±S.E.M. Statistical significance (*p* value \< 0.05) was evaluated either by unpaired student's *t* test between two groups or one-way analysis of variance (ANOVA) to compare multiple groups using dose and genotype as factors. Statistical analyses were performed using Graph Pad Prism 5 software (GraphPad Software Inc., La Jolla, CA, USA) and SAS Enterprise Guide 6.1 (SAS Institute Inc., Cary, NC, USA). For gene analyses, significances for functional enrichment of specific genes were determined by a right-tailed Fisher's exact test as the negative log of the probability that the number of focus genes is not due to random chance. Additional Information ====================== **How to cite this article**: Shon, W.-J. *et al.* Severity of DSS-induced colitis is reduced in Ido1-deficient mice with down-regulation of TLR-MyD88-NF-kB transcriptional networks. *Sci. Rep.* **5**, 17305; doi: 10.1038/srep17305 (2015). Supplementary Material {#S1} ====================== ###### Supplementary Information This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) Grant funded by the Ministry of Science, ICT & Future Planning (NRF-2013R1A1A1007858) and grants Korea Healthcare Technology R&D project, Ministry of Health, Welfare, and Family affairs of Korea (HI12C0213). **Author Contributions** D.S. and E.C designed the study. W.S., Y.L. and J.S. conducted the experiments. W.S., Y.L., E.C. and D.S. analyzed data and interpreted the results. W.S. and D.S. wrote the manuscript. All authors reviewed the manuscript. ![Differential gene-expression profiles of *Ido1*^−/−^ mice compared with *Ido1*^+/+^ mice detected by microarray analysis.\ (**a**) 3-D view of PCA scores plot of *Ido1*^−/−^ group (n = 3) *versus Ido1*^+/+^ group (n = 3). Groups are shown by different colors and dots represent individual strains. Blue spots represent *Ido1*^+/+^ group and the red spots represent *Ido1*^−/−^ group. On a 3D-PCA plot of *Ido1*^−/−^ group can be clearly distinguished from *Ido1*^+/+^ group. (**b**) Hierarchical clustering and heat map of up- or down-regulated genes that are differentially expressed (\>2-fold, 1% \[FDR\]) in the absence of IDO1. Red indicates high relative expression and green indicates low expression of genes as shown in the scale bar. Cluster analysis functionally categorized by IPA; Canonical pathway significantly detected in *Ido1*^−/−^ mice compared to *Ido1*^+/+^ mice. Statistical significance of pathway modulation was calculated via a right-tailed Fisher's exact test in Ingenutity Pathway. (**c**,**d**) Gene-pathway networks using IPA. (**c**) Gene Expression and (**d**) Inflammatory Response pathway are displayed as networks. Green and red symbols denote down-regulated and up-regulated genes in *Ido1*^−/−^ compared with *Ido1*^+/+^ mice, respectively. Arrows with unbroken lines indicate a direct interaction between two molecules, with the mode of action in the direction of the arrow; arrows with broken lines denote an indirect interaction. (**e**) We performed gene-set-enrichment analysis (GSEA) to determine whether the filtered gene list from *Ido1*^−/−^ mice *versus Ido1*^+/+^ mice showed specific enrichment in the inflammatory response in the rank-based analysis. Rank of 102 genes in our data sets ordered by expression level with enrichment plots for the up-regulated and down-regulated genes. ES (Enrichment Score) is a value that represents how well the gene set is enriched within the selected gene list. The FDR q value \<0.01 for specific enrichment of the gene set is as indicated. The leading edge analysis of our microarray data identified that the gene is highly correlated with Inflammatory Response.](srep17305-f1){#f1} {#f2} ![Transcriptional Profiling of inflamed colon tissues in *Ido1*^+/+^ and *Ido1*^−/−^ mice after DSS treatment.\ (**a**) 3-D view of PCA scores plot of data obtained on *Ido1*^−/−^ mice *versus Ido1*^+/+^ mice treated with DSS. On a 3D-PCA plot of 2% DSS treated *Ido1*^−/−^ group can be significantly seperated from 2% DSS treated *Ido1*^+/+^ group. Each spots represents individual mouse in the group. Blue spots, *Ido1*^+/+^ group; Red spots, *Ido1*^−/−^ group; Purple spots, *Ido1*^+/+^ + 1% DSS treatment group, Green spots, *Ido1*^−/−^ + 1% DSS treatment group; Sky blue spots, *Ido1*^+/+^ + 2% DSS treatment group; Orange spots. *Ido1*^−/−^ + 2% DSS treatment group. (**b**) Functional categories by IPA; Canonical pathway significantly detected in *Ido1*^−/−^ mice compared to *Ido1*^+/+^ mice after DSS-induced colitis. Statistical significance of pathway modulation was calculated via a right-tailed Fisher's exact test in Ingenutity Pathway and represented as --log (p-value). A larger value on the x axis indicates a higher degree of significance, i.e., a smaller p value. The leading edge analysis of our microarray data identified gene highly correlated with Inflammatory Response. (**c**) Heat map of cytokine and chemokine genes that are differentially expressed \[5% FDR and 2-fold change restriction\] in DSS-challenged *Ido1*^−/−^ *versus Ido1*^+/+^. The given values are the average of normalized intensities (3--5 mice per group) with red representing higher levels of expression and green indicating lower levels. Color code: blue, pro-inflammatory cytokines; yellow, anti-inflammatory cytokines.](srep17305-f3){#f3} {#f4} {#f5} [^1]: These authors contributed equally to this work.

Introduction {#Sec6} ============ Selective logging is the dominant commercial activity in forested areas of the tropics with strong conservation potential (Frumhoff, [@CR11]; Struhsaker, [@CR38]). This potential is arguably strongest in central Africa, where a combination of weak communications and infrastructure, lack of logistics systems, and high transportation costs result in highly selective timber extraction, focusing on a few economically valuable species occurring at low densities (Boardman et al., [@CR6]; Clark et al., [@CR8]). During the last decade, the timber industry has thrived in central Africa and many of the remaining forests have been allocated to timber concessions. With extraction rates of 1--4 trees harvested per hectare, damage to remaining forest is relatively limited and disrupts only 7--20% of the canopy (Hall et al., [@CR18]; Van Gemerden, [@CR42]). This is compared to extraction rates of 5--10 trees per hectare and associated disruption of 25--50% of the canopy typical of other forested regions of the tropics (Skorupa, [@CR36]; Chapman et al., [@CR7]). Whereas some species of wildlife thrive in such low-intensity operations, even low-level extraction appears to have a negative effect on some primate species, including chimpanzees (White, [@CR43]; Barnes and Lahm, [@CR4]; White and Tutin, [@CR44]; Matthews, [@CR24]). Although logging is known to reduce the abundance of some primate species due to associated hunting and the loss of food sources for frugivores (Wilkie and Carpenter, [@CR45]; Chapman et al., [@CR7]), the potential role of parasite infections in such primate population declines remains largely unexplored. Logging results in a suite of alterations in host ecology forest structure and human-wildlife overlap that may alter infection prevalence and infection risk in resident populations. Patterns of parasitism in wildlife populations are influenced by host ranging patterns, density, intraspecific and interspecific contact rates, diet, and immune function (Hudson et al., [@CR21]; Nunn et al., [@CR28]; Gillespie et al., [@CR13], [@CR14]; Beldomenico et al., [@CR5]). Studies on a variety of species have demonstrated that these characteristics can be affected by changes in forest structure (Olupot et al., [@CR29]; Heydon and Bulloh, [@CR19]; Patriquin and Barclay, [@CR30]). *Giardia* and *Cryptosporidium* are protozoal pathogens that infect humans, domestic animals, and wildlife worldwide (Appelbee et al., [@CR1]). In humans and livestock, these parasites cause diarrhea and other enteric disorders that can contribute to nutritional deficiencies and impaired weight gain (Savioli et al., [@CR34]; Thompson, [@CR40]). Little information exists about the clinical affects of these parasites on their wildlife hosts; however, environmental pollution with human fecal material is recognized as a potential pathogen pathway for wildlife infections with zoonotic protozoal parasites, such as *Giardia* and *Cryptosporidium* (Appelbee et al., [@CR1]). Considering the capacity of such infections to put wildlife populations at risk, it is important to monitor the health status of wildlife in general and endangered species in particular. It is equally important to identify potential pathogen transmission pathways from humans and domestic animals to wildlife. To better understand the affect of selective logging on the transmission and persistence of these protozoal pathogens, we examined noninvasively collected fecal samples from 134 sympatric western lowland gorillas (*Gorilla gorilla gorilla*) and chimpanzees (*Pan troglodytes troglodytes*) in the northern Republic of Congo. We compared patterns of infection between ape samples from the Kabo Logging Concession with those sampled from the adjoining undisturbed forest of the Nouabalé-Ndoki National Park. Methods {#Sec1} ======= Study Site {#Sec2} ---------- The Sangha River Tri-National Conservation Area (36,236 km^2^) is notable for its extensive tracks of forest and transboundary protected area network: Nouabalé-Ndoki National Park in Republic of Congo, Lobéké National Park in Cameroon, and Dzanga-Ndoki National Park and Dzanga-Sangha Special Reserve in Central African Republic (Fig. [1](#Fig1){ref-type="fig"}). This region is largely unpopulated with human densities estimated at 0.7 to 0.8 people per km^2^ (Auzel and Wilkie, [@CR2]). Semi-nomadic Ba'Aka pygmies and Bantu agriculturalist-fishermen live in complex, interdependent economic and social relationships (Eves and Ruggiero, [@CR10]). Most villages are along major waterways leaving the remote interior forest with little human development and associated pressures. However, commercial logging is rapidly expanding along the periphery of the Tri-National Conservation Area. Timber exploitation in these remote areas is relatively labor-intensive and can result in the transient relocation of skilled workers and their families, introducing the risk of novel opportunities for pathogen transmission from people to resident wildlife.Figure 1Land use in the Sangha Tri-National Conservation Area at the boundary of Republic of Congo, Cameroon, and the Central African Republic. The Goualougo Triangle is located in the southern portion of the Nouabalé-Ndoki National Park. The Kabo Logging Concession is contiguous to the national park's southern boundary. The Kabo Logging Concession (2800 km^2^), operated by the *Congolaise Industrielle des Bois,* has been logged to varying degrees for the past 40 years. The pristine forests of the adjoining Goualougo Triangle (310 km^2^) were initially part of the Kabo Concession until biological surveys by the Wildlife Conservation Society revealed the conservation significance of this region's great ape populations and their habitat. In June 2003, the Congolese government announced that it would annex the majority of the Goualougo Triangle to the Nouabalé-Ndoki National Park (4400 km^2^, 2.05′--3.03′N, 16.51′--16.56′E). However, the adjacent forest remains in an active logging concession. As part of the Goualougo Triangle Ape Project's mandate to evaluate the impacts of logging on wild apes, this area was stratified into study zones (A, B~1~, B~2~, C, and D) with relation to the national park boundaries and logging activities (Fig. [2](#Fig2){ref-type="fig"}). Zones A, B~1~, and B~2~ are undisturbed forest within the national park. Zone D was logged in the early 1970s, and zone C is slated for logging in the near future. All study zones are characterized by similar terra firma forest dominated by mixed species interspersed with monodominant *Gilbertiodendron dewevrei*, seasonally flooded forest, and open swamp forests (Morgan et al., [@CR25]).Figure 2Location of study zones within the Goualougo Triangle (A, B~1~, and B~2~) and adjacent logging concession (C and D) in Republic of Congo. Sample Collection and Analysis {#Sec3} ------------------------------ Between 2004 and 2009, we conducted repeated reconnaissance and line transect surveys for ape signs in zone D of the Kabo Logging Concession and zones A and B~1~ within the Goualougo Triangle (Morgan et al., [@CR25], [@CR26]). Our first surveys of zone D were conducted after forestry activities had been dormant for more than 30 years. Ape fecal samples were collected opportunistically in the field and preserved in 10% neutral buffered formalin (Gillespie, [@CR12]). Fecal samples were characterized as chimpanzee or gorilla based on morphological features. At similar sites using similar methods, samples characterized in this fashion were verified to the correct species more than 94% of the time by genetic methods, with field researchers able to correctly identify gorilla dung (318/324 = 98%) more often than chimpanzee fecal remains (200/211 = 94%) (Arandjelovic, 2009, personal communication). When collected, fecal material was characterized as soft or firm depending on the consistency, assuming that sick apes will evacuate soft stools (Gillespie et al., [@CR14]). A Merifluor *Cryptosporidium*/*Giardia* Direct Immunofluorescent Detection Kit (Meridian Bioscience, Inc., Cincinnati, OH) was used to detect both parasites (Johnston et al., [@CR23]). Fecal samples were scored both for presence or absence of the pathogens as well as quantification of *Cryptosporidium* sp. oocysts and *Giardia* sp. cysts in feces. Fecal samples were concentrated and then resuspended in 1 g/1 ml suspensions. Counts were calculated by analyzing 10 μl of the 1 g/ml solution of concentrated feces from each sample. Oocysts or cysts were then quantified by counting total numbers in 150 microscope fields (400× magnification) and extrapolating results to the entire sample. This method was validated by spiking negative fecal samples with known numbers of *Cryptosporidium* sp oocysts and *Giardia* sp. cysts. Results were analyzed using EpiInfo, Version 3.3.2 (Centers for Disease Control, Atlanta, GA). A power analysis was conducted using the Finite Population Correction for Proportions (Israel 1992). The corrected sample sizes are calculated using the original sample size results calculated using the equation for a large population. These numbers are based on a confidence level of 95% and level of precision of 0.05. Results {#Sec4} ======= All 134 samples were negative for *Cryptosporidium* sp. (Table [1](#Tab1){ref-type="table"}). Three of 34 chimpanzee samples (8.82%) from the logged site, 1 of 33 chimpanzee samples from the undisturbed site (3.03%), 0 of 31 gorilla samples from the logged site (0%), and 1 of 36 gorilla samples from the undisturbed sites (2.78%) were infected with *Giardia* sp. (Table [1](#Tab1){ref-type="table"}). Prevalence of infection with *Giardia* sp. was not significantly different between logged and undisturbed forest for chimpanzees (χ^2^ = 1.00, *P* = 0.32, degrees of freedom (*df*) = 1) or gorillas (χ^2^ = 0.87, *P* = 0.35, *df* = 1).Table 1Prevalence and intensity of *Cryptosporidium* and *Giardia* spp. in chimpanzees and gorillas in the Goualougo Triangle of Nouabalé-Ndoki National Park and Kabo Logging Concession, Republic of CongoSpeciesPopulationSamples*CryptosporidiumGiardia*Prevalence (%)Mean intensityPrevalence (%)Mean intensityChimpanzeeGoualougo33003.031000Kabo34008.821833 ± 573.49GorillaGoualougo36009.78900Kabo310000Mean intensities are expressed as oocysts (*Cryptosporidium*) or cysts (*Giardia*) per gram, standard error of the mean; negative samples are not included in this calculation. Of all 67 chimpanzee samples, one was characterized as soft (1.5%) and the remainder as firm (98.5%). Of all 67 gorilla samples, 3 were characterized as soft (4.5%) and the remainder as firm (95.5%). There was no relationship between fecal consistency and *Giardia* infection status for chimpanzees (χ^2^ = 0.00, *P* = 0.78, *df* = 1) or gorillas (χ^2^ = 0.07, *P* = 0.78, *df* = 1). Our power analysis demonstrated that additional zone D chimpanzee sampling would be required to ensure that the lack of significance observed in intersite comparisons of chimpanzees *Giardia* sp. prevalence is indeed valid (47 zone D compared with the 34 collected thus far). Samples were collected during repeated line transect surveys during a period of 5 years. Collection of fresh ape feces in this system requires a tremendous amount of effort. Three to five samples can be collected in a day by a team of five trackers. Samples are collected by backtracking sighted apes to find feces voided within the past hour. Aged feces, such as those from night nests are not collected. Due to these limitations, we regard the lack of significant differences in *Giardia* sp. prevalence between sites for chimpanzees to be preliminary. Additional sampling will be required to definitively examine this comparison. We also compared the intensity of infection for each ape species between logged and undisturbed sites. The mean intensity of infection was highest in chimpanzees from the disturbed site (1833 ± 573.49 cysts/g), followed by chimpanzees in the undisturbed site (1000 cysts/g). Mean intensity of infection was slightly lower for gorillas from the undisturbed site (900 cysts/g), and no infections were found in gorillas from the disturbed site, resulting in an intensity of zero (Table [1](#Tab1){ref-type="table"}); mean intensities of infection could not be compared statistically across sites due to rarity of infection. Discussion {#Sec5} ========== Our results demonstrate that *Giardia* sp. is present at low prevalence in chimpanzees and lowland gorillas in undisturbed forests and forests with a history of selective logging. In these same populations, there is no evidence of infection with *Cryptosporidium* sp. A lack of *Cryptosporidium* sp. in all samples examined suggests that *Cryptosporidium* sp. is not a natural component of ape parasite communities at this site and that historical land-use patterns are not exposing apes to zoonotic sources of this pathogen. Low prevalence of *Giardia* sp. in the study area regardless of historical land-use patterns suggest that a low level of circulating *Giardia* is a natural component of Congolese ape parasite communities and that historical land-use patterns are not exposing apes to zoonotic sources of this pathogen. Our study area is located only 32 km from the nearest village and the large river that serves as a major transportation source for the region. However, the forest is continuous and the encounter rate of human signs in the study zones was low, which confirms that these forests have remained relatively undisturbed despite past timber exploitation. This post-logging scenario is much different than the high human densities and forest fragmentation typical in east and west Africa. *Giardia* and *Cryptosporidium* are notable for cross-species transmission (Thompson, [@CR40]). The propensity of these protozoans to cross species barriers results from the host nonspecificity of trophozoites of either genus, and the high infectivity and environmental stability of *Giardia* cyst and *Cryptosporidium* oocysts (Smith and Nichols, [@CR37]). Previous field-based studies confirm this pattern, demonstrating that high levels of primate--human overlap enhance transmission of potentially pathogenic protozoa, including *Giardia* sp. and *Cryptosporidium* sp. (Nizeyi et al., [@CR27]; Graczyk et al., [@CR15], [@CR16]; Graczyk et al., [@CR17]; Salzer et al., [@CR33]). The results of the current study are the first to examine the long-term effects of historical primate--human overlap on contemporary patterns of infection with these pathogens. Considered among the most common human intestinal protozoa, *Giardia* ranges in clinical presentation from asymptomatic to highly pathogenic, causing chronic malabsorptive diarrhea in some patients (Thompson, [@CR40]). Both host factors (e.g., nutrition, immunity, coinfection with other agents) and pathogen factors (e.g., strain, infectious dose) are thought to contribute to the severity of clinical disease (Thompson, [@CR39]). Giardiasis is now considered by the World Health Organization to be a "neglected disease," due to its ubiquity in equatorial latitudes, its chronic detrimental health effects (especially in children), its propensity to infect economically disadvantaged populations, and the paradoxical availability of relatively inexpensive and effective pharmaceuticals for its treatment (Savioli et al., [@CR34]). In our study, we found no relationship between infection status and fecal consistency (soft vs. firm). The lack of a strong measurable association between infection and acute gastrointestinal symptoms in our study population was surprising, but it is concordant with recent studies of humans. Although *Giardia* is associated with clinical diseases in developed economies, evidence is nevertheless mounting that it may be commensal in rural settings where people may normally be exposed to diverse parasite communities from an early age. For example, Peréz Cordón et al. ([@CR31]) found *Giardia* at 28.1% prevalence in diarrheic Peruvian children, as well as in 19.5% of nondiarrheic children, emphasizing the importance of asymptomatic patients in *Giardia* transmission where hygiene and sanitation are poor (Peréz Cordón et al., [@CR31]). Other studies in India (Traub et al., [@CR41]), Ethiopia (Ayalew et al., [@CR3]), Bangladesh (Dib et al., [@CR9]), and Peru (Hollm-Delgado et al., [@CR20]) have found similarly weak or nonexistent associations between *G. duodenalis* infection status and gastrointestinal symptoms, suggesting that *G. duodenalis* may coexist benignly with human hosts under conditions where acquired immunity to the pathogen can develop. Lack of early exposure and acquired immunity to *G. duodenalis* may, in fact, account for the role of the pathogen as a cause of sporadic diarrheal outbreaks in developed countries (Istre et al., [@CR22]), as well as a major cause of traveler's diarrhea for people from developed countries who visit endemic areas (Reinthaler et al., [@CR32]; Shlim et al., [@CR35]). The interplay between clinical symptoms and *Giardia* infections in wildlife are far less known. We hope to explore in detail the association between symptom and infection in apes in more detail in future work. Our results from the Goualougo Triangle provide a rare opportunity to observe baseline patterns of infection in undisturbed ape populations. Baseline patterns of parasitism are necessary to understand both individual animal health status and to assess what would constitute an "outbreak": a level of disease or parasitism in excess of what is normally observed. Thus, these results provide a first step toward an index of population health and disease risk assessment for conservation and management plans of these endangered African ape populations. Our results from the Kabo Concession suggest that a legacy of low-intensity selective logging did not result in the persistence of elevated prevalence of *Cryptosporidium sp*. and *Giardia* sp. in these ape populations. It is encouraging that the resident ape populations are not suffering from infections of these protozoa, because this suggests such potential health threats may be mitigated. However, if these protozoa are exceptionally pathogenic in wild ape populations, infected individuals may have been removed from the population long ago due to high associated mortality. We have yet to examine how patterns and risk of parasitism may differ during the various stages of the active logging process (i.e., exploration, road construction, and timber harvesting) from those that are the product of a legacy of selective logging. The expansion of timber activities in the neighboring undisturbed forests of the Kabo concession will enable us to address these questions through an ongoing progressive monitoring program that involves documenting ape populations, human disturbance, and the health implications during their interactions. Permission and logistical support for this research was provided by the Government of the Republic of Congo and the Congo Program of the Wildlife Conservation Society. C. Eyana assisted with data collection and a dedicated team of assistants from the villages of Bomassa and Makao provided invaluable assistance in the field. The authors thank N. Hauser and E. Canfield for laboratory assistance and two anonymous reviewers for helpful comments on an earlier draft of this manuscript. Funding for this research was provided the National Institutes of Health (NIH Grant T35 2006), the U.S. Fish and Wildlife Service (Great Ape Conservation Fund), and the National Geographic Society.

1.. Introduction {#s1} ================ Most of the world\'s population now lives in countries where fertility rates are plummeting, with a global demographic transition to low fertility projected by the end of this century \[[@RSPB20132732C1]\]. Yet explaining why and how fertility declines is a profound theoretical challenge \[[@RSPB20132732C2],[@RSPB20132732C3]\], because it remains unclear whether the fundamental causes are economic or cultural in nature \[[@RSPB20132732C2]--[@RSPB20132732C9]\]. Do individuals optimize their reproductive output based mainly on their own characteristics, or are their decisions driven by the frequency of a particular behaviour in their local community? Researchers agree that multi-causal, multi-level processes drive the onset and pace of demographic transitions \[[@RSPB20132732C4]--[@RSPB20132732C9]\]. Nonetheless, most research is focused on either micro or macro levels of analysis, with little integration between the individual predictors of fertility, for example education \[[@RSPB20132732C10]\], and the wider socio-cultural influences on reproductive decision-making. Studies comparing individual characteristics across a population often favour economic models of fertility decline; these emphasize the optimization of reproductive output based on the costs and benefits of investing in the 'quality' over the quantity of children \[[@RSPB20132732C11]--[@RSPB20132732C16]\]. However, studies comparing neighbouring populations have long suggested that people take their reproductive cues from the behaviour of others, and that social transmission and the diffusion of new cultural norms about fertility may be more important \[[@RSPB20132732C17],[@RSPB20132732C18]\]. Women\'s education is the key predictor of fertility decline \[[@RSPB20132732C10]\]. And yet, at both an individual and an aggregate level, it is unclear whether education is an economic or a cultural variable \[[@RSPB20132732C10],[@RSPB20132732C11]\]. Demographers have shown that province-, district- or sampling-area-level education is correlated with faster parity-transition rates \[[@RSPB20132732C19]--[@RSPB20132732C22]\], higher rates of contraceptive uptake \[[@RSPB20132732C19],[@RSPB20132732C23]--[@RSPB20132732C26]\] and lower fertility desires \[[@RSPB20132732C19]--[@RSPB20132732C24]\], independent of individual education. Neither micro- nor macro-level analyses explicitly incorporate the most important, intermediate-level social structures that define peoples' social interactions: their local communities and ego-networks. It remains unknown whether aggregated educational effects on individual fertility are due to social transmission of new cultural norms, or to ecological or economic differences between study areas that might incentivize fertility decline. Strong tests of cultural transmission hypotheses require community-level analysis, which most available sources of demographic data do not provide. Social network research can partly bridge this gap, but research thus far has tended to focus on contraceptive diffusion and not on fertility outcomes \[[@RSPB20132732C27]--[@RSPB20132732C30]\], and little is known about how networks themselves may change in the course of demographic transition. Measuring individual access to contraceptives or healthcare, exposure to migrants or mass media, or macro-level fertility rates may capture the general diffusion of information in a population over time \[[@RSPB20132732C11]\], but these do not necessarily capture the interpersonal cultural transmission of fertility norms and behaviour in local contexts. Cultural dynamics operate, by definition, at levels beyond the individual and should therefore be examined at the local community level. We know little about how the characteristics of other people in local villages and networks influence women\'s fertility outcomes, and the causal mechanisms by which communities may affect individuals have not been well tested. Cultural evolutionary theory provides plausible mechanistic links between individual decision-making and information flow in populations \[[@RSPB20132732C31],[@RSPB20132732C32]\]. This theory assumes that the frequency with which individuals in a group embrace different beliefs, or behave, is a good proxy for a cultural/social norm, and further, that summary measures of these frequency distributions (such as the mean and variance) can be used to compare cultural/behavioural norms between groups. Importantly, these frequencies can feed back to influence individual decisions, such that a person with similar individual characteristics may make radically different decisions in different communities, depending on what others do. Formal modelling shows that the average level of education in a community can theoretically alter the cultural dynamics of fertility decline, by modifying the social channels (modes) of communication and changing the rates by which fertility preferences are socially transmitted \[[@RSPB20132732C33]--[@RSPB20132732C35]\]. For example, higher rates of horizontal (peer-to-peer) and oblique (e.g. teacher-to-pupil) transmission may speed up the diffusion of low fertility norms even in populations that are homogeneously high in fertility, less economically developed and/or subject to natural selection pressure \[[@RSPB20132732C33]--[@RSPB20132732C35]\]. Increased frequencies of social interaction with more highly educated people and with non-kin, who are assumed to promote anti-natal values, are predicted to have a fertility-reducing effect \[[@RSPB20132732C32],[@RSPB20132732C36]\]. Higher average education in a community increases the chances that any person, irrespective of their own characteristics, is exposed to low-fertility behaviour, to contraceptive users or to other peoples' preferences for educating children \[[@RSPB20132732C31]--[@RSPB20132732C36]\]. The level of education in a community can therefore alter the speed of fertility decline by subtly affecting who we interact with, and thus the preferences we are likely to encounter. These cultural dynamics also theoretically work at a between-group level (i.e. between neighbouring populations or communities of a larger metapopulation). High rates of social transmission are predicted to homogenize behavioural norms within particular groups \[[@RSPB20132732C9],[@RSPB20132732C37],[@RSPB20132732C38]\], leading to between-group variation in the speed of fertility decline. This can expose individual communities themselves to low-fertility norms while at lower levels of education than would be expected if they were isolated \[[@RSPB20132732C35]\]. Cultural evolutionary processes can therefore potentially explain two important patterns that have been observed by demographers. First, fertility declines often differ among neighbouring populations at similar levels of economic development \[[@RSPB20132732C4],[@RSPB20132732C17],[@RSPB20132732C18]\], especially when there are social or cultural boundaries to social transmission, such as linguistic, religious or ethnic affiliations. Second, and conversely, fertility declines can spread across economic boundaries when neighbouring regions share a language or religion \[[@RSPB20132732C17],[@RSPB20132732C18],[@RSPB20132732C39]\]. Examining which mechanisms drive these patterns requires data at the appropriate level of aggregation and a representative sample of communities. In this study, we show that the average level of education in a community substantially accelerates fertility decline independent of individual- and other community-level characteristics, most likely by altering the social interactions of less educated women and thereby the types of fertility norms they encounter. We examined multi-level effects of education on both fertility decline and social network structure and composition, using new data from a randomized study of 1995 women living in 22 randomly selected communities (21 villages and one town) in a rural agricultural area of Poland. Our data provide a rarely observed but valuable insight into a mid-transitional population and our samples are representative of the villages women inhabit (see electronic supplementary material). Under-five mortality is low (see electronic supplementary material, table S1) and completed fertility is high (grand mean = 3.81 (±s.d. 2.15); [figure 1](#RSPB20132732F1){ref-type="fig"}), varying significantly across communities from an average of 3.03 (±s.d. 1.85) to 4.75 (±s.d. 2.56) children per woman at the end of her reproductive career (ANOVA, *p* = 0.024; see [figure 1](#RSPB20132732F1){ref-type="fig"} for raw data). Completed fertility is high relative to the Polish total fertility rate (TFR) when the data were collected (1.38 in 2010; electronic supplementary material) and to the TFR for the most relevant birth cohort (2.16 in 1949). Given that Poland is a developed European country, where fertility decline began around 1910 and which has had a below replacement TFR since the early 1990s \[[@RSPB20132732C40]\], such high and varied fertility is remarkable. Figure 1.Raw data showing between-community variation in completed fertility ± s.e. (*n* = 907 women aged 45+). Grand mean completed fertility ± s.d. is 3.81 ± 2.15 (solid line). Fertility is high relative to the cohort TFR for the most relevant (1949) birth cohort (2.16, dotted line), and to the national TFR when the data were collected in 2010 (1.38, dashed line). Community ID numbers give the order of sampling, with communities shown in order of increasing population density from left to right. There is no clear trend for decreased fertility in denser communities. The size of each point indicates the sample size of post-reproductive women. Note that group 12 is the town. Our study communities are characterized by centuries of peasant subsistence farming, but are now rapidly transitioning to exclusive dependence on labour-market activities. More than 65% (*n* = 1255) of respondents continue to live in subsistence farming households. The communities are geographically proximate (within an area of approx. 30 km^2^), at effectively equivalent levels of economic modernization, and are highly homogeneous with respect to religion and ethnicity. This enables us to isolate the effects of education from religious or ethnic differences between communities---this is important because fertility declines tend to spread among co-religionists and language speakers independent of economic modernization \[[@RSPB20132732C17],[@RSPB20132732C18],[@RSPB20132732C39]\]. In addition to individual variation, our data capture community-level variation in population size and density, proportion of inhabitants engaging in subsistence farming, average household material wealth and market integration (Material and methods; electronic supplementary material, table S1). We test a number of predictions designed to disentangle the contribution of cultural transmission hypotheses from purely individualistic optimizing models of fertility decline. If cultural transmission models are correct, then: (i) higher average levels of education in the community, rather than other community-level measures of economic modernization, such as wealth, market integration or population density, should accelerate fertility decline over and above individual characteristics; and (ii) less educated individuals should have lower fertility when living among a higher frequency of highly educated women. Women in highly educated communities should interact with (iii) more highly educated network partners, (iv) more sources of peer-to-peer (horizontal) rather than parent-to-child (vertical) transmission or (v) more non-kin. On the other hand, if individuals make reproductive decisions that are optimal relative to their own characteristics, then: (vi) variation in fertility should decrease along with variation in education; and (vii) community-level education should matter more for highly than poorly educated women, because factors like better educational opportunities, higher employment rates and more intensive reproductive competition may provide 'runaway' incentives and advantages to investing in 'quality' over quantity \[[@RSPB20132732C2],[@RSPB20132732C11]--[@RSPB20132732C16]\]. We standardized our measures to enable comparison at different levels of analysis. Our measures were also group-mean centred so that an individual\'s score represents her deviation from her community mean (i.e. her education relative to others in her local context rather than to others in the general population---this differs from most previous work, which compares randomly sampled individuals across a wider area). We then use group means to measure community effects---this has the advantage of cleanly partitioning the variation in the effects into within- and between-community components (see electronic supplementary material). 2.. Results {#s2} =========== Education was universally associated with reductions in fertility. Surprisingly, though, individual education was not a significant predictor in all communities, and the effect was larger at the community than at the individual level. A 1 s.d. increase in individual education was associated with a 13% reduction in fertility (95% CI (−0.19, −0.08), Wald *z*-test, *p* \< 0.001; electronic supplementary material, table S3), when considering the average ('fixed') effect across all communities. This association varied by community and was not always significant, ranging from a 6 to a 20% decrease in fertility for every 1 s.d. increase in education (see electronic supplementary material, table S4). Independent of individual factors, a 1 s.d. increase in average education in the community was associated with an 18% reduction in individual fertility (95% CI (−0.26, −0.10), Wald *z*-test, *p* \< 0.001; electronic supplementary material, table S3; [figure 2](#RSPB20132732F2){ref-type="fig"}*a*). A 1 s.d. increase in education at the community level therefore had a 1.3 times larger effect than a comparable increase at the individual level. [Figure 2](#RSPB20132732F2){ref-type="fig"}*a* shows that, independent of their own characteristics relative to others, women in the most educated community were predicted to have fewer than half as many children (1.64 ± s.e. 0.33) as those in the least educated community (3.25 ± s.e. 0.22). Note that these results are independent of community-level controls for modernization and a range of individual-level controls (see the electronic supplementary material). Figure 2.Community-level education is associated with a reduction in (*a*) mean (±s.e.) and (*b*) variance (given as s.d.) in fertility, independent of individual differentials and of controls for between-community variation (*n* = 1972). However, there is no clear relationship between (*c*) variance (given as s.d.) in fertility and variance (given as s.d.) in education. In (*a*) and (*b*), each interval on the *x*-axis corresponds to 1 s.d. in average education. All panels show model-adjusted relationships. Family sizes were more homogeneous in more highly educated communities. [Figure 2](#RSPB20132732F2){ref-type="fig"}*b* shows that the variance in predicted fertility (here, given as s.d. to retain the same units) declined as average education in the community increased, and these variances differ significantly across communities (Brown--Forsythe test, *p* \< 0.001). An intuitive explanation of this pattern might be that variance in fertility decreases along with variance in education. However, [figure 2](#RSPB20132732F2){ref-type="fig"}*c* shows clearly that this pattern of convergence has no relationship with how varied educational levels were in the community: family sizes were more similar independent of variation in individual education. This convergence is not predicted by purely economic approaches that rely on individual characteristics as the primary drivers of fertility decline. However, evolutionary-economic approaches do often argue that there may be 'runaway' benefits to investing in 'quality' over quantity in more educated populations, or reproductive competition to obtain the perceived fitness pay-offs to producing fewer high-quality children \[[@RSPB20132732C2],[@RSPB20132732C11]--[@RSPB20132732C16]\]. We examined this possibility by modelling a cross-level interaction between individual and average community education, but this revealed no significant relationship, and the model fit was not improved by including the interaction term (see electronic supplementary material, table S3). This suggests that between-community differences in fertility are not driven by the fertility-reducing behaviour of highly educated women. In other words, highly educated women do not necessarily reduce fertility more dramatically when the local context is conducive to greater investments in offspring education. Instead, it appears that social transmission may drive the differences between our study communities. Dividing each community into high and lower education brackets, with 'high' representing individuals with tertiary education and 'lower' representing everybody else, we re-examined the predicted fertility scores of only the lower-educated women, against the proportion of tertiary-educated women in the 22 communities. This allows us to examine how lower-educated women, who are not expected to reduce fertility substantially based on their individual characteristics, may be altering their reproductive strategies based on what other women around them do. As [figure 3](#RSPB20132732F3){ref-type="fig"}*a* shows, less educated women had lower fertility in communities with a higher proportion of tertiary-educated women. This implies that simply living among a higher frequency of tertiary-educated women may have influenced less educated women to reduce their own fertility. Figure 3.Community education effects on less educated women only. (*a*) Fertility decline is more dramatic among less educated women when the proportion of tertiary-educated women in the community is larger (*n* = 1667), and (*b*) less educated women living in highly educated communities have more educated social network partners (*n* = 1667). Both panels show model-adjusted relationships. What are the potential mechanisms driving these effects? Cultural evolutionary theory suggests that the level of education in a community may independently drive changes in ego-network size, density, proportion of alters who are kin, proportion of alters who are horizontal sources of transmission (neither parents nor individuals from the parental generation) and average education of alters. We tested these hypotheses using the same individual- and community-level controls, but found no evidence that community education is associated with social network changes in size, density, proportion of alters who are kin or the proportion who are horizontal sources of transmission (see electronic supplementary material, table S6). However, network alters were significantly more educated when ego lived in a highly educated community. Independent of ego\'s education, every 1 s.d. increase in community-level education was associated with a 12% increase in average education among the alters in the network (95% CI (0.04, 0.21), *t*-test, *p* = 0.006; electronic supplementary material, table S6). [Figure 3](#RSPB20132732F3){ref-type="fig"}*b* shows that this effect holds true when focusing again on less educated women only; alters in the network were significantly more educated when ego lived in a community with higher average education. This clearly indicates that simply living in a highly educated community may influence the characteristics of the friends that women choose to interact with, and thus the kinds of social information they are likely to receive. It is important to note that 48% of all network partners were not resident in the same community as ego (see electronic supplementary material, table S2). This result thus shows that less educated women have strong ties with more highly educated women both within and beyond the boundaries of their local communities. 3.. Discussion {#s3} ============== Individual characteristics are only part of the story of demographic transition, and studies that exclusively rely on them are likely to underestimate the importance of cultural transmission in driving fertility decline. Cultural norms are by definition factors beyond the individual; explicitly measuring and analysing community-level characteristics is therefore crucial for testing hypotheses about the local cultural transmission of fertility preferences. Community-level characteristics, for example the frequency of highly educated women, do not simply tell us about the distribution of resources and opportunities in a population; they themselves may drive interesting and important social dynamics, which cannot be captured using individual characteristics alone. While demographers have long been interested in measuring contextual effects on reproductive behaviour, these have not often been measured with appropriate community-level data and thus could not determine whether community effects existed at socially relevant levels of aggregation, or indeed if they were independent of alternative predictors of local community modernization. Our study shows that they are, and furthermore examines some of the causal pathways by which average education might feed back to influence individual behaviour. Among 22 groups of a culturally and ethnically homogeneous population at effectively equivalent levels of economic modernization, the education level of other people in the community affected individuals' reproductive outcomes. Indeed, the community average had a larger effect than the individual effect; a poorly educated woman was predicted to have twice as many children when living in a less educated community compared with living in a highly educated one. Family sizes were also less varied in highly educated communities, but this lack of variation in fertility was not due to homogeneity in educational levels. Reproductive norms can therefore converge on a small family size independent of individual variation in education. Thus, although highly educated women do tend to have lower fertility than less educated women within communities, between-community differences were driven by the behaviour of less-educated women only. It follows that comparing individual characteristics across large swathes of a population will be insufficient for a full understanding of the mechanisms of fertility decline. Less educated women had strong ties with significantly more educated women, both within and beyond their communities, when surrounded by more educated neighbours---this result is all the more striking because the other social network characteristics do not appear to change with community education. Any person in a more educated group is exposed to a greater variety of social learning models, as well as a greater number of individuals exhibiting low-fertility behaviour. Increasing average education may therefore alter both the content and structure of social interactions, consistent with cultural evolutionary models assuming that it leads to higher rates of social transmission, and thereby speeds up the diffusion of low fertility norms \[[@RSPB20132732C31]--[@RSPB20132732C36]\]. Interestingly, the fact that networks contained highly educated alters living outside the immediate community also points to a mechanism for spreading low-fertility norms between communities: higher average education within a community leads to interaction with more educated women across communities---this is consistent with formal modelling \[[@RSPB20132732C35]\]. Our results strongly suggest that social transmission from highly to poorly educated women accelerates the pace of fertility decline, once a critical mass of educated women is reached. The principal mechanism driving this effect does not appear to be a greater presence of non-kin \[[@RSPB20132732C36]\] or of horizontal sources of transmission \[[@RSPB20132732C33]--[@RSPB20132732C35]\] among the strong ties of ego-networks, but rather an increased frequency of interaction with more highly educated women. The individual adoption of low fertility norms and behaviour may therefore be prestige- or conformist-biased \[[@RSPB20132732C32],[@RSPB20132732C37],[@RSPB20132732C38],[@RSPB20132732C41]\], or indeed simply the result of a more neutral process of randomly copying the most frequent behaviour in the population \[[@RSPB20132732C31],[@RSPB20132732C32]\]. Less educated women in more educated communities are more likely to interact with and observe the reproductive behaviour of highly educated neighbours and friends, who may be more likely to endorse behaviours consistent with lower fertility. This exposure may change the expectations that less educated women have about the most appropriate behaviour in their local community, while also legitimizing and facilitating them to change their reproductive preferences within their private social networks. This can then contribute to a faster fertility decline in the community and lead to between-community variation in the pace of demographic transition. It is important to note both that the predicted differences in fertility between these geographically proximate communities are large, and that the ego-networks reported here are mainly distributed in the local villages and towns (see electronic supplementary material, table S2). District- or province-level analysis that aggregates across these kinds of small-scale social structures cannot adequately capture these dynamics. We do not wish to imply that rational-actor decision-making is unimportant in driving fertility decline, or that cultural transmission mechanisms are the only important causes of this phenomenon. On the contrary, our results demonstrate the need to integrate theory and evidence on cultural dynamics at the group level with individual decision-making that optimizes the perceived costs and benefits to reducing fertility. We simply emphasize that those perceptions are partly driven by cultural dynamics beyond the individual, with important implications for the speed of fertility decline. The approach outlined here provides a mechanistic link between the individualist, cost--benefit decisions that people appear to make, and which a plethora of micro-level data support, and the larger ideational shifts that make smaller family sizes acceptable and desirable. Irrespective of whether explicit 'norms' are being transmitted or whether individuals simply respond to the easily observable correlation between fewer children and higher social status, social transmission of fertility decline should homogenize behaviour within groups \[[@RSPB20132732C9],[@RSPB20132732C37],[@RSPB20132732C38]\]; which is a prediction that our data supports. Whether the initial fertility-reducers in a population are responding to the real or perceived economic costs of children is essentially a different question to whether cultural transmission can spread low-fertility norms, and our analysis focuses only on the latter question. All that is needed for the process to take off is that a certain proportion of individuals invest in education and that education is reliably associated with lower fertility. Individuals do not need to know what the mean level of education is in their community for this characteristic to affect their probability of encountering low-fertility norms and behaviour. Community-level characteristics can alter the social cues to the most appropriate behaviour in ways that are neither predicted by a person\'s individual characteristics nor by the level of economic or structural modernization in a particular group. Community-level factors may therefore have independent causal force \[[@RSPB20132732C35]\]. These results raise important theoretical questions about the extent to which individuals are influenced by their neighbours in perhaps the most important decisions of their lives---how many children to have---decisions that are expected to be under selection pressure \[[@RSPB20132732C42]\]. Our findings also have important implications in mid-transitional populations. Apart from the fact that improving average education in a community will accelerate fertility decline, even among less educated individuals \[[@RSPB20132732C19]--[@RSPB20132732C26]\], our results imply that the mean may be more important than the mode. It is possible that universally high education may not be as important as ensuring that a critical mass of highly educated women is reached, by which point the changing structure of social transmission will facilitate the diffusion of low-fertility norms to all individuals in the community. That feedback between community- and individual-level education ratchets up the speed of fertility decline may also have significant implications for global fertility projections. More fine-grained data and multi-level analysis are needed to explicitly test hypotheses about the mechanisms driving these effects. Although only one study to date has examined the influence of education on fertility at multiple higher levels of aggregation \[[@RSPB20132732C20]\], more local contexts mattered most and our study supports this assertion. Thus, in the era of 'big data', there remains an important place for smaller-scale anthropological demographic research in elucidating the mechanisms that underlie fertility change. 4.. Material and methods {#s4} ======================== Data were collected between 2009 and 2010 using a standardized anonymous questionnaire and structured interview (see the electronic supplementary material for further discussion). We used Poisson, linear and binomial multi-level regression \[[@RSPB20132732C42]\] techniques, which explicitly account for the non-independence of individuals living in the same local context, to simultaneously model the effects of individual- and community-level education on (i) fertility, and on the following social network characteristics of (ii) size, (iii) density, (iv) proportion of kin, (v) proportion of horizontal sources of transmission and (vi) average education of network partners, excluding ego (see the electronic supplementary material, tables S3--S6). At the individual level, all models included controls for age, farmer status, household material wealth and household market integration. In the fertility analysis, we additionally controlled for age squared and experience of under-five mortality, as well as including an additional random term at the individual level to account for over-dispersion in the data. At the community level, all models included controls for population density, proportion of the population engaging in farming, average household material wealth and average household market integration (see the electronic supplementary material). These are important controls for alternative measures of modernization at the relevant local levels of analysis because all of these factors might jointly drive higher education and lower fertility preferences. We examined the following alternative model specifications to check the robustness of our results in predicting fertility outcomes. None of these altered our findings (see electronic supplementary material, table S5): (i) excluding unmarried women (*n* = 374) from the analysis, (ii) including nonlinear parametrizations of individual education on fertility within communities, (iii) using uncentred variables, (iv) removing non-significant variables or substituting average education for any other community-level predictor; the model was a significantly worse fit whenever we excluded average community education. We thank all the women who took part in our study as well as our local field assistants from all stages in the project, and the Wojtas and Markiewicz families for support in the field. We also gratefully acknowledge the parish Rectors who helped to publicize our study. We thank A. Powell, A. Kandler, K. Magid, D. Lawson, K. Snopkowski and C. Moya for discussion of earlier manuscript drafts, and two anonymous reviewers for insightful comments. {#s5} Informed consent was obtained from all participants and the study protocols were approved by the Ethics Committee of the Department of Anthropology at University College London. Funding statement {#s6} ================= This work was financially supported by grants from the UK Economic and Social Research Council, the Wenner Gren Foundation for Anthropological Research and Osmundsen Initiative (Dissertation Fieldwork grant no. 8182), UCL Graduate School, the Gay Clifford Fund and the European Research Council (grant no. AdG12 249347).

Introduction ============ The devastating prevalence of type 2 diabetes mellitus (T2DM), together with better knowledge of pathogenesis of the disease, have prompted search for new treatment strategies. Dipeptidyl peptidase-4 (DPP-4) inhibitors offer novel opportunities for the management of T2DM as they are endowed with a new mechanism of action, potentiating the physiological function of endogenous incretin, glucagon-like peptide 1 (GLP-1). Although cleavage of GLP-1 is recognized as the only clinically relevant enzymatic effect of DPP-4, this enzyme recognizes as target glucose-dependent insulinotropic polypeptide (GIP), another incretin that shares insulinotropic activity with GLP-1, but is devoid of other properties such as reduction of glucagon release, delay of gastric emptying, and suppression of appetite (Baggio and Drucker, [@B2]). In addition, besides incretins, other substrates of DPP-4, not directly involved in the control of glycemic homeostasis have been identified. These include different peptides with recognized renal and cardiovascular effects such as Brain Natriuretic Peptide (BNP), Neuropeptide Y (NPY), Peptide YY (PYY), and type 1 Stromal-Derived Factor (SDF-1α; Hocher et al., [@B30]). Although we are far from identifying a direct link between inhibition of DPP-4, modified cleavage of these peptides and clinical correlates, it cannot be excluded that changes in their metabolism account for additional outcomes of DPP-4 inhibition that are increasingly emerging from preclinical studies. Drugs belonging to the class of DPP-4 inhibitors are virtually devoid of any hypoglycemic effect and have no significant impact on body weight. In addition, with respect to other drugs belonging to the incretin family, DPP-4 inhibitors offer the undeniable advantage of oral administration. Five DDP-4 inhibitors are available, including sitagliptin, saxagliptin, vildagliptin, and the more recent linagliptin and alogliptin (Russell-Jones and Gough, [@B44]). At present sitagliptin, linagliptin and saxagliptin are marketed in US, sitagliptin, saxagliptin, and vildagliptin have been approved in Europe and alogliptinis already present only in the Japanese market and has just received approval by US Food and Drug Administration (FDA). Linagliptin represents the most recently approved of this class of anti-diabetic drugs both in US and Europe. Rather than analyzing the clinical properties of linagliptin, when used either as monotherapy or as add-on therapy to other anti-diabetic drugs, this review is aimed at emphasizing specific features of this molecule from preclinical studies to pharmacokinetics and interaction with other drugs. Particular attention is given to linagliptin effect in animal models, as this may represent a unique opportunity to know in depth features of the molecule that can help to disclose properties not so evident from clinical studies. Chemistry ========= (8-(3-(R)-Aminopiperidin-1-yl)-7-but-2-ynyl-3-methyl-1-(4-methyl-quinazolin-2-ylmethyl)-3,7dihydropurine-2,6-dione; linagliptin) is a DPP-4 inhibitor with a xanthine-based structure. Its nature makes it different from other compounds of the same class (saxagliptin, sitagliptin, vildagliptin, with the exception of alogliptin) that are all peptidomimetic molecules. Linagliptin binds tightly to the core of the DPP-4 enzyme forming three hydrogen bonds between the amino function on the piperidine ring and acceptor groups on residues Glu205, Glu206, Tyr662. A fourth hydrogen bond is formed between the C-6 carbonyl of the xanthine scaffold and the backbone amide of residue Tyr631. Finally, aromatic stacking interactions are established between the xanthine ring and Tyr547 as well as between the quinazoline ring and Trp629 (Eckhardt et al., [@B9]). Pharmacology: Pharmacodynamics ============================== Linagliptin inhibits DPP-4 activity *in vitro* with an IC50 value of 1 nM, exhibiting a potency higher than all other DDP-4 inhibitors which yield IC50 values of 19 (sitagliptin), 24 (alogliptin), 50 (saxagliptin), and 62 (vildagliptin) nM, respectively (Thomas et al., [@B49]). Maximal efficacy for DPP-4 inhibition is similar for linagliptin and other compounds of the same class. The kinetics of linagliptin interaction with DPP-4 reveals a slow dissociation from the enzyme with a calculated *k*~off~ rate (3 × 10^−5^/s) that is approximately 10-fold slower than the off-rate for vildagliptin (Thomas et al., [@B49]). Linagliptin exhibits a 10,000-fold higher selectivity for DPP-4 than for other dipeptidyl peptidases such as DPP-2, DPP-8, and DPP-9, a property that makes this molecule, together with alogliptin, the highest selective compound vs. DPP-4. Similarly, the selectivity of linagliptin vs. peptidases N and P, prolyl oligopeptidases, and proteases such as trypsin, plasmin, and thrombin is much more than 10,000-fold. Linagliptin also inhibits fibroblast activating protein (FAP) with an IC50 of 89 nM (about 90-fold higher selectivity vs. DPP-4; Gao et al., [@B17]; Thomas et al., [@B49]; Deacon, [@B6]). This effect appears interesting as FAP is a protease with high potentials: it is considered in fact a marker of fibroblasts in tumor growth and is also expressed in activated fibroblasts involved in tissue remodeling (O'Brien and O'Connor, [@B40]). Whether and how the inhibition of FAP may contribute to the clinical effects of linagliptin is not known at present, but it is of interest that linagliptin exerts a role in reparative processes following wound injury. Pre-Clinical Studies ==================== Pharmacokinetics ---------------- Pharmacokinetics of linagliptin is characterized by its binding to the DPP-4 target, not only in plasma, but mainly in tissues, where the enzyme is bound to membranes (Greischel et al., [@B27]). After single or repeated oral administration in male Fischer rats, linagliptin is highly present in kidney, liver, and lung, coincident with the distribution of DPP-4 (Fuchs et al., [@B15]; Retlich et al., [@B42]). Absorption of linagliptin is modified by intestinal P-glycoprotein. In Wistar rats receiving a single oral dose of linagliptin, administration of a P-glycoprotein inhibitor produced an increased bioavailability of the drug (Fuchs et al., [@B14]), suggesting that P-glycoprotein activity limits absorption of linagliptin. Intravenous administration produces a non-linear increase in linaglipin tissue concentrations with increasing doses, suggesting the presence of saturable binding sites. These correspond with peripheral DPP-4, either in plasma and tissues, as also suggested by the fact that, in DPP-4-deficient rats, the amount of linagliptin in tissues increases linearly with dose and is eliminated rapidly whereas, in wild-type rats, increase in tissues is less dose proportional and its peripheral decline is much slower compared to DPP-4-deficient rats (Retlich et al., [@B42]). Following administration of a low-dose of linagliptin, AUC is significantly reduced and terminal half-life is significantly shorter in DPP-4 deficient rats compared to wild-type animals (Retlich et al., [@B42]). Hence, binding to DPP-4 in plasma and tissues potently affects linagliptin pharmacokinetics leading to a non-linear profile, a long terminal half-life, and a prolonged exposure to the drug. Based on these observations, and as also confirmed in human studies, tissue accumulation of linagliptin is unlikely and steady state conditions are reached very quickly. Metabolism does not primarily affect linagliptin that is rather mainly eliminated unmodified. After repeated oral administration, the great part of linagliptin is excreted through the intestinal tract, a major part through the bile and a minor component (about 12%) directly into the gut, independently of biliary excretion (Fuchs et al., [@B15]). In contrast only traces are found in urine (Fuchs et al., [@B14],[@B15]). Furthermore, following a single oral administration, the majority of drug is eliminated within 24 h, although about 4% is still present after 96 h. In vivo studies in animal models -------------------------------- Administration of linagliptin to Han Wistar rats produces inhibition of DPP-4 in plasma that appears after 30 min, is maintained 7 h after treatment and is still present at 24 h. Maximal effects are observed with doses of 3 and 10 mg/kg (90% inhibition), whereas more than 70% inhibition is achieved following 1 mg/kg. Calculated ED50 values are 0.3 and 0.9 mg/kg, 7 and 24 h post-dose, respectively (Thomas et al., [@B49]). If inhibition of plasma DPP-4 activity is compared with that achieved using the same dosage of other DPP-4 inhibitors, linagliptin exhibits a similar efficacy vs. other compounds, at the 7 h time point. An exception is sitagliptin, that appears less effective at both time points examined, 7 and 24 h. In contrast, at 24 h, linagliptin exhibits a sustained and maximal inhibition of DPP-4 activity, not observed with the other drugs (Thomas et al., [@B49]). Similar to other DPP-4 inhibitors, linagliptin (1 mg/kg) reduces blood glucose levels and suppresses glucose AUC between 0 and 120 min (about 50% reduction), when administered 45 min before glucose challenge. In contrast to other compounds, linagliptin maintains this effect (between 20 and 30% AUC reduction) when administration of the drug precedes by 16 h glucose load (Thomas et al., [@B49]). In diabetic Zucker rats, administration of linagliptin 30 min or 24 h before an oral glucose tolerance test (OGTT) reduces blood glucose peak levels and plasma glucose AUC. This effect is accompanied by increase in GLP-1 and insulin levels (Thomas et al., [@B49]). In animals treated with linagliptin, while the increase of GLP-1 levels precedes glucose load and is maintained thereafter, insulin levels are not elevated in the fasting state, but sharply increase after feeding. In an established animal model of diabetes, obtained with administration of a low-dose streptozotocin (STZ) and high-fat diet (HFD), the HFD/STZ rat exhibiting both hyperglycemia and hyperinsulinemia, oral daily administration of linagliptin (1, 3, 10 mg/kg), for 4 weeks, produces a dose-dependent inhibition of plasma DPP-4 (by 59, 78, and 87% with increasing doses) vs. levels at time 0, when measurements are carried out at the end of the study period (day 27), 21 h after last administration of the drug (Thomas et al., [@B50]). Reduction of DPP-4 activity is paralleled by decrease in blood glucose concentrations, glucose AUC, and post-prandial glucose levels, the latter indicating improved glucose tolerance (Thomas et al., [@B50]). An increase in active GLP-1 is also observed following multiple, once daily dosing of linagliptin, with no change of insulin levels. Furthermore, daily administration of linagliptin causes a significant, dose-dependent decrease of HbA1c with changes of 0.8, 0.9, and 1.0% at linagliptin doses of 1, 3, and 10 mg/kg, respectively. Similar effects after chronic treatment with linagliptin are observed in a more drastic model of diabetes, the Zucker diabetic rat that exhibits marked insulin resistance, progressive beta-cell failure with age and a severe evolution of the disease. In this animal model, linagliptin acutely reduces glucose concentrations after OGTT and elevates GLP-1 levels, differently from vildagliptin, in a sustained and long-lasting manner (Thomas et al., [@B49]) without affecting basal insulin levels. In Zucker diabetic rat, chronic treatment with linagliptin does not cause change in body weight, but reduces cumulative food intake by 6%, in contrast to vildagliptin that does not modify this parameter. In NOD mice that spontaneously develop a form of insulin-dependent diabetes caused by an autoimmune T cell-dependent destruction of islets, thus resembling human type 1 diabetes, linagliptin affects all the above described parameters. Specifically, exposure to linagliptin, given with food at doses of 3--10 mg/kg for 8 weeks, delays the onset and reduces the incidence of diabetes; it in fact induces reduction of basal blood glucose levels, decreases water intake and enhances GLP-1 levels. In addition, it significantly increases beta-cell mass in the pancreas without modifying the non-beta-cell component (Jelsing et al., [@B34]). Interestingly, recent evidence suggests a positive interaction of linagliptin with an inhibitor of sodium-glucose co-transporter 2 (SGLT2) that is emerging as a novel target for treatment of T2DM. In genetically diabetic mice, linagliptin improves in fact the effects induced by the SGLT2 inhibitor, not only in terms of glycemic control and islet function, but also with regards to the expression of genes associated with the inflammatory response (Chen et al., [@B4]). Additional effects of linagliptin in animals models --------------------------------------------------- The effects of linagliptin are not restricted to direct control of glycemic levels. As already mentioned, linagliptin in fact increases beta-cell mass component in the pancreas (Jelsing et al., [@B34]). In human cell islets *in vitro*, linagliptin (100 nM for 48 h) protects from apoptosis induced by exposure to insults such as palmitic acid, interleukin-1/interferon-γ, or hydrogen peroxide (Shah et al., [@B47]), restores beta-cell function and reverts oxidative stress responsible for islet dysfunction. When administered, in a period comprised from 4 weeks to 3 months (3 or 30 mg/kg), to diet-induced obese rats and HFD-fed mice, linagliptin significantly reduces plasma leptin levels, as well as hepatic steatosis, measured as hepatic red oil staining and/or by magnetic resonance spectroscopy (Kern et al., [@B37]; Klein et al., [@B38]). The latter effect is particularly relevant as liver fat content correlates with metabolic disturbances, and signs of improved hepatic steatosis, as those observed following linagliptin, can be predictor of a better controlled glucose homeostasis. In addition, in these obese mice, high doses of linagliptin significantly reduce inflammatory response in adipose tissue, as measured by the number of infiltrating macrophages, and negatively affect the expression of several genes related with fatty acid synthesis and oxidation and primarily involved in inflammation (Kern et al., [@B37]). It is highly plausible that all these effects on lipid metabolism significantly contribute to the global improved insulin sensitivity induced by linagliptin. In obese rats, however, linagliptin does not modify total body weight, cumulative food intake or total body fat (Klein et al., [@B38]), but interestingly, it reduces weight regain following exenatide withdrawal (Vickers et al., [@B51]). Linagliptin also seems to be endowed with vascular protective properties as it induces vasodilation in isolated aortic rings under basal conditions, in a concentration-dependent manner. This effect is mimicked by vildagliptin, but not by other DPP-4 inhibitors. It also affects vascular dysfunction-related reactive oxygen species and inflammation, thus exhibiting anti-oxidative and anti-inflammatory properties (Kröller-Schön et al., [@B39]). Of note, linagliptin exhibits a protective effect against diabetes-dependent renal impairment. In diabetic endothelial nitric oxide synthase (eNOS) knockout mice, an established model of diabetic nephropathy, linagliptin, given as add-on therapy to the standard treatment telmisartan, improves albuminuria, a predictor of diabetic nephropathy, and reduces plasmatic levels of osteopontin, a marker of vascular calcification. These effects are accompanied by reduction of glomerulosclerosis and renal oxidative stress, as measured by accumulation of malondialdehyde, as well as reduction of tumor necrosis factor α (TNF-α) levels, a marker of systemic inflammation (Alter et al., [@B1]). Inhibition of DPP-4 has been reported to have beneficial effects on protection from myocardial ischemic damage and ensuing improvement of cardiac function. Among possible mechanisms of action, a reduced degradation of myocardial SDF-1α, a substrate for DPP-4, with subsequent increased recruitment of circulating CXCR-4+ stem cells that may favor tissue repair. Alternatively, increased incretin acting at its receptor could activate pathways improving tissue repair. Treatment with linagliptin (given once daily for 7 days, starting 2 days before ischemia/reperfusion) induced less severe histological features of cardiac damage and reduced infarct size and myocardial fibrosis after ischemia reperfusion in rats (Hocher et al., [@B31]). A similar reduction of infarct size is observed when linagliptin treatment is started 4 weeks before ischemia/reperfusion and is prolonged for 8 weeks thereafter. Linagliptin treatment also increased the expression of CD34, c-kit, CXCR-4, and SDF-1α, suggestive of enhanced recruitment of CXCR-4+ circulating progenitor cells that may be responsible for a better reparative process (Hocher et al., [@B31]). In *ob/ob* mice, that are known to exhibit delayed and impaired wound healing, treatment with linagliptin (3 mg/kg/day) while reducing hyperglycemia, improves wound closure and wound morphology. However, even at very high doses (30 mg/kg/day), it does not affect wound healing process in normoglycemic mice. The effects observed in *ob/ob* mice are accompanied by reduced influx and persistence of polymorphonuclear neutrophils, as well as infiltration of macrophages in wounds (Schürmann et al., [@B46]). Such effects appear extremely important and identify a potential additional effect of linagliptin. Reduction of hyperglycemia in fact does not necessarily correlate with improvement of wound healing. In addition, the high expression of DPP-4 in the skin and its down-regulation following injury suggest a role for the enzyme in reparative processes. Interestingly enough, in *ob/ob* mice, DPP-4 expression in the skin is absent at early times following injury, but is increased in the late phases of wound repair (Schürmann et al., [@B46]). Finally, very recent data have shown that treatment with linagliptin produces a neuroprotective effect, independent of its glucose lowering action. Specifically, in a model of stroke induced by middle cerebral artery occlusion in normal and diabetic mice, linagliptin is effective in reducing stroke volume (with a greater effect in normal mice) and in increasing the number of viable neurons in the peri-infarct area (Darsalia et al., [@B5]). Studies in Humans ================= Pharmacokinetics ---------------- First-human study was carried out to test pharmacodynamic and pharmacokinetic features as well as safety and tolerability of linagliptin (Hüttner et al., [@B32]). A single dose of oral linagliptin ranging from 2.5 to 600 mg/day produced different *t*~max~ values, from 0.7 to 3 h, depending on the dose. The 5 mg dose resulted in a *C*~max~ of 5.71 nmol/L, *t*~max~ of 1.47 h, *t*~1/2~ of 69.7 h, and very limited excretion in urine. In dose groups comprised between 5 and 50 mg, the terminal half-lives were approximately 70--80 h (69.7--79.9) whereas they ranged from 128 to 184 h in dose groups from 100 mg and above (Hüttner et al., [@B32]). Comparable pharmacokinetic profiles were observed in type 2 diabetic patients following single administration of linagliptin (2.5, 5, and 10 mg). Linagliptin was rapidly absorbed with a *t*~max~ of approximately 1.5 h. AUC and *C*~max~ increased with dose, but, as already observed in healthy subjects (Hüttner et al., [@B32]), the increase was less than dose proportional. This is likely due to high-affinity saturable binding of linagliptin to DPP-4. At low concentrations (1 nmol/L), in fact, binding of linagliptin to plasma protein is complete (99%) but it decreases to 70--80% at concentrations higher than 100 nmol/L (Fuchs et al., [@B16]). At low plasma concentrations, linagliptin free fraction is thus very low whereas, at higher plasma concentrations, linagliptin saturates DPP-4, its free fraction increases and pharmacokinetic parameters, including volume of distribution and clearance, are modified. As a consequence, terminal half-life is long, ranging from 110 to 130 h, depending on the dose administered. However, daily administration induced only moderate accumulation and a steady state level was yielded after 4--6 days following 1, 2.5, and 5 mg dose (Heise et al., [@B29]). Linagliptin showed a low rate of hepatic metabolism and renal excretion was minimal, being less than 1% on day 1 and between 3 and 6% on day 12 (Heise et al., [@B29]). Food intake does not modify linagliptin pharmacokinetics in a clinically relevant manner. In fact, in subjects administered a high-fat meal, AUC~0--72~ of linagliptin was similar to that of control subjects. Although *t*~max~ was delayed by about 2 h (from 1.02 to 2.99 h) and *C*~max~ was reduced by 15%, these changes did not affect the overall exposure to the drug and did not modify its pharmacodynamics (Graefe-Mody et al., [@B21]). It can then be concluded that linagliptin can be taken independent of meals as food does not seem to have any clinically relevant effect on the efficacy of the drug. Pharmacodynamics ---------------- Linagliptin induced marked inhibition of DPP-4 activity in both healthy subjects (Hüttner et al., [@B32]) and type 2 diabetic patients (Heise et al., [@B29]). According to Hüttner et al. ([@B32]), a dose as low as 2.5 mg linagliptin caused about 70% inhibition of DPP-4 activity, whereas almost complete inhibition was observed at doses above 25 mg. The time needed to achieve maximum inhibition was variable depending on the dose and was about 3 h with the lowest doses. The inhibition was long-lasting and DPP-4 activity did not return to baseline levels after 96 h. In diabetic patients, with the 5 mg dose, 92.3% DPP-4 inhibition was observed at steady state and over 80% inhibition over a 24 h-interval after drug intake was still present (Heise et al., [@B29]). This level of inhibition (\>80%) was maintained also after long-term treatments (12 and 24 weeks), when mean linagliptin trough concentrations were constant over time, at concentrations around 6.5 nmol/L (Del Prato et al., [@B8]). When patients were subjected to OGTT, linagliptin significantly reduced plasma glucose levels in a dose-dependent manner, starting at the 2.5 mg dose. This effect was present at 26 h, but was more pronounced at steady state, measured at 13 days (Heise et al., [@B29]). Pharmacokinetics and pharmacodynamics in patients with hepatic or renal impairment ---------------------------------------------------------------------------------- In a phase I, parallel group-comparison study, 5 mg/day linagliptin was administered for 7 days to healthy controls and subjects with mild, moderate, and severe hepatic impairment. None of the groups showed significant changes in linagliptin pharmacokinetic parameters (AUC, *C*~max~, and *t*~max~), so that no dose adjustment is required in these patients. In fact, about 80% DPP-4 inhibition was present in all groups examined (Graefe-Mody et al., [@B26]). This observation appears particularly relevant in view of the potential correlation of type 2 diabetes with altered hepatic function and mainly with respect to the predominant non-renal excretion of linagliptin. Pharmacokinetic profiles following linagliptin administration were investigated, both after single dose and at steady state, in patients with renal impairments of various degrees. Absorption of linagliptin after single and multiple dosages was similar to that observed in healthy and diabetic subjects, i.e., approximately or less than 3 h to reach maximal plasma concentrations. In addition, plasma concentrations 24 h after a single dose were similar in patients with mild, moderate, and severe renal impairment and were not modified by the concomitant presence of type 2 diabetes. AUC and maximal plasma concentrations at steady state were similar or only slightly increased in patients with moderate renal impairment when compared to healthy controls. In addition accumulation *t*~1/2~ of linagliptin following multiple 5 mg/day dose did not show any significant change in control and diabetic subjects, regardless of renal function (Graefe-Mody et al., [@B20]). Linagliptin present in urine was very low, around 1% in healthy controls, and reached only 4% in patients with renal impairment. Renal dysfunction did not modify linagliptin pharmacodynamics, as about 80% inhibition of DPP-4 was present, irrespective of renal function conditions. These features make linagliptin unique among DPP-4 inhibitors as all other compounds are mainly excreted by the kidney and exhibit modified kinetics under conditions of renal impairment. After a single oral administration, approximately 87% of sitagliptin, 75% of saxagliptin, and 85% of vildagliptin are excreted in urine. The three drugs show profound changes in their plasma levels with increasing renal impairment, exhibiting already a twofold increase when a moderate renal dysfunction is present (Graefe-Mody et al., [@B20]). Clinical Studies ================ A detailed report of all the clinical studies that have been carried out with linagliptin is beyond the scope of present review. Linagliptin has in fact shown a clinical efficacy that is comparable to that observed with other DPP-4 inhibitors. A summary of the clinical trials with linagliptin is reported in Table [1](#T1){ref-type="table"}. ###### **Summary of major results obtained with linagliptin in clinical studies**. Treatment Adjusted mean (SE) change from baseline in HbA1c (%) Adjusted mean (SE) change from baseline in FPG (mmo/l) Achievement of target HbA1c \<7% (% of patients) Reference --------------------------------------------------------------- ------------------------------------------------------ -------------------------------------------------------- -------------------------------------------------- -------------------------- **4 WEEKS OF TREATMENT** Linagliptin, 2.5 mg (*n* = 26) −0.31\* −16.6\*\* -- Forst et al. ([@B11]) Linagliptin, 5 mg (*n* = 16) −0.37 −21.4 -- Linagliptin 10 mg (*n* = 19) −0.28 −16.6 -- Placebo (*n* = 16) -- +3.2 -- **24 WEEKS OF TREATMENT** Linagliptin, 5 mg once daily (*n* = 336) −0.44 (±0.05) −0.5 (±0.1) 25.2 Del Prato et al. ([@B8]) Placebo (*n* = 167) +0.25 (±0.07) +0.8 (±0.2) 11.6 **18 WEEKS OF TREATMENT** Linagliptin, 5 mg once daily (*n* = 151) −0.39 (±0.14) −0.7 (±0.3) 27.9 Barnett et al. ([@B3]) Placebo (*n* = 76) +0.21 (±0.16) +0.4 (±0.3) 15.1 **26 WEEKS OF TREATMENT** Linagliptin, 5 mg once daily (*n* = 159) −0.13 (±0.7) −0.3 (±0.1) 30.2 Kawamori et al. ([@B36]) Linagliptin, 10 mg once daily (*n* = 160) −0.19 (±0.7) −0.4 (±0.1) 34.4 Voglibose, 0.2 mg tid (*n* = 162) +0.19 (±0.7) +0.1 (±0.1) 22.2 **12 WEEKS OF TREATMENT** Placebo (*n* = 80) +0.63 (±0.08) +0.4 (±01) 10.0 **12 WEEKS OF TREATMENT** Linagliptin, 1 mg once daily (*n* = 65) −0.15\*\*\* −0.36\*\*\* 15 Forst et al. ([@B10]) Linagliptin, 5 mg once daily (*n* = 66) −0.48 −1.22 15 Linagliptin, 10 mg once daily (*n* = 66) −0.42 −0.90 21 Glimepiride, 1 mg once daily (*n* = 65) −0.90 -- -- Placebo (*n* = 71) +0.25 +0.7 1.4 (Add-on therapy to metformin) **24 WEEKS OF TREATMENT** Linagliptin, 5 mg once daily (*n* = 524) −0.49 (±0.04) −0.6 (±0.1) 26 Taskinen et al. ([@B48]) Placebo (*n* = 177) +0.15 (±0.06) +0.6 (±0.2) 9 (Add-on therapy to metformin) **24 WEEKS OF TREATMENT** Linagliptin, 5 mg once daily (*n* = 142) −0.5 (±0.1) −0.5 (±0.2) -- Haak et al. ([@B28]) Metformina, 500 mg bid (*n* = 144) −0.6 (±0.1) −0.9 (±0.2) -- Metformina, 1000 mg bid (*n* = 147) −1.1 (±0.1) −1.8 (±0.2) -- Linagliptin, 2.5 mg + metformina 500 mg bid (*n* = 143) −1.2 (±0.1) −1.8 (±0.2) -- Linagliptin, 2.5 mg + metformina 1000 mg bid (*n* = 143) −1.6 (±0.1) −2.7 (±0.2) -- Placebo (*n* = 72) +0.1 (±0.1) +0.6 (±0.3) -- **24 WEEKS OF TREATMENT** Linagliptin, 5 mg + pioglitazone 30 mg once daily (*n* = 130) −1.06 (±0.06) −1.8 (±0.1) 30.5 Gomis et al. ([@B18]) Placebo + pioglitazone, 30 mg once daily (*n* = 259) −0.56 (±0.09) −1.0 (±0.2) 42.9 **24 WEEKS OF TREATMENT** Linagliptin, 5 mg once daily (*n* = 793) −0.72 (±0.03) −0.3 (±0.1) 29.2 Owens et al. ([@B41]) Placebo (*n* = 265) −0.10 (±0.5) +0.4 (±0.1) 8.1 (Add-on therapy to metformin and a sulphonylurea) *SE, standard error; HbA1c, glycated hemoglobin; FPG, fasting plasma glucose*. *\*Placebo-corrected mean change in HbA1c; \*\*mg/dl; \*\*\*mean (+SE) change from baseline*. Co-administration of linagliptin with other anti-diabetic drugs --------------------------------------------------------------- No reciprocal interaction was evident when linagliptin and metformin were administered together (Graefe-Mody et al., [@B25]). Metformin *per se* did not modify DPP-4 activity and did not affect inhibition of the enzyme induced by linagliptin in healthy subjects. No clinically relevant pharmacokinetic interactions at steady state between the two drugs were observed. Metformin increased AUC by 20% and *t*~max~ of linagliptin by about 50%, but these effects are unlikely to have clinical relevance. Conversely, linagliptin co-administration only slightly reduced metformin *C*~max~ and delayed its *t*~max~. A randomized open-label, two-way crossover study examined the effects of co-administration of multiple oral doses of linagliptin (5 mg/day × 6 days) and a single dose of glyburide (1.75 mg/day administered for 1 day) on the reciprocal pharmacokinetic parameters of the two drugs. Glyburide did not affect any pharmacokinetic feature of linagliptin and conversely, linagliptin slightly reduced glyburide AUC and *C*~max~ without achieving a clinically relevant effect (Graefe-Mody et al., [@B22]). As glyburide is a recognized substrate of CYP2C9 and CYP3A4, this lack of interaction confirms that linagliptin does not modify the activity of these metabolic enzymes. Linagliptin did not either affect pharmacokinetics of pioglitazone when the two drugs were administered together. This may appear particularly significant since pioglitazone is a substrate for CYP2C8. Hence, pioglitazone (45 mg/day) did not modify linagliptin (10 mg/day) AUC and *C*~max~, whereas linagliptin only slightly reduced pioglitazone *C*~max~ (Graefe-Mody et al., [@B24]). The two drugs appear fully administrable in combination with no relevant pharmacokinetic interaction. Interaction of linagliptin with other drugs not directly correlated with diabetes --------------------------------------------------------------------------------- An open-label, multiple-dose study examined, in healthy subjects, the pharmacokinetic interaction of linagliptin (10 mg/day) and simvastatin (40 mg/day), that is classified as a sensitive substrate of CYP3A4. Changes in CYP3A4 activity may significantly modify simvastatin metabolism, leading even to the occurrence of severe adverse effects. Co-administration of linagliptin increased (by 10--30%) both exposure and *C*~max~ of simvastatin and its active analog simvastatin acid (Graefe-Mody et al., [@B23]), without however achieving clinical relevance in terms not only of pharmacokinetic parameters, but also of safety and appearance of adverse effects during co-administration. Eighteen healthy subjects were enrolled to evaluate the reciprocal pharmacokinetic interaction of linagliptin and warfarin, a CYP2C9 substrate. No pharmacokinetic parameter of linagliptin, S-warfarin, and R-warfarin was modified and co-administration did not significantly change the coagulation parameters prothrombin time (PT) and international normalized ratio (INR; Graefe-Mody et al., [@B19]). Together with the study in which linagliptin was co-administered with glyburide (Graefe-Mody et al., [@B22]), this study supports the concept that linagliptin does not significantly affect CYP2C9 activity and that substrates for this enzyme, such as warfarin, can be administered in conjunction with linagliptin without dose changing. To test the interaction of linagliptin with drugs that act as substrate for P-glycoprotein, an open-label cross over study enrolling 20 healthy subjects was carried out to evaluate co-administration of linagliptin and digoxin (Friedrich et al., [@B13]). *In vitro* studies have in fact indicated linagliptin as a weak *P*-glycoprotein inhibitor (Ishiguro et al., [@B33]). There was no significant change in digoxin pharmacokinetics and all parameters were within the bioequivalence range 80--125%, a critical aspect due to the low therapeutic index of digoxin. Due to the wide use of anti-diabetic drugs in women of fertile age, the potential of impact of linagliptin on pharmacokinetics of an oral contraceptive combination composed of ethinyl estradiol (30 μg) and levonorgestrel (150 μg) was examined in an open-label, multiple-dose study carried out in healthy women. None of the parameters examined including both AUC and *C*~max~ at steady state were significantly affected by linagliptin co-administration, with values within the standard acceptance limit of bioequivalence 80--125% (Friedrich et al., [@B12]). Safety and Tolerability ======================= In a phase I study carried out by Hüttner et al. ([@B32]), linagliptin was shown to be well tolerated in healthy male volunteers when administered in doses ranging from 2.5 to 600 mg per day. The overall incidence of adverse effect was similar in placebo- and linagliptin-treated subjects and there were no serious adverse effects (Hüttner et al., [@B32]). Hypoglycemia was never present and no changes of electrocardiographic or other cardiac function parameters could be detected (Hüttner et al., [@B32]). In this respect, a randomized, double-blind, placebo-controlled study was carried out in healthy subjects to evaluate the potential effect of linagliptin on QT interval. At the therapeutic dose (5 mg/day) and the 20-fold therapeutic dose (100 mg/day) administered, linagliptin did not cause any prolongation of the QT interval (Ring et al., [@B43]). However, with regards to general cardiovascular risk in patients treated with linagliptin, a large meta-analysis evaluated cardiovascular events observed in eight phase III studies using linagliptin and different comparator anti-diabetic drugs (Johansen et al., [@B35]). It can be concluded that a reduction of cardiovascular events is observed in patients with type 2 diabetes treated with linagliptin vs. pooled comparators, suggesting a potential cardiovascular benefit of linagliptin. Safety and tolerability data obtained with linagliptin in eight phase III clinical trials have been collected in a recent review article and data obtained from about 3,500 patients were analyzed altogether (Schernthaner et al., [@B45]). The overall incidence of adverse events and serious adverse events was similar in placebo- and linagliptin-treated patients (55 vs. 55.8 and 2.7 vs. 2.8%, respectively). Hypoglycemia was the most important adverse effect occurring in 5.1% of patients receiving placebo vs. 8.2% of patients on linagliptin treatment. Severe hypoglycemia was absent in both groups. The higher incidence of hypoglycemia in linagliptin group was most likely attributable to the concomitant treatment with a sulfonylurea, as incidence of hypoglycemia was drastically reduced when the sulfonylurea was not allowed (0.6 vs. 1% with linagliptin and placebo, respectively). Among other adverse events, the gastrointestinal disorders were more common, occurring in 10.5% of linagliptin-treated patients, very similar to placebo--treated (11.4%). Diarrhea appeared in 2.1% of patients on either linagliptin or placebo. Other adverse events occurring at a frequency \>2% were infections of the upper respiratory tract, urinary tract or nasopharyngitis, headache, backpain, and hypertension. Minor adverse events, including increase of hepatic enzyme and serum creatinine, blood, and lymphatic disorders, occurred at an incidence \<2% and were similar in patients treated with linagliptin or placebo. Cardiac disorders, mainly palpitation and tachycardia prevailed in linagliptin-treated patients, but always at an incidence \<0.5% (Schernthaner et al., [@B45]). An important issue that has been evaluated in several different studies is weight gain, as it represents a main adverse event with the use of other anti-diabetic drugs. DPP-4 inhibitors have generally a neutral effect on weight and, accordingly, the major part of studies on linagliptin reports no significant change of body weight or waist circumference at 12 or 24 weeks when the drug is administered alone (Hüttner et al., [@B32]; Del Prato, [@B7]; Del Prato et al., [@B8]), or in combination with other anti-diabetic drugs (Owens et al., [@B41]; Taskinen et al., [@B48]; Haak et al., [@B28]). Exceptions were provided by studies carried out by Forst et al. ([@B10], [@B11]) that reported a slight reduction in body weight after 4 or 12 weeks of treatment with linagliptin in monotherapy (Forst et al., [@B11]) or in addition to metformin (Forst et al., [@B10]). In contrast, an increase in body weight was reported after 24 weeks of treatment with linagliptin plus pioglitazone (2.3 kg) vs. pioglitazone plus placebo (1.2 kg; Gomis et al., [@B18]). Conclusion ========== Linagliptin is an oral, highly selective DPP-4 inhibitor causing over 80% DPP-4 inhibition throughout the 24-h dosing interval. It exhibits non-linear pharmacokinetic properties with a less than dose proportional profile and is almost completely eliminated via the enteric system, with less than 5% found in urine. This particular feature makes linagliptin unique among other DPP-4 inhibitors, as no dosage adjustment is required in patients with renal impairment and the drug can be used safely. The efficacy of linagliptin in improving glycemic control and reducing HbA1 has been analyzed in several different studies with linagliptin used in monotherapy or as add-on therapy to other commonly used anti-diabetic drugs. Finally linagliptin has a good tolerability profile, with the vast majority of adverse events of mild to moderate severity. Besides the pharmacokinetic and clinical features of linagliptin, more impressively, preclinical studies have highlighted beneficial effects on the cardiovascular system, and in skin reparative processes that deserve further investigations. Conflict of Interest Statement ============================== Maria Angela Sortino and Pier Luigi Canonico have been consultants of Boehringer Ingelheim Pharma GmbH. Tizian Sinagra has no conflict of interest to declare. [^1]: Edited by: Romesh Khardori, Eastern Virginia Medical School, USA [^2]: Reviewed by: Undurti Narasimha Das, UND Life Sciences, USA; Sarah Louise Gray, University of Northern British Columbia, Canada [^3]: This article was submitted to Frontiers in Diabetes, a specialty of Frontiers in Endocrinology.

1. Introduction {#sec0005} =============== Neural representations of the body formed within the human brain, known as the body representation system (BRS), is central to the understanding of motor functions ([@bib0095], [@bib0230]). The BRS is constantly updated using sensory information, especially proprioception that encompasses the perception of positional changes and movements of body parts ([@bib0305]). Using vibration-evoked proprioceptive illusions, neuroimaging studies suggested that two networks constitute the cerebral basis of the [BRS]{.ul} ([@bib0275], [@bib0280], [@bib0290], [@bib0270]): (i) a sensorimotor control network -- i.e. motor and somatosensory cortical regions, basal ganglia, thalamus, cerebellum -- that contributes to the formation of the body representations, [a]{.ul}nd is involved in on-line control (fast corrections) of movement, and (ii) a fronto-parietal network extending from the inferior frontal gyrus to the posterior parietal cortex (e.g. inferior parietal lobule) that integrates environmental information together with bodily information into a single percept, thereby [providing a corporeal representation]{.ul} adjusted to the environmental context. [Furthermore]{.ul}, the fronto-parietal network in the right hemisphere has also been found to be involved in corporeal self-awareness ([@bib0050], [@bib0285]). Although it is obvious that the [BRS]{.ul} must be updated during development due to many factors such as morphological changes, acquisition of motor skills, and cognitive practice, age-related changes in its cerebral correlates have never been investigated directly. Indirect information on the developmental trajectory of the [BRS]{.ul} comes from resting state functional magnetic resonance imaging (fMRI) studies. The sensorimotor control network is already topologically adult-like by the age of two while higher-order networks including the fronto-parietal network are topologically incomplete, presenting a less specialized architecture compared to adults ([@bib0130]). Further studies reported [that]{.ul} within-network connectivity changes [stop in]{.ul} late childhood (∼10 years old) for the sensorimotor network and that they continue until adulthood for the frontal and parietal networks ([@bib0195], [@bib0200]). Therefore, the fronto-parietal network supporting the [BRS]{.ul} presents an extended development compared to the sensorimotor network. A recent fMRI study using tendon vibration also demonstrated that the proprioceptive brain network still undergoes refinements during and beyond adolescence, mostly the fronto-striatal connections that exhibit functional prunin[g]{.ul} leading to a more restricted topology ([@bib0055]). Therefore, although the sensorimotor network is likely earlier to mature compared to the fronto-parietal network, it may not yet be mature by late childhood. Outcomes from structural MRI studies would support this view. Studies on cortical grey matter development from childhood to adulthood reported a maturational sequence from sensorimotor to higher-order association regions, specifically from the precentral gyrus to the prefrontal cortex in the frontal lobe and from the postcentral gyrus to the angular/supramarginal gyri in the parietal lobe (e.g. [@bib0160], [@bib0350]). Likewise, a study by [@bib0410] showed that grey matter networks establishing sensory and motor regions were already well-developed in early childhood, although not yet adult-like. In contrast, higher-level cognitive networks were undeveloped in early childhood and showed an important [amount]{.ul} of change during adolescence. However, regional variations in the grey matter maturation pattern of the sensorimotor network appear to exist, especially [in]{.ul} the subcortical regions that demonstrate important age-related changes in grey matter density during adolescence ([@bib0370]). Considering white matter maturation, it is less clear whether the sensorimotor network reaches maturity before the fronto-parietal network. Diffusion tensor imaging (DTI) studies showed that the microstructural characteristics of association fibres, and especially the superior longitudinal fasciculus that connects the parietal cortex to the frontal gyrus, become adult-like by late adolescence ([@bib0015], [@bib0220], [@bib0225], [@bib0215], [@bib0355]). This supports an extended development of the fronto-parietal network. However, late to mature in adolescence are also projection fibres, such as the corona radiatia connecting the basal ganglia to the cortex, as well as cerebellar connections ([@bib0015], [@bib0220], [@bib0355]). Moreover, connections at terminal grey matter sites in basal ganglia were found to mature even in later adulthood ([@bib0220], [@bib0355]). Therefore, a set of white matter fibres involved in the sensorimotor network continue to mature during and beyond adolescence, suggesting once again that this network is not completely mature by late childhood. Using a protocol of kinaesthetic illusions in children (7--11 years) and adults (25--40 years) in fMRI, the aim of the present study was to evaluate the degree of maturation of the sensorimotor and fronto-parietal networks subtending the BRS by late childhood. We expected to find larger differences in activation levels between children and adults [in]{.ul} fronto-parietal regions, i.e. a more immature fronto-parietal network by late childhood. A secondary objective was to examine the extent to which structural brain maturation influences the functional development of the networks that implement the [BRS]{.ul}. To this end, we investigated group differences in fMRI results while statistically controlling for differences in grey and white matter between children and adults. 2. Methods {#sec0010} ========== 2.1. Participants {#sec0015} ----------------- Forty seven healthy right-handed individuals including fifteen adults (mean age ± SD: 32.4 ± 4.5; 9 females) and thirty two children (from seven to eleven years old) took part [in]{.ul} the experiment. Data from 9 of the adult participants were previously reported by [@bib0050] and six additional adult participants were recruited to complete the adult group. The children sample was finally restricted to twenty-two individuals (mean age ± SD: 9.0 ± 1.4; 15 females) due to the exclusion from the analysis of 10 children with excessive head movements during fMRI scanning (cf. Section [2.5.1](#sec0040){ref-type="sec"} fMRI data analysis). Sex distribution did not differ between the two age groups (χ^2^ (4, n = 37) = 0.3; p = 0.6). Adult participants, parents of minors, and children gave written informed consent. The study was approved by the research ethics committee CPP Sud-Méditerranée 1. 2.2. Paradigm {#sec0020} ------------- The cerebral correlates of the BRS were examined using a tendon vibration paradigm, which consisted in vibrating the tendons of the right and left tibialis anterior muscles to excite the muscle spindle primary endings (e.g. [@bib0050], [@bib0055]). To this end, custom-made pneumatic vibration devices, driven by constant air pressure, were placed perpendicularly to the anterior right and left ankles using elastic straps (contact area = ∼ 6 cm^2^). Frequencies were delivered at 30 Hz and at 100 Hz with a 0.5 mm amplitude, leading to four vibration conditions: right and left tendon vibration at 30 Hz (R30 and L30; the control 'no illusion' condition) and at 100 Hz (R100 and L100; the 'illusion' condition). These stimulation parameters were selected based on the fact that (i) 20--40 Hz frequencies drive weak discharges of the primary endings, which are not likely to elicit kinaesthetic illusions, and (ii) ∼100 Hz frequency optimally activates primary endings, generally providing consistent illusory movements (i.e. plantar-flexion) ([@bib0050], [@bib0055], [@bib0275], [@bib0315], [@bib0335], [@bib0340]). 2.3. Pre-scanning session {#sec0025} ------------------------- In the present study, participants first experienced a pre-scanning session lying supine outside the scanner with the eyes closed, during which they were presented with the four vibratory stimulations (12-s long R30, L30, R100, and L100 vibrations; each stimulation presented twice). The vibration conditions were presented in a random order and each vibration was followed by a rest period during which the participants were questioned as to whether the stimulation generated illusory movements, and were then requested to verbally describe them. The six additional adults reported illusory movement at 100 Hz but not at 30 Hz stimulation. Finer-grained evaluation was conducted in children to increase confidence in self-reports on illusions. After the 12 seconds period of vibration, the children had to report about what they felt and were asked to reproduce (if any) the illusory percept by moving their foot. In the 100 Hz condition, all children felt their foot 'going down' and produced a plantar-flexion. In the 30 Hz condition, 6 children reported plantar-flexion illusion while the other 16 children reported a 'vibrating' sensation. Children were also asked to (i) score the vividness defined as how realistic the illusion was, between 0 and 4, a score of 0 corresponding to the absence of illusion and a score of 4 to a plantar-flexion illusion similar to an actual movement ([@bib0275]), and (ii) to report the amplitude of the illusory movement, using an angle measurement scale. The angle measurement scale corresponded to a lying individual with a degree of freedom about the ankle. The experimenter rotated the foot until the estimated position was reached, as indicated by the children. An angle of 0° corresponded to the absence of illusory plantar-flexion and a positive value to a plantar-flexion, the larger the value the more important the plantar-flexion. In sum, none of the adults reported experience of illusory movement in the 30 Hz conditions while all of them experienced illusory plantar-flexion in the 100 Hz conditions. Six children out of twenty-two reported illusory movements in both the 30 Hz and the 100 Hz conditions, with a tendency for less vivid and important illusions in the former conditions (vividness R30: 1.5 ± 1.4; vividness L30: 1.3 ± 1.2; angle R30: 17.5 ± 10.3°; angle L30: 19.2 ± 9.3°) compared to the latter conditions (vividness R100: 2.5 ± 1.4, L100: 3 ± 1; angle R100: 25.2 ± 10.3°, L100: 23.5 ± 13.8°). The other 16 children reported illusions in the 100 Hz conditions only (vividness R100: 2.8 ± 0.7, L100: 3.1 ± 0.7; angle R100: 13.6 ± 2.9°, L100: 18.5 ± 5.1°). 2.4. MRI data acquisition {#sec0030} ------------------------- After the familiarization pre-scanning session, the subjects were placed head first and supine into a 3T MRI scanner (Medspec 30/80 AVANCE, Bruker, Ettlingen, Germany). Supports were used to minimize head movement, including a soft strap attached around the head and air pads placed between the lateral sides of the head and the radio-frequency head coil. The participants also wore a headphone for communication purpose. fMRI time series were acquired with a T2\*-weighted gradient echo-planar imaging sequence (42 interleaved axial slices acquisition; 3 mm thickness; 0.5 mm interslice gap; reconstruction matrix = 64 × 64; field of view = 192 mm × 192 mm; repetition time = 2.8 s; echo time = 30 ms; flip angle = 84°). The scanning planes were parallel to the anterior commissure − posterior commissure and covered the top of the cortex down to the base of the cerebellum. The scanning session was composed of five runs each lasting 5 min and including 12-s long conditions (epochs) of vibration (R30, L30, R100, L100) and REST. Each vibration condition was repeated three times per run. The order of vibration conditions was randomized within a run, and REST epochs were inserted between all vibration conditions to ensure the relaxation of the muscle spindles. At the end of each run, subjects were questioned about the illusory movements via headphones. For each stimulation condition, we asked whether it induced illusory movements, and if yes, whether all three repetitions (each stimulation condition having been repeated three times within a run) induced illusory movements. Subjects answered the questions using a finger button response system. Answers were identical to the verbal report of the pre-scanning session for all participants. The six additional adults and all children reported consistent illusion for the R100 and the L100 in all runs. In addition, the six children who reported having experienced an illusion at 30 Hz during the pre-scanning phase also reported consistent illusory movements for the R30 and L30 stimulation in all runs. We preferred asking the subjects to report the presence of illusions after each run instead of after each stimulation to avoid methodological issues related to movement preparation. Indeed, reporting illusions on an epoch basis would have involved pre-movement activity of the responding finger in areas that were processing proprioceptive information, including precentral and parietal areas ([@bib0255], [@bib0345], [@bib0380]). Structural MRI data were also acquired using a 12 min three-dimensional T1-weighted scanning sequence (MP-RAGE; repetition time = 9.4 ms; echo time = 4.4 ms; inversion time = 800 ms; field of view = 179 mm × 256 mm × 180 mm, reconstruction matrix = 256 × 256 × 180). Finally, the participants underwent [an 8]{.ul} [min]{.ul} MR-diffusion scanning. MR-diffusion images were acquired using a dual spin-echo, single shot echo-planar imaging sequence (TE = 86.659 ms; TR = 10400 ms; 52 axial slices, 2.3 mm thickness, without gap; field of view = 242 mm x 242 mm and matrix size = 128 × 128). Each MR-diffusion data set included 8 non-diffusion-weighted and 36 diffusion-weighted images acquired with a b-value of 1000 s/mm^2^. 2.5. Data analyses {#sec0035} ------------------ ### 2.5.1. fMRI {#sec0040} Image pre-processing and statistical fMRI data analysis were conducted with SPM8 (Wellcome Department of Imaging Neuroscience, London, UK) running in Matlab 7.5 environment (Mathworks, Inc., Sherbon, MA, USA) and custom-made Matlab scripts. Each run included 113 images, including six dummy images of magnetic field saturation that were discarded. The remaining images were first (i) slice-time corrected, (ii) realigned to the first image of the time series to correct for head movement between scans, (iii) unwrapped to remove residual movement-related variance ([@bib0010]), and (iv) co-registered to high-resolution structural data. The structural image was normalized to the MNI (Montreal Neurological Institute) T1 template image and the resulting parameters were used for normalization of the functional images, which were resampled to 3-mm isotropic voxel size and smoothed with an 8-mm FWHM Gaussian kernel. Ten children in which translational and rotational head movement as obtained from realignment (step ii) exceeded 3 mm and 3°, respectively, were discarded from the study. We applied this rule-of-thumb since higher amount of motion are not corrected properly by applying motion correction procedures ([@bib0115]). Accordingly, the final children population considered for analysis included twenty two individuals. Pre-processed data were afterwards analysed using the univariate two-stage summary statistics random effect model ([@bib0060], [@bib0120], [@bib0180]). Task-dependent changes in BOLD signal were modelled as boxcar regressors time-locked to the onsets of the vibrations (R30, L30, R100, and L100) and REST conditions. These regressors were convolved with the canonical hemodynamic response function (HRF) of SPM8 and entered into the general linear model (GLM). Constant terms and parameters for head movement estimated by motion realignment procedure were entered at the individual level as covariates of no interest to account for shifting signal levels across runs and influence of head motion on BOLD signal, respectively. Specifically, a 24-parameter autoregressive model including current and past position parameters along with the square of each parameter was used to account for the cumulative effects of motion on spin magnetization ([@bib0125]). It has been demonstrated that this modelling-based strategy is the most efficient to account for motion-related variations in the BOLD signal ([@bib0400]). [Finally]{.ul}, a high-pass filter (cut-off period = 128 s) was applied to remove low-frequency drifts in the data. Maps of parameter estimates were computed from the GLM to reveal the magnitude of activation of the regions subtending the BRS, i.e. individual SPM{t} maps of the contrast RL100 (illusion condition) \> RL30 (no illusion condition). Note that such contrast pooled the right and left stimulations. Individual maps were then entered into a second-level random-effects GLM to evaluate group differences. [A]{.ul} summary measure of head motion (i.e. mean framewise displacement, labelled mean FD) was also included as a covariate in the second-level random-effects GLM. The mean FD was continuous and corresponded to the mean head displacement over the 5 functional runs and was calculated as the summed absolute values of the derivatives of the translational and rotational realignment estimates (after converting rotational estimates to displacement at 50 mm radius; [@bib0300]), averaged over all scans. Associating subject-level correction with the realignment parameters and group-level correction for mean FD appears to be an optimal strategy to regress out motion-related artefacts from the data ([@bib0110], [@bib0400]). Multiple comparisons correction of statistical maps at the second-level was conducted using a cluster-based extent thresholding of p \< 0.05 (FWER-corrected) calculated based on the Gaussian random field method and following a primary threshold set at p \< 0.005. We set the primary threshold at a relatively liberal threshold given that our working hypotheses dealt with large-scale networks (i.e. fronto-parietal network and motor network including cortical and subcortical structures), or equivalently large clusters of activation, and not anatomically localized activity pattern ([@bib0395]). The extent to which brain structure influenced BOLD contrast differences between the two groups was examined from individual white (DTI) and grey (grey matter mask from T1 scan) matter information implemented into the second-level random effects GLM as covariates, thereby adjusting functional results for structural differences between children and adults. Details on how these covariates have been obtained are provided below. ### 2.5.2. White matter {#sec0045} Processing of MR-diffusion data was conducted using tools in FSL5.0 ([@bib0360]). As a first step, non-brain tissues were deleted from diffusion- and non-diffusion-weighted images (BET tool; [@bib0365]) and diffusion-weighted images were aligned to the first non-diffusion-weighted image using affine registration (FLIRT tool; [@bib0190]). Affine registration was used to reduce misalignment between the images due to head motion and eddy currents. The effects of subject motion on diffusion-weighted images were assessed afterwards using the procedure developed in [@bib0405], which relied on four different measures: (i) the average image-by-image translation, (ii) the average image-by-image rotation, (iii) the percentage of slices with signal drop-out, and (iv) the signal drop-out severity. The average image-by-image translation and rotation indicated acceptable (i.e. ∼3 mm translation and ∼3° rotation) between-image motion in all subjects. These four measures enable the capture of the global frame-to-frame motion as well as the frequency and severity of rapid slice-to-slice motion, and was used to detect volume data corrupted by motion. In particular, one image among the 36 diffusion-weighted images demonstrated more than 5% of drop-out for three children. These images were therefore removed from subsequent analysis. Finally, a tensor model was fit to the images (DTIFIT tool) at each voxel and the fractional anisotropy (FA) was obtained voxelwise. FA images of each participant were aligned to a standard space (FMRIB58 FA image) using a nonlinear registration and then, using an affine registration, aligned to the MNI152 template. Individual FA values were finally extracted from 8 tracts of interest (averaged over the left and right hemispheres for each tract) of the Johns Hopkins University (JHU)-ICBM-DTI-81 white matter labels atlas ([@bib0265], [@bib0385]). These tracts included association (i.e. corticospinal tract, corona radiata, external capsule, cingulum, superior longitudinal fasciculus, and superior fronto-occipital fasciculus) and projection (i.e. anterior limb of the internal capsule) fibres as well as cerebellar pathway (i.e. cerebral peduncle) that interconnect the different cortical, subcortical and cerebellar regions most likely involved in [BRS]{.ul} processing. Finally, the set of FA values were decomposed into principal components and the first component (accounting for 67% of variance; Fig. 1) served as the covariate that was carried forward to the second-level GLM analysis of fMRI data. ### 2.5.3. Grey matter {#sec0050} Grey matter information was obtained using voxel-based morphometry ([@bib0020], [@bib0165]) implemented in VBM8 toolbox. As a first step, all T1 images were inspected for artefacts and the centre point was placed on the anterior commissure. All images were subsequently (i) normalized (affine registration) to the MNI152 standard space (using a Bayesian framework), (ii) segmented in grey matter, white matter and cerebro-spinal fluid partitions, and (iii) smoothed with an 8-mm FWHM Gaussian kernel. Grey matter density (dGM) was extracted from regions of interest (ROIs) of the probabilistic Harvard-Oxford cortical and subcortical structural atlases that cover 48 cortical and 21 subcortical structural areas ([@bib0070]; [@bib0250]). Specifically, individual dGM was obtained from fourteen ROIs (averaged over the left and right hemispheres for each ROI) that most likely compose the sensorimotor and the fronto-parietal networks of the BRS, including the inferior frontal gyrus, the insular cortex, the caudate, the putamen, the thalamus, the parietal operculum, the postcentral gyrus, the precentral gyrus, the cingulate gyrus, the superior parietal lobule, the precuneus, the superior frontal gyrus, the prefrontal cortex, and the middle frontal gyrus. As for the white matter, principal component analysis was finally run to decompose the set of dGM values. The first component accounted for 65% of variance of the data set (see [Fig. 1](#fig0005){ref-type="fig"}), and served as the covariate in second-level GLM analysis of fMRI data.Fig. 1Grey and white matter maturation. Mean ± SE of the first principal component (1st PC) for FA (filled line) and dGM (dashed line) for the children and the adults. The 1st PC accounted for 67% and 65% of the variance of the FA and dGM data sets, respectively. Group comparison revealed a significant difference both for FA and dGM (t~35~ = −3.8 and t~35~ = 7.9; \*\*p \< 0.001, respectively), reflecting an increase of the FA and a decrease of the dGM with age. FA: fractional anisotropy; dGM: grey matter density.Fig. 1 3. Results {#sec0055} ========== 3.1. Brain regions contributing to the BRS in children and adults {#sec0060} ----------------------------------------------------------------- The RL100 \> RL30 contrast, or equivalently the comparison between the illusion and the no illusion conditions, revealed significant activations (p \< 0.05 FWER-corrected) in several cortical and subcortical areas for both groups ([Fig. 2](#fig0010){ref-type="fig"}). As illustrated in [Fig. 2](#fig0010){ref-type="fig"}, clusters of activation were located in the basal ganglia (putamen, caudate) and neighbouring regions (claustrum, anterior insula), motor-related regions (anterior cingulate cortex -- BAs 24 and 32, supplementary motor area -- BA 6, primary somatosensory cortex -- BA 3), and associative regions (inferior frontal gyrus − BA 44, and inferior parietal lobule -- BA 40) in the two hemispheres. These results support the idea that both the sensorimotor and the frontoparietal regions are already contributing to processing bodily information in late childhood. Detailed information on the clusters of activation (spatial localization, number of active voxels, intensity) are provided in [Table 1](#tbl0005){ref-type="table"}.Fig. 2Activation maps for the RL100 \> RL30 contrast for (A) the children (green) and (B) the adults (red). Maps are thresholded at p \< 0.05 family-wise error rate (FWER)-corrected with cluster-based extent thresholding at p \< 0.005 and k = 20 voxels.IFG: inferior frontal gyrus; FPC: frontopolar cortex, IPL: inferior parietal lobule, ACC: anterior cingulate cortex, SI: primary somatosensoy cortex; SMA: supplementary motor area. The Insula-Basal ganglia complex involved the right and left anterior insula, putamen, and caudate. This complex also overlapped with the claustrum in both right and left hemisphere.Fig. 2Table 1Activation peaks during right and left vibratory stimulus (RL100 \> RL30 contrast) for adults and children. Z-values refer to significant activation peaks at p \< 0.005 (uncorrected for multiple comparisons). All cluster reported are significantly active at p \< 0.05 (FWER-correction). Coordinates are reported in the MNI space; L: left hemisphere, R: right hemisphere.Table 1RL100 \> RL30Peak locationBASidex(mm)y(mm)z(mm)z-valuesCHILDRENCluster \# 1(3751 voxels)Insula13L−392136.07R45−1135.56Ant. Cing. Gyrus24R6−7465.75L−6−7435.46Precentral Gyrus44L−51−1105.67Primary Motor Area4L−6−34705.63Sup. Motor Area6L−3−19735.61ClaustrumR33−1135.49Inf. Parietal Lobule40R60−34315.32L−51−31254.99Sup. Front Gyrus6L−9−19765.23Cluster \# 2(227 voxels)Frontopolar Cortex10L−304444.919L−3035314.39Cluster \# 3(99 voxels)Inf. Front. Gyrus44R334114.07Ant. Cing. Cortex32R154473.13Frontopolar Cortex10R1556−22.97

ADULTSCluster \# 1(558 voxels)Inf. Parietal Lobule40L−51−34226.07L−51−31344.28Postcentral Gyrus1;2L−60−28403.87Cluster \# 2(884 voxels)Insula13R362314.93Precentral gyrus44R545104.90ClaustrumR33−1104.70Inf. Frontal Gyrus44R5411223.96PutamenR24814.30Cluster \# 3(568 voxels)Insula13L−42−1104.65Inf. Frontal Gyrus44L−5714284.53PutamenL−27−13163.88Middle Frontal Gyrus6L−452433.87ClaustrumL−302043.86Precentral Gyrus44L−541173.79Cluster \# 4(317 voxels)Inf. Parietal Lobule40R39−37404.5R63−34314.28Postcentral Gyrus1;2R63−25404.35Cluster \# 5(317 voxels)Sup. Motor Area6L−311494.04R62583.17Ant.Cing. Cortex24;32L020433.78R35313.68 Yet, the spatial extent of the pattern of the children's network was larger than in the adult's network ([Table 1](#tbl0005){ref-type="table"}). Visual inspection of [Fig. 2](#fig0010){ref-type="fig"} clearly highlights such difference between children and adults. Moreover, a significant cluster of activation was found in the frontopolar cortex (BA 10) in the children only (cf. top part of [Fig. 2](#fig0010){ref-type="fig"}). 3.2. Differences in brain activation with age {#sec0065} --------------------------------------------- Contrast between children and adults (i.e. Children \> Adults) revealed a significant cluster in the frontopolar cortex ([Fig. 3](#fig0015){ref-type="fig"}A; [Table 2](#tbl0010){ref-type="table"}). This is a direct consequence of the exclusive activation of this region in children relative to adults previously reported[.]{.ul} The inverse contrast, namely Adults \> Children, revealed two significant clusters with activation peaks localized in the right and left inferior parietal lobules and primary somatosensory cortices ([Fig. 3](#fig0015){ref-type="fig"}B; [Table 2](#tbl0010){ref-type="table"}). As showed by the barplots of [Fig. 3](#fig0015){ref-type="fig"}B, which represent the percent signal change (PSC) in the RL30 and RL100 conditions for the activation clusters, the PSC in the adults was found to be lower in the RL30 condition compared to the RL100 condition while there was no clear PSC difference between both conditions in the children. The RL30 condition seems to elicit for those regions a ceiling effect on the BOLD signal in children. Altogether, these results indicate an immaturity of the BRS in late childhood that manifests at the brain level through (i) the additional recruitment of a prefrontal region to the sensorimotor and fronto-parietal networks, and (ii) a parietal activation less sensitive to variations in proprioceptive inputs (i.e. a ceiling effect for weak discharges of primary endings).Table 2Activation peaks for the contrast Children \> Adults (top) and Adults \> Children (bottom) with and without correction for group differences in white (FA) and grey (dGM) matter. Z-values refer to significant activation peaks at p \< 0.001 (uncorrected for multiple comparisons). All cluster reported are significantly active at p \< 0.05 (FWER-correction). Coordinates are reported in the MNI space; L: left hemisphere, R: right hemisphere, FA: fractional anisotropy, dGM: grey matter density.Table 2CHILDREN \> ADULTSPeak locationBASidex(mm)y(mm)z(mm)z-valuesfMRICluster \# 1 (480 voxels)FrontoPolar Cortex10L05374.38R155643.03

fMRI-FACluster \# 1 (511 voxels)FrontoPolar Cortex10L05374.35R−356−84.22R653−54.22

fMRI-dGMCluster \# 1 (147 voxels)FrontoPolar Cortex10L−1253193.77R95073.74R356133.60

ADULTS \> CHILDRENfMRICluster \# 1 (327 voxels)Postcentral gyrus1; 2L−57−28404.63Inf. Parietal Lobule40L−36−43524.00Cluster \# 2 (143)Inf. Parietal Lobule40R39−34405.57R51−34404.39Postcentral Gyrus3R60−25405.17[^1]Fig 3Comparison of the adults and children group activity for the RL100 \> RL30 contrast. A: Children \> Adults contrast (Green) and B: Adults \> Children contrast (Red). Maps are thresholded at a voxel-wise threshold of p \< 0.001 uncorrected and a cluster extend threshold of p \< 0.05 FWE-corrected. Barplots show the percent of signal change (PSC) ± SE for each significant cluster. PSC were extracted using MarsBaR ([@bib0035]). Dashed bars represent the RL30 condition and filled bars the RL100 condition. FPC: Frontopolar Cortex, SI: Primary Somatosensory Cortex; IPL: Inferior Parietal Lobule.Fig 3 Correction of the fMRI results for the white matter or the grey matter (i.e. FA or dGM used as the covariate in second-level fMRI analysis, respectively) still revealed the cluster of activation in the frontopolar cortex that initially resulted from the contrast Children \> Adults. However, a substantial reduction of the cluster size was induced by the dGM correction ([Fig. 4](#fig0020){ref-type="fig"}, [Table 2](#tbl0010){ref-type="table"}). Hence, the functional difference between children and adults with respect to this cluster is not simply explained by structural maturation, although changes in local grey matter density with age contributes to the effect. Inversely, clusters found using the contrast Adults \> Children, whose activation spread over the primary somatosensory cortices and the inferior parietal lobules, did not remain significant when using either FA or dGM as a covariate in the analysis. Therefore, functional differences between the two groups about the parietal cortex emerge as a consequence of structural maturation.Fig. 4Activation cluster in the frontopolar cortex revealed by the contrast Children \> Adults (red) corrected for differences with age in FA (blue) and dGM (green). Voxel-wise threshold was set at p \< 0.001 uncorrected and cluster extend threshold at p \< 0.05 FWE-corrected.Fig. 4 3.3. Supplementary analyses {#sec0070} --------------------------- Since the children group included six individuals who reported illusions in the 30 Hz condition (i.e., a 'biased' control condition) and was therefore not as homogeneous as the adult group, we replicated the previous group-level analyses removing these children. Both the RL100 \> RL30 contrast in children (see supplementary Fig. 1 and Table 1) and group comparisons (see supplementary Fig. 2 and Table 2) revealed similar statistical parametric maps as those previously observed with the 22 children. Thus, we are confident in saying that second-level outcomes reflected differences between a no illusion condition and an illusion condition, therefore providing relevant information on the BRS (i.e. mapping sensory inputs into body representations). 4. Discussion {#sec0075} ============= Using kinaesthetic illusions in fMRI, the present study investigated the degree of maturity of the sensorimotor and fronto-parietal networks implementing the BRS by late childhood, with the assumption that larger differences in activation should be about regions of the latter network when comparing children to adults. A secondary objective was to investigate the causes of age-related changes in functional activation by examining the extent to which maturational changes in white and grey matter account for it. Overall, the results showed similar brain activation patterns in adults and children involving sensorimotor (SMA, basal ganglia) and fronto-parietal (IPL, IFG) regions, although the spatial extent of activation was larger in the children (i.e. more diffuse activation). Such a shift from more diffuse to more focal cortical activity appears to be a general principle of brain development ([@bib0040], [@bib0055], [@bib0090]). Outcomes are twofold with regard to group differences, involving (i) a more intense activation in the right and left inferior parietal lobules and primary somatosensory cortices in the adults compared to the children, and (ii) the additional engagement of a prefrontal region (frontopolar cortex, BA10) in children. Altogether, these findings suggest that while the basic configuration of the sensorimotor and fronto-parietal networks of the BRS is established by the age of 7--11, functional specialization of both networks -- and not only of the fronto-parietal network -- continue during adolescence. Importantly, such functional specialization is fully explained by structural grey and white matter maturation for the parietal/somatosensory regions but not for the prefrontal region. 4.1. Age-related changes in somatosensory and parietal activations {#sec0080} ------------------------------------------------------------------ The higher magnitude of activity in adults compared to children, as reported here both in the primary somatosensory and inferior parietal lobule regions, is commonly interpreted as immature neural mechanisms not processing as efficiently as they do in the mature system ([@bib0245]). Accordingly, this finding suggests that children were not able to access a mature and presumably optimal processing in these regions related to BRS and corporeal awareness. Indeed, there is ample evidence that the posterior parietal cortices, and especially the inferior parietal lobules, in the two hemispheres play a crucial role in monitoring/sustaining body representations (see [@bib0065] and [@bib0290] and references therein) and accessing consciously to these representations ([@bib0050], [@bib0080], [@bib0075]). Likewise, several works support the idea that body representations are hosted within the primary somatosensory cortices ([@bib0100], [@bib0030]). A study by [@bib0085] even reported cortical modifications in both the primary somatosensory cortex and in the posterior parietal cortex following surgical extension of the lower limbs. Although crucial to body representations, the primary somatosensory cortex and the posterior parietal cortex also likely play different roles, the latter region being more concerned with higher-order aspects related to the body such as awareness and updating body representations according to the environment ([@bib0025], [@bib0290]). The examination of the PSC provided further information about how to interpret the above developmental regional difference. The less marked increase in PSC from the RL30 condition to the RL100 condition in children compared to adults suggested a lower ability in the former to discriminate between sensory inputs, or equivalently to map them into accurate body representations. This interpretation is also consistent with behavioural findings that show a lower ability of the children compared to the adolescents and the adults to estimate limbs\' positions using proprioception ([@bib0155], [@bib0145]). However, it doesn't coincide with the children's behavioural outcomes. Indeed, children reported to experience illusory movement at 100 Hz, and few or none of them reported illusory movement at 30 Hz. This suggests a perceptual sensibility at a behavioural level which is not reflected by the parietal activity. Nevertheless, the focus on the posterior parietal areas activation assessment appears insufficient to completely explain the functional correlates eliciting illusory movement. Therefore, a more complete vision covering the functional connectivity within the BRS would be required in order to capture the generalization of illusory processes in its entirety ([@bib0170]) More importantly, the increase [with age]{.ul} in the magnitude of activation within these two regions disappeared when accounting for grey and white matter. Such a relationship between structure and function in the parietal cortex is not surprising given the significant changes that occur in its structural morphology during adolescence. There is evidence for a significant decrease in grey matter density and thickness in the parietal cortex starting at about 10 (e.g. [@bib0160], [@bib0350]), which support the development of mature activation patterns. For instance, several studies have shown that increased activation over development is accounted for by cortical thinning ([@bib0235], [@bib0390]). White matter tracts also become more structured and myelinated with development to support higher efficiency of signal propagation, and thus improve information transfer between distant regions ([@bib0295]). In particular, certain association fibres that pass through the parietal lobe, such as those of the superior longitudinal fasciculus that plays a role in corporeal awareness ([@bib0005]), are still immature during adolescence ([@bib0015], [@bib0215]). In sum, our study suggests a substantial influence of the brain structure to the development of the BRS network. 4.2. Disengagement of the anterior prefrontal cortex with development {#sec0085} --------------------------------------------------------------------- Another important finding was the disengagement of the frontopolar cortex from the BRS from late childhood to adulthood. The frontopolar cortex is an anterior prefrontal region that forms the apex of the executive function system in humans (e.g. [@bib0260], [@bib0205]). This region has been reported to contribute to several high-level functions including motor intention, motor imagery, and decision making ([@bib0175], [@bib0330], [@bib0240]), and hence a supervisory function. Along this line, it has been shown that the prefrontal cortex modulates the magnitude of neural activity in distant brain regions -- a mechanism of control known as top-down modulation -- to establish high fidelity representations of task-relevant stimuli and to facilitate their internal maintenance ([@bib0140], [@bib0135]). Accordingly, the frontopolar cortex may serve in children as the 'top' in top-down modulation of the activity of the other BRS regions to facilitate the establishment of body representations from sensory (here, proprioceptive) inputs. This would signal a transition over the course of development from a rather top-down control scheme to a more bottom-up (i.e. automatic) control scheme for the generation of body representations from sensory inputs. Although the most common finding is an increase in activation of prefrontal regions with the development of cognitive control ([@bib0245]), our finding fits with previous evidence that some prefrontal regions show an age-related decrease in activation likely reflecting a decreased 'effort' required to exert cognitive control ([@bib0375]). Moreover, the frontopolar cortex is robustly involved in learning new behavioural cognitive sequences and progressively disengages over the course of learning ([@bib0210]). Specifically, this region disengages once an expectation or equivalently an internal representation has become reliable for a given sequence. Likewise, activation of the frontopolar cortex in children may relate with uncertain body representations given the proprioceptive inputs, and progressively disengages until adulthood as the amount of uncertainty related to mapping sensory inputs into body representation decreases. Again, this supports the idea that the activation of the anterior prefrontal cortex in children is an 'add-on' to the BRS system, perhaps to assist the other regions in building the more plausible representation given the ongoing sensory inputs. Finally, the disengagement of the frontopolar cortex also marks a BRS system that becomes more segregated (or differentiated) with development, which is a well-established principle of brain development (e.g. [@bib0055], [@bib0105], [@bib0310]). Phenomena such as myelination, synaptic pruning and experience-dependent regional specialization are candidates to explain the increased segregation of cortical systems. Here, using white and grey matter as covariates still revealed frontopolar activation in the children. Therefore, disengagement of this region may be more related to experience-dependent changes in the activation of the BRS system. However, disambiguating structural from functional factors is difficult in that evoked activity plays a key role in determining which synaptic connections persist and which are eliminated during development ([@bib0045], [@bib0185], [@bib0320], [@bib0325]). Accordingly, stating that the age-related disengagement of prefrontal activation is completely unrelated to structural factors is likely too simplistic. This is also supported by the fact that grey matter correction affected the prefrontal activation. 4.3. Methodological considerations {#sec0090} ---------------------------------- Although the study makes a unique contribution to the understanding of the maturation of the BRS and of the structural factors that influence it, methodological considerations have to be notified. First of all, six children experienced unanticipated illusory movements during the 30 Hz stimulation condition, indicating that the control condition included to some extent a body representation component and therefore may not have been optimal to investigate the BRS. However, we replicated group-level results when removing those subjects with illusory movements at 30 Hz. Accordingly, we are confident in saying that 100 Hz tendon stimulation vs. 30 Hz tendon stimulation was an adequate contrast to tap into the BRS, although the frequency difference between the two conditions also embedded a proprioceptive component. Another possibility would have been to contrast the 100 Hz tendon stimulation with a 100 Hz bone stimulation as done in previous studies ([@bib0150], [@bib0285]). However, the resulting activations from this latter contrast would have been much more related to proprioceptive processing, making any conclusion of the BRS intricate. Closely related to the previous issue was the absence of quantitative data about adult's illusory percept in our study. As a consequence, we could not have evaluated between-group difference in the extent to which amount of activation correlated with the degree of the illusion, which would have brought a finer-grained analysis of the cerebral correlates of body representations and the way these correlates develop. In particular, such an analysis would have been important to completely disentangle activations more related to the illusory percept from those more related to proprioceptive processing per se. Thus, it is of outmost importance that future investigations on the development of the BRS include careful psychophysical assessment of the illusion and link it to activation data. Finally, the fact that we did not record electromyographic (EMG) activity during vibration may be considered a further limitation of our study. Indeed, vibration may have induced involuntary muscular contractions whose re-afferences might have blurred group results. However, previous studies reported either very limited EMG activity in a few subjects (e.g. [@bib0315]) or no EMG activity ([@bib0275], [@bib0005]) during the vibratory stimulation. Therefore, this limitation is not likely to question the validity of the study outcomes. 5. Conclusion {#sec0095} ============= Although the network subtending the BRS is already well established as early as 7--8, our study also demonstrated immaturity including a decreased activation in both sensorimotor (primary somatosensory cortex) and parietal (inferior parietal lobule) regions, as well as an exclusive anterior prefrontal (frontopolar) activation. The former differences were found to be entirely related to white and grey matter properties while the latter difference did not. This lends credence to the idea that maturation of the BRS network is complex and presents an extended development, relying on structural factors and a functional process that results in the disengagement of an executive region not classically involved in body processing. This expressed a functional neurodevelopmental strategy. Future studies will have to examine how the multimodal (i.e. modulation of proprioceptive-based representations by other sensory channels) and modular (i.e. interactions between brain regions) features of the BRS develop. Conflict of interest {#sec0100} ==================== The authors declare no conflict of interest in relation to the present manuscript. Funding {#sec0105} ======= This work was supported with funds from 'Fondation Yves Cotrel - Institut de France' and 'Fondation de France'. Appendix A. Supplementary data {#sec0115} ============================== The following are Supplementary data to this article: The authors are grateful to the children, their parents, and the adults for their precious collaboration. The authors are also grateful to Doris Mathisen who revised the English. Supplementary data associated with this article can be found, in the online version, at [http://dx.doi.org/10.1016/j.dcn.2017.02.010](10.1016/j.dcn.2017.02.010){#intr0005}. [^1]: The significant clusters reported in fMRI did not survive to correction for group differences in grey and white matter.

Introduction ============ The molecular mechanisms of seed development are extensively studied, and many genes involved have been already identified. *SHB1*,[@R1] *IKU1*,[@R2] *MINI3*,[@R3] *IKU2*,[@R3] AP2,[@R4]^-^[@R6] and MADS-box transcription factors *AGL61*,[@R7]^,^[@R8] AGL62,[@R9] and *AGL80*[@R10] mainly regulate endosperm development to affect seed size/weight. *ARF2*,[@R11] TTG2,[@R12] and *TT16*[@R13] regulate integument development to affect seed development. *FIS2*, *FIE*, *MEA*, *MSI1*, SWN,[@R14] and *MET1*[@R15] influence seed development by epigenetic regulation of endosperm development and paternal imprinting. *ANT*[@R16] and *RGE1*[@R17] are involved in regulating embryo proliferation and affect seed development. Hormones play crucial roles in seed development,[@R18] including auxins, cytokinins, gibberellins, and BR. Also many genes that take part in ovule development and regulate the seed number have been identified in *Arabidopsis*, such as *SUP*,[@R19] *INO*,[@R20] *SIN1*,[@R21] *SIN2*,[@R22] *TSO1*,[@R23] *BEL1*,[@R21]^,^[@R24] *AG*,[@R25] *HLL*,[@R26] *ANT*,[@R26] and *AP2*.[@R24] Steroid hormone BR is required in seed development. Deficiency in BR synthesis and signal transduction pathway leads to severe phenotypes in reproductive development. Many excellent articles and reviews covered BR synthesis, signal transduction, and related plant growth and development. Here we briefly summarize our current understanding of BR functions and regulation mechanisms in *Arabidopsis* seed development. BR Plays Key Roles in Seed Size/Mass Regulation =============================================== The functions of BR in seed development have been demonstrated by studies of BR deficient and insensitive mutants of *Arabidopsis*, *Oryza sativa*, *Pisum sativum*, and *Vicia faba*. *Arabidopsis* BR deficient mutant *dwf5* produces small seeds.[@R27] Another *Arabidopsis* dwarf mutant *shk1-D* (overexpression of a P450 monooxygenase family gene *CYP72C1*[@R28]) has lower endogenous BR level and produces general short organs and small seeds. Rice BR deficient mutant *brd2* exhibits shortened and smaller grains.[@R29] A cytochrome P450 gene *DWF11* of rice is implicated in BR biosynthesis, and the *dwf11* mutant also bears reduced seeds.[@R30] Rice dwarf mutant *d61* produces small grain size, which is caused by loss of function of a rice *AtBRI1* ortholog, *OsBRI1*.[@R31] Contrarily, overexpression of BR-biosynthetic gene increases rice seed filling and yield.[@R32] BR-deficient *Vicia faba* mutant also produces small seeds,[@R33] and dwarf mutant *lk*, a severe BR deficient mutant of pea, has smaller seeds.[@R34] These facts indicate that BR is required for normal seed development and seed size/mass determination. So far, the molecular mechanisms of BR involved in seed development remain unclear. Our recent studies introduce advanced understanding of BR regulation in seed size/mass.[@R35] First, we found the mature dry seeds of the BR-deficient mutant *det2* and the BR insensitive mutant *bri1--5* (a weak allele of *bri1* mutant) were smaller than corresponding wild type seeds. Furthermore, the exogenous BR partially rescued seed size and weight of *det2*, confirming BR positive regulation of seed size/mass. Detailed studies showed that the reduced seed size of *det2* was possibly caused by reduced seed cavity and endosperm volume, delayed embryo development, decreased embryo cell size, number, and integument cell length. Our results are in according with the theory that final seed size depended on interactions among zygotic embryo and endosperm, maternally derived seed coat, and mother plant.[@R11] We further analyzed the effects of BR synthesis and signaling in specific tissues on seed size determination, and deduced that BR produced by the embryo and endosperm contributed to the increased seed size. *SHB1*, *IKU1*, *MINI3*, and *IKU2* are reported to be positive regulators of seed size and act in the same pathway.[@R1]^-^[@R3]^,^[@R36] We demonstrated they were positively regulated by BR and *SHB1*. *IKU1* and *IKU2* were found as direct targets of BZR1,[@R35] confirming ChIP results of Sun et al.[@R37] Unlike MINI3, its interacting protein IKU1[@R2] is a direct target of BZR1. Genetic experiments showed these genes acted downstream of *DET2* and *BZR1* in seed size regulation. We proposed the model that BR regulated embryo and endosperm development to determine seed size through binding to *SHB1* promoter region by BZR1, consequently regulated the expression of *MINI3* and *IKU2*,[@R1] or through BZR1 binding to *IKU1*, which interacted with MINI3[@R2] to regulate the activity of *MINI3* and the expression of *IKU2*, or alternatively through BZR1 binding to *IKU2* and regulated the transcription of *IKU2* directly ([Fig. 1](#F1){ref-type="fig"}). *FIS2*, *ANT*, and *RGE1* were detected to be direct targets of BZR1 and positively regulated by BR. We proposed that BR mediated the development of embryo and endosperm through regulating the expression of these BZR1 target genes to affect seed development ([Fig. 1](#F1){ref-type="fig"}). {#F1} Our studies illustrated that BR was implicated in the regulation of integument development as well. The first evidence observed was the significant decrease of integument cell length in *det2*, indicating that BR positively regulated integument cell elongation. The second evidence was that *ARF2* gene mutation (*arf2*) resulted in dramatically increased seed size/mass due to the extra cell division in the integuments.[@R11] *ARF2*, a direct target gene of BZR1, was negatively regulated by BR. Genetic evidences demonstrated *ARF2* acted downstream of *BZR1* and BR signal in regulating seed size. These results rose up a hypothesis that BR might regulate the seed size/mass also through BZR1 binding and repressing *ARF2* to positively regulate the integument development ([Fig. 1](#F1){ref-type="fig"}). Because cytochrome P450 CYP78A5 (KLU) was reported to act as a maternal regulator of seed size by stimulating the integument cell proliferation,[@R38] and we demonstrated KLU was a direct target gene of BZR1, we gained another evidence supporting the hypothesis that BR regulated the integument development to influence seed development through BZR1 ([Fig. 1](#F1){ref-type="fig"}). Besides, *AP2* acts as an important negative regulator of seed development by inhibiting integument growth (and embryo/endosperm development) to reduce the seed size.[@R4] *AP2* also affects seed size/mass by coordinating the endosperm/embryo growth.[@R4]^-^[@R6] Our results showed *AP2* was a target of BZR1, too, and negatively regulated by BR, and *AP2* acted downstream of BZR1 and BR signal in regulating seed size ([Fig. 1](#F1){ref-type="fig"}). Above all, we demonstrated that BR regulated seed size by expanding seed cavity and endosperm volume, promoting embryo development, and increasing embryo cell size, number, and integument cell length through transcriptionally modulating specific seed developmental pathways ([Fig. 1](#F1){ref-type="fig"}). BR Functions in Seed Shape Determination ======================================== Shape is an important seed trait. To date, there are a few reports mentioning seed shape regulation. *Arabidopsis* seed shape mutant *ats* is isolated from an ethyl methanesulfonate-treated population. Its length/width ratio is significantly lower than wild type. *ATS* functions in integument development and directly affects seed shape.[@R39] A series of rice QTLs for grain weight are identified. Among them, *GW3*, *GW6*, and *GS3* correlate with grain shape.[@R40]^,^[@R41] MADS genes also contribute to seed shape. *Osmads6--1* seeds seem to be more roundish than elliptical.[@R42] However, the underlying mechanisms of seed shape determination are mostly unknown. BR is involved in seed shape determination. *dwf5* and *shk1-D* produce aberrantly shaped *Arabidopsis* seeds.[@R27]^,^[@R28] *dwarf1* and *dwarf11* mutants bear small, round rice seeds that are shorter length but equal width to wild type.[@R30]^,^[@R43] Rice *brd2* mutant also exhibits shortened grains.[@R29] *lk* mutant produces irregularly shaped pea seeds.[@R34] Although BR has major effect on seed shape and BR effect cannot be recovered by other signals, the detailed mechanism of BR regulating seed shape is still unclear. Our recent studies made progress in explaining the mechanism of seed shape regulation by BR. We found that the length to width ratio of seeds produced by BR-deficient and insensitive mutants was notably lower than of wild type. The ratio of wild type *Arabidopsis* seeds resembles the Golden Ratio of 1.61803399.[@R44] Owing to the fact seeds come from fertilized ovules, we further analyzed ovule shape before pollination and found ovule shape of BR-deficient and insensitive mutants was normal. Thus we concluded BR regulated seed shape after fertilization. Furthermore, our results suggested that BR synthesis and signaling in the maternal tissues integument determined the seed shape ([Fig. 1](#F1){ref-type="fig"}). The reciprocal crosses demonstrated that seeds shape was kept roundish when maternal tissue contained less BR or abnormal BR signal, even if BR level and signal have been rescued in embryo/endosperm and seed size/mass already restored. These results suggested that BR level in zygote determined the seed size/mass and had little effect on seed shape, which was consistent with the opinion that BR could not be transported through tissue boundary.[@R45]^,^[@R46] Although *SHB1*-*MINI3*-*IKU2* pathway, *AP2*, and *ARF2* mediate the BR regulation of seed size/mass, the BR functions in seed shape determination is independent on these genes. Herein, the evidences we got suggested that mechanism of BR regulating seed shape was different from that for seed size. Furthermore, our studies indicated BR regulated distinct pathways in specific cell types to control seed size and shape. However, the detailed mechanism and unknown concerned genes of BR regulating seed shape (maybe a good model for studying BR transport), are worth further investigation in the future ([Fig. 1](#F1){ref-type="fig"}). BR is Involved in Seed Number Determination =========================================== There are several reports about seed number determination in flowering plants; however there is a significant lack of deeper insight into the mechanism of this regulation. First study describes the ectopic overexpression of a BR-biosynthetic gene *DWF4* in *Arabidopsis* transgenic plants results in increased seed yield due to a greater total seed number.[@R47] Constitutive expression of *ZmDWF4* also increases the seed numbers.[@R48] Enhancement of BR biosynthesis in transgenic rice enhances the grain yield due to more tillers and seeds and higher seed weight.[@R32] These studies suggest that BR causes an increase of branches and sillique number that leads to increased seed yield. Due to the fact that many factors contribute to total seed number, we first focused on seed number per sillique in *Arabidopsis*. The ovule number determines the maximal number of seeds in *Arabidopsis*. Many genes have been identified in each phase of ovule development. *INO* and *SUP* promote and suppress integument, and also affect the asymmetric form of integuments.[@R19]^,^[@R20] *SIN1* and *SIN2* are required for cell division during integument development.[@R21]^,^[@R22] *TSO1*, which encodes a nuclear protein, is required for the orientation of cell elongation and cytokinesis in integument development.[@R23]^,^[@R49] *BEL1* and *AG* regulate integument identity.[@R21]^,^[@R24]^,^[@R25] *HLL* and *ANT* both promote ovule primordial growth.[@R26] Interestingly, *AP2* also affects early ovule formation.[@R24] Our recent studies showed the ovule or seed number per sillique of *det2* and *bri1--5* were significantly decreased, whereas *bzr1--1D* contained obvious increased ovule and seed number.[@R50] Thus we demonstrated that BR positively regulated ovule and seed number determination, and BR functioned in ovule number determination through BZR1 regulating *HLL*, *ANT*, and *AP2* transcription directly or indirectly ([Fig. 1](#F1){ref-type="fig"}).[@R50] *AP2* did not act downstream of BZR1 in ovule number determination, but functioned antagonistically with BZR1, which seemed to be different to seed size/mass regulation. The research of *Arabidopsis* seed number determination would provide opportunities for further use in yield enhancement of crops like *Brassica napus* and *Glycine max*. Apart of ovule number, *Arabidopsis* seed number per sillique determination is also affected by paternal tissues, fertilization, seed development, and other factors. These processes and their regulation will be worth further investigation. Total amount of seeds depends on seed number per sillique and sillique number, which reflects the meristem differentiation and maintenance from early stages of reproductive development. Although rice seed number determination differs from *Arabidopsis*, the signals in reproductive meristems required for seed number determination process in both monocot and eudicot may be conserved. How BR functions in reproductive meristem and affects flower, fruit, and seed number is an attractive field in future research. Conclusions and Perspectives ============================ The economic potentiality of BR in agricultural production was recognized as early as the 1980s, and the ability of exogenous BR to increase yield was shown in a variety of plant species. Combining previous reports and our results, we conclude that BR has crucial function in seed development, including seed size/mass, shape, and number determination. The regulation mechanisms of BR in these processes are independent and different from each other. We thank Dr. Frantisek Baluska for kindly inviting this review, and Ms. Ivana Ferjentsikova (University of Nottingham) for language editing. The concerned research work was supported by the National Basic Research Program of China (Grant 2014CB943404 and 2009CB941503). BR : brassinosteroid *SHB1* : *SHORT HYPOCOTYL UNDER BLUE1* *IKU1* : *HAIKU1* *MINI3* : *MINISEED3* *IKU2* : *HAIKU2* *AP2* : *APETALA2* *AGL* : *AGAMOUS-LIKE* *ARF2* : *AUXIN RESPONSE FACTOR2* *TTG2* : *TRANSPARENT TESTA GLABRA2* *TT16* : *TRANSPARENT TESTA 16* *FIS2* : *FERTILIZATION INDEPENDENT SEED 2* *FIE* : *FERTILIZATION INDEPENDENT ENDOSPERM* *MEA* : *MEDEA* *MSI1* : *MULTICOPY SUPRESSOR OF IRA* *SWN* : *SWINGER* *MET1* : *METHYL TRANSFERASE* *ANT* : *AINTEGUMENTA* *RGE1* : *RETARDED GROWTH OF EMBRYO1* *SUP* : *SUPERMAN* *INO* : *INNER NO OUTER* *SIN* : *SHORT INTEGUMENTS* *BEL1* : *BELL* *AG* : *AGAMOUS* *HLL* : *HUELLENLOS* *AP2* : *APETALA2* *dwf* : *dwarf* *shk1-D* : *shrink1-dominant* *brd2* : *brassinosteroid-deficient dwarf2* *BRI1* : *BRASSINOSTEROID INSENSITIVE1* *AHK* : *ARABIDOPSIS HISTIDINE KINAS* *CKI1* : *CYTOKININ INDEPENDENT 1* *ats* : *aberrant testa shape* QTLs : quantitative trait loci No potential conflicts of interest were disclosed.

Introduction {#s1} ============ Prostate cancer (PCa) is the most prevalent solid tumor in men from industrialized nations and is the second largest cancer-related killer (Center et al., [@B14]; Siegel et al., [@B58]). Age is a significant risk factor for the disease. The incidence in USA jumps from 1 in 7,964 to 1 in 8 when comparing men under 40 years of age with men older than 70, respectively (Siegel et al., [@B58]). Most of the patients with advanced PCa are initially sensitive to traditional treatments of androgen ablation therapy. This is the mainstay of treatment, and leads to the regression of PCa tumors (Suzuki et al., [@B61]; Feldman and Feldman, [@B23]). However, with the progression of the disease, PCa often develops resistance to therapy and patients may eventually die of this metastatic castration-resistant prostate cancer (mCRPC). As many as 50% of PCas will progress from an androgen-dependent (AD) to a hormone refractory state of disease, and will metastasize to bone and pelvic lymph nodes (Thalmann et al., [@B62]). There are still no curable treatments for CRPC (Karantanos et al., [@B34]). Drugs such as abiraterone, enzalutamide and TOK-001, bone-targeted therapies (such as bisphosphonates, denosumab, and Radium-223), and immunotherapies all have questionable efficacy (Chaturvedi and Garcia, [@B17]). Hence, novel treatment strategies are a necessity to improve the quality and span of life for PCa patients. Understanding the underlying mechanisms behind progression of PCa to CRPC and its metastasis is crucial in defining new therapeutic targets and prophylactic therapies for this type of cancer. While the role of non-steroidal anti-inflammatory drugs (NSAIDs) in preventing colorectal cancer has been well-established (Muscat et al., [@B48]), numerous epidemiological studies have shown that they are protective against other cancers as well (Baron and Sandler, [@B7]; Cha and DuBois, [@B15]). One study reported nearly a 63% drop in the relative risk for colon cancer, 39% for breast cancer, 36% for lung cancer, and 39% for PCa with the increasing intake of NSAIDs (mainly aspirin or ibuprofen; Harris et al., [@B32]). One meta-analysis looked specifically at the association between aspirin and PCa, combining the results of 39 studies (20 case-control and 19 cohort studies; Liu et al., [@B43]). It was found that aspirin use was significantly associated with lower PCa incidence (OR = 0.92, 95% CI = 0.87--0.97) and lower PCa-specific mortality (HR = 0.86, 95% CI = 0.78--0.96). While evidence suggests a protective effect of aspirin, the processes underlying this remain unclear. The aim of this paper is to review some of the proposed mechanisms relating aspirin to the pathobiology of PCa neoplasms, with a main focus on the basic science research done in this context. Decreased PSA levels in chronic aspirin users among prostate cancer patients {#s2} ============================================================================ The majority of PCa patients first learn they might have the disease through a blood test that looks for increased or rising levels of PSA protein (Hamilton et al., [@B31]), produced by luminal cells in the prostate (Feldman and Feldman, [@B23]). In these settings, PSA can indicate the presence or recurrence of PCa. The concern is that drugs that artificially lower PSA levels might mask this marker, which normally flags the development of a prostate neoplasm. One study analyzed PSA levels in a cohort of over 1,000 men and found that PSA levels in the sample of men taking aspirin were nearly 10% lower than a control sample not taking aspirin (Hamilton et al., [@B31]). Researchers question whether this decrease in PSA is artificial or whether it might be a direct result of anti-tumorigenic properties of aspirin. Evidence points in both directions. Multiple epidemiological studies have reported an increase in high-grade PCa (HGPCa) in aspirin users (Bosetti et al., [@B10]; Olivan et al., [@B49]). One suggested explanation found in the literature is that aspirin and other NSAIDs lower PSA levels, but is not itself anti-tumorigenic (Berg et al., [@B9]; Schroder et al., [@B55]). If true, this would likely delay diagnosis of the disease, allowing it to progress unnoticed. Alternatively, it was also suggested that these drugs might relieve cancer-associated pain that normally motivates patients to investigate their symptoms earlier on. In contrast, it must also be noted that other studies show decreased levels of HGPCa in chronic aspirin users (Brasky et al., [@B11]; Dhillon et al., [@B21]). More conclusive research needs to be conducted to determine the mechanism by which aspirin decreases PSA: whether this is a by-product of the drug\'s anti-tumorigenic properties or truly a masking effect. Nonetheless, physicians should be weary of the accuracy of the PSA blood test in patients taking aspirin chronically. Aspirin and the COX pathway in prostate cancer {#s3} ============================================== NSAIDs inhibit the metabolism of arachidonic acid by blocking the cyclooxygenases (COXs) pathway and the prostaglandins (PGs) synthase pathway, thus suppressing PG synthesis and inflammation (Majima et al., [@B46]; Figure [1](#F1){ref-type="fig"}). COXs are key enzymes in prostanoid synthesis, existing in two isoforms: COX1 and COX2. COX1, referred to as "constitutive isoform," is expressed in several tissues under basal conditions. COX2 is believed to be undetectable in normal human tissues, but can be induced by mitogens, cytokines, and tumor promoters under various, mainly pathological condition. It is thus referred to as "inducible isoform" (Katori and Majima, [@B37]; Gupta and Dubois, [@B28]; Subbaramaiah and Dannenberg, [@B60]). COX2 activation hence promotes enhanced PGs synthesis in both inflamed and neoplastic tissues (Bennett, [@B8]; Rigas et al., [@B52]). {#F1} Aspirin is unique in that it irreversibly blocks both COX1 and COX2 activities through acetylation of significant enzyme serine residues. Ergo, new COX activity can only be achieved following aspirin treatment through *de novo* synthesis of COX. The main mechanism by which NSAIDs are thought to prevent the growth of neoplasms is the blocking of COX2 activity (Thun et al., [@B64]), though studies have shown that NSAIDs like aspirin have anticancer effects through both COX-dependent and independent cascades (Grosch et al., [@B27]; Alfonso et al., [@B2]). Several studies have demonstrated higher expression of COX2 in PCa tumor tissues than in benign prostate tissues (Gupta et al., [@B29]). It has been shown that both LNCaP and PC3 PCa cell lines express COX2. High COX2 expression in PCa cells has also been associated with poor prognosis (Khor et al., [@B38]). This has further corroborated the suggestion that NSAIDs could play a role in reducing PCa risk specifically through inhibiting the COX pathway. *In vivo*, anti-inflammatory doses of aspirin (a daily dose of \>2,000 mg) do bring systemic concentrations high enough to inhibit both COX1 and COX2. However, in nucleated cells, due to *de novo* synthesis, inhibition can only be prolonged with repeated daily dosing (Thun et al., [@B65]). It has been suggested in that same paper that aspirin in lower doses might still effectively inhibit COX2 due to partial dependence of COX2 expression in monocytes on activated platelets. Consequently, aspirin permanently inactivates COX in platelets, thus indirectly inhibits COX2 expression (Thun et al., [@B65]). The blockage of COX prevents the production of downstream PG products, known as prostanoids, such as TXA~2~, PGI~2~, PGE~2~, PGF~2α~, and PGD~2~. These prostanoids have roles in decreasing apoptosis and increasing cellular proliferation (Thun et al., [@B65]). One PCa-specific study reported that aspirin-treated LNCaP and PC3 PCa cells had the same proportion of dead cells as non-treated cells, signifying that aspirin might not induce apoptosis but instead suppresses proliferation (Olivan et al., [@B49]). The literature is not conclusive on this, however. In addition, this paper reported decreased colony formation and significant inhibition of invasion and migration capacities in aspirin-treated cells (PC3 cells in particular) with higher effects when aspirin is combined with simvastatin, a cholesterol-lowering drug (Olivan et al., [@B49]). Among the five PGs that have been identified in the COX pathway, PGE~2~ is the most common and ubiquitously produced PG, contributing to tumorigenesis via cell proliferation induction (Tjandrawinata et al., [@B66]), angiogenesis (Wang and Klein, [@B72]; Jain et al., [@B33]), invasion (Sheng et al., [@B57]; Buchanan et al., [@B13]), and metastasis (Konturek et al., [@B39]; Fulton et al., [@B25]). PGE~2~ levels are 10-fold higher in human malignant PCa tissues than in benign prostatic tissues (Chaudry et al., [@B18]). PGE~2~ works through EP~1~, EP~2~, EP~3~, and EP~4~, four G-protein coupled receptors (Kashiwagi et al., [@B36]). Human prostate epithelial cells express EP~2~ and EP~4~ receptors, while EP~1~ and EP~3~ receptor expression in these cells is not detected (Wang and Klein, [@B72]). EP~3~ is distinct from EP~2~ and EP~4~ in that it is not a stimulatory but instead an inhibitory G-protein. Thus, EP~3~ decreases levels of the secondary messenger cAMP when activated. A study by Kashiwagi et al. reported that aspirin decreases Androgen Receptor (AR) mRNA and protein levels in dose-and time-dependent manners (Kashiwagi et al., [@B36]), which is thought to be related to the proliferation of PCa. Interestingly, the same study reported upregulation of EP~3~ expression and a consequent downregulation of AR and EP~2~ expression in PCa cell lines upon aspirin treatment. This domino effect was confirmed using both pharmacological and knockdown methods. The results are supported by another study that found that EP~3~ signaling inhibits the NF-κB pathway (Wang et al., [@B71]), which decreases AR expression levels in PCa cells (Zhang et al., [@B76]). This was not the first paper to claim this connection to the NF-kB pathway. Lloyd et al. previously showed that aspirin inhibits NF-κB, resulting in diminished urokinase-type plasminogen activator (uPA) secretion---one of the crucial molecules involved in cancer metastasis---from the highly invasive human PC3 PCa cells (Lloyd et al., [@B44]). The inhibition of COX in platelets might also be significant, since experimental evidence has shown that platelets are significant in cancer metastasis through the blood (Labelle et al., [@B40]; Dudeja et al., [@B22]). This effect is mediated through their ability to aggregate and allow cancer cells to escape immune detection as well as the pro-angiogenic factors, such as VEGF, that they release (Usman et al., [@B68]). Thus, the EP~3~ receptor might represent a potential molecular target for developing therapy in PCa. Aspirin and COX-independent regulation of the cell cycle in prostate cancer {#s4} =========================================================================== Aspirin might influence regulation of the cell cycle, which is dependent on a family of proteins called cyclins and another group of protein kinases called cyclin-dependent kinases (CDKs). When combined with statins in treatment, aspirin was shown to decrease proliferation of LNCaP cells with a reduction in cyclin D1 levels---which modulates cell cycle progression (Olivan et al., [@B49]). Aspirin, on its own, was shown to cause ubiquitin-dependent degradation of cyclin D1 in colorectral cancer cells (Thoms et al., [@B63]). Further research is needed to deduce whether aspirin can instigate the same mechanism in PCa cells in the absence of statins, since on their own, statins were also shown to be associated with low expression of cyclin D1 in a breast cancer trial (Feldt et al., [@B24]). Another proposed mechanism of control---as suggested by epidemiological studies in colorectal cancer patients (Seiler, [@B56]; Laukaitis and Gerner, [@B41])---is the induction of polyamine catabolism and subsequent regulation of cell proliferation and cancer progression (Arisan et al., [@B5]). Polyamines are small cationic molecules, formed from the decarboxylation products of ornithine and S-adenosyl-methionine. They are present in high concentrations in rapidly dividing tumor cells (Agostinelli et al., [@B1]). Although intracellular levels of polyamines are elevated in normal prostate gland (Karr et al., [@B35]), abnormal regulation of their metabolism results in rapid cell proliferation and PCa progression (Arisan et al., [@B5]). In fact, when PCa cells were treated with CDK inhibitors purvalanol and roscovitine, which induce apoptosis by promoting cell cycle arrest in cancer cells, upregulation of polyamine catabolic enzymes (SSAT, SMO, and PAO) was induced. This caused the depletion of intracellular polyamine levels (Arisan et al., [@B5]). In the same study, silencing of SSAT prevented CDK inhibitors-induced apoptotic cell death in PCa cells (Arisan et al., [@B5]). Accordingly, aspirin has been recognized as an inducer of SSAT by allowing NF-κB binding on the *Sat1* gene (Babbar et al., [@B6]). However, another study showed that treating LNCaP PCa cells with aspirin decreased induced SSAT activity in these cells (Li et al., [@B42]). Authors of this study concluded that SSAT and its related polyamine metabolism may play a significant role in the susceptibility of PCa to aspirin therapy (Li et al., [@B42]). The potential relevance of these mechanisms needs to be further explored, especially using *in vivo* trials and feasible, non-toxic doses of aspirin. One study found that aspirin promotes "tumor necrosis factor-related apoptosis inducing ligand" (TRAIL)-induced apoptosis in both androgen-dependent LNCaP cells and other LNCaP derived cells (C4, C4-2, and C4-2B), which represent CRPC, through decreased survivin protein---a versatile modulator of cell division and apoptosis in cancer (Altieri, [@B3])---expression in these cells (Yoo and Lee, [@B75]). And finally, researchers have also evaluated the effect of new nitric oxide (NO) donating NSAIDs, including NO-aspirin and NO-ibuprofen, on LNCaP and PC3 PCa cell lines. They found these drugs to be potent inhibitors of proliferation and inducers of apoptosis via enhanced caspase-3 expression (Royle et al., [@B54]). One reason for the importance of those novel NSAIDs over classical ones lies in the presence of NO, which when endogenous, contributes to the action of immune cells against foreign pathogens and tumor cells. NO was additionally suggested to play a role in the modulation of cell death by apoptosis, though this effect depends on a multitude of factors, including the concentration of NO and the cell type (Wallace and Soldato, [@B70]). Interestingly, NO-aspirin had been shown to be much more potent, even at lower concentrations, at inducing apoptosis and inhibiting proliferation in those PCa cells than conventional aspirin (Royle et al., [@B54]). In accordance, NO-aspirin inhibited proliferation of PC3 and DU145 PCa cells through blocking Wnt/β-catenin signaling in those cells (Lu et al., [@B45]). Other studies have implicated the lipoxygenase (LO) pathway of arachidonic acid metabolism in the progression of PCa. Yang et al. showed that LO products, including 12-HETE, were significantly higher in malignant prostate tissue than non-malignant tissue. The role of an NSAID like aspirin in this process is unclear (Yang et al., [@B74]). While one might intuitively warn that the inhibition of COX redirects arachidonic acid to the LO pathway, the evidence in the literature is not conclusive. Gray et al. looked at COX and LO activity in whole blood, noting that the blocking of COX was not actually associated with an increase in LO products. Furthermore, they reported that NO-aspirin even reduced LO activity, a notion supported by other publications as well (Gray et al., [@B26]). Brunn et al. for instance, reported that endogenously released NO inhibits the production of 5-LO metabolites in macrophages (Brunn et al., [@B12]). However, these precise mechanisms of NO-aspirin still remain the subject of investigation. Thus, with the failure of androgen ablation therapies and emergence of hormone-refractory states in PCa, enhancing tumor cell death via facilitating apoptosis of cancer cells using aspirin may be an effective promising chemopreventive therapy for the disease in the future. Aspirin and COX-independent regulation of metastasis of prostate cancer cells {#s5} ============================================================================= Aspirin treatment has been associated with decreased migration of PCa cell lines and increased levels of α2 integrin (Olivan et al., [@B49]), which may be a metastasis suppressor as suggested by Ramirez et al. ([@B51]). These results are controversial, however, as other studies reported conflicting data. Other literature found that the expression of integrin α2β1 actually induces PCa metastasis to the bone (Hall et al., [@B30]; Van Slambrouck et al., [@B69]; Sottnik et al., [@B59]). These studies suggest that the expression of this protein is in fact correlated with the different stages of cancer progression (Hall et al., [@B30]; Van Slambrouck et al., [@B69]; Mitchell et al., [@B47]). This might be the cause of reported higher levels of HGPCa in patients treated chronically with aspirin (Olivan et al., [@B49]); however, further research is needed to clarify the role of integrins in PCa tumors and whether they can be a molecular target for therapy. It has been demonstrated that cell migration, and the process of cancer metastasis, is regulated or influenced by different molecular mechanisms. Another mechanism explored specifically in regards to PCa concerns p75^NTR^, a member of the tumor necrosis factor (TNF) receptor superfamily and tumor suppressor highly expressed in normal prostate epithelial cells (Chao, [@B16]). This high expression diminishes as the tumor progresses (Pflug et al., [@B50]). Reports have shown that NSAIDs like aspirin induce p75^NTR^ expression through the p38 mitogen-activated protein kinase (MAPK) pathway (Wynne and Djakiew, [@B73]). Correlating with the induction of p75^NTR^ by NSAIDs is the induction of Nag-1, a member of the TGF-B superfamily that inhibits cell migration, possibly through blocking the activity of uPA, and matrix metalloproteinases MMP2 and MMP9 (Wynne and Djakiew, [@B73]). Thus, Wynne and Djakiew et al. proposed that NSAID suppression of cell migration might be mediated by Nag-1 induction, downstream of p75^NTR^. The effect of aspirin on chemotherapy in prostate cancer {#s6} ======================================================== Aspirin may promote resistance to treatment in PCa in three different mechanisms. First, many chemotherapeutic agents work by targeting rapidly-dividing cells, thus decreased cellular proliferation---as was shown to be an effect of aspirin on PCa cells---might decrease the efficacy of these anticancer treatments. Second, one study outlined how aspirin might produce resistance against chemotherapy by looking at how the drug affects P-glycoprotein (P-gp) expression (Rotem et al., [@B53]). It was found that aspirin and similar drugs induce protein kinase C (Zhu et al., [@B77]; Zimmermann et al., [@B78]), which enhances the activity of a nuclear factor for IL-6 expression (Trautwein et al., [@B67]; Combates et al., [@B19]). This consequently increases the activity of the MDR1 promoter (Combates et al., [@B20]). MDR1 encodes for an efflux pump called P-gp, which removes a number of anticancer drugs from the cell, thereby causing the chemotherapy agent to be ineffective at normal concentrations. Rotem et al. concluded that although aspirin reduces cellular proliferation in all 3 PCa cell lines studied (DU-145, PC-3, and LNCaP), it induces a three-fold increase in the percentage of cells expressing P-gp in LNCaP cell lines on the other hand (Rotem et al., [@B53]). This expression of P-gp was reversible, only persisting around 3 days, implying that it was not mediated by changes at the genetic level (Rotem et al., [@B53]). Third, it has been noted that aspirin causes cells to become more thermotolerant by increasing heat shock protein (HSP)-70 expression in these cells (Amici et al., [@B4]). The administration of this drug might thus also interfere with hyperthermic treatment, which is commonly used in conjunction with chemotherapy or radiotherapy to enhance the effectiveness of these forms of treatment. Conclusion and perspectives {#s7} =========================== While large epidemiological studies have significantly shown an inverse correlation between aspirin intake and cancers like PCa, tests and assays using cell lines have revealed desirable and undesirable outcomes that need to be explored more thoroughly. It is thus clear that there are many reasons why clinicians need to at least be aware if their patients with PCa are taking aspirin. Whether or not aspirin can be used as an adjuvant to therapy for PCa is yet undecided. This review strongly warrants further consideration of the signaling cascades activated by the aspirin (Table [1](#T1){ref-type="table"}), which may lead to new knowledge that might be applied to improve the diagnosis, prognosis and treatment of PCa (Figure [1](#F1){ref-type="fig"}). ###### **Proposed molecular mechanisms relating aspirin use in prostate cancer**. **Mechanism of Action** **References** --------------------------------------- ----------------------------------------------------------------- --------------------------- Cancer detection Decreases PSA mRNA and protein levels Kashiwagi et al., [@B36] Regulation of proliferation/apoptosis Inhibition of SSAT Li et al., [@B42] Reduction in cyclin D1 Olivan et al., [@B49] Decreases AR, upregulation of EP~3~ and downregulation of EP~2~ Kashiwagi et al., [@B36] Blocks Wnt/B-catenin signaling Lu et al., [@B45] Decreases survivin expression, TRAIL-induced apoptosis Yoo and Lee, [@B75] Enhances caspase 3 expression Royle et al., [@B54] Inhibition of NF-κB pathway and decrease in AR expression Lloyd et al., [@B44] Regulation of metastasis Increases a2-integrin expression Olivan et al., [@B49] Inhibition of NK-kB pathway and decrease in uPA expression Lloyd et al., [@B44] Induction of p75^NTR^ through MAPK pathway and Nag-1 Wynne and Djakiew, [@B73] Resistance to treatment Increases expression of MDR1 Rotem et al., [@B53] Increases HSP-70 expression Amici et al., [@B4] *SSAT, spermidine/spermine N(1)-acetyltransferase; AR, androgen receptor; EP~3~, E-type prostaglandin receptor 3; EP~2~, E-type prostaglandin receptor 2; TRAIL, tumor necrosis factor-related apoptosis inducing ligand; NF-κB, nuclear factor kappa-light-chain-enhancer of activated B cells; uPA, urokinase-type plasminogen activator; p75^NTR^, p75 neurotrophin receptor; MAPK, mitogen-activated protein kinase; Nag-1, N-acetylglucosamine-1; MDR1, multidrug resistance protein 1; HSP-70, heat shock protein 70*. Author contributions {#s8} ==================== All authors listed were involved in the concept, literature screening, and writing of the article, and approved it for publication. Funding {#s9} ======= This research was supported by funding from the Medical Practice Plan (MPP) at AUB-FM. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Conflict of interest statement ------------------------------ The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer JL and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review. We would like to thank all members in Dr. Abou-Kheir\'s Laboratory (The WAK Lab) for their help on this work. [^1]: Edited by: Shanmugasundaram Ganapathy-Kanniappan, Johns Hopkins School of Medicine, USA [^2]: Reviewed by: Gaetano Facchini, Istituto Nazionale Tumori Fondazione G. Pascale (IRCCS), Italy; Jian Lu, Johns Hopkins University, USA [^3]: This article was submitted to Cancer Molecular Targets and Therapeutics, a section of the journal Frontiers in Pharmacology

Introduction {#Sec1} ============ Pelvic organ prolapse (POP) is a significant health issue in females worldwide \[[@CR1], [@CR2]\]. There are approximately 250,000 procedures annually in the USA for POP, with as many as 29% of women having to undergo repeat surgery \[[@CR2]\]. Traditional anterior repair of cystocele or anterior compartment prolapse utilizing the patient's own tissue is a compensatory procedure that utilizes weakened and/or damaged tissue and has reported failure rates in the range of 40--60% \[[@CR3]\]. Additionally, plication or colporrhaphy techniques address only midline defects in the anterior compartment and do not provide any apical support which may also contribute to the failure rates seen with this type of repair \[[@CR3]\]. Abdominal Y-mesh sacralcolpopexy has the highest cure rates in the literature for vault prolapse, and the benefit of utilizing mesh in this particular repair seems to outweigh the risks \[[@CR4]\]. It results in anatomic repair with minimal tension and does not rely on the patients weakened tissue to maintain support. With the success of apical graft use, more recently, graft augmentation of prolapse repair has been utilized via the vaginal route. A recent Cochrane review has confirmed that mesh use in the anterior compartment has a lower failure rate versus traditional repair \[[@CR5], [@CR6]\]. Mesh trocar-based kits such as Perigee and Prolift were developed to try to simplify the placement of mesh, and initial studies have shown cure rates in the range of 87--96% \[[@CR7]--[@CR9]\]. These first-generation kits all required needles to be passed through the groins and obturator space with mesh arms subsequently pulled through this same space to attach the body of the graft to the levator muscles. Although many studies have shown successful cure rates when used by experienced surgeons in the right patient population, complications resulting from the external needle passes such as visceral and vascular injury as well as post-operative mesh complications have been concerning \[[@CR9]--[@CR11]\]. Vaginal or pelvic pain from the mesh arms being too tight, as well as issues such as rates of mesh extrusion as high as 15%, have been reported \[[@CR12]--[@CR14]\]. Additionally, a shortfall of most of these first-generation kits, is that they did not offer true level I or apical support in the anterior compartment, which may have led to apical failure. In an attempt to both benefit from the advantages of mesh use in the anterior compartment and also decrease some of the associated risks seen in the first-generation kits, a new single-incision vaginal approach was developed that utilizes a lighter and less-dense type I polpypropylene mesh (Intepro Lite, AMS) and eliminates all blind external needle passes. It also obtains true level I support through an anterior approach, by attaching the mesh, in an adjustable, tension-free matter to the sacrospinous ligaments via a novel and very minimally invasive fixation technique involving small self-fixating tips attached to the apical mesh arms. The purpose of the current study is to report the outcomes on a group of patients that underwent this procedure for treatment of symptomatic anerior compartment and/or apical prolapse at our specialty urogynecology center. Materials and methods {#Sec2} ===================== This study is a descriptive retrospective case series of 60 consecutive women with symptomatic stage 2 or greater anterior compartment prolapse (cystocele) that underwent anterior repair with mesh graft augmentation and vaginal vault suspension through a single transvaginal incision utilizing the Anterior Elevate by AMS (American Medical Systems, Minnetonka, MN, USA) system over a consecutive 12-month period at our center. Comprehensive pre-operative urogynecologic exams were completed including prolapse quantification utilizing the International Continence Society Pelvic Organ Prolapse Quantification (POP-Q) staging system. Inclusion criteria were patients with symptomatic anterior compartment prolapse ≥stage 2 that received the Anterior Elevate procedure over a consecutive 12-month timeframe at our center. Exclusion criteria included any patient that had any other vault suspension procedure at the time of the surgery. Additional testing included complex urodynamic testing to evaluate for the presence of concomitant stress urinary incontinence (SUI) (or detrusor instability) with and without the patient's prolapse reduced. If SUI was documented on urodynamic testing with or without the prolapse reduced, the patient was scheduled for a sub-urethral mesh tape sling procedure at the time of surgery. A sling was not placed prophylactically in any patient if SUI was not seen on testing. Procedure technique {#Sec3} ------------------- The procedure is begun with injection of a hydrodissection solution (25 cc of 1% lidocaine with epinephrine 1:200,000 diluted in 250 cc of saline) into the anterior vaginal wall. Typically, 40--60 cc of this solution is utilized. A 3 to 4-cm vertical incision is then made in the anterior vaginal wall starting at the level of the bladder neck towards the vaginal vault or the cervix. The incision is kept as small as possible and is NOT taken all the way to the apex of vagina. If the uterus was in situ, it was left in place in the current trial. A full-thickness dissection is then completed in the relatively avascular vesico-vaginal space out to the pelvic sidewalls and then up to the ischial spines bilaterally. Once the spine is reached, a medial sweep with the index and/or middle finger is completed to isolate the sacropinous ligament (SSL). It should be noted that the retropubic space does not need to be entered, nor does the ligament need to be visualized. The bladder should also be dissected off of the cervix or vaginal vault. The bladder neck arms of the graft are placed into the obturator internus muscle at the level of the bladder neck with their small self-fixating tips. The separate apical arms are then placed in the SSLs bilaterally approximately 2 cm medial to the ischial spine (Fig. [1](#Fig1){ref-type="fig"}). The apical portion of the graft is then fed over the arms attached to the SSLs and the vault adjusted into place in a tension free manner with an adjusting tool provided. Two 2--0 PDS sutures are used to attach the graft in the midline to the pericervical ring or the vaginal cuff. Once the final adjustment is reached, the graft is locked in place with small locking eyelets and the excess mesh cut off of the apical arms. Minimal to no vaginal epithelium is excised and the incision closed with a 2--0 delayed absorbable suture in a running-locked fashion. Cystoscopy is completed, foley catheter and vaginal packing placed for 24 h.Fig. 1Apical arm of the Elevate system placed into the sacropinous ligament 2 cm medial to the ischial spine. (Reproduced with permission from AMS) If an incontinence procedure was completed concomitantly, a separate sub-urethral incision was made and the tension-free sling placed utilizing standard technique. Remaining prolapse procedures were then completed as necessary. If stable, patients were discharged home on post-operative day 1. Vaginal estrogen cream was started 1 week post-operatively and used every other day. Patients were evaluated in the office at 4 weeks, 3 months, 6 months and then every 6 months thereafter. ICS POP-Q staging was completed as well as subjective assessment of prolapse (feeling or seeing a bulge), incontinence and urinary urgency and frequency symptoms. Quality of life questionnaires utilized were the Incontinence Impact Questionnaire-Short form (IIQ-7) and the Urogenital Distress Inventory-Short form (UDI-6). Objective cure was defined if the midline anterior vaginal wall (points Aa and Ba) was \<1.0 − 1.0 cm inside the hymenal ring and the apex was also less than or equal to stage I. Results {#Sec4} ======= Patient demographics are presented in Table [1](#Tab1){ref-type="table"}. Of the 60 patients, 11 (18.4%) had stage 2 prolapse, 43(71.6%) stage 3, and 6 (10%) had stage 4 on pre-operative pelvic examination and POP-Q scoring. The mean Ba value (±SD) was +2.04 ± 1.3 cm outside the vaginal opening, average C value (±SD) was −2.7 ± 2.9. Sixteen (26.6%) patients had previous anterior vaginal wall repairs and had recurrent cystoceles. Seventy-seven percent of patients had clinically significant apical prolapse. SUI was present in 61.6% of patients and tension-free tape slings were placed at the time of their surgeries. Twenty-six patients (43.3%) presented with SUI and 11 (18.3%) were discovered to have occult stress leakage with their prolapse reduced during urodynamic testing. Concomitant procedures at time of surgery included, 26/60 patients (43.3%) with posterior repair (16 with mesh grafts, 10 no graft used). No patient had any other vault support besides the Anterior Elevate and no patient underwent hysterectomy (27% of patients had their uterus in situ) (Table [2](#Tab2){ref-type="table"}). Nineteen patients (31.6%) were sexually active prior to surgery.Table 1Demographics*N* = 60Age (years ± SD)70.8 ± 6.1 (range 51--81)Parity2.8 ± 1.6 (range 0--8)Previous hysterectomy*N* = 44 (73%)Menopausal*N* = 59 (98.4%)Estrogen use*N* = 19 (31.6%)Previous repair*N* = 16 (26.6%)\>1 Previous repair*N* = 5 (8.3%)Prolapse stageII*N* = 11 (18.4%)III*N* = 43 (71.6%)IV*N* = 6 (10.0%)Table 2Concomitant proceduresProcedure*n*%Hysterectomy00Sling Mini-sling3355 TOT23.3 TVT23.3Posterior repair No graft1016.6 Mesh graft1626.6 Average blood loss was 47 cc (range 25--125 cc) and was deemed minimal. There were no post-operative bleeds or hematomas and no patient had to be taken back to the O.R. for bleeding or pain. There was one midline cystotomy above the trigone that occurred during the dissection of the anterior wall. This was closed in a two-layer fashion with absorbable sutures and the mesh still placed and ureteral patency confirmed. The average hospital stay was 23 h (range 18 h to 2 days). Average time to void was 3.6 days (range 1--10 days). Foley catheters were taken out on post-operative day 1 and voiding trial attempted. If patient did not pass the voiding trial, she was sent home with an indwelling catheter and voiding trial re-attempted on post-operative day 3. Twelve patients required catheterization for more than 3 days with ten of these having sling procedure at time of surgery. One patient who had the complication of cystotomy had her catheter for 10 days and had no sequalae from the injury. Four patients suffered from post-operative urinary tract infections (UTI); however, one of these had a history of recurrent UTIs. Average follow-up was 13.4 months (range 3--24 months). Only one patient has symptoms of recurrent prolapse for a subjective cure rate of 98.4%. Objective/anatomic cure rate was 91.7% using a definition of Ba \< −1.0 (≤stage I) and no patients have required repeat surgery for prolapse. Mean Ba value was −2.45 ± 0.9, point C was −8.3 ± 0.9, and TVL was 9.1 ± 0.3 cm. Mean vaginal length did not change statistically from pre-operative values (Table [3](#Tab3){ref-type="table"}).Table 3Pre-operative versus post-operative POP-Q measurements (mean)PreoperativePostoperative*P* value (*t* test)Mean POP-Q measurementsPoint Aa (cm)+1.4 ± 1.4−2.4 ± 0.8\<0.001Point Ba (cm)+2.0 ± 1.3−2.5 ± 0.9\<0.001Point C (cm), cervix−2.7 ± 2.9−8.3 ± 0.9\<0.001Point Ap−1.1 ± 1.3−2.4 ± 0.6\<0.001Point Bp−0.9 ± 1.6−2.3 ± 0.6\<0.001Total vaginal length (cm)9.09 ± 0.59.16 ± 0.30.343 There have been no patients that reported any significant buttock or leg pain that would be consistent with a pudendal or obturator type of neuropathy. Three patients suffered from post-operative levator myalgia or vaginal pain that required short-term treatment with NSAIDS and/or physical therapy with resolution of all patients' pain. In this predominantly elderly post-menopausal patient group, 31% were sexually active pre-operatively with one of these patients reporting clinically significant dyspareunia that is being treated with conservative measures (physical therapy) and is improving. Thirty-two patients (53%) complained of significant urge symptoms pre-operatively, and 20 of these patients (62%) had resolution of these symptoms post-operatively following their surgery. One patient (1.6%) developed de-novo urge symptoms post-operatively requiring treatment with anticholinergic agents. Overall IIQ-7 and UDI-6 scores improved significantly from baseline to follow-up (Table [4](#Tab4){ref-type="table"}). The UDI-6 subdomains of urge, stress, and obstructive symptoms were also evaluated and a significant improvement in each domain was also seen in the overall group (Table [4](#Tab4){ref-type="table"}). Patients that had a concomitant sling procedure or that had a previous repair did not statistically differ in improvement in these QOL indices compared to those that did not (Tables [5](#Tab5){ref-type="table"} and [6](#Tab6){ref-type="table"}). The UDI-6 subdomain scores were also evaluated in the patients that had a concomitant sling or not and these results can be seen in Table [6](#Tab6){ref-type="table"}.Table 4Overall UDI-6 and IIQ-7 scores (Pre-operative vs. Post-operative) and UDI-6 subscalesVariable (*n* = 58)^a^Baseline mean ± SDFollow-up mean ± SDFollow-up vs. baseline*P* value (*t* test)UDI-642.9 ± 26.922.4 ± 24.4−20.6 ± 30.8\<0.001IIQ-731.5 ± 25.715.4 ± 23.4−16.1 ± 26.9\<0.001UDI subscalesIrritative (56)52.4 ± 31.726.9 ± 32.3−24.4 ± 38.5\<0.001Stress (56)44.6 ± 29.825.4 ± 30.4−20.2 ± 36.5\<0.001Obstructive (58)40.4 ± 30.415.5 ± 22.7−24.1 ± 32.9\<0.001^a^58/60 patients filled out both pre- and post-operative questionnairesTable 5IIQ-7/UDI-6 scores in patients with previous anterior compartment repairs compared to those that did notPrevious repairIIQ-7 PreIIQ-7 PostIIQ-7 ChangeUDI-6 PreUDI PostUDI-6 ChangeNo = 54Mean29.310.6−18.727.713.6−14.1StDev26.016.825.019.015.323.4Yes = 16Mean36.428.1−8.343.525.6−17.9StDev25.732.931.319.723.423.2*T* test (between groups)0.350.010.190.010.020.58Table 6IIQ-7/UDI-6 scores (and UDI-6 domain scores) in patients with and without concomitant sling at time of surgeryQuestionnaire (*n*)Pre-operative scorePost-operative scoreMean improvement*P* valueIIQ-7Sling (37)33.2 ± 26.016.6 ± 24.616.6 ± 28.80.001No sling (21)28.1 ± 26.013.6 ± 22.114.5 ± 24.30.013*T* test(between groups)0.480.650.72UDI-6Sling (38)45.1 ± 23.626.0 ± 27.019.0 ± 28.6\<0.001No sling (22)38.5 ± 32.216.4 ± 19.022.1 ± 35.10.008*T* test(between groups)0.370.150.72UDI subscalesIrritativeSling (37)53.2 ± 31.426.8 ± 33.526.6 ± 37.6\<0.001No sling (21)48.4 ± 32.927.3 ± 31.120.6 ± 40.80.031StressSling (36)49.1 ± 27.028.4 ± 31.619.9 ± 38.40.004No sling (20)36.7 ± 33.620.5 ± 28.120.8 ± 33.70.012ObstructiveSling (35)44.0 ± 33.617.6 ± 24.825.7 ± 29.0\<0.001No sling (21)34.1 ± 35.512.1 ± 18.721.4 ± 39.10.021 Three patients suffered from significant SUI post-operatively. One had a concomitant tension-free sling (TVT type) at time of surgery that failed and the other two did not have a concomitant sling as they did not suffer from SUI pre-operatively and tested negative on UDTs. The patient that had the sling already placed is scheduled for treatment with injectable bulking agents and the other two are being scheduled for outpatient sling procedures. There have been no mesh extrusions noted to date. There were no post-operative infections of the mesh and no mesh had to be removed secondary to infection or pain or allergic reaction. No patients had to be taken back to the operating room for revision of the mesh or release of any of the mesh attachment points for pain or dyspareunia. Discussion {#Sec5} ========== The current series utilizing the Anterior Elevate system, showed excellent anatomic results and an objective cure rate of 92% with up to 24 months of f/u (mean f/u 13.4 months). Subjectively, only one patient complained of symptomatic prolapse (98.4% subjective cure rate) and no patient has had prolapse of the anterior wall outside of the introitus. These results are consistent with other series utilizing synthetic mesh for anterior compartment repair; however, we believe the Anterior Elevate procedure to be less invasive and a more simplified technique to place an anterior wall graft. Additionally, it allows for concomitant apical support at the time of anterior compartment repair. Goldberg et al. showed the anterior approach to SSL to result in anatomic restoration of the vaginal vault with cure rates consistent with the posterior approach \[[@CR15]\] which our anatomic results support this as well. The Anterior Elevate system utilizes small self-fixating tips for attachment of the graft to the obturator internus muscle at the level of the bladder neck and into the SSLs at the apex (Figs. [2](#Fig2){ref-type="fig"} and [3](#Fig3){ref-type="fig"}). The attachment of the graft to the SSLs involves minimal dissection with placement of the small tip into the mid-portion of the ligament. Secondary to this, risk of nerve injury or post-operative pain pudendal nerve type pain syndromes is minimal and this is supported in the current trial with no patients developing a post-operative pudendal nerve pain type syndrome and no patient requiring any surgical intervention for post-operative pain or dyspareunia. The elimination of lateral mesh arms penetrating through and through the levators also should help reduce the risk of vaginal or pelvic pain from the arms being placed too tight and "banding" in the vagina. This and/or mesh "bunching" seems to be the cause of most cases of dyspareunia with the first generation trocar-based mesh kits \[[@CR12], [@CR13], [@CR16]\]. We did have three patients develop post-operative pain remote from surgery that required treatment; however, it was all deemed muscular and responded to pelvic floor physical therapy and did not need surgical intervention. This percentage is consistent with other forms of prolapse treatment. One does need to be careful; however, in the adjustment of the graft at the apex as if too much tension is placed then this could potentially cause pain that may require release.Fig. 2Close-up view of the small self-fixating tip attaching into the ligament (Reproduced with permission from AMS)Fig. 3Final adjustment of the graft into place. The bladder neck portion of the graft has been fixated to the levators and the apical portion of the graft is slid up the arms in a tension-free manner to elevate the anterior wall and vault. (Reproduced with permission from AMS) Concerning the safety of the procedure, bleeding was minimal in most cases with average blood loss at 47 cc. No patients developed post-operative hematoma nor did any require blood transfusion. One patient suffered a midline cystotomy during the procedure; however, this was during the dissection of the anterior wall and was not related to the mesh or needle passes. We repaired the cystotomy with a double-layered closure, still placed the mesh, and she recovered without sequelae. Overall, with the elimination of blind external needle passes the risk of intra-operative bladder or vascular injury is reduced. Our mesh extrusion rate in the current trial was very low, and we attribute that to two variables: mesh quality and surgical technique. The mesh used in the current trial (Intepro Lite) is a type I macroporous polypropylene mesh that is 50% less dense and lighter than the first generation type I mesh (Intepro) used in Apogee/Perigee. In a previously published trial from our center, we reported mesh extrusion rate of 6.5% with the use of the Perigee procedure for treatment of cystocele in a similar group of patients \[[@CR17]\], which is consistent with other reports in the literature. Utilizing a less dense and softer mesh, results in a smaller mesh load and, most likely, less inflammatory response during healing which may reduce mesh extrusion rates. Additionally, our surgical technique of utilizing a deeper dissection plane, i.e., a full-thickness dissection utilizing hydrodissection in a more avascular space, allows for less bleeding during the procedure, less risk of post-operative hematoma, and a thicker vascular flap healing over the mesh. All of these factors may contribute to the lower extrusion rate seen in our trial. Other lighter meshes have also been introduced, however, have not shown this low of extrusion rate. Prolift-M utilizes a soft polypropylene mesh that has a similar density to Intepro-Lite however is 50% Monocryl, and therefore, 50% of the mesh is absorbable. The thought was that this would decrease vaginal exposure rate; however, in a recent study, Cosson et al. reported a relatively high extrusion rate of 10.5% \[[@CR18]\]. Sixteen patients (26.6%) had previous repairs and would be considered higher risk for failure; however, their cure rate was consistent with patients that had not had prior repair and there was no difference in complications seen in these patients either. In our opinion, this is a group of patients that benefit the greatest from a graft, given a previous failure using their own tissue, and the results of these patients in our series is very encouraging. The current study is limited by its retrospective nature and its medium-term follow-up. Another limitation, again inherent to a retrospective single-center trial, is surgeon bias in evaluating their own surgical outcomes. A limitation of any surgical trial that also has to be considered is surgeon experience with the particular procedure or similar procedures as well as the anatomy of the dissection involved. Complications in the current trial may have been kept to a minimum secondary to this variable. Conclusion {#Sec6} ========== In conclusion, we have found the vaginal repair of anterior/apical wall prolapse utilizing a lightweight soft type I anterior wall mesh placed via the Elevate system---a safe, minimally invasive, and effective procedure for the treatment of anterior wall prolapse in this subset of mostly post-menopausal patients. We feel that the role of mesh in vaginal repairs is in its infancy, and much study still needs to be done to determine the ideal material to be utilized and the optimal way to place and attach the graft vaginally and the proper patient to utilize it in; however, it can be expected that improvements in technology and techniques will continue to improve outcomes. The small self-fixating tips on the arms of the Elevate procedure that hold the mesh in place are one of the least invasive approaches to fixate a mesh graft in place vaginally to date. The elimination of trocar-based, blind needle passes through the groins also seems to decrease the risk of complications. We do recommend further prospective studies with longer-term follow-up to further help delineate its role in clinical practice. **Conflicts of interest** RD Moore is a consultant and speaker for the research grants of American Medical Systems (AMS). GK Mitchell has nothing to disclose. JR Miklos is a consultant of the research grants of AMS and a speaker for Coloplast. No funding, research grants, stipends, or any other outside assistance was received in support of the current paper. **Open Access** This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.

Disclosure: No potential conflicts of interest were disclosed. Introduction {#j_raon-2014-0040_s_001} ============ Ectopic spleen tissue takes two forms, it is either congenital and presents as accessory spleens or an acquired condition, called splenosis. Splenosis occurs as un-encapsulated splenic tissue localized outside the spleen. It is a benign condition, most commonly linked to splenic trauma or splenectomy. Post traumatic splenosis is believed to be a rare state; however, it is more likely only to be under-recognized. Patients are in fact most often asymptomatic and the majority of cases are diagnosed by coincidence.[@j_raon-2014-0040_ref_001] As a rule, there are multiple splenic tissue implants embedded in serous surfaces of abdominal cavity such as mesentery, omentum, surfaces of the colon or diaphragm. After penetrating abdominal trauma, extraperitoneal splenosis has also been reported in subcutaneous tissue, pleural cavity or pericardium.[@j_raon-2014-0040_ref_002]-[@j_raon-2014-0040_ref_005] Rarely, splenic tissue can implant in parenchyma of a visceral organ such as liver. In that case, the condition is defined as hepatic splenosis. Mechanism of splenic tissue spread is not completely understood. One of the hypotheses postulates invagination of splenic implants into sub-capsular area of the liver after spleen trauma. Alternatively, deep-seated hepatic splenic nodules would results from the microembolisation via the splenic vein. Nodular lesions in liver parenchyma are non-specific on imaging and can mimic malignant as well as benign lesions.[@j_raon-2014-0040_ref_006] Exploratory laparoscopy is the least invasive method for reaching definitive diagnosis and is the most suitable, especially in patients with history of malignant disease where liver metastases are suspected. Case report {#j_raon-2014-0040_s_002} =========== A 22-year old Caucasian male was presented after operative procedure due to left-sided inguinal hernia, during which a cystic mass inside of hernial sac was revealed. Histology of the tumour disclosed it to be mature cystic teratoma. The patient was treated for immature teratoma with a prevalence of neuroepithelial components and with high mitotic activity 22 years prior and the tumour most likely originated from retained right testicle. Surgical procedure was performed on day 6 after birth. The tumour had perforated and was macroscopically removed. He received 3 cycles of chemotherapy with vincristine, actinomycin D, cyclophosphamide and cisplatin in the adjuvant setting. On follow-up examinations there were no signs of relapse. At the age of 4 he underwent splenectomy after a motor vehicle accident. For the past year he had been treated for Henoch--Schönlein purpura with renal impairment, he was receiving methylprednisolone. After the operative procedure for hernia, patient's alpha-fetoprotein (AFP), Beta unit of human Chorionic Gonadotropin (Beta hCG) and Lactate dehydrogenase (LDH) levels were within normal ranges: AFP 2.2 kU/L (\< 5.8), Beta hCG \< 0.1 U/L (\< 2.0), LDH 2.25 ukat/L (\< 4.13). Additional laboratory testing revealed thrombocyte count 491 \*109 /L (140--340), neutrophil count 77.7% (40--75%), lymphocyte count 17.3% (20--40%), eosinophil count 0.7% (1--6%), and sedimentation rate 29 mm/h (\< 19). Electrolyte concentrations, liver function tests, LDH and thyroid hormones were within normal limits. Imaging diagnostics were performed for disease staging. Scrotal ultrasound revealed small left-sided hydrocele, and no suspicious lesions in left testicle. Left epididymis seemed appropriate. Computed tomography (CT) scan of thoracic organs demonstrated no signs of disease progression. However, on abdominal CT scan two poorly demarcated areas in the 6th liver segment and under the capsule were seen in portal phase of contrast enhancement ([Figure 1](#j_raon-2014-0040_fig_001){ref-type="fig"}). There was also small perihepatic oval shaped peritoneal solid lesion of same appearance. Liver size was normal. {#j_raon-2014-0040_fig_001} With a view to define perihepatic lesions, magnet resonance imaging (MRI) with hepatospecific contrast medium (Gd-EOB-DTPA disodium, i.e. Primovist, Bayer Pharma AG) was performed. Five round lesions ranging from 0.7 to 2.6 cm in diameter were visible in 2nd, 6th segment and on the border of 6th and 7th segment. They were subcapsularly in liver parenchyma and on the surface of the liver. Lesions were hypointense in T1 weighted images (WI), slightly hyperintense in T2 WI, after administration of hepatospecific contrast medium enhanced during arterial phase, after that they remained hypointense in the late phase images ([Figures 2](#j_raon-2014-0040_fig_002){ref-type="fig"}‒[4](#j_raon-2014-0040_fig_004){ref-type="fig"}). Lesions were suspected to be metastases. {#j_raon-2014-0040_fig_002} {#j_raon-2014-0040_fig_003} {#j_raon-2014-0040_fig_004} The patient underwent two ultrasound guided fine-needle aspiration biopsies. Sonographically lesions were barely seen, mildly hyperechoic ([Figure 5](#j_raon-2014-0040_fig_005){ref-type="fig"}). Obtained samples were unfit to warrant a cytologic diagnosis. {#j_raon-2014-0040_fig_005} In order to verify suspicious lesions, laparoscopic excision was carried out and intraoperative frozen section analysis evinced spleen tissue. Histology of the obtained sample confirmed hepatic splenosis. Tumour, a coincidental finding at operative procedure in left inguinal region, was most likely a metastasis of childhood germ cell tumour. There were no signs of disease spread. Treatment of asymptomatic hepatic splenosis was not indicated. Discussion {#j_raon-2014-0040_s_003} ========== Hepatic splenosis is heterotopic autotransplantation of splenic tissue, as a rule it is a consequence of spleen trauma or splenectomy carried out from other reasons. Ectopic splenic tissue in abdominal cavity is present in more than 60% of patients after traumatic splenic rupture; however, isolated hepatic localization is described only in individual cases.6-10 The diagnosis is usually an incidental finding following diagnostics in the scope of another disease, as patients are most often asymptomatic. Mescoli *et al*. report an average time of 29 years between splenectomy and the liver nodules detection.6 The patient in our report underwent splenectomy 18 years before the hepatic splenosis was established. He was subjected to regular follow- up after the treatment for childhood immature teratoma, and intermittent abdominal sonography revealed no lesions in hepatic parenchyma. However, none of the imaging techniques is specific enough to identify hepatic splenosis; ultrasound sensitivity is very poor, CT imaging is moderately sensitive and MRI has high sensitivity. Hypervascular nodular hepatic lesions are most commonly haemangioma, hepatic metastases, hepatic adenoma, focal nodular hyperplasia, hepatocellular carcinoma.11 Strictly peripheral lesions in hepatic parenchyma with peritoneal deposits in spleenless patient should warrant differential diagnosis of splenosis. With respect to childhood tumour, hepatic metastases and/or peritoneal dissemination seemed a plausible differential diagnostic option in presented case. The patient was subjected to operative treatment for abdominal tumour few days after birth. Tumour was linked to undescended right testicle and was histologically diagnosed to be immature teratoma. Data shows a link between intraabdominal teratoma and cryptorchidism in children, also, tumour localization at the *annulus inguinalis profundus* is believed to cause testicular retention.[@j_raon-2014-0040_ref_012] Mature teratoma presenting in our patient's adulthood was considered to be a metastasis of primary childhood tumour and less likely metastatic germ cell tumour originating in left testicle. It is known that contralateral testicular tumour is more often found in patients with history of teratoma and *in situ* disgenesis is present in 9% of the patients.[@j_raon-2014-0040_ref_012] However, clinical examination, biochemistry and imaging excluded left testicle pathology with high probability. Considering differential diagnosis, nodular hepatic and perihepatic lesions could reflect metastases of immature childhood teratoma as well as mature adulthood teratoma. Immature teratomas are known to metastasize to the solid organs, including liver. Metastatic foci can contain histologically more mature elements than found in primary tumour. One of the reasons is retroconversion of metastases to differentiated mature teratoma under the influence of chemotherapy.[@j_raon-2014-0040_ref_014] Mature teratomas are usually asymptomatic and slow growing, 1.8 mm per year on average. That explains the longtime interval between primary tumor management after birth and finding the mature teratoma at the age of 22 in our case. Also, slow growth could explain presence of nodular hepatic lesions, assessed as possible metastases. On the other hand, the origo of metastases could be the adulthood teratoma, as malignant alteration and metastasizing occur in 1 to 2% of mature teratomas.[@j_raon-2014-0040_ref_015] Differential diagnosis of hepatic splenosis is wide and indirect diagnostic procedures are unreliable as opposed to histologic evaluation. Hematological evaluation can be useful in assessing any persistence of functioning splenic tissue and the absence of Howell-Jolly bodies, Heinz bodies or pitted cells on blood smears may be helpful in diagnosing splenosis. However, sensitivity of the test is low, especially when there is only small amount of ectopic splenic tissue. There are no typical radiological features of intrahepatic splenosis. Sonographic appearance is completely unspecific, similar to the current case. Hypoechoic, homogeneous, solid and well circumscribed implants are described in literature, however, in our case lesions were sonographically mildly hyperechoic.[@j_raon-2014-0040_ref_010] In contrast-enhanced CT scans, intrahepatic splenosis is generally revealed as round, oval, or lobular, well circumscribed, non-calcified and homogeneously enhancing. Before injection of contrast medium the implants are usually hypodense or isodense to the liver. In the literature, there are few published descriptions of hepatic splenosis on MRI.[@j_raon-2014-0040_ref_010],[@j_raon-2014-0040_ref_016]-[@j_raon-2014-0040_ref_021] On MRI, the characteristics of splenosis are limited to anecdotal cases that have found such lesions as homogeneous, of low signal intensity in T1 WI and moderately hyperintense in T2 WI. Sometimes hypointense thin layer of capsule around the lesion is found in T1 and T2 WI. In arterial or/and portal phase of enhancement lesions are mostly hyper-vascular and hypointense in delayed images.[@j_raon-2014-0040_ref_017] According to published reports intrahepatic splenosis is most commonly manifested as nodular, solitary lesions, and ranging from 2 to 6 cm in size.[@j_raon-2014-0040_ref_006],[@j_raon-2014-0040_ref_008],[@j_raon-2014-0040_ref_010] Described characteristics are true in our case with the exception of the number of lesions. In contrast with presented case Mescoli *et al*. reported that solitary splenic nodules were found in 24 out of 27 patients with the hepatic splenosis. In only one patient more than three nodular lesions were found.[@j_raon-2014-0040_ref_006] Difference may stem from an underestimation of the results based on radiological investigations. In our case five intrahepatic implants were identified with MRI and two with CT, what is in concordance with multiple studies, which have shown that MRI is superior to CT in sensitivity and accuracy of detecting hepatic lesions.[@j_raon-2014-0040_ref_022] Upon ultrasonography the lesions were particularly indistinguishable and mildly hiperechoic. In the literature, all the lesions were described as hypoechoic. Splenosis is thought to be uncommon, but the incidence is probably underreported since the majority of patients are asymptomatic. Distinguishing the aetiology of hepatic nodular lesions is important because it significantly alters therapeutic procedures. Typical imaging modalities such as US, CT or MRI will not differentiate splenosis from other entities and a histologic specimen needs to be obtained to reach definitive diagnosis. Unfortunately, result of fine-needle aspiration biopsy can be inconclusive as it was in presented case. On the other hand, laparotomy is an excessive operational procedure with potential complications in patients with hepatic splenosis. We presented a case in which hepatic splenosis has been confirmed by explorative laparoscopy. A laparoscopic approach is minimally invasive for the visualization of suspected intrahepatic masses, and allows access for potential liver biopsy or resection. Similar experiences have been described, but in only few published reports.[@j_raon-2014-0040_ref_009],[@j_raon-2014-0040_ref_023] Conclusions {#j_raon-2014-0040_s_004} =========== Due to the significant impact on treatment decisions intrahepatic splenosis must be considered in the diagnostic spectrum of nodular liver lesions, especially in patients with prior splenic trauma or surgery.

Introduction {#Sec1} ============ Glycan analysis is increasingly applied in biological research, clinical analysis, and pharmaceutical biotechnological production. Specific glycosylation patterns have been associated with states of health and disease \[[@CR1]--[@CR4]\]. Moreover, glycosylation changes may modulate the biological activity of proteins, as demonstrated, for example, for the glycosylation of the Fc moiety of recombinant immunoglobulin G \[[@CR5], [@CR6]\]. Many approaches for analysis of oligosaccharides from glycoproteins have been described \[[@CR7]--[@CR9]\], and strategies for the analysis of the glycosylation of immunoglobulin G (both natural and recombinant) have recently been summarized \[[@CR10]\]. A large number of methods focus on the analysis of released and subsequently derivatized glycans. These approaches allow in-depth analysis of the oligosaccharide structure independent of the carrier glycoprotein, but provide no information on the attachment site of the glycan. As the application of separation methods with optical or mass-spectrometric detection for the characterization of glycans is hindered by the lack of a chromophore and their poor ionization properties, a wide variety of derivatization strategies with a large number of different labels have been applied. In this review an overview of the techniques applicable for glycan derivatization is given, e.g., reductive amination, Michael addition, hydrazide labeling, and permethylation, followed by some aspects on purification of derivatized glycans. Additionally, strategies for the analysis of tagged oligosaccharides using chromatographic and electromigrative separation techniques as well as mass spectrometry (MS) and their use in quantitation will be addressed. Criteria for the choice of derivatization strategies and detection methods will be discussed. Derivatization strategies {#Sec2} ========================= A large number of glycan derivatization strategies have been described in the literature. The most common reaction employed is the reductive amination. Permethylation is also often applied. Alternatively, Michael addition or hydrazide labeling may be applied. Various compounds which provide the required functional group for the labeling reaction can be used. The application of a specific release method may restrict the labeling options. Reductive amination, Michael addition, and hydrazide labeling all require the reducing end of the glycan, which is not present on *O*-glycans released by reductive β-elimination. Labeling reagents are usually present in large excess, and most of the analytical techniques require a sample workup to remove excess labeling reagents. Most frequently, size-exclusion chromatography and solid-phase extraction (SPE) are applied. Release methods {#Sec3} --------------- An efficient, reliable method for the release of oligosaccharides from protein or peptide backbones is a prerequisite for their detection \[[@CR7]\]. Several methods for *N*-glycan or *O*-glycan release have been reported, and are represented in Fig. [1](#Fig1){ref-type="fig"} \[[@CR7]--[@CR19]\]. Fig. 1Strategies for *N*-glycan and *O*-glycan release from glycoproteins and glycopeptides *N*-Glycans are commonly released enzymatically using PNGase F or A (see Fig. [1](#Fig1){ref-type="fig"}). The hydrazinolysis approach is applicable for *N*-glycans as well as *O*-glycans \[[@CR7], [@CR18], [@CR19]\], yielding glycans with a free reducing terminus. Even though the reaction mechanism of hydrazinolysis is not fully understood, it is generally accepted that glycan release occurs via an initial β-elimination step prior to a reaction with hydrazine resulting in a hydrazone derivate \[[@CR19]\]. The method has some major drawbacks, such as the necessity to use anhydrous hydrazine, which is a highly toxic and explosive chemical also used in rocket fuel \[[@CR20]\]. Several other disadvantages of the hydrazinolysis method have been given by various groups: one example is the release of *O*-glycans by hydrazinolysis, which has to be followed by re-N-acetylation and a concomitant acetohydrazone cleavage step, which is crucial in order to obtain intact glycans with a free reducing terminus \[[@CR7], [@CR9], [@CR18]--[@CR23]\]. Enzymatic release of *O*-glycans has a much more limited use owing to the very narrow substrate specificity of the enzymes available. Only two *O*-glycanases are commercially available \[[@CR7]\], the use of which is unfortunately restricted to the liberation of the unsubstituted core 1 disaccharide Galβ1-3GalNacα linked to either serine or threonine. Accordingly, chemical release is the major method for O-linked oligosaccharide liberation and several options have been reported (see Fig. [1](#Fig1){ref-type="fig"}), all of which, however, have the inherent risk that a possible, unwanted degradation step called a "peeling" reaction can occur \[[@CR19]\]. Peeling refers to a base-catalyzed elimination reaction which results in the loss of the 3-substituent of the innermost residue of an oligosaccharide \[[@CR19]\]. Reductive amination {#Sec4} ------------------- ### Reaction {#Sec5} Glycans can be labeled at their reducing end using reductive amination. In this reaction, a label containing a primary amine group reacts in a condensation reaction with the aldehyde group of the glycan, resulting in an imine or Schiff base, which is reduced by a reducing agent to yield a secondary amine (see Fig. [2a](#Fig2){ref-type="fig"}). The reaction is often performed in dimethyl sulfoxide containing acetic acid \[[@CR24]\], but alternative approaches using tetrahydrofuran \[[@CR25]\] and methanol \[[@CR26]\] have been described. An advantage of this labeling approach is the stoichiometric attachment of one label per glycan, allowing a direct quantitation based on fluorescence or UV-absorbance intensity. Fig. 2Labeling of glycans. **a** 2-Aminobenzoic acid (2-AA) labeling via reductive amination, **b** 1-phenyl-3-methyl-5-pyrazolone labeling via a Michael-type addition, **c** labeling with phenylhydrazide, and **d** glycan permethylation ### Labels {#Sec6} Various labels have been used for the reductive amination of glycans. In two reviews, an extensive, though not complete list of labels for reductive amination has been given \[[@CR27], [@CR28]\]. The most widely applied labels are 2-aminobenzamide (2-AB), 2-aminobenzoic acid (2-AA), 2-aminopyridine (PA), 2-aminonaphthalene trisulfonic acid (ANTS), and 1-aminopyrene-3,6,8-trisulfonic acid (APTS). Labeling kits are available for the tags 2-AB, 2-AA, and PA (see, e.g., <http://www.ludger.com>) as well as for labeling with APTS (see <http://www.beckmancoulter.com>) and ANTS (see <http://www.prozyme.com>). 2-AB is a label that lacks negative charges and is widely applied in chromatographic analysis. An extensive database has been developed which uses the standardized elution positions of 2-AB-labeled glycans in hydrophilic interaction liquid chromatography (HILIC) with fluorescence detection for structural assignment \[[@CR29]\]. PA is also widely used in high performance liquid chromatography (HPLC) profiling \[[@CR30], [@CR31]\], and databases for structural assignments based on standardized elution positions have been developed as well \[[@CR32]\]. A major drawback of PA is the necessary recrystallization step prior to its use, since the commercially available compound lacks sufficient purity. The 2-AA label carries one negative charge, which makes it very versatile. It is used in HPLC and capillary electrophoresis (CE) separations as well as in positive-mode and negative-mode matrix-assisted laser desorption/ionization (MALDI) analysis, allowing detection of both neutral and sialylated glycan species \[[@CR33]--[@CR36]\]. APTS has three negative charges and is therefore very suitable for CE and capillary gel electrophoresis (CGE; see later) \[[@CR37], [@CR38]\]. However, the analysis of APTS-labeled oligosaccharides by MALDI appears to be difficult \[[@CR39]\]. Other labels have been explored for their potential in quantitation: A \[D~4~\]PA label was introduced for glycosylation analysis \[[@CR40]\]. Promising results were obtained; however, slight differences in retention times were observed between \[H~4~\]PA- and \[D~4~\]PA-labeled glycans when they were separated on a C~18~ column. No other stationary phases have been tested \[[@CR40]\]. In an alternative attempt, oligosaccharides from chondroitin sulfates were labeled with \[H~4~\]2-AA or \[D~4~\]2-AA \[[@CR41], [@CR42]\], and separation by size exclusion was performed to avoid possible shifts in retention time. A mass shift of only 4 Da was achieved, which may result in a partial overlap of glycan pair isotope envelopes in mass-spectrometric detection. Another option is to label oligosaccharides with \[^12^C~6~\]alanine and \[^13^C~6~\]alanine \[[@CR43], [@CR44]\]. With use of this approach, a difference of 6 Da is generated between the two different glycans, which is sufficient to have full resolution of the isotopic patterns of glycan pairs in MS. Moreover, the use of deuterium is avoided, so no retention shifts on C~18~ columns are observed. ### Reducing agent {#Sec7} The most widely applied reducing agent is sodium cyanoborohydride. Very high yields of labeled oligosaccharides have been reported using this compound \[[@CR24]\]. Alternatively, sodium triacetoxyborohydride \[[@CR45]\] and borane-diethylamine \[[@CR46]\] have been used; however, these compounds are not widely used within the glycoanalytical field. As an alternative, we recently described the use of 2-picoline borane for the reductive amination of oligosaccharides with 2-AA and 2-AB \[[@CR47]\]. The use of sodium cyanoborohydride results in the release of the toxic compound hydrogen cyanide, 2-Picoline borane is an efficient, nontoxic alternative. When comparing the two compounds, we observed equal efficacies for reductive amination using various tags such as APTS (unpublished results), 2-AA, and 2-AB \[[@CR47]\]. Moreover, 2-picoline borane can be employed in both aqueous and nonaqueous conditions \[[@CR47]\]. Owing to its superb properties, it has replaced sodium cyanoborohydride completely in the glycoanalytical protocols used in our laboratory, and we expect this to take place in other laboratories too. ### Optimal reaction conditions {#Sec8} The optimal conditions for the reductive amination reaction of *N*-glycans with the 2-AB and 2-AA labels using NaCNBH~3~ as the reducing agent were reported by Bigge et al. \[[@CR24]\] in 1995. The concentration of the labeling agent was recommended to be 0.25 M or greater, whereas more than 1 M reducing agent was required. The derivatization is enhanced by the addition of glacial acetic acid up to a content of 30% (v/v). A reaction temperature of 60 °C was found to be optimal. With use of these conditions, all glycans are derivatized within 2 h, and glycan degradation reactions such as acid-catalyzed loss of sialic acid are minimized. In a different study, the influence of different acids was evaluated for the derivatization of oligosaccharides with APTS \[[@CR48]\]. Acids with lower p*K*~a~ values, such as citric acid, malic acid, and malonic acid, were shown to result in higher yields than acetic acid when the derivatization was performed at 37 °C for 16 h. When higher temperatures and very acidic conditions are applied, partial degradation of the glycans will be induced with loss of sialic acids \[[@CR49]\]. Michael addition {#Sec9} ---------------- Labeling of the reducing end of glycans with reagents such as 1-phenyl-3-methyl-5-pyrazolone (PMP) by Michael addition is performed under alkaline conditions \[[@CR50]\], thereby avoiding the considerable risk of loss of sialic acids under acidic conditions which is intrinsic to reductive amination reactions (see earlier). The reaction itself is the nucleophilic addition of a carbanion to an α,β-unsaturated carbonyl compound, and is one of the most useful methods for the mild formation of C--C bonds. A detailed structural description of this glycan labeling procedure including the overall reaction scheme has been given by You et al. \[[@CR50]\] and can also be found in a simplified version in Fig. [2b](#Fig2){ref-type="fig"}. Labeling reagents such as PMP \[[@CR51], [@CR52]\] or its methoxy analog 1-(*p*-methoxy)-phenyl-3-methyl-5-pyrazolone are most often applied \[[@CR53]\]. The derivatization reaction itself is a base-catalyzed Michael-type addition involving a two step labeling process in which Michael donor molecules of the labeling reagent are formed and consecutively added to the reducing end of the glycan with a stoichiometry of two label molecules per glycan (II in Fig. [2b](#Fig2){ref-type="fig"}; \[[@CR50]\]). After the addition of the first label molecule, a water molecule is lost \[[@CR50]\], which results in the formation of an α,β-unsaturated carbonyl compound. The following step yields a Michael 1,4-addition product by conjugation of the second label molecule. You et al. \[[@CR50]\] introduced the novel labeling reagent 1-(2-naphthyl)-3-methyl-5-pyrazolone for glycan labeling in the presence of an ammonium catalyst, which showed increased sensitivity in UV detection as well as a faster reaction time as compared with the previously applied PMP \[[@CR50]\]. The use of the PMP label and related tags is restricted as no fluorescent labels have hitherto been reported. Moreover, as two label molecules are added per glycan, the chromatographic properties of the derivatized oligosaccharides are dominated by the tags, which may be disadvantageous in isomer separation. Hydrazide labeling {#Sec10} ------------------ Another labeling procedure can be accomplished by a reaction of glycans with labels providing a hydrazide end group (for the reaction mechanism, see Fig. [2c](#Fig2){ref-type="fig"}). Hydrazide labeling may be performed for mass-spectrometric detection \[[@CR54]\] in combination with chromatography \[[@CR55]\] or to allow biomolecular interaction analysis \[[@CR56]--[@CR59]\]. Lattova and Perreault \[[@CR54]\] demonstrated that the conversion of monosaccharides and oligosaccharides to the corresponding phenylhydrazones is quantitative, meaning that no unreacted sugars were observed in the mass spectra after the labeling procedure \[[@CR54]\]. In addition, it was shown that low picomolar sample concentrations were sufficient for structural identification by MALDI-MS and electrospray ionization (ESI) MS \[[@CR55]\]. Carboxymethyl trimethylammonium hydrazide (Girard's T reagent) is a label which places a permanent positive charge at the reducing end of a glycan \[[@CR60]\]. This label is suitable for sensitive analysis of glycans by MALDI time of flight (TOF) MS. Shinohara et al. \[[@CR58]\] were among the first to introduce hydrazide tagging of oligosaccharides for the introduction of a biotin tag, which allows further biomolecular interaction studies by immobilization of the glycan to streptavidin-coated carriers \[[@CR58]\]. Several applications of this approach, including the preparation of various hydrazide derivates followed \[[@CR56], [@CR57]\]. Leteux et al. \[[@CR56]\] described the preparation of biotin hydrazide derivates of oligosaccharides employing biotinyl-[l]{.smallcaps}-3-(2-naphthyl)-alanine hydrazide (BNAH) and 6-(biotinyl)-aminocaproyl hydrazide (BACH) in which the ring of the reducing end monosaccharide is nonreduced. In this study BACH, BNAH, and 2-amino-(6-amidobiotinyl)pyridine derivatives (the latter one being obtained under reductive amination conditions) were compared. The BNAH tag combines two advantageous features, i.e., the presence of a chromophore/fluorophore \[similar to 2-amino-(6-amidobiotinyl)pyridine\] with the presence of an intact pyranose ring of the labeled glycan (similar to BACH derivates) \[[@CR56]\]. A constitutive study by the same authors followed, where the same glycan tags conjugated to oligosaccharides on a streptavidin matrix were used to investigate their interaction with monoclonal antibodies, adhesion molecules, and E-, P-, and L-selectins \[[@CR57]\]. (Per-)methylation {#Sec11} ----------------- In permethylation, hydrogens on hydroxyl groups, amine groups, and carboxyl groups are replaced by methyl groups, yielding a hydrophobic sugar derivative (Fig. [2d](#Fig2){ref-type="fig"}) \[[@CR61]\]. This is an advantageous derivatization step as it enhances the glycans' signal strength in MS with both ESI and MALDI \[[@CR8]\]. Permethylation facilitates the spectral interpretation since both acidic and neutral structures can be measured in the positive ion mode \[[@CR62]\] as the sialic acids are stabilized after the procedure \[[@CR22]\]. Moreover, tandem MS of sodium adducts of permethylated glycans provides detailed information on linkage positions \[[@CR63]\]. For structural characterization of glycans, permethylation is often preceded by reduction of the reducing end aldose with sodium borohydride to achieve an open-ring alditol, thereby providing a mass tag which facilitates spectral interpretation \[[@CR63]\]. Ciucanu and Kerek \[[@CR64]\] have introduced the basic permethylation protocol which has become a standard procedure and is applied in many cases for *O*-glycan as well as *N*-glycan analysis with only minor alterations \[[@CR50], [@CR62], [@CR63], [@CR65]--[@CR68]\]. Morelle and Michalski \[[@CR22]\] provided a detailed step-by-step protocol and have summarized advantages and disadvantages of the procedure. Recently, Kang et al. \[[@CR69]\] proposed packing a solid phase (either sodium hydroxide mesh beads or powder) in microspin columns or in fused-silica capillaries to allow effective and quantitative microscale derivatization of oligosaccharides in a short time \[[@CR69]\]. The authors subsequently adapted this miniaturized approach to create a high-throughput method and combined it with an isotope labeling strategy \[[@CR70], [@CR71]\]. Recently, an elegant approach for sequential double permethylation for the analysis of sulfated glycans was introduced. Briefly, sulfated glycan samples are permethylated prior to the methanolytic cleavage of their sulfate groups. These oligosaccharides are then subjected to a second permethylation step using deuteromethyl iodide to label the hydroxyl groups resulting from methanolysis. By comparing the collision-induced dissociation (CID) MS^2^ spectra of permethylated and permethylated plus deuteromethylated samples, one can calculate the initial number of attached sulfate groups on the glycan. This method allows the detection and structural characterization of sulfated glycans using MS, avoiding the problem of signal suppression due to the presence of sulfate groups \[[@CR72]\]. Alternative methylation protocols solely transform the carboxyl groups of monosaccharides such as sialic acids into methyl esters, thereby stabilizing sialic acid linkages for subsequent mass-spectrometric analysis \[[@CR73]\]. Strategies for purification {#Sec12} --------------------------- After derivatization, labeled glycans often have to be purified prior to analysis. Not only does the excess of salts have to be removed (e.g., for MALDI analysis), but also the concentration of the labeling reagent, normally present in large excess during the labeling step, should be reduced. Five major strategies are applied for this purpose: SPE \[[@CR22], [@CR29], [@CR34], [@CR74], [@CR75]\], liquid--liquid extraction \[[@CR64]\], gel filtration \[[@CR76], [@CR77]\], paper chromatography \[[@CR23]\], and precipitation \[[@CR75]\]. Typical examples of purification strategies are summarized in Table [1](#Tab1){ref-type="table"}. SPE is mainly performed after reductive amination and hydrazide labeling \[[@CR55]\], whereas liquid--liquid extraction is the method of choice after permethylation \[[@CR61], [@CR64], [@CR69]\] or Michael addition \[[@CR51], [@CR52], [@CR78]\]. In permethylation procedures, liquid--liquid extraction may be followed by C~18~ SPE \[[@CR79]\]. Paper chromatography may be used after reductive amination and hydrazide labeling \[[@CR19], [@CR80]\]. Precipitation with acetone is not widely applied, but may be used to remove proteins \[[@CR75]\]. The strategies strongly differ in complexity as well as time consumption of sample handling. Among the five strategies, SPE is the method of choice for analyzing larger sample cohorts as it can be adapted to high-throughput setups \[[@CR34]\]. The diversity of stationary phases for SPE is relatively large, and its use in purification of derivatized glycans is extensive. Several stationary phases have been used for purification of derivatized glycans by SPE: reverse phase (RP), porous graphitized carbon (PGC), HILIC, and anion-exchange chromatography. Table 1Sample purification after glycan labelingMethodMaterialLabelCommentsReferencesSPERP, C~18~RA, aniline\[[@CR43]\]SPERP, C~18~PermethylationTFA may be added\[[@CR22], [@CR72], [@CR85]\]SPERP, C~18~HL, BACH\[[@CR55]\]SPEPGCRA, 2-ABSample application in water, elution using 25% (v/v) ACN\[[@CR83]\]SPEPGCRA, benzylamineSample application in water, elution using 20% (v/v) ACN\[[@CR82]\]SPEHILIC, nylon filter, Oasis HLB or amide-2RA, 2-AASample application in 95% ACN, elution using water\[[@CR33], [@CR86]\]SPEHILIC, celluloseRA, 2-AASample application in 80% ACN, elution using water\[[@CR34]\]SPEHILIC, microelution plateRA, 2-ABSample application in 80% ACN, elution using water\[[@CR29]\]SPEHILIC, DPA-6SRA, APTSSample application in 90% ACN, elution using water\[[@CR87]\]Gel filtrationSephadex G10RA, APTSGel filtration in 96-well format in filter plates\[[@CR38]\]Gel filtrationToyopearl HW-40FRA, APTS\[[@CR88]\]Gel filtrationSephadex G15RA, PA\[[@CR89], [@CR90]\]Liquid--liquid extractionWater/chloroformPermethylationPermethylated glycans are in the organic (lower) phase\[[@CR62], [@CR64], [@CR69], [@CR91]\]Liquid--liquid extractionWater/dichloromethanePermethylationPermethylated glycans are in the organic (lower) phase\[[@CR61]\]Liquid--liquid extractionWater/ethyl acetateHL, phenylhydrazineLabeled glycans are in the aqueous phase\[[@CR54]\]Liquid--liquid extractionWater/butyl etherMA, PMPLabeled glycans are in the aqueous phase\[[@CR51]\]Liquid--liquid extractionWater/chloroformMA, PMPLabeled glycans are in the aqueous phase\[[@CR52], [@CR78]\]Paper chromatography3 MM Whatman filter paperRA, 2-AB, DAPGlycans are eluted using water\[[@CR19], [@CR80]\]PrecipitationAcetoneRA, PAPrecipitation with water-free acetone was conducted 3 times\[[@CR75]\]*SPE* solid-phase extraction, *RP* reverse phase, *PGC* porous graphitized carbon, *HILIC* hydrophilic interaction liquid chromatography*, RA* reductive amination, *HL* hydrazide labeling, *BACH* 6-(biotinyl)-aminocaproyl hydrazide, *2-AB* 2-aminobenzamide, *2-AA* 2-aminobenzoic acid, *APTS* 1-aminopyrene-3,6,8-trisulfonic acid, *PA* 2-aminopyridine, *MA* Michael addition, *PMP* 1-phenyl-3-methyl-5-pyrazolone, *DAP* 2,6-diaminopyridine, *TFA* trifluoroacetic acid, *ACN* acetonitrile The introduction of hydrophobic properties by derivatization of native glycans (e.g., using the 2-AB label, see also later) is necessary for their adsorption to RP material, since nonderivatized glycans do not show sufficient retention on this material. PGC is most widely and originally applied for nonderivatized glycans \[[@CR81]\], but its use with labeled glycans has also been reported \[[@CR82], [@CR83]\]. Even though the mechanism is not well understood \[[@CR84]\], the optimized method is highly specific for glycans. Elution is usually performed with acetonitrile/water mixtures containing trifluoroacetic acid (Table [1](#Tab1){ref-type="table"}). The drawbacks of this method are the relatively high costs of the PGC and the fact that excess label may not be removed, depending on the properties of the label. An upcoming approach for purification of mainly reductively aminated glycans is HILIC \[[@CR29], [@CR33], [@CR34]\]. Several HILIC stationary phases have been described. In this approach, glycans are retained on the basis of their hydrophilic properties, whereas less hydrophilic excess label may be removed. We recently described the application of cellulose for purification of 2-AA-labeled glycans on a high-throughput platform \[[@CR34]\]. Since elution of the derivatized glycans can be performed with water, samples can be stored immediately without the risk of the acid-catalyzed hydrolysis of sialic acids. Glycan purification strategies may also include fractionation by anion-exchange chromatography of oligosaccharides labeled with PA or 2-AB, allowing the separation of neutral from sialylated glycans \[[@CR92], [@CR93]\]. Separation and detection {#Sec13} ======================== Electromigrative separation techniques {#Sec14} -------------------------------------- Electromigrative separation techniques are now frequently applied in glycan and carbohydrate analysis, providing high-speed separations and high resolution as has been summarized in a number of reviews \[[@CR39], [@CR94]--[@CR101]\]. The separations are based on differences in the effective electrophoretic mobility of the analytes in the separation medium, when an external voltage is applied. Whereas "acidic" glycans carrying sialic acid, glucuronic acid, sulfate, or phosphate are negatively charged at low pH (p*K*~a~ of sialic acids is 2.0--2.8 \[[@CR102]\]), "neutral" glycans will only exhibit partial negative charges at very high pH owing to the deprotonation of hydroxyl groups (see also "High-pH anion-exchange chromatography"). The electromigrative separation techniques applied so far include CE, micellar electrokinetic chromatography (MEKC), gel electrophoresis (also called "fluorescence-assisted carbohydrate electrophoresis), CGE, and capillary electrochromatography (see Table [2](#Tab2){ref-type="table"}). CE gives rise to a charge-based separation; therefore, sialoforms are relatively easy to separate with high resolution \[[@CR34]\]. An interesting approach for the separation of APTS-labeled glycans was presented by Zhuang et al. \[[@CR103]\]. The authors used chip technology and CE \[background electrolyte (BGE) 1 mM phosphate and 20 mM *N*-(2-hydroxyethyl)piperazine-*N′*-ethanesulfonic acid at pH 6.8\]. A very fast analysis (less than 3 min) was possible with high resolution (see Fig. [3](#Fig3){ref-type="fig"}). Further separation based on the size and the shape of the glycans needs a more careful optimization of the BGE, but can be achieved by optimizing the pH and the concentration of organic modifiers in the BGE \[[@CR104], [@CR105]\]. Alternatively, size-dependent separation can be introduced using MEKC or CGE, enabling the separation of structural isomers. Sometimes borate is used to further enhance the separation selectivity \[[@CR106], [@CR107]\]. A number of methods for 2-AA- and APTS-labeled glycans have been published, mainly including polyethylene glycol (molecular weight 20,000--300,000) as the sieving matrix \[[@CR36], [@CR108]--[@CR110]\]. Multiplexing using a DNA sequencer has been developed for high-throughput glycoprofiling of APTS-labeled glycans \[[@CR37], [@CR38], [@CR77]\]. The major disadvantage of CGE-based methods is their incompatibility with MS, so the identification of signals is complicated and can only be achieved by standards or additional experiments such as the reanalysis of the sample after exoglycosidase treatment \[[@CR36], [@CR37], [@CR110]\]. In recent years there has thus been an increase in the use of capillary electrochromatography for the analysis of glycans \[[@CR111]--[@CR117]\]. This technique allows charge-based separation driven by the external electric field, but also provides orthogonal chromatographic interaction for enhanced selectivity. Further growth of this approach can be anticipated, especially when it is used in combination with MS. Table 2Some typical methods and applications of electromigrative separation techniques for oligosaccharide analysisApplicationDerivatizationSeparation techniqueDetectionBackground electrolyteCommentsReferences*N*-Glycome mappingAPTSCGELIF 488 nmPOP-6 and POP-7 (Bio-Rad)Sequencer technology\[[@CR38]\]Glycoprotein and glycan standardsAPTSCGELIF 488 nmPolyacrylamide polymer^a^ in 89 mM Tris, 89 mM borate, 2.2 mM EDTASequencer technology\[[@CR37]\]Exoglycosidases and standards for identificationInfluenza A virus hemagglutininAPTSCGELIF 488 nmPOP-6 (Bio-Rad)Sequencer technology\[[@CR88]\]Glycoprotein standardsAPTSCGELIF 488 nm25 mM acetate pH 4.75 with 0.4% PEO^b^\[[@CR108]\]Recombinant monoclonal antibodyAPTSCGELIF 488 nmKit system based on gel electrophoresisCoated capillary, enzymatic digestions for identification\[[@CR135]\]Glucose ladder standardANTSCGELIF 325 nm25 mM ammonium acetate, 0--1% PEO (300,000)Neutral capillary coating\[[@CR136]\]Glucose ladder standard, RNase BAPTSCGELIF 488 nm25 mM ammonium acetate, 0-1% PEO^b^Neutral capillary coating\[[@CR137]\]FetuinAPTSCGELIF 488 nm25 mM ammonium acetate, 0.4% PEO (300,000)Neutral capillary coating, enzymatic digestions for identification\[[@CR138]\]α~1~-Acid glycoprotein2-AACGELIF 325 nm100 mM Tris--borate (pH 8.3), 10% PEG (70,000)Neutral (DB-1) capillary\[[@CR110]\]Linkage differentiationMonoclonal antibody2-AACGELIF 325 nm100 mM Tris--borate (pH 8.3), 10% PEG (35,000)Neutral (DB-1) capillary\[[@CR36]\]Linkage differentiationAntibody pharmaceuticals3-AA, APTSCGELIF 488 and 325 nm3-AA: 100 mM Tris--borate pH 8.3 with 10% PEG (70,000)Neutral (DB-1) capillary\[[@CR109]\]APTS: 50 mM Tris--acetate pH 7.0 with 0.5% PEG (70,000)Serum IgGAPTSCGELIF 488 nmPOP-7 gel bufferSequencer technology\[[@CR77]\]Clinical applicationFetuinAPTSCGELIF 488 nm25 mM phosphate pH 2.5 or 25 mM acetate pH 4.75 with 0--0.8% PEO (300,000)Neutral capillary coating\[[@CR139]\]Chondroitin sulfate type proteoglycans2-AminoacridoneCGELIF 488 nm100 mM Tris--borate pH 8.0 with 1% PEG (70,000) orUse of a neutral capillary coating\[[@CR140]\]2-AALIF 325 nm100 mM Tris--borate pH 8.3 with 10% PEG 70,000Chondroitine sulfate, trypsin inhibitor ulinastatinNoneMEKC, CGEUV 214 nm50 mM borate pH 9.3, 100 mM SDS orPartial use of a neutral capillary coating\[[@CR141]\]2-AminoacridoneLIF 488 nm100 mM Tris--borate pH 8.0 with 1% PEG (70,000) or2-AALIF 325 nm100 mM Tris--borate pH 8.3 with 10% PEG (70,000)RNase B, hen egg albumin, bovine serum fetuin2-AminoacridoneMEKCLIF 488 nm500 mM sodium borate pH 8.87, 80 mM taurodeoxycholate\[[@CR127]\]IgG2-AminoacridoneMEKCLIF 442 nm300 mM boric acid, 80 mM taurodeoxycholic acid, pH 9.2 (NaOH)\[[@CR129]\]RNase B2-AminoacridoneMEKCLIF 442 nm300 mM boric acid , 80 mM taurodeoxycholic acid, pH 9.2 (NaOH)\[[@CR128]\]Glycosaminoglycan disaccharides2-Aminoacridone, ANTSFACEFluorescenceFACE carbohydrate analysis kit from Prozyme (<http://www.prozyme.com>)\[[@CR142], [@CR143]\]RNase B, α~1~-acid glycoprotein2-ABMEKC and CD-CEUV 254 nm50 mM phosphate, 150 mM SDS, pH 6.7 or\[[@CR98]\]50 mM phosphate, pH 6.7 (NaOH/TEA), 4% SBE-γ-CDMaltoheptaose standard, RNase BAPTSCE and CGEOff-line MALDI; UV 254 nm10 mM ammonium actetate pH 4.75 orNeutral poly(vinyl alcohol) coating\[[@CR39]\]25 mM ammonium acetate pH 4.75 with 0.4% PEO^b^CE with fraction collectorFetuin, α~1~-acid glycoprotein, IgG, transferrin9-Fluorenylmethyl chloroformateCEMS50 mM ammonoium acetate\[[@CR130]\]Maltose ladder standardAminonaphthalene monosulfonic, disulfonic, and trisulfonic acidCEUV 235 nm50 mM triethylammonium phosphate, pH 2.5Reverse EOF\[[@CR144]\]High-mannose *N*-glycans RNase B, xyloglucan, acetylchitooligosaccharides6-Aminoquinoline, PACEUV 240 nm100 mM sodium phosphate, 50 mmM Bu~4~N^+^, pH 5.0Polyether coating\[[@CR131]\]Dextran ladder, monosaccharidesAPTSCELIF 488 nm100 or 200 mM borate pH 10.2 or 50 mM phosphate pH 2.2Linkage differentiation\[[@CR126]\]Monclonal antibody characterizationAPTSCELIF 488 nm and MS40 mM EACA pH 4.5 (HOAc)Hydroxypropylmethylcellulose or poly(vinyl alcohol) coating\[[@CR105]\]RNase B, asialofetuine, α~1~-acid glycoproteinAPTSChip-CELIF 488 nm1 mM phosphate, 20 mM HEPES, pH 6.8\[[@CR103]\]Cellobiohydrolase, RNase BAPTSCEMS25 mM ammonium acetateSeparation of phosphorylated isomers\[[@CR145]\]Dextran ladderANTSCEMS1% acetic acid/NH~4~OH, pH 3.4\[[@CR146]\]Standards, fetuin, ovalbuminANTSCELIF 325 nm50 mM phosphate pH 2.5\[[@CR147]\]Standards and beveragesNoneCEAmperometric100 mM NaOH\[[@CR123]\]Wine samplesNoneCEMS300 mM diethylamine\[[@CR122]\]Fetuin, α~1~-acid glycoprotein, erythropoietinNoneCEMS100 mM EACA, 900 mM NH~3~, 70% MeOHMS/MS characterization\[[@CR104]\]Disaccharide standardsNoneCEIndirect UV 400 nm; or direct UV 195 nm6 mM *p*-nitrophenol, 175 mM sodium borate pH 10.0Borate complexation\[[@CR118]\]Mono- and disaccharidesNoneCECMSAcetonitrile/water/ammonium formate mobile phase\[[@CR113]\]*O*-Glycans from bovine mucin, bile salt-stimulated lipaseNoneCECMS2.4 mM ammonium acetate pH 3, 0.2 mM sodium acetate in 50:50 acetonitrile/waterDesialylation prior to analysis\[[@CR117]\]RNase B, human milk bile salt-stimulated lipaseNoneCECMSAcetonitrile/water/ammonium formate mobile phaseMS/MS characterization\[[@CR114]\]RNase B2-ABCECLIF 325 nmAcetonitrile/water/ammonium formate mobile phase\[[@CR111]\]α~1~-Acid glycoprotein, ovalbumin2-ABCECUV 210 nmAcetonitrile/water/ammonium formate mobile phase, sulfated β-CD\[[@CR116]\]*ANTS* 2-aminonaphthalene trisulfonic acid, *GCE* capillary gel electrophoresis, *MEKC* micellar electrokinetic chromatography, *FACE* fluorescence-assisted carbohydrate electrophoresis, *CD* cyclodextrin, *CE* capillary electrophoresis, *LIF* laser-induced fluorescence, *MALDI* matrix-assisted laser desorption/ionization, *MS* mass spectrometry, *Tris* tris(hydroxymethyl)aminomethane, *PEO* poly(etheylene oxide), *PEG* poly(ethylene glycol), *3-AA* 3-aminobenzoic acid, *SDS* sodium dodecyl sulfate, *TEA* tetraethylammonium, *SBE* sulfobutyl ether, *EACA* ε-aminocaproic acid, *HEPES N*-(2-hydroxyethyl)piperazine-*N′*-ethanesulfonic acid, *EOF* electroosmotic flow^a^12% solution of acrylamide/bisacrylamide (19:1) polymerized with ammonium persulfate and *N*,*N*,*N′*,*N′*-tetramethylethylenediamine^b^Molecular weight not statedFig. 3Separation of 1-aminopyrene-3,6,8-trisulfonic acid labeled glycans on a microfluidic device in 1 mM phosphate and 20 mM *N*-(2-hydroxyethyl)piperazine-*N′*-ethanesulfonic acid pH 6.8. Glycans were released from a blood sample of a stage IV breast cancer patient. (Taken from \[[@CR103]\] with permission) Most work using electromigrative separation techniques has focused on the identification of glycans. However, CE has also been used to semiquantitatively compare glycan peak area ratios in CE-MS \[[@CR104]\]. Several strategies for glycan and oligosaccharide analysis using electromigrative separation techniques have been presented in the literature, differing in their way to achieve separation: *Direct analysis*: High-pH BGEs are applied to achieve the dissociation of the oligosaccharide hydroxyl groups to obtain electrophoretic mobility for the glycans. Alternatively, a focus on sialylated and phosphorylated glycans is possible; the neutral glycans are then only transported by electroosmotic flow and are not further separated. The analysis of nonlabeled glycans is possible using direct UV detection at a very low wavelength \[[@CR118]\] or indirect detection with appropriate probes added to the BGE \[[@CR118]--[@CR121]\]. Also MS \[[@CR104], [@CR122]\] and electrochemical detection \[[@CR123]\] can be applied.*Introduction of charged carriers*: Several metal acids have been shown to complex with vicinal diol groups mainly via *cis*-diols, but also via *trans*-diols \[[@CR106], [@CR107]\], and thus they introduce a charge on the glycan. Borate or boronic acids are mostly used \[[@CR124], [@CR125]\]. These complexing agents can also be added to the BGE at lower concentration to enhance resolution by changing the separation selectivity, when labels are applied \[[@CR36], [@CR109], [@CR110], [@CR126]\]. MEKC can in principle be used for the analysis of unlabeled glycans; however, owing to detection restrictions, it has mainly been used for the separation of labeled glycans (neutral labels were applied such as 2-aminoacridone \[[@CR98], [@CR127]--[@CR129]\] and 2-AB) by charged cyclodextrins as separation carriers \[[@CR98]\].*Introduction of labels*: Glycan labeling mainly has two functions in electrophoretic separation techniques: On the one hand, labels enable or facilitate fluorescence or MS detection; on the other hand, they enhance or in the case of neutral glycans allow the separation by introducing a charge on the glycans, thus allowing electrophoretic mobility. We can distinguish between neutral tags (only applied for detection purposes), cationic tags, and anionic tags. Neutral tags are only rarely applied for electromigrative separation techniques as the electrophoretic mobility has then to be achieved by other means; only sialylated glycans may be considered, or alternatively MEKC using charged micellar carriers may be applied. Examples are the use of 2-AB \[[@CR98]\] or 9-fluorenylmethylchloroformate on the amine function after release \[[@CR130]\]. Several cationic labels have been applied for the analysis of monosaccharides, such as 6-aminoquinoline \[[@CR131], [@CR132]\], dansylhydrazine \[[@CR133]\], benzoylhydrazine \[[@CR134]\], aminopyridine \[[@CR131]\], and aminoacridone \[[@CR127]--[@CR129]\]. Their use for complex glycans containing sialic acids is not recommended, as it would be difficult to have all glycans with an observed electrophoretic mobility towards the detector: neutral glycans (regarding only the glycan core) will be positively charged, whereas for sialylated glycans an overall cationic charge cannot always be achieved, leading to a mixture of cations and anions to be separated and transported towards the detector. Most work regarding the analysis of glycans with electromigrative separation techniques has thus been done using anionic labels, with APTS, ANTS, 3-aminobenzoic acid, and 2-AA most frequently applied (see Table [2](#Tab2){ref-type="table"}). APTS, with three charged sulfonic acid groups, can be used over a large pH range, whereas for the analysis of aminobenzoic acid labeled glycans containing sialic acids a more basic pH is recommended to achieve an overall negative charge for all glycans. Chromatographic separation {#Sec15} -------------------------- Chromatographic separation of glycans may be performed by adsorption chromatography, which is mostly done in the HPLC mode, by gel permeation chromatography (size-exclusion chromatography \[[@CR148]\]), or by (capillary) gas chromatography \[[@CR85], [@CR149]\]. The latter technique is used for the analysis of monosaccharides in glycoconjugate composition analysis \[[@CR149]\] or linkage analysis \[[@CR85]\]. This section is restricted to adsorption chromatography for glycan analysis, as most of the recent chromatographic developments in glycan analysis have occurred in this field. Moreover, the focus will be on HPLC separations, and other, mostly older adsorption chromatographic techniques will be omitted, such as paper chromatography and thin layer chromatography of monosaccharides combined with radioactive detection using reductive labeling of the reducing end with tritium \[[@CR150], [@CR151]\]. ### Hydrophilic interaction liquid chromatography {#Sec16} HPLC-mode HILIC is a widely applied technique for oligosaccharide separation \[[@CR152]--[@CR154]\]. The most commonly used stationary phases comprise amine-bonded silica, amide-bonded silica, and ZIC HILIC phases \[[@CR31], [@CR152]--[@CR154]\]. Also, diol phases are widely used \[[@CR155]\]. Hydrogen bonding is often the major retention mechanism. To be used in a pure HILIC mode (without anion exchange present), amine-bonded phases require rather high ionic strengths, which lead to the suppression of ionic interactions of acidic groups on the glycan such as sulfate and carboxyl groups of sialic acids with the ammonium groups of the stationary phase. Typically, volatile buffers such as acetic acid/triethylamine are applied at concentrations of up to 200 mM \[[@CR156]\]. The use of amine-bonded phases at lower ionic strength results in a mixed-mode HILIC/anion-exchange separation, which is discussed later. With ZIC-HILIC, where silica beads or polymer beads are functionalized with zwitterionic sulfobetaine groups, the retention properties of sialylated PA-labeled glycans change when 20 mM ammonium acetate is used instead of 5 mM ammonium acetate in the solvent, whereas the retention properties of neutral glycans are not influenced, pointing to a combination of ionic and nonionic interactions on these (mixed-mode) phases \[[@CR31]\]. With amide and diol phases, ionic interactions play only a minor role and are mainly caused by the remaining bare silica groups of the carrier material. Therefore, these phases may be used in pure HILIC mode also at low salt concentrations such as 50 mM ammonium formate, pH 4.4 \[[@CR29]\]. Typical equilibration conditions in HILIC HPLC of monosaccharides and oligosaccharides are 10--25% water in acetonitrile with a low concentration of acid or salt (mostly below 100 mM). Elution is accomplished by a gradient which increases the aqueous fraction of the mobile phase to typically 50%. To allow for sensitive detection in HILIC, glycans are almost always labeled with a chromophore or a fluorophore. The most commonly used labels are PA \[[@CR93]\], 2-AB \[[@CR29]\], and 2-AA \[[@CR34]\], but other tags such as 3-(acetylamino)-6-aminoacridine \[[@CR157]\] have also been used. As most of these labels have hydrophobic characteristics, the derivatized glycans show slightly less retention than native, unlabeled glycans. Fluorescence detection is mostly performed using detectors equipped with a xenon lamp and both excitation and emission monochromators can be set to the optimal wavelengths of the chosen label to achieve high sensitivity. HILIC is sometimes referred to as "size separation", owing to the correlation between glycan size and its retention time. However, the contribution of added monosaccharides to oligosaccharide retention may vary considerably, depending on the composition of the sugars of the glycan. Generally speaking, hexoses tend to give a larger (increase of) retention than *N*-acetylhexosamines of deoxyhexoses (fucoses) \[[@CR23], [@CR29]\]. HILIC also has a considerable potential to separate structural isomers \[[@CR158]--[@CR162]\]. For example, core-fucosylated biantennary monogalactosylated *N*-glycans which carry the galactose on the 6-antenna (isomer 1) or 3-antenna (isomer 2) are partially separated by HILIC after labeling with 3-(acetylamino)-6-aminoacridine \[[@CR157]\]. HILIC separations with fluorescence detection can be performed in a very robust manner. Fluctuations in retention times can be compensated by performing a standardization of elution properties using a glucose ladder (partial dextran hydrolysate). The time axis may then be normalized to an axis of glucose units (GUs), relating the oligosaccharide migration properties to the elution positions of these glucose oligomers \[[@CR29]\]. Rudd and coworkers have established a database (GlycoBase and autoGU) which allows the structural assignment of 2-AB-labeled oligosaccharides on the basis of migration positions on a TSK-amide-80 column. In this database, GUs are listed for 2-AB-labeled glycans before and after cleavage by exoglycosidases \[[@CR29], [@CR163]\]. Recently, the use of smaller (3 μm instead of 5 μm) amide-coated HILIC particles has led to a marked reduction of analysis times \[[@CR29], [@CR157]\], which makes HILIC in combination with fluorescence detection a suitable method for intermediate- to high-throughput glycosylation analysis \[[@CR29], [@CR164]\]. A detailed overview of glycosylation analysis by HILIC with mass-spectrometric detection has recently been provided \[[@CR152]\]. Mass-spectrometric detection of both labeled and unlabeled glycans is compatible with HILIC separation, particularly when amide or diol phases are used, owing to the low concentrations of acid or volatile salt combined with the relatively high percentages of acetonitrile, which is beneficial for ESI \[[@CR152], [@CR155]\]. To allow high-sensitivity detection, HILIC separation in combination with ESI-MS is often performed in miniaturized form down to the nanoscale \[[@CR152], [@CR159], [@CR160], [@CR165]\], allowing the detection of low femtomole amounts of glycan (absolute amount injected) in both native \[[@CR160]\] and 2-AB-labeled form (Fig. [4](#Fig4){ref-type="fig"}) \[[@CR159]\]. Similar to HILIC with fluorescence detection, HILIC with nano-ESI--ion trap--MS can make use of GUs when a 2-AB-labeled dextran ladder is applied as an internal standard. In this case, however, the primary information for glycan identification and structural elucidation is glycan mass and tandem MS data, and the GUs have a supporting role \[[@CR159]\]. Alternatively, HILIC fractions may be analyzed by with MALDI-TOF(/TOF)-MS(/MS) detection, which may be combined with high-energy fragmentation of sodium adducts \[[@CR161], [@CR162]\]. HILIC-MS is particularly useful for the in-depth analysis of complex oligosaccharide mixtures owing to its potential to separate glycans of different structures and identify isomers \[[@CR158]--[@CR162]\]. The proven usefulness of HILIC-MS for the analysis of complex oligosaccharide mixtures \[[@CR156]--[@CR160]\] will surely lead to its widespread application in the future. Fig. 4Subfemtomole sensitivity of nanoscale liquid chromatography (LC)--electrospray ionization (ESI)--mass spectrometry (MS) in the analysis of a 2-aminobenzamide (2-AB)-labeled hexamannosidic *N*-glycan (Man~6~GlcNAc~2~). A 2-AB-labeled hexamannosidic *N*-glycan was analyzed using nanoscale hydrophilic interaction LC (HILIC)--ESI--ion trap--MS. Base-peak chromatograms of a dilution series show that the detection of as little as 2 fmol of 2-AB-tagged Man~6~GlcNAc~2~ injected is possible (*A*). Sum mass spectra of the elution range of Man~6~GlcNAc~2~ (25.2--26 min; *B*--*E*) revealed the detection of proton adducts (*m*/*z* 1,517) and sodium adducts (*m*/*z* 1,539) down to 0.5 fmol of injected standard (*E*). Combined selected ion chromatograms for *m*/*z* 1,517 and *m*/*z* 1,539 allowed the detection of the standard after injection of a 2-fmol as well as a 0.5-fmol aliquot (*inset* in *A*). (Reproduced from \[[@CR159]\] with permission) ### Mixed-mode HILIC/anion exchange separation {#Sec17} Some authors have chosen to combine ionic interaction with hydrophilic interaction in mixed-mode HILIC/anion exchange separations of 2-AA-labeled glycans \[[@CR74], [@CR166]\]. These separations are performed on amine-bonded columns, which owing to (partial) protonation exhibit anion-exchange properties \[[@CR74]\], or on a strong anion exchange HPLC column functionalized with quaternary ammonium groups \[[@CR166]\]. The contribution of ionic interactions of both the tag and the glycan can be modulated by changing the ionic strength of the eluents (see "Hydrophilic interaction liquid chromatography"), whereas hydrophilic interactions (hydrogen bonding) with the stationary phase are influenced by the water content of the eluents. By carefully tuning the two interaction modes, optimized gradients result in the separation of glycan groups that differ in charge (Fig. [5](#Fig5){ref-type="fig"}). Within each charge group, separation is accomplished on the basis of HILIC. The separations obtained for glycans with different degrees of sialylation resemble somewhat those obtained by high-pH anion-exchange chromatography (HPAEC) of unlabeled glycans \[[@CR74]\]. Fig. 5Mixed-mode HILIC/anion exchange separation of α~1~-acid glycoprotein oligosaccharides before (**a**) and after (**b**) desialylation. *S1*−*S5* refer to the number of sialic acid residues present in the oligosaccharides. *Bi*, *Tri*, *Tri + F*, *Tetra*, and *Tetra + F* refer to biantennary, triantennary, fucosylated triantennary, tetraantennary, and fucosylated tetraantennary N-linked oligosaccharides, respectively. (Reproduced from \[[@CR166]\] from with permission) ### Anion-exchange chromatography {#Sec18} Many glycan fractionation schemes include preparative anion-exchange chromatography with fluorescence detection of glycans labeled with PA or 2-AB \[[@CR92], [@CR93]\]. Anion-exchange chromatography of glycans is typically performed at neutral to slightly alkaline pH. Under these conditions, the charge of most glycans is determined by the presence of carboxylic acids, sialic acids, glucoronic acids, sulfates, and phosphates. The separation is based on the (negative) charge of the glycans determined by these substituents. Elution is often accomplished by a salt gradient, which hinders the combination with mass-spectrometric detection. Anion-exchange chromatography may be followed by HILIC separation of the charged groups obtained after desialylation to the simplify elution patterns \[[@CR167]\]. ### High-pH anion-exchange chromatography {#Sec19} In contrast to classic anion-exchange chromatography, HPAEC is performed at a pH of approximately 13. Under these conditions, hydroxyl groups on glycans are partially deprotonated, resulting in a partial negative charge, which is then used for the separation of the glycans. Glycans with a reduced end exhibit less retention than reducing-end glycans, as they lack the anomeric hydroxyl group which is a major contributor to the partial negative charge. Usually electrochemical detection (pulsed amperometric detection, PAD) is applied for the analysis after HPAEC separation. HPAEC is suitable for the separation of nonderivatized glycans as well as glycans with various types of labels and modifications. Kotani and Takaski \[[@CR168]\] reported the analysis of N-linked oligosaccharides released from various glycoproteins labeled with 2-AB using HPAEC with PAD followed by in-line neutralization of the eluent with an anion micromembrane suppressor to allow for fluorescence detection. In their hands, high pH led to quenching of the fluorescence intensity, which made postcolumn neutralization necessary. Moreover, they observed a better chromatographic resolution for 2-AB-labeled asialooligosaccharides released from human α~1~-acid glycoprotein by HPAEC than by HILIC (Fig. [6](#Fig6){ref-type="fig"}). Fig. 6Analysis of N-linked oligosaccharides from human α~1~-acid glycoprotein. 2-AB-labeled oligosaccharides obtained were analyzed by high-pH anion-exchange chromatography with fluorescence detection (**a**) or high-performance LC (HPLC) with an amide-80 column and fluorescence detection (**b**). (Reproduced from \[[@CR168]\] with permission) Ridley et al. \[[@CR169]\] reported a method for biotin labeling of biologically active oligogalacturonides using a stable hydrazide linkage after first creating the water-instable hydrazone derivative. With use of HPAEC-PAD, optimization of the reaction conditions was monitored, product yields were determined, and a subsequent stability test was performed. It could be concluded that the yield of the oligogalacturonide-biotin hydrazide derivatives is dependent on the size of the oligogalacturonide. Stubbs et al. \[[@CR170]\] described a semipreparative purification of tetraantennary oligosaccharides from human α~1~-acid glycoprotein. Six *tert*-butoxycarbonyl-[l]{.smallcaps}-tyrosine oligosaccharide derivatives were purified and the subsequent labeling process was monitored by HPAEC-PAD. The chromatographic resolution of *tert*-butoxycarbonyl-[l]{.smallcaps}-tyrosine derivatives of the oligosaccharides with HPAEC-PAD was comparable to that of the free oligosaccharides. Closely related to this publication \[[@CR170]\] is the work by Collard et al. \[[@CR171]\], in which HPAEC-PAD was applied to monitor a time course of the elimination of the *tert*-butoxycarbonyl (BOC) group from a BOC-tyrosinamide-labeled triantennary *N*-glycan, followed by reductive alkylation/formylation resulting in dimethylation of the tyrosinamide moiety. ### PGC chromatography {#Sec20} PGC stationary phases are often used for the separation of oligosaccharide alditols, fluorescently labeled glycans, and even permethylated glycans \[[@CR66]\], as reviewed recently \[[@CR84]\]. Retention is mediated by a combination of predominantly ionic and hydrophobic interactions \[[@CR84]\], and elution is often achieved by a gradient of organic solvents with addition of small amounts of volatile salts, acids, or bases, making PGC of oligosaccharides compatible with mass-spectrometric detection \[[@CR84]\]. Both small and large oligosaccharides show retention on PGC. Moreover, glycans with a reducing-end tag as well as native glycans may be applied. PGC is the separation method of choice for direct liquid chromatography (LC)--MS analysis of *O*-glycans which have been released by reductive β-elimination, as these glycans are often rather small and do not contain a reducing end for labeling \[[@CR172], [@CR173]\]. Next to native or reduced glycans, fluorescently labeled glycans have often been analyzed using PGC \[[@CR84]\]. The fluorescent labels contribute to the retention of glycans on PGC: PA, which is a label with only modest retention in PGC, results in superb PGC separations of labeled glycans, whereas derivatization with a label such as 2-AB, which has pronounced influence on glycan retention on PGC, results in inferior resolution \[[@CR174]\]. Notably, PGC is very efficient in separating structural isomers. Retention times in combination with glycan mass analysis are often sufficient for structural assignment \[[@CR175], [@CR176]\]. ### RP chromatography {#Sec21} The aspect of hydrophobic interactions between (labeled) oligosaccharides and the stationary phase is common for both PGC and RP separations on C~18~ phases. In contrast to PGC separation, however, RP-HPLC of oligosaccharides requires derivatization, because the retention of underivatized glycans is often insufficient \[[@CR177]\]. For labeled glycans, the retention properties are determined by a combination of the hydrophobic properties of the tag and the contribution of monosaccharide residues \[[@CR75]\]. The effect of the addition of monosaccharides on the retention of a labeled glycan may differ depending on the tag: *N*-glycans labeled with PA, which is a low-retention tag, showed increasing retention with a higher degree of antenna galactosylation, whereas the same set of *N*-glycans labeled with 2-aminobenzoic acid ethyl ester showed decreasing retention with a higher degree of antenna galactosylation \[[@CR75]\]. For fluorescence detection in RP-HPLC, the most commonly applied tags are PA \[[@CR93], [@CR178]\] and 2-AB \[[@CR179], [@CR180]\]. Together with HILIC and anion-exchange chromatography, RP-HPLC with fluorescence detection is used in multidimensional mapping strategies for structural assignment of PA-labeled *N*-glycans \[[@CR93]\]. Next to fluorescence detection, RP-HPLC is often combined with online ESI-MS analysis, which has been shown for glycans labeled with various tags such as PMP, phenylhydrazine, and the biotin label BACH \[[@CR55], [@CR78], [@CR181], [@CR182]\]. Moreover, both 2-AB-labeled and 2-AA-labeled *N*-glycans can be analyzed by RP-nanoLC-MS/MS, and separations of structural isomers have been demonstrated \[[@CR83], [@CR179], [@CR180], [@CR183]\]. Chen and Flynn \[[@CR180]\] have impressively demonstrated the separation power of RP-HPLC of 2-AB-labeled glycans, and have combined this with both positive-mode and negative-mode multistage tandem MS on an ion trap mass spectrometer for structural assignment (Fig. [7](#Fig7){ref-type="fig"}). Fig. 7Reverse-phase HPLC with fluorescence detection of 2-AB-labeled glycans released from a 10 μg RNase B, b 30 μg ovalbumin, and c 30 μg fetuin. The structures of the labeled oligosaccharide peaks are shown in d. Species that were coeluted in the same fluorescence peak are indicated by letters *a* and *b*. The peaks corresponding to the species that were eluted early (less than 20 min) containing two or more sialic acids are not shown. *Circles* mannose, *squares* GlcNAc, *diamonds* galactose, *triangles* fucose, *stars* sialic acid, *dashed lines* α linkage, *solid lines* β linkage, *vertical lines* 1--2 linkage, *horizontal lines* 1--4 linkage, *forward slashes* 1--3 linkage, *backslashes* 1--6 linkage; *wavy lines* unknown linkage. (Reproduced from \[[@CR180]\] with permission) Alternatively, nanoscale RP-HPLC-MS/MS of labeled glycans may be performed without fluorescence detection. A major advantage of this approach is that the necessary analytical equipment is available in many proteomics and metabolomics laboratories: columns (nano-RP C~18~) as well as solvents (water/acetonitrile mixtures containing typically 0.1% of formic acid) which are used for the LC-MS analysis of complex peptide mixtures can be directly used for the analysis of 2-AB-labeled \[[@CR83], [@CR183]\] and 2-AA-labeled glycans (unpublished results) without any adjustment. Permethylated glycans may likewise be analyzed by RP-HPLC-MS. Structural isomers of permethylated glycans are separated using this approach \[[@CR184]\]. This method in particular has potential for in-depth structural characterization of glycans by tandem MS of sodium adducts, which has hitherto predominantly been applied without a preceding separation \[[@CR63]\]. A comparative study analyzing PMP- and ANTS-labeled glycans was performed by Saba et al. \[[@CR78]\], where the ANTS labeling was accomplished by reductive amination. Overall, PMP-modified oligosaccharides yielded better normal and RP separation and improved sensitivity than the ANTS-labeled sugars for LC-ESI-MS analysis \[[@CR78]\]. This approach could facilitate the structural investigation of various oligosaccharides in the future. Mass spectrometry {#Sec22} ----------------- ### Glycan ionization and detection {#Sec23} Both ESI-MS and MALDI-MS are often used for mass-spectrometric analysis of glycans. ESI of glycans is most often used in-line with HPLC or CE (see "Separation and detection"), which may be used for quantification in combination with (isotope) labeling (see "Comparative glycomics by MS"). Here, off-line ESI-MS and MALDI-MS of glycans are discussed with a focus on derivatization techniques. The ionization efficacies of oligosaccharides are markedly influenced by charged substituents of the glycan such as carboxyl groups of sialic acid or glucoronic acid and sulfate groups, which promote negative-mode ionization. Moreover, derivatization influences the ionization efficacies by introducing additional charges. Glycans with negatively charged substituents are more efficiently ionized in negative-ion mode than their neutral counterparts, whereas native as well as labeled glycans lacking acidic substituents are predominantly analyzed using positive-mode ionization \[[@CR185]\]. Notably, the influence of charged substituents on ionization efficacies is much more pronounced in MALDI than in ESI. In addition, derivatization often gives the oligosaccharide a more hydrophobic character, which usually increases ESI efficacies. Next to their pronounced effect on ionization, labels influence the fragmentation behavior of glycans and may help with the assignment of fragmentation spectra by tagging the reducing end. ### Off-line ESI-MS {#Sec24} Off-line ESI-MS of glycans is performed on both native and permethylated glycans. Off-line ESI-MS of native glycans is mostly applied for structural analysis by tandem MS and is possible in both negative-ion and positive-ion mode. In positive-ion mode, some sodium salt is usually added to the spray solvent consisting typically of a water/methanol mixture containing a volatile acid (formic acid or acetic acid) \[[@CR186]\]. The sodium ions suppress the formation of proton adducts. The resulting sodium adducts are more stable than proton adducts and less prone to in-source decay during ion transfer from atmospheric pressure to a vacuum. Thus, a better sensitivity and less complex structures are obtained. Moreover, whereas tandem MS of proton adducts by CID usually exclusively results in the fragmentation of glycosidic bonds, CID of sodium adducts additionally results in cross-ring cleavages providing linkage information \[[@CR186]\]. Negative-mode ionization and fragmentation has been applied by Pfenninger et al. \[[@CR187], [@CR188]\] and more recently by Harvey \[[@CR189]--[@CR191]\], who optimized the off-line ESI conditions, resulting in the registration of (multiple) deprotonated species and/or anion adducts (phosphate, nitrate, or chloride). Tandem MS of negatively charged oligosaccharide ion species usually provides specific cross-ring cleavages allowing structural elucidation. Notably, fragmentation of sodium adducts and deprotonated species of specific glycans result in very different patterns of cross-ring cleavages \[[@CR192]\]. Negative-mode tandem MS of oligosaccharides is characterized by a considerable stability of fucose glycosidic linkages, in contrast to positive-mode tandem MS, where fucose bonds are labile \[[@CR192]\]. The most commonly applied labeling technique for off-line positive-mode ESI is permethylation. Compared with native glycans, permethylated glycans generally give higher MS signals. Moreover, permethylation allows purification/desalting of the glycans by liquid--liquid extraction or RP-SPE (Table [1](#Tab1){ref-type="table"}). Tandem mass spectra of sodium as well as other alkali ion adducts of permethylated glycans provide very informative tandem MS spectra: the methyl substitution pattern allows the discrimination between terminal, internal, and branching monosaccharide residues. Cross-ring cleavages are extensive, providing linkage information. The mass spectrometer of choice is the ion trap, which allows multiple ion isolation/fragmentation cycles for in-depth structural characterization. The power of this approach was demonstrated by Prien et al. \[[@CR63]\] by characterizing isomers of RNase B *N*-glycans, some of which were described for the first time. ### Matrix-assisted laser desorption/ionization MS {#Sec25} In contrast to ESI-MS, which is often used online with separation techniques, MALDI-MS of glycans is predominantly used as a stand-alone technique. MALDI-MS of native glycans is mostly performed in positive-ion mode using 2,5-dihydroxybenzoic acid (DHB) as the matrix substance, resulting in the registration of sodium adducts. Negative-mode detection of neutral native glycans, in contrast, is difficult to achieve and may require specific matrices and additives \[[@CR192]\]. Permethylation is often used in combination with MALDI-TOF-MS detection, resulting in the very sensitive detection of glycans, mostly as sodium adducts \[[@CR22], [@CR72], [@CR85], [@CR193]\]. Permethylation transforms the carboxyl groups of sialic acid residues into methyl esters, which are far less labile than the underivatized sialic acid groups, thereby allowing the MALDI-TOF-MS analysis of glycans without loss of sialic acids by in-source and metastable decay \[[@CR22]\]. Another labeling method which is used in combination with MALDI-TOF-MS analysis of glycans is the derivatization with carboxymethyl trimethylammonium hydrazide (Girard's reagent T) \[[@CR60]\]. This label induces a permanent positive charge at the reducing end and thereby allows very efficient generation of glycan ions. Labeling of glycans by reductive amination has several consequences for MALDI: the (secondary) aromatic amine group introduced in this reaction (see Fig. [2a](#Fig2){ref-type="fig"}) makes glycans labeled with PA, 2-AB, and 2-AA readily protonable. Thus, MALDI-MS of these glycans often results in the detection of proton adducts next to sodium adducts. With regard to negative-mode MALDI, the 2-AA tag---owing to its carboxyl group---makes neutral glycan structures readily ionizable, thereby allowing the joint registration of sialylated and nonsialylated glycans in one MALDI spectrum \[[@CR34]\]. When sialylated glycans are analyzed, reflectron-mode MALDI-TOF-MS often shows partial or complete loss of sialic acids, which is due to in-source decay and metastable decay. Loss of sialic acids is somewhat less pronounced in negative-ion mode. Moreover, a reduction of decay problems may be achieved by applying alternative matrices such as 2,4,6-trihydroxyacetophenone, which is colder than DHB \[[@CR194]\]. In contrast, a hot matrix such as α-cyano-4-hydroxycinnamic acid leads to a more pronounced desialylation. Notably, loss of sialic acid is hardly seen when MALDI-TOF-MS is performed in linear mode \[[@CR34]\]: metastable decay still occurs under these conditions, but the resulting fragments are not visible. Acceleration starts immediately after ionization/desorption, and fragment ions arising thereafter in the field-free region will not be resolved. Although MALDI-TOF-MS instruments generally require a source pressure in the range of 10^-6^ mbar, MALDI with other MS detectors may alternatively be performed at intermediate pressures (typically submillibar range). Under such conditions, the loss of sialic acids is suppressed: using DHB as a matrix, intermediate-pressure MALDI--Fourier transform ion cyclotron resonance (FTICR)--MS allows the high-resolution analysis of both sialylated and nonsialylated, 2-AA-labeled *N*-glycans from human plasma with virtually no in-source decay (Fig. [8](#Fig8){ref-type="fig"}). Our results show that loss of sialic acids is avoided by collisional cooling of ions in the ion source at intermediate-pressure conditions. Moreover, compared with MALDI-TOF-MS, relatively low extraction voltages are applied in the ion source. Fig. 8Analysis of 2-AA-labeled total plasma *N*-glycans by matrix-assisted laser desorption/ionization (MALDI)--Fourier transform ion cyclotron resonance (FTICR)--MS. Samples were prepared as described previously \[[@CR34]\], desalted by porous graphitized carbon solid-phase extraction, and analyzed by MALDI-FTICR-MS using a 2,5-dihydroxybenzoic acid matrix. The low-mass range (*top*) and the high-mass range (*bottom*) were measured using different ion-transfer times. The *inset* shows the isotope pattern of registered glycan species. Yellow circles *galactose*, *green circles* mannose, *blue squares N*-acetylglucosamine, *purple diamonds* sialic acid, *red triangles* fucose ### Comparative glycomics by MS {#Sec26} Most often, relative quantification of glycoforms within one sample is achieved using MS (with or without preceding separation) or by LC or CE with fluorescence detection (see "Separation and detection"). Comparison of a large number of glycan samples by mass-spectrometric analysis may be compromised by changes in the performance of the mass spectrometer over time, and also because of the large number of sample preparation steps necessary, possibly introducing unwanted bias, e.g., loss of sialic acids. Especially the labeling step itself can be problematic with respect to labeling efficiency, interference by matrix components, or the subsequent purification step. We have recently established a high-throughput method for glycosylation profiling of the total plasma *N*-glycome by negative linear-mode MALDI-TOF-MS after glycan release and 2-AA labeling. Similarly, N-glycosylation profiling can be performed with permethylated glycans, which has the advantage that sialic acid groups are stabilized by methyl ester formation \[[@CR22], [@CR69]--[@CR71]\]. In proteomics approaches, labeling with isotopic or isobaric tags is often employed to obtain good quantification results \[[@CR195]\]. Some research groups have now started to incorporate the idea of isotopic \[[@CR40]--[@CR44], [@CR196]--[@CR198]\] as well as isobaric \[[@CR61], [@CR91]\] labeling also for glycomics analysis (see also "Reductive amination"). Isotopic derivatives share the same atomic composition, but have different masses owing to the introduction of heavy isotopes, most often D and ^13^C. Several reports of use of this approach for comparative glycomics have been published: Many of the isotopic derivatives, such as \[H~4~\]PA versus \[D~4~\]PA \[[@CR40]\], \[H~4~\]2-AA versus \[D~4~\]2-AA \[[@CR41]\], and \[^12^C~6~\]alanine versus \[^13^C~6~\]alanine \[[@CR43], [@CR44]\], are introduced by reductive amination (see "Reductive amination"). The PA-labeled monosaccharides were analyzed by PGC-LC-MS, and the observed relative standard deviations (RSDs) were less than 7.2% \[[@CR40]\]. Derivatization of a dextran ladder with aniline and subsequent analysis using HILIC-LC-MS resulted in RSDs below 10% for Hex~3~-Hex~5~. Larger oligosaccharides had larger RSDs \[[@CR44]\]. In a second study, an RSD of 4.85% was observed for lacto-*N*-neotetraose derivatized with aniline followed by HILIC-LC-MS analysis with a large dynamic range \[[@CR43]\]. The incorporation of an isotopic tag by permethylation has been described by Alvarez-Manilla et al. \[[@CR196]\]. Oligosaccharides are permethylated using either ^12^\[C\]H~3~I or ^13^\[C\]H~3~I, resulting in a 1 Da mass difference per permethylation site. Derivatization of Hex~3~-Hex~5~, followed by MALDI-TOF-MS analysis revealed an average RSD of 12.7% over a linear dynamic range of 2 orders of magnitude. Differentially labeled glycans from α~1~-acid glycoprotein were mixed 1:1, resulting in an average RSD for seven glycoforms of 6%. However, these approaches only allow for the comparison of one sample pair at a time, in contrast to proteomics labeling techniques such as iTRAQ enabling the comparison of up to eight samples \[[@CR199]\]. Recently, Bowman and Zaia \[[@CR197]\] synthesized different tags on the basis of 2-aminobenzoic acid applicable to reductive amination, which are suitable for the direct comparison of four samples which, based on the incorporation of deuterium versus hydrogen atoms, give rise to an isotopic variety with a mass difference of 4 Da between the glycans from different samples. Upon analysis by ESI--quadrupole TOF--MS, the RSDs were below 10% for compounds with a ratio of more than 1:3 between the differentially labeled glycans. These tags are, however, not compatible with LC separation, owing to the large number of deuterium atoms, resulting in retention time shifts. The previously described labeling techniques are all based on the same principle: introduction of an isotopic tag in vitro. Recently, Orlando et al. \[[@CR198]\] reported on the use of an in vivo metabolic isotopic labeling approach using aminosugars containing ^15^N-glutamine. All GlcNAc, GalNAc, and sialic acids become labeled with ^15^N, resulting in a mass increase of 1 Da. With use of this approach, differentially labeled cell glycans can be mixed prior to all sample preparation procedures, thus reducing the risk of bias due to sample handling. Isobaric derivatives have the same nominal mass, whereas the accurate mass is different. An example is the derivatization via permethylation using either ^13^CH~3~I or ^12^CH~2~DI, resulting in a mass difference (Δ*m*) of 0.003 Da per permethylation site as applied. This approach, called "quantification by isobaric labeling", has been described recently for comparative glycomics of both *N*-glycans and *O*-glycans \[[@CR61], [@CR91]\]. All glycans have multiple methylation sites and, owing to the multiplication of the mass-shift, mass differences for the discrimination of differentially labeled glycan pairs to be analyzed using high-resolution mass spectrometers (e.g., Fourier transform MS) are obtained. Microarrays {#Sec27} ----------- Although most of the methods discussed in this review deal with glycan labeling and analysis for identification, purification, and structural characterization, this section is dedicated to labeling approaches which combine structural characterization with biomolecular interaction analysis. In glycan binding assays using microarrays and microtitration plates, glycans are often immobilized, which may be achieved in a covalent or noncovalent manner. Carbohydrate-binding proteins (CBPs) such as animal or plant lectins and immunoglobulins are then applied. Fluorescence detection of attached CBPs is achieved by either directly labeling the CBPs or using fluorescently labeled secondary antibodies. For noncovalent immobilization, Liu et al. \[[@CR200]\] have established two types of neoglycolipid probes for glycan microarrays: glycans may be conjugated to 1,2-dihexadecyl-*sn*-glycero-3-phosphoethanolamine via reductive amination or to aminooxy-functionalized 1,2-dihexadecyl-*sn*-glycero-3-phosphoethanolamine via oxime ligation. Neoglycolipid probes are purified by silica chromatography, structurally characterized by MS, and used for preparing glycan microarrays. Several biotin-containing tags are available for glycan labeling (see "Reductive amination" and "Hydrazide labeling") offering the possibility to perform microtitration-plate binding assays \[[@CR57], [@CR58]\] as well as glycan microarray analyses combined with surface plasmon resonance detection \[[@CR201]\]. BACH-labeled glycans have recently been shown to be suitable for detailed structural analysis by LC-MS \[[@CR55]\], which, together with the aforementioned binding assays, offers the possibility to identify natural ligands of CBPs. Xia et al. \[[@CR80]\] introduced 2,6-diaminopyridine, which carries two aromatic amine groups, as a versatile tag. After reductive amination, the free amine groups can be further coupled to any kind of *N*-hydroxysuccinimide-activated reagent, allowing immobilization on array surfaces or biotin labeling \[[@CR80]\]. We have recently shown that glycans isolated from natural sources and labeled with fluorescent tags such as 2-AA and 2-AB using reductive amination are not only suitable for many separation techniques with fluorescence and mass-spectrometric detection (see "[Separation methods](#Sec30){ref-type="sec"}"), but can also directly be used for covalent immobilization on a natural glycan microarray \[[@CR202]\] (Fig. [9](#Fig9){ref-type="fig"}). The key to this approach is the reactivity of epoxy-activated slides which react with the secondary amine group linking the fluorescent tag with the oligosaccharide moiety, thereby allowing the oriented immobilization of fluorescently labeled glycans on microarray slides (Fig. [9d](#Fig9){ref-type="fig"}). Microarrays generated in this manner have been probed for interaction with CBPs by both fluorescence detection \[[@CR202]\] and surface plasmon resonance \[[@CR203]\]. Fig. 9A natural glycan microarray approach with reductively aminated glycans. **a** Glycans are derivatized, **b** fractionated by HPLC, **c** analyzed by MALDI time of flight MS(/MS), **d** immobilized on microarray epoxide slides, **e** assayed for protein interaction, and **f** the data obtained are interpreted. (Reproduced from \[[@CR202]\] with permission) Discussion {#Sec28} ========== In this section we will discuss some aspects regarding the choice of a glycan label such as the introduction of changes in glycan properties upon derivatization, sample preparation, and requirements for the analytical method. Sample preparation {#Sec29} ------------------ Next to detection and separation properties of a specific label, the derivatization process together with possible matrix effects also have to be taken into account when choosing a label. Some labels require optimized reaction conditions as the equilibrium is on the side of the nonlabeled glycans and the free label. This holds true for the common labeling scheme for 2-AB, which requires organic solvents as the reaction solution, and for APTS, for which only poor labeling yields may be achieved \[[@CR75]\]. Most analytical techniques require a purification step after sample derivatization. MALDI and HPAEC need samples with low amounts of salts, which is also the case for some electromigrative separation techniques. Acetone purification as well as paper chromatography leave the sample with quite some matrix compounds and salts; therefore, SPE methods may be preferred. The removal of excess label is taken into account in most strategies, as problems with overload, column contamination, or quenching of signals (in MS) can occur. Hydrophilic labels are generally difficult to remove from the glycans, as their properties resemble more closely those of the glycans. An example is APTS, which is difficult to remove using HILIC and RP-SPE. Very hydrophobic labels most likely require easy-to-use RP-SPE; however, they are not compatible with many subsequent separation techniques. In addition, introduction of hydrophobic properties to the glycans can give rise to interferences with proteins and peptides in both purification and separation as these biomolecules usually show a higher hydrophobicity than the glycans. As the mechanism of binding on PGC-SPE is not fully understood, the optimization is dominated by trial and error as the adsorption and the loading capacity cannot easily be judged. In general, with both PGC-SPE and HILIC-SPE, both labeled and unlabeled glycans are retained and can be analyzed with appropriate analytical techniques. This is not the case in RP-SPE, where the label dominates binding and unlabeled glycans are usually not retained. In recent years there has been an onset of commercialization of purification strategies. More effort in this field can be expected in the coming years. Separation methods {#Sec30} ------------------ Most applications use separation techniques to gain information on the glycan content and to identify the glycans. Charge and hydrophobicity are major criteria in choosing a particular label. Charge plays a major role in several separation techniques, such as electromigrative separation techniques, anion-exchange chromatography, mixed-mode anion exchange/HILIC separations, and HPAEC. In CE-based separations the charge is necessary to provide the glycan with an effective electrophoretic mobility for migration. Glycan labeling with APTS and ANTS is often applied for CE: with their three sulfonic acid groups, both labels provide a nearly pH independent high anionic charge, giving rise to low analysis times. Laser-induced-fluorescence detection is applied at 488 and 325 nm, respectively \[[@CR105], [@CR108], [@CR109], [@CR126], [@CR147]\]. Good detection limits can also be obtained by mass-spectrometric detection by taking advantage of the highly sensitive and selective negative ionization mode and the high negative charge of the labeled glycans \[[@CR105]\]. Good results can also be obtained using 2-AA-labeled glycans with either laser-induced-fluorescence (325 or 355 nm) or ESI-MS detection in negative ionization mode using CE or CGE (see Table [2](#Tab2){ref-type="table"}). HPAEC separation of glycans requires a (partial) negative charge, which may be achieved by partial deprotonation of glycan hydroxyl groups at high pH. Under these conditions, a charge-based separation of native glycans as well as glycans derivatized with a neutral label can be achieved. The introduction of labels providing an additional negative charge leads to an increase in retention times, but a change or enhancement in selectivity is not to be expected. Glycan separation using HILIC is dominated by the physical properties of the glycan moiety. The label often has little influence on the separation. HILIC can be used as an SPE method for the purification of glycans after labeling, as small aromatic molecules such as 2-AA and 2-AB are hardly retained \[[@CR34]\]. Notably, the APTS label, which owing to the three sulfonic acid groups has pronounced hydrophilic properties, leads to longer migration times of labeled glycans in HILIC compared with native glycans (unpublished results). In RP-HPLC native glycans are hardly retained, and retention is largely determined by the label. Separation of 2-AB-labeled *N*-glycans has been demonstrated with resolution of structural isomers and detection by (tandem) MS, showing the potential of RP-HPLC \[[@CR180]\]. Separation by PGC somehow resembles RP-HPLC, yet with significant retention of native glycans and charge effects contributing to retention. Summarizing, there are a few labels that are suitable for a large number of different applications. 2-AB, for example, has been shown to be applicable for MALDI, HILIC, RP-LC, PGC separation, and HPAEC. It gives rise to reasonable signal intensities in ESI and can be used for fluorescence detection. 2-AA shows a somewhat similar broad applicability; it can, however, also be applied in electromigrative separation techniques as it introduces an anionic charge at intermediate to high pH. Moreover, in contrast to 2-AB, it can be used for MALDI-TOF-MS analysis of both neutral and sialylated glycans in negative ionization mode \[[@CR34]\]. However, complex mixtures of *N*-glycans tend to be less efficiently separated by RP-HPLC in 2-AA-labeled form than in 2-AB-labeled form (unpublished observations). Regarding the determination of labeling yields, those methods which allow the simultaneous separation and detection of nonlabeled and labeled glycans have to be considered. MS-based detection techniques as well as PAD are the few ones available. For separation, HPAEC, PGC and HILIC can be used. CE with mass-spectrometric detection will allow the analysis of sialylated glycans with and without a reducing end label, but not of the nonlabeled neutral glycans. Optical detection {#Sec31} ----------------- Several glycan labels are suitable for UV detection, for example, 2-AA, with an absorption maximum at 214 nm \[[@CR204]\], and PMP, with an absorption maximum at 245 nm \[[@CR205]\]. However, as many other biomolecules likewise absorb UV light, the specificity is generally low, and alternative detection techniques are usually preferred also to achieve better sensitivity. Fluorescence detection is the most commonly used optical detection method for glycan analysis. Whereas lamp-based fluorescence can be applied in chromatographic techniques, CE usually requires lasers as excitation sources owing to the short optical path length for detection in the capillary. Therefore, fluorescent labels have to be chosen which exhibit excitation/emission properties that match the wavelengths of common lasers, such as the argon laser at 488 nm (APTS), the He--Cd laser at 325 nm (2-AA, 2-AB), and the frequency-tripled Nd:YAG laser at 355 nm (2-AA, 2-AB) (see also Table [2](#Tab2){ref-type="table"}). Amperometric detection {#Sec32} ---------------------- Amperometric detection is almost exclusively used in combination with HPAEC \[[@CR206]\]. It enables the analysis of both labeled and unlabeled glycans. However, this detection method is not very selective, and compounds from the sample matrix such as amino acids also give rise to signals. With use of a membrane desalter, capillary-format HPAEC with amperometric detection was successfully combined with on-line ESI-MS for glycan characterization \[[@CR207]\]. Mass-spectrometric detection {#Sec33} ---------------------------- Mass-spectrometric detection of glycans by both ESI-MS and MALDI-MS is usually enhanced by glycan labeling compared with the analysis of unlabeled glycans. Both the hydrophobicity and the charge brought in by the label can be advantageous. Some tags are particularly suitable for detection in the positive ionization mode (e.g., PMP, 2-AB); others are preferably used with negative ionization mode (e.g., ANTS, APTS). Other labels such as 2-AA have been shown to give rise to acceptable signal intensities in both ionization modes \[[@CR75]\]. Tandem MS of glycans as alkali adducts or in deprotonated form by (high-energy) CID is particularly valuable for structural characterization owing to the observed cross-ring cleavages which provide linkage information \[[@CR8], [@CR9], [@CR22], [@CR192]\]. Obviously, extensive cross-ring cleavages often lead to rather complex spectra, making spectral assignment challenging \[[@CR161], [@CR162]\]. Sialic acids {#Sec34} ------------ The analysis of sialylated glycans is still a challenge. Many researchers decide not to record the degree of sialylation and use neuraminidases to remove the sialic acids from the glycans \[[@CR38]\]. This largely simplifies glycan patterns, and at the same time increases the repeatability of the method. If the determination of sialoforms is of interest, labeling and purification conditions should be chosen that reduce the risk of sialic acid hydrolysis: steps including low pH and high temperature and especially the combination of both have to be avoided. Users might consider labeling strategies other than acidic reductive amination, e.g., PMP labeling, to reduce the risk of hydrolysis. In addition, the acidic conditions under which storage takes place might catalyze hydrolysis. Loss of sialic acids also occurs when glycans are stored on MALDI plates with DHB as the matrix (unpublished results). Permethylation can largely stabilize sialic acids; however, the choice of the subsequent analytical techniques is then reduced owing to the increase in hydrophobicity of the labeled glycans. Loss of sialic acids during LC or CE separations is usually not a problem, even when very acidic conditions are applied. However, in combination with heating of the separation system, loss of sialic acid might become noticeable. Detection with MS using ESI or MALDI can likewise give rise to loss of sialic acids by in-source decay or metastable decay. Using a separation method capable of separating sialylated from nonsialylated glycoforms is advantageous, as it allows one to discriminate on the basis of retention/migration times between sialylated glycan species which are present in the sample and those species which are formed by desialylation during MS. Quantification and profiling {#Sec35} ---------------------------- Relative quantification of labeled glycans by mass-spectrometric analysis has been shown to yield consistent results \[[@CR34], [@CR193]\]. One important aspect to take into consideration is the occurrence of in-source decay, where mainly sialic acids are lost, but also the loss of full antennae can be observed. The use of specific ion sources such as intermediate-pressure MALDI may successfully minimize in-source decay \[[@CR185]\], whereas sialic acids may also be stabilized using permethylation \[[@CR22]\]. In MALDI, DHB is widely used as the matrix. DHB usually results in macrocrystalline sample preparations. Recrystallization using ethanol leads to smaller, more equally distributed crystals, which enhances shot-to-shot reproducibility and is therefore advantageous for quantification, also in combination with automatic measurements. Next to relative quantitation, absolute quantitation with internal standards is possible for comparative glycomics. For this purpose, stable isotope labels may be used in combination with MS (see "Comparative glycomics by MS"). The use of isotope-labeled PA or 2-AA may be easy to implement in laboratories where the nonisotopic derivatives are already in use \[[@CR40], [@CR41]\]. Unfortunately, deuterium may result in retention shifts in chromatography, and Δ*m* between glycan pairs for these labels is only 4 Da. Both disadvantages may be overcome by the use of aniline, which is labeled by ^13^C, and generates Δ*m* of 6 Da \[[@CR43], [@CR44]\]. Aniline derivatives cannot be detected on the basis of fluorescence, but only by UV detection. However, as absolute quantitation using isotope-labeled internal standards will rely on mass-spectrometric detection, the lack of fluorescence detection should not represent a major drawback. As an alternative, glycans may be isotope-labeled by permethylation \[[@CR70], [@CR71], [@CR196]\]. Again, this will be easy in laboratories where permethylation is already employed; however, since the number of permethylation sites may differ per glycan, the Δ*m* generated is also different per glycan. When this method is applied to compare complex mixtures, it will result in rather complex mass spectra. Recently an attempt to implement four-plex isotope labeling has been published \[[@CR197]\]. Unfortunately, mass shifts are generated by the introduction of deuterium, which makes the use of chromatography difficult. Moreover, the labels are not commercially available yet. As a counterpart to the stable isotope labelling by amino acids in cell culture method, which is widely applied in proteomics, ^15^N metabolic labeling of oligosaccharides via glutamate may be employed \[[@CR198]\]. Since only 1 Da mass difference is generated per GlcNAc, GalNAc, and sialic acid residue, the overall Δ*m* per glycan is often quite small. Moreover, this method may be applied in cell cultures, but is not suitable for the comparison of patient material. The use of isobaric labeling has only very recently been introduced to glycomics \[[@CR61], [@CR91]\]. With use of this approach, pairs of labeled glycans are detected at the same nominal mass. As only very small mass differences are generated, the use of high-resolution MS, which is not available in every laboratory, is necessary. All of these methods have not been used in larger sample sets yet, and we expect further labels to be developed, as well as commercialization of such tags. L.R.R. was supported by IOP Grant IGE05007, G.Z. by the Netherlands Genomics Initiative (Horizon Breakthrough Project 93518016), and C.H. by a fellowship within the Postdoc-Program of the German Academic Exchange Service (DAAD). **Open Access** This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.

Summary ======= Aortic pulse wave velocity is altered by normal aging, as well as vessel wall pathology. We evaluated aortic compliance and its relationship to age in normals and patients with heart failure. Background ========== Aortic compliance (AC) can be evaluated noninvasively and its reduction with aortic pathology and age in normals has been demonstrated with both MRI and Doppler echo methods. Aortic pulse wave velocity (PWV), a measurement of the flow pulse traveling along aorta as a surrogate for AC, can be assessed using a single breath-hold phase contrast (PC) imaging technique. Congestive heart failure is often associated with a chronic cardiac remodeling process in which the myocardium either cannot eject blood very well (systolic heart failure); or the myocardium is stiff and ventricular chambers do not fill with blood easily (diastolic heart failure). We hypothesize that aortic stiffness is increased in the CHF population and its age dependency differs from that in normals. Methods ======= As normal controls, 196 healthy volunteers gave informed consent (96 male, age: 61.3±14.2) and were screened to exclude hypertension, hyperlipidemia and cardiovascular disease. Twenty two CHF patients had been referred for clinical CMR study with LVEF \< 55%. Using the 'candy cane' view of aorta, an axial plane through the ascending and descending aorta at the pulmonary artery level was prescribed and a through-plane velocity encoded PC cine imaging was acquired with VENC of 150 cm/s, TR/TE/FA = 98ms/2.9ms/15° and voxel spatial resolution 1.3×2×6 mm3 on a 1.5T MRI scanner. The distance traveled by the aortic pulse wave, ΔD, was determined as the distance along the center line between the axial sections as imaged in the 'candy cane' image. For flow pulse onset, the cross correlation between the first halves of the ascending and descending aortic flow curves was calculated by varying the relative time shift between them. The Δt was the time shift at the maximal correlation. We then calculated PWV=ΔD/Δt. Linear regression was used to determine the relationships between PWV and age at both groups. Results ======= PWV in CHF and in normals correlate with age, as shown in the Figure as left graph and right graph of scatter plots, respectively. The linear regression in normals is: PWV (m/s) = -5.548 + 0.2601\*Age; while in CHF patients: PWV (m/s) = -1.277 + 0.1879\*age; However, the relationship is stronger in normals. Young and elderly patients with CHF had lower PWV than normals, likely due to reduced stroke volume and blood pressure in the CHF group. {#F1} {#F2} Conclusions =========== Aortic stiffness increases with age in CHF patients, but the slope seems to differ from CHF patients to normals. Funding ======= None.

Introduction ============ Pseudoaneurysm (PA) or false aneurysm is a relatively rare but threatening clinical disease \[[@B1]\]. PAs are always the results of trauma \[[@B2]\], inflammation \[[@B3]\] or iatrogenic procedures \[[@B4],[@B5]\], presenting with pulsatile masses, compressive feelings or hemorrhage as the most notable manifestations \[[@B1]\]. We report a case of a patient troubled with fever of unknown origin (FUO) that was solely due to the spontaneous formation of a huge PA in the bifurcation of his left common iliac artery. Combined endovascular managements using coils, balloon and stent graft were successfully performed. Case presentation ================= A 79-year-old Chinese man was admitted to our hospital complaining of episodes of repeated fever and some uncertain discomfort for 10 days. He had a nonproductive cough, but no chill, pharyngalgia, arthralgia, chest pain, dyspnea or diarrhea. He denied experience of trauma or drug abuse. He previously had been well but had a medical history of hypercholesterolemia and a long duration of two packs/day cigarette smoking. His father died from a sudden heart attack and he also had two brothers suffering from hypertension. On physical examination, he was 69 inches (1.73m) tall and weighed 178lb (81kg). He was febrile with a heart rate of 90 beats per minute, blood pressure 135/85mmHg, and respiratory rate 22 breaths per minute. Head and neck examinations were significant for pale conjunctivae and a left carotid bruit. His chest was clear to auscultation. His heart rhythm was regular with a nondisplaced apical impulse, an S4 gallop, no murmurs or rubs. His abdomen was benign, with no tenderness or masses. He had symmetrical palpable pulses in the femoral, popliteal and pedal arteries bilaterally without changes of skin appearance, temperature or sensations. Laboratory assessments revealed a high level of white blood cell count (20.7×10^9^/L), neutrophils (85%), C-reactive protein (132mg/L), erythrocyte sedimentation rate (45mm/hour) and a low level of hemoglobin (94g/L). A lipid level test showed total cholesterol, triglycerides, high-density lipoprotein and low-density lipoprotein were 342, 280, 35 and 263mg/dL respectively. Liver and kidney functions, urine and stool tests were normal. A chest X-ray showed no recent infections. Conventional antibiotics using cefoxitin and levofloxacin were given for 10 days in the out-patient department and 4 days in our hospital for suspected sepsis, but his temperature still fluctuated and mostly remained above 38°C. However, bacteriological cultures of blood, urine, and sputum were negative. Immune markers like complements, interleukins, self-antibodies and globulins showed no abnormalities. Tumor and human immunodeficiency virus biomarkers were also negative. However, a brief scanning of computed tomography (CT) and magnetic resonance imaging (MRI) were done to search for the focus of the fever and suggested edemas and abnormalities around his left greater psoas and peritoneum, suggesting inflammatory changes. Ceftizoxime and clindamycin were regulated to use on advice from microbiologists and kept for another 2 weeks, but still without any conspicuous relief of symptoms. Based on the ambiguity of the CT and MRI done earlier, a contrast-enhanced computed tomographic angiography (\[CTA\], Figures [1](#F1){ref-type="fig"}a and [1](#F1){ref-type="fig"}b) was performed and demonstrated the existence of an 81.9mm×61.6mm PA, which originated from his left iliac arteries and covered the bifurcation of his left common iliac artery and proximal ends of both his internal and external iliac arteries. {#F1} The endovascular procedure was performed 1 day after the CTA. Heparin 4000IU was administered intra-arterially once access had been obtained. A 6-F artery sheath punctured his contralateral femoral artery. Then a 5-F Rösch inferior mesenteric catheter (Cordis) and 0.035 3mm "J" standard guide wire (Cook) were used together to enter his left common iliac artery. Hand injections of contrast demonstrated the existence of a huge PA (Figure [2](#F2){ref-type="fig"}a) and also identified both the rupture of PA and the proximal end of his left internal iliac artery (Figure [2](#F2){ref-type="fig"}b). Another catheter and guide wire were inserted through an ipsilateral femoral artery puncture. Via the catheter passway, metal coils (Cook: 8mm diameter, Figure [2](#F2){ref-type="fig"}c) were released and fixed at the distal end of his left internal iliac artery. Then the left femoral artery was exposed and a 14-F artery catheter (Cordis) was inserted. A 14mm×7cm Bard self-expandable polytetrafluoroethylene-covered Fluency stent (Boston Scientific) was placed after dilating with an 8mm×4cm angioplasty balloon. A selective left iliac arteriography (Figure [2](#F2){ref-type="fig"}d) 5 minutes after deployment of the stent graft confirmed complete exclusion of the PA and normal flow to the arteries of his ipsilateral lower extremity. {#F2} The postprocedural course went well. The patient received subcutaneous low-molecular-weight heparin, urokinase and prophylactic cefminox. Furthermore, he was given long-term medications for hypercholesterolemia with atorvastatin, aspirin and omega 3 cod liver oil. Initially there was eradication of fever. His subsequent clinical course was uneventful and he was discharged 7 days after the endovascular treatment. CTA performed after 30 days showed a well-expanded stent with no signs of a new PA (Figure [3](#F3){ref-type="fig"}) and a decreasing and consolidated residual hematoma. He received oral administration of warfarin for 3-month anticoagulation, maintaining an international normalized ratio of 2.0 to 3.0. At 1.5-year clinical follow-up, his recovery was good. {#F3} Conclusions =========== Isolated PAs are always secondary to blunt trauma \[[@B2]\], inflammation \[[@B3]\], and iatrogenic procedures such as organ transplantation \[[@B4]\], interventional procedures \[[@B5]\] or neoplasia \[[@B6]\] that cause laceration of part of the vessel wall and extravasation of blood into surrounding tissues, followed by tamponade and clot formation. Typically, PAs present with pulsatile masses, compressive symptoms, secondary hemorrhage and neurologic deficit as the most common clinical manifestations. However, even spontaneous PAs or PAs without clear origins are clinically rare; they have been reported to occur in the facial artery \[[@B7]\], tibioperoneal trunk and anterior tibial artery \[[@B6]\], lumbar artery \[[@B8]\] and superficial femoral artery \[[@B9]\], presenting as pulsatile masses in the former two cases and as ruptured hemorrhage in the latter two respectively. FUO is defined as a temperature higher than 38.3°C on several occasions and lasting longer than 3 weeks, with a diagnosis that remains uncertain after 1 week of investigation. There are well over 200 different reported causes of FUO and five major categories of conditions: infections, neoplasms, connective tissue diseases, miscellaneous disorders and undiagnosed conditions \[[@B10],[@B11]\]. To the best of our knowledge, spontaneous formations of PAs originating from iliac arteries with atypical clinical manifestations are extremely rare. The mechanism of this case may be correlated with iliac arteriosclerosis. Underlying brittleness of the arterial wall due to extensive lesions of calcification (Figures [1](#F1){ref-type="fig"}a and [1](#F1){ref-type="fig"}b) both in the abdominal aorta and bilateral iliac arteries made them predisposed to rupture. Some cases of PAs were reported to be associated with the process of calcification or atherosclerosis \[[@B12]-[@B17]\]. Other than images, the patient's high body mass index, cigarette smoking, family history of heart attack and medical history of hypercholesterolemia all contribute to risk factors for arteriosclerosis. Besides, in this case the deep location and pelvic surroundings made the PA present without a remarkable pulsatile mass or ruptured hemorrhage. As for treatments of PAs, in recent years, several interventional techniques have been developed to achieve occlusion of PAs, including ultrasound-guided compression, direct percutaneous thrombin injection, transcatheter embolization with coils, glue, gelfoam or sclerosing agents and stent graft placement, showing a shorter hospitalization time, more favorable success rates and minimal morbidity \[[@B1]\]. Although traditional opening surgery may be necessary in some instances, the pendulum has now swung to endovascular therapy \[[@B2]\]. In this case, the PA is a complex condition as some episodes of giant size, wide neck and bifurcation lesion existed simultaneously. Considering those situations, we tend to opt for combined endovascular managements sequentially. Embolization of his internal artery with coils ensured no recurrence of blood flow into the PA, and then occlusion of the ruptured PA and support of the vessel that supplied blood to his lower extremity with a self-expandable covered stent graft were performed successfully. The result of a 1.5-year follow-up confirmed the combined treatment to be effective and safe. Consent ======= Written informed consent was obtained from the patient for publication of this case report and any accompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal. Abbreviations ============= CT: Computed tomography; CTA: Computed tomographic angiography; FUO: Fever of unknown origin; MRI: Magnetic resonance imaging; PA: Pseudoaneurysm. Competing interests =================== The authors declare that they have no competing interests. Authors' contributions ====================== SL participated in the design of the study and wrote the paper. DJH participated in the treatment and helped write the paper. KG participated in the treatment. YWX conceived of the study, and participated in its design and coordination and helped to draft the manuscript. All authors read and approved the final manuscript. Acknowledgments =============== The authors gratefully acknowledge the assistance of Qing Xia of the Department of Cardiology of Shanghai Tenth People's Hospital for help in writing and language review.

1.. Introduction {#s1} ================ Despite a wealth of clinical and health services research, there is a well-recognized *"failure to translate research into practice and policy"* [@b1]. This failure has created a growing and significant interest in the field of implementation science (IS) which is defined as "*the scientific study of methods to promote the systematic uptake of research findings and other evidence-based practices into routine practice, and hence, to improve the quality and effectiveness of health services"* [@b2]. Over the last decade, IS has yielded an array of theories and frameworks to help clarify which factors influence the implementation and sustainability of evidence-based practice in health care [@b3],[@b4]. Given the extensive theoretical work in the field of IS, we asked how theoretical developments in IS might be of value to population health intervention research (PHIR) which supports the development and uptake of evidence-based policies and programs aimed at improving population level conditions of risk and reducing health inequities. Our research, therefore, sought to explore, if, and how key IS theories and frameworks might inform the development, implementation and sustainability of population health interventions [@b5]. This is not to suggest that PHIR is atheoretical [@b6],[@b7]; but rather the intent was to explore if recent theoretical developments in the field of IS might be of value to researchers and decision makers working within population health. 1.1.. Implementation science and population health {#s1a} -------------------------------------------------- There are similarities across the fields of IS and population health. IS investigates the best approaches to move research into practice "*to improve the quality and effectiveness of health services"* [@b2] focusing mostly on changing healthcare professional and organizational behavior [@b2],[@b8],[@b9]. A population health approach focuses on improving the health of the entire population through action toward determinants that contribute to inequalities among population groups [@b10]. Examples of population health interventions include affordable housing policies, immunization programs, and tobacco taxation, and PHIR aims to contribute relevant, credible, and timely evidence to enable decision makers to develop and implement policies and programs at the population-level [@b5]. In both, success in achieving intended impacts relies on evidence that is generated through meaningful and explicit engagement with the population of interest (e.g., patients/community) and the recognition of the importance of context to ensure interventions are relevant for individuals/patients, providers/practitioners, organizations and communities [@b11],[@b12]. Further, just as IS supports the need to develop and draw from relevant theories and theoretical constructs, PHIR supports the application of appropriate theories to ensure that evidence-based interventions reach whole populations/communities [@b13],[@b14]. Population health interventions are often directed at large groups or populations, targeting large-scale policies or entire communities, and focus on social-cultural factors and environmental conditions [@b5],[@b15]. Interventions that are aimed specifically at improving the social determinants of health are also often fundamental considerations in a population health approach [@b12],[@b16], including interventions that target key determinants such as income, housing, social exclusion, and food insecurity [@b17]. Considering these similarities and differences between IS and population health, the purpose of this synthesis was to critically assess the potential relevance of existing theories and frameworks developed in the field of IS to population health interventions. 2.. Methods {#s2} =========== A critical interpretive synthesis approach was used to identify key IS theories and frameworks, and to explore if, and how, they might be applicable to population health interventions [@b18]. For the purposes of this research, "theories" were defined as a set of analytical principles or statements with explanatory power [@b3],[@b19],[@b20], and "frameworks" as a structure without explanatory power that includes descriptive categories outlining various constructs that may explain the relations between them [@b3],[@b21]. The steps outlined in a recent scoping review of knowledge synthesis methods [@b18] were used to inform the operationalization of the review (see [Figure 1](#publichealth-05-01-013-g001){ref-type="fig"}). {#publichealth-05-01-013-g001} The critical interpretive synthesis method offered an iterative approach to refine the research question, search and select articles, and define and apply codes and categories [@b18]. A core review team (JLM, GW, LL, RU, and a research assistant) conducted the primary review of relevant theories, and the larger authorship team (SP, JG, MM, LJ) reviewed the analysis, contributed feedback to preliminary results and revised the manuscript. A number of strategies were used to collect literature on IS theories and frameworks, which is supported by the emergent and exploratory nature of the review method [@b18]. Drawing on existing comprehensive research, the search began with an examination of two recent reviews of IS theoretical approaches that provided a comprehensive synthesis of current IS theories and frameworks [@b3],[@b4]. An initial list of theories was developed using the more recent review by Nilsen [@b3] using their categories of "determinant frameworks" and "implementation theories" as these categories emphasized theoretical explanations to implementation. To validate and expand the initial list, theories categorized as having a focus on "implementation" (i.e., putting evidence-based interventions into practice/policy) in the review by Tabak et al. [@b4] were also included. These theories were determined to be relevant for this review due to their focus on theoretical constructs to inform implementation. Thirty-one theories and frameworks were initially identified. Theories cited by others more than 1000 times in three bibliographic databases (Scopus, Web of Science, Google Scholar) provided the basis for further narrowing the list to those considered "seminal" or "formative" for the field of IS. Ten theories and frameworks were identified for inclusion through this process, and consensus on their strength of conceptual support for implementation and merit for the review was obtained from the authorship team. A standardized data extraction form was developed and used to identify theory development and evolution, key theoretical components and mechanisms, and other descriptive features for each of the ten theories ([Supplementary Table](#s01){ref-type="supplementary-material"}). The theories were also assessed using a data extraction process developed a priori to appraise if and how they might inform PHIR. The data extraction process was guided by three questions, and coding definitions (see [Table 1](#publichealth-05-01-013-t01){ref-type="table"}). Data were extracted by the core review team, with at least two reviewers supporting extraction and verification of each theory. The review team appraised the resulting data extraction table following the secondary review to ensure consistency in the results. The information extracted enabled the review and authorship teams to critically interpret the results, in an iterative manner, by identifying emerging similarities and differences across the theories and frameworks in terms of their relevance to inform population health interventions. Although not the main focus of this study, a secondary search within the published literature was completed to further understand where the final 10 theories/frameworks to population health were being utilized. Using each theory\'s foundational paper and applying population health search terms based on the Population Health Search Strategy developed by the United States National Library of Medicine [@b23], a search of relevant bibliographic databases (MEDLINE, Embase, CINAHL, Web of Science) was conducted. The results were screened by two reviewers (JLM, MM); examples of theories applied to population health research were identified (see [Table 2](#publichealth-05-01-013-t02){ref-type="table"}). ###### Data extraction questions, focus and coding developed for the review. -------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Key questions Could the theory help to inform interventions seeking to address populations (e.g., change conditions of risk?) Could the theory help to inform interventions seeking to address social determinants of health? Could application of the theory generate evidence to inform policy/practice change at a population level? Key focus Population health provides insights into the needs of populations, rather than the needs of individuals [@b22]. Population health interventions are intended to change the social-cultural and environmental conditions of risk to support the overall health of populations and reduce health inequities [@b5],[@b15]. The social determinants of health influence the health of populations. Although variously defined, one definition includes income and social status; social support networks; education; employment/working conditions; social environments; physical environments; personal health practices and coping skills; healthy child development; gender; and culture [@b16]. Population health research is aimed at generating relevant, contextually sensitive, credible and timely knowledge to enable decision makers to use evidence to improve policies and programs that prevent disease at a population level [@b5]. Coding definitions Yes, if macro (e.g., societal, policy) and community-level factors are considered, including a focus on populations and systems. Possibly, if remains focused on more micro (e.g., individual) or organizational levels. Yes, if social determinants of health are considered and integrated into components of the theory/framework. Possibly, if social determinants of health are not explicitly considered in the components of the theory/framework. Yes, if application of theory/framework could result in useful information to inform policy/practice change. Possibly, if application of theory/framework focused more on research utilization and does not seem likely to be useful for policy/practice change. -------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ 3.. Results {#s3} =========== A comprehensive description of the development of all ten theories is available in the [Supplementary Table](#s01){ref-type="supplementary-material"}; the results of the critical appraisal are available in [Table 2](#publichealth-05-01-013-t02){ref-type="table"}. ###### Results of the critical interpretive synthesis based on a priori data extraction criteria. Theory Could the theory help to inform interventions seeking to address populations (e.g., change conditions of risk?) Could the theory help to inform interventions seeking to address social determinants of health? Could application of the theory generate evidence to inform policy/practice change at population level? Examples of theory application in population health identified through secondary review. ----------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Absorptive Capacity (ACAP) (Zahra & George, 2002) Possibly; remains focused on organizational level. Does explore larger systems factors, with the exception of structural, behavioural or political "social integration mechanisms". Possibly; could potentially be taken into consideration upon application of the theory through the activation triggers and social integration mechanisms. Yes; could provide information on what is contributing to the success of an organization and what can improve its success. One identified related to knowledge brokering in the health sector [@b24]. Active Implementation Framework (Fixsen et al. 2005) Possibly; consistent focus on the community level but does not fully consider populations or systems. Possibly; could be applied to implementation in areas that address SDH (e.g. education). Yes; but the framework focuses more on community-level implementation. Several identified related to sexual health [@b25], school-based positive behavioral supports [@b26], and HIV [@b27]. Consolidated Framework for Implementation Research (Damschroder et al. 2009) Yes; approaches implementation through a multilevel lens and recognizes factors existing at multiple levels of the system. Possibly; does consider the importance of the organization knowing and prioritizing patient needs Yes; could help to explore practice and policy change as it relates to implementing new programs. Many identified related to various population health issues, such as tobacco cessation [@b28], child mental health [@b29] and nutrition [@b30], oral health [@b31], HPV vaccine [@b32]. Diffusion of Innovations for Service Organizations (Greenhalgh et al. 2004) Yes; although focused primarily at the organizational level, considers system-level factors that affect diffusion, dissemination and implementation of innovations. Possibly; sub-factors are associated with the innovation so could be adapted to address SDH. Yes; provides a wide variety of factors that could be explored to generate evidence for practice and policy change. Two identified related to HIV testing [@b33] and public health policy [@b34]. Ecological Framework (Durlak & Dupre, 2008) Yes; takes a multilevel ecological approach by considering individuals in the context of their environments. Yes; characteristics of the innovation consider community needs, values and cultural norms and examples of SDH interventions provided. Yes; relevant factors described that could become levers for changes to policy and practice. None identified. Implementation Effectiveness Model (Klein & Sorra, 1996) Possibly; focused at organizational level and does not explicitly address populations. Possibly; does consider how climate within an organization influences implementation. Possibly; it may generate evidence at the organizational level, this would not be widely applicable at a population level. One identified related to mental health amongst low-income women and health care practitioners [@b35]. Multilevel Change Framework (Ferlie & Shortell, 2001) Yes; multi-level and highlights the importance of the legal, political, and economic environment. Possibly; however, factors focused on health care quality. Yes; system-wide changes are considered in the scope of multilevel change and could provide useful evidence to inform practice/policy change. Two identified related to HIV testing [@b36] and quality improvement in public health [@b37]. Promoting Action on Research Implementation in Health Services (Kitson et al., 1998) Possibly; conceived as an organizational framework for health care and does not explicitly address populations. Possibly; SDH could be considered as part of the context where the evidence is being implemented. Yes; could generate evidence to inform practice change within particular settings but may have fewer implications for policy change. Several identified but mostly related to health care settings: Oral health in home care setting [@b38]; community-based mental health care [@b39]; community-based health for people with disabilities [@b40]. Sticky Knowledge (Szulanski, 1996). Possibly; focused more on the organizational level, but could be adapted to help inform the transfer of knowledge to populations. Possibly; developed within a relatively narrowly-defined organizational context. Yes; could be used to identify different barriers to implementing policies/programs, but developed in a business context may limit transferability to a larger or more diverse setting. None identified. Theoretical Domains Framework (Michie et al., 2005) & Behaviour Change Wheel (Michie et al., 2011). Yes; although TDF remains largely focused on individuals; the policy categories of the BCW provide an explicit focus at the level of populations. Possibly; the TDF does consider the role of resources and environment on behaviour change and the policy categories of the BCW could help to focus interventions relevant to SDH. Yes; use of BCW could inform policy and practice, but the focus of the TDF on practice change suggests its influence on policy may be limited. Many identified related to a various population health issues, such as tobacco cessation [@b41], childhood nutrition [@b42] and physical activity [@b43], oral health [@b44]. None of the theories provided an explicit or comprehensive framework for how to address population health interventions. However, six theories had underlying constructs/principles that were considered applicable to population health: The Ecological Framework [@b45], the Active Implementation Framework [@b46], Diffusion of Innovations for Service Organizations [@b47], Multi-level Change Framework [@b48], the Consolidated Framework for Implementation Research [@b49], and the Theoretical Domains Framework/Behaviour Change Wheel [@b50],[@b51]. These six theories considered influences beyond the organizational level, such as macro or system-level contexts (e.g., societal, policy and community factors) that are relevant for interventions targeting populations. Further, although none of these theories explicitly referred to key social determinants of health or population-level policies and practices, their identified constructs could be of value to PHIR. In comparison, the remaining four theories appear to require significant adaptation or expansion to be applicable to population health given their emphasis on the organizational level within healthcare services (Promoting Action on Research Implementation in Health Services framework [@b52]) or the business/management sector (Absorptive Capacity [@b53], Implementation Effectiveness Model [@b54] and Sticky Knowledge [@b55]). Although organizational level considerations are an important component of population health, these frameworks did not provide explicit consideration of broader community, policy or societal factors that play central roles in population health interventions. 3.1.. Theories appraised as relevant for population health {#s3a} ---------------------------------------------------------- Durlak and DuPre\'s Ecological Framework was identified as one theory that did, to some extent, refer to populations and changing the conditions of risk, which is likely due to its origin in community-based health promotion and disease/substance prevention settings [@b45]. Based on their review of 542 relevant studies (including 5 meta-analyses), the authors concluded that five categories of influencing factors are critical to implementation, and provide detail about these factors. The central two categories are the innovation delivery system (e.g., organizational capacity), and the innovation support system (e.g., training, technical assistance), which represent an organization\'s capacity for effective implementation. Their ecological lens posits that successful implementation is dependent on factors in three other categories: Innovation characteristics (e.g., compatibility with existing practice, adaptability to new contexts or needs); characteristics of those providing the innovation (e.g., perceived need, benefit, delivery skills); broader community factors (e.g., current state of the research, available funding, policies, politics/ideological climate). These five categories are interrelated and were also dependent and may change based on different local contexts. The Ecological Framework provides relevant core components for population health that could be helpful in generating evidence for policy and practice change (e.g., guidance for strengthening intervention delivery and support systems). Also the Ecological Framework constructs were similar to constructs identified in other IS theories in this review, including factors such as training [@b46],[@b47], managerial support, shared vision and organization norms regarding change [@b47]. However, there were limitations noted as the framework was not based on a systematic review of the literature, nor have its components been subsequently tested and refined. There were no examples of the Ecological Framework applied to population health research identified through the secondary review of literature. The Active Implementation Framework [@b46] considers evidence from multiple fields with the aim of supporting interventions related to mental health, social services, justice, education and early childhood. The importance of community involvement and assessing community readiness for change is a focus of the theory, which is detailed in the framework\'s essential components (source, destination, communication link, feedback mechanism, sphere of influence) and relevant for population health interventions. However, the authors\' reference to essential implementation outcomes remains focused on changes in individual behavior, relationships and organizational structure. Although these outcomes are sometimes important to changing the conditions of risk in a population, the emphasis on individual and organizational characteristics may limit its applicability to population-level change. There is some recognition of system-level partnerships and the stages of implementation that suggest appropriate times to deliver interventions, but the Active Implementation Framework would require further details on the operationalization of the components if it was to inform implementation of evidence into policy and practice, particularly among changing population-level circumstances with dynamic or unsupportive "sphere of influences" (e.g., social, economic or political factors that influence people, organizations and systems). There were several applications of this framework identified to population health interventions [@b25]--[@b27] in the secondary review of the literature. The four additional theories deemed as relevant for population health originated from health services and care delivery but their recognition of external influences and explanation for implementation offer important insights for population health. For example, Greenhalgh and colleague\'s framework for Diffusion of Innovations for Service Organizations is focused primarily at health service organizations yet its comprehensive description of factors and determinants relevant for the development, coordination and delivery of interventions offers insight into how an intervention might achieve broader population level change through its assimilation into larger systems [@b47]. The "Outer Context" domain in this theory would also be relevant for large-scale initiatives in population health (e.g., sociopolitical climate and interorganizational networks) and the domain of "System Readiness for Change" could help to articulate readiness for change across communities or populations. Decision makers could also use information generated from this theory to identify/evaluate factors that affect the spread of population health interventions, but taking into account the multitude of factors, it may be complex for decision makers to operationalize this framework in a way that informs the generation of evidence for policy and practice change. Congruently, there were only two examples identified of the application of this theory in population health [@b33]. The Multilevel Change Framework [@b48] focuses on quality improvement in health care by advancing previous work [@b56] and captures multiple levels of change for improving quality (individual, groups/team, organization, larger system/environment). In an effort to identify factors that might influence adaptation and associated properties, the resulting framework includes essential core properties to support change, such as leadership, organizational culture, microsystem/team development, information technology to inform continuous improvement and accountability. The emphasis on intervening, or at least considering, all levels of change in interventions suggest that it could provide useful information to inform population-level policy and practice, especially as it considers the impact of legal, political and economic environments. However, although the authors examined various quality strategies to develop the framework, the review methods were not systematic which may limit the comprehensiveness of factors influencing implementation. This theory was not commonly applied in the population health literature, with only two examples identified through the secondary search [@b36],[@b37]. The Consolidated Framework for Implementation Research (CFIR) was create to synthesize existing implementation theories such as Greenhalgh et al. [@b47],[@b57] framework as well as eighteen other models and frameworks. CFIR conceptualizes implementation from a multilevel lens. The framework identifies influences from "Outer" and "Inner Settings" that consider external policy and cultural factors, acknowledging that these factors exist at multiple levels of a system. CFIR also considers the importance of socio-political conditions or circumstances that can influence implementation. This is relevant to the context of population health (e.g., nature and quality of social networks both inside and outside the system, governmental policy and regulations) and could help to inform an understanding of the reasons for successful policy and practice change initiatives [@b49]. However, it was noted that the explanations for these various external/system factors were somewhat superficial, suggesting that further clarification is needed to inform its usefulness in population health. One example of this is the construct "Patient Needs & Resources" in the "Outer Setting" domain. This construct could be clarified and expanded to represent a more inclusive definition of the population-level influences on implementation processes, including the values, preferences, and circumstances of individuals within communities. The secondary literature review identified numerous examples of the application of CFIR to a range of population health interventions (e.g., [@b28]--[@b32]). Finally, although originating from health psychology and using the lens of individual behavior change, the Theoretical Domains Framework (TDF) [@b50] and the subsequent development of the Behavior Change Wheel (BCW) [@b51],[@b58] provides relevant reflections for population health interventions when considered as interconnected theories (i.e., considering the BCW as an extension of TDF). On its own, the TDF is used to identify individual-level determinants that impede or support individual behavior change, although the domains of "Social Influences" and "Environmental Context and Resources" are relevant for population health research. The strength of the TDF/BCW lies in their combined use: When the population health issue is viewed as a behavioral issue (e.g., individuals within a particular population not seeking cancer screening), the TDF can guide a needs assessment to identify the main barriers to change. Once barriers are identified, they are situated within the "COM-B" (Capability, Opportunity, Motivation, Behavior) system in the centre of the BCW. Depending on where the barriers lie, they can then be linked to evidence-based "Intervention Functions" (e.g., education, persuasion, training, environmental restructuring) within the BCW, which is based on previous research on addictions [@b59]. In turn, these intervention functions can be linked to specific "Policy Categories" of the BCW (e.g., fiscal measures, social planning, legislation), making it the only theory we reviewed that provided clear guidance on how to target individual-level health behaviors at the population level. Thus, information generated by use of the BCW could be used to inform population level policy and practice through application of "Intervention Functions" and "Policy Categories". The secondary literature review identified numerous population health interventions that applied this framework [@b41]--[@b44]. 3.2.. Theories appraised as requiring adaptation to inform population health {#s3b} ---------------------------------------------------------------------------- The remaining four theories included in the review were determined to have less relevant components for population health and would require significant adaptations or expansions [@b52]--[@b55]. Common among these theories was their limited consideration of macro-level factors, often focusing only at the organizational level. Although organizational-level influences are important considerations for population health programs, it is important that these are considered in the broader context of communities and populations. Similar to the theories described in the previous section, the social determinants of health were not specifically considered. The Promoting Action on Research Implementation in Health Services framework [@b52] was conceived as an organizational framework in health care, and as a result the core components of "Evidence", "Context" and "Facilitation" remain tightly focused on health services organizational settings, making it unclear how the framework might be of value for interventions targeting populations. Absorptive Capacity [@b53], Implementation Effectiveness Model [@b54] and Sticky Knowledge [@b55] were all developed from the field of management and primarily focused on organizational change. These four frameworks may be able to generate evidence to inform practice change within particular settings (e.g., determinants of implementation and what leaders/teams might do to facilitate implementation in their contexts), but they appear to have fewer implications for policy change. Components of these theories could be adapted to improve their relevancy; for example, although the definition of context in Sticky Knowledge referred more to organizational characteristics, this could be adapted for contextual considerations at the population level [@b55]. Application of these theories could generate evidence to inform change at the organizational level (e.g., considerations of "Competitive Advantage" from Absorptive Capacity [@b53] and "Innovation/Implementation Effectiveness" from the Implementation Effectiveness Model [@b54], but as they were developed from a business/management perspective, these factors may not be widely applicable at a broad population health level. The secondary review of literature identified few examples of the application of these theories to inform population health intervention. One example was identified from both Absorptive Capacity [@b24] and Implementation Effectiveness [@b35] and several from the Promoting Action on Research Implementation in Health Services framework, although these latter examples were primarily oriented toward health care [@b38]--[@b40]. 4.. Discussion {#s4} ============== Given the growing burden of chronic and infectious diseases globally, there is a need for researchers and decision makers to identify effective and efficient interventions that can be scaled-up improve population health. Theories and frameworks can help guide population health interventions by providing a comprehensive explanation of how multi-level factors affect successful implementation. The purpose of this review was to explore if and how key IS theories and frameworks might be applicable to interventions aimed at improving population level conditions of risk and reducing health inequities. It is evident from our findings that IS theories and frameworks could help to identify potential constructs or components to inform implementation strategies for population health interventions. However, there were some limitations to their application when considering the importance of factors beyond the organizational level, particularly at the macro level which is a level that often influences policies and programs targeting populations. Broadly defined, context shapes research utilization in policy and program development [@b60] and is relevant to both IS and PHIR. Although all of the reviewed theories/frameworks identified various contextual elements as being important, they often limited their focus to the organizational-level and did not provide a comprehensive description of the community, and socio-cultural and political contexts that need to be considered in PHIR. Population health interventions need to not only consider the socio-demographic and health characteristics of the population (e.g., age, disease status, education level) but also the characteristics of communities (e.g., rurality, culture, resource availability), as these can all affect implementation and sustainability. The potential levels targeted by the intervention are also important for various reasons not the least of which is because it influences how researchers work with relevant stakeholders to design, implement, evaluate and scale-up evidence-based interventions [@b61]. For example, community-level characteristics may also be relevant for some health service interventions such as primary healthcare, but these factors are a core consideration in population health interventions to ensure relevancy to the targeted population. Several reviewed theories/frameworks provided some considerations for interventions at a broader community-level, including specific mention of factors at this level in Durlak and Durpre\'s framework through politics, funding and policy. This could be used to consider how, for example, funding for public health interventions shapes the feasibility of implementation strategies (e.g., was the reason for deciding on a given implementation strategy guided primarily by budget constraints at the expense of effectiveness?). Other theories/frameworks deemed as relevant also included some mention of community-level intervention, such as CFIR\'s "outer" and "inter" constructs (e.g., external policies and incentives, culture) and the policy categories in the BCW (e.g., fiscal measures, legislation) but the information was not as detailed. Similarly, these external factors can help explore implementation strategies, such as how closely they align with existing external policies (CFIR) or influenced by legislative changes which may or may not align with current practices (BCW). Additional clarification of community-level and macro-level factors would be important adaptations to the reviewed IS theories/frameworks to inform population-level interventions. In PHIR there is a focus on understanding how social, economic, political, and environmental conditions influence the health of a community or population and effective interventions should address multiple levels at the same time [@b5],[@b10]. This multi-level planning that is required to simultaneously target micro-, meso-, and macro-levels was not fully addressed in the reviewed IS theories/frameworks, although most did acknowledge the complexity of interventions. Despite their potential theoretical insight to population health, we identified several theories that were not commonly applied to population health interventions, which may be a result of their broad nature (e.g., Durlak and Dupre\'s Ecological Framework, Multilevel Change Framework) or the multitude of complex factors in the frameworks may present pragmatic limitations (e.g., Greenhalgh et al. Diffusion of Innovation). In comparison, the theories that were more commonly applied in the population health literature were those that built upon previous theories and also seemed to have investigator teams promoting its use through websites, books and publications (e.g., CFIR, TDF and BCW and Promoting Action on Research Implementation in Health Services framework). A recent study explored the use of IS theories in the field and reported that implementation scientists use various criteria to select theories, but suggested that the process may be informed by convenience or prior exposure [@b62]. Future research is needed to explore uses of IS theories and frameworks in PHIR to identify the reasoning of their use, how they have been adapted and applied in different population-level contexts. It is also important to note, that Durlak and Dupre\'s Ecological Framework was the only theory/framework to explicitly include factors relevant to the social determinants of health and provide relevant examples of social policy interventions that would address these concerns. Further examination of the theories/frameworks and their application could help to identify further constructs relevant to the social determinants of health that were less explicitly defined in the source documents. There were numerous similarities across the reviewed IS theories and frameworks, likely a result of the developing nature of the field. Many of the theories/frameworks built upon prior work. For example, inherent in its name the *Consolidated* Framework for Implementation Research [@b49] brought together learnings from previous theories, including four of those explored in this review [@b46],[@b47],[@b52],[@b54]. Another example is the extension of the TDF to the BCW, representing the development and refinement of behavior change research over a decade. The more recent BCW provides explicit guidance in intervention design that can bridge the gap between individual health behaviors and large-scale intervention strategies [@b50],[@b51]. There was also a clear difference between factors identified as relevant for implementation, depending on the field the theories and frameworks originated from. For example, those developed in business settings are more mechanistic in that they did not account for external factors which might make implementation difficult, and thus are less useful for PHIR [@b53]--[@b55]. This does not mean that these theories cannot lend themselves to population health interventions; conversely, it is possible that their focus on organizational-level factors might provide added depth or insight that broader frameworks or theories do not. However, based on the results of this review, we feel that the absence of external factors is a critical gap which makes them less suitable for use in population health interventions, and that alone they are less relevant than other IS theories or frameworks that do not account for external factors. In contrast, most of the reviewed theories from a health care context acknowledged multiple levels of action [@b47]--[@b50],[@b58] or at least contextual factors [@b46],[@b52]. Moreover, three theories/frameworks that we identified to be useful to PHIR [@b45],[@b47],[@b49] agreed on the importance of five factors when assessing implementation viability which might offer potential value to PHIR, including: (1) innovation characteristics; (2) individuals involved with its implementation; (3) organizational capacity to support implementation; (4) broader-level policy/politics/financial/social contexts; (5) the implementation process. The overlap of important factors across theories and frameworks may limit the operational use to a "checklist" of factors to consider in implementation research. This is likely due to the complexity of implementation in health settings, as the theories/frameworks that were deemed less useful to PHIR were also the ones that provided more details on the operationalization of theoretical components [@b53]. Unfortunately, beyond indicating the relative importance of factors related to effective implementation, few theories provided direction on how the components of these different frameworks might work together, or if they need to be applied as a whole theory/framework. The CFIR is sole theory that specified it is meant to be a menu of constructs the researcher chooses from based on the salience of the construct to the context, so it is not necessary it be applied as a whole model [@b49]. This "menu" approach presents an additional challenge for researchers who cannot target all relevant factors: How will constructs be prioritized and selected? Tools such as the APEASE criteria (Affordability, Practicality, Effectiveness, Acceptability, Safety, Equity) [@b58] or worksheets developed by Flottorp and colleagues [@b63] offer guidance on prioritization and assessment of factors within an implementation strategy. These tools enhance understanding of the implementation context(s), which the interventionist can use to identify what factors are critical to address. This concept of "critical capacities" has been noted in the political science literature; many factors may contribute to success in complex, dynamic systems (e.g., population health interventions, federal policy making), but not at all times or across all conditions [@b64]. It also aligns with realist philosophy that supposes that different contexts can leverages aspects of interventions to make them successful [@b65]. 4.1.. Strengths and limitations {#s4a} ------------------------------- One of the strengths of this review is that the authors involved with this paper provided varying perspectives due to their diverse backgrounds (e.g., public health, health promotion, population health, health services research, implementation science), which provided safeguards for potential disciplinary biases in analytic approaches. An additional level of rigor was added by having a larger authorship team review and provide feedback on nascent findings. It is important to note the review team broadly interpreted the terms used in the frameworks and theories as many of the terms were specific to health care settings (e.g,. patients/clients could also refer to community members, or providers could refer to practitioners). However, it was acknowledged that this broad interpretation of terms may not be relevant or applicable to certain communities or populations. Several limitations to the review should be noted, including the fact that full transparency of the review procedures are not possible due to the interpretive nature of the process [@b18]. However, the various steps taken by the team to ensure rigour were used to frame the analysis have been explained. A second limitation is that the review only included seminal theories; this process was deemed necessary to keep the review focused and manageable yet as a result, we may have missed relevant theories that were not as widely cited in the literature. 5.. Conclusions {#s5} =============== Theories and frameworks are important to help clarify factors influencing the implementation and sustainability of evidence-based population health interventions. This study explored how key theories from the field of IS might inform policies and programs seeking to reduce the burden of illness across populations. The theories most relevant to PHIR were expansive frameworks that considered macro-level contexts, but these theories were still limited in their practical guidance to inform population level interventions. A plethora of constructs were offered by the reviewed theories and frameworks as factors to consider or assess in implementation, with five categories of factors common appearing throughout all theories: (1) innovation characteristics; (2) individuals involved with its implementation; (3) organizational capacity to support implementation; (4) broader-level policy/politics/financial/social contexts; (5) the implementation process. However, there was limited explanation of how they might be of practical value to address structural factors such as the social determinants of health. ###### Click here for additional data file. This research was funded by the Faculty of Health (Dalhousie University). The authors would like to acknowledge support from the Healthy Populations Institute (Dalhousie University), Leah Boulos and the Maritime SPOR (Strategy of Patient-Oriented Research) SUPPORT Unit. Jessie-Lee D McIsaac acknowledges funding from the Canadian Cancer Society (grant \# 703878) while the review was completed. **Conflict of interest:** The authors declare no conflict of interest.

Questions of the accommodation of specific Muslim practices[^1^](#fn1){ref-type="fn"} and rights (e.g., headscarf, Mosques, Islamic education) within the limits of liberal societies are at the center of the polarized debate in Western Europe and other Western societies (e.g., [@c7]). Research has focused on understanding attitudes toward Muslim practices and rights in terms of group-based prejudice and Islamophobia (e.g., [@c36]; [@c56]; [@c59]). Support for banning the headscarf, for example, reflects anti-Muslim feelings ([@c25]; [@c57]). However, some studies suggest that the relationship between prejudice toward Muslims and acceptance of their religious practices is not straightforward (e.g., [@c63]; [@c74]; [@c75]). The level of acceptance or rejection of Muslim practices may depend on the nature of the specific practice in question, and individuals can be intolerant because of more principled objections to that particular practice rather than due to prejudicial feelings. Considering general feelings toward Muslims as well as the acceptance of a range of Muslim practices allows us to examine the possibility that people reject particular Muslim practices and the related rights (e.g., Islamic primary education) while thinking well of Muslims as a group, and vice versa. Further, if we were to investigate only one specific practice we risk missing the attitudes of those who would like to ban that practice but accept other Muslim practices, or conversely those who might accept the practice but want to forbid all else. Therefore, and in going beyond previous research on anti-Muslim attitudes, our primary goal is to jointly examine majority members' general feelings toward Muslims and their acceptance of a range of controversial but legal Muslim practices. Using four large datasets from the Netherlands and a person-centered approach we examined whether there are distinct groups of individuals within the majority population with different combinations of feelings and acceptance, and with different demographic and social psychological characteristics. Anti-Muslim Reactions {#SS1-1} ===================== Research in Europe (e.g., [@c66]; [@c69]) and in the United States (e.g., [@c36]) indicates that anti-Muslim feelings are more wide-spread than negative feelings toward other immigrant groups. [@c36] found that an empirical distinction can be made between people's attitudes toward categories that are defined by racial, ethnic, and religious background and their attitudes toward cultural groups that are defined by dissenting practices and behaviors. Anti-Muslim feelings were found to be connected to both attitudes, and most strongly to the latter ones. Using survey data, several studies try to examine the extent to which these anti-Muslim feelings reflect group-based prejudice and the extent to which these reflect specific forms of (religious) critique ([@c6]; [@c23]; [@c30]; [@c36]; [@c63]). For instance, research has demonstrated that Enlightenment values (e.g., [@c22]; [@c30]), secularism ([@c72]), and universalistic notions ([@c17]; [@c57]) predict anti-Muslim feelings over and above the statistical effect of generalized prejudice. These findings indicate that criticism of Islam cannot be reduced to anti-Muslim feelings. Another strategy that can be used to understand anti-Muslim feelings is to distinguish between group-based prejudice and (in)tolerance of specific practices. Research in Western Europe has argued that people with more prejudicial attitudes toward Muslim minorities also more strongly oppose, for example, the building of Mosques, the wearing of headscarves, and Islamic schools. Thus, opposition to and rejection of these specific practices would reflect a general dislike of Muslims. In support of this perspective, several studies have found that higher prejudice is indeed associated with higher intolerance of Muslim practices and that the rejection of dissenting practices is used to justify anti-Muslim feelings (e.g., [@c25]; [@c57]; [@c74]; [@c76]). However, there is also research that supports an alternative perspective, indicating that (political) intolerance and prejudicial attitudes are distinct phenomena ([@c9]; [@c23]; [@c37]; [@c75]; [@c83]). For example, in the context of Quebec, Canada, while those who hold prejudicial views supported a ban on religious symbols, a majority of the people supporting the ban did so out of principled secularism rather than prejudice ([@c6]). In another study in Quebec it was found that feelings of cultural threat and generalized prejudice predicted support for banning *minority* religious symbols whereas holding liberal values predicted support to ban *all* religious symbols ([@c4]). Similarly, analyzing data from six European countries, [@c25] found that Europeans with secular liberal values felt positively toward Muslims as a group, but felt torn regarding the legislation of religious practices such as the wearing of the headscarf. In addition, among national samples in the UK, France, Germany, and the Netherlands, a substantial portion of people with a positive attitude toward Muslims supported a ban on headscarves ([@c73]; see also [@c57]) and also rejected Islamic education and the building of Mosques ([@c74]). Drawing on these two competing perspectives, an objection to a specific practice may represent (1) an expression of one's negative attitude toward Muslims as a group, (2) an expression of disapproval of that particular practice, or (3) a combination of these two. For example, majority members can reject the founding of Islamic schools because they dislike Muslims, or because they believe that religion in general has no place in education, or a combination of the two. In other words, the distinction between group-based attitudes toward Muslim minorities and practice-specific disapproval is important and requires greater attention ([@c4]; [@c14]; [@c29]). One way to address this is by considering a range of Muslim practices in combination with general feelings toward Muslims. This allows us to develop a more detailed understanding of particular combinations of group-based feelings of Muslim minorities and the acceptance of specific Muslim practices. Intolerance and Multiple Practices {#SS1-2} ================================== The combination of negativity toward a minority group with nevertheless accepting this minority group's civil rights is central in the literature on political tolerance ([@c19]; [@c70]). However, negative out-group feelings are not a precondition for (political)*in*tolerance because one can reject a specific practice (e.g., ritual slaughter of animals) of people or groups (Jews, Muslims) to whom one has neutral or even positive feelings ([@c29]; [@c64]). Considering various dissenting practices simultaneously makes it possible to take into account the distinction between majority members who consistently accept or rather reject various Muslim practices, and people who do not consistently accept or reject the different practices. Cross-practice consistency can indicate a general like or dislike of Muslim minorities and inconsistencies can indicate principle considerations or rather social normative concerns about specific practices ([@c14]; [@c64]). Specifically, some people may display positive feelings toward Muslims and are consistent in accepting the various practices ("liking"), while others have negative feelings and consistently reject all practices ("disliking"). The existence of these two groups is in agreement with the literature that links support for Muslim minority practices with group-based feelings. Additionally, however, there is the possibility that these positive or negative feelings go together with the acceptance of some Muslim practices but not of other practices. This practice-based inconsistency indicates that people do not only consider information about whom they are asked to tolerate but also on the nature of the dissenting practices (e.g., [@c20]; [@c53]). The consideration of multiple practices might demonstrate that people have objections to a particular practice (e.g., wearing of a headscarf) but not toward another practice (e.g., Islamic education). A particular practice might raise specific moral concerns, such as the wearing of a headscarf which might evoke the issue of gender equality and religious education in public schools that can evoke concerns about the secular nature of the state ([@c46]; [@c58]). However, the rejection of some Muslim practices might also result from social normative standards that make it socially acceptable to express one's prejudiced feelings by rejecting these practices and not others ([@c8]). Thus, some individuals may be practicing a form of principled intolerance, where their disapproval emerges from specific views held about specific practices, while other individuals may be practicing a prejudiced intolerance, where the disapproval of specific practices emerges from a prejudice toward the group that they may be unwilling to express. Using data from four representative samples of the Dutch majority and a person-centered approach, we will examine whether these four groups of individuals do indeed exist and how many people can be described as practicing, for example, a principled or prejudiced intolerance. Furthermore, as a matter of construct validity we consider whether some key demographic and social psychological constructs characterize the different groups. Person-Centered Approach {#SS1-3} ======================== In general, social psychological research typically investigates associations using a variable-centered approach in which the evaluation of an out-group and its practices is considered to have a common underlying dimension that ranges individuals from low to high prejudice (see [@c42]). This approach ignores the possibility that individuals are not equally negative toward distinct sets of practices and therefore that a single score does not accurately reflect the stances that majority members take. Further, it is possible that people combine their feelings and objections in different ways leading to groups of individuals with distinct constellations of ratings. In other words, a variable-centered approach ignores the complex constellation of characteristics that make up individuals and precludes the possibility that there are profiles of people which would provide a nuanced and more detailed understanding of how majority members perceive and evaluate Muslims as a group and the various practices they are engaged in. Taking a person-centered approach makes it possible to consider these combinations of out-group feelings and the evaluation of different out-group practices simultaneously (e.g., [@c3]; [@c42]; [@c47]). This type of analysis seeks to identify unobserved groups, or categories, of individuals who differ in the particular ways in which they combine their out-group feelings and objections to a range of out-group practices. For example, in a research on political tolerance, the best model for the data required four categories of individuals rather than a continuum of tolerance ([@c39]). In addition to a group of individuals who were consistently positive toward various minority groups and three different practices and a group of people who were consistently negative, two other categories of people were found (see also [@c26]; [@c40]; [@c44]; [@c64]). Individuals in these two categories accepted some groups and some practices but rejected others. These four categories could not be readily placed on a unidimensional positive-negative continuum because there was no monotonic change across the four groups of individuals. Rather, they formed four latent classes of political tolerance. Thus, a person-centered approach allows us to investigate whether feelings toward Muslims as a group of people and objections to various Muslim practices are combined in different ways by different groups of individuals. This makes it possible to examine whether, and how many, majority members have, for example, a liking-based, a disliking-based, or a principled or prejudiced intolerant profile concerning their attitudes toward Muslims and Muslim practices. One group of individuals might reject almost all Muslim practices while another group of individuals might reject the building of Mosques but accept the establishment of Islamic schools. Thus, rather than a unidimensional continuum, by considering different dissenting practices simultaneously it is possible to assess potentially different patterns among different groups of individuals that may speak to the underlying attitudes that affect which practices they accept. While it might be reasonable for someone to have a fair objection to a single specific practice, it seems unlikely that they would have such objections to all outgroup practices. Rather someone who is generally prejudiced toward Muslims as a group will be less likely to differentiate based on the specifics of individual practices but will tend to be broadly intolerant of a range of Muslim practices. Profiles and Their Correlates {#SS1-4} ============================= Beyond identifying groups of individuals based on their acceptance of outgroup practices and general out-group feelings, it is important to examine whether the different groups of individuals differ for some key characteristics that are typically considered in research on minority group prejudice. This is important because both principled and prejudiced intolerance involve practice-inconsistent rejection and therefore are largely indistinguishable in terms of observed responses. However, several factors should strengthen or weaken a tendency to principled tolerance and these can be examined as a matter of construct validity. We will look into the role of educational background and political orientation as two main demographic predictors of prejudicial attitudes and of important social psychological constructs. Research on the effect of education on prejudicial attitudes suggests a complex relationship. The ideological refinement perspective ([@c31]) views education as endowing majority members with more advanced cognitive skills and ideological commitments to support abstract ideas of equality and justice, while simultaneously using their cognitive skills to protect the status quo by rejecting social policies designed to overcome group-based inequalities. Research also suggests that the association between higher education and a more positive attitude toward minority groups is not explained by a greater tendency of the higher educated to respond in a socially desirable way ([@c24]; [@c51]; [@c82]). This does not mean that the higher educated have less spontaneous negative reactions toward ethnic minorities. Research on aversive racism ([@c12]) has shown that aversive racists show relatively strong prejudice on implicit but not on explicit measures ([@c65]), and the higher educated have been found to have lower explicit but not implicit prejudices ([@c38]). However, the association between higher education and a more positive self-reported attitude toward ethnic and cultural minority groups is one of the most replicated findings in the social sciences ([@c32]) and has been predominantly explained in terms of cognitive development and the learning of liberal values. Education is associated with cognitive ability and flexibility ([@c5]; [@c49]) and is a strong correlate of political sophistication ([@c2]; [@c11]; [@c27]; [@c28]). When the educational level in the population increases, the ideological sophistication also increases ([@c71]). Education implies political socialization which involves a better understanding of the values and beliefs that underlie political-ideological differences ([@c50]; [@c67]) and making the higher educated better able to understand the importance of basic norms and values of equality and tolerance underlying the democratic culture ([@c81]). The higher educated are not only more likely to be tolerant in general but also as a matter of principle ([@c64]). Thus, the literature leads us to expect that the "general liking" and "principled intolerant" groups are more highly educated than the "prejudiced intolerant" and "generally disliking" groups. In addition to education, we also considered a direct measure of cognitive sophistication as the tendency to consider how issues that one feels strong about can have multiple perspectives. Based on our reasoning for the role of education we can expect that the "general liking" and "principled intolerant" will display higher cognitive sophistication than the prejudiced intolerant and the disliking groups. Social psychological research on the social cognition model has argued and demonstrated that two core aspects capture the most important differences between the political right and the left ([@c35]). Political orientation is manifested in a specific ideological configuration in which respect for tradition and acceptance of inequality are central ([@c33], [@c34]). The first dimension concerns attitudes toward cultural tradition and social deviance, while the second relates to (in)equality and egalitarianism. Individuals on the political right tend to prefer traditions and social conformity, and to accept inequalities. In contrast, those on the left more strongly embrace socio-cultural change and equality. Extensive empirical research in political psychology and in different Western countries ([@c54]) confirms that these two core dimensions capture the most important ideological differences between right-wing and left-wing political orientations ([@c33], [@c34]; [@c61]). This means that we expected that the Muslim liking and principled intolerant groups have a stronger left-wing political orientation compared to the Muslim disliking and prejudiced intolerant groups. We further investigated Social Dominance Orientation (SDO), Right Wing Authoritarianism (RWA), and perceived out-group threat as three key social psychological constructs that have been extensively linked to tolerance as well as prejudicial attitudes toward minority groups. Examining these constructs allows us to assess whether the expected groups of individuals do not only differ in their attitude toward Muslims and Muslim practices but also in these important correlates. Specifically, the "generally like" group and the "principled intolerance" group should have lower SDO, lower RWA, and lower perceived threat compared to the "prejudiced intolerance" and "generally dislike" groups. Furthermore, we also sought to determine whether the groups of individuals differ in their internal motivation to control and not express prejudices. This measure can give a further understanding of the underlying difference between principled and prejudiced intolerant individuals. Specifically, we can expect that the "general liking" and "principled intolerant" will display a stronger internal motivation to control prejudice than the prejudiced intolerant and the disliking groups. Finally, we tried to extend beyond attitudes and ideologies to behavioral intentions. Specifically, we considered the willingness to engage in anti-discrimination activities. We selected this variable because someone who is intolerant of specific practices for more principled reasons should not accept discriminatory treatment of minority members or members of other minority groups. Therefore, support for anti-discrimination activities is expected to serve as a useful identifier for making a distinction between the principled and prejudiced intolerant groups of individuals. Summing Up {#S2} ========== In this article, we drew on data from four large representative datasets of majority Dutch participants collected in the Netherlands between 2014 and 2018 which allows us to investigate how well the data aligns over time and across sampling error. We examine feeling thermometer ratings toward the two most prominent and typical groups of Muslims in the Netherlands (of Turkish and Moroccan background), together with the acceptance of five different Muslim practices that are strongly debated in society (public expression of Muslim religion, wearing of the headscarf, celebration of Islamic holidays, building of Mosques, founding of Islamic schools). Our first aim is to investigate whether there are groups of individuals who are characterized by particular combinations of general feelings and acceptance of specific practices. More specifically, we expect to identify the four types of dislike-based (prejudiced feelings and rejection of all practices), like-based (non-prejudiced feelings and acceptance of all practices), principled intolerant (non-prejudiced feelings and differential rejection of practices), and prejudiced intolerant (non-prejudiced feelings and general rejection of practices). Second, we expect that the groups of individuals differ in terms of several well-known correlates of prejudice toward minority groups. Based on existing variable-centered research we expect that the groups differ in level of education, political orientation, RWA, SDO, perceived out-group threat, the internal motivation to control prejudice, cognitive sophistication, and the tendency to be involved in anti-discrimination actions. Specifically, we expected that the liking group and the principled intolerance group differ on these correlates from the disliking group and the prejudiced intolerance group. Method {#S3} ====== Participants {#SS1-5} ------------ For this paper, we analyzed data of four large representative datasets that contain various measures and that have been used for other purposes in previous research (e.g., [@c43]; [@c79]). In all four studies the data were collected online among probability samples drawn from nationally representative pools of the majority Dutch population. The response rate of the different studies was around 55% which is similar to other research in the Netherlands ([@c68]). The samples covered various segments of the Dutch public in terms of age, gender, education, household size, and the region of residence. The samples were selected by research consultancy companies which maintain a database of majority Dutch people who regularly participate in surveys in return for remuneration. In Study 1, 469 majority Dutch participants in the 2014 dataset (Study 1) completed all seven key measures and were retained for analysis. Participants were identified as majority Dutch based on self-identification and if both of their parents were born in the Netherlands. Participants who identified as Muslim were excluded in the analysis (two participants each in Studies 2 and 3). Similarly, 800 participants in the 2015 dataset (Study 2), 590 in the 2017 dataset (Study 3), and 563 participants in the 2018 dataset (Study 4) completed all key measures and were retained for analysis. The participants were split relatively evenly along gender lines (2014: 52.2% male; 2015: 50.0% male; 2017: 54.56% male: 2018: 50.3% female), and from a wide range of ages (2014: *M* = 50.26, *SD* = 16.98, range = 18--88; 2015: *M* = 50.65, *SD* = 17.16, range = 18--87; 2017: *M* = 55.46, *SD* = 14.60, range = 18--87;2018: *M* = 51.08, *SD* = 17.52, range = 18--91), education levels and political orientation (descriptive statistics reported below). Materials {#SS1-6} --------- The materials used are discussed below and the English (translated) versions of all items can be found in the Electronic Supplementary Material (ESM 1). Complete data and analytic scripts can be found on the Open Science Framework at <https://osf.io/7j2zm/>. ### Out-Group Feelings {#SS2-1} Respondents were presented with feeling thermometers to indicate their coldness or warmth to members of the two main Muslim minority groups in the Netherlands: Turks and Moroccans. Both groups are over 95% Muslim, and are widely recognized as the two prototypical Muslim groups in the Netherlands. Using feeling thermometers with wider ranges of responses than Likert-type scales generates a more reliable measure ([@c1]), and these explicit measures tend to correlate with subtler methods of assessing prejudice ([@c13]). Overall and reflecting the ethnic hierarchy in the Netherlands ([@c60]), feelings toward Turks hovered around the midpoint of the 1--11 scale, neither warm nor cold (*M*~Study1~ = 6.05, *SD*~Study1~ = 2.29; *M*~Study2~ = 6.20, *SD*~Study2~ = 2.20; *M*~Study3~ = 6.45, *SD*~Study3~ = 2.27; *M*~Study4~ = 5.13, *SD*~Study4~ = 2.21), while feelings toward Moroccans were consistently lower (*M*~Study1~ = 4.36, *SD*~Study1~ = 2.23; *M*~Study2~ = 4.28, *SD*~Study2~ = 2.12; *M*~Study3~ = 4.66, *SD*~Study3~ = 2.47; *M*~Study4~ = 4.54, *SD*~Study4~ = 2.24). More negative attitudes toward Turks in 2018 (Study 4) may be due to a 2017 political and diplomatic crisis between Turkey and the Netherlands. Across all three studies, the two measures were positively correlated (*r* ranging from .49 to .61) and in a recent study among a Dutch representative sample, the feelings toward Turks and Moroccans correlate strongly with the feeling toward Muslims as a category (.73 and .77, respectively). ### Acceptance of Muslim Practices {#SS2-2} The specific practices presented to the respondents were partially adapted from previous research (e.g., [@c21]; [@c62]). These relate to different types of civil liberties that, however, are subject to much debate in Dutch society.[^2^](#fn2){ref-type="fn"} Using 7-point scales, one item asked participants' agreement that Muslims can express their faith in public ("Muslims in the Netherlands must be able to show and experience their own faith in public life"; *M*~Study1~ = 4.31, *SD*~Study1~ = 1.60; *M*~Study2~ = 4.44, *SD*~Study2~ = 1.53; *M*~Study3~ = 4.16, *SD*~Study3~ = 1.77; *M*~Study4~ = 4.72, *SD*~Study4~ = 1.54), the second referred to Muslim women's ability to wear the headscarf ("Muslim women should have the opportunity to wear a headscarf anywhere in the Netherlands"; *M*~Study1~ = 3.34, *SD*~Study1~ = 1.77; *M*~Study2~ = 3.78, *SD*~Study2~ = 1.80; *M*~Study3~ = 3.63, *SD*~Study3~ = 1.92; *M*~Study4~ = 3.87, *SD*~Study4~ = 1.77), the third asked about Muslim's rights to celebrate their festivals in public ("Muslims in the Netherlands should not only be able to celebrate their Islamic holidays at home, but also in public"; *M*~Study1~ = 4.04, *SD*~Study1~ = 1.64; *M*~Study2~ = 4.13, *SD*~Study2~ = 1.60; *M*~Study3~ = 3.99, *SD*~Study3~ = 1.82; *M*~Study4~ = 4.46, *SD*~Study4~ = 1.62), the fourth about the right to build mosques ("Muslims must have the right to build mosques in the Netherlands"; *M*~Study1~ = 3.72, *SD*~Study1~ = 1.76; *M*~Study2~ = 3.94, *SD*~Study2~ = 1.71; *M*~Study3~ = 3.91, *SD*~Study3~ = 1.87; *M*~Study4~ = 4.29, *SD*~Study4~ = 1.72), and the fifth about the right to establish Islamic schools ("Muslims must have the right to establish Islamic schools"; *M*~Study1~ = 3.09, *SD*~Study1~ = 1.65; *M*~Study2~ = 3.08, *SD*~Study2~ = 1.65; *M*~Study3~ = 2.84*SD*~Study3~ = 1.71; *M*~Study4~ = 3.35, *SD*~Study4~ = 1.75). Thus, these items range from the less objectionable (showing and experiencing faith in public) to the more objectionable (establishing Islamic schools that may perceived to prevent social and cultural integration). ### Predictor Variables {#SS2-3} *Educational background* was measured using a single-item in which participants indicated their highest educational achievement on a scale ranging from 1 (= *no higher education*) to 8 (= *doctorate or advanced masters*) in Studies 1 and 3, and 1 (= *no higher education*) to 7 (= *doctorate or advanced masters*)in Study 2 and 4 (Study 1: 23.4% low, 48.9% middle, 27.7% high, *M*~study1~ = 5.14, *SD*~study1~ = 1.68; Study 2: 25.4% low, 49.7% middle, 25.0% high, *M*~study2~ = 5.03, *SD*~study2~ = 1.68; Study 3: 23.5% low, 45.1% middle, 31.4% high,*M*~study3~ = 4.24, *SD*~study3~ = 1.75; Study 4: 17.1% low, 47.4% middle, 35.5% high, *M*~study4~ = 4.45, *SD*~study4~ = 1.72). The distinction between these levels of achieved education is comparable to the International Standard Classification of Education (ISCED)-measure that is used, for example, in the European Social Survey. Similar to other research in the Netherlands (e.g., [@c10]; [@c77]), education was treated in the analysis as a continuous variable which allows us to investigate the difference between lower and higher educated participants. *Political orientation* was measured with the well-known self-placement question ([@c33]). A 5-point scale was used ranging from politically left, center-left, center, center-right to right. In all four samples, the overall mean for political ideology fell in the center (*M*~study1~ = 3.00, *SD*~study1~ = 1.09; *M*~study2~ = 2.92, *SD*~study2~ = 1.05; *M*~study3~ = 3.02, *SD*~study3~ = 1.18; *M*~study4~ = 2.93, *SD*~study4~ = 1.27) with an equal distribution to the political right and to the political left. A measure of *social dominance orientation* was available in two datasets. Eight (in Study 2; *M* = 3.24, *SD* = 0.86; α = .76), and six (in Study 3; *M* = 3.20, *SD* = 1.02; α = .72) items of a short version of SDO were used that was validated and translated to Dutch by [@c16] and that has been used in other research ([@c15]; [@c41]). *Right-wing authoritarianism (RWA)* was measured using a short measure previously utilized in the Netherlands ([@c63]) and that focuses on the conformity aspect of RWA. This measure was used in Study 2 (3 items, *M* = 4.61, *SD* = 1.12, α = .70) and Study 3 (4 items; *M* = 5.09, *SD* = 0.98; α = .70). *Perceived out-group threat* was measured in three datasets in a reliable but not identical way. In Studies 1 and 2, symbolic threat was measured using four and three items, respectively, that asked about Muslims in the Netherlands undermining the Dutch identity and way of life (2014: *M* = 4.00, *SD* = 1.49, α = .90; 2015: *M* = 3.99, *SD* = 1.64, α = .96). In Study 3, in light of the refugee crisis, the four threat items were expanded to refer to refugees who are from predominantly Muslim nations rather than Muslims as a group (*M* = 3.97, *SD* = 1.89, α = .97). Study 3 also included a measure of *internal motivation to control prejudice* ([@c55]), with the goal of better understanding the motivations of people who might not explicitly declare prejudice but nonetheless be unaccepting of Muslim practices. The scale was comprised of four items measured on a 1--7 scale (*M* = 5.01, *SD* = 1.09, α = .97). *Cognitive sophistication* was also measured in Study 3 using a 4-item scale (*M* = 5.18, *SD* = 0.99, α = .83) that focused on whether and how frequently participants sought to understand alternative perspectives on issues that they felt strongly about. *Behavior intentions.* Following previous research in the Netherlands ([@c78]), in Study 3 respondents were asked how likely it is that they would engage in a set of actions in response to discrimination against immigrants in the Netherlands (*M* = 2.34, *SD* = 0.95, α = .85). Analyses {#SS1-7} -------- For the analysis we used a latent profile analysis approach (e.g., [@c48]) which we conducted using R software. Latent profile analysis is analogous to latent class analysis (LCA; [@c47]) and identifies patterns across a set of continuous variables that can be identified as different profiles of individuals. We included the two feeling thermometers measuring attitudes toward Muslims (Turks and Moroccans) and the five items about Muslim practices into latent profile analysis to determine how general feelings and acceptance of specific practices combine to best identify different profiles of individuals in Dutch society. We conducted this same analysis across four separate datasets to triangulate in on a set of profiles that best represent Dutch society. The LPA analyses performed identify the best models of profiles using different sources of empirical information about the appropriate number of profiles. First, the Bayesian information criteria (BIC) and Akaike information criteria (AIC) indicate how well a model with the selected number of profiles fits the data, with the lowest numbers indicating the best fit. Secondly, the Bootstrappped Likelihood Ratio Test (BLRT) conducts significance tests comparing the selected model with a model including one fewer profile (comparison between *k* and *k* − 1 profiles). Lastly, entropy scores indicate how uniquely the datapoints belong to one profile and not others. Low entropy scores indicate that datapoints (participants) could be classed into more than one profile, whereas high entropy scores indicate that participants are uniquely classed into one profile and not others. A final component of determining profiles is based on interpretability. One important nuance of interpretability is the need not to overfit the data by generating a model with profiles that represent specific variances unique to the dataset rather than generalizable profiles of participants. The suggested ways to avoid this is to ignore small classes and to inspect the profiles that emerge for interpretability. In our analysis, we used all of these approaches to arrive at the best profile fit for the data. Results {#S4} ======= Latent Profiles {#SS1-8} --------------- [Table 1](#tbl1){ref-type="table"} shows that across four independent datasets collected over a 4-year period, the combination of respondents' general feelings toward Muslims minorities and their support for specific Muslim practices reveal four groups of individuals. Investigation of the best type of model across the four datasets revealed that a model that allowed for varying means while holding variance and covariance equal best fit the data. As can be seen in [Table 1](#tbl1){ref-type="table"}, while the BIC criteria in Studies 2--4 indicate that six profiles provide a slightly better fit, in both cases the improvement over the previous iteration is relatively small, and the criterion of theoretical interpretability suggests a four profile solution across the four datasets. Specifically, when we investigated profiles generated by the six-profile solution, the resultant profiles were inconsistent across the studies, generating profiles of less than 50 people (*n* ranging from 31 to 45 in Studies 2 and 4), and somewhat larger, though still inconsistent profiles in Study 3 (*n* = 61--78). In all studies, the six-profile solution returned groups consistent with the Like, Dislike, and Prejudiced Intolerant groups (see below), and returned fractured sections of what we term the Principled Intolerant group, usually in which some participants displayed greater or lesser concern about individual practices like wearing the headscarf or opening Islamic schools. These differences, while likely reflecting a range of concerns among that group, did not consistently return distinct profiles across the datasets, so we elected to use the more interpretable four profile solution. Thus, while the six profile solution performed better statistically, there was no consistent pattern across the six-profile solutions that improved interpretability above the four-profile model. ###### Model fit indices across three studies {#tbl1a} Study \# Profile \# BIC AIC BLRT Entropy ---------------- ------------ --------------- ---------------- ------------------------- ---------- Study 1 (2014) 3 11,644.66 11,432.98 65.56, *p* \< .001 0.93 **4** **11,563.52** **11,318.633** **130.35, *p* \< .001** **0.92** 5 11,697.80 11,419.71 −85.07, *p* = .999 0.87 6 11,650.53 11,339.23 NA 0.87 Study 2 (2015) 3 19,845.43 19,606.42 57.85, *p* \< .001 0.88 **4** **19,689.68** **19,413.29** **209.23, *p* \< .001** **0.86** 5 19,698.98 19,385.11 44.18, *p* \< .001 0.84 6 19,663.19 19,311.85 89.26, *p* \< .001 0.89 Study 3 (2017) 3 14,842.74 14,619.35 142.92, *p* \< .001 0.89 **4** **14,719.83** **14,461.41** **173.95, *p* \< .001** **0.88** 5 14,712.99 14,419.52 57.89, *p* = .001 0.90 6 14,712.11 14,383.61 51.91, *p* = .002 0.86 Study 4 (2018) 3 13,773.84 13,552.84 63.82, *p* \< .001 0.90 **4** **13,615.86** **13,360.20** **208.64, *p* \< .001** **0.92** 5 13,625.82 13,335.49 40.71, *p* \< .001 0.91 6 13,628.64 13,303.65 47.84, *p* \< .001 0.90 [Figure 1](#fig1){#fgc1-1 ref-type="fig"} (panels A--D) presents the mean levels of Muslim group feelings and acceptance of the specific practices across the four studies for the four profiles identified. Note that for ease of interpretation these scores were subtracted from the neutral midpoints of the scales of the items used. Thus, positive scores indicate attitudes that are higher than the midpoint and negative scores indicate attitudes below the midpoint. This allows us to differentiate between relative differences in attitudes that indicate negativity as compared to neutrality. {#fig1} A first group consists of individuals with relatively positive feelings toward the two Muslim minority groups combined with relatively high levels of acceptance for all the Muslim practices ("liking"; between 27.8% and 36.5% across the datasets).The second ("disliking") consists of individuals with negative feelings toward Muslim minorities combined with a tendency to reject the different Muslim practices (between 13.8% and 22.0%). In addition, the latent profile analysis indicates that almost half of the participants do not appear to be less prejudiced against Muslim minority groups than the disliking group but nonetheless are unwilling to tolerate Muslim practices. This category of individuals emerges as two profiles: one in which all or almost all Muslim practices are rejected without apparent distinctions made between them, and another in which some practices are rejected but not others. The first of these subgroups we label as "prejudiced intolerance" (between 28.5% and 34.5%). While they do not appear to have particularly negative feelings toward Muslims, displaying neutral to slightly positive attitudes toward the Turkish minority group for example, their rejection of all Muslim practices (most evident in Studies 2 and 3) without differentiating much between them suggests that this rejection may be driven by a general dislike of Muslims. In contrast, members of the second of these subgroups show substantial differences in attitudes toward distinct practices. For example, in Study 1 they are opposed toward the headscarf, and across all of the studies they are especially opposed to Islamic schools. At the same time, their support for religious freedom (i.e., building Mosques) and public expression is usually very similar to that of the "liking" group. Thus, this fourth group is distinguished both from those that are generally positive across the board and those that are neutral or negative across the board, and we label this group "principled intolerant" (between 17.6% and 29.2%). Their intolerance of some Muslim practices but not others suggests a rejection based on specific objections rather than a generalized dislike. Thus, while responses on some of the less controversial practices are broadly similar in pattern to those of other groups, it is precisely the differences in one or two practices that differentiate the groups. Across all four studies, the group we identify as principled intolerant deviates from the pattern found among the other three groups, specifically by displaying positivity or neutrality to most practices, but a strong and consistent objection to specific other practices. Thus, these findings support our expectation of the existence of four specific profiles, with two subgroups of intolerance without prejudice emerging: one which appears to disapprove of all practices and thus seems to more closely fit the description of prejudiced intolerance, and another which is intolerant of some practices but not others and thus appears to resemble a more principled intolerance. Predictors of the Groups of Individuals {#SS1-9} --------------------------------------- Further evidence for the construct validity of these groups is provided by investigation of differences in important correlates. Therefore, we next looked at whether the four groups of people identified across these datasets differ on key demographic, social psychological, and behavioral variables. To do this, we generated multinomial logistic regression models using SAS software to predict membership in these categories. We created the model using three steps. In the first step, we looked at education and, the related concept of cognitive sophistication, as well as political orientation as key characteristics. In the second step, we included SDO, RWA, and internal motivation to control prejudice to see whether and how ideological world view differences predicted membership in these four profiles. Then we added in out-group threat and the behavioral measure of willingness to act regarding discrimination against immigrants in the Netherlands, to see how those added to our understanding of how the types of people who fall into these profiles differ. We used the three-step model since we expect there to be a substantial overlap between demographic and world view predictors and with threat and behavioral predictors. By conducting the analysis in three steps we are able to see the effects of demographic variables independent of the other predictors, as well as demographic and world view predictors independent of threat and behavioral predictors. In the third step, we can also see which predictors play a meaningful unique role when included with all other predictors. For the analyses and following our predictions, the principled intolerant group was set as the referent. All variables included in these analyses were rescaled from 0 to 1 which means that the log odds beta coefficients and the odds ratios indicate the relative change in likelihood of belonging to each of the groups as a function of a full scale increase in a given predictor. [Table 2](#tbl2){ref-type="table"} shows how, while there is variation across the four datasets, there also appears to be a distinction between unique predictors for the four profiles. Overall the principled intolerant group appears to be distinct from the prejudiced intolerant and disliking groups, and in some cases appears to be slightly more similar to the liking profile than the prejudiced intolerant profile. First, across the four studies and as expected, people in the principled intolerance group tended to be among the best educated: the disliking and prejudiced intolerant groups are less well educated than the principled intolerant group and the liking group. Second, consistent with the education variable, the principled intolerant group was more likely to engage in cognitively sophisticated thought than the prejudiced intolerant or disliking groups. Third, members of the principled intolerant group are politically more left-leaning than people in the prejudiced intolerance and disliking groups. Fourth, the principled intolerant group has consistently RWA and had lower SDO tendencies than the disliking and prejudiced intolerant groups, with mixed differences from the liking group. Lastly, the principled intolerant group has marginally less motivation to control prejudice than the liking group and (marginally) more than the prejudiced intolerance group and the disliking group. ###### Multinomial logistic regression results of a demographics-only and demographics and personality predictors models with the principled intolerant group as the referent {#tbl2a} Liking Prejudiced intolerant Disliking ------------------------------------------------- ------------------ ----------------------- -------------------- ------- -------------------- -------- Panel A: Demographic predictors  Education   Study 1 −0.19 (0.50) 0.83 −0.69 (0.49) 0.50 −1.99\*\*\* (0.54) 0.14   Study 2 1.20\*\* (0.38) 3.31 −1.26\*\* (0.40) 0.29 −1.58\*\*\* (0.48) 0.21   Study 3 0.39 (0.38) 1.48 −1.87\*\*\* (0.42) 0.15 −1.86\*\*\* (0.52) 0.16   Study 4 0.64 (0.50) 1.89 −0.80^†^ (0.48) 0.45 −2.16\*\*\* (0.59) 0.12  Cognitive sophistication   Study 3 1.54^†^ (0.83) 4.67 −2.08\*\* (0.79) 0.13 −4.79\* (0.91) 0.01  Political orientation   Study 1 −1.57\*\* (0.51) 0.21 −0.44 (0.50) 0.65 1.13\* (0.56) 3.10   Study 2 −0.27 (0.39) 0.76 1.30\*\* (0.41) 3.68 0.83^†^ (0.50) 2.30   Study 3 0.44 (0.41) 1.55 2.56\*\*\* (0.46) 12.90 2.66\*\*\* (0.57) 14.36   Study 4 −0.84^†^ (0.45) 0.43 1.09\* (0.44) 2.97 2.94\*\*\* (0.57) 18.97 Panel B: Demographic and personality predictors  Education   Study 2 1.14\*\* (0.39) 3.13 −0.89\* (0.41) 0.41 −0.83 (0.51) 0.44   Study 3 0.23 (0.41) 1.26 −1.40\*\* (0.45) 0.25 −1.02^†^ (0.58) 0.36  Cognitive sophistication   Study 3 1.21 (0.93) 3.35 −1.30 (0.93) 0.27 2.02^†^ (1.10) 0.13  Political orientation   Study 2 −0.22 (0.43) 0.80 0.71 (0.45) 2.04 0.14 (0.54) 0.87   Study 3 0.57 (0.43) 1.77 2.33\*\*\* (0.47) 10.30 1.81\*\* (0.61) 6.08  SDO   Study 2 0.92 (0.76) 2.52 2.88\*\* (0.80) 17.78 4.61\*\*\* (1.01) 100.20   Study 3 0.82 (0.77) 2.28 1.18 (0.83) 3.25 2.96\*\* (1.09) 19.22  RWA   Study 2 −0.82 (0.62) 0.44 1.01 (0.67) 2.75 3.18\*\*\* (0.87) 24.07   Study 3 −1.18^†^ (0.70) 0.31 1.98\* (0.80) 7.22 3.81\*\*\* (1.05) 45.11  Internal motivation to control prejudice   Study 3 1.21 (0.91) 3.37 −0.71 (0.88) 0.49 −3.19\*\*\* (1.06) 0.04 [Table 3](#tbl3){ref-type="table"} shows how the addition of threat and behavioral predictors adds to our understanding of the different groups. Individuals in the principled intolerant group tend to perceive somewhat less threat from Muslims than the prejudiced intolerant group, further differentiating them, although those in the principled intolerant group nonetheless perceived greater threat than those in the liking group. Importantly, the distinctions between the two intolerant groups also emerged for behavioral intentions. A measure of willingness to engage in anti-discrimination activities showed that while people in the liking group were equally likely to engage in anti-discrimination activities as those in the principled intolerant group, those in the prejudiced intolerant and disliking groups were less likely to engage in such activism compared to the principled intolerant group (see ESM 1 for additional tables). ###### Multinomial logistic regression results of a full predictor model with principled intolerant as the referent group {#tbl3a} Liking Prejudiced intolerant Disliking --------------------------------- -------------------- ----------------------- ------------------- ------- -------------------- -------- Education  Study 1 −1.06^†^ (0.55) 0.35 −0.93^†^ (0.51) 0.40 −1.33\* (0.60) 0.26  Study 2 0.97\* (0.39) 2.64 −0.82\* (0.41) 0.44 −0.63 (0.53) 0.53  Study 3 0.14 (0.42) 1.16 −1.08\* (0.46) 0.34 −0.38 (0.63) 0.69  Study 4 0.64 (0.50) 1.89 −0.80^†^ (0.48) 0.45 −2.16\*\*\* (0.59) 0.12 Cognitive sophistication  Study 3 1.23 (0.95) 3.43 −1.18 (0.96) 0.31 −1.67(1.16) 0.19 Political orientation  Study 1 −0.56 (0.58) 0.57 −0.23 (0.53) 0.80 0.16 (0.61) 1.17  Study 2 0.11 (0.45) 1.12 0.51 (0.46) 1.67 −0.78 (0.56) 0.46  Study 3 0.68 (0.47) 1.97 1.73\*\*\* (0.50) 5.66 0.81 (0.65) 2.25  Study 4 −0.84^†^ (0.45) 0.43 1.09\* (0.44) 2.97 2.94\*\*\* (0.57) 18.97 SDO  Study 2 1.78\* (0.82) 5.94 2.44\*\* (0.84) 11.47 3.05\*\* (1.05) 21.22  Study 3 0.91 (0.78) 2.48 0.66 (0.87) 1.94 1.82 (1.17) 6.15 RWA  Study 2 −0.27 (0.65) 0.76 0.54 (0.71) 1.71 0.99 (0.95) 2.68  Study 3 −0.98 (0.72) 0.38 1.06 (0.83) 2.88 2.09^†^ (1.15) 8.11 Motivation to control prejudice  Study 3 0.99 (0.89) 2.69 0.36 (0.94) 1.43 −1.26 (1.14) 0.28 Threat  Study 1 −4.69\*\*\* (0.76) 0.01 −1.14^†^ (0.69) 0.32 5.76\*\*\* (0.96) 317.23  Study 2 −1.67\*\*\* (0.51) 0.19 0.94^†^ (0.53) 2.56 4.27\*\*\* (0.77) 71.59  Study 3 −0.50 (0.51) 0.61 1.55\*\* (0.50) 4.72 3.80\*\*\* (0.74) 44.81 Anti-discrimination  Study 3 0.00 (0.60) 1.00 −1.49\* (0.67) 0.23 −3.36\*\*\* (1.01) 0.04 Discussion {#S5} ========== The accommodation of Muslim practices in Western societies tends to evoke much political and public debate whereby some sections of the population argue for the acceptance of these practices and others are in favor of banning them ([@c7]; [@c45]). While previous research has examined how the public evaluates these sorts of practices, this research tends to consider the rejection of these practices as an expression of anti-Muslim feelings (e.g., [@c25]; [@c57]; [@c74]). However, while people can reject certain practices because of their prejudicial feelings toward Muslims as a group, they might also be opposed to these practices because of more principled objections (e.g., [@c22]; [@c30]; [@c72]). Individuals can be intolerant of specific practices while having either negative or positive feelings toward a group ([@c23]; [@c29]; [@c64]). We examined majority Dutch reactions to the Muslim minority group and different Muslim practices. Drawing on data from four large datasets covering 4 years, we used latent profile analysis to identify groups of individuals across the datasets, and then tested the construct validity of those groups by considering important demographic, psychological, and behavioral correlates of these groups. The advantage of these analyses is that the varying levels of anti-Muslim feelings and rejection to a range of Muslim practices are taken into account and that the heterogeneity of the population is identified. Across the four datasets and similar to research on political tolerance in different national contexts ([@c39]; [@c40]), we identified four latent profiles. In addition to those who generally like and generally dislike Muslims and their practices, we found evidence of two distinct groups of intolerant people that do not explicitly appear motivated by strong negative feelings toward both Muslim groups (but less negative toward Turks compared to Moroccans). One group was intolerant of all or almost all Muslim practices and we labeled this group "prejudiced intolerant." The second one was intolerant of some but not all practices and was labeled "principled intolerance." The principled and prejudiced intolerant groups appeared to represent large sections of society with up to half of the participants in our samples falling into these two groups, indicating the importance of investigating (in)tolerance when seeking to understand intergroup attitudes ([@c80]). We considered several demographic and social psychological variables to examine the meaningfulness of the distinction between the groups. The findings supports the expectation that there are individuals whose rejection of specific practices seems to be guided more by principled considerations than a general dislike toward Muslims. For example, the principled intolerant group is higher educated and more likely to engage in cognitively sophisticated thought than the prejudiced intolerant group. Interestingly, an additional analysis which looked only at the predictive power of cognitive sophistication above and beyond education indicated that adding cognitive sophistication as a predictor had no effect on the role of education in predicting category membership (see ESM 1). In light of past research suggesting that education improves intergroup attitudes through increased cognitive sophistication, this might suggest that education may improve intergroup attitudes by conveying liberal and accepting values or by decreasing feelings of intergroup competition. Level of education is also known as a strong correlate of political sophistication ([@c27]; [@c28]; [@c71]) and of being (in)tolerant as a matter of principle ([@c64]). Similarly, compared to the prejudiced intolerant group, the principled intolerant group is more left-wing politically and research in different Western countries has found that political orientation organizes people's values and beliefs about equality and social deviance ([@c33], [@c34]; [@c54]). Moreover, the principled intolerant group did not only endorse social dominance and authoritarianism less than the prejudiced intolerant group but was also more willing to address the unjust treatment of minority groups. Our research offers greater nuance than the common distinction between more or less prejudice that is typically used as an underlying continuum for understanding people's attitude toward Muslim minority groups and the different practices they engage in. By making a distinction between people's feelings toward the group of people and toward a range of out-group practices, it is possible to identify a more complex constellation of evaluations. For some individuals, their (un)acceptance of Muslim practices corresponds to their anti-Muslim feelings, but for others it does not. For the principled intolerant group, generally positive group feelings are associated with positivity toward some Muslim practices with disapproval of other practices (i.e., founding Islamic schools and, to a lesser extent, wearing the headscarf). Moreover, not all Muslim practices are rejected to the same extent which indicates that a relative interpretation of rejection is more appropriate than an interpretation in terms of generalized rejection. These findings indicate that rejection of a particular practice (headscarf or Islamic schools) cannot simply be taken to indicate prejudice toward Muslims, and that acceptance of a particular practice does not have to indicate non-prejudicial feelings. Research on anti-Muslim attitudes has examined the extent to which these attitudes reflect prejudice or specific forms of critique based on the endorsement of secularism, and Enlightenment and universalistic values (e.g., [@c6]; [@c17]; [@c72]). Indeed, our analyses indicate that while some people may not openly express prejudice, their objection appears to be guided by hidden prejudice, while others do not express prejudice and appear to be rejecting specific practices as a function of principled objections. Following research on political tolerance, we have tried to argue and demonstrate that it is also useful and important to consider group-based attitudes together with the acceptance of group-specific practices. One can tolerate certain practices of a disliked minority group and when the practice itself is controversial one can be intolerant toward the group one likes or dislikes ([@c29]; [@c64]). By also evaluating a range of predictors, we were able to identify differences in attitudes, background, and behavioral intentions between the principled and prejudiced intolerant groups. Thus, despite the group identified as prejudiced intolerant showing a relative absence of prejudice in explicit attitudes toward the Muslim outgroup, their consistent rejection of Muslim practices coupled with differences in predictive attitudes suggests that they may harbor implicit prejudice or have prejudice that they are aware of but are unwilling to express (e.g., [@c52]). However, although we found similar patterns in four different datasets we need to be careful about generalizing the specific content and the size of the different profiles. The findings of latent profiled analyses and person-centered approaches are context-specific and sensitive to the practices that are considered. The practices used in this research reflect issues that are broadly debated in Western Europe and the profiles indicate how people tend to group these issues. However, different profiles might emerge if different practices were considered. For instance, the consideration of more demanding issues (e.g., arranged marriage, Sharia ruling in the Netherlands) could result in very skewed distributions of answers with different profiles as a consequence, such as the "liking" and the "principled intolerant" groups not sustain their overall acceptance of various practices ([@c18]). Further research is needed to test whether this pattern is broadly replicated across a range of other Muslim minority practices. Furthermore, although we considered familiar predictors of prejudice and tolerance, it is important to note that our indicators were not explicitly developed for the current analysis. Additionally, while our analysis of predictors of membership in the different groups drew on a wide range of predictors, these predictors did not always use identical items and were not present in all of the four datasets. Therefore, while some predictors have evidence from across multiple datasets, others are present in only one dataset and thus provide weaker evidence. Further research using similar items in new datasets may allow us to further update these findings. In addition, future research could consider other important predictors such as intergroup contact, need for closure, and cultural diversity beliefs as these might provide a further understanding of the differences between the groups of individuals. Lastly, it is important to note that we focused our discussion on the more stable patterns we found in our datasets and not on the differences between the datasets. While differences are relevant and may be informative, it can be difficult to know with any degree of certainty whether they are due to random noise or whether they reflect socio-political factors (see, e.g., our discussion on attitudes toward Turks in Study 4 in the Out-Group Feelings section). Therefore, we focus on between-sample similarities that we find across multiple datasets collected over a 4-year period. This allows us to assess the probability that the results we discuss represent more general patterns of population characteristics. Conclusion {#SS1-10} ---------- In light of the important academic and societal debates on the acceptance and accommodation of Muslims in Western societies, it is critical to parse between forms of intolerance of particular practices that represent more principled positions on complex matters of policy and those which tend to justify the disliking of Muslim minorities. In this research, we found support for both sides of competing perspectives on the source of intolerance and we have explained in more detail how these groups differ. Many majority members are struggling with questions around immigration and Muslim minorities, and the acceptance of dissenting minority practices in particular. Psychologically various types of feelings, beliefs, norms, and values come into play and the weighting and balancing of these considerations against each other is not easy ([@c80]). A social psychological perspective that tries to understand the rejection of specific minority practices only in terms of prejudicial attitudes is limited, as is a perspective that ignores the justification of prejudicial feelings and negative beliefs ([@c8]). Ordinary people are influenced by their group-based likes and dislikes but are also capable of considering different principles and values, including the importance of tolerance. Using a person-centered approach makes it possible to identify unobserved groups of individuals who differ in the particular ways in which they try to combine their out-group feelings and evaluations of a range of out-group practices ([@c39]; [@c42]). These groups cannot be placed on a unidimensional prejudice continuum but rather form latent classes of majority group members who differently combine their general feelings toward Muslim groups and their acceptance of Muslim practices. In this way a more nuanced understanding of majority members' evaluation of minority groups and minority practices can be provided which is critical for the continuing social and theoretical debates. Electronic Supplementary Material {#S6} ================================= The electronic supplementary material is available with the online version of the article at <https://doi.org/10.1027/1864-9335/a000380> 1. **ESM 1**. Questionnaires, Tables, and Mokken analysis. When we refer to "Muslim practices" here, we are referring to practices linked to Muslims in Western Europe and frequently debated in broader society. Thus, this is not to say that the practices are essential or defining Muslim practices, as many of them are widely debated within the Muslim community. In Studies 1--3, the five practices listed here are the only practices included in the surveys. In Study 4, one additional practice asked about the right for Muslims to create political parties. As that item measures political rather than social tolerance, and as that item differed from the other three studies, it was not included in the analyses. **Authorship:** Maykel Verkuyten collected the data; both authors were involved in all other parts of the research. **Open Data:** Complete data and analytic scripts can be found on the Open Science Framework at <https://osf.io/7j2zm/>.

Introduction ============ Pancreatic ductal adenocarcinoma (PDAC), whose 5-year survival rate is as low as 6%,^[@bib1],\ [@bib2]^ is one of the most aggressive malignancies, because the disease is often diagnosed at a late stage, and its treatment options are limited. PDAC has a very poor prognosis.^[@bib3],\ [@bib4],\ [@bib5]^ Therefore, a better understanding of the mechanisms driving the progression of this cancer is needed. Approximately 90% of all PDACs acquire *KRAS* mutations,^[@bib6]^ and the progression of these tumors is also accompanied by an increase in cellular oxidative stress levels.^[@bib7],\ [@bib8],\ [@bib9]^ Mitochondria are the main source of reactive oxygen species (ROS), and their functional state is modified during tumor progression.^[@bib10],\ [@bib11],\ [@bib12],\ [@bib13]^ Mitochondrial ROS play an essential role in cell proliferation and tumorigenesis in PDAC.^[@bib14],\ [@bib15]^ In particular, mitochondrial fragmentation, a phenomenon known as 'fission', is associated with increased energy demands and increased ROS production.^[@bib16],\ [@bib17]^ Mitochondrial fission is also associated with the generation of new organelles. Fission is mainly regulated by dynamin-related protein 1 (DRP1). DRP1 recruitment around mitochondria results in the formation of spirals, which draw together both the inner and the outer mitochondrial membranes to allow mitochondrial division.^[@bib18]^ Conversely, 'fusion', which is required to reduce stress, is regulated by mitofusins 1 and 2 (MFN1/2), which fuse the outer membrane, and optic atrophy 1, (OPA1), which fuses the inner membrane, creates elongated mitochondria.^[@bib19],\ [@bib20],\ [@bib21]^ Metabolic changes in cells lead to the regulation of fission and fusion.^[@bib22],\ [@bib23],\ [@bib24]^ Family with sequence similarity 49 member B (FAM49B) is encoded by a highly conserved gene in mammals. In humans, the *FAM49B* gene is localized on chromosome 8q24, encodes for a 37-kDa protein composed of 324 amino-acid residues,^[@bib25]^ and contains a characteristic DUF1394 domain. Another FAM49B isoform of \~20 kDa lacks the first 123 amino acids due to alternative splicing of its transcript. None of the isoforms contain any other known functional motifs. To date, no functional data regarding this protein have been published, and its role in cancer is unknown. In this study, we investigated the expression and role of FAM49B in PDAC. We demonstrated that FAM49B is highly expressed in PDAC cell lines and that this expression is downregulated *in vivo* by the surrounding tumor environment. In PDAC cells, FAM49B is predominantly localized in the mitochondria, and *FAM49B* gene knockdown leads to oxidative stress that enhances tumor proliferation and invasiveness. Thus, we have identified a novel 'tumor suppressor' gene that links the inflammatory environment to mitochondrial dynamics. Results ======= FAM49B expression in PDAC ------------------------- FAM49B expression levels in PDAC biopsy tissue samples (*n*=93) were analyzed by immunohistochemistry ([Figure 1a](#fig1){ref-type="fig"}). Most of the PDAC biopsies (71%) displayed negative staining for FAM49B ([Figure 1a](#fig1){ref-type="fig"}). However, the remaining biopsies (29%) displayed positive staining for FAM49B ([Figure 1a](#fig1){ref-type="fig"}). We subsequently evaluated the association between FAM49B expression and PDAC patient clinical and pathological characteristics ([Supplementary Table 1](#sup1){ref-type="supplementary-material"}). There were no significant differences in FAM49B expression levels among patients of different ages or survival times ([Supplementary Figure 1A](#sup1){ref-type="supplementary-material"}). However, there was a significant difference in the rate of tumor lymph node (LN) invasion between patients with positive and negative FAM49B expression. Specifically, the rate of LN invasion was much higher in patients who displayed negative FAM49B expression (*P*=0.004) than in patients who displayed positive FAM49B expression ([Figures 1b and c](#fig1){ref-type="fig"}). A similar pattern of low FAM49B expression was observed in the genetically engineered mouse (GEM) strains that spontaneously develop PDAC (LSL-KrasG12D/+, Pdx-1-Cre (KC), LSL-KrasG12D/+, LSL-Tp53R172H/+ and Pdx-1-Cre (KPC))^[@bib26]^ ([Figure 1d](#fig1){ref-type="fig"}). To determine whether FAM49B expression is downregulated during disease progression we examined KC and KPC mouse tissues collected at different time points, namely 3, 6, 9 and 12 months of age (*n*=8/age). FAM49B expression was low in normal pancreatic ducts ([Figure 1d](#fig1){ref-type="fig"}, left panels) and strongly expressed in most of the ADM/ early pancreatic intraepithelial neoplasia (PanIN) lesions ([Figure 1d](#fig1){ref-type="fig"}, mid panels), which represents early stages of the disease. By contrast FAM49B staining was very faint in well-established PDAC ([Figure 1d](#fig1){ref-type="fig"}, right panels). These results suggest FAM49B expression is lost during the progression from early PanIN to adenocarcinoma and the loss of correlates with its invasiveness. FAM49B expression is downregulated by the tumor microenvironment ---------------------------------------------------------------- Microarray analysis of FAM49B mRNA expression showed that FAM49B was expressed at similarly low levels in normal pancreatic duct samples and PDAC biopsy samples from both patients and xeno-transplanted immunosuppressed (xeno-PDAC) mice. In contrast, FAM49B was highly expressed in the Paca44 PDAC cell line ([Figure 2a](#fig2){ref-type="fig"}). Additional mRNA analysis showed that FAM49B expression levels were two- to threefold higher in six human PDAC cell lines (BXPC3, T3M4, PT45, Mia-PA-CA2, CFPAC1 and Hs766T) and elevated to a lesser extent in another two cell lines (L3.6pl and PANC1), as well as in the primary human pancreatic duct epithelial (HPDE) non-tumor cell line ([Figure 2b](#fig2){ref-type="fig"}), compared to normal human pancreatic duct tissues. Western blot analysis confirmed that FAM49B expression levels were higher in all the PDAC cell lines than in the HPDE cell line ([Figure 2c](#fig2){ref-type="fig"}) and that FAM49B expression was not detectable in the normal pancreatic duct ([Supplementary Figure 1B](#sup1){ref-type="supplementary-material"}). Our finding that FAM49B expression levels were higher in PDAC cell lines than in tumor tissues suggests that FAM49B expression is downregulated by the tumor environment. To confirm this hypothesis, we kinetically evaluated FAM49B *ex vivo* expression in orthotopically injected PDAC cells. KPC-derived K8484 murine PDAC cells expressing FAM49B ([Supplementary Figure 1C](#sup1){ref-type="supplementary-material"}) were orthotopically injected into syngeneic mice. After 30 days, the tumors were excised and dissociated, and the cells were analyzed for FAM49B expression ([Figure 2d](#fig2){ref-type="fig"}). On day 0, *ex vivo* mRNA analysis showed that FAM49B transcription was almost completely absent. However, when the K8484 cells were cultured again over 7--14 days, FAM49B expression increased significantly ([Figure 2e](#fig2){ref-type="fig"}). The extracellular matrix (ECM) can interact with tumor cells to influence their cellular behavior, such as migration, adhesion and proliferation. To evaluate the regulation of FAM49B expression by the ECM, we cultured CFPAC1 and T3M4 PDAC cell lines in a three-dimensional (3D) culture, by embedding cells in Matrigel or seeding cells on Matrigel coated plates, as the cell-cell and cell-ECM interactions that characterize this environment more closely mimic those of the natural environment found *in vivo*.^[@bib27]^ After spheroid formation ([Supplementary Figure 1D](#sup1){ref-type="supplementary-material"}), RNA and protein were isolated from the cells and analyzed for FAM49B expression. PDAC cells cultured in 3D displayed an almost complete downregulation of FAM49B compared to PDAC cells cultured on a 2D monolayer ([Figure 2f](#fig2){ref-type="fig"}) or normal HPDE cells ([Figure 2g](#fig2){ref-type="fig"}). Normal HPDE cells did not display the same downregulation of FAM49B when cultured in 2D and 3D conditions ([Figure 2f](#fig2){ref-type="fig"}). PDAC cells cultured on Matrigel coated plates showed a similar trend in the downregulation of FAM49B expression, while no changes were observed for HPDE cells ([Supplementary Figure 1E](#sup1){ref-type="supplementary-material"}). Even the murine PDAC cell line K8484 cultured in 3D conditions showed no detectable levels of FAM49B transcript at both 7 and 14 days ([Figure 2e](#fig2){ref-type="fig"}). To check possibility that downregulation of FAM49B expression in tumor microenvironment was due to accelerated protein degradation, we treated the PDAC cell lines CFPAC1, T3M4, K8484 and normal HPDE cells with the proteasome inhibitor MG132. There was no increase in the levels of FAM49B upon proteasome inhibition both in 2D and 3D cultures ([Supplementary Figures 2A and B](#sup1){ref-type="supplementary-material"}). These data strongly suggest that FAM49B downregulation in tumor cells occur at transcript levels rather than due to its protein stability. FAM49B knockdown enhances PDAC proliferation and invasion --------------------------------------------------------- The low *in vivo* FAM49B expression levels, mentioned above, correlate with higher PDAC LN invasion ([Figure 1c](#fig1){ref-type="fig"}). In addition, culturing *ex vivo*-cultured PDAC cells *in vitro* strongly enhances FAM49B expression. Those data prompted us to investigate whether FAM49B expression limits PDAC progression and invasiveness. We stably knocked down FAM49B in CFPAC1, T3M4 and HPDE cells using a short hairpin RNA (shRNA) for FAM49B (shFAM49B) or a scrambled control (shCTRL) ([Supplementary Figures 2C and D](#sup1){ref-type="supplementary-material"}) and evaluated their proliferation and invasiveness Using MTT and BrdU assays, we observed that both FAM49B-silenced PDAC cell lines proliferated significantly more than shCTRL cells ([Figures 3a and b](#fig3){ref-type="fig"}). Wound healing and Matrigel invasion assays revealed that FAM49B-silenced cells displayed enhanced migration compared to control cells ([Figures 3c and d](#fig3){ref-type="fig"}). Notably, FAM49B-silenced HPDE cells did not display increased proliferation or migration ([Supplementary Figures 3A and B](#sup1){ref-type="supplementary-material"}). Finally, to investigate the effects of FAM49B silencing *in vivo*, we injected shFAM49B or shCTRL CFAPC1 cells into the tail veins of immune compromised mice. The FAM49B-silenced cells spread to the lungs of these mice, where they grew to a much greater extent than the shCTRL cells ([Figure 3e](#fig3){ref-type="fig"}). FAM49B silencing alters mitochondrial dynamics ---------------------------------------------- The intracellular distribution of FAM49B was also investigated. FAM49B-GFP fusion protein were overexpressed in HEK293 cells and subjected to fluorescence protease protection assay, which utilizes digitonin and trypsin digestion. Both treatments eliminated the majority of the cytoplasmic proteins in the above-mentioned cell line, but the cells overexpressing FAM49B-GFP partially retained the fluorescence signal, suggesting that in addition to localizing in the cytoplasm, FAM49B also localizes in organelles ([Supplementary Figure 4A](#sup1){ref-type="supplementary-material"}), particularly in mitochondria, as demonstrated by electron microscopy ([Supplementary Figure 4B](#sup1){ref-type="supplementary-material"}). Furthermore, confocal analysis showed that in CFPAC1 and T3M4 cells, as well as in HPDE cells, FAM49B (green) co-localized with the MitoTracker dye (red), a finding that confirmed its mitochondrial localization ([Figure 4a](#fig4){ref-type="fig"}). Of note, HPDE cells showed greater cytoplasmic FAM49B expression than PDAC cells. Western blot analysis of FAM49B expression confirmed its predominant mitochondrial localization ([Figure 4b](#fig4){ref-type="fig"}). We employed MitoTracker RED staining to assess whether FAM49B silencing affected mitochondrial morphology. We found that shCTRL cells exhibited many more interconnected, elongated and tubular mitochondria than their counterparts, while shFAM49B cells, including HPDE cells, displayed more punctuated mitochondria than other cells ([Figures 5a and b](#fig5){ref-type="fig"}). Mitochondrial dynamics are driven by fission and fusion machinery. DRP1 phosphorylation, a hallmark of fission,^[@bib28]^ was induced in shFAM49B CFPAC1 cells and was increased in shFAM49B T3M4 and HPDE cells ([Figure 5c](#fig5){ref-type="fig"}), whereas fusion-related protein, MFN1 and MFN2 expression,^[@bib29]^ was not modified by FAM49B silencing ([Figure 5c](#fig5){ref-type="fig"}). These data suggest that FAM49B maintains mitochondrial health status, as its absence favors fission in normal and PDAC cells. These findings were also confirmed by our observation of a decrease in mitochondrial membrane potential (Δψm), which is considered a consequence of fission ([Supplementary Figure 4C](#sup1){ref-type="supplementary-material"}), but FAM49B silencing is not associated to the induction of apoptosis as no changes were observed in the cytochrome *c* release or caspase 9 and 3 activation ([Supplementary Figures 4D and E](#sup1){ref-type="supplementary-material"}). In addition, FAM49B re-expression in silenced cells restored the elongated and interconnected mitochondrial morphology ([Supplementary Figure 4F](#sup1){ref-type="supplementary-material"}) suggesting that FAM49B expression regulates the mitochondrial dynamics. FAM49B regulates oxidative stress --------------------------------- As FAM49B affects mitochondrial dynamics, and mitochondria are key regulators of redox balance,^[@bib30]^ we investigated whether FAM49B silencing was accompanied by an increase in intracellular ROS levels and antioxidant pathway activation. Compared to shCTRL cells, FAM49B-silenced PDAC cells displayed much higher levels of intracellular ROS, as well as elevated superoxide dismutase (SOD) activity levels in their mitochondria ([Figure 6a](#fig6){ref-type="fig"}), but not in their cytosol (data not shown). A similar trend was also observed in HPDE cells, wherein ROS levels and SOD activity levels were lower than in PDAC cells ([Figure 6a](#fig6){ref-type="fig"}). Glutathione (GSH) expression levels decreased in shFAM49B cells concurrently with the above increases in ROS levels ([Figure 6b](#fig6){ref-type="fig"}). No significant changes were observed in NADPH oxidase activity ([Figure 6c](#fig6){ref-type="fig"}), which indicated this enzyme is not the source of the increased ROS levels that were detected in shFAM49B cells. As efficient oxidative phosphorylation is associated with increased mitochondrial ROS production,^[@bib31]^ we investigated electron transport chain (ETC) complex activity. shFAM49B cells displayed aberrant complex III activity ([Figure 6d](#fig6){ref-type="fig"}). Consistent with the above findings, the extent of this phenomenon was greater in PDAC cells than in HPDE cells. Mitochondrial oxygen consumption and ATP generation were significantly decreased in FAM49B-silenced PDAC cells, but not in FAM49B-silenced HPDE cells ([Figures 6e and f](#fig6){ref-type="fig"}), suggesting that silencing FAM49B in PDAC cells uncouples electron transport and ATP synthesis. The pentose phosphate pathway (PPP), which is activated in response to intracellular ROS, fuels lipid and nucleotide biosynthesis and sustains antioxidant responses to support cell survival and proliferation.^[@bib32]^ PPP activity levels were increased in FAM49B-silenced PDAC cells but not in FAM49B-silenced HPDE cells ([Figure 7a](#fig7){ref-type="fig"}). ROS act as secondary messengers that favor tumor proliferation and invasion through ERK and AKT signaling.^[@bib14],\ [@bib33],\ [@bib34],\ [@bib35],\ [@bib36]^ Western blot analysis revealed the presence of increased ERK1/2 and AKT phosphorylation in shFAM49B PDAC cells ([Figure 7b](#fig7){ref-type="fig"}), but not in shFAM49B HPDE cells ([Figure 7b](#fig7){ref-type="fig"}), suggesting that the absence of FAM49B significantly upregulated the pro-tumoral functions of ROS in PDAC cells, but not in HPDE cells. FAM49B ablation cooperates with KRAS to increase PDAC invasiveness ------------------------------------------------------------------ As activated *KRAS* is present in \>90% of PDACs,^[@bib37]^ to determine if FAM49B inactivation synergizes with KRAS oncogene activation, FAM49B was silenced the HPDE cell line expressing KRAS^G12V^ (KRAS+shFAM49B). In constitutive KRAS activated HPDE, FAM49B inactivation was able to induce an increased cell proliferation, cell motility, invasiveness, and EMT marker expression ([Figures 7c--g](#fig7){ref-type="fig"}). These KRAS-mutated shFAM49B HPDE cells also displayed a decreased expression of key intracellular junction components, namely, the epithelial marker E-cadherin, and an increased expression of the mesenchymal marker Vimentin ([Figure 7g](#fig7){ref-type="fig"}). By contrast, no changes were observed in EMT marker expression with FAM49B-silenced or not parental HPDE ([Supplementary Figure 3C](#sup1){ref-type="supplementary-material"}). In addition, the mesenchymal-associated transcriptional factor SNAIL was mostly localized into the nuclei of these cells compared to KRAS-mutated shCTRL HPDE cells, which in general displayed a more epithelial phenotype ([Figures 7f and g](#fig7){ref-type="fig"}). These findings indicate that FAM49B inactivation in the presence of abnormal KRAS activity promotes tumor progression. Discussion ========== This study showed for the first time that the FAM49B protein is involved in mitochondrial fission/fusion regulation and acts as a metastasis suppressor in PDAC. Notably, FAM49B expression was downregulated by the PDAC tumor microenvironment, as demonstrated by the results of our *ex vivo* analysis of its transcription in tumor cells injected into mice. FAM49B expression was significantly decreased in cells analyzed immediately after excision and then sharply increased as early as 7 days after culture initiation. Tumor cells exhibit different morphological characteristics when cultured in 3D as opposed to a 2D monolayer, as 3D culture is known to preserve cell proliferation rates, gene expression profiles and drug sensitivities.^[@bib38],\ [@bib39]^ Downregulation of FAM49B expression in PDAC cells cultured in a 3D Matrigel environment provided evidence indicating the ECM interacts with PDAC cells, a critical interaction that subsequently regulates FAM49B expression. Therefore, the dynamic and reciprocal interactions between tumor and stromal cells, as well as between tumor cells and the ECM, are crucial for the downregulation of FAM49B expression, which is associated with PDAC metastasis. Most of the patients who had lower LN invasion percentages were positive for FAM49B, while the patients with higher LN invasion percentages were negative for FAM49B. We show that the loss of FAM49B expression occurring in human PDAC correlates with the invasion of the PDAC. We used two mouse models of spontaneous and autochthonous PDAC, namely KC and KPC that fully recapitulate the human disease, Both mouse models shows similar progressive loss of FAM49B expression during pancreatic lesion progression: FAM49B expression was at lower levels in normal acinar cells consistent with results in human tissues, and increased in acinar-to-ductal metaplasia (ADM) and early PanIN. The cells undergoes metabolic changes, stress and plasticity during the ADM^[@bib40]^ as later diminished FAM49B expression strongly correlated with increasing severity of later PanIN lesions and FAM49B was not detected in the tumor samples. The tumor microenvironment is known to affect the regulation of genes, and the expression other tumor suppressor genes, such as *BRAC1* and *PTEN*, is downregulated in the tumor environment.^[@bib41],\ [@bib42]^ Here, we showed that FAM49B plays a suppressive role in PDAC, as shFAM49B PDAC cells displayed higher proliferation and invasive ability than their counterparts, both *in vitro* and *in vivo*, confirming our initial hypothesis. A unique feature of FAM49B is its prevalent mitochondrial localization. This observation was confirmed by confocal microscopy and western blot analysis. Interestingly, FAM49B silencing resulted in punctuated mitochondria, a finding reflective of an increase in fission, in conjunction with higher DRP1 phosphorylation, which is considered the hallmark of this process. However, MFN1/2 protein expression, which is associated with the fusion process, was not modified by FAM49B silencing, suggesting that the absence of FAM49B has a stronger impact on fission rather than on fusion. Mitochondrial dynamics have been shown to play an important role in maintaining mitochondrial morphology, stress responses, mitochondrial DNA stability, respiratory capacity and cellular apoptosis.^[@bib21],\ [@bib43],\ [@bib44]^ It is worth noting that FAM49B re-expression after FAM49B silencing led to restoration of the elongated and tubular mitochondrial morphology. Another phenomenon induced by FAM49B silencing in non-tumoral and neoplastic pancreatic cells is the modification of mitochondrial ETC complex activity, particularly complex III activity, which results in the uncoupling of this complex from subsequent ATP synthesis. This uncoupled oxidative phosphorylation may explain the increase in mitochondrial ROS, which has occurred in several cell models,^[@bib30],\ [@bib45]^ and the parallel attempts by cells to counter balance this ROS activity through increase in mitochondrial SOD and PPP activity. In addition to buffering ROS, the PPP also promotes tumor cell proliferation, prevents oxidative cell damage and paradoxically transforms the small amounts ROS still present from damage-inducing agents into pro-survival signals.^[@bib46]^ Interestingly, we demonstrated that in the context of KRAS mutations, depleting FAM49B triggered the pro-tumoral functions of ROS, as demonstrated by the increases in ERK1/2 or AKT phosphorylation in PDAC cells. The consequences of triggering the pro-tumoral effects of ROS are higher cancer cell proliferation rates and invasiveness but not in non-tumoral HPDE cells, which has no mutation in the most commonly mutated genes as PDAC cell lines though p53 is nonfunctional and there is loss of the Rb gene.^[@bib47]^ However, when KRAS is constitutively activated in non-tumoral HPDE cells, FAM49B depletion induced an increase in proliferation and invasion. Regulation of 'fusion-fission' mitochondrial dynamics is involved in several physiological and deregulated processes that are classical 'hallmarks' of cancer,^[@bib48]^ such as tumor metastasis.^[@bib10],\ [@bib13]^ Other tumor suppressors that participate in mitochondrial functions have been described previously. The *Fus1* gene, which modulates inflammatory responses and mitochondrial functions, such as ROS production and mitochondrial membrane potential, is downregulated, mutated or lost in most inflammatory thoracic malignancies.^[@bib49]^ A mitochondrial outer membrane channel known as VDAC1 stabilizes HIF-1alpha by inducing ROS accumulation and favors the growth of Ras-transformed MEF, which indicates that the protein plays a tumor suppressive role.^[@bib50]^ Another member of the 'family with sequence similarity' is FAM96A, which has been identified as a pro-apoptotic tumor suppressor in gastrointestinal stromal tumors and increases mitochondria-mediated apoptosis by binding apoptotic protease activating factor 1 (APAF1).^[@bib51]^ In addition, the cyclin-dependent kinase inhibitor 1B (P27) maintains mitochondrial membrane integrity.^[@bib52]^ These data are relevant to our findings regarding a novel tumor suppressor gene that is responsible for regulating mitochondrial dynamics and highlight the role of mitochondrial integrity and metabolism in the promotion of tumor growth and invasion. Collectively, our findings indicate that FAM49B activity is important for preserving mitochondrial integrity and for protecting cells from oxidative stress. FAM49B is an emerging metastasis suppressor in PDAC cells, as its downregulation leads to mitochondria deregulation, increased cellular ROS levels and, consequently, genomic instability and thus benefits cancer cells. FAM49B is involved in regulating tumor metabolism and mitochondria morphology, which has not been previously described, and FAM49B expression is suppressed in PDAC tumors as a selective metabolic growth advantage. An important finding of our study is that *FAM49B* gene is controlled by the tumor microenvironment, which influences tumor metastasis by downregulating FAM49B expression. ECM stiffness has been described as being critical for tumor progression, as it mechanically activates pro-tumorigenic signaling pathways and also alters the metabolism of the cancer cells.^[@bib53],\ [@bib54]^ ECM components negatively regulated FAM49B expression, which we have shown to reprogram the metabolic states of cancer cells by affecting mitochondrial dynamics, and the distribution of mitochondria as a bioenergetics source have been described to enhanced ECM-dependent invasion, migration and metastasis.^[@bib13],\ [@bib55]^ As mitochondrial dynamic changes have an impact on cancer metastatic potential and chemo-resistance,^[@bib10],\ [@bib11],\ [@bib56]^ mitochondrial fission and other vulnerable metabolic functions that are controlled by FAM49B expression may be promising targets of new therapeutic approaches designed to treat PDAC. Materials and methods ===================== Cell culture ------------ PDAC cell lines CFPAC1, MIAPaCa2, BxPC3, PANC1 (obtained from ECACC), T3M4, PT45, L3.6pl, HS766T, HPDE (kindly provided by Dr P Nisticò, Regina Elena National Cancer Institute, Rome, Italy), normal human ductal cells (kindly provided by Dr Lorenzo Piemonti, San Raffaele Hospital, Milan, Italy). K8484 murine PDAC cell line (Kindly provided by Dr K Olive, Columbia University, NY, USA). Cell lines were cultured in DMEM, RPMI and KSF (Lonza, Basel, Switzerland) accordingly and supplemented with 10% FBS (Lonza, Basel Switzerland) plus 50 μg/ml gentamicin (Sigma-Aldrich, St Louis, MO, USA). Tissue microarray analysis -------------------------- For tissue microarrays (TMAs) including a retrospective series of 93 formalin-fixed paraffin-embedded (FFPE) PDACs and 20 non-neoplastic controls were retrieved from the ARC-Net biobank at Verona University Hospital. All specimens were from naïve-treatment patients underwent surgery for curative intent. The materials have been collected under the Program 1885 approved by the Integrated University Hospital Trust (AOUI) Ethics Committee (Comitato Etico Azienda Ospedaliera Universitaria Integrata). Four 1-mm tissue-cores per case were analyzed. Immunohistochemical staining for FAM49B was performed as described.^[@bib57]^ Two pathologists (AS, BR) scored each tissue core in a blinded manner. Cases were considered negative when having absent (score=0) or weak (score=1) staining, and positive when having intermediate (score=2) to strong (score=3) staining. Immunostaining was homogeneous among tissue-cores from the same case. Murine IHC analysis ------------------- Immunohistochemical staining for FAM49B was done on 4-*μ*m-thick FFPE Pancreatic tissues obtained from KC and KPC mouse at 3, 6, 9 and 12 months of age (*n*=8/age) as previously described.^[@bib26]^ Slides were analyzed in double blinded fashion. *In vivo* studies ----------------- Orthotropic injection into the pancreases of C57B16 mice (*n*=6) was performed with 1 × 10^5^ K8484 cells. Thirty days later tumors were collected immediately processed or cultured them *in vitro* after subjected to standard enzymatic disaggregation. For *in vivo* metastasis NOD-SCID IL2Rgamma^null^ (NSG) were injected into the tail vein (i.v.) with shFAM49B or shCTRL cells 1 × 10^5^ (in 0.1 ml PBS), After 28 days, mice were killed, necropsied, and examined for the presence of tumor masses. Mice (*n*=5) were used in each group. All animals were handled in accordance with European and institutional guidelines (Legislative Order No. 116/92, under protocol No 265/2015 PR) and maintained at the animal facility of the Molecular Biotechnology Center, University of Turin, Italy. Three-dimensional culture ------------------------- A total of 1.5 × 10^5^ cells were seeded on transwell apparatuses in six-well plates containing a 1:1 ratio of cell preparation and Matrigel (Corning GF reduced) or were plated on plastic plates pre-coated with Matrigel. For the embedded 3D culture the cells were incubated at 37 °C for 20--30 min to allow the matrix to set before 1.8 ml of medium was added to the culture plates beneath the transwell apparatus. The 3D culture was then allowed to grow at 37 °C for \~7 days or 14 days, and fresh medium was added to the culture every 2 days. Imaging was performed via inverted microscopy. For RNA and protein extraction, the cells were isolated from the Matrigel by dispase (Sigma-Aldrich) digestion for \~2 h. Transfection ------------ Cells were transfected using calcium phosphate precipitation method and Lipofectamine 3000 (Life Technologies, Carlsbad, CA, USA), according to the manufacturer's instructions. Knockdown experiment -------------------- MISSION shRNAs targeting human FAM49B mRNA cloned into the pLKO.1-puro vector were purchased as glycerol stocks from Sigma-Aldrich (SHCLNG-NM_016623). The efficiency of five shRNA with respect to facilitating decreases in FAM49B mRNA levels was assessed by qRT-PCR in transiently transfected cells. The two most efficient shRNAs (TRCN0000168446 and TRCN0000168778) and control shRNA (SHC002, Sigma-Aldrich) were chosen to generate stable cell lines using lentiviral transduction as described.^[@bib58]^ Protease protection assay ------------------------- A GFP tag was added to the C terminus of the FAM49B delta stop codon to facilitate protein fusion using pEGFP plasmids. HEK293T cells were transfected with GFP-fusion FAM49B, FPP assay was performed as described.^[@bib59]^ Quantitative RT--PCR -------------------- Total RNA was extracted using an RNeasy Mini Kit (Qiagen, Hilden, Germany), and reverse transcription was performed using 2 μg of total RNA using an iScript cDNA Synthesis Kit (BioRad, Hercules, CA, USA), according to the manufacturer's instructions. Quantitative RT--PCR was performed using SYBR Green Dye (Life Technologies) and SsoAdvanced Universal SYBR green mix (BioRad) on a Thermal iCycler (BioRad). The PCR reactions were performed in triplicate, and the relative amounts of cDNA were calculated using the ΔΔCT method. Protein lysate and western blot analysis ---------------------------------------- Protein lysate and western blot was performed as described.^[@bib60]^ Migration assay --------------- Wound-healing assay using Ibidi culture inserts (Thistle Scientific Ltd, UK) according to manufacture instructions. Images were acquired along the cell-free zone between time 0 and 24 h every 2 h. Cell migration was assessed using ImageJ software by measuring the area covered by the cells. Cell-free gap encroachment was determined by comparing the results at any time point to time 0. Invasion assay -------------- The cell lines, with or without FAM49B silencing were evaluated with respect to their invasion capacity using transwell Matrigel assay as described.^[@bib58]^ MTT assay --------- Cells were seeded in 96-well plates at a density of 2 × 10^3^ cells/well and synchronized using serum deprivation for 24 h cells were released using complete medium and were treated with 20 μl of MTT solution (Sigma-Aldrich; 5 mg/ml) according to manufacture instructions. BrDu assay ---------- Cells were seeded as MTT assay labeling for BrDu assay was performed by adding 10 μ[M]{.smallcaps} BrDU (BD Biosciences, Franklin Lakes, NJ, USA) according to the manufacturer's instructions. Hoechst was used to stain the nuclei. Image J was used to quantify the numbers of positive cells. Mitochondrial membrane potential analysis ----------------------------------------- Cytofluorometric analysis of the mitochondrial membrane potential was performed on the shCNTRL and shFAM49B cell lines with 2.5 μ[m]{.smallcaps} 5, 5′, 6, 6′-tetrachloro-1, 1′, 3, 3′-tetraethyl- imidacar-bocyanine iodide (JC-1) (Sigma Aldrich). Staining was performed in accordance with the manufacturer's instructions, and 1 μ[m]{.smallcaps} valinomycin (Sigma Aldrich) was used as a control. Electron microscopy ------------------- Electron microscopy was performed on HEK293 cells expressing FAM49B GFP fusion protein. These cells were grown on Thermanox TMX coverslips (NUNC). Embedding and staining were performed as previously described.^[@bib61]^ Proteasome Inhibitor analysis ----------------------------- MG132 (M7449 Sigma-Aldrich) used at 5 μ[m]{.smallcaps}, at 6, 12, 24 and 48 h in 2D culture, For the treatment of 3D culture of PDAC cells and Normal HPDE spheroids were grown till 7 days in Matrigel and treated for 24 h with 5 μ[m]{.smallcaps} of MG132. Microarray analysis ------------------- RNA was obtained from three PDAC primary tumor samples and three adjacent normal pancreas (NP) tissue samples from the Surgery Department of the University of Verona, as well as from eight PDAC xenografts and one PDAC cell line (PaCa44). Microarray analysis was performed as described.^[@bib62]^ Mitochondria staining and analysis ---------------------------------- The mitochondrial structural network staining was done using MitoTracker Red (Invitrogen) according to manufacture's instructions. Images were recorded with an LSM5 Pascal confocal microscope and super-resolution microscopy. The images were analyzed as previously described.^[@bib63]^ Immunofluorescence ------------------ Cells were fixed permeabilized using Image It Kit (Life technologies) Followed with overnight incubation of primary antibodies, at dilution 1:500 and with an Alexa Fluor secondary antibody (Life Technologies) at a dilution of 1:200 for 1 h. Nuclei were stained with Hoechst (Life Technologies). Mitochondrial fraction extraction --------------------------------- The mitochondrial fractions were extracted via sequential-detergent extraction with a cell fractionation kit (Mitosciences MS861, Eugene, OR, USA), according to the manufacturer's instructions. ROS, SOD, NADPH oxidase, ATP and PPP activity measurements ---------------------------------------------------------- We evaluated ROS, SOD, NADPH, ATP and PPP activity levels with the shCTRL or shFAM49B transfected PDAC cell lines CFPAC1, T3M4, and HPDE, as previously described.^[@bib64]^ Mitochondrial respiratory chain activity measurement ---------------------------------------------------- Complex I-III activity levels were measured using non-sonicated extracted mitochondria, as previously reported.^[@bib65]^ Cytochrome c release -------------------- 10 μg of proteins from cytosolic and mitochondrial extracts, prepared as reported above, were probed with an anti-cytochrome c antibody (mouse, clone 6H2.B4; BD Biosciences). Caspase activity ---------------- The activity of caspase 9 and 3 was measuring by incubating 20 μg cell lysates with the respective fluorogenic substrates Ac-LEHD-7-amino-4-methylcumarine (LEHD-AMC) or DEVD-7-amino-4-methylcumarine (DEVD-AMC), as reported.^[@bib66]^ Antibodies and primers ---------------------- Antibodies used for this study are stated in ([Supplementary Table 2](#sup1){ref-type="supplementary-material"}). Primers stated in ([Supplementary Table 3](#sup1){ref-type="supplementary-material"}). Statistical analysis -------------------- All experiments were repeated at least three times. Data are expressed as the mean±s.e.m. Statistical analysis were conducted using the two-tailed *T*-test or ANOVA with GraphPad Software. *P* values ⩽0.05 were considered statistically significant. We thank Roberta Curto for technical support in the *in vivo* experiments, Elena Frola for technical support in the electron microscopy and Dr Radhika Srinivasan for critically reading the manuscript. The nature research language editing service edited this work. This work was supported by grants from the Associazione Italiana Ricerca sul Cancro (5 × mille no. 12182 to AS and FN, IG no. 15257 to FN, IG no. 15232 to AS); University of Turin-Progetti Ateneo 2014-Compagnia di San Paolo (PC-METAIMMUNOTHER to FN and PANTHER to PC), Italian Ministry of Health-Progetti Ricerca Finalizzata (RF-2013-02354892 to FN), Fondazione Ricerca Molinette Onlus, Fondazione Nadia Valsecchi to FN, European Community FP7 Grant Cam-Pac (agreement no: 602783) to AS; FIMP, Ministero Salute (CUP_J33G13000210001) to AS. MSC was supported by Fondazione Ursula e Giorgio Cytron, Torino. **Author contributions** MSC designed and performed the majority of the experiments, interpreted the results, generated the figures and tables and wrote the manuscript; CR designed and performed the metabolic functional assays and interpreted the results; MS performed the CFL and electron microscopy studies; MP, MMS, and CR performed the *in vitro* experiments; CC performed the immunohistochemistry on the mouse tissues; AE performed the biostatistics analysis of the PDAC patients; PA generated the HPDE overexpressing KRAS cell line; AS, RS and BR recruited PDAC patients, performed immunohistochemistry and analyzed results. PC supervised and designed the *in vivo* experiments, interpreted the results and wrote the manuscript; FN supervised the study, interpreted the results and wrote the manuscript. [Supplementary Information](#sup1){ref-type="supplementary-material"} accompanies this paper on the Oncogene website (http://www.nature.com/onc) The authors declare no conflict of interest. Supplementary Material {#sup1} ====================== ###### Click here for additional data file. ###### Click here for additional data file. ###### Click here for additional data file. {#fig1} {#fig2} {#fig3} {#fig4} {#fig5} {#fig6} {#fig7}

Introduction {#sec1-2331216517744915} ============ Hearing loss in older adults is often left untreated ([@bibr4-2331216517744915]; [@bibr16-2331216517744915]), despite abundant evidence showing that hearing aid use improves communication and health outcomes ([@bibr5-2331216517744915]; Mulrow, Tuley, & Aguilar, 1992). The factors associated with low hearing aid uptake that were reported in observational studies were summarized in two systematic reviews ([@bibr21-2331216517744915]; [@bibr31-2331216517744915]). For a number of factors, one or both reviews concluded that they significantly influenced uptake. In particular, a higher likelihood of hearing aid uptake was associated with older age, greater self-reported disability, and greater measured hearing loss ([@bibr21-2331216517744915]; [@bibr31-2331216517744915]). Furthermore, persons who acknowledged more benefits than barriers to hearing aid use, who had no or limited perceptions of hearing aid stigma, who experienced greater social support and greater social pressure to get a hearing aid, and whose family and friends had positive attitudes toward hearing aids were more likely to take up a hearing aid ([@bibr31-2331216517744915]). In both reviews, inconsistent or sparse evidence was found for many other factors. Examples include level of education, manual dexterity, coping style, technology use, cognitive functioning, and employment status ([@bibr21-2331216517744915]; [@bibr31-2331216517744915]). There were also factors that were judged as potentially important based on clinical experience or anecdotal evidence, such as patient--professional interaction (e.g., mismatches in views between clinicians and patients; Laplante-Lévesque et al., 2012) and self-efficacy of hearing aid handling. Both [@bibr21-2331216517744915] and [@bibr31-2331216517744915] mentioned two important caveats in the research performed so far. First, many of the included studies did not use appropriate multivariable prediction modeling. When following the successive steps of building a multivariable model ([@bibr33-2331216517744915]), negligibly weak predictors and confounders (i.e., factors without true predictive value) will be discarded from the model. This is in contrast to a univariate approach, in which these factors could---incorrectly---be considered relevant predictors, based on their univariate statistical significance. In addition, when the final multivariable model has been built, the unique predictive value of each of the included predictors is reflected because each predictor is adjusted for all other predictors with which they share variance. Second, all previous studies had a retrospective, cross-sectional design (i.e., uptake status and predictor factors were measured at the same moment in time). This approach hampered any strong conclusions about the factors' true *predictive* value (see also [@bibr7-2331216517744915]). Since the publication of these two reviews, two observational studies have addressed the caveats outlined earlier. [@bibr32-2331216517744915] retrospectively examined a wide range of audiological and nonaudiological factors in a sample of older Australian hearing aid candidates. Their multivariable model showed that the following factors were associated with hearing aid uptake: relatively positive attitudes toward hearing aids, high self-efficacy of hearing aid handling, high social support toward hearing aids, severe hearing loss, low cognitive reasoning skills, and receiving a pension. [@bibr44-2331216517744915] examined the predictive value of a smaller list of nonaudiological factors in older American veterans and were the first to apply a longitudinal design. Their multivariable model showed that greater baseline hearing loss severity and higher self-reported readiness for change significantly predicted uptake at 6 months' follow-up. In various countries, including the United States, Australia, and The Netherlands, a hearing aid trial period can be part of the hearing aid uptake process. In Australia, whether trials are offered varies from clinic to clinic, while in most states in the United States, a trial period is mandated. In general, during the trial period, clients can return the device for a full or partial refund of the costs. In contrast, in The Netherlands, every client has the right to enter a 2-month trial period *before* having to decide to purchase the hearing aid or not. This evaluation period is noncommittal and free of costs in all cases. To the best of our knowledge, the predictors of entering a hearing aid trial period as a distinct outcome have never previously been analyzed. Understanding of such predictors is especially relevant for health-care systems (such as in The Netherlands) in which a trial period is a standard part of the purchase process, and precedes the purchase of the device. To distinguish the Dutch hearing aid trial period from trial periods in which the hearing aid has already been purchased, we henceforth refer to it as a hearing aid evaluation period (HAEP). In first-time hearing aid users, it is plausible that the decision to enter a HAEP may be predominantly driven by prefitting expectations, while the decision to take up hearing aids may be driven by actual experiences with the device during a HAEP. Examples of factors for which expectations might differ from experience are as follows: hearing aid benefits, wearing comfort, self-efficacy of hearing aid handling, and monetary costs. The fact that expectations may differ from experience was recently supported by [@bibr44-2331216517744915]. They found positive attitude changes in first-time help seekers who decided to take up a hearing aid, presumably as a consequence of their positive experiences with the devices. These changes did not occur in those who did not take one up. Both [@bibr32-2331216517744915] and [@bibr44-2331216517744915] incorporated the health belief model (HBM; [@bibr19-2331216517744915]; [@bibr41-2331216517744915]) into their study designs. The HBM is a common health behavior change theory that can be used to identify the determinants of particular health-behavior changes and is increasingly applied to explain the hearing help-seeking process ([@bibr32-2331216517744915]; [@bibr42-2331216517744915]; [@bibr43-2331216517744915]; [@bibr44-2331216517744915], [@bibr45-2331216517744915]; [@bibr51-2331216517744915]). When we apply the HBM to entering a HAEP (see [Figure 1](#fig1-2331216517744915){ref-type="fig"}), a person would be more likely to enter a HAEP if he or she perceives high *severity* of the consequences of hearing loss and high *susceptibility* to developing a more severe hearing loss. These two factors together cover *perceived threat*. Also, a person who expects relatively many *benefits* (and few *barriers*) would be more likely to enter a HAEP. In addition, strong *internal cues to action* (e.g., self-reported hearing disability) and strong *external cues to action* (e.g., incentives by others, hearing loss as measured through the audiogram) would move people to enter a HAEP. Figure 1.Schematic representation of constructs of the Health Belief Model along with examples of possible predictors of entering a hearing aid evaluation period. Adapted from [@bibr49-2331216517744915]. In the HBM, individual, modifying demographic and sociopsychological variables are defined. These can modify the effects that personal perceptions of *threat*, *benefits*, and *barriers* have on the health behavior change (see [Figure 1](#fig1-2331216517744915){ref-type="fig"}). Examples of such variables are age, gender, personality, and self-efficacy of hearing aid use. [@bibr15-2331216517744915] found that greater hearing loss severity in the better ear was significantly associated with uptake in women, while in men, only less denial and lower concerns over costs were significantly associated with uptake. In other words, they found supportive evidence that gender was a modifying variable for these factors. To our knowledge, no other researchers have examined whether there are individual factors that act as effect modifiers. Instead, individual factors (such as age and gender) were studied as *determinants* of uptake. We argue that it would be important to examine whether gender and age also modify the effects that predictors have on entering a HAEP. With increasing age, changes in societal participation (e.g., retirement) and health (e.g., more chronic diseases, more cognitive problems) occur ([@bibr6-2331216517744915]; [@bibr13-2331216517744915]). We hypothesize that the predictive strength of factors like having paid work and self-reported hearing disability decrease with increasing age relative to factors like social leisure activity, comorbidity, and cognitive function (their strength would increase with increasing age). In addition to age and gender, predictors of HAEP may be modified by a person's degree of readiness to do something about their hearing. Readiness for behavior change originates from the stages construct of the transtheoretical stages of change model (TTM; [@bibr39-2331216517744915]; [@bibr40-2331216517744915]). The TTM assumes that people progress via various stages toward adopting and maintaining a particular behavior change. For the purposes of this article, only the first three stages are relevant: (a) precontemplation (problem denial), (b) contemplation (problem awareness and ambivalence regarding the pros and cons of change), and (c) action (healthy behavior acquisition). Previous studies have shown that the stages are predictive of various hearing help-seeking outcomes ([@bibr18-2331216517744915]; Laplante-Lévesque, Hickson, & Worrall, 2012, 2013; [@bibr43-2331216517744915], [@bibr44-2331216517744915]). However, the question has never been studied whether the effects of certain predictors are different for persons with low intrinsic readiness (i.e., high precontemplation relative to contemplation and action) as opposed to persons with high intrinsic readiness (i.e., low in precontemplation relative to contemplation and action). Taking the HBM and the TTM as guiding frameworks, the current study aims to identify the predictors of entering a HAEP in a large sample of older adults eligible for hearing aids who had never tried a hearing aid before. This was done by using a prospective study design, by including a wide range of candidate predictors, applying multivariate prediction modeling, and studying effect modification by age, gender, and readiness for change. Methods {#sec2-2331216517744915} ======= Sample and Procedures {#sec3-2331216517744915} --------------------- The target population of this study comprises older adults without complex hearing problems, as defined in the Dutch field norm 'Nationaal Overleg Audiologische Hulpmiddelen (NOAH)-4' ([@bibr36-2331216517744915]). Roughly, all hearing losses except for presbyacusis are considered complex. Examples of complex hearing problems include psychosocial problems due to hearing loss, poor speech recognition in quiet, and bothersome tinnitus. In The Netherlands, all hearing aids are obtained via a hearing aid dispenser (HAD). The official hearing aid prescription for older adults without complex hearing problems is carried out by Ear, Nose, and Throat (ENT) specialists, but this population is also allowed to directly visit a HAD for eligibility assessment and subsequent start of a HAEP. Whether a formal prescription by an ENT specialist is necessary depends on the requirements of the patient's health insurance company. In the HAD practice that participated in the study, care usually consisted of the following: a preparatory appointment (screening audiogram and assessment of interest regarding pursuit of a hearing aid), an intake appointment (speech audiometry and comprehensive pure tone audiometry; assessment of client needs; choice of a hearing aid), a fitting appointment, one or more fine-tuning appointments, and a purchase appointment (if applicable). The HAEP formally starts at the time of the fitting appointment. Usually, the cost of a hearing aid is partially (∼75%) reimbursed by a person's health insurance if the pure tone average hearing threshold is 35 dB or greater, averaged across 1, 2, and 4 kHz in the ear to be aided. Therefore, the following inclusion criteria were applied for participant inclusion: self-initiated consultation with an ENT specialist or a HAD for a hearing loss assessment, absence of medical problems hindering hearing aid use (e.g., skin allergy), no prior hearing aid use, a minimum pure-tone average hearing threshold of 35 dB HL averaged across 1, 2, and 4 kHz in the ear(s) to be aided, and age 55 years or older. Specialists from six ENT departments recruited participants. ENT patients were invited to participate in the study by their specialist at the end of their consultation. Patients who expressed interest in the study were phoned by the researchers 1 to 3 days later to further describe the study. Subsequently, a postal package was sent, consisting of a participant information letter, an informed consent form, a questionnaire, an instruction sheet to perform the telephone speech-in-noise test (see Candidate predictors), and a stamped return envelope. The research team recruited HAD clients originating from 118 HAD practices for the study. Clients were approached via a postal invitation shortly (around 3 to 5 days) after they had visited the HAD for their preparatory appointment. Their postal package was sent directly along with this invitation. We invited roughly equal numbers of clients who did and did not have an intake appointment scheduled, as we knew from the HAD that having an intake appointment scheduled was strongly associated with entering a HAEP. We thus attempted to recruit a sample with optimal spread in HAEP status. During recruitment, it was made clear to the participants that they did not need to know or disclose yet whether they wanted to pursue a hearing aid fitting. All participants were asked to fill out the questionnaire within 1 week of receipt and before their fitting appointment, if applicable. Four months after recruitment, participants were phoned and asked if they had entered a HAEP or not since baseline. Since there were no significant waiting lists, 4 months was considered sufficient to move from having been pronounced eligible for hearing aid fitting to entering a HAEP. [Figure 2](#fig2-2331216517744915){ref-type="fig"} shows the participants' flow through the study. The data of 110 ENT patients and 267 HAD clients could potentially be included in the analyses. Informed consent was obtained from all participants. Written approval for the study (name: PredictEAR, reference number: 2013.464) was obtained from the Dutch Institutional Review Board of the VU Medical University Center, Amsterdam (registered with the US Office for Human Research Protections as IRB00002991; FWA number: FWA00017598). Figure 2.Participants' flow through the study. Outcome Measure {#sec4-2331216517744915} --------------- HAEP status was defined as did or did not enter a HAEP between baseline and 4 months' follow-up. Seven participants who had the hearing aid fitting appointment scheduled at the time of the follow-up interview but who had not been fitted yet were regarded as having entered a HAEP. Candidate Predictors {#sec5-2331216517744915} -------------------- Given the conceptual overlap between hearing aid uptake and entering a HAEP as outcomes, the choice of the candidate predictors originates from the influencing factors for uptake as reviewed by [@bibr21-2331216517744915] and [@bibr31-2331216517744915]. Below, 'likely' and 'novel' candidate predictors are described. This categorization was used in the analyses (see Statistical Analyses section) and was based on the empirical evidence for the particular factor being associated with hearing aid uptake. If one or both of the reviews concluded that there was strong evidence that a given factor influenced uptake, that factor was labeled a 'likely' predictor of entering a HAEP. In line with this, the following six predictors were regarded as 'likely' predictors: higher age, greater hearing loss severity, greater self-reported hearing disability, greater hearing aid stigma, and greater social pressure to get a hearing aid. The rationale for including each of the 'likely' predictors was already provided in the Introduction section. How the predictors were measured in the current study is described later. For 'novel' predictors, the same reviews found that the evidence was sparse (i.e., only one study had shown a significant association with uptake), inconsistent (roughly equal numbers of studies showed significant associations and nonsignificant associations with uptake), or absent (no studies reported). The evidence found in the later studies by [@bibr44-2331216517744915] and [@bibr32-2331216517744915] did not alter the categorization of 'likely' versus 'novel'. For the 'novel' predictors, the rationale for including them in the current study is provided in the Appendix (see Supplementary Material). How the predictors were measured is described later. ### Likely predictors {#sec6-2331216517744915} Unless stated otherwise, variables were collected via self-report (the postal questionnaire) and were analyzed as continuous variables. -- *Age* was included as age in years and was calculated from self-reported date of birth.-- *Hearing loss severity* was defined as the better-ear, pure-tone average hearing threshold (in dB HL) across 1, 2, and 4 kHz (3F-BEA). Audiograms were retrieved via ENT specialists or HADs. Air conduction hearing thresholds were obtained via a modified Hughson--Westlake procedure for octave frequencies between 250 and 8000 Hz.-- *Self-reported hearing disability* was measured using the Dutch, validated, 28-item version of the Amsterdam Inventory for Auditory Disability and Handicap ([@bibr22-2331216517744915]; [@bibr3-2331216517744915]). Scores can range from 0 to 74 (higher scores indicate more severe hearing disability). The scale showed an excellent internal consistency in our study sample (Cronbach's α = .95).-- *Benefits of hearing aids* (in short: *benefits*), *Hearing aid stigma* (in short: *stigma*), *Sound quality and cost of hearing aids*, *Social pressure to get a hearing aid* (in short: *social pressure*), and *Evaluation of hearing aids by others* were measured using subscales of the Attitude Questionnaire (AQ; [@bibr50-2331216517744915]). The AQ was developed based on the HBM and validated on older Dutch hearing-impaired individuals ([@bibr50-2331216517744915]). *Benefits* included expected benefits of hearing aid use and were measured using the 10-item Benefits subscale. Scores can range from 10 to 50 (higher scores indicate greater benefit). *Stigma* was measured using the six-item Stigma subscale. Scores can range from 6 to 30 (higher scores indicate greater stigma). Expected barriers with regard to *sound quality* and *cost of hearing aids* were measured using the three-item Sound subscale. Scores can range from 3 to 15 (higher scores indicate more expected barriers). *Social pressure* was measured using the five-item Social Pressure subscale. Scores can range from 5 to 25 (higher scores indicate greater social pressure). The *evaluation of hearing aids by others* was measured using the three-item Evaluation of Aid subscale*.* The scale measures the respondent's perception of others' evaluation of the pros and cons of hearing aids in general (two items), and whether the respondent thought others would discourage him or her to get a hearing aid (one item). Scores can range from 3 to 15 (higher scores indicate more negative evaluation of hearing aids by others). Four AQ subscales showed a reasonable to good internal consistency in our study sample. Cronbach's αs were .89 (Benefits), .83 (Stigma), .82 (Social Pressure), and .67 (Evaluation of Aid). The Sound subscale showed an unacceptable, low internal consistency (Cronbach's α = .18) and was therefore excluded from the analyses. ### Novel predictors {#sec7-2331216517744915} -- *Passive acceptance of hearing loss* was measured by the two-item AQ subscale Passive Acceptance ([@bibr50-2331216517744915]). It refers to the perceived lack of need to do something about hearing problems in old age. Scores can range from 2 to 10 (higher scores indicate greater passive acceptance). The Spearman--Brown coefficient of this scale was 0.27 in our sample.-- *Precontemplation*, *Contemplation*, and *Action* were measured using the three subscales of the University of Rhode Island Change Assessment of the same name (URICA; [@bibr29-2331216517744915]). Each subscale consists of eight items and refers to a separate stage of change. The scores on each subscale can range from 8 to 40 (higher scores indicate a higher weight on the particular stage). The URICA was translated and validated for a Dutch nonclinical adult population by [@bibr8-2331216517744915]. Items in the original URICA refer to "the problem," which can be replaced by a specific health problem. For the purposes of this study, these items were adapted for hearing health behaviors by two of the authors (MP and SK) following the method applied by [@bibr25-2331216517744915] for the English URICA. [@bibr25-2331216517744915] showed good psychometric properties for all the URICA subscales in an older audiological population. The Precontemplation, Contemplation, and Action subscales showed reasonable to good internal consistency within our study sample. Cronbach's αs were .68, .67, and .85, respectively.-- *Readiness for change* (in short: *readiness*) was determined by calculating the readiness composite score. This was done by adding a participant's Contemplation and Action scores and subtracting his or her Precontemplation score (see [@bibr25-2331216517744915]). Scores could range from −24 (lowest readiness) to +72 (highest readiness).-- *Speech-in-noise recognition* was measured using the digit triplet speech-in-noise test administered over the telephone ([@bibr46-2331216517744915]). The test determines an individual's speech-reception threshold in noise (SRT~n~) defined as the signal-to-noise ratio (SNR) in dB SNR corresponding to 50% intelligibility. It correlates highly (*r* = .87) with the standard Dutch sentences speech-in-noise test ([@bibr38-2331216517744915]; [@bibr46-2331216517744915]) and has shown satisfactory test--retest reliability in a sample of older participants (intraclass correlation coefficient = .70; [@bibr35-2331216517744915]).-- *Age at onset of hearing problems* was measured by the item "At what age did you first notice your hearing problems? When I was ... years old."*Maladaptive behavior*, *Verbal strategies*, *Nonverbal strategies*, *Self-acceptance*, *Acceptance of loss*, and *Stress and withdrawal* included coping behaviors and were measured using the six subscales of the 35-item Dutch short form Communication Profile for the Hearing Impaired (CPHI; [@bibr34-2331216517744915]). The six subscales reflect the degree of particular hearing coping behaviors that a person applies (i.e., use of communication strategies and personal adjustment). The CPHI was validated on Dutch hearing-impaired adults and has good psychometric properties ([@bibr34-2331216517744915]). The Maladaptive Behavior subscale has seven items and scores can range from 7 to 35 (higher scores indicate less use of maladaptive behaviors such as avoiding conversations). The Verbal Strategies subscale has seven items and scores can range from 7 to 35 (higher scores indicate more use of verbal strategies such as asking for repetition). The Nonverbal Strategies subscale has five items and scores can range from 5 to 25 (higher scores indicate more use of nonverbal strategies such as positioning in a room). The Self-Acceptance subscale has four items and scores can range from 4 to 20 (higher scores indicate more self-acceptance of oneself with hearing loss). The Acceptance of Loss subscale has three items and scores can range from 3 to 15 (higher scores indicate more acceptance of hearing loss). The Stress and Withdrawal subscale has nine items and scores can range from 9 to 45 (higher scores indicate less stress and social withdrawal). All CPHI subscales showed acceptable to excellent internal consistency in our study sample. Cronbach's αs were .78 (Maladaptive Behavior), .76 (Verbal Strategies), .82 (Nonverbal Strategies), .73 (Self-Acceptance), .72 (Acceptance of Loss), and .91 (Stress and Withdrawal).*Self-efficacy of hearing aid handling* (in short: *self-efficacy*) was measured using the seven-item Basic Hearing Aid Handling subscale of the Measure of Audiologic Rehabilitation Self-Efficacy for Hearing Aids (MARS-HA). It possesses good psychometric properties ([@bibr54-2331216517744915]). Scores can range from 0 to 100 (higher scores indicate greater self-efficacy). The questionnaire was translated from English to Dutch for the purposes of this study using the forward-backward method ([@bibr2-2331216517744915]). Six persons were involved in this process. Excellent internal consistency was found in our study sample (Cronbach's α = .96).*Prompting consultation* referred to the form of the primary prompt for the initial visit to the HAD or ENT. It was defined as (a) self-initiated or encouraged by family, (b) referred or directed by another health-care professional (general practitioner or Audiological Center or ENT specialist or HAD), or (c) prompted by the HAD (i.e., advertisement: postal or media invitation).*Agreement or discrepancy in views between the participant and the health-care professional about the necessity of a hearing aid* (in short: *Agreement or discrepancy*) was measured using an item that was based on excerpts collected in a study by [@bibr24-2331216517744915]. Responses were collected for the following statement: "My health care provider and I both think I need a hearing aid." Response categories were "no, (s)he thinks I need a hearing aid, but I do not"; "yes, we both think I need a hearing aid"; "no, we both think I do not need a hearing aid"; "no, (s)he thinks I do not need a hearing aid, but I do"; "I do not know what (s)he thinks about the benefit of a hearing aid for me"; and "I don't find any HCP's opinion important. I know best how severe my hearing problems are."*Level of education* (highest level completed) was categorized into: low (uncompleted elementary, elementary, lower vocational), medium (general intermediate, intermediate vocational, general secondary, higher vocational), and high (college and university).*Hours of paid work* were categorized into 0 hours per week, 1 to 20 hours per week, or 21 hours or over per week.*Living situation* was defined as currently living with other people in the household or not.*Country of birth* was dichotomized into the Netherlands or other.*Social network size* (family and friendship networks) was measured using the six-item Lubben's Social Network Scale, which has shown good psychometric properties in older European community*-*dwelling adults ([@bibr27-2331216517744915])*.* Scores can range from 0 to 30 (higher scores indicate larger social networks). The scale showed good internal consistency in our study sample (Cronbach's α = .86).*Social participation* was measured using the nine-item Maastricht Social Participation Profile, which has been validated in older Dutch persons with chronic conditions ([@bibr28-2331216517744915]). It measures frequency and diversity of consumptive (e.g., eating out) and formal social participation (e.g., organized day outings). Total scores can range from 0 to 27 (higher scores indicate higher levels of social participation). The scale showed reasonable internal consistency in our study sample (Cronbach's α = .68).*Personal computer (pc) use* was dichotomized into using a pc (desktop or laptop or palmtop or iPad or tablet) or not.*Comorbidity of chronic diseases or conditions* (in short: *comorbidity*) was measured using the question: "Besides your hearing problems and possible vision problems or osteoarthritis, do you have any *other* longstanding illnesses or conditions?" Response options were *no* or *yes*. This item was based on an item that is used in the EU-Statistics on Income and Living Conditions instrument ([@bibr12-2331216517744915]).*Self-rated health* was measured using the question "How is your health in general?" ([@bibr52-2331216517744915]). Response options were *very good* (0), *good* (1), *fair* (2), *sometimes good, sometimes poor* (3), or *poor* (4).*Osteoarthritis of the hands* was assessed using the question "Do you suffer from osteoarthritis (degenerative arthritis) in your hands?" Response options were *no* or *yes*.*Vision status* was measured using two items (near- and far-sighted vision) of the Organisation for Economic Co-operation and Development disability indicator ([@bibr30-2331216517744915]). The summed score can range from 0 to 6 (worst to best vision). The Spearman--Brown coefficient of this scale was 0.42 in our sample.*Mastery* was measured by the abbreviated, five-item version of the Pearlin Mastery Scale ([@bibr37-2331216517744915]). The Pearlin Mastery Scale has shown good psychometric properties in adult samples ([@bibr37-2331216517744915]; [@bibr11-2331216517744915]). Scores can range from 5 to 25 (higher scores indicate a higher sense of mastery). The scale showed good internal consistency in our study sample (Cronbach's α = .82).*Cognitive dysfunction* was measured by the 20-item Cognitive Dysfunction Questionnaire ([@bibr53-2331216517744915]). Scores could range from 20 to 100 (higher scores indicate higher cognitive dysfunction). The questionnaire was translated from Swedish into Dutch using the forward-backward method ([@bibr2-2331216517744915]). Five persons were involved in this process. Excellent internal consistency was found in our study sample (Cronbach's α = .90). Candidate Effect Modifiers {#sec8-2331216517744915} -------------------------- *Gender*, *age*, and *readiness for change* were regarded as candidate effect modifiers. *Age* and *readiness for change* were analyzed as continuous variables. Statistical Analyses {#sec9-2331216517744915} -------------------- ### Descriptive analyses {#sec10-2331216517744915} For each scale, the internal consistency was calculated using the full study sample. The Cronbach's α was calculated for scales with three or more items and the Spearman--Brown coefficient for scales with two items ([@bibr9-2331216517744915]). Mean values and standard deviations (*SD*s), stratified by HAEP status, were calculated for continuous candidate predictors that followed a normal distribution. For continuous candidate predictors that followed a skewed distribution, medians and 25th and 75th percentile points were calculated. For dichotomous or categorical candidate predictors, as well as for the effect modifier *gender* and the corrective factor *source of recruitment* (defined later), proportions across HAEP status were calculated. Further, we tested the univariate associations between HAEP status and each of candidate predictors, *gender* and *source of recruitment*. ### Collinearity analysis {#sec11-2331216517744915} Collinearity between candidate predictors was examined by calculating correlations and the predictors' variance inflation factors for multicollinearity. This was done both for the final and for the starting prediction model before the backward selection procedure (see Prediction modeling section). The highest correlation found was .50 (between *benefits* and *social pressure*). All variance inflation factor values were well below 5, indicating no relevant collinearity. ### Prediction modeling {#sec12-2331216517744915} Logistic multivariable regression models were built. Odds ratios (ORs) for each predictor were determined. When the OR was larger than 1, the OR indicated the increase in odds to enter a HAEP for 1 point increase in the predictor score. When the OR was smaller than 1, the OR indicated the decrease in odds for 1 point increase in the predictor score. Five steps were followed to build the models which are described later. A two-step approach (Steps 1 and 2) was chosen to take into account the previously obtained evidence on 'likely' predictors and build further on this evidence base. The variable *source of recruitment* was kept in the model in each step (forced) to adjust all predictors for this factor. Three sources of recruitment were defined: via an ENT department, via a HAD (persons who had an intake appointment scheduled), or via a HAD (persons who had no intake appointment scheduled). The adjustment for this factor was done because the different sources showed different univariate associations with HAEP status (see [Table 2](#table2-2331216517744915){ref-type="table"}). *Step 1*: *Determination of the basic model*. In this step, it was evaluated whether the seven 'likely' predictors were significant predictors of HAEP status. This was done by including the predictors in the starting model and performing a backward selection of variables using a *p~removal~* of .157. With this criterion, there is less optimism in regression coefficients as compared with a more strict α of .05 ([@bibr33-2331216517744915]). The α of .157 corresponds to the use of the AIC for predictor selection, which is a method that accounts for model fit while penalizing for the number of parameters being estimated (see [@bibr33-2331216517744915]).*Step 2: Determination of the extended model*. In this step, it was evaluated whether 'novel' predictors significantly added predictive value to the basic model. This was done by creating reclassification tables (which stratify individuals into risk categories of high risk or low risk of showing the outcome in question) and examining changes in categorization under a new model. Twenty-nine new models were created by separately adding the 29 'novel' candidate predictors to the basic model. By evaluating the statistical significance (*p* \< .05) of the Net Reclassification Improvement (NRI; [@bibr47-2331216517744915]), it was determined whether participants whose HAEP status were wrongly predicted by the basic model shifted to the correct outcome when a 'novel' predictor was added to the model. The NRI reflects the net percentage of improved classification (percentage improved minus percentage worsened). A threshold of 66% was used to distinguish between low and high risk in the reclassification tables because this was the prevalence of entering a HAEP within our study sample (see Descriptives subsection). The prevalence of the outcome can be used as a threshold when no clinically well-accepted threshold is available ([@bibr48-2331216517744915]). When more than one 'novel' predictor showed statistically significant added predictive value, the strongest 'novel' predictor (i.e., the one with the highest NRI) was added to the basic model first. Then, the second strongest predictor was added, after which the new NRI was tested. This procedure was repeated for all significant 'novel' predictors.*Step 3*: *Testing effect modification*. Effect modification by *gender*, *age*, and *readiness* was tested for each of the predictor variables that was included in the extended model. This was done by separately including an interaction term for each of the predictor variables (e.g., *benefits* × *gender*) and determining its statistical significance (*p* \< .05).*Step 4*: *Determination of model performance.* The variance explained by the final prediction model was determined by calculating the Nagelkerke *R*^2^. The discriminatory power was determined by assessing the area under the curve of the receiver operator characteristic.*Step 5*: *Determination of internal validity*. The final prediction model was internally validated using bootstrapping. This procedure results in an adjusted linear predictor, which is presented. The adjusted linear predictor includes the predictors' effect sizes adjusted for the model's shrinkage factor ([@bibr33-2331216517744915]). Internal validation via preshrinkage is appropriate because prognostic models usually perform better in the patients who were used to build the model than in new patients, due to optimism in regression coefficients and performance measures ([@bibr33-2331216517744915]). The bootstrapping procedure was carried out using the statistical software package R. For all other analyses, SPSS Statistics version 22 was used. Results {#sec13-2331216517744915} ======= Descriptives {#sec14-2331216517744915} ------------ Of the full sample (*n* = 377), 129 (34%) participants did not enter a HAEP at 4 months' follow-up against 248 (66%) who did. The mean *age* of participants was 72.6 years (*SD* = 8.0), and 60% were men. Their *level of education* was reported as low (*n* = 90, 24%), medium (*n* = 238, 63%), and high (*n* = 48, 13%). *Hearing loss severity* ranged from 10 to 68 dB HL (mean: 39.8, *SD* = 9.1) and SRT~ns~ ranged from −7.4 dB SNR to +5.0 dB SNR. [Table 1](#table1-2331216517744915){ref-type="table"} shows the distribution of the participants across the candidate predictor variables and effect modifiers and across the *sources of recruitment*. Univariate associations are displayed in [Table 2](#table2-2331216517744915){ref-type="table"}. Except for *age* and *stigma*, significant univariate associations with entering a HAEP were found for all 'likely' predictors (higher odds were associated with greater *hearing loss severity*, greater *self-reported hearing disability*, more *benefits*, greater *social pressure*, and more positive *evaluation of hearing aids by others*). Some of the 'novel' predictors also showed a significant association with HAEP status (see [Table 2](#table2-2331216517744915){ref-type="table"}). Table 1.Characteristics of the Study Sample (*n* = 377).*n* = 377Candidate predictors and effect modifiers (possible range)Did not enter a hearing aid evaluation period (*n* = 129)Entered a hearing aid evaluation period (*n* = 248)mv*n* (%^[\*](#table-fn2-2331216517744915){ref-type="table-fn"}^)Mean (*SD*) or median \[25th; 75th percentile\]*n* (%^[\*](#table-fn2-2331216517744915){ref-type="table-fn"}^)Mean (*SD*) or median \[25th; 75th percentile\]Likely predictors Age^[⋄](#table-fn1-2331216517744915){ref-type="table-fn"}^012972.7 (8.2)24872.5 (8.3) Hearing loss severity, 3F-BEA012937.1 (8.6)24841.2 (9.0) Self-reported hearing disability (0--74)1512121.1 (12.8)24131.7 (14.2) Benefits (10--50)812834.4 (5.7)24140.0 (5.0) Stigma (6--30)112913.7 (4.7)24714.3 (5.0) Social pressure (5--25)512914.4 (3.8)24817.8 (3.4) Evaluation of hearing aids by others (3--15)101277.7 (1.9)2407.1 (2.1)Novel predictors Passive acceptance (2--10)61284.0 (1.3)2433.9 (1.4) Precontemplation (8--40)1612717.7 (5.4)23415.4 (5.4) Contemplation (8--40)1712727.9 (5.5)23331.4 (4.9) Action (8--40)1412526.6 (7.3)23832.4 (5.5) Readiness^[◊](#table-fn1-2331216517744915){ref-type="table-fn"}^ (−8 to 70)2412537.1 (15.5)22848.4 (11.1) Speech-in-noise recognition, SRT~n~17124−3.3 (2.5)236−2.7 (2.6) Age of onset of hearing problems1111865 \[56.5--72.0\]24865 \[58.0--70.0\] Maladaptive behavior (7--35)112933.0 \[31.0--34.0\]24731.0 \[31.0--34.0\] Verbal strategies (7--35)012914.4 (4.0)24817.2 (5.2) Nonverbal strategies (5--25)112914.5 (5.1)24715.8 (4.9) Self-acceptance (4--20)012917.3 (2.9)24816.6 (3.1) Acceptance of loss (4--15)1112811.5 (2.6)23810.6 (2.6) Stress and withdrawal (9--45)312836.0 (7.0)24632.5 (7.1) Self-efficacy of hearing aid handling (0--100)512767.0 \[55.0--70.0\]24567.0 \[57.5--70.0\] Prompting consultation4  self-initiated or encouraged by family97 (44%)--123 (56%)  referred by another HCP/prompted   by the HAD29 (19%)124 (81%)-- Agreement/ discrepancy22  R does not know what HCP thinks37 (58%)--27 (42%)--  disagreement---HCP: HA necessary;  P: HA not necessary16 (64%)--9 (36%)--  disagreement---HCP: HA not necessary;  P: HA necessary8 (62%)--5 (38%)--  agreement---HCP & P: HA necessary26 (12%)--187 (88%)--  agreement---HCP & P: HA not necessary23 (85%)--4 (15%)--  R only finds own beliefs about HA needs   important7 (50%)7 (50%) Level of education1  low27 (30%)--63 (70%)--  medium82 (35%)--156 (65%)--  high19 (40%)--29 (60%)-- Hours of paid work5  0 hours per week110 (35%)--204 (65%)--  1--20 hours per week12 (52%)--11 (48%)--  21 hours or over per week7 (20%)--28 (80%)-- Living situation  alone in the household439 (38%)--63 (62%)--  with others in the household87 (32%)--184 (68%)-- Country of birth  The Netherlands120 (34%)--236 (66%)--  other country18 (40%)--12 (60%)-- Social network size (0--30)528717.8 (10.0)24016.6 (8.8) Social participation (0--21)81256.7 (4.3)2446.2 (4.0) Pc use  no125 (36%)--44 (64%)--  yes103 (34%)--204 (66%)-- Comorbidity  no776 (35%)--141 (65%)--  yes51 (33%)--102 (67%)-- Self-rated health (0--4)41271.1 (0.7)2461.2 (0.8) Osteoarthritis in hands  no588 (33%)--180 (67%)--  yes38 (37%)--66 (63%)-- Vision status (0--6)21276.0 \[5.0--6.0\]2486.0 \[5.0--6.0\] Mastery (5--25)612710.2 (3.8)24410.4 (4.0) Cognitive dysfunction (20--100)712730.5 (8.2)24831.3 (8.5)Effect modifier^[◊](#table-fn1-2331216517744915){ref-type="table-fn"}^ Gender0  men70 (31%)--155 (69%)--  women59 (63%)--93 (37%)--Corrective factor Source of recruitment0  ENT25 (23%)--85 (77%)--  HAD---intake appointment planned   at baseline22 (14%)--132 (86%)--  HAD---no intake appointment planned   at baseline82 (73%)--31 (27%)--[^1][^2][^3] Table 2.Univariate Associations (Odds Ratios).Odds to enter a hearing aid evaluation period[\*](#table-fn4-2331216517744915){ref-type="table-fn"}Candidate predictors and effect modifiers (possible range / reference category)OR95% CI*p*Likely predictors Age^[◊](#table-fn5-2331216517744915){ref-type="table-fn"}^1.000.97--1.020.847 Hearing loss severity, 3F-BEA**1.051.03--1.08\<0.001** Self-reported hearing disability (0--74)**1.061.04--1.08\<0.001** Benefits (10--50)**1.221.16--1.28\<0.001** Stigma (6--30)1.030.98--1.070.258 Social pressure (5--25)**1.301.21--1.40\<0.001** Evaluation of hearing aids by others (3--15)0.86**0.77--0.960.008**Novel predictors Passive acceptance (2--10)0.910.78--1.060.241 Precontemplation (8--40)**0.930.89--0.96\<0.001** Contemplation (8--40)**1.141.09--1.19\<0.001** Action (8--40)**1.151.11--1.20\<0.001** Readiness ^[◊](#table-fn5-2331216517744915){ref-type="table-fn"}^ (−24 to +72)**1.071.05--1.09\<0.001** Speech-in-noise recognition, SRT~n~**1.101.00--1.200.042** Age of onset of hearing problems1.000.98--1.010.796 Maladaptive behavior (7--35)**0.930.88--0.990.029** Verbal strategies (7--35)**1.141.08--1.20\<0.001** Nonverbal strategies (5--25)**1.051.01--1.100.020** Self-acceptance (4--20)**0.930.86--1.000.059** Acceptance of loss (4--15)**0.870.80--0.950.002** Stress and withdrawal (9--45)**0.930.90--0.96\<0.001** Self-efficacy of hearing aid handling (0--100) Prompting consultation (reference = self-initiated  or encouraged by fam.)  referred by another HCP/prompted by the HAD3.372.08--5.47**\<0.001** Agreement/ discrepancy (reference = R does not know  what HCP thinks)**\<0.001**  disagreement---HCP: HA necessary; P: HA not necessary0.770.30--2.000.593  disagreement---HCP: HA not necessary; P: HA necessary0.860.25--2.910.804  agreement---HCP & P: HA necessary**9.865.18--18.76\<0.001**  agreement---HCP & P: HA not necessary**0.240.07--0.770.016**  R only finds own beliefs about HA needs important1.370.43--4.370.594 Level of education (reference = low)0.516  medium0.820.48--1.380.445  high0.650.31--1.360.257 Hours of paid work (reference = 0 hours per week)**0.046**  1--20 hours per week**0.490.21--1.160.104**  21 hours or over per week**2.160.91--5.100.080** Living situation---with others in the household  (reference = alone)1.310.82--2.100.265 Country of birth -- other country (reference =   the Netherlands)0.760.30--1.920.564 Social network size (0--30)1.020.98--1.050.395 Social participation (0--27)0.970.92--1.020.273 Pc use (reference = no pc use)1.130.65--1.940.671 Comorbidity (reference = no comorbidity)1.080.70--1.670.736 Self-rated health (0--4)**1.280.94--1.740.118** Osteoarthritis of the hands (reference =   no osteoarthritis)0.850.53--1.360.498 Vision status (0--6)0.900.70--1.150.384 Mastery (5--25)1.010.96--1.070.664 Cognitive dysfunction (20--100)1.010.99--1.040.385Effect modifier^[◊](#table-fn5-2331216517744915){ref-type="table-fn"}^ Women^[◊](#table-fn5-2331216517744915){ref-type="table-fn"}^ (reference = men)**0.710.46--1.100.123**Corrective factor Source of recruitment (reference = ENT)**\<0.001** HAD---intake appointment planned at baseline**1.760.94--3.330.079** HAD---no intake appointment planned at baseline**0.110.06--0.20\<0.001**[^4][^5][^6][^7] Prediction of Entering a Hearing Aid Evaluation Period {#sec15-2331216517744915} ------------------------------------------------------ *Step 1*: *Determination of the basic model*. Five out of the possible seven 'likely' predictors constituted the basic model following backward selection. *Age* (*p = *.461) and *evaluation of hearing aids by others* (*p = *.306) were subsequently excluded from the model, as such leaving the following significant predictors: *benefits* (*OR* = 1.17, CI = 1.10-1.26\], *p* \< .001), *social pressure* (OR = 1.15, CI = 1.03--1.28, *p* = .010), *hearing loss severity* (OR = 1.04, CI = 1.00--1.08, *p* = .044), *self-reported hearing disability* (OR = 1.02, CI = 1.00--1.05, *p* = .076), and *stigma* (OR = 1.07, CI = 1.00--1.14, *p* = .048).*Step 2: Determination of the extended model*. Four out of the 29 'novel' predictors showed a statistically significant (*p*-values \< .157) association with HAEP status after adding them to the basic model. However, their NRIs were small and nonsignificant, indicating that none of them significantly added predictive value to the model (NRIs: *agreement or discrepancy*: 3%, *prompting consultation*: 1%, *comorbidity*: 4%, *vision status*: 1%, *action*: 1%). To illustrate this further: *Comorbidity* showed an OR of 1.67 (CI = 0.87--3.24, *p* = .126) when it was added to the basic model but showed a nonsignificant NRI of 4%. For the remaining 25 'novel' predictors, no significant association with HAEP status was found, nor were their NRIs significant. In conclusion, an extended model was not applicable.*Step 3*: *Testing effect modification (by gender, age, readiness)*. No significant interaction by *age* was found (*p*-values of all interaction terms ≥.567). Significant interaction by *gender* was found for two predictors: for *hearing loss severity* (OR~Hearing Loss Severity×Gender~ = 1.09, CI = 1.01--1.18, *p = *.027) and for *stigma* (OR~Stigma×Gender~ = 1.16, CI = 1.01--1.32, *p = *.030). [Table 3](#table3-2331216517744915){ref-type="table"} displays the final gender-specific prediction model. Greater *hearing loss severity* was significantly predictive of entering a HAEP in women (*p = *.005) but not in men (OR = 1.01, CI = 0.97--1.06, *p = *.618). For women, the odds to enter a HAEP were 1.10 times (CI = 1.03--1.17) greater for each 1-dB increase in *hearing loss severity*. Likewise, greater *stigma* was significantly predictive of entering a HAEP in women (*p = *.015) but not in men (OR = 0.99, CI = 0.91--1.08, *p = *.836). For women, the odds to enter a HAEP were 1.13 times (CI = 1.03--1.26) greater for each 1-point increase on the stigma scale. For both men and women, higher odds to enter a HAEP were significantly associated with more expected *benefits* (OR = 1.19, CI = 1.11--1.27, *p* \< .001), greater *social pressure* (OR = 1.15, CI = 1.03--1.28, *p* = .011), and greater *self-reported hearing disability* (OR = 1.03; CI = 1.00--1.05, *p* = .066). Table 3.Final Multivariable Prediction Model Including the Modifying Effect by Gender of the Predictors Severity of Hearing Loss and Stigma. Odds Ratios for These Predictors are Presented Separately for Men and Women.*n* = 351^[◊](#table-fn10-2331216517744915){ref-type="table-fn"}^Odds to enter a hearing aid evaluation period[\*](#table-fn9-2331216517744915){ref-type="table-fn"}Predictor (possible range / reference category)OR95% CI**p**Source of recruitment (reference category = ENT)----**\<0.001** HAD---intake appointment planned at baseline**2.231.00--4.980.052** HAD---no intake appointment planned at baseline**0.180.09--0.39\<0.001**Benefits (10--50)**1.191.11--1.27\<0.001**Social pressure (5--25)**1.151.03--1.280.011**Hearing loss severity, 3F-BEA~men~1.010.97--1.060.618Hearing loss severity, 3F-BEA~women~**1.101.03--1.170.005**Self-reported hearing disability (0--74)**1.031.00--1.050.066**Stigma~men~ (6--30)0.990.91--1.080.836Stigma~women~ (6--30)**1.131.03--1.260.015***Explained variance:(Nagelkerke R^2^)*0.59*Calibration:(Hosmer-Lemeshow test; p; H~0~ = good fit)p* = 0.694*Discrimination:(AUC of ROC; 95% CI)*0.90(0.87--0.94)[^8][^9][^10][^11][^12] Significant interaction by *readiness for change* was found for one predictor: *self-reported hearing disability* (see [Table 4](#table4-2331216517744915){ref-type="table"}). The OR of *self-reported hearing disability* was 0.92 (CI = 0.84--1.00, *p* = .059) and that of the interaction between *readiness* and *self-reported hearing disability* was 1.002 (CI = 1.000--1.004, *p* = .015). This indicated that for persons with relatively low readiness (*readiness* score ≤ 42), an increase in *self-reported disability* predicted decreasing odds to enter a HAEP (because the OR*~self-reported\ disability~* \< 1), and this effect became somewhat weaker with increasing levels of readiness (because the OR*~self-reported\ disability~* then approached 1). At the same time, this result indicated that for persons with relatively high readiness (*readiness* score \> 42), there was a positive predictive effect of *self-reported disability* (because for them the OR*~self-reported\ disability~ \>* 1), and this effect became somewhat stronger with increasing levels of readiness. So for the latter group, greater severity of *self-reported hearing disability* predicted higher odds to enter a HAEP, with this effect being stronger for persons who were more ready for change. Table 4.Final Multivariable Prediction Model Including the Modifying Effect by Readiness of the Predictor Self-Rated Hearing Disability.*n* = 334^[◊](#table-fn15-2331216517744915){ref-type="table-fn"}^Odds to enter a hearing aid evaluation period[\*](#table-fn14-2331216517744915){ref-type="table-fn"}Predictor (possible range / reference category)OR95% CI*p*Source of recruitment (reference category = ENT)**----\<.001** HAD -- intake appointment planned at baseline**2.351.04--5.31.040** HAD -- no intake appointment planned at baseline**0.170.08--0.30\<.001**Benefits (10--50)**1.181.09--1.27\<.001**Social pressure (5--25)**1.131.01--1.26.026**Hearing loss severity, 3F-BEA**1.051.01--1.09.017**Self-reported hearing disability (0--74)**0.920.84--1.00.059**Self-reported hearing disability × Readiness**1.0021.00--1.00.015**Readiness**0.960.91--1.01.114**Stigma (6--29)**1.091.01--1.16.018***Explained Variance:(Nagelkerke R^2^)*0.60*Calibration:(Hosmer--Lemeshow test; p; H~0~ = good fit)p* = .318*Discrimination:(AUC of ROC; 95% CI)*0.90(0.87--0.94)[^13][^14][^15] For the remaining four significant predictors, there was no interaction by *readiness*, so their predictive effects were independent of readiness (see [Table 4](#table4-2331216517744915){ref-type="table"}). Those who expected more *benefits of hearing aids* (OR = 1.18, CI = 1.09--1.27, *p* \< .001), experienced greater *social pressure to get a hearing aid* (OR = 1.13, CI = 1.01--1.26, *p* = .020) or experienced greater *stigma* (OR = 1.09, CI = 1.01--1.17, *p* = .020) had significantly higher odds to enter a HAEP. *Step 4*: *Determination of model performance.* The variances explained by the gender-specific model ([Table 3](#table3-2331216517744915){ref-type="table"}) and the readiness-specific model ([Table 4](#table4-2331216517744915){ref-type="table"}) were similar and quite high (Nagelkerke *R*^2 ^= 0.59 and 0.60, respectively). This was also the case for the models' discriminatory power (area under the curve = 0.90 for both models). *Step 5*: *Determination of internal validity*. The adjusted linear predictors of the models are presented at the bottom of [Tables 3](#table3-2331216517744915){ref-type="table"} and [4](#table4-2331216517744915){ref-type="table"}, respectively. Similar shrinkage factors were found for the gender-specific and the readiness-specific models (0.89 and 0.91, respectively). These factors were reasonably high, thus indicating that the original models had reasonably good internal validity. Discussion {#sec16-2331216517744915} ========== The current study aimed to identify variables that drive entering a HAEP. This is an important step in the hearing help-seeking journey of older persons with hearing problems in health-care systems in which an evaluation period is part of the hearing aid purchase process, like in The Netherlands. We built on established predictors of hearing aid uptake, and used the HBM and TTM as theoretical frameworks to select candidate predictors and to examine effect modification by *age*, *gender*, and *readiness for change*. Across all study participants, significant predictors of entering a HAEP were more expected *benefits of hearing aids* and greater *social pressure to get a hearing aid*. Greater experienced *hearing aid stigma* and greater *hearing loss severity* were modified by *gender* such that they were only predictive of entering a HAEP for women and not for men. Greater *self-reported hearing disability* was modified by *readiness for change* such that its positive predictive effect was stronger for persons who were more ready to do something about their hearing than for persons who were less ready. No modifications by *age* were found. The fact that self-reported hearing disability, hearing loss severity, and social pressure emerged as significant predictors is not surprising as several studies have shown their relevance for driving other help-seeking steps (i.e., seeking professional help and taking up a hearing aid; [@bibr21-2331216517744915]; [@bibr32-2331216517744915]; [@bibr51-2331216517744915]). With regard to self-reported hearing disability and hearing loss severity, [@bibr44-2331216517744915] and [@bibr32-2331216517744915] only found greater hearing loss severity (4F-BEA) and not self-reported hearing disability to predict hearing aid uptake. In contrast, in the current study, both hearing loss severity and self-reported hearing disability explained unique variability in the models, presumably by increasing the exposure to perceived threat of hearing loss (see HBM, [Figure 1](#fig1-2331216517744915){ref-type="fig"}). Moreover, the more ready a person was, the more this person's self-reported hearing disability predicted his or her likelihood of entering a HAEP. This is a novel finding. Although it is known that more advanced stages of change are associated with greater self-reported hearing disability, hearing loss severity, and hearing loss duration ([@bibr25-2331216517744915]; [@bibr23-2331216517744915]), the present study is the first to indicate that readiness acts as an effect modifier of self-reported hearing disability. It should be noted that the model also indicated that within individuals who did not feel ready, the predictive effect of self-reported hearing disability was in the unexpected direction: Greater self-reported hearing disability predicted *lower* odds to enter a HAEP (and this effect became weaker with higher readiness). We do not know whether this effect is real or an artifact of the statistical model. In the former case, it may be speculated that the participants who did not feel ready may have been persons with very poor acceptance of their hearing loss who coped with this by turning away from hearing aids and thus from entering a HAEP (and they did this more strongly with increasing severity of hearing disability). However, then one would expect that correcting for the coping behaviors Acceptance of loss and Self-acceptance would weaken or nullify the effect. We performed this explorative analysis and this was not the case: The effect of self-reported hearing disability remained unchanged. Replication of the results in other studies should elucidate this further. An important next step would be to identify the factors causing low readiness for change so that they can be influenced favorably as part of the help-seeking journey. In our view, such studies should take into account---but at the same time go beyond---factors that were previously identified as correlates of readiness (e.g., severity and duration of hearing loss and self-reported disability). Such factors can provide valuable information on target groups for intervention (e.g., persons with relatively mild hearing losses and short duration of complaints), but they do not necessarily reveal the true underlying mechanisms that cause low readiness. From an intervention perspective, the modifiable factors within these mechanisms then seem especially relevant. Researchers and hearing health-care professionals may be guided by the processes of change that the TTM describes in that respect (e.g., consciousness raising and self-reevaluation). According to the TTM, different types of processes should be emphasized during the successive stages of change ([@bibr1-2331216517744915]; [@bibr10-2331216517744915]). Surprisingly, the barrier of strong perceived hearing aid stigma did not predict low odds to enter a HAEP. Rather, for women, we found that greater perceived stigma predicted a *greater* likelihood of entering a HAEP. Further inspection of the data showed that the stigma scores of women who did not enter a HAEP were low (mean = 12.7), as compared with women who did enter a HAEP (mean = 14.8), men who did not enter a HAEP (mean = 14.3), and men who did enter a HAEP (mean = 14.1). It is possible that women who did not enter a HAEP unconsciously or consciously anticipated their choice not to enter a HAEP, and therefore either did not give stigma much thought or felt that stigma did not apply to them, in contrast to men who did not enter a HAEP. The gender difference we found may be surprising in the light of [@bibr15-2331216517744915] results. They found that male nonadherents (who rejected professional advice to obtain hearing aids) found hearing aids more stigmatizing than male adherents (who followed such advice). This difference was not found in women. Replication of the current study's findings is needed, as well as further research into the mechanism behind the counterintuitive relationship between stigma and HAEP status. Another factor that emerged as a significant predictor in women, but not in men was better-ear severity of hearing loss. We do not have an explanation for this finding. Interestingly, [@bibr15-2331216517744915] reported a similar result with respect to hearing aid uptake. Unfortunately, they did not explain this finding in their article. Further research is needed to elucidate this gender difference. Although our final models showed reasonably high-explained variances, they did leave room for prediction by other factors (i.e., gender-specific model: *R*^2 ^= 0.59; readiness-specific model: *R*^2 ^= 0.60). The result showing that none of the examined 'novel' predictors added significantly to predictive strength was unexpected, as the reviews by [@bibr21-2331216517744915] and [@bibr31-2331216517744915] had identified many of them as promising candidates. In particular, for two of these promising candidates (cognition and self-efficacy of hearing aid handling), [@bibr32-2331216517744915] showed that they significantly predicted hearing aid uptake. [@bibr32-2331216517744915] found a significant predictive effect by one specific cognitive factor only (cognitive reasoning skills). We did not include this cognitive factor, but [@bibr32-2331216517744915] emphasized that it explained very little variance, suggesting minor relevance. Based on the nonsignificant prediction by self-efficacy that was found in this study and the contrasting significant prediction by this factor found by [@bibr32-2331216517744915], we hypothesize that expectations about one's capacity to handle a hearing aid may be less relevant for the decision to enter a HAEP. This may be explained by the fact that an evaluation period in the Netherlands allows persons to try out a hearing aid without commitment and free of any costs. As such, expected poor handling skills may be considered harmless and no reason to forego the HAEP, whereas in hearing aid uptake, it is. Study Limitations {#sec17-2331216517744915} ----------------- This study has some limitations that deserve discussion. First, previous studies showed that perceived barriers toward hearing aids, and, in particular, concerns over hearing aid cost, are key barriers to hearing aid uptake ([@bibr14-2331216517744915]; [@bibr15-2331216517744915]; [@bibr20-2331216517744915]; [@bibr24-2331216517744915]; [@bibr32-2331216517744915]). The reimbursement system in the Netherlands is quite complicated and cost of a hearing aid depends on a person's health insurance coverage and the particular hearing aid category that the purchased hearing aid falls under. The assessment of the hearing aid category and the choice of the particular hearing aid would occur after the baseline measurement of the current study; and therefore, it was decided that *expectations* about cost was the only possible and sensible construct to measure as a predictor. However, the Sound subscale showed an unacceptably low-internal consistency so that it had to be excluded from the analyses. As such, the predictive value of the barriers of expected high cost and poor sound quality could not be assessed within this study. [@bibr50-2331216517744915] developed the AQ and found a much higher internal consistency for the Sound subscale (i.e., Cronbach's α = .48, unpublished data). In contrast, [@bibr32-2331216517744915] performed a factor analysis on the AQ and did not find that the items of the original Sound subscale formed a separate factor. [@bibr50-2331216517744915] and [@bibr32-2331216517744915] samples partly consisted of hearing aid users, and this may explain why the Sound-items behaved differently in their samples as compared with the sample of the current study. Nonetheless, the results do show that the Sound subscale that we used was invalid, and a different scale should have been used to capture the barriers related to expected hearing aid sound quality and cost. This study incorporated the HBM and the TTM. Both psychological models have received much criticism over the years, but this mainly concerned critiques expressed in other health behavior fields (see [@bibr7-2331216517744915], for an overview). One major concern for the TTM has been the lack of convincing evidence of the existence of distinct stages of readiness and thus whether a progression from one stage to another actually takes place. Another is the possible instability of stage scores over time (people may change stages quickly). However, these critiques can be refuted for hearing help-seeking, as several studies have shown that the stages of change have satisfactory construct and concurrent validity and can predict help-seeking, intervention take-up, and intervention outcomes across periods of 6 to 18 months (e.g., [@bibr18-2331216517744915]; [@bibr25-2331216517744915]; [@bibr43-2331216517744915], [@bibr44-2331216517744915], [@bibr45-2331216517744915]). Another much-cited critique of both the TTM and the HBM is that they rely on rational reasoning only, and unconscious influences on behavior are barely considered (see [@bibr7-2331216517744915]). Along similar lines, [@bibr44-2331216517744915] stated that automatic motivational processes such as emotions and habits may play an important role in hearing help-seeking and it may be worthwhile to use models that do take such factors into account. [@bibr17-2331216517744915] suggested that Leventhal's self-regulatory model would be such a model. We subscribe to these ideas. Conclusion, Clinical Implications, and Future Directions {#sec18-2331216517744915} -------------------------------------------------------- This study aimed to comprehensively and prospectively examine the predictors of entering a HAEP in older persons who were eligible for a hearing aid. The results show that many well-known predictors of hearing aid uptake are also predictive of HAEP status. The HBM proved valuable in identifying factors reflecting the perceived threat of hearing loss (self-reported hearing disability, hearing loss severity, social pressure to get a hearing aid) and the perceived benefits of starting a HAEP (expected benefits of hearing aids). Furthermore, the current study led to the notion that the relevance of some of these predictors (hearing loss severity, stigma, and self-reported hearing disability) may depend on gender and the level of readiness for behavior change (TTM). As this was the first study examining predictors prospectively while applying effect modification analyses, we strongly advocate for future longitudinal studies to test the external validity of the models. In addition, we recommend researchers to explore the underlying mechanisms of some of the unexpected results we found (prediction by stigma and hearing loss severity in women). After assuring external validity, an important next step would be to develop prediction rules that could be used in clinical practice. The factors that would cause a particular patient to be unlikely to enter a HAEP in the near future can then be identified early and subsequently addressed using counseling (mapping a person's hearing problems, creating awareness, and acceptance of hearing loss, thereby possibly influencing self-reported hearing disability), education (for realistic expectations about hearing aid benefits), and involvement of significant others (to facilitate social support). As previously described, an important research gap to fill in this context would be the identification of the determinants of readiness for change (and their interplay), so that these can be targeted through intervention as well. The authors thank the participants for their participation in the study and Schoonenberg Hoorcomfort (in particular Vera Jansen) for their assistance with the participant recruitment and the data collection among their clients. The authors thank the clinical and supporting staff of the Ear, Nose, and Throat departments of the following hospitals for their help with participant recruitment and audiogram data collection among their patients: Westfriesgasthuis (Hoorn), Spaarne Hospital (Hoofddorp), Tergooi (Blaricum, Hilversum), Van Linschoten Specialisten (Hilversum), Amstelland Hospital (Amstelveen). The authors also thank Daphne van der Veen-Hobé for her efforts as the primary research assistant on this project; Evelien Leffelaar and Rachel Williams for their efforts as research assistants; Dr. Jana Besser, Nicolette Karst, Anneke Spekenbrink-Spooren, Dr. Kelly Rijs, and Dr. Suzan van der Pas for their assistance in translating some of the questionnaires; Dr. Ariane Laplante-Lévesque, professor Thomas Lunner, Dr. Rob van den Brink, and Dr. Carly Meyer for their advice in the design phase of the study. Declaration of Conflicting Interests {#sec19-2331216517744915} ==================================== The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. Funding {#sec20-2331216517744915} ======= The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was financially supported by the Oticon Foundation, Denmark (case no. 12-0608). Co-author G. N. was supported by grant U135097131 from the UK Medical Research Council, and by a grant from the Chief Scientist Office. [^1]: Age and readiness were tested both predictors and effect modifiers. Gender was analyzed as an effect modifier only. [^2]: Percentages only apply to dichotomous and categorical variables. Per category of the particular the variable (e.g., separately for men and women), the percentages indicate the proportion (of men/women) that did not enter a HEAP (left column), and the proportion (of men/women) that did enter a HAEP (right column). [^3]: SRT~n~: speech-reception threshold in noise in dB signal-to-noise ratio; ENT: ear nose and throat specialist; HAD: hearing aid dispenser; HCP: health care practitioner; HA: hearing aid; 3F-BEA: better-ear, pure tone average hearing threshold (dB HL) across 1, 2, and 4 kHz; mv: total number of missing values for this variable; P: participant; \|: or; --: not applicable. [^4]: Entered a hearing aid evaluation period = 1; did not enter a hearing aid evaluation period = 0 (reference category). [^5]: Age and readiness were tested both as a predictor and an effect modifier. Gender was analyzed as an effect modifier only. [^6]: **Bold:** Statistically significant (*p* \< 0.157). [^7]: SRT~n~: speech-reception threshold in noise in dB signal-to-noise ratio; ENT: ear nose and throat specialist; HAD: hearing aid dispenser; HCP: health care practitioner; HA: hearing aid; 3F-BEA: better-ear, pure tone average hearing threshold (dB HL) across 1, 2, and 4 kHz; P: participant; OR: odds ratio; CI: confidence interval; fam.: family; reference: reference category. For each of the candidate predictors, effect modifiers, and for source of recruitment, the odds to enter a hearing aid evaluation period is indicated. [^8]: ENT: ear, nose, and throat specialist; HAD: hearing aid dispenser; 3F-BEA: better-ear, pure tone average hearing threshold dB HL across 1, 2, and 4 kHz; OR: odds ratio; CI: confidence interval; AUC: area under the curve; ROC: receiver operator characteristic. Intercept of model: Beta = −8.43. [^9]: Entered a hearing aid evaluation period = 1; did not enter a hearing aid evaluation period = 0 (reference category). [^10]: Sample size does not equal 377 due to missing values. [^11]: **Bold**: Statistically significant (*p* \< 0.157). [^12]: Adjusted linear predictor after bootstrapping: Intercept of model: Beta = −7.41, OR~HAD\ --\ f.u.\ appointment\ planned~ = 2.03, OR~HAD\ --\ no\ f.u.\ appointment\ planned~ = 0.22, OR~Benefits~ = 1.16, OR~Social\ pressure~ = 1.13, OR~3F-BEA\ men~ = 1.01, OR~3F-BEA\ women~ = 1.09, OR~Self-reported\ hearing\ disability~ = 1.02, OR~Stigma\ men~ = 0.99, OR~Stigma\ women~ = 1.12. Nagelkerke R^2 ^= 0.54; AUC = 0.89. [^13]: ENT: Ear Nose and Throat specialist; HAD: hearing aid dispenser; 3F-BEA: better-ear, pure tone average hearing threshold (dB HL) across 1, 2, and 4 kHz; OR: odds ratio; CI: confidence interval; AUC: area under the curve; ROC: receiver operator c haracteristic. Intercept of model: Beta = −8.76. \*Entered a hearing aid evaluation period = 1; did not enter a hearing aid evaluation period = 0 (reference category). [^14]: Sample size does not equal 377 due to missing values. **Bold**: Statistically significant (*p* \< .157). [^15]: Adjusted linear predictor after bootstrapping: Intercept of model: Beta = −7.98, OR~HAD\ --\ f.u.\ appointment\ planned~ = 2.18, OR~HAD\ --\ no\ f.u.\ appointment\ planned~ = 0.20, OR~Benefits~ = 1.16, OR~Social\ pressure~ = 1.12, OR~3F-BEA~ = 1.04, OR~Self-reported\ hearing\ disability~ = 0.93, OR~Self-reported\ hearing\ disability\*Readiness~ = 1.002, OR~Readiness~ = 0.96, OR~Stigma~ = 1.08, Nagelkerke *R*^2 ^= 0.56; AUC = 0.89.

All data is available in the supplements of this article. Introduction {#sec004} ============ Reverse transcription quantitative PCR (RT-qPCR) has become the method of choice for quantification of RNA molecules. This powerful tool enables the detection of minimal starting amounts of specific nucleic acids. However, because of its power, this technique is subject to a high degree of technical variation that can severely bias the data. This issue has been thoroughly described and guidelines were recently published that aim to set minimal standards of data reporting to enable an adequate interpretation of published data and provide sufficient details to allow reproduction of experiments by independent researchers \[[@pone.0122515.ref001]\]. Moreover, these guidelines can help researchers with the set-up of qPCR experiments to conform to the posed guidelines and standardize qPCR experimental set-up. Alas, despite a wide acceptance of these guidelines within the specialized field of qPCR, recent literature reviews indicate that these guidelines have not been sufficiently adopted by a larger group of researchers \[[@pone.0122515.ref002],[@pone.0122515.ref003]\]. One of the most crucial points in RT-qPCR data analysis is the choice of a proper normalization method. The parallel quantification of endogenous reference genes is accepted as the most reliable method for normalization of samples \[[@pone.0122515.ref004]\]. This strategy compensates for most of the variation introduced by pre-PCR and PCR processing. However, in biological systems there are no endogenous reference genes that are truly stably expressed across tissues. Reference gene stability can even vary within the same tissue/cell type between conditions \[[@pone.0122515.ref005],[@pone.0122515.ref006]\]. Consequently, the optimal strategy to normalize raw qPCR data is to perform an initial comparison of a set of independently regulated reference genes to assess the most stable ones in each specific experiment or biological setting. This subset of stably expressed reference genes is then used to calculate a normalization factor based on the geometric mean of the most stable reference genes \[[@pone.0122515.ref005],[@pone.0122515.ref007],[@pone.0122515.ref008]\]. Strikingly, this strategy is still not sufficiently applied in the general literature. Recent reviews indicate that in most investigations only one reference gene is used \[[@pone.0122515.ref002],[@pone.0122515.ref003]\]. To facilitate the use of multiple reference genes, a series of algorithms were developed to enable a comparison of the stability of the reference genes. Three of these algorithms, i.e. BestKeeper, geNorm, and NormFinder, have been incorporated in free to use Excel based software packages \[[@pone.0122515.ref004],[@pone.0122515.ref007],[@pone.0122515.ref009]\]. The development of these packages, along with newly developed R-based packages using these algorithms has resulted in a rise of research papers in which reference genes are compared. Recently, these algorithms were combined in a free to use web-tool (RefFinder) that, in combination with a fourth comparison termed the comparative CT method \[[@pone.0122515.ref010]\], enables the assessment of the most stable reference gene. This web-tool uses raw Cq values as raw input and does not enable the input of the specific PCR efficiency of each separate assay. Although this web-tool has been frequently used in the recent literature \[[@pone.0122515.ref011]--[@pone.0122515.ref025]\], its implementation has not been sufficiently validated to the original described software packages. In the present paper, we describe the assessment of the most stable reference genes in biopsies of testicular tissue containing carcinoma in situ (CIS), as well as cells from a co-culture model of the human adult Sertoli cell line FS1 and the seminoma-like TCam-2 cells. For reference gene validation, we used the three main algorithms, i.e. BestKeeper, geNorm, and NormFinder and compared these results to the RefFinder output and to the outputs of three R-based software packages incorporating the geNorm and/or NormFinder algorithm. Materials and Methods {#sec005} ===================== Cell lines and cell culture {#sec006} --------------------------- The human adult Sertoli cell line FS1 \[[@pone.0122515.ref026]\] was cultured in Medium I (DMEM (Invitrogen, Karlsruhe, Germany), supplemented with 4.5 g/l D-glucose, 1% 4 mM L-glutamine (Sigma, Deisenhofen, Germany), 1% sodium pyruvate (41966, Gipco, by Life Technologies, Eggenstein, Germany), 20% fetal calf serum (PAA, Pasching, Austria), 1% nonessential amino acids, and 1% penicillin/ streptomycin (Biochrom, Berlin, Germany). The human seminoma-like cell line TCam-2 \[[@pone.0122515.ref027]\] was cultured in Medium II (RPMI 1640 (Gipco) with 2 mM L-glutamine, 10% fetal calf serum, and 1% penicillin/ streptomycin). Each cell line was incubated in 75-cm^2^ flasks (1x10^6^ cells/ flask) at 37°C in a humidified incubator with 5% CO~2~ with medium changes every two or three days. Cells reaching 80% of confluence were used for the further experiments. Cell co-culture model {#sec007} --------------------- FS1 cells (passages 14--16) were indirectly co-cultured with TCam-2 cells in six well ThinCert (Greiner Bio One, Frickenhausen, Germany) cell culture inserts with translucent membranes and 0.4 μm pores. 50,000 FS1 cells were seeded into the apical chamber on the upper side of the membrane and 50,000 TCam-2 cells were seeded on the underside of the membrane. The cells were co-cultured for 3 weeks in a mixture of 1.5 ml Medium I and 2 ml Medium II at 37°C and 5% CO2 in 2--3 day rotation. One control group consisted of FS1 cells indirectly co-cultured with equine bone marrow derived mesenchymal stromal cells (eBM-MSC). A second control group consisted of FS1 in monoculture. Both control groups were cultured under the same conditions as the experimental group for 3 weeks. Each of the three groups was cultured in four replicates. Patient samples {#sec008} --------------- Testicular biopsies from six patients (ages 20--35 years; mean 26 years) were taken under general anaesthesia after written informed consent was obtained. The study was approved by the ethics committee of the Faculty of Medicine of the Justus Liebig University of Giessen (refnr 105/10). Biopsy specimens were cut into two equal pieces. One part was fixed in Bouin solution and embedded in paraffin wax. Sections (5μm) were stained with hematoxylin and eosin and scored according to Bergmann and Kliesch (2010) \[[@pone.0122515.ref028]\]. Diagnosis of CIS was established by placental alkaline phosphatase immunostaining \[[@pone.0122515.ref028]\]. Diagnosis of CIS was established by placental alkaline phosphatase immunostaining \[[@pone.0122515.ref029]\]. The other part was snap-frozen in liquid nitrogen and stored at -80°C until further processing for RT-qPCR. RNA isolation and reverse transcription {#sec009} --------------------------------------- For RNA isolation, cells were sampled after three weeks of culture in the different conditions. Total RNA was extracted using the peqGold Total RNA kit (Peqlab Biotechnology, Erlangen, Germany) following the kit instructions. From the cryo-preserved biopsies, total RNA was isolated with Trizol (Gibco) according to the manufacturer's instructions. Quantity and purity (260/280 ratio) of the RNA was assessed by a BioPhotometer (Eppendorf AG, Hamburg, Germany). RNA integrity was assessed by capillary electrophoresis (Bioanalyzer 2100, Agilent Technologies, Boblingen, Germany). DNase digestion was performed incubating 6.65 μl of total RNA (200 ng/μl) with RNase-free DNase I (1 U/μg; Roche, Mannheim, Germany) for 25 min at 37°C, 10 min at 75°C, and cooled at 4°C. Reverse transcription was performed with the MultiScribe Reverse Transcriptase kit (Applied Biosystems, Darmstadt, Germany) using 0.2 μg RNA in a final volume of 10 μl. Samples were incubated for 25 min at 37°C, 5 min at 75°C and cooled at 4°C. RT-qPCR analysis {#sec010} ---------------- For RT-qPCR, four replicates of each of the three *in vitro* cell cultures were used along with the six CIS samples. 14 putative reference genes were selected for evaluation of their expression profile ([Table 1](#pone.0122515.t001){ref-type="table"}): TATA box-binding protein (*TBP*), Ubiquitin C (*UBC*); Succinate dehydrogenase, subunit A (*SDHA*), Ras-like protein 13 (*RLP13*), Tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein, zeta polypeptide (*YHWAZ*), Hydroxymethylbilane synthase (*HMBS*), ß2-Microglobulin (*B2M*), Hypoxanthine phosphoribosyltransferase 1 (*HPRT1*), Glyceraldehyde-3-phosphate dehydrogenase (*GAPDH*), Ribosomal protein S18 (*RPS18*, formerly *S18*), ß-actin (*ACTB*), topoisomerase (DNA) II beta (*TOP2B*), insulin-like growth factor 1 receptor (*IGF1R*), and SRY (sex determining region Y)-box 9 (*SOX9*). 10.1371/journal.pone.0122515.t001 ###### Primers used in the study (Ta: annealing temperature). {#pone.0122515.t001g} Oligo Name Sequence direction Gene symbol Accession No. length bp Ta --------------- --------------------------- ----------- ------------- --------------- ----------- ---- GAPDH122-for aatcccatcaccatcttccag forward GAPDH NM_002046.4 122 59 GAPDH122-rev aaatgagccccagccttc reverse BACT90-for ttccttcctgggcatggagt forward ACTB NM_001101.3 89 59 BACT90-rev tacaggtctttgcggatgtc reverse B2M135-for ggcattcctgaagctgacag forward B2M NM_004048.2 135 59 B2M135-rev tggatgacgtgagtaaacctg reverse SDHA85-for tggttgtctttggtcggg forward SDHA NM_004168.2 85 59 SDHA85-rev gcgtttggtttaattggaggg reverse UBC74-for gccttagaaccccagtatcag forward UBC NM_021009.5 74 59 UBC74-rev aagaaaaccagtgccctagag reverse RLP13-127-for caaactcatcctcttccccag forward RLP13 NM_000977.3 127 59 RLP13-127-rev ctccttcttatagacgttccgg reverse YHWAZ178-for atgcaaccaacacatcctatc forward YWHAZ NM 003406.3 178 59 YHWAZ178-rev gcattattagcgtgctgtctt reverse TOP2B137-for aactggatgatgctaatgatgct forward TOP2B NM_001068.3 137 59 TOP2B137-rev tggaaaaactccgtatctgtctc reverse HMBSlOO-for ctgtttaccaaggagctggaac forward HMBS NM_001258208 100 59 HMBSlOO-rev tgaagccaggaggaagca reverse S18-88-for aaaaccaacccggtcagcc forward PRS18 X03205 88 59 S18-88-rev cgatcggcccgaggttatct reverse HPRT94-for aggaaagcaaagtctgcattgtt forward HPRT NM_000194. 94 59 HPRT94-rev ggtggagatgatctctcaactttaa reverse TBP-143-for gagagttctgggattgtaccg forward TBP NM_003194.4 143 59 TBP-143-rev atcctcatgattaccgcagc reverse SOX9-F gagcgaggaggacaa forward SOX9 NM_000346.3 151 59 SOX9-R catgaaggcgttcatggc reverse IGF1R-F tgatgacacggggcgatct forward IGF1R NM_000875 82 59 IGF1R-F gcttggaggtgctaggactgg reverse All samples were run in triplicate and each run included three no template controls. Standard dilution curves were generated to determine PCR efficiency using cDNA of normal testicular tissue. RT-qPCR was performed in 20 μl final volume containing 1 μl cDNA, 0.6 μl of primers each (10 μM), and 10 μl iQ SYBR Green Supermix (Bio-Rad, Hercules, CA). RT-qPCR was performed on a CFX 96 Real-Time system (Bio-Rad) with a two-step method. The hot start enzyme was activated (95°C for 3 min), and cDNA was amplified for 40 cycles consisting of denaturation at 95°C for 10 s and annealing/extension at 59°C for 30 s. Afterwards a melt curve assay was performed (59°C of 1 min and then the temperature was increased until 94°C by an increment of 0.5°C every 5 s) to detect the formation of non-specifically amplified products. Software programs used for statistical analysis and RT-qPCR data processing {#sec011} --------------------------------------------------------------------------- Baseline correction and threshold setting were performed using the automatic calculation of the CFX Manager Software (Bio-Rad). Cq determination was performed with the CFX Manager Software (Bio-Rad) using the Single Threshold mode ([S1 Table](#pone.0122515.s003){ref-type="supplementary-material"}). Linear relative values were assessed by the comparative Ct method ΔCq taking the amplification efficiency (E) into account ([S2 Table](#pone.0122515.s004){ref-type="supplementary-material"}) and using the sample with the lowest Cq value as reference Cq value ([Eq 1](#pone.0122515.e001){ref-type="disp-formula"}). Relative quantities of all samples were assessed by using their validated efficiency into account as well as by assuming 100% overall efficiency, by setting E at 2. {#pone.0122515.e001g} RQ = E − ( m i n C q − s a m p l e C q ) Both the efficiency adjusted as well as the non-adjusted RQs were used for validation of the most stable reference genes using different algorithms. The geNorm algorithm was used through the originally described geNorm package (Version v3.5) \[[@pone.0122515.ref004]\], through the R-based NormqPCR package (Version1.8.0) and the SLqPCR package (Version 1.0.0) \[[@pone.0122515.ref030],[@pone.0122515.ref031]\]. The NormFinder (Version 0.953) algorithm was used as an Excel add-in and as an R-script (NormFinder for R version 2014-08-23; available at <http://moma.dk/normfinder-software>)\[[@pone.0122515.ref032]\] and BestKeeper (Version 1) was used through the BestKeeper software \[[@pone.0122515.ref007]\]. Finally raw Cq values were used to assess the output of the three software packages using the web-based RefFinder platform (<http://www.leonxie.com/referencegene.php>). Results {#sec012} ======= Comparison of geNorm, NormFinder and BestKeeper analysis to different software packages {#sec013} --------------------------------------------------------------------------------------- The ranking of the reference genes was compared between RefFinder and the three different algorithms, geNorm, NormFinder, and BestKeeper, used as described in their original papers. The gene rankings from the original software packages of geNorm and NormFinder differed from their outputs on the RefFinder platform ([S1 Document](#pone.0122515.s001){ref-type="supplementary-material"}, [S3](#pone.0122515.s005){ref-type="supplementary-material"} and [S4](#pone.0122515.s006){ref-type="supplementary-material"} Tables). As RefFinder only requires raw Cq values without any possibility to include PCR efficiency, the original geNorm and NormFinder packages were run with relative values, assuming 100% efficiency for all genes. This data was in agreement with the RefFinder output. The raw data output of the BestKeeper software was in agreement with the RefFinder software, regardless of the efficiency used for the reference genes. The results of the geNorm package were in agreement to that of the geNorm algorithm as provided with the R-based NormqPCR package and the SLqPCR package (data not shown). In addition, the ranking of the genes by the original Excel based NormFinder add-in were similar compared to the NormFinder algorithm within the R-based NormqPCR package and within the NormFinder for R script. However, the measure of the absolute stability value was different for all tested Normfinder applications ([S3](#pone.0122515.s005){ref-type="supplementary-material"} and [S4](#pone.0122515.s006){ref-type="supplementary-material"} Tables). In concordance to the original geNorm package, the NormqPCR as well as the SLqPCR package provide the pairwise variation of using n/n+1 reference genes to estimate the optimal number of reference genes to be used for data normalization. For the FS1 dataset, the two most stable reference genes can be considered sufficient for normalization based on the 0.15 cut-off for the pairwise variation ([S1 Fig](#pone.0122515.s002){ref-type="supplementary-material"}). For the CIS data, the optimal number was the three most stable reference genes ([S1 Fig](#pone.0122515.s002){ref-type="supplementary-material"}). This analysis revealed that the optimal number of reference genes for the FS1 and CIS samples was two and three reference genes respectively ([S1 Fig](#pone.0122515.s002){ref-type="supplementary-material"}). This data was not provided by the RefFinder output. Effect of efficiency adjusted values on NormFinder and geNorm outputs {#sec014} --------------------------------------------------------------------- To assess the influence of using efficiency corrected versus non-corrected values in the validation of reference genes, a comparison of the rankings was made between efficiency corrected and non-corrected relative quantities in sets of samples ([Table 2](#pone.0122515.t002){ref-type="table"}). This comparison revealed that 71% (20/28) and 50% (14/28) of the rankings were different using geNorm ([Table 2](#pone.0122515.t002){ref-type="table"}, [Fig 1](#pone.0122515.g001){ref-type="fig"}) and NormFinder ([Table 2](#pone.0122515.t002){ref-type="table"}, [Fig 2](#pone.0122515.g002){ref-type="fig"}), respectively. A classification based on minor changes (position switches of one rank up or down) or major changes (position switch of more than one rank up or down) revealed that 16 out of 20 discrepant rankings were major changes with the geNorm algorithm, whereas only 5 out of 14 position changes were major with the NormFinder algorithm. In the FS1 data set the three most stable reference genes as appointed by the geNorm algorithm, SDHA, HMBS and UBC, were replaced by TOP2B, B2M and HPRT1 when gene specific efficiency was not taken into account. Similarly, in the CIS samples, YWHAZ, TOP2B and HMBS would be replaced by ACTB, HMBS and SDHA. The effect on the top three most stable genes with the NormFinder algorithm was less affected by the exclusion of the gene specific efficiency. In the FS1 samples the original top three, HMBS, TBP and UBC, would change to HMBS, TBP and HPRT1 and in the CIS samples the initial top three, HMBS, ACTB and YWHAZ would change to HMBS, IGF1R and ACTB. 10.1371/journal.pone.0122515.t002 ###### Rankings of the reference genes based on geNorm, NormFinder and BestKeeper, showing both efficiency corrected and non-corrected data used in the original software and the refFinder output for the different algorithms, for the geNorm and NormFinder software. {#pone.0122515.t002g} FS1 samples geNorm NormFinder BestKeeper ----------------- ---------------- ------------------------------ --------------- -------------------------- ------------------------------ --------------- ----------------------------- --------------- ACTB 13 13 13 13 **12** **12** 14 **1** B2M 9 **1/2** **1/2** 9 9 9 4 **12** GAPDH 11 11 11 11 11 11 10 **13** HMBS 1/2 **7** **7** 1 1 1 5 **3** HPRT1 7 **3** **3** 7 **3** **3** 1 **9** IGF1R 8 **5** **5** 8 8 8 6 **8** RLP13 14 14 14 14 14 14 12 **14** RPS18 12 12 12 12 **13** **13** 13 **4** SDHA 1/2 **9** **9** 4 4 4 8 **2** SOX9 10 10 10 10 10 10 11 **10** TBP 4 4 4 2 2 2 3 **6** TOP2B 6 **1/2** **1/2** 5 **6** **6** 2 **11** UBC 3 **6** **6** 3 **5** **5** 7 7 VWHAZ 5 **8** **8** 6 **7** **7** 9 **5** **CIS samples** **geNorm** **NormFinder** **BestKeeper** **Gene** **Efficiency** **No efficiency correction** **RefFinder** **Efficiency corrected** **No efficiency correction** **RefFinder** **Correlation Coefficient** **RefFinder** ACTB 4 **1/2** **1/2** 2 **3** **3** 7 **4** B2M 11 **8** **8** 11 **10** **10** 10 **12** GAPDH 7 7 7 7 7 7 1 **11** HMBS 3 **1/2** **1/2** 1 1 1 4 **2** HPRT1 12 **13** **13** 12 12 12 14 **5** IGF1R 5 **4** **4** 4 **2** **2** 5 **6** RLP13 13 **10** **10** 13 13 13 9 **13** RPS18 10 **9** **9** 10 **11** **11** 12 **3** SDHA 6 **3** **3** 6 **4** **4** 6 **1** SOX9 8 **11** **11** 9 **8** **8** 11 **8** TBP 9 **12** **12** 8 **9** **9** 8 **9** TOP2B 1/2 **5** **5** 5 5 5 2 **7** UBC 14 14 14 14 14 14 13 **14** YWHAZ 1/2 **6** **6** 3 **6** **6** 3 **10** Switches in ranking from the original algorithm are marked in bold. BestKeeper rankings are based on the correlation coefficient or on the RefFinder output which is only based on the standard deviation of the Cq values. {#pone.0122515.g001} {#pone.0122515.g002} BestKeeper ranking {#sec015} ------------------ The raw data output from the RefFinder platform for the BestKeeper algorithm was equal to the output obtained by the original BestKeeper software. However, the final ranking made by RefFinder was different, because RefFinder ranking is based on the standard deviations of the reference gene Cq-values, whereas the final ranking of the BestKeeper software is usually performed by assessing the correlation coefficients of each individual gene with the geometric mean of all genes (the BestKeeper Index). This different basis for determining gene stability resulted in a substantially different gene ranking ([Table 2](#pone.0122515.t002){ref-type="table"}, [Fig 3](#pone.0122515.g003){ref-type="fig"}). {#pone.0122515.g003} Comparison of the three algorithms {#sec016} ---------------------------------- A comparison of the gene rankings between geNorm, NormFinder, and BestKeeper using efficiency corrected values among the sample sets revealed that in most cases the software packages largely ranked the genes in a similar fashion. Discussion {#sec017} ========== Because of the increasing attention on a proper normalization of qPCR data, there are an increasing number of methods and software packages that have been developed for the validation of the most stabile reference genes. The data of the present manuscript suggests that different software packages should be carefully validated prior to their use in research. The present paper shows that the use of qPCR efficiency corrected relative values versus non-corrected values impacts the ranking of the reference genes based on their expression stability. Although these differences are in many cases discrete, the use of raw Cq values results in a different set of reference genes that would be selected for normalization. PCR assays with efficiencies varying only slightly from the optimal 100% will have little influence on the final ranking. However, in practice, an optimal PCR efficiency is not always attainable. Here the efficiencies of four assays (RPS18, TBP, UBC and YWHAZ) differed by more than 10% from the optimal 100% efficiency ([S2 Table](#pone.0122515.s004){ref-type="supplementary-material"}), indicating that efficiency correction in these cases is absolutely necessary. Interestingly, the geNorm output was more affected by different PCR efficiencies compared to the NormFinder output. The NormFinder software has been described to be less robust with small sample sizes compared to the geNorm algorithm \[[@pone.0122515.ref009]\]. However, in the present study, the differences between the NormFinder and geNorm algorithm did not vary in the FS1 samples (n = 11) versus the CIS samples (n = 6). The RefFinder platform is a popular tool for reference gene validation, since it is free and performs a quick analysis using the three most popular algorithms for reference gene validation starting from a single input of the Cq values only. However, despite its application in a set of published studies, this software tool has not been thoroughly validated. Our data shows that the underlying algorithms in the RefFinder software are largely in agreement with the originally described software packages except for one major point. Because of the inability to use gene specific efficiencies, the outputs are based on non-corrected raw Cq values, which biases the gene ranking. The analysis of the BestKeeper data constitutes an additional concern for the RefFinder software. The BestKeeper software provides two measures that can be used for assessing the stability of the reference genes. The first one is the raw standard deviation of the Cq values, which should be low in the case that an equal amount of input material is used for all samples. These data can be used to exclude specific reference genes when their standard deviation is too high (\>1.5). Subsequently, BestKeeper calculates the BestKeeper Index from the geometric mean of the remaining reference genes and performs Pearson correlation of each of the reference genes to the BestKeeper Index to indicate the correlation of that gene with the Index \[[@pone.0122515.ref007]\]. Hence, the correlation coefficient is a better measure to assess the most stable genes than the standard deviation. The differences between the underlying algorithms of the three software packages make a direct comparison impossible. Therefore, a mere arithmetic assessment of the most stabile reference genes as performed by the RefFinder platform has no scientific basis. This platform should rather be used as complementary tool to assess reference gene stability, and interpreted by taking their strengths and weaknesses into account. The pairwise correlation of the geNorm algorithm is known to be a strong algorithm for small sample sizes, but is biased towards selecting genes that are mutually correlated (e.g. expressed in the same pathway). The model-based approach by NormFinder has the strength that it can differentiate intragroup variation from intergroup variation and is therefore a suitable tool for identifying candidate genes when different sample groups are to be assessed, yet it requires bigger sample sizes compared to geNorm (\>8). Despite the differences among the three algorithms, the outcome of most stable and least stable reference genes was largely comparable for each sample set, indicating that each software package on its own is capable of differentiating the most stable from the least stable reference genes. Differences in final ranking are mainly observed between genes that do not differ much in stability ranking. This shows that the use of uncorrected data will lead to a suboptimal choice of reference genes, but the risk of using the least stable genes will not arise in any the three software tools. From the three studied algorithms, only the geNorm algorithm provides a data driven method to define this optimal number based on the pairwise variation obtained by comparing n versus n+1 reference genes. In its original publication, a cut-off of 0.15 was proposed for geNorm, based on the data used in that paper \[[@pone.0122515.ref004]\]. When the pairwise variation of n versus n+1 reference genes falls below this cut-off, n genes are considered sufficient (2 and 3 for the FS1 and CIS samples respectively). However, it should be noted that this cut-off is a rule of thumb, providing a means to minimize the number of reference genes, while maximizing the stability of the estimated normalization factor. A visual interpretation of the trend of the pairwise variations can also be informative to establish the optimal number of reference genes. In the present project, both sample sets would profit from using more reference genes for normalization, as can be observed from the pairwise variation, which becomes smaller when more reference genes are included ([S1 Fig](#pone.0122515.s002){ref-type="supplementary-material"}). The data from the present report also clearly show that different rankings are observed among the two sample sets. Indeed, expression levels of commonly used reference genes are known to vary across different cell or tissue types, and even within one cell or tissue type when subjected to different conditions \[[@pone.0122515.ref005]\]. Thus, the most stable genes found in the present report cannot be directly applied in other studies using the same cells without prior re-validation of the reference gene stability. Consequently, the current report should be interpreted as a guideline how to perform reference gene validation. Furthermore, it shows that different software packages should be carefully validated prior to their use in research. Supporting Information {#sec018} ====================== ###### Output from the RefFinder software using the raw Cq values as input data. (DOCX) ###### Click here for additional data file. ###### Bar plots with pair-wise variations for using n/n+1 reference genes for data normalization to estimate the optimal number of reference genes for normalizing data from the FS1 data set (A) and CIS data set (B). (TIF) ###### Click here for additional data file. ###### Raw Cq values. (XLSX) ###### Click here for additional data file. ###### Efficiencies of the different primer sets. (XLSX) ###### Click here for additional data file. ###### Stability measures from all algorithms for the FS1 samples. (XLSX) ###### Click here for additional data file. ###### Stability measures from all algorithms for the CIS samples. (XLSX) ###### Click here for additional data file. We thank Prof. Dr. Carsten Staszyk for providing the eBM-MSC cells. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: WDS CF. Performed the experiments: JDW RW. Analyzed the data: WDS JDW RW. Contributed reagents/materials/analysis tools: VS HS DN MB SK. Wrote the paper: WDS CF RB LV.

These authors contributed equally to the study. This work was supported by grants from the Shanghai Public Health Clinical Center (Nos. 2016-27 and KY-GW-2017-17), the National Key Research and Development Program of China (Nos. 2016YFC1201000 and 2016YFC1200405 to S. J., 2016YFC-1202901 to L. L.), the Sanming Project of Medicine in Shenzhen to L. L. and S. J., and the Hi-Tech Research and Development Program of China (863 Program) (No. 2015AA020930 to L. L.).We also thank Guangzhou Sagene Biotech Co., Ltd. for its aid in the preparation of the figures.

In the paper by Sen *et al.* (2018[@bb1]), the address of the correspondence author, Galyna G. Tsapyuk, should be 'Taras Shevchenko National University of Kyiv, Department of Chemistry, 64 Vladimirska Str., Kiev 01601, Ukraine', as given above.

Introduction {#Sec1} ============ An electromagnetic tracking system is a navigation system that is based on the measurement of electromagnetic fields with a known distribution. Based on those measurements, the position and orientation of the field sensor can be calculated. Electromagnetic tracking systems are not dependent on line-of-sight or rigid connections like in optical and mechanical tracking systems. It is used in computer-assisted surgery to track the position of the instrument relative to the patient's anatomy \[[@CR1]\]. It has been used successfully for navigation in ear, nose and throat surgery (ENT) as well as in cranial and spine applications. The availability of microsensors has also allowed for the use of EMTS in tightly controlled cardiac mapping and pulmonary applications. EMTS has also shown its benefit in laparoscopic ultrasonography, needle aspiration biopsy, ultrasound freehand tracking, brachytherapy, ultrasound bone registration, and has also been introduced in endoscopic and bronchoscopic procedures \[[@CR1]--[@CR7]\]. EMTS is often applied in environments that typically contain metallic objects (conductive, para- or ferromagnetic) or electronic devices that may affect the magnetic reference fields. Ferromagnetic materials alter the shape of the reference fields and consequently distort both the direct current (DC) and the alternating current (AC) EMTS \[[@CR8]--[@CR10]\]. The eddy currents induced by conductive materials generate secondary AC magnetic fields, which distort the reference field pattern \[[@CR8], [@CR11]--[@CR13]\]. The resulting fields are difficult to determine or characterize analytically \[[@CR8], [@CR14]\]. Stray fields produced by electronic devices \[[@CR4], [@CR14]\] or power lines \[[@CR9]\] also distort the reference field. Internal electronic components of EMTS may also have an influence on the accuracy of position measurements \[[@CR14]\]. Distortions originating from electronic devices and eddy currents additively overlay the reference field. Errors caused by nearby metallic objects are the most serious problem in clinical applications \[[@CR14]\]. In a minimally invasive surgery operating table, LCD displays, light, metal studded walls, C-arms, operating microscopes and surgical instruments placed in proximity to the field generator affect the accuracy of EMTS \[[@CR9], [@CR14]--[@CR16]\]. Distortions in EMTS could be eliminated by placing potential sources of distortions far away from the region of interest (ROI). This is typically not possible in clinical applications without changing the medical workflow. There are many methods presented in literature for the compensation of distortions caused by metal devices fixed in position and orientation. A survey of such methods of calibration has been published \[[@CR17]\]. Calibration techniques, however, cannot compensate dynamic distortions originating from mobile objects such as surgical tools or movable imaging equipment. In such circumstances, measurements could be averaged or filtered (e.g., with a Kalman filter) \[[@CR18]\]. Another method utilizing reverse tracking has been claimed, whereby a magnetic field source is integrated within the medical device, i.e., the catheter to be tracked contains the magnetic field source while a sensor coil array is distributed outside the patient body \[[@CR19]\]. This approach is less sensitive to external distortions by conductive objects but suffers from very weak magnetic fields which could be generated within a small volume inside the patient's body. Another approach is the shielding of at least one side of the tracking system \[[@CR20]\]. Aside from strategies to avoid distortions, there are also approaches for the detection and correction of the errors caused by conductive objects. The detection of field distortion can be accomplished using two field sensors attached rigidly to the surgical tool. The known distance between the sensors is compared with the measured value. If the deviations exceed a given tolerance, an error message may be presented \[[@CR9]\]. The correction of metallic disturbances can be based on redundant measurements \[[@CR21]\]. In this manner, another approach measured the position using two frequencies \[[@CR22]\]. The measurement with lower frequency was interpreted as less distorted and was used to correct the measurement with higher frequency. In order to decrease its influence on the accuracy of the tracking system, the medical instrument in use could be made from laminated conductive materials \[[@CR23]\] similar to transformer cores. In the patent \[[@CR24]\], the authors model distortion with RL circuits and claim to correct the measured voltage with the calculated parameter of the conductive distorter. The author of the patent \[[@CR25]\] proposes to represent the conductive distorter as lag or lead--lag networks. The excitation of the emitting coil with at least two frequencies leads to the calculation of the parameters of this lead--lag network. The influence of the conductive distorters is removed by adjusting the magnetic fields based on the estimated parameters. In contrast, the method of distortion compensation presented in this paper does not require modeling of the parameter of the conductive distorter. In proposed approach, the emitting coil is supplied with a quadratic excitation, which therefore allows the voltage induced in the sensing coil by the emitting coil to be distinguished from the voltage induced by the conductive distorter. Only the undistorted voltage induced in the sensing coil directly by the emitting coil is used for the position and orientation calculation. Methods {#Sec2} ======= Equivalent circuit {#Sec3} ------------------ The presented method of the distortion minimization is based on a model of the tracking system and on a model of a conductive distorter. AC EMTS typically uses a sensing coil as a field sensor, which measures the magnetic field of the emitting coil for further position estimation. The equivalent circuit of the magnetically coupled emitting and sensing coil is presented in Fig. [1](#Fig1){ref-type="fig"}. In this equivalent circuit, the emitting coil is supplied with a voltage $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U}_{1}$$\end{document}$ and the voltage $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U}_{2}$$\end{document}$ is measured on the sensing coil. The amplitude of the voltage on the sensing coil is used for the position estimation. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{L1}, {R}_{L2}, {L}_{1}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${L}_{2}$$\end{document}$ are resistances and inductances of the emitting coil and of the sensing coil, respectively. The currents $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${I}_{1}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${I}_{2}$$\end{document}$ are flowing in the emitting and in the sensing coil, respectively. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M_{12}$$\end{document}$ is the mutual inductance between the emitting and the sensing coil.Fig. 1Equivalent circuit of the coupling between the emitting and the sensing coil The output of the model may be considered as an open loop due to the high impedance of the AD card. The current flow in the sensing coil $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${I}_{2}$$\end{document}$ is therefore approximately equal to zero. Such a system may be also considered as an air transformer with a very low efficiency. The resulting transmittance $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${G}_{1{-}2}(s)$$\end{document}$ linking the input voltage $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U}_{1}$$\end{document}$ and the output voltage $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U}_{2}$$\end{document}$ is given in the Laplace domain in Eq. ([1](#Equ1){ref-type=""}), where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tau }_{1}$$\end{document}$ is a time constant depending on the inductance and on the resistance of the emitting coil ([3](#Equ3){ref-type=""}).$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&{G}_{1{-}2}({s})=\frac{{U}_2 ({s})}{{U}_1 ({s})}=\frac{M_{12}^{\prime } \cdot {s}}{\left( \frac{1}{{\tau }_{1}}+{s}\right) },\end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&M_{12}^{\prime }=\frac{M_{12}}{{L}_{1}},\end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&{\tau }_1 =\frac{{L}_{1}}{{R}_{L1}}, \end{aligned}$$\end{document}$$According to the transmittance, the voltage on the sensing coil is proportional to the derivative of the voltage on the emitting coil. The value of the mutual inductance $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M_{12}$$\end{document}$ is used for the calculation of the magnetic field of the emitting coil and for solving the position and orientation algorithm. Conductive distorters in the EMTS may be modeled as an additional RL circuit \[[@CR24], [@CR25]\]. The equivalent circuit of two magnetically coupled coils with an additional conductive distorter is presented in Fig. [2](#Fig2){ref-type="fig"}. Along with the coefficients common with the previous model, Fig. [2](#Fig2){ref-type="fig"} comprises the current flow within the conductive distorter $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${I}_{\mathrm{D}}$$\end{document}$, inductance $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${L}_{\mathrm{D}}$$\end{document}$ and resistance $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{\mathrm{D}}$$\end{document}$ of the conductive distorter. The model also includes the mutual inductance between the emitting coil and the distorter $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M_{1{\mathrm{D}}}$$\end{document}$ and the mutual inductance between the distorter and the sensing coil $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M_{2{\mathrm{D}}}$$\end{document}$.Fig. 2Equivalent circuit of two magnetically coupled coils with conductive distorter modeled as RL circuit The resulting transmittance $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${G}_{1-D-2}(s)$$\end{document}$ linking the input voltage $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U}_{1}$$\end{document}$ and the output voltage $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U}_{2}$$\end{document}$ of the two coils coupled indirectly across the RL circuit is presented in the Laplace domain in Eq. ([4](#Equ4){ref-type=""}). The value of the time constant $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tau }_\mathrm{D}$$\end{document}$ of the conductive distorter ([6](#Equ6){ref-type=""}) depends on the size, conductivity, form and the orientation of the conductive object relative to the magnetic field of the emitting coil. The voltage on the sensing coil is proportional to the second derivative of the voltage on the emitting coil.$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&{G}_{1-D-2} ({s})=\frac{{U}_2 (s)}{{U}_1 (s)}=\frac{\frac{M_{1\mathrm{D}}}{{L}_1}\cdot {s}}{\left( {\frac{1}{{\tau }_1 }+{s}} \right) }\cdot \frac{\frac{M_{\mathrm{D}2}}{{L}_\mathrm{D}}\cdot {s}}{\left( {\frac{1}{{\tau }_\mathrm{D}}+{s}}\right) }\nonumber \\&\qquad \qquad \qquad \quad =\frac{M_{1\mathrm{D}2}^{\prime } \cdot {s}^{2}}{\left( {\frac{1}{{\tau }_1}+{s}} \right) \cdot \left( \frac{1}{{\tau }_{\mathrm{D}} +{s}} \right) },\end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&M_{1\mathrm{D}2}^{\prime } =\frac{M_{1\mathrm{D}}}{{L}_{1}}\cdot \frac{M_{1\mathrm{D}}}{{L}_\mathrm{D} },\end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&{\tau }_\mathrm{D} =\frac{{L}_\mathrm{D}}{{R}_\mathrm{D} }, \end{aligned}$$\end{document}$$The transmittance $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${G}_\mathrm{E}(s)$$\end{document}$ linking the input voltage $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U}_{1}$$\end{document}$ and the output voltage $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U}_{2}$$\end{document}$ including both the directly and indirectly induced voltages ([7](#Equ7){ref-type=""}) is a sum of both transmittances ([1](#Equ1){ref-type=""}) and ([4](#Equ4){ref-type=""}).$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {G}_\mathrm{E} (s)=\frac{{U}_2 (s)}{{U}_1 (s)}=\frac{M_{12}^{\prime } \cdot {s}}{\left( {\frac{1}{{\tau }_1}+{s}} \right) }+\frac{M_{1\mathrm{D}2}^{\prime } \cdot {s}^{2}}{\left( {\frac{1}{{\tau }_1 }+{s}} \right) \cdot \left( \frac{1}{{\tau }_\mathrm{D}}+{s} \right) }, \end{aligned}$$\end{document}$$The proposed model of the conductive distorter exhibits properties of a high-pass filter and therefore contributes additional filtered voltage to the voltage induced directly by the emitting coil. The transfer function is similar to the transfer function proposed by Schneider \[[@CR25]\]. However, the model presented in this work links the voltage on the emitting coil with the voltage on the sensing coil. This difference is crucial for understanding the approach proposed in the following section. The response of the system to a sinusoidal excitation without distortions is presented in the Laplace domain ([8](#Equ8){ref-type=""}) and in the time domain ([9](#Equ9){ref-type=""}), respectively.$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {U}_{\mathrm{WS}} (s)&= {U}_1 (s)\cdot {G}(s)=\frac{{U}\cdot {\omega } }{\left( {{\varpi } ^{2}+{s}^{2}} \right) }\cdot \frac{{s}\cdot M_{12}^{\prime } }{\left( {\frac{1}{{\tau }_1 }+{s}} \right) }\nonumber \\&= \frac{{U}\cdot {\omega } \cdot {s}\cdot M_{12}^{\prime } }{\left( {{\varpi }^{2}+{s}^{2}} \right) \cdot \left( {\frac{1}{{\tau }_1 }+{s}} \right) },\end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {U}_{\mathrm{WS}} ({t})&= {U}\cdot M_{12}^{\prime } \nonumber \\&\quad \cdot \frac{{\omega } \cdot {\tau } _{1}\cdot \left( {\cos \left( {{\varpi } {t}} \right) \!-\!\mathrm{e}^{-\frac{{t}}{{\tau }_1 }}\!+\!{\omega } \cdot {\tau }_1 \cdot \sin ({\varpi } {t})} \right) }{\left( {{\varpi }^{2}\cdot {\tau }_1^2 +1} \right) },\nonumber \\ \end{aligned}$$\end{document}$$The response may be distinguished between a steady-state response comprising a sum of sinusoidal functions and a transient response comprising an exponential function. After a time period depending on the time constant $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tau }_{1}$$\end{document}$, the transient response decays to zero and the response comprises only the sum of sinusoidal voltages. The response of the system to a sinusoidal excitation caused by distortions only is presented in the Laplace domain ([10](#Equ10){ref-type=""}) and in the time domain ([11](#Equ11){ref-type=""}), respectively, where U is the amplitude of the sinusoidal excitation.$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {U}_{\mathrm{DS}} (s)&= {U}_1 (s)\cdot {G}_{1{-}2} (s)=\frac{{U}\cdot {\omega }}{\left( {{\varpi }^{2}+{s}^{2}} \right) }\nonumber \\&\quad \cdot \frac{{s}^{2}\cdot M_{1\mathrm{D}2}^{\prime } }{\left( {\frac{1}{{\tau }_1}+{s}} \right) \cdot \left( {\frac{1}{{\tau }_\mathrm{D}}+{s}} \right) }\nonumber \\&= \frac{{U}\cdot {\omega } \cdot {s}^{2}\cdot M_{1\mathrm{D}2}^{\prime } }{\left( {{\varpi } ^{2}+s^{2}} \right) \cdot \left( {\frac{1}{{\tau }_1}+{s}} \right) \cdot \left( {\frac{1}{{\tau }_\mathrm{D}}+{s}}\right) }, \end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&{U}_{\mathrm{DS}} ({t})={U}\cdot M_{1\mathrm{D}2}^{\prime } \cdot \frac{{\omega } \cdot {\tau }_1\cdot {\tau }_\mathrm{D} \cdot \mathrm{e}^{-\frac{{t}}{{\tau }_1}}}{\left( {{\varpi }^{2}\cdot {\tau } _1^2 \!+\!1}\right) \cdot \left( {{\tau }_1 \!-\!{\tau }_\mathrm{D}}\right) }\nonumber \\&\!\quad -{U}\cdot M_{1\mathrm{D}2}^{\prime } \cdot \frac{{\omega } \cdot {\tau }_{1} \cdot {\tau }_\mathrm{D} \cdot \mathrm{e}^{-\frac{{t}}{{\tau }_\mathrm{D}}}}{\left( {{\varpi } ^{2}\cdot {\tau }_\mathrm{D}^2 +1} \right) \cdot \left( {{\tau }_1 -{\tau }_\mathrm{D}} \right) }\nonumber \\&\!\quad + {U}\cdot M_{1\mathrm{D}2}^{\prime }\cdot \frac{{\omega }^{2}\cdot {\tau } _1 \cdot {\tau }_\mathrm{D} \cdot \left( {\omega } \cdot {\tau }_{1} \cdot \cos \left( {\varpi } {t} \right) -\sin ({\varpi } {t})\right) }{\left( {\varpi }^{2}\cdot {\tau }_{1}^{2} +1\right) \cdot \left( {\varpi }^{2}\cdot {\tau }_\mathrm{D}^2 +1\right) }\nonumber \\&\!\quad +{U} \cdot M_{1\mathrm{D}2}^{\prime } \cdot \frac{{\omega }^{2} \cdot {\tau }_{1} \cdot {\tau }_\mathrm{D} \cdot \left( {\omega } \cdot {\tau }_\mathrm{D} \cdot \cos \left( {\varpi } {t} \right) \!+\!\omega ^{2}\cdot {\tau }_{1} \cdot {\tau }_\mathrm{D} \cdot \sin \left( {\varpi } {t} \right) \right) }{\left( {\varpi }^{2}\cdot {\tau }_{1}^{2} \!+\!1\right) \cdot \left( {\varpi }^{2} \cdot {\tau }_\mathrm{D}^2 \!+\!1\right) },\nonumber \\ \end{aligned}$$\end{document}$$The sum of the voltage induced directly ([9](#Equ9){ref-type=""}) and indirectly ([11](#Equ11){ref-type=""}) in the sensing coil builds a resulting voltage. The voltage induced by the conductive distorter has the same frequency as the voltage induced directly by the emitting coil, and therefore, the distortion due to the conductive distorter cannot be filtered out of the measured sum of signals. The only feature which distinguishes the voltage induced in the sensing coil by the emitting coil from the voltage induced by the conductive distorter is a slightly different phase shift. Quadratic excitation {#Sec4} -------------------- As a solution to the problem of distinguishing the voltage induced in the sensing coil directly by the emitting coil and indirectly by the conductive distorter, an EMTS with a quadratic excitation on the emitting coil ([12](#Equ12){ref-type=""}) is proposed.$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {U}_\mathrm{Q} ({t})={U}\cdot {t}^{2}, \end{aligned}$$\end{document}$$The response of the sensing coil on the quadratic excitation of the emitting coil without conductive distorter is presented in the Laplace domain ([13](#Equ13){ref-type=""}) and in the time domain ([14](#Equ14){ref-type=""}), respectively.$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {U}_{\mathrm{WQ}} (s)&= X(s)\cdot {G}_{1{-}2} (s)=\frac{2\cdot {U}}{{s}^{3}}\cdot \frac{{s}\cdot M_{12}^{\prime } }{\left( {\frac{1}{{\tau } _{1}}+{s}}\right) }\nonumber \\&= \frac{2\cdot {U}\cdot {s}\cdot M_{12}^{\prime } }{{s}^{3}\cdot \left( {\frac{1}{{\tau }_1 }+{s}} \right) },\end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {U}_{\mathrm{WQ}} ({t})&= 2\cdot {U}\cdot {\tau }_{1} \cdot M_{12}^{\prime } \cdot \left( {t}-{\tau }_{1} +{\tau }_{1} \cdot \mathrm{e}^{-\frac{{t}}{{\tau }_{1}}} \right) , \end{aligned}$$\end{document}$$The system response on the sensing coil to the quadratic excitation caused by the conductive distorter only is presented in the Laplace domain ([15](#Equ15){ref-type=""}) and in the time domain ([16](#Equ16){ref-type=""}), respectively.$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {U}_{\mathrm{DQ}} (s)&= X(s)\cdot {G}_{1{-}2} (s)\nonumber \\&= \frac{2\cdot {U}}{{s}^{3}}\cdot \frac{{s}^{2}\cdot M_{\mathrm{1D2}}^{\prime } }{\left( {\frac{1}{{\tau }_{1}}+{s}} \right) \cdot \left( \frac{1}{{\tau }_\mathrm{D}}+{s} \right) }\nonumber \\&= \frac{2\cdot {U}\cdot {s}^{2}\cdot M_{\mathrm{1D2}}^{\prime } }{{s}^{3}\cdot \left( {\frac{1}{{\tau }_{1}}+{s}} \right) \cdot \left( {\frac{1}{{\tau }_\mathrm{D}}+{s}}\right) }, \end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&{U}_{\mathrm{DQ}} ({t})=2\cdot {U}\cdot M_{\mathrm{1D2}}^{\prime }\nonumber \\&\quad \quad \cdot \left( \frac{1}{{\tau }_1 \cdot {\tau }_\mathrm{D}}\!+\!\frac{1}{{\tau }_1 \cdot \left( {{\tau }_{1} -{\tau }_\mathrm{D}}\right) }\cdot \mathrm{e}^{-\frac{{t}}{{\tau }_{1}}}-\frac{1}{{\tau }_{\mathrm{D}} \cdot \left( {\tau }_{1} -{\tau }_\mathrm{D} \right) }\cdot \mathrm{e}^{-\frac{{t}}{{\tau }_{\mathrm{D}}}} \right) ,\nonumber \\ \end{aligned}$$\end{document}$$ The directly induced voltage ([14](#Equ14){ref-type=""}) comprises a saw tooth, a rectangular pulse and an exponential transient response. The voltage induced indirectly ([16](#Equ16){ref-type=""}) by the distorter comprises a rectangular pulse and an exponential transient response only. This difference is caused by a double derivative of the signal induced by the emitting coil induced in the sensing coil indirectly across a conductive distorter ([4](#Equ4){ref-type=""}). According to the model, a saw tooth signal form of the voltage is induced in the conductive distorter. After the second derivative, the signal form is changed into a rectangular wave with an additional exponential transient response. The missing saw tooth signal form distinguishes the signal induced in the sensing coil directly by the emitting coil from the signal induced indirectly across a conductive distorter in the sensing coil. It also provides a measurement of the amplitude of the signal induced directly by the emitting coil while neglecting the voltage induced by the source of distortion. However, the measured signal also includes the exponential transient response, which overlays the steady-state response. After a time period dependent on the properties of the conductive distorter, the transient response decays to zero and the signal comprises the saw tooth and the rectangular pulse only. The measurement of the saw tooth wave allows calculation of the mutual inductance between the emitting and the sensing coil. This value of the mutual inductance allows estimation of the position of the sensor without an influence from the conductive distorter. When the emitting coil is excited by quadratic excitation, the transient response in the form of the sum of exponential functions overlays the steady-state response comprising the saw tooth function. This problem could be solved by increasing the period of the quadratic excitation. However, the amplitude of the signal in the sensing coil depends on the time derivative of the current in the emitting coil. The increase in the length of the period would decrease the amplitude of the measured voltage in the sensing coil and therefore significantly reduce the working volume of the electromagnetic tracking system. In the presented paper, the transient response of the system is estimated in the first step using the response on the excitation by an additional rectangle pulse (Fig. [3](#Fig3){ref-type="fig"}c) supplied into the emitting coil shortly after supplying it with the quadratic excitation (Fig. [3](#Fig3){ref-type="fig"}a). The resulting response of the system ([17](#Equ17){ref-type=""}) and of the conductive distorters ([18](#Equ18){ref-type=""}) on a rectangle input voltage is as follows:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {U}_{\mathrm{WR}} ({t})&= {U}\cdot M_{\mathrm{\mathrm{1D2}}}^{\prime } \cdot \mathrm{e}^{-\frac{{t}}{{\tau }_{1}}},\end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {U}_{\mathrm{DR}} ({t})&= {U}\cdot M_{\mathrm{1D2}}^{\prime } \nonumber \\&\quad \cdot \left( {\frac{{\tau }_{1}}{{\tau }_{1} -{\tau }_{\mathrm{D}}}\cdot \mathrm{e}^{-\frac{{t}}{{\tau }_1 }}-\frac{{\tau }_\mathrm{D} }{{\tau }_{1} -{\tau }_\mathrm{D}}\cdot \mathrm{e}^{-\frac{{t}}{{\tau }_\mathrm{D}}}} \right) , \end{aligned}$$\end{document}$$This response to a rectangular excitation contains information about the tracking system and the conductive distorter and comprises the transient response only (Fig. [3](#Fig3){ref-type="fig"}d). The emitting coil will be supplied with a rectangular signal directly after supplying it with the quadratic signal form. Therefore, it is assumed that the position of the sensing coil and the position of the conductive distorter relative to the emitting coils are constant over the entire position and orientation estimation. The response of the system to a rectangular input signal leads to the calculation of the transient response of the system to a quadratic input signal. This calculation is based on a relationship between the rectangular and the quadratic signal, that is, the second time derivative of the quadratic signal produces a rectangular form of the signal.Fig. 3Extraction of the steady-state response from the response on the quadratic excitation of the emitting coil measured by ADC: **a** quadratic excitation fed into the emitting coil measured by ADC, **b** response on the sensing coil to the quadratic excitation of the emitting coil measured by ADC, **c** rectangular excitation fed into the emitting coil measured by ADC, **d** response on the sensing coil to the rectangular excitation of the emitting coil measured by ADC, **e** transient response of the system on the rectangular excitation estimated with double integration of the rectangular excitation of the emitting coil, **f** response on the sensing coil to the quadratic excitation with subtracted transient response Therefore, the second numerical integration of the response on the rectangular wave with suitable amplitude is equal to the transient response of the system to the quadratic signal when the constant resulting from the numerical integration is removed. After the calculation, the estimation with the described method of the transient response of the system on the quadratic excitation (Fig. [3](#Fig3){ref-type="fig"}e) will be subtracted from the voltage measured on the sensing coil (Fig. [3](#Fig3){ref-type="fig"}f). The transient response on the quadratic excitation is calculated with double integration of the response on the rectangle excitation:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&\int \int {{U}_{\mathrm{WR}} ({t})} \mathrm{dt}^{2}={U}\cdot M_{12}^{\prime } \cdot {\tau } _{1}^{2} \cdot \mathrm{e}^{-\frac{{t}}{{\tau }_{1}}},\end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&\int \int {{U}_{\mathrm{DR}} ({t})} \mathrm{dt}^{2}={U}\cdot M_{\mathrm{1D2}}^{\prime } \nonumber \\&\quad \cdot \left( \frac{1}{{\tau }_{1} \cdot \left( {\tau }_{1} -{\tau }_{\mathrm{D}} \right) }\cdot \mathrm{e}^{-\frac{{t}}{{\tau }_{1}}}-\frac{1}{{\tau }_{\mathrm{D}} \cdot \left( {\tau }_{1} -{\tau } _\mathrm{D}\right) }\cdot \mathrm{e}^{-\frac{{t}}{{\tau }_\mathrm{D}}}\! \right) \!, \end{aligned}$$\end{document}$$The estimation of the position of the sensing coil requires calculation of the mutual inductance between every coupled emitting and sensing coil. The known dimensions of the emitting and sensing coil and the mutual inductance between the emitting and the sensing coil lead to the estimation of the position of the sensing coil. When the emitting coil is fed with the sinusoidal excitation, the mutual inductance is calculated based on the voltage measured in the sensing coil and the known amplitude of the voltage in the emitting coil. On the other hand, when the emitting coil is fed with the quadratic excitation, the mutual inductance between the emitting and the sensing coil will be calculated based on the gradient of the saw tooth signal induced in the sensing coil. Its curve is fitted with the least squares method. The gradient of the saw tooth signal leads to the calculation of the mutual inductance that consequently leads to the estimation of the position of the sensing coil. Experimental evaluation {#Sec5} ======================= Experimental setup of the tracking system {#Sec6} ----------------------------------------- For an experimental evaluation of the proposed method, an experimental setup of the tracking system was used (Fig. [4](#Fig4){ref-type="fig"}). It includes 8 emitting coils each with a diameter of 40 mm supplied with a voltage generated by an analog-to-digital converter NI PXI 6281 (ADC) and amplified by an amplifier. The experimental setup utilizes the NDI field sensor with a diameter of 0.3 mm as a sensing coil. The voltage induced in the sensing coil is amplified by an additional amplifier and measured by ADC. The resulting position of the sensing coil is estimated in Matlab by solving the position and orientation algorithm \[[@CR26]\].Fig. 4Experimental setup of the electromagnetic tracking with attached optical marker for the evaluation of the EMTS with quadratic excitation Measurement setup {#Sec7} ----------------- In multiple interventions in computer-assisted surgery, the sensing coil of the EMTS is attached to surgical instruments which are made of conductive materials. In the experimental validation of the proposed method, an aluminum disk with a radius of 0.05 m and a thickness of 0.03 m has been utilized as such a conductive distorter. This disk represents the surgical tools used in minimal invasive surgery, e.g., hammer, trocar, endoscope or ultrasonic probe. Those instruments are usually made of less electrically conductive and non-ferromagnetic steel. Therefore, in this experiment, the more electrically conductive aluminum disk may be considered as a worst-case scenario of distortion. The goal of the performed experiment was to examine the influence of the conductive distorter on the accuracy of the tracking system with the quadratic excitation and compare it to the accuracy of the tracking system with the sinusoidal excitation. The sensing coil was moved within the working surface by a non-conductive plastic robotic arm in 200 random positions in proximity to the conductive distorter. The working surface had a rectangular shape ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0.1\hbox {m}\times 0.1\hbox {m}$$\end{document}$), where the middle of the rectangle is placed in the middle of the coordinate system parallel to the surface of the field generator at a height of 0.15 m. The working surface was chosen to be relatively small because the experiment was designed in order to focus on the influence of the conductive distorter on the accuracy of the EMTS. The measurements were therefore performed in direct proximity to the conductive distorter in order to achieve the highest influence on the accuracy of the electromagnetic tracker. A larger distance between the sensing coil and emitting coils reduces the accuracy of the EMTS. Furthermore, the voltage on the sensing coil was measured for 8 s for the sinusoidal excitation (1 kHz) and 32 s for the quadratic excitation and averaged in order to increase the accuracy of the experimental setup of the tracking system. After each movement of the robotic arm, the position of the sensing coil was measured by the electromagnetic tracking system and by the optical tracking system Polaris Spectra. Accuracy of this optical tracking system (root mean square error $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$<$$\end{document}$0.25 mm) exceeds the expected accuracy of the EM tracking and therefore was used as a reference for the evaluation of the accuracy of the position estimation by the EMTS. The EMTS uses sinusoidal excitation in the first position measurement and shortly after, while in the same position, uses quadratic excitation. After 200 random measurements in the presence of the aluminum disk, the entire experiment was repeated without the presence of any conductive distorter. In order to evaluate the electromagnetic tracking system, the coordinate system of the EMTS was registered with the coordinate system of the optical tracking system prior to the accuracy evaluation. For the performed registration, 27 additional measurements within a working volume of 0.2 m $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times $$\end{document}$ 0.2 m $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times $$\end{document}$ 0.2 m with both the electromagnetic and the optical tracking system were performed. The registration of both coordinate systems causes additional inaccuracy that increases the mean root square error of the position estimation with the optical tracking system. The registration error was calculated according to the work by Fitzpatrick \[[@CR27]\]. The registration was carried out by fiducial registration via the software 3D Slicer. Results and discussion {#Sec8} ====================== Results {#Sec9} ------- The registration error introduced by registering the optical and electromagnetic tracking system is equal to 0.7 mm. The experimental errors in the position estimation described in the previous chapter are presented in Fig. [5](#Fig5){ref-type="fig"}. The errors were calculated in four different scenarios. Figure [5](#Fig5){ref-type="fig"}a, b presents the errors in the position estimation without the aluminum disk with the quadratic and sinusoidal excitation, respectively. Furthermore, Fig. [5](#Fig5){ref-type="fig"}c, d presents the errors in the position estimation with the aluminum disk with the quadratic and sinusoidal excitation, respectively.Fig. 5The influence of the aluminum disk on the accuracy of the electromagnetic tracking system as a function of the distance between aluminum disk and the sensing coil: **a** tracking with quadratic excitation without the aluminum disk, **b** tracking with sinusoidal excitation without the aluminum disk, **c** tracking with quadratic excitation nearby to the aluminum disk, **d** tracking with sinusoidal excitation nearby to the aluminum disk The maximal error, the root mean square error (RMSE) and the standard deviation (std) for four different scenarios within the performed experiment are presented in Table [1](#Tab1){ref-type="table"}.Table 1Accuracy of the electromagnetic tracker with and without the aluminum disk in the proximityMax error (mm)RMSE (mm)Std (mm)Sinusoidal input voltage without conductive distorter1.50.90.3Quadratic input voltage without conductive distorter3.41.70.6Sinusoidal input voltage with conductive distorter10.92.41.7Quadratic input voltage with conductive distorter3.82.00.6 Discussion {#Sec10} ---------- According to Fig. [5](#Fig5){ref-type="fig"}, when the aluminum disk is placed in proximity of the sensing coil, the maximal error of the position estimation in the scenario with the sinusoidal excitation increased from 1.5 to 10.9 mm. The root mean square error of the position estimation increased from 0.9 to 2.4 mm. This shows a high influence of the conductive aluminum disk on the accuracy of the electromagnetic tracker. Despite the high increase in the maximal error of the position estimation in the sinusoidal excitation scenario with the conductive distorter in proximity to the sensing coil, the maximal error in the scenario with quadratic excitation changed only slightly from 3.4 to 3.8 mm. The RMSE with quadratic excitation changed from 1.7 mm without any distorters to 2.0 mm in the presence of conductive distorters. This increase in the RMSE in the presence of the distorter is smaller than the standard deviation of the measurements. Additionally, Fig. [5](#Fig5){ref-type="fig"}c does not indicate the existence of a trend in the error measurement as the distance between the surface of the aluminum disk and the sensing coil increased for the quadratic excitation scenario. A declining trend could be seen, however, in Fig. [5](#Fig5){ref-type="fig"}d, where the sinusoidal excitation has been used. The results indicate very low influence of the conductive distorter on the accuracy of the electromagnetic tracker with quadratic excitation. The root mean square error of the position estimation without the additional distorter for the EMTS with quadratic excitation (1.7 mm) is higher than with the sinusoidal excitation (0.9 mm). The lower accuracy in the position estimation is caused by the noise originating from, e.g., the amplifier, the discretization, the electromagnetic field generated by surrounding electronic devices or noise with a frequency of 50 Hz induced by the power supply. Due to the analysis of the system response on the quadratic excitation in the time domain, the response of the system was not filtered out. Only a low-pass filter of the ADC with a cutoff frequency equal to 80 kHz was used. In the proposed method, the emitting coils have been supplied with both the quadratic and rectangular signal form, which has a negative effect on the measuring rate. Conclusion {#Sec11} ========== In this paper, a novel method of electromagnetic tracking is proposed, which by using quadratic excitation for position estimation reduces the error in the position estimation in the presence of the conductive distorter. The EMTS is able to compensate for dynamic distortions even when previous calibrations have not been made in the presence of the fixed distorting object. The proposed method is based on the difference in the transmittance between the voltage induced directly in the sensing coil and the voltage induced indirectly in the sensing coil across a conductive distorter. The undistorted voltage in the emitting coil depends on the first derivative of the voltage in the emitting coil. In contrast, the distorted voltage depends on its second derivative. In many applications of computer-assisted surgery, the sensing coil is placed in direct proximity of conductive surgical instruments or even attached to such instruments. The proposed method was evaluated using an aluminum disk to represent those surgical instruments used in minimal invasive surgery like hammer, scalpel or ultrasonic probe. In the performed experiment, the influence of the conductive distorter on the accuracy of the electromagnetic tracking system was measured simultaneously with a quadratic and a sinusoidal excitation. The measurements in the distance between the sensing coil and aluminum plate smaller than 20 mm show a non-significant impact of the conductive distorter on the accuracy of the electromagnetic tracking system when supplied with a quadratic excitation (Mann--Whitney $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U$$\end{document}$-test with the significance level 0.01). The same experiments performed with the sinusoidal excitation show a significant influence of the conductive distorter on its accuracy. On the other hand, it should be noted that the position estimation with the quadratic excitation is less accurate without additional distortions due to the conductive objects. The methods proposed in the literature require parameter calculations of the conductive distorter equivalent circuit. These calculations are complicated and lead to high inaccuracies, especially when there are multiple distorters to be modeled as multiple RL circuits. In the proposed method, it is not necessary to calculate these parameters of the conductive distorter equivalent circuit, and therefore, the proposed method is applicable when multiple distorters with complicated geometry are present---a common situation in surgical and radiological suites. The proposed method focuses on the distortions due to the conductive distortions and does not take into consideration distortions due to ferromagnetic distorters. The Tomasz Bien, Mengfei Li, Zein Salah and Georg Rose declare that they have no conflict of interest.

Background ========== Global growing concerns about energy shortages and the environmental pollution have led to worldwide use of renewable energy. Hydrogen is considered as a viable energy carrier for the future which could play an important role in the reduction of emissions of greenhouse gases \[[@B1],[@B2]\]. Recently, biological hydrogen production processes, especially photo-fermentative hydrogen production by PFB has been attracting more and more attention, as it utilizes various renewable sources like biomass and sunlight to produce an ideal, renewable and carbon-free energy for the future \[[@B3]\]. However, it should be realized that most of photo-fermentative processes are based on suspended culture \[[@B4]-[@B6]\] in which it is difficult to achieve high biomass concentration, effective retention and separation of PFB biomass, resulting from poor flocculation of PFB \[[@B7]\]. For steady-state operation of photobioreactor, due to the poor flocculability, PFB cells cannot be efficiently separated from supernatant and rush out with effluent from reactor continuously. This increased the effluent turbidity, meanwhile led to increases in pollutants like chemical oxygen demand, total nitrogen, and total phosphate, causing poor effluent water quality. Furthermore, to replenish the biomass washout, substrate was continuously utilized for cell growth rather than hydrogen production \[[@B8],[@B9]\]. Thus, the poor flocculability not only deteriorated the effluent quality, but also decreased the potential yield of hydrogen from substrate. Therefore, enhancing the flocculability of PFB is urgent necessary to further develop photo-fermentative process. Previous studies also tried to isolate self-flocculated PFB or enhance flocculation of PFB, but successful case was rare. Watanabe \[[@B10]\] first and only reported that photosynthetic bacteria *Rhodovulum sp*. PS88 has a self-flocculating activity. And high density cell culture was obtained under continuous cultivation in a single-tower fermenter \[[@B11]\]. However, there was no report about photo-hydrogen production and flocculation mechanism of *Rhodovulum sp*. PS88. According to the DLVO theory, the PFB, *Rhodopseudomonas acidophila*, could not overcome the total energy barrier to flocculate effectively, because contribution of van der Waals interaction energy to the total interaction energy could be neglected resulting from the small effective Hamaker constant (2.27×10^-23^ J) \[[@B12]\]. As a result, *R. acidophila* could not overcome the total energy barrier to flocculate effectively. So far, the information about PFB could flocculate and simultaneously improve hydrogen production have been not yet reported, and effective method and detailed mechanism of flocculation in photo-fermentation hydrogen production is still lacking. In this work, we first time found that the L-cysteine induced the obvious bioflocculation of *Rhodopseudomonas faecalis* RLD-53 and at the same time promoted hydrogen production. Traditionally, flocculability of biological cells highly depended on the extracellular polymeric substances (EPS) \[[@B13]\], bacterial surface characteristics \[[@B14]\] and electrolyte concentration \[[@B15]\]. However, L-cysteine is unique natural amino acids containing a thiol group, which could form disulfide bond. Disulfide bonds are crucial to the folding and stability of many proteins \[[@B16],[@B17]\], usually proteins secreted to the extracellular medium. As a predominant component in EPS, proteins have been demonstrated to play a crucial role in the bacterial aggregation \[[@B18],[@B19]\]. Therefore, the mechanism of aggregation triggered by L-cysteine was explored through combination biological function of L-cysteine and traditional flocculation theory. EPS, surface properties and zeta potential of PFB were investigated for better understanding flocculation characteristics of strain RLD-53 under different concentration of L-cysteine. Effect of disulfide bonds on components of EPS production and conformational changes of proteins in EPS were also determined. Furthermore, contribution of specific EPS protein conformation and cell surface functional groups to bacterial aggregation were further discussed. Finally, the DLVO theory was used to evaluate the flocculability of *R. faecalis* RLD-53. Results and discussion ====================== Hydrogen production and bioflocculation of *R. faecalis* RLD-53 --------------------------------------------------------------- Hydrogen productions were carried out at different L-cysteine concentrations (0.5, 1.0 and 1.5 g/l) in batch culture and the control was no addition of L-cysteine. After cumulative hydrogen production was obtained, modified Gompertz equation was used as kinetic model to determine the hydrogen production kinetics of *R. faecalis* RLD-53 at different L-cysteine concentrations (Figure [1](#F1){ref-type="fig"}). The hydrogen production kinetic parameters at different L-cysteine concentrations were shown in Table [1](#T1){ref-type="table"}. The maximum cumulative hydrogen production (*H*~*max*~) and maximum production rate (*R*~*max*~) increased with increasing L-cysteine concentration from 0.0 to 1.0 g/l, but decreased with further increasing the L-cysteine concentrations from 1.0 to 1.5 g l/l. At 1.0 g/l of L-cysteine, hydrogen production started rapidly with the lag phase of about 27.46 h, reaching the maximum hydrogen yield of 2.58 mol H~2~/mol acetate and production rate of 32.85 ml/l/h, respectively. However, at 1.5 g/l of L-cysteine, hydrogen production started slowly, did not reach the maximum rate of 20.81 ml/l/h until lag time 46.10 h and hydrogen yield was only 1.66 mol H~2~/mol acetate. The results indicated that a proper concentration of L-cysteine (1.0 g/l) could promote the hydrogen production, while excessive L-cysteine depressed the hydrogen productivity. {#F1} ###### Effect of L-cysteine concentration on hydrogen production kinetics and nitrogenase activity **L-cysteine (g/l)** **H**~**2**~**yield (mol H**~**2**~**/mol acetate)** ***H***~***max***~**(ml/l)** ***R***~***max***~**(ml/l/h)** ***λ*(h)** ***r***^***2***^ **Nitrogenase activity (nmol C**~**2**~**H**~**4**~**/ml/h)** ---------------------- ------------------------------------------------------ ------------------------------ -------------------------------- ------------ ------------------ --------------------------------------------------------------- 0.0 2.33±0.106 2610±118 24.97 36.22 0.998 957±85 0.5 2.46±0.123 2755±137 29.35 31.77 0.997 1125±78 1.0 2.58±0.129 2878±144 32.85 27.46 0.994 1374±94 1.5 1.66±0.114 1859±127 20.81 46.10 0.996 765±69 In order to further demonstrate above results, acetylene reduction was used to determine the activity of nitrogenase, which catalysed the hydrogen production in photofermentation \[[@B3]\]. As shown in Table [1](#T1){ref-type="table"}, nitrogenase activities increased with the L-cysteine concentration, reached maximum (1374 nmol C~2~H~4~/ml/h) at 1.0 g/l. However, with further increase of L-cysteine to 1.5 g/l, the nitrogenase activity sharply decreased to765 C~2~H~4~/ml/h. Thiol group from L-cysteine is a key part of active sites in nitrogenase, which play an important role in structure and function of nitrogenase \[[@B20],[@B21]\]. In chemical evolution of a nitrogenase model, the ratio of thiol and molybdenum significantly influenced the catalytic activity, and the maximum catalytic activity was obtained at ratio of 1:1 \[[@B22]\]. In this study, nitrogenase activity was enhanced by increasing of thiol from L-cysteine, but strongly depressed by excessive L-cysteine. Figure [2](#F2){ref-type="fig"} showed the cell growth and flocculability of *R. faecalis* RLD-53 at various L-cysteine concentrations. Cell biomass increased with the concentration of L-cysteine from 0 to 1.0 g/l, reached maximum (1.08 g/l) at 1.0 g/l, and then decreased sharply with further increase of L-cysteine to 1.5 g/l. The results suggested that excessive L-cysteine depressed cell growth of RLD-53. In the previous study, L-cysteine was also found to inhibit the growth of *Neurospora crassa* at high concentration, due to itself rather than its metabolism products \[[@B23]\]. The flocculability increased from 16.73 to 40.86% with an increase in L-cysteine concentration from 0.0 to 1.0 g/l, and then decreased. Above results indicated that suitable concentration of L-cysteine could significantly enhance the flocculability of *R. faecalis* RLD-53 and also promote photo-hydrogen production. After standing for 5 minutes, the absorbance of the supernatant decreased with increasing L-cysteine concentration. This suggested that after a short settlement, the biomass rush out could be significantly reduced and the effluent quality could be greatly improved in the continuous operation. Therefore, the described flocculation behaviour of *R. faecalis* RLD-53 during photo-hydrogen production is potentially suitable for practicable application. {#F2} As seen from Figure [3](#F3){ref-type="fig"}a, bioflocculation formation increased with the concentration of L-cysteine, and best floc was observed at 1.0 g/l. And then, bioflocculation decreased with L-cysteine, due to weak cell growth at high concentration of L-cysteine. The result showed that L-cysteine could cause the remarkable flocculation of *R. faecalis* RLD-53. After formation (about 48 hours), the floc was stable in whole hydrogen production, which indicated that flocculation of *R. faecalis* RLD-53 could be applied in the continuous hydrogen production. Scanning electron microscope (SEM) analysis also showed that there were clear morphological differences of bioflocculation at different concentration of L-cysteine (Figure [3](#F3){ref-type="fig"}b). Loosely structure was observed in the control (0 g/l of L-cysteine) and there were almost not formation of floc. Interaction adhesion among cells increased and formed a stable structure through covered and tightly linked together by EPS at 1.0 g/l of L-cysteine. The amount of EPS on the cell surface was increased with the L-cysteine from 0.0 to 1.0 g/l, indicating that proper concentration of L-cysteine may promote the production of EPS. The porous structure of the floc was also observed at 1.0 g/l. Such structure was likely to facilitate the passage of substrate and the release of hydrogen gas as well as light penetration. {#F3} EPS of *R. faecalis* RLD-53 --------------------------- EPS have been reported as a major component in microbial aggregates and play a crucial role in bioflocculation formation \[[@B24]\]. Characteristics and the contents of the EPS have significant effect on the formations and properties of microbial aggregates \[[@B13]\]. Figure [4](#F4){ref-type="fig"} illustrated the effect of L-cysteine concentrations on EPS chemical compositions. The total EPS content increased from 24.23 to 60.47 mg/g dry cell with the increase of L-cysteine from 0 to 1.0 g/l. A further increase in L-cysteine concentrations decreased of EPS content, due to inhibition from excessive L-cysteine. The polysaccharides, proteins and humic substances were the major component of EPS from strain RLD-53. Compared with the control (0 g/l), the EPS at 1.0 g/l of L-cysteine increased by 36.24 mg/g dry cell, mainly (18.03 mg/g) from increase of protein secretion. Previous studies reported that with high content of negatively charged amino acids, protein is more interacted than polysaccharides with multivalent cations for stabilizing floc structure \[[@B18]\]. Removal of proteins from the sludge floc resulted in deflocculation, which demonstrated the important function of protein in formation of floc \[[@B19]\]. These results indicated that L-cysteine promoted the production of extracellular polymeric substances, especially protein secretion, which facilitated bioflocculation formation. {#F4} In addition, humic substances also increased significantly (Figure [4](#F4){ref-type="fig"}). Humic substances are a natural organic matter, resulting from the biodegradation of dead biomass, which are resistant to degradation \[[@B25]\]. In the pure culture of photofermentative bacteria, humic substances mainly came from the dead cell decomposition. In free cell culture, cell debris and humic substances from the dead cell decomposition may disperse into the culture broth. However, after the bioflocculation formation, humic substances from dead cell decomposition may be retained in the EPS matrix. As a result, the humic substances also significantly increase, due to the floc formation caused by L-cysteine. Disulfide bonds formation from L-cysteine are crucial to the folding and stability of many proteins \[[@B16],[@B17]\], usually proteins secreted to the extracellular medium. Figure [4](#F4){ref-type="fig"}b presented disulfide bonds concentrations in the EPS at different L-cysteine concentration. The disulfide bonds content in EPS increased with the increase of L-cysteine concentration, and reached maximum at 1.0 g/l of L-cysteine (Figure [4](#F4){ref-type="fig"}b), corresponding to the above mentioned maximum proteins production in EPS (Figure [4](#F4){ref-type="fig"}a). Disulfide bonds detected in the control may come from the inoculation containing a small amount of L-cysteine. The relationships between disulfide bonds and components of EPS production were also investigated, as shown in Figure [5](#F5){ref-type="fig"}. Compared with other components of EPS, proteins had the most significant positive linear relationship with disulfide bonds based on correlation coefficients (*R*^*2*^=0.924) (Figure [5](#F5){ref-type="fig"}). In the previous studies, bovine pancreatic trypsin inhibitor (a model protein to investigate protein structure and folding pathways) secretion efficiency was significantly decreased by disulfide removal \[[@B26]\], while proteins secretion in *Pichia pastoris* were enhanced by overexpression of protein disulfide isomerase, which helped in rearrangement of incorrect disulfide pairings \[[@B27]\]. Therefore, our results implied that the formation of disulfide bonds from L-cysteine stimulated the secretions of proteins and it is beneficial to stabilize floc. {#F5} In addition, the biological function of disulfide bonds includes stabilization of protein structure as well as determining the pathway and efficiency of protein folding \[[@B28],[@B29]\]. Thus, the potential impacts of L-cysteine on the conformations of EPS proteins were further explored based on amide I region (1600--1700 cm^-1^) in the Fourier Transform Infrared Spectroscopy (FTIR) spectra \[[@B30]\] (Figure [6](#F6){ref-type="fig"}). The percentages of secondary structure were determined from the areas of the individual assigned bands and their fraction of the total area in the amide I (Table [2](#T2){ref-type="table"}). In the control (0 g/l), random coil was predominant in EPS proteins, while 3-Turn helix, α-Helix, β-Sheet and antiparallel β-sheet/aggregated strands also had considerable proportion. With the increasing of L-cysteine, the content of random coil and antiparallel β-sheet/aggregated strands significantly decreased, while α-Helix and β-Sheet increased. These results indicated that the protein conformations were greatly changed by disulfide bonds from L-cysteine. {#F6} ###### **Conformation changes of EPS proteins from*R. faecalis*RLD-53 at different concentration of L-cysteine** **Secondary structures** **Wavenumber (cm**^**-1**^**)** **L-cysteine (g/l)** ---------------------------------------------- --------------------------------- ---------------------- ------- ------- ------- Aggregated strands (%) 1625-1610 1.55 1.21 2.51 2.01 β-Sheet (%) 1640-1630 11.33 14.73 21.09 25.31 Random coil (%) 1645-1640 37.50 27.30 3.28 5.52 α-Helix (%) 1657-1648 16.96 25.52 49.40 31.03 3-Turn helix (%) 1666-1659 20.13 27.63 21.35 34.54 Antiparallel β-sheet/ aggregated strands (%) 1680-1695 12.53 3.62 2.38 1.59 Conformation of protein could be influenced by contact time with the matrix \[[@B31]\], hydrophobicity of surface \[[@B32]\] and curvature of matrix \[[@B33]\]. Such changes of protein molecules on bacterial surface were not reversible. So, the conformational changes of proteins were suggested as driving force for bacterial adhesion to solid matrix \[[@B34]\]. In this study, the contribution of specific conformation of EPS protein to flocculability of PFB were further analysed in Figure [7](#F7){ref-type="fig"}. Aggregated strands, β-sheets and α-helices promoted bioflocculation formation (Figure [6](#F6){ref-type="fig"}a, b and d), but random coils and antiparallel β-sheet/aggregated strands deteriorated flocculability (Figure [6](#F6){ref-type="fig"}c and f). However, there was no obvious corresponding relationship between 3-Turn helix and flocculability (Figure [6](#F6){ref-type="fig"}e). {#F7} Cell surface functional groups of *R. faecalis* RLD-53 ------------------------------------------------------ With increasing protein and EPS covering on the cell surface, the surface elemental composition and functional groups could be greatly influenced by L-cysteine. The surface elements and functional groups of *R. faecalis* RLD-53 were studied by X-ray photoelectron spectroscopy (XPS) (Additional file [1](#S1){ref-type="supplementary-material"}: Figure S1), which detected the outermost molecular layers (mainly EPS) of the cell surface (2--5 nm). The major peaks in the spectra identified by XPS were the C 1s and O 1s, and N 1s peak, with minor peaks of P, Na, Cl and Si. The functional groups on the cell surfaces were illustrated by high-resolution XPS spectra of the C 1s, O 1s and N 1s region in Additional file [2](#S2){ref-type="supplementary-material"}: Figure S2. The C 1s spectra were resolved into four individual component groups: C~G1~, 284.6 ev, [C]{.ul}-(C, H) mainly from hydrocarbons; C~G2~, 286.2 eV, [C]{.ul}-(O, N) from proteins and alcohols; C~G3~, 287.8 eV, [C]{.ul}=O or O-[C]{.ul}-O from carboxylate, carbonyl, amide, acetals, or hemiacetals, and C~G4~, 289.2 eV, O=[C]{.ul}-OH and O=[C]{.ul}-OR commonly from uronic acids. The O 1s peak was decomposed into two peaks, O~G1~, 531.3 eV, [O]{.ul}=C from carboxylate, carbonyl, ester, or amide, and O~G2~, 532.7 eV, [O]{.ul}-(C, H) from hydroxide, acetal, and hemiacetal. The N 1s peak was also resolved into two component peaks, N~G1~, 399.9 eV, O=C-[N]{.ul}H-R from amines and amides, and N~G2~, 401.3 eV, C-[N]{.ul}H~2~ mainly from basic amino acids. The percentages of surface functional groups were determined from the XPS peaks area after subtraction of a linear background (Table [2](#T2){ref-type="table"}). The results showed that the functional groups on the cell surface of *R. faecalis* RLD-53 were significantly affected by L-cysteine concentration. The contribution of cell surface functional groups to flocculability of PFB was presented in Figure [8](#F8){ref-type="fig"}. Result showed that the present of [C]{.ul}-(O, N) and [C]{.ul}=O or O-[C]{.ul}-O on the cell surface promoted photo-fermentative bacteria flocculation, while compounds with [C]{.ul}-(C, H) and O=[C]{.ul}-OH, O=[C]{.ul}-OR on the cell surface deteriorated the flocculability (Figure [8](#F8){ref-type="fig"}a-d). The flocculability of *R. faecalis* RLD-53 and content of [O]{.ul}-(C, H) followed the same trends changed with concentration of L-cysteine, but no significant linear correlation (Figure [8](#F8){ref-type="fig"}e and f). In addition, the C-[N]{.ul}H~2~ on the cell surface promoted bioflocculation whereas the O=C-[N]{.ul}H-R decreased bioflocculation ability (Figure [8](#F8){ref-type="fig"}g and h). Previously, some reports suggested that O=[C]{.ul}-OH groups play an important role in flocculation \[[@B35]\], but others emphasized the importance of C-[N]{.ul}H~2~\[[@B36]\]. In this work, the presence of C-[N]{.ul}H~2~ promoted microorganism flocculation whereas O=[C]{.ul}-OH, O=[C]{.ul}-OR hampered bioflocculation (Figure [8](#F8){ref-type="fig"}e and f). Therefore, the change of cell surface functional groups caused by L-cysteine was conducive to flocculation of PFB. ![**Contribution of cell surface functional groups to flocculability of*R. faecalis*RLD-53.** (**a**), [C]{.ul}-(C, H); (**b**), [C]{.ul}-(O, N); (**c**), [C]{.ul}=O, O-[C]{.ul}-O; (**d**), O=[C]{.ul}-OH, O=[C]{.ul}-OR; (**e**), C=[O]{.ul}; (**f**), C-[O]{.ul}H, C-[O]{.ul}-C; (**g**), C-[N]{.ul}H~2~; (**h**), O=C-[N]{.ul}H-R.](1754-6834-6-64-8){#F8} Zeta potential of *R. faecalis* RLD-53 -------------------------------------- Zeta potential is also an important parameter which characterizes the physicochemical properties of the bacterial cell envelope and plays an important role in aggregation and disaggregation processes \[[@B37]\]. The absolute zeta potential decreased with the L-cysteine concentration, and reached minimum at 1.0 g/l (Additional file [3](#S3){ref-type="supplementary-material"}: Figure S3). This was mainly attributable to the changes in cell surface characteristics, especially changes in the functional groups caused by L-cysteine. Bacterial surface charge mainly originated from ionization of surface groups \[[@B38]\], including negatively charged (e.g., carboxyl, phosphoryl or sulfhydryl groups of carbohydrates and proteins) and positively charged groups (amine groups of amino acids, amino sugars) \[[@B37]\]. In this study, cell surface of *R. faecalis* RLD-53 mainly contained [C]{.ul}OOH and C-[N]{.ul}H~2~ as charged groups (Table [3](#T3){ref-type="table"}). With increasing of [C]{.ul}OOH, [C]{.ul}OOR content, the absolute zeta potential increased, due to the negative contribution to cell surface charge from ionization of [C]{.ul}OOH group (Figure [9](#F9){ref-type="fig"}). On the contrary, absolute zeta potential decreased with the content of C-[N]{.ul}H~2~ group, because negative net surface charge decreased by positive charge from ionization of NH~2~ group. ![**Effect of charged groups on the zeta potential of*R. faecalis*RLD-53.** (**a**), [C]{.ul}OOH, [C]{.ul}OOR; (**b**), C-[N]{.ul}H~2~.](1754-6834-6-64-9){#F9} ###### Results of the high-resolution XPS analysis of the C 1s, O 1s and N 1s peak region from cell surface **L-cysteine (g/l)** **C 1s (%)** **O 1s (%)** **N 1s (%)** ---------------------- -------------- -------------- -------------- ------ ------- ------- ------- ------- 0.0 69.55 20.17 6.89 3.38 48.09 51.91 99.36 0.64 0.5 68.79 20.03 6.95 4.23 38.47 61.53 86.19 13.81 1.0 57.10 27.18 13.73 1.98 24.98 75.02 76.20 23.80 1.5 59.38 27.45 9.78 3.39 70.64 29.36 83.57 16.43 Flocculability evaluated by DLVO theory --------------------------------------- DLVO theory has been widely applied as both qualitative and quantitative models to explain microbial adhesion and aggregation \[[@B39],[@B40]\]. Here, DLVO theory was applied to predict the potential energy barrier that hindered aggregation of *R. faecalis* RLD-53 at different concentration of L-cysteine. Application of the DLVO approach requires the surface thermodynamic parameters. Surface thermodynamic properties of *R. faecalis* RLD-53 cell were calculated through the measurement of the contact angles with of three different liquids (water, formamide, and 1-bromonaphthalene) (Figure [10](#F10){ref-type="fig"}). Cell surface hydrophobicity was determined by water contact angle *θ*~*W*~. Traditionally, surfaces are divided into two categories: wetting (*θ*~*W*~ \< 90°) and non-wetting (*θ*~*W*~ \> 90°). It is worth noting that hydrophobic interactions between surfaces become effective at *θ*~*W*~ \> 65° and hydrophilic interactions at *θ*~*W*~ \< 65° \[[@B41]\]. In this study, the all of water contact angles (*θ*~*W*~ \<65°) indicated that cell surface was hydrophilic and hydrophilic interactions were effective between cells of *R. faecalis* RLD-53. Hydrophilic interaction is the proclivity of strongly polar chains, molecules or particles for repelling each other in aqueous \[[@B41]\]. The hydrophilic interactions decreased with increasing the water contact angle resulting from different L-cysteine concentrations. This was favourable for the formation of biofloc. {#F10} The total interaction energies between the microbial cells were calculated as a function of the separation distance between the cell and cell in 0.01 mol/l NaCl (Figure [11](#F11){ref-type="fig"}). In the control, interaction energy barrier was 389.77 KT. The high energy barrier indicated stable cell suspension, corresponding to poor flocculability of *R. faecalis* RLD-53 at 0.0 g/l of L-cysteine (Figure [2](#F2){ref-type="fig"}). The energy barrier dropped from 389.77 to 127.21 kT with an increase in L-cysteine concentration to 1.0 g/l. Hence, the flocculability of *R. faecalis* RLD-53 increased with the increasing L-cysteine concentration (Figure [2](#F2){ref-type="fig"}). However, when the L-cysteine concentration was further increased to 1.5 g/l, bacterial cell growth and EPS production were depressed, resulting in increase of zeta potential and electrostatic repulsion (*W*~*EL*~). Consequently, the interaction energy barrier rebounded to 193.21 KT. Therefore, the flocculability of *R. faecalis* RLD-53 decreased at 1.5 g/l of L- cysteine. {#F11} The poor flocculability of photosynthetic H~2~-producing bacterium, *R. acidophila*, was attributed to its inherent surface characteristics \[[@B12]\]. The effective Hamaker constant between *R. acidophila* and water was only 2.27×10^-23^ J, resulting in the negligible contribution of van der Waals interaction energy (*W*~*LW*~) to the total interaction energy. As a result, the interaction energy barrier between cells was up to 1665 KT in 0.01 mol/l NaCl solutions. Consequently, the bacterial cells could not overcome the total energy barrier to flocculate effectively. In this study, the effective Hamaker constant between *R. faecalis* RLD-53 and water was 5.54×10^-21^ J at 1.0 g/l of L-cysteine, suggesting that van der Waals interaction energy was important for total interaction energy. In addition, the electrostatic repulsive energy (*W*~*EL*~) decreased with decreasing of absolute zeta potential, resulting from changes of surface charged groups caused by L-cysteine. The interaction energy barrier between cells of *R. faecalis* RLD-53 was 127.21 KT. Therefore, *R. faecalis* RLD-53 flocculate effectively at 1.0 g/l of L-cysteine. Conclusions =========== In this work, L-cysteine was found to promote the effective flocculation and photo-hydrogen production of *R. faecalis* RLD-53. This finding suggested L-cysteine can be applied as flocculant for continuous photo-hydrogen production. Results showed that proper L-cysteine concentration (1g/l) improved flocculability and hydrogen productivity of *R. faecalis* RLD-53. The reasons of flocculation also were analysed. Through formation of disulfide bonds, L-cysteine not only promoted production of EPS, in particular the secretion of protein, but also stabilized the final confirmation of protein in EPS. Research also noted that the cell surface covered by EPS have been changed by L-cysteine, thus absolute zeta potential decreased with L-cysteine and reached minimum at 1.0 g/l, which greatly decreased electrostatic repulsion interaction energy based on DLVO theory. Further analysis indicated that total interaction energy barrier decreased from 389.77 KT at 0.0 g/l of L-cysteine to 127.21 kT at 1.0 g/l. This led to the *R. faecalis* RLD-53 overcome the total interaction energy barrier and flocculate effectively. Therefore, forming stable floc caused by L-cysteine offers great advantages for the realization of enhancing production yield and scale-up application in bio-hydrogen production. A better understanding the flocculation behaviour of PFB triggered by L-cysteine not only could help the design of subsequent hydrogen production process by flocculation of PFB, but also might favour the further understanding of the bioflocculation mechanism. Methods ======= Bacterium, medium and culture conditions ---------------------------------------- The photo-hydrogen producer used in this study was *Rhodopseudomonas faecalis* RLD-53 \[[@B42]\]. Acetate was used as the sole carbon source, and glutamate was used as nitrogen source in the medium for hydrogen production. The culture medium of strain RLD-53 was prepared as described in previous report \[[@B42]\]. The batch culture experiments were carried out in triplicate with 80 ml of the medium in 100 ml sealed reactors and filled with argon to maintain anaerobic conditions \[[@B43]\].The reactors were autoclaved at 121°C for 15 min. *R. faecalis* RLD-53 in the mid-exponential growth phase was inoculated into reactors. The light intensity on the outside surface of the reactors was maintained at 150 W/m^2^ by incandescent lamps (60 W). The reactors were stirred at 120 rpm at constant temperature of 35°C. Hydrogen production kinetics ---------------------------- Modified Gompertz equation has been widely accepted and used to determine the kinetic parameters of hydrogen production \[[@B44]\]. After cumulative hydrogen production curves were obtained over the course of an entire batch experiment, modified Gompertz equation was used to describe the hydrogen production kinetics at different L-cysteine concentrations: $$H = H_{\textit{max}}\textit{exp}\left\{ {- \textit{exp}\left\lbrack {\frac{R_{\textit{max}}e}{H_{\textit{max}}}\left( {\lambda - t} \right) + 1} \right\rbrack} \right\}$$ Where *t* is culture time (h); *H* is cumulative H~2~ production (ml/l medium); *H*~*max*~ is maximum cumulative H~2~ production (ml/l medium); *e*=2.71828; *R*~*max*~ is maximum H~2~ production rate (ml/l/h); and λ is the lag-phase time (h). Scanning electron microscope (SEM) ---------------------------------- Surface morphology of the bioflocculation samples was evaluated by a scanning electron microscope \[[@B45]\]. The bioflocculation samples were fixed with 2.5% glutaraldehyde and left for 1.5 h in a 4°C refrigerator. The samples were gently washed with phosphate buffer solution and then dehydrated by successive passages through 50%, 70%, 80%, 90%, and 100% ethanol. Each rinsing and dehydrating step took 10 min. The samples were refreeze dried (Hitachi E-2030, Japan) for 4 h, subsequently coated with gold powder by Sputter Coater (Hitachi E-1010, Japan) and finally attached on to the microscope supports with silver glue. Scanning electron microscope images were taken at 5 kV using an SEM (Hitachi S-3400N, Japan). X-ray photoelectron spectroscopy (XPS) -------------------------------------- Surface elements concentrations and functional groups on cell surface were determined by the XPS method, which detected the outermost molecular layers (mainly EPS) of the cell surface (2--5 nm) \[[@B46]\]. After 72 h cultivation, *R. faecalis* RLD-53 cultured at different L-cysteine concentrations were harvested by centrifugation at 12000 rpm for 10 min and washed twice with double distilled water. Collected cell samples were placed in a freeze-dryer for about 48 h (until freeze-dried). XPS was carried out on a PHI-5600 equipped with a monochromatic Al Kα source and data acquisition and processing were conducted using the PC Access ESCA version 7.2A program. The anode voltage and power were 12.5 kV and 250 W, respectively. The pressure in analysis chamber was maintained at 10^-9^ Torre during each measurement. All binding energies were referenced to the C 1s neutral carbon peak at 284.6 eV. Spectra were analysed using XPSPeak software (Version 4.1). Zeta potential and flocculability tests --------------------------------------- The bacterial cells cultured at different L-cysteine concentrations were harvested by centrifugation at 12000 rpm for 10 min and washed twice with 0.9 % NaCl solution. The bacterial cells were resuspended in 0.01 mol/l NaCl solution. These cell suspensions were used for the zeta potentials measurement (Nano-ZS, Malvern Co., UK) and flocculability tests. The absorbance of prepared cell suspensions was also measured with a spectrophotometer (Shimadzu UV-2550; Shimadzu, Kyoto, Japan) at 650 nm (*A*~*0*~). Thereafter, the cell suspensions were centrifuged at 1000 rpm for 2 min, and the supernatant optical density was measured again at 650 nm (*A*~*t*~). Thus, the flocculability of *R. faecalis* RLD-53 value can be calculated using the following equation. $$\mathit{F}\% = \left( {1 - \frac{A_{t}}{A_{0}}} \right) \times 100\%$$ EPS extraction -------------- EPS was extracted using cation exchange resin \[[@B47]\] (Dowex Marathon C, 20--50 mesh, sodium form, Fluka 91973). The bacterial cells were collected by centrifugation at 12000 rpm for 10 min. And then the cells were washed twice with 0.9% NaCl solution. Subsequently, the cells were re-suspended in ddH~2~O and transferred to an extraction beaker. And then the beaker was added resin (70 g/g-VSS) and stirred at 600 rpm for 12 h at 4°C. The samples was centrifuged at 12000 g for 30 min followed by filtration using a 0.45 μm cellulose acetate membrane to remove resin, microorganisms, and residual debris to obtain an EPS sample for further analysis.?\> Fourier transform infrared spectroscopy (FTIR) ---------------------------------------------- The FTIR spectra of EPS samples were determined using a Fourier transform infrared spectrophotometer (Spectrum One-B, Perkin Elmer, U.S.). The freeze-dried EPS samples were ground with infrared grade KBr and press into pellets and used for FTIR measurement. For each sample, 350--400 scans were collected over the spectral range of 400--4000 cm^-1^ at a resolution of 4 cm^-1^. The protein conformation was analysed from the amide I region \[[@B30]\]. Component peaks were fitted with Gaussian band profiles using the frequencies of the components deduced from the second derivatives. Interaction energy evaluated by DLVO approach --------------------------------------------- Application of the DLVO approach required the surface thermodynamic parameters, which was determined by measuring the contact angles and using the Lifshitz van der Waals acid--base approach \[[@B48]\]. The DLVO interaction energy (*W*~*Total*~) between two bacterial cells can be calculated as the sum of the van der Waals (*W*~*LW*~) and electrostatic (*W*~*EL*~) interaction energies \[[@B49],[@B50]\]: $$W_{\textit{Total}} = W_{\textit{LW}} + W_{\textit{EL}}$$ where $$W_{\textit{LW}} = - \frac{A_{\textit{BLB}}R}{12H}$$ $$W_{\textit{EL}} = 2\mathit{\pi\epsilon R}\psi_{s}{}^{2}\ln\left( {1 + \text{exp}\left( {- \mathit{\kappa H}} \right)} \right)$$ Where *A*~*BLB*~ is the effective Hamaker constant. *H* is the separation distance between the cells. *R* is the cell radius of *R. faecalis* RLD-53, determined by the Malvern Mastersizer 2000 (Malvern Instruments Ltd., UK). *ψ*~*s*~ and κ represent the stern potential and inverse of the Debye length respectively, which are related to the electric double layer interaction *W*~*EL*~. *ψ*~*s*~ could be replaced by zeta potential measurement and *κ* can be calculated from different electrolyte concentrations. Analytical method ----------------- Light intensity was measured at the surface of reactor with solar power meter TENMARS TM-207 (Tenmars Electronics CO., LTD., Taiwan, China). Biogas was sampled from the head space of the photobioreactor by using gas-tight glass syringes and hydrogen content was determined by using a gas chromatograph (Agilent 4890D, Agilent Technologies, USA). The gas chromatograph column was Alltech Molesieve 5A 80/100. Argon was used as the carrier gas with a flow rate of 30 ml/min. Temperatures of the oven, injection, detector, and filament were 35°C, 120°C, 120°C, 140°C, respectively. Residual acetate in culture broth was determined using a second gas chromatograph (Agilent 7890 A, Agilent Technologies, USA) equipped with a flame ionization detector. The liquor samples were firstly centrifuged at 12,000 rpm for 5 min, and filtered through a 0.2 μm membrane before free acids were analyzed. The operational temperatures of the injection port, the column and the detector were 220, 190 and 220°C, respectively. Nitrogen was used as carrier gas at flow rate of 50 ml/min. Whole-cell nitrogenase activity was assayed by acetylene reduction following the procedure in our previous report \[[@B51]\]. The polysaccharide content in EPS was determined by the anthrone method \[[@B52]\] using glucose as a standard. The protein and humic substance in EPS were measured followed the modified Lowry method \[[@B53]\] using bovine serum albumin and humic acid (Fluka Chemical Corp., USA) as the respective standards. The nucleic acid content was measured by the diphenylamine colorimetric method \[[@B54]\] using fish DNA as the standard. Thiol (SH) and disulfide bond (SS) in EPS were determined using 5-5′-dithio-bis (2-nitrobenzoic acid) (DTNB) according to the method of Ellman \[[@B37]\] and the procedure reported by Kalapathy et al. \[[@B55]\]. Abbreviations ============= PFB: Photo-fermentative bacteria; DLVO: Derjaguin--Landau--Verwey--Overbeek; EPS: Extracellular polymeric substances; FTIR: Fourier Transform Infrared Spectroscopy; SEM: Scanning electron microscope; XPS: X-ray photoelectron spectroscopy; DTNB: 5-5^′^-dithio-bis (2-nitrobenzoic acid). Competing interests =================== The authors declare that they have no competing interests. Authors' contributions ====================== All authors contributed intellectually via scientific discussions during the work and have read and approved the final manuscript. Guo-Jun Xie designed the study, executed the experimental work, data interpretation and drafted the manuscript. Bing-Feng Liu participated in experimental design and data interpretation, and reviewed the manuscript. De-Feng Xing helped the determination of nitrogenase activity. Nan Jun made a Hypothesis to explain humic substances increase. Ding Jie commented on the manuscript and contributed to the design of the study. Nan-Qi Ren contributed to the design of the study, data interpretation and reviewed the manuscript. Supplementary Material ====================== ###### Additional file 1: Figure S1 XPS spectra of *R. faecalis* RLD-53 at different concentration of L-cysteine (g/l). (a), 0.0; (b), 0.5; (c), 1.0; (d), 1.5. ###### Click here for file ###### Additional file 2: Figure S2 High-resolution fitted C 1s, O 1s and N 1s spectra of *R. faecalis* RLD-53 at different concentration of L-cysteine. ###### Click here for file ###### Additional file 3: Figure S3 Absolute zeta potential of *R. faecalis* RLD-53 at different concentration of L-cysteine. ###### Click here for file Acknowledgements ================ This research was supported by the National Natural Science Foundation of China (No. 51106040 and 51178140), Shanghai Tongji Gao Tingyao Environmental Science & Development Foundation, China Postdoctoral Science Foundation (No. 2012T50366), Heilongjiang Postdoctoral Financial Assistance (No. LBH-Z11120), Project 51121062 (National Creative Research Groups), 863 Program (No. 2011AA060905), the open Project of State Key Laboratory of Urban Water Resource and Environment, Harbin Institute of Technology (No. HC201212).

1. Introduction {#sec1} =============== Alpha 2-HS glycoprotein (AHSG), also designated as fetuin-A, is a 46 kDa serum protein that is synthesized mainly in hepatic cells \[[@B1]\]. The protein is known to exhibit multiple functions \[[@B2]\]. AHSG acts as a natural inhibitor of insulin receptor tyrosine kinase \[[@B3]\]. Goustin and Abou-Samra \[[@B4]\] summarized its molecular mechanism and pathogenesis. AHSG was recently revealed to be an endogenous ligand for Toll-like receptor 4 playing a crucially important role in stimulating adipose tissue inflammation, resulting in insulin resistance \[[@B5]\]. These evidences show the close relation of AHSG to occurrence of type 2 diabetes \[[@B6]\]. Moreover, AHSG is bound to calcium phosphate with high affinity. It is a major component of the extracellular matrix of the bone \[[@B7]\]. The major physiological function is to prevent extracellular calcification \[[@B8]\]. The low plasma level of AHSG from hemodialysis accelerates the stiffening and calcification of arteries \[[@B9]\]. AHSG used to be a conventional genetic marker for use in isoelectric focusing in anthropology \[[@B10]\]. Two major alleles of *AHSG* ^*∗*^ *1* and *AHSG* ^*∗*^ *2* comprise double nonsynonymous substitutions of T230M (rs4917) in exon 6 and T238S (rs4918) in exon 7 \[[@B11]\]. In addition to the amino acid replacement, the plasma AHSG level in the *AHSG* ^*∗*^ *2* homozygote is approximately 20% lower than that of the *AHSG* ^*∗*^ *1* homozygote \[[@B12]--[@B14]\]. This phenomenon means that the linked nucleotide substitutions are directly attributable to biosynthesis or degradation of mRNA. In particular, SNP (rs2248690) of AP-1 binding site in the promotor region affects the transcriptional activity \[[@B15]\]. This SNP has been nominated as the most potent association with type 2 diabetes \[[@B16]\]. However, the mechanism of the*cis*-acting expressional differences has not been elucidated. To assess the abundance of allele-specific mRNA, intraindividual assay of the heterozygote is preferred over interindividual assays, which are subject to several potential environmental errors such as variation in mRNA quality as a result from specimen degeneration \[[@B17]\]. For the intracellular assay, human hepatoma cell lines that have heterozygous AHSG genotype were searched \[[@B18]\]. In order to develop such intra-assays, the specific detection of coexisting alleles in the heterozygote is crucially important, relying on the ability to distinguish one allele from the other with no cross-reaction. Buckland \[[@B19]\] described several molecular procedures to assess allele-specific expressional differences. In the present study, we developed a novel RT-PCT procedure to quantitate the amounts of two allelic transcripts of*AHSG* separately, in which primers were designed with incorporation to the two closely located SNPs of rs4917 and rs4918. The allelic expression differences in the heterozygote were examined in the liver tissue and the culture cells of hepatic carcinoma cell lines. Finally, to elucidate the effects of degradation, mRNA levels were examined after the addition of actinomycin-D synthesis as a potent inhibitor of RNA synthesis. 2. Materials and Methods {#sec2} ======================== 2.1. Subjects {#sec2.1} ------------- To examine amounts of AHSG mRNA, four specimens of postmortem liver tissues obtained during autopsy at the Department of Forensic Medicine, Tokai University School of Medicine, were available for mRNA extraction. Then they were stored at −80°C until use \[[@B11]\]. All specimens were heterozygous in the*AHSG* genotype. After being approved by our institutional ethics committee, informed consent was obtained from the family members of deceased subjects. DNA from unrelated individuals was used to ensure the linkage of SNPs. Buccal swabs were obtained from 52 Japanese individuals for DNA extraction, as described previously \[[@B20]\]. 2.2. Cell Culture {#sec2.2} ----------------- Cells of HepG2, PLC/PRF/5, and HeLa cell lines were cultured in DMEM medium supplemented with penicillin at 100 mg/L, streptomycin at 100,000 U/L and 10% fetal bovine serum (Biological Industries Ltd., Israel) in a humidified atmosphere containing 5% CO~2~ at 37°C. The cells for culture were dispersed in 60 mm dishes with 7.5 × 10^6^ cells in 5 mL of the medium at overnight and were collected for DNA and total RNA extraction. After washing with phosphate buffered saline, the confluent cells were treated in 0.25% w/v trypsin and 1 mM EDTA at 37°C for 5 min, followed by adjustment to 1 × 10^5^ cells in 1 mL of DMEM medium. To arrest RNA synthesis, the medium was replaced with fresh medium containing 5 *μ*M actinomycin-D (Wako Pure Chemical Industries Ltd., Osaka, Japan). Cells were harvested at 0, 2.5, 5, 7.5, and 10 h after addition of the transcriptional inhibitor. Then they were examined using real-time PCR. 2.3. Extraction of DNA and RNA {#sec2.3} ------------------------------ Homogenized liver tissue of 1 g and the suspended culture cells were dissolved in Trizol solution (Invitrogen Life Technologies, Carlsbad, CA, USA). From 500 *μ*L of the cell lysates, RNA was extracted using as Rneasy Lipid Tissue Mini Kit (Qiagen Inc., Hilden, Germany) according to instructions provided by the manufacturer. Total RNA was reverse-transcribed to cDNA using Superscript III (Invitrogen) priming with oligo-dT combined with T7 RNA polymerase promoter. DNA was extracted using spin columns of the QIAamp mini kit (Qiagen). Genotyping was performed using PCR-restriction fragment length polymorphism (RFLP), as described previously \[[@B11]\]. 2.4. Allele-Specific PCR {#sec2.4} ------------------------ To detect mRNA expression levels of *AHSG* ^*∗*^ *1* and *AHSG* ^*∗*^ *2* separately in the intra-assays, allele-specific amplification was performed in the real-time PCR procedure. We designed specific oligonucleotide primer sets of 5′-AGGTTGCAGTGACCTGCA[C]{.ul}-3′ and 5′-TCTGGTTGGGGCTGTGAG[G]{.ul}-3′ for *AHSG* ^*∗*^ *1*, and 5′-AGGTTGCAGTGACCTGCA[T]{.ul}-3′ and 5′-TCTGGTTGGGGCTGTGAG[C]{.ul}-3′ for *AHSG* ^*∗*^ *2*, in which substitutions of rs4917 and rs4918 are incorporated into the 3′-terminal of both forward and reverse oligonucleotide primers. The amplified fragments were detected using Power SYBER Green PCR Master Mix (Applied Biosystems, Life Technologies) in optical 96-well MicroAmp plates using 7300 real-time PCR system (Applied Biosystems). To generate standard curves, the plasmid vectors recombined by the *AHSG* ^*∗*^ *1* and *AHSG* ^*∗*^ *2* cDNA fragments were used as the template \[[@B11]\]. Fivefold serial dilutions at 0.4, 0.08, 0.016, 0.0032, and 0.00064 ng were added to the reaction mixture of 25 *μ*L, in which 1 ng of the template corresponded to 4.18 × 10^−16^ mol and 2.52 × 10^8^ copies. To confirm the cross-reactivity to the other allele, the allele-specific primer pairs operated against the other allelic vector as the template. For the control, *β*-actin amplification using a primer set of 5′-GGACTTCGAGCAAGAGATGG-3′ and 5′-AAGGAAGGCTGGAAGAGTGC-3′ was employed using pGEMT-Easy plasmid recombined by the PCR fragment. The critical threshold (Ct) values for each PCR reaction were calculated automatically using the ABI prism sequence detection system (ver. 1.6). The starting concentration of DNA fragment present in each reaction was calculated by comparing Ct values of unknown samples to those of standards with known amounts of target DNA fragment; Ct values were plotted against the log of the initial concentration of DNA to produce a standard curve. The Ct value that is dependent upon the starting copy number of the target is defined as the cycle number at which a statistically significant increase of the reporter fluorescence can first be detected. The mean of triplicates was selected for each measurement. To compare two groups statistically, data from the experiments were analyzed using Student\'s *t*-test. Results for which *P* \< 0.01 were inferred as significant. 2.5. Linkage Analysis {#sec2.5} --------------------- PCR amplification was performed with extracted DNA from each of 52 individuals. The three SNPs examined here were presented in [Figure 1](#fig1){ref-type="fig"}. Linkage analysis was conducted using software (SNPAlyze ver. 7 standard; Dynacom Co. Ltd., Yokohama, Japan); *r* square (*r* ^2^) and the linkage disequilibrium (*D*′) were calculated. 3. Results {#sec3} ========== 3.1. Allele-Specific Detection of mRNA {#sec3.1} -------------------------------------- The first objective was comparison of the relative expression levels of two*AHSG* alleles within the same individual and the single culture cell line. The comparison in these intra-assays of the heterozygote depends on specific detection of the alleles. As [Figure 1](#fig1){ref-type="fig"} shows, the highly linked nucleotide substitutions of rs4917 and rs4918 are closely located in adjacent exons 6 and 7. In the present attempt, these two substitutions are incorporated into the 3′ end of both forward and reverse PCR primers, which yields 61 bp DNA products in length from cDNA. As presented in [Figure 2(a)](#fig2){ref-type="fig"}, no cross-reactivity was observed using the other template like the *AHSG* ^*∗*^ *1* amplification against *AHSG* ^*∗*^ *2* allelic fragment as the template. The efficiencies of amplification might be mutually distinct. Therefore, the standard curve was constructed separately to obtain the absolute copy number. However, no apparent difference of fluorescence intensity in the amplification curves was observed in the subcloned DNA fragments of *AHSG* ^*∗*^ *1* and *AHSG* ^*∗*^ *2* as the template. The standard curve was generated by regression in which amplification of the template of recombinant plasmid at various concentrations showed linearity over a range of four orders of magnitude ([Figure 2(b)](#fig2){ref-type="fig"}). 3.2. Intraindividual Assay {#sec3.2} -------------------------- The real-time PCR procedure quantitated the abundance of transcript amounts of *AHSG* ^*∗*^ *1* and *AHSG* ^*∗*^ *2* mRNA in the hepatic tissues from four heterozygous individuals. As depicted in [Figure 3(a)](#fig3){ref-type="fig"}, a relative amount of ^*∗*^ *1* mRNA was significantly greater from 2.2-fold to 3.2-fold than that of ^*∗*^ *2* with a mean and SD of 2.5 ± 0.4. As the control, the allele-specific PCR amplification was applied to genomic DNA, which caused equal amplification of ^*∗*^ *1* and ^*∗*^ *2*. 3.3. Intracellular Assay {#sec3.3} ------------------------ The allelic imbalance was evaluated in cells of culture cell lines, as well. Cells of the HepG2 cell line, of which genotyping from the genome sequence revealed *AHSG* ^*∗*^ *1* homozygote, expressed only *AHSG* ^*∗*^ *1* mRNA abundantly but not *AHSG* ^*∗*^ *2* ([Figure 3(b)](#fig3){ref-type="fig"}). The HeLa cells did not express*AHSG* mRNA, which served as the negative control. Another hepatic PLC/PRF/5 cell line was genotyped to the heterozygote. Allele-specific amplification revealed that the *AHSG* ^*∗*^ *1* mRNA count was approximately 6.2-fold greater than that of *AHSG* ^*∗*^ *2* (*P* \< 0.01) in the PLC/PRF/5 cells ([Figure 3(c)](#fig3){ref-type="fig"}). 3.4. Effect of Actinomycin-D {#sec3.4} ---------------------------- To assess the cause of allelic imbalance of*AHSG* transcripts further, actinomycin-D, which arrests RNA synthesis, was added to the culture medium for the PLC/PRF/5 cells. Cells were harvested at 2.5, 5, 7.5, and 10 h after addition of the transcription inhibitor. Then the mRNA level was quantitated using real-time PCR analysis. [Figure 4](#fig4){ref-type="fig"} demonstrates the degradation of mRNA as the ratio to the amount at time zero. No significant change was evident in the degradation curves between *AHSG* ^*∗*^ *1* and *AHSG* ^*∗*^ *2* mRNA transcripts, indicating that a major factor that affects the allelic imbalance may be attributable to transcription, but not to degradation. 3.5. Linkage Analysis from Genotyping {#sec3.5} ------------------------------------- The A/T SNP at rs2248690 in the promoter regulatory region potentially affects transcriptional activity of*AHSG* \[[@B15]\]. The genotype of rs2248690 in cells of the PLC/PRF/5 cell line was a homozygote of (A, A), although the PLC/PRF/5 cells exhibited the*AHSG* heterozygote. This result indicated no apparent interaction of the allelic imbalance of*AHSG* with the variety of AP-1 binding in the promoter region. Therefore, the linkage between the SNP and the phenotypic determinants of two SNPs (rs4917 and rs4918) was confirmed in the unrelated Japanese individuals. Incomplete linkage was found: the linkage disequilibrium values were 0.78 in *D*′ and 0.52 in *r* ^2^ ([Figure 1](#fig1){ref-type="fig"}). Moreover, in the four heterozygous specimens using the intraindividual assay included three homozygotes of (A, A) at rs2248690. 4. Discussion {#sec4} ============= Our previous work demonstrated that the AHSG serum level in *AHSG* ^*∗*^ *1* homozygote is approximately 20% higher than that of the *AHSG* ^*∗*^ *2* homozygote \[[@B12]\]. For many years, association studies in a population group showed that this polymorphism is related to various physical statures such as bone mineral density \[[@B21], [@B22]\] and leanness \[[@B13]\]. Moreover, a number of association studies with major diseases such as type 2 diabetes \[[@B6], [@B16], [@B23]\], lipid levels \[[@B24], [@B25]\], and ischemic heart disease \[[@B26]--[@B28]\] have been reported extensively. In the present study, to examine the origin of the major phenotypic differences, we developed a quantitative real-time PCR assay for the sensitive and specific detection of the two*AHSG* alleles, where two separated titrations for absolute quantification were necessary to analyze the respective alleles. This series of experiments confirmed a significant difference in the mRNA transcription level of two major alleles. In the intraindividual assay using liver tissues of the heterozygote, the *AHSG* ^*∗*^ *1* mRNA level was significantly, over two times, higher than that of *AHSG* ^*∗*^ *2* mRNA in the real-time PCR detection. In the intracellular assay using PLC/PRF/5 cells of the heterozygote, ^*∗*^ *1* mRNA level was approximately six times higher than that of ^*∗*^ *2* mRNA as well. It was readily apparent that the AHSG protein level in serum was correlated with mRNA level in the hepatocytes. This result means that some*cis*-acting mechanism operates in the process of protein biosynthesis. A number of other functional*cis*-regulatory variations are known. Among them, allelic differences of twofold or greater in the rate of transcription are common, and even more than 20-fold differences are not uncommon \[[@B29], [@B30]\]. In the experiment using actinomycin-D that inhibits mRNA synthesis, no apparent difference was observed in the ratio of ^*∗*^ *1*/^*∗*^ *2* allele numbers, suggesting that the allelic imbalance appeared not to be derived from the difference in the degeneration process of mRNA. In terms of the allelic imbalance of mRNA level, the molecular cause was unknown. However, the possibility exists of linkage of the ^*∗*^ *1* and ^*∗*^ *2* alleles with SNP in the promoter region, in which transcriptional factor AP-1 binding site is present. Inoue et al. \[[@B15]\] demonstrated that SNP at −768 nucleotide position in the AP-1 binding site affects the transcription of AHSG in the*cis*-acting manner. The AP-1 binding type ([T]{.ul}GTGTCA) is bound to AP-1 more tightly than the other AP-1 nonbinding type ([A]{.ul}GTGTCA). Their luciferase assay showed that tight binding of T-type reduces the transcriptional activity. This SNP (rs2248690) in the promoter region potentially links to *AHSG* ^*∗*^ *1* and *AHSG* ^*∗*^ *2* polymorphism and affects the allelic imbalance of mRNA by changing the transcription. The homozygote of (A, A) at −768 n.p. was observed in the heterozygous cells of PLC/PRF/5, indicating that the promoter SNP associated with AP-1 binding is unlikely to be the main cause for allele-specific imbalance of transcription. We deduce that the allelic imbalance originated from the two SNPs themselves of exons 6 and 7 or other closely linked SNPs within the haploblock. Because the two critical SNPs of rs4917 and rs4918 are present in the vicinities of exon-intron junctions, the linked substitutions potentially affect splicing of transcripts \[[@B31]\]. To examine another mRNA species caused by alternative splicing, RT-PCR using several primer pairs was performed extensively, but no evidence was obtained (data not shown). Moreover, another extensive search of the Expression Sequence Tag (EST) database at the NCBI site (<http://www.ncbi.nlm.nih.gov/nucest/>) failed to reveal any potential cDNA fragment to infer the occurrence of alternative splicing. These preliminary results suggest that the allelic imbalance was not caused by alternative splicing. We have speculated that the imbalance originated from biosynthesis of mRNA or inhibited splicing activity in *AHSG* ^*∗*^ *2*, which reduces the amount of mature mRNA, but the exact mechanism has not been elucidated through this series of experiments. It seems certain that the imbalance derived from the relation with the two major SNPs and other highly linked SNPs, at least. Although *AHSG* ^*∗*^ *2* has been derived from the older ^*∗*^ *1* allele lately, this ^*∗*^ *2* allele distributes rapidly to all examined population group, as shown previously \[[@B32]\]. This polymorphism is potentially related to natural selection in the human evolution, but its true physiological significance remains unclear. Supplementary Material {#supplementary-material-sec} ====================== ###### Genotypes of three SNPs in 52 unrelated individuals. The authors thank Professor Yutaka Inagaki, Department of Regenerative Medicine, Tokai University School of Medicine, for his kind provision of the cells of the three culture cell lines. This work was supported by a Grant-in-Aid for Scientific Research from the Ministry of Education, Culture, Sports, Science and Technology of Japan (no. 15K08884). Conflict of Interests ===================== All authors declared no conflict of interests. {#fig1} {#fig2} {#fig3} {#fig4} [^1]: Academic Editor: Ralf Lichtinghagen

Background ========== A wealth of research about the health effects of economic insecurity has provided important evidence about the impact of unemployment on mental and physical well-being \[[@B1]-[@B11]\]. Yet, investigations into the short and long-term effects of unemployment on health behaviours such as smoking, drinking and body weight have yielded mixed results. It is increasingly being recognized that the dichotomous categories of employed/unemployed inadequately explain the current unemployment situation; unemployment must be studied in the context of social assistance programs, temporary employment and other atypical employment situations \[[@B12]-[@B16]\]. Mixed findings regarding the impact of unemployment on health behaviours may partially be attributed to heterogeneity within unemployment groups defined by the receipt of unemployment benefits. Alcohol ------- Socio-environmental and stress-based theories of alcoholism propose that alcohol abuse and addiction develop as coping reactions to stressful socio-environmental conditions such as unemployment\[[@B17]\]. However, it is also known that risky behaviours, such as problem drinking, result in increased unemployment\[[@B18]\]. Thus, it is likely that both causation and selection effects are involved in the relationship between alcohol consumption and unemployment. Cross-sectional studies in Scotland\[[@B19]\] and France\[[@B20]\] have found an increased frequency of alcohol abuse among the unemployed compared to the employed. Longitudinal studies have yielded mixed results; studies in the US\[[@B21]\] and Sweden\[[@B22]\] suggest that unemployment may increase alcohol consumption, but British and Norwegian studies demonstrate no such effect \[[@B23]-[@B25]\]. There is some evidence that the relationship between unemployment and alcohol use may be time-dependent; a recent investigation suggested that short-term unemployment decreases alcohol use while longer unemployment increases it\[[@B26]\]. Smoking ------- Cigarette smoking, like alcohol use, has been found to be associated with stressful socioeconomic conditions such as unemployment\[[@B27]\]. Cross-sectional studies generally show that unemployed people are more likely to smoke than employed people\[[@B28]\], although unemployed smokers sometimes report smoking fewer cigarettes a day than their employed counterparts\[[@B29]\]. Longitudinal studies on unemployment and smoking have presented mixed findings\[[@B24]\] and are not consistent across countries\[[@B30]\]. Yet, there is some evidence that job loss is a risk factor for increasing smoking and that these effects may be long lasting\[[@B31]\]. Body Weight ----------- There is evidence from longitudinal studies that unemployment may impact body weight; data from the British Regional Heart Study\[[@B24]\] showed that men who had experienced some unemployment during the study period were more likely than continuously employed men to either lose or gain more than 10% of their body weight. In the study of a factory closure in Michigan, job loss was shown to have an impact on body weight. Subjects who lost their job showed greater instability in their weight over the two years of observation, even after re-employment\[[@B32]\]. In a nationwide representative sample of Finnish subjects, overweight in women was associated with long-term unemployment\[[@B33]\]. Yet, the association between food insecurity and body mass index (BMI) has been found to be curvilinear (i.e., thin and overweight people display different behaviours from normal weight subjects while facing periods of economic and food insecurity)\[[@B34]\]. Objectives ---------- As is the case with any epidemiological research short of a clinical trial, it is virtually impossible to control for all of the variables that could determine both employment instability and health behaviours simultaneously. However, when facing contradictory evidence on the relationship between unemployment and health behaviours, it would be precipitous to conclude that differences are due simply to a better or worse adjustment for the effects of selection. A key factor in understanding these mixed findings could be the social context in which the unemployment occurs. Systems of formal support and social benefits are known to help maintain the health status of individuals exposed to unemployment. Perceived health status has been shown to differ between groups of unemployed people defined by the types of benefits they receive\[[@B30],[@B35]\] Mental health is also affected by social assistance programs with the receipt of benefits being associated with a reduction in depression symptoms\[[@B35]\]. Our aim is to examine the impact of unemployment on smoking, drinking and body weight changes among re-employed individuals. This study expands upon previous work by investigating whether unemployment compensation influences these behaviours. Receiving unemployment compensation may help alleviate the stress of unemployment and thereby reduce the negative behavioural effects for two reasons. One reason is because receiving benefits reduces the psychological stress associated with a sharp decline in income. The second, in line with research on the study of income inequality, social capital and health\[[@B36]\], is because unemployment compensation might act as a form of social support for the unemployed. While the benefits conferred in unemployment compensation are primarily financial, previous research has shown that the individual\'s perception of support, in addition to the extent of actual supportive behaviours, mediates its health protective effects\[[@B37]\]. Thus, by reducing the financial, psychological and sociological strains of unemployment, the receipt of unemployment compensation may help reduce the negative behavioural changes associated with job-instability. Methods ======= Participants ------------ Data for this study were provided from the Panel Study of Income Dynamics (PSID) conducted by the University of Michigan, Ann Arbor. The PSID is a longitudinal survey of a representative sample of US individuals and the families with whom they reside. The study emphasizes the dynamic aspects of economic and demographic behaviour but also contains information on mental and physical health. Beginning with a national sample of 4,800 families in 1968, the PSID has traced individuals from those original households and their offspring since that time. Because the original focus of the study was the dynamics of poverty, the 1968 sample included a disproportionately large number of low income households. To make the sample more representative of the population, additional families were added in 1995 and 1999 resulting in a sample size of 6,997 households in 1999. The PSID has a very high response rate ranging between 96.9% and 98.5%. Approximately 92% of the sample is interviewed by telephone each year; the remaining 8% of the sample is interviewed in person\[[@B38]\]. The household head, defined as the husband in a husband-wife pair or the primary wage earner, is most often the respondent. Measures -------- ### Unemployment We define employment as working full-time (\> = 40 hours/week). Unemployment is defined as not working while still actively looking for work. Individuals who reported that they did not have a job and were not looking for work were not considered to be unemployed. In order to reduce the influence of prior unemployment and increase the homogeneity of the study group, only heads of household who were employed continuously in 1998, 1999 and at the time of the survey in 2001 were included in the study group. A total of 3,451 respondents were used in the analysis. The subjects were divided into three groups: those who were employed continuously in 2000 (N = 3,321), those who experienced unemployment in 2000 and received unemployment compensation (N = 51) and those who experienced unemployment in 2000 and who did not receive unemployment compensation (N = 79). The latter two groups were compared to the continuously employed group to examine their relative risk for increasing their smoking and drinking and either increasing or decreasing their body weight. Unemployment insurance (UI) is a US. federal-state system that provides partial, temporary wage replacement to eligible workers. It is the largest unemployment compensation program in the US. The eligibility for (UI) varies by state and a comprehensive review of its conditions can be found elsewhere\[[@B39]\]. Briefly, eligibility has two major components: a monetary standard and a non-monetary standard. The monetary eligibility criteria can exclude some low-wage workers and temporary workers and, as a result, many states have adopted an alternative eligibility standard based on hours worked\[[@B40]\]. The main purpose of the non-monetary standard is to ensure that individuals who quit their jobs voluntarily or are fired cannot collect UI. Additional standards ensure that workers are available for work, are actively seeking employment and have not unduly refused work. Nearly all wage and salary workers are covered by the unemployment insurance system; the main exceptions are individuals who are self-employed and seasonal workers\[[@B41]\]. The weekly benefit allowance provided by UI again varies by state, but in general the maximum amount workers can claim is between 50 and 70% of the their previous average weekly wage\[[@B41]\]. Most states provide benefits for up to 26 weeks for workers with substantial work experience\[[@B41]\]. ### Smoking, Alcohol and Body Weight In the PSID, alcohol consumption is defined as the average number of drinks per day (none, less than one, 1--2, 3--4, or 5 or more per day) over the past year. Cigarette usage is defined as the average number of cigarettes smoked per day over the past year (1--100 cigarettes). Because of the relatively small numbers of individuals in the two unemployment categories, a binary dependent variable was used as the outcome variable to increase the power of the analysis. We created binary variables to determine whether the number of cigarettes smoked per day, the number of alcoholic drinks drunk per day or body weight had either increased or decreased between 1999 and 2001. An increase was defined as a 2001 self-reported value of smoking, drinking and body weight greater than the self-reported value in 1999; a decrease was defined as the inverse. To facilitate the interpretation of results, every outcome was modelled separately. Individuals with missing data for smoking, drinking or body weight in 1999 or 2001 were removed from the analysis. In the analysis of an increase in alcohol consumption, individuals who responded that they drank five or more drinks per day in 1999 (the highest category included in the survey) could not have answered drinking more in 2001 and were removed from the model. To control for prior body weight in the analysis of an increase or a decrease of body weight, we created a variable calculating the individual\'s BMI in 1999 using the standard formula\[[@B42]\]. Statistical Analysis -------------------- A logistic regression analysis was performed to investigate whether a period of unemployment in 2000 resulted in a change in smoking, drinking and body weight among re-employed people in 2001 in comparison with baseline levels recorded in 1999. All of our models adjust for potential confounders of the relationship between the likelihood of a change in smoking, drinking and body weight and unemployment status. These include: age, sex, race, education, health status, income, number of household members and marital status of the respondents. In addition, we created different variables to control for prior smoking, drinking and weight for each of the outcome variables as described above. The income variable used was the total household post-government income created for use in the Cross-National Equivalent File\[[@B38]\]. This represents the combined income after taxes and government transfers. Likelihood ratio tests comparing models assuming the exposure showed a log-linear effect with a more general model including the exposure as a categorical variable were used to determine the most appropriate means for modelling, income, number of household members and, BMI in 1999. None of the likelihood ratio test statistics for these variables were significant for any of the outcome variables and in order to increase our power to detect an association, income, number of household members and BMI in 1999 were modelled as continuous variables. We calculated the mean increase in the number of cigarettes smoked per day between 1999 and 2001. We also calculated the mean increase and decrease in body weight (measured in both pounds and BMI) between 1999 and 2001. Because alcohol consumption was measured as a categorical variable, we were only able to determine if consumption had increased by 1, 2 or 3--4 categories (equivalent to increasing consumption by 1--2 drinks per day, 3--4 drinks per day or 5 or more drinks per day). For all three outcome variables, we investigated any differences in the magnitude of changes by employment status in 2000. The statistical analysis was performed using binary logistic regression analysis with SPSS 11.5 V and SPSS 13.0. Correlation and colinearity analyses were performed to confirm the appropriateness of the models. Results ======= Descriptive ----------- Women account for 19% of the continuously employed household heads included in the study sample, but they represent 27% and 28% of those who experienced a period of unemployment in the year 2000 with and without receiving unemployment compensation respectively. As described in Table [1](#T1){ref-type="table"}, the average age of the continuously employed was 43.4 years, 40.6 years for those who experienced a period of unemployment with unemployment compensation and 40.4 years for those who did not receive compensation. Respondents who experienced unemployment and received unemployment compensation in 2000 were unemployed for an average of 11.4 weeks, and those who did not receive compensation were unemployed for an average of 8.2 weeks. The number of weeks of unemployment was included in a preliminary analysis as a controlling variable, but it was not included in the final model because it did not reach a significant level with any of the variables of interest. ###### Descriptive Sample Characteristics by 2000 Employment Status ----------------------------------------------------------------------------------------------------------------------------------------------------------------------- Continuously employed\ Experienced unemployment with compensation\ Experienced unemployment without Compensation\ (N = 3,321) (N = 51) (N = 79) ----------------------------------------------- ------------------------ --------------------------------------------- ------------------------------------------------ Mean Age in 2001\ 43.5\ 40.6\ 40.4\ (Median, SD) (43.0, 9.9) (40.5, 9.5) (40.0, 10.1) Weeks unemployed in 2000, Mean\ 0\ 11.4\ 8.2\ (Median, SD) (0, 0) (8.0, 9.0) (4.5, 8.2) \% Women 19.2% 27.1% 28.2% %African American 25.6% 31.3% 34.6% Individual labor earnings 1999: Mean \$45,319 \$37,589 \$27,980 (Median, SD) (35,000, 45,692) (29,368, 30,832) (22,750, 20,581) Hours worked in 1999: Mean 2,327 2,259 2,139 1999 Baseline Drinking, Smoking and Weight:\*  % Did not drink alcohol in 1999 33.% 33% 30%  Mean drinking (if drinker) category\ 1.4\ 1.3\ 1.7\ (Median, SD) (1, 0.6) (1, 0.6) (2, 0.5)  % Did not smoke in 1999 78% 78% 74%  Mean \#cigarettes (if smoker) per day\ 16.3\ 16.3\ 20.8\ (Median, SD) (15.0, 10.6) (20.0, 10.1) (20.0, 12.0)  Mean BMI in 1999\ 27.3\ 27.3\ 27.7\ (Median, SD) (26.6, 4.6) (27.7, 4.7) (26.6, 5.5) ----------------------------------------------------------------------------------------------------------------------------------------------------------------------- SD = standard deviation. \*None of the differences in drinking, smoking, and body weight were statistically significant in a GLM analysis controlling for age, gender and ethnicity We investigated other possible differences among the three groups at baseline. Although the number of weeks and the hours per week worked in 1999 were similar, annual earnings were substantially different. The continuously employed in 2000 reported an average individual labour earning of \$45,316 in 1999 (for a total of 2,115 hours of work), those who experienced unemployment in 2000 but received unemployment compensation reported an average earning of \$37,589 in 1999 (for 2,259 hours of work) and those who experienced a period of unemployment in 2000 but did not receive unemployment compensation reported an average earning of \$27,980 in 1999 (for 2,139 hours of work). The income difference between the groups was not statistically significant in a GLM analysis when controlling for the variables included in our analytical models. The subjects who experienced unemployment in 2000 reported slightly higher levels of smoking, drinking and BMI in 1999 than those who were continuously employed in 2000. However, the difference was not significant when controlling for age, gender and ethnicity in a GLM analysis. Alcohol ------- Overall, 15.7% of respondents reported an increase in alcohol consumption between 1999 and 2001. Table [2](#T2){ref-type="table"} presents the results of binary logistic regression analysis to assess the impact of unemployment in the year 2000 and receipt of unemployment compensation on the likelihood of increasing alcohol consumption between the years 1999 and 2001. Prior history of alcohol consumption appeared to have the greatest effect on whether the number of alcoholic beverages consumed per day increased between 1999 and 2001. The higher the level of consumption in 1999, the less likely subjects were to increase their consumption in 2001. Respondents who experienced unemployment without receiving benefits had a greater likelihood than those continuously employed to increase their alcohol consumption (odds ratio = 1.76, 95% CI: 1.01 to 3.07). Women were less likely than men to increase their consumption. Non-married and separated individuals were more likely to increase their drinking than married individuals. Respondents with more than a high school education were less likely to increase their drinking, as were older respondents. ###### Likelihood of Increasing Alcohol drinking between 1999 and 2001 Frequency Sig. OR 95.0% C.I. ---------------------------------- ----------- -------- ------ ------------ ------ Employment status in 2000  Continuously employed 3,204 1.00  Unemployed with compensation 48 .91 1.05 .47 2.31  Unemployed without compensation 78 .04 1.76 1.01 3.07 Alcohol use in 1999  0 drinks per day 1,100 1.00  Less than 1 drink per day 1,703 \<.001 .41 .33 .50  1--2 drinks per day 394 \<.001 .28 .19 .41  3 or more drinks per day 133 \<.001 .23 .12 .44 Education  Less than high school 210 1.00  High school 947 .04 .66 .45 .98  More than high school 1,358 .01 .58 .40 .86  Missing 815 .22 .78 .52 1.16 Marital status  Married or partner 2,212 1.00  Single/living alone 419 .01 1.70 1.16 2.48  Widowed, sep. or divorced 699 .04 1.41 1.01 1.95 Income level\* 3,330 .15 1.04 .99 1.09 \# household members 3,330 .09 .93 .86 1.01 Female 652 \<.001 .49 .35 .68 Age 3,330 .11 .99 .98 1.00 Race  White 2,196 1.00  African American 861 .58 .93 .73 1.19  Other 273 .79 1.05 .72 1.52 Health status  Excellent 885 1.00  Very good 1,255 .11 1.22 .95 1.57  Good 942 .21 1.19 .90 1.56  Fair/Poor 248 .94 1.02 .67 1.55 \* The income variable used was the total household post-government income created for use in the Cross-National Equivalent File\[[@B38]\]. This represents the combined income after taxes and government transfers. A Box-Cox transformation was performed on this variable to normalize it. Among the 604 participants who reported an increase in alcohol consumption, 88.6% increased by 1--2 drinks per day, 8.9% increased by 3--4 drinks per day and 2.5% increased by at least 5 drinks per day. The proportion of individuals increasing their drinking by more than 3 drinks per day was slightly greater among those unemployed in 2000; among those employed, 11.2% increased by 3 or more drinks per day compared to 12.5% and 16.7% of individuals who were unemployed with and without compensation, respectively. The magnitude of the increase in alcohol consumption was not seen to differ greatly between those who were light and moderate/heavy drinkers at baseline in 1999. Smoking ------- In the study sample, 7.8% of participants reported an increase in smoking between 1999 and 2001. In logistic regression analysis, individuals who smoked in 1999 were significantly more likely to have increased their smoking in 2001 compared to individuals who reported smoking zero cigarettes per day in 1999 (Table [3](#T3){ref-type="table"}). Employment status in 2000 also had an effect on smoking; both groups of unemployed individuals in 2000 were twice as likely to increase their smoking by 2001 compared to individuals who were continuously employed in 2000, but this was not statistically significant. Age had an effect on smoking changes over time, with older people being less likely to increase their smoking between 1999 and 2001. ###### Likelihood of Increasing Tobacco Smoking between 1999 and 2001 Frequency Sig. OR 95.0% C.I. ---------------------------------- ----------- -------- ------- ------------ ------- Employment status in 2000  Continuously employed 3,205 1.00  Unemployed with compensation 48 .06 2.23 .95 5.20  Unemployed without compensation 78 .09 1.90 .90 4.01 Cigarette use in 1999  0 per day 2,606 1.00  1--10 per day 303 \<.001 15.52 11.12 21.67  11--20 per day 306 \<.001 5.08 3.43 7.53  21 or more per day 116 \<.001 4.08 2.24 7.44 Education  Less than high school 210 1.00  High School 947 .74 .91 .54 1.55  More than high school 1,359 .67 .89 .51 1.53  Missing 815 .87 1.05 .60 1.82 Marital status  Married or partner 2,213 1.00  Single living alone 419 .63 1.14 .67 1.94  Widowed, sep. or divorced 699 .06 1.52 .98 2.37 Income level\* 3,331 .99 1.00 .93 1.07 \# household members 3,331 .04 .88 .78 .99 Female 652 .47 .86 .56 1.31 Age 3,331 .001 .97 .96 .99 Race  White 2,197 1.00  African American 861 .29 .83 .58 1.17  Other 273 .48 1.22 .70 2.11 Health Status  Excellent 885 1.00  Very good 1,255 .75 1.06 .73 1.56  Good 943 .17 1.37 .89 1.98  Fair/Poor 248 .56 1.19 .66 2.14 \* The income variable used was the total household post-government income created for use in the Cross-National Equivalent File\[[@B38]\]. This represents the combined income after taxes and government transfers. A Box-Cox transformation was performed on this variable to normalize it. Among those who increased smoking, the average increase was by 8.5 cigarettes per day (95% CI 7.7--9.4). The average increase was slightly higher among those who were unemployed in 2000 and did not receive compensation. Individuals who experienced unemployment without compensation had an average increase of 9.5 cigarettes per day (95% CI 5.9--13.0). Those who were employed had an average increase of 8.4 (95% CI 7.4--9.3) and those who were unemployed with compensation had an average increase of 5.1 (95% CI 2.3--8.0). Body Weight ----------- Overall, 30.0% of participants reported a decrease in body weight between 1999 and 2001, 47.6% reported an increase and 22.7% reported no change. As reported in Table [4](#T4){ref-type="table"}, unemployed respondents who did not receive unemployment compensation were significantly more likely to lose weight than those who were continuously employed. Individuals who were unemployed and received unemployment benefits showed no significant difference. Women were less likely to decrease their weight than men. Age also had an effect, with older groups increasing their probability of decreasing weight. Baseline BMI was significant: those with a higher BMI were more likely to experience a decrease in their body weight. Respondents living in households with more members were less likely to decrease their body weight. ###### Likelihood of Increasing or Decreasing Weight between 1999 and 2001 a: Likelihood of Increasing Weight between 1999 and 2001 ---------------------------------------------------------- --------------- -------- ----- ------------ ------- Frequency Sig. OR 95.0% C.I. Total = 3,331 Lower Upper Employment status in 2000 Continuously employed 3,205 1 Unemployed with compensation 48 0.77 0.9 0.51 1.64 Unemployed without compensation 78 0.25 0.8 0.48 1.21 BMI in 1999 3,331 \<.001 1 0.94 0.97 Education Less than high school 210 1 High School 947 0.73 1 0.7 1.29 More than high school 1,359 0.77 1.1 0.77 1.42 Missing 815 0.69 0.9 0.68 1.29 Marital status Married or partner 2,213 1 Single living alone 419 0.97 1 0.75 1.34 Widowed, sep. or divorced 699 0.75 1 0.75 1.23 Income level\* 3,331 0.46 1 0.98 1.05 \# household members 3,331 0.25 1 0.98 1.09 Female 652 0.06 1.3 0.99 1.6 Age 3,331 0.03 1 0.98 1 Race White 2,197 . 1 African American 861 0.02 1.2 1.04 1.47 Other 273 0.72 1.1 0.8 1.39 Health Status Excellent 885 1 Very good 1,255 0.38 1.1 0.91 1.29 Good 943 0.15 1.2 0.95 1.4 Fair/Poor 248 0.02 1.4 1.04 1.9 b: Likelihood of Decreasing Weight between 1999 and 2001 Frequency Sig. OR 95.0% C.I. Total = 3,331 Lower Upper Employment status in 2000 Continuously employed 3,205 . 1 Unemployed with compensation 48 0.4 1.3 0.71 2.38 Unemployed without compensation 78 0.02 1.7 1.09 2.76 BMI in 1999 3,331 \<.001 1.1 1.05 1.09 Education Less than high school 210 1 High School 947 0.27 0.8 0.6 1.15 More than high school 1,359 0.14 0.8 0.57 1.09 Missing 815 0.94 1 0.7 1.38 Marital status Married or partner 2,213 1 Single living alone 419 0.72 0.9 0.69 1.3 Widowed, sep. or divorced 699 0.28 1.2 0.89 1.51 Income level\* 3,331 0.6 1 0.95 1.03 \# household members 3,331 0.05 0.9 0.88 1 Female 652 0.02 0.7 0.56 0.95 Age 3,331 0.04 1 1 1.02 Race White 2,197 1 African American 861 0.48 1.1 0.88 1.3 Other 273 0.15 1.2 0.93 1.67 Health Status Excellent 885 1 Very good 1,255 0.1 0.9 0.7 1.03 Good 943 0.44 0.9 0.74 1.14 Fair/Poor 248 0.2 0.8 0.58 1.12 \* The income variable used was the total household post-government income created for use in the Cross-National Equivalent File\[[@B38]\]. This represents the combined income after taxes and government transfers. A Box-Cox transformation was performed on this variable to normalize it. Employment status in the year 2000 did not significantly affect the risk of increasing body weight. Reporting fair or poor health significantly increased the risk of reporting body weight increases. African Americans were also at a significantly greater risk of increasing their body weight than were whites. Among those whose weight decreased, the mean weight-loss was 7.1 pounds (95% CI 6.6--7.6) or 0.9 BMI units (95% CI 0.8--1.0). The mean decrease in body weight did not greatly differ according to employment status in 2000; in the employed group, the mean decrease was 6.9 pounds (95% CI 6.4--7.4) while in the unemployed groups the mean decrease was 8.0 (95% CI 4.2--11.7) and 8.3 (95% CI 5.2--11.5) for those with and without benefits, respectively. Among those who increased weight, the mean increase was by 12.8 pounds (95% CI 12.2--13.4) or 1.8 BMI units (95% CI 1.74--1.94). For those employed in 2000, the mean increase was 12.6 pounds (95% CI 11.9--13.2). For those unemployed with compensation the mean increase was 11.5 pounds (95% CI 8.5--14.6) and among those without compensation the mean increase was 14.8 (95% CI 8.2--21.6). The mean increase and decrease in BMI when stratified by employment status showed a similar pattern to weight change in pounds. We used multiple regression models to investigate possible differences among the employment groups of interest and did not find statistically significant differences in the percentage of body weight change. Discussion ========== Our main findings are that employment and benefit status had an effect on the likelihood of changes in drinking and body weight. Those who experienced unemployment in the year 2000 and did not receive unemployment compensation were more likely to increase daily alcohol consumption by 2001 and decrease body weight. Those who experienced unemployment in 2000 were more likely to increase daily smoking by 2001 compared to the continuously employed but this was not statistically significant. We did not find evidence of large differences in the magnitude of smoking, drinking and body weight changes among employment groups. Among those who decreased body weight, the magnitude of weight-loss was similar in those who were unemployed without compensation and those who were continuously employed in 2000. Similarly, the magnitude of increases in alcohol consumption did not greatly differ between the two groups. Rather, a higher proportion of people who experienced unemployment and did not receive compensation reported weight loss or an increase in alcohol consumption compared to the continuously employed. As drinking and smoking patterns show some degree of correlation, investigation into the impact of unemployment on increases in both smoking and alcohol consumption would be of use. However in this study, very few unemployed individuals increased both their smoking and alcohol consumption and such an analysis would be underpowered. Although there is strong consensus in considering smoking increases as unhealthy behaviours that increase the risk of early death, we are more tentative in our interpretation of a decrease in body weight. A decrease in body weight is not a healthy sign necessarily because unintended body weight reductions could be a symptom of ill-health, and it is unknown if the observed weight drop was intentional or unintentional. Our findings are consistent with previous research reporting instability in body weight after unemployment, which may reflect a stress reaction. While the association between unemployment compensation and body weight change did not persist when body weight changes were restricted to a change of \>5%, we think it useful to report even small changes since modest fluctuations in weight (such as the mean 7 pound weight-loss found in our study) are known to be associated with stress\[[@B43]\]. Moderate increases in alcohol consumption, as seen in our study, cannot be uniformly interpreted as detrimental to health. Studies generally report a J-shaped relationship between alcohol use and risk of total mortality with light to moderate drinkers showing a decreased risk compared to abstainers and heavy drinkers\[[@B44]\]. A health-protective effect of alcohol is generally defined as one or two drinks per day for men and one drink per day for women, although some disagreement still exists regarding the cut-offs for healthy drinking\[[@B45]\]. We explored the possibility of limiting our analysis to increases which resulted in unhealthy drinking in 2001 (alcohol consumption of at least 3 drinks a day for men and at least 2 drinks a day for women). However, the number of respondents available was too small and the data available precluded further investigation due to lack of power. While the higher likelihood of increasing drinking following a period of unemployment without compensation may not translate into an increased risk of poor health outcomes for all members of this group in the short term, we do find our results of concern as it is not possible to predict in which individuals modest increases in drinking will eventually lead to unhealthy drinking patterns\[[@B46]\]. Our finding that job loss increases the long-term risk of increasing alcohol consumption is consistent with previous literature showing an association between unemployment and an increase in alcohol consumption \[[@B21],[@B22]\] but differs from other studies that report no such association \[[@B23]-[@B25]\]. This discrepancy could be due to differences in selection, but could also be due to differences in the social context of unemployment, particularly when comparing this study to studies performed in other countries. Our results indicate that those who experience periods of unemployment are more vulnerable and at a higher risk of adapting potentially unhealthy behaviours than the continuously employed. The difference was especially pronounced for those who did not receive unemployment compensation, even though, on average, they experienced fewer weeks of unemployment than those who received benefits. Overall, our findings suggest that higher levels of stress exist among those who experience unemployment without compensation. This is an area that has not been previously explored and we hope that our findings will lead to further exploration of the association between unemployment compensation and body weight, smoking and alcohol consumption. This study has certain limitations. As in any self-reported panel data our results could be influenced by information bias, particularly related to the use of self-reported smoking, drinking and body weight. The impact of possible information bias on our results is difficult to predict but seems unlikely to fully explain our findings. Because the variables on self-reported smoking and drinking used in this study reflect usual consumption patterns over the past year, we were not able to capture binging patterns or other fluctuations in substance use. The PSID is an established survey with a high participation rate which decreases the likelihood of selection bias. We attempted to control for the impact of a wide variety of potential confounders in our models. However, as little is known regarding the relationship between unemployment compensation and lifestyle changes, we cannot completely rule out the possibility of residual confounding. In order to increase the generalizability of the PSID sample to the general US population, we replicated the analysis, weighting the cases by the 2001 longitudinal weight variable created by the PSID\[[@B38]\]. While the main results remained the same, the significance levels for the explanatory and control variables were higher. We attenuated the possible effects of reverse causation by using prospective longitudinal data. In order to reduce the effects of unknown individual characteristics that might predispose individuals to unemployment, we analyzed only heads of household who were employed in 1998, 1999 and 2001. Those restrictions limited our sample size to 3,451 respondents, of which 51 experienced unemployment and received unemployment compensation in 2000 and of which 79 were unemployed but did not receive compensation. Because of this small sample size, we were unable to provide precise estimates of the magnitude of changes in smoking and body weight seen in the unemployed groups and limited in our ability to study heavy drinking patterns. Despite these limitations, this study provides important evidence that clears the way for further investigations. In our study population, only 38% of respondents who were unemployed in 2000 reported receiving unemployment compensation which is consistent with national averages for receipt of unemployment insurance \[[@B47]\]. Information on why the unemployed workers did not receive benefits is not available. Failure to meet income requirements was probably not a primary reason since the average earnings of those unemployed who did not receive benefits exceeds the minimum wage requirements for UI in most states\[[@B39]\]. Those who did not receive unemployment compensation could have been more likely to have voluntarily left their job than those who did receive benefits. While this could represent a possible source of heterogeneity between employment groups, we do not consider it a source of confounding in our results given that a predicable change in employment status due to voluntary resignation is likely less harmful than an uncontrollable loss of income\[[@B48]\]. It is estimated that that over half of workers who are eligible for unemployment insurance do not file for benefits and we think it unlikely that ineligibility completely explains why a large portion of our study population did not file for UI benefits\[[@B47],[@B49]\]. A recent report by the United States Government Accountability Office found that the most important predictor of whether eligible workers filed for UI benefits was if they had received them in the past. The report concludes that \"a worker\'s perception of UI when faced with unemployment is key to whether that will worker will ever use the program\[[@B47]\].\" Perceptions of support are known to be crucial in determining if any beneficial health effects are achieved from systems of social support \[[@B50]\]. Thus, the same favourable perceptions of UI that leads workers to file for benefits may also be responsible for the beneficial health effects seen in our results. This study focuses on the US and it will be interesting to replicate it in different countries with different compensation schemes. Previous work looking at the impact of unemployment benefits on health status has shown remarkable similarities among the US and European countries with different compensation schemes such as those found in Germany and the UK\[[@B35]\]. Conclusion ========== The stated goals of UI revolve around financial stabilization both for the individual worker and the overall economy. However, unemployment assistance programs should not only aim to reduce the financial impact but also the detrimental health effects of unemployment. This study suggests that UI may help to alleviate the drinking and body weight changes associated with employment instability. Recipiency rates for unemployment insurance have dropped steadily over the past 40 years, sparking recent legislative changes to make the UI program more responsive to the needs of US workers\[[@B41]\]. Continuing to develop an understanding of the effects of unemployment compensation on health will be increasingly important as rapid and substantial changes occur in the UI program and the overall structure of employment. Further research in this area is necessary to explore possible differential behavioural impacts of workers in different occupational categories and possible effects on concurrent policies geared toward decreasing tobacco and alcohol consumption in the general population. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= Both KLB and ER conceived of the design of study, participated in the statistical analysis and drafted the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2458/9/77/prepub> Acknowledgements ================ We would like to acknowledge Paul Pharoah and Jonathon Tyrer for their statistical support and Kirsten Colquhoun for her editorial support.

Introduction ============ The elimination of preventable blinding diseases as a public health problem by the year 2020, is one of the three components of the World Health Organization (WHO) "VISION 2020 initiative: the Right to Sight." According to WHO, about 284 million people in the world suffer from visual impairment \[[@r1]\]. Recent studies and patented new ophthalmic compositions suggest a potential prevention of visual dysfunction and a potent medicine to treat blinding ocular diseases using antioxidants in the eye \[[@r2]-[@r4]\]. Such specificity may be explained by the fact that diet rich in antioxidants (vitamins) and developed pro drugs prevent reactive oxygen species (ROS)-induced oxidation of lipids and proteins in the inner mitochondrial membrane in vivo and outside mitochondria in the cellular and tissue structures of the lens and eye compartments \[[@r2]\]. However, under pathological conditions, light exposure, aging, ROS, pro-oxidants (free radicals, atoms or clusters of atoms with a single unpaired electron determined by external and internal factors under aerobic metabolism), can increase, surpassing the body's detoxification capacity and thus contribute to molecular level organic pathology \[[@r5]\]. The literature provides exhaustive catalogs of all available oxidative stress measures. Available biomarkers of oxidative stress in our context proceed by lipid peroxidation (8-isoprostane and oxidized low-density lipoprotein) and by DNA oxidation (8-hydroxy-deoxyguanosine). Oxidation of lipid molecules may result in reduced structural fluidity of these compounds with loss of integrity of cellular membrane, while oxidation of DNA molecules can result in mutation \[[@r5]\]. Because of epidemiologic transition (emergence of non-communicable diseases such as hypertension, metabolic syndrome, obesity, and diabetes mellitus versus decrease of infectious diseases), lifestyle changes (physical inactivity, smoking, alcohol intake), urbanization, rural-urban migration, poverty, advanced nutrition transition \[[@r6]\], and demographic transition (Aging), Central Africans in Kinshasa town are exposed to external sources of free radicals such as environment with ionizing radiation and pollution (heavy metals in soils and food plants, dioxins in clays; unpublished), cigarette smoke, as well as dietary intake of excessive alcohol, unsaturated fat, refined sugars, and salt, but lack or very low intake of fruits and vegetables \[[@r7]\]. They are also at higher risk of cardiometabolic and ocular pathologies often associated with oxidative stress and/or with visual disability \[[@r8]-[@r14]\]. Up to now, no published data in the Democratic Republic of Congo (DRC), Central Africa, addressed these issues. Therefore, this study aimed to estimate the normal values of oxidant status defined by oxidized low-density lipoprotein (Ox-LDL), 8-isoprostane, and 8-hydroxy-deoxyguanosine (8-OHdG), and to determine their pathogenic role in the prevalence and the severity of visual disability among these Central Africans. Methods ======= Design, period, and setting --------------------------- This was a cross-sectional, analytic and community-based survey conducted in Kinshasa town, RDC, from July to September, 2010, from a case-control study. Kinshasa, the largest city (7 million inhabitants) enjoys a tropical climate and constitutes an Hinterland with 4 administrative districts (Mont Amba, Funa, Lukunga, Tshangu). Permission for the study was obtained from the authorities of St. Joseph Hospital in Kinshasa Limete, RDC. The study protocol was approved by the University of Kinshasa Medical Ethics Committee and the study was performed in full compliance with the Declaration of Helsinki II. Participants and sampling ------------------------- Two groups were divided on type 2 diabetes and non-diabetic participants. A 10% simple random of adults with type 2 diabetes mellitus (T2DM, n=156 cases) was drawn from the list of all T2DM patients managed at the Ophthalmology division, St Joseph hospital, DRC, Kinshasa between May and June 2010. These cases were matched for gender and age to fifty two apparently healthy controls from Kinshasa general population; 13 participants randomly selected from each district. Among eligible cases, six refused to participate in, while two individual eligible controls did not accept to participate in the study. Data collection --------------- The structured and standardized questionnaire, administered to each participant during 30 min, sought relevant information on age, gender, rural-urban migration, residence, education level, cigarette smoking, alcohol intake, socioeconomic status (SES), ethnicity and diet. Dietary assessment ------------------ To assess dietary intake, a-24 H recall of eating taro leaves (*Colocasia Antiquorum* ; Yes or No) and a 10-Month recall of eating Safou fruit (*Dacryodes edulis*; Yes or No) were collected. Clinical measurements --------------------- Weight, height and waist circumference (WC) were measured in a standard fashion by trained, certified observers. Height was measured with a portable stadiometer and recorded to the closest 0.1 cm. WC was measured with a non stretchable tape measures to the nearest 0.1 cm. Weight was measured with a Soehnle beam scale (Soenle-Waagen Gmbh Co, Murrhardt, Germany) to the nearest 0.1 Kg. All instruments were calibrated once weekly. Blood pressure including systolic blood pressure (SBP) and diastolic blood pressure (DBP), was measured from the right arm of the seated participant after 15 min with an electronic validated digital devices (OMRON M7; Intelle/Sense, Kyoto, Japan). Laboratory data --------------- Fasting (10--12 overnight fast and Post-prandial fast) blood glucose (hexokinase glucose-6-phosphate dehydrogenase reaction), triglycerides, total cholesterol, low density lipoprotein cholesterol (LDL-C), and high density lipoprotein cholesterol (HDL-C) were measured on commercially available kits (Biomerieux, Marcy l'Etoile, France) and a Hospitex autoanalyzer (Hospitex Diagnostic, Florence, Italy). Antibodies against Oxidized LDL-cholesterol (Ox-LDL) were measured using solid phase two-side immunoassay based on the direct sandwich technique (Mercodia AB, Uppsala, Sweden). Commercial Cayman's kits (Cayman Chemical Company, Ann Arbor, MI) were used to measure 8-Isoprostane in plasma. Serum 8-OHdG levels were measured using a competitive enzyme linked immunosorbent assay method on Biomerieux Reader version and commercial Kits supplied by Northwest Laboratories (Northwest Life Science Specialties, LLC, Vancouver, Canada). Eye examination --------------- Eye examination of each participant included visual acuity measurement, ocular alignment and motility, pupil reactivity and function, visual fields, intraocular pressure, slit lamp examination of the cornea, iris, lens and vitreous, and dilated fundus examination. This fundus examination was detailed and performed at the best possible mydriasis, after dilating the pupils with tropicamide 1% and phenylephrine 10%, by indirect ophthalmoscopy at the slit lamp (Haag Steit 900, Koeniz, Switzerland) with 90 D lens. Participants were refracted with the use of standard subjective refraction techniques. Visual acuity (VA) was measured separately for each eye and was defined according to the lowest line on the Snellen chart for which the majority of letters were read correctly, with the full required distance correction as determined by the decimal optometric scale. Best corrected VA was defined as the VA in the better eye with full distance correction. Due to limited resources, retinal photography was excluded as a diagnostic tool in sub-Saharan Africa \[[@r15]\]. Definitions ----------- Diagnosis of T2DM was based on criteria established by the American Diabetes Association Expert Committee \[[@r16]\]: exhibiting either a fasting plasma glucose concentration of ≥126 mg/dl (7.0 mmol/l) on more than one occasion, and/or pharmacological treatment of diabetes. Total obesity was defined by body mass index (weight in Kg/height in m^2^) ≥30 Kg/m^2^ \[[@r17]\]. Metabolic syndrome (Mets) was diagnosed according to the IDF criteria: WC≥94 cm for men, WC≥80 cm for women, triglycerides levels ≥1.7 mmol/l, HDL-cholesterol \<1.03 mmol/l (male) or \<1.29 mmol/l (female), SBP ≥130 mmHg or DBP ≥85 mmHg, or treatment for lipid abnormalities, and fasting plasma glucose ≥100 mg/dl (5 mmol/l) \[[@r18]\]. Arterial hypertension was defined as SBP ≥140 mmHg or DBP ≥90 mmHg or under antihypertensive drug treatment as recommended by the ISF/WHO guidelines Committee \[[@r19]\]. Aging was defined as age ≥60 years. The intake of ≥4 glasses of beer/day was considered as excessive alcohol intake. Smoking habits were classified into 2 groups: never smoker and current smoking ≥1 cigarette per day. Longer duration of diabetes was ≥5 years (median). A diagnosis of diabetic retinopathy (DR) was made only for participants who had a minimum of one microaneurysm in any field, in addition to exhibiting hemorrhages (dot, blot, or flame shaped) and maculopathy (with and without clinically significant macula edema). For classification of DR, the modified Airlie House classification as introduced by Early Treatment Diabetic Retinopathy Study (ETDRS) \[[@r20]\] was used as follows: non proliferative (NPDR), proliferative (PDR), and maculopathy. Visual Disability (VD) was defined as blindness and visual impairment using the World Health Organization (WHO) definition and the recommendation for the revision of classification \[[@r21]\]. Blindness was defined as VA \<0.1 (\<6/60) and visual impairment (low vision) was defined as VA \<0.3≥0.1 (\<6/18≥6/60). Normal vision was defined by a VA between 1.0 and 0.3 (6/6--6/18). Uncontrolled diabetes was defined by fasting plasma glucose ≥126 mg/dl at evaluation. Self-reported ethnicity included Kongo tribe from South-West of DRC, Ngala tribe from the North-West, Luba tribe from the Centre and Swahili people from the Eastern part of DRC. According to the 4 administrative districts of Kinshasa town, the environment of residence of participants was defined as rural for Lukunga and Tshangu districts and urban for Funa and Mont Amba districts. The education level included illiteracy, primary school, high school, and university levels. Low education level was defined as illiteracy and primary school levels. High education level included secondary school (High school) and university levels. Socioeconomic status (SES) included low (lack of income, unemployment) and high SES. Statistical analysis -------------------- Qualitative data were expressed as frequency (number=n) and proportions (%). Continuous data were presented as mean±standard deviation (SD). Comparisons of variables between two groups were performed with Student *t*-test and Chi-square test for continuous and categorical variables, respectively. Comparisons of variables between three groups were performed with one way ANOVA (ANOVA) and Bonferroni Post-Hoc test for continuous variables, while P for trend test was used in comparisons of proportions of qualitative variables. The univariate risk of visual disability was assessed in calculating Odds ratio (OR) with 95% confidence intervals (95% CI). Multivariate analyses such as logistic regression models, were used to assess the independent effect of each biomarker of oxidative stress on the presence of visual disability after adjusting for the effect of confounding factors and avoiding collinearity in each logistic regression model. The receiver operating characteristic curve for Ox-LDL, 8-Isoprostane, and 8-OHdG to predict the presence of VD, was plotted. The optimal cut off point of biomarkers were calculated by plotting sensitivity against 1-specificity. A p-value \<0.05 was considered significant. Data analysis was performed using the Statistical Package for Social Sciences (SPSS) for Windows version 19 (SPSS Inc., Chicago, IL). Results ======= A total of 200 participants (response rate of 96.2% in type 2 diabetics and response rate of 96.2% in controls) including 90 males and 110 females (sex ratio of 1:1) were examined. The prevalence rates were 8.5% for blindness (n=17), 20.5% for visual impairment (n=41) and 29% for visual disability (n=58). The mean values of continuous variables of participants according to the presence of visual disability are shown in [Table 1](#t1){ref-type="table"}. The mean values of WC, SBP, blood glucose, 8-isoprostane, and 8-OH-dG were significantly higher in people with visual disability than participants without visual disability. [Figure 1](#f1){ref-type="fig"} and [Figure 2](#f2){ref-type="fig"} showed the ROC curves that served to calculate AUC and elevated cut-offs for 8-isoprostane level\>55 ng/ml (sensitivity=78% and specificity=100%) and 8-OH-dG level\>40 mg/ml (sensitivity=76% and specificity=100%). ###### Comparisons of continuous variables according to Visual Disability (VD) in all participants. **Variables of interest** **Presence of VD mean±SD** **Absence of VD mean±SD** **p-value** ------------------------------------------------------------------- ---------------------------- --------------------------- ------------- Age (years) 56.7±15.3 53.3±11.8 0.086 Weight (Kg) 67.2±11.1 66.4±14.1 0.687 Height (m) 1.643±0.074 1.659±0.474 0.796 BMI (Kg/m^2^) 24.9±4.4 25±5 0.972 WC (cm) 94.8±16.6 90±14.6 0.040 SBP (mmHg) 131.4±24.2 124.3±19.6 0.032 DBP (mmHg) 79.5±12.8 77.8±12.2 0.389 Ox-LDL (U/L) 62.6±18.6 56.8±20.8 0.67 PG in post prandial for non diabetics or FPG in diabetics (mg/dl) 186.1±82.8 158±73.7 0.019 8-isoprostane (ng/ml) 101.3±51.4 58.8±32.3 \<0.0001 8-OHdG (mg/ml) 73.7±27.8 46.3±21.8 \<0.0001 Total cholesterol (mg/dl) 188.3±48.1 198.3±56.9 0.238 Triglycerides (mg/dl) 125.8±46.4 114.2±46.5 0.110 LDL-C (mg/dl) 81.3±41 78.5±45.3 0.675 HDL-C (mg/dl) 51.5±14.7 53.9±15.5 0.312 {#f1} {#f2} Age ≥60 years, migration, low education level ([Table 2](#t2){ref-type="table"}), as well as no intake of safou fruit, presence of T2DM, elevated Ox-LDL, elevated 8-Isoprostane, and elevated 8-OH-dG had individual positive and significant effect on development of visual disability. ###### Univariate risk of visual disability (VD) in all participants: role of gender, age, smoking, excessive alcohol intake, education level, and migration. ** ** **Presence of** ** ** ------------------------------ ----------------- ----------------- ------- **Gender** Males 31 (34.4) \- ** ** Females 31 (28.2) \- 0.341 **Aging** ≥60 years 28 (36.8) 1.8 (1.01--3.4) 0.040 \<60 years 30 (24.2) ** ** ** ** **Smoking** Yes 5 (45.5) \- 0.180 No 56 (29.8) ** ** ** ** **Excessive alcohol intake** Yes 17 (35.4) \- 0.448 No 45 (29.6) ** ** ** ** **Education level** Low 43 (33.6) 1.9 (1.02--3.8) 0.039 High 15 (20.8) ** ** ** ** **Migration** Yes 46 (34.6) 2.4 (1.2--5) 0.014 No 12 (17.9) ** ** ** ** [Figure 3](#f3){ref-type="fig"} shows a significant (p=0.046) and unequal distribution of visual disability rates across Kinshasa districts; the highest rates observed in the two urban districts of Funa and Mont Amba gave a sum of 37.4% versus 19% for rural districts of Lukunga and Tshangu. {#f3} Univariate risk factors of visual disability such as no intake of safou, taro leaves and red beans, total obesity, abdominal obesity, T2DM, elevated Ox-LDL, elevated 8-isoprostane, and high 8-OHdG are shown in [Table 3](#t3){ref-type="table"}. ###### Univariate risk of visual visability (VD) in all participants conferred by dietary intake, cardiometabolic conditions and elevated levels of biomarkers of oxidative stress. ** ** **Presence of VD** ** ** ------------------------------------------ -------------------- -------------------- ---------- **Safou fruit intake** No 40 (88.9) 60.9 (21.3--174.3) \<0.0001 Yes 18 (11.6) ** ** ** ** **Taro leaves (*Colocasia antiquorum*)** No 47 (31.3) \- 0.860 Yes 15 (30.0) ** ** ** ** **Red beans intake** No 23 (35.4) \- 0.352 Yes 39 (28.9) ** ** ** ** **Total obesity** Yes 37 (29.1) \- 0.956 No 21 (28.8) ** ** ** ** **Abdominal obesity** Yes 30 (32.6) \- 0.299 No 28 (25.9) ** ** ** ** **T2DM** Yes 53 (35.3) 4.9 (1.8--13.1) \<0.0001 No 5 (10.0) ** ** ** ** **Elevated Ox-LDL** Yes 44 (34.6) 2.2 (1.1--4.4) 0.020 No 14 (19.2) ** ** ** ** **Elevated 8-isoprostane** Yes 47 (40.2) 4.4 (2.1--9.2) \<0.0001 No 11 (13.3) ** ** ** ** **Elevated 8-OHdG** Yes 48 (39.3) 4.4 (2.1--9.4) \<0.0001 After adjusting for taro leaves (*Colocasia antiquorum*) intake, red beans intake, T2DM, aging, WC, and SBP, the regression logistic model 1 identified low education level, migration, and elevated Ox-LDL as the independent and significant determinants of visual disability in this population ([Table 4](#t4){ref-type="table"}). ###### Independent deleterious role of low education, migration and elevated oxLDL on visual disability (VD) in Central Africans. **Independent variables** **B coefficient** **Standard error** **Wald χ^2^** **OR (95% CI)** **p-value** ------------------------------ ------------------- -------------------- --------------- ------------------ ------------- **Education level** Low versus High 1.192 0.400 8.877 3.3 (1.5--7.2) 0.003 **Migration** Yes versus No 0.945 0.395 5.724 2.6 (1.2 -- 5.6) 0.017 **Oxidized LDL-C** Elevated versus Normal level 0.820 0.375 4.788 2.3 (1.1 -- 4.7) 0.029 Constant \- 3.049 0.555 30.174 \<0.0001 [Table 5](#t5){ref-type="table"} shows that no intake of safou fruits, age ≥60 years, and elevated 8-isoprostane were the independent determinants of visual disability in the population when the logistic regression model 2 was adjusted for education level, intake of red beans, intake of taro leaves, triglycerides, and T2DM. ###### Independent role of 8-isoprostane on the presence of visual disability in Central Africans. **Independent variables** **B coefficient** **Standard error** **Wald χ^2^** **OR (95% CI)** **p-value** ----------------------------- ------------------- -------------------- --------------- -------------------- ------------- **Safou fruit intake** No versus Yes 3.764 0.584 41.610 43.1 (13.7--135.4) \<0.0001 **Migration** Yes versus No 1.348 0.598 5.083 3.9 (1.2--12.4) 0.024 **Aging** ≥60 years versus \<60 years 1.225 0.496 6.100 3.4 (1.3--9) 0.014 **8-isoprostane** High versus Normal level 2.397 0.607 15.609 11 (3.4--36.1) \<0.0001 Constant \- 5.413 0.900 36.163 \<0.0001 No intake of safou fruits, migration, age ≥60 years and elevated 8-OH-dG were significantly, positively, and independently associated with the presence of visual disability in the study after adjusting for education level, intake of red beans, intake of taro leaves, triglycerides, and T2DM ([Table 6](#t6){ref-type="table"}). ###### Independent role of 8-OHd guanosine on the presence of visual disability in Central Africans. **Independent variables** **B coefficient** **Standard error** **Wald χ^2^** **OR (95% CI)** **p-value** ----------------------------- ------------------- -------------------- --------------- -------------------- ------------- **Safou fruit intake** No versus Yes 3.925 0.613 41.004 50.7 (15.2--168.5) \<0.0001 **Migration** Yes versus No 1.358 0.616 4.849 3.9 (1.2--13) 0.028 **Aging** ≥60 years versus \<60 years 1.247 0.499 6.255 3.5 (1.3--9.3) 0.012 **8-OHd guanosine** High versus Normal level 2.685 0.670 16.032 14.7 (3.9--54.5) \<0.0001 Constant \- 5.774 0.985 34.332 \<0.0001 There was a significant, dose--response and linear relationship between significant biomarkers of oxidative stress (8-isoprostane and 8-OH-dG in each y-axis) and severity of visual disability(x-axis): increase in their values from normal vision toward visual impairment and blindness ([Figure 4](#f4){ref-type="fig"}). {#f4} Discussion ========== The present survey was conducted in a population at higher risk of chronic diseases. To our knowledge, this is the first African study to assess the levels of Biomarkers using ROCs and their independent role in the development of visual disability. The current burden of visual disability in both non-diabetic and diabetic participants and its severity from visual impairment to blindness, was consistent with our previous paper \[[@r22]\], studies from Cameroon \[[@r23]\] and from Jordan \[[@r24]\], but higher than that reported from DRC in 1994 \[[@r25]\] and 2009 \[[@r12]\]. This burden may be explained by nutrition transition \[[@r6]\], explosing T2 DM \[[@r7]\], political crisis-induced poverty, aging, lifestyle changes and oxidative stress. In univariate analysis, higher levels of WC, SBP, ox-LDL, blood glucose, 8-isoprostane, and 8-OH-dG, as well as no intake of safou fruit, age ≥60 years, urban residence, low education level, and migration were identified as the factors significantly and positively associated with the presence of visual disability among these Congolese. Gender and the rest of variables were not associated with visual disability. However, after avoiding collinearity and confounding factors such as T2DM, we identified age ≥60 years, migration, low education level, no intake of safou fruit, and higher oxidant status as independent determinants of the presence of visual disability in these Central Africans. This suggests a synergistic, isolated or interactive action of non modified factors (aging), endogenous attributes of people, and environmental factors in the development of visual disability as observed in multifactorial and chronic diseases. In healthy individuals, the eye needs specific nutrients such as safou fruit rich in antioxidants and vitamins \[[@r26],[@r27]\]. In Kinshasa town, fruits are very expensive in general and during seasons of fruits in particular. No intake of safou, a season fruit, conferred the highest risk of the development of visual disability because of possible inefficient antioxidant status. Aging, however, reduces the levels of these nutrients, and the eyes' capability to remove free radicals. The essential antioxidants within safou fruit may keep ocular tissues healthy. After rural-urban migration with settlement in urban residence, the eye is exposed to Kinshasa environment with pollution, lack of traditional diet rich in vegetables, fibers and fruits. Kinshasa town is characterized by few green spaces, but longer exposure to sunlight. Low education level may reflect lack of education on nutrition, poverty limiting access to health care for T2DM, and cultural beliefs. In DRC, green and crude vegetables in general and cabbages in particular are considered foods for goats, while fruits are left for children. The demographic, environmental, and dietary factors identified in this study may facilitate the initiation and the acceleration of imbalance of oxidant/antioxidant status and visual disability. Indeed, there was an increase in levels of oxidant status (elevated 8-Isoprostane and elevated 8-OH-dG) with the presence and the severity (progression) of visual disability in these Congolese. Increase in OxLDL and 8-isoprostane is determined by lipid peroxidation which occurs in response to elevated levels of free radicals. Chronic hyperglycemia in diabetics and acute hyperglycemia induced by post-prandial time may be responsible for the elevated OxLDL and 8-Isoprostane in these Congolese with visual disability. Oxidation of LDL is a complex process taking place in both the extra- and intra-cellular space. Numerous studies have clearly acknowledged that the levels of circulating plasma of Ox-LDL are associated with obesity-related metabolic disturbance such as metabolic syndrome and diabetes mellitus \[[@r28]\]. In this study, both high levels of WC, SBP, and blood glucose, components of metabolic syndrome, as well as T2DM were associated with the presence of visual disability in univariate analysis, while elevated oxLDL was associated with the presence of visual disability in multivariate analysis. The present study showed that visual disability was associated with increased DNA damage defined by high levels of 8-OH-dG. Aldebasi et al. \[[@r29]\] indicate significantly higher concentrations of serum 8-OH-dG in Saudi type 2 diabetic patients with retinopathy compared to type 2 diabetic patients without retinopathy. Diabetic retinopathy, one of the leading causes of visual disability worldwide among adults and associated with oxidative stress \[[@r29]\], may be involved in the development of observed visual disability. The present study showed that elevated levels of Ox-LDL, 8-isoprostane, and 8-OH-dG, identified as independent determinants of the presence of visual disability, were also significantly associated with the progression and severity of vision loss. It illustrated that the levels of Ox-LDL, 8-Isoprostane, and 8-OHdG in participants with blindness were higher than those in participants with visual impairment and participants with normal vision, respectively. The levels of these biomarkers of oxidative stress in participants with visual impairment were also higher than those in people with normal vision. Clinical implications and perspectives for public health -------------------------------------------------------- The sensitive and specific levels of 8-isoprostane and 8-OH-dG will be used in different diseases related to oxidative stress. The present findings will impact on nutrition education, integrating ophthalmic practice into Primary Health care systems, and the WHO "VISION 2020 initiative" in Central Africa. Training ophthalmologists with skills in clinical nutrition, infrastructures for early diagnosis of visual disability and recommended diet rich in antioxidants are needed to eliminate preventable blinding diseases as public health problems. This calls for the implementation of more preventive strategies among this population with low education level, low intake of fruits and vegetables, facing urban inequalities after rural-urban migration, nutrition transition and aging. The current management strategy for vision loss requires early detection of nutrients deficiency and oxidative stress and optimal control of WC, blood pressure, and blood glucose to slow the progression of visual dysfunction. Most of these Central Africans require early and good binocular visual acuity screening. These results suggest the need of drugs for treating several oxidative stress-related ocular diseases of aging eye, protecting against vitamins deficiency at early stages of ocular diseases, and delaying the progression of vision loss \[[@r2]-[@r4]\]. Study limitations ----------------- This study may be limited to some degree because of its cross-sectional design which is not able to demonstrate a causal association between the identified independent determinants and visual disability presence and progression. Conclusion ---------- These findings emphasize the important and independent role of aging, no intake of safou fruit, rural-urban migration, low education level, and oxidative stress in the presence and the progression of visual disability. Antioxidant supplements, nutrition education, control of migration, and blocking of oxidative stress are crucial steps for delayed onset, progression and severity of visual disability in general and in aging people in particular. We thank *M*~r~ Simon Stimela Mathabatha in Memoriam, Analytical chemistry Laboratory from Walter Sisulu University, Mthatha, South Africa and the staff of Lomo Laboratories, Kinshasa, Limete, DRC for their assistance with blood samples analyses. We pay tribute to the Division of Ophthalmology, Saint Joseph hospital, Kinshasa, Limete, DRC and all participants for their total support in performing the study. The study was financed by intramural funds of LOMO MEDICAL Center (Prof LONGO-MBENZA), Kinshasa Limete, DRC.

Introduction {#eji3970-sec-0010} ============ Testicular cancer is the most common tumor among young men of 20--34 years of age and is increasing in incidence [1](#eji3970-bib-0001){ref-type="ref"}. Testicular germ cell tumors (TGCTs) account for the majority of testicular cancers and consist predominantly of classical seminomas and nonseminomatous germ cell testicular tumors (NSGCTTs). Seminomas are morphologically homogeneous whilst NSGCTTs are more heterogeneous and can be composed of multiple components including embryonic carcinoma, choriocarcinoma cells and teratoma [2](#eji3970-bib-0002){ref-type="ref"}. In addition, some testicular germ cell tumors can contain both seminoma and NSGCTT elements and these are classified as mixed germ cell tumors (mGCTs). All post‐pubertal TGCTs are thought to originate from a pre‐invasive lesion termed germ cell neoplasia in situ (GCNIS) [3](#eji3970-bib-0003){ref-type="ref"}. Genome‐wide expression profiling studies suggest that GCNIS cells are derived from pluripotent gonocytes that have survived within the post‐natal testis [4](#eji3970-bib-0004){ref-type="ref"}, [5](#eji3970-bib-0005){ref-type="ref"} and subsequent post‐pubertal signals from surrounding somatic cells may initiate the development of an invasive tumor. Cancer/testis antigens (CTAg) are proteins that are expressed in germ cells but are usually silenced in somatic cells [6](#eji3970-bib-0006){ref-type="ref"}. They include the melanoma‐associated antigen (MAGE) family and NY‐ESO‐1 antigen which are attractive candidates for cancer vaccine trials and adoptive cellular immunotherapy [7](#eji3970-bib-0007){ref-type="ref"}, [8](#eji3970-bib-0008){ref-type="ref"}, [9](#eji3970-bib-0009){ref-type="ref"}. Importantly, CTAg expression is observed in many tumor subtypes and incomplete clonal deletion of CTAg reactive thymocytes in the thymus can lead to the development of CTAg‐specific T‐cell responses [10](#eji3970-bib-0010){ref-type="ref"}, [11](#eji3970-bib-0011){ref-type="ref"}, [12](#eji3970-bib-0012){ref-type="ref"}. Study of CTAg expression in testicular cancer has revealed that MAGE‐A family proteins are expressed frequently in classical seminoma and seminomatous elements of mGCTs but are generally absent in GCNIS [13](#eji3970-bib-0013){ref-type="ref"}, [14](#eji3970-bib-0014){ref-type="ref"}, [15](#eji3970-bib-0015){ref-type="ref"}, [16](#eji3970-bib-0016){ref-type="ref"}, [17](#eji3970-bib-0017){ref-type="ref"}, [18](#eji3970-bib-0018){ref-type="ref"}, [19](#eji3970-bib-0019){ref-type="ref"}, [20](#eji3970-bib-0020){ref-type="ref"}. Conversely NY‐ESO‐1 expression is limited to pre‐invasive GCNIS [21](#eji3970-bib-0021){ref-type="ref"} and is usually negative in both seminoma and NSGCTT [16](#eji3970-bib-0016){ref-type="ref"}, [17](#eji3970-bib-0017){ref-type="ref"}, [18](#eji3970-bib-0018){ref-type="ref"} indicating a pattern of downregulation during transition to carcinoma. As testicular cancer is itself a germ cell tumor we investigated if cellular immunity against CTAg proteins develops in patients with testicular tumors. We find that strong CD8^+^ and CD4^+^ CTAg‐specific T‐cell responses are indeed found in many patients and that the global memory T‐cell pool is also substantially increased at diagnosis. The magnitude of CTAg‐specific responses decreases substantially after treatment and coincides with a reduction in the T‐cell memory pool. These data indicate that natural CTAg‐specific immunity is established in many patients with testicular cancer and comprises a population of short‐lived effector T cells that persist poorly in the absence of antigen. Results {#eji3970-sec-0020} ======= Spontaneous T‐cell responses against MAGE‐A proteins develop in many patients with testicular cancer {#eji3970-sec-0030} ---------------------------------------------------------------------------------------------------- Forty‐five patients with a diagnosis of testicular cancer were recruited prior to orchidectomy or within 2 weeks of the operation. PBMCs were stimulated with overlapping peptide pools derived from CTAg proteins prior to analysis using IFN‐γ ELISPOT assay. T‐cell responses were determined against peptide pools representing MAGE‐A1, MAGE‐A3, MAGE‐A4 and NY‐ESO‐1 and were examined in all patients as well as 17 age‐matched male donors. T‐cell responses against MAGE proteins were observed in many patients with testicular cancer. A representative ELISPOT is shown in Fig. [1](#eji3970-fig-0001){ref-type="fig"}A. Specifically, T‐cell responses to MAGE‐A1 were detected in 17% (1/6) and 31% (9/29) of patients with mixed germ cell tumors (mGCT) and seminoma respectively but were absent in patients with non‐seminomatous germ cell testicular tumors (NSGCTT) (Fig. [1](#eji3970-fig-0001){ref-type="fig"}B). MAGE‐A3‐specific responses were detected in patients across all testicular tumor types (Fig. [1](#eji3970-fig-0001){ref-type="fig"}C) with 31% (9/29) of seminoma, 50% (3/6) of mGCT and 30% (3/10) of NSGCTT patients demonstrating an immune response to this protein. MAGE‐A4 responses were detected in 28% (8/29) of seminoma and 33% (2/6) of mGCT patients (Fig. [1](#eji3970-fig-0001){ref-type="fig"}D), but were not identified in patients with NSGCTT. Interestingly, NY‐ESO‐1‐specific T‐cell responses were undetectable in all TGCT patients examined, irrespective of the tumor subtype (Fig. [1](#eji3970-fig-0001){ref-type="fig"}E). The majority of patients and healthy donors responded to the CEFT positive control (Fig. [1](#eji3970-fig-0001){ref-type="fig"}F). Overall, CTAg‐specific T‐cell responses were detectable in 44% (20/45) of patients prior to adjuvant therapy. {#eji3970-fig-0001} Further analysis revealed that a substantial proportion of patients with seminoma and mGCT exhibited a T‐cell immune response against multiple MAGE‐A family proteins (Fig. [1](#eji3970-fig-0001){ref-type="fig"}G). In particular, of those exhibiting at least one response, 50% (7/14) of patients with seminoma and 66% (2/3) of patients with mGCT had responses to at least 2 of the 3 MAGE‐A family antigens that were examined. In contrast, NSGCTT patients generated responses against only the MAGE‐A3 protein. Interestingly, analysis of the immune response in relation to clinical parameters revealed that all NSGCTT patients with a MAGE‐A3 directed immune response had evidence of metastatic disease outside the testis. There was no association between the frequency or magnitude of CTAg‐specific responses and tumor stage in patients with seminoma or mGCT (data not shown). MAGE‐specific T cells comprise both CD4^+^ and CD8^+^ T cells and demonstrate a Th1 cytotoxic phenotype {#eji3970-sec-0040} ------------------------------------------------------------------------------------------------------- In order to further characterize the CTAg‐specific immune response we next went on to define the T‐cell subtype and effector function in a group of patients with an established MAGE‐specific response (*n* = 10). PBMCs were stimulated with overlapping peptides in short‐term cultures and then intracellular cytokine staining was used to examine IFN‐γ and TNF‐α production (Fig. [2](#eji3970-fig-0002){ref-type="fig"}A). As expected given their previous identification through IFN‐γ ELISPOT, all patients demonstrated intracellular IFN‐γ production following peptide stimulation (data not shown). However, it was noteworthy that all patients also exhibited a high frequency of MAGE‐specific cells which produced TNF‐α but not IFN‐γ and this IFN‐γ^−^TNF‐α^+^ phenotype was more pronounced within the CD8^+^ T‐cell compartment (Fig. [2](#eji3970-fig-0002){ref-type="fig"}B). The frequency of MAGE‐specific CD8^+^ T cells was 6‐fold greater than MAGE‐specific CD4^+^ T cells, yet a strong positive relationship within individual donors was observed (r = 0.7697, *p* = 0.0156) (Fig. [2](#eji3970-fig-0002){ref-type="fig"}C) suggesting the development of a coordinated CD4^+^ and CD8^+^ T‐cell response against such antigens. {#eji3970-fig-0002} Finally, we determined the cytotoxic capacity of CD8^+^ T cells using CD107a surface mobilization as a marker of degranulation, and analyzed this in relation to the cytokine secretion profile. Importantly, although the majority of IFN‐γ^+^TNF‐α^+^ and IFN‐γ^−^TNF‐α^+^ CD8^+^ T cells demonstrated surface mobilization of CD107a in response to peptide stimulation (mean: 97.5% vs. 85.3%) (Fig. [2](#eji3970-fig-0002){ref-type="fig"}D), nearly 15% of the latter subset failed to undergo degranulation (*p* = 0.0054). MAGE‐specific CD8^+^ T cells can be identified through staining with immunodominant pMHC‐I multimers {#eji3970-sec-0050} ---------------------------------------------------------------------------------------------------- Following identification of MAGE‐specific CD8^+^ T cells through cytokine production we next investigated if CTAg‐specific T cells could be identified through the use of staining with HLA‐peptide dextramers containing immunodominant peptides from MAGE‐A1, MAGE‐A3, MAGE‐A4 and/or NY‐ESO‐1/LAGE‐1 (Table [1](#eji3970-tbl-0001){ref-type="table-wrap"}). Freshly isolated PBMCs were expanded with peptide for 10 days prior to dextramer staining (Fig. [3](#eji3970-fig-0003){ref-type="fig"}A) and a positive response was reported when the frequency of MAGE‐specific T cells was greater than 0.05% of the total CD8^+^ T‐cell pool (1 in 2000). MAGE‐specific responses were detected in 36% (8/22) of patients (Fig. [3](#eji3970-fig-0003){ref-type="fig"}B). MAGE‐A1 (RVRF)‐specific T‐cell responses were detected in 9% (2/22) of patients and almost a third (32%) of patients had detectable MAGE‐A3‐specific T cells (EVDP and/or KVAE). Importantly, whereas both patients who demonstrated MAGE‐A1 (RVRF)‐specific responses had a diagnosis of seminoma, MAGE‐A3 responses (EVDP and KVAE) were detected in patients across all tumor subtypes (data not shown). MAGE‐A4 (GVYD) and NY‐ESO‐1/LAGE1 (MLMA)‐specific T‐cell responses were not detected in this patient cohort. These results confirm that CD8^+^ T‐cell responses against a range of MAGE proteins are present in patients with testicular cancer and that these can be identified within the peripheral circulation through the use of pMHC‐I multimers. ###### Immunodominant peptides used in dextramer assays Protein HLA restriction HLA frequency (%) Peptide sequence Position ----------------- ----------------- ------------------- ------------------ -------------- MAGE‐A1 A2 42.6 KVLEYVIKV 278‐286 B7 21.4 RVRFFFPSL 289‐298 MAGE‐A3 A1 29.1 EVDPIGHLY 168‐176 A2 42.6 KVAELVHFL 112‐120 MAGE‐A4 A2 42.6 GVYDGREHTV 230‐239 NY‐ESO‐1/LAGE‐1 A2 42.6 MLMAQEALAFL 1 ‐11 (ORF2) John Wiley & Sons, Ltd. {#eji3970-fig-0003} MAGE‐specific CD8^+^ T‐cell clones display cytotoxic activity against tumor cells {#eji3970-sec-0060} --------------------------------------------------------------------------------- In order to investigate the avidity and function of CTAg‐specific T cells in patients with testicular cancer we next went on to generate a primary CTAg‐specific T‐cell clone from a patient with seminoma. A CD8^+^ T‐cell clone specific for the EVDP peptide derived from MAGE‐A3 and restricted by HLA‐A1 was isolated from peripheral blood taken at disease presentation (Fig. [3](#eji3970-fig-0003){ref-type="fig"}C, inset). Interestingly, the clone exhibited a relatively low avidity for peptide with 50% maximal IFN‐γ release (EC~50~) at a peptide concentration of 4.58 × 10^−7^ M (Fig. [3](#eji3970-fig-0003){ref-type="fig"}C). Despite this the CD8^+^ clone demonstrated strong and specific recognition of both peptide‐pulsed LCLs and the HT‐29 cell line, indicating the ability to recognize tumor cells which express endogenous CTAg protein (Fig. [3](#eji3970-fig-0003){ref-type="fig"}D). Finally, the cytotoxic potential of the clone was examined through the use of the CD107a mobilization assay. T cells were co‐cultured with either EVDP peptide‐loaded LCL or HT‐29 cells and expression of CD107a was then examined on effector cells. CD107a expression was observed in both cases, indicating both that the T‐cell clone has cytotoxic capacity and that this can be induced following endogenous antigen presentation of the MAGE protein (Fig. [3](#eji3970-fig-0003){ref-type="fig"}E and F). The frequency and magnitude of the CTAg‐specific T‐cell response decrease markedly after treatment {#eji3970-sec-0070} -------------------------------------------------------------------------------------------------- We next went on to investigate the profile of CTAg‐specific immune responses during and after chemotherapy. Blood samples were obtained at several time points from a cohort of seminoma patients in whom we had detected a positive CTAg‐specific T‐cell response prior to treatment. The frequency of MAGE‐specific T cells was assessed at least 3 months after completion of initial treatment and compared to that prior to adjuvant therapy (Fig. [4](#eji3970-fig-0004){ref-type="fig"}A). Interestingly, the MAGE‐specific immune response remained detectable in only 22% of cases (4/18) at the later timepoint. {#eji3970-fig-0004} We next investigated the temporal kinetics of CTAg‐specific immunity to individual MAGE proteins within the observed period of decline during follow up. Interestingly, a progressive decrease in the frequency of antigen‐specific T cells against each MAGE protein was observed (Fig. [4](#eji3970-fig-0004){ref-type="fig"}B). Specifically, the magnitude of the MAGE‐specific T‐cell response fell by 89% overall during follow‐up compared to pre‐treatment values (median: 29.6 vs. 273, *p* = 0.0002). Importantly, we also studied if MAGE‐specific T‐cell responses were induced following adjuvant treatment in any patients in whom they had not been present at diagnosis but no such responses were observed (data not shown) suggesting that chemotherapy does not induce CTAg‐directed anti‐tumor immunity. Testicular cancer patients have a marked expansion in the T‐cell memory pool at disease presentation {#eji3970-sec-0080} ---------------------------------------------------------------------------------------------------- In order to explore potential mechanisms that might underlie the marked attrition of MAGE‐specific T cells after treatment we went on to investigate the composition of the peripheral T‐cell memory pool in patients both at diagnosis and following treatment. These values were then compared to the proportions of memory cells in age and gender matched healthy controls. The pattern of CD45RA and CCR7 expression was used to define the proportion of naive (CD45RA^+^CCR7^+^, T~Naive~), central memory (CD45RA^−^CCR7^+^, T~CM~), CD45RA^−^ effector/memory (CD45RA^−^CCR7^−^, T~EM~) and CD45RA^+^ revertant effector (CD45RA^+^CCR7^−^, T~EMRA~) cells within the CD4^+^ and CD8^+^ populations of healthy donors and TGCT patients (Fig. [5](#eji3970-fig-0005){ref-type="fig"}A). {#eji3970-fig-0005} Interestingly, patients with testicular cancer were shown to have a markedly increased proportion of T cells within the memory pool and a significant reduction in the proportion of naïve cells. Specifically, within healthy donors the proportion of CD45RA^+^CCR7^+^ T~Naive~ cells represented 56% and 51% within the CD4^+^ and CD8^+^ repertoires respectively. These values were reduced by 25--40% to levels of 38 and 30% within the patient group (*p* = 0.003 and *p* = 0.0008, respectively) (Fig. [5](#eji3970-fig-0005){ref-type="fig"}B). In contrast, the proportion of CD4^+^ T~EM~ cells was increased from 23% within healthy donors to 32% within TGCT patients (*p* = 0.0028). CD8^+^ T~EM~ cells were also increased by around 35% from values of 32% within healthy donors to 45% within the patient group (*p* = 0.0284). Furthermore, the CD8^+^ T~EMRA~ population was also increased in the patient group compared to the healthy cohort (*p* = 0.0354). These data suggest that tumor development is associated with the generation of large numbers of antigen‐experienced T cells that inflate the peripheral blood memory T‐cell pool. Treatment of testicular cancer leads to a reduction in the peripheral blood memory T‐cell pool {#eji3970-sec-0090} ---------------------------------------------------------------------------------------------- Given our observation that patients with TGCT exhibited an average 30% increase in the size of the CD4^+^ and CD8^+^ T‐cell memory pool at the time of presentation, we next went on to assess the effect of treatment on the proportion of naïve and memory T cells. Interestingly, prospective analysis revealed that the proportion of T cells within the effector/memory T‐cell pool gradually declined after treatment whereas the naïve T‐cell pool increased (Fig. [5](#eji3970-fig-0005){ref-type="fig"}C). In particular, the naïve CD4^+^ and CD8^+^ subsets increased significantly from 38 and 29% to values of 46 and 34% respectively at ≥6 months after treatment. Reciprocally, the CD4^+^ and CD8^+^ effector memory (T~EM~) pools fell from 33 and 54% to values of 28 and 46% respectively. Furthermore, we investigated the absolute numbers of memory T‐cell subsets before and after treatment and found a substantial reduction in both the CD4^+^ (*p* = 0.0488) and CD8^+^ (*p* = 0.0059) T~EM~ populations (Fig. [5](#eji3970-fig-0005){ref-type="fig"}D). Importantly, there was no significant change in the number of T~Naive~ cells or the other memory subsets suggesting an overall loss of T~EM~ cells from the T‐cell pool during follow‐up. These data indicate that the inflation within the memory T‐cell pool that was observed at the time of diagnosis is comprised largely of short‐lived effector cells that are eroded relatively rapidly following treatment. Discussion {#eji3970-sec-0100} ========== Burnet\'s original model of cancer immune surveillance [22](#eji3970-bib-0022){ref-type="ref"} has now been developed and extended to include the phases of tumor elimination, equilibrium and escape [23](#eji3970-bib-0023){ref-type="ref"}. Immune‐mediated elimination of transformed cells requires the recognition of cancer‐specific epitopes and cancer testis antigens (CTAgs) are one such class of potential target. Cellular and humoral CTAg‐specific immune responses have been observed in many tumor subtypes and are suggested to play an important role in the control of disease progression [24](#eji3970-bib-0024){ref-type="ref"}. In this study, we frequently observed spontaneous CTAg‐specific T‐cell responses ex vivo (by IFN‐γ ELISPOT) in patients with testicular cancer but not in healthy male individuals. In addition, CD8^+^ T cells specific to known immunogenic epitopes of MAGE‐A family proteins were detectable in testicular cancer patients (by pMHC‐I multimer staining) following short‐term in vitro peptide stimulation cultures. It is interesting to speculate on the mechanisms that underlie the priming of immune responses against CTAg proteins in patients with testicular cancer. CTAg expression has been reported in testicular germ cell tumors and appears to be largely restricted to tumors containing seminomatous elements. Indeed, classical seminomas express several MAGE‐A family proteins but expression is absent or rare in NSGCTTs [14](#eji3970-bib-0014){ref-type="ref"}, [15](#eji3970-bib-0015){ref-type="ref"}, [17](#eji3970-bib-0017){ref-type="ref"}, [19](#eji3970-bib-0019){ref-type="ref"}, [20](#eji3970-bib-0020){ref-type="ref"}, [25](#eji3970-bib-0025){ref-type="ref"}. Interestingly, this pattern reflects the profile of CTAg‐specific immunity that was observed in the patient cohort, with T‐cell immunity being detected frequently in patients with seminoma and mixed germ cell tumors. One intriguing observation was that patients with seminoma and mGCT often developed T‐cell responses against several MAGE‐A family proteins whereas patients with NSGCTT only exhibited immune responses with specificity for MAGE‐A3. Furthermore, these responses to MAGE‐A3 were seen only in patients with NSGCTT who had metastatic disease outside the testis. MAGE‐A3 has been described as a mediator of extracellular matrix protein function which promotes tumor cell migration [26](#eji3970-bib-0026){ref-type="ref"} and it is therefore possible that expression of MAGE‐A3 is activated during NSGCTT metastasis and that a specific immune response is subsequently induced at the secondary tumor site. In addition, it is unlikely that immunogenic CTAg protein was derived from normal testicular tissue since NY‐ESO‐1‐specific immune responses were not observed in the present study even though NY‐ESO‐1 is expressed strongly by normal testicular spermatogenic cells [21](#eji3970-bib-0021){ref-type="ref"}. The lack of NY‐ESO‐1‐specific T‐cell responses in our patients could be explained by potential down regulation of NY‐ESO‐1 protein during the malignant transformation from pre‐malignant GCNIS to clinically overt disease [21](#eji3970-bib-0021){ref-type="ref"}. Cross‐presentation of tumor antigens by dendritic cells at the tumor site or within tumor‐draining lymph nodes is generally considered the primary mode for priming naive tumor‐specific CD8^+^ T cells [27](#eji3970-bib-0027){ref-type="ref"}, [28](#eji3970-bib-0028){ref-type="ref"}. Effective antigen presentation combined with the high level of CTAg protein that is available in patients with testicular cancer is likely to lead to the profound level of T‐cell expansion that is observed in this setting. In this regard, it is noteworthy that a large immune infiltrate is seen in virtually every case of seminoma [20](#eji3970-bib-0020){ref-type="ref"}, which is likely to allow substantial exposure of tumor antigens to the immune system. We observed that a large proportion of MAGE‐specific CD8^+^ T cells secreted TNF‐α alone following stimulation with CTAg peptides and that this subset was marginally less cytotoxic than cells which also produced IFN‐γ. Previous studies in hepatocellular carcinoma [29](#eji3970-bib-0029){ref-type="ref"}, [30](#eji3970-bib-0030){ref-type="ref"} and breast cancer [10](#eji3970-bib-0010){ref-type="ref"} patients have shown a similar phenomenon whereby tumor antigen‐specific CD8^+^ T cells demonstrated a lack of IFN‐γ production. As such, tumor‐specific CD8^+^ T‐cell responses may exhibit unique patterns of functional competence and at a practical level this indicates that the choice of IFN‐γ ELIPSOT as a functional test is likely to have underestimated the true frequency of CTAg‐specific T cells. A further interesting aspect of MAGE‐specific T‐cell immunity was the detection of both cytotoxic CD8^+^ and CD4^+^ T‐cell immunity in many patients. Antigen‐specific CD4^+^ T cells are required for the induction of cross‐primed CD8^+^ T cells responses [31](#eji3970-bib-0031){ref-type="ref"} and also support the functional activity of secondary immune responses [32](#eji3970-bib-0032){ref-type="ref"}, [33](#eji3970-bib-0033){ref-type="ref"}, [34](#eji3970-bib-0034){ref-type="ref"}. These data suggest that immunotherapeutic approaches which seek to induce or expand CTAg‐specific CD8^+^ T‐cell immunity should include appropriate epitopes for CD4^+^ T cells. Longitudinal analysis in TGCT patients demonstrated a dramatic reduction in the number of MAGE‐specific T cells within the blood following treatment. The initial blood samples had been taken shortly after tumor removal but before adjuvant chemotherapy. It is possible that adjuvant chemotherapy may trigger apoptosis of proliferating tumor‐reactive immune cells, including the loss of MAGE‐specific T cells. Alternatively, the decline of CTAg‐specific immunity is likely to reflect the absence of available CTAg antigen following orchidectomy and as such is similar to the kinetics of decline of adaptive immunity following the clearance of acute viral and bacterial antigens [35](#eji3970-bib-0035){ref-type="ref"}, [36](#eji3970-bib-0036){ref-type="ref"}. In support of this, a recent publication showed that placental‐specific T‐cell immune responses can be established during pregnancy but that these are short‐lived and became undetectable in the periphery after delivery [37](#eji3970-bib-0037){ref-type="ref"}. It also remains possible that CTAg‐specific immunity does in fact persist but becomes unresponsive to antigenic stimulation due to increased expression of inhibitory checkpoint receptors such as PD‐1 on the cell surface, or an increase in regulatory cells such as T‐regulatory cells (Tregs) or myeloid‐derived suppressor cells (MDSCs) in the blood following treatment, as has been described previously for HPV‐specific responses in oropharyngeal patients [38](#eji3970-bib-0038){ref-type="ref"}. Intriguingly, the fall in CTAg‐specific immunity correlated with the concurrent decline in memory T‐cell subsets and so it is tempting to speculate that MAGE‐A‐specific T cells represent a component of a potentially sizeable tumor‐specific effector T‐cell repertoire in patients at the time of diagnosis. Patients with testicular cancer have an excellent clinical outcome with a cure rate of over 95%. In addition, patients with seminoma respond very well to therapy even in the setting of metastatic disease. It is tempting to consider the potential role of tumor‐specific immune responses within this response as the efficacy of chemotherapy may be partially mediated through immunological mechanisms [39](#eji3970-bib-0039){ref-type="ref"}. The expression of HLA proteins on testicular tumors is low which raises important questions regarding the potential mechanisms by which a tumor‐specific T‐cell response may have some benefit in this disease [40](#eji3970-bib-0040){ref-type="ref"}, [41](#eji3970-bib-0041){ref-type="ref"}. The lack of surface HLA expression on tumor cells would mitigate against direct tumor cell lysis but recognition of local tissue could lead to inflammatory responses that may be capable of mediating non‐specific activity against tumor tissue. Unfortunately, due to the limited availability of fresh autologous tumor tissue, we were unable to correlate expression of MAGE antigens with the presence of peripheral MAGE‐specific T‐cell populations, or determine if MAGE‐specific T cells were present at the tumor site. However, a previous study demonstrated the presence of MAGE‐A3‐specific CD8^+^ T cells amongst TIL in a patient with seminoma [25](#eji3970-bib-0025){ref-type="ref"}, indicating that CTAg‐specific T cells can infiltrate tumor tissue. In summary, our data demonstrates that MAGE‐specific T‐cell responses frequently develop in many patients with testicular cancer, yet these progressively decline as the disease comes under control. As such patients with testicular cancer represent an excellent opportunity to investigate novel aspects of tumor‐specific immune responses. Moreover, it is possible that strong tumor‐specific immunity plays an important role in mediating the excellent clinical outcome for patients with this condition. Materials and methods {#eji3970-sec-0110} ===================== Study participants {#eji3970-sec-0120} ------------------ Venous blood samples were obtained from patients prior to, and after completion of, post‐orchidectomy therapy for the treatment of testicular cancer at the New Queen Elizabeth Hospital, Birmingham, UK (*n* = 72). All patients had recently undergone radical orchidectomy to remove the involved testicle. Up to 30 mL heparinized blood was obtained from patients prior to chemotherapy and at regular intervals following treatment, which coincided with clinic visits. Up to 50 mL of blood was donated by healthy male volunteers (*n* = 22), which were used as controls. Patient and healthy donor characteristics are shown is Supporting Information Table 1. Written informed patient consent and local ethical committee approval (South Birmingham research ethics committee LREC reference 09/H1207/161, study reference RRK3953) were obtained prior to sample collection. Patients were 18 years or above and competent to give full informed consent. Isolation of peripheral blood mononuclear cells (PBMCs) {#eji3970-sec-0130} ------------------------------------------------------- PBMCs were isolated from heparinized blood by density gradient centrifugation over Lymphoprep (Axis‐Shield) within 4 h of collection. PBMCs were either assayed fresh (for dextramer analysis) or cryopreserved in liquid nitrogen (for memory phenotyping and ELISPOT analysis) in media containing 90% FCS and 10% DMSO. Memory phenotyping of T cells {#eji3970-sec-0140} ----------------------------- PBMCs (5 × 10^5^) were resuspended in 100 μL MACS buffer (PBS, 0.5% BSA, 2 mM EDTA), and surface stained with fluorochrome‐conjugated antibodies on ice for 30 min to identify memory T‐cell subsets (CD3‐APCCy7, CD4‐PerCPCy5.5, CD8‐AmCyan, CCR7‐PE and CD45RA‐efluor450). Propidium Iodide (PI) was added prior to cytometric analysis to exclude non‐viable cells. Acquisition was carried out with an LSRII flow cytometer (BD Biosciences) using FACSDiva software. IFN‐γ ELISPOT Assay {#eji3970-sec-0150} ------------------- T‐cell responses to CTAgs were measured for each individual patient in duplicate across all available timepoints using IFN‐γ ELISPOT assays conducted against pools of overlapping 15mer peptides (pepmixes; JPT Peptide Technologies) spanning the entire amino acid sequence of each antigen. Previously frozen PBMCs (5 × 10^6^) from TGCT patients and healthy male donors were seeded in a 24‐well cell culture plate and rested overnight at 37°C to allow cells to recover from cryopreservation. A Multiscreen 96‐well plate (Millipore) was coated with IFN‐γ capture antibody (Mabtech) at 4°C overnight. PBMCs were harvested the following day for ELISPOT assay setup and seeded at 2.5--3.5 × 10^5^ cells/well. For background spot determination, a negative control containing DMSO solvent only was used for each patient sample. Test condition wells were incubated with MAGE‐A1, MAGE‐A3, MAGE‐A4 or NY‐ESO‐1 pepmixes (1 μg of each peptide/mL). A CEFT peptide mix (immunogenic peptides derived from CMV, EBV, Flu and tetanus antigens) was also included as a positive control. The ELISPOT plates were incubated for 16--18 h at 37°C and developed as per manufacturer\'s instructions (Mabtech). IFN‐γ spots were counted using an AID automated ELISPOT reader. Each spot is representative of a single reactive IFN‐γ secreting T cell. Mean spot counts for negative control wells were subtracted from those for the test wells to determine the frequency of antigen specific T cells. Positive CTAg responses were defined as the recognition of pepmixes for which the mean adjusted counts were ≥2‐fold higher than that of the highest adjusted spot count observed in our healthy male control cohort. Intracellular cytokine staining and CD107a mobilization assay following antigen restimulation {#eji3970-sec-0160} --------------------------------------------------------------------------------------------- Flow cytometry was used to determine whether responses were attributable to antigen‐specific CD4^+^ and/or CD8^+^ T cells. Production of effector cytokines IFN‐γ and TNF‐α, and CD107a mobilization was simultaneously examined following restimulation with pepmixes. CD107a mobilization to the surface of responding T cells is a prerequisite for degranulation and was used here as a marker of cytotoxic potential of CD8^+^ T cells. Briefly, previously cryopreserved PBMCs of patients (*n* = 10) with known CTAg responses by ELISPOT were resuspended in RPMI 1640 media containing 10% FCS at a concentration of 2 × 10^6^ cells/mL. Cell were stimulated with pepmixes overnight, then cultured for 10 days in the presence of 50 U/mL IL‐2 from day 3. Cells were then restimulated with pepmixes in the presence of 20 ng/mL CD107a‐FITC antibody. After 1 h, protein transport inhibitors (5.3 mM brefeldin‐A, 1 mM monensin) were added to each well. Cells were incubated for a further 4 h, and then surface stained with CD3‐APC‐Cy7, CD4‐PerCP‐Cy5.5 and CD8‐AmCyan. Cells were washed in PBS, then fixed and permeabilised with 1% PFA and 0.5% saponin, respectively. Cells were stained intracellularly for the cytokines IFN‐γ‐AF700 and TNF‐α‐PECy7, and then analyzed by flow cytometry. Unstimulated cells (DMSO only) and CEFT peptide pool served as negative and positive controls, respectively. Expansion of antigen‐experienced CTAg‐specific CD8^+^ T cells in short‐term T‐cell line cultures {#eji3970-sec-0170} ------------------------------------------------------------------------------------------------ Fresh PBMCs were incubated with 10 μg/mL of each peptide (Table [1](#eji3970-tbl-0001){ref-type="table-wrap"}) in serum‐free RPMI 1640 media for 1 h with gently agitation. Cells were washed and resuspended in RPMI 1640 media supplemented with 10% human serum, IL‐7 (25 ng/mL), and IL‐15 (5 ng/mL) at a concentration of 2 × 10^6^ cells/mL. IL‐2 (50 U/mL) was added to the cultures from day 3. Cells were cultured for between 10 and 12 days to allow for sufficient T‐cell expansion. Dextramer (pMHC‐I multimer; Immudex) staining was used to identify antigen‐specific CD8^+^ T cells following short‐term culture. Identification of CTAg‐specific CD8^+^ T cells using pMHC‐I dextramers {#eji3970-sec-0180} ---------------------------------------------------------------------- T‐cell line cultures were stained with pMHC‐I dextramers for 20 min at RT in MACS buffer, and washed once prior to surface staining. Cells were surface stained with CD3‐APC‐Cy7, CD4‐FITC and CD8‐PC5 on ice for 30 min, then washed twice before cytometric analysis. PI was added just prior to cytometric analysis to exclude non‐viable cells. Cells were first gated for lymphocytes (SSC‐A vs. FSC‐A), followed by gating for viable CD8^+^ T cells (PI^−^CD3^+^CD4^−^CD8^+^). Dual CD8 and pMHC‐I dextramer expression was then determined from this gated population. A positive response was reported when the frequency of antigen‐specific dextramer‐stained CD8^+^ T cells was ≥0.05% of total CD8^+^ T cells, a threshold described in a number of previous studies [42](#eji3970-bib-0042){ref-type="ref"}, [43](#eji3970-bib-0043){ref-type="ref"}. Generation of MAGE‐A3‐specific T‐cell clones {#eji3970-sec-0190} -------------------------------------------- MAGE‐A3~EVDP~ dextramer positive cells were enriched using anti‐PE beads and a double MS column isolation procedure under sterile conditions, following the manufacturer\'s instructions (Miltenyi Biotec). The isolated cells were then cloned by limiting dilution over irradiated (40 Gy) PHA‐stimulated allogeneic PBMCs and peptide‐loaded partially‐matched lymphoblastic cell lines (LCL) in RPMI 1640 supplemented with 5% human serum, 5% FCS, IL‐2 (100 U/mL), IL‐15 (5 ng/mL), and IL‐21 (2 ng/mL). For the maintenance of the T‐cell clones, media containing IL‐2 (100 U/mL) and IL‐15 (5 ng/mL) was added twice weekly. Functional assessment of an EVDP‐specific T‐cell clone {#eji3970-sec-0200} ------------------------------------------------------ Potential MAGE‐A3~EVDP~‐specific T‐cell clones were initially screened by staining with dextramer and a single positive clone was studied further. Effector function was measured by the release of IFN‐γ into culture supernatant following 16 h incubation with targets cells. Target cells were LCL loaded with relevant peptide (EVDP), LCL loaded with irrelevant peptide (KVAE) and a cell line expressing HLA‐A1 and MAGE‐A3 (HT‐29) [44](#eji3970-bib-0044){ref-type="ref"}, [45](#eji3970-bib-0045){ref-type="ref"}. Assay controls included T‐cell clone only, LCL only and HT‐29 only. The culture supernatant medium was harvested and assayed for IFN‐γ by ELISA (Mabtech) in accordance with the manufacturer\'s instructions. For TCR avidity assays, peptide was titrated from 10^−5^ to 10^−11^ M and expressed as a percentage of maximal IFN‐γ release. CD107a mobilization assay was performed as described above. T cells were stimulated with LCL loaded with either 10 μg/mL EVDP or KVAE peptides, or with the HT‐29 cell line ± EVDP peptide. Data handling and statistical analysis {#eji3970-sec-0210} ====================================== Statistical analysis was performed using GraphPad Prism version 5 (GraphPad Software). To determine differences between two independent groups, a non‐parametric Mann--Whitney test was performed. A Wilcoxon matched‐pairs signed‐rank test was used to compare non‐parametric paired data. A linear regression was performed to assess the relationship between two variables. Spearman\'s nonparametric test was used to determine correlations. Normal distribution was assessed using the D\'Agostino & Pearson test, where appropriate. A *p* value \<0.05 was considered statistically significant. Conflict of interest {#eji3970-sec-0230} ==================== The authors declare no commercial or financial conflict of interest. TGCT : testicular germ cell tumor NSGCTT : non‐seminomatous germ cell testicular tumor mGCT : mixed germ cell tumor MAGE : melanoma‐associated antigen LCLs : lymphoblastic cell lines PBMCs : peripheral blood mononuclear cells pMHC‐I : peptide major histocompatibility antigen class‐I Supporting information ====================== ###### Supporting Information Table 1 Patient characteristics ###### Click here for additional data file. ###### Peer review correspondence ###### Click here for additional data file. This work was supported by the MRC Centre for Immune Regulation, University of Birmingham.

High-fidelity self-recognition and self-discrimination are of principal importance in living systems[@b1], which enable biological macromolecules, such as proteins and DNA, to self-organize into uniform quaternary and double helical structures with a controlled handedness, respectively, through noncovalent interactions even within complex mixtures of subunits or molecular strands with a similar shape, size and sequence, thereby providing sophisticated functions that are essential for human life[@b1][@b2]. Such an incredible self-sorting performance observed in biological systems ultimately relies on the preprogrammed monomer sequences and homochirality of their building blocks and components[@b2][@b3]. Recent advances in the total chemical synthesis of unnatural [D]{.smallcaps}-proteins[@b4][@b5] and [L]{.smallcaps}-DNA[@b6] clearly revealed the indispensable role of the homochirality along with the monomer sequences of natural [L]{.smallcaps}-proteins and [D]{.smallcaps}-DNA towards chiral specificities in enantioselective reactions and complementary double-helix formations, respectively. In organic and supramolecular chemistry, the control of the size and shape[@b7][@b8][@b9][@b10][@b11][@b12][@b13], and topology[@b14][@b15] as well as the handedness[@b16][@b17][@b18][@b19][@b20][@b21][@b22][@b23][@b24][@b25][@b26][@b27][@b28][@b29][@b30][@b31] of supramolecular assemblies from mixtures of a variety of components via spontaneous self-sorting has become one of the urgent and emerging topics[@b10][@b32][@b33] because it will not only enrich our understanding of the principles underling the precise recognition behaviour of biological macromolecules but also contribute to the development of novel supramolecular catalysts[@b34], optoelectrical devices[@b35] and sensors[@b36]. In most cases[@b32][@b33], however, self-sorting has been achieved in a size- and shape-[@b7][@b8][@b9][@b10][@b11][@b12][@b13] or topology-selective way[@b14][@b15][@b28] through the formation of macrocycles[@b11][@b15], capsules or cages[@b13][@b17][@b22] and helices[@b23][@b24][@b27][@b30][@b31][@b37], while a limited number of supramolecular systems undergoes chiral self-sorting, producing either homochiral[@b16][@b17][@b18][@b19][@b20][@b21][@b22][@b23][@b24][@b25][@b26][@b27][@b28][@b29][@b30][@b31][@b32] or heterochiral assemblies[@b32][@b38][@b39][@b40], which is mostly relevant to the biological processes as well as the state-of-the-art asymmetric catalysis[@b41] and optical resolution on crystallization[@b42]. Here we report artificial chirality- and sequence-selective chiral self-sorting of dimeric strands consisting of carboxylic acid or amidine groups joined by chiral amide linkers with different sequences (NHCO or CONH) through specific homoduplex and subsequent hetero- (complementary-) duplex formations with implications for biological double helices in DNA. DNA-like double helices have a significant advantage in the precise recognition of the monomer sequence[@b14][@b43] and chain length between the complementary strands[@b44]. Although hydrogen-bond-driven peptide nucleic acids[@b45] and metal-coordinate-bonded helicates[@b7][@b8][@b9][@b46] are known to show self-sorting with respect to their sequences and/or chain lengths, chirality- and sequence-selective successive self-sorting is currently unknown. Results ======= Design and synthesis of carboxylic acid and amidine dimers ---------------------------------------------------------- Our molecular design is mainly based on the previously reported heterodouble helices composed of complementary dimer strands with *m*-terphenyl backbones intertwined through amidinium--carboxylate salt bridges ([Fig. 1a](#f1){ref-type="fig"}, **AA·CC**)[@b44][@b47][@b48][@b49][@b50]. Owing to the high tolerance of the salt bridges towards various functional groups, a variety of linkers (L) could be introduced while maintaining the double-stranded helical structures with a one-handed helical sense biased by the chirality introduced on the amidine residues[@b49]. During the course of our study, we also found that an achiral carboxylic acid dimer joined by a *p*-diethynylbenzene linker with *n*-octyl (R^A^) substituents self-associated into a racemic homodouble helix through interstrand hydrogen bonds between the carboxy groups ([Fig. 1a](#f1){ref-type="fig"}, (**CC**~**2**~))[@b51]. We anticipated that either a right- or left-handed homodouble helix would be induced in the carboxylic acid dimers when chiral linkers[@b52], such as (*S*,*S*)- or (*R*,*R*)-*trans*-1,2-cyclohexane-based (*c*Hex) *bis*-amide derivatives, were introduced between the monomer units (**1** and **2**, [Fig. 1b](#f1){ref-type="fig"}). Incidentally, we have found a complete sequence-selective chiral self-sorting in a mixture of racemic (*rac*)-**1a** and *rac*-**2a** mediated by unique interstrand multihydrogen-bond-driven homoduplex formations, the structures of which are significantly different from our expected self-associated homodouble helices, and are further different from each other depending on the linker amide sequences (NHCO-*c*Hex (**L1**) or CONH-*c*Hex (**L2**), [Fig. 1c](#f1){ref-type="fig"}). The further addition of an enantiopure amide-linked amidine dimer (**3** or **4**) with a particular linker amide sequence to a mixture of *rac*-**1a** and *rac*-**2a** results in the formation of a single optically pure complementary duplex with a 100% diastereoselectivity and sequence specificity stabilized by the amidinium--carboxylate salt bridges ([Fig. 1c](#f1){ref-type="fig"}). In addition, we show that the diastereoselectivities during the complementary duplex formations, however, greatly rely on the sequence and chirality of the amide linkers of the amidine dimers relative to those of the carboxylic acid dimers and are almost independent of the amidine chirality. We also demonstrate a vital role of multihydrogen bonds formed between the amide linkers that determines the observed unique chirality- and sequence-selective homo- and complementary self-sorting. A series of chiral dimers of carboxylic acids (**1**, **2**) and amidines (**3**, **4**) linked through chiral or *meso* amide residues with different sequences was prepared according to the reported methods (see [Supplementary Methods](#S1){ref-type="supplementary-material"})[@b47][@b50][@b53]. Self-sorting behaviour of carboxylic acid dimers ------------------------------------------------ We first investigated the self-association behaviour of the carboxylic acid dimers (*R*,*R*)- or (*S*,*S*)-**1a** and -**2a**. The circular dichroism (CD) spectra of the enantiomeric (*R*,*R*)- and (*S*,*S*)-strands of **1a** and **2a** in CDCl~3~ showed intense split-type Cotton effects that are mirror images of each other ([Fig. 2a](#f2){ref-type="fig"}). However, the CD intensities significantly decreased in dimethyl sulfoxide (DMSO; [Supplementary Fig. 1](#S1){ref-type="supplementary-material"}) because of dissociation into single strands in DMSO because DMSO is a strong hydrogen-bond acceptor and strongly hampers the interstrand hydrogen bonds; therefore, the CD spectral patterns and intensities of (*R*,*R*)-**1a** as well as (*R*,*R*)-**2a** in DMSO were similar to those of the carboxy-protected (*R*,*R*)-**1b-OMe** and -**2b-OMe** in CDCl~3~, respectively. The formation of homochiral duplexes **1a**~2~ and **2a**~2~ in CDCl~3~ was also supported by the corresponding dimer peaks \[**1a**~2~+Na\]^+^ and \[**2a**~2~+Na\]^+^ observed in their matrix-assisted laser desorption/ionization time-of-flight mass spectra ([Supplementary Fig. 2](#S1){ref-type="supplementary-material"}). The dimerization constants (*K*~d~) of (*R*,*R*)-**1a** and (*S*,*S*)-**2a** in CHCl~3~ at 25 °C were then estimated to be *ca.* 3.2 × 10^7^ and *ca.* 5.0 × 10^5 ^M^−1^ by extrapolation of the *K*~d~ values measured in CHCl~3~/THF (tetrahydrofuran) and CHCl~3~/CH~3~CN mixtures with various ratios, respectively, because the *K*~d~ values in pure CHCl~3~ are too high to accurately estimate ([Supplementary Fig. 3](#S1){ref-type="supplementary-material"}). These significantly higher *K*~d~ values as compared with those of the model compounds (*R*,*R*)-**M1** and -**M2** lacking the *m*-terphenyl moieties (*K*~d~=*ca.* 20 M^−1^, [Supplementary Fig. 4](#S1){ref-type="supplementary-material"}) and **CC** (*K*~d~=*ca.* 1.4 × 10^3 ^M^−1^)[@b51] suggest the strong multiple interstrand hydrogen bonds, which were consistent with the significant downfield shifts of the amide NH proton resonances of (*S*,*S*)-**1a** and (*R*,*R*)-**2a** in their ^1^H NMR spectra in comparison with the corresponding carboxy-protected (*S*,*S*)-**1b-OMe** and (*R*,*R*)-**2b-OMe** ([Fig. 2b](#f2){ref-type="fig"}). The variable-temperature ^1^H NMR spectral changes of (*R*,*R*)-**1a** and -**2a** also support the strong interstrand hydrogen bonds; the amide NH signals of both (*R*,*R*)-**1a** and -**2a** showed larger negative temperature coefficients (Δ*δ*/Δ*T* (ppb per °C)) in DMSO-*d*~6~ (−5.8 and −4.9) than those in CDCl~3~ (−2.0 and −3.1; [Supplementary Figs 5--8](#S1){ref-type="supplementary-material"})[@b54]. The X-ray crystallographic analyses of analogous (*S*,*S*)-**1a′** and -**2a′** whose structures are assumed to be almost identical to those of (*S*,*S*)-**1a** and -**2a**, respectively, except for the pendant and/or terminal substituents (see [Fig. 1b](#f1){ref-type="fig"}), revealed a unique deeply intertwined duplex structure ([Fig. 2c,d](#f2){ref-type="fig"}). Unexpectedly, these structures are completely different from a self-associated homodouble helix-like **CC**~2~ (see [Fig. 1a](#f1){ref-type="fig"}). Each strand adopts a similar 'U-shape\' structure resulting from the (*S*,*S*)-*trans*-1,2-cyclohexyl linker residue, binding together through the eight interstrand hydrogen bonds, in which each amide group at the linker moiety is sandwiched between the two carboxy groups of the other strand. The remarkable structural difference between (*S*,*S*)-**1a′** and -**2a′** in the solid state is that the homoduplex ((*S*,*S*)-**2a′**)~2~ has a well-packed structure due to the almost parallel orientation of the strands ([Fig. 2d](#f2){ref-type="fig"}), while the intertwined strands of (*S*,*S*)-**1a′** are oriented perpendicular to each other, resulting in a less sterically hindered structure ([Fig. 2c](#f2){ref-type="fig"}). This difference in the steric hindrance between ((*S*,*S*)-**1a′**)~2~ and ((*S*,*S*)-**2a′**)~2~ is reasonably correlated with the difference in their self-association affinities; the *K*~d~ value of (*S*,*S*)-**1a** is ∼10^2^ times greater than that of (*S*,*S*)-**2a**. The ^1^H two-dimensional (2D) nuclear Overhauser effect spectroscopy (NOESY) analysis of (*R*,*R*)-**1a** and (*S*,*S*)-**2a** in CDCl~3~ ([Supplementary Figs 9--13](#S1){ref-type="supplementary-material"}) showed characteristic interstrand NOE cross-peaks including those between the linker *c*Hex protons and the terminal aromatic protons of the *m*-terphenyl moieties along with the interstrand aromatic protons, indicating that such unique intertwined structures in the solid state were retained in solution. The observed large downfield shifts of the amide NH and CO~2~H protons of (*S*,*S*)-**1a** and (*R*,*R*)-**2a** in their ^1^H NMR spectra ([Fig. 2b](#f2){ref-type="fig"}) also support the structures. We next investigated the chiral self-sorting behaviour of *rac*-**1a** and *rac*-**2a** using ^1^H NMR in CDCl~3~ at 25 °C. The ^1^H NMR spectrum of *rac*-**1a** showed two sets of signals with an integral ratio of 2:1 (0% enantiomeric excess (e.e.), [Fig. 3a](#f3){ref-type="fig"}), which did not coalesce even at 50 °C ([Supplementary Fig. 14](#S1){ref-type="supplementary-material"}), whereas a single set of signals was observed in DMSO ([Supplementary Fig. 15](#S1){ref-type="supplementary-material"}), in which the duplexes undergo the anticipated dissociation into the single strands. The NOESY spectrum of *rac*-**1a** in CDCl~3~ also exhibited interstrand NOEs similar to those for the enantiopure (*R*,*R*)-**1a** without chemical exchange cross-peaks between the two species except for the labile CO~2~H signals ([Supplementary Figs 16 and 17](#S1){ref-type="supplementary-material"}), suggesting that the exchange rate between the species is slower than the present NMR timescale. The major set of signals of *rac*-**1a** is identical to the signals of the homochiral duplexes ((*R*,*R*)-**1a**)~2~ and ((*S*,*S*)-**1a**)~2~, and the minor set of the signals decreased with an increase in the % e.e. of **1a** ([Fig. 3a](#f3){ref-type="fig"}). Therefore, the minor signals are unambiguously assigned to the heterochiral duplex (*R*,*R*)-**1a**·(*S*,*S*)-**1a**. The diastereomeric excess (d.e.) values (%) of **1a**~2~ (homochiral duplex versus heterochiral duplex) were then plotted versus the % e.e. of **1a**, which matched well with the simulated curve using the equation (*Y*=100{−2\[4−3(*X*/100)^2^\]^0.5^+5}/3) (inset in [Fig. 3b](#f3){ref-type="fig"}; equation A) obtained by a modification of the reported equation ([Fig. 3b](#f3){ref-type="fig"})[@b55][@b56] (for details, see [Supplementary Methods](#S1){ref-type="supplementary-material"}), suggesting no preference in the diastereomeric duplex formation, thus forming the homochiral and heterochiral **1a**~2~ duplexes in a 2:1 molar ratio in CDCl~3~ at 25 °C. As anticipated, however, the diastereomeric duplex formation is sensitive to temperature and its molar ratio changed on heating or cooling ([Supplementary Fig. 14d](#S1){ref-type="supplementary-material"}), being attributed to a subtle change in the relative interstrand hydrogen-bonding strength between the homo- and heterochiral **1a**~2~, which was in accordance with the observed differences in their amide NH temperature coefficients ([Supplementary Fig. 14c](#S1){ref-type="supplementary-material"}). In sharp contrast, *rac*-**2a** was completely chiral self-sorted to form only the homochiral duplexes ((*R*,*R*)-**2a**)~2~ and ((*S*,*S*)-**2a**)~2~, giving one set of ^1^H NMR signals in CDCl~3~ independent of the % e.e. of **2a** ([Fig. 3c](#f3){ref-type="fig"}) and temperature from −20 to 50 °C ([Supplementary Fig. 19](#S1){ref-type="supplementary-material"}). This was also supported by the following mixing experiments of (*S*,*S*)- or (*R*,*R*)-**2a** with its trimethylsilyl (TMS)-deprotected analogue (*S*,*S*)-**2c**; an equimolar mixture of (*S*,*S*)-**2a** and (*S*,*S*)-**2c** in CDCl~3~ gave a new set of signals corresponding to the heteroduplex (*S*,*S*)-(**2a**)·(*S*,*S*)-**2c**, whereas no trace amount of such a distinguishable heteroduplex derived from (*R*,*R*)-**2a** and (*S*,*S*)-**2c** was observed for an equimolar mixture of (*R*,*R*)-**2a** and (*S*,*S*)-**2c** ([Fig. 3d](#f3){ref-type="fig"}), indicating that *rac*-**2a** completely self-sorts to form only the homochiral duplexes ((*R*,*R*)-**2a**)~2~ and ((*S*,*S*)-**2a**). In addition, these observations exclude the possibility of the fast exchange between the hetero- and homochiral duplexes of *rac*-**2a** on the present NMR timescale. The difference in the chiral self-sorting behaviour between **1a** and **2a** may be due to the difference in the twist angles between the two rigid diphenylethynylene groups connecting to the chiral amide linkers with a different sequence, in which the average twist angle in ((*S*,*S*)-**2a′**)~2~ (*ca.* 48°) is much larger than that in ((*S*,*S*)-**1a′**)~2~ (*ca.* 18°) in the solid state ([Fig. 2c,d](#f2){ref-type="fig"}). The density functional theory (DFT) calculations revealed that the homochiral duplex ((*S*,*S*)-**1a**)~2~ is only 17.0 kJ mol^−1^ more stable than the heterochiral duplex (*R*,*R*)-**1a**·(*S*,*S*)-**1a**, whereas, for **2a**, the homochiral duplex formation ((*S*,*S*)-**2a**)~2~ is much more favourable than the heterochiral duplex formation (*R*,*R*)-**2a**·(*S*,*S*)-**2a** by 51.7 kJ mol^−1^ ([Supplementary Fig. 21a--d](#S1){ref-type="supplementary-material"}), being reasonably consistent with the previously discussed experimental results. The observed large energy difference (Δ*E*) between the homochiral and heterochiral duplexes for **2a**~2~ (−51.7 kJ mol^−1^) as compared with that for **1a**~2~ (−17.0 kJ mol^−1^) is attributed to a distorted structure of the heterochiral homoduplex (*R*,*R*)-**2a**·(*S*,*S*)-**2a** that lacks one interstrand hydrogen bond between the amide NH and carboxy C=O groups ([Supplementary Fig. 21d](#S1){ref-type="supplementary-material"}), while all of the possible interstrand hydrogen bonds are retained in both the homo- and heterochiral **1a**~2~ during the calculations because of the perpendicular orientation of the strands as seen in the solid state ([Fig. 2c](#f2){ref-type="fig"}), resulting in a similar duplex structure ([Supplementary Fig. 21a,b](#S1){ref-type="supplementary-material"}). Therefore, predominant homochiral self-sorting could not take place for *rac*-**1a** ([Fig. 3a](#f3){ref-type="fig"}). Of further interest is that an equimolar mixture of *rac*-**1a** and *rac*-**2a** showed complete self-sorting in terms of their linker amide sequences, giving only the enantiomeric pairs of the **1a**~2~ and **2a**~2~ duplexes as explicitly observed in their ^1^H NMR spectra in CDCl~3~ ([Fig. 4a](#f4){ref-type="fig"}). Each duplex of **1a** and **2a** is unable to exchange, that is further supported by the ^1^H NMR and CD measurements of an equimolar mixture of (*S*,*S*)-**1a** and (*S*,*S*)-**2a** in CDCl~3~ ([Fig. 4a,b](#f4){ref-type="fig"}); the ^1^H NMR spectrum of an equimolar mixture of (*S*,*S*)-**1a** and (*S*,*S*)-**2a** in CDCl~3~ displayed only two sets of the signals corresponding to ((*S*,*S*)-**1a**)~2~ and ((*S*,*S*)-**2a**)~2~ ([Fig. 4a](#f4){ref-type="fig"}). Moreover, the mixture showed a CD spectrum precisely identical to the simulated one ([Fig. 4b](#f4){ref-type="fig"}), indicating that the complete sequence-selective chiral self-sorting took place within mixtures of *rac*-**1a** and *rac*-**2a**. The DFT calculations demonstrate the important role of the interstrand hydrogen bonds; the energy-minimized duplex structure of (*S*,*S*)-**1a**·(*S*,*S*)-**2a** with the different linker amide sequences has a mismatched arrangement of the interstrand hydrogen bonds ([Supplementary Fig. 21e](#S1){ref-type="supplementary-material"}), and is 11.1 and 32.6 kJ mol^−1^ less stable than those of the chiral self-sorted homoduplexes ((*S*,*S*)-**2a**)~2~ and ((*S*,*S*)-**1a**)~2~, respectively ([Supplementary Fig. 21a,c](#S1){ref-type="supplementary-material"}). On the basis of the total energies of the calculated structures ([Supplementary Fig. 21](#S1){ref-type="supplementary-material"}), the stabilities of the duplexes decrease in the following order: ((*S*,*S*)-**1a**)~2~\>(*R*,*R*)-**1a**·(*S*,*S*)-**1a**\>((*S*,*S*)-**2a**)~2~\>(*S*,*S*)-**1a**·(*S*,*S*)-**2a**\>(*R*,*R*)-**2a**·(*S*,*S*)-**2a**, which are in good agreement with the experimental results, and the energetically disfavoured duplexes (*S*,*S*)-**1a**·(*S*,*S*)-**2a** and (*R*,*R*)-**2a**·(*S*,*S*)-**2a** were not detected at all under the present experimental conditions. Chirality- and sequence-selective heteroduplex formation -------------------------------------------------------- With all the above results taken together with our previous findings of the complementary double-helix formations through amidinium--carboxylate salt bridges[@b44][@b47][@b48][@b49][@b50], we envisaged that the optically active amidine dimers linked through the chiral or *meso* amide linkers with a different sequence (NHCO-*c*Hex (**L1**) or CONH-*c*Hex (**L2**)) (**3a--e** and **4a--c**, [Fig. 1b](#f1){ref-type="fig"}) could selectively recognize the carboxylic acid dimers (**1a** and **2a**) according to the linker chirality and sequences, thereby leading to a unprecedented diastereo- and sequence-selective complementary duplex formation. On mixing *rac*-**2a** and (*R*,*R*,*R*,*R*,*R*,*R*)-**3a** composed of the dimeric (*R*,*R*,*R*,*R*) amidine linked through the (*R*,*R*)-**L2** linker in a 2:1 molar ratio in CDCl~3~, the ^1^H NMR spectrum immediately changed to that consisting of complementary diastereomeric duplexes of **3a**·(*R*,*R*)-**2a** and **3a**·(*S*,*S*)-**2a** together with the remaining **2a** ([Supplementary Fig. 25](#S1){ref-type="supplementary-material"}). The nonequivalent N--H proton signals appeared at a low magnetic field (13--13.5 p.p.m.), suggesting the preferred-handed duplex formation stabilized by salt bridges[@b44][@b47][@b48][@b49][@b50]. Because the chain exchange rate between the complementary dimer strands is slower than the NMR timescale[@b49], the diastereoselectivity between (*R*,*R*,*R*,*R*,*R*,*R*)-**3a** and *rac*-**2a** was estimated to be d.e.=58% ((*R*,*R*)-**2a**-rich, run 1, [Table 1](#t1){ref-type="table"}) from their integral ratio, indicating the homochiral (*R*,*R*) selectivity with respect to their *c*Hex linker chirality. This diastereoselectivity was further confirmed by CD; the observed CD spectrum is almost identical to the simulated CD ([Supplementary Fig. 26](#S1){ref-type="supplementary-material"}). Interestingly, the amidine dimer (*R*,*R*,*S*,*S*,*R*,*R*)-**3b** bearing the opposite (*S*,*S*)-**L2** linker showed a perfect diastereoselectivity towards *rac*-**2a**, producing the complementary duplex of **3b**·(*S*,*S*)-**2a** with d.e.\>99% as well as (*R*,*R*)-**2a** (e.e.\>99%) remaining as a homochiral duplex, as evidenced by the ^1^H NMR and identical experimental and simulated CD spectra (run 2, [Table 1](#t1){ref-type="table"}, [Fig. 5a--c](#f5){ref-type="fig"} and [Supplementary Figs 27 and 28](#S1){ref-type="supplementary-material"}). It should be noted that highly disfavoured **3b**·(*R*,*R*)-**2a** was hardly formed even in the presence of a large excess of (*R*,*R*)-**2a** (12 equivalents), indicating that the binding affinity of (*S*,*S*)-**2a** to **3b** is at least 10^5^ times higher than that of (*R*,*R*)-**2a** ([Supplementary Fig. 29](#S1){ref-type="supplementary-material"}). This surprisingly high diastereoselectivity[@b57][@b58] of **3b** towards *rac*-**2a** enabled the separation of *rac*-**2a** into the (*S*,*S*)- and (*R*,*R*)-**2a** enantiomers by facile silica gel column chromatography, giving the corresponding optically pure enantiomers in 85% and 63% yield, respectively ([Supplementary Fig. 30](#S1){ref-type="supplementary-material"}). (*R*,*R*,*S*,*S*,*R*,*R*)-**3b** could be recovered for further separation of *rac*-**2a**. Moreover, the resolved optically pure (*R*,*R*)-**2a** (e.e.\>99%) could be also used to separate *rac*-**3b** in principle. More practically, the chemical bonding of (*R*,*R*,*S*,*S*,*R*,*R*)-**3b** to chromatographic supports including silica gel will enable more efficient and preparative separation of *rac*-**2a** as well as other enantiomers as a novel chiral stationary phase (CSP) for chromatographic enantioseparation. The complementary 1:1 duplex formation of **3b**·(*S*,*S*)-**2a** was further evidenced using an electron-spray ionization mass spectroscopic (ESI-MS) measurement and vapour pressure osmometry experiment ([Supplementary Figs 31--33](#S1){ref-type="supplementary-material"}). More interestingly, (*R*,*R*)-**3c** and (*R*,*R*)-**3d**, in which the chiral amidine residues were replaced by achiral isopropyl (**3c**) and cyclohexyl (**3d**) substituents, respectively, also performed a perfect diastereoselectivity towards *rac*-**2a** to form duplexes only with (*R*,*R*)-**2a** (d.e.\>99%) having the same linker chirality, which also resulted in the optically pure (*S,S*)-**2a** (e.e.\>99%) as the remaining homochiral duplex (runs 3, 4, [Table 1](#t1){ref-type="table"} and [Supplementary Figs 34--37](#S1){ref-type="supplementary-material"}), suggesting that the diastereoselective duplex formation is governed by the linker chirality, and the chiral amidine residues may be no longer required. In other words, an anticipated complementary double-stranded helix formation stabilized by chiral amidinium--carboxylate salt bridges that would occur on both sides of the linker may not be prerequisite in order to achieve diastereoselective duplex formations. Therefore, the amidine dimer having the chiral amidine residues linked through achiral *meso*-linker (*R*,*R*,*meso*,*R*,*R*)-**3e** completely lost its diastereoselectivity (d.e.=0%; run 5, [Table 1](#t1){ref-type="table"} and [Supplementary Figs 38 and 39](#S1){ref-type="supplementary-material"}). Various attempts to obtain crystals of the complementary duplex dimers suitable for an X-ray analysis produced only amorphous solids. Therefore, the energy-minimized structures of the duplexes of (*R*,*R*,*S*,*S*,*R*,*R*)-**3b**·(*S*,*S*)-**2a** and (*R*,*R*,*S*,*S*,*R*,*R*)-**3b**·(*R*,*R*)-**2a** were constructed using the semiempirical molecular orbital calculations followed by the DFT calculations based on an analogous crystal structure[@b47]. The initial model structures of (*R*,*R*,*S*,*S*,*R*,*R*)-**3b**·(*S*,*S*)-**2a** and its diastereomer (*R*,*R*,*S*,*S*,*R*,*R*)-**3b**·(*R*,*R*)-**2a** with a right-handed twisted conformation, in which the pendant 1-octynyl groups are replaced by hydrogen atoms, were constructed so as to satisfy the following experimental results: (1) as shown in their ^1^H NMR spectra ([Fig. 5d](#f5){ref-type="fig"} and [Supplementary Fig. 62](#S1){ref-type="supplementary-material"}), all of the amide NH resonances of (*R*,*R*,*S*,*S*,*R*,*R*)-**3b**·(*S*,*S*)-**2a** showed significant downfield shifts compared with those of the amidine strand (*R*,*R*,*S*,*S*,*R*,*R*)-**3b** (Δ*δ*=0.53 or 0.73 p.p.m.; [Table 2](#t2){ref-type="table"}, entry 2), indicating that all of the amide protons participate in hydrogen bonds. In contrast, the amide NH resonances of (*R*,*R*,*S*,*S*,*R*,*R*)-**3b**·(*R*,*R*)-**2a** were slightly shifted downfield (Δ*δ*=0.12 or 0.09 p.p.m.), suggesting weak hydrogen bonds. (2) An analogous complementary dimeric duplex composed of the identical (*R*,*R*)-amidine and carboxylic acid dimer strands linked by diacetylene residues bound together through salt bridges was determined to have a right-handed helical structure using the single-crystal X-ray analysis[@b47]. The resultant energy-minimized structures of (*R*,*R*,*S*,*S*,*R*,*R*)-**3b**·(*S*,*S*)-**2a** and (*R*,*R*,*S*,*S*,*R*,*R*)-**3b**·(*R*,*R*)-**2a** with their total energies are depicted in [Fig. 6a,b](#f6){ref-type="fig"}, respectively, which revealed that they take a largely bent-shaped, right-handed double-helix-like structure, in which (*R*,*R*,*S*,*S*,*R*,*R*)-**3b**·(*S*,*S*)-**2a** is 57.4 kJ mol^−1^ more stable than the other. All of the linker amide protons in **3b**·(*S*,*S*)-**2a** form inter- or intramolecular hydrogen bonds with the amide C=O groups (average NH···O distance=2.14 Å; [Fig. 6a](#f6){ref-type="fig"}), whereas only two of the amide protons of **3b**·(*R*,*R*)-**2a** participate in such hydrogen bonds with a longer average NH···O distance (2.30 Å; [Fig. 6b](#f6){ref-type="fig"}). The observed difference in the hydrogen-bonding networks between **3b**·(*S*,*S*)-**2a** and **3b**·(*R*,*R*)-**2a** was supported by their ^1^H NMR spectra ([Fig. 5d](#f5){ref-type="fig"} and [Supplementary Fig. 62](#S1){ref-type="supplementary-material"}). Therefore, the calculated structures reasonably explain the present unexpectedly high diastereoselective duplex formation between **3b** and (*S*,*S*)-**2a**. The 2D NOESY spectra of **3b**·(*S*,*S*)-**2a** in CDCl~3~ showed interstrand NOE cross-peaks including those between the terminal TMS protons and the phenyl moieties of the amidine residues along with the interstrand aromatic protons, while interstrand NOEs between the linker *c*Hex protons were not identified because of the same amide linker sequence (NHCO-*c*Hex) whose chemical shifts were quite similar to each other ([Supplementary Figs 68--70](#S1){ref-type="supplementary-material"}). On the other hand, an interstrand NOE was clearly observed between the linker *c*Hex protons of **4b**·(*R*,*R*)-**2a** because of the different linker amide sequences ([Supplementary Figs 71 and 72](#S1){ref-type="supplementary-material"}). Considering all the results including the 2D NOESY, ESI-MS, vapour pressure osmometry, salt-bridge formations and large downfield shifts of the linker amide NH resonances of **3b**·(*S*,*S*)-**2a**, the calculated structure ([Fig. 6a](#f6){ref-type="fig"}) is most likely retained in solution. The amidine dimers **3a**--**3d** also formed a duplex in a diastereoselective manner towards *rac*-**1a** with the linker amide sequence (NHCO-*c*Hex) different from that of *rac*-**2a** and **3a**--**3d** (CONH-*c*Hex; runs 6--9, [Table 1](#t1){ref-type="table"} and [Supplementary Figs 40--47](#S1){ref-type="supplementary-material"}). However, their diastereoselectivities significantly decreased with the opposite heterochiral selectivity with respect to the *c*Hex linker chirality, for example, (*R*,*R*)-**3c** preferentially formed a duplex with (*S*,*S*)-**1a** in 64% d.e., which is in significant contrast to the homochiral selectivity observed between **3a**--**3d** and *rac*-**2a** (runs 1--4, [Table 1](#t1){ref-type="table"}). Again, the linker chirality plays a critical role so that (*R*,*R*,*meso*,*R*,*R*)-**3e** showed no diastereoselectivity (run 10, [Table 1](#t1){ref-type="table"} and [Supplementary Figs 48 and 49](#S1){ref-type="supplementary-material"}). The reason for this heterochiral preference between **3a-d** and *rac*-**1a** is not totally understood, but may be due to the difference in the interstrand hydrogen-bonding networks that could be more efficiently formed between the heterochiral linker amide residues than between the homochiral counterparts. This speculation is supported by the fact that the amide NH resonances of, for instance, the heterochiral (*R*,*R*,*S*,*S*,*R*,*R*)-**3b**·(*R*,*R*)-**1a** showed higher downfield shifts (Δ*δ*=0.56 p.p.m.) as compared with those of the homochiral (*R*,*R*,*S*,*S*,*R*,*R*)-**3b**·(*S*,*S*)-**1a** (Δ*δ*=0.03 p.p.m.), respectively ([Supplementary Fig. 64](#S1){ref-type="supplementary-material"} and [Table 2](#t2){ref-type="table"}, entry 1). We also investigated the diastereoselective duplex formations using a series of amidine dimers (**4a**--**4c**) with the NHCO-*c*Hex linker amide sequence towards *rac*-**1a** and *rac*-**2a**. In all the duplex formations, moderate diastereoselectivities (64--80% d.e.) were observed except for that between **4b** and *rac*-**2a** (34% d.e.; runs 11--16, [Table 1](#t1){ref-type="table"} and [Supplementary Figs 50--61](#S1){ref-type="supplementary-material"}), which were in contrast to the perfect diastereoselectivities achieved between **3b**--**3d** and *rac*-**2a**. As for the selectivities with regard to the *c*Hex amide linker chirality, there is the same tendency; **4a**--**4c** favourably formed duplexes with **1a** and **2a** with the same and opposite configurations, respectively. As a typical example, (*R*,*R*)-**4c** preferentially formed a duplex with (*R*,*R*)-**1a**, but with (*S*,*S*)-**2a** in 70 and 80% d.e., respectively. The observed diastereoselectivities also rely on the difference in the interstrand hydrogen-bond strengths between the diastereomeric duplexes as revealed by more downfield shifts of the amide NH protons for the major duplex diastereoselectively formed during the complexations ([Supplementary Figs 65--67](#S1){ref-type="supplementary-material"} and [Table 2](#t2){ref-type="table"}, entries 3 and 4). On the basis of these results, we anticipated that particular chiral amidine dimers would simultaneously recognize the chirality and sequence within a mixture of complementary carboxylic acid dimers via specific duplex formations. In fact, the mixing of two equivalents of each of *rac*-**1a** and *rac*-**2a** with (*R*,*R*,*S*,*S*,*R*,*R*)-**3b** in CDCl~3~ resulted in the formation of only the (*R*,*R*,*S*,*S*,*R*,*R*)-**3b**·(*S*,*S*)-**2a** duplex (d.e.\>99%), while no duplex formation was observed towards *rac*-**1a**. As a result, an optically pure (*R*,*R*)-**2a** (e.e.\>99%) quantitatively remained as a homochiral duplex together with free *rac*-**1a** ([Fig. 7](#f7){ref-type="fig"}); thus, the perfect sequence- (CONH-*c*Hex or NHCO-*c*Hex) and chirality- ((*R*,*R*) or (*S*,*S*)) selective duplex formation was achieved as unambiguously evidenced using their ^1^H NMR analysis ([Fig. 7a](#f7){ref-type="fig"}) and the observed and simulated CD spectra ([Fig. 7b,c](#f7){ref-type="fig"}). The linker chirality rather than the amidine chirality also plays a vital role in the sequence- and chirality-selective complementary duplex formation; therefore, (*R*,*R*)-**3d** composed of achiral amidine residues formed a duplex only with (*R*,*R*)-**2a** with a complete diastereoselectivity (d.e.\>99%) and sequence specificity in a mixture of *rac*-**1a** and *rac*-**2a** ([Supplementary Figs 74 and 75](#S1){ref-type="supplementary-material"}). Discussion ========== We demonstrate here an unprecedented successive chiral self-sorting during the unique homo- and subsequent heteroduplex formations through interstrand multihydrogen bonds that take place in a perfect sequence-selective way accompanied with an extraordinary high diastereoselectivity in the latter case, which enables the separation of the racemic strands into the enantiomers. The present chirality- and sequence-selective successive self-sorting is most likely achieved because of the rigid chiral geometries generated by the chiral amide-linked dimeric strands that arrange functional groups in a suitable way for specific duplex formations. In our previous studies, we showed that a carboxylic acid dimer such as **CC** formed an intertwined homodouble helix via interstrand association of the carboxylic acids[@b51], which further formed a double-stranded helix (**AA·CC**) with the complementary amidine dimer such as **AA** ([Fig. 1a](#f1){ref-type="fig"})[@b44][@b47][@b48][@b49]. The helical handedness of the complementary double helix is fully controlled by the chirality introduced on the amidine residues[@b49]. On the basis of the present studies, however, it appears that the chirality and sequence (NHCO-*c*Hex or CONH-*c*Hex) of the amide linkers of dimeric carboxylic acid and amidine strands are of primary importance and dictate the overall chirality- and sequence-selective self-assemblies of the dimer strands, resulting from the unique multihydrogen-bonding networks formed between the linker amide residues. Importantly, the amidine chirality is not almost involved in the present diastereoselective complementary duplex formations that mostly rely on the sequence and chirality of the amide linkers of the amidine dimers relative to those of the carboxylic acid dimers. The dimers of amidines and carboxylic acids possess reactive trimethylsilylethynyl groups at the ends, which allows longer oligomers joined by chiral amide linkers with specific sequence and chirality to be synthesized, which would provide a unique strategy for asymmetric template synthesis and an artificial replication system[@b59][@b60] based on chirality- and sequence-selective multihydrogen-bond-assisted duplex formations, and also provide a clue towards a better understanding of the biological chiral self-sorting process. Methods ======= General procedures for the complementary duplex formations ---------------------------------------------------------- A typical procedure for the diastereoselective duplex formations between amidine and carboxylic acid dimers is described below. Stock solutions of (*R*,*R*,*R*,*R*,*R*,*R*)-**3a** (2.0 mM; solution I) and *rac*-**2a** (2.0 mM; solution II) were prepared in dry CDCl~3~. Aliquots of I (0.40 μmol, 200 μl), II (0.80 μmol, 400 μl) and dry CDCl~3~ (200 μl) were added to an NMR tube, and its ^1^H NMR spectrum was measured at 25 °C to determine the % d.e. of the **3a**·**2a** duplexes ([Supplementary Fig. 25](#S1){ref-type="supplementary-material"}). The solution was also used for measuring the CD ([Supplementary Fig. 26](#S1){ref-type="supplementary-material"}). In a similar way, other diastereoselective duplex formations were preformed and their % d.e. values were estimated ([Table 1](#t1){ref-type="table"}). Optical resolution of *rac*-2a via the heteoduplex formation ------------------------------------------------------------ (*R*,*R*,*S*,*S*,*R*,*R*)-**3b** (5.44 mg, 3.58 μmol) and two equivalents of *rac*-**2a** (9.46 mg, 7.16 μmol) were dissolved in CHCl~3~ (2 ml), which produced an equimolar mixture of **3b**·(*S*,*S*)-**2a** and (*R*,*R*)-**2a** judging from its ^1^H NMR spectrum ([Fig. 5a](#f5){ref-type="fig"}) and were separated into the first and second fractions, respectively, using flash column chromatography (SiO~2~, 2 cm (i.d.) × 12 cm; eluent; CHCl~3~/MeOH=1/0--50/1, v/v). The % d.e. of the **3b**·(*S*,*S*)-**2a** duplex obtained (6.3 mg) was \>99% as estimated using its CD and ^1^H NMR spectra ([Supplementary Fig. 30a,c](#S1){ref-type="supplementary-material"})), while the second fraction mainly contained (*R*,*R*)-**2a** along with a small amount of **3b**·(*S*,*S*)-**2a** and an unknown compound probably generated during the column chromatography. Thus, the second fraction was purified again using flush column chromatography, yielding **3b**·(*S*,*S*)-**2a** (2.3 mg, \>99% d.e., total yield; 8.6 mg, 85%) and (*R*,*R*)-**2a** containing a small amount of the unknown compound. Further purification using recycle size-exclusion chromatography (eluent; CHCl~3~) afforded pure (*R*,*R*)-**2a** (2.99 mg) in 63% yield; its e.e. value was estimated to be \>99% on the basis of the CD and ^1^H NMR measurements ([Supplementary Fig. 30b,c](#S1){ref-type="supplementary-material"}). Additional information ====================== **Accession codes:** The X-ray crystallographic coordinates for structures ((*S*,*S*)-**1a′** and (*S*,*S*)-**2a′**) reported in this Article have been deposited at the Cambridge Crystallographic Data Centre (CCDC), under deposition number CCDC 1036590 and 1036591. These data can be obtained free of charge from The Cambridge Crystallographic Data Centre via [www.ccdc.cam.ac.uk/data_request/cif](http://www.ccdc.cam.ac.uk/data_request/cif). **How to cite this article:** Makiguchi, W. *et al*. Chirality- and sequence-selective successive self-sorting via specific homo- and complementary-duplex formations. *Nat. Commun.* 6:7236 doi: 10.1038/ncomms8236 (2015). Supplementary Material {#S1} ====================== ###### Supplementary Information Supplementary Figures 1-117, Supplementary Tables 1-2, Supplementary Methods and Supplementary References ###### Supplementary Data 1 cif file for x-ray crystal structure of (S,S)-1a\' ###### Supplementary Data 2 cif file for x-ray crystal structure of (S,S)-2a\' This work was supported in part by a Grant-in-Aids for Scientific Research (S) from the Japan Society for the Promotion of Science (JSPS; no. 25220804) and by the Nanotechnology Platform Program (Molecule and Material Synthesis) of the Ministry of Education, Culture, Sports, Science and Technology, Japan. W.M. and J.T. express thanks for a JSPS Research Fellowship for Young Scientists (no. 480 for W.M. and no. 8886 for J.T.). **Author contributions** E.Y. designed and directed the project, and W.M., J.T. and H.Y. performed the experiments. H.I. confirmed the validity of the X-ray crystallographic analysis. N.O. calculated the duplex structures by DFT. All authors analysed the data and discussed the results. N.O. and E.Y. co-wrote the manuscript. {#f1} {ref-type="supplementary-material"}). (**c**,**d**) X-ray crystal structures of (*S*,*S*)-**1a′** (**c**) and (*S*,*S*)-**2a′** (**d**). Hydrogen atoms and solvent molecules are omitted for clarity. Black dotted lines represent interstrand hydrogen bonds. Average twist angles between the diphenylethynylene groups (red dotted lines) connecting to the amide linker, in which the twist angles are defined as the dihedral angles formed by the four carbon atoms (marked with red open circles), and interstrand hydrogen-bonding schemes are also shown.](ncomms8236-f2){#f2} {ref-type="supplementary-material"}. (**b**) Plots of % d.e. of **1a**~2~ versus the % e.e. of **1a** in CDCl~3~ at 25 °C. The plots were fitted using *Y*=100{−2\[4−3(*X*/100)^2^\]^0.5^+5}/3 (equation A), where *X* and *Y* are % e.e. of **1a** and % d.e. of **1a**~2~, respectively (for details, see [Supplementary Methods](#S1){ref-type="supplementary-material"}). (**c**) ^1^H NMR spectra of carboxylic acid CO~2~H and amide NH proton resonances of **2a** (0.50 mM) with different % e.e. in CDCl~3~ at 25 °C. Full-scale spectra are shown in [Supplementary Fig. 18b](#S1){ref-type="supplementary-material"}. (**d**) Partial ^1^H NMR spectra (0.50 mM) of (*S*,*S*)-**2a**, (*S*,*S*)-**2c**, an equimolar mixture (1.0 mM) of (*S*,*S*)-**2a** and (*S*,*S*)-**2c**, and an equimolar mixture (1.0 mM) of (*R*,*R*)-**2a** and (*S*,*S*)-**2c** in CDCl~3~ at 25 °C. Red circles denote the signals because of the duplex (*S*,*S*)-**2a**·(*S*,*S*)-**2c**. Full-scale spectra are shown in [Supplementary Fig. 20](#S1){ref-type="supplementary-material"}.](ncomms8236-f3){#f3} {ref-type="supplementary-material"}. (**b**) CD and absorption spectra of (*S*,*S*)-**1a** (0.50 mM), (*S*,*S*)-**2a** (0.50 mM) and those observed and simulated for the mixture of an equimolar amount of (*S*,*S*)-**1a** and (*S*,*S*)-**2a** (0.50 mM) in CDCl~3~ at 25 °C.](ncomms8236-f4){#f4} {ref-type="supplementary-material"}. (**b**) CD (upper) and absorption (bottom) spectra (0.50 mM) of **3b**, (*R*,*R*)-**2a**, (*S*,*S*)-**2a**, **3b**·(*R*,*R*)-**2a** and **3b**·(*S*,*S*)-**2a** in CDCl~3~ at ambient temperature. (**c**) Experimental and simulated CD (d.e.=100%) and absorption spectra for a mixture of **3a** (0.50 mM) and two equivalents of *rac*-**2a** (1.0 mM) in CDCl~3~ at ambient temperature. For the simulated CD and absorption spectra, see [Supplementary Methods](#S1){ref-type="supplementary-material"}. (**d**) Partial ^1^H NMR spectra (0.5 mM) of **3b**, **3b**·(*R*,*R*)-**2a** and **3b**·(*S*,*S*)-**2a** in CDCl~3~ at 25 °C. The linker amide NH resonances were assigned using ^1^H 2D NMR spectroscopy ([Supplementary Fig. 62](#S1){ref-type="supplementary-material"}).](ncomms8236-f5){#f5} {#f6} {ref-type="supplementary-material"}. (**b**) CD (upper) and absorption (bottom) spectra (0.50 mM) of **3b**·(*S*,*S*)-**2a**, (*R*,*R*)-**2a** and *rac*-**1a** in CDCl~3~ under ambient temperature. (**c**) Experimental and simulated CD (d.e.=100%) and absorption spectra for a mixture of **3b** (0.50 mM) and two equivalents of each of *rac*-**1a** (1.0 mM) and *rac*-**2a** (1.0 mM) in CDCl~3~ under ambient temperature.](ncomms8236-f7){#f7} ###### Results of diastereoselective duplex formations between chiral amidine dimers (3a-4c) and racemic carboxylic acid dimers (*rac*-1a and *rac*-2a). **Amidine dimer (amide sequence)** ***rac*****-2a (CONH-*****c*****Hex)** ***rac*****-1a (NHCO-*****c*****Hex)** ---------------------------------------------- ---------------------------------------- ---------------------------------------- ---- ---- -------------- ---- (*R,R,**R**,**R**,R,R*)-**3a** (CONH-*c*Hex) 1 58 (*R*,*R*) 25 6 14 (*S*,*S*) 40 (*R,R,**S**,**S**,R,R*)-**3b** (CONH-*c*Hex) 2 \>99 (*S*,*S*) 27 7 64 (*R*,*R*) 42 (***R**,**R***)-**3c** (CONH-*c*Hex) 3 \>99 (*R*,*R*) 34 8 64 (*S*,*S*) 44 (***R**,**R***)-**3d** (CONH-*c*Hex) 4 \>99 (*R*,*R*) 36 9 68 (*S*,*S*) 46 (*R,R,meso,R,R*)-**3e** (CONH-*c*Hex) 5 0 38 10 0 48 (*R,R,**R**,**R**,R,R*)-**4a** (NHCO-*c*Hex) 11 70 (*S*,*S*) 50 14 64 (*R*,*R*) 56 (*R,R,**S**,**S**,R,R*)-**4b** (NHCO-*c*Hex) 12 34 (*R*,*R*) 52 15 74 (*S*,*S*) 58 (***R**,**R***)-**4c** (NHCO-*c*Hex) 13 80 (*S*,*S*) 54 16 70 (*R*,*R*) 60 ^\*^Estimated using ^1^H NMR (CDCl~3~, 25 °C). ^†^Supplementary Fig. no. ###### Relationships between the diastereoselective complementary duplex formations and downfield chemical shifts (Δ*δ*) of the linker amide NH resonances of the amidine strands of **3b·1a--4b·2a** duplexes from the monomeric amidine strands (**3b** and **4b**). **Entry** **Combination** **Duplex** **Δ*****d***~**NH**~ **(p.p.m.)** **d.e. (%)** **Supplementary Fig no.** ----------- --------------------------------------------------------------------------------- ------------------------------------ ------------------------------------------ ------------------------------ --------------------------- 1 (*R*,*R*,***S***,***S***,*R*,*R*)-**3b** (CONH-*c*Hex) and **1a** (NHCO-*c*Hex) **3b**·(*R*,*R*)-**1a**favoured 0.56 64 ((*R*,*R*)-**1a**-rich) 64     **3b**·(*S*,*S*)-**1a**disfavoured 0.03     2 (*R*,*R*,***S***,***S***,*R*,*R*)-**3b** (CONH-*c*Hex) and **2a** (CONH-*c*Hex) **3b**·(*R*,*R*)-**2a**disfavoured 0.12 or 0.09[\*](#t2-fn2){ref-type="fn"} \>99 ((*S*,*S*)-**2a**-rich) 62     **3b**·(*S*,*S*)-**2a**favoured 0.73 or 0.53[\*](#t2-fn2){ref-type="fn"}     3 (*R*,*R*,***S***,***S***,*R*,*R*)-**4b** (NHCO-*c*Hex) and **1a** (NHCO-*c*Hex) **4b**·(*R*,*R*)-**1a**disfavoured 0.52 74 ((*S*,*S*)-**1a**-rich) 67     **4b**·(*S*,*S*)-**1a**favoured 1.16 or 1.08[\*](#t2-fn2){ref-type="fn"}     4 (*R*,*R*,***S***,***S***,*R*,*R*)-**4b** (NHCO-*c*Hex) and **2a** (CONH-*c*Hex) **4b**·(*R*,*R*)-**2a**favoured 0.90 34 ((*R*,*R*)-**2a**-rich) 66     **4b**·(*S*,*S*)-**2a**disfavoured 0.59     *c*Hex, (*R*,*R*)-*trans*-1,2-cyclohexane. ^\*^The amide NH resonances of the amidine strands of duplexes could not be distinguished from those of the carboxylic acid strands because of overlapping with the phenyl and aliphatic proton signals. For the assignments of the amide NH proton signals, see [Supplementary Figs 62--67](#S1){ref-type="supplementary-material"}.

Introduction ============ Black phosphorus (BP), as a novel 2D material, has attracted global attention owing to outstanding optoelectronic properties and wide applications ([@B17]; [@B56]). Bulk BP can be easily exfoliated into nanosheets (NSs) with different thicknesses ([@B7]; [@B47]). Compared with other 2D materials such as graphene and MoS~2~, BP has a larger specific surface area to adsorb large amounts of theranostic agents or antitumor drugs, thereby being potentially applicable to drug delivery ([@B34]). Unlike other 2D materials, the bandgap voltage of BP is related with its number of layers, ranging from 0.3 eV (a bulk value) to ∼2.0 eV (a monolayer value), ([@B20]) so it has absorptions in both UV and near-infrared (NIR) regions. Thus, BP NSs have unique optoelectronic performance ([@B28]) to work as either an excellent nano-optoelectronic device or an effective photothermal agent for photothermal therapy (PTT) due to high photothermal conversion efficiency and NIR extinction coefficient ([@B16]; [@B32], [@B31]; [@B46], [@B45]). Therefore, antitumor drug doxorubicin (DOX) or paclitaxel are often loaded onto BP NSs to fabricate multifunctional drug delivery systems for synergistic cancer chemotherapy/PTT. Nevertheless, the application of BP is greatly hindered, because it is prone to degradation into P*x*O*y* in the air and aqueous solutions ([@B21]; [@B8]). To this end, researchers have endeavored to stabilize BP NSs through surface modification strategies such as ligand surface coordination ([@B55]; [@B15]), covalent aryl diazonium functionalization ([@B27]) and capping layer protection ([@B41]), which, however, are unsuitable for human drug delivery for either introducing toxic substances or weakening photothermal outcomes. As a surface-adherent biomimetic material formed through the oxidative self-polymerization of dopamine under alkaline conditions, polydopamine (PDA) is inspired by marine mussel and has been widely used a coating on nanomaterial surfaces owing to high biodegradability, biocompatibility and pH responsiveness at low pH values ([@B19]; [@B22]; [@B4]). We have previously elevated the stability of BP NSs in aqueous solution by simply modifying their surface with PDA safely and effectively, without attenuating the photothermal effects ([@B10]), based on the photothermal conversion efficiency of PDA ([@B22]; [@B5]; [@B24]). Besides, BP NSs can be readily phagocytosed and cleared by the mononuclear phagocytic system (MPS) after being injected *in vivo*. Conventionally, the *in vivo* circulation of nanocarriers is prolonged through surface modification with hydrophilic or zwitterionic polymers among which polyethylene glycol (PEG) is most investigated and utilized due to excellent biocompatibility and hydrophilicity, especially for the polymers based drug delivery ([@B2]; [@B9]; [@B14]; [@B39]). This technique is also known as PEGylation ([@B51]; [@B6]; [@B23]; [@B53]; [@B43]; [@B33]). However, it suffers from the following issues. First, PEGylated therapeutic agents, after being administered repeatedly, cannot fully escape from being phagocytosed by cells in MPS, and the immunogenicity is bound to induce obvious humoral immune response. Therefore, they are recognized and removed by the immune system. Second, PEGylated liposome and nanoparticles can be immunoreactive to induce an accelerated blood clearance (ABC) phenomenon ([@B40]; [@B36]). Third, the *in vivo* stability of PEG is affected, because its polyether main chain easily undergoes oxidative degradation. [@B1] found the cleared PEG and PDLLA fragments of (ethylene glycol)-block-poly([D,L]{.smallcaps}-lactide) (PEG-b-PDLLA) micelles in mouse urine. Furthermore, it is difficult to conjugate the surface of PEGylated nanocarriers with functional ligands because PEG has limited reactive groups, thus requiring an alternative technique to stabilize BP NSs for *in vivo* biomedical applications ([@B42]). In recent years, poly(2-ethyl-2-oxazoline) (PEOz) has been verified as a high-molecular weight, long-chain polymer with high water solubility, flexibility and biocompatibility, and approved by the United States Food and Drug Administration. Meanwhile, PEOz is capable of long circulation *in vivo*, inhibiting protein adsorption and decreasing blood clearance, as a qualified substitute for PEG. Compared with PEG, PEOz has a more stable main chain which facilitates the introduction of various active groups and provides a chemical basis for further linking to target molecules. Notably, with unique tertiary amide groups in the main chain, PEOz has a similar p*K*a to that of physiological pH, which can be adjusted by varying the molecular weight. At pH lower than its pKa, PEOz is reversed from negatively to positively charged through ionization of tertiary amide groups along the PEOz chain. As a result, PEOz-modified drug nanodelivery system can be enriched and charge-reversed in the weakly acidic environment of tumor tissue (pH ∼6.8), allowing endocytosis and pH-responsive drug release after being induced by the low pH (∼5.0) of endosomes and lysosomes. Finally, the drug release rate is controllable and the tumor-targeting ability is improved, managing to enhance the antitumor activities and to reduce the side effect simultaneously ([@B11],[@B12],[@B13]; [@B54]; [@B37]). As a broad-spectrum antitumor agent, DOX can be intercalated into DNA of tumor cells to suppress nucleic acid synthesis, exerting therapeutic effects on acute leukemia and a variety of solid tumors. Nevertheless, it is highly cytotoxic and easily degradable, without selectivity or specificity ([@B3]; [@B44]; [@B52]). Bortezomib (BTZ), on the other hand, is a common clinical antitumor agent applicable to patients with multiple myeloma, which can inhibit tumor cell growth by binding the threonine residues of active sites of several proteases and induce apoptosis mainly via the mitochondrial pathway. However, the therapeutic effects of BTZ on many types of solid tumors are limited, because it is non-specifically bound to normal cell proteins and can thus be rapidly cleared by the blood, accompanied by dose-related cytotoxicity ([@B30]). Consequently, we herein designed a nanodrug BP-DOX\@PDA-PEOz-BTZ carrying DOX and BTZ simultaneously for the chemo/photothermal combination therapy of breast cancer. The PDA layer enhanced the system stability before reaching the tumor site, and maintained the remarkable photothermal effects for subsequent modification. Afterward, PEG was replaced by PEOz to prolong *in vivo* drug circulation and to increase cellular uptake. In the meantime, a pH-targeted controlled release trigger system was constructed to remedy the deficiency of chemical drugs in solid tumor therapy, and to further boost the antitumor effects relying on high photothermal conversion efficiency. This system is conducive to chemo/photothermal combination therapy by not only raising the drug loading content, cellular uptake and pH-responsive release rate, but also exhibiting high photothermal activity against tumor cells. Materials and Methods {#s1} ===================== Materials --------- Bulk BP was purchased from Smart-Elements (Austria) and stored in a 4°C refrigerator. Dopamine hydrochloride was bought from Sigma-Aldrich (St. Louis, MO, United States). H~2~N-PEG was obtained from Shanghai Yare Biotech, Inc. (China). H~2~N-PEOz was bought from Xi'an Ruixi Biological Technology Co., Ltd. (China). The molecular weights (M*~w~*) of PEG and PEOz were both 2 kDa. DOX was purchased from Dalian Meilun Biology Technology Co., Ltd. (China). BTZ was obtained from Beijing Zhongshuo Pharmaceutical Technology Development Co., Ltd. (China). Other analytical-grade reagents and solvents were used as received. HyPure Molecular Biology Grade Water (Hyclone^TM^) was used to prepare all solutions. Preparation of BP NSs --------------------- Black phosphorus NSs were fabricated by simply exfoliating the corresponding bulk BP sample in liquid. Briefly, 20 mg of BP was dispersed in 20 mL of 1-methyl-2-pyrrolidinone (NMP) which was argon-bubbled to decrease oxidation by eliminating dissolved oxygen molecules during exfoliation. The mixture was thereafter sonicated in an ice water bath for 8 h (amplifier: 25%, on/off cycle: 5 s/5 s). The low system temperature was kept by the ice water bath. Afterward, unexfoliated bulk BP was removed by centrifuging the brown dispersion at 2,000 rpm for 10 min, and the supernatant containing BP NSs was carefully collected for further use. DOX Loading Onto BP NS Surface ------------------------------ Doxorubicin (2 mg) was mixed with 2 mL of 1 mg mL^-1^ BP NSs solution in water, and the solution pH was adjusted to 8.5 with sodium hydroxide. After being stirred vigorously in dark overnight, the obtained DOX-loaded BP (BP-DOX) NSs were collected by centrifugation and washed by water. The LC (%) of DOX were calculated using the following equation. Drug LC ( \% ) = w e i g h t o f d r u g i n t h e n a n o p a r t i c l e s w e i g h t o f n a n o p a r t i c l e s × 100 \% PDA Coating on BP NS Surface ---------------------------- BP-DOX NSs were dispersed in 2 mL HyPure Molecular Biology Grade Water at 1 mg mL^-1^. Then pH was adjusted to 8.5 by adding sodium hydroxide, and the solution was added 10 μL dopamine hydrochloride (100 mg mL^-1^) and stirred for 2.5 h in dark at room temperature. Finally, BP-DOX\@PDA particles were collected by 10 min of centrifugation at 12,000 rpm and washed by deionized water. Conjugation of H~2~N-PEOz or H~2~N-PEG Onto BP-DOX\@PDA Surface --------------------------------------------------------------- PDA-coated NSs (2 mg) were first resuspended in 2 mL of HyPure Molecular Biology Grade Water with pH adjusted to 8.5 by an appropriate amount of sodium hydroxide. After 2 mg of H~2~N-PEOz was added into the BP\@PDA suspension, the mixture was vigorously stirred for 3 h in dark at room temperature. Then the H~2~N-PEOz-modified NPs (BP-DOX\@PDA-PEOz) were purified by 10 min of centrifugation at 12,000 rpm and washed by deionized water. With a similar procedure, BP-DOX\@PDA-PEG was fabricated by using H~2~N-PEG instead of H~2~N-PEOz. BTZ Loading Onto PDA-Coated NPs ------------------------------- In brief, 50 mg of BP-DOX\@PDA-PEOz NPs were suspended at 1 mg mL^-1^ in deionized water with pH adjusted to 8.5 by sodium hydroxide, and 6 mg of BTZ powders were dispersed in 200 μl of DMSO. Under stirring, the latter solution was then dropwise added to the former solution. Afterward, the mixture was stirred overnight and centrifuged with the same process as that described above. After 48 h of lyophilization, the product was referred to as BP-DOX\@PDA-PEOz-BTZ. Characterizations of BP NSs --------------------------- Transmission electron microscopy (TEM) images were acquired by Tecnai G2 F30 transmission electron microscope (FEI, Hillsboro, OR, United States). BP NSs were observed after being dropped onto a copper grid-coated carbon membrane and air-dried. Fourier transform infrared (FTIR) spectra were recorded with Nicolet iS 50 spectrometer (Thermo Scientific, United States). Raman spectra were recorded at room temperature by LabRAM HR800 high-resolution confocal Raman microscope (HORIBA, United States). X-ray photoelectron spectroscopy was performed with Axis HSi X-ray photoelectron spectroscope (Kratos Ltd., United Kingdom) employing Al Kα radiation (150 W, 1486.6 eV photons) as the excitation source. Zeta potential and size were measured by Malvern Mastersizer 2000 particle size analyzer (Zetasizer Nano ZS90, Malvern Instruments Ltd., United Kingdom). All measurements were conducted three times independently and averaged. Photothermal Effects of Different BP NSs ---------------------------------------- The aqueous solutions (0.1 mg mL^-1^) of different NSs (BP NSs, BP\@PDA, BP\@PDA-PEOz) were added into microfuge tubes. With the same process, water was utilized as control. The middle of each solution was irradiated at 808 nm with KS-810F-8000 fixed fiber-coupled continuous semiconductor diode laser (Kai Site Electronic Technology Co., Ltd., Xi'an, China) at the power density of 1.0 W cm^-2^. To evaluate the effects of concentration changes, BP\@PDA-PEOz solutions with various concentrations of NSs (10--200 μg mL^-1^) were tested by recording the temperature changes under the above-mentioned irradiation. Also, BP\@PDA-PEOz solution (100 μg mL^-1^) was tested at various power densities (0.5--2.0 W cm^-2^) to monitor the temperature changes. Ti450 IR thermal imaging camera (Fluke, United States) was used for temperature recording. pH and Photothermal-Induced Drug Release Profiles ------------------------------------------------- To evaluate the DOX release profile of BP-DOX\@PDA-PEOz-BTZ NSs, 5 mg of NSs were resuspended in 1 mL of phosphate-buffered saline (PBS, pH = 5.0, 6.8 or 7.4, containing 0.1% w/v Tween 80). Subsequently, the dispersion was transferred into a dialysis membrane bag \[MWCO = 3500, Sangon Biotech (Shanghai) Co., Ltd., China\] that was then incubated in 10 mL of PBS at pH 5.0, 6.8 or 7.4 in an orbital water bath and shaken at 37°C. At dedicated time points, 0.5 mL of the solution outside was collected to detect the amount of released DOX with UV--vis spectrometer at 490 nm, which was supplemented by 0.5 mL of fresh PBS. Under identical conditions, photothermal-triggered drug release was tested at pH 5.0 with 6 min of 808 nm laser irradiation at the power density of 1.0 W cm^-2^. *In vitro* BTZ release and photothermal-triggered drug release from BP-DOX\@PDA-PEOz-BTZ NSs were detected by LC 1200 HPLC system ([@B23]) (Agilent Technologies, Santa Clara, CA, United States). Compounds were separated by a reverse-phase C18 column (5 μm, 150 × 4.6 mm; Agilent Technologies, Santa Clara, CA, United States) using a mobile phase comprising deionized water and acetonitrile (20/80 for BTZ, v/v). The flow rate was 1.0 mL/min and the injection volume was 20 μl. BTZ amount was measured by UV--vis spectroscopy at 270 nm. Finally, the accumulative release versus time profiles of BTZ and DOX were plotted. Cell Culture Assays ------------------- Breast cancer cell line MCF-7 was chosen to study the endocytic behaviors of the above NPs. The cells were incubated in Dulbecco's modified Eagle's medium (DMEM) (Life Technologies, Carlsbad, CA, United States) containing 100 μg mL^-1^ streptomycin, 100 U mL^-1^ penicillin and 10% (v/v) fetal bovine serum at 37°C with 5% CO~2~. Cellular Uptake of NPs ---------------------- MCF-7 cells were inoculated into 20 mm glass-bottomed Petri dishes and thereafter incubated for 24 h. BP\@PDA-PEG and BP\@PDA-PEOz loading 10 μg mL^-1^ DOX were added into the wells simultaneously at pH 6.8 or 7.4, followed by 4 h of incubation at 37°C. The cells were washed three times by PBS, and observed with Fluoview FV-1000 confocal laser scanning microscope (Olympus, Tokyo, Japan) at 488 and 590 nm as the excitation and emission wavelengths, respectively. Cytotoxicity Assay ------------------ The cytotoxicities of BP\@PDA-PEG and BP\@PDA-PEOz were determined by the MTT assay. MCF-7 cells were seeded into a 96-well plate at the density of 5 × 10^3^ and thereafter incubated for 24 h. Afterward, the medium was replaced by 100 μL of fresh medium containing different concentrations of BP\@PDA-PEG or BP\@PDA-PEOz (10, 25, 100 μg mL^-1^), and the cells were further incubated for 48 h. Then the medium was replaced by MTT solution in DMEM (5 mg mL^-1^, 100 μL), followed by another 4 h of incubation. The supernatant was removed from each well into which 100 μL of DMSO was then added to dissolve the formed formazan crystals. The absorption of each well was detected at 570 nm by Model 680 microplate reader (Bio-Rad, United Kingdom). *In vitro* PTT Study -------------------- MCF-7 cells were inoculated into a 96-well plate and incubated for 24 h. After the medium in each well was refreshed, the cells were incubated with various concentrations (10, 25, 50 μg mL^-1^) of BP\@PDA-PEG and BP\@PDA-PEOz at 37°C for 4 h, irradiated for 10 min by 808 nm laser (1.0 W cm^-2^) and incubated again for 12 h. The cell viability was assessed by the MTT assay. *In vitro* Cell Viabilities at Different pH Values -------------------------------------------------- MCF-7 cells were incubated overnight after being inoculated into 96-well plates. Subsequently, they were incubated by BP-DOX\@PDA-PEG/BP-DOX\@PDA-PEOz/BP-DOX\@PDA-PEOz-BTZ with equivalent DOX concentrations at pH 6.8 or 7.4. The MTT solution was added after 48 h, and the cell viability was also detected with the microplate reader at 570 nm. The viability of untreated cells was set at 100%. *In vitro* Combined Antitumor Therapy ------------------------------------- MCF-7 cells were inoculated into a 96-well plate at 1 × 10^4^/well, and incubated overnight. The adherent cells were treated with DOX, BTZ, DOX + BTZ (1:1), BP-DOX\@PDA-PEG, BP-DOX\@PDA-PEOz, BP-DOX\@PDA-PEOz-BTZ, BP-DOX\@PDA-PEG + NIR, BP-DOX\@PDA-PEOz + NIR, and BP-DOX\@PDA-PEOz-BTZ + NIR (808 nm laser irradiation, 1.0 W cm^-2^) at 0.25, 1, 2.5, and 5 μg mL^-1^ equivalent DOX concentrations for 24 and 48 h. After treatment with the NSs for 24 or 48 h in the presence or absence of NIR laser irradiation, the cell viability was tested by the MTT assay. The optical density of each well at 570 nm was measured by the microplate reader. The viability of untreated cells was set at 100%, and the absorbance of the control group was zero. *In vivo* IR Thermal Imaging ---------------------------- All animal experiments have been approved by the Administrative Committee on Animal Research in Sun Yat-sen University, and *in vivo* experiments were conducted according to corresponding guidelines. Female severe combined immunodeficient (SCID) mice aged 5--6 weeks old were provided by the Guangdong Medical Laboratory Animal Center and given free access to water and food. Each mouse was subcutaneously injected with 100 μL PBS suspension of MCF-7 cells (∼2 × 10^6^) on the dorsal side to induce tumors. Every 2 days, the length and width of tumors were measured with a digital vernier caliper to estimate the volumes using the formula: 0.5 × (length) × (width)^2^. SCID mice bearing MCF-7 tumors were employed as the animal model. IR thermal imaging was carried out when the tumor volumes reached approximately 180 mm^3^. The mice were thereafter intravenously injected with PBS, BP\@PDA-PEG, BP\@PDA-PEOz or BP\@PDA-PEOz-BTZ, and the BP dose is 5 mg/kg in 100 μl PBS. Twenty-four hours later, the tumor sites were irradiated for 5 min by 808 nm laser (1.5 W cm^-2^). IR thermographic maps and temperature changes were recorded with the IR thermal imaging camera. Statistical Analysis -------------------- Unless otherwise stated, all experiments were performed at least in triplicate, and the data were represented as mean ± standard deviation. Statistical analysis was carried out by using SPSS 22.0 software for one-way analysis of variance and subsequent Bonferroni test. ^∗^*P* \< 0.05 and ^∗∗^*p* \< 0.01 were considered statistically significant and extremely statistically significant, respectively. Results and Discussion ====================== Morphology and Characterizations -------------------------------- The entire synthesis procedure of the BP-based drug delivery platform, including DOX loading, PDA coating, PEOz conjugation and BTZ loading, is shown in [Figure 1](#F1){ref-type="fig"}. According to the modified liquid exfoliation technique reported by our group previously ([@B34]), BP NSs were prepared from bulk BP in NMP. Then DOX was absorbed onto the corrugated surface of BP NSs by non-covalent bonding such as Van der Waals force and electrostatic attraction. The DOX loading capacity of BP can reach over 300% in a weakly alkaline condition ([@B10]). In a weakly alkaline solution, dopamine monomer underwent oxidative self-polymerization into PDA, ultimately adhering to the surface of NPs ([@B19]; [@B25]; [@B18]). PDA coating enhanced the system stability in physiological medium, probably as a universal bond between NSs and ligand by being reactive with thiol and amine groups. The terminal amine group of NH~2~-PEOz conjugated to NSs coated with PDA through a simple Michael addition reaction. The oxidative self-polymerization mechanism of dopamine and the conjugation mechanism between H~2~N-PEOz and PDA coating were shown as [Supplementary Figure S1A](#SM1){ref-type="supplementary-material"}. As a long-chain molecule, PEOz was introduced into the system as a substitute for PEG to maintain long-term circulation, to attenuate the ABC phenomenon and to allow pH-responsive drug release. In alkaline solutions, the boronic acid active site in BTZ reacted with catechol in PDA, which may suppress the activity of BTZ and thus decrease non-specific cellular drug endocytosis. Additionally, BTZ was released under acidic conditions, possibly also facilitating drug release at the tumor site and selectively augmenting antitumor activity ([Supplementary Figure S1B](#SM1){ref-type="supplementary-material"}) ([@B30]). In short, DOX-loaded BP\@PDA-PEOz-BTZ may be applicable to synergistic chemotherapy/PTT by prolonging *in vivo* circulation as well as elevating cellular uptake efficiency, pH responsiveness and dual drug loading capacity. {#F1} TEM images ([Figure 2A](#F2){ref-type="fig"}--[D](#F2){ref-type="fig"}) exhibit that BP NSs have sheet-like morphology, and bare BP and modified BP NSs have the lateral sizes of approximately 200--250 nm, being consistent with the results of dynamic light scattering analysis. After PDA and PEOz coating, the NS surface became rough and slightly thickened. {#F2} [Figure 3A](#F3){ref-type="fig"} presents the FTIR spectra of BP NSs, BP\@PDA, BP\@PDA-PEG, and BP\@PDA-PEOz. The adsorption peak at ∼1,625 cm^-1^ represents P=O stretching vibration ([@B29]). After PDA coating, a broad and intense band between 3,150 and 3,600 cm^-1^ appears, corresponding to N-H/O-H stretching vibration. The peak at ∼1,500 cm^-1^ can be assigned to the bending vibrations of benzene ring and N-H in PDA. In the spectrum of BP\@PDA-PEG, the peak at about 2,900 cm^-1^ represents C-H stretching vibration, suggesting that PEG had been successfully modified. The peak at 1,640 cm^-1^ is related to the C=O stretching vibration of imide bond in PEOz ([@B26]). Moreover, the spectrum of BP\@PDA-PEOz shows a peak at ∼2,900 cm^-1^ corresponding to C-H stretching vibration ([@B48]), indicating successful modification of PEOz. {#F3} The structures of PDA- and PEOz-modified BP NSs were studied by Raman spectroscopy ([Figure 3B](#F3){ref-type="fig"}). In the spectrum of bare BP, there are three obvious peaks at ∼360.7, 437.5, and 466.1 cm^-1^ which correspond to the A^1^~g~, B~2g~, and A^2^~g~ modes of BP, respectively. The peaks of modified NSs (BP\@PDA, BP\@PDA-PEG, BP\@PDA-PEOz, BP\@PDA-PEOz-BTZ) shift toward lower wavenumber slightly, which can be ascribed to the mild ultrastructural changes after PDA coating and further modification, demonstrating successful modification of PDA and PEOz. The surface modification of NSs loading BTZ was confirmed by X-ray photoelectron spectroscopy ([Figure 3C,D](#F3){ref-type="fig"} and [Supplementary Figure S2](#SM1){ref-type="supplementary-material"}). As evidenced by the intensity increase of nitrogen peak (N1s) at 399.49 eV ([Figure 3D](#F3){ref-type="fig"}), both PDA coating and PEOz/BTZ loading were successful. The P2p peak (129.6 eV) intensities of bare BP, BP\@PDA, BP\@PDA-PEG, and BP\@PDA-PEOz gradually drop ([Supplementary Figure S2A](#SM1){ref-type="supplementary-material"}) because of the coverage of P element. Collectively, corresponding compounds had indeed been successfully modified ([@B38]). The results of dynamic light scattering are listed in [Table 1](#T1){ref-type="table"}. The hydrodynamic sizes increased slightly owing to layer-by-layer modification with PDA, PEG or PEOz, being in accordance with the TEM results. Additionally, the appropriate size and narrow size distribution may be beneficial to NSs accumulation in tumors through the EPR effect ([@B35]). Besides, the zeta potential of bare BP NSs was -18.9 mV, while that after surface modification with PDA became -16.6 mV, probably because phenolic hydroxyl groups on the PDA layer were deprotonated at neutral pH ([@B6]; [@B23]). After NH~2~-PEOz and NH~2~-PEG modification, the zeta potentials of BP\@PDA-PEOz and BP\@PDA-PEG in deionized water were measured to be -10.5 mV and -13.2 mV, respectively. ###### Characterization of BP-based NSs in deionized water. Sample Size (nm) PDI ZP (mV) ---------------------- -------------- ------- -------------- BP 217.1 ± 15.3 0.145 --18.9 ± 3.2 BP\@PDA 224.5 ± 18.1 0.150 --16.6 ± 2.7 BP\@PDA-PEG 232.7 ± 20.2 0.174 --13.2 ± 1.1 BP\@PDA-PEOz 236.3 ± 19.4 0.163 --10.5 ± 1.8 BP\@PDA-PEOz-BTZ 239.2 ± 21.8 0.127 --9.3 ± 1.5 BP-DOX\@PDA-PEG 238.4 ± 25.6 0.161 --8.1 ± 1.0 BP-DOX\@PDA-PEOz 241.5 ± 23.1 0.152 --6.4 ± 0.7 BP-DOX\@PDA-PEOz-BTZ 248.6 ± 22.0 0.133 --4.9 ± 0.5 PDI, polydispersity index; ZP, zeta potential, n = 3. Effects of pH on Size and Zeta Potential ---------------------------------------- To further evaluate the influence of pH, drug-free and DOX-loaded BP\@PDA-PEG/BP\@PDA-PEOz NSs (1 mg mL^-1^) were exposed in 10 mM PBS at pH 5.0, 6.8 or 7.4 and sonicated for 30 min at 37°C before their sizes and zeta potentials were measured. As shown in [Figure 4A](#F4){ref-type="fig"}, the sizes of all NSs barely change with reducing pH. BP\@PDA-PEOz became slightly negatively charged under physiological conditions (pH 7.4), with the zeta potential of -10.2 mV ([Figure 4B](#F4){ref-type="fig"}). As pH decreased from 7.4 to 5.0, the surface charge was reversed from negative to positive, and the zeta potential rose to 2.4 mV at pH 6.8 and 5.2 mV at pH 5.0. The charge reversal can be attributed to ionization of the amide groups from PEOz in the outer layer, inducing partial charge neutralization. Taken together, the surface charge of BP\@PDA-PEOz and BP-DOX\@PDA-PEOz are positive while the pH values decreased from 7.4 to 6.8 and 5.0. {#F4} *In vitro* Photothermal Effects ------------------------------- To clarify the photothermal performance of the prepared co-delivery platform, the temperature variations were tested under 808 nm laser irradiation for 10 min. As displayed in [Figure 5A](#F5){ref-type="fig"}, the temperatures of bare BP NSs, BP\@PDA and BP\@PDA-PEOz solutions (0.1 mg mL^-1^) all soar compared with that of distilled water under identical irradiation conditions. The photothermal efficiency of BP\@PDA (Δ*T* = 24.1°C) exceeded that of bare BP (Δ*T* = 18.1°C), and the temperature of BP\@PDA-PEOz was elevated by 22.9°C. PDA coating may be responsible for the augmented photothermal response of BP\@PDA, accompanied by considerable photothermal conversion efficiency. Furthermore, the photothermal properties of PDA-coated BP NSs were both concentration- and laser power-dependent ([Figure 5B,C](#F5){ref-type="fig"}). After five cycles of NIR laser irradiation, the temperature no longer changed evidently ([Figure 5D](#F5){ref-type="fig"}), so the sample was highly photostable. Moreover, the photostability of BP\@PDA-PEOz NSs surpassed that bare BP NSs, without obviously losing photothermal conversion efficiency in 1 week ([Supplementary Figure S3](#SM1){ref-type="supplementary-material"}). Hence, PDA coating boosted the photothermal performance of BP NSs, and rendered them suitable for PTT due to high photostability and photothermal conversion efficiency. {#F5} *In vitro* pH- and Photo-Responsive Drug Release Profiles --------------------------------------------------------- The sustained and controlled DOX release profiles of BP-DOX\@PDA-PEOz-BTZ were tested at pH 7.4 for simulating normal physiological microenvironment, pH 6.8 for simulating tumor extracellular microenvironment and pH 5.0 for simulating the acidic microenvironment of tumor endosome/lysosome, in the presence or absence of NIR laser irradiation. At pH 5.0, nearly 30% of DOX was released from BP-DOX\@PDA-PEOz-BTZ within 48 h, whereas only 11% of DOX was released at pH 7.4 ([Figure 6A](#F6){ref-type="fig"}), which may be ascribed to the pH sensitivity of PEOz coating after tertiary amide groups along the PEOz chain were ionized at a pH value lower than its pKa ([@B23]; [@B37]). The positive charges on the nitrogen atoms of PEOz main chains may result in electrostatic repulsion, which loosened the outer shell in the slightly acidic tumor cell microenvironment ([@B37]) and accelerated the release of inner hydrophobic anticancer drugs into tumor tissues while reducing that into the normal blood circulation. As a result, the anticancer effect was increased, and the side effects of common anticancer drugs were relieved ([@B11]; [@B54]; [@B37]). {#F6} Moreover, the photo-responsive drug release behaviors were studied. After 808 nm laser irradiation (1.0 W cm^-2^, 6 min for each pulse), the temperature of BP-DOX\@PDA-PEOz-BTZ increased gradually at pH 5.0, which significantly raised the cumulative DOX release amount, reaching above 40% after irradiation four times. Presumably, the PDA layer decomposed and released the loaded drug after NIR laser irradiation. In the meantime, BP decomposed gradually due to NIR exposure, further inducing drug release ([@B49]). Furthermore, we investigated the drug release behaviors of BTZ from BP-DOX\@PDA-PEOz-BTZ NSs at different pH values with or without IR irradiation ([Figure 6B](#F6){ref-type="fig"}). Merely approximately 25% of BTZ was released at pH 7.4 after 24 h. At pH 6.8 and 5.0, BTZ release was significantly accelerated. After NIR laser irradiation for 6 min, about 95% of BTZ was released from DOX-loaded BP\@PDA-PEOz-BTZ within 48 h at pH 5.0. In other words, the pH sensitivity of catechol-BTZ bond contributed to BTZ accumulation at tumor sites, so the treatment outcomes were improved. The drug BTZ was loaded onto the surface of nanoplatform through the reversible covalent bond between catechol and phenylboronic acid. However, the drug DOX was absorbed onto the corrugated surface of BP NSs by non-covalent bond. And the DOX-loaded BP NSs was covered by the PDA. That is, the drug DOX is at the interior of the nanoplatform and BTZ is at exterior. Therefore, the release rate for DOX is in general much lower compared to the release rate of BTZ. Similar results were reported by our previous research ([@B23]). This release behavior of BTZ highlights the pH sensitivity of the catechol-BTZ bond contributing to the accumulation of BTZ at the tumor sites. The release pattern was very important for better tumor killing effect, as it reduces the drugs leakage during the circulation in blood and increases the drugs enrichment in tumor sites or endosomes. Overall, the pH-sensitive drug release triggered by NIR laser irradiation markedly enhanced the antitumor efficacy and minimized the side effect. Cellular Uptake of NSs ---------------------- The uptake of BP-DOX\@PDA-PEG and BP-DOX\@PDA-PEOz NSs by MCF-7 cells in the weakly acidic tumor microenvironment (pH 7.4 and 6.8) was observed by confocal laser scanning microscopy. After treatment with different NSs for 4 h, the intracellular fluorescent intensities of BP-DOX\@PDA-PEG NSs at pH 6.8 and 7.4 were similar, but the intensity of BP-DOX\@PDA-PEOz NSs at pH 6.8 significantly exceeded that at pH 7.4 ([Figure 7](#F7){ref-type="fig"}). The results validated the hypothesis that PEOz promoted the cellular uptake of DOX compared with PEGylated copolymer did in the mildly acidic endosomal/lysosomal and tumor extracellular environment, which can be attributed to the charge reversal of PEOz after tertiary amide groups along the PEOz chain were ionized ([@B5]; [@B37]; [@B50]). {#F7} Cell Viability -------------- The *in vitro* cytotoxicities of drug-free BP\@PDA-PEG and BP\@PDA-PEOz as well as DOX-loaded BP\@PDA-PEG, BP\@PDA-PEOz and BP\@PDA-PEOz-BTZ NSs were evaluated by the MTT assay. Drug-free BP\@PDA-PEG and BP\@PDA-PEOz NSs were also tested to eliminate the potential toxic characteristics of drug delivery capsule. Given that all drug-free BP-based NSs displayed negligible cytotoxicities against MCF-7 cells ([Supplementary Figure S4](#SM1){ref-type="supplementary-material"}), they were highly biocompatible. The photothermal cytotoxicities of different NSs were also evaluated by the MTT assay. Cell growth was barely affected by individual NIR laser irradiation, whereas BP\@PDA-PEG and BP\@PDA-PEOz NSs exerted concentration-dependent photothermal effects ([Figure 8A](#F8){ref-type="fig"}). Moreover, over 80% of MCF-7 cells were killed in the presence of 50 μg mL^-1^ BP\@PDA-PEOz under 808 nm laser irradiation. Taken together, BP\@PDA-PEOz nanocapsule may be an effective PTT agent with desirable biocompatibility. {#F8} Furthermore, the BP-DOX\@PDA system exhibited pH-dependent cytotoxicity after PEOz modification ([Figure 8B](#F8){ref-type="fig"}). BP-DOX\@PDA-PEOz was significantly more toxic at pH 6.8 than at pH 7.4, but BP-DOX\@PDA-PEG had almost the same inhibitory effects at pH 7.4 and 6.8, potentially allowing selective killing of cancer cells that were more acidic than normal cells/tissues *in vivo*. Collectively, PEOz modification was conducive to cellular uptake and pH-sensitive drug release in the mildly acidic tumor microenvironment, thereby promoting tumor inhibition and alleviating side effects during chemotherapy. In addition, MCF-7 cells were treated by DOX, BTZ, DOX + BTZ (1:1) and drug-loaded NSs with the DOX concentrations of 0.25, 1, 2.5, and 5 μg/mL for 24 or 48 h ([Figure 8C,D](#F8){ref-type="fig"}). First, the cytotoxicities of free DOX, BTZ, DTX + BTZ (1:1) and drug-loaded NSs were time- and dose-dependent. Second, compared to individually administered DOX or BTZ, directly co-administering DTX + BTZ was more cytotoxic, as suggested by the raised inhibition rate. Furthermore, the survival rate of cells treated with DOX-loaded BP\@PDA-PEOz NSs was apparently lower than that of the DOX-loaded BP\@PDA-PEG NSs group after incubation for 24 or 48 h. Therefore, PEOz was more conducive to long-term drug circulation *in vivo* than PEG, extending the half-lives of drugs and enriching pH-sensitive drugs at tumor sites by recognizing the acidic tumor microenvironment. Most importantly, the group treated by DOX-loaded BP\@PDA-PEOz-BTZ in combination with 808 nm laser irradiation (1.0 W cm^-2^) had the lowest survival rate after 48 h of incubation, demonstrating that chemotherapy plus PTT exerted the strongest cytotoxic effects. In short, the antitumor effects of drug were effectively boosted with this pH-sensitive release pattern triggered by NIR laser irradiation, accompanied by minimal side effects. *In vivo* IR Thermal Images --------------------------- The photothermal efficacy of this BP-based nanoplatform was further studied by acquiring IR thermal images ([Figure 9A](#F9){ref-type="fig"}). 24 h after intravenous injection of BP\@PDA-PEG, BP\@PDA-PEOz, and BP\@PDA-PEOz-BTZ, the tumor sites were irradiated for 5 min by 808 nm laser (1.5 W cm^-2^), and the temperatures in tumor site were measured every 15 s. The tumor surface temperatures significantly increased after NIR irradiation, which can be attributed to the remarkable photothermal effects of PDA coating and BP. The tumor temperatures of BP\@PDA-PEOz group and BP\@PDA-PEOz-BTZ group rapidly increased to 51.4 and 50.1°C, respectively, which were sufficiently high for effective ablation. The *in vivo* local tumor temperatures of BP\@PDA-PEOz and BP\@PDA-PEOz-BTZ groups changed more obviously than that of the BP\@PDA-PEG group did, because PEOz underwent charge reversal from negative to positive upon tertiary amide group ionization along the PEOz chain in tumor tissue with a low pH, ultimately benefiting the uptake by cancer cells. On the contrary, the temperature of PBS-treated tumor did not rise evidently after irradiation under identical conditions, so cancer cells remained intact. Based on long-term circulation *in vivo*, the PEOz-modified, BP-based NS drug delivery platform was photothermally active, being capable of pH-triggered targeting of tumor tissues. The quantified tumor temperature variations are presented in [Figure 9B](#F9){ref-type="fig"}. {#F9} Conclusion ========== In summary, we have successfully developed an effective pH-responsive and dual drug co-delivery nanoplatform (DOX-loaded BP\@PDA-PEOz-BTZ) for combined chemotherapy and PTT. Not only did we show PDA coating can enhance both biostability and photothermal activity of the BP NS, we also demonstrated that PEOz conjugation can improve the targeted, long circulation *in vivo* as well as pH- and photo-responsive drug release, indicating that PEOz is an excellent substitute for PEG. Given the high drug encapsulation efficiency, the strong cellular uptake and cytotoxicity, together with the photo-responsive, rapid drug release triggered by low pH, our versatile PDA- and PEOz-modified, BP-based dual drug co-delivery nanoplatform has great potentials for synergistic cancer treatment. Author Contributions ==================== ZP designed the research project. NG, CX, and LF had full controlled the experiments, data analysis, and preparation of article. HW, XZ, and LM were involved in planning the analysis and drafting the article. The final draft article was approved by all the authors. Conflict of Interest Statement ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. **Funding.** This work was supported by National Natural Science Foundation of China (61671308 and 81701819) and the Science and Technology Innovation Commission of Shenzhen (JCYJ20170817094728456 and JCYJ20170302153341980). Supplementary Material ====================== The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fphar.2019.00270/full#supplementary-material> ###### Click here for additional data file. [^1]: Edited by: Jianxun Ding, Changchun Institute of Applied Chemistry (CAS), China [^2]: Reviewed by: Kun Zhou, Boston Children's Hospital, United States; Houliang Tang, Southern Methodist University, United States; Ruinan Yang, Pharmaceutical Product Development, United States [^3]: This article was submitted to Experimental Pharmacology and Drug Discovery, a section of the journal Frontiers in Pharmacology

Introduction {#s01} ============ The presynaptic active zone (AZ) is a highly specialized subcellular compartment, where neurotransmitter-containing synaptic vesicles dock within a few tens of nanometers from voltage-gated calcium channels (Ca~V~) and are prepared to fuse with the plasma membrane in a Ca^2+^-dependent manner. Synaptic signal transduction is coordinated by protein complexes at the pre- and post-synaptic sites. In this framework, the presynaptic release machinery is regulated by cytomatrix at the AZ (CAZ) proteins, including Munc13, RIM, Bassoon, CAST (also named ELKS2α or ERC2), and ELKS (ELKS1α or ERC1; [@bib16]; [@bib48]; [@bib39]) that compose the presynaptic density ([@bib18]). These proteins are believed to perform a variety of roles such as formation and maintenance of synapses, tethering and docking synaptic vesicles at AZ release sites, and recruitment of Ca~V~ channels to the AZ. In addition to studies investigating the functional properties of the individual CAZ proteins in various synapses ([@bib48]; [@bib19]), work using combinatorial deletion of proteins, such as ELKS (CAST/ELKS) and RIM, or RIM and RIM-BP, has shown a strong reduction of docked vesicles or presynaptic dense projections---classical morphological markers of the AZ ([@bib1]; [@bib51]). Elaborate electron-dense structures are found at invertebrate T-bar synapses and ribbon synapses of the vertebrate eye and ear ([@bib57]; [@bib54]; [@bib35]; [@bib41]). These so-called "synaptic ribbons" are mainly composed of RIBEYE and CAZ proteins, including Bassoon, Piccolo, RIM, and CAST ([@bib43]; [@bib11]; [@bib25]; [@bib39]; [@bib35]; [@bib22]). In this framework, genetic deletion of RIBEYE eliminated the ribbon and disrupted both fast and sustained neurotransmitter release from bipolar cells (BCs; [@bib35]). In contrast, in auditory hair cells, ribbon loss upon RIBEYE deletion led to elaborate developmental compensation that resulted in the formation of multiple ribbonless AZs at each synaptic contact with spiral ganglion neurons that sustained basic release rates ([@bib4]; [@bib22]). Bassoon, another multi-domain CAZ protein, exerts an essential role in anchoring the synaptic ribbon at the AZ membrane, and loss of Bassoon results in impaired transmission at retinal and cochlear synapses ([@bib12]; [@bib25]; [@bib5]). Deletion of RIM2 reduced Ca^2+^ influx and affected release from rod terminals without changing rod ribbon synapse anatomy ([@bib15]; [@bib33]). At hair cell synapses, RIM2 disruption reduced the number of presynaptic Ca^2+^ channels and tethered synaptic vesicles at the AZ membrane. Conversely, deletion of CAST, a molecular scaffold and protein interaction hub, reduced rod photoreceptor AZ size, ultimately leading to impaired electroretinogram (ERG) responses and attenuated contrast sensitivity ([@bib50]). While the presynaptic function of CAST has been analyzed in various preparations over recent years ([@bib49]; [@bib23]; [@bib50]; [@bib20]; [@bib30]), other work on invertebrate CAST/ELKS homologues in *Caenorhabditis elegans* (ELKS) and *Drosophila melanogaster* (bruchpilot) suggest additional roles in synapse formation and the promotion of AZ assembly, respectively ([@bib8]; [@bib27]). In contrast, the role of presynaptic ELKS remains largely enigmatic, mainly owing to the fact that in vertebrates ELKS isoforms are ubiquitously expressed and constitutive ELKS knockout (KO) mice are embryonic-lethal ([@bib9]; [@bib31]). ELKS is characterized by higher solubility than CAST in spite of its 92% sequence similarity with the neuron-specific CAST ([@bib40]; Fig. S1). Since ELKS is also found at retinal ribbon synapses ([@bib10]), we hypothesized that ELKS may---at least partially---compensate for the loss of CAST from the synapse, thereby masking the individual contributions of CAST/ELKS family proteins in synaptic processes. To elucidate the roles of ELKS and CAST in retinal photoreceptor AZ formation and maintenance as well as in synaptic processing of visual information, the present study established ELKS conditional KO (cKO) and CAST KO/ELKS cKO (double KO \[dKO\]) mice using a Crx-Cre line ([@bib38]) for retina-specific deletion of ELKS. Comprehensive functional and anatomical analysis revealed that CAST and ELKS have both redundant and unique roles at the rod photoreceptor ribbon synapse. Remarkably, the acute depletion of ELKS from the mature retinal network induced a concomitant loss of ribbon synapses and photoreceptors. Therefore, we conclude that organization of release machinery with CAST and ELKS plays functional and anatomical roles in photoreceptor neurotransmission, the disruption of which may cause progressive blindness due to synaptic and neural remodeling. Results {#s02} ======= Combined ablation of ELKS and CAST from retinal photoreceptors causes impaired visual responses {#s03} ----------------------------------------------------------------------------------------------- First, we generated an ELKS cKO crossing *ELKS^flox/flox^* mice with Crx-Cre mice for targeted deletion of exon 11 in the retina ([Fig. 1 A](#fig1){ref-type="fig"} and Fig. S1; [@bib38]; [@bib13]). Deletion of exon 11 causes a frame shift in exons 12 and 13, resulting in the premature termination of all ELKS isoforms including α, β, γ, δ, and ε (Fig. S1 C; [@bib37]), and hence elimination of all functional ELKS protein from the retina. ELKS cKO mice showed normal development, while mice carrying *CAST^−/−^* and *ELKS^flox/flox^* alleles as well as Crx-Cre (CAST/ELKS dKO) exhibited slightly smaller body weight and eye size (Fig. S2 A and Table S1). In accordance with a previous study ([@bib10]), immunohistochemical analysis of ELKS expression in the retina confirmed the presence of ELKS in the outer plexiform layer (OPL), where rod and cone terminals form synapses with BCs and horizontal cells (HCs). In contrast, in the OPL of ELKS cKO mice, only a small number of nonspecific speckles could be observed ([Fig. 1 C](#fig1){ref-type="fig"}). Similarly, the inner plexiform layer (IPL), containing the synapses formed by BC, amacrine, and ganglion cells, also showed ELKS immunofluorescence in control retinae that was absent in ELKS cKO mice ([Fig. 1 C](#fig1){ref-type="fig"}). Immunoblotting of synapse-related proteins from retinal lysates confirmed the targeted deletion of ELKS and CAST in the respective individual KOs and the removal of both proteins from the dKO ([Fig. 1, D and E](#fig1){ref-type="fig"}). As previously reported ([@bib50]), we found a threefold, potentially compensatory up-regulation of ELKS in the CAST KO mice. Conversely, CAST expression increased in ELKS cKO mice ([Fig. 1, D and E](#fig1){ref-type="fig"}; and Fig. S2, B--E), indicating mutual compensatory up-regulation of both proteins. Interestingly, expression of RIM1/2, which directly interacts with CAST/ELKS ([@bib40]; [@bib52]; [@bib9]), was decreased in the ELKS cKO and dKO mice, but not in the CAST KO mice ([Fig. 1, D and E](#fig1){ref-type="fig"}). We speculate that the reduction in RIM1/2 protein levels reflects the loss of an ELKS-RIM interaction, because the observed reduction of RIM1/2 was comparable between ELKS cKO and dKO mice ([Fig. 1 E](#fig1){ref-type="fig"}) despite the prominent up-regulation of CAST in ELKS cKO mice. {#fig1} To characterize visual function in mice of the different genotypes, we measured the ERG under scotopic conditions ([Fig. 1, F--I](#fig1){ref-type="fig"}). In these experiments, dark-adapted mice were exposed to dim light flashes of increasing intensity to trigger progressively stronger retinal responses. Here, the a-wave, indicating photoreceptor transduction, was unaltered across all genotypes. In contrast, the scotopic b-wave, reflecting rod photoreceptor transmission, was reduced in all mutants, from near normal in ELKS cKO to about half in the CAST KO mice, and nearly abolished in dKO. The oscillatory potentials, primarily reflecting the activity of the inner retina, were reduced in CAST KO and even further in dKO mice. Hence, the reduction in scotopic ERG signals indicates impaired synaptic transmission in the OPL and IPL in response to loss of CAST, a phenotype that is further aggravated by additional deletion of ELKS. Loss of photoreceptor ribbon synapses and enhanced formation of ectopic ribbon synapses {#s04} --------------------------------------------------------------------------------------- Next, we analyzed the formation of retinal layers. The thickness and the density of nuclei layers (outer nuclear layer \[ONL\] and inner nuclear layer \[INL\]) were not significantly different across the genotypes in the adult retina (∼10 wk old; Fig. S2, F and G). In contrast, both CAST KO and dKO showed a reduction in OPL thickness (Fig. S2, F and G). In addition, dKO led to significant reduction in IPL thickness (Fig. S2, F and G), while the thicknesses of OPL and IPL were normal in the ELKS cKO. The reduced thickness of the synaptic layers suggests extensive remodeling of retinal synapses without major cell loss in the absence of CAST and ELKS. The distribution of presynaptic proteins (i.e., RIBEYE and vGluT1) and synaptic connectivity with BCs and HCs was assessed by immunohistochemistry using cell type--specific markers (protein kinase C and calbindin, respectively; [Fig. 2, A and B](#fig2){ref-type="fig"}). Both CAST KO and dKO mice exhibited an increase in the formation of ectopic ribbon synapses in the ONL, which were accompanied by abnormal sprouting of BCs and HCs dendritic processes ([Fig. 2, A and B](#fig2){ref-type="fig"}). Here, we quantified ectopically formed synapses in the ONL by measuring the displacement of RIBEYE-positive spots from the OPL to the ONL ([Fig. 2 C](#fig2){ref-type="fig"} and Fig. S3, A and B). Immunolabeled ribbons were isolated from background noise by their fluorescent intensity and the spot size (see Materials and methods). While ∼10% of synapses mislocalized to the ONL in the CAST KO, a staggering ∼40% of synapses formed ectopically in the dKO ([Fig. 2 C](#fig2){ref-type="fig"}). Similar to the ribbon marker, the presynaptic terminal marker vGluT1 also exhibited ectopic synapse formation in CAST KO and dKO mice (Fig. S3 C). Age-related synaptic remodeling was apparent from the increase in ectopic synapse probability with advancing age in the dKO: ectopic synapses were identified as early as 5 wk of age and further increased up to \>40 wk ([Fig. 2 D](#fig2){ref-type="fig"} and Fig. S3 D). A similar observation was made in CAST KO mice, but to a lesser extent ([Fig. 2 D](#fig2){ref-type="fig"} and Fig. S3 D). {#fig2} We also calculated the density of RIBEYE spots within the OPL and ONL ([Fig. 3, A--C](#fig3){ref-type="fig"}). The densities of such spots in ONL plus OPL were similar for control and ELKS cKO mice. However, in the CAST KO and dKO mice, the density of RIBEYE spots decreased by ∼30% ([Fig. 3 A](#fig3){ref-type="fig"}). Inspection of the ONL only showed a several-fold increase in density of RIBEYE spots in the dKO, while the CAST KO was less strongly affected ([Fig. 3 B](#fig3){ref-type="fig"}). Finally, the density of RIBEYE spots in the OPL of CAST KO and dKO mice was greatly reduced to ∼50% and 40% compared with controls, respectively ([Fig. 3 C](#fig3){ref-type="fig"}). Since RIBEYE spot detection was set to the size threshold, the resultant spot size was comparable between genotypes (control = 0.48 ± 0.045 µm^2^, ELKS cKO = 0.53 ± 0.035, CAST KO = 0.45 ± 0.022, and dKO = 0.46 ± 0.018). However, a previous study ([@bib50]) reported that loss of CAST reduces rod ribbon size. Hence, due to threshold exclusion of smaller ribbons in the mutants, our current analysis of RIBEYE immunofluorescence spots did not reveal a reduction in ribbon size (see Roles of CAST and ELKS in the formation of rod triads and [@bib50]) and likely underestimated the density. Therefore, we also performed conventional EM to examine rod ribbon density in the OPL at higher resolution. The density of ribbon-bearing rod terminals in the OPL was drastically reduced in both CAST KO and dKO to a similar extent, while it was unchanged in the ELKS cKO ([Fig. 3, D and E](#fig3){ref-type="fig"}). Furthermore, we investigated the synaptic vesicle distribution at rod ribbon synapses taken from randomly selected ultrathin sections. The density of ribbon-associated synaptic vesicles was slightly decreased in dKO, but did not differ significantly from control or CAST KO ([Fig. 3, F and G](#fig3){ref-type="fig"}). The Pearson's correlation coefficients between ribbon length and vesicle number were ∼0.4 for all genotypes (control = 0.41, ELKS cKO = 0.40, CAST KO = 0.46, and dKO = 0.43), indicating a scaled reduction of vesicle tethering capacity with decreasing ribbon size. Moreover, we analyzed the densities of freely floating synaptic vesicles in the cytoplasm of randomly selected rod terminals that possessed compartments of triads including ribbon synapses, and found them to be comparable between all genotypes (control = 130.5 ± 6.7 /µm^2^ \[*n* = 5\], ELKS cKO = 144.5 ± 8.9 \[*n* = 6\], CAST KO = 155.4 ± 6.2 \[*n* = 5\], and dKO = 153.9 ± 6.2 \[*n* = 5\]). {#fig3} Roles of CAST and ELKS in the formation of rod triads {#s05} ----------------------------------------------------- In a previous study, CAST was shown to promote AZ size, light responses, and contrast sensitivity ([@bib50]). The present study further assessed the roles of CAST and ELKS in photoreceptor synapse formation and maintenance by investigating the ultrastructure of rod terminals, which form so-called "triads" between photoreceptor ribbon synapses and BC and HC processes ("tips"), using focused ion beam milling combined with scanning EM (FIB-SEM; [@bib26]). Three terminals in the OPL from three different animals of each mouse strain were randomly chosen for 3D reconstruction ([Fig. 4](#fig4){ref-type="fig"}, Fig. S4, and Tables S2 and S3). Rod terminals possessed single mitochondria, and volume of neither individual terminals nor mitochondria was significantly different between the genotypes (Tables S2 and S3). While we found BC tips to be comparable across all genotypes ([Fig. 4 D](#fig4){ref-type="fig"} and Fig. S4), the majority of HCs only exhibited a single tip in CAST KO and dKO mice. In contrast, control and ELKS cKO mice always possessed at least two tips ([Fig. 4 E](#fig4){ref-type="fig"}). Intriguingly, 3D reconstructions revealed that the single HC tip in the CAST KO and dKO mice branched before contacting the synaptic ribbon, forming a structure similar to that of the two HC tips that normally target the ribbon ([Fig. 4 B](#fig4){ref-type="fig"} and Fig. S4). Although the individual tip sizes (surface area and volume) remained unaltered across all genotypes (Tables S2 and S3 and Fig. S4), the loss of the second HC tip from the triad of CAST KO and dKO retinae led to a significant reduction of the accumulated surface area and volume of HC tips ([Fig. 4 E](#fig4){ref-type="fig"}, Fig. S4, and Tables S2 and S3). In CAST KO and dKO mice, overall HC counts, as measured by calbindin staining, remained unaltered (Table S5). However, 3D reconstructions of HC branching from a tip within a triad revealed fewer branches with filopodia-like structures that were lacking tips (Fig. S4 E). {#fig4} Photoreceptor ribbons have been proposed to play an important role in replenishing synaptic vesicles to the AZ ([@bib21]; [@bib35]). Hence, we next investigated synaptic ribbon morphology from 3D reconstructed images. In these experiments, reconstructed ribbons showed plate-like structures with a horseshoe-shaped appearance in the control and ELKS cKO ([Fig. 4 C](#fig4){ref-type="fig"}). However, compared with control, ribbon lengths were significantly decreased in all deletion mutants, with the highest and comparable severity in CAST KO and dKO ([Fig. 4 G](#fig4){ref-type="fig"} and Table S4). Moreover, we measured the shortest distance from the ribbon to the surface of BC tips, and found it to be significantly longer in the dKO compared with control ([Fig. 4 H](#fig4){ref-type="fig"}). The CAST KO fell in between, showing no statistical difference in ribbon--BC tip separation for the CAST KO versus dKO, nor the CAST KO versus control. Overall, the 3D reconstructions of FIB-SEM data revealed impaired triad formation that can be summarized as follows for the CAST KO and dKO: (a) reduced HC branching, (b) decreased surface area of the HC processes in the rod triads, (c) diminution of ribbon size (and to a lesser extent in ELKS cKO), and (d) increased width of the synaptic cleft between rod ribbons and BC tips. In other words, CAST and ELKS codetermine the structure of the rod ribbon synapses and synaptic connectivity with HCs and BCs ([Fig. 4 I](#fig4){ref-type="fig"}), with CAST exerting a more prominent role than ELKS. Combining these findings with conventional EM, the rod AZs of CAST KO and dKO mice are similar in many ways: both show truncated ribbons and comparable vesicle density. The increased distance from the ribbon to the BC tips might not suffice to explain the severely attenuated scotopic b-waves and the enhanced formation of ectopic synapses in the dKO. Although ELKS itself seems to contribute in a minor way to the presynaptic morphology, it may affect AZ function. Therefore, we next probed rod Ca~V~ channels to assess if their function behaves differently in the CAST KO and dKO. Roles of CAST and ELKS in the voltage-dependent calcium influx at rod terminals {#s06} ------------------------------------------------------------------------------- L-type Ca~V~1.4 channels localize to rod ribbon synapses ([@bib32]; [@bib55]), and previous work suggests that its pore-forming α1 subunit ([@bib50]) and α2δ4 auxiliary subunit ([@bib53]) traffic to the downsized ribbons formed by CAST KO rods. Therefore, we performed whole-cell voltage-clamp recordings from rods to test the behavior of the Ca~V~1.4 channels in acute retinal slices from adult control, CAST KO, and dKO mice (4 mo of age). Calcium currents (I~Ca~) were recorded from rod somata located on the ONL border to the OPL, where the synaptic ribbon is contained in the soma compartment ([Fig. 5 A](#fig5){ref-type="fig"}; [@bib15]). While the example control recording showed an inward I~Ca~ with a peak amplitude of approximately −9 pA when the voltage was ramped from −70 to +50 mV, the response acquired from a CAST KO was essentially flat ([Fig. 5, B and C](#fig5){ref-type="fig"}). To illustrate the excitability of cells without I~Ca~, the cells were also given steps from −70 to −90 mV to activate the hyperpolarizing inward current (I~h~) carried by hyperpolarizatoin-activated cyclic nucleotide-gated channels ([@bib2]). The hyperpolarizing voltage step followed by a depolarizing voltage ramp activated an I~h~ current in rods ([Fig. 5 D](#fig5){ref-type="fig"}), but I~Ca~ was seldom detected in CAST KO and dKO ([Fig. 5 E](#fig5){ref-type="fig"}). In addition, rod somata located in the ONL (about a single row of soma away from the OPL) were also patched, thus allowing the visualization of the rod's short axon linked to the small terminal (Fig. S5 A). Again, the I~Ca~ was either strongly reduced or almost absent in CAST KO and dKO mice, while I~h~ currents as well as an outward current related to the inner segment ([@bib3]; [@bib15]) were routinely observed (Fig. S5, B and C). The summarized data show a drastically reduced I~Ca~ in CAST KO and dKO relative to control, while the KO mice were not significantly different from one another ([Fig. 5 F](#fig5){ref-type="fig"} and Fig. S5 G). In addition, we estimated the voltage where the I~Ca~ reaches half maximal amplitude on the voltage ramp (V~1/2~) in recordings with measurable I~Ca~, which can indicate modulation of Ca~V~ gating. The V~1/2~ values were similar across the genotypes: control = −29 ± 1.5 mV (*n* = 9 cells), CAST KO = −30 ± 1.4 mV (*n* = 4 cells), and dKO = −31 ± 1.9 mV (*n* = 7 cells). However, it should be noted that only half of CAST KO (four of eight cells) and dKO (7 of 16 cells) had sufficient I~Ca~ amplitudes (\>0.5 pA) to estimate the V~1/2~. These results indicate that CAST plays an essential role in maintaining Ca^2+^ entry into rod terminals. {#fig5} Acute deletion of ELKS induces photoreceptor remodeling and elimination of ribbon synapses {#s07} ------------------------------------------------------------------------------------------ ELKS is ubiquitously expressed, including in nonneural tissues, and constitutive ELKS KO is embryonic-lethal ([@bib9]; [@bib31]). Therefore, developmental compensation by other presynaptic terminal proteins might mask the ELKS KO phenotype in our preparation. To address this issue, we next investigated ELKS function in photoreceptors after intact synapse formation and maturation. In these experiments, acute ELKS depletion was achieved by intravitreal injection of adeno-associated virus serotype 5 (AAV5) encoding Cre (AAV5-CAGGS-nCre) into ∼5-wk-old *ELKS^fl/fl^* animals. At 3 wk after injection, robust Cre recombinase expression was observed in photoreceptors (Cre-positive; [Fig. 6](#fig6){ref-type="fig"} and Fig. S6). Acute ELKS cKO induced a reduction in ONL thickness, suggesting substantial photoreceptor loss ([Fig. 6](#fig6){ref-type="fig"}). Ribbon synapse density in the OPL (as assessed by RIBEYE immunolabeling) decreased by 64% in the Cre-positive region compared with the proximal Cre-negative regions ([Fig. 6, B, D, and F](#fig6){ref-type="fig"}). The acute depletion of ELKS in CAST KO retinae led to partial ablation of the ONL, as illustrated by an almost complete loss of ribbon synapses, i.e., 89% ([Fig. 6 D](#fig6){ref-type="fig"} and Fig. S5). Moreover, we have confirmed that acute ablation of ELKS by AAV-Cre expression caused loss of ELKS protein at the OPL in *ELKS^fl/fl^* mice ([Fig. 7 A](#fig7){ref-type="fig"}), which difference was even more drastic in *CAST KO ELKS^fl/fl^* mice ([Fig. 6 G](#fig6){ref-type="fig"}). Because ELKS expression was up-regulated in the CAST KO retina ([Fig. 1 D](#fig1){ref-type="fig"}), acute dKO after synapse formation might induce photoreceptor degeneration more severely than acute ELKS cKO. To establish whether these effects were indeed a direct result of ELKS elimination from photoreceptors and exclude adverse effects due to the virus application, we also injected AAV5-CAGGS-nCre into *ELKS ^fl/fl^ x Crx-Cre^+/−^* (ELKS cKO) mice. Importantly, acute Cre expression had no quantifiable effects on ELKS KO photoreceptors, ONL thickness, or RIBEYE density ([Fig. 6, E and F](#fig6){ref-type="fig"}; and Fig. S6 B). Consistent with these findings, injection of an AAV5 encoding the fluorescent transduction marker Venus showed no effect on ONL thickness or RIBEYE density ([Fig. 6, E and F](#fig6){ref-type="fig"}; and Fig. S6, C--E). ![**Acute, AAV-mediated, ELKS depletion in photoreceptors induces photoreceptor degeneration. (A--D)** Acute induction of ELKS cKO and dKO via AAV5-CAGGS-nCre. Subretinal injection of AAV5 produces mosaic Cre expression in the retina (Cre\[+\], indicated by arrowheads in A and C). (B and D) Immunolabeling of RIBEYE (green) in the OPL is reduced in the Cre-positive area (expansion of area indicated by white rectangle in A and C), when directly compared with adjacent Cre-negative regions (Cre\[−\] in B and D). Bars, 200 µm in A and C, 50 µm in B and D. **(E and F)** Quantitative analysis of ONL thickness and RIBEYE density. In *ELKS^fl/fl^* and *CAST KO; ELKS^fl/fl^* mice, ONL thickness (E) and RIBEYE density (F) are significantly decreased in Cre-positive regions, although Venus expression does not affect these parameters. Importantly, acute Cre expression in *ELKS^fl/fl^ x Crx-Cre^+/−^* (ELKS cKO) mice affects neither ONL thickness nor RIBEYE density, indicating a CAST/ELKS-dependent mechanism as the underlying cause; mean ± SEM. ONL: *ELKS^fl/fl^*; Cre (*n* = 5 mice), Venus (*n* = 2), *ELKS^fl/fl^ x Crx-Cre^+/−^*; Cre (*n* = 3), Venus (*n* = 3), *CAST KO; ELKS^fl/fl^*; Cre (*n* = 3), and Venus (*n* = 3). RIBEYE: *ELKS^fl/fl^*; Cre (*n* = 5), Venus (*n* = 2), *ELKS^fl/fl^ x Crx-Cre^+/−^*; Cre (*n* = 3), Venus (*n* = 2), *CAST KO; ELKS^fl/fl^*; Cre (*n* = 4), and Venus (*n* = 3). \*, P \< 0.05; \*\*, P \< 0.01 (two-sided Student's *t* test). **(G)** Ablation of ELKS in the OPL in *CAST KO; ELKS^fl/fl^* Cre (+) retina (arrows). Bar, 50 µm.](JCB_201704076_Fig6){#fig6} ![**Acute depletion of ELKS in retina induces apoptotic cell death in photoreceptors, which can be prevented by simultaneous expression of CAST. (A)** Acute induction of ELKS cKO by injection of AAV5-CAGGS-nCre in *ELKS^fl/fl^* retina (Cre^+^) showed multiple TUNEL positive cells in the ONL, while no TUNEL-positive cells were observed in Cre negative retinae (Cre^−^). ELKS expression in the OPL was ablated in *ELKS^fl/fl^* Cre^+^ retina (arrows). **(B--E)** Acute depletion of ELKS by AAV5-CAGGS-nCre in *CAST KO; ELKS^fl/fl^* mice were effectively rescued by simultaneous injection of AAV5-CAGGS-CAST-IRES-Venus. The ONL thickness was not significantly decreased (C; Cre^−^ \[*n* = 18\], Cre^+^ CAST-Venus \[*n* = 19\]). In contrast, the simultaneous injection of AAV5-IRES-Venus failed to rescue the significant reduction of ONL thickness (E; Cre^−^ \[*n* = 10\], Cre^+^ Venus \[*n* = 16\]); mean ± SEM; \*, P \< 0.05 (two-sided Student's *t* test). Bars, 50 µm in A, 20 µm in B and D. **(F and G)** Simplified schematic model summarizing the findings from photoreceptors by ELKS ablation. Acute ELKS depletion after maturation of retinal network induces degeneration of photoreceptors (light green and white cells), which was further enhanced in the CAST KO retina.](JCB_201704076_Fig7){#fig7} Finally, to further characterize photoreceptor degeneration, we performed TUNEL assays to detect apoptotic cells in retinal cryosections obtained from injected mice 2 wk after AAV5-Cre administration. Here, besides the TUNEL reagents, retinal serial cryosections were simultaneously subjected to anti-ELKS and anti-Cre antibodies. In contrast to the nontransduced control parts of these retinae, where we mostly detected TUNEL negative cells, clear TUNEL staining was observed in Cre expressing photoreceptors within the ONL, accompanied by a reduction of ELKS in the OPL ([Fig. 7 A](#fig7){ref-type="fig"}). In contrast, injection of AAV5-Cre into ELKS cKO did not generate a large proportion of TUNEL-positive cells in the ONL, as confirmed by the maintained ONL thickness (Fig. S6 F). Therefore, the reduction of the ONL in acute ELKS KO by AAV5-Cre injection into *ELKS^fl/fl^* eyes was likely caused by photoreceptor apoptosis. The reduction in the thickness of the ONL induced by AAV5-Cre could be prevented by the coinjection of AAV5-CAST-IRES-Venus, but not with AAV5-Venus in the *CAST KO ELKS^fl/fl^* retina ([Fig. 7, B--E](#fig7){ref-type="fig"}). In these experiments, the expression of the transduction marker Venus was substantially weaker in AAV5-CAST-IRES-Venus transduction. We speculate that this finding can be attributed to the construct design and less efficient translation of Venus due to the IRES motif. To clarify this point, we assessed the transduction efficiency of AAV-CAST-IRES-Venus in cultured neurons, in which CAST was strongly expressed in Venus-positive neurons (Fig. S6, F and G). The acute depletion of ELKS in retinae lacking CAST led to partial ablation of the ONL, as illustrated by an almost complete loss of ribbon synapses ([Fig. 6, D and F](#fig6){ref-type="fig"}). Hence, acute ablation of ELKS after synapse formation might induce photoreceptor degeneration more severely in the absence of CAST, indicating overlapping roles between ELKS and CAST, as suspected from the observed compensatory up-regulation of ELKS expression in CAST KO retinae ([Fig. 1 D](#fig1){ref-type="fig"} and [Fig. 6 G](#fig6){ref-type="fig"}). Discussion {#s08} ========== Photoreceptors transform light into membrane voltages, which in turn drives signal transmission to the inner retina. The molecular physiology of synaptic transmission at rod photoreceptors remains incompletely understood. In the current study, we analyzed structural and functional changes in the retina upon deletion of the highly homologous multi-domain proteins of the AZ, CAST, and ELKS. This study revealed that both CAST KO and CAST/ELKS dKO rod terminals exhibited altered triad ribbons and HC processes, drastically reduced calcium influx and comparable loss of synaptic ribbons from the OPL, and, finally, exhibited extensive synaptic remodeling involving ectopic synapses in the ONL. In addition, rod ribbon synapses of the dKO exhibited fewer ribbon-associated synaptic vesicles and a greater separation of the ribbon from the BC dendrite. Moreover, dKO mice exhibited the largest number of ectopic ribbon synapses in the ONL. Taken together, these additional deficits are likely to account for the stronger reduction in the scotopic ERG b-wave in dKO mice, compared with that observed in the CAST KO mice. In contrast to CAST, ablation of ELKS early in development did not significantly alter ERG responses or ribbon anatomy. However, ELKS deletion after synapse formation caused massive loss of ribbon synapses and photoreceptors, which was partially halted by CAST. In summary, our data indicate that CAST and ELKS are critical for visual processing by exhibiting both redundant and complementary functions in synapse formation, function and maintenance in retinal rod photoreceptors. A role for CAST/ELKS in presynaptic Ca^2+^ influx {#s09} ------------------------------------------------- Using whole-cell voltage-clamp recording from rod photoreceptors ([@bib15]; [@bib53]), which gives unprecedented access to a synaptic ribbon, we demonstrated that CAST elimination greatly reduces the Ca^2+^ influx via Ca~V~ channels. This finding is consistent with previous studies in which CAST deletion not only decreases ribbon lateral expansion but also reduces presynaptic Ca~V~1.4 abundance at rod AZ ([@bib50]; [@bib53]). Interestingly, RIM2, which interacts with CAST ([@bib52]) and is located at rod ribbons near Ca~V~1.4 channels, also promotes Ca^2+^ influx into mouse rods ([@bib15]). A role of CAST/ELKS in regulating presynaptic Ca^2+^ influx has also been reported in cultured hippocampal neurons. One previous study reported that action potential--driven changes in Ca^2+^ levels at the nerve terminals were lower than in control mice despite apparently normal expression of Ca~V~ channels ([@bib31]). In accord with this finding, another study reported a direct interaction of CAST with Ca~V~ α1 and β subunits, and further reported a CAST-dependent hyperpolarizing shift in the voltage dependence of presynaptic Ca^2+^ channels upon coexpression in a heterologous expression system ([@bib29]). Loss of CAST from rods greatly reduced the magnitude of Ca^2+^ currents. Taken together, these results suggest that CAST directly or indirectly regulates Ca~V~ abundance and function. Morphological abnormalities of rod synapses in CAST KO and dKO mice {#s10} ------------------------------------------------------------------- The impact of the loss of CAST/ELKS on the molecular architecture of rod triads was investigated by 3D reconstruction from a set of serial images using FIB-SEM ([@bib26]). In these experiments, we observed varying degrees of ribbon size reduction in all deletion mutants ([Fig. 4](#fig4){ref-type="fig"}), consistent with a previously proposed scaffolding role of CAST/ELKS that facilitates the lateral expansion and structural integrity of retinal ribbons ([@bib50]). Interestingly, neither the limited reduction of ribbon size (\<30%) nor the observed lateral fragmentation of ribbon bodies (Table S4) significantly affected photoreceptor neurotransmission or ERG performance in ELKS cKO mice ([Fig. 1, F--I](#fig1){ref-type="fig"}). In contrast, the major reduction in ribbon length of ∼50--70% in CAST KO and dKO mice severely impaired synaptic function and visual processing. According to the 3D reconstruction of the rod terminal, a single HC tip triad was formed in the CAST KO and dKO mice ([Fig. 4](#fig4){ref-type="fig"}). Further 3D reconstruction of the HC tips revealed fewer branches and the occurrence of filopodia-like structures (Fig. S4 E). We assumed an HC morphological impairment caused by the absence of CAST in HCs, mostly because Crx-driven Cre recombinase could eliminate ELKS only in non-HC neurons in the INL (Fig. S4, F--H). Therefore, the phenotypes seen in HCs could be explained by the cell-type--independent depletion of CAST in CAST KO and dKO mice. In this context, CAST localization at the growth cone was previously suggested to play a role in maturation of neural circuits in primary cultured neurons ([@bib40]). Previous reports also showed comparable single HC tip morphology in deletion mutants of PlexinA4 and netrin-G ligand 2, which are selectively localized to HC neurites and tips, respectively ([@bib34]; [@bib46]; [@bib7]). Enhanced ectopic synapse formation in retinae of CAST/ELKS deletion mutants {#s11} --------------------------------------------------------------------------- We previously found that retinae of CAST KO exhibited ectopic ribbons ([@bib50]). In the current study, we found that the compound deletion of ELKS and CAST further confounded synapse stability. Several previous studies have shown that mutation or deletion of various proteins involved in synaptic transmission was associated with ectopic ribbon formation ([@bib12]; [@bib44]; [@bib47]; [@bib35]). As an extreme example, depletion of synaptic vesicle protein cystein string protein α causes ectopic synapse formation in concert with massive photoreceptor degeneration ([@bib44]). Bassoon is a prominent example of an AZ protein, which is essential for anchoring the photoreceptor ribbon to the AZ, and its deletion is associated with ectopic synapses ([@bib12]). Moreover, mouse models of congenital stationary night blindness that carry mutations in the Ca^2+^ channels and/or its binding proteins are in accord with hypothesized causal relationship between loss of AZ structural as well as functional integrity and ectopic synapse formation ([@bib17]; [@bib6]; [@bib32]; [@bib42]). Thus, current data from CAST KO and dKO mice confirmed that impairment of neurotransmission drove ectopic synapse formation in the ONL, which was exacerbated with aging in CAST KO and dKO mice. Synaptopathies and retinal degenerative disease {#s12} ----------------------------------------------- The remodeling of photoreceptors is a hallmark of various acquired retinal diseases, such as retinitis pigmentosa (RP), which is mostly classified as inherited retinal dystrophy. The progression of RP is variable, although continued photoreceptor degeneration ultimately leads to functional blindness ([@bib24]). Although several gene mutations have been identified from severe RP forms ([@bib14]), and gene therapeutic approaches are currently under development ([@bib56]), the precise molecular mechanisms of RP progression are not well understood. Recent advances in stem cell therapy hold great promise for the treatment of advanced RP by transplantation of embryonic and induced pluripotent stem cells ([@bib45]; [@bib56]). Nevertheless, it is of utmost importance to suppress disease progression and retain normal vision for as long as possible. Thus, it is important to (a) understand the factors that initiate and facilitate neural remodeling and (b) identify molecular targets that are amenable to genetic therapy. In the current study, we found that the loss of ELKS in mature ribbon synapses induces photoreceptor degeneration, which was not apparent in the Cre-mediated ELKS cKO model, suggesting a novel molecular mechanism that specifically induces photoreceptor remodeling and degeneration ([Fig. 7, G and H](#fig7){ref-type="fig"}). Thus, to develop novel therapies for late-onset retinal neurodegenerative disease, future studies should (a) identify potential relationships between RP mutant genes and AZ proteins and (b) analyze functional roles of RP mutant genes in neurotransmitter release. Materials and methods {#s13} ===================== Generation of ELKS cKO mice {#s14} --------------------------- The use of the animals was approved by the Institutional Committee for the Care and Use of Experimental Animals at the University of Yamanashi, Yamanashi, Japan. Inducible retina-specific ELKS mutant mice were obtained by crossing *ELKS^flox/flox^* mice ([@bib13]) with *Crx-Cre* mice carrying Cre recombinase under control of the *Crx* promoter, which is an otx-like homeobox gene expressed abundantly in the retina (provided by T. Furukawa, Osaka University, Osaka, Japan; [@bib38]). The *ELKS^flox/flox^* mice were further crossed with *CAST^−/−^* ([@bib50]) to generate *CAST^−/−^; ELKS^flox/flox^* mice. The mice derived from crossing *ELKS^flox/flox^* with *ELKS^flox/flox^; Crx-Cre^+/−^* mice and *CAST^−/−^; ELKS^flox/flox^* with *CAST^−/−^; ELKS^flox/flox^; Crx-Cre^+/−^* mice were used for subsequent studies. Genotyping of ELKS cKO, CAST KO, and dKO mice by PCR was performed using the following primers: ELKS flox: ELKS flox forward, 5′-AAGGCCCAAACAGAAGTTGA-3′, ELKS flox reverse, 5′-ATGATTTGCTTTCCCATGCT-3′; CAST WT/KO: CAST WT forward, 5′-GTCACCACGTCTGCCAAGGT-3′, CAST KO forward, 5′-GACATAGCGTTGGCTACCCGT-3′, CAST reverse, 5′-GGGCTTGAAGATCCAACATCG-3′; and Crx-Cre: LCB836-Cre1, 5′-GTCGATGCAACGAGTGATGA-3′, LCB836-Cre2, 5′-AGCATTGCTGTCACTTGGTC-3′, mLC3ex3GT-Cre3, 5′-TGAGCGAGCTCATCAAGATAATCAGGT-3′, and mLC3ex3GT-Cre4, 5′-GTTAGCATTGAGCTGCAAGCGCCGTCT-3′. Immunoblotting {#s15} -------------- Retina homogenates from the adult mutant mice were analyzed using Western blotting, and the signal intensity was quantified using ImageJ software (National Institutes of Health) from more than three independent experiments. The primary antibodies used for Western blotting were anti-ELKS (1:500, rabbit; [@bib9]), anti-CAST (1:1,000, guinea pig; [@bib40]), anti-RIM1/2 (1:1,000, rabbit; provided by S. Kiyonaka, Kyoto University, Kyoto, Japan; [@bib28]; although it was originally described as RIM1, the antigen was designed from the consensus region of RIM1/2), anti-RIBEYE (1:1,000, rabbit; 192003; SYSY), anti-synaptophysin (1:2,000, mouse; AB5258; Chemicon), anti-GAPDH (1:5,000, HRP conjugate; 3683; Cell Signaling), and anti-tubulin (1:500, mouse; CP06; Oncogene). Immunohistochemistry and image analysis {#s16} --------------------------------------- Under deep pentobarbital anesthesia, mice were transcardially fixed with 4% paraformaldehyde and 10% picric acid in 0.1 M phosphate buffer (pH 7.4). The retinae were dissected and cryoprotected in 30% (wt/vol) sucrose in PBS for \>2 h. Pieces of retinae were mounted in OCT compound (Tissue-Tek) and sectioned at a thickness of 10 µm on a cryostat (HM525 NX; Thermo Fisher Scientific). The collected sections were blocked for 1 h in blocking solution (2% normal goat or horse serum, 10% block ace \[Dainippon Pharmaceutical\], and 0.2% Triton X-100 in PBS) and incubated in primary antibodies in blocking solution overnight at room temperature. The following primary antibodies were used: anti-RIBEYE (1:1,000, rabbit), anti-PKC (1:1,000, mouse; ab31; Abcam), anti-vGluT1 (1:1,000, guinea pig), anti-calbindin (1:2,000, mouse; C9848; Sigma-Aldrich), and anti-Cre (1:800, mouse; MAB3120; and 1:500, rabbit; 69050-3; Millipore). A guinea pig antiserum against vGluT1 was raised against mouse vGluT1 C-terminal 531--560 aa (NM020309; [@bib36]). The sections were further processed with appropriate Alexa Fluor--conjugated secondary antibodies for 1 h at room temperature. Immunolabeled samples were viewed using a confocal laser microscope (FV1200; Olympus) at room temperature. Distribution of ribbon synapses {#s17} ------------------------------- RIBEYE-immunolabeled images were obtained with oil-immersion 60× objective (NA 1.35) with 2× digital zoom, and then binary-transformed (threshold ∼1,300) images were analyzed using the ImageJ particle analysis program (particles smaller than 20 pixels were eliminated; Fig. S3). The total number of particles in OPL and ONL were divided by the area to calculate the RIBEYE density. The distance of each particle from the bottom of OPL (y axis) was evaluated by generating histograms and plotted as the cumulative frequency (%) for display. The statistical significance of density and probability for each distance from ONL was analyzed using one-way ANOVA or two-way ANOVA, respectively. Imaging with FIB-SEM and 3D reconstruction {#s18} ------------------------------------------ The mice were transcardially perfused with 2% glutaraldehyde and 2% paraformaldehyde in 0.1 M phosphate buffer (pH 7.4). Retinal sections (thickness, 400 µm) were further en bloc stained with a reduced OTO staining method: 1.5% potassium ferrocyanide, 2% OsO~4~, 1% thiocarbohydrazide7, and then 2% OsO~4~. For counterstaining, sections were further treated with 4% uranyl acetate and Walton's lead aspartate solution. After staining, samples were dehydrated and embedded in epoxy resin (EPON812; TAAB). Specimens were subjected to FIB-SEM tomography (Quanta 3D FEG; FEI) as previously described ([@bib26]). In brief, the embedded slices were placed on a metal stub and further trimmed with glass and diamond knives in an ultramicrotome (Ultracut E microtome; Leica). The slices were coated with an electroconductive layer of carbon, which prevented any charge. The metal stub with the slices was set on the FIB-SEM stage, and then serial section images were automatically obtained from the OPL of the retina. Serial images of the block face were acquired by repeated cycles of sample surface milling and imaging using the Slice & View G2 operating software (FEI). The milling was performed with a gallium ion beam at 30 kV with a current of 1.0 nA. The milling pitch was set to 15 nm/step. The images were acquired at an accelerating voltage of 2.5 kV. The other acquisition parameters were as follows: dwell time = 6 s/pixel and pixel size = 7.3 nm/pixel. The serial section images were reconstructed to 3D images and analyzed using Amira 5.5 software (FEI). ERG recordings {#s19} -------------- Mice were adapted to the dark overnight and anesthetized by intraperitoneal injection of ketamine (0.125 mg/g) and xylazine (2.5 µg/g), and one pupil was dilated with 1% atropine sulfate. All preparations were done under a dim red light. A ring-like AgCl wire electrode, moistened by methylcellulose and NaCl solutions, was placed on the cornea, and a needle reference electrode was subcutaneously inserted above the nose with a ground electrode near the tail. A custom-designed Ganzfield illuminated by 25 white LEDs was used to produce light flashes of incremental luminances (0.0003 to 0.278 candela/m^2^; measured by Mavolux, IPL 10530). Scotopic responses were recorded for a stimulus duration of 0.1 ms with an inter-stimulus interval of 5 s. Recorded potentials were amplified, filtered (\<400 Hz for a-waves, \<20 Hz for b-waves, and 30--400 Hz for oscillatory potentials), and sampled at a rate of 24 kHz. 10 responses were averaged per light intensity. For analysis, a-wave amplitudes were measured relative to the baseline, whereas b-wave amplitudes were estimated relative to the trough of the a-wave. All data were analyzed using custom-written MATLAB-based analysis routines. Statistical significance was assessed by repeated measures two-way ANOVA followed by post hoc Tukey corrected for multiple comparisons. EM {#s20} -- For conventional EM, mice were perfused transcardially with 2% paraformaldehyde and 2% glutaraldehyde in 0.1 M phosphate buffer (pH 7.4). Retinal sections (thickness, 400 µm) were further post-fixed with 2% osmium tetroxide in phosphate buffer for 2 h, stained with 2--4% uranyl acetate for 1 h, and then embedded in Durcupan ACM (Fluka). Ultra-thin sections (thickness, 70 nm) were prepared with an ultramicrotome (Ultracut EM UC7; Leica) and counterstained with uranyl acetate and lead citrate. For analysis, images were captured at OPL with an electron microscope (H-7500; Hitachi). Density of ribbon bearing rod terminal was analyzed from EM images of 2,500 magnification (single image size: 125 µm^2^) in which cone terminals containing several mitochondria and ribbons could be identified by chance. The probability of finding the cone terminal was almost 10% of the control rod terminal, which was not significantly different between genotypes. Synaptic vesicle distribution was analyzed at higher magnification (e.g., 100,000, 2.25 µm^2^), and quantitative parameters were measured using ImageJ software. AAV5-mediated Cre expression and image analysis {#s21} ----------------------------------------------- For AAV-mediated Cre, Venus, or CAST-Venus expression in the retina, plasmid encoding pAAV5-CAGGS-nCre, pAAV5-CAGGS-IRES-Venus, and pAAV5-CAGGS-CAST-IRES-Venus were prepared to generate the recombinant AAV5. To inject the virus into eyes, mice were anaesthetized at ∼5 wk of age with ketamine/xylazine by intraperitoneal injection. A 30-gauge needle was used to make a small hole in the temporal eye, below the cornea, and ∼1 µl of AAV virus was injected into the vitreous humor over the retina using a glass pipette. For rescue experiments, a mixture of pAAV5-CAGGS-nCre with pAAV5-CAGGS-CAST-IRES-Venus or pAAV5-CAGGS-IRES-Venus was injected. Because Cre or Venus expression in photoreceptor neurons with AAV5 was "patchy" in the injected retinae, we took at least three Cre or Venus expression images (Cre\[+\], Venus\[+\]) and non-Cre or -Venus expression images (Cre\[−\], Venus \[−\]) from the same retina. For ONL thickness, the images were obtained using a 40× objective lens (NA 0.95). RIBEYE density was calculated from images taken with a 60× objective and a 2× digital zoom. ONL thickness and RIBEYE densities were analyzed using ImageJ and compared between Cre-positive and -negative regions using the Student's *t* test. TUNEL assay {#s22} ----------- For the assessment of apoptosis, the DeadEnd Fluorometric TUNEL system (Promega) was applied to 10-µm-thick PFA-fixed retinal cryosections according to the manufacturer's instructions. To identify the Cre expression area in the retina, another serial cryostat section was PFA fixed and then immunolabeled with anti-ELKS (1:100) and anti-Cre (1:800) antibodies. Whole-cell voltage clamp recording {#s23} ---------------------------------- Mice between 14 and 18 wk of age were tested. Preparation of retinal slices and whole-cell voltage clamp recordings were conducted as described previously ([@bib15]). Briefly, retinal slices (∼200 µm thickness) were prepared in the following mouse extracellular solution, which had a low Ca^2+^ concentration (in mM): 135 NaCl, 2.5 KCl, 0.5 CaCl~2~, 1 MgCl~2~, 10 glucose, and 15 Hepes, pH 7.35, with osmolality adjusted to 295 mOsm. Slices in the recording chamber were then continually perfused in mouse extracellular solution with 2.5 mM Ca^2+^ at a temperature of 30°C to 32°C. The intracellular solution contained the following reagents (in mM): 105 CsCH~3~SO~4~, 20 TEA-Cl, 2 MgCl~2~, 5 MgATP, 0.2 NaGTP, 10 Hepes, 10 EGTA, and 10 µg/ml Alexa Fluor 488, with pH adjusted to 7.30 with CsOH to an osmolality of 285 to 290 mOsm. The theoretical junction potential generated with these intra- and extracellular solutions is estimated to be −12 mV. The voltage-clamp data are not presented with the −12 mV added (V~1/2~ and current traces plotted over voltage are raw values). The hyperpolarizing I~h~ current was monitored to confirm the intactness of the recording. The ramp protocols were preceded by steps from −60 to −70 mV or −70 to −90 mV (to activate I~h~), and then ramped to +50 mV at a speed of 1 V/s. Since these two ramp types gave very similar I~Ca~ and V~1/2~ values within cells, data were summarized by averaging together the different voltage ramps. Average values from five control cells for ramps starting at −70 and −90 mV were as follows: I~Ca~, 12.1 vs 11.8 pA (P = 0.95), and V~1/2~, −28 vs −30 mV (P = 0.61). Statistical analysis {#s24} -------------------- Data are presented as mean ± SEM. Data distribution was assumed to be normal, but this was not formally tested. Statistical significance was analyzed using one-way ANOVA, two-way ANOVA, or two-sided Student's *t* test as indicated in each figure legend and in Table S6. Online supplemental material {#s25} ---------------------------- Fig. S1 shows the details of ELKS cKO with amino acid sequence of CAST and ELKS and their different solubility. Fig. S2 and Table S1 show the characterization of each genotype of mice including body weight and eyeball size, original data of immunoblotting, and retinal layer formation. Fig. S3 shows the analysis of ectopic ribbon synapse immunolabeled with RIBEYE and vGluT1. Fig. S4 and Tables S2, S3, and S4 give the additional data from the 3D-reconstructed rod terminal from each genotype of mice. Fig. S4 demonstrates lower expression of Cre recombinase in HC and Table S5 shows the line density of HC was not significantly different between genotypes. Fig. S5 shows the recoding of Ca^2+^ current from rod photoreceptors with inner segments. Though the traces of Ca^2+^ current with inner segments showed outward current, the peak amplitude and V~1/2~ were not significantly different from the soma without inner segments. Then the data were pooled and statistically analyzed between genotypes. Fig. S6 shows the AAV-mediated Cre and Venus expression in *ELKS^fl/fl^*, ELKS cKO (*ELKS^fl/fl^ x Crx-Cre^+/−^*), and *CAST KO; ELKS^fl/fl^* mice. Fig. S6, G and H, also showed the AAV-mediated expression of CAST and Venus in primary culture neurons for rescue experiments. Table S6 shows summary of statistical analysis. Supplementary Material ====================== ###### Supplemental Materials (PDF) We thank N. Sugiyana, K. Moriya, and all members of the Ohtsuka Laboratory for their helpful discussions and technical assistance, and S. Thom for expert technical assistance with the ERG recordings. We also thank Dr. T. Furukawa for the Crx-Cre mouse and Dr. S. Kiyonaka for the RIM antibody. This work was supported by Japan Society for the Promotion of Science KAKENHI grants 25830008 (to A. Hagiwara), 15H04272 (to T. Ohtsuka), and 16H05135 (to A. Nishi); a Grant-in-Aid for Scientific Research on Innovation Area "Dynamic Regulation of Brain Function by Scrap and Build System" from the Ministry of Education, Culture, Sports, Science and Technology, grant 17H05741 (to T. Ohtsuka); Japan Science and Technology Agency CREST, grant JPMJCR1751 (to T. Ohtsuka); and the University of Yamanashi. Work on the Göttingen Campus was supported by the German Research Foundation through Collaborative Research Center 889 (to T. Moser) and the Max Planck Society (fellowship to T. Moser). C. Vogl is a Creutzfeldt Fellow of the Elisabeth and Helmut Uhl Foundation. Production of AAV vectors was supported by the Cooperative Study Program of the National Institute for Physiological Sciences (K. Kobayashi). The authors declare no competing financial interests. Author contributions: A. Hagiwara designed and performed the biochemical, morphological, and optical imaging analysis, with the support of R. Kitta, and drafted the manuscript. T. Ohtsuka directed the study and edited the manuscript. Y. Kitahara, K. Ohta, K. Nakamura, and A. Nishi performed FIB-SEM and 3D reconstruction analysis. C. Vogl and T. Moser performed physiological analysis and edited the manuscript. C.P. Grabner performed and analyzed the whole-cell recordings and edited the manuscript. M. Abe and K. Sakimura generated the ELKS floxed mouse. All authors read and approved the final manuscript. [^1]: Y. Kitihara and C.P. Grabner contributed equally to this paper.

INTRODUCTION ============ Type II restriction--modification (R--M) systems are comprised of (i) a restriction endonuclease that recognizes a specific DNA sequence and introduces double-stranded breaks at or around the recognition site and (ii) a methyltransferase (methylase) that recognizes the same DNA sequence and methylates it first on one of DNA strands to produce hemimethylated DNA, and then on the other strand to produce fully methylated DNA. Methylation protects DNA from cleavage by the endonuclease ([@B1],[@B2]). In several cases that have been studied experimentally, bacterial cells carrying R--M genes become resistant to infection by bacteriophages containing unmethylated (unmodified) DNA, which may explain their wide dissemination. R--M systems are often carried on mobile genetic elements capable of horizontal spread between different bacterial species ([@B3]). Premature production of endonuclease upon the entry of a genetic element carrying R--M system genes into a naïve host can lead to host DNA degradation and death of the host. To minimize the likelihood of such an outcome, which would also destroy the R--M system and the mobile element that carries it, R--M systems evolved special mechanisms to coordinate expression of their genes, ensuring that endonuclease expression is activated only after the host DNA is fully methylated by the methylase. Several distinct mechanisms of such activation have been described ([@B4]). Eight R--M systems---*Ahd*I ([@B5],[@B6]), *Bam*HI ([@B7]), *Bgl*II ([@B8]), *Eco*72I ([@B9]), *Eco*RV ([@B10]), *Esp*1396I ([@B11]), *Pvu*II ([@B12]) and *Sma*I ([@B13])---have been experimentally shown to rely on specialized control (C) proteins ([@B7],[@B12]) for coordinated expression of their genes. All C proteins are related through common ancestry and are also distantly related to phage helix-turn-helix DNA-binding transcription factors, including the well-studied phage λ repressor. The structures of three C proteins (C.BclI, C.AhdI and C.Esp1396I) that have been studied by crystallography ([@B14; @B15; @B16]) reveal that these proteins form dimers and that each monomer is similar to the DNA-binding domain of the lambda cI repressor, belonging to the Xre family of transcription factors ([@B15]). Genes coding for C proteins are often located upstream of, and partially overlap with, the endonuclease (*R*) gene ([@B17]), forming a single operon. Upstream of and partially overlapping with the *CR* operon promoter, two C protein-binding sites are located ([@B17]). When a C protein dimer binds to the high-affinity promoter-distal site, transcription is activated, leading to increased C protein (and endonuclease) gene expression ([@B6],[@B10]). The exact mechanism(s) of activation is not known and may vary in different R--M systems. In the few cases that have been studied, the promoter-distal C protein-binding site is located immediately upstream of the −35 promoter element of the *CR* operon promoter ([@B6],[@B10],[@B18]). Thus, C protein-dependent activation may involve protein-protein contacts between the C protein and the RNA polymerase σ subunit region 4, which specifically recognizes the --35 promoter element ([@B19]). C protein binding to the weaker, promoter-proximal site, occludes the --35 element of the *CR* operon promoter and inhibits transcription ([@B6],[@B15],[@B20]), most likely by excluding the RNA polymerase σ subunit region 4 from the --35 element. The dual (activation and repression), concentration-dependent mode of transcription regulation of the *CR* promoter by C protein ensures a delayed appearance of the endonuclease activity during establishment of C protein-dependent R--M systems in a naïve host and allows to maintain constant steady-state levels of endonuclease during the stable maintenance of R--M system in the host \[see, for example, ref. ([@B6]) for kinetic modeling of the process\]. The high cooperativity of C protein dimer interactions with DNA observed in at least some studied systems ([@B5],[@B6]) affords sharp regulatory responses of C protein-dependent autoregulatory loops. In this work, we used a bioinformatics approach to answer the following questions. First, we wanted to systematically identify genes coding for C protein homologs. Second, we wanted to predict DNA-binding sites of bioinformatically identified C proteins. Lastly, we sought to determine if proteins homologous to C proteins from R--M systems are specific to such systems or are also found in other genetic contexts. METHODS ======= The *Rebase* database contains 48 C protein sequences. One of the proteins (C.MjaVP) is more than twice as long as the rest of the C proteins. Another protein (C.AmaFACHORFAP) resulted from a formal translation of a pseudogene. These two proteins were excluded from the analysis. The remaining 46 C protein sequences were used as queries in the *BLAST* ([@B21]) search against the non-redundant nucleotide database of GenBank ([@B22]) (*tblastn*, threshold 1*e*--05). This search yielded 245 unique hits, which were considered as genes encoding putative C proteins, or, more exactly, C protein-family regulators (although, naturally, in the absence of experimental data even this general functional assignment is only preliminary). Starts of the genes were manually corrected using the standard bacterial ORF analysis rules and the fact that the average C protein length is about 70 amino acids (aa). Multiple alignment of all 291 proteins (46 proteins from *Rebase* and 245 hits generated by the *BLAST* search) was built using the *muscle* program ([@B23]) and the unrooted maximum likelihood tree with molecular clock was constructed using the *proml* procedure from the *PHYLIP* package ([@B24]). Both programs were run with the default parameters. The tree was split into several large subfamilies which were analyzed independently. For each group of closely related proteins, short (100 bp) regions upstream of the corresponding genes were aligned using *muscle* with the default parameters. The following procedure was used to extend this alignment by including upstream regions of more distantly related C protein genes. Genes were added to the alignment one by one in the order dictated by the tree, and at each step the upstream regions were re-aligned. When the alignment started to degrade, such 'extension' process was stopped and putative binding motifs were manually predicted by the analysis of the remaining conservation islands. Further, each remaining upstream region was compared with its nearest neighbors on the tree, for which the binding sites had already been predicted. The multiple alignment, which included this remaining region and its tree neighbors, was forced to align the predicted sites and again was analyzed manually. If the conserved island covering the putative site did not deteriorate upon inclusion of the new sequence, the latter was also predicted to be a binding site. To account for a possibility that some sites were missed because of mis-annotation of gene starts or positioning of the site outside of the 100 bp upstream region, all C protein-family genes for which the procedure described above failed to reveal a putative binding site were analyzed further. First, *hmmer* (<http://hmmer.wulst.edu>) profiles of candidate sites were built for each constructed alignment (hmmbuild --g, nucleotide mode). Second, the *hmmsearch* procedure was applied and these profiles were used to scan regions from --100 bp to +50 bp relative to translation starts of putative C protein genes without predicted sites. The best candidates were added to the set of predicted binding sites. However, this procedure resulted in few additional binding sites, showing that the overall approach is a robust one. The whole set of putative binding sites was split into clusters, which are further referred to as motifs, using the *ClusterTree-RS* procedure ([@B25]). The procedure yielded 10 stable motifs, which contained 181 (90%) of 201 predicted sites. Since *ClusterTree-RS* generates nested clusters, we were able to subdivide motifs 2 and 6 into motifs 2, 2^b^, 2^c^ and 6 and 6^b^, respectively. The resulting motifs 2 and 6 contain the most conserved members of the original motifs, while 2^b^, 2^c^ and 6^b^ contain sequences more distantly related to motif 2 and motif 6 consensus sequences. Genome loci containing C protein-family genes with predicted binding sites were studied in more detail. These loci were defined formally as genomic regions from 3000 bp upstream to 3000 bp downstream of each C protein-family gene. Candidate genes were defined as ORFs longer than 150 codons with potential start codons. These genes were translated and compared to the non-redundant protein database (*blastp*, threshold 1*e*--06), and the *pfam-a* seed database ([@B26]) (*hmmer*, global mode, no calibrate-mode, threshold --E 0.01 --Z 1). All hits were classified into three functional categories: 'phage-related', 'R--M-related' and 'the rest'. To do that, a list of relevant pfam families and words in protein annotations was compiled. This list was used to scan the pfam assignments and lists of *BLAST* hits for each ORF, followed by manual verification of the ORF status. A locus containing at least one phage-related or R--M-related ORF was labeled correspondingly. To identify hypothetical genes consistently appearing in the vicinity of C protein-family genes, all ORFs were further clustered by similarity using a two-step procedure. At the first step, groups of highly similar ORFs were identified using the standard *blastclust* procedure ([@B21]) (length coverage *L* = 0.50, identity percentage *S* = 0.90). Sixty groups which contained three or more sequences were collapsed and only one representative of each group was used for further clustering. At the second step, total pairwise *BLAST* search (*blastp*, threshold 1*e*--10) was performed and clusters were determined by a single linkage procedure. Proteins from the resulting 38 clusters were aligned using *muscle* (default parameters). However, since the analysis did not reveal any genes significantly associated with the candidate C-protein genes, the clusters were not considered further (data not shown). At the end, each candidate binding site was assigned the following data: its sequence; site motif (if any); candidate C protein sequence; similar *Rebase* C proteins that contained this protein in the *BLAST* similarity search output, see above; map of the genomic locus with all ORFs; list of *pfam* families which matched the orfs; list of *BLAST* hits for each ORF. These data were collected into a specially developed database that can be accessed online (<http://iitp.bioinf.fbb.msu.ru/vsorokin>). Logos were produced using the *weblogo* 2.8 package ([@B27]). The tree was visualized using the web-based tool *iTOL* (<http://itol.embl.de/>). RESULTS ======= Identification of new C protein family members ---------------------------------------------- Using 46 annotated C proteins from *Rebase* as a starting point for database similarity search with *BLAST*, we obtained 245 additional putative members of the C protein family. The parameters of the search, described in the Materials and methods section, were set such that distant relatives of known C proteins, such as phage repressors, were not retrieved by the search. To identify closely related C protein sequences, an unrooted likelihood tree of all 291 (46 annotated sequences + 245 newly found sequences) members of the family was constructed. A slightly smaller variant of this tree, containing proteins whose binding motifs could be identified (see below), is shown in [Figure 1](#F1){ref-type="fig"}. A version of the tree with bootstrap values resulting from 100 pseudoreplications is available as [Supplementary Data](http://nar.oxfordjournals.org/cgi/content/full/gkn931/DC1). It should be noted that the tree reflects protein, rather than species, evolution as in large parts it does not match accepted taxonomy divisions. This likely indicates that C protein genes are subject to extensive horizontal transfer, an expected result, given the biological function and genetic contexts of known C proteins. Figure 1.Maximum likelihood tree built of REBASE and newly discovered putative C proteins. Color indicates C proteins whose predicted binding sites fall into distinct motifs (1 through 10). The subdivided motifs (2, 2b, 2c and 6, 6b) are marked with similar, but not identical colors. The asterisk preceding the protein id indicates a relatively lower level of prediction confidence. Identification of putative C protein DNA-binding sites ------------------------------------------------------ On the basis of published information about transcription regulation by a few C proteins that have been studied experimentally, we expected that C protein-binding sites would be located in close vicinity to translation start points of C protein genes. We also hypothesized that the binding sites for closely related C proteins will be similar. The following unbiased iterative procedure was used to search for evolutionary conservation of DNA sequences upstream of C protein genes (see also Materials and methods section). The tree of C proteins was split into several large branches, which were analyzed independently. For every branch, upstream (100 bp upstream of manually curated annotated translation start codon) DNA regions for most closely related annotated or putative C protein genes were aligned using *muscle*. Next, the alignment was extended by inclusion of upstream regions of more distantly related C protein genes from the same branch. At each step, the upstream regions were realigned. When the alignment started to degrade, the 'extension' process was terminated. In this way, ∼60% of sequences from each branch of the tree could be aligned. Putative binding motifs were next predicted by visual analysis of conservation islands in the alignments. It is noteworthy that in all cases, only one continuous conserved motif-like element was detected. Some additional sequences, which did not 'naturally' fit in the alignments, were subsequently added manually using conserved consensus sequences derived from the iterative procedure. In total, conserved motifs were detected upstream of 201 of the 291 C protein family genes (69.1%). An alignment of the proteins with predicted binding sites is available as [Supplementary Data](http://nar.oxfordjournals.org/cgi/content/full/gkn931/DC1). Of the 46 C protein genes listed in *Rebase*, 32 (69.5%) contained conserved upstream motifs. These included eight R--M systems where C protein-binding sites have been identified experimentally (C.BamHI, C.BglII, C.PvuII, C.SmaI, C.AhdI, C.MunI, C.EcoRV, C.EcoO109I), as well as 24 additional C protein-dependent R--M systems (C.Pde1222ORF1578P, C.MspMCORF1281P, C.BcnI, C.SptAI, C.SbaI, C.SonORF4P, C.NmeSI, C.VeiORF3519P, C.BstLVI, C.Sse9I, C.Lci22RP, C.SspMR4ORF3202P, C.SgrAI, C.Csp231I, C.BfrYORF1158P, C.BfrLV23P, C.BfrYORF1980P, C.BfaSORF1835P, C.ChuAORF2941P, C.BfaSORF1077P, C.EcoT38I, C.LlaDI, C.SnaBI, C.LgaORF1464); putative C protein-binding sites in some of these systems were also identified earlier by Naderer *et al.* ([@B28]). Most importantly, for eight of the 46 *Rebase*-listed C protein genes for which C protein-binding sites are known, the conserved upstream motifs coincided with experimentally determined binding sites. Since the binding site information was ignored during our analysis, we conclude that identification of upstream conserved sequences leads to identification of C protein-binding sites with high confidence. The fact that the search procedure did not result in identification of other regulatory elements likely to be present upstream of C protein genes, such as C protein gene promoters, which in cases that have been studied in experiment overlap with C protein-binding sites ([@B6],[@B10],[@B11],[@B18]), indicates that the evolutionary conservation of promoter elements is significantly lower than that of the C protein-binding sites (see also below). The structure of putative C protein-binding sites ------------------------------------------------- Previous limited analysis of C protein-binding sites by Blumenthal and colleagues identified three types of sites ([@B20]). The first type contained six distinct sites including the C.PvuII site and was called non-palindromic; two remaining types were represented by only one site each (of C.EcoRV and C.EcoO109I) and were considered palindromic. A collection of putative C protein-binding sites revealed by our analysis allowed us to extend the previous classification and to identify new motifs ([Table 1](#T1){ref-type="table"}, [Figure 2](#F2){ref-type="fig"}a--c). Table 1.Motifs, the number of candidate C proteins and the gene content of respective lociMotifsTotal members*Rebase* membersMotif typeNumber of loci with RM-related orfsNumber of loci with phage-related orfsMotif 1210C.PvuII-like (double-box motif)715Motif 23312C.PvuII-like (double-box motif)2123Motif 2^b^146C.PvuII-like (double-box motif)810Motif 2^c^121C.PvuII-like (double-box motif)26Motif 3180C.PvuII-like (double-box motif)018Motif 480C.PvuII-like (double-box motif)37Motif 5100C.PvuII-like (single-box motif)39Motif 650C.PvuII-like (single-box motif)02Motif 6^b^100C.PvuII-like (single-box motif)17Motif 7133C.EcoRV-like (palindromic motif)811Motif 8142C.EcoO109I-like (palindromic motif)67Motif 9156new (non-palindromic motif)711Motif 1080new (palindromic motif)14 Figure 2.(**a**) Logos of C.PvuII-like ([@B1; @B2; @B3; @B4; @B5; @B6]) motifs. The total number of members and the number of REBASE members (if any) are indicated for every motif. Paired palindromic boxes (consensus sequences) are marked with light green squares. Palindromic elements of motifs' architecture are underlined with colored arrows. Conserved trinucleotides found at the outside of the motifs are underlined with orange arrows. (**b**) Logos of palindromic (7, 8 and 10) motifs. The total number of members and the number of REBASE members (if any) are indicated for every motif. Paired palindromic boxes (consensus sequences) are marked with light green squares. Palindromic elements of motifs' architecture are underlined with colored arrows. (**c**) Logo of new motif 9. The total number of members and the number of REBASE members are shown. The C.PvuII-like sites were assigned to six related but clearly distinct motifs (motifs 1--6, [Figure 2](#F2){ref-type="fig"}a). All these motifs have the same length (35 bp) and share a common architecture, which consists of three conserved bases at the outside of the motif and some almost invariant positions in the core of the motif. The three bases at the outside boundary of the motif are often complementary. A typical motif consists of two copies of the same palindromic consensus, which we term 'operator', with the 5′ copy (with respect to C protein gene translation start point) being much closer to the consensus than the 3′ copy. The two copies are separated by a highly conserved non-symmetrical GTG sequence. Thus, the overall typical motif architecture is Z-X-N-X\*-GTG-x-n-x\*-Z\*, where X denotes internal boxes forming a palindrome, N denotes internal, non-palindromic positions in operators, Z denotes external three-nucleotide boxes and asterisks denote complementary boxes; uppercase denotes highly conserved nucleotides or boxes, while lower-case denotes weakly conserved boxes or nucleotides. It should be noted, however, that some individual conserved putative C protein-binding sites within a group do not match this idealized scheme. Moreover, distinct variations in symmetry patterns of different motifs exist. For instance, in motifs 1, 3 and 5, the overall symmetry is odd (the center of symmetry at the T of the central GTG) while the individual operators have the even symmetry (X-X\*) ([Figure 2](#F2){ref-type="fig"}a). On the contrary, the overall symmetry of motif 2 is even (with symmetry center between the G and the T of the central GT), while the operators are odd palindromes ([Figure 2](#F2){ref-type="fig"}a). It is worth mentioning that a sequence identical to motif 2 was previously identified by Vijesurier *et al.* ([@B29]) The outer elements (TG/CA) in a motif 2 sequence from the *AhdI* system were recently shown to be contacted by the cognate C protein ([@B16]). In motif 4, the symmetry of operators extends and includes the internal conserved trinucleotide, in this case, CTG ([Figure 2](#F2){ref-type="fig"}a); resulting in the richest symmetry pattern for putative binding sites identified here, which can be codified as Z3-X6-X6\*-T-X6-X6\*-Z3\*, where numbers denote lengths of respective elements. In motifs 5 and 6, no major symmetry pattern could be observed besides the palindromic ATG-CAT trinucleotides at the outside flanks of the motif. Together, these observations suggest that molecular details of C protein interactions with different motifs are subject to considerable variation. Our analysis extends the total number of palindromic C.EcoRV-type and C.EcoO109I-type binding sites to 13 (for C.EcoRV-type sites) and 14 (for C.EcoO109I-type sites). The C.EcoRV-type and C.EcoO109I-type binding sites form motifs 7 and 8, respectively. We also identified a new motif of a similar palindromic structural type (motif 10, [Figure 2](#F2){ref-type="fig"}b). The newly identified motif 9 ([Figure 2](#F2){ref-type="fig"}c) also may be tentatively assigned to this type, although its properties are rather unusual. It is short and lacks any discernible symmetry; in particular, it does not have complementary terminal trinucleotides. Thus it might be a case of a false positive. On the other hand, motif 9 is the only highly conserved sequence found upstream of 15 C protein homologs, of which six are annotated in *Rebase*. The branch of C proteins corresponding to motif 9 is the most distant one in the phylogenetic tree ([Figure 1](#F1){ref-type="fig"}), which may explain the unique properties of the putative binding site. Still, given the uncertainty about this motif, we exclude it from further consideration. The structure of genomic loci containing C protein family genes --------------------------------------------------------------- We analyzed 6 kbp genomic loci centered at putative C protein genes. The size of the window was arbitrarily selected; however, type II R--M loci from *Rebase* are almost half the length of the window selected (BamHI ∼ 2.3 kb; PvuII ∼ 1.8 kb; AhdI ∼ 3.5 kb) and we therefore expected that this analysis should reveal association of putative C proteins with R--M systems genes. Each ORF in the locus was used as a query for the *BLAST* similarity search against the non-redundant protein database and the search against the curated database of protein families *pfam*. The resulting hits were clustered by similarity and classified into R--M-related ORFs (candidate restriction endonucleases and methyltransferases), phage-related ORFs, or neither of the above as described in the Materials and methods section. The results are given in [Table 2](#T2){ref-type="table"}. Naturally, 100% of C protein genes from *Rebase* belong to R--M loci. In contrast, only ∼25% of newly defined putative C protein genes belong to such loci. However, the number of putative C protein genes with adjacent phage-related ORFs in *Rebase* and in newly defined loci is similar (80% and 70%, correspondingly). Twenty-seven of 169 newly defined putative C protein genes have both R--M-related and phage-related genes in their vicinity. Table 2.Distribution of candidate C protein genes in loci containing RM and phage-related genesGenomic lociRM-related orfsPhage-related orfsBothTotalNew C protein- family genes39 (23%)115 (68%)27 (16%)169 (100%)C protein genes from Rebase with binding sites predicted32 (100.0%)26 (78%)26 (81%)32 (100%) Since a conspicuously large number of newly defined putative C protein genes did not contain putative R--M genes, in particular, easily recognizable methyltransferase genes, we addressed a possibility that some branches in our C protein family may in fact contain *bona fide* phage repressors. Visual analysis of the tree shown in [Figure 1](#F1){ref-type="fig"} demonstrates that C protein genes from purely phage-related loci are interspersed with those from R--M-related loci, and, with an exception of 18 putative C proteins that are associated with motif 3, all branches contain interspersed phage- and R--M-related loci. This observation argues against a hypothesis that a large proportion of newly identified C protein genes encode phage repressors. High frequency of phage-related genes in the vicinity of putative C protein genes may indicate that a considerable fraction of putative C protein genes identified by our analysis are remnants of previously functional R--M systems destroyed by genome recombination/phage insertion events. In eight cases, two putative C proteins genes were found within 3 kb from each other ([Table 3](#T3){ref-type="table"}). Such 'paired' C protein genes may result from large-scale genomic rearrangements, such as duplications of C protein genes-containing loci or multiple insertions of C protein genes-containing elements in the same genomic location. The fact that two out of eight paired C protein genes loci encode C proteins associated with different motifs, while the other six loci contain genes encoding putative C proteins whose predicted binding sites belong to same motifs, indicates that both scenarios are realized. While known C proteins control simple autoregulatory loops individually, it is attractive to speculate that some clustered C protein genes may jointly control more complex regulatory circuits. Table 3.Loci containing two C protein genesPutative C protein gene 1StartEndBinding motif class of C protein 1Putative C protein gene 2StartEndBinding motif class of C protein 2Genbank IDDistanceMlo243  5425354495Motif 7Mlo0935471454923Motif 4AL672113219Bxe227453842454198Motif 8Bxe226453455453781Motif 8CP00027261Ccr19729195292919741Motif 2^b^Ccr19629198372920043Motif 2^b^AE00567396Mlo09249438384944044Motif 4Mlo10549443504944562Motif 1BA000012306Plu192157322157555Motif 2Plu191154190154423Motif 2BX5718732899Brl14924596642459906Motif 5Brl14724582762458518Motif 5CP0000851146Bps148521079521321Motif 5Bps146522475522717Motif 5BX5719661154 We also wondered whether any of the 39 putative R--M genes associated with predicted C protein genes correspond to known R--M systems from *Rebase*. To address this possibility, the coordinates and Genbank IDs of R--M-related genes associated with a particular putative C protein gene were compared with Genbank IDs of *bona fide* R--M systems genes present in REBASE. The match of Genbank IDs and coordinates would indicate that R--M genes associated with a putative C protein gene are identical to genes of an R--M system from REBASE that, however, lacks an annotated C protein gene. The results of this analysis are shown in [Table 4](#T4){ref-type="table"}. As can be seen, 14 of the 39 putative R--M loci correspond to already known REBASE R--M systems. In 2 of 14 cases (Hne199, Gur068), the newly discovered putative C proteins correspond to a short uncharacterized ORF of a *Rebase*-annotated R--M system. Our independently identified putative C protein gene Ent115 corresponds to the recently annotated C.Esp1396I. In all remaining cases, the discovered putative C protein genes are adjacent to *Rebase*-annotated R--M systems (which, however, are not annotated as having any genes other than the restriction endonuclease or methyltransferase genes). Our analysis predicts that these R--M systems are in fact C protein-dependent. Table 4.Genomic co-occurrence of C protein genes and known Rebase R--M systemsNo.Putative system IDMotifOrganism descriptionGenbank IDPutative C protein start-endREBASE system IDGene annotationGene start-endDistance1Pst154Motif 1Pseudomonas stutzeri A1501, complete genomeCP000304.1749928--750170PstA1501ORF647PMethylase743080--74505648722Lho088Motif 2^c^Laribacter hongkongensis plasmid pHLHK8, complete sequenceAY858987.11735--1983LhopHLHKPRestrictase3832--480618493Nha178Motif 2^b^Nitrobacter hamburgensis X14, complete genomeCP000319.12773100--2773321NhaXORF2515PMethylase2770295--277100220984Hch122Motif 2^b^Hahella chejuensis KCTC 2396, complete genomeCP000155.12547264--2547500HchORF2488PMethylase2547622--25488121225Hne199Motif 2Hyphomonas neptunium ATCC 15444, complete genomeCP000158.12696435--2696641HneORF2545PUnannotated short protein2696432--269664106Swi193Motif 2Sphingomonas wittichii RW1, complete genomeCP000699.11752458--1752658SwiRWORF1578PMethylase1754713--175616420557Nwi091Motif 4Nitrobacter winogradskyi Nb-255, complete genomeCP000115.1922287--922499NwiORF847PMethylase924490--92524819918Ent115 (C.Esp1396I)Motif 2Enterobacter sp. RFL1396 plasmid pEsp1396, complete sequenceAF527822.11481--1717Esp1396IRecently annotated C protein1481--171709Lpn060Motif 6^b^Legionella pneumophila str. Corby, complete genomeCP000675.1226951--227193LpnCMrrPRestrictase233982--234959678910Cef187Motif 2Corynebacterium efficiens plasmid pCE3 DNA, complete sequenceAP005226.116363--16605CefpCE3MrrPRestrictase10596--11558504711Gur068Motif 2Geobacter uraniumreducens Rf4, complete genomeCP000698.11337269--1337499GurRORF1135PHTH-domain protein1337269--1337499012Pgi032Motif 9Porphyromonas gingivalis W83, complete genomeAE015924.1595797--595997PgiTORF544PMethylase596192--59813819513Bth033Motif 9Bacteroides thetaiotaomicron VPI-5482, complete genomeAE015928.15932179--5932379BthVORF4518PRestrictase5934256--5936961187714Nha244Motif 7Nitrobacter hamburgensis X14, complete genomeCP000319.1883322--883612NhaXORF803PMethylase885888--8871982276 Regulatory mechanisms --------------------- C proteins bind DNA as dimers ([@B5],[@B6]). Thus, their binding sites should be palindromic. In addition, in all cases that have been investigated experimentally, two adjacent C protein dimer binding sites are present in the regulatory regions of the C protein genes. The upstream site has a higher affinity for the C protein dimer and the interaction with this site activates transcription of the C protein gene ([@B6],[@B10]). The downstream site has lower affinity for C protein dimer and the interaction with this site decreases transcription of the C protein gene. These general considerations lead to certain constraints that putative C protein-binding sites should conform to (assuming that the regulatory mechanisms in the newly identified cases are similar to those already described). A complete C protein-binding site should contain a direct repeat of a palindromic sequence. Cooperative interaction of a C protein with its two binding sites necessitates that a distance between the binding sites is conserved. In C.PvuII-like motifs 1--6, a highly conserved GTG trinucleotide is observed between the putative palindromic operators. The special geometry and/or bendability of this trinucleotide may promote cooperative interactions between bound C protein dimers. Indeed, mutational and structural analysis of C protein interactions with two motifs of this group ([@B16],[@B20]), confirmed the importance of this trinucleotide in cooperative C protein binding and the existence of a bend in this sequence when bound by two C protein dimers. The analysis presented above demonstrates that four out of six C.PvuII-like motifs (1, 2, 3 and 4) contain two copies of the operator ([Figure 3](#F3){ref-type="fig"}a). Figure 3.(**a**) Logos of 1--4 motifs' consensus sequences. Palindromic boxes are marked with light green squares and underlined with colored arrows. The upper logo represents the 5′ (distal) copy, while the lower logo represents the 3′ (proximal) copy. (**b**) Logos of palindromic (7, 8 and 10) motifs' consensus sequences. Palindromic boxes are marked with light green squares and underlined with colored arrows. The upper logo corresponds to annotated binding sites, while the lower logo corresponds to their weak downstream copies. Motifs 7, 8 and 10, as defined by our procedure, are single palindromes. One of the reasons why the second operator was not identified could be that the initial procedure implied a fixed distance between the operators. Thus, we used scanning of the 100 bp regions upstream of putative C protein genes with profiles constructed from already identified sites, and also performed additional manual searches for conservation islands. This resulted in identification of additional, downstream conservation islands containing weaker copies of the same motif ([Figure 3](#F3){ref-type="fig"}b). As expected, no preferred length of the spacer between the sites could be seen (data now shown). In several cases that have been studied in experiment, the C protein RNA is leaderless ([@B10],[@B18]), i.e. the transcription start site coincides with or is located 1--3 nucleotides upstream of adenosine (or guanosine) of C protein ORF initiating ATG (or GTG) codon. Presumably, the less efficient translation of a leaderless message ([@B30]) causes a delay in C protein-dependent activation of transcription of the toxic restriction endonuclease gene located downstream of the C protein gene ([@B31]). In the cases where the existence of leaderless C protein mRNA has been experimentally demonstrated, the distance between the C protein-binding site and the C protein ORF initiating codon is necessarily short, the typical spacer length being 18 nt ([Figure 4](#F4){ref-type="fig"}a). We considered all instances of sites forming motifs 1--6 and calculated the distance between the motif and the C protein start codon. The results are presented in [Figure 4](#F4){ref-type="fig"}b, where, indeed, a very strong maximum at a distance of 17--18 nt between the C protein gene start codon and the putative C protein-binding site is evident. We take this result as a strong indication that leaderless translation is a common feature of C protein regulation. Figure 4.(**a**) The structure of a region upstream of a typical C.PvuII-like C protein gene. The binding site and the ATG start codon are marked with black color. The palindromic elements of the site are underlined. (**b**) The histogram of distances between the candidate binding sites and start codons of C protein genes. Only C.PvuII-like motifs (1, 2, 2b, 2c, 3, 4, 5, 6, 6b) are considered. Horizontal axis and numbers above the bars: distances, vertical axis: frequency of such distance. The presence of a pool of C protein genes whose mRNA is translated through a leaderless mechanism prompted us to bioinformatically search for conserved promoter elements upstream of these genes, since the distance between the promoter element and transcription start (defined by the position of the initiating codon) should be fixed. However, we failed to identify any reliable --10 promoter consensus elements in sequences preceding initiation codons of apparently leaderless putative C protein genes, nor did not observe any over-representation of TATAAT-like hexamers in the region preceding the putative transcription start point areas of these genes (data not shown). This agrees with the absence of easily identifiable --10 promoter elements in known *CR* operon promoters ([@B6],[@B10],[@B20]) and supports a model, which posits that these promoters are weak and require C protein binding for activity. DISCUSSION ========== Using a combination of phylogenetic footprinting and bioinformatic motif searches, we have identified 201 putative C protein-binding sites, 181 (90.0%) of which fall into ten distinct motifs. The remaining binding sites do not belong to motifs, however, most of them resemble C.PvuII-like motifs ([@B1; @B2; @B3; @B4; @B5; @B6]): the sites are also 35 bp long, some sites contain the central GTGG tetranucleotide and some of them contain self-complementary trinucleotides at the outside flanks. The genes of *Rebase*-annotated C proteins are preceded only by sites belonging to motifs 2, 7, 8 and 9. Among REBASE R--M systems with previously unannotated C proteins, we observed three C protein-homologous genes preceded by sites from other motifs: motif 1 (Pst154), motif 4 (Nwi091) and motif 6^b^ (Lpn060). Thus, the apparent limitation in the kinds of C protein-binding sites present in *bona fide* R--M systems is likely because of a bias in experimental analysis centered on a number of close homologs rather than some biological reason. Still, there remains a possibility that a group of identified candidate transcription factors from the C protein family are not involved in the regulation of R--M systems. One hundred and sixty nine sites belong to new motifs that have not been described previously. Despite the fact that we have not used experimental data during our searches, all eight experimentally verified C protein-binding sites were identified correctly, indicating that our search procedure is robust. C proteins with putative binding motifs from the same class typically cluster in the phylogenetic tree ([Figure 1](#F1){ref-type="fig"}). The few exceptions could be caused by both low reliability of deep branches of the tree or by *bona fide* convergent evolution of the motif. A more detailed analysis of co-evolution of C proteins and their binding motifs is required to characterize the molecular events and their structural consequences in detail. This work is currently ongoing in our laboratories. The motifs identified in this work can be subdivided into two types, and the first type consists of two subtypes. The C.PvuII-like motifs 1, 2, 3 and 4 are characterized by rich symmetries. This group of motifs comprises two palindromic operators separated by highly conserved spacers and framed by highly conserved complementary trinucleotides. The related motifs 5 and 6 retain some features of the former subtype (conserved nucleotides in the middle, short inverted repeats at the termini), but do not contain operators. The symmetry and conservation beyond the palindromic operators suggest that there exist additional functional and/or structural forces shaping the motif. Indeed, the observed pattern of conservation is in good agreement with the recently published structure of a complex of two dimers of the C.Esp1396I protein bound to the binding site (motif 2) ([@B16]). In the structure, the outside complementary trinucleotides Z-Z\* (A)TG-CA(T) form extensive contacts with the protein. Similarly, the highly conserved, non-symmetrical central dinucleotide (G)TG also contacts the protein. In contrast, no direct interactions with palindromic operators are evident, though the recognition helix of the helix-turn-helix motif of each C.Esp1396I monomer is positioned in the major grove of operator half-sites. The occurrence of a large number of highly conserved positions that do not appear to form contacts with the protein makes it likely that factors such as structural constraints on the DNA or additional modes of protein binding (i.e. as single dimers or even monomers) may be involved in shaping the motif. The second major type of sites is formed by palindromic motifs 7, 8 and 10, each consisting of a single operator with a downstream weaker copy located at variable distance from the initially identified one. On the basis of the only biochemically studied example, of C.EcoRV ([@B10]), motif 7, interactions of C proteins with such binding sites are not characterized by highly cooperative interactions between C protein dimers observed with C.PvuII-like sites with fixed distances between the operators. Although genomic contexts of some identified candidate C proteins contain genes encoding putative restriction endonucleases and methylases, in the majority of cases no such genes could be identified. On the other hand, the immediate vicinity of many candidate C proteins genes contains phage- or transposon-related genes such as resolvases, integrases, transposases, recombinases and other genes annotated as phage-, plasmid- and conjugation-related. This leaves open a possibility that some of identified C protein-family factors regulate functions other than restriction-modification. Although C proteins share distant homology and structural similarity with phage repressors ([@B15]), our database search was sufficiently restrictive, and the pool of putative C proteins was not contaminated with *bona fide* phage repressors. Despite the lack of a universal association of putative C protein genes with R--M systems, essential features of autoregulatory loops controlled by characterized C proteins from known R--M systems such as (i) the presence of two binding sites, a high affinity one leading to activation of transcription of C protein gene as well as any gene that is coupled to it, and another, low affinity site whose occupancy leads to transcriptional repression and (ii) translation from leaderless transcripts appears to be a common feature of at least C protein genes associated with motifs 1--6. In R--M systems, these features allow highly regulated, time-delayed expression of the highly toxic restriction endonuclease. It remains to be determined whether newly identified C protein genes that are not associated with any R--M systems also control expression of genes toxic to the cell. SUPPLEMENTARY DATA ================== [Supplementary Data](http://nar.oxfordjournals.org/cgi/content/full/gkn931/DC1) are available at NAR Online. FUNDING ======= This study was partially supported by grants from the Howard Hughes Medical Institute (55005610 to M.G.), the Russian Foundation of Basic Research (09-04-01098-a to V.S.) and National Intitutes of Health (RO1 GM59295 to K.S.). M.G. and K.S. are partially supported by grants from the program 'Molecular and Cellular Biology' of the Russian Academy of Sciences. *Conflict of interest statement*: None declared. We are grateful to Anna Karyagina, Andrei Alexeevsky and Sergei Spirin for useful discussions.

Background {#Sec1} ========== Pattern recognition receptors (PRRs) are well known to function in the early stages of the innate immune response and recognize pathogen-associated molecular pattern (PAMPs) from external microorganisms or damage-associated molecular patterns (DAMPs) from damaged tissues \[[@CR1]\]. There are four types of PRRs, including C-type lectin receptors (CLRs), RIG I-like receptors (RLRs), NOD-like receptors (NLRs), and Toll-like receptors (TLRs) \[[@CR2]--[@CR4]\]. TLRs recognize various ligands (e.g. DNA, RNA, proteins), and activate intracellular signaling pathways (e.g., MAPK, NF-кB) and produce pro-inflammatory cytokines in antigen presenting cells (APCs), e.g., dendritic cell (DCs) \[[@CR5]--[@CR7]\]. Among the TLRs, TLR4 is the mostly well-studied, and it was the first TLR found in humans that can recognize lipopolysaccharide (LPS), a component present in gram-negative bacteria \[[@CR8]\]. Its ligands also include HMGB1, heat shock proteins, fusion proteins, as well as other proteins \[[@CR2]\]. DCs express TLRs to which numerous ligands can bind, thereby inducing the expression of co-stimulatory molecules and the secretion of pro-inflammatory cytokines \[[@CR9]\]. DCs have been shown to be important not only in the innate immune response, but also in the adaptive immune response where mature DCs can present uptaken antigen to naïve T cells leading to their activation \[[@CR10], [@CR11]\]. DC-based cancer vaccines have been produced, and of the different cancer vaccines available for use in humans (antigen/adjuvant vaccines, anti-idiotype vaccines, DNA vaccines, tumor cell vaccines) are the most preferred \[[@CR12]--[@CR14]\]. As an example, Provenge, a DC-based cancer vaccine drug, is the only such vaccine approved by the FDA \[[@CR15], [@CR16]\]. In order to generate DC-based cancer vaccines, an adjuvant that binds to TLRs that can induce maturation and activation of DCs is needed. It is well known that LPS, a TLR4 ligand, has a significant effect on inducing maturation and activation of DCs through the MyD88 and TRIF cell signaling pathways \[[@CR17]\]. However, LPS is not appropriate for use in humans because it is an endotoxin, which is derived from the cell walls of gram-negative bacteria. Instead, the TLR9 ligand CpG, or the TLR3 ligand I:C, are commonly used, but require a large amount of CpG or I:C to induce DC maturation and activation. Therefore, we need to find a new adjuvant that can bind to TLR4 to effectively induce DC maturation. To identify an efficient adjuvant that binds to TLR4, intracellular proteins in human tumor cells were screened through a pull-down assay. About twenty TLR4-binding proteins were found, half of which were ribosomal protein families. These ribosomal proteins were purified and tested for their ability to induce maturation and activation of DCs. Among the ribosomal proteins studied, 40S ribosomal protein S3 (RPS3) was identified as a possible adjuvant for use in DC-based vaccines. RPS3 is a component of the 40S/60S ribosome. RPS3 is involved in the regulation of apoptotic signaling pathways and the control of gene expression \[[@CR18], [@CR19]\]. Specifically, it has been shown that DNA damage is repaired and that NF-κB driven gene expression is activated after binding to the KH domain, the DNA binding site, that resides at the N-terminal of RPS3 (a.a. 21--92). Moreover, RPS3 has also been shown to bind to Hsp70, Hsp90, several enzymes, and kinases, as well as other proteins \[[@CR20]--[@CR27]\]. In addition to the many intracellular functions of RPS3, it has been shown that RPS3 is secreted after its N-linked glycosylation, although how it acts extracellularly is not known \[[@CR28]\]. In the present study, we screened RPS3 from human tumor cells as a new TLR4 ligand. We examined the effects of RPS3 on maturation and activation of mouse and human dendritic cells in vitro. Furthermore, DC vaccination with RPS3 and tumor cell specific peptides confirmed the utility of RPS3 as an adjuvant in vivo. In addition, we showed that the adjuvant effects of RPS3 were dependent on TLR4. Methods {#Sec2} ======= Mice {#Sec3} ---- Female C57BL/6 mice, 6--8 weeks of age, were purchased from Orient Bio (Seongnam, Korea). TLR4 knockout mice (C57BL/10ScNF) were purchased from Jackson Laboratories (Maine, USA). All mice were kept under specific pathogen-free conditions in accordance with the animal care guidelines approved by the Institutional Animal Care and Use Committee (IACUC) of Konkuk University. Cells {#Sec4} ----- TC-1 is a transformed mouse lung epithelial cell expressing the HPV-16 E7 gene \[[@CR29]\]. The human cell lines CFPAC-1, SNU17, and SKOV3, as well as the mouse lymphoma EG.7 cell line (mouse EL4 cells transfected with the OVA cDNA) were obtained from ATCC (Virginia, USA). The mouse melanoma cancer cell line B16F10 was obtained from the Korean Cell Line Bank. Cells were incubated at 37 °C in an atmosphere of 5% CO~2~ and cultured in RPMI-1640, DMEM, or IMEM (Biowest, Rue de la Caille, France) medium supplemented with 10% fetal bovine serum (Biowest, Rue de la Caille, France) and 50 U/mL penicillin streptomycin (Biowest, Rue de la Caille, France). HEK293 cells expressing hTLR4-MD2-CD14 were cultured according to the protocol provided by InvivoGen (CA, USA). Human monocytic THP-1 cells were differentiated by treating with PMA (200 μg/mL) (Sigma, Missouri, USA) \[[@CR30]\]. Mouse dendritic cells were isolated from mice bone marrow and cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum, 50 U/mL penicillin streptomycin, 1000 μM 2-mercaptomethanol (Gibco, MA, USA), recombinant mouse IL-4, and granulocyte-macrophage colony-stimulating factor (JW CreaGene, Seongnam, Korea). The culture medium was changed every two days and the cells were differentiated for six days. After DC differentiation, CD11c + DCs were used in our experiments after gating through Flow Cytometry. Human DCs were differentiated after the isolation of monocytes from peripheral blood mononuclear cells (PBMCs). And due to the excess of number of characters, the detailed protocols are described in supplementary data (Additional file [1](#MOESM1){ref-type="media"}). Purification of recombinant RPS3 protein {#Sec5} ---------------------------------------- The RPS3 expression vector was constructed from a cDNA derived from the human pancreatic adenocarcinoma CFPAC-1 cell line. The insert DNA along with cloning sites (NdeI and NotI) was synthesized using a CFPAC-1 cDNA and the RPS3 primer (forward primer; 5′-CATATG GCAGTGCAAATATCCAAGAAG-3′, reverse primer; 5′-GCGGCCGC TTATGCTGTGGGGACTGG-3′). This DNA was then inserted into the pET28b bacterial expression vector. Two fragments of the recombinant RPS3 protein (1--95 and 91--243) were synthesized using the 95-amino acid reverse primer (5′-GCGGCCGC TTAACCTCTAGTGGCCACCTT-3′) and the 91-amino acid forward primer (5′-CATATG GTGGCCACTAGAGGTCTGTGT-3′). The detailed protocols are described in supplementary data (Additional file [1](#MOESM1){ref-type="media"}). Luciferase assay {#Sec6} ---------------- Luciferase activity was assessed using the Dual-Glo Luciferase Assay System (Promega, Wisconsin, USA). The pGL4.32\[luc2P/NF-κB-RE/Hygro\] vector (Promega) and the pRL-TK vector (Promega) were transfected into 293/hTLR4A-MD2-CD14 cells to assess the activation of NF-κB activity by the RPS3 protein. Cells (2 × 10^4^/well were seeded into a 96-well white plate (SPL, Pocheon, Korea) and then transfected with each of the two plasmids. Cells were then incubated at 37 °C, after 16 h, RPS3 (1 μg/mL), GFP (5 μg/mL), or LPS (100 ng/mL) were added to the culture and the incubation was continued for a further 2 h. Firefly luminescence, as a measure of NF-κB activity value and Renilla luminescence as a normalization value, were measured according to the protocol provided by the assay kit manufacturer (Promega). Luciferase activity was expressed as the ratio of firefly luminescence:Renilla luminescence for each well. Enzyme-linked immunosorbent assay (ELISA) {#Sec7} ----------------------------------------- The levels of secreted cytokines derived from mouse BMDCs and THP-1 cells were measured using ELISA kits and followed the individual ELISA kit's manufacturer's instructions for mouse and human TNF-α, IL-1β, IL-6, IL-10, IL-12p70 (ebioscience, MA, USA) and IFN-β (pbl). The levels of cytokines (IL-10 and IL-12p70) secreted from human DCs were determined using ELISA kits obtained from BD Bioscience (New Jersey, Korea). Flow cytometry analysis {#Sec8} ----------------------- Mouse BMDCs, or THP-1 cells, were stained with the following surface markers; FITC-CD11c, PE-CD40, PE-CD80, PE-86, PE-MHC-I, PE-CD83, or PE-CCR7 (BioLegend). Human DCs were stained with FITC-labeled anti-HLA-DR, PE-labeled anti-CD80, and FITC-labeled anti-CD86 (Miltenyi). Cell activation was analyzed using a FACSCalibur with CELLQuest (Becton Dickinson, New Jersey, USA). To assess CD8^+^ T cell generation in vivo, splenocytes from treated or vaccinated mice were stained with PE-labeled anti-CD8a (BioLegend) and FITC-labeled anti-IFN-γ (BioLegend). FITC-labeled IFN-γ staining was performed using an intracellular staining kit (BD Bioscience). TLR4 blocking antibody {#Sec9} ---------------------- MAb mTLR4/MD2 (InvivoGen) was pre-treated on mouse BMDCs 1 h before the treatment of RPS3 of LPS. Western blotting {#Sec10} ---------------- Mouse BMDCs were treated with RPS3 (1 μg/mL) over a time course (0, 10, 20, 30, 40, 50, and 60 min) and the cells harvested. Cells were lysed using RIPA buffer (0.5% NP-40, 1 mM EDTA, 50 mM Tris-Cl pH 8.0, 120 mM NaCl, protease inhibitor cocktail, 100 mM PMSF, 0.5 M NaF) on ice for 1 h. The detailed protocols are described in supplementary data (Additional file [1](#MOESM1){ref-type="media"}). Immunoprecipitation {#Sec11} ------------------- His-tagged human recombinant TLR4 was bound to the Ni-NTA resin at 4 °C using a binding buffer of 1× PBS (pH 7.4). After 12 h, the TLR4-Ni-NTA beads were washed five times with PBS, and then the B16F10 tumor cell supernatant was added and the incubation continued for 12 h at 4 °C. The B16F10 tumor cells were treated with 10 μg/mL of doxorubicin. After 2 h, the media was changed to Opti-MEM (Gibco, MA, USA). Two days later, 500 μL of the supernatant was collected and concentrated to 100 μL using a Centricon filter (MERK, Darmstadt, Germany). Following this the beads were washed five times with PBS. Residual buffer was then removed from the resin, and 50 μL of 5 × SDS loading buffer was then added and the beads were boiled. The eluted proteins were separated by electrophoresis on 10% SDS-PAGE gels and the separated protein transferred to a PVDF membrane. In vivo experiments {#Sec12} ------------------- Tumor antigen-specific CD8^+^IFN-γ^+^ T cell generation, tumor prevention, and treatment experiments were performed using mice injected with different cell formulations: no vaccination, PBS-treated immature DCs, immature-DCs loaded with tumor antigen peptide, RPS3-treated mature DCs, RPS3-treated mature DCs loaded with tumor antigen peptide, and LPS-treated mature DCs loaded with tumor antigen peptide. Mice were vaccinated with BMDCs (2 × 10\^6) on the footpad twice at intervals of one week. The tumor-specific antigen peptides OVA (SIINFEKL) and E7 (RAHYNIVTE) (Anygen, Gwangju, Korea) were used. Tumor antigen-specific CD8^+^IFN-γ^+^ T cell generation in the mouse spleens was measured using flow cytometry. The tumor prevention experiment was performed by subcutaneous injection of mouse cancer cells (5 × 10^6^ EG.7 cells, 2.5× 10^5^ TC-1 cells) seven days after the last DC injection. In the tumor treatment experiments, cancer cells (1 × 10^6^ EG.7, 2 × 10^5^ TC-1) were injected subcutaneously. After five days, BMDCs were injected and the tumor size was measured every 2 to 3 days. The size of the tumor mass was calculated using the formula (length × width × width × 1/2). Measurement of tumor size and mouse survival was performed until the tumor diameter was over 2 cm or the mice were dead. Mice were vaccinated with RPS3-treated BMDCs pulsed with OVA twice at intervals of one week. The memory T cell experiment was performed using the following injection groups: no vaccination with either tumor injection or not, and vaccination with either tumor injection or not. Seven weeks from the last BMDC injection, 1 × 10^6^ EG.7 tumor cells were injected subcutaneously. Seven days later, CD8^+^IFN-γ^+^ T cells generation in the spleen was measured by flow cytometry and the mice were observed until the tumor diameter was over 3 mm. \(A\) Identification of intracellular RPS3 in human and mouse tumor cells by western blot. (B) Human and mouse tumor cells were treated with doxorubicin (10 μg/mL). After 2 h, the media was changed to Opti-MEM media and the cells were incubated overnight. Release of RPS3 into the culture supernatant was assessed by western blot. (C) After doxorubicin treatment (10 μg/mL) of B16F10 cells, the binding of released RPS3 in the culture supernatant and recombinant TLR4 was confirmed by immunoprecipitation. (D) Purification of recombinant RPS3 from *E. coli* assessed by Coomassie Brilliant Blue (CBB) staining and western blotting. (E) TLR4-MD2 expressing HEK293 cells were treated with recombinant RPS3 (0.01, 0.1, 1 μg/ml), GFP (5 μg/mL) or LPS (100 ng/mL) and NF-кB activity was measured by luciferase assay (\*\*; *p* \< 0.01). (F) The binding affinity between recombinant RPS3 (1 μg) or LPS (1μg) and TLR4 (1 μg) was assessed using BLITz. All the experiments were performed three times. Mouse BMDCs were treated with RPS3 (0.01, 0.1, or 1 μg/mL), GFP (5 μg/mL), or LPS (100 ng/mL) and (A) pro-inflammatory cytokines (TNF-α, IL-1β, IL-6, IL-10, and IL-12p70), as well as IFN-β secretion levels, in the cell culture supernatant were measured by ELISA (\*; *p* \< 0.05, \*\*; p \< 0.01 and \*\*\*; *p* \< 0.001). (B) The expression of co-stimulatory molecules (CD40, CD80, CD86, and MHC-I) in CD11c + BMDCs were assessed by flow cytometry. (C) Mouse BMDCs were treated with RPS3 protein (1 μg/mL) for different periods of time (0, 10, 20, 30, 40, 50, or 60 min) and activation of the MAPK (ERK, P38, and JNK), AKT and NF-κB cell signaling pathways were analyzed by western blot. All the experiments were performed three times. All cells were treated with RPS3 (0.1, 1 μg/mL), GFP (5 μg/mL) or LPS (100 ng/mL). (A) The expression levels of co-stimulatory molecules (CD80, CD83, and CD86) and CCR7 in human THP-1 cells were analyzed by flow cytometry. (B) The levels of cytokines (TNF-α, IL-6, IL-10, and IFN-β) secreted from human THP-1 cells were measured by ELISA. (C) The expression of co-stimulatory molecules (HLA-DR, CD80, and CD86) in human DCs were analyzed. (D) The levels of cytokines (IL-10 and IL-12p70) secreted from human DCs were measured. All the experiments were performed three times. (\*; *p* \< 0.05, \*\*; *p* \< 0.01 and \*\*\*; *p* \< 0.001). (A, B) Mouse BMDCs were separated into the following groups; no vaccination, PBS-treated immature DCs, immature DCs loaded with tumor antigen peptide, RPS3-treated mature DCs, RPS3-treated mature DCs loaded with tumor antigen peptide, and LPS-treated mature DCs loaded with tumor antigen peptide. DCs were treated with RPS3 (1 μg/mL) or LPS (100 ng/mL) and then pulsed with OVA (A) and/or E7 (B) tumor specific antigen peptides. The treated DCs were then injected into the footpads of mice. After 1 week the last injection of BMDCs, the spleen was isolated and the number of tumor specific CD8^+^IFN-γ^+^ T cells produced was counted by Flow Cytometry (C, D). After vaccination by the different DCs, EG.7 (C) or TC-1 (D) cells were subcutaneously injected and tumor formation was observed for 60 days. All the experiments were performed three times. IBM SPSS Statistics Base 22.0 was used as statistical method to analyze differences between experimental groups. (\*\*; *p* \< 0.01 and \*\*\*; *p* \< 0.001) (A, B) EG.7 (A) or TC-1 (B) tumor cells were injected subcutaneously into mice. Following this, mouse BMDCs were separated into the following groups: no vaccination, PBS-treated immature DCs, immature DCs loaded with tumor antigen peptide, RPS3-treated mature DCs, RPS3-treated mature DCs loaded with tumor antigen peptide, and LPS-treated mature DCs loaded with tumor antigen peptide. Five days after tumor injection, the tumor size was measured and mouse survival was observed until the tumor diameter was over 2 cm or the mouse had died. (C, D) Mouse BMDCs were separated into the following groups: no vaccination with either tumor injection or not and vaccination with either tumor injection or not. Seven weeks after the last BMDC injection, EG.7 tumor cells were injected subcutaneously into mice. The spleen was isolated from mice and the number of tumor-specific memory T cells produced was determined by Flow Cytometry (C) and tumor formation was observed (D). All the experiments were performed three times. IBM SPSS Statistics Base 22.0 was used as statistical method to analyze differences between experimental groups. (\*\*; *p* \< 0.01 and \*\*\*; *p* \< 0.001) (A, B) WT and TLR4−/− mouse BMDCs were treated with RPS3 (1 μg/mL), GFP (5 μg/mL), or LPS (100 ng/mL) and the levels of secreted cytokines (TNF-α, IL-1β, IL-6, IL-10, IL-12p70, and IFN-β) as well as the expression of co-stimulatory molecules (CD40, CD80, CD86, and MHC-I) were measured. (C) RPS3 (1 μg/mL) was used to treat WT and TLR4−/− mouse BMDCs for different periods of time (0, 10, 20, 30, 40, 50, and 60 min) and the activation of the MAPK (ERK, P38 and JNK), AKT, and NF-κB cell signaling pathways were analyzed by western blot. Total MAPK and AKT levels were used as normalization controls. (D) WT or TLR4−/− mouse DCs were treated RPS3 (1 μg/mL) and then pulsed with the E7 tumor antigen peptide. Following this the DCs injected were injected into the footpads of mice. The number of tumor-specific CD8^+^IFN-γ^+^ T cells was then determined by flow cytometry. (E) After vaccination with the prepared WT or TLR4−/− DCs, TC-1 cells were subcutaneously injected and tumor mass and mouse survival were analyzed. All the experiments were performed three times. IBM SPSS Statistics Base 22.0 was used as statistical method to analyze differences between experimental groups. (\*; *p* \< 0.05, \*\*; *p* \< 0.01 and \*\*\*; *p* \< 0.001) \(A\) Schematic showing the two fragments of the RPS3 protein (1--95 and 91--243). (B) The two His-tagged fragments of the RPS3 protein was assessed for purity and size by western blot. (C) The different forms of the RPS3 protein (WT, 1--95, and 91--243; 1 μg/mL), GFP (5 μg/mL), or LPS (100 ng/mL) were used to treated 293/hTLR4A-MD2-CD14 cells, after which luciferase activity was measured. (D, E) The different forms of the RPS3 protein (WT, 1--95, and 91--243) (1 μg/mL), GFP (5 μg/mL), or LPS (100 ng/mL) were used to treat mouse BMDC and the levels of secreted cytokines (TNF-α, IL-1β, IL-6, IL-10, IL-12p70 and IFN-β), as well as the expression of co-stimulatory molecules (CD40, CD80, CD86, and MHC-I) were measured. All the experiments were performed three times. IBM SPSS Statistics Base 22.0 was used as statistical method to analyze differences between experimental groups. (N.S.: not significant, \*; *p* \< 0.05, \*\*; *p* \< 0.01 and \*\*\*; *p* \< 0.001) Results {#Sec13} ======= Release of RPS3 from tumor cells and the binding of recombinant RPS3 from *E. coli* to recombinant TLR4 {#Sec14} ------------------------------------------------------------------------------------------------------- To identify protein candidates in human cancer cells that can associate with TLR4, we screened human cancer cells using a luciferase assay and three cancer cell lines were selected in which NF-kB activity could be observed (Additional File [2](#MOESM2){ref-type="media"}: Figure S1). Following this, lysates from three cancer cells were used in pulled-down experiments with recombinant TLR4 (Additional File [2](#MOESM2){ref-type="media"}: Figure S2). Among the various ribosomal protein families that were found to bind to TLR4, ribosomal protein S3 (RPS3) was selected for use in our experiments because it had the greatest effects when used to treat BMDCs. An initial experiment revealed that RPS3 is expressed in various cancer cells (Fig. [1](#Fig1){ref-type="fig"}A). Furthermore, RPS3 was released from not only B16F1 and B16F10 tumor cells (Fig. [1](#Fig1){ref-type="fig"}B) but also normal cells like BMDCs (Additional File [2](#MOESM2){ref-type="media"}: Figure S3) when they were treated with doxorubicin and the released RPS3 could bind to TLR4 (Fig. [1](#Fig1){ref-type="fig"}C). SKOV3 supernatant does not seem to release RPS3 due to merely effects of doxorubicin occurring cell death in SKOV3 compared to other tumor cells (Fig. [1](#Fig1){ref-type="fig"}B). Recombinant RPS3 was then purified (Fig. [1](#Fig1){ref-type="fig"}D) and the interaction between RPS3 and TLR4 was confirmed by the increased luciferase activity seen after TLR4-MD2 cells were treated with RPS3 (Fig. [1](#Fig1){ref-type="fig"}E). Using a blitz assay, the Kd (M) value for the RPS3 and TLR4 interaction was lower (i.e. of higher affinity) than the BSA and TLR4 interaction (Fig. [1](#Fig1){ref-type="fig"}F). It is concluded that RPS3 is released from tumor cells when treated with an anticancer drug and that RPS3 can associate with TLR4.Fig. 1The association of TLR4 with released or recombinant RPS3 RPS3 induces maturation and activation of mouse BMDCs, THP-1 cells, and human DCs {#Sec15} --------------------------------------------------------------------------------- To confirm the effects of RPS3 which binds to TLR4, we used recombinant RPS3 from E.coli in all experiments due to productivity and convenience. Firstly, we determined the effects of RPS3 on mouse BMDCs by characterizing the expression of pro-inflammatory cytokines and IFN-β. The levels of TNF-α, IL-1β, IL-6, IL-10, IL-12p70, and IFN-β were all significantly increased in a dose-dependent manner in DCs treated with RPS3 compared to untreated DCs (Fig. [2](#Fig2){ref-type="fig"}A). In addition, the expression levels of co-stimulatory molecules CD40, CD80, CD86, and MHC class I were also increased in DCs treated with RPS3 compared to untreated DCs (Fig. [2](#Fig2){ref-type="fig"}B). To evaluate the activation of signaling pathways, mouse BMDCs were treated with RPS3 in a time course over an hour. The levels of p-ERK, p-P38, and p-JNK and p-AKT were all found to be elevated (Fig. [2](#Fig2){ref-type="fig"}C). To assess NF-κB activation, we measured the degradation of the inhibitory IκB-α protein (Fig. [2](#Fig2){ref-type="fig"}C). The activation of DCS by RPS3 was carried out using protein purified from *E. coli* in such a way that endotoxins were removed from the preparation. As a negative control, GFP prepared in an identical manner was used; LPS was used as a positive control since it is a known TLR4 ligand. We assessed LPS contamination of the RPS3 protein preparation used to treat DCs and showed that it was not as a result of endotoxin contamination from *E. coli*. Treatment with polymyxin B (PMB) decreased the effects of LPS in DCs, whereas the effects of RPS3 in DCs were not affected but instead were decreased by proteinase K treatment (Additional File [2](#MOESM2){ref-type="media"}: Figure S4). These results indicate that RPS3 can induce the secretion of pro-inflammatory cytokines, increase the expression of co-stimulatory molecules, and activate signaling pathways in mouse BMDCs.Fig. 2Maturation and activation of mouse dendritic cells by recombinant RPS3 Next, we assessed the effects of RPS3 on the human monocytic cell line THP-1, as well as human DCs. As shown in Fig. [3](#Fig3){ref-type="fig"}A, the expression levels of CD80, CD83, CD86, and the migration factor CCR7 were all increased in a dose-dependent manner in THP-1 cells treated with RPS3 compared to untreated THP-1 cells. Moreover, the secretion levels of the pro-inflammatory cytokines TNF-α, IL-6, IL-10, and IFN-β were increased in THP-1 cells treated with RPS3 compared to untreated THP-1 cells (Fig. [3](#Fig3){ref-type="fig"}B). Likewise, the expression of the co-stimulatory molecules MHC class II, CD80, and CD86, as well as the of the pro-inflammatory cytokines IL-10 and IL-12, were all increased in RPS3-treated human DCs (Fig. [3](#Fig3){ref-type="fig"}C and D). Interestingly, the expression levels of co-stimulatory molecules and the secretion levels of pro-inflammatory cytokines were higher in RPS3-activated THP-1 cells and human DCs than in LPS-activated THP-1 cells or human DCs. These results suggest that RPS3 can induce the activation and maturation of human monocytes and human DCs.Fig. 3Maturation and activation of human monocytic THP-1 cells and human dendritic cells by RPS3 protein RPS3-activated BMDC vaccine produces antigen-specific CD8^+^ T cells in vivo and leads to tumor prevention, effective tumor treatment, and the generation of memory T cells {#Sec16} --------------------------------------------------------------------------------------------------------------------------------------------------------------------------- We next assessed the effects of mature DCs activated by RPS3 as a DC vaccine to assess whether RPS3-activated DCs can generate antigen-specific CD8^+^ T cells. Mouse BMDCs (2 × 10\^6) were treated with RPS3 or LPS after which the DCs were pulsed with a tumor specific peptide and then injected into the footpad of mice once a week for two weeks. Following this, the number of IFN-γ secretory CD8^+^ T cells was measured. As shown in Fig. [4](#Fig4){ref-type="fig"}A and B, tumor specific CD8^+^IFN-γ^+^ T cells were significantly generated in a mouse vaccinated with RPS3-activated DCs pulsed with either OVA or E7 antigenic peptides, as compared to mice vaccinated with immature DCs or non-peptide pulsed mature DCs (OVA and E7 are well known antigens expressed in EG.7 and TC-1 cells respectively). Next, we assessed if these RPS3-activated DC vaccines could prevent tumorigenesis. The RPS3-activated DCs were pulsed with the tumor specific antigens OVA or E7 and were then injected into the footpads of mice. After the last vaccination, EG.7 or TC-1 cells were then injected subcutaneously into the mice. As shown in Fig. [4](#Fig4){ref-type="fig"}C and D, all the mice vaccinated with RPS3-activated DCs pulsed with the OVA or E7 peptides remained tumor-free for 60 days. In contrast, tumor formation occurred within 15 days in mice vaccinated with immature DCs pulsed with peptide or mature DCs that were not pulsed with peptide or mature DCs pulsed with non-specific peptide (Additional File [2](#MOESM2){ref-type="media"}: Figure S5 A). Interestingly, the RPS3-activated DCs pulsed with OVA or E7 peptide vaccine had a slightly greater effect than the LPS-activated DCs pulsed with OVA or E7 peptide vaccine.Fig. 4In vivo study showing tumor specific CD8^+^IFN-γ^+^ T cell induction and tumor prevention effects with a DC-based cancer vaccine using the RPS3 protein as adjuvant Next, we evaluated the effect of the RPS3-treated DC vaccine on tumor treatment. EG.7 or TC-1 cells were injected subcutaneously into mice, and after establishment of a tumor mass on day 5, RPS3-treated DCs pulsed with OVA or E7 were used to vaccinate the mice using a footpad inoculation once a week for two weeks. As shown in Fig. [5](#Fig5){ref-type="fig"}A and B, a suppression of tumor growth and the long-term survival was seen in mice injected with RPS3-treated DCs pulsed with OVA or E7 compared to mice injected with untreated DCs or RPS3-treated DCs not pulsed with peptide. More importantly, the tumor treatment effects in mice vaccinated with the RPS3-treated DCs pulsed with OVA or E7 were greater than in mice vaccinated with LPS-treated DCs pulsed with OVA or E7 (Fig. [5](#Fig5){ref-type="fig"}A and B). To understand the function of memory T cells, the tumor cells were injected seven weeks after DC vaccination. The number of CD8^+^IFN-γ^+^ T cells was dramatically increased in mice which had been vaccinated with BMDCs treated with RPS3 and pulsed with the OVA peptide (Fig. [5](#Fig5){ref-type="fig"}C) rather than non-specific peptide E7 (Additional File [2](#MOESM2){ref-type="media"}: Figure S5 B). Long-term prevention effects were also shown as vaccinated mice remained tumor free. However, tumors continued to grow in unvaccinated mice (Fig. [5](#Fig5){ref-type="fig"}D). Therefore, a vaccine comprised of DCs treated with RPS3 and pulsed with a tumor specific peptide has a significant effect in producing memory T cells. Furthermore, we clarified which cytotoxic immune cell make the most significant contribution to the effects of the RPS3-based DC vaccine on tumor prevention using depletion with antibodies against CD4, CD8, and NK (Additional File [2](#MOESM2){ref-type="media"}: Figure S6). As a result, the survival of mice vaccinated with RPS3-activated DCs decreased when the mice were also injected with a CD8-depleting antibody, indicating that CD8^+^ T cells function to prevent tumor formation and have an antitumor effect.Fig. 5In vivo study showing the tumor treatment effect and the generation of memory T cells by the DC-based cancer vaccine using RPS3 as adjuvant RPS3-mediated activation and maturation of DCs and its adjuvant effect in DC vaccine are dependent on TLR4 {#Sec17} ---------------------------------------------------------------------------------------------------------- Next, we examined the TLR4-dependency for DC activation and maturation, and its adjuvant effects in the DC vaccine. Mouse BMDCs from either wild type or TLR4−/− mice were treated with RPS3 or LPS. After 16 h of treatment, the levels of the secreted pro-inflammatory cytokines TNF-α, IL-1β, IL-6, IL-10, IL-12, and IFN-β were evaluated by ELISA. As shown in Fig. [6](#Fig6){ref-type="fig"}A, DCs treated with RPS3 or LPS showed the expected increases in cytokine levels and co-stimulatory molecules (CD40, CD80, CD86, and MHC class I) compared to untreated cells. These increases were not observed in TLR4−/− DCs. And the same effects were shown with a TLR4 blocking antibody as was shown with TLR deficient mice. The expression of co-stimulatory molecule CD40 and the secretion of pro-inflammatory cytokines TNF- α and IL-6 were decreased after pre-treated by TLR4 blocking antibody (Additional File [2](#MOESM2){ref-type="media"}: Figure S7). In addition, the levels of p-ERK, p-P38, p-JNK, and p-AKT were increased and the levels of IкB-α were decreased in wild type DCs after incubating with RPS3 in a time course over an hour, there was no effect on signaling pathways in TLR4 −/− DCs. Finally, the effect of the DC-based cancer vaccine was assessed using wild type and TLR4−/− BMDCs to see whether its adjuvant effects are dependent on TLR4. The number of tumor specific CD8^+^IFN-γ^+^ T cells was greatly increased when mice were vaccinated with RPS3-activated wild type DCs pulsed with E7, but not with RSP3 treated TLR4−/− DCs pulsed with E7. To examine the therapeutic anti-tumor effects of these two vaccines, mice were first injected with TC-1 tumor cells to establish tumor mass. After the tumor mass was established, the mice were vaccinated with either RPS3 treated wild type or TLR4−/− DCs pulsed with E7 peptide once a week for two weeks. As shown in Fig. [6](#Fig6){ref-type="fig"}E, the tumor treatment effect was absent in mice injected with the RPS3-treated DCs lacking TLR4 and pulsed with E7. These observations show that TLR4 is necessary for DC activation and maturation by RPS3 and for its ability to function as an adjuvant in the DC vaccine.Fig. 6Comparison of RPS3 protein activity in WT and TLR4−/− mouse BMDCs Two fragments of RPS3 (1--95 and 91--243) have different effects on BMDCs {#Sec18} ------------------------------------------------------------------------- After characterizing the ability of RPS3 to induce maturation and activation of DCs, we sought to perform further functional studies to assess which portion of RSP3 is important for the adjuvant effects. The KH domain, (amino acids from 22 to 92 in RPS3 (1--243)), is a well-known DNA & RNA binding domain. On the basis of the presence of this KH domain, two fragments of RPS3, a fragment (1--95) and another fragment (91--243), were cloned and generated (Fig. [7](#Fig7){ref-type="fig"}A). Each His-tagged fragment (1--95, 91--243) was purified from *E. coli* and confirmed by western blot (Fig. [7](#Fig7){ref-type="fig"}B). TLR4-MD2 overexpressing HEK293 cells were then treated with RPS3, one fragment (1--95), or the other fragment (91--243). As shown in Fig. [7](#Fig7){ref-type="fig"}C, luciferase activity was increased by RPS3 and the (91--243) fragment, whereas there was no effect of the (1--95) fragment. The levels of secreted pro-inflammatory cytokines TNF-α, IL-1β, IL-6, IL-10, IL-12 and IFN-β were significantly increased in DCs treated with RPS3 or the (91--243) RPS3 fragment, but not following treatment with the (1--95) fragment (Fig. [7](#Fig7){ref-type="fig"}D). The expression levels of the co-stimulatory molecules CD40, CD80, CD86, and MHC class I were also increased in DCs treated with the (91--243) RPS3 fragment (Fig. [7](#Fig7){ref-type="fig"}E). Therefore, these results suggest that the effect of RPS3 as an adjuvant for DC stimulation is mediated mainly by the 91--243 fragment (which does not include the KH domain). This small protein fragment protein might therefore be useful in immunotherapy.Fig. 7Identification the RPS3 protein domain associated with TLR4 signaling Discussion {#Sec19} ========== In this study, we sought to find new ligands for TLR4. First, a number of intracellular proteins extracted from different human cancer cells showed a significant increase in NF-кB activity when used to treat HEK293 cells expressing TLR4-MD2. Using these human cancer cells, numerous ribosomal family proteins were found in pull-down assays conducted with recombinant TLR4. After purification and treatment of mouse BMDCs to assess the TLR4 ligand candidates, ribosomal protein S3 (RPS3) stood out as the strongest inducer of maturation and activation of DCs. Accordingly, we used RPS3 as an adjuvant in a DC-based vaccine to generate CD8^+^ T cells and to prevent and treat tumors. Notably, RPS3 had large effects on DCs, even at very low concentrations (0.1 μg/mL) compared to HMGB1, a DNA binding protein, which requires 50\~100 μg/mL to induce maturation and activation of DCs \[[@CR31]\]. Since HMGB1 is well-known TLR4 binder, HMGB1 was used directly to compare with the effects of RPS3 by activating BMDCs (Additional File [2](#MOESM2){ref-type="media"}: Figure S8). RPS3 exists intracellularly in both mouse and human cancer cells, but it is secreted after its N-linked glycosylation \[[@CR28]\]. Based on previous studies, the intracellular protein HMGB1 is known to be secreted from tumor cells following chemotherapy or radiotherapy, and it is also known that HMGB1 can bind to TLR4 on DCs and activate downstream signaling pathways \[[@CR32]\]. In this study, we also demonstrated that RPS3 is released from tumor cells following treatment with an anti-cancer drug, and that this released RPS3 can affect the surrounding DCs expressing TLR4 in the tumor microenvironment. To confirm that RPS3 binds to TLR4 and activates DCs, recombinant human RPS3 was purified and used to treat human and mouse DCs. As a result, human-derived RPS3 had an excellent effect on both species of DCs even though a human-derived protein can sometimes have different effects on mouse DCs. Human RPS3 is 99% identical to mouse RSP3 and so the effects of RPS3 in human and mouse dendritic cells are likely to be the same \[[@CR33]\]. Furthermore, since RPS3 is expressed and released by tumor cells, DC would be activated by tumor lysate which contains RPS3. However, tumor lysates contain varied DAMPs or inflammatory factors even except RPS3 that can activate DCs (Additional File [2](#MOESM2){ref-type="media"}: Figure S9). Next, we demonstrated that the number of tumor specific CD8^+^IFN-γ^+^ T cells is increased and that both tumor prevention and tumor treatment effects could be achieved using a tumor-specific antigen pulsed DC-based vaccine with RPS3 as an adjuvant. Even though the DC-based vaccine had a large effect on tumors, larger size or terminal cancers might require alternative treatments. We expect better effects when immune checkpoint inhibitors are used together with a DC-based vaccine. Immune checkpoint inhibitors such as Ipilimumab, a CTLA-4 inhibitor for the treatment of melanoma, is already approved by the FDA. In addition, several other immune checkpoint inhibitors, such as PD-1 and PD-L1, are undergoing clinical trials for the treatment of several cancers, including melanoma, lung, and breast cancer \[[@CR34]\]. Currently, we are investigating the use of this DC-based vaccine together with immune checkpoint inhibitors to examine if they have synergistic or greater effects. Furthermore, RPS3 is confirmed to be safe when using as an adjuvant in tumor-specific antigen DC-based vaccines via TLR4 since RPS3 is nucleoprotein that could induce humoral immunity, producing autoantibodies against itself (Additional File [2](#MOESM2){ref-type="media"}: Figure S10). We also examined the effects of the RPS3 protein on DCs and the dependency on TLR4 using TLR4−/− and wild type mice. We suspected that if DNA binds to the KH domain of RPS3 this could induce responses through TLR3, TLR7, or TLR9. Therefore, two fragments (1--95 and 91--243) were constructed based on the location of the KH domain in RPS3 to understand whether the KH domain of RPS3 directly participates in DC maturation and activation. When the two RPS3 fragments (1--95 containing the KH domain and 91--243 lacking the KH domain) were used to treat HEK293-MD2-TLR4 cells or BMDCs, only by the (91--243) fragment increased luciferase activity and stimulated BMDCs. Therefore, based on the TLR4 dependency, and the fact that the (91--243) fragment, which lacks the DNA binding KH domain, is active we can conclude that DNA binding is not related to the effects on DCs. Based on the findings of this study, we conclude that RPS3 from human cancer cells has a very good effect on the immune system. RPS3 induces maturation and activation of dendritic cells, acting as a new ligand for TLR4 in the innate immune system, and significantly increases CD8^+^ T cell production in the presence of a tumor-specific antigen in the adaptive immune system. Furthermore, since RPS3 is released when tumors are treated by chemotherapy, it is necessary for us to develop RPS3 neutralizing antibody to block TLR4/RPS3 interaction in vivo to demonstrate a more specific mechanism. In addition, RPS3 is a human-derived protein that is safe for in vivo use, unlike LPS, a bacterial endotoxin, and has been shown to be highly effective in very small doses when compared to other adjuvants that can cause maturation and activation of DCs. Thus, RPS3 is a novel and potential adjuvant for DC-based cancer vaccines. In our previous studies, not only RPS3, but also several ribosomal proteins that bind to TLR4 were screened from human tumor cells. A study on the effects of these ribosomal proteins to induce maturation and activation of DCs has revealed previously unknown functions of these other ribosomal proteins on DCs. Therefore, we are currently assessing the potential of these other ribosomal proteins for use as novel adjuvants for the development of DC-based vaccines. Additional files ================ {#Sec20} Additional file 1:Supplementary Materials and methods. (DOCX 20 kb) Additional file 2:**Figure S1.** NF-κB activity was measured using cancer cell lysates to assess TLR4-expressing tumor cells. **Figure S2.** To identify the TLR4 binding proteins, two groups were divided and assessed through pull-down assay. **Figure S3** Mouse BMDCs were treated with doxorubicin (0, 0.1, 1, 10 μg/mL). **Figure S4.** For endotoxin contamination test, recombinant RPS3 (1 μg) and LPS (100 ng) were incubated with proteinase K (100 μg/ml) or with polymyxin B (10 μg/ml). **Figure S5** Mouse BMDCs were separated into RPS3 treated mature DCs loaded with OVA peptide or E7 peptide. **Figure S6.** To confirm the dominancy of T cells in adaptive immune responses to clear tumors, T cell or NK cell depletion antibodies were injected into mice before the injection of cancer cells. **Figure S7** We examined the TLR4-dependency for DC activation and maturation by using TLR4 blocking antibody. **Figure S8** The purification of recombinant HMGB1 protein was confirmed by CBB staining and by western blot (A). **Figure S9** HEK293 cells were transfected by shRNA (GFP) as a negative control or shRNA (RPS3). **Figure S10** Mice serum with vaccination or not were used to confirm that RPS3 does not induce humoral immunity, producing autoantibodies against itself. (DOCX 1509 kb) Funding {#FPar1} ======= This study was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (NRF-2016R1A5A2012284 and (NRF-2018R1A2B6008455). This study was also supported by a grant of Korea Health Technology R&D project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic of Korea (grant number: HI15C2524). Availability of data and materials {#FPar2} ================================== The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. HJP, GYJ, THK, and YMP performed the literature search. HJP, GYJ, YSK, JHP, SEL, MCV, JJL, HDH, IDJ, THK, and YMP designed the study, collected, analyzed, and interpreted data, and wrote the manuscript. HJP, GYJ, THK, and YMP analyzed data and provided statistical interpretation. HJP, GYJ, YSK, JHP, SEL, MCV, JJL, HDH, IDJ, and THK contributed to data collection. All authors read and approved the final manuscript. Ethics approval and consent to participate {#FPar3} ========================================== Not applicable. Consent for publication {#FPar4} ======================= Not applicable. Competing interests {#FPar5} =================== The authors declare that they have no competing interests. Publisher's Note {#FPar6} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Core tip:** Endosonography-guided biliary drainage is an effective alternative in the failure of endoscopic retrograde cholangiopancreatography, with the potential to provide the least invasive and the lowest risk therapeutic modality for biliary drainage when compared to percutaneous transhepatic biliary drainage or surgery. For this procedure, access to the biliary tree can be obtained by transhepatic or transduodenal approaches. However, the transhepatic approach offers a good acoustic window for puncture of the biliary tree, a straight and easier to work with position of the echoendoscope, a better positioning of the guidewire, and a lower chance of bleeding or choleperitoneum. INTRODUCTION ============ Traditionally, endoscopic retrograde cholangiopancreatography (ERCP) is the standard approach to biliary drainage\[[@B1],[@B2]\]. However, the procedure fails in up to 10% of patients, especially owing to anatomic variations, malignant duodenal obstructions and previous surgeries\[[@B3],[@B4]\]. For these cases, percutaneous transhepatic biliary drainage (PTBD) or surgery has been used, despite the high morbidity and not negligible mortality caused by these procedures\[[@B5],[@B6]\]. More recently, endosonography-guided biliary drainage (EUS-BD) has emerged as an effective alternative, with the potential to provide the least invasive and lowest risk therapeutic modality for biliary access and drainage\[[@B7],[@B8]\]. A recent meta-analysis has reported technical and clinical success of 90% and 94%, respectively\[[@B9]\]. We aimed to evaluate the role of different EUS-BD techniques in case of ERCP failure, and to propose a systematic routine for EUS-BD according to the feasible access routes to the biliary tree. MATERIALS AND METHODS ===================== Study design ------------ This was a retrospective study with prospective data collection about the role of EUS-BD conducted at two tertiary-referral centers. Between February 2010 and December 2016, 3528 ERCPs were performed at these centers. Eligible cases included patients older than 18 years with unresectable biliopancreatic neoplasia, and patients with benign conditions referred to EUS-BD when access to the biliary tree and internal biliary drainage by ERCP were not possible. ERCP failure was considered when biliary cannulation could not be achieved even after advanced techniques (cannulation in addition to a pancreatic guidewire or stent, needle-knife access papillotomy over a pancreatic stent, cannulation through a duodenal stent, and back-loading of the duodenoscope over a duodenal guidewire to pass a luminal stricture). Exclusion criteria were an international normalized ratio (INR) \> 1.5 or platelet count \< 50000/μL, ascites around the puncture area, absence of an adequate acoustic window for hepatic or choledochal puncture, total gastrectomy, and patient refusal. After EUS-BD, four follow-up visits were scheduled for each patient during the first 90 d, or until their death. The study was approved by the Institutional Review Board (Approval No. 2.191.319), and all patients gave written informed consent for ERCP and EUS-BD before enrollment. Technical aspects ----------------- All EUS-BD procedures were performed by the same experienced endoscopist with Fujinon (FujiFilm Corporation, Nishiazabu 2-chome Minato, Ku, Tokyo) duodenoscopes (ED-530XT) and curvilinear array echoendoscopes (EG530UT2) coupled to SU-7000 or SU-8000 ultrasound units. The sequential EUS-BD procedures proposed for all patients were as follows: first, transhepatic puncture with a 19 gauge aspiration needle (EUSN-19 T, Cook, Winston Sallen, NC, United States) was tried. The EUS-RV technique was successful when the guidewire could be passed through the papilla and seized in the second portion of the duodenum. In case of papillary benign disease or absence of duodenal stenosis, retrograde treatment with a duodenoscope or echoendoscope was performed. An anterograde approach was attempted when tumoral duodenal infiltration or duodenal stenosis did not allow the capture of the guidewire in the duodenum. If the anterograde approach failed, Endosonography-guided hepatogastrostomy (EUS-HG) was the next alternative. In case of failure of the intrahepatic puncture due to unfavorable anatomy, cirrhosis or difficulty in maintaining the adequate position of the guidewire, patients were submitted to endosonography-guided choledocoduodenostomy (EUS-CD). If all approaches for EUS-BD were unsuccessful, patients were submitted to PTBD. Duodenal self-expandable metallic stents (SEMS) were used in all stenoses obstructing access to the papilla. The procedures were always performed with the patient in the left lateral decubitus position, under deep sedation with the assistance of an anesthesiologist. After the procedure, patients were monitored for two hours, and intravenous antibiotics (ciprofloxacin and metronidazole) were given for 7 d. Routine for EUS-BD approaches ----------------------------- **Endosonography-guided rendez-vous:** When the duodenoscope could reach the major papilla, EUS-RV was tried and a curvilinear echoendoscope was used to obtain biliary access. The tip of the echoendoscope was positioned in the gastric fundus to access the intrahepatic bile duct. A 19 gauge EUS aspiration needle was used to puncture the bile duct close to the hepatic hilum, and to insert a large-caliber guidewire to deploy the stent. After fluoroscopic confirmation of the needle inside the bile duct, the guidewire was inserted through the obstruction and passed to the duodenum. Once the guidewire crossed the papilla, the guidewire was retrieved with a biopsy forceps or snare. Next, a metal stent was deployed by means of the over-the-wire technique\[[@B10]\]. **Endosonography-guided anterograde stent insertion:** In the presence of neoplastic duodenal stenosis, when the guidewire could not be seized in the duodenum, the stent was placed in an anterograde way. Access to the intrahepatic bile duct was obtained using a 19 gauge aspiration needle. Once puncture of the bile duct was confirmed by fluoroscopy, the guidewire was inserted through the duodenal major papilla and positioned in the second portion of the duodenum. At this point, a SEMS was inserted through the gastric wall across the papilla. **Endosonography-guided hepatogastrostomy:** EUS-HG was tried after failure of the EUS-RV and EUS-anterograde stent insertion (EUS-ASI) techniques, in those cases whose hepatic puncture was successful but the guidewire could not be passed through the papilla. The dilated intrahepatic bile duct was punctured, and the guidewire was placed through the stenosis. The tract was dilated with a 6 Fr cystostome, and a fully covered metal stent was deployed, with care taken to leave more than 3 cm of the stent in the gastric lumen to avoid food obstruction. **Endosonography-guided Choledocoduodenostomy:** In patients for whom a transhepatic approach was not feasible, EUS-CD was performed with the identification of the extrahepatic bile duct from the duodenal bulb. Once the insertion of the guidewire into the bile duct was confirmed by cholangiography, the tract was dilated with a 6 Fr cystostome, and a fully covered self-expandable metal stent was inserted. Technical and clinical success ------------------------------ Technical success was defined as adequate positioning of the stent as shown by endoscopic and fluoroscopic images. Clinical success was defined as a decrease of at least 50% in serum total bilirubin levels. Statistical analysis -------------------- A linear model was adjusted for the calculation of the technical success prevalence ratios, generalized by Poisson distribution and by the linking logarithmic function using the Proc Genmod of SAS 9.3 software (SAS Institute Inc., Cary NC, United States) to determine whether the different approaches had any impact on efficacy, compared to the EUS-RV technique (*P* \> 0.05). RESULTS ======= Patient demographics and technical aspects ------------------------------------------ During the study period, it was not possible to cannulate the biliary tree in 24 of 3528 (0.68%) patients submitted to ERCP. Thirteen men and 11 women with a mean age of 67.8 years old were included in the study. The most common symptom was jaundice in 96% of the patients, followed by abdominal pain and acute biliary pancreatitis in 21% and 8.3% of cases, respectively. The demographics, reasons for ERCP failure, indications for EUS-BD, as well as technical and clinical success are listed in Table [1](#T1){ref-type="table"}. ###### Demographics and treatment success of patients submitted to endosonography-guided biliary drainage due to endoscopic retrograde cholangiopancreatography failure **EUS-BD** **EUS-RV** **EUS-ASI** **EUS-HG** **EUS-CD** ---------------------------------------------------------------------- -------------- -------------- -------------- -------------- -------------- *n* (%) 24 (100) 7 (29) 5 (21) 6 (25) 6 (25) Sex (M/F) 13/11 5/2 1/4 4/2 3/3 Age (range), yr 67.8 (42-91) 67.7 (42-84) 60.8 (42-70) 68.2 (50-81) 73.5 (52-91) Reasons for ERCP failure (*n*) \- \- \- \- \- Malignant duodenal stenosis 8 2 3 2 1 Malignant papillary infiltration 7 1 2 1 3 Impossibility of access to the common bile duct or intrahepatic duct 7 2 0 3 2 Giant duodenal diverticulum 1 1 0 0 0 Billroth II gastrectomy without access to the duodenal papilla 1 1 0 0 0 Indications for EUS-BD \- \- \- \- \- Malignant 20 3 5 6 6 Pancreatic cancer 13 3 4 2 4 Liver metastases of colon cancer 4 0 0 3 1 Cholangiocarcinoma 1 0 0 1 0 Duodenal lymphoma 1 0 1 0 0 Papillary cancer 1 0 0 0 1 Benign 4 4 0 0 0 Common bile duct stones 2 2 0 0 0 Biliary necrotizing acute pancreatitis 1 1 0 0 0 Recurrent acute pancreatitis due to sphincter of Oddi dysfunction 1 1 0 0 0 Technical success *n* (%) 20 (83.3) 5 (71.4) 5 (100) 5 (83.3) 5 (83.3) Clinical success (%) 18 (75) 4 (57.1) 5 (100) 4 (66.7) 5 (83.3) Complications (%) 3 (12.5) 2 (28.5) 0 (0) 1 (16.7) 0 (0) EUS-BD: Endosonography-guided biliary drainage; EUS-RV: Endosonography-guided rendez-vous; EUS-ASI: Endosonography-guided anterograde stent insertion; EUS-HG: Endosonography-guided hepaticogastrostomy; EUS-CD: Endosonography-guided choledochoduodenostomy. Endosonography-guided rendez-vous --------------------------------- The EUS-guided transhepatic approach was tried in all patients (Figure [1](#F1){ref-type="fig"}). In 18/24 (75%) cases, puncture of the bile duct was possible, but the passage of the guidewire through the papilla occurred only in 12 (50%) cases. The guidewire could be recovered in 5/7 cases, and the passage of the stent was performed by means of an EUS-RV technique (Figure [2](#F2){ref-type="fig"}). The complication rate for these cases was 28% (2/7), consisting of an intracavitary hemorrhage and a choleperitoneum, both managed conservatively. In 5 other cases the guidewire could not be recovered in the duodenum owing to duodenal stenosis (3) or papillary infiltration (2). For these cases, an EUS-ASI technique was the next option. In 6 other cases, the guidewire did not cross the papilla, and was positioned in the proximal common bile duct (4), and in the right lobe (1) and left lobe of the liver (1). For these cases, an EUS-HG was the next alternative. The remaining 6 patients for whom transhepatic approaches were not possible underwent EUS-CD. {#F1} {#F2} EUS-guided anterograde stent insertion -------------------------------------- Even after passage of the guidewire in the second duodenal portion, the recovery of the guidewire was not possible in 5 patients due to malignant duodenal stenosis (3) or papillary infiltration (2). For these cases, anterograde deployment of the biliary SEMS was performed (Figure [3](#F3){ref-type="fig"}). After passage of the biliary SEMS, a duodenal SEMS was delivered in 3 patients with neoplastic duodenal stenosis. The overall technical success was 100%. {#F3} Endosonography-guided hepatogastrostomy --------------------------------------- EUS-HG through transhepatic puncture was tried in 6 patients in whom the guidewire was positioned in the common bile duct (4), right lobe (1) and left lobe of the liver (1) (Figure [4](#F4){ref-type="fig"}). In 5/6 (83.3%) cases, an uneventful passage of the biliary SEMS was possible. For a single patient with recurrent liver metastasis from colon cancer after hepatectomy, the introduction of the transhepatic guidewire was impossible. The technical success rate was 83.3%, with one patient developing a pneumoperitoneum after the procedure. {#F4} Endosonography-guided choledochoduodenostomy -------------------------------------------- The insertion of the biliary stent through the duodenal puncture was tried in 6 patients as a rescue EUS-guided procedure for biliary drainage (Figure [5](#F5){ref-type="fig"}). All of these cases presented malignancies (Table [1](#T1){ref-type="table"}). The correct positioning of the guidewire was achieved in 5/6 (83.3%), and one case was referred to PTBD. There was no complication. {#F5} Technical and clinical success ------------------------------ The overall technical success for EUS-BD was 83.3% (20/24). There was no significant difference among the various techniques (*P* = 0.81). Prior to EUS-BD, the mean levels of serum total and direct bilirubin were 13.3 mg/dL (5-29.9) and 9.1 (3-20.4) mg/dL, respectively. Ten days after EUS-BD, the mean levels were 2.3 (1.3-33) mg/dL, and 1.7 (0.6-22) mg/dL, respectively. The overall clinical success of EUS-BD was 75%. Complications ------------- Three (12.5%) complications occurred in patients submitted to EUS-BD: a pneumoperitoneum, a choleperitoneum, and an intracavitary liver hemorrhage. All of them were a consequence of the liver puncture in the hilum and were treated conservatively (Table [1](#T1){ref-type="table"}). The patient with liver hemorrhage died three days after the PTBD due to acute respiratory and renal failure. DISCUSSION ========== In our experience, an alternative to ERCP failure for biliary drainage was necessary in 0.68% of the cases, a finding similar to the rate of 0.62% in the experience of Holt et al\[[@B11]\]. Elderly people with malignant biliary obstruction are the most common candidates for the procedure\[[@B11]\], which was the case in our study, with patients at a median age of 68 years and with malignancies representing 83% of the cases. Endosonography-guided biliary drainage has been an alternative therapy to PTBD and surgery in ERCP failure\[[@B8],[@B12]\]. PTBD, despite its satisfactory results, has a complication rate of about 30%, and surgery, although regarded as the definitive treatment for biliary drainage, is associated with high morbidity and mortality, especially for cases with terminal neoplastic disease\[[@B11],[@B13],[@B14]\]. Overall, the therapeutic success of EUS-BD ranges from 73% to 100%\[[@B15]-[@B19]\]. However, there is no consensus about the best EUS-BD technique\[[@B9]\]. Regarding particular EUS-BD techniques, there is a scarcity of comparative studies. Ogura et al\[[@B20]\] compared EUS-HG and EUS-CD for patients with jaundice and duodenal obstruction. Patients submitted to the transhepatic approach exhibited a longer patency of the biliary stent than those submitted to the transduodenal approach. In addition, the EUS-CD technique revealed a higher rate of complications, especially reflux cholangitis (OR = 10.285; 95%CI: 1.686-62.733; *P* = 0.012). Artifon et al\[[@B21]\] also evaluated the two techniques in a randomized clinical trial. There was no significant difference in effectiveness or safety between the two procedures. Technical and clinical success, as well as complications rates were 96%, 91%, and 20% for EUS-HG, respectively, and 91%, 77% and 12.5% for EUS-CD, respectively. In an attempt to demonstrate the value of EUS-RV as the initial therapeutic option for biliary drainage in ERCP failure, Iwashita et al\[[@B22]\] performed the procedure using the transduodenal approach and using the transhepatic approach after failure of the transcholedochal approach. The authors concluded that EUS-RV is an effective and safe procedure, as also observed in our own experience. However, in contrast to the cited study, we began EUS-BD by the transhepatic approach, leaving the transduodenal approach only for the rescue option in the failure of the transhepatic approach. In our experience, the transhepatic approach allows us to choose among three EUS-BD techniques according to the recovery or not of the guidewire, *i.e*., the EUS-RV, EUS-ASI and EUS-HG techniques. Our group has adopted a systematic EUS-BD routine starting with the transhepatic access to initially perform the EUS-RV or EUS-ASI technique. This approach offers a good acoustic window for puncture of the biliary tree, a straight and easier to work position of the echoendoscope, a better positioning of the guidewire, and a lower chance of bleeding or choleperitoneum, with both complications amenable to tamponade by the liver parenchyma\[[@B19],[@B23]\]. In our study, beginning with the transhepatic approach, the overall technical success was 83%, and the clinical success (intention-to-treat) was 75%, similar to literature results\[[@B23]\]. On the other hand, the transduodenal approach permits an easier execution of only the EUS-CD or, although more laborious and time-consuming, the EUS-RV. In the failure of this approach, the transhepatic approach should be the rescue therapy. Nevertheless, despite the good results of EUS-BD when using the transhepatic approach, the literature still mentions some concern about the risk of complications with the intrahepatic access\[[@B18],[@B20],[@B24]\]. The needle must traverse the peritoneal cavity, a procedure that might increase the risk of pneumo- and choleperitoneum. This complication occurred in one of our patients and was managed conservatively. Another issue is the movement of the stomach and liver during breathing and peristalsis, which might induce stent migration, trauma to the bilioenteric tract, and bile leakage. Finally, small-caliber intrahepatic ducts may not accommodate wider 8-mm to 10-mm metal stents, possibly predisposing to pneumoperitoneum and bile leakage due to incomplete sealing of the bilioenteric fistula\[[@B25],[@B26]\]. For this reason, our goal during EUS-BD by means of the transhepatic approach is to obtain an intrahepatic duct of larger caliber as close as possible to the hepatic hilum. In all of our cases in which the guidewire could not be reached in the duodenum due to stenosis or papillary infiltration, EUS-ASI succeeded without complications. The good performance and low complications rate of the EUS-ASI technique has been demonstrated in the literature\[[@B27]\]. On the other hand, if the patient has only a dilated biliary tree where the hepatic puncture is feasible but the guidewire could not reach the papilla, EUS-HG should be the next option. The greatest limitation in patients undergoing EUS-HG is the access to the right intrahepatic biliary tract and the progression of the guidewire to the common bile duct or its passage through the duodenal papilla. However, many authors justify selective drainage of the left intrahepatic biliary tract compared to the extrahepatic approach\[[@B7],[@B28],[@B29]\]. Both approaches have been shown to be effective and to involve low complications rates\[[@B21],[@B26]\]. Nonetheless, EUS-BD by transhepatic approach may not be possible in some cases, depending on the patient anatomy\[[@B19],[@B30]\]. We observed EUS-RV failure due to the impossibility of puncturing the liver or the inability to maintain the stability of the guidewire, and the difficulty to seize the guidewire in the duodenal lumen. In such cases, an extrahepatic approach must be adopted. The transcholedochal approach has the benefit of being feasible in patients whose papilla cannot be reached and has the advantage of being close to the duodenum\[[@B7],[@B31],[@B32]\]. In the current study, the technical success rates were the same (83.3%) for EUS-HG and EUS-CD, in agreement with published series\[[@B20],[@B21]\]. Except for a pneumoperitoneum in the intrahepatic group, no difference in major complications was found between EUS-HG and EUS-CD (16.6% *vs* 0%; *P* = 0.81). As a whole, EUS-BD is a safer technique than PTBD and surgery, with complication rates ranging from 10% to 20%, although the severity of most cases is mild to moderate\[[@B10],[@B13]\]. Our complication rate also agreed with that reported in other studies\[[@B10],[@B13]\]. Three of our cases developed complications, representing an overall rate of 12.5%. All of these cases were managed conservatively, but a patient with intracavitary bleeding was submitted immediately to PTBD after EUS-BD failure, and died three days later. Despite the small number of our patients, this study did not demonstrate any significant difference in technical success or complication rates among different techniques of EUS-BD, in agreement with other studies\[[@B19],[@B23]\]. In summary, a rational algorithm for EUS-BD in case of obstructive biliary diseases and ERCP failure might begin with the transhepatic approach, followed by particular EUS-BD techniques based on the patient's anatomy and feasibility to recover the guidewire. ARTICLE HIGHLIGHTS ================== Research background ------------------- Endoscopic retrograde cholangiopancreatography (ERCP) is the standard approach to biliary drainage, and, in the failure of the procedure, percutaneous transhepatic biliary drainage or surgery must be used. However, endosonography can guarantee the least invasive and lowest risk treatment for biliary drainage of these cases. This study presents the results of different techniques for endosonography-guided biliary drainage in case of ERCP failure. Research motivation ------------------- In case of ERCP failure, patients must be submitted to surgery or percutaneous transhepatic biliary drainage at different places in the hospital and with a long delay in treatment, conditions which can increase the morbidity and risks for the patient. Endosonography-guided biliary drainage can be performed immediately after ERCP failure, decreasing the time and risk of definitive treatment of the patient. Research objectives ------------------- The main objectives of the study were to evaluate the success rates of endosonography (EUS)-guided biliary drainage techniques after ERCP failure for the management of biliary obstruction, and to propose a rational approach based on the access to the biliary tree and feasibility to recover the guidewire. Research methods ---------------- In our experience, an alternative to ERCP failure for biliary drainage was necessary in 24 of 3538 (0.68%) cases. Elderly people with malignant biliary obstruction were the most common candidates for the procedure. The sequential endosonography-guided biliary drainage (EUS-BD) procedures proposed for all patients were transhepatic puncture in order to perform the EUS-guided rendez-vous technique. An anterograde approach was attempted when the capture of the guidewire in the duodenum was not possible. If the anterograde approach failed, EUS-guided Hepatogastrostomy was the next alternative. In case of failure of the intrahepatic puncture, patients were submitted to EUS-guided choledochoduodenostomy (EUS-CD). Research results ---------------- Patients were submitted to EUS-guided rendez-vous (7), EUS-guided anterograde stent insertion (5), EUS-guided hepaticogastrostomy (6), and EUS-CD (6). Success rates did not differ among the various EUS-BD technique. Overall, technical and clinical success rates were 83.3% and 75%, respectively. The technical success for each technique was 71.4%, 100%, 83.3%, and 83.3%, respectively (*P* = 0.81). Complications occurred in 3 (12.5%) patients. All of these cases were managed conservatively, but one patient died after a rescue percutaneous transhepatic biliary drainage. Regarding particular EUS-BD techniques, there is a scarcity of comparative studies, and a consensus about the best technique has not been established. Research conclusions -------------------- A rational approach to EUS-guided biliary drainage in case of obstructive biliary disease and ERCP failure should begin with the transhepatic approach, followed by particular EUS-guided biliary drainage techniques based on the patient's anatomy and feasibility to recover the guidewire in the duodenum. Research perspectives --------------------- EUS-guided biliary drainage should be included in the therapeutic arsenal for the management of malignant biliary obstruction in case of ERCP failure, and should be the choice rather than surgery or percutaneous transhepatic biliary drainage. Institutional review board statement: The institutional review board statement was approved by protocol No. 2.191.319. Informed consent statement: All study participants, or their legal guardians, provided informed written consent prior to study enrollment. Conflict-of-interest statement: No potential conflicts of interest relevant to this article were reported. Data sharing statement: No additional data are available. Manuscript source: Unsolicited manuscript Peer-review started: January 10, 2018 First decision: January 23, 2018 Article in press: March 15, 2018 Specialty type: Gastroenterology and hepatology Country of origin: Brazil Peer-review report classification Grade A (Excellent): 0 Grade B (Very good): B, B Grade C (Good): C Grade D (Fair): 0 Grade E (Poor): 0 P- Reviewer: Andrianello S, Garg P, Govindarajan GK S- Editor: Cui LJ L- Editor: A E- Editor: Li D [^1]: Author contributions: Ardengh JC performed the procedures; Lopes CV designed the study and wrote the manuscript; Kemp R and dos Santos JS provided the collection of all human material. Correspondence to: César Vivian Lopes, MD, PhD, Doctor, Department of Gastroenterology and Digestive Endoscopy, Santa Casa Hospital, Rua Prof. Cristiano Fischer 668/1001, Porto Alegre 91410-000, Brazil. <drcvlopes@gmail.com> Telephone: +55-51-33388054

Introduction ============ Biology is an integrated study, which aimed to discover the various mechanisms in complex network. Although conventional experiments based on biochemical experiment can suggest \'clue\' of individual mechanism in certain condition, but they restricted to reveal the \'real\' phenomenon in whole cell or organism level. Molecular imaging is a powerful tool to discover various biological mechanisms, providing temporal and spatial information of cell and organism level \[[@B1],[@B2]\]. The spatio-temporal analysis could provide the more exact mechanism in complex and changeable system, living organism. As researches become more complex with our knowledge of biological system, the role of molecular imaging are more increasing. Infectious disease is one of the required fields with molecular imaging. Infection is complex process in whole organism related with various molecules in pathogens and host. As the biological effects of pathogen and immunological responses of host are varies at each time course and each organ \[[@B3]\], study of infectious disease requires analysis with various conditions. In conventional analysis with animal experiment as host system, enormous labor, time and animals are needed with long time-consuming experiment at separated time point. But, molecular imaging analysis which can chase the pathogens in organism level, pathogen can overcome the limitation of conventional analysis and biological and pathological process and be monitored. In recent study, it was discovered that bubonic plaque is colonized at lymph node first, and disseminated at liver and spleen by visualization of bioluminescence *Yersinia pestis* \[[@B4]\]. In similar analysis system, the primary germination site of *Bacillus anthracis* spore in organism was revealed as lung \[[@B5]\]. Vaccine is one of the ideal solutions against infectious disease. In order to develop efficient vaccine, efficacy evaluation methods with high accuracy are required. In conventional method for efficacy evaluation of vaccine have been dependent on detection of immunoglobulin for specific antigens or observation of animal lethality \[[@B3],[@B6]\]. But, the former cannot provide direct efficacy against disease, the latter requires a lot of labors and animals. In vaccine efficacy evaluation, visualization of pathogens in whole organism at each time point would be able to reduce the consuming animals and provide the *in vivo* information within consistent background with identical organism. In this study, we verified the application of *in vivo* molecular imaging system as vaccine efficacy evaluation method using fluorescent *Escherichia coli*, and suggest that the possibility of novel method for vaccine evaluation system. Materials and Methods ===================== Bacterial strain and growth condition ------------------------------------- *E. coli*-green fluorescent protein (*E. coli*-GFP) strain MC1061 with GFP expression plasmid obtained from Korea National Institute of Health (KNIH; Osong, Korea). *E. coli*-GFP was cultured in Luria Bertani broth (Invitrogen, Carlsbad, CA, USA) in 37℃. One hundred µg/mL ampicilin (Invitrogen) was used as antibiotics to selection of bacteria containing GFP expression plasmid. Measurement of bacteria density and relative fluorescence intensity ------------------------------------------------------------------- Bacteria density was determined by measurement of OD^600^ and calculated with matched bacteria density-OD^600^ index. Relative fluorescence intensity of *E. coli*-GFP was measured using SpectraMax (Molecular Devices, Sunnyvale, CA, USA) with 485 nm of excitation and 510 nm of emission filters \[[@B7],[@B8]\]. Whole-animal fluorescent imaging -------------------------------- Five to six weeks old of female BALB/c mice were purchased and used in bacteria injection experiment. All experiments were accomplished under the provision of KNIH Institutional Animal Care and Use Committee. *E. coli*-GFP was injected with intradermal (i.d.) of intraperitoneal (i.p.) route to BALB/c mice and injected mice were analyzed with IVIS spectrum whole live-animal imaging system (Perkin Elmer Inc., Waltham, MA, USA). Mice were anesthetized with isoflurane using a vaporizer, and fluorescent image was obtained GFP filter set (excitation wave length, 488 nm; emission wave length, 510 nm). Results ======= Fluorescent signal was increased by bacterial replication --------------------------------------------------------- Compared to wild type *E. coli* strain, fluorescent intensity was increased proportionately by concentrating *E. coli* MC1061 strain which expresses GFP (*E. coli*-GFP), which suggest quantified fluorescent intensity can indicate the number of bacteria ([Table 1](#T1){ref-type="table"}, [Fig. 1A](#F1){ref-type="fig"}). Fluorescent signal of *E. coli*-GFP was visualized by IVIS spectrum analysis and signal intensity could be represented by pseudo-color distribution ([Table 1](#T1){ref-type="table"}, [Fig. 1B](#F1){ref-type="fig"}). Increasing of fluorescent signals was matched with the number of bacteria measured by optical density at 600 nm. Injected bacteria was traced by fluorescent signal on the molecular imaging system ---------------------------------------------------------------------------------- Visualizing of bacteria which reside inside of whole animal is most essential factor for *in vivo* live imaging. After confirmation of fluorescent signal increased by bacterial replication, GFP-tagged strain was injected at different colony-forming unit (CFU) through the various routes and the fluorescent signal was measured. To verify the possibility of applying fluorescence expressing pathogen to *in vivo* monitoring system of pathogen infection, we injected *E. coli*-GFP to mice with i.d. and i.p. route ([Fig. 2](#F2){ref-type="fig"}). Over 10^9^ CFU, injected *E. coli*-GFP in both cased was clearly visualized in IVIS analysis system, suggesting fluorescent signal can be pass through skin of animal and be detected. Although auto-fluorescent signal from naïve animal was strong, high intensity of fluorescent signal from *E. coli*-GFP was distinguishable and can be normalized with background signal. Visualizing disseminated pathogen showed different signals in each organ ------------------------------------------------------------------------ Infected pathogen can disseminated to several organs in animal, and targeting of pathogen toward specific organ is related with pathogenic disease. So, investigation of disseminated localization of pathogen is important to study of disease and also evaluation of vaccine efficacy. In anatomical condition, each organ which directly injected with *E. coli*-GFP was visualized by IVIS analysis system ([Fig. 3A](#F3){ref-type="fig"}). Detached organ showed more obvious fluorescent signal, and intestine showed strongest fluorescent signal ([Fig. 3B](#F3){ref-type="fig"}). This seems to be due to its cavity structure that could contain bacteria. Organ of infectious situation, which provide the condition of bacterial infiltration to cells of organ, would be represented with more obvious and strong signal. Discussion ========== Vaccine efficacy is usually tested using IgG titer after vaccination or protection assay with animals. Of course those methods are very effective to confirm the efficacy and safety for newly developed vaccine candidates and currently available vaccines on the improvement. However we still can\'t get further information about the distribution and clearance of infected pathogens after vaccination. To fill the gap of information about the works of vaccines, *in vivo* imaging using GFP-conjugated pathogens can be applied for pathogen tracing and immune cell trafficking. Injected bacteria was clearly distinguished organ by organ presumably representing bacterial replication levels or integrated bacterial number in each organ ([Fig. 3](#F3){ref-type="fig"}). It was clear that fluorescence-tagged bacterial strain was replicated both on the culture plates and in the animals ([Figs. 1](#F1){ref-type="fig"}, [3](#F3){ref-type="fig"}), however, *in vivo* tracing was not completely possible only with fluorescence-tagging on the current *in vivo* imaging apparatus due to the penetration capacity of fluorescence and the intensity of current fluorescence-tagging molecules \[[@B9]\]. There are several limits to use the GFP-conjugated bacteria to trace the pathogen distribution and clearance. First of all, GFP signals are usually not strong enough to be analyzed on currently available *in vivo* imaging systems, luciferase or other stronger signals could be considered for better signal tracing for infected pathogens \[[@B10],[@B11]\]. Even though autofluorescence from animal body itself can be normalized using imaging software, still fluorescence signal seems not enough to show the track and localization of infected bacteria or other pathogens in each organ. Second, the sensitivity of fluorescence is not high enough to detect small number of infected pathogens, especially for chronic infection diseases without exponential replication such as acquired immunodeficiency syndrome or tuberculosis. Visualization of signal-tagged pathogens can be traced and measured after several multiplications in animal body, but the quick multiplication of infected pathogens cannot be expected in chronic infectious diseases. This low multiplicity of infection in chronic infection could be another limit for pathogen tracing in infected animals. Even though these several limits, the visualizing methods using signal-tagged pathogens can be a good model system to enforce the efficacy tests of vaccines or new drug candidates because it can supply related information about the pathogen localization and distribution after infection, also about the pathogen clearance in each targeted organ \[[@B12]-[@B15]\]. With more strong visualizing signals such as luciferase or optical nanomolecules, the molecular imaging methods for pathogen tracing and clearance could be a highly confident way to evaluate the vaccine efficacy in the future. No potential conflict of interest relevant to this article was reported. This study was supported by Korea National Institute of Health (grant number 2011-N45003-00). {ref-type="table"}. (B) Visualization of serial diluted *E. coli*-GFP. In pseudo-color covered image, the high intensity of fluorescence appeared as red color and low intensity of fluorescence as blue. *E. coli*-GFP, *Escherichia coli* MC1061 strain which expresses green fluorescent protein.](cevr-1-83-g001){#F1} {#F2} {#F3} ###### Concentration of diluted *E. coli*-GFP for optimizing bacterial counts  *E. coli*-GFP, *Escherichia coli* MC1061 strain which expresses green fluorescent protein; CFU, colony-forming unit. [^1]: ^\*^Both authors equally contributed to this research.

Introduction {#sec1-1} ============ Insertion of laryngeal mask airway (LMA) requires sufficient depth of anesthesia for jaw muscles to relax and suppression of airway reflexes for the device to be tolerated within the hypopharynx without undue coughing, gagging, and patient movement.\[[@ref1]\] Such conditions are provided by a generous dose of an intravenous (IV) anesthetic induction agent. Propofol itself does not have analgesic activity and when used alone, the high doses required for induction may cause adverse cardiovascular effects.\[[@ref2]\] Fentanyl has been reported to provide satisfactory conditions for insertion of LMA without administration of muscle relaxants. We compared insertion conditions for LMA silicone classic using propofol in a fixed dose (2.5 mg/kg) and fentanyl in two doses, i.e., 1 and 2 mcg/kg. Materials and Methods {#sec1-2} ===================== After approval from the hospital Research and Ethical Committee, a prospective, randomized double-blind study was conducted on patients of American Society of Anesthesiologists (ASA) grades 1 and 2, between 18 and 60 years of age, undergoing elective surgery (modified radical mastectomy, mastectomy, or superficial surgery of the upper limb). The exclusion criteria were risk of aspiration and regurgitation, mouth opening \< 2.5 cm, weight \< 40 kg or \>110 kg, respiratory tract pathology, cervical spine disease, and an anticipated difficult airway. An informed consent was obtained from all patients. Patients were randomly allocated using sealed envelope technique to two groups of 52 each. Fentanyl was used in a dose of 1 mcg/kg in the F~1~ group and 2 mcg/kg in the F~2~ group. All patients were fasted for over 6 h and premedicated with oral diazepam 0.1 mg/kg. Intraoperative monitoring included electrocardiogram, noninvasive blood pressure, oxygen saturation, end-tidal carbon dioxide, and airway pressure. Following preoxygenation, ondansetron 0.1 mg/kg IV was given. The study drug, fentanyl 1 or 2 mcg/kg was given IV over 10 s by an anesthesiologist blinded to the drug dose. Two minutes after administration of this, propofol 2.5 mg/kg was injected IV over 1 min. The jaw thrust was used as an indicator of adequate depth of anesthesia. In all patients, a classic silicone LMA was inserted by an experienced anesthesiologist using index finger technique. After adequate depth of anesthesia was attained, the patient\'s head was extended, neck flexed, jaw opened, and the LMA was inserted. The cuff was inflated with air in 3 ml increments until the LMA tube was seen to rise slightly out of patient\'s mouth. The anesthesiologist was blinded regarding dose of fentanyl. On attempting LMA insertion, if mouth opening was not adequate, further bolus doses of propofol were given in increments of 10 mg IV. Time taken from induction and number of attempts taken were noted in both study groups. After three failed attempts, the patient\'s trachea was intubated. In case LMA insertion took more than 20 s or more than three attempts were required, it was considered as failure and the case was excluded from statistical analysis. The heart rate and blood pressure were recorded by a blinded assessor. Optimal ventilation was assessed by the following criteria:\[[@ref3]\] adequate chest expansion, square wave capnography, and stable oxygenation. Following successful LMA insertion, position of LMA was assessed using fiberoptic bronchoscope and graded.\[[@ref4][@ref5]\] LMA insertion conditions were graded on a three point scale using six variables\[[@ref6]\] -- mouth opening, ease of LMA insertion, swallowing, coughing, patient movements, and laryngospasm \[[Table 1](#T1){ref-type="table"}\]. ###### LMA insertion score  Anesthesia was maintained with isoflurane, nitrous oxide in oxygen, and vecuronium bromide using a standard technique. On completion of the surgery neuromuscular block was antagonized with neostigmine and glycopyrrolate. LMA was removed after deflating the cuff when the patient regained consciousness. Patients were evaluated for sore throat in the immediate postoperative period and for 24 h thereafter. In a pilot study, fentanyl 1 mcg/kg with propofol 2.5 mg/kg given 3 min prior to LMA insertion, was found to provide excellent insertion conditions in 80% of patients. Based on this data, a sample size of *n* = 96 was required to show an improvement of 15% (95%) in LMA insertion success rate with α = 0.05 and a power of 80% with fentanyl in a dose of 2 mcg/kg. Sample size was calculated using Medcalc (Medcalc Software, Mariakerke, Belgium). Statistical analysis was done using chi-square method. Hemodynamic parameters were analyzed by "*t*" test. Results {#sec1-3} ======= The patients in both groups were comparable in terms of age, weight, ASA status, and airway (Mallampati grade). Three patients in each group were excluded from analysis as they needed tracheal intubation as LMA insertion took more than 20 s or more than three attempts were required. Overall LMA insertion and fiberoptic score was comparable \[Tables [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}\] as was time taken and number of attempts \[[Table 4](#T4){ref-type="table"}\]. ###### LMA insertion score  ###### Fibreoptic scoring of the position of LMA\[[@ref4][@ref5]\]  ###### Time taken and attempts taken to insert LMA  There was a significant (a fall of 20% from baseline value) decrease in systolic blood pressure (*P* \< 0.05) at 1, 2, 3, 4, and 5 min and mean arterial pressure at 2 and 3 min after LMA insertion in F~2~, but this fall in blood pressure was not clinically relevant and did not need any intervention. The heart rate did not show any significant changes in either group though the incidence of bradycardia (heart rate \< 60 beats per minute) was higher (16%) in group F~2~ compared to group F~1~. Two of these patients needed atropine IV to reverse bradycardia. Incidence of sore throat was comparable immediately after recovery and 24 h after recovery in both groups. The overall conditions for LMA insertion, mean time taken, total number of attempts taken for insertion, fiberoptic view, and incidence of sore throat were comparable in both study groups. There was a significant fall in systolic blood pressures and mean arterial pressure and higher incidence of bradycardia in the group receiving fentanyl 2 mcg/kg. The patients who received fentanyl 1 mcg/kg remained more hemodynamically stable compared to those receiving fentanyl 2 mcg/kg in our cancer patient population. Discussion {#sec1-4} ========== Induction regime for LMA insertion should provide a depth of anesthesia with a relaxed jaw and suppressed airway reflexes. The optimal insertion conditions for LMA insertion include a relaxed jaw, adequate mouth of opening, and suppressed airway reflexes. This study was designed to compare the insertion conditions when propofol is used with two doses of fentanyl. A dose response study\[[@ref7]\] conducted to predict an optimal dose of fentanyl (placebo and fentanyl 0.5, 1.0, 1.5, and 2.0 mcg/kg) coadministered with propofol 2.5 mg/kg for inserting the LMA Classic™ laryngeal mask airway,) concluded that a standard fentanyl dose of 1.0 mcg/kg coadministered with propofol 2.5 mg/kg provided optimal conditions in 65% of cases. In our study 32 cases (66.7%) in F~2~ group and 24 cases (50%) in F~1~ group had excellent insertion conditions. Satisfactory insertion conditions were found in 25% cases and 43% cases with fentanyl 2 and 1 mcg/kg, respectively. More patients responded adversely to LMA insertion in the F~1~ group but on statistical analysis both groups had comparable results. In this study, four patient responses to LMA insertion were graded. These were swallowing, coughing-gagging, movement (purposeful), and laryngospasm. Patient movement was encountered most frequently and was seen in 20% in group F~1~ and 14.5% in group F~2~. This differs from that of a study comparing LMA Classic insertion conditions following coadministration of alfentanil--propofol and fentanyl--propofol,\[[@ref6]\] which found swallowing to be the most common patient response to LMA insertion. The incidence of coughing-gagging in our study was low with seven patients (14%) in fentanyl 1 mcg/kg group and four patients (8%) in the second study group developing it. Wong *et al*.\[[@ref7]\] found that higher doses of fentanyl were associated with a notable increase in the incidence of coughing and laryngospasm and attributed these episodes to fentanyl rather than to LMA insertion. Laryngospasm was not encountered by us in either study group possibly because we ensured adequate depth of anesthesia and obtundation of airway reflexes, by waiting for 3 min after administration of fentanyl, before attempting LMA insertion. We also used the jaw thrust to assess adequate depth of anesthesia before attempting to insert the LMA. Kodaka *et al*.\[[@ref8]\] studied the effective concentration of propofol for 50% of attempts to secure laryngeal mask insertion (predicted EC50 LMA) using a target-controlled infusion. The predicted EC50 LMA of all fentanyl groups (0.5, 1.0, and 2 mcg/kg) was significantly lower than that of control group, but neither of the fentanyl groups showed a significant difference. Bi-spectral index (BIS) values after fentanyl 1 and 2 mcg/kg were significantly greater than in control and fentanyl 0.5 mcg/kg groups. There were no differences in hemodynamic responses among any groups for any trend. We found that with a fixed dose of propofol there was a statistically significant decrease in systolic blood pressure (*P* \< 0.05) at 1, 2, 3, 4, and 5 min and mean arterial pressure at 2 and 3 min after LMA insertion in group F~2~. The fall was however not clinically relevant and no intervention was needed. The heart rate did not show any significant changes in either group even though the incidence of bradycardia (heart rate \< 60 beats per minute) was higher (16%) in group F~2~ compared to group F~1~. This difference in the incidence of bradycardia was not statistically significant. In patients with poor hemodynamic profiles (e.g., ASA 3 and 4, patients with history of ischemic heart disease, patients with valvular heart disease/using beta blockers), where a tight control of blood pressures and heart rates would be required, the same fall in pressures could become clinically significant. Bradycardia would also have a deleterious effect in such patients. In such cases fentanyl 1 mcg/kg would be a better option, as it would provide optimum LMA insertion conditions along with a more stable hemodynamic profile. Our results indicate that as the preadministered dose of fentanyl was increased from 1 to 2 mcg/ kg, the supplementary doses of propofol required for facilitating LMA Classic™ insertion decreased, even though this decrease was not statistically significant. The supplementary dose of propofol needed by group F~1~ was 7.083 + 18.1 mg and group F~2~ it was 5.625 + 15.97 mg. Kodaka\'s\[[@ref8]\] study is in concordance with our findings. The fibreoptic view depends on the alignment of LMA bowl and glottis and the extent to which the epiglottis is down folded during insertion.\[[@ref9]\] We did not find any statistically significant difference in the fiberoptic view scores in either study group. We had a low rate of immediate postoperative sore throat as compared to Brimacombe and others (0--50%):\[[@ref9][@ref10]\] 33% in fentanyl 1 mcg/kg and 20% in fentanyl 2 mcg/kg group. There was no significant difference in the time taken for LM\[A insertion as well as the number of attempts required in both the groups even though the time taken for LMA insertion was 5 s in two patients in the fentanyl 2 mcg/kg group. In majority of patients (about 90% in each group) the time taken for LMA insertion was in the range of 6--10 s. We were unable to find any published data, comparing time taken for LMA insertion using two doses of fentanyl. However, in a study by Lee *et al*.\[[@ref11]\] on adult patients breathing 8% sevoflurane for induction of anesthesia, pretreatment with low dose fentanyl (1 mcg/kg) and propofol (0.5 mg/kg) was found to significantly decrease the time taken for LMA insertion when compared with induction with sevoflurane alone. The combination of propofol and fentanyl facilitated Classic LMA insertion. Optimal conditions for insertion were obtained with both doses of fentanyl, i.e., 1 and 2 mcg/kg, when propofol was used in a dose of 2.5 mg/kg. Fentanyl 1 mcg/kg provided a more stable hemodynamic profile, with fewer episodes of hypotension and bradycardia. **Source of Support:** Nil **Conflict of Interest:** None declared.

Background {#Sec1} ========== Idiopathic inflammatory myopathies (IIMs) are a group of heterogeneous autoimmune diseases that are accompanied by progressive symmetric muscle weakness, elevated serum levels of muscle enzymes, electromyographic abnormalities, and inflammatory infiltrates observed via muscle biopsy \[[@CR1], [@CR2]\]. Dermatomyositis (DM; juvenile, adult), polymyositis (PM), sporadic inclusion body myositis (sIBM), and immune-mediated necrotizing myopathy (IMNM) are the most common subtypes of IIMs \[[@CR3]\], of which DM and PM are probably the best known \[[@CR4]\]. Classic histopathologic findings for DM are CD4^+^ T cells, B cells, and plasmacytoid dendritic cells (pDCs) infiltrating muscle fibers. Unlike DM, the infiltrating inflammatory cells of PM are CD8^+^ T cells and macrophages \[[@CR5], [@CR6]\]. The infiltrating CD4^+^ T cells can differentiate into Th1 and Th17 cells; Th1 can activate macrophages by producing IFN-γ, and Th17 can facilitate the migration of mononuclear cells to the muscle by producing IL-17 \[[@CR7], [@CR8]\]. The infiltrating CD8^+^ T cells mediate myocytotoxicity via release of perforin \[[@CR9]\]. The infiltrating macrophages can present antigens to T cells and produce multiple cytokines and chemokines \[[@CR10]\]. Additionally, infiltrating pDCs may be the main source of type I interferon \[[@CR11]\]; however, the exact mechanism by which these inflammatory cells migrate into muscles is not fully understood and requires further study. Neutrophil serine proteinases (NSPs) are granule-associated enzymes that play an essential role in blood coagulation, apoptosis, inflammation, and immune responses \[[@CR12]--[@CR14]\]. Recent studies have shown that three NSPs---cathepsin G (CTSG), neutrophil elastase (NE), and proteinase 3 (PR3)---play roles in inflammatory cell migration. These NSPs can cleave the endothelial VE-cadherin, are involved in the possible disruption of endothelial integrity \[[@CR15], [@CR16]\], and can disturb the endothelial cell cytoskeletal architecture to increase vascular permeability \[[@CR17], [@CR18]\]. Moreover, CTSG is a chemoattractant for mononuclear cells and neutrophils \[[@CR19]\]. There are few studies about NSPs and their roles in DM/PM. In our previous microarray analysis, we used total RNA from peripheral blood mononuclear cells (PBMCs) of DM and PM patients, and Illumina HumanHT-12 v4.0 Expression Beadchips (Illumina, Inc.) for mRNA transcription profiling. The platform contained 47,323 transcripts. We identified a total of 2006 differentially expressed genes in patients with DM/PM compared to normal controls (NC) (*P* \< 0.05 and fold change ≥ 2), among which 908 genes were upregulated and 1098 genes were downregulated. CTSG, NE, and PR3 were increased in DM and PM patients (data not shown). In our previous genome-wide DNA methylation analysis, we used the Illumina Human-Methylation 450 K BeadChip array for DNA methylation profiling as described previously \[[@CR20]\]. This platform contains 485,000 CpG sites across the whole genome and covers 99% of RefSeq genes with an average of 17 CpG sites per gene region distributed across the promoter, 5′ UTR, first exon, gene body, and 3′ UTR. It covers 96% of CpG islands with additional coverage in island shores and the flanking regions. The results showed that CTSG, NE, and PR3 were hypomethylated in DM and PM patients; the exact methylation sites are presented in Additional file [1](#MOESM1){ref-type="media"}: Table S1. Previous studies have proven that purified NSPs can increase vascular permeability by cleaving the endothelial VE-cadherin. We hypothesize that in DM and PM patients the upregulation of NSPs cleaves the endothelial VE-cadherin, thus disrupting endothelial integrity and increasing the permeability of vascular tissue and the migration of inflammatory cells to extravascular tissue. To better understand the function of these proteinases, we measured their expression in PBMCs, serum, and muscle tissues, and further explored serum NSPs in the pathogenesis of DM and PM. We found that the expression of NSPs was increased in PBMCs, serum, and muscle tissues of DM and PM patients; additionally, the NSPs were hypomethylated in PBMCs in patients. Moreover, serum NSPs from DM/PM patients cleaved the endothelial VE-cadherin, disrupted endothelial integrity, and increased the migration of PBMCs. Methods {#Sec2} ======= Patients and controls {#Sec3} --------------------- We studied 48 DM patients, 16 PM patients, and 39 normal controls (NC). All patients met the Bohan and Peter diagnostic criteria for DM and PM \[[@CR21], [@CR22]\]. This study was approved by the institutional review board at Xiangya Hospital, Central South University of Changsha (Changsha, Hunan, China). All of the participants in the study signed a written informed consent form prior to participation. RNA isolation and real-time quantitative PCR {#Sec4} -------------------------------------------- Peripheral blood samples were obtained from patients and controls as described previously \[[@CR23]\]. We isolated PBMCs in Ficoll-Paque Plus (GE Healthcare) by density gradient centrifugation. We used Trizol (Invitrogen Life Technologies) to isolate RNA from PBMCs and a Reverse Transcription System (Promega) to obtain cDNA. The relative expression of genes was measured by gene-specific primers (shown in Additional file [2](#MOESM2){ref-type="media"}: Table S2) with SYBR Green (SYBR Premix Ex Taq RT-PCR kit; Takara) and the 7500 real-time PCR system analyzer (Applied Biosystems). DNA isolation and bisulfite pyrosequencing {#Sec5} ------------------------------------------ Genomic DNA was isolated from PBMCs using genomic DNA extraction kits (Life Technologies, Gaithersburg, MD, USA). We used pyrosequencing to validate the microarray methylation data, and DNA samples were bisulfite-converted by the EpiTect Plus DNA Bisulfite Kit (Qiagen). Bisulfite pyrosequencing was performed on a PyroMark Q96 MD pyrosequencing system with the PyroMark Gold Q96 CDT reagent kit (Qiagen). Gene-specific PCR and sequencing primers were designed by the PyroMark Assay Design 2.0 software (Qiagen). The PCR reaction mixture consisted of 12.5 μl of 2× EPIK Amplification Mix, 0.6 μl (10 μM) primer, 20 ng bisulfite-converted DNA, and added water to a total volume of 25 μl (EPIK Amplification Kit; Bioline). Amplifications were performed with an initial denaturation step at 95 °C for 2 min, 40 cycles of 95 °C for 15 s, 56 °C for 15 s, and 72 °C for 30 s, and finally 4 °C for 20 min. The Pyro Q-CpG software (Qiagen) was used for data analysis. Enzyme-linked immunosorbent assay {#Sec6} --------------------------------- The expression levels of CTSG, NE, and PR3 in serum were quantified using the human ELISA kit (Abnova) according to the manufacturer's instructions. The absorbance was measured at 450 nm with a microplate reader. Histological analysis and immunohistochemistry {#Sec7} ---------------------------------------------- We obtained the muscle biopsy specimens from the bicipital muscles of patients and controls. Muscles were frozen in isopentane cooled with liquid nitrogen. For histological examination, we stained the prepared histological sections (8 μm thick) with hematoxylin and eosin (HE). For immunohistochemistry, frozen sections were fixed by acetone, incubated in 3% hydrogen peroxide to block endogenous peroxidase activity, blocked with 10% goat serum for 30 min, and incubated with rat anti-human CTSG (Abcam) overnight at 4 °C. This was followed by application of the secondary antibody for 40 min, avidin--horseradish peroxidase for 45 min, and DAB chromogen. Finally, the slides were counterstained with hematoxylin as described previously \[[@CR24]\]. Immunofluorescence {#Sec8} ------------------ After blocking the frozen sections with 5% bovine albumin, the slides were incubated overnight with rat anti-human NE (Abcam) or rabbit anti-human PR3 (Abcam) at 4 °C followed by the secondary antibody for 1 h without light. We then stained the slides with DAPI for 5 min. Cell culture and treatment {#Sec9} -------------------------- Human dermal microvascular endothelial cells (HDMECs), which were bought from Cellbio (Catalog \#CBR130858), were cultured in Dulbecco's Modified Eagle's medium (DMEM; Gibco) supplemented with 10% heat-inactivated fetal bovine serum (Hangzhou Sijiqing Biological Technology) in an atmosphere containing 5% CO~2~ and 95% air at 37 °C as described previously \[[@CR25]\]. When cells reached 80--90% confluence, we incubated them in serum-free medium overnight. We then exposed the cells to serum from patients and controls or serum from patients pretreated with phenylmethylsulphonyl fluoride (PMSF) for 30 min. Serum was diluted with culture medium at a 1:5 ratio. Western blot analysis {#Sec10} --------------------- A bicinchoninic acid kit was used to calculate the concentration of total proteins. RIPA buffer was used to lyse HDMECs and the total proteins were then denatured at 100 °C for 10 min. Sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) was carried out to isolate the denatured proteins, and the proteins were then transferred to a polyvinylidene difluoride (PVDF) membrane. Five percent skim milk in Tris-buffered saline with Tween (TBST) was used to block the membrane for 1 h, and then the membrane was incubated overnight with rat anti-human VE-cadherin (Abclonal) at 4 °C. The next day, the membrane was washed and incubated with the secondary antibodies. The enhanced chemiluminescence (ECL) method was used to quantify the western blotting results. Endothelial cell tube formation {#Sec11} ------------------------------- Seventy-five microliters of Matrigel matrix (Corning) were added into each well of 96-well plates and then incubated at 37 °C for 30 min to allow the basement membrane to gel. One hundred microliters of HDMECs (10^6^ cells) were added to each well, and cells were incubated in a 5% CO~2~ humidified incubator for 4 h at 37 °C. The tubes were then observed with an inverted microscope. Transendothelial migration of PBMCs {#Sec12} ----------------------------------- A 24-well Transwell with an 8-μm-pore membrane insert (Corning) was used. PBMCs (10^6^ cells) were added to the top chamber on a confluent monolayer of treated HDMECs. The bottom chamber was filled with 500 μl 1640 Media supplemented with 10% heat-inactivated fetal bovine serum (Hangzhou Sijiqing Biological Technology) and 10 ng/ml MIP-1α (Peprotech). Cells were cultured for 24 h at 37 °C in a 5% CO~2~, humidified incubator. Transmigrated cells were counted in the lower membranes after staining with toluidine blue. Statistical analysis {#Sec13} -------------------- Data are shown as the mean ± SEM. Univariate comparisons were made using a one-way ANOVA or two-sample *t* test. Count data comparisons were made using a chi-squared test. Correlations were made using Pearson's *r* coefficient. *P* \< 0.05 was considered statistically significant. The chi-squared test was performed using SPSS; the other tests were performed using GraphPad Prism software. Results {#Sec14} ======= Clinical and laboratory features of DM/PM patients {#Sec15} -------------------------------------------------- All patients and NC were matched for age, sex, and ethnicity (Table [1](#Tab1){ref-type="table"}). DM patients have higher incidence of interstitial lung disease (ILD) than PM patients and PM patients had higher levels of lactic dehydrogenase (LDH) than DM patients, but there was no difference in creatine kinase (CK), erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), immunoglobulin A (IgA), IgM, and IgG, component 3 (C3) and C4, anti-Jo-1 autoantibody, and anti-Ro-52 autoantibody. More DM patients were treated with thalidomide and hydroxychloroquine, which are both effective in the treatment of rashes.Table 1Clinical manifestations and laboratory data of DM patients, PM patients, and controlsClinical characteristicNC (*n* = 39)DM (*n* = 48)PM (*n* = 16)*P* value^a^DM vs NCPM vs NCDM vs PMAge, mean ± SD48.87 ± 12.4050.25 ± 11.9744.41 ± 11.990.6060.21700.0889Sex (male/female)9/3011/376/100.9860.2750.253Disease duration (months)--15.41 ± 21.3423.00 ± 34.25----0.3138CK (U/L)^b^--842.46 ± 1969.144368.46 ± 7149.77----0.061LDH (U/L)^c^--409.43 ± 267.63612.02 ± 432.16----0.032\*ESR (mm/h)^d^--57.93 ± 95.6743.29 ± 34.00----0.542CRP (mg/L)^e^--14.92 ± 20.7212.99 ± 23.47----0.766IgA (mg/L)^f^--2625.22 ± 1408.362660.25 ± 1308.82----0.932IgM (mg/L)^g^--1848.37 ± 1546.841715.38 ± 874.10----0.748IgG (g/L)^h^--29.58 ± 98.3416.70 ± 7.69----0.616C3 (mg/l)^u^--888.65 ± 238.43967.25 ± 319.90----0.318C4 (mg/l)^j^--243.07 ± 127.62299.75 ± 347.79----0.534Jo-1 (positive/negative)--11/375/11----0.505Ro-52 (positive/negative)--16/322/14----0.108ILD (yes/no)--23/233/13----0.029\*Prednisone (yes/no)--46/216/0----0.407Cyclophosphamide (yes/no)--6/421/15----0.488Methotrexate (yes/no)--8/405/11----0.209Azathioprine (yes/no)--9/395/11----0.295Mycophenolate mofetil (yes/no)--4/400/16----0.212Thalidomide (yes/no)--25/231/15----0.001\*Hydroxychloroquine (yes/no)--29/194/12----0.014\*Total glucosides of *Paeonia* (yes/no)--16/327/9----0.452There were no significant differences in age and gender between NC, DM patients, and PM patients*NC* normal controls, *DM* dermatomyositis, *PM* polymyositis, *SD* standard deviation^a^*t* test, \**P* \< 0.05 significant^b^Normal range of creatine kinase (*CK*): 40--200 U/L^c^Normal range of lactate dehydrogenase (*LDH*): 120--250 U/L^d^Normal range of erythrocyte sedimentation rate (*ESR*): 0--20 mm/h^e^Normal range of C-reactive protein (*CRP*): 0--8 mg/L^f^Normal range of immunoglobulin A (*IgA*): 690--3820 mg/L^g^Normal range of immunoglobulin M (IgM): 630--2770 mg/L^h^Normal range of immunoglobulin G (*IgG*): 7.23--16.85 g/L^i^Normal range of complement component 3 (*C3*): 850--1930 mg/L^j^Normal range of complement component 4 (*C4*): 120--360 mg/L Expression of NSPs in DM/PM patients and methylation of NSPs {#Sec16} ------------------------------------------------------------ Real-time PCR was performed to confirm the results of our previous microarray analysis. Our results showed that the relative mRNA levels of CTSG (DM patients 11.09 ± 3.02, PM patients 25.36 ± 8.16), NE (DM patients 5.80 ± 1.54, PM patients 2.50 ± 0.69), and PR3 (DM patients 12.64 ± 2.74, PM patients 9.21 ± 2.54) were significantly upregulated in DM/PM patients compared to those of normal controls (CTSG 1.29 ± 0.32, NE 0.44 ± 0.09, PR3 0.71 ± 0.16) (Fig. [1a](#Fig1){ref-type="fig"}--[c](#Fig1){ref-type="fig"}). The relative expression of CTSG was higher in PM patients than in DM patients (*P* = 0.0046), but there was no difference in the expression of NE (*P* = 0.239) and PR3 (*P* = 0.4828). The results of bisulfite pyrosequencing showed that CTSG (DM patients 22.17 ± 2.70, PM patients 18.44 ± 2.18), NE (DM patients 17.22 ± 1.74, PM patients 14.88 ± 1.67), and PR3 (DM patients 14.28 ± 1.82, PM patients 13.50 ± 2.59) were hypomethylated in DM/PM patients compared to normal controls (CTSG 42.17 ± 1.02, NE 31.26 ± 1.05, PR3 24.05 ± 2.05), but the difference between DM patients and PM patients was not obvious (CTSG *P* = 0.2975, NE *P* = 0.3388, PR3 *P* = 0.8017) (Fig. [1d](#Fig1){ref-type="fig"}--[f](#Fig1){ref-type="fig"}).Fig. 1Expression of CTSG, NE, and PR3 in DM/PM and their methylation. Relative expression of CTSG, NE, and PR3 increased in DM and PM PBMCs compared to controls at RNA level (**a**--**c**). CTSG, NE, and PR3 hypomethylated in PBMCs of DM/PM patients (**d**--**f**). No significant difference in serum levels of CTSG in DM, PM, and controls (**g**). Serum levels of NE and PR3 higher in DM patients and PR3 higher in PM patients than in controls (**h**, **i**). Values are mean ± SEM. CTSG cathepsin G, DM dermatomyositis, GAPDH glycerol-3-phosphate dehydrogenase, NC normal control, NE neutrophil elastase, PM polymyositis, PR3 proteinase 3 CTSG, NE, and PR3 could be released to extracellular space when cells are stimulated. We found that the serum levels of NE (*P* = 0.0005) and PR3 (*P* = 0.0017) were higher in DM patients than in normal controls (Fig. [1h](#Fig1){ref-type="fig"}, [i](#Fig1){ref-type="fig"}), and the levels of PR3 (*P* = 0.0112) were higher in PM patients than in controls. However, the levels of CTSG were almost the same in all three groups (Fig. [1g](#Fig1){ref-type="fig"}). Expression of NSPs in DM/PM patient muscle tissues {#Sec17} -------------------------------------------------- DM patients displayed typical perifascicular atrophy and inflammatory cell infiltration in the perimysial area. In PM patients, the inflammatory cells mainly existed around or invaded non-necrotic muscle fibers (Fig. [2a](#Fig2){ref-type="fig"}). Immunohistochemistry showed that the expression of CTSG was higher in DM/PM patients, especially around myofibers in the muscle tissues. There was no obvious CTSG expression in the controls (Fig. [2b](#Fig2){ref-type="fig"}). The expression of NE and PR3 in muscle tissues was also significantly higher in DM/PM patients (Fig. [2c](#Fig2){ref-type="fig"}, [d](#Fig2){ref-type="fig"}), mainly around myofibers and in perivascular areas.Fig. 2Expression of CTSG, NE, and PR3 in DM/PM muscle tissues. DM sections displayed typical perifascicular atrophy and inflammatory cell infiltration in perimysial area, while inflammatory cells mainly infiltrated around or invaded non-necrotic muscle fibers in HE (**a**). Immunohistochemistry showed an increase of CTSG protein in DM/PM muscle tissues, especially around muscle bundles (**b**). Expression of NE (green) and PR3 (green) also significantly higher in muscle bundles and perivascular areas of DM/PM patients shown by immunofluorescence (**c**, **d**). Nuclei (blue) stained by DAPI. CTSG cathepsin G, DM dermatomyositis, HE hematoxylin and eosin, NC normal control, NE neutrophil elastase, PM polymyositis, PR3 proteinase 3 Correlations between serum NSPs and clinical indicators {#Sec18} ------------------------------------------------------- To determine the relationship between NSPs and clinical indicators of the diseases, correlation analysis was carried out. The results indicated that the serum levels of CTSG in DM/PM patients were positively correlated with LDH, ESR, IgG, and IgA (Fig. [3a](#Fig3){ref-type="fig"}), NE showed positive correlation with CRP and IgM (Fig. [3b](#Fig3){ref-type="fig"}), and PR3 had positive correlation with CRP, ESR, IgG, and IgM (Fig. [3c](#Fig3){ref-type="fig"}). Patients with anti-Jo-1, with anti-Ro-52, or without ILD had lower levels of PR3 (Fig. [3d](#Fig3){ref-type="fig"}). However, the serum levels of CTSG and NE were not significantly different between anti-Jo-1 antibody-positive patients and anti-Jo-1 antibody-negative patients, between anti-Ro-52 antibody-positive patients and anti-Ro-52-negative patients, or between patients with ILD and patients without ILD.Fig. 3Correlations between serum CTSG, NE, and PR3 levels and clinical indicators. Serum levels of CTSG in DM/PM correlated positively with levels of LDH, ESR, IgG, and IgA (**a**). Serum levels of NE in DM/PM correlated positively with CRP and IgM (**b**). Serum levels of PR3 correlated positively with CRP, ESR, IgG, and IgM in DM/PM patients (**c**). Patients with anti-Jo-1, with anti-Ro-52, or without ILD had lower levels of PR3 (**d**). CRP C-reactive protein, CTSG cathepsin G, ESR erythrocyte sedimentation rate, Ig immunoglobulin, ILD interstitial lung disease, LDH lactic dehydrogenase, NE neutrophil elastase, PR3 proteinase 3. \*represents compared with patients, whose anti-Ro-52 antibody or anti-Jo-1 antibody are negative, or patients with ILD *P* \< 0.05 Serum NSPs degraded VE-cadherin and increased the permeability of HDMECs {#Sec19} ------------------------------------------------------------------------ Serum from the patients was applied to HDMEC monolayers to assess its effect on the integrity of junctional proteins. The results showed that serum NSPs significantly degraded VE-cadherin, which could be weakened by their nonspecific inhibitor PMSF (Fig. [4a](#Fig4){ref-type="fig"}, [b](#Fig4){ref-type="fig"}). Serum NSPs could disrupt the tube formation of HDMECs (Fig. [4c](#Fig4){ref-type="fig"}, [d](#Fig4){ref-type="fig"}). Serum NSPs increased the permeability of HDMECs to human PBMCs, yet this increase was attenuated by PMSF (Fig. [4e](#Fig4){ref-type="fig"}).Fig. 4Serum NSPs degraded VE-cadherin, disrupted tube formation, and increased permeability of HDMECs. After stimulating HDMECs with 20% serum from patients for 24 h, expression of VE-cadherin (110 kDa) decreased. VE-Cadherin expression could be neutralized by nonspecific inhibitor PMSF (**a**, **b**). After treating HDMECs with 20% serum from patients, tube formation ability of HDMECs decreased. This function of serum could be alleviated by PMSF (**c**, **d**). After treating HDMECs with 20% serum from patients, permeability of HDMECs to human PBMCs was significantly increased. This increase could be lessened by PMSF (**e**). Values are mean ± SEM. DM dermatomyositis, NC normal control, PM polymyositis, PMSF phenylmethylsulphonyl fluoride Discussion {#Sec20} ========== We reported that the relative expression of NSPs was significantly higher and hypomethylated in the PBMCs of DM/PM patients. NE and PR3 were elevated in DM/PM serum, and there was positive correlation between serum NSPs and clinical indicators, such as ILD, LDH, ESR, CRP, IgG, IgM, IgA, anti-Jo-1 autoantibody, and anti-Ro-52 autoantibody. Serum NSPs degraded the VE-cadherin of HDMECs while increasing the permeability of HDMECs. This study is the first to measure the expression of CTSG, NE, and PR3 in DM/PM patients and to show human dermal microvascular endothelial junctional protein degradation and increased permeability. In our study, DM patients have the higher incidence of ILD. The serum level of LDH was significantly higher in PM patients than in DM patients, and the level of CK had an increasing trend. These findings were in accordance with previous studies because PM patients typically have more severe muscle disease \[[@CR26], [@CR27]\]. DM can be distinguished from PM by its typical cutaneous features, which include heliotrope rash, Gottron's papules, V-sign rash, shawl sign rash, and so on \[[@CR28], [@CR29]\]. Thalidomide has biologic effects on cytokines and cell-mediated responses, and is beneficial in systemic lupus erythematosus, pyoderma gangrenosum, erythema nodosum leprosum, and skin manifestation of DM \[[@CR30], [@CR31]\]. Hydroxychloroquine also has a positive effect on the cutaneous manifestations of DM \[[@CR32]\]. Our DM patients had higher utilization rates of thalidomide and hydroxychloroquine for the treatment of skin lesions related to DM. CTSG, NE, and PR3 are three serine proteinases that are stored in azurophil granules of neutrophils, monocytes, mast cells, and so on \[[@CR33]\]. When the aforementioned cells were stimulated by immune complexes, certain pharmacological agents, or phagocytosis, the three NSPs were either released to the extracellular space or bound to the surface of those cells \[[@CR33], [@CR34]\]. These three NSPs exert several effects, such as processing inflammatory mediators and extracellular matrix proteins, thrombus formation, and engaging protease-activated receptors \[[@CR14]\]. Our study proves for the first time that NSPs were increased in PBMCs, serum, and muscle tissues of DM/PM patients; that the serum levels of NSPs had a positive correlation with the levels of LDH, ESR, CRP, IgG, IgA, and IgM; and that patients with ILD have higher levels of PR3. It has been reported that the serum LDH level has significant correlation with the degree of muscle weakness, muscle destruction, and the degree of muscle inflammation \[[@CR35]\], and that the levels of LDH tend to indicate disease activity \[[@CR27]\]. Serum levels of ESR and CRP are risk factors of interstitial lung disease (ILD) in DM/PM \[[@CR36]--[@CR39]\], elevated ESR is associated with increased mortality in patients with DM \[[@CR37], [@CR40]\], and CRP has positive correlation with the global activity scores in DM \[[@CR41]\]. So, to an extent, these parameters can reflect the activity or prognosis of DM/PM. Further study is needed to identify the exact functions of NSPs in DM/PM patients. Recently, studies have shown that NSPs can enter the endothelial cell, disturb the endothelial cytoskeletal architecture, increase endothelial cell apoptosis, and reduce viability \[[@CR17], [@CR42]\]. NE can disrupt junctional proteins such as E-cadherin and VE-cadherin \[[@CR43]\] and β-catenin \[[@CR44]\], and can break down components of the extracellular matrix \[[@CR45]\]. NE also activates matrix metalloproteinase 9 \[[@CR46]\], thus potentiating the destructive effects on the vasculature and facilitating neutrophil transmigration \[[@CR47]\]. PR3 has been shown to induce the release of CXCL-8 from endothelial cells \[[@CR48]\], to activate matrix metalloproteinases \[[@CR49]\] and to potentiate neutrophil transmigration across the endothelium \[[@CR50]\]. Serum NSPs degraded VE-cadherin in the HDMECs in vitro. VE-cadherin is a critical component of the adherens junction \[[@CR51]\]. Homogeneous VE-cadherin interactions with adjacent cells maintain cell--cell adhesion and contribute to endothelial permeability. Loss of VE-cadherin function greatly reduces microvascular stability and leads to increased permeability and hemorrhage \[[@CR52], [@CR53]\]. Degradation of protein junctions would result in increased permeability of endothelial cells. What is the clinical relevance of our findings? The infiltration of inflammatory cells into muscles plays a key role in the pathogenesis of DM/PM. We identified NSPs as key molecules that degrade primary junctional proteins of the HDMECs, thereby increasing permeability of the HDMECs and facilitating the migration of inflammatory cells. In light of these findings, selective NSP antagonists, with or without antioxidants, may offer future vascular protection and new treatment for DM/PM. Despite the novel and clinically relevant findings in this study, there are some limitations. First, the patients included in our study differed in the course and severity of their disease, and they had different treatment options. Second, we used the serum of patients to stimulate HDMECs, which would not completely rule out the influence of other elements in the serum and the NSPs produced by other cells. Third, the function of NSPs was not confirmed in vivo. Thus, further studies should be done to confirm whether the expression and roles of NSPs in patients have relationships with the course and/or severity of the diseases, and the function of NSPs in the model of DM/PM. Conclusions {#Sec21} =========== Our study showed that NSPs were elevated in DM/PM patients both at RNA and protein levels. The serum NSPs had some relation with clinical indicators, although the exact relationship was not fully understood. Most importantly, serum NSPs degraded the VE-cadherin of HDMECs and increased the permeability of HDMECs to PBMCs. These interactions may play key roles in the development and progression of DM/PM. Additional files ================ {#Sec22} Additional file 1:**Table S1.** Exact methylation sites of CTSG, NE, and PR3 in genome-wide DNA methylation analysis. (DOCX 14 kb) Additional file 2:**Table S2.** Sequences of specific primers of CTSG, NE, PR3, and GAPDH gene used in our study. (DOCX 14 kb) C3 : Component 3 C4 : Component 4 CK : Creatine kinase CRP : C-reactive protein CTSG : Cathepsin G DM : Dermatomyositis DMEM : Dulbecco's modified Eagle's medium ECL : Enhanced chemiluminescence ELISA : Enzyme-linked immunosorbent assay ESR : Erythrocyte sedimentation rate HDMEC : Human dermal microvascular endothelial cell HE : Hematoxylin and eosin IgA : Immunoglobulin A IgG : Immunoglobulin G IgM : Immunoglobulin M IIM : Idiopathic inflammatory myopathy ILD : Interstitial lung disease IMNM : Immune-mediated necrotizing myopathy LDH : Lactic dehydrogenase NC : Normal controls NE : Neutrophil elastase NSP : Neutrophil serine proteinase PBMC : Peripheral blood mononuclear cell pDC : Plasmacytoid dendritic cell PM : Polymyositis PMSF : Phenylmethylsulphonyl fluoride PR3 : Proteinase 3 PVDF : Polyvinylidene difluoride SDS-PAGE : Sodium dodecyl sulfate-polyacrylamide gel electrophoresis sIBM : Sporadic inclusion body myositis TBST : Tris-buffered saline with Tween **Electronic supplementary material** The online version of this article (10.1186/s13075-018-1632-x) contains supplementary material, which is available to authorized users. This work was supported by grants from The National Key Research and Development Program of China (2016YFC0903900), the National Natural Science Foundation of China (81771766, 81771765, 81373206, 81401357, 81671622, 81701621, 81671621), and the Independent Innovation Projects of Central South University (2016zzts131). SG participated in the study design, performed data analysis, and drafted the manuscript. XZ carried out the RNA and DNA isolation. DL carried out the enzyme-linked immunosorbent assay, immunohistochemistry, and immunofluorescence. YX carried out the cell culture and treatment. HZhu coordinated the sample collection, conceived the study, and revised the manuscript. HZha participated in sample collection, real-time quantitative PCR, bisulfite pyrosequencing, and drafted the manuscript. HL conceived the study, participated in the study design, and revised and finalized the manuscript. All authors read and approved the final manuscript. Ethics approval and consent to participate {#FPar1} ========================================== This study was approved by the institutional review board at Xiangya Hospital, Central South University of Changsha (Changsha, Hunan, China). All of the participants in the study signed a written informed consent form prior to participation. Competing interests {#FPar2} =================== The authors declare that they have no competing interests. Publisher's Note {#FPar3} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

1. Introduction {#sec1} =============== Pulmonary edema, a state of abnormal fluid accumulation in the alveolar space of the lung, can interfere with normal oxygenation and ventilation. At baseline, the heart, lungs, and vascular system work in concert to facilitate fluid movement through the capillary and lymphatic beds of the lung ensuring adequate oxygen transport with limited transudation of fluid from the lung interstitium into the alveolar space. In disease states causing acute lung injury (ALI), both direct and indirect, this harmony can be disrupted. Pulmonary edema affects the morbidity and overall mortality of critically ill patients. Patients with pulmonary edema have longer hospital stays, duration of mechanical ventilation, and rates of pneumonia in both adults and children \[[@B1]--[@B3]\]. In adult and pediatric populations, pulmonary edema can complicate the hospital course of any patient, but has notable effects on those with primary respiratory failure, transplant surgery, cardiothoracic surgery, or traumatic brain injury. More recently, clinicians are appreciating the impact of acute kidney injury (AKI) on ventilation and water balance in the lung. As clinical and experimental evidence continues to mount for organ crosstalk, the relationship between the kidney and the lung in critical illness emerges as pathologically significant \[[@B4]\]. The purpose of this brief narrative is to discuss the potential impact of AKI on lung homeostasis, specifically water balance. Numerous laboratory models of isolated AKI have demonstrated deleterious effects on lung pathology and support the existence of kidney-lung crosstalk. Using this background, we propose the concept of nephrogenic pulmonary edema. Though fluid overload is often seen in patients with acute renal failure, we highlight the evidence pointing to a state of pulmonary edema induced by extrarenal microscopic effects of kidney injury. Timely identification of these effects of AKI may aid in the treatment of this disease process which has significant associated morbidity and mortality \[[@B5]\]. 2. Pathophysiology {#sec2} ================== Pulmonary edema occurs when net flow across the alveolar-capillary membrane is positive. Under normal conditions, fluid in the alveolar space is drained by pulmonary lymphatic vessels and then contained within the interstitial space. If the alveolar epithelial barrier is disrupted or if the interstitial fluid volume of the lungs increases by more than 50%, flooding of the alveolar space can occur \[[@B6]\]. The movement of fluid is dependent on both hydrostatic and oncotic pressures and is modeled by the Starling equation $$Q_{f} = K_{f}\left\lbrack \left( P_{C} - P_{T} \right) - \sigma\left( \pi_{C} - \pi_{T} \right) \right\rbrack,$$ where *Q* ~*f*~ is the net flow across the alveolar-capillary membrane; *K* ~*f*~ is the filtration coefficient (dependent on the permeability and the surface area of the membrane); *P* ~*C*~ is the capillary hydrostatic pressure; *P* ~*T*~ is the interstitial fluid hydrostatic pressure; *σ* is the reflection coefficient of the alveolar-capillary membrane (*σ* = 1 if the membrane is impermeable to protein, and *σ* = 0 if the membrane is completely permeable to protein); *π* ~*C*~ is the capillary oncotic pressure; *π* ~*T*~ is the interstitial oncotic pressure \[[@B7]\]. In normal conditions, capillary hydrostatic pressure (*P* ~*C*~) approximates interstitial fluid hydrostatic pressure (*P* ~*T*~), (*P* ~*C*~≃*P* ~*T*~), such that net fluid movement (*Q* ~*f*~) is low and fluid stays in the capillary lumen. Similarly, capillary oncotic pressure (*π* ~*C*~) is normally greater than interstitial oncotic pressure (*π* ~*T*~), and since the reflection coefficient (*σ*) is generally near 1 (0.7--0.95), *Q* ~*f*~ is generally low. Implicit in the equation is that any force or mechanism that creates a large difference between the hydrostatic forces and the oncotic forces will increase *Q* ~*f*~, and thus flux of fluid from the capillary bed into the alveolar space. The lung has safety measures in place to prevent such leakage. The lymphatic system has high capacitance and flow to clear excess fluid from the interstitium. The capillary endothelium between the interstitium and the alveolar beds can tolerate increasing amounts of hydrostatic and interstitial fluid pressure \[[@B8]\]. However, in conditions of excess lung water such as pulmonary edema, overall oncotic pressure becomes dilute (thereby increasing the discrepancy between the hydrostatic and the oncotic pressures) favoring net flow of fluid into the alveolar space ([Figure 1](#fig1){ref-type="fig"}). Aside from bulk fluid and protein effects, changes at the microscopic level change permeability of the alveolar capillary endothelium. Decreased expressions of sodium transport molecules, such as the epithelial sodium channel (ENaC), the sodium-potassium-ATPase pump (Na^+^/K^+^-ATPase), or the cystic fibrosis transmembrane conductance regulator (CFTR), all contribute to increased alveolar fluid accumulation in experimental models of pulmonary edema \[[@B9], [@B10]\]. ENaC promotes sodium absorption from the alveolar space into the lung epithelial cells. Similarly, the sodium ATPase pump allows for sodium extrusion from the alveolus. Aberrant expression of either has significant ramifications for fluid balance as water passively follows sodium. Additionally, the regulation of fluid clearance by endogenous catecholamines, particularly the *β*-adrenergic system, has been highlighted \[[@B11]\]. Variations in vascular preload, capillary endothelial integrity, and aberrations in osmotic pressure can alter the interstitial-alveolar fluid balance. Increased vascular preload leading to pulmonary edema is commonly observed in patients with left-sided cardiogenic shock and disorders of incomplete pulmonary vascular drainage. In these situations, the capacitance of the interstitial lymphatic system is overwhelmed, the *P* ~*C*~ ≫ *P* ~*T*~, and fluid spills from the vascular space down a hydrostatic gradient. Acutely negative intrathoracic pressure raises left ventricular stroke work and likely leads to backup of fluid into the pulmonary bed (increasing *P* ~*C*~) in postobstructive pulmonary edema. The integrity of the alveolar capillary endothelium is altered in acute respiratory distress syndrome (ARDS), sepsis, and stress states \[[@B12], [@B13]\]. Bacterial contribution to type II pneumocyte surfactant production has been linked to increased capillary permeability in sepsis \[[@B14]\]. Such disease states alter the reflection coefficient (*σ*), decreasing the ability of oncotic pressure to balance hydrostatic pressure in the Starling equation. Neurogenic pulmonary edema, commonly observed in intracranial hypertension, may both alter the sympathetic catecholamine regulation of the systemic circulation, thereby diverting blood to the alveolar capillary bed (↑*P* ~*C*~), and disrupt capillary integrity \[[@B15]\]. Finally, acute or chronic disease states commonly carry the comorbidity of hypoproteinemia. Low circulating serum total protein levels directly reduce the capillary oncotic pressure, favoring fluid efflux from the luminal space. The interstitial-alveolar interface in the lung is under tight control and can be affected by disease states that cause dysregulation in total body fluid status, levels of inflammatory mediators, and aberrant protein homeostasis. 3. AKI and Fluid Balance {#sec3} ======================== Acute kidney injury complicates critical illness in many ways. Concomitant with the diagnosis of AKI was fluid overload, which has been found retrospectively to independently increase morbidities such as duration of mechanical ventilation and hospital length of stay, and overall mortality \[[@B16]--[@B18]\]. Fluid overload carries increased risk of morbidity in mortality in both adults and children \[[@B19]\]. On retrospective review, children with greater degrees of fluid overload had higher mortality on initiation of renal replacement therapy \[[@B20]\]. The kidney is central to many homeostatic mechanisms in the body, and as AKI progresses dysregulation occurs in various locations in the body. The kidney communicates with the lung in several ways: regulating acid-base balance, increasing oxygen carrying capacity through erythropoiesis, and regulating blood pressure through the renin-angiotensin-aldosterone axis. All of these processes may be altered in AKI. The deleterious connection between the kidney and the lung in disease states, first reported by Bass and Singer in 1950, was assumed to be secondary to uremia and fluid overload \[[@B21]\]. However, despite the use of dialysis to limit uremia and to control fluid balance, high mortality rates continue in patients with both AKI and ALI \[[@B22]\]. In progressive ALI and ARDS, impaired global oxygenation will necessarily decrease renal oxygenation leading to a vicious cycle of progressive lung and kidney injury \[[@B23]\]. The crosstalk between the kidney and the lung in the critically ill patient is of extreme importance to all critical care physicians \[[@B4]\]. 4. AKI and ALI {#sec4} ============== The impact of AKI on the lung can be seen on many levels. One of the major effects of AKI is to effect lung water balance. While fluid overload from anuric or oliguric AKI can upset the hydrostatic-oncotic balance in the pulmonary interstitium, inflammation and endothelial injury which may be triggered by AKI also upset this equilibrium. The contributions of the latter, inflammation, changes in expression of regulators of the pulmonary fluid and electrolyte chemical gradient, and altered regulation of oxidative stress, may be greater than previously appreciated. AKI triggers gene expression changes which may alter lung vascular stability. Global gene expression mapping of lung tissues in experimental murine kidney ischemia identifies proinflammatory and proapoptotic gene upregulation in the lung transcriptome \[[@B24]\]. Proinflammatory genes such as Cd14, lipocalin-2, chemokine ligand-2 (CXCL2), and IL-6 are all upregulated after ischemia \[[@B24]\]. These mediators, notably IL-6, trigger acute phase responses to antigens and initiate the inflammatory cascade. Subsequent injury is caused by many mechanisms including: disruption of endothelial integrity, aberrant signaling of the coagulation and contact cascades, and direct cellular toxicity in vital organs. Reduction of the IL-6 effect using chemical inhibition of IL-6 or use of IL-6-deficient mice reduces lung inflammation after ischemic AKI \[[@B25]\]. Caspase-3, a marker of cellular apoptotic activity, is upregulated in type II pneumocytes after experimental AKI \[[@B26]\]. Macrophages, mediators of global injury and repair, may be involved; use of macrophage activation inhibitor decreases the amount of pulmonary capillary leakage in murine ischemia \[[@B27]\].Uremia may contribute to such upregulation, as lung inflammation and apoptosis are abrogated after nephrectomy \[[@B28]\]. In repeated animal models of AKI, proinflammatory mediators are expressed within hours of injury. Circulating inflammatory cytokines may affect lung capillary endothelium, changing permeability. Apoptosis of either capillary endothelial cells or pneumocytes responsible for producing surfactant will also change the permeability of the interstitial-alveolar interface in the lung. Cytokine, chemokine, and apoptotic dysregulation is therefore a prime suspect of ALI after AKI and demonstrates kidney-lung crosstalk. Fluid and electrolyte conduction in the lung is altered in AKI. Rabb and colleagues published the seminal work describing the downregulation of pulmonary ENaC, Na/K ATPase, and aquaporin-5 after ischemic injury and nephrectomy in rats \[[@B29]\]. As the primary interstitial solute, sodium is the central determinant of fluid shift based on osmolar gradients. Changes in sodium balance on either side of the capillary membrane, therefore, may have marked effects on fluid shifts between the interstitium and the alveolar bed. Aquaporin-1 message upregulation, in a sheep model of cardiopulmonary bypass and hypothermic circulatory arrest, was correlated with increased rates of pulmonary edema \[[@B30]\]. Similarly, aquaporin-5 has been identified in sepsis as a contributor to dysregulation in tissue and vascular permeability \[[@B31]\]. In experimental models, osmotically driven water permeability between the alveolar and capillary components of the lung is changed tenfold by deletion of aquaporin-1 or aquaporin-5 \[[@B32]\]. While changes in aquaporin expression may have only minor clinical ramification, the evidence for kidney-lung crosstalk is robust \[[@B33]\]. Aberrant salt and water handling by the lung is altered in even mild models of experimental kidney ischemia. The impact of this on pulmonary edema formation is obvious. Oxidative stress plays a role in the kidney-lung crosstalk seen during AKI. Hepatic levels of superoxide dismutase, glutathione, and catalase are decreased after ischemic AKI in mice, which may decrease host response to oxidative stress \[[@B34]\]. Additionally, TNF-*α* levels were significantly higher in ischemic mice than in sham controls pointing to a systemic oxidative/inflammatory response of AKI \[[@B35]\]. The nitric oxide signaling pathway is critical to oxidative balance and is altered in AKI. Patients with chronic kidney disease have lower basal production of nitric oxide compared to controls, and mice subjected to subtotal nephrectomy have lower levels of nitric oxide synthase \[[@B36], [@B37]\]. Finally, mice lacking heme-oxygenase-1 (HO-1), critical for reducing oxidative stress and generation of antioxidant metabolites, have increased levels of inflammation after ischemic AKI versus sham controls \[[@B38]\]. Nitric oxide, HO-1, and TNF-*α* all affect lung capillary stability \[[@B39]--[@B41]\]. Therefore, the alteration in these regulators of oxidative balance after AKI can affect lung vascular stability. The conclusion that significant lung pathology results from ischemic AKI models must be tempered by the fact that, in practice, renal hypoperfusion without concurrent pulmonary hypoperfusion is rare. Indeed, it is rarely known if ischemia to the kidney occurs before or after ischemia to the lungs. Models of hemorrhagic and septic shock do not adequately address this issue. Additionally, in the common case of decreased renal preload (i.e., prerenal azotemia), the microscopic effects on the lung have not been clearly described. Still, the findings described above highlight the potential contribution of the injured renal endothelium to lung injury and edema formation. 5. Nephrogenic Pulmonary Edema {#sec5} ============================== In the context of ischemic AKI and kidney-lung crosstalk, we propose the concept of *nephrogenic pulmonary edema.*Resting on the evidence of experimental extrarenal effects of AKI and the acknowledged clinical effects of oliguric volume overload, nephrogenic pulmonary edema (NPE) could represent net aberrant fluid handling at the interstitial-alveolar lung interface. On a microscopic level, NPE could result as a consequence of damage of pulmonary endothelium by AKI-mediated upregulated inflammation, disruption of pulmonary sodium transport, and activation of apoptosis in pulmonary cells responsible for maintaining the homeostatic integrity of lung water balance. AKI should be counted amongst the numerous disease processes that alter both the volume load in the pulmonary capillary bed and the endothelial integrity of the alveolar capillary barrier. Experimental manifestations of kidney injury are manifest in multiple extrarenal locations \[[@B42]\]. Experimental AKI has deleterious effects on the central nervous system affecting glial cell viability, blood-brain barrier permeability, and neurocognitive status \[[@B43]\]. Systolic ventricular function is negatively affected by ischemic kidney injury \[[@B44]\]. Leukocyte trafficking is aberrant, and a host of proinflammatory genomic and proteomic responses occur after experimental AKI \[[@B24], [@B45]\]. The experimental evidence for lung injury after AKI is robust \[[@B4]\]. Critically ill patients with acute lung injury with AKI have worse morbidity and mortality. Fluid overload and uremia may be regulated by extracorporeal therapy such as dialysis, but lung injury and kidney injury are progressive and mortality rates in patients with both disease processes have not significantly improved with these therapies. Use of stratification criteria has retrospectively demonstrated improved outcomes for patients with less severe AKI \[[@B46]\]; biomarker research seeks to identify kidney injury in its early stages. Appreciation of the impact of kidney injury during this epoch, before the onset of oliguria and fluid overload, may be critical. The impact of kidney-lung crosstalk on the molecular level may be creating a milieu of lung inflammation, mediated by proinflammatory, oxidative, chemical, and apoptotic signals. Ultimately, the accumulated signals overpower the ability of the lung to regulate fluid balance, leading to nephrogenic pulmonary edema. {#fig1} [^1]: Academic Editor: Charlie Strange

1. Introduction =============== Some luminescent materials continue emitting light for minutes or hours after removal of the excitation. This phenomenon, known as **persistent luminescence**, is sometimes undesired, as it limits the use of these materials in applications that require high switching speeds, such as LEDs or information displays. On the other hand, the very long lifetime of the emission without the need of a constant energy input opens up possibilities in many other fields such as, but not limited to, emergency signage \[[@B1-materials-04-00980]\], traffic safety, medical imaging \[[@B2-materials-04-00980]\], dials and displays, and decoration. In 1996, the discovery of bright green afterglow in SrAl~2~O~4~:Eu,Dy \[[@B3-materials-04-00980]\] boosted the search for new and better persistent luminescent materials. Today, several efficient persistent phosphors are known and described \[[@B4-materials-04-00980]\], mainly aluminates such as Sr~4~Al~14~O~25~:Eu,Dy \[[@B5-materials-04-00980]\] and (di)silicates like Sr~2~MgSi~2~O~7~:Eu,Dy \[[@B6-materials-04-00980]\]. Most of these emit light in the blue or green region of the visible spectrum. The lack of potent red-emitting afterglow phosphors, which are desired for emergency signage and medical imaging \[[@B2-materials-04-00980]\], is caused by two reasons: Firstly, the sensitivity of the human eye is low in the orange-to-red region of the visible spectrum. This effect is even more dramatic at low light conditions, which are typical for persistent luminescent applications. Below 1 cd/m², in the mesopic and scotopic regime, the eye sensitivity shifts to shorter wavelengths, a phenomenon known as the Purkinje effect. This makes it very hard to detect red light in dark environments. Because of this, a potential red-emitting persistent phosphor needs a much higher radiance in order to achieve the same apparent brightness as a green- or blue-emitting one \[[@B7-materials-04-00980]\].Secondly, the large majority of persistent luminescent materials are based on divalent europium as luminescent centre. The position of the broad emission spectrum of Eu^2+^ strongly depends on the interaction with the host compound \[[@B8-materials-04-00980]\]. However, in the aforementioned aluminates and silicates, and by extension in almost all oxides, it is difficult to achieve a red-shift that is large enough to obtain red emission from divalent europium. The first problem is inherent to human vision and cannot be solved, but the second one can be addressed by turning to other host materials such as sulfides \[[@B9-materials-04-00980]\] or, in the present case, nitrides. During the past few years, the europium-doped alkaline earth nitrido-silicates M~2~Si~5~N~8~:Eu (M = Ca, Sr, Ba) have gained popularity as wavelength conversion phosphors in white LEDs, because of their large chemical and thermal stability and high quantum efficiency \[[@B10-materials-04-00980]\]. Additionally, promising afterglow properties have been reported in some of these materials \[[@B11-materials-04-00980],[@B12-materials-04-00980],[@B13-materials-04-00980],[@B14-materials-04-00980]\]. Previously, we have shown that Ca~2~Si~5~N~8~:Eu has an orange afterglow, which can be greatly enhanced by Tm codoping \[[@B15-materials-04-00980]\]. To study the influence of the lattice cations on the persistent luminescence we have extended this investigation to include Sr~2~Si~5~N~8~:Eu and Ba~2~Si~5~N~8~:Eu, which have emission spectra very similar to Ca~2~Si~5~N~8~:Eu. As we describe in this paper, the afterglow is not only influenced by codoping, but also depends on the starting materials and the dopant and codopant concentrations. 2. Experimental =============== All powders were prepared using a solid state reaction at 1,400 °C for 3 hours, under reducing atmosphere of forming gas (90% N~2~, 10% H~2~). The host material was made by mixing appropriate amounts of M~3~N~2~ (99%, Alfa Aesar for Ca~3~N~2~ and Cerac for Ba~3~N~2~ and Sr~3~N~2~) and α-Si~3~N~4~ (99.85%, Alfa Aesar). Dopants (europium) and codopants (rare earths, R) were added to the starting mixture in oxide (R~2~O~3~) or fluoride (RF~3~) form (typically 99.9%, Alfa Aesar). Unless mentioned otherwise, the powders were prepared with 1 mol% of Eu and R (*i.e*., 1% substitution of the alkaline earth ions). All materials were weighed, ground and mixed under a protective N~2~ atmosphere in a glove box prior to the solid state reaction. Crystallographic phases of the obtained powders were checked using X-ray diffraction (XRD) with a Bruker D5000 θ--2θ diffractometer using Cu-Kα radiation and compared with literature data \[[@B16-materials-04-00980],[@B17-materials-04-00980]\]. The band gap was estimated through diffuse reflection measurements using a Varian Cary 500 spectrophotometer equipped with an integrating sphere. The photoluminescent properties (emission, excitation) of the samples were measured with a fluorescence spectrometer (FS920, Edinburgh Instruments). Afterglow decays were measured with both the aforementioned spectrometer and a calibrated photometer (ILT 1700, International Light Technologies). Microsecond decay measurements were performed using a pulsed nitrogen laser setup (λ~exc~ = 337 nm, pulse length 800 ps, repetition rate 1 Hz) and a 1024-channel intensified CCD (Andor Technology) attached to a 0.5 m Ebert monochromator. Thermoluminescent curves were obtained at Delft University of Technology, The Netherlands. The samples were exposed to two minutes of 300 nm irradiation at room temperature and the emission during heating was monitored using an OceanOptics QE65000 CCD-based spectrometer. 3. Results and Discussion ========================= 3.1. Structural Properties -------------------------- The M~2~Si~5~N~8~ compounds crystallize in two different structure types. The orthorhombic Sr~2~Si~5~N~8~ and Ba~2~Si~5~N~8~ are isostructural, both having space group Pmn2~1~ \[[@B17-materials-04-00980]\]. In contrast, Ca~2~Si~5~N~8~ has a monoclinic structure of space group Cc \[[@B16-materials-04-00980]\]. Because of this, the properties of the different compounds cannot be readily compared. For example, the position of the emission spectrum of M~2~Si~5~N~8~:Eu^2+^ does not shift continuously by changing the alkaline earth ion. Also the excitation spectrum of Ca~2~Si~5~N~8~:Eu differs significantly from those of Sr~2~Si~5~N~8~:Eu and Ba~2~Si~5~N~8~:Eu \[[@B10-materials-04-00980]\]. Upon (co)doping, the rare earth ions occupy the lattice sites of the alkaline earth cations in the host. Due to the different size of the rare earth ions, the lattice parameters slightly change upon (co)doping. For example, in Ca~1.94~Eu~0.02~Tm~0.04~Si~5~N~8~ (where Eu^2+^ is somewhat larger than Ca^2+^, and Tm^3+^ is notably smaller \[[@B18-materials-04-00980]\]) the volume of the unit cell is reduced by around 1.7% compared to the undoped compound. The type of dopant or codopant has no significant influence on the crystal structure. Diffuse reflection spectra on undoped samples (not shown) were performed to verify the band gap of the materials. Ca~2~Si~5~N~8~ has the largest band gap (around 250 nm), followed by Sr~2~Si~5~N~8~ (265 nm) and Ba~2~Si~5~N~8~ (280 nm). These values are the same as found by Li *et al.* \[[@B10-materials-04-00980]\], except for Ba~2~Si~5~N~8~ where they found a band gap of 270 nm. 3.2. Photoluminescent Properties -------------------------------- [Figure 1](#materials-04-00980-f001){ref-type="fig"}a shows the steady state emission spectra (emission spectra recorded during excitation) of Ca~2~Si~5~N~8~:Eu, Sr~2~Si~5~N~8~:Eu and Ba~2~Si~5~N~8~:Eu. In all three cases, the spectrum consists of one broad Eu^2+^-based band (FWHM of about 100 nm). The broadness of the emission can be explained by the existence of two similar cation sites in the host lattice \[[@B16-materials-04-00980],[@B17-materials-04-00980]\]. Since the europium ions occupy these cation sites, and the position of the emission band is closely related to the europium surroundings, the spectrum will consist of two bands, in this case mostly overlapping \[[@B10-materials-04-00980],[@B11-materials-04-00980]\]. The emission maximum shifts from 580 nm (yellowish orange) for M = Ba, over 610 nm (orange) for M = Ca, to 620 nm (reddish orange) for M = Sr. It could be expected that the position of the emission maximum shifts continuously when changing the cation from Ca over Sr to Ba. However, Ca~2~Si~5~N~8~:Eu does not follow this trend because of its different crystal structure compared to Sr~2~Si~5~N~8~:Eu and Ba~2~Si~5~N~8~:Eu. For concentrations below 5%, the wavelength position of the maxima is in all powders only slightly dependent on the europium concentration \[[@B10-materials-04-00980]\]. The steady state excitation spectra are given in [Figure 1](#materials-04-00980-f001){ref-type="fig"}b. These are very broad, and extend well into the visible region, up to 550 nm. On the short wavelength side, the samples show exciton absorption in the region just below the band gap. The remaining bands can be attributed to the 4f^7^ → 4f^6^5d^1^ transition in the Eu^2+^ ions, with a maximum in a broad region around 400 nm. {#materials-04-00980-f001} 3.3. Afterglow -------------- After excitation, all materials show some form of persistent luminescence. [Figure 2](#materials-04-00980-f002){ref-type="fig"} shows the afterglow intensity as a function of time after 1 minute excitation with an unfiltered xenon arc lamp at 1000 lux. The decay curves follow straight lines in a log-log plot, implying that they can be modeled by a power law with negative scaling exponent. In other words, the decay is very fast initially, but slows down over time. This type of behavior is typical for most persistent luminescent materials. {#materials-04-00980-f002} The afterglow duration can be defined as the time between the end of the excitation and the moment when the afterglow intensity drops below 0.32 mcd/m², about 100 times the sensitivity of the human eye (a threshold value often used in industrial standards \[[@B7-materials-04-00980],[@B19-materials-04-00980]\]). The afterglow duration is longest for Ba~2~Si~5~N~8~:Eu, around 400 seconds, followed by Ca~2~Si~5~N~8~:Eu, around 150 seconds. Sr~2~Si~5~N~8~:Eu has a very short afterglow duration of 80 seconds. This rapid afterglow decay is an advantage, since Sr~2~Si~5~N~8~:Eu is a popular conversion phosphor in GaN or InGaN-based white LEDs, where a long afterglow is undesired \[[@B20-materials-04-00980]\]. The afterglow spectra of Ca~2~Si~5~N~8~:Eu and Ba~2~Si~5~N~8~:Eu are red-shifted about 5--10 nm compared to the steady state spectra ([Figure 3](#materials-04-00980-f003){ref-type="fig"}a; for Sr~2~Si~5~N~8~:Eu the afterglow was too weak to measure this with sufficient accuracy. A similar red-shift in the persistent spectrum is also seen in Ca~2~SiS~4~:Eu,Nd \[[@B21-materials-04-00980]\] and CaAl~2~Si~2~O~8~:Eu \[[@B22-materials-04-00980]\]. Measurements of the decay behavior in the first moments after the end of the excitation show that this red-shift occurs in the first 5--10 µs after the end of the excitation ([Figure 3](#materials-04-00980-f003){ref-type="fig"}a), after which the spectrum remains unchanged during the remainder of the afterglow. This means that both crystallographic sites available for europium take part in the persistent luminescence. If only one of the sites showed persistent luminescence, the decay time of the other would be determined solely by the lifetime of Eu^2+^ (around 1 µs \[[@B23-materials-04-00980]\]), and the persistent emission spectrum would reach its final position on a much faster timescale than the observed 5--10 µs. Secondly, the integrated emission intensity in the first microseconds after the excitation does not show a clear exponential component with a lifetime of 1 µs ([Figure 3](#materials-04-00980-f003){ref-type="fig"}b), which would be expected if one of the sites did not take part in the persistent luminescence. Translating these observations to a defect model is not straightforward and remains somewhat speculative. A first possibility is formed by defects (which serve as trap state) being proximate to the europium ions, thus altering the local environment. This could then lead to the observed red-shift in the afterglow spectrum. Another possibility could be the affinity for traps to be dominantly coupled to only one of the europium sites. Differences in energy transfer behavior between both sites for the steady state emission and the afterglow might occur also, although this is less likely given that the steady state emission spectrum of Ca~2~Si~5~N~8~:Eu is invariable for changes in excitation wavelength or intensity. Clearly more in-depth research is required to explain this observation. {#materials-04-00980-f003} 3.4. Rare Earth Codoping ------------------------ In many persistent luminescent materials the afterglow can be greatly influenced by adding traces of various rare earths. The most notable example is SrAl~2~O~4~:Eu,Dy, where the presence of Dy codoping increases the afterglow brightness by almost two orders of magnitude \[[@B3-materials-04-00980]\]. The influence of each rare earth differs for every host material. To optimize the afterglow, Dy is the most common codopant, closely followed by Nd. However, other choices are possible for specific hosts, such as Tm, Ho, Y or Ce \[[@B4-materials-04-00980]\]. The reason behind this positive (or negative) influence on the afterglow is the subject of an ongoing debate. Most authors agree that the codopant ions influence the charge carrier traps in the material, which are responsible for the persistent luminescence. However, some researchers believe the rare earth ions merely influence the depth of pre-existing traps in the host crystal, such as oxygen vacancies (for example, Clabau *et al.* \[[@B19-materials-04-00980]\]), while others assume that they act as electron traps themselves (such as Dorenbos \[[@B24-materials-04-00980]\] and Aitasalo *et al.* \[[@B25-materials-04-00980]\]). In the former case, the trap depth is related to the ionization potential of the different codopant ions \[[@B19-materials-04-00980]\]. In the latter case, the 4f^7^ ground state levels of the doubly ionized rare-earths (trivalent codopant with a trapped electron) determine the depth of the traps \[[@B24-materials-04-00980]\]. To study the influence of rare earth ions on M~2~Si~5~N~8~:Eu, we codoped samples with 1% of dysprosium, neodymium, samarium or thulium, and compared the persistent luminescence with the material without codoping. This codoping has no effect on the emission spectrum. [Figure 4](#materials-04-00980-f004){ref-type="fig"}a shows the afterglow intensity 1 minute after excitation with 300 nm (violet) light. As can be seen, the codopant strongly influences the afterglow in Ca~2~Si~5~N~8~:Eu,R and Ba~2~Si~5~N~8~:Eu,R. For Sr~2~Si~5~N~8~:Eu,R no afterglow of importance could be achieved with any of the codopants. For Ba~2~Si~5~N~8~:Eu,R codoping with dysprosium does not considerably influence the afterglow. The other codopants have a negative effect on the afterglow in Ba~2~Si~5~N~8~:Eu,R. In Ca~2~Si~5~N~8~:Eu,R a strong influence of the codopant is seen. Dysprosium and neodymium enhance the afterglow intensity considerably, while samarium reduces it by more than 40%. The most spectacular effect is obtained with thulium codoping. In this case, the afterglow intensity after one minute is nearly six times as high as in the non-codoped sample. [Figure 4](#materials-04-00980-f004){ref-type="fig"}b shows the effect of all rare earth codopants (including yttrium, but excluding promethium) on the afterglow in Ca~2~Si~5~N~8~:Eu,R (λ~exc~ = 400 nm) proving again that thulium codoping is the best choice for persistent luminescence in this host. The afterglow duration after 1 min excitation with a Xe arc lamp at 1000 lux is around 2,500 s ([Figure 2](#materials-04-00980-f002){ref-type="fig"}). The other rare earths, except for neodymium and dysprosium, have a negative or no effect on the afterglow intensity. {#materials-04-00980-f004} 3.5. Thermoluminescence ----------------------- The charge carrier traps, which are presumably created or affected by the codopant ions, can be investigated by thermoluminescence (TL) experiments. To collect a TL spectrum, the sample is excited at low temperature, after which it is heated (with a fixed heating rate β). During heating, the light emission is monitored, resulting in one or more broad peaks ('glow peaks') in an intensity-*vs*-temperature diagram. In this way an estimate of the trap depths can be made, since deeper traps require more thermal energy to be emptied, and hence will cause peaks at higher temperatures. It is usually assumed that the ideal trap depth for persistent luminescence at room temperature is about 0.65 eV \[[@B4-materials-04-00980]\], corresponding to glow peaks around 70--100 °C, depending on the heating rate. [Figure 5](#materials-04-00980-f005){ref-type="fig"}a compares the glow curves obtained during a TL measurement for Ca~2~Si~5~N~8~:Eu and the thulium-codoped variant, after 300 nm excitation at room temperature for 2 minutes, with a heating rate β of 2.5 °C/min. The curves were analyzed with the *TL Glow Curve Analyzer* software \[[@B26-materials-04-00980]\]. Nevertheless, care should be taken when analyzing these glow curves. The broadness of the peaks, especially on the high-temperature side, indicates that a single first- or second-order trap description is insufficient and multiple traps or trap distributions might be present. Therefore, the trap depths presented here are only indicative. In both the non-codoped and the thulium-codoped case the shape of the peak is similar, suggesting that the same trap is responsible for the afterglow. However, upon codoping with thulium the peak location shifts to higher temperatures, around 100 °C, indicating that the trap depth has increased to a more suitable depth for persistent luminescence. Assuming general order kinetics, the trap depth is estimated to shift from 0.87 eV to 0.91 eV upon codoping. {#materials-04-00980-f005} To illustrate that the same codopant can have different effects in different host crystals, the glow curves for Ba~2~Si~5~N~8~:Eu and Ba~2~Si~5~N~8~:Eu,Tm are shown in [Figure 5](#materials-04-00980-f005){ref-type="fig"}b. In this case, the glow peak shifts to lower temperatures, in other words, the charge carrier trap becomes shallower (0.71 eV in the non-codoped case, 0.68 eV in the thulium-codoped sample, again assuming general order kinetics). 3.6. Influence of the Starting Materials ---------------------------------------- The afterglow intensity of the samples is not only influenced by the codopants, but also by the starting materials chosen during the preparation. For example, when the dopant and codopant ions are added to the starting mixture in their fluoride form (RF~3~), the afterglow 10 minutes after excitation is over 12 times brighter than if oxides (R~2~O~3~) are used. This is most probably due to the more favorable chemical properties of fluorides compared to oxides, for example their lower melting point. This makes it easier for the materials to diffuse into the host crystal during the preparation. XRD measurements show an increased crystallinity when using fluorides, indicating that the fluorine also acts as a flux material during the reaction. EDX investigations of Ca~2~Si~5~N~8~:Eu,Tm have revealed the formation of undesired micron-sized Tm- and O-rich aggregates upon preparation with oxides (not shown). Analysis of these aggregates indicates that they contain comparable concentrations of Tm, Si and N, and almost no Ca. This suggests the formation of the non-luminescent TmSiO~2~N phase. When using fluorides some Tm-rich aggregates were still formed, but in a lower concentration. Here, the equal concentrations of Ca and Tm and the hexagonal structure of the aggregates suggest the formation of CaTmSi~4~N~7~. No fluorine was encountered in any of the samples. A second effect of the starting materials is of a stoichiometric nature. A 5% deficiency of Ca~3~N~2~ in the starting mixture for Ca~2~Si~5~N~8~:Eu,Tm increases the afterglow intensity nearly threefold. This might be due to the fact that the dopant and codopant ions are located at Ca-sites. Hence an increase of Ca-vacancies, caused by the Ca~3~N~2~ deficiency, might facilitate the incorporation of rare earth ions in the host, because of charge compensation effects. However, as noted above, the nature and location of the traps that lead to persistent luminescence is still debated, and further research is necessary to study the incorporation of the rare earths and their influence on the afterglow. At even higher deficiencies of Ca~3~N~2~, the brightness of the persistent luminescence decreases again. XRD measurements show that a (non-luminescent) α-Si~3~N~4~ phase remains in these samples. 3.7. Dopant and Codopant Concentration -------------------------------------- A final point of interest concerns the concentration of the dopant and codopant ions in the host crystal. As with normal fluorescence, a high concentration of luminescent centers is usually undesired in persistent luminescence, due to concentration quenching. This is confirmed by [Figure 6](#materials-04-00980-f006){ref-type="fig"}a, showing the afterglow intensity in Ca~2~Si~5~N~8~:Eu,Tm after 10 minutes as a function of Eu concentration, where the Tm codopant concentration is kept at 1%. Clearly, it is best to keep the dopant concentration below 1%. {#materials-04-00980-f006} The afterglow brightness as a function of codopant concentration is shown in a similar fashion in [Figure 6](#materials-04-00980-f006){ref-type="fig"}b, where the Eu concentration is kept constant at 1%. The effect is even more dramatic here: a Tm concentration of 2% reduces the afterglow intensity to 10% of the value at a concentration of 0.5%. Previous SEM and EDX investigations have revealed the formation of (non-luminescent) Tm-rich aggregates in the material when the concentration becomes too high \[[@B15-materials-04-00980]\]. This undesired phenomenon can be partially countered by increasing the preparation duration and temperature. Still, a Tm concentration above 1% is undesired for persistent luminescence in Ca~2~Si~5~N~8~:Eu,Tm. 4. Conclusions ============== The M~2~Si~5~N~8~ (M = Ca,Sr,Ba) family of materials was synthesized using a solid state reaction, doped with Eu as luminescent centers, and codoped with Nd, Dy, Sm and Tm. All samples show persistent luminescence, but for all M = Sr samples this is very weak. The afterglow spectra are red-shifted by about 5--10 nm compared to the steady state spectrum. In Ba~2~Si~5~N~8~:Eu the afterglow lasts about 400 s, which can be slightly increased by Dy codoping. Ca~2~Si~5~N~8~:Eu,Tm shows by far the brightest and longest afterglow, lasting over 2,500 s. Thermoluminescence measurements indicate that the addition of Tm in Ca~2~Si~5~N~8~:Eu deepens the relevant charge trap to a more suitable depth. For optimal persistent luminescence, the dopant and codopant should be added to the starting mixture in their fluoride form, in concentrations not higher than 1 mol%. During the preparation, a deficiency of Ca~3~N~2~ of about 5% compared to the stoichiometric amount triples the intensity of the afterglow. Koen Van den Eeckhout is supported by the Special Research Fund (BOF) of Ghent University. Pieter Dorenbos, Adrie Bos and Erik van der Kolk from Delft University of Technology, The Netherlands, are acknowledged for assistance in TL measurements.

INTRODUCTION ============ The advent of microsurgery in the 1960\'s is considered to be one of the most important milestones in recent plastic and reconstructive surgery \[[@B1]\]. Although there is no generally agreed definition, microsurgery may be defined as surgery requiring an operating microscope. The first to coin the term \"microsurgery\" was the vascular surgeon, Jules Jacobson, who anastomosed blood vessels with a diameter smaller than 1.4 mm with the aid of a microscope \[[@B2]\]. Nowadays many surgical subspecialties, such as plastic surgery, trauma surgery, neurosurgery, and maxillofacial surgery, use microsurgery in their clinical routine. Successful microsurgical operations require sufficient training and experience using delicate instruments and suture material of 8/0 and less. In addition, optical magnification should be mandatory for precise handling of tissue and sutures. Optical magnification can be achieved using loupes \[[@B3]\] or regular operating microscopes \[[@B4]\]. However, optical magnification may be neglected for several reasons such as lack of availability of devices, laborious preparation of the operating microscope, or a surgeon\'s unwillingness to use an operating microscope. During a surgical procedure, in which certain delicate structures such as peripheral nerves are involved, it is not legally mandatory to use optical magnification in the form of loupes or an operating microscope. Nevertheless, exact approximation of neural or vascular tissues is imperative for a successful microsurgery. The purpose of this paper is to objectively evaluate the impact of optical magnification on the success of microsurgery. The presented results should serve as a guideline for the novice microsurgeon on when to use optical magnification in various clinical situations. METHODS ======= Neurophysiological fundamentals ------------------------------- The eye has a visual acuity threshold below which an object will go undetected. The standard definition of normal visual acuity (20/20 vision) is the ability to resolve a spatial pattern separated by a visual angle of one minute of arc (MOA). Because one degree contains 60 minutes, a visual angle of one MOA is 1/60 of a degree. Furthermore, one MOA equals 60 arcseconds. The spatial resolution limit is derived from the fact that one degree of a scene is projected across 288 µm of the retina by the eye\'s lens. In this 288 µm, there are 120 color sensing cone cells packed. Thus, if more than 120 alternating white and black lines are crowded side-by-side in a single degree of viewing space, they will appear as a single gray mass to the human eye. Microsurgical suture material of size 11/0 presents a thread thickness of 50×10^-6^ m. For instance with a working distance of 50 cm, the thread is depicted on the retina with a visual angle of 20.6 arcseconds according to the following formula: alpha \[degree\]=(180/Pi)×arctan (50×10^-6^/0.5). Considering solely the optical abilities of the human eye, the 11/0 thread may be impossible to visualize. The thread is merely visible because it is contrasted, creating a diffraction/flexion phenomena, which are perceptible as a misalignment (Vernier acuity). Vernier acuity refers to the ability to determine when two parallel straight lines are exactly in line \[[@B5]\]. The perceptibility depends on the brightness, wavelength, place, size, and duration of the manipulation as well as of the retinal adaptation. The contrast sensitivity of the human eye depends on the light-dark changes per visual angle. A ×6 magnification would lead to an image or depiction of the above-mentioned thread on the retina with a visual angle of 120 arcseconds, which is sufficient for a purely optical perception. Thus, the discrimination of two structures lying next to each other is made possible, and consequently, the surgical manipulations are under absolute visual control. Practical experiments --------------------- The authors are members of maz, which is a microsurgical training and research center. Persons with an interest in microsurgery are trained at various levels of expertise, starting from complete novices, to participants who have had some exposure to microsurgery learning basic microsurgical techniques, to medical doctors in a surgical specialty dealing with various microsurgical problems learning operative dissection. In this study, during the basic microsurgical courses, 16 voluntary participants (age range, 22 to 28 years old; mean, 25.8 years) with comparable levels of expertise in surgery were asked to perform the following experiments. ### Experiment 1 In the first experiment, the participants had to coapt an intentionally cut ischiadic nerve 3 cm distal to the femoral head on the chicken leg \[[@B6],[@B7]\] without using a magnifying optical device (exercise 1), with 2.5×magnifying glasses (EyeMag Smart, Zeiss, Oberkochen, Germany) (exercise 2), and with an operating microscope (OPMI pico, Zeiss) (exercise 3). The mean diameter of the chicken nerve at the thigh level was 1.28±0.13 mm. The magnification of the microscope could be adjusted randomly but had to be above 6×magnification. The exercise, which had to be performed first, was randomized by lot. Standard microsurgical instruments (Aesculap, Tuttlingen, Germany) from our training courses and suture material 10/0 (Monosof, Covidien, Dublin, Ireland) were used. All coaptations had to be carried out using interrupted stitches. The specimen (chicken legs), with a weight of 248 to 291 g (average, 273 g), were purchased from a supermarket. Chicken legs are a standard training model in our center \[[@B8]\]. ### Experiment 2 In the second experiment, the participants had to anastomose a porcine coronary vessel 7-9 with a vascular lumen of approximately 1.5 mm. For all of the anastomosis, interrupted stitches had to be used. Again, the end-to-end anastomosis was performed without a magnifying optical device (exercise 1), with 2.5 magnifying glasses (exercise 2), and with an operating microscope (exercise 3). The magnification of the microscope could be adjusted randomly but had to be above 6×magnification. Which exercise among the three had to be performed first was randomized by lot. Standard microsurgical instruments from our training courses and suture material 9/0 (Monosof, Covidien) were used. The pig hearts were purchased fresh and unopened from a slaughterhouse and had a weight of 353 to 487 g (average, 405 g). Pulsatile perfused pig heart coronary vessels are a standard training model in our center \[[@B9]\]. Assessment ---------- Evaluation of the sutured nerves and vessels was performed by 2 experienced microsurgeons using an operating microscope. Each mistake was counted only once for each participant. The specimens (chicken legs and porcine hearts) were blinded to the evaluators by two persons not involved in the experiment or evaluation. The anastomoses were cut open to be examined on all of the sides, and the nerve was also dissected. Mistakes were noted and documented with a digital camera connected to the microscope. The types of mistakes collected after the nerve repair included (experiment 1): 1) Insertion depth: an insertion depth mistake means that not only the epineurium but also the axon was captured with the stitch; 2) Rotation: there is a noticeable rotational mistake in aligning the ends of the cut nerve; 3) Nerve damage through pressure and suture asymmetry: there is noticeable damage in the integrity of the nerve through inadequate forceps pressure or misalignment of the interrupted sutures. On the other hand, the types of mistakes recorded after porcine coronary vessel anastomosis exploration included (experiment 2): 1) Backwall sutures: any stitches that captured the opposing backwall and thereby occluded the lumen of the vessel; 2) Distance between stitches and leakage: distances between each interrupted stitch that are too wide and will cause leakage; leakage was checked separately by perfusing the vessel with an ink-colored fluid; 3) Suture asymmetry and rotation: visible difference between the edges of the vessel that were captured within a single interrupted suture, and a gross rotational mistake causing kinking of the vessel; 4) Vessel damage through pressure: noticeable alterations of the integrity of the vessel wall through pressure injury caused by forceps. Statistical analysis -------------------- Statistical analysis was performed using the McNemar test to compare the unadjusted categorical data of the exercise groups obtained during the two experiments. All of the tests were two-sided. The differences were considered statistically significant at P≤0.05. RESULTS ======= All of the 16 participants completed all the necessary experiments. The results for each experiment for the 16 students are given in [Tables 1](#T1){ref-type="table"}, [2](#T2){ref-type="table"}. By choosing 16 voluntary participants of the basic microsurgical courses with comparable surgical levels, we were able to minimize the surgical-related variables in performing the described experiments. Experiment 1 group ------------------ Within this group, the number of mistakes made during the three exercises (without a magnifying optical device, with ×2.5 magnifying glasses, and with an operating microscope) did not differ significantly in rotation, nerve damage through pressure, or suture asymmetry. On the other hand, comparisons across the outcomes obtained within the three types of exercises yielded significant differences in insertion depth, with far greater superiority of the operating microscope over the surgical loupes and simple naked-eye sutures (P\<0.01). Finally, the analysis showed a significant difference in the number of accurate neurorrhaphies of the nerve stumps in favor of the operating microscope over the surgical loupes (P\<0.01) and sutures without magnification (P\<0.01) ([Table 1](#T1){ref-type="table"}). Experiment 2 group ------------------ Leakage, rotation, and vessel damage through pressure did not differ significantly among the three sets of experiments for vessel sutures. Anastomosis performed with any kind of magnification (loupes or microscope) presented a significantly better constant distance and symmetry between the stitches, whereas the operating microscope allowed a significantly lower number of backwall sutures (P\<0.01) than the other methods. Comparisons of accuracy of the specimen showed significant superiority (P\<0.01) of the operating microscope over the surgical loupes and simple naked-eye sutures ([Table 2](#T2){ref-type="table"}). DISCUSSION ========== Microsurgical techniques have revolutionized many fields of modern surgery. In the history of microsurgery, the development and advancement of magnifying devices has always played a pivotal role. However, it is important and remarkable to note that the pioneers of microvascular surgery, such as Carrel \[[@B10]\], Hopfner \[[@B11]\], and Guthrie \[[@B12]\], did have the benefits of intraoperative magnification. The history of optical magnification started with Lippershey and Janssen in 1590 when they first built a compound microscope by reversing a telescope \[[@B13]\]. Anton van Leeuwenhoek created the next hallmark in the development of magnifying devices, building powerful transportable microscopes with a basic design used without significant changes until the 20th century \[[@B14]\]. Carl Zeiss and Ernst Abbe combined their craftsmanship with their physics knowledge to create high quality microscopes with apochromatic lenses, which reduced chromatic aberrations even more than had been achieved with achromatic lenses; this was another important developmental step \[[@B15]\]. The microscope entered the operating theater when Nylen \[[@B16]\] treated a patient with chronic otitis media using a microscope. The event that marked the historical beginning of microvascular surgery was credited to Jacobson and Suarez \[[@B17]\] who first used a microscope for microvascular anastomosis. This hallmark innovation laid the foundation for all the great achievements in modern reconstructive microsurgery, such as thumb replantation \[[@B18]\] or free flap transplantation \[[@B19]\]. All of these remarkable developments made clear that adequate optical magnification is mandatory for successfully performing microsurgery. However, the application of such magnifying devices, such as the operating microscope or loupes, may pose an additional workload. Thus, surgery is performed without the aid of these devices in order to complete the operation within a reasonable amount of time. Typical examples of such clinical scenarios might include a cutting injury to the finger with laceration of the digital nerve that is sown in the emergency room or a severed digital artery during fasciectomy that has to be repaired. Although most readers know that such clinical problems call for surgery under magnification, at times external or internal causes prevent the surgeon from making the extra effort to use magnification. Another factor worth mentioning is that there is no legal requirement to use optical magnification under certain circumstances. Thus, it is not mandatory for a surgeon repairing a digital nerve to at least wear loupe magnification. The results of our study showed that performing routine microsurgical tasks without any magnification device will produce significantly more mistakes. The current literature provides abundant information that learning and practicing microsurgery yields better results for a novice microsurgeon \[[@B20]-[@B24]\]. The goal and intention of our investigation was to show that different surgical tasks require different magnifying devices for achieving optimal results. To the best of our knowledge, this type of information has not been published yet. The results should serve as a guideline for new surgeons to understand what type of magnifying device is appropriate for a given surgical problem. During our basic microsurgical courses, we try to offer novice microsurgeons a sense and feeling for a delicate approach to specific tissues even if they do not pursue a career directly involved with microsurgery. We firmly believe that it is a great success when surgeons are aware of the fact that some clinical scenarios suggest a microsurgical approach for achieving superior results. This also means that a surgeon may refer patients to an appropriate institution for microsurgery. We understand that the results of this study do not make the use of magnification devices mandatory or even legally binding for surgeons. However, this is the first experiment to demonstrate that the eye has limited abilities in certain surgical situations, and this limitation will lead to mistakes with a surgical outcome that is inferior to that which could be achieved with a magnification device. Therefore, our study should increase awareness of microsurgery and the intrinsic ties to adequate optical magnification in this surgical specialty. By conducting practical experiments in a standardized manner with novice microsurgeons, we were able to prove that microsurgical success is directly related to optical magnification. The human eye\'s ability to discriminate potentially important anatomical structures is limited, which might be detrimental to clinical results. Although not legally mandatory, microsurgical operations such as reparative surgery after hand trauma should be conducted with magnifying devices for achieving an optimal patient outcome. No potential conflict of interest relevant to this article was reported. ###### Overview and comparison of results and for 3 sets of experiments for nerve coaptation  Values are presented as number of mistakes. WM, without magnification; SL, surgical loupes; OM, operating microscope. ^a)^Values of P ≤ 0.05 were considered statistically significant. ###### Overview and comparison of results for 3 sets of experiments for vessel anastomosis  Values are presented as number of mistakes. WM, without magnification; SL, surgical loupes; OM, operating microscope. ^a)^Values of P ≤ 0.05 were considered statistically significant.

Introduction {#sec1} ============ The copolymerization of carbon dioxide (CO~2~) and epoxides is an attractive means to valorize an abundant, renewable feedstock and to produce useful polymers with reduced greenhouse gas emissions.^[@ref1]−[@ref3]^ Coupling reactions between CO~2~ and epoxides yield cyclic carbonates as the thermodynamic products,^[@ref4],[@ref5]^ but with a suitable catalyst the kinetic polycarbonate product may also be accessible.^[@ref6]−[@ref8]^ Catalyst selection, therefore, is critical to governing reaction selectivity and rates, as well as providing a means to control polymer properties such as molar mass and dispersity.^[@ref9],[@ref10]^ In this field of homogeneous catalysis, the majority of the research has focused on metal complexes of cobalt(III), chromium(III), and zinc(II), with the best catalysts showing high activity and excellent polycarbonate selectivity.^[@ref11]^ In many cases, the metal complexes are applied together with an ionic cocatalyst with the combination being referred to as a binary catalyst system.^[@ref6],[@ref10]^ The cocatalyst significantly improves the catalyst activity and selectivity, although its role is somewhat complex. Different researchers propose that the cocatalyst is important to labilize the propagating alkoxide and carbonate intermediates, to provide counterions for coordination with metal-free polymer chains, to "hold" the propagating ionic (metal-free) chains close to the metal center, and to limit backbiting reactions to form cyclic carbonates.^[@ref6],[@ref10]^ Here, Ti(IV) catalyst systems are selected for investigation due to their earth abundance, lack of color, and precedence in related catalytic processes.^[@ref12]−[@ref14]^ In particular, tripodal M(IV) complexes have an excellent track record in lactide and lactone ring-opening polymerization (M = Ti(IV), Zr(IV), Hf(IV)).^[@ref13],[@ref15]−[@ref20]^ The complexes combine outstanding rates, high tolerance, operation at low loadings, and stability under industrially relevant high-temperature polymerization conditions.^[@ref13],[@ref16]−[@ref20]^ Although the mechanism for epoxide/CO~2~ ROCOP is different from lactone polymerization, there are some similarities in catalyst requirements. For example, both cycles apply Lewis acidic metal centers to bind the monomers (epoxide or lactone), and both invoke labile metal alkoxide intermediates as propagating species. To deliver catalysts relevant for larger-scale deployment, the metal applied should be colorless, thermally stable, and earth abundant, show low toxicity, and lack redox chemistry under the conditions of catalysis; Ti(IV) complexes fit these criteria, and furthermore, there is already some precedent for their application in cyclohexene oxide (CHO)/CO~2~ ROCOP.^[@ref21]−[@ref28]^ For example, Nozaki and co-workers pioneered group IV catalyst systems, applying a tetradentate Ti(IV) complex, with bis(triphenylphosphine)iminium chloride (PPNCl), to achieve a turnover frequency (TOF) of 76 h^--1^ (0.05 mol %, 50 °C, 20 bar of CO~2~) with quantitative polycarbonate selectivity (\>99%) ([Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}).^[@ref27]^ Ti(IV) complexes coordinated by phenolate-N-heterocyclic carbene^[@ref26]^ or salalen^[@ref24],[@ref29]^ ligands also showed good catalytic activities of 55 h^--1^ (0.04 mol %, 80 °C, 10 bar of CO~2~) and 41 h^--1^ (0.2 mol %, 90 °C, 40 bar of CO~2~), respectively ([Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}). Encouraged by these prior successes for Ti(IV) complexes, we reasoned that Ti(IV) amino-tris(phenolate) complexes could be promising catalyst systems for CHO/CO~2~ ROCOP. The target Ti(IV) complexes are also related to the successful Al(III) and Fe(III) catalyst systems for epoxide/CO~2~ ROCOP developed by Kleij and co-workers ([Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}).^[@ref30]−[@ref34]^ Kleij's catalysts show outstanding activity for CO~2~/epoxide coupling to cyclic carbonates and a CHO/CO~2~ ROCOP TOF of ∼6 h^--1^ (0.5 mol %, 85 °C, 2 bar of CO~2~).^[@ref31],[@ref34]^ The Al(III)--amino tris(phenolate) catalyst system is also notable, as it is one of only two catalysts active for CO~2~/limonene oxide ROCOP.^[@ref33],[@ref35]^ ![Structures of reported Ti(IV) catalysts along with an active Al(III)−amino tris(phenolate) complex for ROCOP of CO~2~/CHO.^[@ref24],[@ref26],[@ref27],[@ref32]^](om9b00845_0001){#fig1} Results and Discussion {#sec2} ====================== Two ligands were targeted, H~3~L^tBu^ and H~3~L^Me^, and their syntheses and subsequent complexation reactions to Ti(IV) to yield complexes **1**--**3** were carried out according to literature procedures (see the [Supporting Information](http://pubs.acs.org/doi/suppl/10.1021/acs.organomet.9b00845/suppl_file/om9b00845_si_001.pdf) for full experimental protocols and [Figures S1--S12](http://pubs.acs.org/doi/suppl/10.1021/acs.organomet.9b00845/suppl_file/om9b00845_si_001.pdf)).^[@ref14],[@ref19],[@ref36],[@ref37]^ The successful ligand complexation to Ti(IV) was confirmed using ^1^H NMR spectroscopy; one diagnostic feature is the formation of diastereotopic benzylic resonances at 3.96 and 2.49 ppm. Complexes **1**--**3** show ^1^H and ^13^C NMR spectra that are fully consistent with previous reports,^[@ref37]^ and elemental analysis results indicate acceptable analytical purity. The catalysts were tested under a standard set of conditions using 40 bar of CO~2~ pressure, 80 °C, a catalyst loading of 0.05 mol % (1:2000 catalyst:CHO) in neat cyclohexene oxide (CHO) (9.9 M), and catalyst:cocatalyst = 1:1 ([Table [1](#tbl1){ref-type="other"}](#tbl1){ref-type="other"}). All of the complexes displayed high CO~2~ and polymer selectivity (\>99%) and showed moderate polymerization activities from 10 to 35 h^--1^ ([Table [1](#tbl1){ref-type="other"}](#tbl1){ref-type="other"}, entries 1, 3, and 5). Complexes **2** and **3**, featuring methyl ligand substituents, have higher activity than **1**, featuring *tert*-butyl substituents, and this higher activity may arise from a reduced steric hindrance at Ti(IV) accelerating rates of epoxide coordination or insertion (vide infra). The similar activities of **2** and **3** suggest that the coligand does not substantially accelerate the rate, and this is consistent with its proposed function as a polymer chain initiating group. When the polymerization temperature was increased to 100 °C, the catalytic activity increased significantly, reaching a TOF of 81 h^--1^ (catalyst **2**), but these conditions also resulted in a significant reduction in polymer selectivity (\<65%) and in the formation of cis and trans cyclic carbonate products ([Table [1](#tbl1){ref-type="other"}](#tbl1){ref-type="other"}, entries 2 and 4). The performance of complexes **1**--**3** showed significantly enhanced activity and selectivity in comparison to the homoleptic titanium alkoxide precursor complex Ti(O^*i*^Pr)~4~ and better performance than a catalyst system comprising Ti(OiPr)~4~/PPNCl ([Table S1](http://pubs.acs.org/doi/suppl/10.1021/acs.organomet.9b00845/suppl_file/om9b00845_si_001.pdf)). This finding demonstrates the importance of the ancillary ligand in mediating Lewis acidity, oxophilicity, and lability of the Ti(IV) active site. ###### ROCOP of CO~2~/CHO Using Catalysts **1**--**4**/PPNCl[a](#t1fn1){ref-type="table-fn"} entry catalyst *T* (°C) time (h) conversn (%)[b](#t1fn2){ref-type="table-fn"} CO~2~ (%)[c](#t1fn3){ref-type="table-fn"} polym (%)[d](#t1fn4){ref-type="table-fn"} TON[e](#t1fn5){ref-type="table-fn"} TOF (h^--1^)[f](#t1fn6){ref-type="table-fn"} *M*~n~ \[*Đ*\] (kg mol^--1^)[g](#t1fn7){ref-type="table-fn"} ------- -------------------------------------------------------- ---------- ---------- ---------------------------------------------- ------------------------------------------- ------------------------------------------- ------------------------------------- ---------------------------------------------- -------------------------------------------------------------- 1 **1** 80 24 13 \>99 \>99 263 11 17.1 \[1.04\]                   7.6 \[1.16\] 2 **1** 100 16 17 \>99 80 342 21 7.5 \[1.21\] 3 **2** 80 24 42 \>99 \>99 840 35 36.8 \[1.05\]                   10.7 \[1.33\] 4 **2** 100 16 65 \>99 65 845 81 37.6 \[1.06\]                   8.5 \[1.56\] 5 **3** 80 24 40 \>99 \>99 800 33 37.9 \[1.11\]                   16.1 \[1.07\] 6 **4** 80 24 32 \>99 \>99 640 27 21.6 \[1.02\]                   9.7 \[1.04\] 7 \[(Boxdipy)Ti^IV^Cl\][h](#t1fn8){ref-type="table-fn"} 60 12 45 \>99 \>99 900 76 13.0 \[1.27\] 8 \[(NHC)Ti^IV^Cl~2~\][i](#t1fn9){ref-type="table-fn"} 80 24 53 \>99 \>99 1325 55 8.3 \[1.53\] 9 \[(Salalen)Ti^IV^Cl\][j](#t1fn10){ref-type="table-fn"} 70 10 44 98 98 220 22 4.2 \[1.11\] Reaction conditions unless specified otherwise: catalyst (0.05 mol %), cocatalyst (0.05 mol %), CHO (8 mL, 9.9 M), 40 bar of CO~2~. Expressed as a percentage of CHO conversion vs the theoretical maximum (100%), determined from the ^1^H NMR spectrum by comparison of the relative integrals of the resonances assigned to the carbonate (4.65 ppm for PCHC and 4.00 ppm for trans cyclic carbonate) and ether (3.45 ppm) linkages against CHO (3.00 ppm). Expressed as a percentage of CO~2~ uptake vs the theoretical maximum (100%), determined by comparison of the relative integrals of the ^1^H NMR resonances due to carbonate (4.65 ppm for the polymer and 4.00 ppm for the trans cyclic carbonate) and ether (3.45 ppm) linkages. Expressed as a percentage of polymer formation vs the theoretical maximum (100%), determined by comparison of the relative integrals of the ^1^H NMR resonances due to the polymer (4.65 ppm) and trans cyclic carbonate (4.00 ppm). Turnover number (TON) = number of moles of cyclohexene oxide consumed/number of moles of catalyst. Turnover frequency (TOF) = TON/time (h). Determined by GPC, in THF, calibrated using narrow-*M*~n~ polystyrene standards. Catalyst (0.01 mmol), cocatalyst (0.01 mmol), CHO (20 mmol).^[@ref27]^ Catalyst (0.008 mmol), cocatalyst (0.008 mmol), CHO (20 mmol), 10 bar of CO~2~.^[@ref26]^ Catalyst (0.2 mol %), cocatalyst (0.2 mol %), CHO (9.9 M), 40 bar of CO~2~.^[@ref29]^ Catalyst systems **1**--**3** all produce poly(cyclohexene carbonate) (PCHC) showing bimodal molar mass distributions, as evidenced by GPC analysis of the crude polymer samples. The bimodal distributions arise due to polymer chains being initiated from both 1,2-cyclohexanediol, formed by reaction between the epoxide and residual water, and from chloride, present in the cocatalyst. The presence of both catalyst-initiated chains and telechelic chain transfer agent-initiated chains gives rise to two molar mass distributions, one of approximately double the molar mass of the other.^[@ref38]^ To understand the influence, if any, of water on the catalysis and on the putative catalytic intermediates, complex **2** was reacted with 10 equiv of water, in THF at 25 °C, and stirred for 2 h. This reaction evolved the known oxo-bridged Ti(IV) dimer complex \[(L^Me^Ti)~2~O\] (**4**) in 32% yield ([Scheme [1](#sch1){ref-type="scheme"}](#sch1){ref-type="scheme"}).^[@ref39]^ The formation of **4** was confirmed using ^1^H and ^13^C NMR spectroscopy, where the loss of the isopropoxide coligand peaks, at 5.34 ppm (^1^H NMR) and 78.1 ppm (^13^C NMR), in comparison to precursor complex **2** was clearly observed ([Figures S11 and S12](http://pubs.acs.org/doi/suppl/10.1021/acs.organomet.9b00845/suppl_file/om9b00845_si_001.pdf)). To establish whether the dimer showed any catalytic activity, it was subjected to identical polymerization conditions, with the addition of a cocatalyst, where it showed performance nearly identical with that of catalyst system **2**/PPNCl ([Table [1](#tbl1){ref-type="other"}](#tbl1){ref-type="other"}, entry 6). This result was somewhat surprising, as the dimeric complex does not feature any obvious initiating groups. {#sch1} The catalytic performance of **2**/PPNCl was compared to three of the most active Ti(IV) catalysts reported in the literature. Although detailed comparisons are premature, as it is not clear that all of the catalysts function with identical rate laws and mechanisms, it can be seen that qualitatively these new catalysts show activity values which are consistent with previous investigations. The most active Ti(IV) catalyst system remains that reported by Nozaki and co-workers, \[(Boxdipy)TiCl\] (1:2000 catalyst:CHO), which shows approximately double the activity of **2** and functions at a lower temperature. The phenolate-carbene Ti(IV) complex \[(NHC)Ti^IV^Cl~2~\] is also slightly more active than **2** and shows a TOF of 55 h^--1^. In contrast, **2** is more active than \[(Salalen)Ti^IV^Cl\] but considerably less active than a related Ti(III) salen catalyst reported by the same group (TOF = 161 h^--1^, 0.1 mol % catalyst, 40 bar of CO~2~, 80 °C).^[@ref24]^ With a successful catalyst system in hand, it was of interest to understand the limits of the catalyst concentration on copolymerization. Holding the catalyst:cocatalyst ratio constant at 1:1 and using progressively greater quantities of epoxide resulted in effective catalysis even at 0.025 mol % catalyst loading ([Table [2](#tbl2){ref-type="other"}](#tbl2){ref-type="other"}, entry 4). ###### Effect of Catalyst Loading and Cocatalyst on the ROCOP of CO~2~/CHO Using Catalyst **2**[a](#t2fn1){ref-type="table-fn"} entry cocat cat:cocat:CHO time (h) conversn (%)[b](#t2fn2){ref-type="table-fn"} CO~2~ (%)[c](#t2fn3){ref-type="table-fn"} polym (%)[d](#t2fn4){ref-type="table-fn"} TON[e](#t2fn5){ref-type="table-fn"} TOF (h^--1^)[f](#t2fn6){ref-type="table-fn"} *M*~n~ \[*Đ*\] (kg mol^--1^)[g](#t2fn7){ref-type="table-fn"} ------- ---------- --------------- ---------- ---------------------------------------------- ------------------------------------------- ------------------------------------------- ------------------------------------- ---------------------------------------------- -------------------------------------------------------------- 1 PPNCl 1:1:500 4 48 \>99 \>99 235 59 15.7 \[1.02\]                   6.8 \[1.05\] 2 PPNCl 1:1:1000 9 42 \>99 \>99 420 47 19.1 \[1.03\]                   8.1 \[1.06\] 3 PPNCl 1:1:2000 24 42 \>99 \>99 840 35 36.8 \[1.05\]                   10.7 \[1.33\] 4 PPNCl 1:1:4000 48 31 \>99 \>99 1240 26 13.1 \[1.05\]                   5.4 \[1.09\] 5 N/A 1:0:2000 24 0 \- \- \- \- \- 6 DMAP 1:1:2000 24 21 \>99 \>99 420 18 8.4 \[1.10\]                   3.1 \[1.14\] 7 PPh~4~Cl 1:1:2000 24 59 \>99 \>99 1180 49 24.5 \[1.04\]                   10.9 \[1.05\] Reaction conditions: 80 °C, neat CHO (8 mL, 9.9 M), 40 bar of CO~2~. Expressed as a percentage of CHO conversion vs the theoretical maximum (100%), determined from the ^1^H NMR spectrum by comparison of the relative integrals of the resonances assigned to the carbonate (4.65 ppm for PCHC and 4.00 ppm for trans cyclic carbonate) and ether (3.45 ppm) linkages against CHO (3.00 ppm). Expressed as a percentage of CO~2~ uptake vs the theoretical maximum (100%), determined from the ^1^H NMR spectrum by comparison of the relative integrals of the resonances assigned to the carbonate (4.65 ppm for PCHC and 4.00 ppm for trans cyclic carbonate) and ether (3.45 ppm) linkages. Expressed as a percentage of polymer formation vs the theoretical maximum (100%), determined from the ^1^H NMR spectrum by comparison of the relative integrals due to the polymer (4.65 ppm) and trans cyclic carbonate (4.00 ppm). Turnover number (TON) = number of moles of CHO consumed/number of moles of catalyst. Turnover frequency (TOF) = TON/time (h). Determined by GPC, in THF, calibrated using narrow-*M*~n~ polystyrene standards. Also, all the copolymerizations proceed with quantitative CO~2~ and polymer selectivity (\>99%) and there is no detectable formation of ether linkages or cyclic carbonate byproducts by ^1^H NMR spectroscopy. The catalytic activity decreases as progressively greater amounts of CHO are used because the overall catalyst concentration falls. It is clearly essential to apply a cocatalyst, since using only complex **2** resulted in no polymerization at all. The cocatalyst selection was also important in controlling activity, and the most effective systems comprised complex **2** in combination with ionic cocatalysts such as PPNCl and Ph~4~PCl. These ionic cocatalyst systems were about twice as active as when Lewis base cocatalysts such as DMAP were applied. The finding that ionic cocatalysts show superior activity is also consistent with previous studies of Ti(IV) or other metal complexes for CO~2~/epoxide ROCOP, but an understanding of the catalyst/cocatalyst speciation remains somewhat ill-defined.^[@ref40],[@ref41]^ Next the influence of copolymerization temperature on the activity of catalyst **2** was assessed ([Table [3](#tbl3){ref-type="other"}](#tbl3){ref-type="other"}). The lowest temperature at which effective catalysis occurred was 60 °C, and as expected, the activity increases upon increasing the temperature to 80 °C (20 to 35 h^--1^). Increasing the temperature from 100 to 120 °C continued to increase the activity but compromised the reaction selectivity with the production of significant quantities of cyclic carbonate byproduct (35--66%, respectively). The cyclic carbonate comprised both cis and trans isomers, which indicates that polymer chain backbiting occurs from both metal alkoxide and carbonate intermediates. The temperature selectivity behavior is consistent with backbiting from the metal carbonate intermediate occurring more quickly than epoxide insertion and forming the *cis*-cyclohexene carbonate and, concomitantly, with backbiting from the alkoxide intermediate to form the *trans*-cyclohexene carbonate. To rule out a change in reaction selectivity being caused by catalyst degradation at elevated temperatures, complex **2** was heated for 24 h over the temperature range 80--120 °C and the speciation monitored by ^1^H NMR spectroscopy. No observable change in the ^1^H NMR spectrum of complex **2** was measured over the temperature range explored, indicating good catalyst stability over the typical conditions of the polymerizations ([Figure S13](http://pubs.acs.org/doi/suppl/10.1021/acs.organomet.9b00845/suppl_file/om9b00845_si_001.pdf)). ###### Effect of Temperature on the ROCOP of CO~2~/CHO Using Catalyst **2**/PPNCl[a](#t3fn1){ref-type="table-fn"} entry temp (°C) time (h) conversn (%)[b](#t3fn2){ref-type="table-fn"} CO~2~ (%)[c](#t3fn3){ref-type="table-fn"} polym (%)[d](#t3fn4){ref-type="table-fn"} TON[e](#t3fn5){ref-type="table-fn"} TOF (h^--1^)[f](#t3fn6){ref-type="table-fn"} *M*~n~ \[*Đ*\] (kg mol^--1^)[g](#t3fn7){ref-type="table-fn"} ------- ----------- ---------- ---------------------------------------------- ------------------------------------------- ------------------------------------------- ------------------------------------- ---------------------------------------------- -------------------------------------------------------------- 1 60 48 47 \>99 \>99 940 20 36.8 \[1.05\]                 15.4 \[1.55\] 2 80 24 42 \>99 \>99 820 35 36.8 \[1.05\]                 10.7 \[1.33\] 3 100 16 65 \>99 65 320 81 37.6 \[1.06\]                 8.5 \[1.56\] 4 120 8 54 \>99 34 160 135 13.8 \[1.07\]                 5.1 \[1.15\] Reaction conditions: **2** (0.05 mol %), cocatalyst (0.05 mol %), CHO (8 mL, 9.9 M), 40 bar of CO~2~. Expressed as a percentage of CHO conversion vs the theoretical maximum (100%), determined from the ^1^H NMR spectrum by comparison of the relative integrals of the resonances assigned to the carbonate (4.65 ppm for PCHC and 4.00 ppm for trans cyclic carbonate) and ether (3.45 ppm) linkages against CHO (3.00 ppm). Expressed as a percentage of CO~2~ uptake vs the theoretical maximum (100%), determined from the ^1^H NMR spectrum by comparison of the relative integrals due to carbonate (4.65 ppm for PCHC and 4.00 ppm for trans cyclic carbonate) and ether (3.45 ppm) linkages. Expressed as a percentage of polymer formation vs the theoretical maximum (100%), determined from the ^1^H NMR spectrum by comparison of the relative integrals due to polymer (4.65 ppm) and trans cyclic carbonate (4.00 ppm). Turnover number (TON) = number of moles of cyclohexene oxide consumed/number of moles of catalyst. Turnover frequency (TOF) = TON/time (h). Determined by GPC, in THF, calibrated using narrow-*M*~n~ polystyrene standards. Controlled polymerization catalysis is useful because it allows for prediction of the polycarbonate molar mass from the starting concentrations of catalyst and monomer conversion. Catalysts showing high polymerization control can also be useful for the preparation of block polymers and, more recently, are important in switch catalysis, where multiple polymerization cycles are accessed by a single catalyst to enchain monomer mixtures into multiblock polymers.^[@ref42],[@ref43]^ To assess the polymerization control, a set of identical copolymerizations, using **2**/PPNCl, was stopped at progressively increasing time periods. The crude polymer, PCHC, was analyzed by GPC and NMR spectroscopy. A plot of PCHC molar mass vs CHO conversion shows a linear fit and the resulting polymers show molar masses with narrow dispersity (*Đ* \< 1.15). Overall the molar mass distributions are all bimodal, consistent with the growth of both α-chloro-PCHC and hydroxyl telechelic PCHC ([Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}). {#fig2} To understand the speciation of the key propagating species, a single crystal, suitable for X-ray diffraction, was obtained by vapor diffusion of pentane into a saturated solution of complex **2** and PPh~4~Cl, in THF at −40 °C under a nitrogen atmosphere. Structural elucidation revealed a hexacoordinate titanium anionic complex whereby the titanium center is coordinated by the ligand, the isopropoxide coligand, and an additional chloride, from the added cocatalyst Ph~4~PCl ([Figure [3](#fig3){ref-type="fig"}](#fig3){ref-type="fig"} and [Figure S15](http://pubs.acs.org/doi/suppl/10.1021/acs.organomet.9b00845/suppl_file/om9b00845_si_001.pdf)). The titanium atom adopts an octahedral geometry and is coordinated to the four donor atoms of the ligand (O, O, O, N), the isopropoxide coligand and the chloride. It is important to note that the formation of such an anionic metal "ate" catalytic intermediate has been proposed in several ROCOP mechanisms but there are only a few examples of the isolation of metal complexes showing this coordination mode.^[@ref26],[@ref44]−[@ref47]^ One recent example, reported by Le Roux and co-workers, features a hafnium "ate" complex, formed by reaction between a hafnium diphenoxy-carbene chloride complex and PPNCl, and the anionic species was proposed as the active site for epoxide ring opening in the ROCOP mechanism.^[@ref48]^  in the Supporting Information).](om9b00845_0003){#fig3} To rationalize the performance of catalyst systems **1**--**3** and the molecular structure of the putative intermediate, a polymerization catalytic pathway is proposed ([Figure [4](#fig4){ref-type="fig"}](#fig4){ref-type="fig"}). The mechanism relates closely to that proposed by Kleij and co-workers to rationalize the performance of Al(III) catalyst systems and which was substantiated by DFT calculations.^[@ref31]^ Accordingly, the pathway requires two Ti(IV) complexes for catalysis and has parallels with bicomponent mechanisms proposed for other metal complex/cocatalyst systems.^[@ref31],[@ref33],[@ref48]^ The initiation reaction involves a neutral Ti(IV) complex coordinating the epoxide, while a second anionic Ti(IV) "ate" complex (\[LTi(OR)Cl\]^−^) attacks the coordinated epoxide using the chloride nucleophile. This reaction generates a titanium alkoxide "ate" intermediate (\[LTi(OR)~2~\]^−^) which can undergo carbon dioxide insertion to form a titanium carbonate "ate" intermediate (\[LTi(OR)(O~2~COR)\]^−^). The carbonate intermediate propagates the chain by attacking and ring-opening an activated epoxide (coordinated at a second Ti(IV) center) and by re-forming the titanium alkoxide intermediate. Propagation reactions involve repeating the cycles between anionic Ti alkoxide and Ti carbonate intermediates. At higher reaction temperatures, the catalytic selectivity is reduced as barriers to chain backbiting reactions are overcome both from the Ti alkoxide (to form *cis*-cyclohexene carbonate) and Ti carbonate (to form *trans*-cyclohexene carbonate) intermediates. Under optimized conditions (i.e., temperature ∼80 °C), the backbiting pathways to cyclic carbonate are not accessible and the propagation cycles continue until the monomer is consumed or until the reaction is quenched. Reactions are quenched by exposure to air and the addition of solvents (not predried) to precipitate the polymer. The model studies into the reaction between **2** and excess water showed that a dimeric titanium μ-oxide species, **4**, formed. This species would be expected to be catalytically inactive, but surprisingly **4**/PPNCl showed performance nearly equivalent to that of the system using **2**. It is clear that there is water present during the copolymerizations, since GPC data show bimodal mass distributions consistent with initiation from 1,2-cyclohexanediol (formed by a reaction between epoxide and water), which is a chain-transfer agent. Thus, a key question that remains unanswered is whether that residual water reacts more quickly with epoxide to form 1,2-cyclohexanediol (CHD) or with catalyst to form the dititanium μ-oxide species **4**. Both possibilities would result in the formation of bimodal PCHC distributions, and on the basis of the catalysis using **4**, both processes allow onward propagation. The mechanism by which **4** propagates could be similar to that proposed for **2**/PPNCl: i.e., **4** reacts with PPNCl to generate the anionic intermediate \[L~2~Ti~2~(O)Cl-alkoxide\]^−^, which coordinates an epoxide and ring-opens it by nucleophilic chloride coligand attack. The anionic dititanium alkoxide intermediate \[L~2~Ti~2~(O)(OR)\]^−^ inserts carbon dioxide to form a carbonate intermediate, \[L~2~Ti~2~(O)(O~2~COR)\]^−^. Another possibility is that the "ate" complexes are in equilibrium with neutral Ti(IV) complexes and anionic polymer chains which propagate "off-metal" ([Figure S16](http://pubs.acs.org/doi/suppl/10.1021/acs.organomet.9b00845/suppl_file/om9b00845_si_001.pdf)). Future investigations should focus on a detailed analysis of the polymerization kinetics and use of operando analytical techniques to fine-tune the mechanistic hypotheses. {#fig4} Conclusions {#sec3} =========== The utilization and transformation of CO~2~ into value-added products remains a significant challenge in order to reduce greenhouse gas emissions and fossil fuel consumption involved in polymer production. The sequestering of CO~2~ into polycarbonate polyols, for subsequent use in resin or polyurethane production, is a credible utilization method and delivers products suitable for a range of applications. Titanium(IV) catalysts confer advantages including low toxicity, high elemental abundance, good temperature stability, high stability to oxidation/reduction side reactions, and a lack of color. The titanium complexes reported herein, with the addition of cocatalyst, are active catalyst systems for the ROCOP of carbon dioxide and cyclohexene oxide. They show activity values which are similar to those of other reported titanium(IV) complexes. The isolation of a rare anionic titanium complex, formed by a stoichiometric reaction with the cocatalyst, supports the proposed role of the cocatalyst as an activated nucleophile. The complexes are the bedrock upon which future ligand and catalyst optimization studies should be conducted to improve activity and allow for yet further insight into the polymerization mechanism. For example, Le Roux and co-workers recently demonstrated higher rates of CO~2~/CHO ROCOP by changing the active metal in the catalysts from Ti(IV) to Hf(IV).^[@ref48]^ They also observed much better activity at low CO~2~ pressure (1 bar). The use of larger ionic radii metals may also allow for mononuclear mechanisms, free from the need for cocatalyst,^[@ref49]^ and investigations into heavier group IV congeners are warranted. Experimental Section {#sec4} ==================== General Experimental Details {#sec4.1} ---------------------------- All manipulations involving air- or moisture-sensitive compounds were carried out using standard Schlenk line or glovebox techniques. 2,4-Dimethylphenol, 2,4-di-*tert*-butylphenol, hexamethylenetetramine, and bis(triphenylphosphine)iminium chloride (PPNCl) were purchased from Sigma-Aldrich. Cyclohexene oxide (CHO) was purchased from Acros Organics and dried over CaH~2~ before being fractionally distilled and degassed, through several freeze--pump--thaw cycles, and stored under an inert atmosphere. Bis(triphenylphosphine)iminium chloride (PPNCl) and tetraphenylphosphonium chloride (Ph~4~PCl) were purified by crystallization from chloroform and hexane. Dimethylaminopyridine (DMAP) was crystallized from toluene. The ligands H~3~L^Me^ and H~3~L^tBu^ and the catalysts **1**--**4** were synthesized by following previously reported methods.^[@ref14],[@ref19],[@ref36],[@ref37],[@ref39]^ NMR spectra were recorded on Bruker AV-400 and Bruker AV-500 instruments. All spectra were processed using MestreNova or Topspin software. GPC analysis was carried out on a Shimadzu LC-20AD instrument, equipped with a refractive index (RI) detector and two PSS SDV 5 μm linear M columns. HPLC-grade THF was used as the eluent, at 1.0 mL/min, at 30 °C. Samples were filtered through 0.2 μm PTFE filters prior to analysis. Narrow-*M*~n~ polystyrene standards were used to calibrate the instrument. Elemental analysis was carried out by Mr. Steven Boyer and Mr. Eric Coleman at the London Metropolitan University. Representative CHO/CO~2~ Copolymerization (ROCOP) {#sec4.2} ------------------------------------------------- All polymerizations were carried out in a 25 mL high pressure Parr reactor, fitted with a mechanical stirrer and aluminum block heating control. Catalyst (0.040 mmol), cocatalyst (0.040 mmol), and cyclohexene oxide (8 mL, 80 mmol) were measured inside a glovebox and transferred into the reactor. The reactor was filled with 20 bar of CO~2~ and heated to 80 °C with continuous stirring. The CO~2~ pressure was then increased to 40 bar. The polymerization was stopped after the desired time by cooling the reactor to room temperature using an ice bath before the excess carbon dioxide pressure was released. The crude product was purified by dissolving it in dichloromethane (10 mL) and by precipitation with the addition of methanol (3 × 100 mL) to form a white solid. The isolated polymer was dried under vacuum at 50 °C for 48 h. General Synthesis of Tris-phenolates {#sec4.3} ------------------------------------ A mixture of hexamethylenetetramine (2.58 g, 18.4 mmol, 1 equiv), 2,4-dialkylphenol (82.8 mmol, 4.5 equiv), and *p*-toluenesulfonic acid hydrate (0.07 g, 0.36 mmol, 0.02 equiv) was heated at 110 °C for 20 h. Then, an additional quantity of 2,4-dialkylphenol (27.6 mmol, 1.5 equiv) was added and the solution heated for a further 20 h. The reaction mixture was cooled to 25 °C, and MeOH (20 mL) was added to the yellow slurry. The solution was sonicated until a pale yellow solid evolved from the solution. The solid was filtered, washed with cold MeOH (−78 °C), and subsequently dried under vacuum. The pure ligand was crystallized from acetone. All spectroscopic data were consistent with the data in the literature ([Figures S1--S4](http://pubs.acs.org/doi/suppl/10.1021/acs.organomet.9b00845/suppl_file/om9b00845_si_001.pdf)). General Synthesis of LTi^IV^X {#sec4.4} ----------------------------- Titanium isopropoxide (1.20 mmol, for complexes **1** and **2**) or chloro-tris(iso-propoxide)titanium(IV) (1.20 mmol, for complex **3**) was added to a solution of the proligand (1.20 mmol) in THF (10 mL) and a stirrer for 16 h. All volatiles were removed in vacuo, and the product was washed with hexane (3 × 10 mL), affording a white solid. ### Complex **1**, (L)^tBu^TiO^i^Pr {#sec4.4.1} Yield: 0.54 g, 0.70 mmol, 58%. ^1^H NMR (400 MHz, C~6~D~6~, 298 K): δ (ppm) 7.45 (s, 3H), 6.83 (s, 3H), 5.47 (hept, *J* = 6.05 Hz, 1H), 4.03 (d, *J* = 13.71 Hz, 3H), 2.62 (d, *J* = 13.71 Hz, 3H), 1.71 (d, *J* = 6.05 Hz, 6H), 1.66 (s, 27H), 1.35 (s, 27H). ^13^C{^1^H} NMR (101 MHz, C~6~D~6~, 298 K): δ (ppm) 161.19, 142.62, 135.63, 124.93, 124.53, 123.22, 79.84, 59.34, 35.43, 34.57, 31.99, 30.02, 27.04. Anal. Calcd for C~48~H~73~NO~4~Ti (775.50 g mol^--1^): C, 74.30; H, 9.48; N, 1.81. Found: C, 74.04; H, 8.94; N, 1.88. ### Complex **2**, (L)^Me^TiO^i^Pr {#sec4.4.2} Yield: 0.38 g, 0.73 mmol, 73%. ^1^H NMR (400 MHz, C~6~D~6~, 298 K): δ (ppm) 6.79 (s, 3H), 6.49 (s, 3H), 5.34 (hept, *J* = 6.12 Hz, 1H), 3.96 (d, *J* = 10.60 Hz, 3H), 2.49 (d, *J* = 10.60 Hz, 3H), 2.34 (s, 9H), 2.17 (s, 9H), 1.63 (d, *J* = 6.12 Hz, 6H). ^13^C{^1^H} NMR (126 MHz, C~6~D~6~, 298 K): δ (ppm) 160.39, 131.81, 129.19, 128.36, 124.54, 124.00, 78.09, 59.44, 25.98, 21.29, 16.60. Anal. Calcd for C~30~H~37~NO~4~Ti (523.22 g mol^--1^): C, 68.83; H, 7.12; N, 2.68. Found: C, 68.47; H, 7.23; N, 2.80. ### Complex **3**, (L)^Me^TiCl {#sec4.4.3} Yield: 0.42 g, 0.84 mmol, 70%. ^1^H NMR (500 MHz, CDCl~3~, 298 K): δ (ppm) 6.89 (s, 3H), 6.74 (s, 3H), 3.99 (br s, 3H), 2.84 (br s, 3H), 2.26 (s, 9H), 2.23 (s, 9H). ^13^C {^1^H} NMR (126 MHz, CDCl~3~, 298 K): δ (ppm) 160.36, 131.42, 131.27, 127.36, 124.51, 122.95, 58.62, 20.82, 15.99. Anal. Calcd for C~27~H~30~ClNO~3~Ti (499.14 g mol^--1^): C, 64.88; H, 6.05; N, 2.80. Found: C, 64.76; H, 6.18; N 2.83. ### Complex **4**, \[(L^Me^Ti)~2~O\] {#sec4.4.4} Water (69 μL, 3.8 mmol) was added to a solution of complex **2** (0.2 g, 0.38 mmol) in THF (10 mL) and the mixture stirred for 16 h. All volatiles were removed in vacuo, and the product was washed with hexane (3 × 5 mL), affording a yellow solid. Yield: 0.11 mg, 0.12 mmol, 32%. ^1^H NMR (400 MHz, C~6~D~6~, 298 K): δ (ppm) 6.72 (s, 3H), 6.54 (s, 3H), 4.07 (br s, 3H), 2.51 (br s, 3H), 2.38 (s, 9H), 2.15 (s, 9H). ^13^C{^1^H} NMR (126 MHz, C~6~D~6~, 298 K): δ (ppm) 160.89, 131.04, 129.30, 128.35, 125.04, 124.36, 58.86, 20.80, 16.20. Anal. Calcd for C~54~H~60~N~2~O~7~Ti~2~ (944.34 g mol^--1^): C, 68.65; H, 6.40; N, 2.97. Found: C, 68.70; H, 6.34; N, 3.00. The Supporting Information is available free of charge at [https://pubs.acs.org/doi/10.1021/acs.organomet.9b00845](https://pubs.acs.org/doi/10.1021/acs.organomet.9b00845?goto=supporting-info).Complete experimental and characterization data for all compounds and polymerizations ([PDF](http://pubs.acs.org/doi/suppl/10.1021/acs.organomet.9b00845/suppl_file/om9b00845_si_001.pdf)) Supplementary Material ====================== ###### om9b00845_si_001.pdf CCDC 1970304 contains the supplementary crystallographic data for this paper. These data can be obtained free of charge via [www.ccdc.cam.ac.uk/data_request/cif](http://www.ccdc.cam.ac.uk/data_request/cif), or by emailing <data_request@ccdc.cam.ac.uk>, or by contacting The Cambridge Crystallographic Data Centre, 12 Union Road, Cambridge CB2 1EZ, UK; fax: +44 1223 336033. The EPSRC (EP/K014668/1; EP/L017393/1), EIT Climate KIC (EnCO~2~re), and Econic Technologies (A.D.) are acknowledged for research funding. The authors declare the following competing financial interest(s): C.K.W. is a director of Econic Technologies.

Introduction {#s1} ============ Memory reconsolidation (ReC) is a recently proposed process explaining the update of long-term memories in the brain. Upon activation, the memory trace enters a state of lability rendering it subject to alteration and permitting integration of new information before being restabalized, or reconsolidated. "Reconsolidation" coined by Sara in 2000 [@pone.0068189-Sara1] has become a widely studied topic in neuroscience. Recent animal and human experiments [@pone.0068189-Dudai1]--[@pone.0068189-Medina1] have presented overwhelming evidence supporting the existence of ReC and identified boundary conditions that characterize and limit this phenomenon [@pone.0068189-Antoine1]. ReC is postulated to strengthen, weaken or extinct memories and update them with new, relevant information. Reconsolidation draws a striking new way of understanding memory and its roles: from a computer-like reliable log, to an adaptive and active part of perception. Recent experiments have also identified reconsolidation as a possible avenue of treatment for phobias and PTSD by effectively allowing the erasure of fear memories. These memories come about through classical conditioning mechanisms that pair aversive stimuli (unconditioned stimuli -- US) with co-occurring, once neutral stimuli (conditioned stimuli -- CS). This coupling is the basis for anxiety disorders and PTSD. The most common treatment for fear related disorders is exposure therapy. Exposure therapy leverages extinction learning mechanisms to create a second safety memory that competes with and suppresses the fear response [@pone.0068189-Golkar1], [@pone.0068189-Barlow1]. This technique, however, does not fully erase the fear memory, allowing it to spontaneously reappear [@pone.0068189-Rescorla1]. Reconsolidation has been demonstrated as a possible method of completely erasing fear associations. In several experiments, fear memories in previously conditioned rats were reactivated, returning the memory traces to labile states. Protein synthesis inhibitors or beta-adrenergic receptor antagonists were then injected into the amygdala, blocking the reconsolidation process. This process resulted in extinction of fear and was not subject to spontaneous recovery [@pone.0068189-Dudai2], [@pone.0068189-Nader1], [@pone.0068189-Nader2], [@pone.0068189-Alberini1]. Cases of reconsolidation of fear memories have also been demonstrated in humans. In these experiments, subjects were exposed to stimuli, which reactivated the fear memory trace rendering it labile. Rather than pharmacological intervention, the normal reconsolidation process was disrupted with competing information which resulted in the memory being updated [@pone.0068189-Schiller1], [@pone.0068189-Agren1]. We propose an adaptive memory model that is consistent with recent findings in ReC. The framework introduces efficient ways to add, remove, and update attractors. Additionally, memories can be strengthened, weakened, or extinguished by controlling the attractor radius. Our memory model builds on an earlier Kernel Associative Memory (KAM) model [@pone.0068189-Nowicki1], [@pone.0068189-Nowicki2] that uses a kernel structure to efficiently compute attractor dynamics. The KAM model is an extension of the attractor based Hopfield network. It has been shown that attractor mechanisms are employed by the brain, notably in the CA3 region of the hippocampus [@pone.0068189-Wills1]. The KAM has several advantages over previous Hopfield models including the number of attractors unbounded and independent of the input dimension, dynamic rewiring of neurons, and the ability to accommodate large real-valued inputs and attractors. This paper derives a ReC algorithm that allows KAM to hold an unbounded number of now flexible attractors, which we call ReKAM. Our approach to the modeling of reconsolidation is based on the principle of robust global update, analogous to psychological findings such as the gang effect where the update of one attractor affects neighboring attractors [@pone.0068189-McClelland1]. We also introduce an approximate ReC algorithm which changes the global updates to local ones, gaining time efficiency at the cost of precision. The relevance of our ReKAM model is demonstrated by replicating three recently found characteristics of ReC seen in human behavioral experiments. First, ReKAM imitates a recent list-learning experiment in which human participants merged new objects into a previously learned list during retrieval. ReKAM also demonstrates fear extinction via the controllable attractor radius. The third experiment follows gradually changing objects resulting in an evolved representation. Finally, a continuous time version of ReKAM is introduced which relates the model to neurobiological studies. This version extends the capabilities of the continuous-time Hopfield network [@pone.0068189-Hopfield1] commonly used to model average firing rate dynamics [@pone.0068189-Hopfield2], [@pone.0068189-Maass1] of adaptive persistent activity. Previous Reconsolidation Models {#s1a} ------------------------------- Reconsolidation\'s significance in explaining the dynamic properties of healthy memory has led to several mathematical models proposing to explain the process. The first ReC model [@pone.0068189-Blumenfeld1] extended the Hopfield model to allow attractors to evolve through weight decay and Hamming-distance terms. Our ReKAM also allows attractors to evolve, but since our attractors lie in high dimensional space, the number of memories is unbounded and inputs are realistic, thus modeling reconsolidation in a more relevant and technologically practical way. The second ReC model to be introduced, called Reconsolidation Attractor Network (RAN) [@pone.0068189-Siegelmann1], takes the approach that attractors do not have to lie in input space and hence an unbounded number of memories are possible. The architecture of the RAN is layered. Attractors appear in the upper level separate from the neural flow and input space. Our ReKAM builds on the same concept of attractors not lying in input space, but it also draws from Hopfield-like networks for mathematical completeness of attractor dynamics. The third model presented in [@pone.0068189-Osan1] is designed to reproduce extinction of fear memories. Like the first model, it is also based on the classical Hopfield network. Attractors can be extinct when an additional binary variable which represents the anisomycin (consolidation-inhibiting) drug is set to 0. Our ReKAM is the only memory model demonstrating all known ReC properties as opposed to a particular architecture demonstrating only one facet of the ReC process; it is also the only one that describes reconsolidation of large memories with real world stimuli. Modeling with Kernels {#s1b} --------------------- Our ReKAM model is based on our KAM architecture [@pone.0068189-Nowicki2]. Kernel representations were introduced by Vladimir Vapnik to the field of Machine Learning when he showed how to transfer input data to a high-dimensional data space called -space (phi-space). The data is classified in -space and then projected back to the original space resulting in the most efficient, optimal, non-linear separation. This is achieved by using the kernel property: a scalar kernel function applied to two inputs is equal to their product in the -space: . This kernel property is the basis of Support Vector Machines (SVM), regarded as the most efficient supervised classifiers [@pone.0068189-Vapnik1]. Support Vector Clustering (SVC) was introduced in a joint work by the third author\'s research group and Vapnik. SVC is an unsupervised extension of SVM (for the case when labels are not available) that groups data into clusters through kernel functions that mimic high-dimensional organization and projections [@pone.0068189-BenHur1]. In Kernel Associative Memory, we follow similar mathematics. However, here the -space is not abstract. Instead, it is based on the output of multiple neurons. Mathematically, Mercer kernels are no longer sufficient. We define the strong Mercer kernels that provide the condition needed to load an unbounded number of attractors (See Materials and Methods 4.4). The use of both low-level and high-level spaces is an effective mathematical way to describe both the synaptic changes of neurobiological memories as well as the behavioral effects of cognitive memories. Model for Reconsolidation based on KAM {#s1c} -------------------------------------- The practical advantages of our ReKAM model include an input space that can be composed of continuous valued vectors rather than binary ones, a number of attractors that is independent of the input dimension, and a variable input length where longer and shorter input vectors are learned with no a priori bound. Furthermore, attractors are efficiently loaded, deleted, and updated. We briefly describe the KAM which is the basis of our ReKAM model (a complete description is given in [@pone.0068189-Nowicki2]. Let and be matrices whose columns represent the input and output space of the memories. Memories are defined by the transformation on these columns through the projective operator. We transfer the input to the higher space (as explained in previous sectionf), so that the current transformation is now : . A connection matrix is defined as: Memory loading is defined byand recall of input by the iterations:where the first iteration is initialized with , each iteration ends with applying any sigmoid-like activation (bounded monotonically increasing) function coordinate-wise to : , and the iterations stop when update is under a chosen threshold. The KAM can be depicted as a neural network, as explained in [@pone.0068189-Nowicki2]. Results {#s2} ======= Model for Reconsolidation and Extinction: ReKAM {#s2a} ----------------------------------------------- Unlike the traditional Hopfield networks, where attractors lie in input space, our ReKAM\'s attractors (stemming from the KAM architecture, see last subsection in Previous Work) lie in a high dimensional manifold. While a Hebbian networks\' (e.g., [@pone.0068189-Blumenfeld1]) synaptic matrices compose a linear space, our use of the efficient pseudo inverse learning method gives rise to Riemannian manifolds in the attractor space. An unbounded number of attractors can exist in the higher dimensional space. Between every two points in a Riemannian manifold there exists at least one geodesic that has a minimal length of all curves joining the two points. The geodesic is analogous to the shortest straight line between two points but in a nonlinear space. Updating an attractor toward a new input is calculated along a geodesic between the new input and the given attractor it recalled. Our ReC algorithm with this manifold makes the memory update global, and capable of representing psychological properties such as the gang effect. This global update is more expensive, although more accurate, and we provide another local algorithm which is faster and just a bit less general. Comparisons between the architectures are provided both for time analysis (in this section below) and in the result "Updating Memories Incrementally". ### The global geodesic ReC Algorithm {#s2a1} We propose a memory update algorithm that assumes that every ReC update has a global effect. Mathematically it is based on geodesic computation in the Reimannian manifold representing the memory attractors. The metric structure of this manifold and a comparison with the special case of the Grassmann manifold are derived in Materials and Methods (4.1--4.3). Suppose that we have an initial memory that contains patterns (concepts) . We then obtain by replacing one of the attractor patterns with a new stimulus that recalls it. The distance between **X** and **X** can be interpreted as a measure of the amount of "surprise" that the memory experiences when it meets a new stimuli. To track these changes, we build a geodesic joining **X** and **X** on the manifold and take a new point . Here is a step parameter related to the size of a shift during each update. When , the memory remains at **X**, when  = 1, the memory is changed to **X**. Using the same process, when a stimulus appears we can track the change from **X** to . The process the continues for future stimuli. The algorithm of memory update using geodesics is shown in [Fig. 1](#pone-0068189-g001){ref-type="fig"}. The exact geodesic calculation is described in the Materials and Methods. Its complexity depends on the optimization algorithm used. The dimension of our manifold is . A "typical" gradient calculation would require operations. The gradient-like minimum search calculation has complexity where is the required tolerance [@pone.0068189-Nocedal1]. This leads to complexity [@pone.0068189-Udriste1]. However, with derivation-free optimization techniques which do not require explicit gradient calculations, we can reduce this complexity estimation to . {#pone-0068189-g001} ### The Local Approximate ReC Algorithm {#s2a2} The exact computation of geodesics may be resource consuming especially for high dimensional data. Here we develop a simplified ReC algorithm with local rather than global updates to attractors. In this linear approximation, we simply replace the geodesic with a straight line in the coordinate space. This leads to the Approximate-Update algorithm in [Fig. 2](#pone-0068189-g002){ref-type="fig"}. The approximation algorithm\'s complexity is , equivalent to the derivative-free version of the geodesic algorithm. However, the approximation algorithm is much easier and faster to implement. Because it requires only a few operations per element, the complexity does not depend on the tolerance. {#pone-0068189-g002} While the approximate algorithm of [Fig. 2](#pone-0068189-g002){ref-type="fig"} shows only one update per reconsolidation, we can easily construct a version of this algorithm that updates any desired number of attractors. For this, we repeat step 3 for the most relevant attractors with the largest values of where , and use the value for the -th attractor. This version of the algorithm demonstrates gang effect properties by updating neighboring attractors. The approximation error is bounded by the following theorem: **Theorem 1** *Let* . Denote *as the solution given by the geodesic algorithm and as the solution given by the approximate algorithm. There then exists a constant such that* **Proof:** Let be a metric tensor on dependent on coordinates and . The straight line between and is a geodesic in the flat space with a constant metric form . Since is a twice differentiable manifold, along the geodesic that lies between and . Denote as the distance between the starting point and a given point along the (geodesic) curve. is called the arc length (see remark 1 below, [@pone.0068189-doCarmo1], or other textbook on Riemannian and differential geometry). Because is a secant of the curve in the coordinate space, when , there exists a constant such that for the given arc length *s*, **Remark 1** *Arc length could also be defined as a parameterization of a curve such that* . ### Controllable Attraction Radius {#s2a3} As part of the ReKAM architecture we include a mechanism for altering the size of an attractor\'s basin of attraction. This affects the probability of recalling an attractor. As the attraction radius goes to zero, the attractor will never be recalled. This is analogous to extinction. **Definition 1** *A Kernel* *is called* ***uniform*** *if it depends only on the difference* If the kernel network has a uniform kernel, Then the attraction radius can be controlled. Assign the *scaling factor* to the -th attractor. We can then divide the -th entry of by where is the temporary vector used in the recall algorithm of ReKAM. This causes the attraction basin to be scaled by . Model Verification with Human Experiments {#s2b} ----------------------------------------- We verified our model\'s ability to describe reconsolidation by comparing the dynamics of our model to those observed in humans. The first experiment simulates the effect of reconsolidation on episodic memories. The second demonstrates the model\'s capability to replicate extinction. The third follows memory changes created by the gradual altering of the associated input. ### List Learning {#s2b1} We first replicate a human experiment investigating reconsolidation of episodic memories [@pone.0068189-Hupbach1]. In the original experiment, participants were split into two groups (A and B). On Day 1, both groups learned a list of 20 objects (List 1) that were associated with a blue basket. On Day 2, Both groups learned a second list of 20 items (List 2). Before learning, group A received a reminder of List 1 in the form of the blue basket; group B did not receive any reminder. On Day 3 both groups were tested on their ability to retrieve List 1. Group A made more errors confusing List 2 items into List 1 than Group B did ([Fig. 3](#pone-0068189-g003){ref-type="fig"}). When the experiment was repeated to test recall of List 2, both groups performed equally well. ![Results of the original list learning human experiment [@pone.0068189-Hupbach1].\ Group A received a reminder cue before learning List 2. This resulted in the List 1 memory becoming labile and updated by integrating some of the new items from List 2. Group B did not receive this reminder and these intrusions were not seen. For our analysis, results were normalized for each group by dividing the number of items recalled per list by the total number of items recalled in both lists together.](pone.0068189.g003){#pone-0068189-g003} In our simulation, all objects were shown as images, rescaled to pixels. Note that the ability of the model to handle large colored images is already beyond the standard Hopfield model used in previous work. Images were represented as real-valued vectors with components . We added an indicator variable to each item, where denotes that the object is unrelated to the -th list, and means that the object belongs to this list with 100% certainty. For computational efficiency, we took the variant of the Gaussian kernel:where and are tuned to balance between the data vector and the list indicator components. In our simulation we modeled the two groups. For each group we created 40 initial attractors corresponding to the items in both lists. In group A we gradually shifted the value of of each item towards 1 when this item was recalled with the blue basket reminder in the background to simulate the effects of reconsolidation. In group B, these updates were not performed. For both groups we tested the memory in recall mode inputting 1000 new vectors per list by taking the attractor and adding uncorrelated white noise (intensity equaled 10% of data STD). For all query vectors, we set the as the initial value. Using our model, we found an exact correspondence between our simulation and the human experiment for values of for Group A and for Group B (see [Table 1](#pone-0068189-t001){ref-type="table"}). 10.1371/journal.pone.0068189.t001 ###### Comparing the results of actual human experiment to our model\'s simulation. {#pone-0068189-t001-1} Human Results --------------- -------- -------- List 1 % 60.399 90.180 List 2 % 39.600 9.819 Simulation Results -------------------- -------- -------- *x* ~01~ 0.75 0.25 List 1 % 60.287 90.105 List 2 % 39.713 9.895 By normalizing the data by the total number of items recalled over both lists, our simulation matched exactly with Group A and Group B of the human experiment. We next simulated more values of which could arise for varying levels of reconsolidation due to differing experimental procedures, memory type, etc. ([Fig. 4](#pone-0068189-g004){ref-type="fig"}). {#pone-0068189-g004} ### Extinction {#s2b2} Many recent experiments have demonstrated the effects of fear extinction in both humans and animals -- e.g. [@pone.0068189-Monfils1], [@pone.0068189-Suzuki1], [@pone.0068189-Pedreira1]. Numerical simulations with Hopfield memory and Hebbian-like learning were presented in [@pone.0068189-Osan1]. Our model has a far larger number of far more detailed memories than previously modeled. We propose to model extinction as a reduction in the attractor\'s radius. To demonstrate, we created a kernel network that memorized 10 images. All images were scaled to pixels. One of the images was randomly chosen to be a "fear" (shock) memory. In our procedure, the scaling factor for the "shock" attractor was gradually decreased. This process is analogous to the weakening of the memory occurring through reconsolidation during extinction training. For each scaling factor value we measured the frequency of retrieval (recall) of the shock memory on 1000 random inputs ([Fig. 5](#pone-0068189-g005){ref-type="fig"}). The decreasing attraction basin radius effectively extinguishes the fear memory trace as its probability of recall goes virtually to 0. {#pone-0068189-g005} ### Updating Memories Incrementally {#s2b3} In an experiment testing the incremental changes of gradually morphing memories [@pone.0068189-Preminger1], participants learned to recognize four faces as "friends." One face was morphed incrementally over a period of days. When the face morphed slowly, participants continually recognized the morphed face as their original friend. By the end of the process, the morphed face was recognized as a friend while the original face was not. The results demonstrated merging of the source and the new face. However, this effect was only observed when the faces were changed gradually, demonstrating that the order in which morphing took place was crucial. A gradual, subtle change was needed to allow for reconsolidation to occur. In our previous work [@pone.0068189-Nowicki2] we published a numerical experiment with morphing face images that replicated the previous result described above. Attractors in the KAM were gradually morphed following the slowly changing face inputs. Here we present a similar experiment aimed at examining the network\'s ability to track images varying gradually over time. Additionally, we compare the performance of the exact and approximate ReC algorithms for this manipulation. We created a training set consisting of 9,000 rotated digits. The rotated digits were created from 100 original MNIST handwritten digits (10 per class from '0' to '9'). Digits were pixel grayscale images which we rotated counterclockwise from to ([Fig. 6](#pone-0068189-g006){ref-type="fig"}). {#pone-0068189-g006} We applied principal-component (PC) preprocessing without considering any specific handwritten digit optimized feature extraction techniques. We took the first PCs which contain 96.77% of the variance. For computational efficiency, the kernel we chose was:where , and is a bias parameter. This is a Gaussian kernel dependent on a weighting metric. The weights were chosen as:We also tried:Where and are the expectation and standard deviation over the -th class, and is the number of classes. Formula (7) yielded better results. Evolution of the classification rate over time for the digit rotation experiment is shown in [Fig. 7](#pone-0068189-g007){ref-type="fig"} with confusion matrices in [Fig. 7](#pone-0068189-g007){ref-type="fig"}. The exact reconsolidation algorithm achieved a recognition accuracy of 96.4+/−0.43%. Results for the local approximate algorithm were 96.32+/−0.26%. The algorithm without reconsolidation performed significantly worse (see [Fig. 7](#pone-0068189-g007){ref-type="fig"}). The CPU time was 142 sec for the approximate algorithm and 54.7 min for the exact geodesic reconsolidation on Intel Centrino Duo 1.4 GHz CPU with no parallelism, in the Matlab environment. The average relative error in attractors was . {#pone-0068189-g007} When inputs were shuffled randomly, gradual reconsolidation was unable to occur. We note that because we are testing on a handwritten digit dataset, there are variations between each test digit: while an ideal number 6 rotated would be equal to an ideal number 9, this will very rarely be the case with the variable hand written digits. Results of no reconsolidation for digits rotated at ([Fig. 8](#pone-0068189-g008){ref-type="fig"}--C) shows that digits such as 0, 1, and 8 remain mislabeled while, to a human eye, these would seem the same. It is possible that using a preprocessing technique specifically designed for hand written digit recognition may allow the system to generalize these specific cases to a greater degree. However, even under these conditions, the reconsolidation algorithm works effectively and allows for accurate classification under constantly changing inputs. {#pone-0068189-g008} Continuous-time ReKAM Models Firing-Rate Dynamics {#s2c} ------------------------------------------------- Up to here, we described the discrete-time form of associative recall. We next relate the ReKAM to biology by introducing a continuous-time version of the kernel memory and comparing it to other firing-rate models. It is important to note that the nature of the time (discrete or continuous) is involved only in dynamical systems of recall, not in the reconsolidation phase. Any step of the reconsolidation (both exact and approximate) depends only on the input and the attractors, not on the time. For this reason both the exact and approximate algorithms of ReC work in continuous time. The Hopfield equation for the -th neuron is:where is the output of the -th neuron, are the elements of the symmetric synaptic matrix , are direct external inputs, is the activation function, and is the "relaxation rate" of the -th neuron. The Hopfield [equation (9](#pone.0068189.e127){ref-type="disp-formula"}) imposes linear and symmetric neuron-to-neuron interactions in the network which can be described by the synaptic matrix . Escalating the model from discrete neurons to neural field (mean field) gives rise to the Wilson-Cowan partial integro-differential equation [@pone.0068189-Wilson1]: If the activation function is simply the Heaviside step function, [equation (10](#pone.0068189.e137){ref-type="disp-formula"}) becomes the Amari field equation [@pone.0068189-Amari1]. If the network\'s activity is a Markov stochastic process (with a vector ), then the first-order approximation of the average firing rate dynamics is (see [@pone.0068189-Buice1]): This equation can have arbitrary dynamics in contrast to (9) that has a Lyapunov function and converges to attractors. We propose a continuous-time version of the Kernel Associative Memory that updates the recall similar to (11):where the components of are: The continuous ReKAM described by [Equations (12](#pone.0068189.e140){ref-type="disp-formula"}) -- (13) with a scalar-product kernel is isomorphic to (9) except for having the synaptic matrix W calculated by the pseudoinverse rule (not the Hebbian rule), or, equivalently, orthogonal Hopfield learning. This continuous memory inherits the Hopfield-like attractor dynamics but is more biologically relevant: the number of attractors is independent of the input dimension and rewiring of the neurons is dynamic. We propose the continuous ReKAM as a model for firing rate adaptive dynamics in the course of persistent activity in various networks in the brain. Discussion {#s3} ========== While the existence of reconsolidation in human memory was once a topic of debate, the accumulation of human experimental results has led to the mechanism becoming widely accepted in the field of neuroscience. Reconsolidation has been dissociated from extinction learning, the latter of which results in a second memory trace rather than the removal of the old one [@pone.0068189-Besnard1], [@pone.0068189-Besnard2]. However, it is not yet entirely clear when or to what extent reconsolidation mechanisms will occur in a given situation. Experimental results have identified numerous boundary conditions involved in determining whether or not a memory will undergo reconsolidation [@pone.0068189-Besnard1], [@pone.0068189-Besnard2]. One such boundary condition is the amount of time between a memory\'s retrieval and the encountering of relevant stimuli. This time window varies depending on the animal tested [@pone.0068189-Tronson1] and in humans begins about 10 minutes after retrieval and lasts for several hours. During this time, the memory is labile and susceptible to new information or experimental interference. If the stimulus is encountered outside of this time window, reconsolidation will not occur. A second boundary condition is the age and strength of the memory trace, affecting the ease in which the memory will undergo reconsolidation. A stronger or older memory may require longer and more frequent reactivation sessions for reconsolidation to occur. A third condition, the predictability of reactivation stimulus, also plays a role in whether or not reconsolidation will occur. If a subject does not correctly predict a novel response to a stimulus, reconsolidation is more likely to occur in order to update an incorrect prediction model [@pone.0068189-Sevenster1]. Another boundary condition is the "trace dominance" -- when a memory stabilizes and becomes resistant to reconsolidation and certain amnesic agents. It would be possible to extend our model in the future to include these observed boundary conditions. The addition of variables that account for time elapsed since retrieval, age of memory, and strength of memory could be implemented to allow for an accurate simulation of the boundary conditions that accompany reconsolidation. Additionally, a mechanism to account for prediction error would allow for a representation of the novelty prediction that has been shown to influence whether or not reconsolidation will occur. These additions could allow for a more accurate simulation of reconsolidation as well as a more biological learning model. We have proposed a mathematical framework of memory reconsolidation, which demonstrates properties as seen in human studies: incremental updates, associations, and extinction. Our ReKAM memory model is far more technologically relevant than previous ones in that it is able to include real-valued inputs as well as massively long inputs; the number of memories is independent of input dimension and hence is practically unbounded. This results in a model providing both a better functional understanding of reconsolidation and the basis for a powerful technology for following changes in real world environments. The mathematical structure has its own beauty: The kernel associative memory has an underlying structure of a Grassman-like manifold in the (feature) -space. Since it is a curved Riemannian manifold, reconsolidation is no longer a linear update, but the creation of geodesics is required. We provided both an exact Reconsolidation algorithm as well as a more efficient one, which is local in update and does not require the exact computation of geodesics. A continuous time version of the memory is introduced with further biological relevance. The kernel method opens the door to reconsolidation of multimodal and dynamical (temporal) memories; this is a subject of our future research. Materials and Methods {#s4} ===================== Defining a Riemannian Structure for the ReC algorithm {#s4a} ----------------------------------------------------- We formulate the distance between two kernel associative networks where both networks have the same kernel and number of memories. Each network contains a different sets of memory attractors. In the -space each kernel memory is a symmetric network whose synaptic matrix is a projective operator . We measure the distance between two projective operators, **X** and **Y** (both of finite rank ), as a Frobenius norm . Taking into account protectiveness and self-conjugatedness of **X** and **Y**, we have: For each projective operator , the singular value decomposition (SVD) leads to the following: For any defined as above, a matrix () is defined as having elements , the pairwise scalar products of the memorized vectors. In matrix notation this is represented as:Since is an orthogonal operator, then Using this template we represent and as follows: , ; , and . So,Here **Q** is an ***mm*** matrix such that . Having a singular decomposition for **XY**, we can now compute the distance as The above defines a Riemannian structure for the KAM manifold. Pseudoinverse Memories and the Grassmann Manifold {#s4b} ------------------------------------------------- We next relate the manifold defined by the ReKAM model to the more well-known and less complex Grassmann manifold. An associative memory with a pseudoinverse learning rule is described in [@pone.0068189-Personnaz1]. This is a Hopfield-type auto-associative memory defined originally for bipolar vectors: . Suppose these vectors are columns of matrix . Then a synaptic matrix of the memory is given by:where is a Moore-Penrose pseudoinverse of . For linearly independent columns of , the pseudoinverse can be computed by or by using the Greville formulae (see, e.g., [@pone.0068189-Albert1]). The resulting weight matrix is projective, i.e. with rank . The Grassmann manifold is a particular type of Riemannian. The Real Grassman manifold is the manifold of all -dimensional subspaces in and is denoted as . To define the Grassman manifold, we first introduce the Stifiel manifold -- a set of orthogonal --matrices , endowed with the Riemannian metric which is induced by the Euclidean norm in the space of -matrices. Next, we say that two matrices are equivalent if their columns span the same -dimensional subspace. This means that two matrices and are equivalent if they are related by right multiplication of an orthogonal matrix . The quotient of the Stiefel manifold with respect to this equivalence relation is called *Grassmann Manifold* [@pone.0068189-Absil1]. For each -dimensional subspace in there exists a unique projective matrix of rank , and vice versa (see [@pone.0068189-Golub1]). Therefore a space of -ranked projective matrices is a Grassmann manifold . Moreover, the Frobenius norm of the difference of two projective matrices gives one possible Riemannian metric over this manifold. The Grassman manifold emerges in our model in the special case of a scalar-product kernel. Other kernels used in our ReKAM model result in manifolds that can be considered generalizations of the Grassmann. Computing Geodesics for the ReC Algorithm {#s4c} ----------------------------------------- To implement the geodesic update algorithm, we have to efficiently compute geodesics on the kernel memory manifolds. Given the metric in explicit form (15) this can be solved as an optimization problem. Let and be points on manifold with metric . Let a point lie on the (minimizing) geodesic segment joining and . divides the segment into two parts with proportions . Let be a point which lies on the manifold but not in the geodesics. The process of finding is stated as follows: The geodesic minimizes the sum of two distances (first line). For the point on the (minimizing) geodesic, the following inequality holds: The Process (17) can be solved numerically using a Gradient Descent Method (or other first-order unconstrained optimization method). Its complexity is for a tolerance . The constant here is typically large due to the hardness of gradient computation. Functions with Mercer Condition {#s4d} ------------------------------- The classical Kernels introduced to the field of Machine Learning by Vapnik [@pone.0068189-Vapnik1] had the Mercer condition. That is, for all square integrable functions the kernel satisfied: The Mercer theorem states that if satisfies the Mercer condition there exists a Hilbert space with a basis and a function , where , such thatand all 0. That is, is a scalar product of and General Mercer kernels are not sufficient for creating the associative memory since our kernel memories require that all attractors are linearly independent in the feature space. Some Mercer kernels, such as the basic scalar-product kernel , do not assure this property. The strong Mercer kernels defined for our kernel memory [@pone.0068189-Nowicki2] provide linear independence of the attractors in the feature space which enables correct association. We express our appreciation to Keen Sung for finding and reviewing the relevant literature as well as for helpful discussions and editing advise. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: DN HS PV. Performed the experiments: DN. Analyzed the data: DN PV. Contributed reagents/materials/analysis tools: DN PV HS. Wrote the paper: DN PV HS. Proved the theorems and derived the algorithms: DN.

Notice of Republication {#s1} ======================= This article was republished on July 1st, 2014, due to panel B of Figure 3 being erroneously removed during the production process. The publisher apologizes for this error. Please download the PDF again to view the corrected article and the complete Figure 3. The originally published, uncorrected article and the republished, corrected article are provided here for reference. Supporting Information {#s2} ====================== ###### Originally published, uncorrected article (PDF) ###### Click here for additional data file. ###### Republished, corrected article (PDF) ###### Click here for additional data file.

INTRODUCTION {#s1} ============ In 2012, hepatocellular carcinoma (HCC) was the third leading cause of cancer mortality worldwide. The incidence of HCC was ranked as the seventh most common cancer worldwide, and its incidence rate was approximately 11.1 per 100,000 persons in 2012 \[[@R1]\]. International Agency for Research on Cancer reported that five-year prevalence is 12.2 per 100,000 persons (633,170 cases) \[[@R1]\]. In Taiwan, HCC is the second leading cause of cancer death \[[@R2]\]. According to Ministry of Health and Welfare in Taiwan, age-standardized mortality of HCC was 22.8 per 100,000 persons in 2015 \[[@R2]\]. Moreover, age-standardized incidence of HCC was 35.0 per 100,000 persons in 2012, whereas mortality and incidence are approximately threefold higher than global occurrences \[[@R1], [@R3]\]. Type 2 diabetes has rapidly become prevalent globally. World Health Organization reports that worldwide occurrence of diabetes is estimated at about 9% in 2014 \[[@R4]\]. Previous cohort studies suggest that type 2 diabetes was associated with an increased risk of HCC \[[@R5]\]. Findings of a systematic review and meta-analysis of 18 cohort studies reveal that patients with diabetes are associated with two times higher risk of developing HCC than individuals without diabetes \[[@R6]\]. Thus, we need further to explore what factors are associated with HCC risks in patients with diabetes by controlling for known diabetes-related risk factors such as hyperglycemia and metformin use in this line of research. Serum alanine transaminase (ALT) is a common biomarker of hepatocellular injury. Elevated ALT is related to several liver diseases such as alcoholic liver disease, nonalcoholic fatty liver disease (NAFLD), hepatitis B/C virus infection, drug-induced hepatotoxicity, and autoimmune and cholestatic liver diseases \[[@R7]\]. Several prior cohort studies report elevated ALT is associated with HCC in patients infected with hepatitis B/C virus \[[@R8], [@R9]\], and a few studies show ALT levels are one of important predictors in predictive models for discriminating HCC patients with hepatitis B/C virus \[[@R10]\]. However, most studies evaluating the relationship between ALT and HCC have focused on patients with B/C virus infection. A recent study examined the relationship between ALT and cancer-specific mortality in diabetic patients but found no such an association existed \[[@R11]\]. Patients with diabetes were been reported to have a higher incidence of liver diseases, including NAFLD, cirrhosis, and acute liver failure \[[@R12]\]. Some studies indicated elevated ALT is more prevalent among diabetic persons \[[@R13]\]. In a community-based study consisting of 11,898 residents, incidence of elevated ALT was 12.5% in type 2 diabetic patients, compared to 7.7% in patients without diabetes \[[@R13]\]. Thus, increased HCC risk in type 2 diabetic patients may be associated with elevated ALT and liver diseases. We conducted a nationwide cohort study of Chinese type 2 diabetic patients with an average follow-up period of 8.17 years to assess whether an elevated ALT level was associated with HCC independently of chronic liver diseases, and to demonstrate its interactive or joint effect with chronic liver diseases on HCC risk in type 2 diabetic patients. Chronic liver diseases considered in this present study were NAFLD, acute liver damage, liver cirrhosis, and hepatitis B and C virus infections. RESULTS {#s2} ======= Incidence rate of HCC was 2.68 per 1,000 person-years (men: 3.45 and women: 1.98 per 1,000 person-years). Mean age was 60.60 years with a standard deviation (SD) of 11.25 years and mean follow-up period was 8.20 years (SD: 1.86 years). The prevalence of ALT level ≥40 U/L was 22.14% in men and 17.93% in women. Baseline characteristics according to clinical criteria cut-off points of ALT were presented in Table [1](#T1){ref-type="table"}. Figure [1](#F1){ref-type="fig"} shows Kaplan--Meier cumulative risks of HCC according to subgroups defined by ALT level. Persons with ALT \> 120 U/L faced the highest risk, followed by persons with ALT 80--120 and 40--80 U/L, and then persons with ALT≤ 40 U/L had the lowest risk (log-rank test *P* \<0.001, Figure [1](#F1){ref-type="fig"}). Table [2](#T2){ref-type="table"} presents HRs of HCC among patients grouped by ALT levels. Adjusted HRs of HCC in patients with ALT levels of 40--80, 80--120, and \> 120 U/L were 3.85 (95% CI 3.34--4.45), 7.05 (5.70--8.71), and 10.85 (8.81--13.36), respectively, after multivariate adjustment (*P* for trend: \<0.001), compared with persons exhibiting ALT levels ≤ 40 U/L. When additionally considering BMI and lipid profiles, these three HRs remained similar. When liver diseases were additionally adjusted, the effects of ALT diminished, but remained significant \[2.85 (95% CI 2.45--3.31) for 40--80 U/L, 3.80 (3.04--4.76) for 80--120 U/L, and 3.89 (3.08--4.91) for \> 120 U/L\]. ###### The comparisons of sociodemographic factors, lifestyle behaviors, diabetes-related variables, drug-related variables, diabetes-related diseases and comorbidity according to ALT level with type 2 diabetes enrolled in the Diabetes Care Management Program, Taiwan (n=46, 369) Variables ALT (IU/L) P value ------------------------------------------------------------------ --------------- --------------- --------------- --------------- --------- *Sociodemographic factors*  Male, n (%)^a^ 17314 (46.64) 3662 (52.53) 738 (54.11) 522 (57.11) \<0.001  Age (years), mean (SD)^b^ 61.31 (11.16) 58.11 (11.06) 56.72 (11.45) 56.59 (11.11) \<0.001 *Lifestyle behaviors*, n (%)^a^  Smoking 5605 (15.10) 1284 (18.42) 260 (19.06) 215 (23.52) \<0.001  Alcohol drinking 3107 (8.37) 714 (10.24) 154 (11.29) 105 (11.49) \<0.001 *Diabetes-related variables* Duration of diabetes (years), mean (SD)^b^ 7.21 (8.09) 5.17 (6.71) 4.69 (7.07) 4.57 (6.11) \<0.001 Type of hypoglycemic drug use, n (%)^a^ \<0.001  No medication 880 (2.37) 200 (2.87) 29 (2.13) 22 (2.41)  Metformin only or metformin plus other oral antidiabetic agents 24412 (65.77) 4710 (67.57) 928 (68.04) 558 (61.05)  Other oral antidiabetic agents 6211 (16.73) 1175 (16.86) 206 (15.10) 159 (17.40)  Insulin 1069 (2.88) 119 (1.71) 23 (1.69) 27 (2.95)  Insulin+ oral hypoglycemic drug 4548 (12.25) 767 (11.00) 178 (13.05) 148 (16.19) *Drug-related variables*, n (%)^a^  Hypertension drug treatment 14007 (37.73) 2637 (37.83) 476 (34.90) 281 (30.74) \<0.001  Statin 10284 (27.7) 1852 (26.57) 292 (21.41) 143 (15.65) \<0.001 BMI^a^ \<0.001  \<18.5 551 (1.48) 56 (0.80) 13 (0.95) 25 (2.74)  18.5-23.9 11246 (30.3) 1250 (17.93) 255 (18.70) 234 (25.60)  24-26.9 13143 (35.41) 2256 (32.36) 411 (30.13) 309 (33.81)  ≥27 12180 (32.81) 3409 (48.90) 685 (50.22) 346 (37.86) Blood biochemical indexes^a^ TG (mg/dL) \<0.001  \<150 20925 (56.37) 3487 (50.02) 763 (55.94) 558 (61.05)  ≥150 16195 (43.63) 3484 (49.98) 601 (44.06) 356 (38.95) FPG (mg/dL) \<0.001  \<110 4015 (10.82) 603 (8.65) 129 (9.46) 87 (9.52)  ≥110 33105 (89.18) 6368 (91.35) 1235 (90.54) 827 (90.48) HbA1c (%) \<0.001  \<7 10911 (29.39) 1700 (24.39) 281 (20.60) 212 (23.19)  ≥7 26209 (70.61) 5271 (75.61) 1083 (79.40) 702 (76.81) HDL (mg/dL) \<0.001  ≥40(male); 50(female) 17880 (48.17) 3079 (44.17) 637 (46.70) 436 (47.70)  \<40(male); 50(female) 19240 (51.83) 3892 (55.83) 727 (53.30) 478 (52.30) LDL (mg/dL) \<0.001  \<100 10636 (28.65) 2054 (29.46) 474 (34.75) 389 (42.56)  ≥100 26484 (71.35) 4917 (70.54) 890 (65.25) 525 (57.44) Comorbidity^a^  Acute hepatitis 49 (0.13) 11 (0.16) 5 (0.37) 7 (0.77) \<0.001  Alcoholic liver damage 84 (0.23) 36 (0.52) 5 (0.37) 9 (0.98) \<0.001  Nonalcoholic fatty liver 469 (1.26) 167 (2.40) 41 (3.01) 28 (3.06) \<0.001  Liver cirrhosis 181 (0.49) 104 (1.49) 35 (2.57) 30 (3.28) \<0.001  Cholelithiasis 524 (1.41) 95 (1.36) 23 (1.69) 9 (0.98) 0.56  Alcohol dependence syndrome 47 (0.13) 14 (0.20) 0 (0.00) 3 (0.33) 0.08  Jaundice 14 (0.04) 7 (0.10) 2 (0.15) 3 (0.33) \<0.001  Hepatitis B 213 (0.57) 109 (1.56) 29 (2.13) 21 (2.30) \<0.001  Hepatitis C 70 (0.19) 60 (0.86) 26 (1.91) 40 (4.38) \<0.001  Cholecystitis 76 (0.20) 9 (0.13) 1 (0.07) 0 (0.00) 0.21  Cholangitis 68 (0.18) 10 (0.14) 4 (0.29) 1 (0.11) 0.62  Gastric ulcer 1053 (2.84) 182 (2.61) 37 (2.71) 21 (2.30) 0.58  Duodenal ulcer 718 (1.93) 116 (1.66) 23 (1.69) 25 (2.74) 0.11  Chronic kidney disease 10260 (27.64) 1407 (20.18) 226 (16.57) 155 (16.96) \<0.001 Number of diagnostic testing^b^ 0.28 (0.82) 0.29 (0.75) 0.34 (0.90) 0.47 (1.09) \<0.001 a: Differences in categorical variables were tested by the Chi-square test. b: Differences in continuous variables were tested by the analysis of variance. ALT: alanine aminotransferase; HCC: hepatocellular carcinoma; LC: liver cirrhosis; ALD: alcoholic liver disease; AFLD: alcoholic fatty liver disease; NAFLD: nonalcoholic fatty liver disease; HBV: hepatitis B virus; HCV: hepatitis C virus. {#F1} ###### Hazard ratios (HRs) of hepatocellular carcinoma according to clinical criteria of baseline ALT level in type 2 diabetic patients enrolled in the NDCMP Variables n Cases Person-years IR Hepatocellular carcinoma (N=1, 018) -------------- ------- ------- -------------- ------- ------------------------------------- ----------------------------- --------------------------- ALT (IU/L)  ≤40 37120 486 304445.57 1.60 1.00 1.00 1.00  40-80 6971 312 57687.96 5.41 3.85 (3.34, 4.45)^\*\*\*^ 4.00 (3.46, 4.62)^\*\*\*^ 2.85 (2.45, 3.31)^\*\*\*^  80-120 1364 107 11025.52 9.70 7.05 (5.70, 8.71)^\*\*\*^ 6.99 (5.65, 8.65)^\*\*\*^ 3.80 (3.04, 4.76)^\*\*\*^  \>120 914 113 7140.59 15.83 10.85 (8.81, 13.36)^\*\*\*^ 10.14 (8.22, 12.50)^\*\*\*^ 3.89 (3.08, 4.91)^\*\*\*^ P for trend∞ \<0.001 \<0.001 \<0.001 Model 1 adjusted for age, sex, duration of type 2 diabetes, smoking, drinking, type of anti-diabetic medication use, anti-hypertension drug treatment and statin use. Model 2 additionally adjusted for body mass index, triglyceride, fasting plasma glucose, HbA1c, high-density lipoprotein and low-density lipoprotein. Model 3 additionally adjusted for alcoholic liver damage, nonalcoholic fatty liver, liver cirrhosis, cholelithiasis, alcohol dependence syndrome, jaundice, hepatitis B, hepatitis C, cholecystitis, cholangitis, gastric ulcer, duodenal ulcer, chronic kidney disease and number of image tests. ALT: alanine aminotransferase; ^\*\*\*^: p\<0.001 for Wald's test in Cox's proportional hazards model. ∞: p for trend was from multivariate Cox's proportional hazards model by treating categorical ALT as an ordinal variable. Table [3](#T3){ref-type="table"} presents sensitivity analyses by ruling out persons with histories of a stroke, hypoglycemia, coronary artery disease, and HBV, and HCV infection. The association between ALT and HCC remained similar despite the exclusion of persons with the mentioned conditions separately and together. Exclusion of all conditions together (n=39,599) resulted in similar significant HRs for HCC among patients with ALT levels 40--80, 80--120, and \> 120 U/L (2.87 \[2.43--3.39\], 3.92 \[3.07-5.02\], and 4.53 \[3.53--5.81\], respectively; *P* for trend\<0.001). ###### Sensitivity analyses for the association between ALT level and hepatocellular carcinoma in type 2 diabetic patients enrolled in the National Diabetes Care Management Program, Taiwan Variables n Cases Person-years IR Hepatocellular carcinoma --------------- ------- ------- -------------- ------- --------------------------- ***Model 1*** ALT (IU/L)  ≤40 35018 455 289088.10 1.57 1.00  40-80 6702 300 55669.49 5.39 2.88 (2.47, 3.36)^\*\*\*^  80-120 1310 104 10633.63 9.78 3.96 (3.16, 4.97)^\*\*\*^  \>120 883 112 6917.34 16.19 4.09 (3.23, 5.18)^\*\*\*^ P for trend∞ \<0.001 ***Model 2*** ALT (IU/L)  ≤40 36956 483 303369.80 1.59 1.00  40-80 6962 311 57631.45 5.40 2.85 (2.45, 3.31)^\*\*\*^  80-120 1364 107 11025.52 9.70 3.80 (3.04, 4.76)^\*\*\*^  \>120 913 113 7131.35 15.85 3.89 (3.08, 4.92)^\*\*\*^ P for trend∞ \<0.001 ***Model 3*** ALT (IU/L)  ≤40 33742 444 278020.00 1.60 1.00  40-80 6423 289 53241.08 5.43 2.84 (2.43, 3.33)^\*\*\*^  80-120 1275 98 10356.68 9.46 3.79 (3.00, 4.78)^\*\*\*^  \>120 865 112 6752.43 16.59 4.08 (3.22, 5.18)^\*\*\*^ P for trend∞ \<0.001 ***Model 4*** ALT (IU/L)  ≤40 36844 469 302318.20 1.55 1.00  40-80 6803 277 56458.66 4.91 2.83 (2.42, 3.31)^\*\*\*^  80-120 1310 96 10642.77 9.02 3.72 (2.94, 4.70)^\*\*\*^  \>120 855 102 6714.97 15.19 4.18 (3.28, 5.32)^\*\*\*^ P for trend∞ \<0.001 ***Model 5*** ALT (IU/L)  ≤40 31602 402 262222.60 1.53 1.00  40-80 6034 247 50341.02 4.91 2.87 (2.43, 3.39)^\*\*\*^  80-120 1180 87 9660.26 9.01 3.92 (3.07, 5.02)^\*\*\*^  \>120 783 100 6165.69 16.22 4.53 (3.53, 5.81)^\*\*\*^ P for trend∞ \<0.001 Model 1 excludes patients with stroke (n=2,456); Model 2 excludes patients with hypoglycemia (n=174); Model 3 excludes patients with coronary artery disease (n=4,064); Model 4 excludes patients with hepatitis B or hepatitis C (n=557); Model 5 excludes patients with stroke, hypoglycemia, coronary artery disease, hepatitis B or hepatitis C (n=6,770). Adjustment for age, sex, duration of T2DM, smoking, drinking, type of diabetes treatment, anti-hypertension drug treatment, statin use, body mass index, triglyceride, fasting plasma glucose, HbA1c, high-density lipoprotein, low-density lipoprotein, alcoholic liver damage, nonalcoholic fatty liver, liver cirrhosis, cholelithiasis, alcohol dependence syndrome, jaundice, hepatitis B (except model 4 and 5), hepatitis C (except model 4 and 5), cholecystitis, cholangitis, gastric ulcer, duodenal ulcer chronic kidney disease and number of image tests. ALT: alanine aminotransferase; ^\*\*\*^: p\<0.001 for Wald's test in Cox's proportional hazards model. ∞: p for trend was from multivariate Cox's proportional hazards model by treating categorical ALT as an ordinal variable. Figure [2](#F2){ref-type="fig"} demonstrates adjusted HRs of HCC for joint effects of ALT \> 40 U/L and ALD, NAFLD, liver cirrhosis, HBV and HCV infection, and any one of these CLD for entire sample, and stratified by insulin use. We observed more significant HRs of HCC for patients with a level of ALT \> 40 U/L with NAFLD, liver cirrhosis, HBV and HCV infection, and any one of these CLDs than those of patients with a level of ALT ≤ 40 U/L and no counterpart comorbidity (4.89, 3.02--7.92; 18.53, 13.41--25.60; 8.47 5.39--13.30; 11.01, 7.51-16.13, and 14.36, 11.51--17.91, respectively). Significant interactions were observed between ALT level with liver cirrhosis, and HBV (p for interaction terms =0.01 and 0.002). Stratified analysis was performed according to insulin use. Similar significant joint effects of ALT \> 40 U/L with NAFLD, liver cirrhosis, HBV and HCV infection, and any one of these CLDs were observed in insulin, and non-insulin users. Main effects of ALT \> 40 U/L were all significant across subgroups of various chronic liver diseases with a narrow 95% CIs, and remained similar in the entire sample, non-insulin users, and insulin users. In general, PERI, AP, and S-index indicated that interaction of ALT \> 40 U/L with liver cirrhosis was positive, but interaction of ALT \> 40 U/L with HBV was negative. {#F2} DISCUSSION {#s3} ========== We investigated the relationship between ALT level and HCC risk among Chinese type 2 diabetic patients in Taiwan NDCMP. This nationwide cohort study had 46,396 type 2 diabetic patients, aged 30 years and over at baseline. This study demonstrated an independent association between ALT level and HCC risk. Findings of sensitivity analysis illustrated this independent association between ALT level and HCC risk was similar after ruling out persons with stroke, hypoglycemia, other cancers, and HBV and HCV infection at baseline. Our consistent findings of sensitivity analysis provided evidence to support robustness of our study results. We identified significant effect modifications of ALT level with liver cirrhosis and HBV. Joint effects of ALT level and chronic conditions were demonstrated by significant combined associations of ALT \> 40 U/L and NAFLD, liver cirrhosis, HBV and HCV infection, or any one of CLDs with HCC risk. ALT is a transaminase enzyme, one of intracellular hepatic enzymes leaking into circulation, and served as part of a diagnostic evaluation of hepatocellular injury. Although an elevated ALT level was uncommon in general population, obese individuals and type 2 diabetic patients were reported to have elevated ALT levels, which can be accounted for by increased body mass index \[[@R14]\]. We observed synergistic effects of ALT and liver cirrhosis on HCC risk. The possible biological mechanisms may explain this effect modification included insulin resistance and inflammation, which were two possible biological mechanisms for diabetes as a risk factor of cancer. The most common cause of a mild elevation of ALT in type 2 diabetic patients was NAFLD \[[@R15]\]. The incidence of NAFLD in diabetes was high and was 100% in patients with obesity. NAFLD was the hepatic manifestation of insulin resistance syndrome \[[@R16]\] with a spectrum of liver disease from fatty infiltration of liver to nonalcoholic steatohepatitis (NASH), consisting of steatosis with inflammation, necrosis, and fibrosis, which further lead to cirrhosis. Insulin resistance was central abnormality in pathogenesis of steatosis. Prior studies reported the association between ALT and HCC in patients with a high risk of HCC such as patients with cirrhosis and hepatitis B/C virus infection \[[@R8]--[@R10]\]. Our study reported an elevated ALT level increased HCC risk in Chinese type 2 diabetic patients, indicating significant interactions for ALT ≥ 40 IU/L with liver cirrhosis and HBV on HCC risk as well as joint associations of ALT ≥ 40 U/L and chronic liver diseases. In a recent study conducted by Williams et al. \[[@R11]\], they have failed to find an association between ALT levels and cancer mortality in patients with type 2 diabetes. The possible reasons that can explain the different findings are the smaller sample size in Williams's study (n=9,795 vs. n=46,369) and the shorter follow-up period (5 years vs. 8 years). Our findings regarding joint effect allow us to distinguish the effects of elevated ALT alone, and cumulative effects of elevated ALT and chronic liver diseases such as hepatitis B/C virus infection, cirrhosis, and NAFLD. The consistent findings that elevated ALT alone may imply that evaluated ALT independently leading to HCC, irrespective of the other known pro-oncogenic effects of hepatitis B/C virus infection, and cirrhosis. The clinical implication of our study findings is that ALT management has to take chronic liver diseases into account. Although elevated ALT is not a life-threatening condition, high ALT does warrant prompt medical attention. Elevated ALT indicates damage to liver caused by life-threatening diseases or infections and mainly reflect underlying liver injury, mainly from NAFLD, HBV, HCV chronic hepatitis and cirrhosis. If high ALT with known cause is identified, the treatments for both chronic liver diseases and high ALT should be applied. Chronic liver disease medications included ursodeoxycholic acid for slowing the progression of primary biliary cirrhosis, antiviral medications for reduction of liver enzymes elevated by hepatitis C, etc. If high ALT without known cause was identified, then medication and diet modification should be considered including corticosteroids and pentoxifylline for liver inflammation reduction, diuretics for removing excess fluid from the body, avoiding intake of salt and alcohol-drinking, and a low-protein diet for reduction of the risk of toxins building up in the body, and weight loss. Prior studies regarding the association of ALT with HCC in type 2 diabetic patients are limited, and most of them had been conducted in general population \[[@R17], [@R18]\] or on individuals with cirrhosis \[[@R19]\], HBV \[[@R8], [@R9], [@R20]\], or HCV infections \[[@R21]\] who are at high risk. Two studies that developed HCC prediction models in Asian general population were conducted \[[@R17], [@R18]\]. One study revealed that ALT levels was a significant predictor of HCC risk in Chinese patients with an unknown or HBV- or HCV-negative infection status \[[@R17]\]. On the contrary, ALT level wasn't incorporated into HCC prediction model in a Japanese population \[[@R18]\]. Among studies conducted in patients with a high risk of HCC, no study has reported ALT level to be a significant predictor in patients with cirrhosis or HCV infection \[[@R19], [@R21]\]. On the contrary, ALT level was a significant predictor among studies that explored factors associated with HCC, or developing scoring systems to predict HCC among patients with chronic HBV infection \[[@R8], [@R9], [@R20]\]. Several studies explored predictors for HCC in type 2 diabetic patients. In a study with a sampling scheme based on insulin use status, HbA1c, was the key predictor for HCC, and ALT level was not considered in the potential factors \[[@R22]\]. Another cohort study focused on the effect modification of HBV infection with lipid profiles and medication use on liver cancer, but it did not consider ALT level \[[@R23]\]. Our study has focused on the association between ALT level and HCC risk, and has directed our research effort on the primary objective to create a strong basis for interpreting the study results. Hence, our study provides credible findings showing joint effect of ALT level and chronic liver diseases on HCC risk in Chinese type 2 diabetic patients. This study has several strengths. First, this study had a nationwide cohort with a large sample size to assess whether elevated ALT levels increased HCC risk in Chinese type 2 diabetic patients enrolled in NDCMP. Second, baseline information was collected before subsequent diagnosis of HCC. This process prevents recall bias inherent in case-control studies with exposure collected after HCC diagnosis. Finally, we considered many factors, including lifestyle habits, anti-diabetes treatment, comorbidity, and biomarkers of FPG, lipid profiles and HbA1c in multivariable models. However, several limitations are noted. First, our datasets do not contain information of aspartate aminotransferase (AST) and alpha-fetoprotein (AFP) markers, which had been demonstrated to be associated with HCC in general population. Thus, we cannot adjust for the confounding effects of these two biomarkers. In addition, the database does not contain information on leisure-time physical activity, and dietary habits, which may also be risk factors for liver cancer. Future studies linking administrative data, personal information is warranted. Second, a potential measurement error because of undiagnosed or misdiagnosed liver cancer cases may exist. However, the likelihood of overdiagnosis would be small because of audit system of clinic and hospital records. NHI program regularly conducts expert reviews of patient charts to confirm the validity of randomly selected claims from all hospitals. False or inconsistent reports will incur severe penalties. In addition, HCC cases were defined as patients with at least three ambulatory claims or at least one inpatient care claim for HCC to improve true positive rates. However, errors arising from miscoding and misclassification may be random, which would result in underestimation of the effect if the association between ALT and HCC exists. This implied that true effect would be stronger, which would be a lesser threat to the validity of our findings. Third, it is very likely to underdiagnose NADLD, HBV, HCV infection and cirrhosis because the tests for HBV and HCV infection, cirrhosis, and NAFLD were not regular check-up items for diabetes care, the tests weren't offered for patients with no probable indication or symptoms. Thus those who didn't have probable symptoms weren't seek for health care and won't be identified. If the un-diagnosis is random according ALT level, the impact of the un-diagnosis would result in underestimation for the independent effects of ALT level and these chronic liver diseases as well as their interactions. If the un-diagnosis is not random, it is likely that patients in the lower levels of ALT had a higher likelihood of un-diagnosis for these chronic liver diseases. The impact of the un-diagnosis would result in underestimating the independent and interactive effects of ALT level and these chronic liver diseases. Because the potential un-diagnosis bias results in the effect that may be toward the null, a lesser threat to validity of the findings. Finally, potential selection bias might be possible because of differential characteristics between type 2 diabetic patients who enrolled and did not enroll in NDCMP. To evaluate the potential selection bias, we made comparisons of age and gender distributions between our study subjects and type 2 diabetes population using NHIRD dataset, and we found similar distributions (difference in mean age was 1.5 years and in female proportion was less than 1%). Non-differential distributions in age and gender imply this kind of selection error might be random; thus, biased results on the effects may be null and would be a lesser threat to validity. In conclusion, our study demonstrates an elevated ALT level increased HCC risk in Chinese type 2 diabetic patients. This result indicates significant interactions were observed for ALT ≥ 40 IU/L with liver cirrhosis and HBV on HCC risk. Our study provides new insights for health professionals to target patients with diabetes who are at higher risks of HCC. Elevated ALT level should warrant medical attention. Diabetes care should provide lifestyle or treatment interventions to manage ALT level, liver cirrhosis and HBV for reducing HCC burden. MATERIALS AND METHODS {#s4} ===================== Study population {#s4_1} ---------------- We carried out a nationwide retrospective cohort study, Taiwan Diabetes Study, among enrollees in National Diabetes Care Management Program (NDCMP), founded by Ministry of Health and Welfare of Taiwan in 2001 for enhancing the quality of diabetes care. NDCMP provides additional financial incentives for care providers to have their eligible patients enrolled in this program and to have continuing clinical education and training programs for certification. Taiwan Diabetes Study, a nationwide cohort study, consisted of 63,084 ethnic Chinese type 2 diabetic patients registered in NDCMP in 2002-2004. We used date of entry into NDCMP as index date. Patients with a clinically confirmed diagnosis of diabetes mellitus based on American Diabetes Association criteria (International Classification of Diseases, ninth revision, Clinical Modification (ICD-9-CM) diagnosis code 250) were invited by their primary care providers to enroll in this program. We ruled out persons who had type 1 diabetes (ICD-9-CM; code 250.x1/x3) and gestational diabetes, patients aged under 30 years, diagnosed with any cancers at baseline, and less than one year of follow-up (n=11,157). Enrollees with missing data on sociodemographic factors, lifestyle behaviors, blood biomarkers, and medication use (n = 5,558) were also excluded from analysis. Finally, 46,369 enrollees were qualified (22,236 men and 24,133 women) in the analysis (Figure [3](#F3){ref-type="fig"}). This study was approved by Ethical Review Board of China Medical University Hospital. Informed consent of study participants was not required because dataset used in this study consists of de-identified secondary data released for research purposes. {#F3} Data sources for baseline and follow-up assessments {#s4_2} --------------------------------------------------- We used datasets of admission for inpatient care and ambulatory care visits during 2002--2011. Every person in Taiwan has a unique personal identification number (PIN). To ensure patients' security and privacy, NHIRD provided data with patient identities being encrypted. Researchers can interlink all NHI datasets through encrypted PIN. Data consists of information on sociodemographic variables, date, treatments and source of diagnosis for ambulatory care and inpatient admission. ICD-9-CM codes were used to identify individual comorbidity status. Proportion of enrollees leaving from NHI program is fairly low because of comprehensive coverage of this program. Thus, loss follow-up bias is negligible. NDCMP is a multidisciplinary case management program run by Ministry of Health and Welfare of Taiwan since 2001. This program provided diet, physical activity, and enhanced self-care education, annual diabetes-specific physical examinations and laboratory tests, as well as continuity of care to decrease diabetes-related complications. On the date of entry into the NDCMP, enrollees underwent a physical check-up, consisting of comprehensive assessment of disease and complication status, along with a series of blood tests, urine tests, and body measurements. Enrollees had to complete a standardized and computerized questionnaire by a case management nurse to record previous or current disease status, medication, and lifestyle habits. Smoking and alcohol drinking status were according to patients' self-report. The smoking group consisted of patients who were current or past smokers and non-smoking group included those who had never smoked. Blood was extracted from an antecubital vein in the morning, after a 12-hour overnight fast, and sent for analysis within four hours after collection. To diagnose a susceptible individual for acute HBV in clinical settings in Taiwan, an HbsAg test is ordered. The test for serum HBV DNA is optional. If the test result for HbsAg is positive, it would be compatible with acute HBV infection. The HBsAg test has to be repeated in six months. If the test result is positive for more than six months, it is defined as chronic HBV based on the definition of clinical guideline for chronic HBV infection that the continued presence of HBsAg in the blood for longer than six months. For HCV infection, an individual with positive anti-HCV antibodies and detectable HCV RNA is defined as hepatitis C infection. In addition, none of these patients have history of hemorrhage from esophageal varices or ascites, or hepatic encephalopathy. Whether an individual is diagnosed as having cirrhosis is determined by peritoneoscopy, biopsy or both, and by clinical symptoms with ultrasonographic findings. An individual with a typical irregular-surfaced liver with coarse internal architecture by the ultrasonography in addition to overt ascites or esophageal varices demonstrated by fiberscopic examination is defined as cirrhosis. NAFLD includes a spectrum of liver disease ranging from simple steatosis (non-alcoholic fatty liver \[NAFL\]) to non-alcoholic steatohepatitis (NASH). For diagnosis of NAFLD, it is required there is evidence of steatosis either by imaging or histology and no secondary causes of steatosis, such as viral hepatitis, increased alcohol consumption, use of steroid medications or other causes. The diagnosis of steatosis is based on a liver biopsy or imaging techniques such as ultrasound. A positive test result for NAFL if a liver biopsy with ≥5--10% of hepatocytes exhibiting macroscopic steatosis or the semi-quantitative image of ultrasound indicating any degree of steatosis alone or steatosis with lobular inflammation but without ballooning. NASH is only diagnosed by liver biopsy with the presence of ballooning injury. Outcome ascertainment {#s4_3} --------------------- Primary outcome was HCC, determined by ambulatory and inpatient care data in NHIRD. HCC incident cases were ascertained by codes (155 for HCC) of ICD-9-CM. All HCC cases met at least one of the following criteria to enhance its true positive rate: at least three ambulatory claims or at least one inpatient care claim. A total of 1,018 patients with HCC incidence were identified from this cohort. Follow-up time began with index date and ended with a newly diagnosed HCC, death, withdrawal from the insurance program, or end of follow-up on December 31, 2011. The diagnosis of HCC in Taiwan is based on clinical, imaging, and histopathological findings. All patients are asked about their medical history to check for risk factors and symptoms, and are examined for signs of liver cancer. If symptoms or the physical exam results suggest patients might have liver cancer, imaging tests are performed during the preoperative period such as ultrasonography, CT scan, or abdominal MRI. Lab tests such as alpha-fetoprotein and liver function are ordered to determine what might have caused the liver cancer, and how well the liver function, which can affect types of treatments. In some cases, a biopsy is needed. For those who have surgery, all specimens are obtained. The validity of the cancer diagnosis in the NHIRD has been reported previously \[[@R24]\] using all newly diagnosed people with cancer between January 1, 2001 and December 31, 2012. The cancer cases from the National Cancer Registry (NCR) in Taiwan had been treated as true cases. The estimates of the validity of cancer diagnoses in the NHI database by sensitivity, specificity, positive (PPV) and negative (NPV) predictive values were 80.07%, 99.99%, 90.39%, and 99.99%, respectively. This study's findings provide the evidence to support NHI database is a valid source for cancer epidemiology study. The data source for true cases was NCR, which is implemented by the Ministry of Health and Welfare and is a compulsory system that asks hospitals caring cancer patients to provide valid personal, clinical, pathology, laboratory, and imaging data. The Taiwan NCR's data quality and validity of diagnostic criteria methods followed the definitions proposed by \[[@R25]\]. It has been reported that the proportions of death certificate-only and morphological verification cases in Taiwan \[[@R26]\] were comparable with cancer registries of Iceland \[[@R27]\], Norway \[[@R28]\], and Singapore \[[@R29]\]. Covariates {#s4_4} ---------- Data on comorbid conditions were extracted from NHIRD for a 24-month period prior to index date by using outpatient and inpatient claim data. Instead of a 12-month period, we specified a 24-month period because a few of these comorbid conditions are not common. To ascertain that we do not miss patients' diagnosis, we required a longer period. For the number of image tests, we extracted outpatient and inpatient claim data from NHIRD for a 24-month period after index date. Data on medication uses prescribed for diseases were calculated for 12-month period prior to cohort entry. Outpatient prescriptions within one year of index date were used to define their anti-diabetes or statin medication use. A patient was a user of anti-diabetes or statin if his/her number of prescription days was greater than 90 days. A patient may have more than one type of anti-diabetes medication use if she/he had more than one medication use. We classified anti-diabetes medications into: no medication, sulfonylurea monotherapy or sulfonylurea plus oral anti-diabetes drug (OAD) monotherapy other than metformin or sulfonylurea (OAD-other), metformin monotherapy or metformin plus OAD-other combination, metformin plus sulfonylurea combination, OAD-other monotherapy or OAD-other combination, insulin monotherapy, and insulin plus one or more OAD. Statistical analysis {#s4_5} -------------------- Baseline measurement of ALT was determined based on datasets of electronic lab records. ALT level at baseline was grouped into four categories according to clinical criteria: ≤40, 40--80, 80--120, and \> 120 U/L. Cox proportional hazards models were utilized to assess the association between ALT level and HCC risk for multivariable adjustment. Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated under three multivariable models. The first multivariable model adjusted for age, gender, smoking status, alcohol drinking status, diabetes duration, statin use, and type of hypoglycemic drug and anti-hypertension drug treatment. The second one additionally adjusted for blood biomarkers including HbA1c, fasting plasma glucose (FPG), high-density lipoprotein (HDL), low-density lipoprotein (LDL), triglyceride (TG), and body mass index (BMI). The third one further included time-varying comorbid conditions. Proportional hazards assumption was verified by the graph of log (−log(survival)) versus log of survival time graph and by statistical significance test of a covariate that allowed time-varying ALT. No significant violation was found. To test the trend of ALT, categorical ALT had been treated as an ordinal variable by coding the four categories of ≤40, 40--80, 80--120, and \> 120 U/L from 1 to 4. To account for the effects of other variables, this ordinal variable was entered into the multivariate Cox's proportional hazards model along with the other variables and p value for this ordinal variable was reported as p for trend. Three dummy indicators were created to assess joint effect of ALT and each chronic liver disease. Using individuals with ALT level ≤ 40 U/L and without chronic liver disease as reference, these three dummy indicators estimated independent effects of ALT level \> 40 U/L only, chronic liver disease only, and both ALT level \> 40 U/L and chronic liver disease. Interactions of ALT level \> 40 U/L with age, gender, ALD, NAFLD, liver cirrhosis, HBV and HCV infection, and any one of chronic liver diseases (CLD) were examined by including their product terms into the full model, and its significance was tested by likelihood ratio test. Moreover, proportion attributable to interaction (AP), relative excess risk due to interaction (RERI), and synergy index (S index) were derived. A zero value of PERI or AP indicates no interaction, a positive value of PERI or AP indicates positive interaction, and a negative value of PERI or AP indicates negative interaction. A value of one for S index indicates no interaction, a value of greater than one indicates positive interaction, and a value of less than one indicates negative interaction. Sensitivity analyses were performed to examine robustness of our findings. The main analyses were repeated by excluding participants with stroke, hypoglycemia, coronary artery disease and HBV and HCV infection separately and together. SAS version 9.4 (SAS Institute Inc., Cary, NC) was used for all analyses. A two-sided level of significance was specified at 0.05. **Author contributions** Conception and design: Tsai-Chung Li, Cheng-Chieh Lin and Chia-Ing Li. Collection and assembly of data: Chiu-Shong Liu, Pao-Hsuan Lin, Wen-Yuan Lin and Chih-Hsueh Lin. Data analysis and interpretation: Chia-Ing Li, Sing-Yu Yang and Jen-Huai Chiang. Manuscript writing: Cheng-Chieh Lin and Tsai-Chung Li. Final approval of manuscript: All authors. **CONFLICTS OF INTEREST** The authors declare no conflicts of interest. **FUNDING** This study was funded primarily by the Bureau of National Health Insurance (DOH94-NH-1007), the Ministry of Science and Technology of Taiwan (NSC 101-2314-B-039 -017-MY3 & NSC 102-2314-B-039-005-MY2 & MOST 104-2314-B-039-016 & MOST 105-2314-B-039 -021 -MY3 & MOST 105-2314-B-039 -025 -MY3) and Taiwan Ministry of Health and Welfare Clinical Trial Center (MOHW106-TDU-B-212-113004) and China Medical University Hospital (DMR-106-094). [^1]: These authors equally contributed as first author

1. Introduction {#sec1-ijms-20-06119} =============== Mycotoxins are extracellular metabolites produced by filamentous fungi that contaminate cereals, grains, fruits, and vegetables. The most important *Fusarium* toxins are trichothecenes, zearalenone (ZEN), and fumonisins (FBs), that are dangerous for human and animal health, and their presence in food is regulated worldwide \[[@B1-ijms-20-06119]\]. Mycotoxin co-occurrence in food is a real and relatively underestimated issue \[[@B2-ijms-20-06119]\], as is the modification of toxins by plant metabolism (creating masked mycotoxins) \[[@B3-ijms-20-06119]\]. Both factors mean that the levels of toxins measured in food, and therefore being ingested, are significantly underestimated. Due to this, it is likely that normative limits will be lowered by the regulatory agencies in the future. Cereals, the staple foods of diets all over the world, are perfect hosts for pathogenic and toxigenic fungi and represent one of the main sources of mycotoxin contamination for humans and animals \[[@B4-ijms-20-06119]\]. Among toxigenic species, *Fusarium* spp. (Division Ascomycota) are major producers of mycotoxins in cereals \[[@B5-ijms-20-06119]\]. Trichothecenes A and B are mainly associated with Fusarium head blight (FHB) and crown rot (FCR) in wheat and barley. The major group of *Fusarium* spp. responsible for these diseases includes *Fusarium graminearum* species complex (FGSC; \[[@B6-ijms-20-06119]\]) that exhibits a diverse distribution of species across the different continents \[[@B7-ijms-20-06119]\]. The most important species are *F. graminearum, F. culmorum*, and *F. pseudograminearum* \[[@B8-ijms-20-06119],[@B9-ijms-20-06119]\]. Grain quality decrease and yield are of concern \[[@B10-ijms-20-06119]\]. The trichothecenes type B are the most prevalent and comprise deoxynivalenol (DON) and nivalenol (NIV) and their acetylated forms 3-ADON, 15-ADON, and 4-ANIV \[[@B11-ijms-20-06119]\]. They are immunosuppressant and neurotoxic and cause intestinal irritation, leading to feed refusal in livestock \[[@B12-ijms-20-06119],[@B13-ijms-20-06119]\]. In maize, *F. graminearum* and other related species were found to be associated with Fusarium ear rot (FER), contaminating grains with ZEN. ZEN displays estrogenic activity, causing reproductive problems in animals, in addition to cytotoxic and immunosuppressive effects \[[@B14-ijms-20-06119],[@B15-ijms-20-06119]\]. Ear rot in maize is also caused by *F. verticillioides* (syn. *F. moniliforme* \[[@B16-ijms-20-06119]\]) and *F. proliferatum*, which produce fumonisins \[[@B17-ijms-20-06119]\]. Fumonisins have been classified as Group 2B carcinogens (i.e., as possibly carcinogenic to humans \[[@B18-ijms-20-06119]\]), and fumonisin B1 (FB~1~) is the most abundant analogue found in contaminated samples \[[@B19-ijms-20-06119]\]. Moreover, *Fusarium* spp. infecting cereals can also produce other minor mycotoxins with cytotoxic effects such as enniatins, beauvericin, and moniliformin. Knowledge gaps regarding the occurrence, toxicity, and toxicokinetic data for these compounds in cereal crops represent a major and immediate problem \[[@B20-ijms-20-06119]\]. *Fusarium* spp. infections of cereals are therefore a major concern for both the growers and the food chains associated with the processing of grains. Several control strategies against this complex group of pathogens have been developed and include host resistance, the application of fungicides, and the implementation of specific agricultural practices \[[@B21-ijms-20-06119]\]. However, effective management of *Fusarium* pathogens and the related toxins cannot be achieved through the use of a single control strategy because each has its own limitations \[[@B22-ijms-20-06119]\]. Therefore, at least in Europe, integrated disease management is urgently needed, favored by European Regulation 1107/2009/EC and European Directive 128/2009/EC \[[@B23-ijms-20-06119],[@B24-ijms-20-06119]\]. Moreover, biocontrol approaches are becoming increasingly important due to the limitation on the use of certain fungicides. Among the biocontrol agents (BCAs) used to control toxigenic *Fusarium* spp. in cereals, bacteria have shown a number of successful outcomes. For instance, strains of *Bacillus* spp. \[[@B25-ijms-20-06119],[@B26-ijms-20-06119],[@B27-ijms-20-06119],[@B28-ijms-20-06119]\], *Brevibacillus* sp. \[[@B29-ijms-20-06119]\], *Pseudomonas* spp. \[[@B27-ijms-20-06119],[@B30-ijms-20-06119]\], and *Lysobacter enzymogenes* \[[@B31-ijms-20-06119]\] were applied to limit pathogen development, reducing disease severity and mycotoxin production. Microbial communities or single strains were also tested to detoxify contaminated substrates as reviewed by McCormick in 2013 \[[@B32-ijms-20-06119]\]. Bacteria of the genus *Streptomyces* display promising plant growth-promoting features and biocontrol efficacy against plant pathogens. They belong to the phylum of Gram-positive Actinobacteria, which is one of the largest taxonomic units within the bacterial domain, and include microorganisms relevant to human and veterinary medicine, biotechnology as well as ecology \[[@B33-ijms-20-06119]\]. Streptomycetes are the most abundant actinobacteria in soil \[[@B34-ijms-20-06119]\]. They display a unique life cycle, and after germination grow through a combination of tip extension and the branching of hyphae. They first form a vegetative mycelium firmly attached to the growth substrate and, subsequently, due to nutrient depletion and under environmental stress signals, develop an aerial mycelium. Each aerial hypha then differentiates into a long chain of pre-spore compartments which subsequently mature into individual spores \[[@B35-ijms-20-06119]\]. The ability to produce a variety of secondary metabolites, including anti-infective agents, has an important ecological role including the inhibition of competitors during the transition from mycelial to aerial growth \[[@B36-ijms-20-06119]\]. These various characteristics enable them to colonize different substrates and establish symbiotic interactions with plant tissues and other eukaryotes \[[@B37-ijms-20-06119]\]. The ability to produce numerous secondary metabolites means they are the most exploited bacterial genus in natural product research. Notably, more than half of all antibiotics in current clinical use are derived from actinobacterial secondary metabolites \[[@B38-ijms-20-06119]\]. Furthermore, *Streptomyces* spp. were evaluated as plant growth-promoting bacteria (PGPB), as they can inhibit pathogen development, enhance nutrient uptake by mineral solubilization, and increase plant growth by nitrogen fixation and phytohormone synthesis \[[@B39-ijms-20-06119]\]. Streptomycetes have therefore been investigated for their possible use in agriculture, including cereal crops \[[@B40-ijms-20-06119]\]. The diversity of secondary metabolite production plus their reported endophytic features make the genus *Streptomyces* a perfect candidate to control toxigenic *Fusarium* spp. development and related toxin production \[[@B41-ijms-20-06119],[@B42-ijms-20-06119]\]. Endophytic microorganisms have been reported as useful antagonists against Fusarium head blight, and are able to reduce disease severity on spikelets \[[@B43-ijms-20-06119]\]. Nevertheless, the incredible diversity and potentiality of these microorganisms against mycotoxigenic fungi and their possible influence on toxin accumulation have been rarely explored and deserve further investigation \[[@B44-ijms-20-06119]\]. This review describes reports in which streptomycetes, or molecules derived from them, were exploited against *Fusarium* spp., and will pay special attention to the possible influence on toxin production. The scientific and patent literature were analyzed from the period 2000--2018. 2. Critical Assessment of Literature {#sec2-ijms-20-06119} ==================================== Despite the huge amount of literature regarding the biological control of *Fusarium* mycotoxigenic isolates in cereals, only two products have found a consistent market niche \[[@B45-ijms-20-06119]\]. These are based on *Pseudomonas chlororaphis* and *Pythium oligandrum* and are marketed in Europe as Cerall^®^ (Belchim Crop Protection) and Polyversum^®^ (Biopreparáty/De Sangosse), respectively \[[@B46-ijms-20-06119]\]. Furthermore, no *Streptomyces* product is officially registered to be used for this purpose \[[@B47-ijms-20-06119]\]. The main obstacles for biocontrol agents are due to the lack of consistency when microbial inoculants are applied under complex environmental conditions, and to the complexity of finding appropriate formulation and timing for application \[[@B48-ijms-20-06119]\]. Biological, ecological, toxicological, and regulatory cost factors also influence the effectiveness and marketability of biological control products \[[@B49-ijms-20-06119]\]. In order to verify the status of research using *Streptomyces* strains, and their derived molecules, to limit toxigenic *Fusarium* spp. infections and/or toxin contamination, we screened the published literature. To critically assess the status of each research paper, a set of definitions describing the type of study and their accuracy was established as follows. *Streptomyces* species definition. Species identification is essential as approximately 10 *Streptomyces* species have been described as plant pathogens, causing economically important diseases on underground plant structures such as tuber/root crops. The best studied and characterized of these is *Streptomyces scabies* which causes potato scab \[[@B50-ijms-20-06119],[@B51-ijms-20-06119]\]. Moreover, from a food safety perspective, *Streptomyces* isolates are able to produce dangerous metabolites for human and animal health, such as antimycin A found on wheat and barley grains \[[@B52-ijms-20-06119]\]. Therefore, it is essential that species and strain characterization is performed accurately.In vitro testing for antifungal activity. This is generally the first step for identifying antifungal microorganisms or molecules produced by them. Such studies help define the mechanism(s) of action of the *Streptomyces* species(s) and lead to the identification of potential interactions with the target organism. Assessment of bioactivity should consider the diversity of targets (verifying if pathogen diversity influences the consistency of the BCA or derived product). Indeed, specific interactions occur among bacterial and fungal strains \[[@B53-ijms-20-06119]\] and this may impact the biocontrol capability of a strain \[[@B54-ijms-20-06119],[@B55-ijms-20-06119]\].The effect of culture media in the bioassays in vitro. Media composition modulates secondary metabolite production in actinomycetes \[[@B55-ijms-20-06119],[@B56-ijms-20-06119],[@B57-ijms-20-06119]\], and optimizing laboratory selection procedures should broaden the number of interesting BCAs that can be identified.The use of fermentation extracts to perform bioassays. During screening procedures, it would be ideal to identify the metabolite(s) responsible for the observed antifungal effect. The screening of crude extracts is generally followed by further steps of purification and chemical analysis, and retesting of purified compound(s) \[[@B58-ijms-20-06119]\].Evaluation of the antifungal mode of action. Risks concerning the use of these antibiotic-producing bacteria associated with events of horizontal gene transfer and the development of antibiotic resistance are still under debate within the scientific community \[[@B44-ijms-20-06119]\]. However, given the current legislative requests \[[@B59-ijms-20-06119]\], understanding the mode of action is essential in order to proceed with the registration of a BCA, in order to avoid risks of spreading dangerous metabolites for human and animal health in the environment \[[@B60-ijms-20-06119]\].Assessment of the ability to colonize treated plant organs. Many BCAs are rhizosphere-colonizing microorganisms and can be applied as seed coatings \[[@B61-ijms-20-06119]\]. However, some *Streptomyces* spp. can exhibit endophytic behavior, colonizing different parts of the plant (e.g., roots, stem, leaves) \[[@B41-ijms-20-06119]\]. Some BCAs exhibit activity both in the rhizosphere and after infection of the plant and function inside the root at the same time. Therefore, the colonization niche of the strain should be investigated in order to warrant a consistent protection \[[@B62-ijms-20-06119]\]. These studies are fundamental to providing an assessment of the durability of the protection warranted by the BCA.Testing the influence of complex environmental conditions. As for pathogens during disease development, antagonist strains are influenced by environmental factors that strongly impact the ability of the BCAs to exert their biocontrol activity \[[@B40-ijms-20-06119]\]. Assessing the impact of environmental parameters on BCAs using both greenhouse and field trials is essential to selecting strains with consistent biocontrol activity.Assessment of antifungal and plant growth-promoting effect in planta. This step is essential, given that the BCA will ultimately be employed in the field. Very often, there is poor correlation between in vitro and in planta trials \[[@B55-ijms-20-06119],[@B63-ijms-20-06119],[@B64-ijms-20-06119]\]. Moreover, the wide range of metabolites produced may have direct influences on plant development, altering growth and plant fitness both positively and negatively \[[@B39-ijms-20-06119]\]. Indeed, negative effects cannot be underestimated---some *Streptomyces* can be pathogens (see before) or produce phytotoxic and herbicidal substances \[[@B65-ijms-20-06119]\].Assessment of the method used for application. Selecting an appropriate delivery system for the BCA as well as an optimized formulation can determine its efficacy in the field \[[@B66-ijms-20-06119]\].The effects of the BCA on the pathogen inoculum in planta. Due to the complex epidemiology of *Fusarium* diseases in cereals, quantification of the pathogen in planta is important to verify if the treatment can, for example, effectively reduce the source of overwintering inoculum, limiting the infection pressure at the subsequent infection season \[[@B67-ijms-20-06119]\].Quantification of the mycotoxin. It is essential to verify if the BCA limits toxin production, specifically given that there is a lack of full correlation between the presence of the fungus and the amount of toxin that is found in the grains \[[@B68-ijms-20-06119],[@B69-ijms-20-06119]\]. Moreover, some secondary metabolites can limit toxin production without impairing growth of the pathogen \[[@B70-ijms-20-06119]\]. Biological interactions can also lead to unexpected crosstalk between the BCA and pathogen that can lead to an overproduction of toxins and secondary metabolites \[[@B71-ijms-20-06119],[@B72-ijms-20-06119],[@B73-ijms-20-06119],[@B74-ijms-20-06119],[@B75-ijms-20-06119]\]. 3. Literature Analysis {#sec3-ijms-20-06119} ====================== To guide future implementation of biocontrol research using *Streptomyces* spp., it is essential to identify the strengths and weaknesses of past and present research in this domain. Therefore, we reviewed the published literature, focusing on the methods used for the selection of promising biocontrol streptomycetes and on the results achieved. We searched the Scopus and Google Scholar databases for articles including the words "*Fusarium*" and "*Streptomyces*" that were published during the timeframe 2000--2018. The resulting articles were read and individually screened leading to the identification of 63 articles that dealt with the ability of *Streptomyces* or their secondary metabolites to limit the growth or toxin production of toxigenic *Fusarium* spp. in cereals ([Table 1](#ijms-20-06119-t001){ref-type="table"}). *Streptomyces* spp. or their derived molecules have been tested mostly against *Fusarium* spp. producing trichothecenes, including DON. The species investigated are all usually found to infect cereals and include *F. graminearum*, *F. culmorum*, *F. poae*, *F. cerealis, F. sporotrichioides*, and *F. equiseti.* The most studied interactions address the wheat--*F. graminearum* pathosystem, which is the most important cause of DON (and derivatives) accumulation in grains \[[@B76-ijms-20-06119]\]. Less frequently, streptomycetes have been tested against fumonisin producers in maize, all belonging to the *F. fujikuroi* species complex \[[@B77-ijms-20-06119]\]. 3.1. *Streptomyces* Identification {#sec3dot1-ijms-20-06119} ---------------------------------- Regarding the identification of *Streptomyces* species, most studies focused on the integration of morphological and molecular characteristics. Given the complexity of streptomycete biology \[[@B33-ijms-20-06119]\], the use of 16S rRNA alone as molecular marker is not sufficient to achieve species discrimination. Multi-locus sequence typing \[[@B78-ijms-20-06119]\] integrated with biochemical and morphological identification would be a preferred option, but none of the studies used this approach. On this basis, all the species identifications reported in the selected papers should be treated with caution. Looking forward, the increasing number of *Streptomyces* strain genomes now available may help in correct species identification \[[@B79-ijms-20-06119]\]. 3.2. Screening for Antifungal Activity: In Vitro Tests {#sec3dot2-ijms-20-06119} ------------------------------------------------------ Among the selected articles, in vitro testing is the most commonly used first-line screening method. Indeed, dual-culture assays on solid media are exploited in all the studies as a preliminary screen, evaluating the inhibition halo between the growth of the streptomycete and the fungal target or measuring the radial growth of the *Fusarium* colony in comparison with an untreated control to obtain a percentage of growth inhibition. Rather than use these standard in vitro inhibition assays, some research groups \[[@B80-ijms-20-06119],[@B81-ijms-20-06119],[@B82-ijms-20-06119]\] characterize the type of interactions occurring in dual culture by using the index of dominance (ID) \[[@B83-ijms-20-06119]\]. The ID consists of visually observing antagonist and pathogen growth in dual culture, testing different media or water activity (*a~w~*) of the culture medium, and classifying the type of interactions occurring based on predefined scores, namely, mutual intermingling (1/1), mutual inhibition on contact (2/2), mutual inhibition at a distance (3/3), dominance of one species on contact (4/0), and dominance at a distance (5/0). This method evaluates whether the inhibition is due to the production of antifungal metabolites diffusible in the media or whether the mycelium is parasitized by the antagonists. Moreover, the negative effect of the target pathogen on the potential antagonists can be noted. Therefore, the selection of biocontrol agents is carried out by evaluating the biocontrol interactions (e.g., mycoparasitism, competition, or antibiosis) established under different growth conditions. For most reports, growth of the *Streptomyces* inoculum to some predefined point usually takes place on agar media before addition of the pathogen in order to allow a complete establishment of these growing bacteria \[[@B80-ijms-20-06119],[@B84-ijms-20-06119]\]. The use of a diverse range of growth media and fungal strains was evaluated in our analysis, given the importance that these criteria have in the estimation of the biocontrol activity in vitro \[[@B55-ijms-20-06119]\]. Interestingly, the influence of growth media was seldomly evaluated in these types of experiments \[[@B84-ijms-20-06119],[@B85-ijms-20-06119]\] as well as the assessment of antifungal activity on different *Fusarium* strains belonging to a single species \[[@B80-ijms-20-06119],[@B86-ijms-20-06119],[@B87-ijms-20-06119],[@B88-ijms-20-06119]\]. Given the lack of a standardized protocol when performing dual-culture assays (e.g., *Fusarium* strains on which the biocontrol activity should be tested, position and distance between streptomycetes and *Fusarium* strain inoculum, timing of observation after pathogen inoculum, culture medium), it is difficult to compare the results between studies. However, here we report some examples of the wide range of activities recorded against mycelial proliferation. For instance, growth inhibition percentages against *F. graminearum* and *F. verticillioides* ranged from the weakest (\<20%) \[[@B84-ijms-20-06119],[@B89-ijms-20-06119]\] up to 60--90% of inhibition \[[@B87-ijms-20-06119],[@B90-ijms-20-06119]\]. Yekkour et al. \[[@B91-ijms-20-06119]\] obtained different levels of inhibition in dual culture for isolated streptomycetes---indeed, only 6 out of 133 isolates displayed an anti-*Fusarium* activity and in particular only *F. culmorum* was significantly inhibited (inhibition halo \>20 mm). Less sensitive fungal species were *F. moniliforme*, *F. sporotrichoides*, *F. graminearum*, and *F. proliferatum* \[[@B91-ijms-20-06119]\]. 3.3. Evaluation of Antifungal Mechanism of Action {#sec3dot3-ijms-20-06119} ------------------------------------------------- The importance of the identification of any antifungal molecules involved in the bioactivity led some researchers to achieve a complete characterization of the compounds involved. The fermentation process and the optimization of all the parameters (e.g., medium, agitation rate, pH, temperature) were strain- and laboratory-dependent \[[@B92-ijms-20-06119],[@B93-ijms-20-06119],[@B94-ijms-20-06119],[@B95-ijms-20-06119]\]. For instance, it has been reported that some of the *Streptomyces* strains which are active against *F. moniliforme* on solid media lack antibiotic production in submerged liquid culture, highlighting the importance of an appropriate optimization of laboratory procedures and media in the stimulation of secondary metabolites \[[@B96-ijms-20-06119]\]. The first attempt of compound purification is commonly carried out by crude extract fractionation \[[@B97-ijms-20-06119],[@B98-ijms-20-06119]\]. Often the bioactivity of the selected strain is not related to a single mechanism, and different metabolites, enzymes, or volatile organic compounds likely contribute to the overall antifungal activity. Many studies exploited the fermentation broth as a source of bioactive compounds \[[@B99-ijms-20-06119],[@B100-ijms-20-06119],[@B101-ijms-20-06119]\]. Therefore, several compounds were purified and tested against toxigenic *Fusarium* spp. For example, strain PAL114 produced saquayamycins A and C which inhibited the growth of *F. culmorum* at the minimum inhibitory concentrations of 75 ug/mL \[[@B102-ijms-20-06119]\]. Three allelochemicals (5,7-dihydroxyflavone, 5-hydroxy-7-methoxyflavone, and di-(2-ethylhexyl) phthalate) able to inhibit mycelial growth of *F. graminearum* were isolated and purified from the fermentation broth of *Streptomyces* sp. 6803 \[[@B103-ijms-20-06119]\]. In vitro cultures of *Streptomyces* sp. 201 produced 2-methylheptyl isonicotinate able to inhibit the growth of *F. moniliforme* more efficiently than a natural analogue (isoniazid) \[[@B98-ijms-20-06119]\]. On the other hand, modest activity was observed by the metabolites extracted from *Streptomyces* LZ35 against *F. verticillioides* \[[@B104-ijms-20-06119]\]. For several studies, chitinase activity, rather than antibiotic production, was shown to play a role in the antifungal mechanism \[[@B105-ijms-20-06119],[@B106-ijms-20-06119],[@B107-ijms-20-06119],[@B108-ijms-20-06119]\]. In addition, new antifungal proteins have been characterized, such as the one isolated from *Streptomyces* sp. C/33-6 culture supernatants which displayed a fungicidal activity, determining complete inhibition of conidia germination of *F. graminearum* \[[@B109-ijms-20-06119]\]. Secondary metabolites exhibiting anti-*Fusarium* activity can also include volatile organic compounds (VOCs). For example, *Streptomyces alboflavus* TD-1 was able to reduce the mycelial growth of *F. moniliforme* when volatile metabolites were applied as fumigants \[[@B110-ijms-20-06119]\]. Inhibition of growth, sporulation, and conidial germination has been recorded when culturing this strain on wheat seeds. In addition, the VOC activity increased the fungal membrane permeability as observed by significant leakage of mycelial materials. Chemical analysis of these VOCs identified a high quantity of 2-methylisoborneol and 2-methyl disulphide, which were further tested for their antifungal activity \[[@B110-ijms-20-06119],[@B111-ijms-20-06119]\]. VOC production was also linked to the antagonist activity of *Streptomyces philanthi* RM-1-138 cultured on wheat seeds, which inhibited mycelium growth of *F. fujikuroi* by 50% \[[@B112-ijms-20-06119]\]. Chemical analysis showed that a complex mixture of volatile metabolites was involved \[[@B112-ijms-20-06119]\]. It is evident from our analysis that the biocontrol activity of *Streptomyces* strains involves a large range of bioactive molecules. The exploitation of *Streptomyces* spp. has been, and will in future also be, hindered by the variability of the production of these metabolites. Therefore, to exploit the huge diversity of streptomycetes for successful disease management, different factors, such as the age of the fungal colony, culture conditions, temperature, and other environmental parameters, will have to be carefully studied, even at the very early stages of investigation. Transferring the outputs of these laboratory studies to the field remains one of the major challenges in exploiting *Streptomyces* spp. as BCAs for tackling toxigenic *Fusarium* spp. 3.4. Assessment of Streptomycete Effects in Planta {#sec3dot4-ijms-20-06119} -------------------------------------------------- The literature reports a lack of durable and consistent effects when streptomycetes or commercially available formulations have been applied in greenhouse experiments and field trials \[[@B40-ijms-20-06119]\]. It is likely that the ability to cope in a complex environment, which comprises the plant, the presence of the pathogens as well as several abiotic factors, varies depending on the fitness of the strain and its formulation in the field. For this reason, verifying the level of colonization achieved by the strain when used as BCA is essential to confirming its ecological fitness. Only a few papers have addressed this question in detail. It is notable that most of these were published recently, which indicates an increasing level of attention regarding *Fusarium*--plant--*Streptomyces* interactions \[[@B113-ijms-20-06119],[@B114-ijms-20-06119]\]. Moreover, in planta experiments are essential during the process of BCA selection to confirm their ability to significantly decrease *Fusarium* spp. infections. Indeed, BCAs can influence crop growth and disease severity, and can reduce *Fusarium* inoculum levels on stubble after harvest as well as ideally the presence of mycotoxins \[[@B115-ijms-20-06119]\]. However, only a limited number of studies (*N* = 16) performed complete in planta studies. The application of streptomycetes was tested on seeds \[[@B84-ijms-20-06119],[@B116-ijms-20-06119],[@B117-ijms-20-06119],[@B118-ijms-20-06119]\], on the main emerged spike \[[@B80-ijms-20-06119],[@B84-ijms-20-06119],[@B86-ijms-20-06119],[@B87-ijms-20-06119]\] as well as wheat stubble \[[@B86-ijms-20-06119]\]. Indeed, these bacteria can contribute to the reduction of FHB on wheat at different times in the *Fusarium* spp. life cycle. In a research study conducted by Palazzini et al. \[[@B80-ijms-20-06119]\] in 2007, isolates from wheat anthers were applied to wheat heads grown in greenhouse and, after 16 days, their influence on FHB severity was estimated. Despite the slight reduction of disease symptoms in comparison to the control, streptomycete BRC 87B decreased the DON content in spikes to below a detectable level. For this reason, in a subsequent study it was tested in the field, showing the ability to decrease FHB severity and DON amounts, as well as the *F. graminearum* inoculum on wheat stubble \[[@B86-ijms-20-06119]\]. Testing the efficacy in the field also requires specific assessments of the way the strains are inoculated. For example, the use of a Korean strain isolated from rice kernels led to a significant reduction of the disease severity after its inoculation using a spore spraying method that was not achieved using the point inoculation method on wheat heads \[[@B84-ijms-20-06119]\]. This is actually the only study where the influence of the BCA application method was taken into account, and shows that, depending on the application of the BCA, different results can be obtained \[[@B47-ijms-20-06119]\]. Differences in the level crop protection have also been reported against other *Fusarium* spp. For instance, two *Streptomyces* strains designated as DAUFPE 11470 and DAUFPE 14632 were isolated from maize rhizosphere in Brazil and tested against maize seed pathogenic fungi. Treatments on seeds with biomass derived from streptomycete fermentation or with cell-free filtrate reduced significantly *Fusarium subglutinans* incidence on stored maize seeds \[[@B119-ijms-20-06119]\]. The same strains were also tested as spore suspension to assess their effects on seedling blight caused by *F. moniliforme* in greenhouse \[[@B120-ijms-20-06119]\]. Bacterial treatments significantly reduced disease incidence compared with the controls, with protection level variable according to the tested pathogen inoculum concentrations. Indeed, the disease incidence was significantly reduced at low and high antagonist and pathogen concentrations, respectively. Moreover, their ability to reduce chlamydospore germination was assessed---the percentage of germinated propagules was evaluated after antagonist treatments in sterilized soil added with glucose, to recreate the natural environment and enhance spore germination. The addition of glucose increased propagule germination in all the treatments, but the presence of the antagonists decreased this parameter up to 65%. This study stressed therefore the important influence of both antagonist and pathogen concentrations and the presence of nutrients in the final biocontrol efficacy obtained in planta \[[@B120-ijms-20-06119]\]. *Streptomyces* strains, as reported above, can be helpful in reducing disease symptoms, acting also as plant growth-promoting bacteria. Despite the wide range of metabolites produced by them, their ability to influence plant development has been seldomly studied by the current literature addressing the biological control properties of the strain. A few positive examples include the report of a negative influence on seed germination and seedling development \[[@B91-ijms-20-06119]\] as well as an improvement in plant growth parameters \[[@B87-ijms-20-06119]\]. 3.5. Evaluation of Streptomycete Activity against Mycotoxin Production {#sec3dot5-ijms-20-06119} ---------------------------------------------------------------------- As noted above, it is essential to accurately determine the concentration of mycotoxins present in grains destined for human or animal consumption. Similarly, verification of the toxin content under experimental conditions is vital for the future of potential streptomycete biocontrol agents. Indeed, it should be possible that the reduction of disease severity does not positively correlate with a reduction of the mycotoxin content in grain samples. So far, only one research group has evaluated the reduction of DON mycotoxins by *Streptomyces* strains isolated from wheat anthers, in comparison to the level of infection, in vitro, in greenhouse, and in the field \[[@B80-ijms-20-06119],[@B86-ijms-20-06119]\]. Indeed, they showed that their streptomycete strains (BRC 87B and BRC 273) were able to significantly reduce DON levels on wheat grains, without influencing disease severity caused by *Fusarium* infections \[[@B80-ijms-20-06119]\]. This suggests the existence of a specific mechanism of inhibition uncoupling fungal fitness and toxin production. Follow-up research by the same group evaluated in the field the use of BRC 87B, which showed strong inhibition of DON production in wheat spikes \[[@B86-ijms-20-06119]\]. Preliminary in vitro studies have also been conducted to verify the ability of streptomycetes to limit fumonisin accumulation. Strains isolated from soil samples amended with different organic manures by Nguyen et al. were tested against fumonisins FB~1~ and FB~2~ production by *F. verticillioides* \[[@B121-ijms-20-06119]\]. They significantly decreased (by up to 98.2%) the level of FB~1~ and FB~2~ in agar plate cultures \[[@B121-ijms-20-06119]\]. Inhibition of FB~1~ accumulation on milled maize agar was also demonstrated in another in vitro study using *Streptomyces* sp. AS1 \[[@B122-ijms-20-06119]\], a strain isolated from peanuts in Egypt. Further, El-Naggar et al. \[[@B123-ijms-20-06119]\] showed the ability of *Streptomyces* isolates to reduce accumulation of a wide range of mycotoxins including total aflatoxins, fumonisin, zearalenone, T-2 toxin, AOH, and AME. However, the identity of the *Fusarium* spp. producers was based only on morphological characteristics and should be considered with caution. 4. Patent Search {#sec4-ijms-20-06119} ================ To have a complete overview of the work using *Streptomyces* against toxigenic fusaria, a research of the major patent databases was carried out. Using both Espacenet and Orbit intelligence, a total of 233 results were obtained using the keywords "*Fusarium*" and "*Streptomyces*". By manually screening the titles and abstracts, a total of 25 patents were retained and added to [Table 2](#ijms-20-06119-t002){ref-type="table"}. Given the use of different languages (most not English), only certain abstracts could be accessed, and it was therefore not possible to apply the same critical criteria used in our literature search. Most of the patents claimed general activity of strains and derived molecules against a large set of microorganisms including toxigenic fusaria. Only a single patent in its claim directly addressed the ability to limit *F. graminearum* growth on cereals \[[@B147-ijms-20-06119]\]. Two documents patented the antifungal metabolites isolated from streptomycete strains and tested them against toxigenic *Fusarium* spp. \[[@B148-ijms-20-06119],[@B149-ijms-20-06119]\]. The other patents are related to specific formulation methods, using live streptomycetes, proposed as biocontrol products against plant pathogens, among them *Fusarium* spp. of cereal crops. Interestingly, most of the patents are concentrated in the last five years ([Table 2](#ijms-20-06119-t002){ref-type="table"}), therefore further developments could also be expected towards novel industrial applications in the near future. 5. Conclusions and Perspectives {#sec5-ijms-20-06119} =============================== Our review of the literature and patents clearly identifies a growing interest in the use of *Streptomyces* spp. as biological control agents against toxigenic *Fusarium* spp., both to inhibit growth and to limit toxin accumulation (contamination). However, it is clear that for the majority of the available studies, the findings are preliminary. In most cases, a clear understanding of the role of the BCA, the identification of the molecules or mechanisms of inhibition, as well as the fungal targets are lacking \[[@B172-ijms-20-06119]\]. Moreover, most of the data are limited to laboratory in vitro experiments and lack validation in planta or in the field. The future of research on streptomycetes as biocontrol agents for *Fusarium* will need to integrate diverse expertise and may profit from new methods able to better mimic in the laboratory interactions occurring in the field \[[@B55-ijms-20-06119]\]. Novel formulation and application techniques will be needed to enable individual beneficial microbes and microbial consortia to exert their activity in a consistent manner for different crops and soils \[[@B173-ijms-20-06119]\]. For instance, one biocontrol approach to further investigate could be combining multiple strains to build consortia able to exert complementary activities \[[@B174-ijms-20-06119]\]. Indeed, understanding the ecological role, including specific interactions with other microorganisms and the host, is essential for developing effective and long-lasting approaches of biocontrol. Reaching a better understanding of microbes--*Fusarium* interactions could help to provide effective biocontrol strains among natural endophytes present in the wheat microbiome \[[@B175-ijms-20-06119]\] and within graminaceous plant rhizosphere \[[@B176-ijms-20-06119]\]. The effect of specific interactions as well as the ability to shift metabolic profiles within the same *Streptomyces* species, niche and also among individuals \[[@B177-ijms-20-06119]\], suggest that studies on the efficacy of strains should encompass a broad range of conditions mimicking the agricultural milieu \[[@B55-ijms-20-06119]\]. Appropriate fitness tests able to predict the behavior in the field are needed at the selection level. Novel BCAs or their metabolites could also be identified and produced integrating appropriate novel genome editing \[[@B178-ijms-20-06119]\] as well as adaptive evolution techniques \[[@B179-ijms-20-06119]\]. A better understanding of secondary metabolite regulation during the interaction with fungi will help to increase their discovery for agricultural purposes \[[@B53-ijms-20-06119]\]. Our analysis of the literature leads to the observation that each single paper only addresses a few aspects of the proposed criteria that would have to be evaluated in identifying effective *Streptomyces*-based BCAs. This review may serve as a proposal for future research efforts which will likely profit from an integrated analysis of the different parameters that we have identified. The increasing interest within industry, proven by the increasing number of patents that address and refer to the use of *Streptomyces* spp. to limit *Fusarium* spp. in grains, is a further indication of the potential role that this powerful group of microorganisms can play in the future of agricultural research. In conclusion, by performing a complete analysis of the literature regarding the use of *Streptomyces* spp. for the biological control of mycotoxigenic fusaria, we identified a set of parameters that we consider essential for enabling their implementation for biological and toxin contamination control. Our review suggests that streptomycetes have the potential to play a crucial role both as BCAs, and as producers of novel inhibitory molecules, for the combined control of *Fusarium* infection and to limit the accumulation of mycotoxins in crops \[[@B180-ijms-20-06119],[@B181-ijms-20-06119]\]. We thank Barrie Wilkinson for the helpful suggestions and the English revision of the manuscript. Conceptualization, M.P. and E.M.C.; Methodology, M.P. and E.M.C.X.; Validation, A.K., M.S., and P.C.; Formal Analysis, E.M.C.; Investigation, M.P. and E.M.C.; Resources, M.S. and P.C.; Data Curation, E.M.C. and A.K.; Writing---Original Draft Preparation, E.M.C. and M.P.; Writing---Review & Editing, M.P., A.K., P.C. and M.S.; Supervision, M.P. and M.S.; Funding Acquisition, M.S., P.C. and M.P. The Department of Food, Environmental and Nutritional Sciences, University of Milan partially covered the open access article processing costs. The authors declare no conflict of interest. ijms-20-06119-t001_Table 1 ###### Published studies regarding the efficacy of *Streptomyces* spp. (and derived molecules) against *Fusarium* toxigenic species in vitro, in planta, and under different environmental conditions. The methods used for the identification of the *Streptomyces* strain are also reported. Data were obtained combining the results of Scopus and Google Scholar searches with the following search words, "*Fusarium*" and *"Streptomyces*", limiting the period of publication from 2000 to 2018. Legend: M (Morphological identification), B (Biochemical identification), BCA/s (Biocontrol agent/s), GC (Growth chamber), G (Greenhouse), F (Field), \* possibly misleading identification. ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- *Fusarium* spp. Studied Streptomycete Identification In Vitro Tests for Antifungal Activity Influence of Pathogen Diversity Influence of Culture Media on BCAs In Vitro Tests Using BCA Extracts Evaluation of Antifungal Mode of Action BCAs' Survival on Plants Environment of Trials in Planta Evaluation of BCA Application BCAs' Effects on Plants BCAs' Effects on Disease BCAs' Effects on Fusaria Inoculum Toxin Measurement References ---------------------------------------------------------------------------------------------------------------- ------------------------------ ---------------------------------------- --------------------------------- ------------------------------------ ----------------------------------- ----------------------------------------- -------------------------- --------------------------------- ------------------------------- ------------------------- -------------------------- ----------------------------------- ------------------- --------------------------- *F. avenaceum* x x x \[[@B95-ijms-20-06119]\] *F. avenaceum* x x \[[@B124-ijms-20-06119]\] *F. avenaceum*, *F. oxysporum*, *F. solani* x x \[[@B125-ijms-20-06119]\] *F. coeruleum*; *Gibberella saubinetii* 16S rRNA x x x \[[@B101-ijms-20-06119]\] *F. crookwellense*; *F. oxysporum* x x \[[@B107-ijms-20-06119]\] *F. culmorum* x x x \[[@B102-ijms-20-06119]\] *F. culmorum* M/B/16S rRNA x \[[@B126-ijms-20-06119]\] *F. culmorum* x x x \[[@B99-ijms-20-06119]\] *F. culmorum* M x x GC x x \[[@B116-ijms-20-06119]\] *F. culmorum* x x GC x x \[[@B114-ijms-20-06119]\] *F. culmorum*, *F. moniliforme*, *F. sporotrichoides*, *F. graminearum*, *F. proliferatum* 16S rRNA x GC x x \[[@B91-ijms-20-06119]\] *F. culmorum, F. equiseti, F. proliferatum, F. graminearum, F. sporotrichioides, F. moniliforme, F. oxysporum* M/16S rRNA x x x \[[@B127-ijms-20-06119]\] *F. culmorum, F. graminearum, F. proliferatum, F. oxysporum* M/B/16S rRNA x x \[[@B128-ijms-20-06119]\] *F. culmorum, F. oxysporum* 16S rRNA x x x \[[@B108-ijms-20-06119]\] *F. culmorum, F. oxysporum* x x x x GC x \[[@B129-ijms-20-06119]\] *F. culmorum; F. graminearum; F. oxysporum* G x \[[@B118-ijms-20-06119]\] *F. fujikuroi* x x x \[[@B112-ijms-20-06119]\] *F. graminearum* M/B/16S rRNA x x x G x x \[[@B130-ijms-20-06119]\] *F. graminearum* x x \[[@B106-ijms-20-06119]\] *F. graminearum* x x x x \[[@B103-ijms-20-06119]\] *F. graminearum* M/B/16S rRNA x x x x x \[[@B113-ijms-20-06119]\] *F. graminearum* M/16S rRNA x x G x x x \[[@B84-ijms-20-06119]\] *F. graminearum* M/B/16S rRNA x x x \[[@B131-ijms-20-06119]\] *F. graminearum* M/B/16S rRNA x x G x x \[[@B80-ijms-20-06119]\] *F. graminearum* x F x x x \[[@B86-ijms-20-06119]\] *F. graminearum* x F x x x x \[[@B88-ijms-20-06119]\] *F. graminearum* x x x x \[[@B85-ijms-20-06119]\] *F. graminearum* G; F x \[[@B132-ijms-20-06119]\] *F. graminearum* x x x G x x \[[@B87-ijms-20-06119]\] *F. graminearum F. culmorum* 16S rRNA x \[[@B43-ijms-20-06119]\] *F. graminearum, F. culmorum, F. oxysporum* M/B x x x \[[@B133-ijms-20-06119]\] *F. graminearum, F. oxysporum* x x x x \[[@B134-ijms-20-06119]\] *F. graminearum, F. oxysporum* x x x \[[@B135-ijms-20-06119]\] *F. graminearum, F. oxysporum, F. solani* x x x \[[@B105-ijms-20-06119]\] *F. graminearum, F. proliferatum, F. sporotrichioides, F. oxysporum* x x x \[[@B109-ijms-20-06119]\] *F. graminearum, F. verticillioides, F. culmorum* M/\ x G x x \[[@B117-ijms-20-06119]\] 16S rRNA *F. graminearum, moniliforme, F. oxysporum, F. solani* M/B/16S rRNA x x x \[[@B100-ijms-20-06119]\] *F. graminearum; F. moniliforme* M/B/16S rRNA x x x x \[[@B136-ijms-20-06119]\] *F. moniliforme* M/B x x \[[@B137-ijms-20-06119]\] *F. moniliforme* x G x \[[@B120-ijms-20-06119]\] *F. moniliforme* x x \[[@B119-ijms-20-06119]\] *F. moniliforme* 16S rRNA x \[[@B89-ijms-20-06119]\] *F. moniliforme* M/16S rRNA x x \[[@B138-ijms-20-06119]\] *F. moniliforme* M/B x x x \[[@B96-ijms-20-06119]\] *F. moniliforme* M/16S rRNA x x x \[[@B93-ijms-20-06119]\] *F. moniliforme* B x x \[[@B94-ijms-20-06119]\] *F. moniliforme* M/B/16S rRNA x x x \[[@B110-ijms-20-06119]\] *F. moniliforme* x x x \[[@B111-ijms-20-06119]\] *F. moniliforme; F. oxysporum* M/B/16S rRNA x x x x x \[[@B139-ijms-20-06119]\] *F. moniliforme; F. oxysporum; F. semitectum* M/B x x x \[[@B98-ijms-20-06119]\] *F. moniliforme; F. oxysporum; F. semitectum; F. solani* M/B/16S rRNA x x x x \[[@B97-ijms-20-06119]\] *F. oxysporum \*; F. solani\** x x \[[@B123-ijms-20-06119]\] *F. poae* M/B x x x \[[@B140-ijms-20-06119]\] *F. poae* M/\ x \[[@B141-ijms-20-06119]\] 16S rRNA *F. poae,*\ x \[[@B142-ijms-20-06119]\] *F. avenaceum, F. culmorum* *F. proliferatum* M/B/16S rRNA x x \[[@B143-ijms-20-06119]\] *F. subglutinans; F. sambucinum* x \[[@B144-ijms-20-06119]\] *F. verticillioides* 16S rRNA x x x \[[@B92-ijms-20-06119]\] *F. verticillioides* x x x \[[@B104-ijms-20-06119]\] *F. verticillioides* 16S rRNA x x x x \[[@B121-ijms-20-06119]\] *F. verticillioides* x x x \[[@B122-ijms-20-06119]\] *F. verticillioides* 16S rRNA x x x \[[@B145-ijms-20-06119]\] *F. verticillioides; F. oxysporum* M/B/16S rRNA x x x \[[@B90-ijms-20-06119]\] *Fusarium graminearum, F. culmorum, F. oxysporum, F. sporotrichiella, F. moniliforme* x x F x x x \[[@B146-ijms-20-06119]\] ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ijms-20-06119-t002_Table 2 ###### Patent lists of *Streptomyces* spp. (and derived molecules) against *Fusarium* toxigenic species. Data were obtained combining the results of Espacenet and Orbit Intelligence with the following search words, "*Fusarium*" and "*Streptomyces*", limiting the period of publication from 2000 to 2018. Publication Number Publication Date Target *Fusarium* spp. Source Reference ------------------------------ ------------------ ----------------------------------------------- -------------------- --------------------------- RU2003100579 A 27/07/2004 *F. moniliforme, F. sambucinum, F. avenaceum* Espacenet \[[@B150-ijms-20-06119]\] KR100914225 B1 26/08/2009 *F. graminearum* Espacenet \[[@B151-ijms-20-06119]\] CN101698827 B; CN101698827 A 28/04/2010 *F. moniliforme* Espacenet \[[@B152-ijms-20-06119]\] CN101822272 A 8/09/2010 *F. avenaceum, F. semitectum* Orbit Intelligence \[[@B153-ijms-20-06119]\] KR101098280 23/12/2011 *F. proliferatum* Orbit Intelligence \[[@B154-ijms-20-06119]\] CN102433281 A; CN102433281 B 2/05/2012 *F. graminearum* Espacenet \[[@B155-ijms-20-06119]\] KR101211681 12/12/2012 *F. fujikuroi* Orbit Intelligence \[[@B156-ijms-20-06119]\] CN102835423 B; CN102835423 A 26/12/2012 *F. nivale, F. graminearum* Espacenet \[[@B157-ijms-20-06119]\] CN103114064 B; CN103114064 A 22/05/2013 *F. moniliforme, F. graminearum* Espacenet \[[@B158-ijms-20-06119]\] CN103820351 A; CN103820351 B 28/05/2014 *F. moniliforme, F. graminearum* Espacenet \[[@B148-ijms-20-06119]\] CN104130965 A 5/11/2014 *F. moniliforme* Espacenet \[[@B159-ijms-20-06119]\] CN104140982 A 12/11/2014 *F. moniliforme* Espacenet \[[@B160-ijms-20-06119]\] CN105060951 A 18/11/2015 *F. moniliforme* Espacenet \[[@B161-ijms-20-06119]\] EP3048890 A1 3/08/2016 *F. culmorum* Orbit Intelligence \[[@B162-ijms-20-06119]\] CN105886428 A 24/08/2016 *F. verticillioides* Espacenet \[[@B163-ijms-20-06119]\] CN106676040 17/05/2017 *F. graminearum* Orbit Intelligence \[[@B164-ijms-20-06119]\] CN107058131 18/08/2017 *F. graminearum* Orbit Intelligence \[[@B147-ijms-20-06119]\] CN107164259 A 15/09/2017 *F. culmorum* Espacenet \[[@B165-ijms-20-06119]\] CN107287130 A 24/10/2017 *F. verticillioides* Espacenet \[[@B166-ijms-20-06119]\] WO201553482 A1 16/04/2018 *F. proliferatum* Orbit Intelligence \[[@B167-ijms-20-06119]\] CN108048380 A 18/05/2018 *F. graminearum* Espacenet \[[@B168-ijms-20-06119]\] CN108102961 A 1/06/2018 *F. graminearum* Espacenet \[[@B169-ijms-20-06119]\] CN108165506 15/06/2018 *F. graminearum* Orbit Intelligence \[[@B170-ijms-20-06119]\] CN108208016 29/06/2018 *F. graminearum* Orbit Intelligence \[[@B149-ijms-20-06119]\] CN108587981 28/09/2018 *F. graminearum* Orbit Intelligence \[[@B171-ijms-20-06119]\]

Introduction {#S1} ============ Resistance training (RT) is widely recommended for subjects seeking improvements of the neuromuscular system, athletic performance, and overall health and wellness ([@B1]; [@B31]; [@B13]; [@B17]; [@B19]). Men and women of all ages display increases in muscle hypertrophy, muscle strength, and functional capacity with the performance of regimented RT ([@B3]; [@B30]; [@B26]). The practice of RT can improve flexibility ([@B29]) especially with the inclusion of stretching exercises ([@B14]); however, there is limited evidence regarding the use of stretching during RT. Several studies have investigated the acute and chronic responses of agonist muscle stretching on strength, flexibility, hypertrophy, total training volume (TTV), and metabolic stress ([@B38]; [@B14]; [@B8]; [@B22]). [@B23] evaluated the effect of antagonist (pectoralis major) muscle stretching, performed during the interset rest period, on seated row performance in resistance trained men. Antagonist muscle stretching resulted in more repetitions in the seated row and higher electromyographic (EMG) activity of the latissimus dorsi and biceps brachii as compared with a non-stretching condition. [@B32] also reported an increase in torque production of the knee extensors and vertical jump performance following static stretching of the antagonist muscles. Moreover, [@B37] demonstrated an increase in fascicle length, lateral and medial gastrocnemius muscle thickness (MT), and a decrease in penation angle following 6 weeks of moderate-intensity static stretch training. Although some stretching approaches during RT may decrease muscle performance, there is a paucity of research investigating the effects of stretching the antagonist muscles on lower limb performance during squatting in trained subjects. Squat exercise is widely used in RT programs to increase lower limb strength and muscle mass ([@B9]; [@B5]). Although the concentric phase of the squat involves hip extension, squatting does not result in significant hypertrophy of the hamstrings ([@B2]; [@B16]), conceivably due to the relative absence of change in the muscle's length-tension curve during the movement ([@B39]). It is speculated that interset stretching to a certain level of mechanical tension results in muscle hypertrophy ([@B24]), while acute, passive stretching could decrease performance during maximal contractions ([@B21]). Several measures are used to evaluate the effects of an RT session. One such measure is volume load, defined as the product of repetitions and lifted load \[number of repetitions × load (kg)\], which is considered a useful marker to estimate workload in a training session ([@B35]). Another common measure, surface EMG, is a non-invasive tool used to evaluate muscle activation ([@B10]). Finally, B-mode ultrasound is a validated method for measuring MT, which is correlated with muscle cross-sectional area, a hypertrophy parameter ([@B7]; [@B40]), as well as acutely assessing skeletal muscle cell swelling in response to a RT bout ([@B4]). Intriguingly, a recent study by [@B12] found that acute measures of muscle cell swelling were positively correlated with chronic increases in muscle hypertrophy. There is a dearth of scientific evidence as to the effects of hamstring stretching prior to performance of the bilateral squat, especially in regard to muscle activation and exercise performance. Given the aforementioned evidence that stretching may elicit muscular adaptations, this remains a gap in the literature. Thus, the objective of the present study was to evaluate the acute effects of hamstrings stretching before performance of the back squat on MT, EMG, and TTV. We hypothesized that the hamstrings stretching performed before sets of the back squat would reduce the EMG, and increase MT of the stretched muscles; and would increase agonist muscles recruitment; accompanied by a higher total training. Materials and Methods {#S2} ===================== Study Design {#S2.SS1} ------------ In an effort to evaluate the effects of hamstrings stretching on muscle performance in an RT session, subjects with previous experience in squat exercise were submitted to two acute training sessions in a randomized order: (1) with stretching before sets and (2) without stretching. All subjects were familiar with the technique and cadence used in the training sessions, which consisted of three sets of the barbell back squat performed to momentary muscle failure. Before and after each experimental session, the thicknesses of the quadriceps muscles (except for the vastus intermedius), and biceps femoris were obtained by ultrasound imaging. During squatting, the activity of the rectus femoris, vastus lateralis, vastus medialis, biceps femoris, and iliocostalis muscles was assessed by EMG. The TTV for each session was also assessed. The load for 10 repetition maximum (RM) was obtained in two separate sessions with a minimum recovery interval of 48 h (see [Figure 1](#F1){ref-type="fig"}). {#F1} Participants {#S2.SS2} ------------ Fourteen men (see [Table 1](#T1){ref-type="table"}) with ∼7.5 years of RT experience participated in this study. All subjects were habituated to perform the barbell back squat exercise, and had no history of osteomyoarticular injuries or limitations that could impair exercise performance. The study was approved by the local Research Ethics Committee (No.: 2,110,224), and followed the guidelines for data collection in human beings established in the resolution No. 466/12, of 12/12/12 by the National Health Council. All subjects were informed about the procedures, risks, and benefits of the investigation, and gave their signed informed consent prior to participation. ###### Subjects' characteristics. Variables Mean (CI) (*n* = 14) ----------------------------- ---------------------- Age (years) 28.46(25−31) Weight (kg) 80.22(74−85) Height (cm) 174(169−179) Body mass index 26.34(25−27) 10RM Squat (kg) 103(91−116) Relative 10RM Squat 1.29(1.1−1.4) Training experience (years) 7.53(3.6−11) 10RM, 10 repetition maximum. 10 Repetition Maximum (10RM) Testing {#S2.SS3} ------------------------------------ The 10RM test was performed in the squat exercise to determine the exact training load for each individual subject. The subjects previously performed a familiarization session in order to reproduce the exercise technique under the supervision of three physical education professionals. For the test, subjects performed a warm-up consisting of 8 min of stationary cycle ergometry at a light intensity, followed by a set of eight repetitions with 60% of the load equivalent to 10RM (estimated). Each subject initiated the test with the suggested load for 10RM. When the subject reached the goal of 10 repetitions, the test was interrupted, and the load increased in 5--10% increments until ascertaining the 10RM. The 10RM was determined within three attempts with a minimum interval of 5 min between attempts. A metronome was used to control the velocity of the movement, establishing a cadence of 2 s for the concentric phase, and 2 s for the eccentric phase, without interruption in the transitions between them. The amplitude of the movement was limited to a depth in which the subject's femur was parallel to the ground, demarcated with the aid of a tape. During the 10RM test attempts, verbal encouragement was provided to encourage volunteers to employ maximum effort. To ensure test reliability, the same procedure was repeated after 48 hours, with a high intraclass correlation coefficient (ICC) found between the first and second tests (*r* = 0.981; *p* \< 0.0001). Training Sessions {#S2.SS4} ----------------- For the experimental sessions, all subjects completed two resistance exercise conditions in a randomized order separated by a 7-day recovery period. Both resistance exercise conditions (stretching and non-stretching) consisted of 8 min of a general warm-up using a cycle ergometer at moderate intensity followed by one set of eight repetitions of the barbell squat with 50% of the subject's 10RM. Subjects then proceeded to complete three sets of the barbell back squat at their 10RM, maintaining a cadence of 2 s for the concentric phase and 2 s for the eccentric, with an interset rest interval of 120 s. Stretching Session {#S2.SS5} ------------------ Subjects completed a static stretch for the hamstrings just before performance of the barbell squat. A research volunteer manually provided support to the participant's shoulder blades, exerting gradual scapular pressure. All participants were required to report a perception of pain higher than 7 (range from 0 to 10) during the active rest with static stretching. The static stretching position was sustained for 40 s, in a sitting position, with the knees extended and hips flexed. Within 10 s after completion of the stretching protocol, participants performed the barbell squat (10RM to failure), then rested for 70 s and repeated the process three times. The total recovery time was 120 s, consisting of 70 s of passive rest, 40 s active rest, and 10 s of transition to performance of the barbell squat exercise. Non-stretching Session {#S2.SS6} ---------------------- For the non-stretching session, the participant completed the barbell squat (10RM to failure) followed by a passive recovery of 120 s for all three sets. Muscle Thickness Assessment {#S2.SS7} --------------------------- Before and after each training session, the thickness of the quadriceps femoris muscles (except for the vastus intermedius muscle), and biceps femoris were evaluated by the same technician by images obtained using B-mode ultrasonography (Medison SA-9900, Live 4D; Samsung Medison Co., Ltd.; Gyeonggi-Do, South Korea) with a probe of 100 mm, 10--15 MHz. Ultrasound images before RT sessions were recorded at a specific joint angle (150°), which corresponds to the approximate total excursion of the knee extension (180°) with the subject in the supine position. Before each scanning session, the subject remained at rest for 20 min. For each site of interest (see [Figure 2](#F2){ref-type="fig"}), the transducer was initially positioned transversely to visualize the greater distance between the internal border of the superior and inferior aponeuroses. The transducer then was aligned in the fascicular plane to visualize the muscle of interest on the ultrasound monitor. To facilitate the identification of the anatomical sites analyzed in each of the evaluated muscles, as well as enhancing reproducibility of the tests (pre and post intervention), the analyzed regions were demarcated with henna ink. MT was measured as the distance between the superficial aponeurosis and the deep aponeurosis (see [Figure 3](#F3){ref-type="fig"}). To ensure accuracy of the measurements, at least three images were obtained from each evaluated site. For each measure of MT, the image was frozen and a preliminary measurement was made in the ultrasound unit. This procedure was repeated two more times and, if a difference superior to 1 mm between measures was found, a fourth image was obtained. After this, the measures were saved on hard drive for posterior analysis in ImageJ 1.42 q software (National Institutes of Mental Health, United States). After obtaining measurements of the quadriceps muscles, the subject was positioned in the prone position to measure the thickness of the biceps femoris muscle. The measures followed this order: RF, VL, VM, and BF. The post-session evaluations were performed approximately 120 s after the last squat set, which equated to the time required for the participant to get positioned on the stretcher (located in the same environment in which the exercises were performed). The total time to obtain all measurements was approximately 20 min. All subjects were instructed to refrain from drinking alcoholic beverages, and practicing physical exercises in the 72 h preceding the exams. {#F2} {#F3} The measurement regions were determined on the basis of the following parameters: 1. Vastus lateralis: 35% of the thigh length; popliteal fold (0%) to the major trochanter (100%); 2. Vastus medialis: 15% of thigh length; 3. Rectus femoris: 50% of the thigh length; 4. Biceps femoris: 50% of the thigh length; popliteal fold (0%) to the major trochanter (100%). Muscle thickness measurements were quantified with the aid of image analysis software ImageJ 1.42 q (National Institutes of Mental Health, United States). Three images were obtained for each site, and the intra-class correlation value was calculated for each measure based on the pre-exercise measures. Electromyographic Activity {#S2.SS8} -------------------------- The measures of EMG activity were evaluated by surface electrodes with an interelectrode spacing of 2 cm fixed on the right side of the body at the same points used for the assessment of MT: vastus lateralis, vastus medialis, rectus femoris, and biceps femoris. EMG activity was also recorded for the iliocostalis muscle, with the electrodes attached three centimeters lateral to the L3 spinous process and the reference electrode positioned on the olecranon process of ulna. The same evaluator fixed pairs of surface electrodes Ag/AGCL (model 2223BRQ, 3M brand), on the bellies of the analyzed muscles, parallel to the striations of the respective muscle fibers, after cleaning and trichotomization of the skin surface. The maximal voluntary isometric contraction (MVIC) was used as a reference to normalize EMG signals. First, the subjects were instructed to perform maximum voluntary isometric contractions for all analyzed muscles. For the vastus lateralis, vastus medialis, and rectus femoris muscles, subjects remained with knee extension limited to 60°; for the biceps femoris, they maintained a knee flexion of 60°, and for the iliocostalis, they maintained hyperextension of the trunk. The MVIC of the quadriceps, biceps femoris, and iliocostalis muscles were obtained by manual resistance in an extensor chair, a flexor table, and a 45° hip extension machine, respectively, all of the same brand (Physicus^®^, São Paulo, Brazil). Furthermore, straps were used to limit the aforementioned angles of flexion and extension, with the exception of trunk extension, for which manual resistance was employed. Finally, two attempts of MVIC with 5 s of duration were performed at each position. The highest RMS-EMG value obtained from the two tests for MVIC was used for EMG normalization. The described positions and angles were obtained with the support of an extensor chair, a unilateral knee flexor, and a back extension free machine, respectively. The RMS-EMG values during performance of three sets of the barbell squat to momentary muscle failure were recorded and the mean of the values was normalized by RMS-EMG during MVIC as %MVIC. EMG activity was evaluated with the aid of Myotool \[Miotec^®^ eight-channel EMG, Porto Alegre/RS, Brazil\], with a sampling frequency of 1000 Hz, 2000-fold gain and bipolar surface electrodes. Data analysis was performed using the Miograph 2.0 software, with a notch filter of 60 Hz, high pass of 20 Hz, and low pass of 500 Hz. Total Training Volume {#S2.SS9} --------------------- Total training volume was calculated using the following equation: TTV = number of sets × number of repetitions × weight. The TTV was assessed for each session to compare the performance obtained by the subject during both training sessions. Data Analysis {#S2.SS10} ------------- The Shapiro--Wilk test was used to verify data normality and Levene's test was performed to check data homogeneity. All data are presented as means (M) and confidence intervals (CI). We applied an ICC, models (2,2) with values (\<0.49 poor, 0.50--0.74 moderate, 0.75--0.89 good, and \>0.90 excellent) ([@B41]; [@B18]; [@B28]) for the measurement of pre-exercise MT and 10RM tests for both conditions. TTV was compared between training sessions by an independent *t*-test. Between-session values of MT (two groups × two moments) and EMG (two groups × three moments) amplitude were compared by the use of a two-way ANOVA. Significant main effects and interactions were further analyzed with the Bonferroni's *post hoc* test. Whenever the sphericity assumption was violated, the Greenhouse-Geisser correction was employed. The partial eta squared (η^2^*~p~*) was used as a measure of effect size and to represent the proportion of variance in the data that is attributed to the independent variable. Statistical analyses were carried out using SPSS v.23.0 (SPSS, Inc., Chicago, IL, United States). A *p \< 0.05* was considered statistically significant. Results {#S3} ======= Total Training Volume {#S3.SS1} --------------------- There was no difference in TTV between conditions stretching vs non-stretching \[*F* = 0.260; (*t*) = 5, (*df* = 26); *p* \< 0.611\]. TTV stretching = 2172 kg (1848--2495); TTV non-stretching = 2278 kg (1973--2582). Intraclass Correlation Coefficient {#S3.SS2} ---------------------------------- The ICCs for pre-exercise MT were: rectus femoris (*r* = 0.998; *p* \< 0.0001); vastus lateralis (*r* = 0.999; *p* \< 0.0001); vastus medialis (*r* = 0.998; *p* \< 0.0001); femoris biceps (*r* = 0.996 *p* \< 0.0001). Muscle Thickness {#S3.SS3} ---------------- There was an effect of the exercise session in increasing MT of the rectus femoris \[*F*~(1;26)~ = 27.51; *p* \< 0.0001; η^2^*~p~* = 0.516\], vastus lateralis \[*F*~(1;26)~ = 40.97; *p* \< 0.0001; η^2^*~p~* = 0.612\], and vastus medialis \[*F*~(1;26)~ = 177.3; *p* \< 0.0001; η^2^*~p~* = 0.872\] for both groups. There was an effect of sets post exercise for the biceps femoris in the stretching condition \[*F*~(1;26)~ = 31.13; *p* \< 0.0001; η^2^*~p~* = 0.545\]. Bonferroni's *post hoc* showed that the MT increased after every session in both conditions (all *p*-values \< 0.05), except for the biceps femoris in the non-stretching condition (*p* \> 0.05). No interaction for time was observed between conditions for the rectus femoris \[*F*~(1:10)~ = 0.169; *p* \< 0.689; η^2^*~p~* = 0.017\], vastus lateralis \[*F*~(1;10)~ = 3.36; *p* \< 0.096; η^2^*~p~* = 0.252\], vastus medialis \[*F*~(1;10)~ = 5.31; *p* \< 0.054; η^2^*~p~* = 0.347\], but a difference was observed for the biceps femoris between post-stretching across time to non-stretching pre and post \[*F*~(1;10)~ = 30.54; *p* \< 0.0001; η^2^*~p~* = 0.753\] (see [Table 2](#T2){ref-type="table"}). ###### Muscle thickness analysis. Stretching session (*n* = 14) Non-stretching session (*n* = 14) --------- ------------------------------- ----------------------------------- ----------------- ------------------- RF (cm) 2.74 (2.5--2.9) 2.80 (2.6--3.0)\* 2.73 (2.5--2.9) 2.80 (2.6--3.0)\* VL (cm) 2.95 (2.7--3.1) 3.08 (2.8--3.3)\* 2.96 (2.7--3.1) 3.09 (2.8--3.3)\* VM (cm) 3.46 (3.1--3.7) 3.90 (3.6--4.2)\* 3.47 (3.2--3.8) 3.79 (3.5--4.1)\* BF (cm) 2.70 (2.4--2.9) 2.84 (2.6--3.0)\*^,^\*\* 2.69 (2.4--2.8) 2.72 (2.4--2.8) CI, confidence interval; RF, rectus femoris; VL, vastus lateralis; VM, vastus medialis; BF, biceps femoris; \*difference across time; \*\*difference across condition. Electromyography Activation {#S3.SS4} --------------------------- An effect of sets was seen in EMG for both conditions (pre vs post) on vastus lateralis \[*F*~(2;52)~ = 12.42; *p* \< 0.0001; η^2^*~p~* = 0.323\], rectus femoris \[*F*~(2;52)~ = 3.69; *p* \< 0.032; η^2^*~p~* = 0.124\], vastus medialis \[*F*~(2;52)~ = 26.25; *p* \< 0.0001; η^2^*~p~* = 0.502\]. Bonferroni's *post hoc* showed that EMG increased RF in set 2 and set 3 compared with set 1 in the stretching condition (all *p*s \< 0.05). Moreover, EMG activity increased for the vastus lateralis in set 3 compared with set 1 in both conditions (*p* \< 0.05). Also, EMG activity increased for the vastus medialis in set 3 compared with set 2 and set 1 in the stretching condition and in set 3 and set 2 compared with set 1 in the non-stretching condition (see [Table 3](#T3){ref-type="table"}). No group or time interaction was observed for the biceps femoris \[*F*~(2;52)~ = 5.29; *p* \< 0.059; η^2^*~p~* = 0.169\] and iliocostalis in any condition \[*F*~(2;52)~ = 0.876; *p* \< 0.422; η^2^*~p~* = 0.033\]. No interaction was observed for conditions vs time for the vastus lateralis \[*F*~(5;65)~ = 0.736; *p* \< 0.599; η^2^*~p~* = 0.054\], rectus femoris \[*F*~(5;65)~ = 1.44; *p* \< 0.100; η^2^*~p~* = 0.221\], vastus medialis \[*F*~(5;65)~ = 2.74; *p* \< 0.105; η^2^*~p~* = 0.174\], biceps femoris \[*F*~(5;65)~ = 0.941; *p* \< 0.461; η^2^*~p~* = 0.067\], and iliocostalis \[*F*~(5;65)~ = 2.13; *p* \< 0.072; η^2^*~p~* = 0.141\]. ###### Electromyographic analysis adjusted for the percent of maximum voluntary isometric contraction (%MVIC). Stretching session (*n* = 14) Non-stretching session (*n* = 14) ------- ------------------------------- ----------------------------------- -------------------- --------------- --------------- --------------- ------------------ ------------------ --------------- --------------- Set 1 47.9 (33--62) 58.6 (45--71) 62.0 (52--71) 21.3 (16--26) 65.5 (50--80) 48.2 (34--61) 58.9 (45--71) 59.2 (49--68) 22.2 (17--27) 61.4 (46--76) Set 2 52.6 (36--68)^A^ 62.6 (48--76) 66.3 (56--76)^A^ 21.9 (16--26) 68.9 (54--83) 49.5 (35--63) 60.5 (46--74) 63.2 (53--73)^A^ 23.7 (18--28) 61.1 (46--76) Set 3 52.1 (38--65)^B^ 66.0 (53--79)^B^ 71.1 (59--82)^B,C^ 22.6 (17--27) 70.6 (55--85) 49.4 (37--61) 64.0 (51--77)^B^ 66.3 (55--77)^B^ 24.2 (19--29) 61.9 (47--76) Note: RF, rectus femoris; VL, vastus lateralis; VM, vastus medialis; BF, biceps femoris; IC, iliocostalis; CI, confidence interval; Different letters mean significant difference between sets within conditions ( p \< 0.05). A Set 2 difference for Set 1; B Set 3 for set 1; C Set 3 for Set 2. Discussion {#S4} ========== This study assessed the influence of static stretching immediately before the squat exercise on TTV, electromyographic activity, and thickness of primary and synergistic skeletal muscles. The main results from the present study partially confirmed our initial hypothesis in that hamstrings stretching before sets of the barbell back squat resulted in an increased biceps femoris thickness, not observed in the condition without stretching. Squatting with or without stretching increased vastus lateralis, vastus medialis, and rectus femoris thickness, without differences in TTV. All muscles displayed an increase in EMG, regardless of whether stretching was employed. These results highlight the possibility that stretching the hamstrings prior to the bilateral squat does not impair performance and may promote a positive stimulus to the hamstring muscles. This strategy is both time-efficient and easily applied during resistance exercise sessions. The acute increase in biceps femoris thickness in the stretching condition may be attributed to an increased hyperemia resultant to stretch-induced blood flow restriction and consequent metabolite accumulation, factors that promote muscle cell swelling (e.g., "the pump") ([@B34]). This phenomenon might be contributing stimuli to muscle growth ([@B12]). Moreover, an association exists between the acute "muscle pump" and integrin activation; a membrane protein that can trigger muscle protein synthesis and thus enhance hypertrophic adaptations ([@B42]). However, despite the correlation between acute and chronic changes in MT ([@B44], [@B43]), it remains to be determined if the increase in biceps femoris thickness found in the present study would result in a chronic increase of muscle mass. In support of the hypothesis, 6 weeks of a static stretching of the triceps surae increased 5.6% the lateral and medial gastrocnemius thickness in detrained subjects ([@B37]). The RT session of the barbell squat without stretching did not increase MT of the biceps femoris. This was expected due to the limited contribution of the hamstrings during bilateral squatting ([@B6]; [@B2]; [@B16]). Muscle length of the biceps femoris remains relatively unaltered during this movement given its dual role in extending the hip and flexing the knee, and thus is not sufficiently stimulated to promote a significant increase in MT ([@B39]). The MT of all quadriceps muscles increased in response to squat training, regardless of condition, probably as a result of the muscle pump mechanisms mentioned earlier ([@B34]; [@B11]). The acute increase in RF thickness in both resistance exercise conditions may be attributed to a swelling effect after performance of the bilateral squat. Although the length-tension curve of the RF is not significantly modified during the bilateral squat, this should not be confused as an absence of recruitment, as opposed to the higher activation of the vastus muscles ([@B39]). Importantly, attempting to extrapolate acute measures to chronic adaptations should be done cautiously as only modest chronic hypertrophic responses of the RF are observed in the back squat regardless of the intensity, speed, and range of motion utilized ([@B9]; [@B5]; [@B16]). Stretching induces alterations in the recruitment pattern of primary muscles and antagonists ([@B23]). Thus, stretching the hamstrings prior to performing the back squat may interfere with quadriceps muscle behavior ([@B32]). Our data revealed that despite the absence of differences between conditions (stretching and non-stretching), EMG amplitude of the rectus femoris increased only in the stretching session. We observed increased EMG activity of the rectus femoris in the second and third sets of barbell squat performed after the stretching of the hamstring. Conceivably, this can be attributed to the synergistic relationship of this muscle with the elongated hip extensors. During squatting, the rectus femoris acts as a transmission arm of the force generated by the hip extensors, especially the gluteus maximus, to the knee joint ([@B15]). Therefore, as the hip extensors are elongated, the rectus femoris begins to withstand greater demands to ensure the maintenance of exercise performance during the supposed reduction in the action of synergic-antagonist hip extensors. A proposed mechanism for this phenomenon is related to mechanical adaptations, such as a gradual decrease in torque production by the hip extensor muscles, as they were stretched before the sets of squat; passively due to viscoelastic stress relaxation, passive torque, and muscle stiffness reduction ([@B36]). Alternatively, the torque might actively decrease as consequence of decreased recruitment or reflex sensitivity ([@B25]) this seems less likely for the biceps femoris, as EMG activity was not decreased even in the stretching condition. This hypothesis remains speculative. Pre-exercise stretching of the hamstrings did not alter TTV during squatting. Both [@B27] and [@B23] reported an increase in total repetitions completed in the seated row when the pectoralis muscle was stretched between sets. Among the explanatory mechanisms, the authors hypothesized that stretching of the antagonists could decrease coactivation of these muscles, increase the storage of elastic energy in the agonists, and promote alterations in the acute sensitivity of specific muscle proprioceptors (Golgi tendon organs and muscle spindles), thereby increasing performance. However, this would not seem to apply to our findings, as the hamstrings can be regarded as synergists during the bilateral squat exercise ([@B6]). Thus, the stretching did not interfere with biceps femoris EMG during squatting, which may have resulted in the absence of changes in iliocostalis activation between conditions (see [Table 2](#T2){ref-type="table"}). Thus, it can be inferred that the acute increase in MT observed in the biceps femoris is not directly associated with the magnitude of EMG amplitude during the stretching session ([@B39]). The EMG represents the neural drive to the muscle ([@B10]) while MT may be associated with metabolite accumulation produced during the stress imposed by the exercise ([@B33]). The present study is not without limitations. First, the acute nature of the thickness measures may not reflect chronic adaptations. Second, measures of MT were taken from a specific muscle site, and it is thus unclear as to whether differential results may have occurred along the length of the muscle belly. Third, although MT measures (post-exercise) were taken immediately after the last set of squat (\<120 s), following the same order (RF, VL, VM, and BF), it is necessary to recognize that the time between the end of the sessions and the measures might interfere in determining acute MT values ([@B12]). Finally, the stretching technique chosen for the present study may not be the most efficient for targeting the hamstrings. Notably, RF, VL, and VM muscles originate from the sciatic tuberosity, and the increased tension during stretching can generate a posterior inclination of the pelvis, which in turn can reduce the intensity of stretching. Moreover, individuals with severe hamstring shortening may increase flexion in the thoracic spine as a compensatory strategy to facilitate performance ([@B20]). Other stretching techniques potentially could produce disparate findings. Future studies should endeavor to investigate EMG and MT in the gluteus maximus and other hamstring muscles, in conjunction with different stretching strategies. Conclusion {#S5} ========== This study suggests that the stretching of the hamstrings immediately before each set of the bilateral squat can be used to optimize time during a resistance exercise session with the inclusion of flexibility exercises, without impairment of exercise performance. Moreover, the stretching of the hamstrings immediately before each set of the squat may be used to acutely increase biceps femoris thickness. It remains to be determined if the acute stretching-induced increase in biceps femoris thickness enhances muscle hypertrophy when performed chronically over time. Data Availability Statement {#S6} =========================== The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation. Ethics Statement {#S7} ================ The studies involving human participants were reviewed and approved by the Research Ethics Committee of Federal University of Rio Grande do Norte, Brazil. The patients/participants provided their written informed consent to participate in this study. Author Contributions {#S8} ==================== All authors contributed significantly to this study and read and approved the submitted manuscript. Conflict of Interest {#conf1} ==================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. [^1]: Edited by: Emiliano Cè, University of Milan, Italy [^2]: Reviewed by: Ryoichi Ema, Shizuoka Sangyo University, Japan; Cassio V. Ruas, Edith Cowan University, Australia [^3]: This article was submitted to Exercise Physiology, a section of the journal Frontiers in Physiology

{#sp1 .17} {#sp2 .18} {#sp3 .19} {#sp4 .20} {#sp5 .21} {#sp6 .22} {#sp7 .23} {#sp8 .24} {#sp9 .24-1} {#sp10 .24-2} {#sp11 .25} {#sp12 .26} {#sp13 .27} {#sp14 .28} {#sp15 .29} {#sp16 .30} {#sp17 .31} {#sp18 .32} {#sp19 .32-1} {#sp20 .33} {#sp21 .34} {#sp22 .35} {#sp23 .36} {#sp24 .37} {#sp25 .38} {#sp26 .39} {#sp27 .40} {#f1 .24-1} {#f2 .24-1} {#f3 .24-1} {#f4 .24-1} {#f5 .24-2} {#f6 .24-2} {#f7 .32-1} {#f8 .32-1} {#f9 .32-1} [^1]: (By Invitation.) [^2]: Read at a Meeting of the Edinburgh Obstetrical Society, 13th December 1933.

Introduction {#sec1} ============ Rapid technological and industrial developments continuously result in the emission of hazardous gases, toxins, and biomolecules. Therefore, sensing of such undesirable chemical or biochemical forms has become a significant research endeavor in recent years.^[@ref1]−[@ref3]^ Certain gases such as NH~3~, NO~*x*~, and CO are contributed to the atmosphere regularly in significant quantities by transportation and chemical industries, which have a deteriorating influence on the environment in various ways.^[@ref4]^ Issues of safety and security also hinge upon detection of trace elements of chemical agents. Some of these pollutants are extremely dangerous, and hence sensing them at ppm levels is crucial for the harmonious survival of mankind as well as the ecosystem at large. Different materials and detection schemes have been adopted in this context to achieve high responsivity combined with less response/recovery time, which are key quality factors that define the sensor performance. In particular, oxides are extensively researched for gas sensing, in view of their robust material properties and their ability to change valence through charge transfer.^[@ref5]−[@ref7]^ Several binary oxides have been examined for effective gas sensing and have yielded promising results.^[@ref5]−[@ref13]^ The interaction of analyte gas molecules depends on the surface atoms present in the metal oxides. The catalytic activity of metal oxides depends on the relative acidic and basic character of the atoms present on the surface which is affected by coordination between the metal atom and oxygen anions. In the case of binary metal oxides, the accessible property regimes in this respect are rather limited. On the other hand, in the case of ternary metal oxides, there are two acidic sites and one basic site; hence, the multitude of their interactions renders wider property windows for the catalytic performance. However, the ternary oxide and sulfide systems, which offer distinct advantages of stoichiometry and valence control, are relatively less explored by far.^[@ref14]−[@ref27]^ Herein, we report the sensing properties of an interesting ternary oxide CuCo~2~O~4~ (CCO) which comprises two earth abundant transition elements, each having an ability to support multiple valence states (Cu 1+ and 2+ and Co 2+ and 3+).^[@ref28]−[@ref30]^ Although the sensing performance of CCO at a high temperature of 300 °C is already reported,^[@ref26]^ its effective gas-sensing response at room temperature is hitherto unexplored. In the present study, we have successfully shown that a controlled stoichiometry manipulation and porous nanoplatelet morphology of CCO were responsible for superior gas-sensing activity for ammonia gas at room temperature. Results and Discussion {#sec2} ====================== Characterization of the Gas-Sensing Material {#sec2.1} -------------------------------------------- A detailed description of the synthesis and characterization of the three different materials, viz., CCO, CCO--Cu(10), and CCO--Co(10) is reported in our previous publication.^[@ref28]^ It was envisaged to achieve porous nanoplatelet morphology of the material that would facilitate effective gas adsorption on account of the enhanced surface area for higher surface accessibility, the proper range of conductivity, and higher density of surface defects. Although there are numerous methods to obtain a single-phase CCO starting with a stoichiometric mixture of precursors; however, these do not yield the appealing morphology of interest required for different applications such as battery electrodes and sensor elements.^[@ref26],[@ref28]−[@ref34]^ In the current study, it was observed that the hydrothermal synthesis route using a stoichiometric mixture of precursors resulted in a product with the desired morphology but with an undesirable secondary phase. As seen in [Figure S2a](http://pubs.acs.org/doi/suppl/10.1021/acsomega.7b01958/suppl_file/ao7b01958_si_001.pdf), the powder product consists of a copper-deficient defect-spinel phase of CCO along with a minor secondary phase of CuO. Hence, to suppress this secondary phase, the synthesis was performed with an off-stoichiometric precursor mixture with excess cobalt. It was found that 10 at. % excess cobalt in the precursor mixture leads to a single-phase sample with almost a negligible secondary CuO phase. It means that the nearly single-phase sample formed has excess cobalt, so by implication, it is also a defect spinel but a single phase. Separately, we also examined the case of excess copper added to the precursor to enhance the CuO component so as to obtain a CuO/defect--CCO composite in the context of another application interest. This gave us an opportunity to examine all these three materials, which represent interesting sample states in terms of the phase constitution, defects states, and biphasic versus single phase nature for the sensing application. In the following section, we give some relevant characterization results performed on the new set of samples used specifically in this work. [Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}a--c shows the field emission scanning electron microscopy (FESEM) images of the three cases of interest, namely CCO, CCO--Cu(10), and CCO--Co(10). The morphology generally appears to be in the form of flower-like nanoplatelets which cross one another. Thus, the off-stoichiometry does not make major changes in the morphology. The high-resolution transmission electron microscopy (HRTEM) data of [Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}d,e reveal that the platelets are assembled through nanoparticles of a rather uniform size of about 10--15 nm via oriented attachment as suggested by their planar assembly with nanoporosity throughout. [Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}f shows lattice fringes with a *d*-spacing of 0.28 nm, which matches well with the (220) plane of CCO. Thus, the nanoporous nature of these nanoplatelets is essential for the sensing application because it provides better access and high surface area for the adsorption of the analyte gas. {#fig1} Sensor Response {#sec2.2} --------------- First, the gas-sensing behavior of the CCO--Co(10) material for different gases at room temperature was investigated, and the resultant data are shown in [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}a. The exposure time for the analytic gas was kept fixed at 3 min. The CCO--Co(10) powder coating used for gas sensing was drop casted on a glass substrate, and corresponding X-ray diffraction (XRD) is shown in [Figure S2b](http://pubs.acs.org/doi/suppl/10.1021/acsomega.7b01958/suppl_file/ao7b01958_si_001.pdf). Basically, it represents the same XRD as that of the CCO--Co(10) powder material. It was found that the material senses all the gases, but it shows recovery only for the case of NH~3~. In the presence of reducing gases such as NH~3~, H~2~S, and CO, the resistance was increased, whereas in the presence of oxidizing gas (such as NO~2~), there was a reduction in the resistance of CCO--Co(10). This confirms the p-type nature of CCO Co(10).^[@ref34]−[@ref36]^ As reported above, CCO with their nanotube morphology was implemented for gas sensing at a high temperature of 300 °C for a few target gases.^[@ref26]^ Although, it is clear that the other gases interact nonreversibly (at least over the time scale examined) with the sample ([Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}); however, the novelty of our work lies in the fact that gas sensing is being carried out at room temperature. Because both the elements in this ternary system, viz., Co and Cu, exist in two valence states, it is not entirely surprising that the material will respond to reducing as well as oxidizing gases. The recovery being an important consideration in a sensing application, further work was mainly focused on NH~3~ sensing. The statements made here about the sensing of different gases are broadly valid for CCO and other off-stoichiometric cases also. {#fig2} [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}b shows NH~3~ gas-sensing properties of all three composites in dry air as carrier gas at 23 °C. It can be seen from [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}a that at room temperature, the CCO--Co(10) sample (single-phase defect spinel) showed a much superior response toward NH~3~ gas as compared to the CCO and CCO--Cu(10) biphasic samples. The CCO--Co(10) exhibits gas response of ∼6% at 400 ppm of NH~3~. There could be three key reasons for this observation of superior performance by a single-phase material as compared to the other two cases, which are in a composite state. First is the potential competition for gas adsorption sites between the two phases depending on the adsorption chemistry; second is the potential siphoning of active carriers to the interface traps between the two constituent phases making them ineffective for transport. Last is the transfer of the carrier to the secondary phase rendering them unavailable for transport because of percolation issues. Further work will be needed to elucidate and separate out these issues. Therefore, on the basis of the high sensitivity of CCO--Co(10) to NH~3~ at room temperature (23 °C), further detailed gas-sensing studies were performed on this CCO--Co(10) material. For practical application, it is necessary to determine gas-sensing properties of the sensor at room temperature, and for this condition, the sensor performance is greatly influenced by the surrounding humidity as well. Hence, measurements were performed in a practically typical relative humidity condition (∼57% RH). [Figure [3](#fig3){ref-type="fig"}](#fig3){ref-type="fig"}a shows NH~3~-sensing behavior of CCO--Co(10) in the presence of 57% RH and dry air at 23 °C. It can be seen that the sensor exhibits enhanced response in the presence of 57% RH as compared to dry air. Importantly, the recovery time is also reduced in the humid atmosphere (14 min compared to 18 min in dry air). The enhanced response in the humid environment suggests that in the presence of humidity, NH~3~ reacts with H~2~O forming NH^4+^ and OH^--^, which facilitates the interaction with the sensor.^[@ref37]^ Therefore, concentration-dependent NH~3~ response, cyclic reproducibility, and also the device stability over the period were carried out in 57% RH at room temperature. {#fig3} [Figure [3](#fig3){ref-type="fig"}](#fig3){ref-type="fig"}b shows the response of CCO--Co(10) over the concentration range from 25 to 400 ppm of NH~3~ in 57% RH at room temperature. Also, the response as a function of NH~3~ gas concentration was noted to be linear up to about 100 ppm suggesting nonsaturation of the active sites over this full range ([Figure [3](#fig3){ref-type="fig"}](#fig3){ref-type="fig"}c). At higher concentrations, however, saturation is noted from the departure from linearity. [Figure [3](#fig3){ref-type="fig"}](#fig3){ref-type="fig"}d shows the performance repeatability of CCO--Co(10) exposed to 400 ppm NH~3~ in 57% RH over seven cycles. A small decrease in the response is noted after one cycle, which can be attributed to the leftover analyte gas on the surface of the material from the previous cycle. However, after the first cycle, the device exhibits quite a stable gas response. We have also investigated the stability of CCO--Co(10) over a period of 15 days ([Figure [3](#fig3){ref-type="fig"}](#fig3){ref-type="fig"}e) and observed that there is a slight gradual drop in the response over the first couple of days, but the response remains relatively constant over the period of 15 days. This suggests a robust gas-sensing performance of the CCO--Co(10) sensor at room temperature in a humid atmosphere (57% RH). [Table S1](http://pubs.acs.org/doi/suppl/10.1021/acsomega.7b01958/suppl_file/ao7b01958_si_001.pdf) compares the response of CCO--Co(10) to NH~3~ with other metal oxide semiconductors. It is evident that CCO--Co(10) renders fairly good gas response values with fast response and lower optimal operating temperature comparable or better than other sensing materials. In general, the sensing properties of metal oxide-based gas sensors depend on many factors such as surface area, microstructures, defect levels, and so forth. The detailed gas-sensing mechanism is incorporated in the [Supporting Information](http://pubs.acs.org/doi/suppl/10.1021/acsomega.7b01958/suppl_file/ao7b01958_si_001.pdf). [Figure S4](http://pubs.acs.org/doi/suppl/10.1021/acsomega.7b01958/suppl_file/ao7b01958_si_001.pdf) shows schematic illustration of the gas-sensing mechanism in the presence of reducing and oxidizing gases. To systematically understand the observed gas-sensing data at room temperature, it is essential to discuss the microstates of the three samples examined, namely CCO, CCO--Co(10), and CCO--Cu(10). As discussed in our previous paper at some length, under the specific hydrothermal growth conditions used to achieve the peculiar platelet-like morphology for the benefit of higher surface area and accessibility, the use of the stoichiometric mixture of precursors leads to a mixed phase; wherein, a small quantity of CuO is present as a secondary phase in addition to the main CCO phase, which is naturally Cu-deficient.^[@ref28]^ Other reports have also discussed the occurrence of the secondary CuO phase.^[@ref31]−[@ref34],[@ref38],[@ref39]^ When an excess cobalt precursor is included in the synthesis vessel, the contribution of the secondary CuO phase diminishes, and at about 10% excess Co, an almost single-phase sample is realized. It must be borne in mind, however, that this single phase is also a defect-spinel phase because it is richer in cobalt than the basic CCO stoichiometry. It is possible that in this case the charge balance will be enforced via a change in the oxygen stoichiometry. Separately, when excess Cu is added in the precursor mixture, it can play a dual role. It could simply add to the separated CuO secondary component or partially eliminate the Cu vacancies. This could again lead to a defect-stabilized CCO phase with a different density of defect types as compared to the other two cases. Thus, it is hard to precisely compare the three cases, although some compelling arguments can be made, particularly in the context of the observed much stronger response of CCO--Co(10) to ammonia as compared to the other two biphasic cases. The Brunauer--Emmett--Teller (BET) surface area measurements were performed on the three samples to examine whether the initial precursor stoichiometry differences affect the surface area significantly, which in turn influences the sensor performance ([Figure S3](http://pubs.acs.org/doi/suppl/10.1021/acsomega.7b01958/suppl_file/ao7b01958_si_001.pdf)). It can be seen that the surface area of CCO, CCO--Cu(10), and CCO--Co(10) was 45, 50, and 55 m^2^/g, respectively; and hence, they do not differ that significantly. However, there will be a little contribution of increased surface area on gas response in CCO--Co(10). To further establish the chemiresistive sensing results, the powders were pelletized to study their bulk electrical conductivity. It was observed that the conductivity was lowest for the CCO--Co(10) case, followed by the CCO--Cu(10) case, and then the CCO case. Because the three-dimensional percolation threshold for the secondary phase to make a major influence on the transport is 18%, in both the cases of CCO and CCO--Co(10), the secondary CuO phase is nonpercolating. Thus, the transport is primarily dominated by the defect-spinel CCO component, and the differences in the defect-borne electronic states in the three cases are responsible for the observed resistivity data. The conductivity (resistivity) is lowest (highest) for the CCO--Co(10) case as compared to those of the other two cases involving the presence of the secondary CuO phase in the form of nonpercolating distributed nanoscale regions. It suggests that this secondary phase may have a role in reducing the interface trap density and hence scattering. The changes in conductivity (resistivity) of the three samples are, however, not as significant as the ammonia sensitivity difference reflected by the CCO--Co(10) case sample vis-a-vis the other two cases. Thus, the search for the origin of this effect has to be sought in another argument. In the two cases of CCO and CCO--Cu(10), wherein the secondary CuO phase is present, it would presumably occupy some grain boundary space and will also produce heterointerfaces. When the sensing gas (NH~3~) is introduced, it will be adsorbed on the surfaces of both the major CCO defect-spinel phase and the secondary CuO phase. The electron donated by the sensing gas will influence the conductivity of the material, which will be decided by the possibility of the transfer of this electron to the major phase because the secondary phase cannot by itself percolate. Our observation of much lower response to ammonia for the biphasic cases implies that there is a hindrance to the carrier transfer process between the major and secondary phases. This hindrance could occur by the band alignment and/or presence of interface trap states. The observed low response in the biphasic composites as compared to the CCO--Co(10) case could also be explained based on the hard soft acid base (HSAB) theory. In these composites, copper is in the form of Cu^2+^, which is a borderline acid; however, cobalt is in the form of Co^3+^, which is a hard acid. The reducing gas NH~3~ is a hard base. According to HSAB theory, hard acid prefers hard base; therefore, in the present case, the Co^3+^ site is comparatively active toward the NH~3~ gas compared to Cu^2+^. As discussed earlier, in the CCO case, CuO emerges as the impurity phase, and the resulting surface contribution of this weak sensing phase will be counterproductive for sensor performance. In the CCO--Cu(10) case, this problem may get accentuated further. In the case of the nanosystem at hand, the unavailability of information regarding the precise defect and interface states precludes a full understanding of the observed effects at this time. However, a message emerges that use of nanocomposites, formed intentionally or unintentionally, especially in the nonpercolating regime for the secondary phase, involves complex considerations and cannot always ensure an enhanced response. Conclusions {#sec3} =========== A ternary oxide, namely CCO, with two multivalent transition elements, is examined for its chemiresistive properties. Unique nanoplatelet-type morphology is realized by the specific choice of the synthetic protocol, but it intrinsically leads to a biphasic state (CuO as a secondary phase in addition to the defect-spinel Cu~1--*x*~Co~2~O~4~). Employing a degree of cation off-stoichiometry, the constitution of the sample is controlled, and the sensing response is found to be modified accordingly. Three cases, viz., 10 at. % excess Co with CCO, 10 at. % excess Cu with CCO, and the right stoichiometric precursor ratios are examined. In particular, a Co 10 at. % excess CCO (CCO--Co(10)) case exhibits a good response (∼7.9% at 400 ppm) for NH~3~ gas with a complete recovery in 57% RH at 23 °C. Although the material exhibited a good response (without any recovery) for H~2~S and NO~2~ gases, the sensing performance was found to be poor in the case of CO. These observations are attributed primarily to the differences in the character of the carrier transport between the cases of the single-phase defect-stabilized material versus the nonpercolating biphasic material. Experimental Section {#sec4} ==================== Material Synthesis {#sec4.1} ------------------ The CCO sample was synthesized by a simple hydrothermal route. All the chemicals used in the present study were of analytical grade. The Cu(NO~3~)~2~·3H~2~O (\>99%) powder was obtained from Merck, whereas Co(NO~3~)~2~·6H~2~O (\>99%) and urea were obtained from LabChem. Initially, copper and cobalt precursors were mixed in deionized (DI) water in 1:2 molar ratio followed by the addition of 0.1 M urea to the solution. This was allowed to stir for 30 min under ambient conditions. The solution thus obtained was transferred into a 160 mL Teflon-lined autoclave and was heated at 120 °C for 16 h. The resultant black-colored solution was washed with DI water and then dried at 80 °C for 6 h. The final product was annealed at 350 °C for 2 h to obtain the desired phase and constitution of CCO. The same procedure was followed for preparing CCO samples with different cation stoichiometry ratios of copper and cobalt in the precursor mixture, namely copper excess case of CCO--Cu (10 at. % excess) and cobalt excess case of CCO--Co (10 at. %. excess). Characterization {#sec4.2} ---------------- The required phase formation of the material was elucidated from powder XRD measurements that were carried out on a Philips X'Pert PRO diffractometer with nickel-filtered Cu Kα radiation (λ = 1.5418 Å). The diffractograms were recorded at a scanning rate of 1° min^--1^ between 10° and 80°. The morphology of the material was established using a high-resolution field emission Nova NanoSEM system. HRTEM (FEI Tecnai G2, F30 TEM microscope operating at an accelerating voltage of 300 kV) was also used for atomistic microstructural elucidation. The gas adsorption experiment (up to 1 bar) was performed on Quantachrome Autosorb-automated gas sorption analyzer. Gas-Sensing Device Fabrication and Measurements {#sec4.3} ----------------------------------------------- The gas-sensing setup is depicted in [Figure S1a,b](http://pubs.acs.org/doi/suppl/10.1021/acsomega.7b01958/suppl_file/ao7b01958_si_001.pdf). An interdigitated (ID) pattern was developed on indium tin oxide (ITO)-coated glass with a CO~2~ laser scriber. A fine paste was made by grinding the as-synthesized CCO powder along with a polyvinylidene fluoride binder and 2-propanol as a solvent. This paste was deposited on the ID-patterned ITO glass by doctor blading and maintained at 150 °C for 2 h. The film was mounted on the alumina plate and placed inside a chamber. Constant dc voltage of 2 V was applied across the film and current was measured by a Keithley 2400 source meter. We used cylinders of test gases with 0.1% of test gas balanced in nitrogen. The desired concentration of the test gas was obtained by controlling mass flow rates of the specific test gas and dry air. The gas-sensing measurements were carried out at 250 sccm of dry air. The gas response was measured at 23 °C in dry air. Because at room temperature the humidity can also affect the gas-sensing performance, the gas-sensing measurements were also carried out in 57% RH at 23 °C. The gas response (*S*) of the p-type sensor is defined as the ratio of the resistance in the presence of gas (*R*~g~) to the resistance in the presence of air (*R*~a~).where, *R*~a~ and *R*~g~ are the resistance of the sensor in air and gas, respectively. Relative percentage response is defined as % *S* = \|*R*~g~ -- *R*~a~\|/*R*~a~ × 100. Other important parameters such as response time and recovery time were also determined from the gas-sensing experiments. It is useful to mention here that it was difficult to achieve a perfect steady state in our case, and hence to control the experimental time, we had to accept a small drift in current before letting the sensing gas in for the measurement. Therefore, the response data are not included for extremely tiny concentrations of gas (e.g., 10 ppm), for which there is a chance of a small degree of ambiguity regarding the precise value of response. The Supporting Information is available free of charge on the [ACS Publications website](http://pubs.acs.org) at DOI: [10.1021/acsomega.7b01958](http://pubs.acs.org/doi/abs/10.1021/acsomega.7b01958).Schematic of the gas-sensing setup; XRD data; BET graphs and pore size distribution of CCO, CCO--Co(10), and CCO--Cu(10); comparative analysis of gas-sensing response of different materials; and discussion and schematic of the gas-sensing mechanism ([PDF](http://pubs.acs.org/doi/suppl/10.1021/acsomega.7b01958/suppl_file/ao7b01958_si_001.pdf)) Supplementary Material ====================== ###### ao7b01958_si_001.pdf ^†^ Equal contribution (S.J. and A.P.). The authors declare no competing financial interest. S.S.B. would like to thank SERB-DST for national postdoctoral fellowship. The authors would like to thank the funding support for this work by the Board of Research in Nuclear Sciences (BRNS), Department of Atomic Energy (DAE), Government of India, and the DST Nanomission (Thematic Unit).

Research in contextEvidence before this studyCircadian rhythm controls behavior, physiology, and mood, and makes them adapt to 24 h day-night periodicity. Dysfunction of circadian clock genes leads to a variety of disorders, such as metabolic syndrome, obesity, diabetes, autoimmunity, and even cancer.Gastric cancer (GC) is the fourth most common cancer worldwide and has a high mortality rate. Most gastric adenocarcinomas start from chronic superficial gastritis, and finally evolve to dysplasia and adenocarcinoma. Gastritis paves the way for malignant transformation of gastric mucosa. Helicobacter pylori (*H. pylori*) was recognized as a type I carcinogen. Several studies uncovered the connection between circadian rhythm, *H. pylori*, and inflammatory diseases. However, the role of circadian rhythm in disease progression induced by *H. pylori* remains elusive.Added value of this studyThrough *in vivo* and *in vitro* experiments, we for the first time found the pathogenic relationship between *H. pylori* infection and circadian rhythm gene BMAL1. And we explored one of the specific pathways by which circadian rhythm genes mediated inflammatory responses induced by *H. pylori*.Implications of all the available evidenceTherefore, the rhythmic gene BMAL1 may serve as a potential diagnostic marker and therapeutic target for the early diagnosis and treatment of diseases related to *H. pylori* infection.Alt-text: Unlabelled Box 1. Introduction {#s0020} =============== Circadian rhythm controls behavior, physiology, and mood, making them adapt to 24 h day-night periodicity. The external daily stimuli regulate the circadian synchronization. Circadian rhythm is determined by the expression of the circadian clock genes which control approximately 10% of all genes \[[@bb0005]\]. Circadian clock genes consist of 3 basic helix-loop-helix/PAS domain containing proteins: histone acetyltransferase CLOCK, transcription factors BMAL1 and NPAS2. These proteins can be assembled as transcriptional complex in heterodimer form, CLOCK/BMAL1 or NPAS2/BMAL1. The complex binds to the *E*-box elements of targeting gene promoters, activating transcription of the period genes (PER1, PER2 and PER3), cryptochrome genes (CRY1 and CRY2), and other core clock genes \[[@bb0010]\]. Dysfunction of circadian clock genes leads to a variety of disorders, such as metabolic syndrome, obesity, diabetes, autoimmunity, and even cancer \[[@bb0015]\]. Gastric cancer (GC) is the fourth most common cancer worldwide with a high mortality rate \[[@bb0020]\]. Most gastric adenocarcinomas start from chronic superficial gastritis, and finally evolve to dysplasia or adenocarcinoma \[[@bb0025]\]. Gastritis paves the way for malignant transformation of gastric mucosa. There is a bacterial community with hundreds of phylotypes in the stomach \[[@bb0030],[@bb0035]\]. However the pH is less than four in the stomach cavity, making it difficult for microbiota to grow excessively under such condition \[[@bb0040]\]. Helicobacter pylori (*H. pylori*) infection leads to decreased acid secretion, which may favor different gastric bacteria to grow \[[@bb0045]\]. In the infected stomach, *H. pylori* accounts for 93% to 97% of the microbiota, and microbial diversity is significantly reduced \[[@bb0035]\]. This implies that *H. pylori* infection triggers a series of abnormal reactions, and contributes most to the gastric diseases. In addition, the global epidemiological distribution of chronic gastritis overlaps that of *H. pylori* infection despite different causes of the diseases \[[@bb0050], [@bb0055], [@bb0060]\]. *H. pylori* has been recognized as a type I carcinogen since 1994, and epidemiological studies have shown its role to increase the risk of GC. Therefore, we focus on the pathogenic effects of *H. pylori* in this study. Two independent groups showed that individual gastrointestinal cells transcribed and translated circadian clock molecules with circadian periodicity in a cell-autonomous manner, out of phase with the SCN \[[@bb0065]\], indicating the influence of local factors on circadian rhythm. Interestingly, *H. pylori* showed effects on circadian gastric acidity which implied the potential relationship between circadian rhythm and *H. pylori* induced disorders \[[@bb0070]\]. Several other studies uncovered the connection between circadian rhythm and inflammatory diseases \[[@bb0075], [@bb0080], [@bb0085], [@bb0090], [@bb0095], [@bb0100], [@bb0105]\]. Therefore, we assumed that circadian rhythm may be crucial for the development of inflammatory disease induced by *H. pylori*. *H. pylori-*induced gastric diseases can be activated by inflammatory cytokines, such as IL-1, TNF-α, IL-6, and IL-8 \[[@bb0110],[@bb0115]\]. TNF-α, IL-6, and IL-8 are the well-known inflammatory factors among the cytokines secreted by gastric epithelial cells, and they can be induced by *H. pylori* \[[@bb0120],[@bb0125]\]. In addition, dysfunction of CRY1 and CRY2 accelerated the progression of arthritis by promoting the secretion of TNF-α \[[@bb0130]\]. Therefore, circadian rhythm change, especially the expression of core circadian clock genes may determine the pathogenic effect of *H. pylori* by regulating the levels of pro-inflammatory cytokines. It is of significance to illustrate the underlying mechanisms. LIN28 is an RNA binding protein which activates cell transformation by inhibiting let-7 to promote inflammatory responses \[[@bb0135]\]. LIN28 inhibited the biosynthesis of let-7, and interestingly it was a target of let-7 itself. There was a bidirectional negative feedback regulation loop between them \[[@bb0140]\]. *H. pylori* inhibited the expression of most molecules of let-7 family, and some let-7 family molecules may be of value as potential biomarkers for the diagnosis of GC or its precancerous diseases \[[@bb0145], [@bb0150], [@bb0155]\]. Furthermore, let-7 regulated circadian rhythm and it was controlled by circadian rhythm genes as well \[[@bb0160],[@bb0165]\]. The above findings suggested the potential critical role of LIN28/let-7 axis for the circadian rhythm and its involvement in the pathogenic process of *H. pylori*. In this study, we demonstrated that *H. pylori* up-regulated the expression of BMAL1 and disrupted its rhythm through activation of LIN28a. BMAL1 enhanced the transcription of TNF-α, thereby aggravating the inflammatory response caused by *H. pylori*. Thus, we for the first time found the pathogenic relationship between *H. pylori* infection and circadian rhythm gene BMAL1. And we explored one of the specific pathways by which circadian rhythm genes mediated inflammatory responses induced by *H. pylori*. Therefore, the rhythmic gene BMAL1 may be a potential diagnostic marker and therapeutic target for *H. pylori* induced inflammatory diseases. 2. Methods and materials {#s0025} ======================== 2.1. IHC staining {#s0030} ----------------- Formalin-fixed, paraffin-embedded (FFPE) sections from mouse or patient samples were subjected to deparaffination and dehydration. After epitope retrieval with citrate buffer and H~2~O~2~ treatment, the samples were blocked with goat serum for 30 min, and then incubated with different primary antibodies overnight at 4 °C. And then the sections were incubated with secondary anti-mouse or -rabbit antibodies and the signals were detected by a DAB staining kit (Vector Laboratories, USA). The IHC scores were counted with Imagepro Plus. Primary antibodies used were BMAL1: AB4140 (Millipore), and LIN28A: ab155542 (Abcam). 2.2. RNA extraction, RT-PCR, and qRT-PCR {#s0035} ---------------------------------------- Total RNA were extracted with Trizol Reagent, and the mRNAs were reverse transcribed with RT reagent Kit gDNA Eraser (TaKaRa) to get cDNAs as the templates for the subsequent qRT-PCR. SYBR-Green (TaKaRa) and the corresponding qRT-PCR protocol were used to detect cDNA expression levels. β-actin was used as internal reference. Primers were shown as follows: hACTIN, Forward (F):5′-AGTTGCGTTACACCCTTTCTTG-3′, Reverse (R):5′-CACCTTCACCGTTCCAGTTTT-3′; mActin, F:5′-GTGACGTTGACATCCGTAAAGA-3′, R:5′-GCCGGACTCATCGTACTCC-3′; hBMAL1, F:5′-ATGTGACCGAGGGAAGATAC-3′, R:5′-GTGCTCCAGAACATAATCGA-3′; mBmal1, F:5′-ACAGTCAGATTGAAAAGAGGCG-3′, R:5′-GCCATCCTTAGCACGGTGAG-3′; hLIN28A, F:5′-CAGGTGCTACAACTGTGGAGGT-3′, R:5′-AGGGCTGTGGATTTCTTCTTCT-3′; hIL-1α, F:5′-TGGTAGTAGCAACCAACGGGA-3′, R:5′-ACTTTGATTGAGGGCGTCATTC-3′; hIL-1β, F:5′-ATGATGGCTTATTACAGTGGCAA-3′, R:5′-GTCGGAGATTCGTAGCTGGA-3′; hTNF-α, F:5′-CTGGGCAGGTCTACTTTGGG-3′, R:5′-CTGGAGGCCCCAGTTTGAAT-3′; hIL-6, F:5′-TCAATATTAGAGTCTCAACCCCCA-3′, R:5′-GCTCCTGGAGGGGAGATAGA-3′; hCXCL8, F:5′-TTTTGCCAAGGAGTGCTAAAGA-3′, R:5′-AACCCTCTGCACCCAGTTTTC-3′; hIL-10, F:5′-GACTTTAAGGGTTACCTGGGTTG-3′, R:5′-TCACATGCGCCTTGATGTCTG-3′; mTnf, F:5′-CAGGCGGTGCCTATGTCTC-3′, R:5′-CGATCACCCCGAAGTTCAGTAG-3′; hCLOCK, F:5′- CTCTTCTCGGAGTTCAAGAAAATCA-3′, R:5′- CATGCCTTGTGGAATTGGTA-3′; hCRY1, F:5′- TCCGATTTGGTTGTTTGTCA-3′, R:5′- GGATTATTTGTTGCTGCTGTAT-3′; hCRY2, F:5′- AACCACGACGAGACCTACGG-3′, R:5′- GGAGGGAGTTGGCGTTCATT-3′; hDBP, F:5′- CTGATCTTGCCCTATCAAGCATT-3′, R:5′- CGATGTCTTCGAGGGTCAAAG-3′; hPER1, F:5′- GCCAACCAGGAATACTACCAGC-3′, R:5′- GTGTGTACTCAGACGTGATGTG-3′; hPER2, F:5′- TCATTGGGAGGCACAAAGTC-3′, R:5′- AGGAGGTCTGGCTCATAAGGT-3′; hPER3, F:5′- CTGTGAGGATTTGAGGAACG-3′, R:5′- CTGATTTAGGTATGGCTGGG-3′; hREV-ERBα, F:5′- GGGAGGTGGTAGAGTTTGCC-3′, R:5′- TGTCTGGTCCTTCACGTTGA-3′; hRORα, F:5′- CACGACGACCTCAGTAACTACA-3′, R:5′- TGGTGAACGAACAGTAGGGAA-3′; hTIMELESS, F:5′- TCTGATCCGCTATTTGAGGCA-3′, R:5′- GGCAGAAGGTCGCTCTGTAG-3′. 2.3. Cell culture and H. pylori infection {#s0040} ----------------------------------------- BGC-823 cells were cultured in RPMI-1640 containing 10% fetal bovine serum (FBS), and AGS cells were cultured in F12 containing 12% FBS. All cells were supported in a humidified incubator with 37 °C, 5% CO~2~. *H. pylori* strains 11,637 and 26,695 were used to challenge gastric epithelial cells with different concentration, and the cells were collected at different time points as shown in the figures. Without special illustration, the gastric epithelial cells were challenged by *H. pylori* with MOI 1:100, and collected after 8 h. 2.4. Western blot {#s0045} ----------------- Total proteins were extracted with RIPA lysis buffer (Beyotime) with phosphatase and protease inhibitors, and the protein concentrations were detected with BCA Protein Assay Kit (Thermo Fisher Scientific). SDS-PAGE was used to separate lysates, and then the proteins were transferred to PVDF membranes. The membranes were blocked with 5% non-fat milk for 2 h at room temperature, and incubated with primary antibody overnight at 4 °C. After being incubated with secondary antibody, the signals were detected with millipore ECL regents. Antibodies used were: β-ACTIN (A1978, Sigma-Aldrich, RRID: [AB_476692](nif-antibody:AB_476692){#ir0015}), BMAL1 (D2L7G,CST, RRID: [AB_2728705](nif-antibody:AB_2728705){#ir0020}), CagA (Abcam, RRID:[AB_2049729](nif-antibody:AB_2049729){#ir0025}), LIN28A (ab155542, Abcam). 2.5. Resetting of circadian time in gastric epithelial cells {#s0050} ------------------------------------------------------------ After stimulating the cells with 50% horse serum, the medium was replaced by 2 ml F12 or RPMI-1640 supplemented with 10% or 12% FBS respectively in 6-well plate \[[@bb0170]\], and then challenged the cells with *H. pylori*. The cells were collected at 0, 4, 8, 12, 16, 20 h time points after the serum shock. 2.6. Mouse model {#s0055} ---------------- 80 C57BL/6 mice were separated into two groups randomly: one group was under a 12 h of light-12 h of dark (LD) cycle (*N* = 40), and another group was under constant darkness (DD, N = 40). They were raised for one week with normal feed to adapt to the environment. Then each group was divided into two subgroups respectively, subjected by intragastric administration (IG) with PBS or *H. pylori* strain *SSI* at 12 pm for one month (*N* = 20). Mice in each group were sacrificed at four time point: 6 am, 12 pm, 6 pm and 12 am (*N* = 5, for each time point). The method to extract gastric epithelial cells from mice was reported in this study \[[@bb0175]\]. All animal experiments were approved by Shandong University Research Ethics Committee. 2.7. H. Pylori staining with May-Grunwald-Giemsa {#s0060} ------------------------------------------------ May-Grunwald-Giemsa kit (G1930, Solarbio) was used to stain *H. pylori.* FFPE sections subjected to deparaffination and dehydration. Stained the sections with May-Grunwald working aliquots for 10 min, and then with Giemsa working aliquots for 20 min. Washed the remaining liquid quickly with anhydrous ethanol, and roasted sections slightly. 2.8. Inflammation score rules {#s0065} ----------------------------- A single pathologist, blind to the treatment, scored the inflammation of the samples. Inflammation was scored according to special inflammatory cells appearing in different parts of stomach, leading to a total score of 12, as previously described \[[@bb0180]\]. 2.9. Patient samples {#s0070} -------------------- Patient samples were from Jinan Central Hospital, Shandong, P. R. China. *H. pylori* infection in patients was detected by 13C urea breath test. The using of patient samples was approved by Shandong University Research Ethics Committee. 2.10. Transfection of plasmids and siRNAs {#s0075} ----------------------------------------- Lipofectamine 2000 (Invitrogen) was used to transfect BMAL1 and LIN28A siRNAs (Ribobio), the corresponding control siRNAs (Ribobio), mimics and inhibitor of let-7a-3p (Ribobio), and the corresponding negative controls (Ribobio) into gastric epithelial cells according to the protocol provided. For transfection of BMAL1 and LIN28A expression plasmids (Addgene), Roche transfection reagent was used according to the instruction book. Specific experimental steps were from manufacturer\'s protocols. BMAL1 expression plasmid was kindly donated by Dr. Steven Reppert (Massachusetts General Hospital, Boston, USA). 2.11. Chromatin immunoprecipitation assay (ChIP) {#s0080} ------------------------------------------------ SimpleChIP® Plus Sonication Chromatin IP Kit (\#56383, CST) was used to detect the binding of the corresponding proteins to DNA. All experimental procedures were performed according to the instructions. Antibodies used were: LIN28A (Abcam), BMAL1 (CST). 2.12. Dual luciferase reporter gene assay {#s0085} ----------------------------------------- Human BMAL1 wt-1 and wt-2 promoters and TNF-α promoter fragments were synthesized by company SYE.Biotech. KOD-Plus-Mutagenesis Kit (TOYOBO) was used to mutate or knock out the corresponding sites. Luciferase Assay System (Promega) was used to measure luciferase reporter activity. 2.13. Enzyme linked immunosorbent assay (ELISA) {#s0090} ----------------------------------------------- Human TNF-alpha ELISA Kit (KE00068, proteintech) was used to detect protein secretion of TNF-α. All experimental procedures were performed according to the instructions. 2.14. Statistical analysis {#s0095} -------------------------- Data of all experimental groups were described as the mean (±) SD. Analysis was evaluated by two-tailed Student\'s *t-*tests or Mann-Whitney *U* test. Statistical analyses were offered by GraphPad PRISM 7. *P* values \<.05 were considered statistically significant. 3. Results {#s0100} ========== 3.1. H. Pylori induced BMAL1 expression and disrupted its circadian rhythm {#s0105} -------------------------------------------------------------------------- *H. pylori* infection is a risk factor for gastric diseases \[[@bb0185]\]. It influenced gastric circadian events, such as gastric acidity, simultaneously \[[@bb0070]\]. To explore the relationship between *H. pylori* infection and circadian clock genes in gastric epithelial cells, we analyzed the relative expression of traditional circadian genes in public microarray data from GEO database (GSE10262). In the mouse gastric epithelial progenitor-derived cell line (mGEP), three repeated infection experiments were performed with two *H. pylori* strains: chronic atrophic gastritis (ChAG)-associated Kx1 and gastric cancer-associated Kx2. Bmal1 expression was significantly increased compared with other genes ([Fig. 1](#f0005){ref-type="fig"}a). The analysis of BMAL1 from other public microarray data (Fig. S1a-c) also showed its up-regulation during *H. pylori* infection. To further verify this, we used diseased 37 patient tissues from superficial gastritis (SG), atrophic gastritis (AG) to Intestinal metaplasia (IM) with or without *H. pylori* infection. The lamina propria cell composition is complex, and gastric epithelial cells are the frontier cells encountered by *H. pylori* so that they are closely related to the initiation of gastritis induced by the bacteria. Therefore, we focused on the effect of *H. pylori* on gastric epithelial cells in this study ([Fig. 1](#f0005){ref-type="fig"}b). There is a close relationship between *H. pylori* infection and the severity of precancerous lesions \[[@bb0190]\]. CagA-positive *H. pylori* showed a lower infection density in patients with milder lesions and the density of *H. pylori* increasesd with the onset of gastritis \[[@bb0195]\]. There was no obvious elevation of BMAL1 in patients with *H. pylori* infection in SG patients, which may result from the low density of *H. pylori* infection. However in AG and IM patients, the expression of BMAL1 in *H. pylori* infected patients was significantly higher than that in uninfected patients. Therefore *H. pylori* infection gradually elevated BMAL1 expression with increasing severity of the gastric diseases ([Fig. 1](#f0005){ref-type="fig"}c). Correspondingly, in another cohort of specimens, BMAL1 mRNA level was also higher in *H. pylori*-positive patients than that in *H. pylori*-negative samples ([Fig. 1](#f0005){ref-type="fig"}d). *H. pylori* infection is continuous and changeable *in vivo*, and the timing of individual cell infection is in random phase of the circadian curve. To investigate the effect of *H. pylori* on BMAL1 in random phase *in vitro*, desynchronized gastric cell lines AGS and BGC-823, which showed nearly constant BMAL1 levels \[[@bb0170]\], were incubated with two *H. pylori* strains (11,637 and 26,695). BMAL1 expression was increased both at mRNA and protein levels in cells with *H. pylori* infection, and the up-regulation was time and concentration dependent ([Fig. 1](#f0005){ref-type="fig"}e and f, Fig. S1d and e).Fig. 1H. pylori induced BMAL1 expression and disrupted its circadian rhythm. a, relative expression of circadian clock genes in mouse gastric epithelial progenitor-derived cell line (mGEP) infected with clinically isolated *H. pylori* strains Kx1 and Kx2, compared with uninfected group. The infection experiments were repeated three times, and represented by the numbers 1, 2, and 3. The data were obtained from the GEO database([GSE10262](ncbi-geo:GSE10262){#ir0005}). b and c, IHC staining for BMAL1 in 37 cases of superficial gastritis(SG), atrophic gastritis(AG) and Intestinal metaplasia(IM) patients respectively with or without *H. pylori* infection, and the relative values of IHC optical density. Scale bars, 100 μm (insets 50 μm). d, qRT-PCR analysis of BMAL1 in 41 cases of gastritis patients with or without *H. pylori* infection. e and f, qRT-PCR and Western blot analysis of BMAL1 in AGS and BGC-823 cells infected with *H. pylori* strain 26,695. The concentration of *H. pylori* and the cell collection time points were shown. g, AGS and BGC-823 cells were reset circadian time. Subsequently, they were infected with *H. pylori* strain 26,695. BMAL1 mRNA level was detected at 0, 4, 8, 12, 16, 20 h after the serum shock.Fig. 1 We had confirmed that *H. pylori* infection up-regulated BMAL1 expression in gastric epithelial cells. Since BMAL1 has stable period and amplitude as a circadian clock gene, it is necessary to identify whether *H. pylori* infection disturbed BMAL1 circadian rhythm or not. The synchronization model showed the eliciting of circadian gene expression by a serum shock. We reset circadian time of gastric epithelial cells with 50% horse serum before adding *H. pylori*, to demonstrate circadian oscillators existence in cell lines \[[@bb0170]\]. The amplitude of BMAL1 expanded dramatically after *H. pylori* infection. Besides, BMAL1 peak expression and the horizontal baseline were all raised in most phases ([Fig. 1](#f0005){ref-type="fig"}g). Therefore, *H. pylori* infection disrupted the circadian rhythm of BMAL1. In addition, we also examined the effect of *H. pylori* on the expression of other clock genes in desynchronized and synchronized gastric epithelial cells. *H. pylori* up-regulated the expression of REV-ERBα and RORα, while showed different effects on other genes. *H. pylori* infection significantly disturbed the rhythm of CRY1, DBP and REV-ERBα (Fig. S1f and g). This suggested the overall circadian rhythm disorders caused by *H. pylori*. 3.2. LIN28A directly activated the transcription of BMAL1, independent of let-7a {#s0110} -------------------------------------------------------------------------------- Studies have shown the role of LIN28/let-7 axis in circadian rhythm. Furthermore, BMAL1 contains a binding site of let-7a-3p in the 3′UTR region ([Fig. 2](#f0010){ref-type="fig"}a). LIN28 family consists of two molecules, LIN28A and LIN28B. To evaluate the relationship between these two molecules and *H. pylori*, we analyzed their expression in the GEO database (GSE10262) and speculated the activation of Lin28a by *H. pylori* (Fig. S2a). As a validation of previous experiments, we demonstrated a negative feedback loop formed by mutual regulation between LIN28A and let-7a-3p (Fig. S2b and c). *H. pylori* infection down-regulated the expression of let-7a-3p (Fig. S2d). To explore the relationship between let-7a-3p and BMAL1, let-7a-3p mimics and inhibitor were used. We found the inhibition of LIN28A and BMAL1 both at mRNA and protein levels by let-7a-3p ([Fig. 2](#f0010){ref-type="fig"}b and c). However, the results of the dual luciferase assay showed the inability of let-7a-3p to activate the transcription of BMAL1 through the binding site of its 3′UTR, which was unexpected ([Fig. 2](#f0010){ref-type="fig"}d).Fig. 2LIN28A directly regulated the transcription level of BMAL1, independent of let-7a. a, binding site of let-7a-3p on the 3′UTR region of BMAL1. *b,* qRT-PCR analysis of let-7a-3p and BMAL1 in AGS and BGC-823 cells transfected with mimics or inhibitor of let-7a-3p. *c,* Western blot analysis of BMAL1 and LIN28A in AGS and BGC-823 cells transfected with mimics or inhibitor of let-7a-3p. *d,* the dual luciferase assay detected the transcriptional regulation of let-7a-3p on the 3′UTR region of BMAL1. e and f, qRT-PCR and Western blot analysis of BMAL1 in AGS and BGC-823 cells transfected with LIN28A siRNA or overexpression plasmid. *g,* the binding sites of LIN28A in the BMAL1 promoter region, and the corresponding base mutation. *h,* ChIP was used to detect the binding between LIN28A and the two sites of the BMAL1 promoter region with and without LIN28A depletion. *i,* the transcriptional activation of the BMAL1 promoter region by LIN28A was detected by using a dual luciferase assay in case of binding sites mutation.Fig. 2 Based on the above results, we hypothesized the regulation of BMAL1 by let-7a-3p through LIN28A. After overexpression or depletion of LIN28A, BMAL1 showed corresponding up- or down-regulation at mRNA and protein levels ([Fig. 2](#f0010){ref-type="fig"}e and f). Although LIN28A is an RNA binding protein, it can directly bind to DNA, promoting transcription of the downstream target genes \[[@bb0200]\]. By analyzing the sequence of the BMAL1 promoter region, we found two potential LIN28A binding sites ([Fig. 2](#f0010){ref-type="fig"}g). Chromatin immunoprecipitation assay (ChIP) showed the direct binding of LIN28A to BMAL1\'s promoter, whereas after LIN28A suppression, the binding was accordingly weakened ([Fig. 2](#f0010){ref-type="fig"}h). To further confirm the transcriptional regulation of BMAL1 by LIN28A, we synthesized the wild type plasmid that contained the two binding sites, as well as the plasmids that the two sites were separately mutated. The subsequent dual luciferase assay was used to examine the effect of LIN28A on the transcriptional activation of BMAL1 ([Fig. 2](#f0010){ref-type="fig"}g). After overexpression or depletion of LIN28A, the transcriptional activity of the BMAL1 promoter regions was up- or down-regulated accordingly. However, these changes disappeared with the mutations of the binding sites ([Fig. 2](#f0010){ref-type="fig"}i). In summary, these results suggested the direct transcriptional regulation of BMAL1 by LIN28A by binding to the two DNA sites, independent of let-7a. 3.3. LIN28A enhanced H. pylori induced BMAL1 expression {#s0115} ------------------------------------------------------- To explore whether LIN28A mediated the expression of BMAL1 caused by *H. pylori*, we detected the effect of *H.pylori* on LIN28A. Indeed *H. pylori* promoted LIN28A expression at both mRNA and protein levels in a time and concentration dependent way ([Fig. 3](#f0015){ref-type="fig"}a and b). Of note, comparing "LIN28A si + *Hp26695*" with "NC + *Hp26695*" group, the induction of BMAL1 by *H. pylori* was alleviated upon LIN28A suppression ([Fig. 3](#f0015){ref-type="fig"}c and d). In addition, the expression of LIN28A was higher in *H. pylori*-positive patients than that in *H. pylori*-negative patients, and the differences were reflected by the relative values of IHC optical density ([Fig. 3](#f0015){ref-type="fig"}e and f). The expression of LIN28A in specimen was very similar to the expression of BMAL1 in the same tissues. Therefore, LIN28A may enhance BMAL1 expression induced by *H. pylori* both *in vitro* and *in vivo*.Fig. 3LIN28A enhanced BMAL1 expression induced by *H. pylori*. a and b, qRT-PCR and Western blot analysis of BMAL1 in AGS and BGC-823 cells infected with *H. pylori*. c and d, qRT-PCR and Western blot analysis of BMAL1 with LIN28A suppression on the basis of *H. pylori* infection. e and f, IHC staining for LIN28A in 37 cases of superficial gastritis(SG), atrophic gastritis(AG) and Intestinal metaplasia(IM) patients respectively with or without *H. pylori* infection, and the relative values of IHC optical density. Scale bars, 100 μm (insets 50 μm).Fig. 3 3.4. Effect of H. pylori-induced rhythm gene change on inflammatory factors {#s0120} --------------------------------------------------------------------------- Circadian rhythm is involved in the development of inflammation, which prompted us to study the potential relationship between circadian rhythm and inflammatory factors upregulated by *H. pylori*. Based on previous results, we selected several inflammatory factors closely related to the pathogenesis of *H. pylori*. We found possible binding sites of BMAL1 (*E*-box region) in the promoter regions of all the selected molecules (Fig. S3a). *H. pylori* significantly activated Tnf expression upon analysis of the GEO database (GSE10262) ([Fig. 4](#f0020){ref-type="fig"}a). And decline of TNF-α was found to be the most pronounced among the selected factors by BMAL1 depletion ([Fig. 4](#f0020){ref-type="fig"}b). Therefore TNF-α may be a potential target of BMAL1. Comparing "BMAL1 si + *Hp26695*" with "NC + *Hp26695*" group, induction of TNF-α RNA expression by *H. pylori* was relived upon BMAL1 inhibition ([Fig. 4](#f0020){ref-type="fig"}c). Accordingly, overexpression of BMAL1 showed the opposite effect ([Fig. 4](#f0020){ref-type="fig"}d). The change of TNF-α secretion was align with that of its RNA ([Fig. 4](#f0020){ref-type="fig"}e and f). Here IL-6 and IL-8 were found to be up-regulated by *H. pylori* infection, however their expression was not mainly regulated by BMAL1 (Fig. S3b).Fig. 4Effect of H. pylori-induced rhythm gene change on inflammatory factors. a, relative expression of inflammatory factors in mGEP. Three repeated experiments data were put in one box. The data were obtained from the GEO database([GSE10262](ncbi-geo:GSE10262){#ir0010}). b, qRT-PCR analysis of inflammatory factors in AGS and BGC-823 cells transfected with BMAL1 siRNA. c, qRT-PCR analysis of BMAL1 and TNF-α with BMAL1 depletion on the basis of *H. pylori* infection. d, qRT-PCR analysis of BMAL1 and TNF-α in AGS and BGC-823 cells transfected with BMAL1 overexpression plasmid. e, ELISA analysis of TNF-α, after BMAL1 inhibition on the basis of *H. pylori* infection. f, ELISA analysis of TNF-α in AGS and BGC-823 cells transfected with BMAL1 overexpression plasmid. g, AGS and BGC-823 cells were reset circadian time, and BMAL1 was depleted on the basis of *H. pylori* infection. BMAL1 and TNF-α mRNA levels were detected at 0, 4, 8, 12, 24 h after the serum shock. h, AGS and BGC-823 cells were reset circadian time, and BMAL1 was inhibited on the basis of *H. pylori* infection. ELISA analysis of TNF-α was detected at 0, 4, 8, 12, 24 h after the serum shock.Fig. 4 Since BMAL1 functions as circadian rhythm gene, TNF-α is supposed to be rhythmic as its target. Similar to BMAL1, the rhythm of TNF-α in rhythm-synchronized cells was destroyed by infection of *H. pylori*. Specifically, its amplitude raised and the expression in most phases increased (Fig. S3c). Correspondingly, when BMAL1 was inhibited, the rhythm change of TNF-α caused by *H. pylori* was restored at RNA and protein levels ([Fig. 4](#f0020){ref-type="fig"}g and h). These results indicated that BMAL1 was involved in the regulation of inflammatory response through targeting TNF-α after *H. pylori* infection. 3.5. BMAL1 directly promoted TNF-α transcriptional expression {#s0125} ------------------------------------------------------------- Based on the distribution of *E*-box elements in the TNF-α promoter region, we knocked them out separately and used double luciferase assay to detect the transcriptional activation of BMAL1 on wild-type plasmid and the five different knockout plasmids. It was found that when the E-box elements of the KO5 group were removed, the transcriptional activation of TNF-α by BMAL1 was not changed even though BMAL1 was inhibited. However, in the other four KO groups, transcriptional activation of TNF-α by BMAL1 was changed in line with the alterations of BMAL1, indicating the dispensableness of the these regions for TNF-α promoter activity regulated by BMAL1 ([Fig. 5](#f0025){ref-type="fig"}a). Since there are two E-box elements in the KO5 group, these two sites were mutated and tested separately in order to further determine the exact transcriptional regulation of TNF-α by BMAL1 ([Fig. 5](#f0025){ref-type="fig"}b). Both sites were found critical for BMAL1 activation since no alteration of promoter activity of TNF-α was observed upon BMAL1 suppression ([Fig. 5](#f0025){ref-type="fig"}c). Furthermore we confirmed the direct binding of BMAL1 to the above two sites in TNF-α promoter by ChIP with or without BMAL1 depletion ([Fig. 5](#f0025){ref-type="fig"}d). In summary, BMAL1 directly promoted TNF-α transcriptional expression.Fig. 5BMAL1 promoted TNF-α expression induced by *H. pylori*. a, a map for knockout of the E-box elements in the promoter region of TNF-α. Correspondingly, the effect of BMAL1 on the transcriptional activity of TNF-α was examined using a dual luciferase assay. *b,* the binding sites of BMAL1 in the TNF-α promoter region, and the corresponding base mutation. c, the transcriptional activation of the TNF-α promoter region by BMAL1 was detected by using a dual luciferase assay in case of binding sites mutation. d, ChIP was used to detect the binding between BMAL1 and the sites of the TNF-α promoter region after BMAL1 suppression.Fig. 5 3.6. Rhythm disorder aggravated inflammatory response caused by H. pylori {#s0130} ------------------------------------------------------------------------- We had confirmed the involvement of BMAL1 in the inflammatory response by modulating TNF-α under condition of *H. pylori* infection *in vitro*. However the relationship between *H. pylori*, gastric rhythm and inflammation *in vivo* was unknown. Since BMAL1 is centrally located in the regulatory loops of the rhythmic genes and its change was evident in *in vitro* studies, we still exploited the change of BMAL1 to represent the magnitude of the overall rhythm disorder of gastric epithelial cells *in vivo*. A mouse model was designed and generated as shown in [Fig. 6](#f0030){ref-type="fig"}a. Briefly there are 4 groups of mice, of which 2 groups were subjected to 12 h of light-12 h of dark (LD) cycle plus intragastric administration (IG) with PBS or *H. pylori* strain *SSI* at 12 pm respectively. The other 2 groups were raised under constant darkness (DD) plus intragastric administration (IG) with PBS or *H. pylori* strain *SSI* at 12 pm respectively. Uninfected mice were analyzed in "LD + PBS" and "DD + PBS" group, and only infected animals were analyzed in "LD + *SSI*" and "DD + *SSI*" group. Comparing "LD + *SSI*" with "LD + PBS" group, *H. pylori* infection up-regulated Bmal1 by two times at both mRNA and protein levels *in vivo*. Although this change was less than that *in vitro*, it still made sense ([Fig. 6](#f0030){ref-type="fig"}b-d). In order to prove the inflammation caused by *H. pylori* infection *via* rhythm change, we further aggravated the rhythm disorder by changing the light input. Comparing "DD + *SSI*" with "LD + *SSI*" group, in the case of the same conditions of *H. pylori* infection, the expression of Bmal1 in gastric epithelial cells increased more significantly when the external light input was artificially changed, indicating the aggravation of circadian rhythm disorder ([Fig. 6](#f0030){ref-type="fig"}b-d). Sydney system was used to define mild colonization which indicated less than one third of the sample surface was covered by *H. pylori* \[[@bb0205]\]. According to this standard, we found that although the degree of colonization of *H. pylori* in each mouse was basically the same, the number of mice colonized by *H. pylori* was significantly increased in the "DD + *SSI*" group with more severe rhythm disorder (Fig. S4a and [Fig. 6](#f0030){ref-type="fig"}e). This suggested the promoting role of circadian rhythm disorder on colonization of *H. pylori*. In addition, the detection and correlation analysis of the IHC scores of Bmal1 and Lin28a indicated the positive correlation between the two molecules *in vivo*, which was in line with our *in vitro* data (Fig. S4b-d).Fig. 6Rhythm disorder aggravated inflammatory response caused by *H. pylori*. a, animal model were designed as described in the methods and materials b, qRT-PCT analysis of Bmal1 in gastric epithelial cells extracted from each mouse. c, IHC staining for Bmal1 in each group. Scale bars, 100 μm (insets 50 μm). d, IHC scores for Bmal1 in each group. e, quantification of mice with or without *H. pylori* infection in all the groups. Total number in each group was 20. *H. pylori*-positive mice number in "LD + SSI" group was 8, and in "DD + SSI" group was 16. f, inflammation scores of each group. g, histogram of each group with different depth of inflammatory cells infiltration. Normal: no inflammatory cells infiltration; M: inflammatory cells infiltrating only mucosal layer; M + SM: inflammatory cells infiltrating both mucosal layer and submucosa. h, specific data displayed in the G chart. i, qRT-PCT analysis of Tnf in gastric epithelial cells extracted from each mouse.Fig. 6 *H. pylori* infection caused infiltration of a variety of inflammatory cells in the stomachs of mice, and the stage of inflammation could be judged according to this (Fig. S4e). Inflammation indexes were scored according to *H. pylori* infection rate as well as the cell type, depth and location of inflammatory cell infiltration. Comparing with "LD + *SSI*" group, "DD + *SSI*" mice showed more severe inflammation, indicating the involvement of rhythm disorder in the process of inflammation induced by *H. pylori* ([Fig. 6](#f0030){ref-type="fig"}f). Our *in vitro* data demonstrated that the longer of the *H. pylori* infection, the higher of the expression of BMAL1 and the more severe of the rhythm disorder. This was further confirmed *in vivo.* Increasing severity of circadian rhythm disorder aggravated the inflammation caused by *H. pylori*. Besides, the "DD + *SSI*" group had the most cases and the highest proportion of mice with inflammatory cells infiltrating both the mucosal layer and the submucosa according to inflammatory response position ([Fig. 6](#f0030){ref-type="fig"}g and h). Similarly, comparing with "LD + *SSI*" group, Tnf expression from mouse epithelial cells stimulated by *H. pylori* was more significantly increased in the "DD + *SSI*" group ([Fig. 6](#f0030){ref-type="fig"}i). The above results strongly suggested the aggravation of *H. pylori* induced inflammatory response by circadian rhythm disorder. 4. Discussion {#s0135} ============= *H. pylori* infection was related to circadian gastric acidity, 24-h gastric pH, and methods of administration to treat acid-related diseases \[[@bb0070],[@bb0210],[@bb0215]\]. In addition, a variety of gastrointestinal diseases might result from unnatural circadian rhythms \[[@bb0220],[@bb0225]\]. These studies highlighted the role of rhythm of the stomach in *H. pylori* induced diseases. We found that *H. pylori* promoted the expression of BMAL1 at the transcriptional level and disrupted its rhythm through LIN28A. Since BMAL1 is in a central regulatory position for the regulation of circadian rhythm genes, this result is instrumental for the understanding of effects of *H. pylori* on the overall rhythm. Gastritis provides the microenvironment niche for malignant transformation. Most gastric adenocarcinomas start from chronic superficial gastritis, and then evolve to atrophic gastritis or intestinal metaplasia, and finally to dysplasia or adenocarcinoma. To provide insights into the interaction between circadian system and inflammation, we found that BMAL1 could directly bind to the *E*-box elements in the promoter of TNF-α, activating its transcription and increasing its synthesis and subsequent secretion. The elevated TNF-α level further promoted the development of inflammation. This finding is in line with the results of previous studies \[[@bb0230]\]. BMAL1 was found to regulate TNF-α expression in the pathological state of intestinal epithelial cells \[[@bb0230]\]. However the underlying mechanism remained elusive since then. In addition, TNF-α regulated the circadian clock in astrocytes and rheumatoid synovial cells, and it also enhanced the mRNA expression of BMAL1 and CRY1 \[[@bb0235],[@bb0240]\]. And we found the transcriptional regulation of TNF-α by BMAL1. Combining these results, it may suggest a feedback loop between TNF-α and BMAL through their mutual regulation, There may be a cascade amplification of rhythm gene-mediated inflammatory response upon *H. pylori* infection, providing a new small field for future research. This also explains the fact that the expression of Lin28a is elevated correspondingly with the increase of rhythm disorder in animal experiments (Fig. S4b-c). Controversially, some studies showed a certain inhibitory effect of BMAL1 on inflammation in heart, lung, or macrophages and bone marrow cells \[[@bb0245], [@bb0250], [@bb0255], [@bb0260], [@bb0265], [@bb0270], [@bb0275]\]. This could be explained by different roles of circadian genes in different diseases. Or this may also be explained by differences in rhythm gene expression caused by different conditions. This is an open field and needs more work to illustrate in the future. In the animal model of this study, we chose to administer the mice at 12 pm in order to investigate the effect of *H. pylori* infection on gastric rhythm *in vivo*. In future experiments, we will administer normal or related rhythm gene knockout mice with *H. pylori* at different time points. By observing the progress of gastric inflammation, immune cell aggregation and inflammatory factor release, we will confirm the regulation of epithelial circadian clock on inflammation induced by *H. pylori*. This will also help us find better treatment options and time points of administration in the future. The *in vitro* data demonstrated that the longer of the *H. pylori* infection, the higher of the expression of BMAL1 and the more severe of the rhythm disorder. In order to prove that rhythm disorders aggravate inflammation induced by *H. pylori in vivo*, we can measure rhythm and inflammation changes at different infection times. We can also choose to change the external light input to aggravate the circadian rhythm disorder under the same infection conditions and measure inflammation scores. With the prolongation of *H. pylori* infection, many pathogenic factors of the bacteria may be aggravated. In order to exclude other pathogenic factors from judging the experimental results, we chose the latter experimental method. In [Fig. 6](#f0030){ref-type="fig"}, compared with "LD + PBS", the increase in inflammation score was not obvious in "DD + PBS" group, indicating that only the change of external light had little effect on inflammation. In the case of the same light change, inflammation score of "DD + *SSI*" was significantly increased compared with "LD + *SSI*" group. It supports that the rhythm can aggravate inflammation under the condition of *H. pylori* infection. Since this experiment aims to find causes that aggravate the inflammatory response and gradually worsen the disease under the premise of *H. pylori* infection. Therefore, we did not explore in this experiment whether the circadian rhythm gene can cause the same degree of inflammation changes when the disorder is excessive under other conditions. We may explore the more complex effects of circadian rhythm disorders on inflammation in future studies. Consistent with our results, studies have shown that inflammatory cytokines such as TNF-α in the gastric mucosa may be involved in the *H. pylori* induced gastric diseases \[[@bb0195],[@bb0280]\]. In addition to BMAL1, *H. pylori* can also induce TNF-α through direct activation of the transcription factors NF-κB and AP-1 \[[@bb0285],[@bb0290]\]. The genes encoding the NF-κB signaling pathway-related proteins belong to the clock-controlling gene family \[[@bb0295]\], and the activation of NF-κB is regulated by the rhythm loop represented by the clock genes BMAL1/CLOCK \[[@bb0300]\]. For example, in bone marrow cells, BMAL1 inhibits the activation of NF-κB \[[@bb0260]\]. CLOCK and NF-κB form a CLOCK/NF-κB complex that promotes NF-κB-mediated transcription and impedes the formation of the CLOCK/BMAL1 complex \[[@bb0305]\]. The above studies demonstrate the existence of mutual repulsion between NF-κB and BMAL1. Besides, in the presence of CLOCK, the RelB subunit of NF-κB binds to BMAL1, inhibiting or modulating the rhythmic target genes downstream of BMAL1/CLOCK in a more complex manner \[[@bb0310]\]. These results suggest that NF-κB and BMAL1 may have mutual regulation or binding and affect the expression of downstream genes such as TNF-α. However, the specific underlying mechanism requires a more in-depth study of various types of cells in different environments *in vivo*. For the relationship between inflammatory response and circadian rhythm, only the unilateral regulation mechanism of inflammatory factors by rhythm genes was explored in our study. And *H. pylori* was found to stimulate epithelial cells or immune cells and promote the secretion of inflammatory factors through other mechanisms \[[@bb0110]\]. The inflammatory factors, such as TNF-α, may promote the progression of the disease by binding to receptors on the surface of epithelial cells and then activate the corresponding signaling pathways to influence the expression of circadian genes. In future, it is significant to explore the relationship between inflammation, circadian rhythm and disease progression in this direction, and further elucidate the underlying mechanisms. There is an interaction between circadian rhythm and gut microbiota. Gut microbiota undergoes diurnal compositional and functional oscillations \[[@bb0315]\], while diurnal microbial behavior in turn drives host circadian transcriptional, epigenetic and metabolite oscillations \[[@bb0320]\]. The disruption of host circadian rhythm causes intestinal microbiota to lose the gut diurnally oscillates \[[@bb0325]\]. Furthermore, the rhythmic destruction of homeostatic microbiome not only eliminates normal chromatin and transcriptional oscillations, but also causes genome-wide *de novo* oscillations in both intestine and liver, thereby affecting the circadian fluctuations of physiology and disease susceptibility \[[@bb0320]\]. In addition, microbial metabolites directly affect circadian gene expression in the host \[[@bb0330]\]. Rhythmic genes, interfered by the disordered gut microbiota, disrupt body health by regulating downstream target genes \[[@bb0335]\]. The above results suggest the maintenance of homeostasis by interaction between rhythm genes and the gastrointestinal microbiota. In our study, we proved the aggravation of *H. pylori*-induced gastritis by rhythm disorders. It would be interesting to explore the specific components or metabolites of *H. pylori* to disrupt rhythm, the effects of rhythm disorder on *H. pylori*, and the influences of their interactions on the gastritis microenvironment. Since *H. pylori* is the most important and initial microbe to induce gastritis, the changes of other gut microbiota, and their relationship with rhythm disorder and progression of gastric diseases during the late stage of infection are worth to illustrating as well. Circadian rhythm regulates physiological homeostasis of the gastrointestinal tract concerning gut motility, gastric acid secretion, maintenance and restoration of the protective mucosal barrier, production of digestive enzymes, and immunologic system of gastrointestinal tract. Disruption of circadian physiology, due to sleep disturbance or shift work, may result in various gastric diseases, such as gastroesophageal reflux disease (GERD), gastric dyspepsia, peptic ulcer disease, metabolic syndrome or cancer \[[@bb0340]\]. Circadian clock disruption results in similar deleterious effects as the chronic inflammation induced by *H. pylori* \[[@bb0345]\]. However, no relationship between GERD and *H. pylori* was found \[[@bb0350]\]. *H. pylori* may be one of the causes of functional dyspepsia. *H. pylori* infection significantly causes chronic mucosal inflammation in the stomach and duodenum, which in turn leads to abnormalities in gastroduodenal movement and sensitivity, affects various endocrine functions of the stomach, and aggravates functional dyspepsia symptom. This suggests the coordination of *H. pylori* infection and rhythm disorders to promote functional dyspepsia \[[@bb0355]\]. The main causes of peptic ulcer include *H. pylori* infection and long-term use of Nonsteroidal Anti-inflammatory Drugs (NSAIDs). The former is a major risk factor for gastric cancer, while the latter shows some preventing effects for gastric cancer. The presence of *H. pylori* infection and precancerous lesions affects the treatment and outcome of the ulcers associated with rhythm disorders \[[@bb0360]\]. Circadian rhythm disorder adversely affects food choices, hunger and appetite, and generates deleterious metabolic consequences that lead to obesity \[[@bb0365]\]. There are increased odds ratios or relative risks of several gastrointestinal complications of obesity: gastroesophageal reflux disease, erosive gastritis, and gastric cancer \[[@bb0370]\]. This suggests that rhythm disorder can promote the development of gastritis through adverse metabolic reactions, and the transformation of inflammation into cancer. Interestingly, a link between the molecular clock machinery and carcinogenesis has been revealed \[[@bb0375]\]. In the future, it would be a good point to study the relationship between *H. pylori*, rhythm disorder and the transformation from inflammation to cancer, and illustrate the underlying mechanisms through which they are connected. In summary, we found that *H. pylori* up-regulated the expression of LIN28A, and LIN28A directly bound to the BMAL1 promoter, activating the transcription of BMAL1 to increase its expression. BMAL1 in turn promoted transcription of TNF-α by directly binding to the *E*-box elements on its promoter to increase its secretion. Therefore, our study revealed the mechanism through which disorder of circadian rhythm aggravated the inflammatory response induced by *H. pylori* ([Fig. 7](#f0035){ref-type="fig"}). This research on circadian rhythm genes may provide new potential targets for the early diagnosis and treatment of diseases related to *H. pylori* infection.Fig. 7Schematic model of the study. *H. pylori* up-regulated the expression of LIN28A, and LIN28A directly bound to the BMAL1 promoter, activating the transcription of BMAL1 to increase its expression. BMAL1 in turn promoted transcription of TNF-α by directly binding to the E-box elements on its promoter to increase its secretion. Therefore, our study revealed the mechanism through which disorder of circadian rhythm aggravated the inflammatory response induced by *H. pylori.*Fig. 7 Appendix A. Supplementary data {#s0160} ============================== Supplementary materialImage 1 Acknowledgements {#s0140} ================ Not applicable. Declaration of interest {#s0145} ======================= The authors indicate no potential conflicts of interest. Author contributions {#s0150} ==================== TL, XL and JJ contributed to the study conception and design. TL, WS, LZ and XL performed the experiment and contributed to data acquisition. TL, WS, LZ, XJ and PS contributed clinical specimens. TL, WS, SL, LM and LZ contributed to the analysis and interpretation of data. TL and XL wrote the manuscript. Funding sources {#s0155} =============== This work was supported by the National Natural Science Foundation of China (Nos. 81571960, 81371781, 81372680, 81471991, 81501720, 81772151 and 81871620), and the Shandong Province Major Science and Technology Innovation Project (2018CXGC1208). Supplementary data to this article can be found online at <https://doi.org/10.1016/j.ebiom.2018.11.043>.

Introduction {#Sec1} ============ Injury is the most common cause of morbidity and mortality in children within the United States. Long-term complications, including depression \[[@CR1]\], chronic pain \[[@CR2]\], and physical functional deficits that include difficulty with mobility and self-care \[[@CR2]--[@CR4]\], have been seen in almost half \[[@CR3]\] of injured children. Victims have also been shown to have a quality of life deficit as compared to their peers \[[@CR2], [@CR5]\]. Types of injuries and outcomes after pediatric trauma vary based on age, sex, income, and area of residence \[[@CR6]\]. Specifically, studies comparing rural and urban pediatric trauma generally demonstrate higher incidence and mortality in rural areas \[[@CR6]--[@CR8]\], while data from Canada \[[@CR9]\] suggests non-accidental trauma is more prevalent in urban areas. Non-accidental trauma in the United States is associated with more severe injury and worse outcomes than unintentional injuries \[[@CR4], [@CR8], [@CR10]\]. Despite existing data demonstrating the impact of environmental circumstances on pediatric trauma, little is known about the difference in epidemiology and outcomes after injury in children treated at urban versus rural pediatric trauma centers. This information would be useful in informing tailored injury prevention efforts and in ensuring pediatric trauma centers are prepared to care for their endemic populations. Our aim was to compare patterns of urban and rural non-accidental trauma, to evaluate for potential trends and identify cases where injury prevention efforts can be focused. Identifying factors associated with non-accidental trauma may function as a cue for providers to conduct further assessment in cases of pediatric trauma potentially preventing future harm. We hypothesized higher mortality in non-accidental compared to accidental trauma and higher severity of injury in urban non-accidental trauma compared to their rural cohort. Main text {#Sec2} ========= Methods {#Sec3} ------- To test our hypothesis, we gathered data from two American College of Surgeons level I adult and pediatric trauma centers, one in an urban setting and one in a rural area of the same state. All pediatric trauma admissions from May 2010 through July 2011 were identified in the trauma registries at the two centers. Trauma registries are maintained by trained abstractors according to guidelines set forth by the American College of Surgeons. Pediatric patients were defined as being between 0 and 14 years old at the time of admission. Patients with burn injury were excluded from the analysis, as only one of the hospitals includes an American Burn Association-verified burn center. Data collected included treatment center (urban or rural), patient age, gender, mechanism of injury (blunt or penetrating), injury severity score (ISS), hospital length of stay, outcome, and payer group. Intent of injury was recorded in the registry based on the Centers for Disease Control matrix of E-code groupings for injury as non-accidental, accidental, self-inflicted, undetermined, or other. We included only patients with intent coded as accidental or non-accidental in our study. The Institutional Review Board for Human Subject Research Committee approved the study at each institution in accordance with the ethical standards of the institution and with the Helsinki declaration. This research involved retrospective data collection, therefore formal consent is not required for this type of study and all identifying information was eliminated following completion of data gathering. We first compared non-accidental and accidental trauma in the entire cohort, followed by a comparison of non-accident trauma in the urban and rural cohorts. To test for similarities and differences we conducted the Chi^2^ test for categorical variables, Student's *T* test for comparison of means, and Mann--Whitney test for comparison of medians. We also conducted regression analysis to identify potential factors associated with non-accidental trauma while controlling for potential confounding variables. Univariable and multivariable logistic regression was conducted to assess factors associated with non-accidental trauma. Statistical significance was set for two-sided tests at P ≤ 0.05. Statistical analysis was conducted using STATA 14.1 (College Station, TX). Results {#Sec4} ======= Of 857 patients, the majority were treated at the rural center (54% vs. 46%). There were no significant differences between the urban and rural patients with regard to age (mean 7.2, SD 4.4), sex (65% male), ISS (median 4), or mortality (1.5%). Length of hospital stay (3.8 vs. 2.7, P = 0.01) and penetrating injury (9% vs. 5%, P = 0.024) were significantly higher in the urban cohort. The most common mechanism of injury for both groups was fall (35%). Rural children were less frequently transported by ambulance than urban children (41.6% vs. 64.2%, P \< 0.001). Children treated at the rural center were more likely to be uninsured compared to urban children (4% vs. 7%, P \< 0.001), while urban children were more frequently covered by government insurance (43% vs. 26%, P \< 0.001). We found no significant difference in mortality between urban (2%) and rural (1.1%) children (P = 0.274). Overall, 10% of injuries were considered non-accidental (Table [1](#Tab1){ref-type="table"}). The mean age for all non-accidental trauma patients was significantly lower than the overall pediatric trauma population (2.6 vs. 7.7, P \< 0.001). Significantly more children treated at the rural center were impacted by non-accidental trauma compared to the urban center (12.3% vs. 7.9%). Additionally, significantly more fatalities occurred in the non-accidental trauma cohort (5.7% vs. 1%, P = 0.001). A higher proportion of children with government insurance were impacted by non-accidental trauma (Table [1](#Tab1){ref-type="table"}).Table 1Comparison of accidental and non-accidental trauma in the upper MidwestAccidental trauma (N = 769)Non-accidental trauma (N = 88)P value^a^Age (y), mean (SD)7.7 (4.2)2.6 (3.2)\< 0.001Male, N (%)510 (66.3)50 (56.8)0.076ISS, median (CI)4 (4, 4)9 (9, 9)\< 0.001LOS days, mean (SD)3.0 (5.6)4.8 (9.8)0.011Mortality, N (%)8 (1.0)5 (5.7)0.001Mechanism of injury blunt, N (%)706 (92.5)87 (98.9)0.026Urban versus rural, N (%)0.037 Urban361 (92.1)31 (7.9) Rural408 (87.7)57 (12.3)Payer group, N (%)0.010 Government246 (32.3)43 (48.9) Commercial447 (62.4)40 (45.5) Uninsured41 (5.4)5 (5.7)*ISS* injury severity score, *LOS* length of (hospital) stay, *y* years^a^Chi^2^ test for categorical variables, Student's *T* test for comparison of means and Mann--Whitney test for comparison of medians Specifically examining children impacted by non-accidental trauma, there was no significant difference in age, gender, or mortality between the patients treated at the rural or urban center. (Table [2](#Tab2){ref-type="table"}). A larger proportion of non-accidental trauma patients were insured by governmental programs in both the urban (71% vs. 43.4%) and rural cohorts (36.8% vs. 26%). Patients that were younger, at the rural center, and receiving government insurance were at higher risk of non-accidental trauma on univariable analysis (Table [3](#Tab3){ref-type="table"}). However, only age remained an independent predictor on multivariable analysis (Table [3](#Tab3){ref-type="table"}).Table 2Characteristics of non-accidental traumaUrban (N = 31)Rural (N = 57)P value^a^Age (y), mean (SD)2.2 (2.8)2.8 (3.4)0.418Male, N (%)20 (64.5)30 (52.6)0.282ISS, median (CI)9 (5, 9)9 (9, 9)0.539Mortality, N (%)2 (6.5)3 (5.3)0.818LOS days, N (%)6.2 (12.2)4.0 (8.3)0.323Mechanism of injury blunt, N (%)31 (100)56 (98.3)0.458Payer group, N (%)0.006 Government22 (71.0)21 (36.8) Commercial7 (22.6)33 (58.9) Uninsured2 (6.5)3 (5.3)*ISS* injury severity score, *LOS* length of (hospital) stay, *y* years^a^Chi^2^ test for categorical variables, Student's *T* test for comparison of means and Mann--Whitney test for comparison of medians Table 3Regression analysis of factors associated with non-accidental trauma in the upper MidwestOR (CI)P value^a^OR (CI)P value^b^Age0.66 (0.59, 0.72)\< 0.0010.67 (0.60, 0.73)\< 0.001Sex male0.69 (0.47, 1.05)0.078--Rural1.63 (1.03, 2.58)0.0381.55 (0.93, 2.59)0.094Mechanism of injury penetrating, N (%)0.14 (0.02, 1.04)0.055--Payer group, N (%) Government2.08 (1.32, 3.29)0.0021.37 (0.82, 2.29)0.230 CommercialRefRef Uninsured1.45 (0.54, 3.89)0.4551.08 (0.37, 3.14)0.885^a^Univariable logistic regression^b^Multivariable logistic regression Discussion {#Sec5} ========== Trauma remains the leading cause of death among children and adolescents in the United States. In Minnesota, the unintentional injury death rate from 2001 to 2005 was 13.6 per 100,000, which is lower than the national average of 15 per 100,000 \[[@CR12]\]. We found that victims of non-accidental trauma had an even higher mortality than victims of unintentional trauma which is similar to other studies comparing the two \[[@CR11]--[@CR13]\]. However, we reject part of our hypothesis; we found no significant differences in injury severity in urban versus rural non-accidental trauma. Only age remained a significant predictor of non-accidental trauma following adjustment for confounding variables. In looking specifically at differences in the rural and urban population, our study cohort contained a higher percentage of patients treated at the rural center. While there were differences in mechanism and payer group, there were no differences in injury severity or mortality when comparing centers. This differs from previous studies that have consistently demonstrated a higher risk of overall injury \[[@CR6], [@CR13]--[@CR15]\] and higher rates of fatal injury for children in rural areas \[[@CR13]\]. National data comparing rural and urban trauma echoes these other studies, demonstrating increasing mortality with increasing rurality, though homicide rates tend to be higher in urban areas \[[@CR16]\]. Across the nation this difference is generally attributed to a lack of pre-hospital care and long transport times in rural areas. However, our data may differ, as both centers in this study are well-established multi-specialty pediatric trauma programs with emergency transport systems. We found that non-accidental trauma occurred at nearly twice the rate in rural setting when compared to the urban setting. Other studies focusing on non-accidental trauma report the opposite; a greater incidence in urban locations \[[@CR9]\], specifically demonstrating more violent pediatric deaths and firearm injuries \[[@CR13], [@CR17]\]. These findings; however, were demonstrated in older age groups than our study population. In our cohort, ages 14 and under, all the firearm-related injuries were considered accidental. Consistent with national data \[[@CR11], [@CR12], [@CR14]\], the non-accidental trauma group in our study tended to be younger than the general population of pediatric trauma patients. For nearly half of all non-accidental trauma patients, the primary payer was a government insurance program, this being even more pronounced in the urban setting. Bogumil et al. reported 64% of patients who were victims of non-accidental trauma across the United States had government insurance \[[@CR15]\]. Interestingly, in our cohort, despite a higher non-accidental trauma rate, a lower percentage of patients at the rural center were covered by government insurance. In an analysis of injury-related emergency department visits in 14 states including Minnesota it was noted that children from low-income communities were more likely to have injury-related emergency department visits than children from higher-income communities. In the Minnesota cohort of this study, 7% of children admitted with injury were uninsured, 20.9% had Medicaid or government insurance, and 70.8% had private insurance \[[@CR7]\]. Despite a higher non-accidental trauma rate, a lower percentage of patients at the rural center were covered by government insurance. Perhaps the rural cohort more accurately reflects the Minnesota population. In our cohort, patients that were younger, treated at the rural location, or receiving government insurance were at higher risk of non-accidental trauma; however, only age remained an independent predictor on multivariable analysis. Other studies have shown this similar relationship between age and risk for non-accidental trauma. From 2014 reports from the US Department of Health and Human Services, 27.4% of children who suffered non-accidental trauma were younger than 3 years old, comprising 70.7% of child fatalities. Most children who die of violence-related injuries have had previous treatment for non-accidental injuries \[[@CR16]--[@CR18]\]. Clinicians have been increasingly called upon to expand their role by recognizing non-accidental injuries, intervening, and educating about prevention of such injuries. This can be difficult in emergency settings where time constraints are limiting and providers are often unfamiliar with the patient \[[@CR19]\]. Without intervention, children returning to abusive homes face a 44% risk of repeated injury \[[@CR20]\]. Limitations {#Sec6} =========== Our data represents patients from two Midwest level I pediatric trauma centers. This population may not represent those in other areas of the country. Even with the combination of two centers' data, our sample size was relatively small. Our sample did not include burn injuries (another common cause of non-accidental injury in children) or children who died in the pre-hospital setting. As in other retrospective studies, data used in the analysis are limited to that which is available in the electronic medical record. It is possible that some instances of non-accidental trauma were missed during the patient's admission. Injury pattern, severity, and patient demographics need to be studied further, to optimize potential preventative strategies. A higher proportion of both urban and rural patients with government insurance are victims of non-accidental trauma than their commercial insurance counterparts. Additional large, multicenter studies are needed to focus on non-accidental trauma in the pediatric trauma population. Outreach efforts therefore should focus on this cohort for educational interventions aimed at reducing childhood trauma in this cohort. ISS : injury severity score LOS : length of (hospital) stay SD : standard deviation y : years APM, RMN, EMC, SFP, AMES, SEW, TE, SM, DDP, MDZ, CJR: Study conception and design. EMC, RMN, SFP, AMES, SEW, TE, SM: Data acquisition. APM, RMN, SFP, DDP, MDZ, CJR: Analysis and data interpretation. APM, EMC, RMN, SFP: Drafting of the manuscript. AMES, SEW, TE, SM, DDP, MDZ, CJR: Critical revision. All authors read and approved the final manuscript. Acknowledgements {#FPar1} ================ We thank the trauma registrars for their assistance extracting data from the trauma registry. Competing interests {#FPar2} =================== The authors declare that they have no conflict of interest. All authors have read and approved the final version of this manuscript. Availability of data and materials {#FPar3} ================================== Please contact corresponding author for data request. Consent to publish {#FPar4} ================== Not applicable. Ethics approval and consent to participate {#FPar5} ========================================== The Institutional Review Board for Human Subject Research Committee approved the study at each institution in accordance with the ethical standards of the institution and with the Helsinki declaration. This research involved retrospective data collection, therefore formal consent is not required for this type of study. Approval of waiver of consent was obtained from Hennepin County Medical Center Institutional Review Board for Human Subjects Research. Approval of waiver of consent was obtained from Mayo Clinic Institutional Review Board for Human Subjects Research. Funding {#FPar6} ======= This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Publisher's Note {#FPar7} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Two-dimensional (2D) radiography and model surgery have traditionally been used for planning orthognathic surgeries. This approach, however, has limitations when presenting and analyzing complex three-dimensional (3D) maxillofacial structures, especially for patients with major facial deformity or asymmetry[@b1][@b2][@b3]. Because 2D cephalometric images cannot provide complete information about the 3D structures, when conventional 2D surgical plans are executed, unexpected problems--such as bony collision in the ramus area; discrepancy in pitch, roll, and yaw rotation; midline difference; and chin inadequacy--may occur[@b4][@b5][@b6][@b7], leading to unsatisfactory outcomes. Three-dimensional imaging has revolutionized orthodontic and orthognathic surgical planning[@b8][@b9][@b10]. For patients with facial asymmetry, a 3D model can demonstrate the extent of yaw rotation in the maxilla and mandible, occlusal plane canting, and differential length on a mandibular body or the ramus. Compared with 2D images, 3D images register more accurate anatomic information and enable more precise quantitative measurement[@b11][@b12][@b13]. Moreover, 3D images are more effective in diagnosing asymmetry and estimating severity and thus may necessitate changes in surgical designs based on traditional 2D modelling[@b14]. Low-dose cone-beam computed tomography (CBCT), which enables the accurate reconstruction of a 3D skeletal model, has been applied widely in treatment planning for orthognathic surgeries[@b15][@b16][@b17][@b18]. Computer-aided design and manufacturing (CAD/CAM) techniques have been adopted to enhance the accuracy of preoperative planning and guidance for surgical procedures[@b19][@b20][@b21][@b22][@b23][@b24]. 3D virtual surgery was started and applied in orthognathic surgery for correction of facial deformities[@b25][@b26][@b27][@b28]. Recently developed 3D computer-assisted orthognathic surgery systems incorporate advanced 3D imaging, computer simulation software, CAD/CAM techniques, and image-guidance technologies and offer the high level of precision essential for optimal treatment planning and intraoperative execution[@b29][@b30][@b31]. Xia *et al*. and Gateno *et al*. have developed a computer-aided surgical simulation system that can create a composite skull model and simulate and transfer a virtual plan for correcting complex craniomaxillofacial deformities[@b32][@b33][@b34]. Similarly, Bell *et al*. combined computer planning and intraoperative navigation[@b35][@b36], and Lucia *et al*. presented a computer-aided surgery system that provides surgical planning and simulation and intraoperative guidance[@b37]. Lin *et al*. proposed a protocol for assessing the surgical simulation, guide positioning, intraoperative navigation, and outcome validation provided by computer-assisted surgery systems[@b38][@b39]. The 3D virtual surgery combined with navigation was proposed and suggested that simulation-guided navigation makes accurate postoperative outcomes possible for maxillary repositioning in orthognathic surgery[@b40][@b41][@b42][@b43][@b44][@b45]. However, the improvement from the 3D technique over the conventional method should be quantitated. A 2D plan can be conveniently revised after 3D surgical simulation. Few studies have investigated the quantitative assessment of such revision in terms of midline correction; chin position; yaw, roll, and pitch rotation; and frontal ramus inclination. This study evaluated how conventional 2D surgical plans were modified after 3D computer-assisted surgical simulation for patients with class III malocclusion and facial asymmetry. Results ======= The ICCs ranged from 0.85 to 0.96 for measurements in final 3D planning model (P2) within and between observers, and ranged from 0.81 to 0.92 for displacements of the 23 landmark locations in comparison of the P1 and P2 models, indicating acceptable intraobserver reproducibility and interobserver reliability. Absolute linear differences in the landmarks of the (x, y, z) coordinates of the P1 (2D) and P2 (3D) models are listed in [Table 1](#t1){ref-type="table"}; landmarks with a difference of more than 0.5 mm were considered to have changed. All absolute linear differences along the mediolateral direction (x axis) were significant and were more than 1 mm, indicating midline correction. The lower molars L6L and L6R had the largest change of 1.7 mm. Absolute linear differences along the inferosuperior direction (y axis) were not as large as those along the x axis. Significant vertical changes (i.e., canting correction) were observed at the U3, U6R, U6L, L6R, L6L, and chin point. Absolute linear differences of landmarks along the anteroposterior direction (z axis) were smaller than those along the x and y axes; significant changes in the anteroposterior direction were observed only at B, U6L, and U6R. Absolute angular differences of planes in the 2D and 3D plans are detailed in [Table 2](#t2){ref-type="table"}. Significant changes were noted in the pitch, yaw, and roll rotation as well as ramus inclination correction. Frontal ramus inclination had the largest correction of 3.37°, followed by yaw (1.88°), pitch (1.73°), and roll (1.06°). In the 30 patients in this study, the distribution and percentage of MMC positional changes from the 2D to the 3D plan were evaluated in terms of 6 parameters: midline, yaw, roll, pitch, ramus inclination, and chin position. Only for 2 patients (7%) were no changes necessary in the 2D plan. For obtaining the best outcomes, for 9 patients (30%), changes were necessary in 1 parameter; similarly, for 3 (10%), 9 (30%), 4 (13%), and 3 (10%) patients, changes were necessary in 2, 3, 4, and 5 parameters, respectively ([Fig. 1](#f1){ref-type="fig"}). Thus, 93% of the patients in this study required changes in at least 1 and up to 5 parameters in their 2D plan; the most frequently changed parameters were frontal ramus inclination (73%) and yaw rotation (60%), followed by roll (30%), midline correction (30%), pitch rotation (23%), and chin position correction (23%) ([Fig. 2](#f2){ref-type="fig"}). The accuracy of the surgical outcome and 3D simulation were evaluated by calculating the root-mean-square difference (RMSD) of the 3D simulation and postsurgical CBCT images ([Table 3](#t3){ref-type="table"}). The maxilla had a lower RMSD (0.63 mm) than did the mandible (0.85 mm). Overall, the accuracy of the surgical execution of the 3D plan was acceptable. An investigation of patient satisfaction revealed that 90% of the patients were happy with the surgical outcome (i.e., facial appearance), and all patients were satisfied with the dental alignment ([Fig. 3](#f3){ref-type="fig"}). No patients in this series were dissatisfied, and none received secondary correction for the initial facial deformity. Discussion ========== Advances in computer-aided surgical simulation has resulted in the wide use of new clinical protocols to evaluate craniomaxillofacial deformity and plan surgical procedures[@b4], especially for orthognathic surgeries as these protocols enable planning through virtual osteotomy in patients' 3D models. In addition, positioning guides and dental splints can be fabricated through computer-aided methods for the accurate fixation of the MMC to the cranium[@b6]. Xia *et al*. reported that the surgical outcomes achieved with the use of computer-aided surgical simulation were better than those achieved through traditional methods. Moreover, they reported that computer-aided techniques enable the surgeon to correct maxillary yaw deformities, place proximal/distal segments, and accurately restore mandibular symmetry. The present authors agree with their report, and the differences or the improvement were further quantitated. An increasing number of orthognathic surgeries are now being planned through computer-assisted methods. Virtual models detect and correct concealed problems and consequently assist in obtaining more favorable outcome. The results of this study suggest that 3D planning outperforms 2D planning in correcting midline deviation, ramus asymmetry, occlusal plane canting, and chin position. Absolute linear differences of landmarks along the x axis were significant (1.11--1.62 mm, [Table 1](#t1){ref-type="table"}) and indicated that the midsagittal landmarks (A, U1, L1, B, Pog, and Me) moved toward the skeletal midline in the 3D plan. The molars required the largest number of changes in the left--right direction, 1.62 mm for maxillary molars and 1.52 mm for mandibular molars, possibly because of their longer radius from the rotation center. The absolute linear differences of landmarks along the y axis ranged from 0.50 to 1.17 mm and were less than those along the x axis, with only 6 landmarks moving significantly. The significant changes of landmarks U6R, U6L, L6R and L6L along the y axis indicated equivalent distance correction from the molars to the FH plane on both sides and more accurate occlusal cant correction. The absolute linear differences of landmarks along the z axis were 0.50 to 1.17 mm and were less significant, except for landmarks B, U6L, and U6R, revealing that the 3D plan did not differ much from the 2D plan along the anteroposterior direction. Generally, after 3D simulation, landmarks along the mandible were more displaced than those on the maxilla. Midline correction varied from 1 to 2 mm ([Table 1](#t1){ref-type="table"}). Previous reports vary in their depiction of the magnitude of midline deviation without reducing smile aesthetics. Pinho *et al*. stated that 1 mm maximum acceptable deviation[@b46]. However, other studies have suggested that a midline deviation of up to 2 mm is acceptable[@b47][@b48][@b49][@b50] and that medical personnel are more sensitive to midline discrepancy than are laypeople. The difference up to 1.6 mm between the two planning systems might not have obvious clinical impact. However, the 3D system is still more accurate than the conventional 2D method, and the extra accuracy could camouflage minor human error during the surgery. In addition to the midline problem, this study revealed that the 2D plan could not completely predict occlusal cant and yaw correction, which may compromise dentofacial aesthetics. Angular changes in the MMC after 3D simulation were significant ([Table 2](#t2){ref-type="table"}). The yaw rotation was changed by 1.88° to obtain proper alignment of proximal/distal segments and more symmetry along the ramus. Yaw rotation is difficult to predict in 2D planning but is vital for managing intraoperative bony collision and cheek asymmetry. Frontal ramus inclination decreased from 4.14° to 0.78° resulting in more cheek symmetry after the simulation because the proximal segment can be moved in to or out from the inner cortex before the bicortical screw fixation. For patients complaining cheek asymmetry, the position of the ramus segments must be changed to achieve symmetry. Pitch rotation was changed by 1.73° as clockwise rotation of the MMC was planned to correct the concave facial profile with class III malocclusion, increasing postoperative stability and obtaining a better smile arc[@b51]. [Figure 2](#f2){ref-type="fig"} presents the distribution of the changes in the parameters; the frontal ramus inclination and yaw rotation were the most altered. During 3D virtual surgery, especially for patients with facial asymmetry, movement of the proximal segment of mandibular ramus is required in order to achieve symmetry on both sides. The authors simulated the axis or fulcrum of the ramus rotation using the condylar center, and degree of movement was decided by the symmetry from both ramus inclinations. We also used a CAD/CAM positioning (spacer) guides to facilitate reproduction of the planned ramus position, allowing precise proximal segment positioning. Yaw rotation of the MMC may change anteroposterior projection on both sides of the maxilla. In our experience, the discrepancy in the maxillary anteroposterior dimension is less noticeable, more tolerated, and easier to correct than the discrepancy in the ramus. Roll, midline, chin position, and pitch were less frequently revised ([Fig. 2](#f2){ref-type="fig"}) because these can be predicted on 2D images. However, the 20--30% incidence of change after simulation indicated that the 3D method is more sensitive and can further improve surgical planning. Occlusal canting on 2D posteroanterior radiograph may have been miscalculated because of the difficulty in precisely defining molar crown position, which can be readily identified on a 3D model. Therefore, canting correction (roll) was required for 30% of the patients. For most patients with facial asymmetry, the original 2D surgical plan must be revised after computer-assisted simulation. In this study, 93% of the patients required revisions in 1--5 parameters. In our experience, severity of facial deformity is proportional to the extent of the change in the parameter: 10% of the patients required changes in 5 parameters because of severe canting and facial asymmetry. Thus, the traditional 2D method is inadequate in treating patients with facial asymmetry. This study focused on how conventional 2D surgical plan was revised after 3D computer-assisted surgical simulation for patients with asymmetric facial prognathism. It is to be noted that there is difference in the planning and outcome assessment between the two systems, and critical analysis of the difference is required. The study did not intend to overlook the well-established 2D conventional planning for orthognathic surgery, but rather to draw attentions to acknowledge the deficit of 2D method and the possible modifications into clinical planning and surgical execution, so that better outcome could be achieved. Considering the additional time, expense and effort required, the 3D computer-assisted surgical simulation system might be considered as an additional or alternative method for planning patients with complex maxillofacial deformity. As for the accuracy of the actual reproduction of the planning on the real patient, the positioning guide combined with the single-splint surgical technique used to control the movement of the MMC proposed in this study was effective and user friendly. The hooked shape design of the guide can be quickly and accurately wedged into the maxillary piriform edge, and the temporary screw fixation of the MMC repositioning guide facilitates intraoperative inspection. A facial aesthetic outcome is as important as functional dental results. A real-time navigation can be used as a definitive tool to determine the final bone position without the physical guides or as an additional tool to guide the bone movement. In this study, the accuracy of reproduction of the 3D plan was tested using the RMSD value between the 3D simulation image and the postoperative model. Accurate registration of the two image models is important before the comparisons. 3D simulation and postoperative models were superimposed using the surface registration method on the basis of the cranial base, which was stable and unaffected by the surgery. The accuracy of the cranial base registration was verified by observing the distance color map between two superimposed images. Based on the color scale, differences between the surfaces was less than 0.3 mm over the forehead and cranial base regions and the RMDS value of forehead and orbital areas were calculated and used to evaluate the front-orbital matching error ([Fig. 4](#f4){ref-type="fig"}). The RMSD value of 0.5 mm or less was considered acceptable to ensure that the corresponding reference areas had the maximum possible accuracy[@b52][@b53]. Overall, the accuracy of the surgical execution achieved a satisfactory precision in positioning the maxilla with an acceptable RMSD of the maxilla (0.63 mm) and the mandible (0.85 mm) ([Table 3](#t3){ref-type="table"}). In summary, the 3D computer assisted surgical simulation helps to improve the planning for patients complaining facial prognathism and asymmetry. The revisions of planning occurred more in the frontal ramus inclination and yaw rotation, but also in other movements of the maxillomandibular complex. The improvement of the surgical planning was measured. The treatment outcome was satisfactory. The information from this study could be used to augment the clinical planning and surgical execution when a conventional approach is applied. Methods ======= The study procedure, outlined in [Fig. 5](#f5){ref-type="fig"}, was divided into five stages: (I) traditional 2D surgical planning, (II) 3D simulation---3D CT images and initial models, 3D models of the 2D plans (P1), modified 3D models after simulation (P2), (III) comparison between P1 and P2, (IV) surgical execution aided by positioning guides, and (V) postoperative validation---superimposition of postoperative 3D images and the P2 models. Ethics ------ This retrospective study was conducted and approved by Chang Gung Craniofacial Center, Taiwan. All experiments were performed with the approval of the Institutional Review Board (IRB) of Chang Gung Memorial Hospital (IRB 103-4038C) and the study methods were carried out in accordance with the approved guidelines of IRB. Written informed consents were obtained from the patients or the guardians of the patients younger than 20 years. Patient collection ------------------ This study included thirty consecutive patients (22 female and 8 male) with class III malocclusion and facial asymmetry undergoing corrective surgery from July 2013 to Feb 2015. The mean age of these patients at surgery was 22.4 years (18--26 years). Patients with congenital or acquired deformities and syndromes, facial cleft, or a history of trauma were excluded. All patients had a concave facial profile, paranasal depression, mandibular prognathism, class III malocclusion, an inverse incisal relationship (overjet −2.5 ± 1.66 mm), and facial asymmetry. The facial asymmetry was characterized by occlusal plane canting, discrepancy between upper and lower dental midline, chin deviation away from the midline, and cheek and jaw irregularity that was readily perceivable by patients and clinicians. Psychosocial evaluation yielded negative findings in these patients. Image acquisition ----------------- Three-dimensional maxillofacial images were acquired using an i-CAT CBCT scanner (Imaging Sciences International, Hatfield, PA) with a low-dose protocol and patient teeth in light-contact condition at 120 kV, 5 mA, and 50 Hz. The extended field of view was 22 (height) × 16 (depth) cm, the scanning time was 40 s, and the voxel size was 0.4 × 0.4 × 0.4 mm. The images were stored in the digital imaging and communications in medicine (DICOM) format and processed with a slice thickness of 0.4 mm. Three-dimensional images and the initial 3D virtual models were obtained using CBCT and the SimPlant O&O (Materialize, Leuven, Belgium) and Dolphin (Dolphin Imaging and Management solutions, Chatsworth, California) software programs. Using the segmentation function, the maxilla and mandible were outlined and the replacement of dentition in 3D CBCT and the virtual occlusion setup were obtained. 2D surgical planning -------------------- All patients received presurgical orthodontic treatment, including leveling, alignment, arch coordination, and dental decompensation. One week before the surgery, 2D cephalogram (lateral and frontal views), photographs, and dental casts were captured and the CBCT examination was performed. Subsequently, 2D surgical planning was conducted according to the cephalometrics for the orthognathic surgery[@b51][@b54]. The dental casts were mounted in an articulator with facebow transfer and bite registration. The model surgery was performed according to the surgical plan drawn from the 2D cephalometrics, and the final occlusal splint was fabricated according to the planned maxilla and mandibular position. All aforementioned procedures were performed by the same orthodontist (CTH). 3D surgical simulation ---------------------- The 2D surgical plan was transferred to the 3D simulation system, and an initial virtual surgical model was created. A Frankfort horizontal (FH) plane was used as the reference plane. A Frankfort horizontal (FH) plane was used as the reference plane. The FH plane was drawn from orbitale to average porion in 2D lateral cephalometric radiographs. The FH plane passed through the midpoint between left and right porion, right orbitale, and left orbitale in 3D model. Although the FH plane of a 2D planning could be somewhat different from the FH of a 3D environment. Comparing 2D and 3D system, some corresponding measurement such as A-B, ANS-Me, N-ANS, S-ANS, ANS-U1, SNA, SNB, MPA and upper occlusal plane to FH plane have found no clinically significant difference. These 2D measurements could be substituted by the corresponding 3D measurements on the 3D model as the registration of the landmarks were inspected in cross-sectional images of axial, coronal, and sagittal slices which 3D Frankfort horizontal (FH) plane was the most concordant with FH plane used for cephalometric radiography[@b11][@b52][@b53]. There were significant differences in gonial angle, lateral ramal inclinations and ramus length between 2D and 3D. As we know, a lateral cephalometric radiograph projects an object slanted toward the left or right onto a sagittal plane, an error in the length or angle depend on how much it slants. Thus, it is less effective in measuring the ramus. It also may be difficult to find asymmetry in case the maxilla yaw on the vertical axis of the skull and this can be solved by 3D measurement. The 3D surgical model included LeFort I osteotomy in the maxilla, bilateral sagittal split osteotomy in the mandibular ramus, and genioplasty. The single-splint two-jaw orthognathic surgery method was applied. The distal mandibular segment was moved by using the final occlusal splint such that it occluded the maxilla, thus forming the maxillomandibular complex (MMC). The MMC was mobilized to the planned position (labelled P1) according to the 2D surgical plan. Data used for transferring 2D plan to 3D system were (1) the midline shift and roll adjustment on the frontal view, (2) pitch rotation, positioning of A, ANS, U1, U6, SNA, and SNB in the right lateral view, and (3) genioplasty in both frontal and right lateral views. The MMC position was evaluated and modified if it was found not perfect. The improvement of the 2D planning in 3D environment was adjusted by using certain measurements that are much closure to Chinese norms of cephalometric analysis[@b55]. The MMC was moved and rotated in the simulation system by the treatment team until the ideal position was achieved and approved. The dental midline (translation) and occlusal plane (roll rotation) were first modified in the frontal view. Subsequently, in the lateral view, the occlusal plane (pitch rotation) and facial profile were adjusted. The basal view of the 3D image was referred to for verifying the symmetry and collision between the ramus segments and the mandibular body contour (yaw rotation). These processes were repeated and were finally used as the virtual 3D planning model (refer to the [video](#S1){ref-type="supplementary-material"}). The new position was labeled P2. Each 3D surgical simulation was performed twice by two operators and measurement in P2 was used to investigate the intraobserver reproducibility and interobserver reliability. In summary, the surgical plan from conventional 2D cephalometry was transferred to the 3D imaging system, producing the P1 model. The P1 images were evaluated and modified by moving the maxillomandibular complex if it was found to have problems in midline, proportion, bony collision, etc. The final maxillomandibular position was named P2. Comparison of the P1 and P2 models ---------------------------------- To compare the differences in P1 and P2, image registration and superimposition of the cranial base were performed in the two models by using the best-fit method. The accuracy of the cranial base registration was verified by observing the distance color map between the registered P1 and P2 images. The deviation value was automatically calculated, and the value of 0.5 mm or less was considered acceptable to ensure that the corresponding reference areas had the maximum possible accuracy[@b52][@b53]. After the registration, the 23 reference landmarks could be located separately in P1 and P2 model for differences comparison. The two models used the same coordinate system to measure changes in hard tissue landmarks. Twenty-three landmarks and 3 reference planes were defined on each 3D model to conduct 14 linear and 4 angular measurements ([Table 4](#t4){ref-type="table"} and [Fig. 6](#f6){ref-type="fig"}). Changes of the landmarks in the FH plane, coronal plane, and sagittal plane were calculated. The 3 planes are defined in [Table 4](#t4){ref-type="table"}. Angular measurements, namely yaw, roll, and pitch rotation and bilateral ramus inclination in the frontal view, were performed from P1 to P2. Six parameters---midline, genioplasty, pitch, yaw, roll and ramus inclination---were used to evaluate the positional changes in the MMC. Changes in these parameters revealed the revisions of the 2D surgical plan (P1) after the 3D simulation (P2). Ten randomly selected cases were used for error assessment. The landmarks were located and digitized in each 3D model. The same examiner repeated the procedure 1 month later, and displacements of the landmarks were measured. Surgical implementation ----------------------- After 3D simulation, the surgery was executed according to the 3D planning by the senior surgeon (LJL). Two-jaw orthognathic surgery by using a single occlusal splint was applied[@b56][@b57]. The maxilla and mandible were mobilized after implementing LeFort I and bilateral sagittal split osteotomy[@b58][@b59]. The maxillary and mandibular segments were placed in the intermaxillary fixation along with the final occlusal splint. The MMC was repositioned using the positioning guides, and the maxilla was fixed on each nasomaxillary and zygomaticomaxillary buttress. The MMC positioning guides were designed using CAD software (Geomagic Wrap software, 3D System, USA) and fabricated using a 3D printer (Objet30 OrthoDesk jets, Stratasys Ltd. Nasdaq: SSYS) according to the 3D plan of the maxillary position[@b60]. Because a single occlusal splint method was used, the MMC location was determined by positioning the maxilla. The positioning guides were matching to the planed movement of the LeFort I segment and creating a hook-shaped wedge in the piriform aperture for stabilization. The hook-shaped design allowed fast and accurate positioning of the maxilla during surgery. An adequate contact area was required, superiorly from infraorbital foramen and inferiorly to gingiva. Drill holes were created for temporary fixation of the guides and the maxilla[@b60]. Postoperative validation ------------------------ CBCT images were captured 1 month after surgery. To evaluate the accuracy of the virtual surgical planning with respect to the actual surgery, the surgical simulation model (P2) and the postoperative CBCT image were registered using surface registration of the cranial base and the upper facial bones. After the initial registration, the images were superimposed to quantify the difference between the images. The accuracy of the surface superimposition was calculated in terms of the root-mean-square deviation (RMSD) of distance between the superimposed model, with RMSD ≤ 0.5 mm considered acceptable ([Fig. 4](#f4){ref-type="fig"})[@b61][@b62]. Differences between the two models were computed for validation. Patient satisfaction -------------------- In follow-up visits at least 6 months after surgery, patients were inquired about their level of satisfaction with the dental alignment and facial appearance. Statistical analysis -------------------- To evaluate intraobserver reproducibility and interobserver reliability of measurements, the intraclass correlation coefficients (ICCs) of the errors as well as the means and standard deviations were calculated. A paired t-test was used to evaluate the significance of the differences in linear and angular measurement, with p \< 0.05 considered significant. Additional Information ====================== **How to cite this article**: Ho, C.-T. *et al*. Three-dimensional surgical simulation improves the planning for correction of facial prognathism and asymmetry: A qualitative and quantitative study. *Sci. Rep.* **7**, 40423; doi: 10.1038/srep40423 (2017). **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary Material {#S1} ====================== ###### Supplementary Information ###### Supplementary Video We thank Ms. Lien-Shin Niu and Ms. Jing-Ling Weng for their assistance with the 3D simulation and the production of positioning guides. The study was supported by a grant from the Ministry of Science and Technology (MOST 103-2314-B-182-028-MY3) and a grant from Chang Gung Memorial Hospital (CRRPG5C0291-3). **Author Contributions** C.-T. Ho conceived, designed, performed the experiments, and wrote the manuscript. H.-H. Lin. helped with the designing and interpreting of the results. Eric J.-W. Liou. analyzed the data. L.-J. Lo. conceived and designed the experiments, and wrote the paper. {#f1} {#f2} {#f3} {#f4} {#f5} {#f6} ###### Absolute linear changes of landmarks in x, y, z axis between 2D plan and 3D plan (3D image models registered at the cranial base). Landmark X-axis Y-axis Z-axis ---------- ---------------- ---------------- -------------- ------ ----------------- ----------------- ------------- ------ ----------------- ----------------- ------------- ------ A 79.30 (3.76) 78.17 (3.98) 1.11 (1.71) 0.03 −121.61 (9.56) −123.11 (9.54) 0.50 (0.95) 0.50 −145.35 (4.40) −146.06 (4.53) 0.71 (1.01) 0.13 B 79.60 (3.80) 78.32 (3.51) 1.24 (1.53) 0.01 −118.03 (10.81) −118.62 (10.49) 0.59 (0.90) 0.30 −184.77 (4.78) −185.78 (4.69) 1.02 (1.43) 0.03 U3L 94.50 (18.22) 93.17 (18.00) 1.35 (1.85) 0.01 −114.22 (23.81) −115.06 (23.90) 0.84 (1.09) 0.05 −161.61 (30.85) −162.47 (31.08) 0.86 (1.17) 0.05 U3R 59.40 (12.04) 57.91 (11.74) 1.53 (2.02) 0.00 −113.27 (23.84) −114.19 (23.89) 0.92 (1.52) 0.07 −160.40 (31.08) −162.04 (30.93) 0.81 (1.35) 0.05 U6L 101.70 (19.56) 100.09 (19.27) 1.61 (1.81) 0.00 −96.34 (21.10) −97.36 (21.23) 1.01 (1.15) 0.01 −157.78 (30.22) −158.73 (30.46) 0.95 (1.32) 0.04 U6R 53.70 (4.68) 52.07 (4.49) 1.62 (1.75) 0.00 −99.79 (11.03) −100.96 (10.94) 1.17 (1.63) 0.02 −162.84 (4.69) −164.08 (4.75) 1.24 (1.50) 0.01 UI 79.50 (4.12) 78.23 (4.41) 1.31 (1.95) 0.02 −125.30 (9.91) −125.95 (9.59) 0.65 (1.24) 0.26 −167.39 (4.68) −168.17 (5.13) 0.77 (1.16) 0.10 L3L 93.70 (3.59) 92.34 (3.80) 1.31 (1.85) 0.01 −116.23(9.31) −117.05 (9.07) 0.82 (1.09) 0.06 −166.12 (5.10) −166.91 (5.51) 0.79 (1.17) 0.09 L3R 65.60 (4.50) 64.30 (4.49) 1.32 (1.81) 0.01 −115.70 (8.71) −116.49 (8.40) 0.78 (1.41) 0.14 −165.95 (5.14) −166.86 (5.59) 0.91 (1.36) 0.06 L6L 99.60 (19.25) 98.11 (18.89) 1.52 (1.62) 0.00 −94.98 (21.27) −95.85 (21.35) 0.87 (1.04) 0.03 −158.69 (30.34) −159.59 (30.56) 0.90 (1.34) 0.06 L6R 54.60 (10.95) 53.10 (10.54) 1.52 ((1.61) 0.00 −94.43 (20.69) −95.32 (20.84) 0.89 (1.25) 0.04 −158.77 (30.32) −159.65 (30.51) 0.89 (1.25) 0.05 LI 79.90 (4.06) 78.65 (3.76) 1.22 (1.66) 0.01 −123.08 (9.55) −123.60 (9.28) 0.52 (0.96) 0.45 −167.97 (4.90) −168.61 (5.22) 0.64 (1.01) 0.23 Pog 79.70 (3.70) 78.38 (3.32) 1.33 (1.64) 0.00 −118.15 (10.33) −118.99 (9.92) 0.84(0.99) 0.03 −202.68 (4.48) −203.33 (4.68) 0.65 (1.07) 0.22 Me 79.70 (3.61) 78.33 (3.24) 1.36 (1.68) 0.00 −115.09 (10.55) −116.04 (10.22) 0.95 (1.09) 0.02 −206.04 (4.63) −206.71 (4.77) 0.66 (1.10) 0.21 ^\*^The difference was considered statistically significant if p \< 0.05. ###### Absolute angular change of the maxillomandibular complex between P1 model (2D plan) and P2 model (3D plan). Parameters 2D planMean (SD) 3D planMean (SD) DifferenceMean (SD) p value[\*](#t2-fn1){ref-type="fn"} ----------------------------------------- ------------------ ------------------ --------------------- ------------------------------------- Pitch 13.50 (4.04) 15.23 (4.03) 1.73 (0.41) 0.00 Roll 1.43 (1.17) 0.37 (0.41) 1.06 (0.23) 0.01 Yaw 0.00 (0.00) 1.88 (1.73) 1.88 (0.32) 0.00 Difference of frontal ramus inclination 4.15 (3.06) 0.78 (1.47) 3.37 (3.21) 0.00 ^\*^The difference was considered statistically significant if p \< 0.05. ###### The absolute root-mean-square distance (RMSD) for the maxilla, mandible and maxillomandibular complex comparing post-surgical 3D image and 3D simulation model.   Forehead+ Orbital Maxilla+ Mandible Maxilla Mandible ---------- ------------------- ------------------- --------- ---------- **Mean** 0.43 0.77 0.63 0.85 **SD** 0.02 0.36 0.25 0.41 The forehead and orbital areas were used for registration of the two images. The data were in mm. ###### Definitions of 3D cephalometric landmarks and reference planes. Landmark Abbreviation Definition ---------------------------- ------------------------------------------------------------------ ----------------------------------------------------------------------------------------------------------------- Orbitale OrLOrR The most inferior point of left infraorbital rimThe most inferior point of right infraorbital rim Porion PoL,PoR The highest points of the left external acoustic meatusThe highest points of the right external acoustic meatus Average orbitale OrA Midpoint between the OrR and OrL. Nasion N The middle point of the frontonasal suture Anterior nasal spine ANS The most anterior midpoint of the anterior nasal spine of the maxilla Point A A The innermost point on the contour of the premaxilla between ANS and U1 Point B B The innermost point on the contour of the mandible between bony chin and L1 Upper incisor U1 Midoint between the crowns of the maxillary central incisors Lower incisor L1 Midpoint contact between the crowns of the mandibular central incisors. Pogonion Pog The most anterior midpoint of the chin on the outline of the mandibular symphysis Menton Me The most inferior midpoint of the chin on the outline of the mandibular symphysis. Maxillary canine U3LU3R Cusp of the left maxillary canineCusp of the right maxillary canine Mandibular canine L3LL3R Cusp tip of the left mandibular canineCusp tip of the right mandibular canine First maxillary molar U6LU6R Mesio-buccal cusp of the left first maxillary molarMesio-buccal cusp of the right first maxillary molar Lateral condyle point Cd latLCd latR The most lateral point of left condyle headThe most lateral point of right condyle head Lateral gonion Go latLGo latR The most lateral point of left gonionThe most lateral point of right gonion Frankfort horizontal plane FH plane A plane passing PoL, PoR, and OrA Coronal plane A plane perpendicular to FH plane, through right and left porion Sagittal plane   A plane perpendicular to the FH and coronal planes, through nasion

1. Trading within Biological Systems {#sec1-cancers-11-00117} ==================================== In economics, barter is the exchange of goods or services without the involvement of money, under the premise that the two sides coincidentally want each other's offer. As a means of trading, barter is now rare, but is historically significant for its role in enabling the division of labor. In early societies, barter would typically take place between partners who were not very familiar to one another \[[@B1-cancers-11-00117]\], perhaps because a complete and immediate transaction was a way of diffusing concerns about the partner's trustworthiness at a time when regulated systems of credit were not available \[[@B2-cancers-11-00117]\]. The introduction of money and credit systems transformed trading and allowed societies to accrue considerable wealth. A biological cell hosts a collection of specialized processes that inherently demonstrate division of labor. Many of these processes require an energetic input, supplied in the form of ATP, which serves as a universally-accepted currency of energy. An analogy can be drawn to an advanced economy, where metabolism generates ATP ('money') to power ('pay') various specialized biological processes ('crafts'), so that these collectively support survival and proliferation ('wealth'). Such a system functions properly because it is in the interest of all constituents that the cell thrives. At the multicellular level, specialization occurs among different tissues which are 'paid' by substrates (e.g., glucose) to engage in activities that benefit the whole body. A mechanism of credit (involving signaling and homeostasis) is implemented to ensure that the components of the system do not seek instantaneous returns. Tumors are, however, quite different. Cancer cells have a strong individualistic survival instinct, with little scope for altruistic gestures towards other cells \[[@B3-cancers-11-00117],[@B4-cancers-11-00117]\]; moreover, cancer cells will often parasitically exploit neighboring cells, such as the host stroma \[[@B5-cancers-11-00117],[@B6-cancers-11-00117]\]. Nonetheless, tumors have the potential to benefit from the division of labor because they are *not* homogenous tissues. Their heterogeneity arises from genetic differences among sub-populations of cancer cells, and also from variation in micro-environmental factors, such as oxygen supply which depends on the distance from the nearest capillary \[[@B3-cancers-11-00117]\]. Reaping the benefits of the division of labor among selfish and suspicious partners may require a primitive exchange arrangement, such as barter. The instantaneous and reciprocal exchange of assets between cancer cells is achievable through inter-cellular conduits called gap junctions, first described in the 1960s \[[@B7-cancers-11-00117]\]. 2. Gap Junctions are Conduits for Exchanging Small- to Medium-Sized Molecules {#sec2-cancers-11-00117} ============================================================================= Gap junctions are assembled from the juxtaposition of two hemichannels expressed at the surface of abutting cells \[[@B8-cancers-11-00117]\]. These hemichannels are made of connexin proteins \[[@B9-cancers-11-00117]\] which are coded by genes belonging to a family of at least 24 members \[[@B8-cancers-11-00117],[@B10-cancers-11-00117]\]. The pore that traverses the complete channel offers a low-resistance pathway for exchanging solutes as big as 1 kDa. Essentially all metabolites, second messengers and inorganic ions are within this range of permeating substances, which results in a functional coupling between cells. The overall strength of this coupling depends on the number of gap junctional channels connecting a pair of cells. Cell-to-cell coupling is essential to the function of many types of tissue (e.g., the spread of electrical excitation in the heart \[[@B11-cancers-11-00117]\]) and, not surprisingly, most cells in the body are junctionally-coupled \[[@B12-cancers-11-00117],[@B13-cancers-11-00117]\]. Many normal epithelia express connexins and manifest cell-to-cell coupling, such as Cx26, 32, 43 in pancreatic ducts and Cx26, 32, 43, 45 in colorectal epithelium \[[@B10-cancers-11-00117],[@B14-cancers-11-00117],[@B15-cancers-11-00117]\]. Notable exceptions to the near-ubiquitous expression of connexins are blood cells, as these are stand-alone units, and fully-developed skeletal muscle, because their contractile units must be electrically-isolated to allow selective recruitment for the purpose of force gradation \[[@B16-cancers-11-00117]\]. Historically, this list of exceptions also included tumors, on the basis that early observations suggested a lack of connectivity \[[@B17-cancers-11-00117],[@B18-cancers-11-00117],[@B19-cancers-11-00117]\], subsequently supported by more recent work \[[@B20-cancers-11-00117]\], as well as evidence for a tumor suppressing effect of connexins \[[@B21-cancers-11-00117],[@B22-cancers-11-00117]\]. Contemporary findings have challenged this dogma \[[@B10-cancers-11-00117]\], and it is now recognized that some cancers do express connexins, at least in some stages of progression \[[@B23-cancers-11-00117]\]. There will be a myriad of roles for connexins in cancer, including some that relate more to their involvement in cell-to-cell adhesion \[[@B24-cancers-11-00117],[@B25-cancers-11-00117],[@B26-cancers-11-00117],[@B27-cancers-11-00117]\], than the provision of a meaningful solute exchange pathway. Clinical studies have not been able to ascertain the direction of correlation between connexin expression and prognosis, as both positive and negative associations have been reported \[[@B10-cancers-11-00117]\]. In colorectal cancer, for example, one study linked high Cx26 expression with poor prognosis and lung metastasis \[[@B28-cancers-11-00117]\], whereas a subsequent study \[[@B29-cancers-11-00117]\] came to the opposite conclusion. On the whole, there appears to be an isoform-specific pattern in the association between expression and prognosis, with Cx26 generally indicating worse prognosis and Cx43 linked to better prognosis \[[@B10-cancers-11-00117]\]. A limitation of the aforementioned studies is that they have not determined the extent to which connexin expression produces meaningful coupling between cancer cells, a principal phenotypic readout of connexin function. Indeed, gap junctional communication is not routinely quantified in most studies of connexins in cancer (with a few exceptions, e.g., \[[@B21-cancers-11-00117],[@B25-cancers-11-00117],[@B30-cancers-11-00117],[@B31-cancers-11-00117],[@B32-cancers-11-00117],[@B33-cancers-11-00117],[@B34-cancers-11-00117]\] and our work \[[@B35-cancers-11-00117],[@B36-cancers-11-00117]\]). With the availability of antibodies and primers, recent studies of connexins in cancer cell lines have tended to quantify their levels in terms of mRNA or protein, which cannot faithfully predict functional coupling. Indeed, mislocalization of connexins to regions other than cell-cell contacts has been described for Cx26 in pancreatic \[[@B37-cancers-11-00117]\] and colorectal cancer \[[@B28-cancers-11-00117]\], putatively. Thus, our appreciation of the physiological role of cell-to-cell diffusive coupling in cancer, particularly in vivo, is still very limited. If, however, a specific type of malignancy manifested strong cell-to-cell diffusive coupling, it would be surprising if solute exchange had no bearing on cancer physiology. One method of quantifying the degree of coupling between cells is based on fluorescence recovery after photobleaching (FRAP). A fluorescent dye, which can freely pass gap junctions (i.e., of molecular weight lower than 1 kDa), is loaded into a confluent monolayer; an exemplar colorectal cancer (CRC) cell line is shown in [Figure 1](#cancers-11-00117-f001){ref-type="fig"}. Fluorescence in one cell is then bleached (e.g., to 50% of starting level) using a high-power laser, and the signal time course registered in the bleached cell and its direct neighbors. The recovery time course is then fitted with a mono-exponential, and the initial slope (arrow in [Figure 1](#cancers-11-00117-f001){ref-type="fig"}D) provides a rate of fluorescence recovery, d*F*/d*t*, which is related to: (*i*) permeability, *P*, (*ii*) the difference in fluorescence between bleached and neighboring cells ΔF (i.e., the diffusion gradient), (*iii*) the surface area of the bleached cell at the junction with the neighboring cells, *s*, and (*iv*) the volume of the bleached cell, *v*: The *s*/*v* ratio can be approximated by quotient *ρ* of the bleached cell's perimeter and its area in the xy plane (this assumes constant monolayer height). Thus, permeability, in units of µm/s, is:$$P = \frac{dF/dt}{\Delta F \times \rho}$$ This permeability relates, of course, to the fluorescent marker, which will be bulkier than many endogenous metabolites, signals or ions. Smaller molecules are thus expected to permeate faster by an amount determined by their size, charge and interactions with the pore area \[[@B38-cancers-11-00117],[@B39-cancers-11-00117]\]. At first sight, solute dissipation across a syncytium of cells may appear to be the antithesis of what individual cancer cells need to compete in the process of somatic evolution because diffusion will blur differences in metabolic, signaling and ionic phenotype between connected cells. In late-stage disease, genetic instability and hyper-proliferation result in clonal heterogeneity \[[@B3-cancers-11-00117]\], but since somatic evolution selects phenotype *not* genotype, the substrate for selection will become less diverse in connected cancer cells \[[@B4-cancers-11-00117],[@B40-cancers-11-00117]\]. The syncytial behaviors of these units limit cell autonomy and constrain cell-to-cell variation, a fundamental determinant of evolutionary change according to game theory \[[@B41-cancers-11-00117]\]. Indeed, mathematical models of somatic evolution routinely consider each cancer cell as a unit \[[@B4-cancers-11-00117]\]. Considering these potential drawbacks, there must be some overriding benefit that intracellular coupling could offer tumors. A plausible beneficiary of coupling is metabolism and pH regulation, two intertwined processes that are fundamental to cancer growth and generate large fluxes of solutes \[[@B42-cancers-11-00117],[@B43-cancers-11-00117]\]. There have been reports for metabolic cooperation between cancer cells, in which substrates are made available to those cells that cannot synthesize them \[[@B44-cancers-11-00117]\]. More recently, we have demonstrated bulk fluxes of lactate \[[@B36-cancers-11-00117]\] and HCO~3~^−^ ions \[[@B35-cancers-11-00117]\] across gap junctions in spheroids grown form pancreatic ductal adenocarcinoma (PDAC) cell lines. 3. Glycolytic Metabolism Establishes Radial Concentration Gradients of Lactate and Bicarbonate in Solid Tumors {#sec3-cancers-11-00117} ============================================================================================================== Cancer cells require a steady flow of ATP to service the demands placed by proliferation, invasion and metastasis. Metabolic reprogramming to a glycolytic phenotype is common in many cancer cells (Warburg effect) \[[@B42-cancers-11-00117],[@B43-cancers-11-00117],[@B45-cancers-11-00117],[@B46-cancers-11-00117]\], although a degree of mitochondrial respiration will also take place if oxygen is adequate \[[@B43-cancers-11-00117],[@B47-cancers-11-00117]\]. The end-product of glycolytic metabolism is the ionized form of lactic acid \[[@B48-cancers-11-00117]\], which must be removed from the cell, particularly if there is insufficient mitochondrial capacity to metabolize lactate further in the Krebs cycle. Typically, cancer cell lines produce lactate at a rate of several millimoles per minute per liter of cell volume (mM/min) \[[@B49-cancers-11-00117],[@B50-cancers-11-00117],[@B51-cancers-11-00117],[@B52-cancers-11-00117],[@B53-cancers-11-00117]\], but cancer cell lines can attain rates as high as 20 mM/min \[[@B54-cancers-11-00117]\]. The sole means by which lactate can cross biological membranes is in co-transport with H^+^ ions aboard H^+^-monocarboxylate transporter (MCTs) \[[@B55-cancers-11-00117],[@B56-cancers-11-00117]\] or, to a much lesser degree, as lactic acid crossing the lipid bilayer; either way, membrane permeation is electroneutral. Once outside the cell, lactic acid dissociates, and its ionic products diffuse towards blood capillaries to complete the process of venting ([Figure 2](#cancers-11-00117-f002){ref-type="fig"}A). Membrane permeation and extracellular diffusion can be considered as resistors in series. For cancer cells that are deep at the hypoxic core of solid tumors, the diffusive resistance will be substantial, and no amount of MCT expression could overcome this. Rate-limiting removal of the products of glycolytic metabolism will lead to their retention at the hypoxic core. The extent of this can be simulated mathematically ([Figure 2](#cancers-11-00117-f002){ref-type="fig"}B,C). The equations used in the model are presented in the [Appendix A](#app1-cancers-11-00117){ref-type="app"}, together with an explanation of the variables and assumptions. For illustrative purposes, we assume the extracellular space occupies 25% of tumor volume, has an intrinsic buffering capacity of 3 mM/min, and its tortuosity reduces ionic diffusivities by a factor of two (\[[@B57-cancers-11-00117],[@B58-cancers-11-00117],[@B59-cancers-11-00117],[@B60-cancers-11-00117]\]). Thus, extracellular lactate, HCO~3~^−^ and CO~2~ diffusion coefficients were 650, 900 and 1200 µm^2^/s, respectively. A similar tortuosity is modelled for the intracellular space; and the apparent H^+^ diffusion coefficient taken as 200 µm^2^/s \[[@B61-cancers-11-00117]\]. MCT activity in cells, radius 7 µm and intracellular intrinsic buffering capacity of 15 mM/pH \[[@B62-cancers-11-00117]\], was set to a high level, equivalent to a permeability to lactic acid of 10^3^ µm/s, and lactic acid production fixed at 20 mM/min \[[@B54-cancers-11-00117]\]. CO~2~/HCO~3~^−^ buffer is included in the modelling, with freely-permeating CO~2~ (permeability constant 10^3^ µm/s) \[[@B57-cancers-11-00117]\], and to facilitate its chemical reactions, exofacial carbonic anhydrase (CA) activity increased spontaneous CO~2~ hydration rate by a factor of 10 (e.g., CAIX/CAXII) \[[@B59-cancers-11-00117],[@B60-cancers-11-00117]\]. The geometry of the modelled tissue is assumed spherical of radius 250 µm to encompass both normoxic regions at the periphery and hypoxic areas at the core. Concentrations at the boundary are set to levels normally found in blood, i.e., CO~2~ at 1.2 mM, HCO~3~^−^ at 24 mM, hence extracellular pH (pH~e~) at 7.4, and negligible lactate. In the model, cells at the tissue boundary were able to fully metabolize lactate (i.e., intracellular lactate is set to nil) and import HCO~3~^−^ ions by active transport (i.e., intracellular \[HCO~3~^−^\] set at 13.5 mM). Lactic acid produced glycolytically by cancer cells ('source') is vented to the boundary of the tissue ('sink'), down the route of least resistance. Essentially all traffic involves permeation across the membrane of the glycolytic cell, followed by a meandering across the extracellular compartment. Results of the simulation indicate substantial gradients of lactate in both the intra- and extracellular spaces at the steady-state. Retention of lactate is also linked to intra- and extracellular acidity, and hence a radial gradient of \[HCO~3~^−^\]. This gradient is notably smaller than the lactate gradient because CO~2~/HCO~3~^−^ is not the only acid-neutralizing buffer present inside cells; a sizable pool of buffering is provided by intrinsic buffers (15 mM/pH). The substantial build-up of intra- and extracellular lactate, and the depletion of HCO~3~^−^ at core regions is consistent with the retention of lactic acid in tumors in vivo \[[@B63-cancers-11-00117]\]. This will likely inhibit glycolysis by the allosteric inhibitory effect of H^+^ ions of glycolytic enzymes (e.g., phosphofructokinase \[[@B64-cancers-11-00117]\]), and trans-inhibition by lactate. Such a slowing of energy provision would be an Achilles heel to proliferating cancer cells. In response to this challenge, cancer cells are required to regulate their pH, and the canonical mechanism involves active transport \[[@B65-cancers-11-00117]\]. An alternative means of reducing lactic acid build-up, without compromising metabolic rate, is to increase the diffusive capacity for venting lactate and HCO~3~^−^ ions. 4. Cancer Cells Can Exchange Lactate and Bicarbonate Ions through Gap Junctions {#sec4-cancers-11-00117} =============================================================================== In the modeled scenario shown in [Figure 2](#cancers-11-00117-f002){ref-type="fig"}A, lactate (and HCO~3~^−^) can only diffuse radially along the extracellular path. In this system, the overall diffusivity for lactate is low, and a large radial gradient is needed to drive an efflux that will balance metabolic production. Transport capacity in the extracellular space is limited because this is the smaller of the two tissue compartments, and after accounting for tortuosity, the diffusion coefficients of small solutes, like lactate and bicarbonate, are of the order of \~10^3^ µm^2^/s. Lactate and bicarbonate ions are also transported in a chemically converted form (lactic acid and CO~2~) but this will be a smaller component of flux by \>3 and \>1 orders of magnitude, respectively. Any means of increasing effective diffusivity would benefit venting, and one way of achieving this is to engage in diffusion across the intracellular syncytium \[[@B35-cancers-11-00117],[@B36-cancers-11-00117]\]. The substantial radial gradient of lactate in the intracellular space provides the energy to drive a dissipating flux, but only if there is sufficient coupling between cells ([Figure 3](#cancers-11-00117-f003){ref-type="fig"}A). On the basis of FRAP experiments, junctionally-coupled CRC and PDAC monolayers can have a permeability constant to calcein of the order of 0.03 µm^2^/s \[[@B36-cancers-11-00117]\]. Anions such as lactate and bicarbonate are smaller than calcein (7- and 10-fold respectively), thus their permeability can be expected to be two orders of magnitude higher \[[@B38-cancers-11-00117],[@B39-cancers-11-00117],[@B66-cancers-11-00117]\], i.e., in the order of \~2 and \~3 µm/s, respectively. The activity of MCT may produce lactic acid permeability as high as 10^3^ µm^2^/s, but this refers to the transport of lactic *acid*, which is \>1000-times less abundant than its ionized conjugate. Lactate transport across the syncytial cytoplasmic volume could be described with an effective diffusion coefficient, D^eff^, given by Equation (3), where D is the diffusion coefficient between permeation barriers of permeability constant P, spaced at intervals d, i.e., approximated by cell diameter:$$\frac{1}{D^{eff}} = \frac{1}{D} + \frac{1}{d \times P}$$ Small solutes like lactate may be expected to have effective diffusion coefficients in the cytoplasmic syncytium of \~30 µm^2^/s. Although this is more than an order of magnitude lower than diffusivity in the extracellular space, the ensuing intracellular flux can be meaningful because of the large volume of the intracellular compartment. Early studies have shown that low pH~i~ uncouples gap junctions \[[@B67-cancers-11-00117],[@B68-cancers-11-00117]\], but more recent work has shown that gap junctions made of Cx43, for example, have a pH optimum that is slightly acidic relative to resting pH~i~ \[[@B69-cancers-11-00117],[@B70-cancers-11-00117]\]; consequently, coupling by Cx43 gap junctions is expected to strengthen as pH~i~ falls modestly, before cells become uncoupled at unphysiologically low pH~i~. Adding junctional lactate permeability to the mathematical model ([Figure 3](#cancers-11-00117-f003){ref-type="fig"}A) significantly improves lactic acid venting and reduces lactate build-up at the core ([Figure 3](#cancers-11-00117-f003){ref-type="fig"}B). However, the diffusive dissipation of lactate, a basic substance, will have an effect on pH \[[@B36-cancers-11-00117]\]. At steady-state, lactate-only junctional efflux leaves behind H^+^ ions that acidify the intracellular space at the core of the tissue ([Figure 3](#cancers-11-00117-f003){ref-type="fig"}D). This generates a substantial radial gradient of \[HCO~3~^−^\] ([Figure 3](#cancers-11-00117-f003){ref-type="fig"}C). To prevent this unwarranted acidification, junctions must also conduct HCO~3~^−^ ions, which would be expected to have a D^eff^, in this example, of 40 µm^2^/s. As shown in the results of a model with junctional lactate and HCO~3~^−^ permeability ([Figure 4](#cancers-11-00117-f004){ref-type="fig"}A), lactate build-up at the core can be reduced by a fifth ([Figure 4](#cancers-11-00117-f004){ref-type="fig"}B), without significantly acidifying the core ([Figure 4](#cancers-11-00117-f004){ref-type="fig"}C), thanks to a concurrent dissipation of any \[HCO~3~^−^\] gradient that develops in response to lactate transport. 5. Concluding Remarks {#sec5-cancers-11-00117} ===================== In this article, we describe a metabolic function for gap junctional coupling between cancer cells in solid tumors. Gradients of hypoxia in solid tumors also produce gradients of other solutes, such as lactate and HCO~3~^−^. These gradients are the driving forces for diffusion, but transport is only possible if there is adequate conductance. If this were to take place in the intracellular compartment of tumors, it must involve an adequate degree of coupling by gap junctions. In at least some cancers, such coupling is possible if connexin proteins are correctly targeted to points of cell-cell contact. Once a low-resistance pathway is established, it can lead to the diffusive exchange of solutes. Quantitatively the largest fluxes will involve those substances that have large gradients, such as lactate and HCO~3~^−^ anions. A lactate/HCO~3~^−^ exchange could benefit both of the coupled cells, as illustrated in [Figure 5](#cancers-11-00117-f005){ref-type="fig"}. For the more hypoxic cell, gap junctions offer a route for lactate efflux, which reduces the retention of this glycolytic end-product. This was shown experimentally in spheroids of Colo357 cells, a Cx43-coupled PDAC cell line \[[@B36-cancers-11-00117]\]. The recipient (more normoxic) cell can benefit from this lactate delivery by obtaining ATP from the mitochondrial respiration of this substrate \[[@B71-cancers-11-00117]\]. Any excess lactate that is not respired can leave the normoxic cell by MCT, which is a means of removing H^+^ ions, another benefit to the recipient shown experimentally in PDAC spheroids \[[@B36-cancers-11-00117]\]. Efflux of lactate, a base, from the hypoxic cell could potentially exacerbate its intracellular acidosis, but this is curtailed by HCO~3~^−^ delivery from the more alkaline normoxic cell, where HCO~3~^−^ is more plentiful. To sustain HCO~3~^−^ delivery, the normoxic cell can engage in active uptake, e.g., by Na^+^-driven HCO~3~^−^ cotransporters of the *SLC4* gene family \[[@B35-cancers-11-00117]\]. From the viewpoint of the hypoxic cell, this active uptake mechanism is an example of remote secondary active transport that we described in PDAC spheroids \[[@B35-cancers-11-00117]\]. Under this arrangement, the energy cost for delivering HCO~3~^−^ to the hypoxic cell is paid by the normoxic cell. In this example of barter, junctionally coupled cancer cells can benefit from the instantaneous and complete ionic trading. We acknowledge the many useful discussions with Tobias Dovmark, Alzbeta Hulikova and Sir Walter Bodmer. This research was funded by the European Research Council, SURVIVE \#723997. The author declares no conflict of interest. The concentration dynamics of intra- and extracellular solutes were modelled with a diffusion-reaction partial differential equation \[[@B57-cancers-11-00117]\] in a sphere of radius *R* = 250 µm, which represents a tumor mass that is sufficient for hosting radial gradients:$$\frac{\partial u}{\partial t} = D \cdot \frac{1}{r^{2}} \cdot \frac{\partial}{\partial x}\left( {r^{2} \cdot \frac{\partial u}{\partial x}} \right) + F\left( u \right)$$ Variable *r* is the radial distance from the core, *u* is a vector of concentrations, *D* is a vector of effective diffusion coefficients, and *F* is a function describing chemical reactions and membrane permeation. Vector *u* describes concentrations in the following order: (1) \[H^+^\]~i~, (2) \[HCO~3~^−^\]~i~, (3) \[CO~2~\]~i~, (4) \[Lac^−^\]~i~, (5) \[HLac\]~i~, (6) \[H^+^\]~e~, (7) \[HCO~3~^−^\]~e~, (8) \[CO~2~\]~e~, (9) \[Lac^−^\]~e~, (10) \[HLac\]~e~, where subscripts *i* and *e* refer to the intra- and extracellular compartments. The initial concentrations, *u^0^*, were 10^−7.2^ M, 13.5 mM, 1.2 mM, 0 mM, 0 mM, 10^−7.4^ M, 22 mM, 1.2 mM, 0 mM, 0 mM. The fractional volume of the combined intracellular space *v~i~* was 0.75. Cells were assumed to have a radius of 7 µm and thus a surface area/volume radio *ρ* of 3/7 µm^−1^. Intracellular \[H^+^\] reactions comprised CA-catalyzed reversible CO~2~ hydration, reversible lactic acid dissociation, and buffering by intracellular intrinsic buffers:$$\frac{\partial u_{1}}{\partial t} = \frac{CA_{i} \cdot \left( {k_{h} \cdot u_{3} - k_{r} \cdot u_{1} \cdot u_{2}} \right) + k_{d} \cdot u_{5} - k_{p} \cdot u_{1} \cdot u_{4}}{1 + \beta_{i}/\left( {\ln\left( 10 \right) \cdot u_{1}} \right)}$$ The spontaneous CO~2~ hydration constant *k~h~* was set to 0.18 s^−1^. The rate constant for the reverse reaction (*k~r~*) was thus *k~h~/K~CO~2~~*, where *K~CO~2~~* is the equilibrium constant equal to 10^−6.15^ M. Intracellular CA (*CA~i~*) activity was set to 10-fold to provide fast reaction kinetics. Lactate protonation rate constant *k~p~* was set to a fast value, 10^−10^ M^−1^ s^−1^, and the reverse rate (*k~d~*) equaled *k~p~*×*K~lac~*, where *K~lac~* is the lactic acid dissociation constant, 10^−3.86^ M. Intracellular buffering capacity *β~i~* was set to 15 mM/pH. Intracellular HCO~3~^−^ reactions comprised CA-catalyzed reversible CO~2~ hydration:$$\frac{\partial u_{2}}{\partial t} = CA_{i} \cdot \left( {k_{h} \cdot u_{3} - k_{r} \cdot u_{1} \cdot u_{2}} \right)$$ Intracellular CO~2~ reactions comprised CA-catalyzed reversible CO~2~ hydration, and diffusive entry from the extracellular compartment:$$\frac{\partial u_{3}}{\partial t} = CA_{i} \cdot \left( {k_{r} \cdot u_{1} \cdot u_{2} - k_{h} \cdot u_{3}} \right) + ~\rho \cdot P_{CO2} \cdot \left( {u_{8} - u_{3}} \right)$$ Membrane CO~2~ permeability *P~CO~2~~* was set to a value that supported rapid exchange, 10^3^ µm/s. Intracellular lactate reactions comprised reversible lactic acid dissociation:$$\frac{\partial u_{4}}{\partial t} = k_{d} \cdot u_{5} - k_{p} \cdot u_{1} \cdot u_{4}$$ Intracellular lactic acid reactions comprised reversible lactic acid dissociation, glycolytic lactic acid production throughout the structure except in the outer rim of thickness *δ*, and MCT-facilitated entry form the extracellular compartment:$$\frac{\partial u_{5}}{\partial t} = k_{p} \cdot u_{1} \cdot u_{4} - k_{d} \cdot u_{5} + J_{lac} \cdot \left( {1 - exp\left( {- \frac{R - r}{\delta}} \right)} \right) + \rho \cdot P_{mct} \cdot \left( {u_{10} - u_{5}} \right)$$ The effective membrane lactic acid permeability *P~mct~* was set to a value that supported fast exchange, 10^3^ µm/s, consistent with a high level of MCT activity. The metabolic rate *J~lac~* was set to 20 mM/min. Thickness of the rim δ was set to 10 µm. Extracellular \[H^+^\] reactions comprised CA-catalyzed reversible CO~2~ hydration and reversible lactic acid dissociation, and buffering by extracellular intrinsic buffers:$$\frac{\partial u_{6}}{\partial t} = \frac{CA_{e} \cdot \left( {k_{h} \cdot u_{8} - k_{r} \cdot u_{6} \cdot u_{7}} \right) + k_{d} \cdot u_{10} - k_{p} \cdot u_{6} \cdot u_{9}}{1 + \beta_{e}/\left( {\ln\left( 10 \right) \cdot u_{6}} \right)}$$ Extracellular CA (*CA~e~*) activity was set to 10-fold to provide fast reaction kinetics. Extracellular buffering capacity *β~e~* was set to 3 mM/pH. Extracellular HCO~3~^−^ reactions comprised CA-catalyzed reversible CO~2~ hydration:$$\frac{\partial u_{7}}{\partial t} = CA_{e} \cdot \left( {k_{h} \cdot u_{8} - k_{r} \cdot u_{1} \cdot u_{2}} \right)$$ Extracellular CO~2~ reactions comprised CA-catalyzed reversible CO~2~ hydration and diffusive entry from the intracellular compartment, with consideration of the intra-to-extracellular volume ratio:$$\frac{\partial u_{8}}{\partial t} = CA_{e} \cdot \left( {k_{r} \cdot u_{6} \cdot u_{7} - k_{h} \cdot u_{8}} \right) + ~\frac{v_{i}}{1 - v_{i}} \cdot \rho \cdot P_{CO2} \cdot \left( {u_{3} - u_{8}} \right)$$ Extracellular lactate reactions comprised reversible lactic acid dissociation $$\frac{\partial u_{9}}{\partial t} = k_{d} \cdot u_{10} - k_{p} \cdot u_{6} \cdot u_{9}$$ Extracellular lactic acid reactions comprised reversible lactic acid dissociation, and MCT-facilitated entry form the intracellular compartment, with consideration of the intra-to-extracellular volume ratio:$$\frac{\partial u_{10}}{\partial t} = k_{p} \cdot u_{6} \cdot u_{10} - k_{d} \cdot u_{9} + \frac{v_{i}}{1 - v_{i}} \cdot \rho \cdot P_{mct} \cdot \left( {u_{5} - u_{10}} \right)$$ Diffusion coefficients for extracellular H^+^, HCO~3~^−^, CO~2~, Lac and HLac were set to half of their values in aqueous media to account for a two-fold tortuosity imposed by the extracellular space, i.e., 6000, 900, 1200, 650 and 650 µm^2^/s, respectively. The effective intracellular diffusion coefficients, as defined by Equation (3), were set to values that depended on the modelling scenarios. In the absence of junctional coupling, as depicted in [Figure 2](#cancers-11-00117-f002){ref-type="fig"}A, intracellular diffusivity for ions was set to zero, and 1200 µm^2^/s in the case of membrane-permeable CO~2~. In the modeling scenario shown in [Figure 4](#cancers-11-00117-f004){ref-type="fig"}A, gap junctions were permeable to all ionic species. In this case, the permeability to HCO~3~^−^, lactate and lactic acid were in proportion to their diffusivity and set to 3, 2.17 and 2.17 µm/s, giving an effective permeability of 40, 29 and 29 µm^2^/s. The diffusive properties of H^+^ ions are unique, as they are set by mobile buffers \[[@B72-cancers-11-00117]\]. Assuming a mobile buffer diffusivity of 200 µm^2^/s, consistent with being larger molecules than lactate, and a permeability of 0.67 µm/s, their effective diffusivity was 8.9 µm^2^/s. In the case depicted in [Figure 3](#cancers-11-00117-f003){ref-type="fig"}A, gap junctions were permeable to lactic acid and lactate only, but not H^+^ or HCO~3~^−^ ions. The boundary conditions at *r* = *R* for extracellular solutes were set to provide direct coupling with a well-stirred compartment representing blood flow, i.e., concentrations were clamped to levels defined in the initial conditions vector *u^0^*. Similarly, the boundary conditions for intracellular solutes was set to values defined in *u^0^* to simulate a scenario of good pH~i~ regulation (and hence constancy of \[HCO~3~^−^\]) and zero lactate (i.e., metabolic degradation of any incoming lactate ions). Thus, boundary conditions took the general form:$$\left. u_{i} \right|_{r = R} = u_{i}^{0}$$ The equations were solved with the pdepe solver (MATLAB) and presented as radial profiles once the system attained a steady-state. {#cancers-11-00117-f001} {#cancers-11-00117-f002} {ref-type="fig"}. Note that diffusion of lactate and lactic acid can take place across the intra- and extracellular spaces. Junctional efflux of lactate (the major permeating species) leaves behind H^+^ ions, and hence a depletion of HCO~3~^−^; this generates a large radial gradient of \[HCO~3~^−^\].](cancers-11-00117-g003){#cancers-11-00117-f003} {ref-type="fig"}. In this system, intracellular lactate diffuses radially out of the core, leaving behind H^+^ ions. This generates an inverse gradient of intracellular \[HCO~3~^−^\], which drives a counter-flux of HCO~3~^−^ i.e., junctional lactate/HCO~3~^−^ exchange. Overall, lactate venting improves by a fifth.](cancers-11-00117-g004){#cancers-11-00117-f004} {#cancers-11-00117-f005}

1. Introduction =============== Longitudinally extensive transverse myelitis (LETM) is characterized by contiguous inflammatory lesions of spinal cord extending to ≥3 vertebral segments.^\[[@R1]\]^ The causes of LETM including various infections, neoplastic reason, and autoimmune disease.^\[[@R2]\]^ Neuromyelitis optica spectrum disorder (NMOSD) is the most common cause of LETM. In recent years, more and more cases of pulmonary tuberculosis (PTB) patients with optic myelitis symptoms were published,^\[[@R3]--[@R5]\]^ suggesting the potential link between mycobacterium tuberculosis infection and NMOSD.^\[[@R6],[@R7]\]^ A retrospective case-control study conducted by Zatjirua et al^\[[@R8]\]^ in South Africa, showed the odds ratio for the presence of active PTB in the neuromyelitis optica (NMO) group versus the control group was 4.6. Moreover, a positive correlation between the activity of PTB and the occurrence of NMO has been found. We report 2 cases of LETM concurrent with PTB. The first case was an anti-aquaporin-4 antibody (AQP4-Ab) positive NMOSD patient with a past history of PTB. The second case was anti-AQP4-Ab negative LETM patient with active PTB. 2. Case presentation ==================== 2.1. Case 1 ----------- A 20-year-old woman was admitted to our hospital for weakness, numbness, and pain in the limbs for 6 months and worsened for 3 days. The patient felt persistent pain in the neck, left shoulder, and the upper left arm 6 months before, and the situation became worse and both upper extremities were affected. The patient visited the local hospital and was diagnosed as myelitis and she received gamma globulin (20 g for 5 days) and dexamethasone (5 mg for 15 days). Her symptoms were improved gradually. However, 1 month later, the weakness, persistent numbness, and pain in both lower extremities occurred again and gradually worsened with time. Therefore, the patient returned to the local hospital, and received vitamin B12 treatment, her symptom became slightly better but not completely recover. Six days before the admission to our hospital, the patient had symptoms of sneezing nose. Three days later, the patient had severe weakness and numbness in the lower extremities. Her past medical history was otherwise negative except for PTB for 6 years. At that time, the patient received systematic anti-tuberculosis treatments and her symptoms were completely disappeared and chest computed tomography (CT) showed chronic pulmonary tuberculosis. Neurological examination revealed decreased muscle strength of both lower extremities (1/5 of the right extremity and 3/5 of the left extremity according to the Medical Research Council \[MRC\] grade) with bilateral hyperreflexia. Bilateral Babinski sign were positive. Hyperalgesia was found below the level of T8, deep sensation of the right lower limb was absent. The Expanded Disability Status Scale (EDSS) of the patient during hospitalization was 6.5. The magnetic resonance imaging (MRI) scan of the cervical spine (6 months before admission) revealed cervical spinal cord lesions without gadolinium enhancement (Fig. [1](#F1){ref-type="fig"}). The serum anti-AQP4-Ab was strongly positive, and the Immunoglobulin G (IgG) oligoclonal bands were negative both in the serum and in the cerebrospinal fluid (CSF). CSF routine test showed elevated protein levels (0.39 g/L, normal range from 0.15 to 0.45 g/L) and increased cell numbers (leukocyte count 25 × 10^6^/L, normal range from 0 to 8 × 10^6^/L), of which 92% were monocytes. Autoimmune diseases related test including anti-nuclear antibodies, anti-dsDNA, anti-streptolysin, rheumatoid factor, thyroid function tests, erythrocyte sedimentation rate (ESR), and hypersensitive C-reactive protein were all negative. T cell spot tuberculosis (TB) testing was positive. No tubercula bacilli were found in sputum TB smear. Anti-TB antibodies of Lipoarabinomannan-IgG and 38 KD/IgG in the serum were positive. Quantitative detection of Brucella polymerase chain reaction was negative in the CSF. Brain MRI after admission demonstrated multiple patchy abnormal signal in pons, the right thalamus, the left basal ganglia, corona radiate and corpus callosum on fluid-attenuated inversion recovery (FLAIR) imaging, diffusion-weighted imaging (DWI), and T2-weighted (T2W) imaging (Fig. [2](#F2){ref-type="fig"}). The second MRI scan of cervical and thoracic spinal cord after admission to the hospital revealed hyperintensity lesions extending from C3 to T8 on T2W image, T1-weighted (T1W) image, and FLAIR image (Fig. [3](#F3){ref-type="fig"}A--C). Chest CT during hospitalization showed left lung upper lobe bronchiectasis and multiple calcifications of upper lobe in both sides (Fig. [3](#F3){ref-type="fig"}D). The patient was treated with methylprednisolone of 500 mg for 3 days, afterwards, the dose of methylprednisolone was decreased to half every 3 days, meanwhile she also received gamma globulin and anti-tuberculosis drugs (isonicotinyl hydrazide 0.3 g/d, rifampin 0.45 g/d, pyrazinamide 1 g/d, and ethambutol 1 g/d). The patient discharged from hospital after 1 months' treatment. The EDSS was decreased to 0 at the time point of follow up 12 months later. The spinal cord MRI showed the lesion size in the cervical and thoracic spine were significantly reduced (Fig. [4](#F4){ref-type="fig"}). {#F1} {#F2} {#F3} {#F4} 2.2. Case 2 ----------- A 21-year-old woman was admitted to our hospital for numbness in bilateral lower limbs and in the chest for 7 days. She had a past medical history of PTB without receiving any treatment for 3 years. On neurological examination, the muscle strength in the upper limbs and lower limbs were 5/5 according to the MRC grade. Hyperreflexia was found in bilateral lower limbs. She was found loss of sensation of pain and touch in the spinal segments below T4. Bilateral Babinski sign were positive. Four days after admission, the muscle strength in bilateral lower limbs were 4/5 according to the MRC grade, with retention of urine. The EDSS of the patient was 4.5 during hospitalization. The MRI scan during hospitalization of cervical and thoracic spinal cord revealed diffuse hyperintense signal extending C3 to T11 on T2W and FLAIR images and hypointense signal on T1W images (Fig. [5](#F5){ref-type="fig"}A--C). The MRI of the thoracic spinal cord revealed altered medullary signal intensity from C6 to T11 levels, appearing hyperintense on T1W and T2W images (Fig. [5](#F5){ref-type="fig"}B). The chest CT showed left upper lobe tuberculosis with cavitation (Fig. [5](#F5){ref-type="fig"}D). CSF routine test showed increased protein levels (0.48 g/L, normal range from 0.15 to 0.45 g/L). The CSF cell numbers were within the normal range (4 × 10^6^/L). The anti-AQP4-Ab was negative both in the serum and in the CSF of the patient. Tuberculosis mycobacterium antibody was negative both in serum and CSF. A combination of corticosteroid and anti-tuberculosis treatment (isonicotinyl hydrazide 0.3 g/d, rifampin 0.45 g/d, pyrazinamide 1 g/d, ethambutol 1 g/d, and dexamethasone 15 mg/d) was given. Neurological dysfunction was ameliorated significantly. The lesion size in the spinal cord and in the lung were significantly reduced 1 month later after discharge from the hospital. The EDSS of the patient dropped to 0 at the follow up time when 3 months after discharge from the hospital. {#F5} 3. Discussion ============= We herein report 2 cases of LETM coincident with PTB. The first case was an anti-AQP4-Ab positive NMOSD patient who had a past history of PTB. The second case was anti-AQP4-Ab negative LETM patient with active PTB. Both patients' clinical symptoms were improved by the combination of corticosteroid and anti-tuberculosis treatment. Further, the EDSS of both patients decreased to 0 and their lesions size in spinal cord was significantly reduced during the follow-up. In case 1, the CSF IgG synthesis rate, the blood--brain barrier (BBB) permeability, and the levels of myelin basic protein (MBP) were found increased. In case 2, the ESR, the levels of CSF IgG, anti-myelin oligodendrocyte glycoprotein (MOG) antibody, and anti-MBP antibody in the serum were found increased. Increased IgG synthesis rate or IgG index reflects the presence of humoral immune response. The elevated levels of MBP, anti-MBP antibody, MOG, and anti-MOG antibody reflect the inflammatory demyelinating lesions.^\[[@R9]\]^ The CSF changes of 2 cases showed the presence of demyelination in the central nervous system (CNS). Firstly, direct CNS infection with tuberculosis can cause NMOSD, evidenced by a study conducted by Liu et al,^\[[@R10]\]^ where the authors injected Bacillus Calmette-Guerin (BCG) into the lateral ventricle of mice, and showed the production of anti-AQP4-Ab and demyelination in the brain. Secondly, mycobacterium tuberculosis (MTB) may have common antigens with MBP. Significant inflammatory response and demyelination were found after injecting BCG into sensitized guinea pigs.^\[[@R11]\]^ Moreover, lymphocytes sensitized by MTB are able to recognize and attack the myelin,^\[[@R12]\]^ indicating delayed hyper sensitivity reaction caused by MTB may be one cause of demyelination. Interestingly, in a retrospective study of NMOSD patients with PTB, Silber et al^\[[@R3]\]^ found that after giving anti-tuberculosis treatment, the corresponding impaired neurologic functions of patients were significantly improved. Both of the patients' symptoms were improved by anti-tuberculosis and corticosteroid therapy. A case-control study conducted by Feng et al^\[[@R13]\]^ where the cases were treated with anti-tuberculosis therapy, while the control group was treated with corticosteroid, plasmapheresis and immune globulin, and other therapies. The follow-up result showed that the EDSS of patients with anti-tuberculosis treatment decreased gradually, and the recurrence rate was lower than that of the control group. In the early course of disease, the first case had remission despite receiving corticosteroid treatment. Considering her past history of PTB, anti-tuberculosis treatments were given. Her clinical symptoms improved rapidly within 1 month. No recurrence was observed during the follow-up period of 12 months. The MRI lesions in the cervical and thoracic spine were significantly absorbed than before. Our experiences in this case together with Feng\'s findings indicate that anti-tuberculosis treatment can improve neurological damage and reduce the relapse rate of the disease NMOSD with PTB patients. The second case was anti-AQP4-Ab negative LETM patient with active PTB. The patient received corticosteroid treatment combined with anti-tuberculosis treatment for 1 month, and the symptoms were improved dramatically. The lesions in the lung and in the spinal cord were significant absorbed than before. In conclusion, a combination of corticosteroid and anti-tuberculosis treatment has better therapeutic effects for patients who had co-occurrence of LETM and MTB. MTB induced activation of the immune system may contribute to the etiology of LETM. Acknowledgments =============== The authors are grateful for the support from the grants from the National Science Foundation of China (No. 31600820); The Health and Family Planning Commission of Jilin Province (No. 2016Q036); The Youth Foundation from The First Hospital of Jilin University (No.00400050050); The Programme for JLU Science and Technology Innovative Research Team (JLUSTIRT) (No: 451170301647). Abbreviations: AQP4-Ab = aquaporin-4 antibody, BBB = blood--brain barrier, BCG = Bacillus Calmette-Guerin, CNS = central nervous system, CSF = cerebral spinal fluid, CT = computed tomography, DWI = diffusion-weighted imaging, EDSS = Expanded Disability Status Scale, ESR = erythrocyte sedimentation rate, FLAIR = fluid-attenuated inversion recovery, Ig = immunoglobulin, LETM = longitudinally extensive transverse myelitis, MBP = myelin basic protein, MOG = myelin oligodendrocyte glycoprotein, MRC = Medical Research Council, MRI = magnetic resonance imaging, NMOSD = neuromyelitis optic spectrum disorder, PTB = pulmonary tuberculosis, T1W = T1-weighted, T2W = T2-weighted, TB = tuberculosis. YZ and MZ have equally contributed to the work. Ethical Considerations: Ethical approval was obtained from the Ethics Committee from The First Hospital, of Jilin University, China. Though written informed consent was not obtained, the patient\'s information was anonymized and de-identified. The authors declare no conflicts of interest.

Introduction {#Sec1} ============ Contact heterogeneity in infectious disease epidemiology {#Sec2} -------------------------------------------------------- Infectious pathogens that spread via contact between people are a major cause of human disease, driving attempts to understand their epidemiology \[[@CR1]\]. Much theoretical work on infectious disease dynamics has been focused on the role of heterogeneity in the human population \[[@CR2]\], which is often conceptualised as a network of epidemiologically relevant contacts \[[@CR3]--[@CR5]\]. Perhaps the most important quantity in any infectious disease outbreak is the basic reproductive ratio, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$R_{0}$\end{document}$, which is defined verbally as the expected number of secondary cases generated by an average primary case early in the epidemic. An epidemic is possible exactly when $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$R_{0}>1$\end{document}$, and typically the efforts required to control such an outbreak grow monotonically with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$R_{0}$\end{document}$ \[[@CR1], [@CR2]\]. In the simplified scenario where each individual picks an integer number of contacts *K* from the same *degree distribution*, but transmission is otherwise homogeneous, $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ R_{0} \propto\mathbf{E}\bigl[K^{2}\bigr] . $$\end{document}$$ This dependence of the basic reproductive ratio on the second moment of the degree distribution has been a 'textbook' result for some time \[[@CR6]\], however work by Pastor-Satorras and Vespignani \[[@CR7]\] and May and Lloyd \[[@CR8]\] raised the question of what might happen for large, or asymptotically divergent, second moments. Such questions can be posed and answered at different levels of mathematical rigour \[[@CR9]\] however in the context of epidemiology it is clear that a highly variable degree distribution is associated with the epidemiologically unrealistic scenario that even the most weakly transmissible pathogen can cause an epidemic, and as a corollary that control of any infectious disease through untargeted vaccination would be impossible. Data {#Sec3} ---- Of course, whether such a theoretical possibility matters for the study of infectious diseases depends on the actual variance in degree for epidemiologically relevant contacts. While 20th century models of infectious disease were often based on strong a priori assumptions about mixing patterns \[[@CR1]\], various methods for measurement of contact patterns now exist and were reviewed by Read et al. \[[@CR10]\]. As well as direct measurement of individuals through surveys \[[@CR11]\] it is possible to improve coverage through snowball and respondent-driven sampling \[[@CR12], [@CR13]\], to make use of the extremely large datasets produced by electronic sensors \[[@CR14], [@CR15]\], and also to combine aggregate data \[[@CR16], [@CR17]\]. Previous empirical studies have seen evidence that for direct (e.g. \[[@CR16], [@CR18]\]) and sexual (e.g. \[[@CR19], [@CR20]\]) contacts, an approximate power-law relationship may hold such that for large *k*, a randomly selected node obeys $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \operatorname{Pr}(\text{node has } k \text{ links}) \approx k^{-\gamma} . $$\end{document}$$ As is the case for almost all biological data, there is much more complexity in the data than such a simple parametric relationships as ([2](#Equ2){ref-type=""}) would imply. For example, Leigh Brown et al. \[[@CR20]\] found that while a power-law was a better functional form than the negative binomial for sexual contacts, the richer Waring distribution was preferable to either. What is hard to dispute, however, is that as better quality data on epidemiologically relevant contacts is obtained the evidence consistently points to a very high level of variance; for example, Read et al. \[[@CR21]\] were able to validate the high numbers of contacts reported by some study participants through direct (rather than postal) surveying. These empirical observations of high heterogeneity in contact number, together with theoretical results about $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$R_{0}$\end{document}$, present a paradox for infectious disease epidemiology: is the extreme heterogeneity in observed contact patterns indicative of PA and does that imply that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$R_{0}>1$\end{document}$ for almost any finite level of person-to-person transmissibility meaning that our theoretical understanding of infectious disease epidemiology is somehow severely lacking? Preferential attachment and power laws in empirical data {#Sec4} -------------------------------------------------------- Recent years have seen a debate about the level of heterogeneity that exists in a variety of observed networks. A particularly influential paper by Barabási and Albert \[[@CR22]\] considered a model of network formation in which many new nodes are added to a small existing network. These new nodes connect preferentially to nodes that have more links in the existing network, leading to the asymptotic result ([2](#Equ2){ref-type=""}) with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\gamma=3$\end{document}$. In this way preferential attachment is intimately linked with, but not always equivalent to, asymptotic power-law behaviour. Simple power-law relationships have been claimed for numerous real-world systems, and a critical review of these claims by Clauset et al. \[[@CR23]\] used maximum-likelihood fitting of distribution tails to power-law distributions to show varying levels of statistical support for claims in the literature. In the context of discrete data, pioneering work by Zipf \[[@CR24]\] found power-laws in word frequencies; considering the count of unique words in *Moby Dick* both Newman \[[@CR25]\] and Clauset et al. \[[@CR23]\] agree that the statistical evidence for Zipf's power-law distribution in this context is strong. On the other hand, the in- and out-degrees of *E. coli* metabolic networks have been claimed to follow a power law \[[@CR26]\], but this is disputed by the analyses of Huss and Holme \[[@CR27]\] and Clauset et al. \[[@CR23]\]. The debate around presence or absence of power laws in real data continues, perhaps most strongly in the context of networks. For example, Barabási \[[@CR28]\] writes that preferential attachment is network science's "most profuse concept," and that "the impact of preferential attachment is hard to miss." At the same time, Stumpf and Porter \[[@CR29]\] argue that "most reported power laws lack statistical support and mechanistic backing." Testing preferential attachment directly {#Sec5} ---------------------------------------- In this work, we attempt to test the hypothesis of preferential attachment in social contact data directly, rather than via asymptotic power law behaviour. We make use of previously collected data on social encounters specifically designed to measure heterogeneity in numbers of contacts amongst the British population, and fit mechanistic models of different complexity to these data. We determine that models with significant levels of preferential attachment have better evidential support from the data than models without. Methods {#Sec6} ======= Social Contact Survey data {#Sec7} -------------------------- A cross-sectional study was conducted between May 2009 and October 2010, recruiting households and individuals through postal and online questionnaires, supported by a large random-address mailshot and a modest online and media promotion \[[@CR30], [@CR31]\]. Questionnaires asked respondents to report on the number of distinct individuals they encountered the previous day: their contacts. Respondents were able to report contacts either as individuals or as members of a group with a reported size. Allowing the reporting of groups of individuals was a deliberate methodological design to permit the easy reporting of large numbers of contacts, to avoid the approach taken by previous studies \[[@CR11]\], which imposed a high burden on respondents with large number of contacts, and to ensure the best capture of the right-hand tail of the degree distribution. In general, we expect that such data will become increasingly available due to the epidemiological importance of this tail (e.g. the study of Read et al. \[[@CR21]\]). In total, completed questionnaires were received from 5,388 participants in Great Britain, 3,901 of which were from postal surveys. There was some bias in demographical representation, most notably younger age groups and males were generally under-represented (see Danon et al. \[[@CR31]\] for more details). The data is available at <http://wrap.warwick.ac.uk/54273/>. Generalised preferential attachment {#Sec8} ----------------------------------- As noted by Durrett \[[@CR32]\], Barabási \[[@CR28]\], and Simkin and Roychowdhury \[[@CR33]\], the basic idea behind the preferential attachment model is close to the population model of Yule \[[@CR34]\]. We consider a Yule-like stochastic process described precisely as follows. In a population of *N* individuals indexed by *i* each individual has an integer-valued random variable $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$K_{i}(t)$\end{document}$ for its number of contacts. Individual *i* starts with $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$K_{i}(0)=0$\end{document}$ and makes new contacts over a time period $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$T_{i}$\end{document}$, which is given by a positive real-valued random variable with probability density function $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\rho(t)$\end{document}$. The generation of new social contacts is modelled by a continuous-time Markov chain with the following events and rates: $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ K_{i} \rightarrow K_{i} + 1 \quad\text{at rate } f_{K_{i}} := 1 + \tau K_{i} . $$\end{document}$$ We take the preferential attachment hypothesis PA to be stated mathematically as $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \text{PA} \Leftrightarrow\tau>0 . $$\end{document}$$ Writing $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$p_{k}(t)$\end{document}$ for the probability that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$K_{i}=k$\end{document}$ at time $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t< T_{i}$\end{document}$, we can use the method of characteristics to derive an expression for the probability generating function of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$K_{i}$\end{document}$, $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ g(t,s) = \sum_{k} p_{k}(t) s^{k} = \bigl(s - (s-1) \mathrm{e}^{\tau t} \bigr)^{-1/\tau} ,\quad s \in[0,1] . $$\end{document}$$ From this, we can derive expressions for the probability mass function, $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document} $$\begin{aligned} p_{k}(t) =& \frac{1}{k!} \frac{\partial^{k} g}{\partial s^{k}} \bigg|_{s=0} = \frac{\Gamma(k+\frac{1}{\tau})}{\tau\Gamma(\frac{1}{\tau })\Gamma(k+1)} \mathrm{e}^{-t}\bigl( \mathrm{e}^{\tau t} - 1\bigr)^{k} \\ \rightarrow& \kappa_{t} \bigl(\mathrm{e}^{\tau t} - 1 \bigr)^{k} k^{{(1-\tau)}/{\tau}} , \end{aligned}$$ \end{document}$$ where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\kappa_{t}$\end{document}$ is a function of *t* but not *k* and the asymptote holds as *k* becomes large. This is not a simple power-law relationship, and so the asymptotic behaviour of the moments is not determined by the power-law exponent, but rather through the moment generating function $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$M(t,z) = g(t,\mathrm{e}^{z})$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$z\in(-\infty,0]$\end{document}$, such that the *r*th moment of the degree distribution, conditional on $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t< T$\end{document}$, is $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ m_{r}(t) = \frac{\partial^{r} M}{\partial z^{r}} \bigg|_{z=0} . $$\end{document}$$ In particular, $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ m_{1}(t) = \frac{1}{\tau}\bigl(\mathrm{e}^{\tau t} -1\bigr) , \quad\quad m_{2}(t) = m_{1}(t) + (\tau+1) \bigl(m_{1}(t)\bigr)^{2} ,\quad\quad \ldots. $$\end{document}$$ Then accounting for the randomness of the times, the *r*th moment of the degree distribution will be $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \overline{m}_{r} = \int_{t=0}^{\infty} \rho(t) m_{r}(t)\,\mathrm{d}t . $$\end{document}$$ We will be interested in the empirical evidence for whether such moments converge or diverge, in light of the epidemiological relationship ([1](#Equ1){ref-type=""}). Phase-type holding times {#Sec9} ------------------------ The question is then posed as to an appropriate distribution from which to draw the holding times $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\{T_{i}\}$\end{document}$ for the amount of time spent making new contacts on the day for which individuals provide data. In previous work \[[@CR30]\] on a related model of contact formation we considered holding times $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$T_{i}$\end{document}$ that were log-normally distributed. This provided a good fit to data, but was computationally intensive and lacked a mechanistic interpretation. We therefore consider here a class of distributions for the holding times that is highly flexible, but which has analytic and numerical benefits - the distributions of *phase type* \[[@CR35]\]. Phase-type distributions are dense in the space of positive-valued probability distributions \[[@CR36]\], meaning that they can be made arbitrarily close to any other distribution. They have a mechanistic interpretation and allow for analytic manipulations that greatly reduce the numerical cost of likelihood evaluation. The basic idea behind the model is shown in Figure [1](#Fig1){ref-type="fig"}. A set of phases is indexed by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$a,b=1,\ldots, m$\end{document}$; the probability of starting in phase *a* is $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\nu_{a}$\end{document}$ (meaning these parameters must sum to unity); the rate of stopping making new social contacts is $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mu_{a}$\end{document}$ for an individual in phase *a*; and the rate of moving from phase *a* to phase *b* is $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$Q_{a,b}$\end{document}$. Note that different rates of making contacts in different phases are not realistically distinguishable from different times spent and so are not included as parameters. The phases have a mechanistic interpretation as the activities that individuals undertake on a given day. Figure 1**A model of** ***m*** **phases.** $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\nu_{a}$\end{document}$ is the probability of starting in phase *a*, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mu_{a}$\end{document}$ is the exit rate of phase *a*, and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$Q_{a,b}$\end{document}$ is the rate of moving from phase *a* to phase *b*. In this model, the probability density function for the holding time is given by the general expression $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \rho(t) = \boldsymbol{\mu}^{\top} \mathrm{e}^{\mathbf{M} t} \boldsymbol{\nu} , $$\end{document}$$ where: $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \begin{aligned} &\boldsymbol{\mu} = (\mu_{a});\quad\quad \boldsymbol{\nu} = (\nu_{a}) ; \\ & M_{a,a} = -M_{a} ;\quad\quad M_{a} =\mu_{a} + \sum_{b} Q_{b,a} ;\quad\quad M_{a,b\neq a} = Q_{b,a} . \end{aligned} $$\end{document}$$ From the expressions ([5](#Equ5){ref-type=""}), ([7](#Equ7){ref-type=""}), ([8](#Equ8){ref-type=""}) and ([9](#Equ9){ref-type=""}) above, in particular through inspection of the form of the moment generating function, it is clear that the *r*th moment of the degree distribution will involve a term like $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \mathcal{I}_{r} = \int_{t=0}^{\infty} \mathrm{e}^{r\tau t} \rho(t) \,\mathrm{d}t = \boldsymbol{\mu}^{\top} \int_{t=0}^{\infty} \mathrm{e}^{(\mathbf{M} + r\tau\mathbf{I})t} \,\mathrm{d}t \boldsymbol {\nu} , $$\end{document}$$ where **I** is the identity matrix. Let $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathbf{A} = \mathbf {M} + r\tau \mathbf{I}$\end{document}$; this matrix is triangular and so its eigenvalues are equal to its diagonal elements; in particular the *a*th eigenvalue of **A** is $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\lambda_{a} = -M_{a} + r \tau$\end{document}$. If we let **R** be a matrix whose *a*th column is the *a*th right eigenvector of **A** and **L** be a matrix whose *a*th row is the *a*th left eigenvector of **A** then $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \mathcal{I}_{r} = \boldsymbol{\mu}^{\top} \int _{t=0}^{\infty} \mathbf{L}^{-1} \mathbf{L} \mathrm{e}^{\mathbf{A}t} \mathbf{R} \mathbf{R}^{-1} \,\mathrm{d}t \boldsymbol{\nu} = \boldsymbol{\mu}^{\top} \mathbf{L}^{-1} \int _{t=0}^{\infty} \mathbf{D} \,\mathrm{d}t \mathbf{R}^{-1}\boldsymbol{\nu} , $$\end{document}$$ where **D** is a diagonal matrix whose *a*th diagonal element is $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathrm{e}^{\lambda_{a} t}$\end{document}$. The integral $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal{I}_{r}$\end{document}$ therefore converges exactly when ∀*a*, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\lambda_{a}<0$\end{document}$, which implies that the condition for divergence of the *r*th moment is $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \overline{m}_{r} \text{ diverges}\quad \Leftrightarrow\quad \exists a \text{ such that } \tau> M_{a}/r . $$\end{document}$$ In general, however, combination of ([10](#Equ10){ref-type=""}) and ([6](#Equ6){ref-type=""}) is not the most numerically efficient method for calculation of the overall probability mass function for final number of contacts $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$K_{i}(T_{i})$\end{document}$ and a different approach is needed. Numerically efficient model solution {#Sec10} ------------------------------------ The model as described above can be solved in a numerically efficient manner using the spectral methods for continuous-time Markov chains developed by Bailey \[[@CR37]\]. We consider the limit as the population size $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$N\rightarrow\infty$\end{document}$ and write down ordinary differential equations (ODEs) for the proportion of the population in phase *a* and with *k* social contacts at time *t*, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$p_{a,k}(t)$\end{document}$. These ODEs take the form $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \frac{\mathrm{d}}{\mathrm{d}t}p_{a,k} = - \biggl(f_{k} + \mu_{a} + \sum_{b>a}Q_{a,b} \biggr) p_{a,k} + f_{k-1} p_{a,k-1} + \sum _{b< a} Q_{b,a}p_{b,k} , $$\end{document}$$ where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$f_{k}$\end{document}$ is the rate at which individuals with *k* social contacts make new contacts, given in ([3](#Equ3){ref-type=""}). We are then interested in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$d_{k}$\end{document}$, the probability mass function for a randomly selected individual having made *k* social contacts by the end of the process. A series of manipulations directly analogous to those of Bailey \[[@CR37]\] shows that $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ d_{k} = \lim_{s\downarrow0} \sum _{a} \mu_{a} \int_{0}^{\infty} \mathrm{e}^{-st} {p}_{a,k}(t)\, \mathrm{d}t =: \sum _{a} \mu_{a} A_{a,k} . $$\end{document}$$ Applying Laplace transformation to ([15](#Equ15){ref-type=""}) subject to the initial condition $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$p_{a,k}(0) = \nu_{a} \delta_{k,0}$\end{document}$ and taking the frequency-space limit $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$s\downarrow0$\end{document}$ then leads to a set of linear equations for $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$d_{k}$\end{document}$ that are triangular and so can be evaluated directly without numerically costly matrix inversion: $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \nu_{a} \delta_{k,0} = - \biggl(f_{k} +\mu_{a} + \sum_{b>a}Q_{a,b} \biggr) A_{a,k} + f_{k-1} A_{a,k-1} + \sum _{b< a} Q_{b,a}A_{b,k} . $$\end{document}$$ These equations are at the root of the numerical efficiency of our model. Note that we use Laplace transformation mainly for technical reasons and our results could be obtained by directly integrating ([15](#Equ15){ref-type=""}) if one were not concerned by all quantities being rigorously defined. Model likelihood, fitting and selection {#Sec11} --------------------------------------- We assume a vector of data $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathbf{y}=(y_{k})$\end{document}$, where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$y_{k}$\end{document}$ is the number of individuals reporting *k* social contacts when surveyed. A model $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal{M}$\end{document}$ is therefore specified by a number of phases *m* and the presence or absence of PA, meaning the general parameters are $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\boldsymbol{\theta} = (\tau,\nu_{a},\mu_{a},Q_{a,b})$\end{document}$, with *τ* present only if there is PA. The number *n* of individuals sampled from the British population *N* is $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ n = \sum_{k} y_{k} = 5388 \ll N \gtrsim6 \times10^{7} , $$\end{document}$$ and so it is appropriate to assume that each individual picks a number of contacts independently from the distribution with pmf given by $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$d_{k}$\end{document}$ as in ([16](#Equ16){ref-type=""}). Accounting for the censoring of zero contacts in the real data, we define $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \tilde{d}_{0} = 0 ,\quad\quad \tilde{d}_{k>0} = \frac{d_{k}}{1-d_{0}} , $$\end{document}$$ meaning that the overall likelihood function is then given by $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ L(\mathbf{y} | \boldsymbol{\theta}) = \frac{n!}{\prod_{l} y_{l} !} \prod _{k} \bigl(\tilde{d}_{k}(\boldsymbol{ \theta})\bigr)^{y_{k}} . $$\end{document}$$ Note that the combinatorial factors do not depend on the parameters, and so need not be calculated during model fitting. We consider the use of the likelihood function ([20](#Equ20){ref-type=""}) using standard statistical methodology. Numerical maximum likelihood estimation was performed using simulated annealing run from multiple starting points to ensure the global optimum was obtained. Model selection was performed using AIC \[[@CR38]\] and BIC \[[@CR39]\], as well as likelihood ratio tests \[[@CR40]\] on pairs of models where this test was informative. This was done since each approach involves different trade-offs between model fit and complexity, and to check that our conclusions about PA are not overly sensitive to the precise method used. Uncertainty in model parameters was quantified using confidence intervals obtained through bootstrapping the data, and uncertainty in model outputs such as the predicted degree distribution was quantified using a parametric bootstrap. Results and discussion {#Sec12} ====================== Table [1](#Tab1){ref-type="table"} shows the models we fitted, their number of parameters, AIC/BIC relative to the minimum, and the first moment that diverges according to ([14](#Equ14){ref-type=""}). Figure [2](#Fig2){ref-type="fig"} shows the results of performing likelihood ratio tests. These show that AIC prefers a 5-phase model with PA as do likelihood ratio tests for any significance level between 0.07% and 20%. BIC penalises complex models more severely and therefore selects a 3-phase model with PA. Figure 2**The results of likelihood ratio tests on the models.** Arrows point towards the model preferred by the likelihood ratio test, with *p* values shown.Table 1**Comparison of models with different numbers of phases, with and without preferential attachment (PA), together with: number of parameters; differences in AIC and BIC values compared to the overall minimum; and the lowest divergent moment for models with PA(Phases, PA)No. ParamsΔAICΔBICDiverge**(1,No)12.2 × 10^3^2.1 × 10^3^--(2,No)42.1 × 10^2^1.5 × 10^2^--(3,No)81.2 × 10^2^83--(4,No)134238--(5,No)192358--(6,No)26271.1 × 10^2^--(1,Yes)21.9 × 10^2^1.1 × 10^2^3(2,Yes)51.3 × 10^2^724(3,Yes)931$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathbf{[0]}$\end{document}$3(4,Yes)1411143(5,Yes)20$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathbf{[0]}$\end{document}$423(6,Yes)279973Preferred models are shown using square brackets and bold type. Therefore, regardless of the number of phases selected by different approaches to model selection, we see that the models with PA are preferred over models without. Figure [3](#Fig3){ref-type="fig"} shows the predictions of the models preferred by different selection criteria, as well as the models with the same number of phases but no PA, against real data. We see in the left column of plots that for the 5-phase models, the main difference is in the tail of the distribution as we would expect. In the 3-phase models shown in the right column of plots, the model without PA also smooths over features in the bulk of the distribution compared to the model with PA. Figure 3**Data at different scales versus (left column) model selected using AIC and likelihood ratio tests (right column) model selected using BIC.** Models are labelled by the number of phases and whether PA is present. Confidence intervals in the data are calculated using bootstrapping for data and parametric bootstrapping for models. For the 3-phase model with PA, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\tau= 0.018\ [0.012,0.026]$\end{document}$; and if we set $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\tau=0$\end{document}$ but leave the other parameters at their fitted values, then the total number of contacts per person is reduced to 64% of its original value. For the 5-phase model with PA $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\tau= 0.026\ [0.019,0.036]$\end{document}$; and if we set $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\tau=0$\end{document}$ but leave the other parameters at their fitted values, then the total number of contacts per person is reduced to 58% of its original value. This shows that in both of these models, we can attribute a substantial fraction of the contacts to PA. We also calculate that the second moment does not diverge in any of the fitted models, which helps to resolve the epidemiological paradox that we introduced at the start of this paper. PA is empirically supported, and is also mechanistically plausible since existing social contacts give more opportunities for future social contact. Combined with a sufficiently detailed phase-based mechanistic model of the contexts in which social contacts are made, however, PA does not imply a divergent second moment for the distribution of contacts relevant for the spread of directly transmitted infections. This means that our understanding of how basic epidemiological quantities like the basic reproductive ratio, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$R_{0}$\end{document}$, are related to contact networks does not need to be revised in the light of empirical evidence. As a final observation, we believe that as computational resources for fitting models to data improve, it will in general be easier to test the hypothesis of PA directly in all kinds of data, rather than looking for asymptotic power laws. Appendix 1: Code {#Sec13} ================ We provide the function phase_ll.c below as a Mex file for use within MATLAB; however, the C syntax is standard and modification to run within R, Python or C should be straightforward. Once compiled, use within MATLAB is: \>\> \[LL, P\] = phase_ll(params,M,kh,K); where params is a vector of the model parameters \[nu(:), mu(:), Q(:), tau\], M is the number of phases, kh is a vector of count frequencies from 0:K, K is the maximum count, LL is the log-likelihood given the parameters and P is a the probability vector for the observed counts given the parameters. 1.1 The function phase_ll.c {#Sec14} --------------------------- 
 \#include \<mex.h\> \#include \<math.h\> 
 void mexFunction (int nlhs, mxArray \*plhs\[\],                   int nrhs, const mxArray \*prhs\[\]) {   /\* Input variables \*/   double \*params, \*kh, \*Mmex, \*Kmex;   size_t lp;   int M, K; 
   /\* Get inputs in correct form \*/   params = mxGetPr(prhs\[0\]);   lp = mxGetN(prhs\[0\]);   Mmex = mxGetPr(prhs\[1\]);   kh = mxGetPr(prhs\[2\]);   Kmex = mxGetPr(prhs\[3\]);   M = (int) Mmex\[0\];   K = (int) Kmex\[0\]; 
   /\* Working variables \*/   double nu\[M\], mu\[M\], Q\[M\]\[M\], s\[M\];   double f\[K\];   double tau, S;   double A\[K\]\[M\];   int fi, fj, k, m; 
   /\* Output variables \*/   double \*P, \*LL; 
   /\* Organise outputs into correct form \*/   plhs\[0\] = mxCreateDoubleMatrix(1,1,mxREAL);   LL = mxGetPr(plhs\[0\]);   plhs\[1\] = mxCreateDoubleMatrix(K,1,mxREAL);   P = mxGetPr(plhs\[1\]); 
   nu\[0\] = 1.0;   for(fi=1; fi\<M; fi++) {     nu\[fi\] = params\[fi-1\];     nu\[0\] -= params\[fi-1\];   }   for(fi=M; fi\<(2\*M); fi++) {     mu\[fi-M\] = params\[fi-1\];   }   m = (2\*M)-1;   for(fi=0; fi\<M; fi++) {     s\[fi\] = 0;     for(fj=(fi+1); fj\<M; fj++) {       Q\[fi\]\[fj\] = params\[m\];       s\[fi\] += params\[m\];       m=m+1;     }   }   if (lp == (m+1))     tau = params\[m\];   else     tau = 0; 
   f\[0\] = 1;   for(k=1; k\<(K-1); k++)     f\[k\] = 1+(tau\*k);   f\[K-1\] = 0; 
   A\[0\]\[0\] = (nu\[0\])/(f\[0\]+mu\[0\]+s\[0\]);   for(k=1; k\<K; k++) {     A\[k\]\[0\] = ( f\[k-1\]/(f\[k\]+mu\[0\]+s\[0\]) )\*A\[k-1\]\[0\];   }   for(m=1; m\<M; m++) {     S=0.0;     for(fi=0; fi\<m; fi++) {       S += Q\[fi\]\[m\]\*A\[0\]\[fi\];     }     A\[0\]\[m\] = (nu\[m\]+S)/(f\[0\]+mu\[m\]+s\[m\]);     for(k=1; k\<K; k++) {       S=0.0;       for(fi=0; fi\<m; fi++) {         S += Q\[fi\]\[m\]\*A\[k\]\[fi\];       }       A\[k\]\[m\] = ((f\[k-1\]\*A\[k-1\]\[m\])+S)/(f\[k\]+mu\[m\]+s\[m\]);     }   } 
   P\[0\] = 0.0;   for(m=0; m\<M; m++) {     P\[0\] += A\[0\]\[m\]\*mu\[m\];   } 
   for(k=1; k\<K; k++) {     P\[k\] = 0.0;     for(m=0; m\<M; m++) {       P\[k\] += A\[k\]\[m\]\*mu\[m\];     }     P\[k\] /= (1-P\[0\]);   }   P\[0\] = 0.0; 
   LL\[0\] = 0.0;   for(k=1; k\<(K-1); k++) {     LL\[0\] += kh\[k\]\*log(P\[k\]);   } 
 } **Competing interests** The authors declare that they have no competing interests. **Authors' contributions** All authors: collected the data, designed the study. TH: performed the analysis. The Social Contact Survey was funded by the Medical Research Council, grant number G0701256. TH and MJK are supported by the Engineering and Physical Sciences Research Council. JMR and MJK are supported by the Economic and Social Research Council, grant ES/K004255/1. LD is supported by the Leverhulme Trust.

Introduction {#Sec1} ============ Using long-lasting insecticidal nets (LLINs) can reduce malaria morbidity and mortality \[[@CR1]--[@CR3]\]. Worldwide funding for malaria control fell by 8% between 2013 and 2014 \[[@CR4]\]. Funding for the provision of LLINs can continue to decrease \[[@CR5]\] and it is necessary to ensure cost-effective and sustainable LLINs delivery models. Spending on malaria control commodities is estimated to have increased globally from about US\$ 40 million in 2004 to about US\$ 1.6 billion in 2014, where LLINs were responsible for 63% of total spending in 2014 (US\$ 1 billion) \[[@CR5]\]. Due to high burden of malaria in the sub Saharan Africa and the strategy of universal LLINs coverage, most of the international funding in 2014 was spent in the WHO African Region. In Mozambique, half of the international funding for malaria intervention went to mosquito nets \[[@CR5]\]. Most of the economic studies do not compare the distribution of LLINs through the same mechanisms or channel. Recently, Ntuku et al. \[[@CR6]\] evaluated a fixed delivery strategy and a door-to-door strategy. Their findings show that the fixed delivery strategy achieved a higher LLIN coverage at lower delivery cost compared with the door-to-door strategy. In 2015, Mozambique piloted a new model of LLIN delivery in campaign \[[@CR7]\]. Two rural districts were intervened with a new LLIN delivery model, and two served as the control, maintaining the standard delivery model. Results of this pilot showed that 87.8% (302,648) of planned LLINs were distributed in the intervention districts compared to 77.1% (219,613) in the control districts \[[@CR7]\]. The objective of this research is to compare the cost-effectiveness of the two delivery models in mass campaigns in rural Mozambique, and establish the most cost-effective LLIN delivery model. Main text {#Sec2} ========= Methods {#Sec3} ------- ### Setting and location {#Sec4} The study was conducted in four rural districts of Mozambique. Two districts served as intervention (with the new delivery model); and two served as control (with standard delivery model). These districts were selected based on pragmatic criteria described elsewhere \[[@CR7]\]. All four districts are rural, with limited access to health services, and low health, social and economic indicators \[[@CR7], [@CR8]\]. ### Study design {#Sec5} An observational and cross-sectional study with cost-effectiveness analysis component was carried out using secondary data from the pilot study conducted between October and December 2015 \[[@CR7]\]. ### Collection of cost data {#Sec6} The campaign costs were retrospectively collected from the providers' perspective. The costs considered were related to training of personnel, allowances, LLINs warehouse storage, LLIN transportation vehicles rental, and materials production (pamphlets, coupons, stickers, etc.). These costs were aggregated into four categories: (1) micro-planning; (2) LLIN storage; (3) LLIN transport; (4) mobilization and training at district level, household registration, and LLIN distribution. Costs were collected in local currency and United States Dollars (US\$). In 2015, the exchange rate was: 1 US\$ = 42.00 Meticais. No adjustment for inflation was undertaken since all cost were paid in 2015. No discount rate was applied since the temporal universe of analysis did not exceed 1 year. ### Comparators: the two delivery models {#Sec7} Both delivery models are community-based. One delivery model allocates LLINs based on the assumption of one LLINs for every two persons in a household (intervention districts), and another the number of LLINs is allocated based on assumptions around households members sleeping patterns (control districts). A comprehensive description of the models and pitfalls associated during implementation are reported elsewhere \[[@CR7], [@CR8]\]. ### Measurement of effectiveness {#Sec8} Two endpoints were used to measure the effects of the campaign in the intervention and control districts: (i) number of LLINs delivered; (ii) households (HHs) achieving universal coverage (UC) target (one LLIN for every two persons). The number of HHs achieving UC was estimated according to the following steps:Step 1: percentage of HHs achieving UC---\[70.8% (95% CI 67.6--74.0) in the intervention districts, and 59.6% (95% CI 56.2--63.0) in the control districts\] \[[@CR8]\];Step 2: registered HHs multiplied the step 1 results (136,985 HHs were registered in the intervention districts, and 120,246 HHs were registered in the control districts \[[@CR7]\]). ### Cost-effectiveness analyses {#Sec9} The following cost-effectiveness measures were calculated: (i) average cost-effectiveness ratio (ACER) per LLIN delivery; (ii) ACER per HH achieving UC; (iii) incremental cost-effectiveness ratio (ICER); and (iv) incremental net benefit (INB).The ACER per LLIN delivered and ACER per HH achieving UC was calculated by dividing the total implementation cost by the number of LLINs delivered and number of HHs achieving UC, respectively.The ICER was calculated by dividing the difference between the total cost by the difference of effects in the intervention and control districts.The INB was calculated by valuing additional effect (∆E) in dollars and then subtracting the associated additional cost (∆C): INB = (∆E × λ) − ∆C, where λ is willingness to pay (WTP) for a 1-unit gain of effect \[[@CR9]\]. ### Willingness to pay (WTP) and decision rule on cost-effectiveness {#Sec10} Three WTP was adopted:WTP 1) for LLINs delivered (US\$ 1.32 per LLIN---adopted by the Global Fund for Mozambique in-country mass free campaign budget planning \[[@CR10]\]);WTP 2) for LLINs delivered plus LLIN purchases cost (US\$ 9.12 per LLIN − US\$ 1.32 + US\$ 7.8 which was the maximum inter-quartile purchase cost for the period 2005--2012 \[[@CR11]\]); andWTP 3) for HHs achieving UC (US\$ 3.30 per household). This third WTP was determined by multiplying the average number of HH members (five) \[[@CR7]\] by US\$ 1.32, and dividing by two (one LLIN for every two persons). The cost-effectiveness decision rule was based on INB results. Two INB was calculated: INB for delivered LLIN (using WTP 1 and 2) and INB for HHs achieving UC (using WTP 3). A positive INB means that the new intervention extra benefits (∆E × λ) outweighs its extra costs (∆C), i.e., the new intervention is deemed cost-effective. Conversely, when INB is less than 0 (negative INB), the new intervention is not cost-effective \[[@CR9]\]. ### Sensitivity analysis {#Sec11} A one-way sensitivity analysis was performed on the following parameters and assumptions: (i) free warehouses; (ii) transport cost (± 50%); and (iii) costs of LLINs purchase (− 25%, − 50%). Base case cost analysis was used to calculate the percentage of deviation. For the first two parameters, the base case was the ACER and ICER per LLIN delivered. For the third parameter, the base case was also ACER and ICER per LLIN delivered but also included the 2014 purchase cost of US\$ 3.63 per LLIN for the planned 344,770 LLINs in the intervention and 284,873 LLINs in the control districts. The US\$ 3.63 per LLIN was based on the unit cost for Disease Control Technologies Royal Sentry^®^ rectangular LLINs 190 × 180 × 180 cm of US\$ 3.19; procurement fee 1.50%; outbound transport charges 11.94%; transport insurance charges 0.14%; question and answer charges 0.09%; and pre-shipment inspection charges 0.12%. Results {#Sec12} ------- ### Financial costs of the LLIN campaign {#Sec13} The total financial cost of the campaign from the providers' perspective was US\$ 231,237.30 and US\$ 174,790.14 in the intervention and the control districts, respectively. Cost activity category 4 and 3 comprised around 43% and 38% in the intervention districts and 50% and 41% in the control districts, respectively. Cost activity category 4 were US\$ 0.06/LLIN higher in the control districts. The ACER per LLIN delivered was US\$ 0.76 and US\$ 0.80 in the intervention and control districts, respectively---Table [1](#Tab1){ref-type="table"}. The ACER for HHs achieving UC was lower in the intervention districts (US\$ 2.38 vs. US\$ 2.43)---Table [2](#Tab2){ref-type="table"}.Table 1Total cost and cost per LLIN delivered (US\$) for each category in the intervention and control districtsIntervention (new delivery model)Control (standard delivery model)LLIN delivered302,648219,613Aggregated and standardized categoriesCost%Cost%1. Micro-planning6184.852.75580.843.2ACER per LLIN delivered0.020.032. LLIN warehouse storage6983.763.06880.333.9ACER per LLIN delivered0.020.033. LLIN transport86,908.6437.671,007.4840.6ACER per LLIN delivered0.290.324. Mobilization, trainings at district level, household registration, and LLIN distribution99,105.5342.986,736.9749.6ACER per LLIN delivered0.330.394.1 Coupons and Stickers production32,054.5213.9NANAACER per LLIN delivered0.11NA4.2 Household registration data analysisNANA4584.522.6ACER per LLIN deliveredNA0.02Total cost231,237.30100.0174,790.14100.0ACER per LLIN delivered0.760.80Household registration data analyses were not necessary in the intervention districts; coupons and stickers were not used in the control districts*NA* not applicable Table 2Comparative cost-effectiveness results in the intervention and control districtsEffects and cost-effectiveness indicatorsIntervention (new delivery model)Control (standard delivery model)Delivered LLINs302,648219,613Registered Households136,985120,446HHs achieving UC (%)70.859.6HHs with UC96,98571,786Total cost (U\$S)231,237.30174,790.14ACER for HHs achieving UC (U\$S)2.382.43∆ Cost (U\$S)56,447.16∆ Effect (LLINs delivered)83,035∆ Effect (HHs achieving UC)25,199ICER for delivered LLIN (U\$S)0.68ICER for HHs achieving UC (U\$S)2.24INB for delivered LLIN (U\$S)+ 53,159.04INB for HHs achieving UC (U\$S)+ 26,709.54 ### Cost-effectiveness and decision rule {#Sec14} The overall ICER to deliver one additional LLIN was US\$ 0.68, with a positive INB (intervention deemed cost-effective), i.e., a saving of more than US\$ 50,000 per LLIN delivered would result from switching from the standard to new delivery model (with U\$S 1.32 as WTP)---Table [2](#Tab2){ref-type="table"}. The overall ICER for one households reaching universal coverage was US\$ 2.24, with a positive INB, i.e., a saving of more than US\$ 26,000 per household reaching universal coverage would result from switching from the standard to new delivery model (with U\$S 3.30 as WTP)---Table [2](#Tab2){ref-type="table"}. ### Sensitivity analysis {#Sec15} After a sensitivity analysis, the ACER per LLIN delivered remained at a lower rate (less sensitive) in the intervention districts rather than in the control districts, i.e., the results of ACER per LLIN delivered remain robust. The cost-effectiveness of the new delivery model also remained sustained for all the parameters tested (positive INB even using WTP 2)---Table [3](#Tab3){ref-type="table"}.Table 3Deterministic one-way sensitivity analysis of cost estimates to key assumptionsParameter testedControl/intervention districtsCost (US\$)ACER~LLIN~% ACER~LLIN~ deviationICER~LLIN~% ICER~LLIN~ deviationINB~LLIN~Free warehouseControl167,909.810.76− 4.43Intervention224,253.540.74− 2.500.68− 0.1853,262.46Transport cost (+ 50%)Control210,293.880.9619.70Intervention274,691.630.9119.420.7814.0945,208.45Transport cost (− 50%)Control139,286.400.63− 20.72Intervention187,782.980.62− 18.360.58− 14.0861,109.62LLIN purchase cost + delivery costControl1,214,576.595.53NAIntervention1,489,647.804.92NA3.31NA482,207.98Less 25% LLIN purchase cost + delivery costControl954,629.984.35− 21.39Intervention1,175,045.183.88− 21.092.65− 19.80536,864.00Less 50% LLIN purchase cost + delivery costControl694,683.363.16− 42.80Intervention860,442.552.84− 42.212.00− 39.69591,520.01*NA* not applicable because is the new base case with the inclusion of LLIN purchase cost into delivery cost Discussion {#Sec16} ---------- This cost-effectiveness study demonstrates that the new delivery model is the more cost-effective strategy for the universal coverage campaign. The positive incremental net benefit shows that important savings could be achieved from adopting the new delivery model (opportunity-cost). The ACER per LLIN delivered was lower in the intervention districts. This was mainly driven by the low relative contribution of the micro-planning, LLINs transport, and district-level activities costs. The ACER per HH achieving UC was also lower in the intervention district. Paintain et al. \[[@CR12]\] found higher financial costs for the distribution of LLINs---US\$ 1.19 (ranging from US\$ 1.08 to US\$ 1.41). However, Grabowsky et al. \[[@CR13]\] found financial costs of around US\$ 0.32, which is considerably lower than what was found in the present study. As for Mueller et al.'s \[[@CR14]\], they incurred an ACER of US\$ 1.6/LLIN. This ACER is higher than what was found in this study, for either the new or standard delivery model. From a health-financing point of view, the high overall cost found for these interventions casts doubt on their long-term sustainability in low-income contexts. Mozambique allocated US\$ 580.9 million (9% of the National budget) to the health sector in 2015 \[[@CR15]\]. For an estimated 25,727,911 inhabitants in 2015 \[[@CR16]\], this budget for the health sector corresponds to US\$ 22.6 per capita (assuming this as Mozambican State WTP). Taking into account the mean cost to distribute one LLIN with the new intervention (US\$ 0.76), and that one LLIN would benefit two persons, the financial sustainability of the intervention would be guaranteed if a campaign were taken stand-alone by the Mozambican State and the WTP were only for the malaria programme (ICER \< WTP). However, considering that the Ministry of Health does not focus exclusively on LLIN campaigns, and other health programmes requires budgetary allocation, the country would not be in a position to guarantee financial sustainability for LLIN distribution. The Mozambique health sector allocated US\$ 4,186,129 to the National Malaria Control Programme in 2014 \[[@CR4]\]. Considering that 100% of the Mozambican population is at risk for malaria, this allocation corresponds to US\$ 0.16 per capita (WTP), i.e., US\$ 0.32 for each two persons (US\$ 0.44 less than the ACER per LLIN in the intervention). This WTP clearly demonstrates the current financial un-sustainability of the country in assuming the LLINs campaign. The same conclusion holds even considering free warehouse storage, less 50% transport costs, and 50% reduction of LLINs' purchase cost. Bed nets campaign is still more cost-effective than indoor residual spraying (US\$ 5.41 per person protected) \[[@CR17]\], RTS,S (US\$ 39.25 per fully vaccinated child) \[[@CR18], [@CR19]\], and treatment (US\$ 2.59 per person tested and treated) \[[@CR20]\]. This is in line with what Winskill et al. \[[@CR3]\] found in their modelling cost-effectiveness study. In conclusion, the new delivery model is worthwhile (INB positive) from a programme provider perspective and current donor economic outlook. However, the long-term sustainability of either delivery models is far from guaranteed in Mozambique's current economic context. Limitations {#Sec17} =========== The WTP is often estimated through extensive surveys and is not always available \[[@CR21]\]. The rationale for using the three ceiling is not only justifiable, but it is also appropriate to the country context. However, it is suggested that each country should adopt their own value for money ceiling, or make use of net benefit approach with cost-effectiveness acceptability curve if the ceiling is unknown plotting probability of cost-effectiveness against variation of the ceiling. Another limitation is the one-way sensitivity analyses. In the "real-world" more than one parameter varies at a time, and correlation between variation in multiple parameters can overstate uncertainty. ACER : average cost-effectiveness ratio HH/HHs : household/households ICER : incremental cost-effectiveness ratio INB : incremental net benefit LLIN/LLINs : long-lasting insecticidal net/long-lasting insecticidal nets OR : odds ratio SA : sensitivity analysis UC : universal coverage WTP : willingness-to-pay **Publisher\'s Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. The authors would like to acknowledge the contributions of World Vision Mozambique and the Foundation for the Community Development for providing the financial records. The corresponding author would like to also acknowledge Prof. Giuliano Russo for encouragement and initiation in the field of health economics. JAHA conceived and designed the protocol, performed data analysis, and wrote the manuscript. BC, ML and CM supported critical manuscript review. MROM supported the protocol design, data analysis, contributed to manuscript writing. All authors read and approved the final manuscript. The LLIN universal coverage campaign in Mozambique was supported through a financial contribution of Global Fund to Fight AIDS, Tuberculosis, and Malaria. The Funding body did not have any role in the design of the study and collection, analysis, and interpretation of data and in writing the manuscript. The datasets used and/or analysed during the current study are available from the corresponding author, upon reasonable request. The study was administratively authorized by the Provincial Health Departments of Zambezia (Ref 2270/DPSZ/512/2015) and Manica (09/11/2015), and also received authorization from the National Committee on Bioethics in Health (Ref 112/CNBS/2015). Not applicable. The authors declare that they have no competing interests.

Introduction {#s2} ============ Macro-autophagy is a complex multistep process that facilitates the degradation of damaged and excess proteins and organelles to generate macromolecular building blocks and fuel metabolic pathways ([@bib30]). The autophagy pathway has critical roles in core biological processes such as mitochondrial function, cell death, immune surveillance, protein homeostasis, stress response, and metabolism. Accordingly, abnormalities in these processes and the disease-associated pathologies have been linked to aberrant autophagic degradation, most notably in aging, neurodegenerative diseases, and multiple forms of cancer. In this review, we focus on the protumorigenic role of autophagy in cancer, highlighting recent insights linking autophagy and apoptosis and other death pathways. With over 60 active clinical trials targeting autophagy in a variety of tumor types, it is critical to understand how the molecular mechanisms that connect these processes can be leveraged to enhance the benefit to patients and prevent relapse. The history of cancer therapy has proven that adaptation and acquired resistance to anticancer therapies represent perhaps the largest obstacle to overcome. Therefore, a critical, as yet incompletely understood, issue is whether autophagy inhibitors will be plagued by these same hurdles. Here we address this and other questions regarding autophagy inhibition as a cancer therapy. Macro-autophagy {#s3} =============== The evolutionarily conserved recycling processes that deliver surplus or damaged cytoplasmic material to lysosomes for degradation can be subdivided into three related processes: micro-autophagy, chaperone-mediated autophagy, and macro-autophagy. Micro-autophagy and chaperone-mediated autophagy involve direct delivery mechanisms to the lysosome, both of which can also be important in cancer; for a detailed discussion, readers are referred to an excellent recent review ([@bib59]). Macroautophagy (hereafter autophagy) is a multistep process involving \>20 core autophagy proteins, called ATGs, that function to envelop cytoplasmic cargo within a double-membrane vesicle structure. These autophagosomes can subsequently fuse with acidic lysosomes, where pH-sensitive enzymes mediate the degradation of the cytoplasmic material ([@bib30]; [Fig. 1](#fig1){ref-type="fig"}). The pathway is initiated by the Unc-51--like kinase (ULK) complex, which phosphorylates a phosphatidylinositol 3-kinase (VPS34), part of the Beclin1 complex necessary for initiation of the phagophore ([@bib91]; [@bib114]; [@bib51]). Extension of the elongating phagophore membrane relies on two ubiquitin-like conjugation systems. The E1- and E2-like enzymes ATG7 and ATG10 conjugate ATG5 and ATG12. The resulting ATG5--12 conjugate binds to ATG16L1, and this complex acts as a E3-like enzyme in coordination with ATG7 as E1 and ATG3 as E2 to conjugate phosphatidylethanolamine (PE) to the GABARAP/light chain 3 (LC3) family of proteins, the most well characterized being LC3B ([@bib122]; [@bib30]). The ATG4 family of cysteine proteases cleave the LC3 family members to create LC3-I, which is conjugated to PE to generate LC3-II ([@bib78]; [@bib67]). Membrane-associated LC3-II associates with the autophagosome membrane and is critical as a target for recognition by adaptor proteins that bring specific substrates into the autophagosome for selective degradation. A handful of adaptor proteins have been identified, including the most well characterized, SQSTM1/p62, but also BNIP3, TAX1BP1, Optineurin, and NIX/BNIP3L, to name a few ([@bib2]). While LC3-II is dispensable for autophagosome formation, it is important for efficient autophagosome closure and fusion with lysosomes ([@bib95]). Consequently, delayed closure and formation of inefficient autophagosomes can still occur in the absence of the conjugation machinery and LC3-II ([@bib134]). Once closure is complete, the double-membrane autophagosome fuses with lysosomes using SNARE proteins, as well as the small GTPases, such as Rab7 ([@bib149]; [@bib49]; [@bib66]; [@bib11]; [@bib153]). Lysosomal enzymes then break down the cytoplasmic contents into amino acids and other macromolecular building blocks that are recycled into new macromolecules and fuel metabolic pathways. {#fig1} Nutrient-sensing pathways tightly regulate autophagy induction, most notably by the mammalian targets of rapamycin complexes, mTORC1 and mTORC2 ([@bib116]). In the case of amino acid deprivation, the inhibitory mTORC complexes can no longer phosphorylate and inhibit the ULK complex, triggering autophagosome membrane nucleation ([@bib53]). Decreased energy availability and ATP:AMP ratios in the absence of sufficient glucose triggers the 5′ AMP-activated protein kinase, which can inhibit mTORC signaling and also directly activate the ULK complex ([@bib63]). Another major mechanism of autophagy regulation is transcriptional and mediated by master transcriptional regulators that activate expression of the core ATGs. These include the transcription factor EB (TFEB; [@bib96]; [@bib118]), microphthalmia-associated transcription factor ([@bib101]; [@bib98]), and the forkhead family of transcription factors (FOXO1 and FOXO3; [@bib3]; [@bib152]; [@bib85]; [@bib117]). Recently, the epigenetic regulator bromodomain-containing protein 4 was identified as a transcriptional repressor of autophagy and lysosomal genes ([@bib115]). Transcriptional regulation of autophagy is thought to regulate the overall basal level of the process, and variations in transcriptional control may explain differences in basal autophagy between different cells/tissues ([@bib98]). On the other hand, posttranslational regulation through kinases controlled by stress and nutrient-sensing pathways such as mTOR largely explain acute changes in autophagy. Additional cellular stressors, including hypoxia, metabolic stress, and ER stress, among others, can induce autophagy through mechanisms that usually converge on these major nutrient-sensing pathways. Moreover, these pathways interact with each other for more coordinated and intricate regulation mediated by feedback loops ([@bib30]). For example, TFEB is negatively regulated by mTORC1 and the Ras-related GTP-binding protein D, which is an mTOR-activating Rag GTPase, a transcriptional TFEB target ([@bib29]). Importantly, autophagy can participate in bulk degradation of cytoplasmic material, but there are also selective forms of autophagy that rely on intricate signaling cascades to target specific proteins or organelles including mitochondria, ribosomes, bacteria, ferritin, ER, and peroxisomes: mitophagy, ribophagy, xenophagy, ferritinophagy, ERphagy, and pexophagy, to name just a few. Details regarding selective autophagy processes have recently been reviewed ([@bib2]). Autophagy and cancer {#s4} ==================== Tumor-cell intrinsic roles of autophagy in tumor growth {#s5} ------------------------------------------------------- Autophagy plays a complex role in tumor development and progression. Autophagy was first suggested to function as a tumor-suppressive process, with the discovery of Beclin1 deletions across multiple tumor types and the ability of Beclin1 to function as a haplo-insufficient tumor suppressor in mice ([@bib150]; [@bib104]; [@bib80]). Loss of other Beclin1 interacting proteins including endophilin B1 and UV radiation resistance-associated gene protein can also cause an increase in spontaneous tumors in mice and are mutated in human tumors, further supporting the Beclin complex as a tumor suppressor ([@bib79]; [@bib128]). It was reported, however, that in human tumors there is no significant loss of *BECN1* independent of the adjacent potent tumor suppressor *BRCA1* ([@bib71]). Additional tumor-suppressive functions of autophagy have emerged, including removal of damaged reactive oxygen species (ROS)--inducing mitochondria, maintenance of genomic stability, and a role in oncogene-induced senescence via degradation of the nuclear lamina ([@bib74]; [@bib139]; [@bib9]; [@bib32]). Together, these studies suggest that autophagy prevents tumor initiation and early steps in tumor progression. Consistent with this, genetic inhibition of other core genes that are themselves only very rarely mutated in cancer, including *Atg5* and/or *Atg7*, confirmed that deletion of autophagy in tumor-prone mouse models following RAS pathway activation or TP53 deletion caused an increase in preneoplastic lesions and even tumor incidence ([@bib112]; [@bib127]; [@bib106]; [@bib144]). However, these studies also showed that after a tumor is fully established, autophagy inhibition often results in less aggressive cancers. Successful tumor cells have to cope with harsh environments characterized by nutrient depletion, hypoxia, and other stresses. Autophagy is able to help cells cope with many of these stressors; thus although autophagy may serve to prevent tumor initiation, it often promotes tumor cell survival in more advanced cancers. Many studies have confirmed that genetic deletion of autophagy regulators (often Atg7 or Atg5) in established tumors from a variety of tissue types causes a dramatic reduction in tumor growth and a corresponding increase in survival of the animal. These effects have been attributed to blocked tumor growth and increased tumor cell death mediated by an accumulation of dysfunctional mitochondria, reduced fatty acid oxidation, reduced glycolytic capacity, elevated DNA damage, and impaired tumor cell metabolism ([@bib46], [@bib47], [@bib48]; [@bib57]; [@bib145], [@bib143]; [@bib102]; [@bib26]; [@bib127]; [@bib142]). Cancer cells vary in their dependence on autophagy. Some cancer cell lines are acutely sensitive to genetic inhibition of autophagy, while others show very little change in viability in the absence of autophagy. While little is known about what drives autophagy dependence in cancer, tumors with RAS-RAF-MEK-ERK pathway activation are often more sensitive to autophagy inhibition. Pancreatic ductal carcinomas (PDACs) for example, where RAS mutations are very common, are exquisitely sensitive to autophagy inhibition ([@bib145], [@bib144]; [@bib98]). TP53 status may also affect autophagy dependence, and loss of p53 in a humanized PDAC mouse model with KRAS activation eliminated the antitumor effects of autophagy inhibition ([@bib112]). Additional studies showed that these effects may apply only to germline deletion of TP53, as somatic loss of heterozygosity of TP53, i.e., the more common form of p53 deletion in human tumors, did not have the same effect ([@bib144]). Pediatric brain cancer cells with a constitutively active BRAF(V600E) mutation are more sensitive to both genetic and pharmacological autophagy inhibition ([@bib75]). Additionally, a variety of cancer cells without RAS pathway mutations have also been identified as extremely autophagy dependent in vitro ([@bib87]; [@bib133]). Many general cytotoxic chemotherapy drugs as well as multiple targeted therapies such as kinase inhibitors have been proposed to induce autophagy as a cytoprotective measure in cancer cells. For example, inhibitors of the MAP kinase pathway can induce autophagy in pancreas cancer, and combinations of these kinase inhibitors with autophagy inhibition can enhance tumor cell killing ([@bib16]; [@bib65]). These effects can even extend to overcoming acquired resistance to specific drugs. For example, patients with BRAF(V600E)-driven brain tumors that had become resistant to the BRAF inhibitor vemurafenib could be resensitized to the kinase inhibitor by addition of an autophagy inhibitor ([@bib93]; [@bib75]). Autophagy also promotes a stemness phenotype, not only in normal stem cells ([@bib140]; [@bib135]; [@bib42]), but also in "cancer stem cells" that can maintain a quiescent or dormant state, self-renew, and regenerate an entire tumor at limiting dilutions ([@bib88]; [@bib123]). In breast cancer models, genetic manipulation of core autophagy proteins can diminish the tumor-initiating properties of specific cancer stem cell populations in mice ([@bib147]). Similarly, in chronic myeloid leukemia (CML), the addition of autophagy inhibition to kinase inhibitors led to the complete elimination of functionally defined CML stem cells ([@bib10]). These effects may be due to autophagy-regulated metabolic changes in the stem cells ([@bib69]). Together, these basic cell biology, preclinical, and clinical studies support a cell-autonomous, tumor-suppressive role for autophagy at early stages in oncogenesis and a tumor-promotional and therapy-resistance role in more established tumors. Non--cell-autonomous roles of autophagy in tumor growth {#s6} ------------------------------------------------------- The tumor-promotional roles of autophagy are both cell autonomous and nonautonomous ([@bib143]) and are becoming increasingly recognized as important in controlling how the tumor microenvironment (TME) affects tumor growth ([Fig. 2](#fig2){ref-type="fig"}). However, as with the tumor cell-autonomous functions, non--cell-autonomous effects can be both pro- and antitumor. For example, autophagy in the tumor cells can affect whether immune cells in the TME recognize the tumor to either enhance or inhibit the antitumor immune response. Autophagy in the tumor cells was shown to be necessary for immunogenic cell killing following chemotherapy treatment to allow efficient tumor infiltration by antitumor dendritic cells and T-lymphocytes ([@bib90]). Moreover, autophagy induction by caloric restriction mimetics before treatment with anticancer drugs could improve anticancer immunosurveillance and decrease tumor burden ([@bib100]). These studies have been interpreted as meaning that tumor cell autophagy should be maintained or even enhanced to elicit a productive antitumor immune response. {#fig2} Several other studies, however, have demonstrated the opposite effect, i.e., tumor cell autophagy may reduce antitumor immune responses, for example, by showing that inhibition of autophagy in the tumor cells can enhance natural killer (NK) cell infiltration and reduce tumor growth ([@bib5]; [@bib89]). Genetic or pharmacological autophagy inhibition caused an increase in the transcript levels of the cytokine CCL5, mediated by decreased PP2A phosphatase activity and subsequent activation of JNK- and c-Jun--mediated transcription of CCL5 to attract NK cells into the TME ([@bib89]). It has also become clear that autophagy in the immune cells themselves is important. Pharmacological inhibition of autophagy with chloroquine (CQ) in tumor-associated macrophages causes a switch from the tumor-promotional M2 phenotype to an M1, tumor-killing, phenotype. CQ-reset macrophages decreased the immunosuppressive T cell populations and enhanced antitumor T cell immunity in mouse models ([@bib20]). Inhibition of autophagy in CD8^+^ T cells can also promote tumor rejection in mice by shifting T cell biology to a more effector memory phenotype and producing more cytokines, including IFNγ and TNFα. These effects were linked to enhanced glucose metabolism and altered epigenetic regulation in the T cells ([@bib28]). Processes that use components of the autophagy machinery but are independent of the degradative process of autophagy itself also affect tumor cell growth. Core autophagy proteins have been implicated in nonconventional secretory pathways ([@bib34]), host--pathogen interactions during infections, immune signaling, and inflammation ([@bib18]; [@bib40]). LC3-associated phagocytosis (LAP) utilizes a subset of core ATGs, in addition to the LAP-specific protein Rubicon, which enhances LAP but inhibits autophagy, to facilitate the conjugation of LC3 to phagosomal membranes. Inhibition of LAP, but not canonical autophagy, in tumor-associated macrophages causes an M1 (antitumor) phenotype that induces a STING-dependent type 1 IFN and antitumor T cell response ([@bib25]). Interestingly, the core LC3 conjugation machinery, including ATG5 and ATG7, is shared between autophagy and LAP; therefore the majority of studies conducted in mice to date using genetic knockout (KO) of these proteins cannot differentiate between LAP and canonical autophagy ([@bib52]). Beyond immune cell--mediated effects, autophagy in other nontumor cells within the TME can increase growth-promoting stimuli. For example, inhibition of autophagy in stroma-associated pancreatic stellate cells causes a decrease in secreted alanine, which is critical for PDAC cell growth. This mechanism involves a significant amount of cross-talk between the different cell types within the TME. The PDAC cells induce autophagy in the stellate cells to increase secretion of nonessential amino acids, which are then used by the PDAC cells to fuel the tricarboxylic acid cycle ([@bib126]). Recently, it was also shown that deletion of host autophagy causes a decrease in circulating arginine, and since many tumor cells are arginine auxotrophs, this can cause a corresponding decrease in tumor growth ([@bib102]). Another study showed that transplantation of autophagy-deficient, dormant tumors into autophagy-proficient hosts caused a reactivation of tumor growth mediated by TNF and IL-6--like signaling in a *Drosophila melanogaster* model ([@bib58]). Importantly, the dormant tumors remained small when they were transplanted into an autophagy-deficient host. These studies suggest that multiple mechanistically distinct mechanisms contribute to autophagy-dependent promotion of tumor growth from the microenvironment ([Fig. 2](#fig2){ref-type="fig"}). Autophagy and metastasis {#s7} ------------------------ Metastasis is a multistep process in which cancer cells acquire invasive phenotypes such as motility, the ability to survive under detached conditions, and stem cell--like properties including a more mesenchymal morphology by way of epithelial-to-mesenchymal transition (EMT; [@bib86]). Pathology studies in breast cancer and melanoma patients identified a correlation between elevated LC3B puncta staining and poorer outcomes and detected especially high LC3 expression in the metastatic lesions ([@bib73], [@bib72]; [@bib151]; [@bib50]). In addition to these correlative studies, autophagy has been shown to play a causative role in metastasis. Both genetic and pharmacological inhibition of autophagy eliminated lung metastasis in a syngeneic breast cancer mouse model after orthotopic injection of the mouse mammary carcinoma 4T1 cell line ([@bib119]). Autophagy has also been associated with EMT, a particular hallmark of metastatic dissemination that is linked to increased invasiveness as well as cancer stem cell phenotypes ([@bib77]; [@bib64]). Inhibition of autophagy-mediated degradation of RNA processing bodies, or p-bodies, mitigates a mesenchymal-to-epithelial transition, ultimately reducing metastatic outgrowth ([@bib121]). An additional mechanism by which autophagy can regulate cell migration and metastasis is dependent on autophagy-mediated regulation of focal adhesion dynamics. Autophagosomes interact with and turn over focal adhesion complexes to destabilize the cell-matrix contact sites, allowing for increased cell motility. This involves the autophagy cargo receptor NBR1 as well as direct interactions between LC3B and the focal adhesion protein, paxillin ([@bib119]; [@bib60]). In line with the competing pro- and antitumorigenic roles of autophagy at the primary tumor site, there have also been studies showing that autophagy can both promote and inhibit metastatic outgrowth at secondary sites. Metastatic cancer cells can be maintained in a dormant state for long periods of time at a secondary site only to initiate growth at a later time, resulting in appearance of metastatic tumors ([@bib125]). Transcriptomic analysis of surviving pancreatic cancer cells after oncogene ablation identified elevated expression of autophagy and lysosome-regulating genes, suggesting a role for these processes in the survival of subpopulations of dormant cells ([@bib137]). Similarly, in breast and ovarian cancer models, autophagy has been shown to promote the survival and outgrowth of dormant cell populations ([@bib84]; [@bib136]). Specifically, in D2.0 mouse mammary carcinoma cells (derived from murine mammary hyperplastic alveolar nodules) that can remain dormant in the lung for months before outgrowth, pharmacological autophagy inhibition caused a drastic reduction in metastatic outgrowth in the lung, suggesting that autophagy is critical for survival of the dormant cells ([@bib136]). However, using the same D2.0 model, a recent study confirmed elevated autophagic flux in the dormant cells, but in contrast concluded that inhibition of autophagy did not eliminate the dormant cells but instead caused them to emerge from the dormant state to reinitiate proliferation, causing an increase in metastatic outgrowth ([@bib70]). The discrepancy between these two studies could depend on differential host immunity (i.e., athymic mice vs. intact mice), different experimental procedures (i.e., time frame for quantification and imaging), or different means used to inhibit autophagy (i.e., pharmacological or genetic). Other mechanisms can also provide competing pro- and antimetastatic signals, even though the same autophagy-related proteins were manipulated. For example, in melanoma, the autophagy cargo receptor SQSTM1/p62, which is reduced in cells where autophagy is active, can interact with select RNA-binding proteins to stabilize specific prometastatic mRNAs ([@bib56]). After autophagy inhibition, the resulting accumulation of p62 can interact with and stabilize the pro-EMT and prometastatic protein, Twist1 ([@bib103]; [@bib13]). Conversely, p62 can also target prometastatic proteins for autophagic degradation, leading to a decrease in metastatic potential ([@bib129]). Targeting autophagy in cancer patients {#s8} -------------------------------------- While additional studies are necessary to fully understand the mechanisms of the tumor-suppressive and tumor-promotional roles of autophagy, both cell autonomously and nonautonomously, pharmacological and therefore systemic inhibition of autophagy is already moving forward in the clinic. A variety of pharmacological autophagy inhibitors and inducers are in development, and it is known that many approved drugs can affect autophagy ([@bib113]; [@bib55]). However, the lysosomal inhibitor CQ and its derivative hydroxychloroquine (HCQ) are the only drugs that are currently used in patients with the deliberate goal of targeting autophagy ([@bib132]). The first wave of clinical trials with CQ/HCQ in combination with other therapies showed promising results, indicating that target doses of the autophagy-inhibitor drugs could be reached with minimal toxicity ([@bib76]). Some antitumor effects were also observed. Glioblastoma patients treated with CQ combined with radiation and the alkylating agent, temozolomide (TZD), tripled their median survival over control patients (33 mo compared with 11 mo; [@bib15]). In melanoma, 41% of patients treated with CQ and TZD showed a partial response or stable disease, and 84% of patients with brain metastasis derived from a variety of solid tumors treated with CQ and radiation reported a 1-yr survival rate compared with 55% of patients treated with radiation alone ([@bib105]; [@bib109]). BRAF(V600E) mutant brain cancers are particularly dependent on autophagy for survival, and combined therapies with the BRAF-targeted therapy vemurafenib and CQ synergistically reduces cancer cell viability ([@bib75]). These results led to the clinical use of CQ in vemurafenib-resistant brain cancer patients who showed favorable results when treated with a CQ/vemurafenib combination ([@bib93]). Despite these initially promising studies, recent clinical trials have provided lackluster results. This is likely due to a variety of factors, including almost no biomarkers to identify the correct patients ideal for treatment, a need for more specific and potent autophagy inhibitors, and a lack in understanding of potential mechanisms of resistance. Novel autophagy inhibitors {#s9} -------------------------- While only CQ and HCQ have been approved for use in patients, other potent autophagy inhibitors are working their way through the pipeline. These include upstream inhibitors that target VPS34, including VPS34-IN1 ([@bib6]), PIK-III ([@bib33]), and SAR405 ([@bib110]). Other agents are in development that target the ULK1 kinase such as SBI-0206965 ([@bib35]), MRT67307, and MRT68921 ([@bib99]). Pharmacological inhibitors of the cysteine protease ATG4B are also being optimized to block LC3 conjugation ([@bib1]). The targets of these agents are annotated in [Fig. 1](#fig1){ref-type="fig"}, but the exact mechanism of action is described in detail elsewhere ([@bib23]). The pharmacological reagents with arguably the most clinical potential are CQ and quinacrine derivatives such as Lys05 and DQ661. Lys05, similar to CQ/HCQ, de-acidifies the lysosome and is thought to have increased antitumor potential over CQ or HCQ ([@bib23]). Up until recently, the exact target of CQ, HCQ, and Lys05 was unknown; however, due to the increased lysosomal specificity and potency of DQ661, an in situ photoaffinity pull-down assay could be used. While palmitoyl-protein thioesterase 1 was identified as the direct target of all the CQ and quinacrine derivatives, DQ661 appears to be the most specific and have the most potent antitumor activity in preclinical models ([@bib108], [@bib107]). Cooperation between autophagy, apoptosis, and necroptosis {#s10} ========================================================= Mechanisms of cell death {#s11} ------------------------ Cell death is a highly regulated and programmed process and has recently been subdivided into 12 interconnected subclasses based on distinct signaling pathways and has been extensively reviewed elsewhere ([@bib41]). The 12 subroutines have been historically grouped into three major categories based on cell morphology, including apoptosis, necrosis, and autophagic cell death ([@bib44]; [@bib62]; [@bib41]). Apoptosis can be initiated either by external stimuli reliant on membrane-bound death receptor signaling or via the intrinsic pathway that is dependent on mitochondrial signaling and is stimulated by internal cues such as DNA damage or ER stress. Both forms of apoptosis share a dependence on caspase cascades converging on cleavage of the executioner caspases 3 and 7, resulting in an orderly cell death process characterized by plasma membrane blebbing, chromatin condensation, cell shrinkage, and nuclear fragmentation. In contrast, necrosis is morphologically distinct and is accompanied by cell swelling and lysis without nuclear condensation. While necrosis can occur after an overwhelming insult lacking controllable signaling cascades, it has become clear in recent years that it can also be programmed and driven by distinct signaling pathways. For example, necroptosis involves activation of receptor-activating protein kinase 3 (RIPK3), which then activates mixed-lineage kinase domain-like pseudokinase (MLKL) that serves as the mediator of cell lysis and death ([@bib41]). The third major form of cell death, autophagic cell death, is the most controversial; it involves the appearance of large vacuoles and is reliant on the core autophagy machinery ([@bib44]). There are distinct contexts where autophagy appears to be necessary for physiological cell death, particularly during *Drosophila* development ([@bib12]; [@bib27]). Autophagic cell death has been reported in cancer cells as a potential safeguard to prevent RAS-induced oncogenic transformation ([@bib36]). ATG5 was reported to be necessary for apoptosis-independent death observed in epithelial cell lines transformed with either HRAS or KRAS ([@bib17]). ROS-inducing agents also stimulate autophagic cell death that cannot be attenuated with caspase inhibitors in different cancer cell lines ([@bib21]). While these initial studies may have reported autophagic cell death, all other forms of cell death were often not properly ruled out. Generally, autophagy is a prosurvival and cytoprotective mechanism, and failed attempts at survival mediated by autophagy often accompany cell death morphologies ([@bib120]). Thus, it is often difficult to exclude the possibility that activation of autophagy in a dying cell was not in fact contributing to the death but was instead an effort by the cell to avoid dying. The current recommendation ([@bib41]) is that the term "autophagic cell death" should refer only to a death process that requires the autophagy machinery but does not involve the other major death mechanisms such as apoptosis or necroptosis. Recently, autophagic cell death was shown to be important in epithelial cells during crisis after telomere dysfunction, implying that activation of an autophagy-dependent form of death acts as a principal tumor-suppressive barrier ([@bib94]). The debate of prosurvival versus prodeath autophagy in cancer cells is, of course, relevant to the tumor-suppressive versus tumor-promotional roles of the process, as well as how cancer cells respond to therapy. Defining stimuli that induce autophagic cell death while identifying strategies that inhibit the cytoprotective functions of autophagy will be critical to optimize the therapeutic use of autophagy inhibitors in the clinic. Additional complications may arise when a single stimulus can induce autophagic cell death but only under some contexts. For example, [@bib4] showed that the DNA damaging agent etoposide can induce apoptotic cell death in normal oxygen conditions; however, the mechanism of cell death switched to autophagic cell death in hypoxia. Interestingly, while it is clear that some cancer cells are prone to die after autophagy inhibition, either genetic or pharmacological, it is unclear exactly how these cells are dying. Caspase activation has been noted in these instances, but any nonapoptotic contributions to the death such as different types of programmed necrosis have not been ruled out. Autophagy promotion of other forms of cell death {#s12} ------------------------------------------------ Beyond autophagic cell death, it is clear that the autophagic machinery can be intertwined with the apoptotic pathways, and the two can either cooperate or inhibit each other depending on the context and inducing stimuli ([@bib31]). While the above experiments suggest that autophagy plays a direct role in cell death, there are a number of experiments indicating a more indirect role where autophagy affects apoptosis. Different anti-apoptotic proteins have been identified as autophagy substrates. For example, some cells with elevated autophagic flux are more sensitive to FAS-induced apoptosis due to selective autophagic degradation of the tyrosine phosphatase, FAP-1/PTPN13. Interestingly, these results were stimulus specific, as FAP-1 specifically modulates FAS-induced apoptosis; accordingly, elevated autophagy in the exact same cells does not sensitize cancer cells to a very similar apoptotic stimulus, such as TNF-related apoptosis-inducing ligand (TRAIL)--induced apoptosis ([@bib45]). In hippocampal astrocytes, autophagic degradation of the anti-apoptotic protein, Cav-1, was shown to be important for autophagy-induced apoptosis in the presence of palmitic acid ([@bib22]). Additionally, in mouse embryonic fibroblasts, caspase-8 can complex with Atg5 and interact with LC3 and p62 on the autophagosome membrane. The intracellular death-inducing signaling complex, including Fas-associated death domain (FADD), a necessary adaptor protein for caspase-8 activation, is also recruited to the complex and interacts with ATG5 on ATG16L- and LC3-positive autophagosomes. These interactions promote caspase-8 self-processing to propagate the caspase signaling cascade and cause cell death ([@bib148]; [@bib130]). Autophagy inhibition of cell death {#s13} ---------------------------------- Although the above examples show that under certain circumstances autophagy can promote apoptosis, autophagy more commonly affects cell death by reducing apoptosis. Genetic inhibition of core ATGs causes an increase in the pro-apoptotic protein, PUMA (p53 up-regulated modulator of apoptosis) and a corresponding increase in complete mitochondrial outer membrane permeabilization, usually the point of no return in apoptosis ([@bib131]). Interestingly, PUMA is not a direct autophagic substrate but instead is transcriptionally regulated by the pro-apoptotic transcription factor, FOXO3a ([@bib140]). The autophagic turnover of FOXO3a mediates changes in PUMA mRNA levels, and deletion of a single intronic FOXO3 binding forkhead response element (FHRE) in the endogenous PUMA gene eliminated both pharmacological and genetic autophagy inhibition-induced PUMA expression ([@bib38]). Consequently, the PUMA FHRE was shown to be critical for synergistic induction of apoptosis when pharmacological autophagy inhibition with CQ was combined with other cytotoxic agents including etoposide, doxorubicin, and the p53-targed therapy, Nutlin-3a. Interestingly, these experiments were conducted in a cell model characterized as autophagy independent, where treatment with CQ alone did not cause cell death. Instead, autophagy inhibition pushed the cells closer to their apoptotic threshold, allowing addition of the other drug to more easily induce apoptosis. Other molecularly distinct mechanisms by which autophagic turnover of a protein can reduce apoptosis sensitivity have also been identified. For example, the initiator caspase and pro-apoptotic protein caspase-8 has also been identified as an autophagic substrate to coordinate cross-talk mechanisms between autophagy and apoptosis after treatment with TRAIL ([@bib54]). Autophagic degradation can also inhibit necroptosis. Proteins critical for necroptosis, including RIPK1, RIPK3, TBP1, and TRIF, all of which contain RIP homotypic interaction motifs, are autophagy substrates. Inhibition of autophagy leads to accumulation of these proteins and subsequent activation of TNF- and Toll-like receptor--mediated necroptosis ([@bib81]). Autophagy regulation of switching between forms of cell death {#s14} ------------------------------------------------------------- The autophagosome can also act as a scaffolding complex to mediate switching between apoptosis and necroptosis. In mouse prostate cells with loss of the tumor suppressor *MAP3K7*, RIPK1 and the necrosome are recruited to the autophagosome membrane in a p62-dependent manner to facilitate TRAIL-induced necroptosis. Interference with RIPK1, RIPK3, or MLKL or recruitment of the complex by knocking down p62 caused the mode of death to switch from necroptosis to apoptosis after treatment with TRAIL ([@bib43]). A similar model of drug-induced necroptosis was described in rhabdomyosarcoma cells, in which treatment with the small-molecule BCL-2 inhibitor obatoclax induced formation of the necrosome, including FADD, RIPK1, and RIPK3, on the outer membrane of autophagosomes via interactions with ATG5 ([@bib8]). Higher ratios of necroptosis compared with apoptosis in dying cancer cells can elicit a greater antitumor immune response and activation of CD8^+^ T cells ([@bib146]; [@bib124]), suggesting that the autophagy machinery through this scaffolding function may play an integral role in immunogenic cell death. These results show that the formation of the autophagosome membrane is a critical event in mediating cell death, suggesting that autophagy inhibition at earlier stages that block autophagosome formation may have opposing effects compared with later-stage autophagy inhibitors that affect autophagic turnover. More work is needed to fully understand the stimuli, cellular contexts, and necessary interacting proteins to fully elucidate the interplay between autophagy, apoptosis, and necroptosis. Autophagy as a mechanism to evade apoptosis in cancer and its treatment {#s15} ======================================================================= One of the key hallmarks of cancer is "evading cell death," particularly evading programmed cell death pathways such as apoptosis. There have been a number of mechanisms to describe how cancer cells can evade apoptosis: loss of the classic death receptors, inactivation of different caspases, and aberrant expression of the BH3 proteins that regulate mitochondrial-mediated intrinsic death pathways. Many of these mechanisms have been reviewed elsewhere ([@bib39]). Up-regulation of autophagy as a prosurvival mechanism has also been implicated as a mechanism for cancer cells to evade apoptosis. Autophagy deficiency increases attachment-induced apoptosis (anoikis) classified by elevated cleaved caspase-3 expression in HRAS(V12)-transformed mouse embryonic fibroblasts ([@bib83]). In addition, immunohistochemistry staining in tumors derived from HRAS(V12)-transformed cells grown in nude mice showed a dramatic increase in active caspase-3 staining when core autophagy genes *Atg5* or *Atg7* were also knocked out ([@bib46]). In both cell lines and primary cells isolated from glioblastoma patients harboring a BRAF(V600E) mutation, pharmacological autophagy inhibition with CQ induced apoptosis ([@bib75]; [@bib93]). Cells without RAS pathway mutations are also particularly dependent on autophagy to evade apoptosis, and shRNA and CRISPR-mediated screens have shown that some human cancer cell lines undergo programmed cell death when core autophagy genes are knocked down/knocked out or after pharmacological autophagy inhibition with CQ ([@bib87]; [@bib133]). A large body of literature indicates that cancer cells can up-regulate autophagy to evade chemotherapy-induced apoptosis, resulting in chemoresistance ([@bib92]). Preclinical cell culture and animal studies showed that pharmacologically targeting autophagy increases cell killing when combined with other chemotherapeutic agents such as cisplatin in esophageal squamous cell carcinoma, the histone deacetylase inhibitor suberoylanilide hydroxamic acid in CML ([@bib19]; [@bib82]), and BRAF inhibitors in brain and pancreatic cancer ([@bib93]; [@bib16]; [@bib65]). These studies and others have provided the rationale to launch dozens of clinical trials combining autophagy inhibition with a variety of different chemotherapies ([@bib76]). While there is little doubt that autophagy inhibition can increase cell killing with other agents, the hypothesized mechanism for why this works is based on the idea that the accompanying chemotherapy induces a protective prosurvival form of autophagy that, upon inhibition, leads to cell death. This hypothesis implies that the increased protective autophagy is downstream of the chemotherapy target, whether that be general DNA damage or specific tumor-driving kinases. An alternate hypothesis based on the mechanistic understanding of how autophagy inhibition enhances drug-induced apoptosis via the FOXO3/PUMA mechanism described above proposes that autophagy may be the upstream mediator and may instead affect how "primed" a cell is to die ([@bib37]). Importantly, these two hypotheses (autophagy inhibition works to enhance cancer therapy because it is blocking cancer drug-induced autophagy that counteracts the death signal; and autophagy inhibition works because it pushes cancer cells closer to their apoptotic threshold) are likely not mutually exclusive. It is important to tease apart the exact mechanisms that link these processes to better leverage autophagy inhibition as a cancer therapy. Adaptation to autophagy inhibition {#s16} ================================== Can cancer cells evade autophagy inhibition? To date there have been few, if any, studies to directly address this question. There have yet to be clinical trials conducted that empirically test for resistance to pharmacological autophagy inhibition in patients. However, previously published trials already show hints of acquired resistance. Most of the trials with published results have been phase I or II, focused on safety and pharmacokinetics-pharmacodynamics, and were not designed for this type of analysis. Nonetheless, in the phase I clinical trial with CQ and TZD, [@bib105] described a patient who initially responded to the therapy but eventually succumbed to a brain metastasis with persistent growth after 4 mo on therapy. Two canine patients in a phase I trial with HCQ and doxorubicin in dogs showed complete initial responses but eventually were removed from the study due to subsequent progressive disease ([@bib7]). In the phase I trial combining the proteasome inhibitor bortezomib with HCQ, 45% of the patients achieved stable disease that eventually progressed after 9--14 wk ([@bib138]). Such results suggest it is plausible that tumors can adapt to autophagy inhibition ([Fig. 3](#fig3){ref-type="fig"}). Purely pharmacological mechanisms of resistance to autophagy-inhibiting drugs such as CQ have been identified. The extracellular pH of the TME can prevent drug uptake ([@bib24]), and cancer cell lines grown in acidic media are resistant to CQ-induced toxicity. Moreover, in tumors grown in mice, normoxic regions with physiological pH showed an expected increase in LC3 expression indicative of a CQ-induced block of autophagy, whereas the hypoxic regions with a more acidic pH were significantly less responsive to CQ ([@bib97]). {#fig3} In addition to pharmacological mechanisms of resistance, there are also likely to be genetic mechanisms of resistance; however, very little work has been done to address this question. A recent study used a rapid CRISPR/Cas9 assay to analyze how KO of 12 different core autophagy genes affects cancer cell viability and growth during the first 7 d after gene loss ([@bib133]). By comparing each gene to loss of known essential and nonessential genes, cell lines where multiple autophagy genes behaved like essential genes could be identified. But, even in autophagy-dependent cell lines where CRISPR-mediated KO of *ATG7* caused the majority of the cells to die within 48 h after editing, at much later time points *ATG7*-deficient clones could be isolated. Intriguingly, the selected *ATG7^−^*^/−^ clones that were derived from autophagy-dependent cells grew at equal rates compared to the WT cells from which they were derived, even in autophagy-inducing conditions such as nutrient starvation or hypoxia ([@bib133]). These adapted cells were also resistant to pharmacological inhibitors of autophagy such as CQ. These results indicate that under enough selective pressure, autophagy-dependent cancer cells can adapt to circumvent autophagy inhibition. *ATG7^−^*^/−^ clones derived from originally autophagy-dependent cells acquired an increased dependence on the master antioxidant transcriptional regulator, nuclear factor erythroid 2-related factor 2 (NRF2), and this was critical to maintain protein homeostasis ([Fig. 3](#fig3){ref-type="fig"}). NRF2 has previously been linked to autophagy, and p62 sequesters and inhibits the NRF2-negative regulator, KEAP1 ([@bib68]). Moreover, a recent genome-wide CRISPR screen showed that KO of a large subset of autophagy genes results in up-regulation of the NRF2 signaling pathway ([@bib61]). As a consequence of NRF2 up-regulation, the autophagy-deficient cells developed an increased sensitivity to pharmacological proteasome inhibitors, a phenotype that was exacerbated with NRF2 knock-down ([@bib133]). These results suggest that tumors that originally start off sensitive to pharmacological autophagy inhibition may be able to adapt and acquire mechanisms of resistance to these therapies. More studies investigating this phenomenon are needed to understand the underlying molecular mechanisms that might provide insight into the correct combinatory therapies to prevent resistance to autophagy inhibition. It will be important to incorporate analysis of these mechanisms into clinical trials, e.g., to investigate if NRF2 expression correlates with clinical response to autophagy inhibition. Conclusions {#s17} =========== Together, the preclinical and clinical studies suggest that autophagy inhibition may be a viable cancer therapy ([@bib76]). However, these initial studies also indicate that both inherent and acquired mechanisms of resistance may be a significant hurdle to overcome before autophagy inhibition can be a truly effective therapy in cancer patients. Additional studies are needed to better identify which patient populations are best suited for autophagy inhibition, and the initial studies in RAS pathway mutated tumors may provide a framework for such studies. It is also unlikely that autophagy is consistently inhibited in cancer patients treated with HCQ ([@bib111]; [@bib141]; [@bib14]). Better pharmacological autophagy inhibitors are needed, such as the dimeric quinacrines, which are more potent and selective autophagy inhibitors ([@bib108]). Additional basic cell biology studies are needed to better understand autophagy's various functions in cancer cell biology and to understand the compensatory mechanisms that are up-regulated when autophagy is inhibited. For example, it is unclear how cells that survive loss of autophagy can maintain mitochondrial homeostasis and turn over ER or ribosomes, all of which use LC3-conjugated autophagosomes. But even larger questions still remain, such as what expression and/or mutational landscape dictates autophagy dependence in the first place? When autophagy is inhibited in autophagy-dependent cancer cells, what is the mechanism of subsequent death? Are there specific therapies that work better in autophagy-independent tumors? Lastly, based on the changes in cell biology that occur in the context of autophagy inhibition and the complex interactions with the TME, what combination therapies will lead to the most robust patient responses? Ultimately, the goal is to have a rational mechanistic basis with which to target autophagy manipulations in tumors to maximize beneficial effects, by, for example, increasing the tumor cell killing capacity of other agents while at the same time circumventing the mechanisms by which cancer cells will adapt to the autophagy manipulation. Achieving this goal may allow long-lasting benefits from autophagy-targeted therapies in patients but will require much deeper understanding of how autophagy controls the cell biological processes that drive cancer cell behavior. This work was supported by National Institutes of Health grants RO1CA150925 (A. Thorburn), RO1CA190170 (A. Thorburn), and 5 T32 CA 190216-2 (C.G. Towers), and the American Cancer Society Postdoctoral fellowship AWD 183648 (C.G. Towers). The authors declare no competing financial interests.

We would like to submit the following corrections to our recently published paper \[[@B1-nutrients-12-01263]\] after discovering the inclusion of three studies in our review that did not meet the inclusion criteria we had specified in our methods: healthy adults (≥18 years old) or individuals with overweight/obesity and/or type 2 diabetes (T2D). Specifically, all three studies included participants with type 1 diabetes (T1D) which was part of our exclusion criteria as described in our manuscript. The reason for the original inclusion of these three studies stems from confusion in the older terminology "insulin-dependent diabetes" that was eventually eliminated and replaced with "T1D" since the term insulin-dependent diabetes mellitus "has been confusing and has frequently resulted in classifying the patient based on treatment rather than etiology" as stated by the Expert Committee on the Diagnosis and Classification of Diabetes Mellitus report \[[@B2-nutrients-12-01263]\]. We feel the correction explained herein is an example of this confusion and the purpose for updating this terminology. Furthermore, this correction could serve as a reminder to readers who do not work in this subject extensively to note the ambiguity surrounding the old terminology. Removal of these studies does not change the overall or specific results or conclusions, since these three studies contributed to the largest section of studies addressing "exercise relative to breakfast/morning meal consumption," and resulted in similar findings. Therefore, revisions did not impact the hypothesis, results, and conclusions of the review. **Abstract:** In the original paper, the total number of included studies was *n* = 20 and is *n* = 17 in the revised version. Also, in the original paper, the total number of participants was *n* = 352, while it is *n* = 332 in the revised paper. **1. Materials and Methods**: In the original paper, Figure 2 specified the inclusion of 20 studies, while the revised paper specifies *n* = 17 included studies. **2. Results**: In the original paper, **page 4, paragraph 1**, we stated "**Twenty** studies met the inclusion criteria for this systematic review (Table 1) **Eighteen** studies included moderate-intensity aerobic exercise, three examined high-intensity. Furthermore, **thirteen** studies investigated exercise relative to breakfast/morning meal \[17,19--22,26--28,31,32,35--37\], four included dinner/evening meal \[24,29,30,33\], and three examined exercise performed at several time points throughout the day \[18,25,34\]." The revised version should read as follows "**Seventeen** studies met the inclusion criteria for this systematic review (Table 1) **Fifteen** studies included moderate-intensity aerobic exercise, three examined high-intensity Furthermore, **ten** studies investigated exercise relative to breakfast/morning meal \[17,19--21,25,28,31--34\], four included dinner/evening meal \[23,26,27,29\], and three examined exercise performed at several time points throughout the day \[18,24,30\]." In the original paper, **Table 1:** summary of studies included in the systematic review, **section:** exercise relative to breakfast/morning meal consumption, included studies by Caron et al. (reference 32), Rasmussen et al. (reference 27), and Ruegemer et al. (reference 20). In the revised paper, we have removed all three studies from this section. Also, in the original version, the study by Nelson et al. (reference 26) specified insulin-dependent diabetes mellitus (IDDM) as the condition with *n* = 16; the revised version specifies healthy individuals with *n* = 7 (This study included healthy subjects and those with IDDM; sample persons and results pertaining to subjects with IDDM were removed). Additionally, the original paper included all three studies (Caron et al., Rasmussen et al., and Ruegemer et al.) in **Table 2:** risk of bias assessment of included studies, which were removed in the revised paper. In the original paper, in the section "exercise relative to breakfast/morning meal consumption" **pages 11--12, paragraph 2**, the sentence "Evidence regarding the glucose lowering effect of exercise performed pre-breakfast was mixed" has been modified in the revised version as "Evidence regarding the glucose lowering effect of exercise performed pre-meal was limited." In the original paper, in the section "exercise relative to breakfast/morning meal consumption" **pages 11--12, paragraph 2**, the sentences "On the other hand, Ruegemer et al. examined the optimal timing of exercise in participants with insulin-dependent diabetes mellitus receiving intensive insulin therapy \[20\]. Participants were provided with standardized meals and were studied on four occasions including two at rest and two with 30 min exercise performed pre-breakfast (7 a.m.) or in the afternoon (4 p.m.). Findings revealed a significant increase in plasma glucose concentration in the pre-breakfast exercise condition compared to exercising in the afternoon or in the no exercise condition (*p* \< 0.05), however, there was no difference in any of the groups 4 h after the meal." have been removed from the revised paper. In the original paper, in the section "exercise relative to breakfast/morning meal consumption", **page 12, paragraph 3**, the sentence "Three crossover trials investigated the effect of moderate-intensity exercise performed post-breakfast on glycemic response using a similar methodological approach \[21,27,32\]." has been removed from the revised version. In the original paper, in the section "exercise relative to breakfast/morning meal consumption", **page 12, paragraph 3**, the statement "Two other studies found a beneficial effect of post-meal exercise on PP glucose levels in participants with T2D \[27,32\] and revealed that this effect persisted at least for the duration of the next meal (p \< 0.05) \[32\]" has been removed from the revised paper. In the original paper, **page 12, paragraph 3**, we stated "Nelson et al. further confirmed these findings in a randomized controlled trial that characterized the metabolic response to moderate-intensity exercise performed post-meal in healthy individuals (controls) and those with insulin-dependent diabetes mellitus \[26\]. Participants were provided with a standardized breakfast after which they either rested for 3 h or exercised for 45 min. Compared to no exercise, PP glycemic response was significantly lower in the exercise condition between 45--75 min and 65--95 min in controls and participants with insulin-dependent diabetes mellitus, respectively (both *p* \< 0.05)." In the revised paper, we edited this paragraph to read "Nelson et al. further confirmed these findings in a randomized controlled trial that characterized the metabolic response to moderate-intensity exercise performed post-meal in healthy individuals \[33\]. Participants were provided with a standardized breakfast after which they either rested for 3 h or exercised for 45 min. Compared to no exercise, PP glycemic response was significantly lower in the exercise condition between 45--75 min (*p* \< 0.05)." We additionally moved this paragraph to the section pertaining to studies on exercise performed post-breakfast among healthy participants. In the original paper, **page 13, paragraph 7**, the sentence "However, results were less consistent in regards to the effect of pre-meal exercise performed in the morning on glycemia in participants with T2D" has been revised to "However, results were limited in regards to the effect of pre-meal exercise performed in the morning on glycemia in participants with T2D". **3. Discussion**: In the original paper, **paragraph 1, page14**, we specified *n* = 352 as the total number of participants and twenty as the total number of studies, while the revised version includes *n* = 332 and seventeen studies as the total number of included studies. In the original paper, **page 14, paragraph 3,** we stated "Studies conducted in the morning resulted in inconsistent findings, with one trial reporting elevated blood glucose in response to pre-breakfast exercise \[20\]; while another revealed a decrease in mean 24 h glucose concentration in exercise conditions compared to control, although, the improvement from exercise was observed in the second (\~4.5 h post-exercise) but not in the first meal (\~30 min post-exercise) \[28\]." This sentence was revised to read as follows "One study was conducted in the morning and revealed a decrease in mean 24 h glucose concentration in exercise conditions compared to control, although, the improvement from exercise was observed in the second meal (\~4.5 h post-exercise) but not in the first meal (\~30 min post-exercise) \[25\]." In the original paper, **pages 14--15, paragraph 3**, we stated "The large heterogeneity of these two studies is an important consideration. Specifically, studies differed in mean age of recruited participants (30.0 vs. 60.1 year), health condition (IDDM vs. T2D), duration of exercise (30 vs. 60 min), and duration of PP blood sampling (up to 3 vs. 24 h) in \[20,28\], respectively." These two sentences were removed from the revised paper. In the original paper, **page 15, paragraph 3,** we stated "Only one study investigated the effect of walking pre-meal in the evening and reported no difference in most examined glycemic outcomes in the exercise condition compared to no exercise \[33\]." This sentence was revised to read as follows "Additionally, Rees et al. investigated the effect of walking pre-meal in the evening and reported no difference in most examined glycemic outcomes in the exercise condition compared to no exercise \[29\]." **4. Conclusions**: No changes were made to this section. The authors apologize to the readers for any inconvenience caused by this amendment. This amendment does not affect the results or conclusion of the manuscript in any way. The original manuscript will remain online on the article webpage with a reference to this correction. This research was supported by the National Cancer Institute of the National Institutes of Health under award number R21CA224764 and Purdue University. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Prior to issuing a press release concerning the outcomes of this research, please notify the NIH awarding IC in advance to allow for coordination. The authors declare no conflict of interest.

{.250} {.251} {.252} {.253} {.254} {.255} {.256} {.257} {.258}

INTRODUCTION {#sec1-1} ============ Histopathologic classification of pancreatic endocrine neoplasms is complex and the recent World Health Organization (WHO) classification of tumors of the digestive system divides pancreatic endocrine tumors into three grades based on mitotic rate and proliferation index (PI) as determined by Ki-67 immunohistochemical staining.\[[@ref1][@ref2][@ref3]\] The WHO classification is formulated for the evaluation of histologic specimens which are paraffin embedded and formalin fixed. Specific recommendations are not given for the evaluation of cytologic specimens. The Papanicolaou Society of Cytopathology (PSC) has developed a set of guidelines for the investigation and diagnosis of pancreatic and biliary tract lesions.\[[@ref4]\] These guidelines recommend a categorization scheme and diagnostic criteria. The scheme includes six categories composed of "nondiagnostic," "negative," "atypical," "neoplastic," "suspicious for malignancy," and "malignant."\[[@ref4]\] The "neoplastic" category is further sub-divided into neoplasm, benign, and "neoplasm, other". Following the PSC criteria, pancreatic endocrine neoplasms are placed in either the "neoplasm, other" or "malignant" categories depending on cytologic features. Several authors have applied the WHO recommendations for classification to formalin fixed paraffin-embedded cell block specimens obtained by fine needle aspiration (FNA).\[[@ref5][@ref6][@ref7][@ref8]\] These studies have shown good correlation between results obtained from cell block specimens and subsequent surgical specimens.\[[@ref6][@ref7][@ref8]\] Analysis of histologic specimens using the WHO criteria appears to have clinical relevance.\[[@ref9]\] Endoscopic ultrasound (EUS) guided FNA biopsies of pancreatic lesions do not invariably result in cell block material and may produce only smear preparations. Such preparations may be stained by either the Diff-Quik or the Papanicolaou methods, and if rapid on-site evaluation is utilized, a significant proportion of the resultant smears are Diff-Quik stained. This may result in the only material available for evaluation being air-dried Diff-Quik stained smears. Little information exists as to criteria for separating cytologic specimens of pancreatic endocrine neoplasms into low-, intermediate-, and high-grade lesions corresponding to the WHO system. The PSC guidelines recommend categorizing pancreatic endocrine neoplasms as high-grade lesions (malignant, pancreatic endocrine carcinoma) and low- or intermediate-grade lesions ("neoplasm, other", pancreatic endocrine tumor).\[[@ref4]\] While the high-grade malignant lesions are invariably aggressive neoplasms and require an aggressive therapeutic approach, neoplasms cytologically characterized as "neoplasm, other (pancreatic endocrine tumor)" have a variable biological behavior with some showing aggressive features including metastases while others have a more indolent behavior. Malignant neuroendocrine carcinomas in the PSC classification are histologically small cell and large cell neuroendocrine carcinomas with aggressive behavior. The PSC guidelines do no offer criteria for separating pancreatic endocrine neoplasms categorized as "neoplasm, other" into those likely to behave aggressively and those with a more indolent clinical course. Prior studies have demonstrated the utility of cell block material for such separation but criteria for the analysis of smear preparations have not been published. We utilized a series of 36 pancreatic endocrine tumors including low-, intermediate-, and high-grade examples to develop a set of cytomorphologic criteria and a related scoring system to identify pancreatic endocrine neoplasms with a poor outcome in smear preparations. Herein, we report the results of that study. METHODS {#sec1-2} ======= The present study was reviewed by the Institutional Review Board and determined to be exempt. The study follows the tenets of the Helsinki agreement. The case files of the Department of Pathology and Anatomical Sciences at the University of Missouri and the Department of Pathology and Laboratory Medicine at the University of Utah/ARUP Laboratories were searched for all FNA biopsies of pancreatic endocrine neoplasms performed between January 2008 and January 2014. Cases had a clinical follow-up of between 1 and 7 years. Thirty-two cases classified as "neoplasm, other, pancreatic endocrine neoplasm" and four classified as "malignant, pancreatic endocrine carcinoma" were identified and the corresponding slides obtained. The malignant cases included two small cell carcinomas and two large cell neuroendocrine carcinomas. Cases with biopsy or resection specimen were graded using the WHO recommendations and classified as neuroendocrine neoplasm Grade 1, neuroendocrine neoplasm; Grade 2 or neuroendocrine neoplasm; Grade 3. The histologic grading was based on Ki-67 index, mitotic counts, the presence or absence of necrosis, and the presence or absence of angioinvasion. Histology, in some cases, was performed on only biopsy specimens while others were resection specimens. All cases had Diff-Quik stained smears, and these smears formed the basis of the present study. For each case, twenty morphologic features \[[Table 1](#T1){ref-type="table"}\] were independently evaluated by two raters. Evaluations were made only on the basis of morphology as seen in Diff-Quik stained slide material. Diff-Quik material was used as this is the preparation type used for rapid on-site evaluation. Raters were blinded to the original diagnoses and did not have access to clinical information. Both raters were board-certified cytopathologists with 8 and 30 years of experience, respectively. The morphologic features scored are listed in [Table 1](#T1){ref-type="table"}. Chromatin stringing is defined as artifactual pulling of nuclear chromatinic material into long chords due to the smearing process. For cytologic evaluation, mitotic activity was graded as absent (no mitotic figures seen) or present (one or more mitotic figures seen in the smear). Morphologic features were classified as absent or present (either focal or diffuse). Outcomes were classified as poor or not poor. Poor outcomes included metastases, death, or high-grade lesions (pancreatic endocrine carcinoma) as determined by histologic evaluation of surgical specimens. ###### Inter-rater agreement statistics for morphologic features  Associations between morphologic features (absent vs. present) and outcomes (poor vs. not poor) were evaluated using hierarchical logistic regression. Features were included as fixed effects and raters were included as random effects. Inter-rater agreement was assessed with absolute agreement and chance-corrected agreement (Kappa). Discriminatory power was assessed using the area under the receiver operating characteristic (AUROC) curve. Statistical calculations (Kappa, logistic regression, ROC analysis) were performed using the Stata 13 (Stata Corporation, College Station, TX, USA). Recursive partitioning was conducted using the JPM 11.0 (SAS Corporation, Cary, NC, USA). RESULTS {#sec1-3} ======= Absolute inter-rater agreement ranged from 51.4% to 97.1% across the 20 morphologic features \[[Table 1](#T1){ref-type="table"}\]. [Table 2](#T2){ref-type="table"} shows the corresponding histologic diagnoses and clinical outcome with the cytologic category as poor or not poor outcome. The absolute agreement exceeded 70% for 14 of the 20 features. The chance-corrected agreement (Kappa) ranged from 0.12 to 0.54. Raters achieved Kappa scores of at least 0.3 for seven morphologic features (chromatin stringing, prominent nucleoli, irregular nuclear membranes, nuclear grooves, mitotic figures, necrosis, and 3-fold variation in nuclear size). ###### Tumor histologic grade, clinical follow-up, and cytologic risk for a poor outcome  Sensitivity and specificity both range from 4% to 100% \[[Table 3](#T3){ref-type="table"}\]. Twelve of the twenty morphologic features showed statistically significant associations with poor outcomes \[[Table 4](#T4){ref-type="table"}\]. Overall discriminatory power was highest for irregular nuclear membranes (AUROC = 0.76), mitoses (AUROC = 0.71), 3-fold variation in nuclear size (AUROC = 0.67), and necrosis (AUROC = 0.65). ###### Accuracy statistics for morphologic features, area under the receiver operating characteristic curve  ###### Strength of association between morphologic features and poor outcomes  Many of the morphologic features had statistically significant associations \[[Table 5](#T5){ref-type="table"}\]. For example, chromatin stringing had a statistically significant correlation with mitotic figures (*r* = 0.50, *P* \< 0.001). Irregular nuclear membranes had statistically significant correlations with nuclear grooves, mitotic figures, 3-fold variation in nuclear size, and the presence of necrosis. ###### Pairwise correlation matrix for selected morphologic features  Recursive partitioning analysis showed that mitotic figures had the highest discriminatory power \[[Figure 1](#F1){ref-type="fig"}\]. All cases with mitotic figures were associated with poor outcomes (specificity of mitoses = 100%). Twenty-two percent of cases without mitotic figures had poor outcomes. Thus, lack of mitoses could not be used to exclude a poor outcome. Cases without mitoses could be further classified on the basis of irregular nuclear membranes and 3-fold variation in the nuclear size. There were no poor outcomes in cases, which lacked mitoses, irregular nuclear membranes, and a 3-fold or greater variation in nuclear size. Thus, the absence of these three markers is specific for nonpoor outcome. We developed two different scoring systems based on odds ratios \[[Table 6](#T6){ref-type="table"}\] and recursive partitioning \[[Table 7](#T7){ref-type="table"} and [Figure 1](#F1){ref-type="fig"}\]. For the odds ratio method, we use the relative odds ratio of three factors: {#F1} ###### Scoring based on odds ratios  ###### Scoring based on recursive portioning  Score = 4\*l (mitoses) + 4\*l (irregular nuclear membranes) + l (3-fold variation in nuclear size) where l (X) is the indicator function and l (X) = 1 if feature X is present and l (X) = 0 if the feature X is absent \[[Table 5](#T5){ref-type="table"}\]. For the partitioning method, scores for the five categories are shown in [Figure 1](#F1){ref-type="fig"} \[[Table 7](#T7){ref-type="table"}\]. ROC analysis showed that there was no statistically significant difference between the two scoring systems. The area under the ROC curve was 0.90 for both methods. Both methods were able to achieve an accuracy rate of 84%. An ROC curve for the score based on odds ratios is presented in [Figure 2](#F2){ref-type="fig"}. Using the regression tree classification shown in [Figure 1](#F1){ref-type="fig"}, cytologic features stratified malignancy risk between 0% and 100% \[[Table 1](#T1){ref-type="table"}\]. Using a risk of malignancy of 60% for assigning a specimen to the poor risk category, cytology correctly assigned specimens to the poor outcome category in 9 of 12 cases (75%) and when high-grade lesions were excluded, correct assignment to the poor outcome category occurred in 5 of 8 cases (62.5%). The scoring system correctly assigned a case to the favorable category (risk 20% or below) in 20 of 24 cases (83%). Histologic grading using the WHO system correctly predicted aggressive behavior in 14 of 13 cases (37.7%) and when high-grade lesions were excluded histologic grading failed to predict aggressive behavior in any case. Figures [3](#F3){ref-type="fig"}--[5](#F5){ref-type="fig"} illustrate Grade 1 (risk of poor outcome 20% or less), Grade 2 (risk of poor outcome \>20 and ≤60%), and Grade 3 (risk of poor outcome 100%). {#F2} {#F3} {#F4} {#F5} DISCUSSION {#sec1-4} ========== In 2010, the WHO published a classification for pancreatic endocrine neoplasms addressing both staging and grading issues.\[[@ref1]\] They proposed a three-tiered categorization system based on mitotic rate and PI as determined by Ki-67 staining.\[[@ref1]\] This classification system has been shown to have clinical utility.\[[@ref9]\] The WHO grading scheme is based on histopathologic material, and specific guidance was not given for the classification\'s application to cytologic material. The PSC developed guidelines for pancreatic and biliary tract cytology which include a categorization scheme with diagnostic criteria.\[[@ref4]\] In the PSC guidelines, pancreatic endocrine tumors are placed into one of two categories depending on cytomorphologic appearance. Grade 1 and 2 neoplasms of the WHO recommendations are placed in the "neoplasm, other" category and further categorized as pancreatic endocrine neoplasms. WHO high-grade lesions are placed in the malignant category with further categorization as pancreatic neuroendocrine carcinoma, either small or large cell variants.\[[@ref4]\] The WHO grading system for pancreatic endocrine neoplasms is most appropriately performed on resection specimens and not small biopsies as the small biopsy may not be representative of the final resection specimen grade. This limitation on histologic grading may be a reason for the superior predictive value of the cytologic grade over the histologic grade obtained from small biopsy specimens in the present study. Assignment of WHO low- and intermediate-grade pancreatic endocrine tumors to the "neoplasm, other" category of the PSC guidelines is based on the desire to maximize therapeutic options for treating clinicians and their patients. Thus, patients with histologically low- and intermediate-grade pancreatic neuroendocrine tumors could be followed if the patient\'s clinical status made operative intervention undesirable. These lower grade neoplasms were not assigned to the malignant category because it was believed such an assignment would make it difficult to clinically follow patients with low- and intermediate-grade pancreatic endocrine neoplasms if such neoplasms were designated cytologically malignant. The PSC guidelines made no attempt to distinguish between low- and intermediate-grade pancreatic endocrine tumors of the WHO classification and assigned both to a single category. A number of authors have recommended performing Ki-67 immunohistochemical staining on formalin fixed paraffin-embedded cell block preparations to prognostically stratify pancreatic endocrine neoplasms and have shown good correlation between such cytologic stratification and subsequent histologic grading of surgical resections.\[[@ref5][@ref6][@ref7][@ref8]\] This approach does require cell block preparations which may not always be obtained during EUS-FNA sampling. In some cases, only smear preparations are available for analysis and little data exists as to the possibility of and the methodology for stratifying pancreatic endocrine tumors in smear preparations. We analyzed a series of 36 pancreatic endocrine neoplasms of which 32 were designated as "neoplasm, other" (pancreatic endocrine tumor) and four were diagnosed as malignant: Pancreatic endocrine carcinoma. We rated the Diff-Quik stained smear preparations for the presence or absence of 20 morphologic features and correlated these features with poor outcome (development of metastatic disease, death or high-grade history morphology on resection). While several of the features analyzed are nuclear features and might be better evaluated on Papanicolaou-stained preparations, the staining preparation generally used for rapid on-site assessment is air-dried Diff-Quik stained material. This preparation appears suitable for assessment of the features utilized in this study. Using these correlations, we developed two different scoring schemes based on odds ratios and recursive partitioning. The statistical analyses revealed that mitotic count is highly specific for poor outcome. Poor outcomes were unlikely (\<20%) if mitoses were not present and the cells demonstrated no nuclear membrane irregularity or 3-fold variation in nuclear size. Cases that lacked mitotic figures on smear preparations but demonstrated nuclear membrane irregularity and a 3-fold variation in nuclear size had a 60% chance of a poor outcome. Morphologic features in our study were highly correlated \[[Table 3](#T3){ref-type="table"}\], so that it is difficult to build a predictive model. Once a feature is selected, correlating other features adds relatively little information to the model. The scoring system developed by the odds ratio method utilized three morphologic factors as shown below: Score = 4\*l (mitoses) + 4\*l (irregular nuclear membranes) + l (3-fold variation in nuclear size) where l (X) is the indicator function and l (X) = 1 if the feature is present and 0 if the feature is absent \[[Table 6](#T6){ref-type="table"}\]. For the partitioning method, scores were based on the five categories shown in [Figure 1](#F1){ref-type="fig"} \[[Table 7](#T7){ref-type="table"}\]. Using recursive partitioning analysis, mitotic figures were shown to have the highest discriminatory power \[[Figure 1](#F1){ref-type="fig"}\]. All cases which had mitotic figures were associated with a poor outcome and specificity for mitoses was 100%. Twenty-two percent of cases without mitotic activity also had poor outcomes. Lack of mitoses could not be used to exclude a poor outcome. Cases lacking mitoses were further classified on the basis of irregular nuclear membranes and 3-fold variation in nuclear size. When mitotic figures, irregular nuclear membranes and 3-fold variation in nuclear size were absent, no poor outcomes were present. The evaluation of specimens for these three markers is a sensitive measure for poor outcome. ROC analysis demonstrated that there was no statistically significant difference between the two scoring systems. CONCLUSION {#sec1-5} ========== The present study demonstrated that a simple scoring system utilizing the presence of mitotic figures, irregular nuclear membranes, and 3-fold or greater variation in nuclear size can subdivide pancreatic endocrine neoplasms into those with poor and those with a more favorable outcome. Given the prominence of mitotic rate and PI (Ki-67) in the WHO classification, it is not surprising that the presence of any mitotic figures is a strong predictor of poor outcome in smear preparations. The proposed scoring system is applicable to Diff-Quik stained smear preparations and is useful for evaluating both of the PSC categories "neoplasm, other (pancreatic endocrine neoplasm)" and malignant (pancreatic endocrine carcinoma). The scoring system is able to achieve an accuracy rate of 84%. The cytologic system had superior predictive value for poor outcome than did the WHO grading system which had an accuracy of classification of 69% (24 of 36 cases). The cytologic system is most useful when cell block preparations are not available for analysis. From the previously published data, cell block immunohistochemical staining for Ki-67 appears to represent the optimal technique to evaluate grade in cytologic specimens obtained from pancreatic endocrine neoplasms. Because our proposed cytologic scoring system yielded higher accuracy than histologic evaluation of small biopsy specimens, it might be considered as superior to evaluation by the WHO system for small samples. However, several studies have demonstrated the high reliability of cell blocks for grading, so we recommend their use when such material is available. When only smears are available for analysis, our scoring system appears to be a helpful alternative. COMPETING INTERESTS STATEMENT BY ALL AUTHORS {#sec1-6} ============================================ The authors declare that they do not have competing interests. AUTHORSHIP STATEMENT BY ALL AUTHORS {#sec1-7} =================================== Lester J. Layfield, M.D. developed the study concept and design, evaluated the specimens for data collection and wrote the draft article. Robert L. Schmidt M.D., Ph.D. performed the data analysis and helped prepare the draft article. Jack Campbell, BA collected the cases for analysis and prepared the data for analysis. Magda Esebua, M.D. analyzed the cases for data collection. All authors acknowledge reading and approving the final draft version. ETHICS STATEMENT BY ALL AUTHORS {#sec1-8} =============================== The present study was reviewed by the Institutional Review Board and determined to be exempt. The study follows the tenets of the Helsinki agreement. LIST OF ABBREVIATIONS (In alphabetic order) {#sec1-9} =========================================== EUS - Endoscopic Ultrasound FNA - Fine Needle Aspiration PI - Proliferation Index PSC - Papanicolaou Society of Cytopathology WHO - World Health Organization EDITORIAL/PEER-REVIEW STATEMENT {#sec1-10} =============================== To ensure the integrity and highest quality of CytoJournal publications, the review process of this manuscript was conducted under a **double-blind model** (authors are blinded for reviewers and vice versa) through automatic online system.

Introduction {#s1} ============ The Collier/Olfactory-1/Early B-cell factor (COE) family of transcription factors is necessary for animal development. COE proteins possess an atypical HLH domain and a unique zinc finger DNA binding domain conserved across metazoans [@pgen.1004746-Daburon1]. Invertebrates encode a single homolog of COE, with roles in mesoderm and ectoderm development [@pgen.1004746-Demilly1], [@pgen.1004746-Jackson1], whereas vertebrates have four COE paralogs with functions in diverse cell types including B-cells and adipocytes [@pgen.1004746-Liberg1]. In the central nervous system (CNS), COE regulates neuronal differentiation, migration, axon guidance, and dendritogenesis during development [@pgen.1004746-Demilly1], [@pgen.1004746-Jackson1], [@pgen.1004746-Crozatier1]--[@pgen.1004746-Hattori1] and maintains neuronal identity throughout adulthood [@pgen.1004746-Kratsios1], [@pgen.1004746-Eade1]. COE proteins have also been proposed to function as tumor suppressors [@pgen.1004746-Zhao1] and are associated with cancers such as acute lymphoblastic leukemia and glioblastoma [@pgen.1004746-Liao1]--[@pgen.1004746-Zardo1]. However, the specific genetic programs regulated by these genes in adult stem cells and mature neurons remain poorly understood. Stem cells can be studied to determine how transcriptional regulators orchestrate developmental processes or cause disease [@pgen.1004746-Zhu1]. An excellent animal model to investigate stem cell regulation *in vivo* is the freshwater planarian *Schmidtea mediterranea* [@pgen.1004746-Elliott1]. *S. mediterranea* has the ability to regenerate all tissue types from a population of adult stem cells (called neoblasts). These cells constitute approximately 10--20% of all the cells in the animal and include pluripotent [@pgen.1004746-Wagner1] and lineage-committed neoblasts [@pgen.1004746-Cowles1]--[@pgen.1004746-Scimone1]. The planarian CNS is composed of two cephalic ganglia and a pair of ventral nerve cords that run along the length of the animal, which are comprised of molecularly diverse neuronal subtypes that are regenerated within days after injury or amputation [@pgen.1004746-Collins1]--[@pgen.1004746-Umesono1]. Functional analysis of transcription factors in planarians using RNA interference (RNAi) has begun to identify regulatory molecules required for the generation and maintenance of specific neuronal subpopulations in the CNS such as serotonergic and cholinergic neurons [@pgen.1004746-Cowles1]--[@pgen.1004746-Mrz1], [@pgen.1004746-Lapan2]--[@pgen.1004746-Sandmann1]. Thus, planarians are outstanding organisms to study basic mechanisms that underlie stem cell-based maintenance and regeneration of the adult CNS. A previous functional screen for transcription factors encoding a helix-loop-helix domain identified a planarian *coe* homolog that is expressed in a small population of neural-committed stem cells (approximately 4--7% of the neoblast pool) and in neurons [@pgen.1004746-Cowles1]. We showed that animals fed dsRNA designed to silence *coe* expression (*coe(RNAi)* animals) regenerated abnormal brains; furthermore, uninjured *coe(RNAi)* planarians displayed behavioral defects and reduced expression of neural subtype-specific genes [@pgen.1004746-Cowles1]. In this study, we sought to identify genes regulated by *coe* with roles in CNS renewal by comparing the transcriptome profiles of uninjured control and *coe(RNAi)* animals, uncovering differentially expressed genes with predicted roles in CNS function. We validated a subset of these genes by testing for loss of expression after *coe* knockdown and visualizing their expression in *coe^+^* cells. These analyses revealed a set of nine candidate targets of *coe* in adult neurons, many of which are important for neuronal subtype identity (e.g., ion channels, neuropeptides, and neurotransmitters). In addition, our findings demonstrate that *coe* functions to drive gene expression in multiple neuronal classes, including excitatory and inhibitory neurons. To gain insights into the roles candidate COE targets play in CNS turnover and repair, we analyzed the function of downregulated transcripts using RNAi. Our functional screen identified several genes required for CNS regeneration, including homologs of a voltage-gated sodium channel α-subunit (*scna-2*) and the transcription factor *pou4l-1*. Our results suggest that COE is required for the expression of neural-specific genes in differentiating and mature neurons, a function that is essential to maintain CNS architecture and regulate neuronal regeneration. Results/Discussion {#s2} ================== *coe* is required for maintenance of nervous system structure {#s2a} ------------------------------------------------------------- Using an optimized whole-mount *in situ* hybridization protocol (WISH) (see [Materials and Methods](#s3){ref-type="sec"}), we found that *coe* mRNA was primarily restricted to neurons in *S. mediterranea* ([Fig. 1A](#pgen-1004746-g001){ref-type="fig"}). In agreement with our previous findings [@pgen.1004746-Cowles1], we also observed *coe* transcripts in a subset of cycling stem cells (*h2b* ^+^) ([Fig. 1B--C](#pgen-1004746-g001){ref-type="fig"}). We previously reported that *coe(RNAi)* animals regenerate cephalic ganglia that fail to connect at the anterior commissure and have significantly smaller brains with fewer *cpp-1^+^*, *npp-4^+^*, and *npy-2^+^* neurons when compared to the controls [@pgen.1004746-Cowles1]. This defect is not restricted to the anterior portion of the animal. Additional experiments showed *coe(RNAi)* animals do not properly regenerate their ventral nerve cords ([Fig. S1A--B](#pgen.1004746.s001){ref-type="supplementary-material"}). Moreover, analysis of the brain patterning defect using anti-VC-1, a marker of the photoreceptor neurons and their axons, revealed that the optic chiasm failed to connect at the midline in *coe(RNAi)* animals ([Fig. S1C](#pgen.1004746.s001){ref-type="supplementary-material"}). These data demonstrate that *coe* is essential for neuronal regeneration at both anterior and posterior facing wounds and that *coe* regulates genes required for reestablishing midline patterning following brain amputation. {#pgen-1004746-g001} In addition, we previously noted that silencing of *coe* in intact uninjured animals results in a reduction of *ChAT^+^* and *pc2^+^* neurons near the anterior commissure and a loss of *cpp-1^+^* neurons. Following the 6^th^ feeding of *coe* dsRNA, 100% of the animals exhibited impaired negative phototaxis [@pgen.1004746-Cowles1]. To investigate the specificity of the *coe* knockdown phenotype on the CNS, we examined the effect of *coe* RNAi on the intestine and muscle as representative endodermal or mesodermal tissues, respectively. We hybridized uninjured control and *coe(RNAi)* animals with riboprobes specific to *ChAT* (as a positive control), *mat* [@pgen.1004746-Wenemoser2], and *collagen* [@pgen.1004746-Witchley1]. As expected, we observed a decrease in *ChAT* ^+^ neurons in the head [@pgen.1004746-Cowles1] and noted a decrease in *ChAT* expression throughout the animal ([Fig. 2A](#pgen-1004746-g002){ref-type="fig"}); by contrast, we did not observe a change in the spatial distribution of *mat* or *collagen* following *coe* knockdown ([Fig. 2B--C](#pgen-1004746-g002){ref-type="fig"}). To quantify the effect of *coe* RNAi treatments on the expression of *ChAT*, *mat* and *collagen*, we measured relative mRNA levels by reverse transcription quantitative PCR (RT-qPCR). First, we confirmed *coe* knockdown led to a significant decrease in the relative expression of *coe* mRNA (down 60%±16% compared to the controls; [Fig. 2D](#pgen-1004746-g002){ref-type="fig"}). Measurement of *ChAT*, *mat* and *collagen* from *coe(RNAi)* planarians revealed that *ChAT* mRNA levels were significantly down (45%±15%) compared to control animals; in contrast to *ChAT*, the relative mRNA levels of *mat* or *collagen* were not affected by *coe* RNAi treatment ([Fig. 2D](#pgen-1004746-g002){ref-type="fig"}). Combined with our previous work [@pgen.1004746-Cowles1], these results strongly suggest that *coe* knockdown specifically affects gene transcription in the nervous system and does not cause obvious defects in other tissues such as the intestine or muscle. Furthermore, our results are consistent with reports demonstrating that COE is required to maintain cholinergic and peptidergic neuronal subtype-specific gene expression in *Caenorhabditis elegans* and *Drosophila melanogaster* [@pgen.1004746-Kratsios1], [@pgen.1004746-Eade1]. {#pgen-1004746-g002} To investigate if the inhibition of *coe* perturbs nervous system architecture downstream of gene expression changes, we labeled neuronal cell bodies and their projections using anti-CRMP-2, which labels a subset of neuronal cell bodies and their axon projections, and anti-β-tubulin to visualize nerve projections ([Fig. 3A--C](#pgen-1004746-g003){ref-type="fig"}). In *coe(RNAi)* animals, we observed a striking decrease in axon projections labeled by anti-CRMP-2 and anti-β-tubulin compared to the controls; however, expression of CRMP-2 was retained in the cell bodies ([Fig. 3C](#pgen-1004746-g003){ref-type="fig"}). In addition, when we labeled sensory neurons using *cintillo* [@pgen.1004746-Oviedo1], *coe(RNAi)* animals exhibited significantly fewer *cintillo^+^* cells ([Fig. 3D](#pgen-1004746-g003){ref-type="fig"}). Our results strongly suggest that nervous system architecture is severely reduced or lost in the absence of *coe*. These structural defects likely underlie the behavioral abnormalities observed in *coe-*deficient planarians. {#pgen-1004746-g003} Identification of genes regulated by *coe* in the planarian nervous system {#s2b} -------------------------------------------------------------------------- Although COE has been shown to drive differentiation of several classes of neurons during development [@pgen.1004746-Green1], the transcriptional programs controlled by this transcription factor in adult nervous system function are poorly defined. We reasoned that the CNS-specific *coe* RNAi phenotype in intact planarians represents an excellent opportunity to identify gene expression programs controlled by COE in the post-embryonic nervous system. Thus, we used comparative mRNA sequencing (RNA-seq; see [Materials and Methods](#s3){ref-type="sec"}) to sequence mRNAs isolated from uninjured controls and *coe(RNAi)* animals one week after the 6^th^ RNAi treatment, which was the point in time we consistently observed behavioral defects and loss of neural-specific gene expression in 100% of *coe-*deficient animals and did not detect overt defects in other tissues ([Fig. 2](#pgen-1004746-g002){ref-type="fig"}). RNA-seq analysis identified 909 differentially expressed genes; 397 were downregulated, and 512 were upregulated ([Table S1](#pgen.1004746.s006){ref-type="supplementary-material"}). Functional annotation using DAVID software showed that the set of downregulated genes was significantly enriched for Gene Ontology (GO) terms associated with "ion channel," "neuronal activities," "nerve-nerve synaptic transmission," "voltage-gated ion channel," and "cell adhesion molecule"; by contrast, the upregulated genes were enriched for GO terms associated with "cytoskeletal protein" and "muscle development" ([Table 1](#pgen-1004746-t001){ref-type="table"}). *coe* mRNAs were not detected in a muscle pattern ([Fig. 1](#pgen-1004746-g001){ref-type="fig"}), nor did we detect overt phenotypes associated with muscle differentiation ([Fig. 2](#pgen-1004746-g002){ref-type="fig"}). However, the RNA-seq data raised the possibility that *coe* might negatively regulate mesoderm specification, which is required for muscle development [@pgen.1004746-Jackson1], [@pgen.1004746-Crozatier2]. It is possible upregulation of muscle genes is an indirect consequence of a loss of nervous system influence such as cholinergic transmission and/or neuropeptide regulation. Previous studies have demonstrated cholinergic neurotransmission is required for coordinated muscle contractions in planarians [@pgen.1004746-Carolei1]--[@pgen.1004746-Nishimura1]. Thus, we speculate that loss of nervous system modulation disrupts muscle homeostasis and leads to changes in expression of muscle-related genes. Although our experiments do not definitively assign the role of COE in muscle differentiation or maintenance, our data do clearly indicate that *coe* is required for expression of nervous system-specific genes in adult planarians. 10.1371/journal.pgen.1004746.t001 ###### Annotation of genes differentially expressed in *coe(RNAi)* animals using DAVID software. {#pgen-1004746-t001-1} Functional Cluster Enrichment Score Differential Expression ----------------------------------- ------------------ ------------------------- Ion channel 7.65 Downregulated Neuronal activities 6.50 Downregulated Voltage-gated ion channel 2.38 Downregulated Microtubule binding motor protein 2.31 Downregulated Nerve-nerve synaptic transmission 1.71 Downregulated Cell adhesion molecule 1.30 Downregulated Neurogenesis 1.27 Downregulated Muscle contraction 4.13 Upregulated Cytoskeletal protein 3.98 Upregulated Mitosis 2.25 Upregulated Based on the annotation of differentially expressed genes, we hypothesized that genes predicted to play roles in nervous system functions in the downregulated category likely include direct COE targets. To test our hypothesis and validate genes found in our RNA-seq dataset, we selected 65 genes that were dramatically downregulated, associated with neural functions, or annotated as transcription factor homologs. First, we performed WISH to determine the tissue-specific pattern of expression of all 65 genes (representative examples are shown in [Fig. 4](#pgen-1004746-g004){ref-type="fig"}). As we expected, the most prominent mRNA expression pattern was in the nervous system (26 of 65 genes; see [Table S2](#pgen.1004746.s007){ref-type="supplementary-material"}), similar to *ChAT* and *cpp-1*, which we had previously found to be putative downstream targets of COE [@pgen.1004746-Cowles1]. In addition, we observed genes that were expressed broadly in the nervous system (such as *neural cell adhesion molecule-2* (*ncam-2*), *vesicle-associated membrane protein like-1* (*vamp*), *gamma-aminobutyric acid receptor subunit beta like-1* (*gbrb-1*), and *voltage-gated sodium channel alpha-1* (*scna-1*)) or in discrete neuronal subpopulations (such as *secreted peptide prohormone-19*, *-18*, *-2* (*spp-19*, *-18*, *-2*), *neuropeptide like* (*npl*), *voltage-gated sodium channel alpha-2* (*scna-2*), and *caveolin-1* (*cav-1*)) ([Fig. 4A--J](#pgen-1004746-g004){ref-type="fig"}). Our list also included transcripts that labeled subsets of neurons in the brain (such as *netrin-1*) ([Fig. 4K](#pgen-1004746-g004){ref-type="fig"}) [@pgen.1004746-Cebri1]. In addition, we found that the transcription factors *iroquios-1* (*irx-1*) and *pou class 4 transcription factor 4 like-1* (*pou4l-1*) were expressed at or near the cephalic ganglia ([Fig. 4L--M](#pgen-1004746-g004){ref-type="fig"}), and their mRNA was detected in *ChAT* ^+^ neurons by fluorescent *in situ* hybridization (FISH) ([Fig. S2](#pgen.1004746.s002){ref-type="supplementary-material"}). Next, we tested the effect of *coe* RNAi on the expression of 33 genes that could be visualized in discrete cell populations by WISH. Knockdown of *coe* led to a marked reduction in the expression of 31 genes ([Table S2](#pgen.1004746.s007){ref-type="supplementary-material"}; representative results are shown in [Fig. 4A′--H′, K′--M′](#pgen-1004746-g004){ref-type="fig"}); for two genes, *scna-2* and *cav-1*, we observed a loss of expression at the midline ([Fig. 4I′--J′](#pgen-1004746-g004){ref-type="fig"}). Furthermore, we quantified the number of cells labeled by *spp-19*, *spp-18*, and *npl* probes. As expected, we found there was a significant reduction in the number of *spp-19^+^*, *spp-18^+^*, and *npl^+^* cells following *coe* RNAi ([Fig. 4N](#pgen-1004746-g004){ref-type="fig"}). {#pgen-1004746-g004} As an additional test to validate the *in situ* hybridization results, we measured the relative expression levels of downregulated genes in control and *coe* RNAi-treated planarians using RT-qPCR ([Fig. S3A](#pgen.1004746.s003){ref-type="supplementary-material"}). All of the genes we tested showed a decrease in relative expression following *coe* RNAi (9 of 14 genes were significantly downregulated; P\<0.05, Student\'s t-test). By contrast, when we measured the relative expression of CNS-expressed genes that were not on our list of differentially expressed genes, none were significantly reduced (11 of 11 genes; [Fig. S3B--C](#pgen.1004746.s003){ref-type="supplementary-material"}). Although some of the control genes we selected were reduced near levels comparable to some genes downregulated following *coe* RNAi (e.g., *ncam2*, *vamp*, and *gbrb1*; [Fig. S3A](#pgen.1004746.s003){ref-type="supplementary-material"}), we noted that *isotig13897* and *npp-2* [@pgen.1004746-Collins1], which are transcripts detected in subsets of neurons or throughout the CNS, respectively, remained unchanged ([Fig. S3B--C](#pgen.1004746.s003){ref-type="supplementary-material"}). It is possible that some changes in gene expression associated with *coe* RNAi are consequence of a reduction in nervous system tissue. We proceeded to perform double-FISH to *coe* and validated genes to determine if any were potential genetic targets of COE. Of the 17 genes we were able to reliably detect by FISH (33 genes were tested; see [Table S2](#pgen.1004746.s007){ref-type="supplementary-material"}), 11 were expressed in *coe^+^* cells (representative results are shown in [Fig. 5](#pgen-1004746-g005){ref-type="fig"} and [Fig. S4](#pgen.1004746.s004){ref-type="supplementary-material"}), including *ChAT* and *cpp-1* [@pgen.1004746-Cowles1]. Together, these results identified nine novel candidate targets of COE in the nervous system, including genes important for maintaining neuronal subtype identity such as ion channels, ion channel receptors, and neuropeptide genes ([Table 2](#pgen-1004746-t002){ref-type="table"}). In addition, our data suggest that COE is essential to maintain genetic programs in multiple classes of adult neuronal subtypes including excitatory (cholinergic) and inhibitory (GABAergic) neurons. {#pgen-1004746-g005} 10.1371/journal.pgen.1004746.t002 ###### Candidate COE targets genes identified in *S. mediterranea*. {#pgen-1004746-t002-2} Gene Name CNS function ------------------------------------------------------------------- --------------------------- *Smed-gamma-aminobutyric acid receptor subunit gamma like (gbrg)* Neurotransmitter receptor *Smed-netrin-1* Axon Guidance *Smed-neural cell adhesion molecule-2 (ncam-2)* Cell adhesion *Smed-neuropeptide like-1 (npl-1)* Novel gene; Unknown *Smed-secreted peptide prohormone 18 (spp-18)* Neuropeptide *Smed-secreted peptide prohormone 19 (spp-19)* Neuropeptide *Smed-secreted peptide prohormone-2 (spp-2)* Neuropeptide *Smed-vesicle-associated membrane protein like-1 (vamp)* Cell adhesion *Smed-voltage-gated sodium channel (scna-1)* Ion Channel Genes downstream of *coe* are required for proper CNS regeneration {#s2c} ------------------------------------------------------------------ Our RNA-seq dataset revealed that *coe* is essential to maintain the expression of hundreds of genes in the adult animal. This change in the neuronal gene expression landscape led to abnormal CNS structure and behavior. To identify genes downstream of *coe* that contribute to CNS differentiation, we took advantage of the experimental ease in examination of gene function in planarian regeneration and analyzed the role of 11 downregulated genes that were expressed in neurons or predicted to encode transcription factors ([Table 3](#pgen-1004746-t003){ref-type="table"}). Following RNAi, animals were amputated pre- and post-pharyngeally and allowed to regenerate for 10 days. We found that 6 out of 11 genes resulted in defective brain regeneration (see [Table 3](#pgen-1004746-t003){ref-type="table"}); *scna-2*, *pou4l-1*, and *nkx2l* caused the strongest phenotypes. Compared to the controls, *scna-2(RNAi)* animals had less eye pigmentation or developed a single eyespot; *nkx2l(RNAi)* animals exhibited photoreceptor defects; and *pou4l-1(RNAi)* animals had less photoreceptor pigment ([Fig. 6A--D](#pgen-1004746-g006){ref-type="fig"}). To examine CNS architecture, we stained *scna-2*, *nkx2l*, and *pou4l-1* RNAi treated planarians with anti-SYNAPSIN and the *coe-*regulated genes *ChAT* and *npl*. Although subtle, all three showed abnormalities in brain morphology ([Fig. 6A--D](#pgen-1004746-g006){ref-type="fig"}). However, when we measured the area of the brain stained by anti-SYNAPSIN, only *scna-2* and *pou4l-1* RNAi animals had a significant reduction in neuropil density ([Fig. 6E](#pgen-1004746-g006){ref-type="fig"}). Consistent with this observation, the *ChAT^+^* brain areas were smaller in *scna-2(RNAi)* and *pou4l-1(RNAi)* animals ([Fig. 6F](#pgen-1004746-g006){ref-type="fig"}) but not in *nkx2l(RNAi)* animals. The smaller brain phenotype was accompanied by fewer *npl^+^* neurons in *scna-2(RNAi)* animals; however, despite their smaller brains, *pou4l-1(RNAi)* animals regenerated significantly more *npl^+^* cells than controls ([Fig. 6G](#pgen-1004746-g006){ref-type="fig"}). These findings demonstrate that *scna-2* is required for CNS regeneration and highlight the importance of ion channels in neurogenesis regulation during CNS development, maintenance, and repair [@pgen.1004746-Tseng1]--[@pgen.1004746-Beane2]. Interestingly, these data suggest that *pou4l-1* plays a role in the specification of certain neuronal lineages. It is possible that in the absence of *pou4l-1*, planarians regenerate the incorrect proportion of neuronal subtypes and have disorganized brains, but this possibility will require further analysis with additional neuronal subtype-specific markers. By contrast, our results suggest *nkx2l* is not required for CNS regeneration *per se*. Following *coe* RNAi, *nkx2l* expression was reduced by *in situ* hybridization and RT-qPCR ([Table S2](#pgen.1004746.s007){ref-type="supplementary-material"} and [Fig. S3A](#pgen.1004746.s003){ref-type="supplementary-material"}), but *nkx2l*, which is primarily expressed in stem cells and in progeny [@pgen.1004746-Labb1], was not detected in the nervous system ([Fig. S5A](#pgen.1004746.s005){ref-type="supplementary-material"}). We hypothesize *nkx2l* functions in early regeneration to establish patterning, which is consistent with the observation that *nkx2l(RNAi)* planarians fail to regenerate properly patterned head ([Fig. 6C](#pgen-1004746-g006){ref-type="fig"}) and tail tissues ([Fig. S5B](#pgen.1004746.s005){ref-type="supplementary-material"}). {#pgen-1004746-g006} 10.1371/journal.pgen.1004746.t003 ###### Functional analysis of genes downregulated following *coe* RNAi. {#pgen-1004746-t003-3} Gene Name RNAi Phenotype DAVID Annotation ------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------ ----------------------------------------------------------- *Smed-gamma-aminobutyric acid receptor subunit beta like (gbrb1)* No phenotype observed (12/15) N-N synaptic transmission, Ion channel, Neural activities *Smed-gamma-aminobutyric acid receptor subunit gamma like (gbrg)* No external phenotype observed (34/35), reduced neuropil density at anterior commissure (6/13) N-N synaptic transmission, Ion channel, Neural activities *Smed-hemicentin-1* No phenotype observed (13/14) Cell Adhesion *Smed-iroquois-1 (irx-1)* No phenotype observed (10/10) Transcription Factor *Smed-nkx2 like-1 (nkx2l-1)* Photoreceptor defects (9/35), abnormal brain architecture (35/35), and reduced or indented tail blastemas (7/35) Transcription Factor *Smed-notch-1* No phenotype observed (10/10) Neurogenesis *Smed-neuropeptide like (npl)* Delayed photoreceptor regeneration (9/33) NA *Smed-pou class 4 transcription factor 3 like-1 (pou4l-1)* Lighter photoreceptors (5/40) and reduced neuropil density (14/20) Transcription Factor *Smed-voltage-gated sodium channel (scna-1)* No phenotype observed (15/15) Voltage-gated ion channel, Ion channel, Neural activities *Smed-voltage-gated sodium channel (scna-2)* Reduced photoreceptor formation (20/41) and neuropil density (10/10); regenerated a single photoreceptor (1/41) Voltage-gated ion channel, Ion channel, Neural activities *Smed-voltage-gated sodium channel (scna-3)* Delayed photoreceptor formation (6/28), reduced neuropil density at anterior commissure (6/14) Voltage-gated ion channel, Ion channel, Neural activities The number of animals showing the phenotype(s) among the total number examined from at least two independent experiments is indicated in parentheses. It is noteworthy that several transcription factors that we identified in our screen are putative COE targets in *Xenopus* development, including *irx-1*, *tal*, *pou4l-1*, and *nkx2l* [@pgen.1004746-Green1]. Of these genes, we found that expression of *pou4l-1* was important for CNS regeneration and *nkx2l* was involved in patterning. NKX and POU orthologs play critical roles during CNS development of invertebrate and vertebrate organisms [@pgen.1004746-Latchman1]--[@pgen.1004746-McMahon1]. These data suggest that regulatory genes downstream of COE are conserved and have roles in CNS regeneration. However, it will be important to experimentally resolve whether these transcription factors are *bona fide* targets of COE in planarians or other animals such as *Xenopus*. Concluding remarks {#s2d} ------------------ COE proteins are known to function as terminal selectors of neuronal identity in adult organisms [@pgen.1004746-Kratsios1], [@pgen.1004746-Eade1], [@pgen.1004746-Deneris1], yet the neuronal subtypes and specific genetic programs regulated by COE in the adult CNS are not well understood. In this study, we exploited the high rate of tissue turnover and regenerative capacity of planarians to expand our understanding of how COE may function in the post-embryonic nervous system. We combined RNAi with RNA-seq analysis and identified a set of differentially expressed genes associated with nervous system biological roles. Expression analysis of a subset of these genes revealed novel candidate targets of *coe* in planarian neurons ([Fig. 7A](#pgen-1004746-g007){ref-type="fig"}), some of which underscored *coe*\'s essential role in maintaining expression of genes vital for neuronal subtype identity and function (such as neurotransmitter receptors, ion channels, and neuropeptide encoding genes) ([Fig. 7A--B](#pgen-1004746-g007){ref-type="fig"}). Decoding which transcriptional changes are direct or indirect consequences of *coe* loss in the planarian model will be vital to further elucidate how mutations in COE proteins cause or contribute to disease pathologies in the CNS. The next step will be to find direct COE binding sites genome-wide using *in silico* and chromatin immunoprecipitation (ChIP) approaches and combining these findings with our differential expression data. In addition, molecular profiling of *coe^+^* cell populations (such as stem cells, postmitotic progeny, and neurons) will be essential to determine how *coe* function alters in cell type-specific contexts. In conclusion, our study demonstrates the importance of COE family proteins in neuronal turnover and repair of the adult CNS and broadens our understanding of the regulatory programs governed by these factors. ![COE function is required for differentiation and maintenance of diverse neuron types.\ (**A**) *coe* is expressed in lineage-committed neoblasts (*smedwi^+^*) and early progeny [@pgen.1004746-Cowles1], and diverse neuron types, including cholinergic (*ChAT*), GABAergic (*gad*), octopaminergic (*tbh*), dopaminergic (*th*), serotonergic (*tph*), and neuropeptidergic (*cpp-1*, *npl*, *spp-18*, *spp-19*, *spp-2*) neurons. Genes in green were identified in [@pgen.1004746-Cowles1]. (**B**) To gain insights into how loss of COE function contributes to defects in nervous system differentiation, we analyzed the function of genes that were downregulated in *coe(RNAi)* animals. These analyses identified additional genes required for CNS regeneration (*gbrb1*, *npl*, *scna-2*, *scna-3*, *pou4l-1*) and patterning (*nkx2l*). In *coe(RNAi)* animals, we also detected upregulated genes enriched for GO terms associated with muscle development ([Table 1](#pgen-1004746-t001){ref-type="table"}), suggesting that COE may also function to repress the expression of mesoderm-specific genes.](pgen.1004746.g007){#pgen-1004746-g007} Materials and Methods {#s3} ===================== Animal husbandry {#s3a} ---------------- Asexual *Schmidtea mediterranea* (CIW4) were reared in 1× Instant Ocean Salts (0.83 mM MgSO~4~, 0.9 mM CaCl~2~, 0.04 mM KHCO~3~, 0.9 mM NaHCO~3~, and 0.21 g/L Instant Ocean Aquarium Salt diluted in ultra-pure water) at 20°C. Animals were starved for one week, and those ranging between 2--5 mm in length were used for experimentation. RNA interference {#s3b} ---------------- Animals were administered six feedings of bacterially expressed dsRNA complementary to the indicated gene over three weeks as previously described [@pgen.1004746-Gurley1]; *gfp* dsRNA was fed as a control. Unless otherwise indicated, all intact RNAi animals were fixed seven days following the 6^th^ dsRNA treatment. For regeneration experiments, planarians were amputated pre- and post-pharyngeally 24 hours following the 6^th^ dsRNA feeding. Whole-mount *in situ* hybridizations and immunostainings {#s3c} -------------------------------------------------------- Animals were processed for colorimetric whole-mount *in situ* hybridization using the protocol described in [@pgen.1004746-King1]. Fluorescent *in situ* hybridization experiments were performed as described in [@pgen.1004746-Cowles1], [@pgen.1004746-King1] and developed using Tyramide Signal Amplification (TSA) as described in [@pgen.1004746-Lauter1]. Briefly, animals were incubated for 5 min. in borate buffer (100 mM borate pH 8.5, 0.1% Tween-20) and then developed in TSA Reaction Buffer (borate buffer, 2% dextran sulfate, 0.1% Tween-20, 0.003% H~2~O~2~), containing fluor-tyramide and 4-iodophenylboronic acid for 30 min. For double-FISH, animals were quenched in 1% H~2~O~2~ for 1 hour. For γ-irradiation experiments, animals were fixed 6 days following a 100 Gy treatment, a time point when both stem cells and postmitotic progenitors are ablated. Accession numbers for the sequences used in this study are listed in [Table S3](#pgen.1004746.s008){ref-type="supplementary-material"}. For immunostaining with anti-SYNORF1 (1∶400, 3C11, DSHB) or anti-VC-1 (1∶10,000; kindly provided by Hidefumi Orii), animals were fixed with Carnoy\'s solution [@pgen.1004746-Umesono2]. For anti-CRMP-2 (1∶50, 9393S, Cell Signaling) or anti-β-TUBULIN (1∶1000; E7, DSHB) labeling, animals were fixed with formaldehyde, processed without a reduction step, and labeled using TSA [@pgen.1004746-King1]. RNA sequencing and DAVID analysis {#s3d} --------------------------------- One week after the final dsRNA treatment, RNA was extracted from three independent control and *coe(RNAi)* animal groups using Trizol (Life Technologies). RNA samples were treated with DNase using the Turbo DNA-free Kit (Life Technologies) and purified using the RNeasy MinElute Cleanup kit (Qiagen). Sequencing libraries were synthesized using the TruSeq RNA Sample Prep Kit v2 and sequenced on a HiSeq 2000 System (Illumina). More than 12 million 100-bp single-end reads were generated for each sample. Sequenced reads were submitted to the Sequence Read Archive (NCBI) under the accession number PRJNA235907. Reads were mapped to the planarian genome using TopHat [@pgen.1004746-Trapnell1]; gene models were predicted using a published transcriptome [@pgen.1004746-Adamidi1], [@pgen.1004746-nal1]. Differentially expressed genes were identified using the R Bioconductor package edgeR [@pgen.1004746-Robinson1] with cutoffs of logCPM score ≥0 and FDR≤0.05. Changes in gene expression detected by RNA-seq were represented as linear fold changes over controls. For the differentially expressed *Schmidtea mediterranea* transcripts, we performed BLASTX against the human UniProt database (cutoff\<1×10^−4^); human accession numbers were then used to assign Gene Ontology terms and perform clustering analysis using DAVID software [@pgen.1004746-Dennis1], [@pgen.1004746-Huangda1] with the "Panther_BP_all" and "Panther_MF_all" gene annotation settings and an Enrichment Score cutoff \>1.3. Gene identification and cloning {#s3e} ------------------------------- For validation studies, transcript sequences were analyzed by BLASTX against protein sequences from human, mouse, fly, and nematode and identified as the top BLAST hit ([Table S3](#pgen.1004746.s008){ref-type="supplementary-material"}). Sequences were obtained from a cDNA collection [@pgen.1004746-Zayas1] or cloned into pJC53.2 [@pgen.1004746-Collins1] or pPR244 [@pgen.1004746-Reddien1] using gene specific primers. GenBank accession numbers and the primers used in this study are listed in [Table S3](#pgen.1004746.s008){ref-type="supplementary-material"}. Reverse transcription quantitative PCR {#s3f} -------------------------------------- Total RNA was extracted and purified as described above. cDNA was synthesized using the iScript cDNA Synthesis Kit (BioRad). Reverse transcription quantitative PCR was performed on a Bio-Rad CFX Connect Real-Time System using SsoAdvanced SYBR Green Supermix (Bio-Rad) with a two-step cycling protocol and annealing/extension temperature of 58.5°C. At least three biological replicates and two technical replicates were performed for each experiment. The relative amount of each cDNA target was normalized to *Smed-β-tubulin* (accession no. DN305397). The normalized relative changes in gene expression, standard deviations, and t-tests were calculated in Bio-Rad CFX Manager Software v3.0. Primers are listed in [Table S3](#pgen.1004746.s008){ref-type="supplementary-material"}. Imaging, cell counts, and statistical analysis {#s3g} ---------------------------------------------- Images of live animals and whole mount *in situ* hybridization samples were acquired using a Leica DFC450 camera mounted on a Leica M205 stereomicroscope. Fluorescent images were acquired with a Zeiss Axio Observer.Z1 equipped with an Axiocam MRm camera and ApoTome; images are displayed as maximum image projections from ten 1-µm optical sections. For all experiments, we counted cells by hand using ImageJ Software [@pgen.1004746-Schneider1], and biological replicates (n≥3) were averaged and shown as mean ± standard deviation. The number of *cintillo* ^+^, *spp-19^+^*, *spp-18^+^*, and *npl^+^* cells ([Fig. 4N](#pgen-1004746-g004){ref-type="fig"}) was normalized to animal length (mm). We used anti-SYNAPSIN staining and *ChAT* expression to determine brain area ([Fig. 6E--F](#pgen-1004746-g006){ref-type="fig"}), normalized to animal length (µm). To quantify *npl^+^* brain-specific neurons following amputation, *npl^+^* cells were counted in the cephalic ganglia and normalized to the average total brain area ([Fig. 6G](#pgen-1004746-g006){ref-type="fig"}). When comparing two groups, we used a Student\'s t-test and significance was accepted at P\<0.05. Supporting Information {#s4} ====================== ###### *coe* is required for proper regeneration of the planarian nervous system. (**A--C**) Control and *coe(RNAi)* animals were amputated pre- and post-pharyngeally, allowed to regenerate for seven days, and the CNS morphology was analyzed in regenerating trunk fragments immunostained with anti-SYNAPSIN or anti-VC-1. Arrows in A and C denote defects in anterior commissure and photoreceptor axon patterning, respectively; arrowheads in B mark reduced anti-SYNAPSIN staining in the ventral nerve cords at the tail region (N = 10). Anterior is up. Scale bar in A = 200 µm; C = 100 µm. (TIF) ###### Click here for additional data file. ###### The transcription factors *irx-1* and *pou4l-1* are detected in brain *ChAT^+^* neurons. Double-fluorescent *in situ* hybridization to *ChAT* and *irx-1* or *pou4l-1* (N≥3 animals). Anterior is up. Scale bar = 200 µm. (TIF) ###### Click here for additional data file. ###### Reverse transcription quantitative PCR validation of downregulated nervous system genes in *coe(RNAi)* planarians. (**A**) RT-qPCR measuring the relative expression of selected genes following *coe* RNAi treatment. (**B**) Whole-mount *in situ* hybridization to genes that are expressed in the nervous system of planarians and did not significantly change expression levels after *coe* RNAi. The genes shown (indicated above each panel) were selected from an *in situ* hybridization screen (unpublished). The expression pattern of the neuropeptide genes *grh-1*, *ilp*, *mpl-1*, *npp-2*, *spp-15* and *spp-16* (see C) were reported in [@pgen.1004746-Collins1]. Anterior is up. (**C**) RT-qPCR measurements of gene expression for nervous system-specific genes in control and *coe* RNAi planarians. All graphs show the mean ± s.d. expression level relative to the controls; \*P\<0.05, Student\'s t-test. Scale bar in B = 200 µm. (TIF) ###### Click here for additional data file. ###### Identification of genes expressed in *coe^+^* neurons. Additional data for [Figure 5](#pgen-1004746-g005){ref-type="fig"}. Double-FISH to *coe* and *vamp*, *gbrg*, or *scna-1* (N≥3 animals). White arrowheads mark cells co-labeled with *coe*. Anterior is up. Scale bar = 200 µm. (TIF) ###### Click here for additional data file. ###### *nkx2l* is required for tissue regeneration. Additional data for [Figure 6](#pgen-1004746-g006){ref-type="fig"}. (**A**) *In situ* hybridization to *nkx2l-1*. (**B**) After 10 days of regeneration, the tail region of *control* and *nkx2l* RNAi animals were imaged live or immunostained with anti-SYNAPSIN. Anterior is up. Scale bars = 200 µm. (TIF) ###### Click here for additional data file. ###### List of differentially expressed genes following *coe* RNAi. (XLSX) ###### Click here for additional data file. ###### Expression analysis of downregulated genes following *coe* RNAi. (XLSX) ###### Click here for additional data file. ###### Accession numbers, primers, and top BLAST hits for the genes analyzed in this study. (XLSX) ###### Click here for additional data file. We would like to thank Jordana Henderson and Kelly Ross for helpful comments on this manuscript; Catherine Adamidi and Nikolaus Rajewsky for providing the gene annotation file used for our RNA sequencing analysis; Steve Head and Lana Schaffer at The Scripps Research Institute DNA Sequencing Core for help designing and analyzing our RNA-sequencing experiments; and Claire Cowles for artwork design. The SYNORF-1 (3C11; developed by E. Buchner) and β-TUBULIN (E7; developed by M. Klymkowsky) monoclonal antibodies were obtained from the Developmental Studies Hybridoma Bank, created by the NICHD and maintained at The University of Iowa, Department of Biology, Iowa City, IA 52242. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: MWC RMZ. Performed the experiments: MWC KCO BNS CGQ. Analyzed the data: MWC KCO CGQ RMZ. Wrote the paper: MWC RMZ.

1. Introduction {#sec1} =============== The objectives of the present investigation are to determine the shape, height, and location of the lingula in relation to the mandibular ramal landmarks and the mandibular first molar in a Turkish pediatric population of 6- to 12-year-old children using cone-beam computed tomography (CBCT). The differences in the parameters were also evaluated between the sexes. Comparisons with previous studies of various ethnic and racial groups were also assessed. The lingula is a tongue-shaped bony projection on the medical surface of the mandibular ramus close to the posterior margin of the mandibular foramen \[[@B1]\]. The exact location of the mandibular foramen on radiographs is not always easy to be established due to its radiolucency and the superimposition of contralateral mandibular structures \[[@B2]\]. The mandibular foramen has often considered to be the most reliable reference point for approaching the inferior alveolar nerve in several anesthesia techniques, which led to several studies about its position and its anatomical relationships to clinically recognizable landmarks. It has been speculated that the mandibular lingula and foramen change the ratio of their positions on the ramus of growing children \[[@B3]\]. Tsai \[[@B4]\] observed in children a variation in the difference between the distance from the mandibular foramen to the anterior border and to the posterior border. This variation is caused by regional growth in different directions in each of Hellman\'s dental developmental stages \[[@B3]\]. Because of its connection to nerve and vascular structures the study of the lingula features provides significant information related to oral and maxillofacial surgical procedures, such as the sagittal split ramus osteotomy, vertical ramus osteotomy, inverted L osteotomy \[[@B2]\], orthognathic surgery, mandibular trauma management, eradication of benign and malignant lesions, preprosthetic surgery, and nerve injury during inferior mandibular nerve block \[[@B3]\]. The lingula is used for identifying the site for injection of local anaesthetics or for excision of nerve for facial neuralgia \[[@B4]\]. If oral-maxillofacial surgeons are unable to identify the lingula correctly, intraoperative complications such as hemorrhage, unfavorable fracture nerve injury and may occur \[[@B5a], [@B5b]\]. The present study is unique in that the mandibles were collected from a single ethnic group (Turkish) with known ages and gender. The number of samples in the present study was large enough to compare the shape of the lingula and its distribution in addition to determining the location of the lingula. The subjects (both girls and boys) have a known and similar age range and the results were subjected to statistical analyses. It was therefore possible to investigate sex and side differences. The researchers analyzed the morphological characteristics of the lingula, and they stated at the conclusion that such structural variability could account for failure to block the inferior alveolar nerve \[[@B6], [@B7]\]. Variations in the shape of the lingula have been reported by various authors \[[@B8]--[@B10]\]; for example, Tuli et al. \[[@B5a], [@B5b]\] classified lingula into four different types based on its shape, namely, triangular, truncated, nodular, and assimilated types. 2. Material and Methods {#sec2} ======================= The study protocol was approved by the Ethics Board of the Medical Faculty of Erciyes University. We designed a retrospective study composed of CBCT images of 2,103 patients who are presented to the Dentomaxillofacial Radiology service at the Erciyes University, Dentistry Faculty. A schematic illustration of the preparation of the samples is demonstrated in [Figure 1](#fig1){ref-type="fig"}. All 269 patients had been referred for CBCT diagnosis and treatment planning and consisted of 19 impacted tooth patients, 178 orthodontic patients, 25 possible pathosis patients, 27 supernumerary tooth patients, and 20 temporomandibular joint disorder patients ([Table 1](#tab1){ref-type="table"}). The shapes of the lingula were classified using the classification proposed by Tsai \[[@B4]\]. To determine the exact location of lingula, the distance of the tip of the lingula from various mandibular ramal landmarks was measured (Figures [2(a)](#fig2){ref-type="fig"}--[2(c)](#fig2){ref-type="fig"}). The CBCT mandibular images were analyzed in the NNT viewer which is a simple version of the NNT software of the CBCT (Newtom 5G, QR, Verona, Italy) machine in a Dell Precision T5400 workstation (Dell, Round Rock, TX, USA) and a 32-inch Dell LCD screen with a resolution of 1280 × 1024 pixels in a darkroom. The contrast and brightness of the images were adjusted using the image processing tool in the software to ensure optimal visualization. All the measurements were carried out by the same person (AES). *Statistical Analyses.* The values obtained were tabulated; and the mean average and respective standard deviations (SDs) were calculated for all distances studied. The data analyses were performed by using the Statistical Package for the Social Sciences (SPSS), version 16.0, (SPSS Inc., Chicago, Ill) and statistical significance was determined at the level of *P* \< .05. The distances were calculated for each of the measurements on right and left sides, and a comparison of the mean values of the right and left sides were made by using *t*-tests. 3. Results {#sec3} ========== The study subjects consisted of 144 (53.5%) girls and 125 (46.5%) boys. The mean age of the patients was 9.17 (SD: 1.87), with ages ranging from 6 to 12 years. The most common shape was the nodular shape of the ML (48.3%, *n* = 260), followed by truncated (23.4%, *n* = 126), assimilated (14.4%, *n* = 78), and triangular (13.7%, *n* = 74) (Figures [3(a)](#fig3){ref-type="fig"}--[3(d)](#fig3){ref-type="fig"}). A bilateral shape (65.7%) was found more often than a unilateral one (34.3%). The MLs were found bilaterally as nodular in 92 sides (52%), truncated in 41 sides (23.2%), triangular in 23 sides (13.0%), and assimilated in 21 sides (1.9%). The distribution of the lingular shapes was also compared between girls and boys ([Table 2](#tab2){ref-type="table"}); there was a statistical difference between gender in triangular and nodular shape. The height, distance of ML from various mandibular ramal landmarks as well as from the distal side of the alveolar socket of the mandibular permanent first molar tooth and ML ratio are shown in [Table 3](#tab3){ref-type="table"}. The ML was located at 13.3 ± 2.3 mm from the anterior border of mandibular ramus, 10.2 ± 1.6 mm from the posterior border of the ramus, and 11.4 ± 2.5 mm from the mandibular notch. The mean distance of the ML from the distal surface of the alveolar socket of the mandibular permanent first molar tooth was 24.7 ±3.7 mm. The mean height (*h* ~1~) of the ML was 5.3 ± 1.6 mm, and the mean ML ratio was determined out to be 0.55 ± 0.04. Statistically significant difference was shown in [Table 3](#tab3){ref-type="table"}. The lingulae were 14.6% (79) at the same level with the occlusal plane, 57.9% (312) above the plane, and 27.5% (97) below the plane. 4. Discussion {#sec4} ============= This is the first study to systematically evaluate the different morphological shapes of the ML in a pediatric population using CBCT images. Pain control during dental procedures is very important in children to maintain a positive relationship between the child and dentist building trust and allaying fear and anxiety \[[@B11]\]. The inferior alveolar nerve block is the most common technique for providing local anesthesia \[[@B12]\] before restorative and surgical procedures of the mandibular posterior teeth \[[@B13]\]. This technique provides anesthesia of teeth, jaw, lip, gingiva, and mucous membrane up to the midline at the related part. However, Malamed identifies the inferior alveolar nerve block as the injection with the highest clinical failure rate, which he reports to be 15 to 20 percent when properly administered \[[@B14]\]. This high failure rate is often attributed to a high degree of variation in the morphology of the mandibular ramus and the location of the ML, especially in childhood and adolescence \[[@B15]\]. Therefore, defining the anatomical characteristics of the ML region plays an important role in successful anesthesia for dental and surgical procedures. Several factors contribute to the reliability of landmark identification in children: the density and sharpness of the images, the anatomic complexity and superimposition of hard and soft tissues, the definition of the landmark, and the training level or experience of the observers \[[@B16], [@B17]\]. CBCT in dentistry has provided an imaging solution that has neither the projection errors associated with magnification nor the superimposition problems associated with traditional panoramic imaging \[[@B18]\]. In addition, CBCT has a wide range of tools, such as 3D reconstructions in any direction to permit the accurate identification of landmarks. Studies have reported excellent accuracy with 3D computed tomography (CT) \[[@B19]\]. Using CBCT (3D) in our study, identification of the AMF reflected a real clinical situation. The frequency of different morphological types of lingula studied by different authors varied among populations and races \[[@B1], [@B5a], [@B5b], [@B20]--[@B25]\] ([Table 4](#tab4){ref-type="table"}). Triangular and truncated shapes of the ML were found most commonly in previous studies \[[@B4]\]. In the present study, a total of 261 nodular shapes were detected in 168 of the 269 patients, and the frequency rate was found to be 48.3%, which differed from the rates generally reported in previous studies. Tuli et al. \[[@B5a], [@B5b]\] observed gender variation of the lingula shapes in their specimens, and in their study, the triangular and assimilated types were the most common and the least common types in males (67.9% and 5%) and females (70.6% and 4.4%), respectively. They observed the truncated type twice as often in males (17.6% sides) as in females (8.8%), and the nodular type was a little less than double in females (16.2%) as compared with males (9.6%). In Jansisyanont et al.\'s \[[@B20]\] study, there was a different genderwise variation observed. Of the 74 lingulae of mandibles belonging to females, from 60 lingulae, the triangular type was found in 21.7%, truncated in 23.3%, nodular in 33.3%, and assimilated in 21.7%, of cases. In case of males, the triangular type was observed in 36.5%, truncated in 31.1%, nodular in 27% and assimilated in 5.4% of cases. In our study, there was statistical difference between gender in the triangular and nodular shapes. The nodular and triangular types were the most and least prevalent types in boys and girls. The height level of the mandibular foramen is an important reference for the inferior alveolar nerve block. Kanno et al. \[[@B3]\] found the mandibular lingula position to be 6 mm above the occlusal plane in 7- to 8-year-old children and 10 mm in 9- to 10- year-old children. Nicholson \[[@B6]\] found the foramen located to be below the occlusal surfaces of the molar teeth. Murlimanju et al. \[[@B22]\] reported in 38 dry, adult, Negroid, Zimbabwean mandibles that 47.1% were at the same level with the occlusal plane, 29.4% were above the and 23.5% were below the plane. Hwang et al. \[[@B26]\] reported that the location of the mandibular foramen changed with age, and that in children, it was located below the occlusal plane, while in adults it was 4.16 mm above the occlusal plane. In Kositbowornchai et al.\'s \[[@B23]\] study, the mandibular foramen, measured from three-dimensional CBCT images, was 10 mm above the occlusal plane. In the present study, the mean height level (*h* ~2~) was measured to be 2.0 ± 1.2 mm above the occlusal plane. A study on Thai mandibles \[[@B27]\] showed that the lingular heights on the right and left sides were 8.7 ± 2.0 mm and 8.2 ± 2.1 mm, respectively. Another study in the Thai population by Jansisyanont et al. \[[@B20]\] reported the height of the lingula to be 8.2 ± 2.3 mm. In the present study, we defined the distance as 5.3 ±1.6 in the pediatric population. Woo et al. \[[@B28]\] reported a study on a Korean population, in which the height of lingula was found to be higher, that is, 10.51 ± 3.84 mm. In a study reported by Samanta and Kharb \[[@B29]\], the lingular height was found to be 5.5 ± 2.02 mm. Another study in a Brazilian population by Monnazzi et al. \[[@B2]\] reported the height of lingula to be 5.82 ± 0.43 mm. Nicholson \[[@B6]\] studied eighty dry, adult human mandibles of East Indian ethnic origin and reported a height of the lingula on the right side to be 8.6 ±4.7 mm and left side to be 9.1 ± 5.7 mm. In the present study, the mean height of the lingula was 5.3 ± 1.6 mm, with a statistical difference between gender on right and left sides ([Table 3](#tab3){ref-type="table"}). The location of the lingula varies among the various ethnic and racial groups \[[@B6], [@B23], [@B30], [@B31]\]. Nevertheless, the common trend was that the distances in females were shorter than or nearly equal to those found in males. This observation is consistent with the knowledge that females generally have larger mandibles than males \[[@B29]\]. In the present study, the mean distance of the lingula to the anterior border of mandibular ramus was 13.3 ± 2.3 mm. The distance of the lingula from the other mandibular ramal landmarks observed in the present study varied from those reported in various populations as shown in [Table 5](#tab5){ref-type="table"}. Jansisyanont et al. \[[@B20]\] and the study by Afsar et al. \[[@B30]\] reported that there was no difference in relation to the mandibular ramal landmarks and the mandibular second molar when comparing sex and side groups except for the distance from the lingula to the distal surface of the mandibular second molar of the female left and the male right groups. In the present study, the statistical differences are shown in [Table 3](#tab3){ref-type="table"}. Although the radiation doses from CBCT are significantly lower than in medical CT, they are generally higher than conventional dental radiography \[[@B32]\]. Recently, the SEDENTEXCT working group proposed provisional evidence-based selection criteria with clinical indications regarding when CBCT should be performed. CBCT should only be used when the clinical question cannot be answered by conventional radiography, and the field of view (FOV) should be limited to the region of interest \[[@B32]\]. Ideally, CBCT equipment should be able to offer a choice of volume sizes to reduce patients\' radiation exposure levels. A risk-benefit analysis must be performed on each individual patient when CBCT is being considered, and in order to assess the risk of CBCT, the effective dose must first be calculated as well. 5. Conclusion {#sec5} ============= The present study provides new information to the literature concerning the shape, height, and location of the mandibular lingual in the Turkish pediatric population. The findings of the present study could be utilized in clinical and dental procedures to localize the lingula and avoid intraoperative complications. The bilateral nodular shape of the lingula was most common in the whole population of study and in each sex. The mean height of the lingula was 5.3 mm. The lingula was located an average of 13.3 mm from the anterior border of the mandibular ramus, 11.4 mm from the mandibular notch, and 24.7 mm from the distal surface of the mandibular permanent first molar. The landmarks for the mandibular nerve block are important for efficient anesthesia during dental treatments. The results from the present study suggest that clinicians or oral surgeons should insert a needle approximately 13.3 mm from the anterior border of the ramus, and approximately 2.0 mm above the occlusal plane due to fact that the lingula in the majority of the samples was found above the occlusal plane. Conflict of Interests ===================== None of the authors have any conflict of interests. {#fig1} ![Three-dimensional CBCT image of the internal surface of the mandible showing the measurement for localizing the position of the lingula, entrance of mandibular foramen from the various landmarks. The "*x*" measurement indicated the distance in millimetres found between the most anterior part of the mandibular lingula points to the anterior border of the mandibular ramus in a straight horizontal line; the "*y*" measurement represented the distance between the most posterior part of the mandibular lingula and foramen points to the posterior border of the mandibular ramus in the same horizontal orientation; the "*w*" measurement determined the distance between the lower point of the mandibular foramen and lingula points to the mandibular base in a vertical straight line; the "*t*" measurement determined the distance most anterior part of the mandibular lingula points to the alveolar socket of first mandibular molar tooth; the "*z*" measurement determined the distance between the same initial points as for the *w* distance to the lower point of the sigmoid notch. The horizontal distances were measured parallel to the occlusal plane (OP) of the molars, whereas the vertical distances were measured perpendicular to the occlusal plane of the molars. The ratio of (*x*) to (*x* + *y*1) was also calculated and used as an additional guide to localize the lingula \[[@B20]\]. Vertical distance from the tip of the lingula to the lower border of the mandibular foramen was measured as height of the lingula (*h* ~1~). The measurement included the distance from the lingula to the occlusal plane of the molars (*h* ~2~).](BMRI2013-825453.002){#fig2} {#fig3} ###### Description of the 269 subjects and their indications for cone beam CT (CBCT) referral. Reason for Scan No. of Subjects ---------------------------------- ----------------- Impaction localization 19 Orthodontic records 178 Other possible pathosis 25 Supernumerary teeth localization 27 TMJ assessment 20 ###### Distribution and incidence (in parentheses) of lingula in girls and boys, bilateral or unilateral. Type Shape Girl   Boy     Total ------ ------------- ----------- ----------- ----- ----------- ---------- ------- --------- ----------- ----------- ----------- ------------ ------------- 1 Triangular 15 (10.4) 20 (6.9) 50 8 (6.4) 8 (3.2) 24 0,021\* 23 (13.0) 16 (17.6) 12 (13.2) 39 (14.5) 35 (13.0) 2 Truncated 17 (11.8) 32 (11.1) 66 24 (19.2) 12 (4.8) 60 0,767 41(23.2) 40 (44.0) 4 (4.4) 81 (30.1) 45 (16.7) 3 Nodular 36 (25.0) 52 (18.1) 124 57 (44.8) 23 (9.2) 137 0,011\* 93 (52.5) 16 (17.6) 59 (64.8) 108 (40.1) 152 ( 56.5) 4 Assimilated 12 (8.3) 24 (8.3) 48 9 (9.6) 11 (4.4) 29 0,069 21 (11.9) 19 (20.9) 16 (17.6) 41 (15.2) 37 (13.8) \*Statistically significance. ###### Distance of lingula from various ramal landmarks of 269 mandibula; distance from distal side of the alveolar socket of mandibular permanent first molar tooth and lingula ratio with their comparison between gender sides. ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------   Side Gender Minimum (mm) Maximum (mm) Mean (mm) Std. deviation *P* value Total -------------------------------------------------------------------------------- ------------ -------- -------------- -------------- ----------- ---------------- ----------- ------- ------- Distance from anterior border of ramus *x*-right 1 10,1 19,4 13,6 2,5 0,13 13,3 2,3 0 9 16,1 12,7 1,9 *x*-left 1 10,2 19,5 14,1 2,1 0,054 0 9,3 18,9 12,8 2,5 Distance from posterior border of ramus *y*1-right 1 7,5 13,2 10,9 1,6 0,000\* 10,2 1,6 0 6,1 12,1 9,2 1,7 *y*1-left 1 8,7 13,8 11,3 1,5 0,015\* 0 7,6 12,8 10,2 1,5 *y*2-right 1 4,2 10,7 7,0 2,3 0,019\* 6,3 2,1 0 2,6 9,9 5,6 1,9 *y*2-left 1 3,2 11,2 7,2 2,3 0,026\* 0 2,4 10,6 5,8 1,9 Distance from mandibular notch *z*1-right 1 8,2 18,2 12,6 2,5 0,06 11,4 2,5 0 7 17,6 11,2 2,5 *z*1-left 1 7,2 16,2 12,4 2,5 0,014\* 0 6,3 15,7 10,6 2,7 *z*2-right 1 13,2 24,4 17,9 3,0 0,041\* 16,7 3,4 0 8,2 23,6 15,7 4,3 *z*2-left 1 8,4 22,5 17,8 3,2 0,005\* 0 6,6 22,5 15,0 3,7 Distance from mandibular base *w*1-right 1 19,9 28,1 23,6 2,2 0,039\* 23,1 3,2 0 11,9 25,9 21,7 4,0 *w*1-left 1 14,7 27,8 23,9 3,0 0,56 0 11,9 26,3 23,5 3,0 *w*2-right 1 12,5 23,8 18,3 2,8 0,29 17,9 3,0 0 12,2 24,3 17,4 3,2 *w*2-left 1 11,9 23,4 18,2 2,9 0,71 0 12,7 22,8 17,9 3,0 Distance from distal side of the alveolar socket of 1st mandibular molar tooth *t*-right 1 18,1 31,1 24,9 3,8 0,23 24,7 3,7 0 16,5 29,4 23,7 3,4 *t*-left 1 16,1 32 25,5 4,2 0,41 0 17,9 30,9 24,7 3,5 Height of the lingula\ *h*1-right 1 3 9,7 5,6 1,4 0,027\* 5,3 1,6 0 2,3 8,3 4,7 1,5 *h*1-left 1 3,6 9,6 6,3 1,6 0,000\* 0 2,7 8,8 4,7 1,5 *h*2-right 1 1,3 5,1 2,1 0,5 0,74 2,0 1,2 0 0,3 5,1 1,9 1,5 *h*2-left 1 0,2 4,7 2,2 1,1 0,2 0 0,2 4,6 1,6 1,3 Lingula ratio: *x*/*x* + *y*1 (%) right 1 0,45 0,59 0,54 0,04 0,48 0,55 0.047 0 0,51 0,63 0,58 0,05 left\ 1 0,44 0,61 0,53 0,05 0,52 0 0,49 0,62 0,55 0,048 --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Groups: 1: boy, 0: girl. \*Statistically significance. ###### The most prevalent shape of lingula in reported studies. Authors Reference Year Population ------------------------ ------------ ------ ------------ Tuli et al. \[[@B5a]\] 2000 Triangular Devi et al. \[[@B1]\] 2003 Truncated Hossain et al. \[[@B21]\] 2001 Triangular Murlimanju et al. \[[@B22]\] 2012 Triangular Kositbowornchai et al. \[[@B23]\] 2007 Truncated Nirmale et al. \[[@B24]\] 2012 Triangular Lopes et al. \[[@B25]\] 2010 Triangular Jansisyanont et al. \[[@B20]\] 2009 Truncated Present study     Nodular ###### Comparison of various studies on location of lingual. Authors Reference Study design Population Year Distance from anterior border of ramus (mm) Distance from posterior border of ramus (mm) Distance from mandibular notch (mm) ------------------------ ------------ -------------- ------------ ------------ --------------------------------------------- ---------------------------------------------- ------------------------------------- Woo et al. \[[@B28]\] Dry mandible Korea 2002 18.6 ± 2.5 16.1 ± 3.5 19.8 ± 5.1 Kositbowornchai et al. \[[@B23]\] Thai 2007 20.7 ± 2.8 15.4 ± 1.9 --- Jansisyanont et al. \[[@B20]\] Thai 2009 20.6 ± 3.5 18.0 ± 2.6 16.6 ± 2.9 Samanta and Kharb \[[@B29]\] India 2012 20.0 ± 2.4 15.0 ± 2.7 15.4 ± 2.7 Monnazzi et al. \[[@B2]\] Brazil 2012 16.5 ± 2.3 14.6 ± 2.13 16.4 ± 2.6 Present study   CBCT Turkey 2013 13.3 ± 2.3 10.2 ± 1.6 11.4 ± 2.5 [^1]: Academic Editor: Mohammadali M. Shoja

1. Introduction {#sec1} =============== Insomnia refers to a subjective experience in which patients are not satisfied with sleep time and quality and affect social function during the day, mainly manifested as difficulty falling asleep, sleep maintenance disorder, early awakening, poor sleep quality, and less total sleep time \[[@B1]\]. Current therapeutic strategies mainly include benzodiazepine receptor agonists, melatonin receptor agonists, and hypnotic antidepressants. However, the current routine pharmacotherapy is less than satisfactory, such that the treatments have many issues, including contraindications, side effects, high costs, and addictions \[[@B2]\]. Recently, the use of complementary and alternative medicine (CAM), such as the Tian Wang Bu Xin Dan, is increasing due to their curative effects, low cost, and wide-range of applications. Treatment with TWBXD has achieved promising clinical effects in terms of better clinical efficacy and fewer adverse effects. However, the strength of this conclusion is limited by the small sample sizes of most trials. In addition, no previous systematic review or meta-analysis has evaluated the effect of TWBXD for treating insomnia. In this study, we systematically evaluated the efficacy and safety of TWBXD in insomnia based on available randomized-controlled trials (RCTs). 2. Methods {#sec2} ========== 2.1. Search Strategy {#sec2.1} -------------------- Systematic searches were conducted with the terms "Sleep Initiation and Maintenance Disorders" or "Early Awakening" or "Insomnia" or "Tianwang Buxin" or "Sleep Initiation Dysfunction" using the following databases: PubMed, EMBASE, Cochrane Central Register of Controlled Trials, Chinese Biomedicine Database (CBM), China National Knowledge Infrastructure (CNKI), Chinese Scientific Journals Database (VIP), and Wan-fang through May 2018. We identified relevant controlled trials on Insomnia, Sleeplessness, or CI using the Cochrane Controlled Trials Register. Other sources searched were conference proceedings, abstracts, thesis dissertations, poster presentations, and materials from professional society meetings. 2.2. Selection Criteria {#sec2.2} ----------------------- Included trials met the following criteria: (1) Parallel-group RCTs, irrespective of blinding or publication status. Crossover trials were only included for the first phase data. Quasi-randomized trials were excluded. (2) Patients diagnosed with insomnia according to internationally accepted diagnostic criteria for insomnia, including the International Classification of Diseases (ICD-10) \[[@B3]\], the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV and DSM-IVR) \[[@B4], [@B5]\], Chinese Classification of Mental Disorders (CCMD-3) \[[@B6]\]. (3) More than 30 participants per group in included trails. 2.3. Types of Outcome Measures {#sec2.3} ------------------------------ The primary outcome measurement was sleep questionnaires such as the Pittsburgh Sleep Quality Index (PSQI) \[[@B7]\]. The secondary outcome measurement was the clinical effective rate based on response evaluation criteria in TCM treatment of insomnia \[[@B8]\] and the adverse events. In Guideline for Clinical Trials of New Patent Chinese Medicines, evaluation standards for clinical therapeutic effects were as follows \[[@B8]\]: (1) clinical cure: sleep time to restore normal sleep time or the nighttime sleep duration of more than 6 hours, deep sleep, and full of energy after waking up; (2) markedly effective: significant improvement of insomnia; sleep time increased over 3 hours compared with the previous sleep time; an increase of the depth of sleep; (3) effective: amelioration in symptoms; sleep time increased less than 3 hours compared with the previous sleep time; (4) ineffective: no significant improvement of insomnia or deteriorated after treatment. 2.4. Data Extraction and Quality Assessment {#sec2.4} ------------------------------------------- Two reviewers independently extracted the data from the selected trials into a standard data extract form. The extracted data included age, gender, trial duration, treatment, outcome, and adverse reactions. We evaluated the methodological quality of the included trials in accordance with the Cochrane risk of bias tool \[[@B9]\]. The judgment of risk of bias includes random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessments, incomplete outcome data, selective reporting, and other sources of bias. Any disagreements were resolved by discussion with a third reviewer. 2.5. Statistical Analysis {#sec2.5} ------------------------- The pooled summary was expressed as odds risk (OR) with 95% confidence interval (CI) for discontinuous variables. Heterogeneity across trials was tested using the I^2^ statistic and Cochrane Q statistic. If I^2^\>50% and p\>0.10 for the Cochrane Q statistic, we selected a random effect model; otherwise, a fixed effect model was applied. Subgroup analysis was performed based on the different treatment durations. To assess potential publication bias, a funnel plot was carried generated. All statistical analyses were conducted using Review Manage software 5.3 (Cochrane Collaboration, Oxford, UK). 3. Results {#sec3} ========== 3.1. Description of Studies {#sec3.1} --------------------------- A total of 790 records were retrieved from the above-mentioned databases and through the manual literature search. Of these, 371 articles were excluded upon exclusion of duplicated publications. After reading the titles and abstracts, another 393 articles were removed. Thus, 26 full-text articles were assessed for eligibility. After a detailed assessment of the full-text papers, an additional 12 articles were excluded mainly because they did not satisfy our predefined inclusion criteria. Finally, 14 RCTs \[[@B10]--[@B23]\] were included in the meta-analysis ([Figure 1](#fig1){ref-type="fig"}). 3.2. Trial Characteristics {#sec3.2} -------------------------- [Table 1](#tab1){ref-type="table"} presents the baseline characteristics of patients in the trials included in the meta-analysis. All the included trials were carried out in China and published from 2006 to 2018. The sample sizes ranged from 60 to 133. A total of 1,256 patients were included in the 14 trials. The TWBXD group consisted of 655 patients, while the conventional Western medicine (WM) treatment group consisted of 601 patients. 11 trials were diagnosed according to CCMD \[[@B10]--[@B14], [@B16], [@B19]--[@B23]\], 2 were based on ICD-10 \[[@B15], [@B17]\], and 1 was diagnosed according to TCM syndrome diagnosis and treatment standard. All trials adopted TWBXD monotherapy or adjunct therapy in the treatment group for insomnia. Among the 14 trails, 12 were monotherapy \[[@B10]--[@B18], [@B21]--[@B23]\] and the rest of 2 studies \[[@B19], [@B20]\] were combined with the WM to treat the insomnia. The duration of treatment was varied from 2 weeks to 8 weeks. Clinical efficacy was observed in 13 studies \[[@B10]--[@B16], [@B18]--[@B23]\]; PSQI score was tested in 6 studies \[[@B11], [@B12], [@B15], [@B17], [@B21], [@B22]\]. Adverse effects were reported in 6 studies \[[@B11], [@B15], [@B21]--[@B23]\], while there is no mention in the other studies ([Table 1](#tab1){ref-type="table"}). 3.3. Methodological Quality of the Included Trials {#sec3.3} -------------------------------------------------- [Figure 2](#fig2){ref-type="fig"} summarizes the methodological quality of the 14 RCTs. 6 trials \[[@B11], [@B13], [@B15], [@B17], [@B18], [@B23]\] were randomized using random number tables to generate a sequence (appropriate), 1 trials \[[@B19]\] used a temporal sequence for randomization (inappropriate), and the remaining trials only mentioned randomization without detailed methods. 1 trial \[[@B15]\] reported the patients\' reasons for withdrawal or loss to follow-up. 3.4. The Primary Outcome {#sec3.4} ------------------------ We use the PSQI as primary outcome, and there were 6 RCTs \[[@B11], [@B12], [@B15], [@B17], [@B21], [@B22]\] comparing TWBXD monotherapy with conventional medicine. As shown in [Figure 3](#fig3){ref-type="fig"}, significant heterogeneity (I^2^=86%, p\<0.001) between trials was observed, and thus, we selected a random effect model. Overall, TWBXD group could reduce PSQI total score better compared with WM treatment alone (WMD=-1.82, 95% CI \[-3.00,-0.64\], P=0.003). We use the methods of eliminating each test one by one to analyze the sources of heterogeneity, showing that the heterogeneity between the trials is still significant (I^2^\>50%), but the result is stable ([Figure 3](#fig3){ref-type="fig"}). 2 trials \[[@B11], [@B12]\] analyze the time required to fall asleep and the quality of sleep. The result indicates that the effect of TWBXD is better than that of WM in shortening the time of falling asleep (WMD=-0.42, 95% CI \[-0.65,-0.19\], P=0.0003) and there was no difference in the quality of sleep between this two groups (WMD=-0.27, 95% CI \[-0.85,0.31\], P=0.36) (Figures [4](#fig4){ref-type="fig"} and [5](#fig5){ref-type="fig"}). 3.5. The Secondary Outcome {#sec3.5} -------------------------- 13 RCTs were assessed by clinical effective rates. Among the 13 trails, 11 \[[@B10]--[@B16], [@B18], [@B21]--[@B23]\] were comparing TWBXD monotherapy with conventional medicine, and the rest 2 studies \[[@B19], [@B20]\] were combined with the WM. The results showed a significant difference in favor of TWBXD with or without WM (OR=5.05, 95% CI \[1.58,16.12\], P=0.006; OR=2.71, 95% CI \[1.67,4.39\], P\<0.0001) (Figures [6](#fig6){ref-type="fig"} and [7](#fig7){ref-type="fig"}). Adverse event monitoring was only reported in 5 studies \[[@B11], [@B20]--[@B23]\]. No serious adverse effects were mentioned in these studies. Li et al. \[[@B11]\] reported that there was no adverse event happened in the TWBXD group or WM group. In the study by Si et al., Liu et al., and Wu et al. \[[@B20], [@B22], [@B23]\], no adverse event happened in the TWBXD with or without WM group; there were 16 cases occurred in the WM group. Among the adverse events, six patients had the headache, one patient had erythra, and the nine patients had thirst. Liu et al. \[[@B21]\] reported that 2 events occurred in TWBXD group, and 9 events happened in the WM group. 4. Discussion {#sec4} ============= 4.1. Summary of Evidence {#sec4.1} ------------------------ To the best of our knowledge, this is the first systematic review and meta-analysis to evaluate the efficacy and safety of TWBXD for the treatment of insomnia. In this review, we analyzed treatment effects in 1,256 patients with insomnia from 14 trials \[[@B10]--[@B23]\]. Meta-analyses of these RCTs over the course of 2-8 weeks show that TWBXD seems to be beneficial for reducing the time to fall asleep. These clinical benefits are mainly associated with TWBXD. The safety of TWBXD treatment was also examined. There were only 1 case with minor adverse events in the TWBXD or TWBXD combined with WM group, which is in agreement with long-term clinical experience. However, the potential benefits of TWBXD in the treatment of insomnia remain uncertain due to methodological deficiencies and limited sample size for inclusion in the study. Therefore, recommendations for clinical practice should be cautious. 4.2. Limitations of the Review {#sec4.2} ------------------------------ First, some methodological deficiencies existed in most of the included trials. Few studies described the generation of a random sequence in detail, the concealment of allocation, and the methods of blinding. Hence, selection bias may exist in our analysis. No placebo was used in the treatment as well, which reduced the possibility of blinding and thus increased the possibility of detection bias or performance bias in the implementation process of the trial, as well as reporting bias in outcome evaluation \[[@B24], [@B25]\]. Dropouts also were not mentioned in all studies. We attempted to contact the authors via telephone or e-mail to obtain additional statistical data or methodological information, but most of the detailed responses could not be obtained. In addition, there were only 5 studies \[[@B11], [@B20]--[@B23]\] that reported adverse events in this systematic review, and the incidences of treatment-effect adverse events were used as the judgment of safety, which could not be considered as an objective indicator for severity evaluation of adverse reactions. A side effect scale, such as TESS, should be adopted and standardized as the safety evaluation method. In addition, no trials conducted pretrial estimation of sample size, which indicated the lack of statistical power to ensure appropriate estimation of the therapeutic effect \[[@B26]\].Thus, specific evidence for efficacy and safety are required in future studies. Secondly, substantial heterogeneity was observed in the pooled total effective rate. The reason for the heterogeneity may be correlated with the use of different types of WM \[[@B27]\] and different doses of the TWBXD. Third, publication bias may be a serious problem in reported herbal trials. Current studies show that Asian countries, including China, typically report a large proportion of positive results, although these studies were independently selected by the two authors and are strictly based on inclusion and exclusion criteria \[[@B24]\]. No negative results were found in this evaluation \[[@B28]--[@B31]\]. All of these studies were conducted and published in China, and most of the studies showed that TWBXD performed better than the WM group. A possible reason for these negative results could be explained by selection bias in the process of participant recruitment \[[@B32]\]. 5. Conclusion {#sec5} ============= There is insufficient evidence regarding the efficacy of TWBXD for the treatment of insomnia because of the poor methodological quality and the small number of trials of the included trials. TWBXD is generally safe, but there is insufficient evidence to make conclusions on the safety because fewer studies reported the adverse events. More high-quality RCTs with large sample sizes are needed to verify the current findings. Conflicts of Interest ===================== The authors declare that they have no conflicts of interests. Authors\' Contributions ======================= Ling Liu and Xi-qian Yang conceived and designed the study. Xi-qian Yang and Shu-ping Ming were responsible for acquisition of data, analysis, and interpretation of data. Xi-qian Yang, Shu-ping Ming, Jie Fang, and Dong-nan Wu drafted the manuscript. All authors read and approved the version to be published. {#fig1} {#fig2} {#fig3} {#fig4} {#fig5} {#fig6} {#fig7} ###### Basic characteristics of the included studies. -------------------------------------------------------------------------------------------------------------------------- Studies Sample Interventions Duration (weeks) Outcome  \ Adverse effects index ------------------- -------- --------------- ------------------- ---------------------- ----------------- ------ ---- ---- Zhangwanting 2006 36 30 TWBXD Estazolam 2 ① \- Limingzi 2012 30 30 TWBXD Bailemian Capsule 4 ①②③④ 0 0 Chenweiming 2012 60 30 TWBXD Estazolam 8 ①②③④ \- Guochangxue 2012 40 40 TWBXD Estazolam 2 ① \- Lipinghua 2013 44 44 TWBXD Estazolam + oryzanol 4 ① \- Hanyanchao 2013 32 32 TWBXD Alprazolam 4 ①② \- Zhuzhenhua 2013 50 50 TWBXD Estazolam + oryzanol 4 ① \- Xiejinping 2014 48 27 TWBXD Diazepam 4 ② \- Jiangaiju 2015 58 62 TWBXD Estazolam 4 ① \- Zhangfanguo 2015 40 40 TWBXD + Diazepam Diazepam 4 ① \- Shichuming 2015 50 50 TWBXD + Estazolam Estazolam 4 ① 0 10 Liuzhanbing 2015 67 66 TWBXD Alprazolam 4 ①② 2 9 Wubaojin 2017 40 40 TWBXD Estazolam +oryzanol 4 ① 0 1 Liushijun 2017 60 60 TWBXD Estazolam 4 ①②③④ 0 5 -------------------------------------------------------------------------------------------------------------------------- ① Clinical effective rate; ② PSQI total score; ③ time required to fall asleep; ④ Quality of sleep. [^1]: Academic Editor: Mario Ledda

Introduction {#s1} ============ Visual attention: models, definitions, and influential factors -------------------------------------------------------------- Our view of the world to a large degree rests on the information we gather by the visual sense, and visual attention plays a central role in the processing of the sensory input. By directing one\'s attention to specific features in the environment, these features are processed in more depth than those environmental characteristics that are not the focus of attention (Chun et al., [@B15]). Several psychological and neuropsychological models of attention have been developed so far (e.g., Broadbent, [@B11]; Kahneman, [@B48]; Posner and Petersen, [@B91]; Mirsky, [@B76]; Coull, [@B21]; Corbetta and Shulman, [@B19]), but nonetheless a precise definition of attention is lacking. Currently, it is generally accepted that several sub-systems of attention can be differentiated. Based on neuropsychological findings and single-cell recordings in monkeys, Posner and Petersen ([@B91]); Petersen and Posner ([@B86]) suggested three neuronal networks which are distinctive but collaborate to direct attention to specific locations in space. The posterior attentional (orienting) system prioritizes the sensory input by selecting a modality and location. The anterior attentional system is involved in detecting stimuli of current relevance by capturing awareness in a specific way and slowing detection of another potential target. This process is related to the limited capacity of the attention system and awareness, hence often labeled as focal attention. Finally, they postulated an alerting system being involved in preparing and sustaining alertness to process high-priority signals. This classical framework has recently been updated on the basis of new findings that deepen or expand the original networks (Petersen and Posner, [@B86]). According to Coull ([@B21]), one sub-system is responsible for directing attention to a stimulus (attentional orientation). A second sub-system mediates selective attention by giving focused attention to a particular stimulus while diminishing the attention to other stimuli (biased competition). A third sub-system provides divided attention between several stimuli even when they are spatially separated (Hahn and Kramer, [@B40]), and hence indicates that the current eye position does not necessarily correspond to the focus of attention. A fourth sub-system is additionally responsible for sustaining the attention to one or several (even non-contiguous) zones of the visual field (Müller et al., [@B78]). Corbetta and Shulman ([@B19]) found evidence for a dorsal frontoparietal network, including intraparietal sulcus, frontal eye field, and middle temporal complex, which was the primary network involved in selective attention to stimuli in the environment. Endogenous (i.e., voluntarily directed) shifts of attention activated the dorsal network, but it was co-activated with a second ventral frontoparietal network (right temporoparietal junction, right ventral frontal cortex, and insula), when unexpected but behaviorally relevant stimuli occurred at unattended locations. Hence, the ventral network was labeled as "stimulus-driven" (and was also often labeled as exogenous; for a detailed description of endogenous and exogenous shifts of attention, see Cormick, [@B20]). Independent of whether these models are sufficient to explain all phenomena of attention, and independent of how the postulated sub-systems specifically interact on the neuronal and functional levels, the core aspect in all models is the bundling of cognitive resources to adequately process stimuli being important at a certain time. Correspondingly, attention often is metaphorically described as a moveable spotlight facilitating the stimulus inside its beam (Eriksen and Yeh, [@B27]; Posner and Petersen, [@B91]), whereby Desimone ([@B23]) pointed out that the enhancing effect of attention on neuronal responses is perhaps better understood in terms of a competition among all of the stimuli in the visual field for control over behavior (biased competition hypothesis). The central question that has motivated much research addresses factors influencing this attentional spotlight or the fate of sensory items during competitive interactions. Although not always explicitly mentioned, the visual system is the primary focus of models and experimental studies. Of course, due to the outstanding significance of the visual sense for human perception and behavior, visual attention adopts a key role in the selective choosing of input that, in turn, substantially constitutes the base for planning and controlling our interactions with the external world. Thereby, multiple factors contribute to the guidance of visual attention: on the one hand, the pure visual characteristics of the outside world, as well as the importance of objects for action control, substantially determine the course of the attention focus. On the other hand, visual attention is also heavily influenced by the internal state of the individual observer, constituted by a huge diversity of factors such as his current needs, the prevailing emotion, and motivational tendencies, as well as more time-independent personality traits. Hence, attention is not only stimulus-driven. In the context of attention research, shifts of attention with saccadic eye movements (overt attention) as well as covert attention phenomena (attentional shifts without eye movements) have moved to the center of interest. However, as Corbetta and colleagues ([@B18], 761) pointed out, "the relationship of visuospatial attention and eye movements is controversial." On the one hand, it has been known for a long time that humans can attend to objects out of their line of sight (James, [@B47]); i.e., it is possible to attend to several locations in the periphery while saccades are not allowed (e.g., Posner, [@B90]). On the other hand, a shift in overt attention seems to require a shift in covert attention (Hoffman and Subramaniam, [@B45]). Correspondingly, functional anatomical data showed that overt and covert shifts of attention not only are functionally related, but they also share parts of the same neuronal networks in the human brain (Corbetta et al., [@B18]; Beauchamp et al., [@B7]). In the present review, we refer to both overt and covert visual attention and their potential interplay with specific higher cognitive functions. In this context, it must be noted that overt attention is not exclusively driven by external stimuli, just as processes of covert attention are not affected only by the inner state of the observer. In order to clarify what we mean when talking about the impact of high-level cognitive mechanisms on overt or covert visual attention, we will briefly describe the two current main research fields dealing with overt attention, as they illustrate nicely the difference between low- and high-level factors in visual attention. Many eye-tracking studies have addressed the impact of low-level image properties on eye movement guidance by comparing spatial image fixation probabilities. In this sense, the bottom-up selection of image locations involves fast stimulus-driven mechanisms such as a compulsory look to unique features (Treisman and Gelade, [@B110]) or at abrupt occurring stimuli (Yantis and Jonides, [@B118]). Indeed, sometimes we consciously recognize the strong impact of high salient objects, which seems to automatically attract our attention even if we try to defend ourselves against it. Hence, in literature, visual saliency is considered one of the main determinants of gaze control by first attracting our attention and then mediating viewing behavior (for a current review, see Schütz et al., [@B103]). Empirical studies suggest the significance of these bottom-up mechanisms on humans\' attention: Several image features were shown to influence fixation behavior, such as spatial contrast (Reinagel and Zador, [@B97]), luminance and edges (Krieger et al., [@B60]), and color (Frey et al., [@B36]). In addition to task dependent information and spatial viewing biases, basic image features were found to make a significant and independent contribution to overt visual attention (Kollmorgen et al., [@B59]). Thereby, the impact of low-level image features was found to be stable across multiple viewings of complex visual scenes (Kaspar and König, [@B51]). Besides this impact of low-level image properties, high-level cognitive mechanisms are suggested as the second important influential factor on viewing behavior (Yarbus, [@B119]). Previous literature showed, for example, that a congruency between a stimulus and the content of the working memory can early and involuntarily attract attention (Soto et al., [@B104]). Also, the current behavioral mode seems to affect the allocation of attention, since drivers and passengers in a virtual environment were found to differ in their sensitivity to changes in the environment (Wallis and Bülthoff, [@B115]). Moreover, in the context of eye-tracking studies, sometimes the effects of the current task on viewing behavior (e.g., browsing mode versus search mode on web pages) are explicitly considered (Triesch et al., [@B111]; Nelson et al., [@B80]; Rothkopf et al., [@B100]; Betz et al., [@B8]; Hamborg et al., [@B41]), as are effects of age on attention (Mather and Carstensen, [@B73]; Acik et al., [@B1]). However, beyond these factors, there are further substantial mechanisms affecting attention control. They range from the current emotional state of the observer and his motivational state to time-invariant personality traits. So far, few studies have addressed these issues, although even saliency models explicitly point out that high-level cognitive processes can significantly affect the formation of a current saliency map. In the following, we give a selective review of studies which focused on these high-level mechanisms in the context of overt as well as covert attention processes. Although a considerable amount of research is devoted to visual attention, a commonly used definition or model of attention is lacking. Several models have been proposed by now, each highlighting important aspects of attention. All in all, attentional processes seem to be more complex, as suggested by the simple spotlight metaphor. The distinction between overt and covert attention shows that shifts in attention are not necessarily paralleled by shifts in eye movements. In this context, overt attention elegantly allows the study of the impact of high- vs. low-level factors in visual attention, as eye movements and their potential correlation with image properties can be controlled while high-level factors are manipulated (and vice versa). While the impact of low-level images on overt attention has been extensively investigated in the last two decades, the impact of high-level factors on attention (overt and covert) has been neglected. The present review intends to give an overview of various, partially heterogeneous, studies on two widely neglected high-level factors in visual attention (emotion and personality) in order to give starting points for future research as well as point out important methodological aspects that have be addressed appropriately. The ambiguous concept of emotion -------------------------------- Is there a point in time when we are completely free of any emotional impact? Probably not! Of course, often we are not aware of the specific emotion that influences our current thinking and behavior. In contrast to other cognitive or physical processes such as motivation, a specific intention or a present pain might not be in the foreground of conscious perception, but might modulate our mental state. Yet, specific emotions do correlate with the arousal level and hence with the degree to which we are sleepy or focused in an attentional sense. Thereby, it seems fruitful to distinguish between the two constructs of emotion and arousal. According to Kensinger ([@B53], 241), "a widely-accepted framework proposes that affective experiences are best characterized in a two-dimensional space." In one dimension, valence ranges from highly negative to highly positive, and in a second orthogonal dimension, arousal ranges from calming to exciting. In several previous studies, it was neither obvious nor explicitly mentioned what dimension was addressed by the experimental manipulation, which hinders comparison or integration with further research. With a view to the current literature, the common labeling appears to be "emotional valence" and "emotional arousal," respectively. However, sometimes the term "affect" seems to be used instead of valence (e.g., Isen et al., [@B46]). This distinction is important, as changes in both arousal and valence do not necessarily lead to identical neuronal activation, as found by Lane et al. ([@B63]). They showed that extrastriate visual cortical and anterior temporal areas were independently activated by emotional arousal and valence. On the other side, emotional arousal and valence were also found to interact under certain circumstances: Adolphs et al. ([@B2]) investigated the recognition of emotional arousal and valence in a subject with complete bilateral damage restricted to the amygdala. Recognition of emotional valence was normal, but recognition of emotional arousal was impaired for unpleasant facial expressions, words, and sentences. Additionally, memory can be enhanced for negative or positive stimuli which do not evoke arousal (Kensinger and Corkin, [@B54]). However, this simple orthogonal relationship between valence and arousal indeed facilitates theory-building and experimental operationalizations, but there is also evidence for a more intermingled relationship between both constructs. Several theorists postulated valence or arousal as fundamental for emotional experience (e.g., Ortony et al., [@B81]; Lazarus, [@B66]); others incorporate both aspects (e.g., Lang, [@B64]). Furthermore, some authors, such as Eysenck ([@B29]), declared that emotion results from arousal. In contrast, others view arousal as a result of emotion (Gray, [@B39]). Moreover, Barrett ([@B5]) showed that people differ in how they label their own emotional states. People high in valence focus but low in arousal focus refer to a dimensional model when labeling their emotional states. In contrast, people high in arousal focus but low in valence focus prefer a discrete emotion model. Finally, Derryberry and Rothbart ([@B22]) pointed out that the unitary nature of both constructs can be questioned. They determined that at almost every level of physiological analysis (cortical, autonomic, and endocrine), arousal was revealed to be a multidimensional set of processes. Similarly, the limbic system, with its numerous circuits, allows a variety of emotional, perhaps separable, systems controlling different emotions such as fear, pleasure, or frustration. In this context, however, neuroscience shows that it could be an obstacle if emotional experiences were assigned to discrete categories such as fear, anger, or happiness. On the neuronal level, the search for co-occurring sets of neuronal features that differentiate between such categories was not successful (Barrett, [@B6]). On the other hand, Talarico et al. ([@B107], 382) emphasized that "although dimensional accounts of emotion are informative, the influence of discrete emotions should not be underestimated." Consequently, as with attention, no common sense seems to exist regarding a definition and theoretical framework of emotion. However, some practical recommendations can be made: One should not only define the type of emotional dimension (arousal and valence) that is manipulated or investigated, but also specify the kind of emotion or affect as clearly as possible. The term *mood* is sometimes used to describe the inner state of participants (as shown below), but it should be avoided against this background, because it is too general and does not allow a clear categorization of the observer\'s current emotional state. Clear labeling will facilitate the interpretation of data, and hence the comparison with other studies will become easier and more unambiguous. Advantages and disadvantages of a dimensional versus categorical framework for emotional experiences have to be assessed (see Section "Interaction Between Emotional and Attentional Processes on the Neuronal Level"). The impact of emotion on attention ================================== In studies dealing with the potential impact of emotion on other cognitive processes, a neutral emotional state is commonly used as a baseline and then contrasted with salient emotional states such as an exalted joy or strong fear. These more pronounced emotional states are in general characterized by higher valence or arousal. For example, the impact of emotional arousal on the sensitivity of several cognitive functions has been investigated in multiple studies: recall of peripheral details of an event is improved when being in a positive emotional state in contrast to a neutral state (Talarico et al., [@B107]). Furthermore, positive affect, in contrast to a negative effect, can boost performance in creative solving tasks (Isen et al., [@B46]) and also enhances performance in solving resource dilemmas (Knapp and Clark, [@B57]). Given these effects on memory and higher cognitive functions, it is only a small step to the question of to what degree the emotional state of an observer influences visual attention and perception. Based on this central question, Wadlinger and Isaacowitz ([@B114]) investigated by means of tracking eye movements how a specific mood affects the attentional preference to emotional stimuli. Individuals induced into positive mood, in contrast to a neutral mood, fixated more on peripheral stimuli than did control participants and displayed a broadened attention distribution. However, this only held true for high-valence positive stimuli, while the arousal ratings of all images were similar. Participants under induced positive mood additionally made more frequent saccades for slides of neutral and positive valence. Hence, they found a significant effect of priming a positive mood. According to Frederickson ([@B35]), the fundamental function of positive emotions is to broaden the momentary thought-action repertoire of an individual, in contrast to negative emotions that narrow this repertoire toward specific actions in order to serve capacities for survival-related actions. However, this broader focus of attention is not limited to positive stimuli in the environment. Several studies have shown that persons being in a positive emotional state attend to negative information when it could be of future advantage to them. For example, Reed and Aspinwall ([@B96]) provided evidence that when recalling positive experiences of the past, the biased processing of self-relevant health-risk information was reduced. This finding is not compatible with an alternative explanation of the broader focus of attention to positive stimuli evoked by a positive mood: Wadlinger and Isaacowitz ([@B114]) also considered the possibility that the broader distributed attention to positive stimuli may act to maintain the current positive emotional state. Here, a general distinction is important when talking about the impact of emotion on attentional processes: on the one hand, many studies focus on the impact of emotional stimulus properties on the attentional focus and the perception of those emotional stimuli. On the other hand, one can also be interested in the impact of the observer\'s current emotional state on the course of attention, irrespective of whether the observed stimulus itself is emotion-laden. In the first case, attention is directly driven by the emotional features of the stimulus and hence is potentially related to so-called bottom-up processes in the visual hierarchy. In the second case, the emotional component playing a role in attentional shifts is located within the observer and hence affects attention processes on subsequent stimuli in a more top-down manner. In the following, we will speak of an externally located influence when referring to the impact of emotion-laden stimuli, and we will speak of an internally located impact when referring to the observer\'s current emotional state. The distinction between internally and externally located emotional influences facilitates the evaluation of attentional processes affected by emotion. However, this distinction is not without problems. The internally located impact of emotion can be undoubtedly assumed as far as the emotional state of an observer precedes and is therefore not dependent on the stimulus. The stimulus-driven way can be assumed for certain if the observer was in a neutral emotional state when confronted with the stimulus. Unfortunately, both cases rarely occur in this pure form. Thereby, investigations on the impact of externally located emotional properties on attention and perception can be handled: given an adequate experimental procedure controlling for a systematic covariance between dependent and confounding variables, inter-individual differences in the current emotional state are balanced out and hence effects between experimental conditions can be derived from the experimental manipulation (i.e., the visual input). For example, in order to investigate the impact of positive vs. negative priming on viewing behavior under natural conditions, potential inter-individual differences in the current emotional state have to be equalized by randomization of subjects to treatments (given a high number of subjects). Otherwise, differences between the priming conditions regarding viewing behavior could be derived from systematic differences between subjects\' pre-experimental emotional states, but not from the treatment. In contrast, the internally located impact of emotion on perception is more difficult to adequately manipulate and control. One critical parameter is the time interval between the experimental induction of a certain emotional state in the observer and the observation process. This interval moderates the strength of the emotional effect on attention. Emotional processes and changes are linked to the hormone system working slower than neural processes. Hence, the strength of an emotion induction reaches its maximum with a temporal delay, and the level of arousal is variable over time. Therefore, some researchers decided to implement an emotional manipulation that works continuously during the whole experimental session to maintain the emotional state. For example, Hirschberger et al. ([@B44]) introduced permanent subliminal emotional priming in order to maintain a context of negative valence while recording subjects\' fixation behavior on a four-image cluster. A further difficulty for investigations on the internally located impact of emotions on attention and perception is the fact that the visual stimuli observed are not emotionally neutral in most cases. Many pictures of the International Affective Picture System (IAPS) Lang et al., [@B65]) depict scenes and objects comparable to those used as target images in many studies, but valence and arousal ratings for the IAPS images show that most pictures are not (and should not) be perceived as neutral. Consequently, it cannot be ruled out that an induced emotional state, i.e., the internally located impact of emotion, interacts with the externally located impact of emotional stimuli. Finally, the observation of specific emotion-laden stimuli in a sequence can result in a more and more internally driven impact of emotion, although the researcher is exclusively interested in the impact of image properties on perception processes. Importantly, in addition to the division of an internally and externally located impact of emotion on attention, researchers sometimes use stimuli showing the emotion of somebody else (e.g., Vermeulen et al., [@B112]; see below), which is often taken as shorthand for an emotionally positive or negative stimulus. It might indeed induce the same emotion, yet it is a very specific choice. These interactions and complexities show that a clear distinction between emotional bottom-up and top-down influences on perception is a difficult venture, and we have to be careful when inferring from study results the exact entity of the emotional impact on perception. The interaction of internally and externally located emotional effects is demonstrated nicely in recent eye-tracking studies (Kaspar et al., [@B49]): Sets of complex scenes with high positive or negative valence as well as neutral images were used as supraliminal emotional primes to build corresponding emotional contexts. Nature scenes served as neutral target stimuli and were shown embedded in the three sets of primes. Subjects did not differentially scan the primes, but viewing behavior on neutral target images was significantly affected by the valence of the emotional context. Consequently, no simple transfer effects from the primes to the targets occurred on the level of eye movements. Rather, the results suggest that viewing behavior was indirectly influenced by the effect of prime valence on the inner state of the observer. Reducing the intensity of emotional priming and embedding single positive and negative primes in a train of neutral images belonging to different categories, negative primes were scanned more actively than their positive counterparts. This applies to fixations of shorter duration, longer saccades, and a more spread-out fixation distribution. Again, the signature of eye movements on primes was not transferred to neutral target images, but viewing behavior on targets was influenced by the valence of the preceding prime. These results illustrate the complex interaction between internally and externally located effects of emotion on attention. Hirschberger et al. ([@B44]) were interested in whether even unconscious negative priming influences the observer\'s viewing behavior when confronted with emotionally negative target images. In the experimental condition, a subliminally presented negative prime (the word *death*) was present across the whole duration of an eye-tracking session. In the control group, a neutral prime was presented and without any specific emotional manipulation. In their first experiment, subjects consecutively observed several four-image patterns depicting three neutral images and one picture showing physical injury. Participants were asked to recognize the images at the end of each trial in order to enhance their motivation to carefully observe the images. The results showed that the negative prime decreased gaze duration toward pictures of physical injury, but did not have an effect on gaze duration toward neutral images. Hence, participants in the negative priming condition spent less time looking at the pictures overall in contrast to subjects in the neutral priming condition. The authors concluded that the results are a signature of motivated unconscious attention, since subliminally presented negative primes evoked a strong influence on the amount of attention that is directed at meaningful stimuli. In addition to the impact of emotion on eye movements, it is only natural to ask for target detection in a search task. Accordingly, researchers also have already investigated whether perceived emotions affect target detection. Vermeulen et al. ([@B112]) used the attentional blink paradigm to demonstrate that processing fearful versus disgusted faces has different effects on attentional processes. In the attentional blink paradigm, a pair of targets is presented consecutively within a small time window. Thereby, the second target is normally missed when presented 200--500 ms after the offset of the first target (e.g., Raymond et al., [@B95]). Vermeulen and colleagues presented either a fearful or disgusted face as primes followed by a rapid sequence of distractors consisting of random strings of symbols and digits. Two target words of the same length were embedded into this sequence. Subjects had to detect both words, but results showed that processing fearful faces (which were previously assumed to enhance the allocation of attentional resources) impaired the detection of the second target word to a greater extent than did the processing of disgusted faces (which were expected to diminish the allocation of attentional resources). Obviously, the kind of emotion that is perceived plays an important role and has to be considered when investigating emotional effects on attention and perception, because the induction of a negative emotion leads to a reduced spread of attention not only in space, but also in time. However, it must be pointed out that it is unclear to what extent the short presentation of one second of emotional faces in this study affected the emotional-state of the observer. Moreover, it is important to note that the observation of emotional faces does not have the same effect as the observation of scenes depicting emotional content. Hariri et al. ([@B42]) compared the response of the amygdala to fearful and threatening faces with the response to corresponding complex scenes. They found a significantly greater amygdala response to faces in contrast to scenes. When we observe emotional faces, we look at another person, but we do not have direct sensory access to what is affecting the other person. The amygdala obviously responds more strongly to such indirectly mediated emotions. However, especially for studies on overt attention, not only the affective impact of images should be considered, but also the image characteristics which revealed a very strong impact on the attention focus (see, e.g., Kaspar and König, [@B50],[@B51]). In general, many studies use human faces to induce different emotion conditions. Perhaps this is due to the fact that facial expressions are relatively easy to interpret for an observer, and hence specific emotions are recognized faster and without ambiguity. Interestingly, it seems that the emotion expressed by a face is more important than that face\'s corresponding physical appearance. Lamy et al. ([@B62]) presented the first report of emotional priming of pop-outs. In the basic pop-out task, a target differs in a specific feature from multiple simultaneously presented distractors (Maljkovic and Nakayama, [@B72]). Remarkably, when in subsequent trials the identity of the target is repeated, detection is accelerated. In the study of Lamy and colleagues, participants repeatedly saw arrays of four faces and had to detect a discrepant target face that differed in the expressed emotion. In line with the classic pop-out effect, the detection of the target face was faster when the same emotional target face was presented on successive trials. However, this effect only occurred for angry and for happy faces embedded in an array of neutral faces, but neutral target faces displayed in parallel with emotional faces were not detected faster. Importantly, this effect completely disappeared in trials in which all faces (target and distractors) were inverted instead of upright. This result indicates that the faces\' emotional category, rather than their physical properties, was the determinant for this pop-out effect. To conclude, the current emotional state of a person does not only affect recall performance, performance in creative solving tasks, or in solving resource dilemmas. Moreover, emotion also was revealed to be a significant factor influencing humans\' visual attention, but the distinction between an internally and an externally located impact of emotion seems necessary in this context. Both externally located emotion-laden stimuli and the internally located current emotional state of an observer affect overt attention. Thereby, the kind of emotional stimuli (faces vs. scenes) is relevant, and the study design as the manipulation of emotions is a challenging task. Perception of emotion outside the focus of attention ==================================================== If no eye movements are recorded, as in the study of Lamy et al. ([@B62]), we will not be able to assess the degree to which overt and covert attentional processes contribute to observed effects. In fact, several studies have shown that covert attention also plays a crucial role in the context of emotion\'s impact on attention and perception processes. To further disentangle the interaction of emotional stimuli and attention, it is a necessary step to present primes outside of the focus of attention. Phelps et al. ([@B87]) investigated the influence emotions have on the interaction of covert attention and perception. Subjects had to perform a two-alternative forced choice orientation discrimination task. One or four faces were presented in the periphery, followed, after a short delay, by four Gabor stimuli; one of these served as the target. Presentation times of all stimuli were short to exclude eye movements and ensure that covert attention was measured. They found that subjects were most sensitive to stimulus contrasts when they previously observed one face in the periphery that had a fearful expression in contrast to a neutral expression. Presently, it cannot be excluded that fearful faces elicited an unspecific arousal; Phelps and colleagues argue, however, that this effect was derived from the specific negative valence of the primes. This implies that emotion facilitates early visual processing. In other words, an internally located impact of emotion can be induced without directing one\'s attention to emotion-laden stimuli. In more natural situations with complex arrangements of objects, the question arises whether humans extract the affective content without fixation. Calvo and Avero ([@B12]) used pleasant and unpleasant complex images as primes that significantly differed in valence and arousal. The primes were shortly presented in parafoveal locations, followed by a blank and then by a foveal probe stimulus being either congruent or incongruent in emotional valence with the preceding prime. Participants had to promptly signal whether this probe was positive or negative by pressing a key. The results showed that congruent prime--probe pairs were detected faster than incongruent pairs when the prime--probe stimulus onset asynchrony was 300 ms and the prime appeared in the left visual field. This priming effect occurred regardless of differences between prime and probe in physical appearance and semantic category, since the assignment of primes and probes was random and both depicted scenes involving either people or animals. No effect was found when the prime was presented in the right visual field or when the prime-probe asynchrony was 800 ms. Hence, this study illustrates an interaction of externally and internally located influences of emotion, because the stimuli themselves were emotion-laden, but the priming also induced some emotional processing in the observers and affected their reactions to the probes. The authors concluded that this result indicates that even complex emotional stimuli can be assessed early by covert attention processes, and that a dominance of the right hemisphere is involved. Other studies scrutinized whether emotional stimuli embedded in a bulk of distractors preferentially attract attention. That is, do emotion-laden stimuli pop out in the visual field and attract attention faster than neutral stimuli do? In a series of experiments by Eastwood et al. ([@B25]), participants had to detect either a unique positive or negative target face while observing displays of several schematic faces. All distractors were neutral faces, whereby the display positions of faces and the number of distractors were randomly varied in each trial. Subjects had to press a key once the target face was localized while maintaining high accuracy. Once the key was pressed, all faces were covered with gray squares and subjects had to indicate the square that corresponded to the location of the target face. Negative faces were detected faster than positive faces, and hence the results suggest a differential guidance of focal attention. To rule out that this effect was derived from the difference in physical image properties between positive and negative faces instead from their valence differences, Eastwood et al. conducted a further control experiment in which all faces were rotated by 180°. This change in face orientation led to a disappearance of differences in reaction time between the positive and negative face condition. Hence, the higher effectiveness with which faces showing negative expressions guided focal attention actually was based on the specific valence, i.e., a holistic processing of the face configuration, and not on local low-level image properties. This study suggests that externally located emotional influences preferentially attract humans\' attention, whereby the valence of the stimulus seems to play a crucial role. However, results are mixed in this regard. When embedded in an array of neutral faces, subjects locate happy or surprised faces faster and with higher accuracy than they do angry or fearful faces (Calvo and Nummenmaa, [@B13]). Moreover, a second experiment revealed that subjects fixated on and localized happy faces more often first and earlier than they did any other face. Faces were compared on several physical, i.e., basic image, properties such as luminance, contrast density, and global energy, as well as color and texture similarity. This analysis revealed that happy faces are more visually salient than the other emotional faces. In a third experiment, the procedure was similar to the first experiment, but all faces were inverted instead of having an upright orientation. This led to a disruption of the holistic configuration but left local facial properties unaltered. The setup with inverted faces replicated the pattern of differences in search time and accuracy found before. Furthermore, the search time and accuracy were worse for inverted faces in general, but the inversion only had a significant influence on the detection of sad, fearful, and angry expressions. Happy, surprised, and disgusted faces were not significantly affected. Hence, these results are consistent with a featural, rather than a configural, explanation. This advantage for happy face detection was attributed to the perception of single features rather than to the emotional meaning of faces. In further studies, Calvo and Nummenmaa ([@B13]) were able to identify which facial features are responsible for this advantage. Their results suggested that the smiling mouth is responsible for facilitating the initial orientation to happy faces. In contrast, a study by Fox et al. ([@B34]) observed that people detected a discrepant face in an array of several faces faster when it displayed angry or sad, rather than happy, expressions. Importantly, this effect disappeared when the faces were inverted or when the mouth was presented in isolation. Interestingly, in a follow-up study, Fox and Damjanovic ([@B33]) showed that the eye region alone can produce this threat of superiority effect in search tasks. The degree of the superiority effect did not increase with whole-face stimuli in contrast to eye-only stimuli. Therefore, the authors concluded that eye configurations are a key signal of threat, whereas Calvo and Nummenmaa ([@B13]) found the smiling mouth to be responsible for the superiority effect of happy faces. Obviously, much more research is necessary to uncover the principles of the superiority effects of emotional facial expressions. Given that emotional faces attract attention more than neutral faces do, researchers have begun to locate brain regions linked to this perceptual phenomenon. Vuilleumier ([@B113]) pointed out that visual responses in the fusiform cortex were enhanced for emotional faces, consistent with their greater perceptual saliency. He also referred to data from event-related evoked potentials and neurophysiology that suggested rapid parallel processing of emotional information of sensory inputs. Moreover, processing of emotional information led to a more detailed perceptual analysis of the sensory inputs and hence biased competition for attention toward the representation of emotionally salient stimuli. However, it is controversial to what extent emotion-laden stimuli are processed without attention. In general, the model of attention proposed by Desimone and Duncan ([@B24]) suggests a competition of stimuli for neural resources. Spatially directed attention can bias this competition among multiple stimuli and enhance the impact of the attended stimulus. For example, Reynolds et al. ([@B98]) have investigated the impact of focused attention on neurons in macaque areas V2 and V4 (see also Reynolds and Heeger, [@B99]). Presenting an additional stimulus in the receptive field reduced the neuronal response to the primary stimulus. However, when the monkey\'s attention was directed to one stimulus, the suppressive impact of the second stimulus was reduced. The reduction of the suppressive influence of such distractors has also been shown in humans by fMRI studies (Kastner et al., [@B52]). Hence, it appears that processing outside the overt focus of attention is weakened or even eliminated under certain conditions (Pessoa and Ungerleider, [@B85]). In contrast to these findings, several studies suggest that the role of attention is different with respect to the processing of emotion-laden stimuli. For example, Whalen et al. ([@B116]) investigated whether the amygdala is activated when human subjects observe masked emotional stimuli. They subliminally presented fearful or happy faces, followed immediately by a neutral face serving as a backward mask. Although subjects reported conscious perception of neutral faces only, the BOLD fMRI signal in the amygdala was higher during the observation of masked fearful faces in contrast to masked happy faces. Pessoa and Ungerleider ([@B85]) interpret these data as showing that the amygdala is specialized for the fast detection of emotionally relevant stimuli in the environment, whereby this detection process can occur even without attention. In contrast to this view, they found that such a differential response of the amygdala to emotional faces significantly depends on the extent to which attentional resources are available for face processing: Participants were confronted with a foveally presented fearful, happy, or neutral face and additional bars located in the left and right periphery (Pessoa et al., [@B84]). Subjects always fixated on the face during the short presentation. In one condition, they had to direct their attention to the face in order to indicate whether the face was male or female. In a second condition, attention was directed to the bars in order to indicate whether they were of the same orientation. This demanding bar orientation task was intended to attract all attentional resources. The results showed that attended faces in contrast to unattended faces elicited greater activations bilaterally in the amygdala for all facial expressions. The authors concluded that contrary to the prevailing view, emotional faces are not immune to the effect of attention. In summary, evidence for an attention-free processing of emotional items is mixed, and future research is mandatory. The reversed way: attention can affect emotional responses ========================================================== Given the extensive evidence suggesting that externally located emotional stimuli as well as internally located affective states determine how visual attention is allocated, the question arises whether this causal effect is also working in the reverse direction. In the following, we discuss an interesting way in which attention can influence emotional perception. In the study by Raymond et al. ([@B93]), participants initially saw two schematic faces showing the same affective expression (positive, negative, or neutral). The faces were placed on the left and on the right of a fixation cross. Both faces were replaced by two complex but meaningless visual patterns differing in structure. At the beginning of each experimental block, one of these two visual patterns was designated as the target and participants were told to locate the target pattern as fast as possible by pressing a corresponding key after stimulus onset. After participants had responded, one of the two visual patterns previously presented occurred in isolation. Now, subjects had to judge whether this meaningless visual pattern appeared to be cheery or dreary. The results showed that the previously attended visual target pattern was judged more positively than the ignored distractor, which in turn was judged more negatively than a novel visual pattern. Hence, not the act of attending, but the active ignoring of the distractor led to its affective devaluation. Obviously, attentional inhibition of task-irrelevant stimuli (distractors) leads to a devaluation when the previously ignored stimulus is encountered again (Fenske and Raymond, [@B31]). In further studies, Raymond et al. ([@B94]) investigated evaluations of Mondrian-like patterns. Participants had to search for a target among a varying number of distractors that appeared simultaneously. Immediately afterwards, participants rated the affective tone of either a prior target or a prior distractor. Relative to the ratings of the prior targets, the magnitude of the distractor devaluation was greater when the distractors were presented at their original screen locations during evaluation, in contrast to the case when the distractors appeared at different locations. Moreover, distractor devaluation was generally enhanced when the distractor was located near the target, and it was attenuated when the target was further away. As shown by Mounts ([@B77]), the potential of distractors to interfere with the current performance is greater when a distractor is near the task-relevant target location. Consequently, to prevent reduced performance, the inhibition of distractors presumably increases with higher proximity to the target. In this context, Fenske and Raymond ([@B31]) interpret the finding that the proximity of a distractor to a target strengthens the devaluation of the distractor in terms of a positive correlation between the amount of inhibition and the amount of distractor devaluation. From our point of view, the attention\'s influence on emotion formation suggested by these findings should motivate further research in this direction. If bottom-driven attention truly constituted emotional ratings or even the emotional state of the observer, it would and should be a high priority to consider this causal effect when investigating subjective data as well as psychophysiological data. Interaction between emotional and attentional processes on the neuronal level ============================================================================= Emotional as well as attentional processes have been attributed to dynamic processes in large-scale networks. With respect to the neuronal level, the central question is, where do these two processes interact? Numerous authors have addressed these processes individually (see Bisley, [@B9]; Lindquist et al., [@B71] for reviews on the neuronal basis of attention and emotion respectively). The interplay between them has been neglected for a long time, as pointed out by several authors (e.g., Carretié et al., [@B14]; Taylor and Fragopanagos, [@B109]), but now has moved into the focus of current research. One functional pathway can be outlined against the background of the original model of attention postulated by Posner and Petersen ([@B91]). The posterior attentional system involved in orienting processes receives strong innervation by pathways of the norepinephrine system arising in the locus ceruleus. The norepinephrine innervation is primarily present in the posterior parietal cortex, the superior colliculus, and the thamalic pulvinar. Hence, it seems to play a crucial role in maintaining alertness. The amygdala, in contrast, is a key structure of the emotional system and has a key role in regulating arousal (Williams et al., [@B117]). Furthermore, it is functionally related to both above-mentioned dimensions of emotion, namely valence and arousal. For an interaction between valence and arousal in the amygdala, see Garavan et al. ([@B38]). Sterpenich and colleagues ([@B106]) found the locus ceruleus to respond during correct recognition of neutral events encoded in an emotional context, whereby this response linearly depends on the autonomic arousal. Furthermore, the LC and the amygdala are functionally more connected during correct face recognition in emotional contexts than in neutral contexts. Obviously, the amygdala modulates emotional memories not only by affecting hippocampal activity, but also indirectly by influencing the central arousal. Perhaps the interplay of the locus ceruleus and the amygdala is a key link for the attention-emotion interaction. With respect to the ventral frontoparietal network in the model of Corbetta and Shulman ([@B19]), a link to emotion was also recently found: the ventral orienting network includes the inferior frontal cortices, which probably process emotional communicative information based on visual or auditory input (Nakamura et al., [@B79]). In line with this, Lim et al. ([@B70]) combined the attentional blink paradigm with aversive conditioning and found evidence that the impact of the amygdala on visual cortical responses was partially mediated by the frontoparietal attention network. This result suggests that the fate of a visual stimulus during competitive interactions is determined by its affective significance, affectively significant stimuli being favored. A further hint for emotion--attention network interactions resulted from a visual oddball task combined with fMRI measurement (Fichtenholtz et al., [@B32]): Participants had to detect rare target stimuli embedded sequentially in a stream of standard stimuli. Fichtenholtz et al. chose squares as the standard stimulus, occurring in 84.4% of trials. Circles, neutral complex scenes, or aversive complex scenes occurred at low rates. One group of subjects counted circle images; a second group counted aversive stimuli. The experiment revealed a higher activation in the amygdala and ventral frontotemporal cortices for emotional rather than for neutral scenes. Furthermore, a main effect of the attention focus in dorsal frontoparietal cortices was observed in terms of a greater activity for task-relevant target images irrespective of emotional content. Importantly, an interaction between the emotional and attentional focus was exclusively located in the anterior cingulated gyrus. When circles were task-relevant, the response of the anterior cingulated gyrus was equal to the response to circle targets and emotional scenes serving as distractors in this group. However, the activity of anterior cingulated gyrus increased when the emotional scenes were task-relevant, but decreased for distracting circles. These results highlight the integrative role of the anterior cingulated gyrus in the context of emotion--attention interactions. Finally, Lane et al. ([@B63]) used the PET technique to measure brain activity while male participants observed emotion-laden IAPS images controlled for valence and arousal. Participants simultaneously performed a distractor task in which they had to respond to auditory tones during image observation. A lowly and a highly distracting task were employed to manipulate the extent to which attention was absorbed by the secondary task. The results showed that the activation of the extrastriate visual cortices and the anterior temporal areas was independent of the emotional valence and arousal of stimuli and independent of the attentional resources. In contrast, there was an overlap in activation patterns associated with emotional arousal, emotional valence, and attention in extrastriate visual cortex centered in the right Brodmann areas 18 and 38 (the right anterior temporal cortex). These results suggest a common modulation of visual processing by emotion and attention at very early as well as late stages of cortical analysis. Some authors devoted to this interaction provide first literature reviews (e.g., Compton, [@B17]; Schupp et al., [@B102]) and develop a model of emotion--attention interaction (Taylor and Fragopanagos, [@B109]). Besides these important works, future research and theory formation would benefit from considering several factors that potentially affect results: First, it seems useful to envision the analogy between the general effects of attention and emotion on the behavioral level (c.f. Lane et al., [@B63]). As outlined above, in terms of the spotlight metaphor, attention enhances the processing of specific target stimuli by reducing the cognitive resources allocated to alternative stimuli. A similar effect can be observed regarding emotion: as Kensinger ([@B53]) pointed out, emotional stimuli are more likely to be processed when attentional resources are limited, suggesting a prioritized processing of emotional information. This conclusion is supported, for example, by the findings of Anderson and Phelps ([@B3]). They used the attentional blink design and found the second target was missed fewer times when it was a word of aversive content in contrast to a neutral word. Moreover, in comparison to neutral scenes, images depicting unpleasant or pleasant content are associated with increased early posterior negativity, late positive potential, and sustained positive slow wave (Schupp et al., [@B102]). The observation of emotion-laden stimuli also enhances the activity in visual brain regions that are associated with object recognition, such as inferotemporal cortices and the fusiform (Sabatinelli et al., [@B101]). Interestingly, patients with damage in the amygdala do not show a modulation of attention by emotional arousal (Kensinger, [@B53]). Correspondingly, Compton ([@B17]) emphasizes a two-stage process in the context of emotion--attention interaction: in the first stage, a subcortical circuit involving the amygdala evaluates pre-attentively the emotional significance of a stimulus. In the second stage, stimuli tagged as emotionally significant are prioritized in the competition for access to selective attention.To conclude, for research on effects of selective attention and biased competition, it appears fruitful to explicitly consider the modulating effect of arousal. Effects found in non-arousing situations change when emotions come into play. Correspondingly, Mather and Sutherland ([@B74]) provide evidence that emotional arousal amplifies salience based on perceptual contrast (bottom-up) as well as the top-down competitive advantage of high- vs. low-priority stimuli. Arousal stimulates the amygdala which, in turn, modulates sensory processing, frontoparietal attentional networks, and memory consolidation. Accordingly, the authors labeled the impact of emotion as "arousal-biased competition" in perception. Consequently, the usage of emotion-laden stimuli, for example, will change the neuronal areas of significant activity compared to emotionally neutral conditions.As outlined above, the distinction between arousal and valence is useful in the context of emotions. Hence the question arises whether emotional valence also affects biased competition. In short, previous studies found arousal more important than valence in biased competition, but positive and negative valence have opposite effects on selective attention, and they additionally interact with the arousal level (for details, see Mather and Sutherland, [@B74]). Moreover, different types of valence are associated with different brain regions (Kensinger and Schacter, [@B55]).The idea that emotional experiences can be assigned to discrete categories such as fear, anger, or happiness could be an obstacle. On the neuronal level, the search for unique signatures of discrete emotions did not produce tangible results. Specifically, co-occurring sets of neuronal features differentiating between such categories were not found (Barrett, [@B6]). Perhaps it is more expedient to assume more basic psychological processes not being directly associated with emotion, but combined in various ways to produce different emotional and affective states (see Kober et al., [@B58]).The kind of attentional process that is addressed by the experimental design significantly influences the potential brain areas that came into question to mediate emotion--attention interactions. Consequently, the kind of attentional process should be specified as precisely as possible because of the heterogeneity of labels and understandings as outlined above. With respect to overt and covert attention, for example, Bisley ([@B9]) provided a recent review focusing on associated brain networks.Future research would benefit from the distinction between an internally located vs. externally located emotional impacts on attention. It is important to differentiate between emotional states within subjects and external stimulation with emotion-laden stimuli as demonstrated nicely in a study by Lee and colleagues ([@B68]) participants observed ambiguous visual stimuli that consisted of emotional faces embedded in different levels of visual white noise. Interestingly, when participants thought to look at an emotional face while they actually observed a neutral face, the type of enhanced activity in posterior visual regions was identical with the activity found during the perception of emotional faces. The impact of personality traits on attention ============================================= Investigations of attentional processes commonly focus on universal mechanisms. Despite ever-present inter-individual variance, the view prevails that attentional processes are comparable between individuals. Given this point of view, sampling from a fictive population of human subjects is an easy step, since it should not significantly matter which subjects constitute the resulting sample. In the end, all values in the dependent variables are averaged across subjects, and an effect of the treatment is eventually observable. However, inter-subject variance may be large, and sometimes subgroups can be identified that differ noticeably from each other but are homogenous within. For example, Pessoa et al. ([@B83]) used a design which leaned on the above-mentioned study by Whalen et al. ([@B116]) who showed that the amygdala responded differentially to masked fearful and happy faces. In contrast, Phillips et al. ([@B88]) did not find any differential response to masked faces in the amygdala, though they also presented emotional target faces for only 30 ms before masking it by a neutral face. Pessoa et al. ([@B83]) showed that no universal objective awareness threshold exists for fear perception. They varied the duration for which a target face was presented before immediately masking it by a neutral face. A substantial percentage of their participants (64%) were able to reliably detect target faces presented for only 33 ms, and some participants even detected fearful faces that were presented for only 17 ms. Obviously, individuals significantly differ in their sensitivity to emotion-laden stimuli, which is perhaps based on individual differences in the general threat sensitivity (Etkin et al., [@B28]). The field of differential psychology explicitly addresses such inter-individual differences. However, even here only few studies make an effort to define a real target population with a representative sample. Instead, including many of our own eye-tracking studies, participants are mostly recruited from local educational programs with multiple resulting biases in the sample composition. Importantly, we do not refer to inter-individual differences induced by context-dependent fluctuations of the physical, mental, or emotional constitution of individuals or plain measurement noise. Rather, we here discuss time- and context-invariant personality traits. Indeed, a whole battery of tests is available to address personality traits in psychological diagnostics. The d2 test, for example, (Brickenkamp, [@B10]) assesses individual differences in attention ability and identifies attentional deficits. However, the two fields of differential psychology and physiological/psychophysical studies of attention and emotion are largely separated. Only a few studies of visual attention control for inter-individual differences in general attention ability in order to eliminate potential confounding variables (e.g., Hamborg et al., [@B41]). In the following, we want to illustrate how manifold the impact of time-invariant personality traits on attention can be. Eizenman et al. ([@B26]) presented slides containing four images. Each of the four images fell into another thematic category: neutral, loss and sadness, threat and anxiety, interpersonal attachment, and social contact. Clinical subjects with depressive disorder and a nonclinical control group observed these slides for an extended period of time. Simultaneous recording of eye movements revealed that the clinical subjects spent more time looking at images depicting loss or sadness than the control group did. Furthermore, mean fixation duration on these images was larger for subjects with depression disorder. Moreover, fixation times of both sample groups correlated with the valence ratings. The authors concluded that individuals with depressive disorder selectively attend to mood-congruent visual material. Consequently, an elaborative processing of mood-incongruent stimuli is prevented when they are simultaneously presented with dysphoric images. Attention research has also focused on the eye movement performance of patients with schizophrenia. They were found to have abnormal smooth-pursuit eye movements; that is, they were not able to adequately follow a moving object with their eyes. This dysfunction in schizophrenia patients was frequently replicated in samples from all over the world, and it was even shown for first-degree biological relatives of schizophrenia patients (Levy et al., [@B69]). Several studies focus on the influence of nonclinical personality traits on humans\' visual attention. Friesen and Kingstone ([@B37]) showed that normal subjects asked to detect, localize, or identify a target letter on the left or right of a centrally presented face were faster when the gaze of the face was toward, versus away from, the target. This was the case even though faces\' gaze direction had no predictive value with respect to the target location and subjects were told so. Based on this finding, Mathews et al. ([@B75]) investigated whether faces showing fearful expressions enhance the effect of another\'s gaze in directing the attention of an observer and whether this effect correlates with the trait scale of the State-Trait Anxiety Inventory (STAI) (Spielberger et al., [@B105]). For that purpose, Mathews and colleagues presented photographs of four men and four women showing either a neutral facial expression or a fearful one. In the original, the eyes of all eight persons looked straight ahead. In two further versions of the photographs, the pupils were moved either to the far left or right corner of both eyes. Each experimental trial started with the original version of a photograph presented in the middle of the screen. It either remained there or was subsequently replaced by the same photograph with eyes shifted left or right. After a varying time interval, one target letter appeared on the left or on the right side of the face, whereby the gaze direction was congruent, neutral, or incongruent to the location of the target letter. Participants had to respond as fast as possible by pressing the corresponding key. Response latencies and errors served as dependent variables and were analyzed by additionally considering inter-individual differences in trait anxiety. Results showed that attention is more likely to be guided by the direction of fearful rather than neutral gaze, but only in anxiety-prone individuals. Overall, faster responses were found in congruent than in incongruent trials, and this congruency effect was similar in both groups with respect to neutral faces. However, when fearful faces were shown, the congruency effect was significantly larger in the group of highly anxious persons in contrast to lowly anxious ones. It can be concluded that the attention focus of highly anxious individuals is more affected by faces showing fearful facial expressions indicating some danger in the environment, whereas lowly anxious persons do not orientate their attention toward the gaze direction of fearful others. Interestingly, in contrast to the highly anxious group, the lowly anxious group made faster responses to neutral faces in congruent as well as incongruent trials. This result pattern suggests a simple argument: internally located fear hampers fast reaction. This slowing of disengaging one\'s attention from a visual cue (i.e., the gaze direction of the faces) shows the complexity of the process. Indeed, a study on the physiology of such disengagement of attention by Khayat et al. ([@B56]) showed that monkeys\' attention can be rapidly allocated to newly appearing objects before their attention disengages from a previously attended object. In a meta-analysis, Bar-Haim et al. ([@B3a]) analyzed a total of 172 studies including 2263 anxious and 1768 non-anxious subjects to reveal the boundary conditions of the threat-related attentional bias, that is, a higher sensitivity to threat-related stimuli than to neutral stimuli in highly anxious (but not in lowly anxious) persons. They found across all studies that a significant threat-related bias was present in highly anxious subjects but not in lowly anxious subjects. Moreover, this bias did not depend on the experimental paradigm and was found under varying experimental conditions. Furthermore, the size of the threat-related bias is comparable between anxious children and anxious adults. The mean effect size is *d* = 0.45, indicating a midsize effect (Cohen, [@B16]). Perlman et al. ([@B82]) conducted an eye tracking experiment to investigate visual scanpath characteristics evoked by emotional facial expressions. In this context, they especially focused on individual differences in personality. Participants had to freely observe prototypic emotional facial expressions, including happy, angry, fearful, sad, surprised, disgusted, and neutral expressions. No further task was implemented. Before the eye tracking session, subjects had to fill out the Neo Five-Factor Inventory to assess the "big" dimensions of personality, namely extraversion, neuroticism, agreeableness, openness, and conscientiousness. Results revealed that the amount of time spent looking at the eyes of fearful faces was positively related to neuroticism. Hence, and in accordance with the above-described findings of Eizenman et al. ([@B26]), a trait congruency model is supported; that is, individuals seem to search for information that is congruent to their personality traits and avoid incongruent material. Moreover, Rauthmann and colleagues ([@B92]) also measured the big five personality traits as well as participants\' motivation in terms of behavioral inhibition and activation. Participants observed abstract images while their eye movements were recorded. The authors demonstrated with linear mixed models that neuroticism, extraversion, openness, and the behavioral activation system predicted the signature of eye movement parameters, namely the number of fixations, the mean duration of fixations, and the dwelling time within specific areas of interest. However, at this point we should emphasize that the obvious impact of certain personality traits on attention processes is often more complex and not as obvious as it seems against the background of the above described studies. We know from differential psychology that a certain behavior is always the result of the interplay between personal dispositions such as needs or personality traits and current situational conditions; these can interact in complex ways (Heckhausen and Heckhausen, [@B43]). Hence, it is probably insufficient to consider only potential main effects of personality traits on attention processes. Rather, it could be fruitful to elucidate the specific interactions between current situational factors and time-invariant personality traits to reveal further interesting effects. In a recent study Kaspar and König ([@B50]), we investigated changes in eye movement parameters during complex scenes repeatedly presented in a long sequence of stimuli. In this context, we also considered the subjects\' global interest in the stimulus set, as well as their general ability to stay within interesting activities without shifting prematurely to alternative activities labeled as the personality trait "action orientation regarding the performance of activities" (AOP; Kuhl, [@B61]). We expected that this ability only affects viewing behavior when the stimulus material becomes successively more familiar to the observer and that it is related to its interestingness. In fact, we found that the attention focus became more and more locally expressed by several saccade parameters, fixation distributions on the level of single subjects, and an increasing inter-subject variance of fixation distributions across repeated image observations. Importantly, this general tendency was weaker for the group of subjects who rated the image set as interesting, compared to those subjects who were not interested in the images. Moreover, the effects were partly mediated by subjects\' ability to stay within interesting activities without shifting prematurely to alternative activities. Besides the role of personality traits in terms of motivational or behavioral tendencies, demographic variables were also found to be influential factors on attention: Feng et al. ([@B30]) provided evidence for gender differences in spatial selective attention. In their study, they used the useful-field-of-view task to measure spatial cognition. This paradigm allows measuring the ability to detect, localize, and identify a target stimuli as well as the spatial distribution of attentional resources over the field of view. As far as subjects having very little video game experience, men performed better than women. The difference diminished when video game experience was high. Interestingly, in a second study, they were able to show that 10 h of training with an action video game improved performance measurements for spatial attention, whereby women benefited more than men. This implies that balancing men and women in between-subject study designs or considering the gender ratio when interpreting results in within-subject designs is important. Moreover, potential inter-individual differences in domain-specific expertise should also be considered if these potentially influence the dependent variables. In this context, Tanaka and Curran ([@B108]) showed that very specific expertise can change EEG signatures. In their study, dog experts as well as bird experts had to categorize objects. In contrast to categorization of objects outside their domain of expertise, categorization of objects within their respective domains of expertise induced an enhanced early negative component. Consequently, it is mandatory to collect useful metadata, such as demographic variables or domain specific expertise, to appropriately control for these confounding variables. This can be done by adapting the study designs or explicitly introducing inter-individual differences as independent variables in the study design. Also, as suggested by Polyhart and Vandenberg ([@B89]) for the field of longitudinal research, it seems necessary to break out of the "take-what-we-can-get" mentality regarding sample selection in cognitive science. In line with this suggestion, we discuss that subjects\' age also can produce substantial artifacts in the results if it is not considered in sample selection or when assigning subjects to different groups in between-subjects designs. Mather and Carstensen ([@B73]) examined age differences in attention to and memory for faces. In their experiment, two faces were placed on the left and on the right side of a screen. One of these faces had an emotional expression (sadness, anger, or happiness); the other face was neutral. After both faces disappeared, one dot was shown either on the left or the right position where the faces were previously presented. The dot remained on the display until the participants pressed a corresponding key. Results showed that older adults responded faster to this dot when it was presented on the same side as a neutral face than when it was located on the same side as a negative face. This attentional bias was not found for younger adults. Moreover, older adults remembered positive faces better than faces with negative facial expressions. Mather and Carstensen concluded that older adults obviously avoided negative information and that this attentional bias was consistent with older adults\' generally better emotional well-being and their tendency to remember negative less accurately than positive information. Acik et al. ([@B1]) compared the viewing behavior of three age groups (7- to 9-year-old children, 19- to 27-year-old young adults, and older adults above 72 years) on natural and complex scenes before performing a delayed patch recognition task. Their results suggested that bottom-up mechanisms play a more important role in younger ages: eye movements of young children were heavily guided by basic image features, whereas older adults\' viewing behavior was less feature-related. In addition to this differential effect of age on feature-fixation correlations, the explorativeness---that is, the spread of fixation distributions---correlated with feature-related viewing negatively in younger ages. In contrast, older adults increased their feature-related viewing behavior by being more explorative, leading to a better performance in the subsequent patch recognition task. Consequently, it seems that basic image features lose their impact on attention guidance as age increases, and this is paralleled by a stronger impact of top-down processes. To conclude, we suggest that future studies on humans\' attention would benefit from experimental designs that are appropriate for revealing personality influences on viewing behavior. For that purpose, study designs have to be ecologically valid, since very artificial experimental paradigms can cap the impact of personality differences. In this context, a pre-experimental definition of the target population and a subsequent sampling of an appropriate (i.e., representative) sample should be considered. Conflict of interest statement ------------------------------ The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Kai Kaspar\'s work was supported by ERC-2010-AdG \#269716 - MULTISENSE. Peter König\'s work was supported by Cognition and Neuroergonomics/Collaborative Technology Alliance \#W911NF-10-2-0022. [^1]: Edited by: John J. Foxe, Albert Einstein College of Medicine, USA [^2]: Reviewed by: Alex R. Wade, Smith Kettlewell Eye Research Institute, USA; Redmond O\'Connell, Trinity College Dublin, Ireland

Background ========== Cerebral vasculitis is a serious but uncommon condition which presents considerable difficulties in recognition, diagnosis and treatment. Moyamoya disease is a progressive occlusive disease of the cerebral vasculature with particular involvement of the circle of Willis. Aim === To describe a case of moyamoya initial misinterpreted as cerebral vasculitis. Methods ======= A three-year-old girl, with a former history of Kawasaki disease, presented with a stroke. Both parents were of Korean heritage. Stenosis of the middle cerebral artery on both sides was seen on MR angiogram. No abnormal blood tests were found. Treatment of suspected cerebral vasculitis was initiated with steroids orally (2mg/kg/d). She developed recurrent transient ischemic attacks in both hemispheres and new strokes despite series of pulse therapy of steroids and Cyclophosphamide and finally Mycophenolic acid in combination with prednisolone daily. Stenosis of the anterior cerebral arteries developed and progressed and the middle cerebral arteries worsened despite treatment. Only minor changes were seen in the internal carotid arteries. Abnormal vascular collateral networks were only limited present. Moyamoya was suspected and confirmed by conventional angiography and the child underwent surgical treatment. Superficial temporal artery to middle cerebral artery bypass ((STA-MCA-Bypass) and encephalomyosynangiosis was performed bilaterally. She recovered without any mental decline and is able to walk despite her hemi paresis. Conclusion ========== Moyamoya is a rare disease affecting the cerebral arteries, and an important differential diagnosis to cerebral vasculitis. Especially in the early course of moyamoya the radiological findings might be difficult to interpret.

Introduction {#Sec1} ============ Effective asthma self-management behaviour (SMB)---e.g., symptom prevention or acute symptom management \[[@CR1]\]---can improve the control of asthma and its prognosis \[[@CR2]\]. The actual implementation of SMB is partially determined by contextual factors \[[@CR1]\], e.g., by the working conditions employees with asthma are exposed to \[[@CR3]\]. A qualitative interview study among employees with asthma suggested that two specific elements of well-established work stress models play a crucial role in the effective implementation of asthma SMB at work \[[@CR3]\]: job decision latitude (JDL; i.e., the control over one's tasks and when and how to complete them \[[@CR4], [@CR5]\]) and social support by colleagues and line-managers \[[@CR6]--[@CR8]\]. The two factors subjectively related to different domains of asthma SMB: high JDL seemed to promote symptom prevention, symptom monitoring, and acute symptom management, whereas support was perceived to facilitate symptom prevention, acute symptom management and communication (e.g., self-disclosure of one's condition) \[[@CR3]\]. Low levels of JDL and social support are considered important contributors to work stress \[[@CR4], [@CR5]\], which has been related to an increased incidence and prevalence of asthma \[[@CR9]--[@CR12]\] and which may also be associated with increased asthma morbidity \[[@CR13], [@CR14]\]. The latter association may be explained by direct physiological stress responses (e.g., decreased corticosteroid sensitivity due to long-term stress exposure \[[@CR15]\]) and/or by behavioural pathways (i.e., poorer asthma self-management). In this epidemiological study, we sought to build on the findings from our prior qualitative study and investigate the associations of JDL and social support with different types of asthma SMB \[[@CR3]\]. Furthermore, we aimed to also test the relationships between the two psychosocial working conditions and asthma morbidity (i.e., asthma control and asthma-specific quality of life) and to investigate possible mediation of these associations by asthma SMB. Methods {#Sec2} ======= Study population {#Sec3} ---------------- We recruited inpatients with asthma from three pulmonary rehabilitation clinics (i.e., Median Klinik Heiligendamm, Nordseeklinik Borkum der DRV Rheinland, and Klinik Bad Reichenhall der DRV Bayern Süd). In Germany, pulmonary rehabilitation aims to stabilize patients with long-term conditions to ensure social participation in private and professional life, including workability. This treatment is mostly financed by the health insurance carrier or the German Pension Fund. Senior physicians screened the records of patients who were newly admitted or about to be admitted for an established asthma diagnosis while excluding patients with comorbid chronic obstructive pulmonary disease (COPD) to strengthen our case definition \[[@CR16]\]. The inclusion and exclusion of participants were solely based on the admission diagnoses as stated in the patients' files. There is evidence that patients with asthma and COPD behave differently concerning their disease self-management at work. In a qualitative study by Boot et al. \[[@CR17]\] who investigated different coping profiles among employees with asthma or COPD, some of the workers with COPD were reported to enter a so-called "worried adaptation profile", which was characterized by preoccupations with their health, adaptation to the experienced limitations and slower working. By contrast, employees with asthma did not display that adaptation profile, but formed, amongst others, a separate group of "eager workers", who were reported to be motivated, leading a healthy life, adjusting their medication when necessary and reluctant to take sick leave \[[@CR17]\]. Further, the authors postulated that patients with progressive diseases like COPD are more likely to suffer from adverse working conditions \[[@CR17]\]. If we had included both illnesses in our study, we would have mixed up two different processes concerning self-management at work and would have reduced the generalizability of our results to either condition. Questionnaires with stamp-addressed envelopes were sent out to the patients as soon as they were registered (Bad Reichenhall) or handed to patients personally by the senior physician in their first week of inpatient treatment (Heiligendamm and Borkum). Out of 755 questionnaires, 221 eligible questionnaires were returned between October 2017 and May 2018 (response rate = 29.3%). Patients' questionnaires were eligible for analyses when the following criteria were met:a diagnosis of and current treatment for asthma, but not COPD, according to admission diagnoses and confirmed by patient reports;employment of at least 20 working hours per week in the last 6 months;having worked with the diagnosis for at least 6 months. Criteria 2 and 3 were applied to ensure that the participants had worked for a reasonable period of time with asthma. Ethical considerations {#Sec4} ---------------------- The participants provided written informed consent before taking part in the study. Our study received ethical approval by the Institutional Review Board of the Medical Faculty of the Heinrich-Heine University of Düsseldorf (no. 5764R). Measures {#Sec5} -------- ### Psychosocial working conditions {#Sec6} We assessed the working conditions that were considered most important by the participants of our prior interview study \[[@CR3]\]---namely, JDL and social support at work. Both constructs are elements of the well-established job demand-control-support model \[[@CR4], [@CR5]\] and were measured by the German version of the Copenhagen Psychosocial Questionnaire \[[@CR18]\]. We combined the scales "Influence at work" (3 items) and "Degree of freedom at work" (2 items) into a sum-score to operationalize JDL (Cronbach's alpha: α = .78). The four-item scale "social support" (Cronbach's alpha: α = .90) was used to measure support. Sum-scores of each variable were dichotomized at the bottom tertile to define adverse working conditions. For methodological details, please see Additional file [1](#MOESM1){ref-type="media"}. We sought to assess additional asthma-specific working conditions that emerged from our prior study \[[@CR3]\]. As an appropriate instrument was lacking, we constructed a novel scale: the "Determinants of work-related asthma self-management (DReAM)" scale. The items were devised based on our qualitative findings and comprised statements concerning asthma-specific working conditions (e.g., "In case of acute asthma symptoms, I promptly take breaks in my everyday work"). They were optimized by cognitive interviews and further reduced in psychometric analyses (see Additional file [1](#MOESM1){ref-type="media"} for details). The final scale comprised seven items all of which loaded on a single factor (Cronbach's alpha = .83). For analyses, the DReAM-items (shown in Table [1](#Tab1){ref-type="table"}) were combined into a sum-score which was dichotomized at the top tertile of the score distribution to define adverse working conditions.Table 1"Determinants of work-related asthma self-management (DReAM)" items (n = 221)ItemParticipants reporting adverse working conditionsn%^a^In my everyday work, I am able to avoid allergens or other triggers that are relevant to me16374.1Since my condition is invisible to others, I do not receive any support in managing my asthma in my everyday work14867.6In my everyday work, I can take a break anytime if necessary for my asthma management13662.1In my everyday work, I can go home early or stay home for one day without a sick note whenever I need to manage my asthma12758.3In my everyday work, I can manage my asthma the way I need to because I can withdraw from others for my asthma management12557.3In my everyday work, I am NOT able to manage my asthma as I need to, because there is nobody to take over my tasks10648.2I feel accepted by my line-manager concerning my asthma and the related limitations6228.8^a^of eligible cases ### Asthma self-management {#Sec7} Based on a model of four domains of asthma SMB \[[@CR1]\], we aimed to measure symptom prevention, acute symptom management, and communication. The fourth domain (i.e., symptom monitoring) did not seem to be related to work (e.g., peak flow metre use) \[[@CR3]\]. The domain symptom prevention was measured by items on trigger avoidance and physical activity. Items assessing acute symptom management covered, for instance, reliever medication use, breathing techniques, or taking breaks. Communication mainly referred to self-disclosure. We used an item of the Survey of Health, Ageing and Retirement in Europe to assess physical activity \[[@CR19], [@CR20]\]: "How often do you engage in vigorous physical activity, such as sports, heavy housework, or a job that involves physical labour?". We defined adequate physical activity as "more than once a week" (vs. once a week, one to three times a month, hardly ever, or never) which represents a level of physical activity that is beneficial to asthma morbidity \[[@CR19], [@CR20]\]. To assess further domains of asthma SMBs \[[@CR1]\], we devised items based on our prior qualitative work \[[@CR3]\]. We used a response format which had been applied before and had been assessed as useful by asthma patients \[[@CR21]\]: "Yes, I do this", "No, but I would like to", and "No, I do not need this". Based on the results of cognitive interviews, we reduced the item pool to its final set of ten items covering trigger avoidance (1 item), acute symptom management (7 items), and communication (2 items). To dichotomize the answers, we defined "No, but I would like to" as an unfulfilled need (scored as 1) and "Yes, I do this" or "No, I do not need this" as a fit of need and working conditions (scored as 0). Also the response option "Item does not apply to me" was scored as 0. For more information on the subscales and the strategy of dichotomization, please see Additional file [1](#MOESM1){ref-type="media"}. ### Asthma morbidity {#Sec8} To assess asthma control, we used the Asthma Control Test \[[@CR22]\]. The resulting sum-score was categorized into uncontrolled (\< 20 points) versus controlled (≥ 20 points) \[[@CR21], [@CR23]\]. We measured asthma-specific quality of life using the Marks Asthma Quality of Life Questionnaire \[[@CR24]\]. The total score was dichotomized at its top tertile to indicate impaired quality of life \[[@CR21]\]. For methodological details, see Additional file [1](#MOESM1){ref-type="media"}. Data analysis {#Sec9} ------------- We used SPSS 25 to conduct logistic regression analyses. The reference categories were high JDL, high support and good working conditions according to the DReAM-scale, respectively. The dependent variables were reported physical activity once a week or less, unfulfilled needs regarding trigger avoidance, acute symptom management, and communication, poor asthma control, and impaired asthma-specific quality of life according to the different dichotomization strategies reported in Additional file [1](#MOESM1){ref-type="media"}. We initially estimated unadjusted odds ratios (ORs) and 95% confidence intervals (CIs), which were subsequently adjusted for age, sex, highest educational degree, body mass index, and smoking status (never, current, former) (i.e., labelled "Model A"). To investigate potential mediating effects of different asthma SMBs on the relationships of JDL or support with asthma control or asthma-specific quality of life, we added the respective asthma SMBs to the adjusted model (i.e., Model A) and re-ran the analyses. Results {#Sec10} ======= Sample description {#Sec11} ------------------ Table [2](#Tab2){ref-type="table"} shows the characteristics of our sample (n = 221). The mean age was 50.6 years (standard deviation = 8.7 years). The gender distribution was fairly balanced (53.8% women). Over 60% of the sample had a secondary school degree or higher. More than one-third of the sample was obese. Almost two-thirds of the sample reported uncontrolled asthma.Table 2Sample characteristics (n = 221)CharacteristicAge (years) Mean (SD)50.6 (8.7) Missing values; n (%)2 (.9)Sex Women; n (%)119 (53.8) Men; n (%)102 (46.2) Missing values; n (%)--Highest school degree Lower than secondary school degree; n (%)82 (37.1) Secondary school degree; n (%)74 (33.5) Higher than secondary school degree; n (%)62 (28.1) Missing values; n (%)3 (1.4)Body mass index (BMI) Underweight, normal (BMI \< 25); n (%)61 (27.6) Overweight (25 ≤ BMI \< 30); n (%)81 (36.7) Obese (BMI ≥ 30); n (%)76 (34.4) Missing values; n (%)3 (1.4)Smoker No, never; n (%)110 (49.8) No, not anymore; n (%)96 (43.4) Current; n (%)13 (5.9) Missing values; n (%)2 (.9)Job Decision Latitude (JDL)^a^ Mean (SD)45.61 (22.83) Missing values; n (%)2 (.9)Support^a^ Mean (SD)53.65 (25.65) Missing values; n (%)12 (5.4)DReAM (working conditions)^b^ Mean (SD)18.50 (5.05) Missing values; n (%)1 (.5)Physical activity More than once a week; n (%)144 (65.2) Once a week or less; n (%)75 (33.9) Missing values; n (%)2 (.9)Trigger avoidance^c^ Mean (SD).6 (.5) Missing values; n (%)1 (.5)Acute asthma symptom management^d^ Mean (SD)2.3 (2.0) Missing values; n (%)5 (2.3)Communication^e^ Mean (SD).3 (.6) Missing values; n (%)4 (1.8)Asthma control^f^ Controlled76 (34.4) Uncontrolled145 (65.6) Missing values--Asthma-specific quality of life^g^ Mean (SD)4.4 (1.8) Missing values; n (%)1 (.5)*SD* standard deviation^a^"Measured by the Copenhagen Psychosocial Questionnaire \[[@CR37]\], range: 0--100, higher values indicating higher levels of JDL and support^b^"Determinants of work-related asthma self-management (DReAM)" measured by 7 self-constructed items, range: 7--28, lower values indicating better working conditions^c^Measured by 1 self-constructed item, range: 0 = fit of need and conditions versus 1 = need unfulfilled^d^Measured by 7 self-constructed items, range: 0 = fit of need and conditions to 7 = all needs unfulfilled^e^Measured by 2 self-constructed items, range: 0 = fit of need and conditions to 2 = both needs unfulfilled^f^Measured by the Asthma Control Test \[[@CR22]\], range: 5--25, higher values indicating higher levels of asthma control^g^Measured by the Marks Asthma Quality of Life Questionnaire \[[@CR24]\], range: 0--10, lower values indicating better quality of life Psychosocial working conditions and asthma SMBS at work {#Sec12} ------------------------------------------------------- Table [3](#Tab3){ref-type="table"} shows estimates of the relationships between working conditions and asthma SMBs. In unadjusted analyses, employees with asthma who experienced low levels of JDL did not report less physical activity than employees with asthma who experienced high levels of JDL (OR = 1.57; 95% CI 0.87--2.84), but they reported more unfulfilled needs concerning trigger avoidance (OR = 2.30; 95% CI 1.23--4.31) and acute symptom management (OR = 5.83; 95% CI 2.76--12.31). There was no evidence of relationship between JDL and communication (OR = 1.57; 95% CI 0.77--3.18). Analyses adjusted for confounders (Model A) yielded comparable results.Table 3ORs and 95% CIs for different domains of asthma SMBs according to JDL and support (n = 221)Asthma symptom preventionAcute asthma symptom management (unfulfilled needs)Communication (unfulfilled needs)Physical activity (once a week or less)Trigger avoidance (unfulfilled need)UnadjustedAdjusted Model A^a^UnadjustedAdjusted Model A^a^UnadjustedAdjusted Model A^a^UnadjustedAdjusted Model A^a^OR95% CIOR95% CIOR95% CIOR95% CIOR95% CIOR95% CIOR95% CIOR95% CIJDL HighRef.--Ref.--Ref.--Ref.--Ref.--Ref.--Ref.--Ref.-- Low1.57.87, 2.841.60.84, 3.032.301.23, 4.312.051.07, 3.945.832.76, 12.316.322.83, 14.131.57.77, 3.181.63.79, 3.39Support HighRef.--Ref.--Ref.--Ref.--Ref.--Ref.--Ref.--Ref.-- Low1.17.64, 2.131.42.74, 2.732.091.13, 3.842.251.18, 4.312.291.08, 4.831.87.85, 4.152.821.37, 5.842.951.38, 6.30DReAM GoodRef.--Ref.--Ref.--Ref.--Ref.--Ref.--Ref.--Ref.-- Adverse1.911.05, 3.462.681.37, 5.235.252.56, 10.796.542.98, 14.375.552.64, 11.694.672.14, 10.212.791.38, 5.632.871.37, 6.02*OR* odd ratio, *CI* confidence interval, *SMB* self-management behaviour, *JDL* job decision latitude, *DReAM* determinants of work-related asthma self-management^a^Model A: adjusted for age, sex, highest educational degree, body mass index, and smoker status Low levels of social support (vs. high support) were associated with more unfulfilled needs concerning trigger avoidance (OR = 2.09; 95% CI 1.13--3.84), acute symptom management (OR = 2.29; 95% CI 1.08--4.83), and communication (OR = 2.82; 95% CI 1.37--5.84). There was no evidence of a relationship between support and physical activity (OR = 1.17; 95% CI 0.64--2.13). One association was attenuated in adjusted analyses, which is between support and acute symptom management (adjusted OR = 1.87; 95% CI 0.85--4.15). Employees with asthma who reported adverse psychosocial working conditions according to the DReAM scale reported less physical activity (OR = 1.91; 95% CI 1.05--3.46) and more unfulfilled needs concerning trigger avoidance (OR = 5.25; 95% CI 2.56--10.79), acute symptom management (OR = 5.55; 95% CI 2.64--11.69), and communication (OR = 2.79; 95% CI 1.38--5.63). Adjusted analyses delivered similar results. Psychosocial working conditions and asthma morbidity {#Sec13} ---------------------------------------------------- Results regarding the relationships between working conditions and asthma control are shown in Table [4](#Tab4){ref-type="table"}. In unadjusted analyses, employees with asthma who experienced poor psychosocial working conditions in terms of JDL, support, and DReAM were more likely to report poor asthma control (OR = 2.14; 95% CI 1.12--4.10, OR = 1.95; 95% CI 1.04--3.68, and OR = 1.90; 95% CI 1.01--3.60, respectively).Table 4ORs and 95% CIs for asthma control according to JDL and support (n = 221)Poor asthma controlUnadjusted modelAdjusted Model A^a^Adjusted Model B^b^Adjusted Model A and physical activityAdjusted Model A and trigger avoidanceAdjusted Model A and acute asthma symptom managementAdjusted Model A and communicationOR95% CIOR95% CIOR95% CIOR95% CIOR95% CIOR95% CIOR95% CIJDL HighRef.--Ref.--Ref.--Ref.--Ref.--Ref.--Ref.-- Low2.141.12,4.102.211.12, 4.351.61.75, 3.442.211.12, 4.371.96.98, 3.891.68.79, 3.572.281.13, 4.59− 27.15% change vs. adjusted Model A0% change vs. adjusted Model A− 11.31% change vs. adjusted Model A− 23.89% change vs. adjusted Model A+ 3.17% change vs. adjusted Model ASupport HighRef.--Ref.--Ref.--Ref.--Ref.--Ref.--Ref.-- Low1.951.04, 3.681.941.00, 3.761.52.73, 3.181.94.999, 3.761.73.88, 3.391.74.87, 3.491.85.93, 3.68− 21.65% change vs. adjusted Model A0% change vs. adjusted Model A− 10.82% change vs. adjusted Model A− 10.31% change vs. adjusted Model A− 4.64% change vs. adjusted Model ADReAM GoodRef.--Ref.--Ref.--Ref.--Ref.--Ref.--Ref.-- Adverse1.901.01, 3.601.86.95, 3.661.16.53, 2.551.89.95, 3.751.49.72, 3.061.34.65, 2.781.79.89, 3.59− 37.63% change vs. adjusted Model A+ 1.61% change vs. adjusted Model A− 19.89% change vs. adjusted Model A− 27.96% change vs. adjusted Model A− 3.76% change vs. adjusted Model A*OR* odd ratio, *CI* confidence interval, *JDL* job decision latitude, *DReAM* determinants of work-related asthma self-management^a^Model A: adjusted for age, sex, highest educational degree, body mass index, and smoker status^b^Model B: adjusted for age, sex, highest educational degree, body mass index, smoker status, physical activity, trigger avoidance, acute asthma symptom management, and communication Analyses adjusted according to Model A yielded comparable results. After additional adjustment for all SMBs (Model B), all associations were attenuated as compared to Model A (ORs dropped by 27.2%, 21.7%, and 37.6%, respectively). Adjustment for single SMBs showed that this attenuation was most pronounced after adjustment for acute symptom management (− 23.9%, − 10.3%, and − 28.0%, respectively) and for trigger avoidance (− 11.3%, − 10.8%, and − 19.9%, respectively). Table [5](#Tab5){ref-type="table"} displays the results regarding the relationships between working conditions and asthma-specific quality of life. In unadjusted analyses, employees with asthma who experienced poor psychosocial working conditions in terms of JDL, support, and DReAM reported impaired asthma-specific quality of life (OR = 2.05; 95% CI 1.13--3.72, OR = 2.56; 95% CI 1.40--4.67, and OR = 2.33; 95% CI 1.28--4.23, respectively) compared to employees with asthma who experienced good working conditions.Table 5ORs and 95% CIs for asthma-specific quality of life according to JDL and support (n = 221)Low asthma-specific quality of lifeUnadjusted modelAdjusted Model A^a^Adjusted Model B^b^Adjusted Model A and physical activityAdjusted Model A and trigger avoidanceAdjusted Model A and acute asthma symptom managementAdjusted Model A and communicationOR95% CIOR95% CIOR95% CIOR95% CIOR95% CIOR95% CIOR95% CIJDL HighRef.--Ref.--Ref.--Ref.--Ref.--Ref.--Ref.-- Low2.051.13, 3.721.77.95, 3.331.08.52, 2.221.70.90, 3.201.65.87, 3.131.15.57, 2.341.72.91, 3.26− 38.98% change vs. adjusted Model A− 3.95% change vs. adjusted Model A− 6.78% change vs. adjusted Model A− 35.03% change vs. adjusted Model A− 2.82% change vs. adjusted Model ASupport HighRef.--Ref.--Ref.--Ref.--Ref.--Ref.--Ref.-- Low2.561.40, 4.672.261.19, 4.291.90.93, 3.862.191.15, 4.182.121.10, 4.072.021.02, 4.012.091.08, 4.03− 15.93% change vs. adjusted Model A− 3.10% change vs. adjusted Model A− 6.19% change vs. adjusted Model A− 10.62% change vs. adjusted Model A− 7.52% change vs. adjusted Model ADReAM GoodRef.--Ref.--Ref.--Ref.--Ref.--Ref.--Ref.-- Adverse2.331.28, 4.232.151.13, 4.111.35.64, 2.861.971.02, 3.822.001.01, 3.961.52.75, 3.071.981.03, 3.84− 37.21% change vs. adjusted Model A− 8.37% change vs. adjusted Model A− 6.98% change vs. adjusted Model A− 29.30% change vs. adjusted Model A− 7.91% change vs. adjusted Model A*OR* odd ratio, *CI* confidence interval, *JDL* job decision latitude, *DReAM* determinants of work-related asthma self-management^a^Model A: adjusted for age, sex, highest educational degree, body mass index, and smoker status^b^Model B: adjusted for age, sex, highest educational degree, body mass index, smoker status, physical activity, trigger avoidance, acute asthma symptom management, and communication Adjusted analyses (Model A) led to an attenuation of all ORs (OR = 1.77; 95% CI 0.95--3.33; OR = 2.26; 95% CI 1.19--4.29; and OR = 2.15; 95% CI 1.13--4.11, respectively). This attenuation was even stronger after additional adjustment for all SMBs (ORs dropped by 39.0%, 15.9%, and 37.2%, respectively), most pronouncedly so when adjusted for acute symptom management (− 35.0%, − 10.6%, and − 29.3%, respectively). Discussion {#Sec14} ========== In this study, we found that adverse psychosocial working conditions were associated with poor asthma SMB. Specifically, we observed that low JDL was related to poorer trigger avoidance and acute symptom management, but not to physical activity and communication. Further, low support was related to poorer trigger avoidance, acute symptom management, and communication, but not to physical activity. Finally, the newly created DReAM-scale, which frames working conditions specifically in the context of asthma, related to all types of asthma SMB. Our study also suggests that the above-mentioned working conditions are generally related to poorer asthma control and impaired asthma-specific quality of life. The fact that those associations were attenuated after additional adjustment for asthma SMBs indicates that asthma SMBs may provide an explanatory mechanism. The results are closely in line with the findings of our qualitative interview-based study \[[@CR3]\]. Specifically, a relationship between JDL and communication was neither suggested by our qualitative findings nor found in this study. The same holds true for an association between support and physical activity. All the expected associations could be confirmed with a single exception, that is, a relationship between JDL and physical activity. This contradicts epidemiological studies \[[@CR25]--[@CR27]\] which suggested an association between low JDL and less physical activity. Whereas those studies assessed leisure-time physical activity, the item used in our study included job-related physical activity. Since physical activity was considered as an asthma SMB in terms of symptom prevention by employees with asthma, regardless of whether performed at work or in leisure-time \[[@CR3]\], our strategy to assess overall physical activity seemed appropriate. The association between support and different SMBs--especially acute symptom management and communication--was investigated and confirmed before, but not specifically for individuals with asthma \[[@CR28]\]. Moreover, psychosocial working conditions were associated with asthma morbidity. This adds to previous results on work stress and asthma morbidity \[[@CR14]\], e.g., by using another work stress model than the effort-reward imbalance model applied in that prior study \[[@CR29]\]. The participants of our earlier qualitative study did not consider the effort-reward imbalance-components as asthma-relevant working conditions \[[@CR3]\]. We present novel insights by documenting that relationships between psychosocial working conditions and asthma morbidity may partly be mediated (and thus explained) by asthma SMB, in particular by acute symptom management and (in case of asthma control) by trigger avoidance. To our knowledge, this is the first study that statistically confirmed the associations of adverse psychosocial working conditions with poorer asthma-specific SMB at work. To date, prior research has either exclusively relied on qualitative methods \[[@CR3], [@CR30]\] or provided statistical estimations which were not specific to employees with asthma \[[@CR28]\]. If prospective studies confirm our results and the suspected temporal sequence of the observed associations, it is conceivable to develop interventions to optimize the working conditions for employees with asthma. For instance, patient education programs addressing return-to-work issues could be devised for pulmonary rehabilitation to empower employees with asthma to influence their working conditions. This could be supported by interventions that increase awareness of the importance of JDL and social support among line-managers and employers. Medical staff could support employees with asthma by explaining how to effectively implement asthma SMB at work and raising the awareness that asthma SMBs comprise more than merely trigger avoidance or taking reliever medication. Employees should especially be given the opportunity to manage their acute symptoms, because this seems to be the most problematic domain of asthma SMB (according to our qualitative study \[[@CR3]\]) and the most influential on asthma morbidity. Limitations {#Sec15} =========== First, this study is cross-sectional and therefore does not provide insights into the temporal nature of the observed associations. Second, since the study focused on chronically ill employees, the results could be affected by the healthy worker effect \[[@CR31]\]. The healthy worker effect suggests that individuals with poor health are more likely to drop out of the workforce than apparently healthy employees (for an overview, see Ref. \[[@CR32]\]). Third, patients with asthma were included and patients with COPD were excluded based on their admission diagnoses. It remains unclear what criteria were applied to establish the respective diagnoses and if and to what extent COPD was actually ruled out. Since both conditions share some characteristics, they are often confused in the diagnostic process \[[@CR16]\]. We originally aimed to check the diagnoses of our participants at discharge from the three cooperating clinics, but this proved unfeasible in the end due to data-security reasons. Therefore, it is possible that some of our participants with asthma do suffer from (comorbid) COPD. Fourth, our overall response rate was low (29.3%) \[[@CR33]\], but this was not unusually low in relation to surveys of patients with respiratory conditions \[[@CR34]\]. This is partly due to the fact that we did not send out reminders to non-responders or did not ask them repeatedly to complete the questionnaire, because this would have meant unacceptable extra work for the administrational and medical staff in the clinics. In terms of the representativeness of our study sample, it seems reassuring that the gender distribution among our participants (53.8% women) did not differ much from the gender distribution among the overall sample groups in the three clinics (55.6% women). However, the mean age was slightly lower in our sample (50.6 years vs. range of 51.3--55.4 years in the three clinics). Lamentably, we do not have any further information on the overall sample in the three clinics. Therefore, we cannot analyse further possibly influential factors such as additional demographic variables, the profession or job situation. It must be noted that the proportion of participants with uncontrolled asthma according to the Asthma Control Test seemed to be rather high in our sample (65.6%) compared to other studies, which did not specifically focus on rehabilitants (ranging from 44.7 to 59.31%) \[[@CR14], [@CR21], [@CR35], [@CR36]\]. Fifth, it is conceivable that physiological conditions at the workplace affect asthma SMB and/or psychosocial working conditions, but we specifically assessed the working conditions which were considered relevant for asthma SMB by the participants of our prior qualitative study \[[@CR3]\], e.g., trigger avoidance. Sixth, the objective measurement of SMBs, e.g., of physical activity by using accelerometers, would have delivered less biased data than self-report information. Seventh, although the DReAM-scale, which was specifically developed for this study, showed promising results, it must be noted that some of the items already implied a relationship between working conditions and asthma SMB at work (e.g., "In my daily work routine, I can NOT manage my asthma the way I need to, because there is nobody to take over my tasks"). Thus, the results concerning the relationships between the DReAM-values and the reported asthma SMBs, which were also assessed by self-constructed items, might overestimate the associations. Conclusions {#Sec16} =========== Our study found that adverse psychosocial working conditions are associated with poorer asthma SMB at work and increased asthma morbidity. The latter relationship may partly be mediated by asthma SMB. Further research---especially longitudinal studies---is now needed to gain more insights into the complex interrelationships between working conditions, asthma SMB and asthma morbidity and to develop suitable interventions. Additional file =============== {#Sec17} **Additional file 1.** Details on measures. CI : confidence interval COPD : chronic obstructive pulmonary disease DReAM : determinants of work-related asthma self-management JDL : job decision latitude OR : odds ratio SMB : self-management behaviour KH: substantial contributions to conception and design, acquisition of data, and analysis and interpretation of data; drafting the manuscript; SH: substantial contributions to acquisition of data; revising the manuscript critically for important intellectual content; JG: substantial contributions to acquisition of data; revising the manuscript critically for important intellectual content; KS: substantial contributions to acquisition of data; revising the manuscript critically for important intellectual content; JL: substantial contributions to analysis and interpretation of data; revising the manuscript critically for important intellectual content; AS: substantial contributions to analysis and interpretation of data; revising the manuscript critically for important intellectual content; AL: substantial contributions to conception and design, acquisition of data, and analysis and interpretation of data; drafting the manuscript. All authors read and approved the final manuscript. Competing interests {#FPar1} =================== The authors declare that they have no competing interests. Acknowledgements {#FPar2} ================ The authors want to thank all the participants in the cognitive interviews and the survey. Further, we appreciate Uwe Hofbauer's help and support regarding data entry. Availability of data and materials {#FPar3} ================================== The datasets used and analysed during the current study are available from the corresponding author upon reasonable request. Consent for publication {#FPar4} ======================= Not applicable. Ethics approval and consent to participate {#FPar5} ========================================== The participants provided written informed consent to take part in the study. Our study received ethical approval by the Institutional Review Board of the Medical Faculty of the Heinrich-Heine University of Düsseldorf (no. 5764R). Funding {#FPar6} ======= This work was funded by refonet - Rehabilitation Research Network of the German Pension Fund Rhineland, Grant Number 14006. The funding body had no influence on the design of the study and collection, analysis, and interpretation of data and in writing the manuscript. Publisher's Note {#FPar7} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

All relevant data have been deposited in the National Center for Biotechnology Information\'s website and can be found at the following links: <https://trace.ncbi.nlm.nih.gov/Traces/sra/?study=SRP120033><https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE105032>. Introduction {#sec005} ============ Strategies for applying liquid biopsy assays to the treatment of pediatric solid tumors is still in the early stages of development \[[@pone.0188835.ref001]--[@pone.0188835.ref004]\]. Studies are hampered by the limited number of patients available for enrollment in sample collection protocols at any one institution. Efforts to determine the prognostic value of liquid biopsy assays in pediatric solid tumors will, therefore, necessitate coordinated efforts across multiple institutions. These studies will require samples to be collected at institutions that may not be capable of rapidly processing samples locally and, instead, samples will need to be shipped and processed uniformly at a central laboratory. For this reason, many collaborative trials are beginning to utilize specialized WBC-stabilizing blood collection tubes for liquid biopsy studies. These tubes contain a nucleated cell-stabilizing preservative that prevent the lysis of WBCs when samples cannot be processed immediately. Stabilizing the WBCs prior to sample processing prevents the release of large amounts of genomic DNA into the plasma that would otherwise dilute circulating tumor DNA (ctDNA) levels to undetectable proportions. In pediatrics, the volume of blood drawn for research is sometimes restricted to ensure the safety of young children who have smaller blood volumes and may be more hemodynamically affected by sample collections. Thus, researchers must often choose between collecting samples for one research project over another. We anticipate that this may limit the usefulness of requesting that additional blood be collected in WBC-stabilizing tubes in ongoing research efforts in pediatric oncology. Therefore, we sought to determine whether samples obtained in WBC-stabilizing tubes could be used simultaneously for two purposes: 1) liquid biopsy studies and 2) germline studies. While the use of WBC-stabilizing tubes for ctDNA studies have been well-validated, to our knowledge there are no published studies describing the quality of germline DNA obtained from these tubes and their usefulness in performing germline genomic analyses. Here, we describe our experience performing whole-exome sequencing (WES) and genotyping by copy number and single nucleotide polymorphism (SNP) array analysis of germline samples obtained simultaneously in standard EDTA and Streck Cell-Free DNA BCT^®^ collection tubes from three patients. Methods {#sec006} ======= Patients and consent {#sec007} -------------------- Blood samples were collected from patients enrolled on a Dana-Farber Cancer Institute IRB-approved banking study (DFCI protocol \#11--104). Written informed consent was obtained from participants or parents of minor participants prior to study inclusion. Patients were selected for participation based on an expectation that they would have normal or near-normal WBC counts based on their clinical course. We also chose children older than 11 years old for the study to ensure that drawing multiple tubes of blood simultaneously would not infringe on the hospital established blood volume limits as it would in younger patients. We chose patients who were expected to require venous access for clinical purposes so that no additional procedures were required to obtain these samples. For each patient, 5 mL of blood was drawn into a purple top EDTA tube immediately followed by the collection of 10 mL of blood into a Streck Cell-Free DNA BCT^®^ tube. Sample processing and DNA extraction {#sec008} ------------------------------------ Blood samples were processed at the Boston Children's Hospital Biorepository on the same day as the specimens were acquired. For blood collected in both EDTA and Streck tubes, samples were first centrifuged at 1000 x g at 4° Celsius for 10 minutes in the collection tube. Plasma was then removed and stored for future use. The remaining blood, a combination of red blood cells and white blood cells, were then removed from the collection tube and subjected to DNA extraction using the Gentra Puregene Blood Kit (Qiagen) according to the manufacturer's instructions. DNA was quantified for each sample using the Quant-iT PicoGreen dsDNA Assay Kit (Thermo Fisher Scientific). Whole exome library construction and sequencing {#sec009} ----------------------------------------------- Control DNA for this experiment was obtained from the human B-lymphocyte cell line CEPH1408 (GM10831), (Catalog No. NA10831; Coriell Institute for Medical Research) \[[@pone.0188835.ref005]\]. A total of 100 ng of DNA from each sample was ultra-sonicated to an average fragment size of 250 base pairs (Covaris) and further size selected using Agencourt AMPure XP beads (Beckman Coulter). DNA was then manually ligated to adaptors using the KAPA HTP Library Preparation Kit (KK8234, KAPA Biosystems). All libraries were purified and size selected using Agencourt AMPure XP beads to perform a double-sided solid phase reversible immobilization (SPRI) cleanup. The bead to DNA ratio of 0.5X was used for the high molecular weight SPRI selection followed by a 2.5X SPRI selection. Libraries were analyzed using a Bioanalyzer (Agilent) to ensure that all sample libraries had the expected DNA fragment size distribution. Library concentrations were quantitated using an Illumina MiSeq. Briefly, 1 μl from each library was pooled and quantitated using a Kapa Biosystems Library Quant Kit (KK4854, Kapa Biosystems), normalized to 4 nM, and then sequenced on a MiSeq nano flow cell (Illumina). The "sequenceable" concentration of each library was calculated by dividing the number of clusters per barcode by the average number of clusters and then multiplying the result by the qPCR value of the pooled samples. The libraries were then normalized and pooled for hybrid capture such that samples were divided into two capture reactions performed with a total of 750 ng of pooled library DNA. Capture was performed using the SureSelectXT Hybrid Capture kit with the Exon v5 bait set (Agilent Technologies). After capture, all samples were pooled together and sequenced across two lanes of a Hiseq 2500 in rapid run mode (Illumina). Sequencing analysis {#sec010} ------------------- Pooled sample reads were deconvoluted and sorted using Picard tools (<http://broadinstitute.github.io/picard/picard-metric-definitions.html>). Reads were then aligned to the b37 edition of the human genome reference sequence (Human Genome Reference Consortium) using bwa (<http://bio-bwa.sourceforge.net/bwa.shtml>) utilizing the parameters "-q 5 -l 32 -k 2 -o 1". Duplicate reads were identified and removed using Picard tools \[[@pone.0188835.ref006]\]. Alignments were further refined using GATK for localized realignment around sites of small insertions and deletions (<https://www.broadinstitute.org/gatk/gatkdocs/org_broadinstitute_gatk_tools_walkers_indels_IndelRealigner.php>). Base quality score recalibration was also performed using GATK (<http://gatkforums.broadinstitute.org/discussion/44/base-quality-score-recalibration-bqsr>) \[[@pone.0188835.ref007], [@pone.0188835.ref008]\]. Unaligned reads were used to confirm performance of both sequencing lanes. We then determined the number of reads per sample, mean target coverage, and percent of targeted bases with at least 30x sequencing coverage to estimate the sequencing quality across all samples. Lastly, DNA fingerprinting analysis was performed using Picard Tools to confirm concordance at 44 polymorphic loci among samples obtained from the same patient. Variant analysis for single nucleotide variants (SNV) was performed using MuTect v1.1.4 and annotated by Variant Effect Predictor (VEP). The SomaticIndelDetector tool from GATK was used for indel calling \[[@pone.0188835.ref009], [@pone.0188835.ref010]\]. MuTect was run in paired mode using sequencing data generated from human cell line CEPH1408 as the project normal. Samples were compared to identify variant calls that were discordant between samples collected in EDTA and Streck tubes from the same patient and to investigate whether any patterns of sequencing discrepancies could be appreciated between the two different collection methods. Aligned BAM files are available at the NCBI Sequencing Read Archive at <https://trace.ncbi.nlm.nih.gov/Traces/sra/?study=SRP120033>. Copy number microarray {#sec011} ---------------------- Microarray analysis was performed on 1 ug of DNA sample. Each test sample was digested in parallel with a sex matched control (Agilent Male/Female Reference DNA) using the restriction enzymes Alu I and RSA I to obtain 200-500bp fragments. Following digestion, the test samples were labeled with Cyanine 5-dUTP and control samples were labeled with Cyanine 3-dUTP. The unincorporated dye was removed. Each test sample and correlating control were combined and hybridized to a custom 4x180K CGH plus SNP microarray (GGXChip+SNP v1.0, Agilent). The copy-number probes on the array (\~135,000) have an average spacing of one probe every 35 kb throughout the genome and one probe every 10 kb in regions known to have clinical significance while the SNP probes (\~67,000) have an average spacing of 40 kb across the genome. Slides were scanned on a SureScan Microarray Scanner (Agilent) and data was analyzed through CytoGenomics v4.0.3.1.2 (Agilent) and Genoglyphix v3.1 (Perkin Elmer). Array data can be accessed at the NCBI Gene Expression Omnibus site <https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE105032>. Results {#sec012} ======= Genomic DNA can be extracted from blood samples collected in Streck tubes {#sec013} ------------------------------------------------------------------------- Genomic DNA was obtained from blood collected in both EDTA and Streck blood collection tubes using a standard extraction kit with no modifications to the protocol for either tube. In all samples, we extracted over 100 μg of genomic DNA material using these standard procedures ([Table 1](#pone.0188835.t001){ref-type="table"}). In most cases, the Streck tube yielded a larger amount of genomic DNA per sample, likely due to the larger starting blood volume compared to samples collected in the EDTA tube. Aliquots of 100 ng of DNA were then used for DNA sequencing and 1 μg was used for copy number analysis. Over 100 μg of genomic DNA material from each sample remained available for additional studies, demonstrating that blood collected in both EDTA and Streck tubes yield ample DNA for multiple genomic studies. 10.1371/journal.pone.0188835.t001 ###### DNA extraction and sequencing metrics. {#pone.0188835.t001g} Subject ID Sample Type DNA extracted (μg) DNA (ug) per mL blood Library Yield (ng) Reads Mean Target Cvg (x) \% Target Bases 30x ------------ ------------- -------------------- ----------------------- -------------------- ------------ --------------------- --------------------- 179 EDTA 230.1 46.0 2242.19 71,417,612 71.7 89.2 Streck 399.0 39.9 3060.21 84,343,168 83.2 92 394 EDTA 349.3 69.9 2615.99 90,165,770 87 93.2 Streck 513.3 51.3 2909.82 90,655,396 87.6 93.2 403 EDTA 158.2 31.6 2985.58 91,215,682 87.2 93.2 Streck 144.1 14.4 3313.12 83,573,048 81.5 91.7 CEPH1408 Cell line     3440.64 81,051,808 80.5 91.8 DNA extraction and sequencing metrics for whole-exome sequencing of germline DNA extracted from blood samples collected in EDTA and Streck tubes. High-quality sequencing can be performed from blood samples collected in Streck tubes {#sec014} ------------------------------------------------------------------------------------- Sequencing libraries were prepared from 100 ng of genomic DNA from each sample. All sequencing libraries were prepared and purified uniformly but with unique barcodes so that samples could be multiplexed for sequencing and then de-multiplexed during analysis. Library yields from all blood samples were remarkably similar and were comparable with the DNA library yields generated from DNA extracted from a human cell line, CEPH1408 ([Table 1](#pone.0188835.t001){ref-type="table"}). Bioanalyzer electropherograms demonstrated that all samples had uniform and expected sample library fragment size (data not shown). Recently published cancer predisposition studies have used both whole-genome sequencing and targeted sequencing approaches to identify recurrent germline variants in patients and families with a high incidence of cancer \[[@pone.0188835.ref011]--[@pone.0188835.ref015]\]. Targeted sequencing often involves a hybrid capture step to enrich sequencing libraries for genomic regions of interest. Thus, hybrid capture of DNA libraries was performed on all samples to enrich for the coding region of the genome using the Agilent Exon v5 capture bait set. Captured libraries were then sequenced in order to achieve an expected target coverage of approximately 80x. There were a similar number of total sequencing reads generated per sample ([Table 1](#pone.0188835.t001){ref-type="table"}). Plasma samples had an average sequencing depth of 83x (71.7x -- 87.6 x) for the target regions, and over 89% of all targeted bases had at least 30x coverage ([Table 1](#pone.0188835.t001){ref-type="table"}). WES metrics were similar regardless of whether genomic DNA samples had originated from blood collected in an EDTA or Streck tubes. When read coverage was binned by GC content of the targeted regions, mean coverage was nearly identical between paired samples obtained in EDTA tubes and samples collected in Streck tubes, and differences in normalized mean coverage were less than 0.02x for all categories of percent GC content ([Fig 1](#pone.0188835.g001){ref-type="fig"}). {#pone.0188835.g001} Sequence variants were identified by comparing each sample to DNA extracted from a human cell line, CEPH1408, which was sequenced simultaneously with the germline samples. A similar number of unfiltered variants were identified for each patient, regardless of whether the germline sample sequenced was collected in an EDTA or Streck tube ([Table 2](#pone.0188835.t002){ref-type="table"}). Since all germline variants can be expected to be either heterozygous or homozygous, we expected true variants to have allelic fractions close to 0.5. However, because hybrid capture sequencing may introduce some sequencing bias between alleles, we included in our analysis all variants with an allelic fraction \> 0.25 and sequencing coverage of greater than 30 reads \[[@pone.0188835.ref016], [@pone.0188835.ref017]\]. All samples had a similar percentage of variants that passed this filter regardless of which tube the sample was used for blood collection (57.1--67.1%; [Table 2](#pone.0188835.t002){ref-type="table"}). When each sample was compared to the matched sample from the same patient, less than 1.2% of the variants were unique to the individual sample, consistent with the published rates of sequencing error observed when repeated WES has been performed \[[@pone.0188835.ref018]--[@pone.0188835.ref021]\]. Furthermore, when variants were filtered by whether they were seen in any of the other 5 samples sequenced, less than 0.58% of variants were unique to that sample, demonstrating an extremely low sequencing error rate that was similar regardless of the type of tube in which the sample was collected. Finally, we compared the pattern of mutations observed in each sample across the three base-pair context and could not detect a bias in mutation pattern between samples collected in EDTA or Streck tubes ([Fig 2](#pone.0188835.g002){ref-type="fig"}). {#pone.0188835.g002} 10.1371/journal.pone.0188835.t002 ###### WES variant calls. {#pone.0188835.t002g} ----------------------------------------------------------------------------------------------------------------------- Subject ID Sample Type Total Variants Filtered (% of Total Variants) Filtered Not in Pair\ Filtered Unique\ (% Filtered) (% Filtered) ------------ ------------- ---------------- -------------------------------- ----------------------- ------------------ 179 EDTA 21672 12379 (57.12%) 122 (0.99%) 59 (0.48%) Streck 22598 14579 (64.51%) 175 (1.2%) 84 (0.58%) 394 EDTA 25478 17153 (67.32%) 182 (1.06%) 89 (0.52%) Streck 25147 16452 (65.42%) 152 (0.92%) 76 (0.46%) 403 EDTA 25213 16909 (67.06%) 176 (1.04%) 85 (0.5%) Streck 24620 15587 (63.31%) 146 (0.94%) 65 (0.42%) ----------------------------------------------------------------------------------------------------------------------- Summary of sample variants called from whole-exome sequencing of germline DNA compared to a cell line control. Total variants were subsequently filtered using three increasingly stringent criteria: 1) Filtered variants were retained if the allelic fraction was \> 0.25 and the coverage was \> 30x 2) Filtered Not in Pair variants were retained if the variant was not seen in the other patient-matched sample 3) Filtered Unique variants were retained if the variant was not seen in any of the other samples. Genome-wide copy number plus SNP microarrays identify the same variants in germline samples regardless of sample collection strategy {#sec015} ------------------------------------------------------------------------------------------------------------------------------------ DNA samples were also analyzed with genome-wide combined copy number/SNP mircroarrays. For each patient, regions of copy number alteration and homozygosity were identified and compared between samples obtained in EDTA tubes and Streck tubes. The quality control metrics were excellent for copy number analysis using the derivative log ratio spread (DLR) metric and SNP analysis using the call rate (CR) metric, regardless of how the sample was collected ([Table 3](#pone.0188835.t003){ref-type="table"}). Copy number analysis showed that clinically relevant data was identical for the Streck and EDTA-derived samples. Similarly, matched samples showed identical regions of homozygosity ([Table 3](#pone.0188835.t003){ref-type="table"}). 10.1371/journal.pone.0188835.t003 ###### Microarray results. {#pone.0188835.t003g} Subject ID Sample Type DLR SNP CR \% Overlapping CN Calls SNP Calls ------------ ------------- -------- ----------------------------------------------------------------------------- ----------------------------------------------------------------------------- ------------------------------------------------------------------------ 179 EDTA 0.1261 0.95 8p11.22 loss, 14q32.33 gain,16p12.2 gain, Xp22.33 loss none Streck 0.1249 0.937 8p11.22 loss, 14q32.33 gain,16p12.2 gain, Xp22.33 loss none 394 EDTA 0.1241 0.971 6p25.3 loss, 7q11.21 loss, 8p11.22 gain,22q11.22 gain none Streck 0.1295 0.972 6p25.3 loss, 7q11.21 loss, 8p11.22 gain,22q11.22 gain none 403 EDTA 0.1292 0.977 1q44 loss, 7p22.3 gain, 8p11.22 loss, 14q32.33 gains, 15q11.2,17q21.31 loss 2p25.1--24.3 ROH, 3p24.3-p24.1 ROH, 6p21.2-p12.1 ROH,7q31.1-q31.31 ROH Streck 0.1322 0.946 1q44 loss, 7p22.3 gain, 8p11.22 loss, 14q32.33 gains, 15q11.2,17q21.31 loss 2p25.1--24.3 ROH, 3p24.3-p24.1 ROH, 6p21.2-p12.1 ROH,7q31.1-q31.31 ROH Germline DNA samples were tested on a custom combined copy number plus single nucleotide polymorphism microarray. Sample-specific quality metrics for copy number (DLR) and SNP (SNP call rate) are shown along with copy-number calls and SNP calls (ROH \> 5 Mb). Discussion {#sec016} ========== The use of liquid biopsy assays that detect, quantify, and profile ctDNA are being routinely incorporated into clinical trials for adult cancers \[[@pone.0188835.ref022]--[@pone.0188835.ref024]\]. These assays can be used to detect targetable mutations in a patient's cancer, sometimes without the need for invasive biopsies \[[@pone.0188835.ref025], [@pone.0188835.ref026]\]. Changes in ctDNA levels often correspond to changes in disease burden and thus, these assays have also been used to track response to therapy and to monitor for relapse \[[@pone.0188835.ref027]--[@pone.0188835.ref029]\]. Profiling of ctDNA has been used to detect the emergence of resistance mutations in tumors for patients on targeted therapy and efforts are now underway to perform more expansive profiling of ctDNA samples to define patterns of clonal evolution and treatment resistance \[[@pone.0188835.ref024], [@pone.0188835.ref025], [@pone.0188835.ref030]\]. Such efforts in pediatric cancers are also beginning to emerge \[[@pone.0188835.ref001]--[@pone.0188835.ref004]\]. Published studies of ctDNA in pediatric solid tumors have utilized relatively small cohorts of patients, justifying further study of ctDNA. However, these initial efforts have been insufficiently powered to study the prognostic value of these assays. Studies in adult cancers confirm that ctDNA levels vary widely between cancer types and assay designs are also dependent on the expected genomic landscape of each histology \[[@pone.0188835.ref022]\]. Thus, studies in pediatric solid tumors will require focused, disease-specific efforts. Due to the relative rarity of most pediatric solid tumors, the development and validation of liquid biopsy technologies will require collaborative efforts that extend beyond single institutions. There are several existing collaborative studies underway that would be uniquely poised to adequately collect blood samples and clinical data, including but not limited to, the Children's Oncology Group (COG) Project: Every Child (<https://clinicaltrials.gov/ct2/show/NCT02402244>), the GAIN Consortium's iCAT2 study (<https://clinicaltrials.gov/ct2/show/NCT02520713>), and the NCI-MATCH study for pediatrics (<https://www.cancer.gov/about-cancer/treatment/clinical-trials/nci-supported/pediatric-match>). There are several logistical hurdles to overcome in developing liquid biopsy technologies in trials enrolling pediatric patients at multiple institutions. One major issue is that samples must be collected and shipped to a central laboratory before being processed. ctDNA is easily extracted from plasma isolated from blood collected in EDTA tubes using standard extraction kits. However, several studies have demonstrated that plasma must be isolated from blood collected in EDTA tubes within eight hours of collection to prevent the chance of lysis of WBCs that will contaminate the plasma with genomic DNA \[[@pone.0188835.ref031]--[@pone.0188835.ref035]\]. This genomic DNA impairs the ability to easily detect and quantify ctDNA from the plasma by lowering the ctDNA abundance relative to germline DNA in the sample. One way to circumvent this problem is to collect blood in preservative-containing tubes designed to stabilize WBCs, thereby allowing samples to be collected at numerous institutions and then be shipped to a central laboratory for processing. Furthermore, liquid biopsy studies have demonstrated that ctDNA can be more easily detected from larger volumes of plasma. Thus, clinical trials designed to study ctDNA in collaborative efforts would benefit from collecting large volumes of blood. However, another logistical hurdle in pediatrics is that many of our patients have much smaller blood volumes, restricting the amount of blood that can be collected at any one-time point. One competing research priority in pediatrics is the collection of a blood sample for germline DNA profiling. DNA for these studies are most often extracted from WBCs obtained from blood collected in an EDTA tube. A potential solution for collecting plasma for ctDNA studies while still prioritizing the collection of germline material for genetic studies would be to utilize a single tube of blood for both germline and liquid biopsy studies. To determine the feasibility of this approach, we collected blood from three patients simultaneously in EDTA and Streck tubes and performed WES and SNP-arrays on the matched germline DNA extracted from cells from each tube. We demonstrated that the WES and SNP-array data was of very similar quality for germline studies, suggesting that both ctDNA and germline study samples can be collected from a single WBC-stabilizing tube for pediatric cancer patients. Similar results were observed when utilizing column-based extraction methods to obtain genomic DNA from EDTA and Streck tubes (data not shown). One potential limitation of this study is that we were not able to test the effects of exposing blood samples to a wider range of temperatures or processing times. However, guidelines for handling Streck tubes suggest that samples be maintained at room temperature during shipment and processed within 7--10 days. We would expect that conforming to those guidelines would ensure the quality of cell-free and germline DNA extracted from blood collected in Streck tubes. We believe this data will facilitate the collection of liquid biopsy studies on multi-institutional trials without compromising germline studies or increasing the blood volumes requested for ongoing research. We thank the Boston Children's Hospital Biorepository for processing blood samples and extracting DNA. We are also grateful to the patients who participated in this study. [^1]: **Competing Interests:**The authors have declared that no competing interests exist.

Introduction {#Sec1} ============ Osteotomy procedures for varus malalignment are frequently performed. Valgus malalignment deformity of the knee however, is much less frequent (2%-2,8%) (Bellemans et al. [@CR1]) and a corrective procedure i.e. a varus producing osteotomy in these valgus knees is most often performed in the distal femur (Haviv et al. [@CR14]; Puddu et al. [@CR22]). Correction in the proximal (or high) tibia is much less frequent. However, if the valgus leg deformity is located in the tibia a tibial correction should be performed to prevent an abnormal knee joint line obliquity (van Egmond et al. [@CR27]; Hofmann et al. [@CR16]). As an osteotomy pioneer, Coventry (1985, 1987) described the varus producing medial closing wedge high tibial osteotomy (CWHTO) to treat valgus malalignment of the knee (Coventry [@CR4]; Coventry [@CR5]). He stated that, by removing a bony wedge on the medial side of the proximal tibia, a laxity of the superficial medial collateral ligament (MCL) is introduced. He therefore suggests to perform a surgical reefing procedure at all times to tighten the MCL (Coventry [@CR4]; Coventry [@CR5]). Shoji and Insall also described this procedure in some patients (Shoji and Insall [@CR24]). However, in our clinic no reefing procedures of the MCL are performed and patients have not reported routinely on knee instability after this procedure. Till date no study has evaluated the MCL-laxity before and after a varus producing medial CWHTO for valgus malalignment of the knee. The MCL consists of a deep part and a superficial part that have different functions while together providing normal laxity during knee range of motion (Wymenga et al. [@CR30]). Instrumented laxity measurements has proven that collateral laxity of the knee differs in flexion and extension in healthy subjects (Deep et al. [@CR9]; Heesterbeek et al. [@CR15]; Yoo et al. [@CR31]). The collateral ligament laxity is variable in different persons and ligaments in women are more lax than in men especially in valgus stress. These reference values in healthy volunteers have been measured with knees positioned in varying flexion angles and with different types of instrumented laxity measurement methods. Whereas laxity can be measured and quantified instability can be defined as the clinical manifestation of patient's perceived increased laxity in the knee and is scored on subjective scales. Self-reported knee instability as well as greater dynamic varus-valgus stress are associated with worse self-reported knee confidence (Skou et al. [@CR25]). This worse self-reported knee confidence has also been shown to predict functional decline in people with OA (Colbert et al. [@CR3]). An absence of change to the MCL laxity as well as an unchanged sense of knee stability therefore is vital to a good outcome for a medial CWHTO. The aim of this pilot study was to evaluate the effect of a medial CWHTO in varus-valgus laxity changes radiologically and subjectively and to report if these correlated to each other. We hypothesize that both MCL and LCL laxity do not change after a medial CWHTO, that self-reported knee instability also does not change and as a result, no additional surgical reefing procedure of the MCL is necessary. Material and methods {#Sec2} ==================== Subjects {#Sec3} -------- Between May 2015 to March 2016 eleven consecutive patients were included in this study. Inclusion criteria were symptomatic valgus malalignment located in the proximal (high) tibia, indication for a medial CWHTO, based on the severity of the complaints and the observed deformity according to Paley (Paley [@CR21]), age 18--65 years and no history of knee ligament injuries. Exclusion criteria were previous MCL surgery, previous ipsilateral total hip replacement because of planned instrumented flexion laxity measurements, BMI higher than 30 kg/m^2^. All patients reported to the outpatient clinic and were seen and operated on by a single orthopaedic surgeon (RvH). The operative procedure consisted of surgical removal of a pre-planned bone wedge after which the created gap was closed to straighten the leg. Fixation after closure was performed with an angle stable Tomofix-plate. The location of the wedge removal is distal to the attachment of the deep part of the MCL and within the attachment of the superficial part of the MCL on the medial proximal tibia (Fig. [1](#Fig1){ref-type="fig"}). This study was approved by the Medical Research and Ethical Committee, and all patients provided written informed consent for participation in the study. Being a pilot study, no power analyses were done.Fig. 1Preoperative and postoperative radiographs of the osteotomy site in relation to the medial collateral ligament. Subscript: The left figure shows a pre-operative long-leg radiographs with valgus malalignment. The right figures show a preoperative (above) and postoperative (below) radiograph of the knee. The osteotomy site (A and B) is below the insertion of the deep medial collateral ligament (dMCL). In the postoperative situation a pseudo laxity of the superficial medial collateral ligament (sMCL) is created Radiographic measurements {#Sec4} ------------------------- Pre- and postoperative long-leg radiographs were obtained for each patient. Subsequently every patient received varus and valgus stress radiographs in 30° and 70° flexion. On the day of surgery, after either spinal or total anaesthetics were applied, radiographs were made before surgery started. The postoperative radiographs were performed three to six months post-surgery in the outpatient clinic without the use of anaesthetics. At that time the subjects were instructed to relax their thigh muscles to minimize the role of the dynamic knee stabilizers when conducting the radiographs in 30° and 70° flexion in the postoperative setting. Varus and valgus laxity of the knee in 30° flexion was assessed using the Telos device (Fa Telos, Medizinisch-Technische GmbH, Griesheim, Germany) with the subject lying in a supine position with leg muscles relaxed. The knee was positioned in an approximately 30° angle by putting a 15 cm diameter plastic pipe under the knee to ensure a reproducible angle pre- and postoperative. The Telos device was applied with 15 Nm load on the leg relative to the level of the knee joint line. While medial and lateral forces were applied, radiographs were obtained in the anteroposterior view. The direction of the X-rays was parallel to the tibia joint surface, centred on the middle of the femoral-tibial joint space. For varus and valgus laxity of the knee in 70° flexion, a custom-made stress device was used to stress the knee and to produce reproducible measurements. An external load of 15 Nm was applied at the knee joint using 50 N on a pulley 0.30 m distal from the joint line. The knee was stressed medially and laterally. Radiographs were made with the X-ray direction parallel to the tibia joint surface in the conditions varus, valgus or no moment applied. This method has previously been described and validated by Heesterbeek et al. (Heesterbeek et al. [@CR15]). The angle between a tangent line on the femur condyles and a line through the deepest tibial joint surfaces was determined on the varus, valgus and neutral radiographs using the measurement tool within the radiographic database program (Sectra workstation IDS7, version 16.1.22.1566 (2015), Sectra AB, Linköping, Sweden) (Fig. [2](#Fig2){ref-type="fig"}). Valgus laxity was defined as the difference between the medial stress radiograph and the neutral radiograph, varus laxity as the difference between the lateral stress radiograph and the neutral radiograph. The measurements were made to the nearest 0.1°. The radiographic measurements were done by two authors (WvL, CM) independently. When the independent measurements differed (\> 1° apart) consensus was made. A third independent experienced orthopaedic surgeon was asked to review a sample of the measurements and reported similar measurements.Fig. 2Valgus and varus laxity stress radiographs with joint space opening. Subscript: Pre-operative stress radiographs of the knee in 30° and 70° of knee flexion in an anaesthetized patient. The upper two figures show radiographs in 30° of flexion with varus (left) and valgus (right) stress. The lower two figures show radiographs in 70° of flexion in varus (left) and valgus (right) stress. The angle between a tangent line on the femur condyles and a line through the deepest tibial joint surfaces was determined and compared to the natural (unstressed) knee joint line congruence angle Questionnaires {#Sec5} -------------- Patients were asked to fill in questionnaires preoperative and at six months postoperative. The validated Dutch Knee Injury and Osteoarthritis Outcome Score (KOOS) was used to assess subjective knee function and quality of life and consists of 5 different subscales: (1) pain, (2) symptoms, (3) function in daily living, (4) function in sport and recreation, and (5) knee-related quality of life. The score for each subscale was calculated, where 0 indicates severe knee problems and 100 indicates no knee problems (de Groot et al. [@CR7]; Roos and Toksvig-Larsen [@CR23]). The validated Dutch International Knee Documentation Committee (IKDC) Subjective Knee Form was used to assess symptoms and limitations in function and sports. A higher score indicates a better function (0--100 scale) (Haverkamp et al. [@CR13]; Irrgang et al. [@CR17]). The IKDC Current Health Assessment Form (SF36 form) was used to measure health status with respect to different dimensions: (1) physical functioning, (2) social functioning, (3) role limitations due to physical problems, (4) role limitations due to emotional problems, (5) bodily pain, (6) mental health, (7) vitality, (8) general health perception, and (10) change in perceived health during the last 12 months. All raw scores were converted to a 0--100 scale, with higher scores indicating higher levels of functioning or well-being (Haverkamp et al. [@CR13]; Ware and Sherbourne [@CR29]). The validated Dutch Oxford Knee Score (OKS) questionnaire was used to assess function and pain after surgery. A lower score indicates a better outcome (range 12--60) (Dawson et al. [@CR6]; Haverkamp et al. [@CR12]). The Lysholm knee scale was used to assess ligament injuries of the knee. A higher score indicates a better outcome (range 0--100) (Eshuis et al. [@CR10]; Kocher et al. [@CR18]). In addition, patients were asked to fill in a visual analogue scale (VAS) Instability, with a higher score indicating more instability of the operated knee (range 0--10). Statistical analysis {#Sec6} -------------------- Differences in varus and valgus laxity of the knee in 30° and 70° flexion pre- and postoperative were assessed using nonparametric Wilcoxon signed-rank tests. Also differences in subjective scores (IKDC, Oxford, Lysholm and VAS Instability) were assessed using nonparametric Wilcoxon signed-rank tests. Results were presented as medians with range. All tests were performed with a level of significance of 0.05. Results {#Sec7} ======= Eleven patients were included in this study, with a median age at the time of operation of 46 years (range 24--66 years). One patient was 66 at the time of operation but was accepted for this study as she was still 65 years old when put on the waiting list for this surgery. Seven patients were female and 4 were male. In one patient preoperative 30° flexion X-rays could not be obtained due to a technical error. All eleven patients completed the study. Preoperative mean hip-knee-ankle (HKA) angle was 6.4° valgus (12° valgus to 1° valgus) and this was restored postoperative to a mean HKA angle of 0.1° (5.8° valgus to 5° varus). In one patient the postoperative long leg radiograph was missing. Stress X-rays in 30° and 70° flexion {#Sec8} ------------------------------------ After separate radiological assessment by the two authors consensus had to be made for 6 patients regarding at least one value. Pre- and postoperative measurements for every single stress X-ray group are presented in Table [1](#Tab1){ref-type="table"}. Overall, in the preoperative setting patients had higher varus laxity values compared to valgus laxity in both 30° and 70° flexion. Postoperative, this differentiated collateral laxity of knee shifted to a more distinguished valgus laxity and a decrease in varus laxity of the knee. This indicated tighter knees on the lateral side and increased laxity on the medial side of the knees.Table 1Differences in varus and valgus laxity in 30° and 70° flexion of the kneePreoperative\*Postoperative\**P* value30° Valgus Laxity2.8° (−4.0--4.3°)5.3° (2.5° -- 7.1°)0.005\*30° Varus Laxity6.7° (3.7° -- 9.3°)3.2° (− 0.9° -- 4.9°)0.005\*70° Valgus Laxity2.0° (−2.6° -- 5.4°)4.8° (0.4° -- 7.6°)0.008\*70° Varus Laxity3.8° (− 3.2° -- 9.0°)1.3° (− 0.1° -- 8.3°)0.113Values are presented as: median (range)\* Significant difference (*P* \< 0.05) When comparing the pre- and postoperative results for each condition, there was a statistically significant difference between the pre- and postoperative 30° valgus laxity (respectively 2.8° and 5.3° (*P* = 0.005)), 30° varus laxity (respectively 6.7° and 3.2° (P = 0.005)) and 70° valgus laxity (respectively 2.0° and 4.8° (*P* = 0.008)). The difference between the pre- and postoperative 70° varus laxity did not reach significance (3.8° versus 1,3° (*P* = 0.113)). Patient reported outcome scores {#Sec9} ------------------------------- Patient-reported outcome results are presented in Tables [2](#Tab2){ref-type="table"}, 3, and [4](#Tab4){ref-type="table"}. Postoperative patient-reported knee instability as measured with the Lysholm questionnaire was significantly improved compared to preoperative instability (*P* = 0.006, Table [2](#Tab2){ref-type="table"}). The VAS knee instability score also improved, but didn't reach significance (8.0 preoperative and 5.5 postoperative (*P* = 0.127, Table [2](#Tab2){ref-type="table"})). Knee function, knee pain and knee-related quality of life as measured with the IKDC Knee evaluation form, Oxford Knee Score and the KOOS all showed significant improvements postoperative as shown in Table [3](#Tab3){ref-type="table"}. Only the quality of life domains 'Physical functioning' and 'Bodily pain' of the IKDC Current Health Assessment Form showed significant improvements (Table [4](#Tab4){ref-type="table"}).Table 2Differences in patient-reported knee instability as measured with the Lysholm and VASPreoperativePostoperativeP valueLysholm42 (26--75)67 (27--90)0.006\*VAS knee instability8.0 (2.5--9.0)5.5 (1.0--8.0)0.127Values are presented as: median (range)\* Significant difference (*P* \< 0.05) Abbreviations: *VAS* Visual Analogue ScaleTable 3Differences in knee function, pain and knee-related quality of life as measured with PROMsPreoperativePostoperativeP valueIKDC Knee Evaluation32 (13--51)60 (24--82)0.004\*Oxford Knee Score35.0 (20--47)23.5 (15--51)0.017\*KOOS Pain42 (17--75)72 (22--92)0.041\*KOOS Symptom43 (12--86)68 (54--96)0.007\*KOOS ADL53 (21--90)81 (35--97)0.008\*KOOS Sport & Recreation0 (0--25)35 (0--70)0.020\*KOOS QoL25 (6--50)38 (6--63)0.044\*Values are presented as: median (range)\* Significant difference (*P* \< 0.05) Abbreviations: *PROMs* patient reported outcome measures, *IKDC* International Knee Documentation Committee; *KOOS* Knee Injury and Osteoarthritis Outcome Score, *ADL* activities of daily living, *QoL* Quality of lifeTable 4Differences in health-related quality of life as measured with the IKDC Current Health Assessment FormPreoperativePostoperativeP valuePhysical functioning55.0 (5--85)75.0 (40--95)0.020\*Physical role functioning62.5 (0--100)87.5 (0--100)0.357Bodily pain51.0 (0--74)62.0 (41--84)0.014\*General Health72.0 (52--97)82.0 (52--95)0.420Vitality65.0 (50--90)65.0 (55--90)0.475Social functioning88.0 (38--100)100.0 (50--100)0.596Emotional role functioning100.0 (67--100)100.0 (0--100)0.157Mental Health88.0 (52--100)88.0 (68--100)0.157Values are presented as: median (range)\* Significant difference (*P* \< 0.05) No correlations were found between radiological findings and patient reported outcome scores. Discussion {#Sec10} ========== This study has shown a significant increase in valgus laxity in both 30° and 70° of flexion after medial closing wedge HTO, which indicates a change of laxity of the medial collateral ligament. Furthermore, a significant decrease in varus laxity in 30° of flexion and a non-significant decrease in 70° flexion, was found. These findings are not correlated to subjective knee stability and outcome scores which all improved after the operation. Therefore, our hypothesis is false. Our preoperative valgus and varus laxity values in 70° of flexion are comparable with the normal values in healthy individuals. Our reported valgus and varus laxity in 70° flexion are 2.0° and 3.8° versus 2.5° and 3.1° as found in the study of Heesterbeek et al. who performed instrumented laxity measurements with the same measurement device and are therefore comparable (Heesterbeek et al. [@CR15]). In a study conducted by Deep the varus and valgus laxity was measured with instrumented laxity measurements in 15° flexion in healthy individuals. This study showed that the femorotibial mechanical angle (FTMA) changed with a mean of 6.9° when 10 Nm varus torque was applied. With valgus torque the mean change in FTMA was 7.9°. The unstressed FTMA was 1.2° varus (Deep [@CR8]). Our reported varus and valgus laxity in 30° flexion preoperative were respectively 6.7° and 2.8°. Our values are therefore lower than reported by Deep, however, we performed the tests in 30° of flexion. This might influence the results. Our postoperative results cannot be compared to the literature since this is the first study to our knowledge to investigate the change in knee laxity after a medial CWHTO. Moreover, due to the difference in radiological measurement settings (i.e. under anaesthetics preoperative and non-anaesthetized conditions postoperative) we have possibly inflicted a non-comparable situation especially as regards the varus laxity measurements. Tsukeoka and Tsuneizumi proved that anaesthesia significantly influenced knee joint laxity on both the medial and lateral side after TKA. They reported a positive correlation between the laxity under anaesthesia and the amount of change in laxity on the lateral side (*r* = 0.57; *p* = 0.0022). They found an increase in laxity for varus stress with mean of 1° under anaesthetics in a 15 degree flexion angle of the knee. Moreover, in 23% of their patients they reported an increase of \> 3° in laxity measured under anaesthesia as compared to non-anesthetized patients (Tsukeoka and Tsuneizumi [@CR26]). The decrease of varus laxity comparing the postoperative non-anesthetized results to the preoperative anesthetized results may be contributed solely to the effect of anaesthesia and may in fact display the same unchanged lateral collateral laxity. Therefore, the different measurement setting is a plausible explanation for the unexpected decrease in varus laxity in this patient population. However, in valgus laxity there is an increase in laxity postoperative even though the postoperative X-rays were performed without the use of anaesthetics. Therefore, the difference in measurement settings cannot explain the increased valgus laxity. A previous study by Mains et al. studied the influence of the separate ligaments with regards to stability of the knee. They have shown that the superficial MCL (sMCL) is the main stabilizer for valgus stress. After sectioning the sMCL the abduction increased with 4.8° in 20--25° of knee flexion and for 40--45° of knee flexion this increase in abduction was 4.3° (Mains et al. [@CR20]). These findings are in correspondence with our increase in valgus laxity after the operation. Hence, we noted less increase in valgus laxity but this may again be explained due to the different setting during the radiological measurements (anaesthesia or not). An explanation for the increase in valgus laxity can be found in the surgical technique for medial CWHTO. The MCL is composed of a superficial and a deep component originating from the medial epicondyle of the femur. The deep component inserts directly into the edge of the tibia plateau and therefore will not be affected by the osteotomy. Mains et al. showed that the deep MCL tightens at 45° flexion but is less powerful as a stabilizing factor for valgus stress compared to the superficial MCL (Mains et al. [@CR20]). Griffith et al. showed that cutting the deep MCL resulted in increased valgus laxity in 60 degrees of knee flexion. This was not recorded for 0, 20, 30 or 90 degrees of knee flexion (Griffith et al. [@CR11]). The superficial component is attached to the tibia approximately 6 cm beneath the joint line and has a broad insertion up to 12 cm below the joint line (LaPrade et al. [@CR19]). The superficial MCL provides the primary resistance against valgus stress and external rotation (Griffith et al. [@CR11]; Mains et al. [@CR20]). When performing the medial CWHTO part of the sMCL is pushed dorsal at the location of the osteotomy when inserting a retractor subperiostially around the posteromedial tibia to protect the soft tissues while performing the osteotomy. Although care is taken not to loosen the MCL fibres during insertion of the retractor the removal of the bone wedge and the gap closure afterwards causes a relative lengthening of the MCL similar to the amount of bone resected on the medial cortex as the insertion of the sMCL remains unchanged (Fig. [1](#Fig1){ref-type="fig"}). So the superficial component fibers themselves have the same length but their underlying surface is shortened due to the removed bony wedge resulting in a pseudo laxity of the MCL and thereby introducing increased valgus laxity. The documented significant changes in laxity in the coronal plane have to be compared with the patient reported outcome score, to identify whether they have any clinical relevance. Patients experience walking with a deformed leg, whether valgus or varus, often as an instability in walking. Patient reported outcome measures after the medial CWHTO showed significant less knee instability and knee instability-related problems 6 months after the surgery that corrected their leg to neutral as measured with the Lysholm scoring tool. The VAS instability score, solely developed to assess the patients subjective stability feeling of the knee, also improved postoperative however, did not reach significance. Due to our short follow up time (6 months) no comparisons can be made with previous studies (Coventry [@CR5], Shoji and Insall [@CR24], Chambat et al. [@CR2], Van Egmond et al. [@CR27]) that reported on outcome after an medial CWHTO. Furthermore, no long-term conclusions can be drawn based on our outcome scores. However, the  findings do support our hypothesis that patients do not experience a disadvantage of the increase in laxity of the MCL. In our department as yet unpublished data among 113 comparable patients showed a 77% satisfaction after a medial CWHTO with a mean follow up of four years (van Lieshout WAM et al. [@CR28]). Our study has several limitations. First, because this study is a pilot study the sample size is small. Nevertheless we reached significant changes for varus and valgus laxity for pre- and postoperative results. Second, the fact that the preoperative laxity measurements were performed after either spinal or total anaesthetics and the postoperative measurements in a normal non-anesthetized state is a confounding factor as mentioned before. However, as we thought it unethical to re-apply spinal or total anaesthesia for the sole purpose of laxity measurements, this was the only way to perform the present study in our patients. Thirdly, our patient reported outcome scores were only available at six months postoperative. Therefore no long-term conclusion regarding outcome can be drawn from them. We conducted these surveys to evaluated patient reported instability before and after medial CWHTO. Since full-weight bearing without the use of walking aids was allowed after 4 weeks we believe that a normal walking pattern and reliable patient reported stability complaints can be obtained after 6 months. A last remark has to be made why we choose to perform a medial CWHTO. Our inclusion criteria were symptomatic valgus malalignment located in the proximal (high) tibia for which an indication for a medial CWHTO is the treatment of choice as deformities localized in the tibia need to be corrected in the tibia. In distal femoral valgus corrections the osteotomies are performed proximal of collateral ligaments attachments and no changes in ligament laxity will result from corrections in the femur. However, when a distal femur correction is performed in a patient with valgus localized in the tibia an unintended obliquity of the knee joint line will be created which may cause new complaints. In conclusion, this study shows a significant increase in postoperative valgus laxity in 30° and 70° of flexion after medial closing wedge HTO which deems reconsidering addition of a MCL reefingplasty to the medial CWHTO although patient reported outcome on subjective stability scores fails to report increase of instability in this study population. Instrumented laxity measurements of medial CWHTO patients treated with additional medial reefingplasty should be performed to prove the value of this procedure. CWHTO : Closing wedge high tibial osteotomy dMCL : Deep portion of the medial collateral ligament FTMA : Femoral tibial mechanical angle IKDC : International Knee Documentation Committee KOOS : Knee Injury and Osteoarthritis Outcome Score LCL : Lateral collateral ligament MCL : Medial collateral ligament OKS : Oxford knee score sMCL : Superficial portion of the medial collateral ligament VAS : Visual analogue scale none. Funding {#FPar1} ======= The first author has received and contribution from the Marti Keuning-Eckhardt Sichting (Stichting MKE) to conduct this research. Availability of data and materials {#FPar2} ================================== The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. WvL performed the radiological procedure and the radiographic measurements and was the first writer of the article; CM was the second author for the radiographic measurement and contributed in writing the article; BvG was responsible for the data acquisition, performed the statistical analysis and assisted in writing of the results section. RvH was the project supervisor and responsible for the study design, performed the surgeries, assisted in writing and revised the manuscript. All authors read and approved the final manuscript. Ethics approval and consent to participate {#FPar3} ========================================== This study was approved by the Medical Research and Ethical Committee, and all patients provided written informed consent for participation in the study. Medical ethical approval number: NL52106.048.15. Consent for publication {#FPar4} ======================= Not applicable. Competing interests {#FPar5} =================== The authors declare that they have no competing interests. Publisher's Note {#FPar6} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Introduction {#Sec1} ============ GaN-based third-generation semiconductor materials and their alloys have attracted great attention \[[@CR1]\] due to their broad applications including light-emitting diodes (LEDs) \[[@CR2]--[@CR4]\] and laser diodes (LDs) \[[@CR5]--[@CR7]\]. Although GaN-based photonic devices are widely commercialized, the relatively low hole concentration and high resistivity of p-type GaN still significantly limit the performance of such devices \[[@CR8], [@CR9]\]. Much research has been down to improve the p-type doping efficiency for III-nitrides \[[@CR10], [@CR11]\]. Hydrogen and carbon are two main residual impurities existing in the metal-organic chemical deposition (MOCVD)-grown Mg-doped GaN epilayers. It is well known that hydrogen impurities can passivate Mg in p-GaN \[[@CR12]\]. On the other hand, carbon impurities can form many kinds of defects and increase the resistivity of Mg-doped p-GaN. Much research has done to decrease the hydrogen and carbon impurities. However, there are few investigations on the interaction of hydrogen and carbon impurities. It is known that too much residual impurity of either hydrogen or carbon can cause high resistivity in as-grown Mg-doped GaN films. On account of the H-containing MOCVD growth environment, Mg is always passivated by hydrogen impurities, and a neutral Mg--H bond complex can be formed during film growth \[[@CR13]\]. Fortunately, in a remarkable way, the group of Nakamura et al. \[[@CR12]\] has firstly demonstrated that rapid thermal annealing in N~2~ ambient at a temperature \> 700 °C can successfully dissociate Mg--H complexes and effectively remove the hydrogen atoms from Mg-doped GaN films. For the past few years, with the research and development of long-wavelength multiple-quantum-well (MQW) devices, high-indium-content InGaN/GaN layers have been widely used as active layers. To avoid the segregation and structural degradation of MQW, relatively low growth temperature (\< 1000 °C) and relatively low rapid thermal annealing temperature are required. However, the unintentionally doped carbon impurity concentration increases with decreasing growth temperature, which leads to a higher concentration of carbon impurity-related defects in GaN, existing in the forms of substitutional defects (C~N~), interstitial defects (C~i~), and complexes \[[@CR14], [@CR15]\]. These defects can act as either donors or deep-acceptor species and increase the resistivity of p-GaN significantly \[[@CR16]\]. As a result, the low-temperature (LT)-grown Mg-doped p-type GaN films often show a higher resistivity than those grown at higher temperature. Contrary to our expectations, our research has found that p-GaN films with both high concentration of hydrogen and carbon impurities show relatively low resistivity. In this work, three sets of Mg-doped GaN films with different concentrations of hydrogen and carbon residual impurities are investigated through secondary ion mass spectroscopy (SIMS), photoluminescence (PL), and Hall measurements. It is found that hydrogen can passivate the carbon impurities in the p-GaN, which points out a new direction to grow high-quality p-type GaN film. Experimental Methods {#Sec2} ==================== It still remains unknown on how to control the residual hydrogen concentration by setting MOCVD growth conditions. So, our samples are divided in different groups basing on SIMS results rather than growth conditions, similar Mg concentration in each group. In this work, numbers of Mg-doped GaN films are grown on a 2-μm-thick unintentionally doped GaN layer template in a metal-organic chemical vapor deposition (MOCVD) system. Trimethylgallium (TMGa), ammonia (NH~3~), and bis-cyclopentadienyl-magnesium (Cp~2~Mg) are used as the precursors for Ga, N, and Mg, respectively. The growth temperature of all p-GaN samples is relatively low at 1020 °C. The Mg doping concentration is mainly adjusted by Cp~2~Mg flow rate. The residual carbon impurity concentration is adjusted mainly by NH~3~ flow rate during MOCVD---more NH~3~ corresponds to less carbon impurity \[[@CR17]\]. The rapid thermal annealing is carried out in a nitrogen environment at a temperature of 800 °C for 3 min to de-passivate the Mg--H complexes. Hall test is carried out to measure the resistivity of p-GaN samples. To make ohmic contact on p-type GaN, molten indium metal is pointed on a sample surface and acts as a metal electrode. To check the concentrations of magnesium, hydrogen, carbon, and oxygen impurities, \[Mg\], \[C\], \[H\], \[O\], secondary ion mass spectroscopy (SIMS) measurements of these p-GaN samples are taken. Seven samples are selected because of the suitable Mg concentration and divided in three groups, similar Mg concentration in each group, named as A1, A2, A3, B1, B2, and C1, C2. Room temperature photoluminescence (PL) measurements of all samples are carried out by the 325-nm wavelength of a He--Cd laser at an excitation density of about 0.4 W/cm^2^. The luminescence intensity is normalized by the near-band-edge emission luminescence intensity (at around 3.44 eV)^1^ for the sake of analysis. Results and Discussion {#Sec3} ====================== The results of Hall test and SIMS measurement are exhibited in Table [1](#Tab1){ref-type="table"}. Based on the SIMS results of Mg, C, and H concentration measurements, the seven samples are divided into three groups A, B, and C. Samples in each group has to be similar to Mg concentration, because Mg is the major acceptor in p-GaN and the conductivity of p-GaN is mostly caused by Mg. So, if we want to investigate the influence of H and C impurity on resistivity, we should keep Mg concentration invariability in each group. The joint influence of doping concentrations of these impurities on the sample property, mainly the p-type electrical resistivity, is analyzed. The doping concentration of magnesium in these samples is very high (in 10^19^\~3 × 10^19^ cm^−3^) and has no remarkable difference for the samples in each group. The concentration of oxygen is low enough (10^16^ cm^−3^) and can be taken out of further consideration. Table 1Resistivity and Mg, C, H, and O concentrations for Mg-doped p-GaN samplesSampleResistivity (Ω cm)Mg (cm^−3^)C (cm^−3^)H (cm^−3^)A11.392.87E+191.17 E+171.70E+18A24.953.14E+198.00E+178.70E+17A347.72.39E+191.12E+194.11E+17B11.491.10E+192.40E+164.60E+17B22.351.08E+194.13E+172.81E+18C11.951.60E+192.90E+171.50E+18C22.951.40E+194.20E+174.59E+18 In group A, the rise of carbon impurities causes a tremendous resistivity increase of the p-GaN, while in group B, the increase of hydrogen along with carbon impurities is found to weaken this trend. And group C is employed to further investigate the influence on BL band. It can be seen in Table [1](#Tab1){ref-type="table"} and Fig. [1](#Fig1){ref-type="fig"} that for samples A1--A3, the concentration of carbon impurity increases dramatically, changing two orders of magnitude from 1.17 × 10^17^ to 1.12 × 10^19^ cm^− 3^, but the concentrations of magnesium, hydrogen, and oxygen change only little. From the previous research, we realized that though the doping concentration of magnesium is very high, actually the hole concentration is still two orders of magnitude lower than magnesium because of the low ionization rate and high possibility of self-compensation \[[@CR18], [@CR19]\]. In GaN, Mg~Ga~ has an acceptor ionization energy of 260 meV \[[@CR20]\], an order of magnitude more than k~B~T (about 26 meV) in room temperature, and defects and impurities exist in GaN can compensate or passivate Mg~Ga~, so the hole concentration in Mg-doped GaN is about two orders of magnitude lower than magnesium. In addition, the residual carbon impurities can also cause negative effects to p-type GaN conductivity \[[@CR16]\]. The resistivity of p-GaN samples in series A raised obviously with increasing carbon concentration (from 1.39 to \~ 47.7 Ω cm). Therefore, the differences between samples A1--A3 can be attributed to the difference of carbon impurities. As described in our previous study \[[@CR16]\], carbon impurities may preferentially play the role of donor-type compensation centers in Mg-doped GaN films. The donors can compensate magnesium acceptors. Therefore, the resistivity of p-GaN increases with the rise of residual carbon impurities concentration. Fig. 1The resistivity of samples changes with the C concentration in groups A and B On the other side, in series B, the concentrations of magnesium and oxygen change little in each group as shown in Table [1](#Tab1){ref-type="table"} and Fig. [1](#Fig1){ref-type="fig"}. The carbon concentration of sample B1 is much higher (about 20 times) than that of sample B2. However, the resistivity of sample B2 is quite close to and not much larger than that of sample B1. This trend is different from what we have observed for group A. Hence, it suggests that this different trend of resistivity variation in two groups may be attributed to the difference in the concentration of hydrogen impurity. For samples A1--A3, hydrogen impurity concentration decreases little, with a factor of \~ 1/3, while carbon impurity concentration increases nearly two orders of magnitude. On the contrary, for samples B1--B2, the concentration of hydrogen impurity increases along with the carbon impurity. Thus, the obtained result suggests that hydrogen incorporation may weaken the influence of carbon on the resistivity of Mg-doped p-GaN, producing a counteraction effect. In order to further investigate how carbon impurity compensates magnesium acceptors and why hydrogen can weaken this process, the room temperature photoluminescence measurements were carried out. In Fig. [2](#Fig2){ref-type="fig"}a, as shown by the results of PL measurement of samples A1\~A3, a luminescence peak at about 2.9 eV can be seen obviously. This blue luminescence (BL) band has been already studied for decades. It is known that the BL band in p-GaN PL spectra around 2.9 eV has a distinct donor--acceptor pair luminescence character. For the candidate of acceptor, isolated Mg substitute of Ga defect (Mg~Ga~) is the natural choice. And the most possible candidate for the deep donor in very heavily Mg-doped GaN is a nearest neighbor complex, which is an associate of Mg~Ga~ and nitrogen vacancy (V~N~), formed by self-compensation \[[@CR21]\]. As the integral intensity of BL band decreases with higher doping of carbon impurities (Fig. [2](#Fig2){ref-type="fig"}b), we can assume that carbon impurities may decrease the number of relevant donor--acceptor pairs by compensating magnesium acceptors, because carbon impurities may preferentially play the role of donor-type compensation centers in Mg-doped GaN films \[[@CR16]\]. The appearance of a strong 2.2-eV peak for sample A3 indicates that there is a larger number of carbon-related defects in sample A3 \[[@CR15]\]. Fig. 2**a** The results of normalized PL intensity of samples A1\~A3. **b** The integral PL (solid triangle) intensity and C (solid square) and H (solid circle) concentrations for samples A1\~A3. **c** The results of normalized PL intensity of samples B1and B2 Meanwhile, regardless of a large increase of both carbon and hydrogen concentrations from B1 to B2 in sample group B, the PL spectra of these two samples are quite similar to each other. Actually, there is no obvious BL band in sample B1 and only a small BL peak in sample B2 (Fig. [2](#Fig2){ref-type="fig"}c), perhaps because of the relatively lower concentration of magnesium in series B samples (nearly 1 × 10^19^ cm^−3^) in comparison with group A samples. Therefore, the data of samples C1 and C2 are employed to check the interaction between hydrogen and carbon impurities further. It is noted that the Mg and C concentrations in sample C1 are similar to those in sample C2, and the resistivity of the two samples is also similar to each other. But it is interesting to note that BL band changes obviously in the PL spectra of sample group C. The H concentration in sample C2 is three times higher than that in sample C1. Figure [3](#Fig3){ref-type="fig"}a shows that the BL band intensity is quite different for samples C1 and C2. The intensity of BL band of C2 is much larger, which is attributed to the larger hydrogen concentration in this sample. In addition, the integral intensity of BL band increases clearly with the rise of the concentration of hydrogen, even though carbon impurity (can decrease BL band) concentration also increases a little at the same time (Fig. [3](#Fig3){ref-type="fig"}b). It implies that the reason for the increase of BL band is the increase of hydrogen impurities instead of carbon. It suggests that hydrogen and carbon may have an opposite effect on BL band of p-GaN. For hydrogen impurities, we assume that the most probable way to enhance the BL band is to form more relevant donor--acceptor pairs by forming C--H complexes with carbon impurity and passivating the carbon impurities in Mg-doped GaN. So, it is speculated that hydrogen can form complexes with carbon in the Mg-doped p-GaN sample, leading to a smaller concentration of donor-type compensation centers. In other words, hydrogen can passivate carbon and improve the conductivity of Mg-doped p-GaN. Further investigation is needed to figure out how to control hydrogen incorporation to preferentially passivate carbon impurity instead of Mg acceptors. Fig. 3**a** The normalized PL intensity of samples C1 and C2. **b** Integral PL intensity and C and H concentration for samples C1 and C2 Conclusion {#Sec4} ========== In summary, the effects of carbon and hydrogen impurities on Mg-doped GaN films were investigated. It is found that carbon impurities may preferentially play the role of donor-type compensation centers and compensate Mg acceptor in Mg-doped GaN films. An increase of the carbon doping concentration can increase resistivity of the p-GaN and weaken blue luminescence (BL) band intensity. However, when hydrogen incorporation increased with carbon doping concentration, the increase of resistivity caused by carbon impurity is weaken and the BL band intensity is enhanced, which suggests that hydrogen not only can passivate Mg~Ga~ acceptors, but also may passivate carbon by forming C--H complex with carbon impurity. GaN : Gallium nitride InGaN : Indium gallium nitride InN : Indium nitride LD : Laser diode LED : Light-emitting device Mg~Ga~ : Mg substitute of Ga defect MOCVD : Metal-organic chemical deposition MQW : Multiple quantum well NH~3~ : Ammonia SIMS : Secondary ion mass spectroscopy TMGa : Trimethylgallium TMIn : Trimethylindium V~N~ : Nitrogen vacancy **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. The authors acknowledge the support from National Key R&D Program of China (Grant Nos. 2016YFB0401801 and 2016YFB0400803), Science Challenge Project (Grant No. TZ2016003), and National Natural Science Foundation of China (Grant Nos. 61904172, 61974162, 61674138, 61674139 and 61604145). YZ and DJ were major contributors in writing the manuscript. FL, DZ, JZ, ZL, JY, and SL help in collecting and analyzing the data. All authors read and approved the final manuscript. All sources of funding for the research reported have been declared. The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. The authors declare that they have no competing interests.

Background {#Sec1} ========== The blood--brain barrier (BBB), and the other blood--tissue barrier sites of the central nervous system (CNS), have been the subject of extensive research since their discovery over 100 years ago. Despite considerable advances in our understanding of the structural and functional interface of the BBB, there remain many gaps in our knowledge particularly regarding its role in disease and the challenges it presents to therapeutic intervention. In recent decades the classic concept of the BBB has also evolved such that it now cannot be considered in isolation from other cellular components of the CNS. Accordingly, the emergence of the concept of the neurovascular unit (NVU) has re-shaped our approach to studying the BBB. In addition, other blood--tissue interfaces, such as the blood--cerebrospinal fluid and blood--retinal barriers, are also providing additional insight into the communication between the blood and the CNS. Our understanding of the normal structure and function of the blood--CNS barriers is well advanced but their roles in many diseases remains incomplete. Whereas blood--CNS dysfunction in some conditions is evident, such as in tumours, multiple sclerosis and stroke, in other diseases such as Alzheimer's disease, Parkinson's disease and epilepsy the involvement is less obvious. Indeed, gross changes, such as loss of structural integrity have clear pathological consequences, whereas subtle changes to function may be more difficult to ascertain and place within the overall pathogenesis of a disease. Whether cause or effect, therapeutic targeting of barrier dysfunction remains an attractive proposition and drives much of the translational research currently underway. However, various questions concerning barrier susceptibility to disease remain outstanding. These include the heterogeneity of the vasculature within the CNS and as a consequence its differential response. Indeed, it is known that in the normal BBB there is endothelial cell heterogeneity that is not only dependent on its position throughout the vascular bed (i.e. artery versus arteriole, versus capillary, versus venule, versus vein) but also within the same region of the vasculature. Moreover, the barrier within different structures of the CNS also differs and together such heterogeneity will undoubtedly impact on the variable response of the barrier to disease. For example, the microvascular pathology observed in diabetes is far more pronounced in the retina than in the brain, the response of white matter vessels and those in grey matter differ in multiple sclerosis, and in meningitis it is the meningeal vessels that are susceptible. Aside from the direct relationship between barrier dysfunction and disease pathogenesis there is another longstanding and major challenge facing those working in the field. This relates to the problems posed by a structurally intact barrier that restricts the delivery of therapeutics to the brain. For almost 50 years this has proved to be largely insurmountable and only recently have advances been made that provide some optimism. In this CNS barrier congress experts from various disciplines were brought together to collectively discuss the best ways to overcome these challenges, and pave the way for progress in the treatment of neurological disease. In recent years, our understanding of barriers has undergone re-evaluation and during the meeting various pressing questions were discussed. These included the role that non-endothelial cells in the NVU play in blood--brain barrier regulation, how much barrier dysfunction really occurs in different CNS diseases, why regional differences exist, and how do immune cells impact barrier function. Contrary to previous dogma, the CNS barriers are now recognised as complex, dynamic, interactive structures that contribute to disease on many levels. Recent advances in drug delivery technologies to the CNS were also presented and discussed at length. Pioneering groups have been perfecting new methods to ferry drugs across the CNS barriers, particularly the blood--brain barrier where access via other routes is problematic. Accordingly, the latest developments in liposome, peptide, antibody, and nanoparticle technology for therapeutic delivery were showcased and how far these technologies have to go before they can become widely available was discussed. This CNS barrier congress allowed for the presentation of unpublished data, the exploration of new technologies, and provided a select platform for academic and industrial leaders in the field to form collaborations, exchange ideas and identify new strategies for development. The specialised vascular barriers of the CNS and their influence on leukocyte migration {#Sec2} ======================================================================================= John Greenwood {#Sec3} -------------- The specialised vascular endothelial cells that line the vessels of the brain and retina form an impermeable but selective barrier between the blood and the neural parenchyma. Under normal physiological conditions this critical interface, termed the blood--brain/retinal barrier, strictly limits the passage of solutes and cells between these two compartments. During disease, however, the endothelial cells become activated resulting in a change of phenotype and an alteration in their regulatory function. Thus, in neuroinflammatory diseases such as multiple sclerosis and posterior uveitis, the function of these vascular barriers changes resulting in an enhanced influx of leukocytes. Accordingly, the endothelial cells of the CNS are recognised as playing a pro-active role in the propagation, maintenance and possibly resolution of CNS inflammatory lesions. Over the last few decades increasing evidence, from our laboratory and that of our collaborators, has shown that the endothelial cell responds to adherent leukocytes in variety of ways resulting in immediate facilitation of diapedesis to the longer-term regulation of gene expression. Many of these outside-in signaling cascades are generated through the engagement of endothelial immunoglobulin superfamily adhesion molecules such as ICAM-1, which act as signal transducers leading to the activation of the small GTPase rho, eNOS, phospholipase C, protein kinase C, src kinase and release of intracellular calcium \[[@CR1]\]. In addition, we have reported more recently that downstream activation of MAP kinases, re-arrangements of the actin cytoskeleton and tyrosine phosphorylation of various cytoskeletal associated proteins results in the activation of divergent inflammatory outcomes \[[@CR2]\]. Finally, we have established that the tightness of the endothelial cell junction and cell cytoskeletal stiffness dictates the route of leukocyte transmigration \[[@CR3]\]. Deciphering the end-points of these signaling networks and identifying potential pharmacological targets, remains a major focus of the laboratory. Non-VEGF mediated breakdown of the blood--retinal barrier: alternative strategies to treat diabetic macular oedema {#Sec4} ================================================================================================================== Alan W. Stitt {#Sec5} ------------- Diabetic macular oedema (DMO) occurs as a symptom of diabetic retinopathy and often leads to significant vision-loss \[[@CR4]\]. The condition is characterised by progressive breakdown of the blood retinal barrier (BRB) as a result of tissue ischaemia and/or inflammation which drive imbalances in vasoactive cytokines and growth factors causing compromise of normal neuroglial--vascular interactions and endothelial dysfunction \[[@CR4]\]. Neutralisation of vascular endothelial growth factor (VEGF) using intravitreal injection of humanised antibodies has become a mainstream treatment for DMO. Unfortunately this is not effective for all DMO patients \[[@CR5]\] and there is a need for additional therapeutic approaches which could be used instead of, or in conjunction with, current anti-VEGF drugs. Our groups have recently focused on the permeability-inducing agent lysophosphatidylcholine (LPC) which is produced through activity of the enzyme lipoprotein-associated phospholipase A~2~ (Lp-PLA~2~). Using a range of in vitro and in vivo approaches, we have recently shown that inhibition of Lp-PLA~2~ can prevent diabetes-induced compromise of the BRB in a manner that is comparable to intravitreal VEGF neutralisation \[[@CR6]\]. Importantly, these protective effects were additive when both targets were inhibited simultaneously. Our mechanistic studies also demonstrated that LPC potently induced permeability, and that there was a coalescence of the LPC and VEGF pathways via a common VEGF-receptor-2 mediated mechanism \[[@CR6], [@CR7]\]. We have concluded that Lp-PLA~2~ may be a useful therapeutic target for patients with DMO, perhaps in combination with currently administered anti-VEGF agents. Such studies demonstrate the utility of studying "real-world" clinical scenarios whereby new approaches can be evaluated alongside the current gold-standard therapies and offer hope for patients who are non-responsive to current treatment regimes. The choroid plexus is an important player in the induction of neuroinflammation {#Sec6} =============================================================================== Roosmarijn Vandenbrouke {#Sec7} ----------------------- The choroid plexus epithelium which forms the blood--CSF barrier is a unique single layer of epithelial cells situated at the interface between blood and cerebrospinal fluid (CSF). The choroid plexus epithelium has several different important functions: it forms a barrier to protect the brain from fluctuations in the blood, produces CSF and is responsible for the active removal of toxic molecules from the brain and thereby assures brain homeostasis. In recent years, the choroid plexus epithelium has gained increasing attention, especially its role in different pathologies. Indeed, subtle changes in the choroid plexus epithelial cells will result in changes in CSF composition, exerting wide-ranging effects on the brain and subsequently affecting disease progression. Therefore, understanding blood--CSF barrier functionality under physiological and pathophysiological conditions might open up new therapeutic strategies to treat inflammatory brain diseases. Our research focusses on the effect of both systemic inflammation (sepsis) and neuroinflammation (the age-related disease Alzheimer's disease) on the blood--CSF barrier. More specifically, we study key molecules that play a role in the activation of detrimental processes at the blood--CSF barrier upon inflammation, focussing on barrier integrity and secretory activity. These studies might allow us to identify novel therapeutic strategies to prevent neuroinflammation. To address the effect of sepsis, we used intraperitoneal lipopolysaccharide (LPS) injection and the caecal ligation and puncture (CLP) mouse model, while Alzheimer's disease was studied using the intracerebroventricular (icv) injection of soluble amyloid β oligomers (AβO) in mice. Our studies showed that both systemic \[[@CR7]\] and central inflammation \[[@CR8]\] induce increased blood--CSF barrier leakage. Interestingly, these effects could be attributed to matrix metalloproteinase (MMP) activity. Additionally, peripheral inflammatory triggers induced an increase in extracellular vesicle (EV) release by the choroid plexus epithelium into the CSF \[[@CR9]\]. Detailed analysis by electron microscopy and inhibitor studies using the neutral sphingomyelinase 2 inhibitor GW4869, revealed that especially the biogenesis of exosomes is increased upon systemic inflammation. Strikingly, these choroid plexus-derived EVs are able to enter the brain parenchyma, are taken up by astrocytes and microglia and transfer a pro-inflammatory message to the brain \[[@CR9]\]. Interestingly, we observed that the icv injection of AβO has similar effects on the extracellular vesicle production of the choroid plexus epithelial cells. Clearly, our data show that both peripheral and central inflammatory triggers affect barrier and secretory activity of the choroid plexus epithelium and these results might open up new therapeutic strategies to treat neuroinflammatory diseases. MicroRNAs and blood--brain barrier function in multiple sclerosis {#Sec8} ================================================================= Ignacio A. Romero {#Sec9} ----------------- Blood--brain barrier dysfunction is a major hallmark of many CNS disorders such as multiple sclerosis. BBB breakdown is characterised by three main features: (1) increased permeability across the endothelium; (2) alteration in the expression of cell-surface receptors and/or transporters; and (3) activation of endothelial cells to support leukocyte extravasation into the CNS parenchyma. Many of these cerebrovascular pathophysiological effects are underpinned by overt acute or chronic changes in gene expression in cerebral endothelial cells and in other cells of the NVU. MicroRNAs (miRs) are novel regulators of gene expression at the post-transcriptional level and may potentially play a key role in cerebrovascular pathophysiology. MiRs mainly suppress the expression of target genes either by blocking translation or by inducing mRNA degradation. We have first identified miRs whose levels in endothelial cells change following inflammatory stimuli. Inflammation induces upregulation of several key inflammatory miRs termed inflammiRs (miR-155 and miR-146) in brain endothelial cells. By contrast, there are other miRs termed brain endothelial housekeeping miRs (miR-125b and miR126) whose levels are elevated under quiescent conditions but are significantly reduced by pro-inflammatory cytokines \[[@CR10]--[@CR12]\]. These inflammation-induced changes in the finely-tuned balance of cerebral endothelial miR levels promote cerebrovascular dysfunction. For example, miR-155 contributes to BBB leakiness by reducing expression of tight junctional and focal adhesion components but it also promotes leukocyte firm adhesion to brain endothelium by indirectly increasing expression of cell adhesion molecules. Conversely, miR-146 inhibits leukocyte firm adhesion to brain endothelium by suppressing expression of key activators of the NF-κB pathway. Brain endothelial miRs could potentially be considered for targeted prophylaxis and therapies for BBB dysfunction. However, we are still far from validating cerebrovascular miRs as potential therapeutic or prophylactic targets for neurovascular dysfunction in inflammatory and/or autoimmune disorders. First, there is the likelihood that several different miRs have combinatorial effects in specific CNS pathologies. Second, individual miRs likely have multiple gene targets and effects on other endothelial cellular pathways that contribute to the pathogenesis of inflammatory disease. In addition, non-specific delivery of miR modulators into other tissues or organs (e.g. liver, kidney) could cause unwanted side effects unless very specific delivery systems targeted at the cerebrovascular bed are available, Nevertheless, the potential for manipulating this novel class of regulators of gene expression for therapeutic purposes is huge and should be given considerable attention in the near future. Dose dependent expression of claudin-5 is a modifying factor in neurological diseases {#Sec10} ===================================================================================== Matthew Campbell {#Sec11} ---------------- Abnormalities in neuronal functioning in psychiatric conditions may derive in part from abnormalities in blood vessel--neuron interactions \[[@CR13]\]. We have shown that the claudin-5 gene, a central component of the paracellular pathway of the BBB is associated with schizophrenia in individuals with the chromosomal disorder 22q11 deletion syndrome (22q11DS), a condition that confers a 30-fold increased lifetime risk of developing schizophrenia. A variant in this gene results in up to 50% less protein product and a targeted suppression of claudin-5 in distinct brain regions, or in an inducible "knockdown" mouse, identifies strong phenotypic correlations with schizophrenia. Additionally, post-mortem human schizophrenia donor brain tissues show evidence of BBB dysfunction, while some of the most common anti-psychotic drugs can directly regulate claudin-5 expression. Identifying the underlying cause of neuropsychiatric conditions at the level of the BBB suggests novel drug entities targeting the integrity of the BBB may have utility in treating these debilitating and socially isolating condition \[[@CR14], [@CR15]\]. All work involving animals was approved by the institutional ethics committee and all national licences were in place prior to commencement of work. Impact of ABC transporters on blood--brain barrier function {#Sec12} =========================================================== Gert Fricker {#Sec13} ------------ The CNS requires a well-balanced homeostasis. Therefore, it is protected by the BBB, which is set up by endothelial cells of brain microvessels. These cells act as a dynamic regulator of ion balance, mediator of nutrient transport and barrier to harmful molecules. A central role accords to ATP-binding-cassette (ABC) export proteins, predominantly P-glycoprotein (P-gp, ABCB1), breast cancer resistance protein (BCRP, ABCG2) and multidrug resistance related proteins (MRPs, ABCCs). ABC transporters are predominantly expressed in microvessel endothelial cells, but are also located on other cell types, such as astrocytes, microglia, and neurons. In microvessel endothelial cells ABC transporters exhibit a polar distribution. P-gp is primarily found in luminal membranes, however, there is evidence that it also localizes to a certain amount to abluminal membranes as well as to pericytes and astrocytes. Studies using a fluorescent labeled construct of P-gp indicate that the export pump is not organized as a single molecule within membranes but forms clusters of several proteins close together \[[@CR16]\]. BCRP appears to be predominantly located at the luminal surface of endothelial cells. Mrps are also expressed at the BBB, however, there is still considerable discussion about the extent of expression, involvement in drug transport and subcellular localization. Inhibition of ABC transporters significantly alters the brain disposition of transporter substrates as illustrated by studies in human: positron emission tomography (PET)-imaging revealed that cyclosporin A modulation of P-gp increased transfer of verapamil into the brain. In another PET-trial uptake of loperamide into the CNS was enhanced after inhibition of efflux pumps by tariquidar. Various mechanisms control expression and function of ABC transporters. Regulation occurs either by ligand-activated transcription factors or by external stress signals which modulate ABC transporter function. The nuclear receptors PXR (pregnane X receptor), AhR (arylhydroxycarbon-receptor), CAR (constitutive andro-stane receptor), VDR (vitamin D receptor) and PPAR-γ (peroxisome proliferator-activated receptor) and related signaling events are of particular interest, since they bind many xenobiotics and subsequently upregulate ABC-transporter expression. ABC transporters also contribute to various CNS diseases. For example, brain tissue from Alzheimer patients showed an inverse relationship between P-gp expression and disease progress and inhibition of P-gp in a rodent Alzheimer model increased amyloid-β levels in brain. Studies showed that St. John's Wort administration to mice resulted in significant reductions of parenchymal amyloid-β accumulation as well as in a moderate increase in cerebrovascular P-gp expression \[[@CR17]\]. In summary, ABC-transporters are of outstanding relevance for the proper function of the blood--brain barrier. They are very sensitive to exogenous and endogenous stimuli and are involved in the progression of various CNS diseases. In vitro models of the blood--brain barrier, pro's and con's {#Sec14} ============================================================ Birger Brodin {#Sec15} ------------- The brain capillary endothelium serves as a gateway for exchange of nutrients, hormones and metabolites between plasma and brain parenchyma, and acts as a barrier for CNS uptake of the majority of drug compounds. Mechanistic in vivo investigations of drug and nutrient transport, signalling and metabolism in the brain endothelium can be difficult to perform, since the brain endothelium is embedded within a complex structure*, the neurovascular unit*, itself within the multi-compartment brain and fluid system. Cell isolation and culture protocols for growing brain endothelial cells in monolayers, either in monoculture or in co-culture with other cell types of the NVU, have been developed over the last 40 years. Although no ideal cell culture model of the blood--brain barrier is yet available, the in vitro models have evolved to become useful tools in the studies of barrier biology and drug delivery. In vitro models based on primary cell cultures of animal origin (typically bovine, porcine or murine models) display endothelial cell morphology, expression of BBB tight junction proteins and high transendothelial electrical resistance, subject to astrocyte induction \[[@CR18]\]. They display vectorial transport of efflux transporter substrates, indicative of luminal expression of ABC-type pumps \[[@CR19]\]. The major drawback of these models may be downregulation of a number of solute carrier family proteins (SLC's) as compared to the expression in vivo. Alternatively, immortalized cell lines of mouse, rat and human origin, in general without astrocyte induction, tend to be considerably more leaky than the in vitro models based on primary cell cultures. Some immortalized cell lines do, however, show relatively higher expression of some BBB-specific SLC transporters than the models based on primary cell cultures and may serve as excellent tools for studies of uptake, receptor activation and some signal transduction systems. A recent approach, generation of in vitro models of the BBB using human stem cells, has resulted in cell monolayers with both high monolayer tightness and expression of BBB-specific marker proteins \[[@CR20]\]. However, the stem cell cultures are not yet widely used, and have not been fully validated with respect to transport properties and functional transporter profiling. In brief, established in vitro blood--brain barrier models all have their advantages and drawbacks. Primary cell lines of bovine, porcine and rodent origin generate tight monolayers whereas immortalized cell lines may have higher transporter expression levels but less tightness. Models derived from human stem cells shows great promise, but are not yet fully characterized. Measuring blood--brain barrier transport of drugs---the hurdle in drug discovery and development {#Sec16} ================================================================================================ Margareta Hammarlund-Udenaes {#Sec17} ---------------------------- Measurement of total drug levels in the brain has been a common but unhelpful practice for many years in drug discovery programs aiming at central drug effects. The paradigm has changed with the introduction of the pharmacologically more important unbound brain interstitial fluid to unbound blood concentration ratio, K~p,uu,brain~, and with more high-throughput methods to estimate this parameter (the brain slice and the brain homogenate techniques). By combining several of these measurements, the combinatory mapping approach (CMA) allows estimation of not only BBB transport, but also intracellular distribution of the pharmacologically active, unbound drug moiety \[[@CR21]\]. CMA can also be used to assess possible lysosomal accumulation that may predict phospholipidosis as a serious side effect. The technique allows estimation of brain regional differences in BBB transport and binding, something that can influence effect/side effect patterns of drugs. In a study of six antipsychotic drugs, we found very different BBB and intracellular distribution patterns of the drugs \[[@CR22]\]. There was a sixfold difference in regional BBB transport of risperidone (a P-glycoprotein substrate), with a more efficient efflux in cerebellum than in frontal cortex in rats, while other drugs such as quetiapine and clozapine showed very small differences in regional transport. In a separate study utilizing the CMA, we did not find any change in the BBB transport of five selected drugs in disease models of Alzheimer's disease (ArcSwe) and Parkinson's disease (A30P) compared with wild-type mice, counteracting the view that the BBB is leaky in disease. Microdialysis was used to study if liposomal delivery would improve the brain uptake of methotrexate \[[@CR23]\]. With microdialysis it was possible to separate the liposomally-encapsulated drug from released drug in blood, and to measure the released compound in brain. While liposomes based on hydrogenated soy phosphatidylcholine did not change BBB uptake at all compared with administering the free drug, the egg-yolk phosphatidylcholine liposomes increased the uptake into the brain threefold, reaching concentrations that could be pharmacologically active also in humans. In conclusion, there are three main conceptual parts of brain drug transport, the rate of transport, measured as the permeability surface area product, the extent of transport measured with K~p,uu,brain~, and the nonspecific intra-brain binding, all contributing to different aspects of brain drug delivery. The extent of transport is considered clinically most relevant, as it measures the steady state relationship across the BBB, quantifying active efflux and influx processes. NEUWAY pharma: qualified for CNS delivery {#Sec18} ========================================= Heiko Manninga {#Sec19} -------------- NEUWAY Pharma GmbH, a German based biotech company, is focusing on the preclinical and clinical development of innovative therapeutics for treatment of brain diseases based on its proprietary CNS Drug Delivery Platform. The presented platform is based on a protein derived from the JC-Virus, which naturally forms capsules, named Engineered Protein Capsules (EPCs). EPCs may be used as carrier to transport highly active drug substances---ranging from small molecules to large nucleic acid strands---across the intact BBB. NEUWAY has demonstrated that its EPC-based proprietary CNS drug delivery platform can deliver plasmids over the blood brain barrier into CNS cells leading to gene expression in the brain. In a proof of concept study it was shown that intravenous injection of EPCs loaded with DNA encoding for luciferase induced an activity of the resulting enzyme, which was detected by bioluminescent signals in the brain (Fig. [1](#Fig1){ref-type="fig"}). More accurate studies of the brains of such treated mice clearly showed that the signals from the brain cells are beyond the blood--brain barrier and not from other cells, e.g., blood vessels.Fig. 1The enzyme luciferase produces light when the conversion of luciferin occurs. If a plasmid coding for luciferase is administered intravenously, no light signal can be detected (left mouse). The enzyme was degraded in the bloodstream. If the luciferase plasmid packaged in Engineered Protein Capsules (EPCs) is applied directly into the brain (intracerebral, i.c., middle mouse), a luminous signal can be detected there after administration of luciferin. This is also observed when the luciferase plasmid is packaged in EPCs and injected intravenously (i.v., right mouse) As EPCs can carry large nucleic acid strands, they may be useful for gene therapy of rare diseases. NEUWAYs current development focuses on lysosomal storage diseases, like metachromatic leucodystrophy. For this and other rare diseases, NEUWAY plans to run its own clinical development programs. NEUWAY is also open to partner its CNS drug delivery platform with pharmaceutical or biotech companies preferably if large indications, like Alzheimer's disease, are addressed. Investigations conformed to the Guide for the Care and Use of Laboratory Animals published by the US National Institutes of Health (NIH Publication No. 85-23, revised 1996). All animal procedures were approved by the appropriate state agency (Protocol Number W17/11). Liposomal technology and the blood--brain barrier {#Sec20} ================================================= Pieter J. Gaillard {#Sec21} ------------------ 2-BBB is successfully harnessing liposomal technology to mediate safe targeting and enhanced drug delivery to the brain. The G-Technology^®^ platform has been shown to enhance transport of therapeutics across the blood--brain barrier. Preclinical studies with the lead product, 2B3-101, containing doxorubicin, demonstrated significantly enhanced delivery and improved survival of mice in comparison to currently available compounds. Phase I clinical trials with 2B3-101 have been completed, with Phase IIa set to determine preliminary antitumor efficacy at the maximum tolerated dose. The study population of this phase I/IIa trial consisted of patients who met all of the inclusion criteria and none of the exclusion criteria, and provided written informed consent. The protocol, any amendments and all other applicable study documents for the study were reviewed by the Independent Ethics Committees (IECs) of the following countries: NL: The "Nederlands Kanker Instituut Antoni van Leeuwenhoek Ziekenhuis"; BE: The "Association Hospitaliere de Bruxelles-Centre des Tumeurs de l'ULB"; USA: The "Office of Human Research Ethics of the University of North Carolina at Chapel Hill"; FR: The "Comite De Protection Des Personnes Ile De France III". Clinical trial identifier is NCT01386580. 2-BBB's second product in development, 2B3-201, containing methylprednisolone, is designed to treat patients with acute neuroinflammation. Superior efficacy, reduced side effects and enhanced plasma circulation half-life have been shown in rodent models, in comparison to competitive compounds. A phase I study in healthy volunteers was completed, addressing safety, tolerability and pharmacokinetics, and markers for pharmacological proof-of-concept. Collectively, preclinical and clinical evidence to date has demonstrated that G-Technology^®^ offers a promising platform to safely enhance delivery of drugs to the brain. The study was approved by the Medical Ethics Committee of the BEBO Foundation (Assen, The Netherlands). Subjects provided written informed consent. Clinical trial identifier is NCT02048358. The blood--brain barrier in gene therapy: hurdle or target? {#Sec22} =========================================================== Markus Schwaninger {#Sec23} ------------------ Gene therapy provides attractive therapeutic options for the many diseases for which no treatment is still available. The adeno-associated virus (AAV) is a safe and efficient tool to transfer genes. However, conventional AAV-based vectors do not cross the BBB when administered systemically. This obstacle can be solved by administering vectors either intrathecally or using vectors based on AAV serotype 9 that cross the BBB to a limited extent. Still another approach is to target the BBB itself. In order to develop a brain endothelial selective vector, a novel strategy for in vivo screening of random ligand libraries displayed on viral capsids was used \[[@CR24]\]. Several rounds of in vivo selection resulted in AAV-BR1, an AAV vector with unprecedented selectivity for brain endothelial cells after systemic intravenous administration. Due to its specific features, this vector allows for modulating and repairing the BBB. Its efficacy was tested in a mouse model of the hereditary disease incontinentia pigmenti \[[@CR25]\], in which a deficiency of the Nemo gene leads to a loss of brain endothelial cells and breakdown of the BBB. Consequently, patients suffer from neurological disability and epileptic seizures. Intravenous injection of AAV-BR1 transferring the Nemo gene was able to largely reduce endothelial cell death and to ameliorate disruption of the BBB. Mice treated with AAV-BR1-Nemo showed less activation of astrocytes. Importantly, the occurrence of focal epileptic seizures was significantly reduced by the gene therapy \[[@CR26]\]. Probably due to its high brain endothelial selectivity, the vector did not induce hepatocellular carcinoma or other adverse effects that have been observed in rodents during gene therapy with AAV vectors. Previous studies suggest that transduced brain endothelial cells may release enzymes that are missing in the CNS \[[@CR27]\]. Thus, transducing brain endothelial cells with gene vectors offers the opportunity to overcome the BBB and to supply diverse proteins to the diseased brain (Fig. [2](#Fig2){ref-type="fig"}). The conceptual progress of this approach is that the BBB is no longer considered as a hurdle but as the target of a successful therapy.Fig. 2After intravenous administration of the vector AAV-BR1-eGFP (enhanced Green Fluorescent Protein) to mice most brain endothelial cells expressed eGFP. eGFP expression in other tissues was low. In exchange of eGFP other genes can be selectively expressed in brain endothelial cells. The figure shows a representative section of the thalamus stained for the endothelial cell marker CD31 and eGFP Antibody therapeutics for CNS diseases and their delivery across the BBB {#Sec24} ======================================================================== Carl Webster {#Sec25} ------------ The BBB protects and regulates the homeostasis of the brain. However, this barrier also limits the access of drugs, including large molecule therapeutics, to the brain and results in sub-therapeutic concentrations of drug reaching CNS targets. MedImmune have utilised their antibody engineering platform to develop a fully human monoclonal antibody that targets the blood brain barrier to deliver therapeutics to the CNS. The antibody was isolated from a phage display library by competitive elution using the known BBB transporting antibody FC5. A combination of pharmacodynamic (PD) and pharmacokinetic (PK) assays were used to confirm central penetration. In mice, the peripheral PK was in line with the expected half-life for a human antibody and showed no target mediated clearance at doses from 0.45 to 45 mg/kg. Brain penetration was around 3% of injected dose and persisted beyond 7 days. To confirm the antibody present in the CNS was able to access central targets, a fusion to interleukin-1 receptor antagonist (IL-1RA) was made. IL-1 is mediator of neuropathic pain and its antagonism is analgesic. Partial ligation of the sciatic nerve results in a neuropathic pain phenotype that manifests through increased sensitivity to mechanical pressure on the hind paw. Peripheral administration of the BBB-IL-1RA fusion resulted in dose dependent analgesia, confirming the ability of the BBB antibody to penetrate the CNS. However, the doses required were high and this fusion protein would not represent a commercially viable treatment for neuropathic pain. Other molecular targets represent better points of intervention in the pain signalling pathways and therefore an antibody against the ATP gated ion channel P2X4 was developed. P2X4 regulates the physiology and pathophysiology of spinal microglia and is implicated in pain signalling. Channel blocking antibodies against mouse P2X4 were obtained by immunising rats with recombinant mouse P2X4. Approximately 5000 hybridoma lines binding to P2X4 were identified of which 28 showed blocking activity. Lead antibodies were tested by intrathecal administration to mice where analgesia in the neuropathic pain model was demonstrated. Peripheral administration resulted in analgesia only when fused to the BBB antibody, confirming central exposure of the drug molecule was required. Doses as low as 3 mg/kg produce statistically significant analgesia, and this offers hope of a treatment to many people suffering from neuropathic pain. All procedures described here were performed in accordance with the Animals (Scientific Procedures) Act 1986 and were approved by the MedImmune local ethics committee. Harnessing bispecific antibodies to overcome the blood--brain barrier {#Sec26} ===================================================================== Krzysztof B. Wicher {#Sec27} ------------------- Penetration of the BBB remains a significant impediment in development of biologics for CNS-related diseases. To identify efficient brain-shuttles, we used shark single domain antibody phage libraries in a combined in vitro and in vivo selection process. To achieve the highest discovery rate of brain-penetrant clones, we employed next generation sequencing of phagemid DNA after subsequent rounds of selection. This way, we identified a panel of TfR1-specific shark antibodies, which are efficiently transported to mouse brain parenchyma. One of these antibodies (B2) fused to human IgG1 Fc showed more than 12-fold better brain uptake than the control, reaching therapeutic concentrations (5 nM) upon single intravenous injection of 25 nmol/kg (\~ 1.9 mg/kg). Most of the brain-associated B2-hFc was found in brain parenchyma. Immunohistochemistry analyses showed the protein present both in the intestinal fluid and in the cells, including neurons. B2 antibody is safe upon administration of high doses in mice, as manifested by lack of significant acute adverse reactions or changes in blood morphology of injected animals. It binds to human TfR1 with comparable affinity as to mouse TfR1, but does not bind to TfR2. Moreover, in silico analysis indicates that B2 antibody would have relatively low immunogenic properties in humans. We next constructed several different variants of bi-specific antibodies composed of B2 and rituximab, a well characterized anti human CD20 specific antibody used to treat peripheral B cell lymphomas. Many of these variants shuttle efficiently to brain and provide up to 14× better exposure than original antibody. The hybrid proteins retain their binding to both TfR1 and CD20 and at least some of them appear to mediate antibody-dependent cell cytotoxicity response on human CD20^+ve^, but not on CD20^−ve^, cells, similar to that of rituximab. Thus, they offer possible therapeutics for multiple sclerosis and cerebral B-cell lymphoma. All research procedures/experiments described here were performed in accordance with Animals Scientific Procedures Act 1986 and European Directive 2010/63/EU. All studies performed were approved by the Royal Veterinary College Animal Welfare and Ethics Review Body and comply with the UK Home Office guidelines and codes of conduct. Drug delivery across the blood--brain barrier using peptide conjugates {#Sec28} ====================================================================== Michel Khrestchatisky {#Sec29} --------------------- Drug delivery to the brain is hindered by the BBB. Receptor-mediated transport/transcytosis (RMT) can be used to shuttle therapeutics into the brain in a non-invasive manner. We developed peptide- and nanobody-based ligands that target specific receptors and that can be used as vector molecules to transport drugs or imaging agents across the BBB. Members of the low density lipoprotein receptor (LDLR) family appear relevant to deliver drugs into cells and organs and we report results on the development of peptide-vectors that target the rodent and human LDLR. Initial screening of complex peptide libraries followed by chemical optimization led to the development of a family of short cyclic peptides (eight natural or non-natural residues) with distinct properties in terms of affinity, stability and biodistribution. Real time two-photon microscopy experiments on mice demonstrated the ability of a lead peptide-vector to transport a non-permeable agent such as RhoRedX across the BBB and the blood--spinal cord barrier. As a further proof of concept, following intravenous (iv) administration in mice, peptide-vectors efficiently transported to the brain molecules such as opiate peptides or neuropeptides, all known to poorly cross the BBB. In particular, a vector-neurotensin conjugate is under preclinical development for its potential to induce pharmacological hypothermia with neuroprotective effects in acute excitotoxic neurodegeneration. Some tumors including glioblastoma are associated with high-level expression of receptors involved in cell metabolism such as the LDLR. Conjugating our peptide-vectors to ^68^Gallium-NODAGA or -DOTA allowed PET imaging of glioblastoma in mouse brain. Others have shown that nanoparticles or liposomes functionalized with one of our lead peptide-vectors (peptide-22) permeate the BBB and exhibit higher glioma distribution than non-functionalized nanoparticles or liposomes; functionalized nanoparticles or liposomes loaded with paclitaxel and doxorubicin respectively elicit significantly prolonged life span of glioma-bearing mice \[[@CR27], [@CR28]\]. The lead peptide vectors also allowed uptake of vectorized protein cargos such as an antibody Fc fragment by brain endothelial cells and transport across an in vitro BBB model, and in vivo into the brain following iv administration in mice \[[@CR29]\]. The peptide-vectors are currently assessed with industrial partners for their potential to transport therapeutic antibodies into the brain. In summary, we have developed a family of chemically optimized peptide-vectors that can be conjugated to a variety of different compounds such as small organic molecules, peptides, siRNAs and proteins including therapeutic antibodies. These peptide vectors bind the rodent and human LDLR and promote transport of cargos across the BBB and brain uptake in rodents. Such peptide-vectors appear promising for CNS delivery of different classes of drugs. Conclusions {#Sec30} =========== From the variety of talks, it was clear that significant progress is being made in understanding control processes at brain barrier interfaces and in novel drug delivery methods. Modification of the normal barrier processes occurs in inflammatory situations through signalling cascades which can be manipulated, barrier gene expression can be modified with micro RNAs and control can be exerted over barrier transporter functions. Techniques for enhancing delivery of therapeutics to the CNS involve cleverly-devised delivery vehicles such as drug-carrying liposomes, endothelial-specific viral vectors, peptide conjugates, engineered protein capsules, nanoparticles, bi-specific antibodies and human monoclonal antibodies. Methods to test the efficacy of therapeutic delivery were discussed using both in vitro and in vivo techniques. Overall this was a stimulating conference bringing together scientists from a number of different disciplines. Authors contributed with an short report of their contribution to the conference. JG, HCJ and MH-U wrote the Abstract, Background and Conclusions. All authors read and approved the final manuscript. Authors' information {#FPar1} ==================== HCJ is a Co-Editor of *Fluids and Barriers of the CNS.* Acknowledgements {#FPar2} ================ MH-U: The contributions of Irena Loryan and Yang Hu are mush appreciated. AWS: The collaboration of Patric Turowski (UCL Institute of Ophthalmology, London, UK) is gratefully acknowledged. IAR: The collaboration of C Cerutti, MA Lopez-Ramirez, and D Wu (Open University, UK) is gratefully acknowledged. HM: The author would like to thank staff at Fraunhofer Institute for Cell Therapy and Immunology (Department of Cell Therapy, Experimental Imaging) for performing the animal studies, particularly: Alexander Kranz and Franziska Werner. MS: The collaboration of Jakob Körbelin (UKE, Hamburg, Germany) is gratefully acknowledged. CW: The author would like to thank staff at MedImmune Cambridge who contributed to this work, particularly: George Thom, Jon Hatcher, Iain Chessell and Wendy Williams. KBW: The collaboration of Paweł Stocki, Jarosław Szary and Mykhaylo Demydchuk (Ossianix Inc., Stevenage UK) and of Frank S. Walsh and J. Lynn Rutkowski (Ossianix Inc., Philadelphia, USA) is gratefully acknowledged. MK: The collaboration of Marion David, Pascaline Lécorché, Cedric Malicet, Yves Molino, Maxime Masse, Guillaume Jacquot and Jamal Temsamani (Vect-Horus, Marseille, France) and Jonathan Nowak, Romy Cohen, Anne Bernard (Aix Marseille Université, CNRS, NICN, Marseille, France) is gratefully acknowledged. Competing interests {#FPar3} =================== JG, MH-U, HCJ, REV, IAR, MC, GF, BB, MS declares that they have no competing interests. AWS: Part of the work described was funded by GlaxoSmithKline plc. HM: All work described in this abstract was funded by NEUWAY Pharma GmbH. HM is employee and shareholder in NEUWAY Pharma GmbH, is named inventor on patent (applications) covering the described work, and therefore has a theoretical competing interests. PJG: All work described in this abstract was funded by to-BBB technologies BV, and is currently owned by 2-BBB Medicines BV. PJG was employee and shareholder into-BBB technologies BV and is currently shareholder in 2-BBB Medicines BV, is named inventor on patent (applications) covering the described work, and therefore has a theoretical competing interests. CW: All work described in this abstract was funded by MedImmune. All collaborators at the time of the work were employees of MedImmune and therefore have a theoretical competing interests. KBW: The author and collaborators were employed by Ossianix, Inc. at the time of completion of this work. MK is Research Director at the CNRS and Director of the NICN laboratory supported by the CNRS and Aix-Marseille Université, and also co-founder, shareholder and scientific counsel of the biotechnology company Vect-Horus. Availability of data and materials {#FPar4} ================================== Not applicable. Consent for publication {#FPar5} ======================= Not applicable. Ethics approval {#FPar6} =============== Ethical approval was obtained for all studies described in this manuscript. Specifically for unpublished data, the approval details are indicated at the end of each contribution. Funding {#FPar7} ======= The papers in this publication were presented at a conference organised by Kisaco Research held March 22--23rd London, UK <http://www.kisacoresearch.com>. Publisher's Note {#FPar8} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

INTRODUCTION ============ Partial nephrectomy has equivalent long-term oncological outcomes and superior renal functional outcomes to radical nephrectomy. Furthermore, the length of the tumor-free margin has minimal clinical significance; a normal tissue margin of 1 mm might be sufficient to prevent local recurrence and disease progression of renal cell carcinoma.^[@B1]--[@B3]^ Minimally invasive nephron-sparing surgery techniques are evolving continuously, and recent comparative studies have demonstrated favorable-to-equivalent outcomes for robot-assisted partial nephrectomy (RAPN) compared with laparoscopic partial nephrectomy (LPN), particularly with respect to warm ischemia time.^[@B4],[@B5]^ These advantages of RAPN are likely a result of the demonstrated advantages offered by the da Vinci robotic platform (Intuitive Surgical, Sunnyvale, CA), such as 3-D vision and 7 degrees of freedom of the surgical instruments, which can overcome the technical limitations of LPN.^[@B6]--[@B8]^ However, it is still unclear whether RAPN has clinical advantages compared with LPN when performed by an experienced laparoscopic surgeon. To address this question, we analyzed outcomes after LPN and RAPN by a single experienced laparoscopic surgeon. MATERIALS AND METHODS ===================== Data for 100 consecutive patients who underwent transperitoneal RAPN (n=48) or LPN (n=52) in a university-based tertiary referral center between October 2007 and June 2009 by a single laparoscopic surgeon were analyzed retrospectively. Inclusion criteria for LPN were a single organ-confined mass of ≤7 cm. All patients had a normal contralateral kidney. Patients younger than 18 years and those with preoperative renal dysfunction (defined as serum creatinine \>1.4 mg/dL), tumor \>7 cm, or preoperative radiographic evidence suspicious for lymph node or distant metastases were excluded from study. For each patient, the choice of operative methods between the 2 minimally invasive approaches was based on a joint decision by the surgeon and patient, who was appropriately informed about the surgical procedures and complications. The preoperative workup included routine laboratory tests and radiological evaluation with contrast-enhanced computed tomography or magnetic resonance imaging to delineate parameters of the renal lesion, including tumor location, size, and extent of protuberance. One radiologist reviewed the imaging studies to categorize the renal mass as exophytic, endophytic, or mesophytic, as previously defined by Finley et al.^[@B9]^ In addition, tumors were classified according to the preoperative aspects and dimensions used for an anatomical classification of renal tumors (PADUA).^[@B10]^ The operation time was defined as the time from when the skin incision was made until the time when the wound was closed. Pathology assessment included histological analysis of the lesion and calculation of the volume of marginal healthy renal tissue. We evaluated the excised volume of the marginal healthy tissue by subtracting the tumor volume from the total volume of the resected specimen calculated using the ellipsoid formula: width × height × length × π/6. To evaluate functional outcomes, serum creatinine levels were measured at 1, 3, 6, and 12 months after surgery, and estimated glomerular filtration rate (eGFR) was estimated using the abbreviated Modification of Diet in Renal Disease study equation.^[@B11]^ Surgical complications were classified according to the Clavien classification system updated in 2004.^[@B12]^ Patient demographics and perioperative parameters were compared between groups using the Mann-Whitney *U* test, Student *t* test, χ^2^ test, analysis of variance, or Fisher exact test as appropriate. SPSS version 18.0 for Windows (IBM, Armonk, NY) was used for statistical analyses, and *P*\<.05 was considered statistically significant. We described our techniques and trocar arrangement for RAPN and LPN in detail in a previous publication.^[@B13]^ Next we highlight some important variations in the techniques we used in the current study. RAPN Technique -------------- A 4-arm technique was used in all cases. Specifically, the camera port was placed approximately 6 cm above the umbilicus to midline (renal hilar level). The 12-mm assistant port was placed just above the umbilicus to the midline. The first 8-mm working port was placed at the junction of the costal margin and the lateral edge plane of the rectus muscle. The second 8-mm working port was placed in the 2-finger medial point of the anterior superior iliac spine at an approximately 120° angle toward the patient leg from the lateral edge of the rectus muscle of the camera port level. The third 8-mm port was placed 5 cm below the umbilicus in the midline. Another 5-mm port was placed for liver retraction on the right side. A laparoscopic ultrasound probe was used to plan the resection margin. The renal hilum was accurately isolated, and then either the artery only or the entire hilum was clamped without cooling in all cases. To reduce the likelihood of ischemic damage, all patients received proper hydration and mannitol infusion (0.25 g/kg) 10 minutes before clamping. LPN Technique ------------- Techniques and trocar arrangement for LPN have been previously described in the literature.^[@B14]^ Three ports (2×12 mm, 1×5 mm) were placed in a triangular shape. Another 10-mm port was placed on the posterior axillary line for assistance, and, if necessary, an additional 5-mm port was used for liver retraction on the right side. The operative technique was similar to that described for RAPN. RESULTS ======= Patient demographic data are provided in **[Table 1](#T1){ref-type="table"}**. There were 33 male and 19 female patients with a mean age of 51.1 (range: 21--72) years in the LPN group, and 34 male and 14 female patients with a mean age of 50.9 (range: 24--75) years in the RAPN group. The median follow-up duration was 16.2 (range: 2--29) months for both groups. There were no significant differences in body mass index; American Society of Anesthesiologists score; tumor laterality, size, or location protuberance; comorbidity including diabetes, hypertension, and others; clinical stage; or peritoneal operation history between the 2 groups. ###### Demographic Characteristics Characteristics LPN (n=52) RAPN (n=48) *P* Value ---------------------------------------- ------------------- ------------------- ----------- Gender (male/female) 33/19 34/14 .662 Mean age (y) (range) 51.1 (21--72) 50.9 (24--75) .992 Median follow-up duration (mo) (range) 16.2 (2--29) 8.9 (2--29) \<.001 Mean BMI (kg/m^2^) (range) 25.2 (18.0--31.5) 24.4 (16.7--31.8) .255 Mean ASA score (range) 1.5 (1--3) 1.5 (1--2) .668 Laterality (left/right) 26/26 23/25 .838 Tumor size (cm) (range) 2.23 (0.8--5.7) 2.47 (0.7--5.1) .102 Tumor location .781     Upper 15 (28.8) 13 (27.1)     Mid 24 (46.2) 21 (43.8)     Lower 13 (25.0) 14 (29.1) Tumor depth .864     Exophytic 16 (30.8) 16 (33.3)     Endophytic 16 (30.8) 13 (27.1)     Mesophytic 20 (38.4) 19 (39.6) PADUA score 7.8 (6--10) 8 (6--10) .095 Clinical stage .700     T1a 49 44     T1b 3 4 No. of comorbidities .218     0 22 (42.3) 30 (62.5)     1 27 (51.9) 16 (33.3)     \>1 3 (5.8) 2 (4.2) Previous operation history 16 (30.8) 9 (18.8) .254 BMI = body mass index; ASA = American Society of Anesthesiologists. In terms of subset analysis based on PADUA score, there was no significant difference between the 2 groups (7.8 vs 8.0, *P*=.095) The perioperative outcomes are shown in **[Table 2](#T2){ref-type="table"}**. Mean operative time, estimated blood loss (EBL), and warm ischemic time (WIT) were similar between the groups (*P*\>.05). With respect to the type of vascular clamping (artery only, total occlusion), more artery-only clamping was performed in the RAPN group than in the LPN group (75 vs 38.5%, respectively, *P*=.001). LPN and RAPN were completed successfully in all patients, and conversion to open surgery was not needed in any case. Intraoperative adverse events occurred in 4 LPN patients (7.7%) (2 transfusions because of bleeding, 1 polar artery ligation, 1 ureter injury requiring primary repair) and in 2 RAPN patients (4.2%) (1 transfusion, 1 spleen minor injury) (*P*=.679). Postoperative complications in the LPN group included 5 Clavien grade I complications (2 cases of pulmonary edema requiring diuretics, one case of hepatotoxicity defined as an elevation in serum aspartate aminotransferase or alanine aminotransferase level from the normal range after surgery, 2 cases of electrolyte unbalance), 1 Clavien grade II complication (seizure requiring medical treatment), and 1 Clavien grade III complication (urine leak requiring insertion of a double-pigtail stent). Adverse events in the RAPN group included 4 Clavien grade I complications (2 cases of hepatotoxicity, 2 cases of electrolyte unbalance), and 1 Clavien grade II complication (postoperative transfusion required because of gross hematuria for intrarenal hematoma rather than an arteriovenous fistula or a pseudoaneurysm). There were no differences in perioperative changes in hemoglobin levels or length of hospital stay between the 2 groups. No recurrences were observed in either the RAPN or LPN group until the final follow-up period. ###### Perioperative Outcomes Characteristics LPN (n=52) RAPN (n=48) *P* Value ------------------------------------------------------------- --------------------------- ------------------ ----------- Mean operative time (min) (range) 263.8 (123--454) 258.6 (119--469) .871 Mean EBL (mL) (range) 207.6 (20--600) 217.4 (50--700) .362 Mean WIT (min) (range) 33.4 (19--74) 32.1 (19--58) .855 Clamping method .001     Artery-only 20 (38.5) 36 (75)     Total occlusion 32 (61.5) 12 (25) Intraoperative complication 4 (7.7) 2 (4.2) .679     Transfusion 2 1     Open conversion 0 0     Other complication Polar artery ligation = 1 Spleen minor Ureter injury = 1 tearing = 1 Postoperative complication[\*](#TF2-1){ref-type="table-fn"} 7 (13.5) 5 (10.4) .761     Grade I 5 4     Grade II 1 1     Grade ≥III 1 0 Hospital stay (days) (range) 8.2 (5--13) 7.8 (6--9) .177 Mean preoperative Hb (g/dL) (range) 14.1 (10.5--16.8) 14.2 (9.3--17.1) .829 Mean postoperative Hb at discharge (g/dL) (range) 13.6 (8.9--16.4) 13.5 (9.0--16.5) .783 \% decrease of Hb --3.9 --5.0 .321 No recurrence (n) 52 (100) 48 (100) NS Hb = hemoglobin. Clavien classification. The pathological results are presented in **[Table 3](#T3){ref-type="table"}**. The final pathology revealed renal cell carcinoma in 86.5% (45/52) of LPN patients and 85.4% (41/48) of RAPN patients. Fuhrmann grade was not significantly different between the 2 groups. In the LPN group, there were 98% (44/45) pT1a and 2% (1/45) pT1b tumors, whereas in the RAPN group, there were 93% (38/41) pT1a, 5% (2/41) pT1b, and 2% (1/41) pT3a tumors. Although two patients in the LPN group had focal positive surgical margins (PSMs), no patients in the RAPN group had PSMs (*P*=.581). One of 2 patients with PSMs underwent radical nephrectomy immediately, and there was no pathological evidence of remnant tumor in the nephrectomy specimen. The other is free of recurrence 29 months after surgery with surveillance. The mean final tumor volume was significantly different between the 2 groups (4.0 cm^3^ in the LPN group vs 8.2 \[range: 1.6--33\] cm^3^ in the RAPN group, *P*=.006). In contrast, the mean resected healthy tissue volume was greater in the LPN group than in the RAPN group (25.1 vs 16.1 cm^3^, respectively, *P*=.044). The mean greatest safety margin was significantly different between the groups (5.3 mm for the LPN group vs 2.2 mm for the RAPN group, *P*=.009). Both mean preoperative serum creatinine (mg/dL) and eGFR (mL/min) were similar between the groups (0.90 vs 0.87 mg/dL and 90.1 vs 94.0 mL/min/1.73 m^2^, *P*=357 and *P*=.241, respectively). There were also no significant differences in serum creatinine levels or eGFR at 1, 6, and 12 months after surgery between the LPN and RAPN groups (*P*\>.05). ###### Pathological Outcomes Characteristics LPN (n=52) RAPN (n=48) *P* Value ----------------------------------------------------- --------------------------- ------------------ ----------- Malignant (RCC) ---     Clear cell 39 (86.7) 35 (85.4)     Papillary 2 (4.4) 3 (7.3)     Chromophobe 3 (6.7) 3 (7.3)     Unclassified 1 (2.2) 0 Other malignancy Ewing\'s sarcoma = 1 --- Benign ---     Oncocytoma 3     Angiomyolipoma 4 4 Urinoma = 1 Inflammatory myofibroblastic tumor = 1 Fuhrmann grade .171     I 1 (2.2) 0     II 22 (48.9) 16 (39.0)     III 22 (48.9) 25 (61.0)     IV 0 0 Positive surgical margin 2 0 .581 Stage (pT) .581     pT1a 44 38     pT1b 1 2     pT3a 0 1 Mean greatest safety margin (mm) 5.3 2.2 .009 Mean resected healthy tissue volume (cm^3^) (range) 25.1 (1.9--92.1) 16.1 (0.6--71.2) .044 Mean tumor volume (cm^3^) (range) 4.0 (0.3--25.6) 8.2 (1.6--33) .006 RCC = renal cell carcinoma. DISCUSSION ========== Although nephron-sparing surgery is not the standard treatment option for small renal tumors from an oncological perspective, it seems to preserve renal function, thereby improving the long-term overall health outcomes of patients.^[@B2],[@B15],[@B16]^ LPN has recently emerged as a viable alternative for the surgical management of small renal masses, with oncological and functional outcomes similar to open partial nephrectomy.^[@B17],[@B18]^ However, LPN is technically challenging; advanced laparoscopic skills are required to perform intracorporeal laparoscopic suturing at often difficult angles and repair the defect under the time constraints of warm ischemia.^[@B19]^ Recently, RAPN has begun to emerge as a paradigm in renal oncological surgery. The development of the da Vinci surgical system (Intuitive Surgical) has enabled more urologists to overcome the challenges associated with complex laparoscopic reconstructive procedures.^[@B5],[@B20]^ Recent studies demonstrated that the early outcomes of RAPN are at least comparable with those achieved using the traditional laparoscopic approach, whereas other studies have reported that RAPN is superior to the traditional laparoscopic approach.^[@B5],[@B20],[@B21]^ We conducted the present study to compare a single surgeon\'s experience of conventional LPN with his experience of RAPN. In a previous report, RAPN had significant advantages compared with LPN, including shorter ischemic times and a shorter hospital stay, based on analysis of the LPN results obtained by a surgeon with intensive laparoscopic training and RAPN results for initial robotic cases.^[@B5]^ In another multi-institutional retrospective review of these 2 approaches (118 LPN cases, 129 RAPN cases) based on procedures performed by 3 experienced minimally invasive surgeons, RAPN was shown to be a safe and viable alternative to LPN, providing the advantages of decreased hospital stay as well as significantly less EBL and shorter WIT.^[@B4]^ However, Jeong et al^[@B22]^ reported a shorter operative time for LPN than for RAPN (139 vs 169 minutes, respectively, *P*=.034), whereas tumor size, EBL, WIT, hospital stay, and associated morbidity were not significantly different between the 2 groups. In our study, no significant difference was found between the 2 groups with regard to age, body mass index, comorbidity, radiological tumor size, location, extent of protuberance, or even PADUA score. Although the perioperative surgical complications measured in this study fail to support a significant advantage of RAPN over LPN, there is a trend toward fewer complications using the robotic approach. Although the pathological tumor volume was larger in RAPN cases than in LPN cases, the resected healthy tissue volume was smaller in the RAPN group ([Table 3](#T3){ref-type="table"}). The reason for this result is not entirely clear. In our opinion, this may be due to more comfortable instrumentation and better visualization during RAPN compared with LPN, such as 3-D stereoscopic vision with easier angles and multijointed wrested instruments.^[@B7],[@B8],[@B21]^ Therefore, RAPN may enable the surgeon to excise closer to the tumor and thus preserve more healthy renal tissue. Moreover, it is possible that the poor ergonomics and technical limitations of LPN increased the surgeon\'s anxiety about positive surgical margins, resulting in a larger margin of normal tissue surrounding the tumor. In a previous multi-institutional study comparing 118 LPN and 129 RAPN cases, although there was a 4-fold increase in the incidence of positive margins in the RAPN group compared with the LPN group, this result was not statistically significant (3.9% vs 0.8%, respectively, *P*=.11).^[@B4]^ The authors of this multi-institutional study suggested that this could have been a result of the learning curve for RAPN not having been surpassed. No positive margin case for our RAPN series may be attributed to the previously cumulative experiences of LPN cases and the robotic technical advantages. Furthermore, our data clearly indicate that regardless of tumor characteristics, RAPN sacrifices less healthy tissue volume than LPN. Therefore, if we can save more renal parenchymal tissue with RAPN, parenchymal suturing at the hilar area can be done more easily because there are sufficient tissues for renorrhaphy. We believe this is one of the technical advantages of RAPN. We found a similar mean WIT between the LPN and RAPN groups. This result may have been contributed by an immature renal hilar control of laparoscopy-naïve assistants during RAPN. The assistants involved in the RAPN procedure are crucial to the success and outcome of surgery. During LPN, the surgeon has to control the renal pedicle by him- or herself, whereas in RAPN, the bedside surgeon places the bulldog clamps on the renal pedicle and creates countertension during tumor excision. Therefore, the bedside surgeon during RAPN should be well experienced in standard laparoscopy and be trained in the use of laparoscopic instruments. We believe that our slightly prolonged WIT of 32 minutes during RAPN compared with that reported by published RLPN series (mean WIT of 19--28 minutes) will decrease as the experience of the operator and assistant increases.^[@B4],[@B19]^ More artery-alone clamping methods were used during RAPN than during LPN in our study. Recent studies have suggested that artery-only occlusion during LPN may provide superior renal preservation, and we have therefore shifted our focus to artery-only occlusion when performing surgery.^[@B22]^ In the present study, focal microscopic positive margins on permanent section were encountered in 2 of 52 (4%) LPN patients but in no RAPN patients. These findings are consistent with previous reports (positive margin rate 1.3% to 6.3% for LPN, 0% for RAPN).^[@B4],[@B5],[@B24],[@B25]^ Partial nephrectomy may still result in some decrease in overall renal function. We found that the extent of postoperative decrease in renal function was similar between the 2 groups in terms of both serum creatinine levels and eGFR (Table 4). However, our renal functional results may potentially have been masked by the compensatory role of the normal contralateral kidney. In addition, renal volume reduction was a significant prognostic factor for functional reduction in previous reports.^[@B12],[@B26]^ We believe that the substantial reduction in healthy tissue volume---from 25.1 cm^3^ overall in the LPN group to 16.1 cm^3^ overall in the RAPN group---represents perhaps the most important parameter that differed between the groups. However, the actual clinical benefit of this reduction in healthy tissue volume remains speculative because long-term functional data are needed to evaluate the actual impact of reduced healthy tissue volume on long-term functional outcomes. Our study had several limitations. It was a retrospective nonrandomized study, which can introduce a number of unknown biases; although there was no difference in tumor size as a single measurement, there was a significant difference in tumor volume, which is a more accurate reflection of tumor size than single greatest dimension. Second, there is the bias of sequential series. Because most of the robotic cases were done after the surgeon was more experienced in pure laparoscopic surgery, the more recent robotic cases should be expected to be faster with fewer complications. Third, our results are based on a single surgeon\'s experiences. Ultimately, a randomized, controlled, multicenter trial is needed to confirm our results. Additional long-term comparative studies to assess cancer control and long-term renal function outcomes are also needed. Finally, our sample size may not have been powered well enough to demonstrate differences in complications, and most of the tumors treated in this study were clinical T1a lesions. Therefore, future studies should explore whether the advantages we noted for these smaller lesions would also apply to larger and more complex lesions. CONCLUSION ========== Our results demonstrate that RAPN is a safe and viable alternative to LPN for small renal tumors, with comparable surgical and functional outcomes. In particular, the resected healthy tissue volume was smaller in the RAPN group than in the LPN group. Although this appears advantageous, there is still no significant difference with respect to serum creatinine/eGFR between the groups, and further follow-up is necessary.

Introduction {#s1} ============ Rhabdomyolysis can be defined as disruption of skeletal muscle leading to leakage of intracellular muscle constituents especially myoglobin into the extracellular fluid [@pone.0007182-HuertaAlardn1]. Myoglobin is a heme containing compound with a molecular weight of 17,400 Daltons and may cause acute renal failure by tubular obstruction, toxicity of heme pigments, renal ischemia and decreased glomerular permeability [@pone.0007182-Smith1]. Bywaters and Beal reported the first clinical syndrome compatible with rhabdomyolysis due to crush injuries in the year 1941 [@pone.0007182-Bywaters1]. Though initially attributed to severe trauma etiology of rhabdomyolysis now includes severe physical exertion, drugs and toxins, muscle hypoxia, hereditary muscle enzyme defects, metabolic and endocrine disorders, infections and temperature alterations [@pone.0007182-Allison1]. Acute renal failure complicates 10 to 67% cases of rhabdomyolysis and is an important cause of mortality [@pone.0007182-Ward1]--[@pone.0007182-Cadnapaphornchai1]. The primary clinical event in the pathogenesis of rhabdomyolysis is elevated creatine kinase (CK) which is often used by clinicians to screen patients who are at risk for rhabdomyolysis. However the extent of CK elevation which may be associated with renal failure is not clear. Published reports have used a value of 1000 to 2500 IU/L (5 to 10 times the upper limit of normal) as a clinically significant elevation which may predispose to acute renal failure [@pone.0007182-Betrosian1], [@pone.0007182-Fernandez1]. But values of CK as low as 519 IU/L have been associated with marked elevation in serum myoglobin and biopsy proven renal failure due to rhabdomyolysis [@pone.0007182-Asao1]. Sepsis is an important cause of morbidity and mortality [@pone.0007182-Dellinger1]. Association between sepsis and rhabdomyolysis has been shown in multiple case reports [@pone.0007182-Ojukwu1]--[@pone.0007182-Bando1]. But only a few original researches are available on the spectrum of infective pathogens associated with rhabdomyolysis [@pone.0007182-Betrosian1]. This study was initiated to identify the epidemiology and laboratory profile of rhabdomyolysis due to bacterial sepsis in a developing nation and analyze the differences with published literature. Materials and Methods {#s2} ===================== Study setting and design {#s2a} ------------------------ Retrospective cohort study of adult patients with community acquired bacterial sepsis complicated by rhabdomyolysis from 2 tertiary care centers in South India from March 2003 to August 2008. The study was approved by the Sri Ramachandra University institutional review board. Written informed consent was obtained from all the study patients. For patients who did not survive when this retrospective study was conducted, a written and informed consent was obtained from the patients nearest relative. Sepsis was defined as systemic inflammatory response syndrome with history or clinical examination suggestive of infection [@pone.0007182-Dellinger1]. Rhabdomyolysis was defined as creatine kinase (CK) more than 2000 IU/L (10 times the upper limit of normal). The presence of myoglobinuria was considered desirable but not mandatory for the diagnosis of rhabdomyolysis in our study cohort since myoglobin released after muscle injury circulates for a brief period (half life -- 2 to 3 hours) and is rapidly excreted by the kidney [@pone.0007182-HuertaAlardn1] making estimation of serum or urine myoglobin a less sensitive test for the diagnosis of rhabdomyolysis. Patients with sepsis due to non-bacterial causes, hospital acquired sepsis, trauma, acute coronary syndrome, cerebrovascular accident, post-operative status, post-cardiac arrest, hyperglycemic crisis, hypothyroidism and malignant hyperthermia were not included. Method of patient evaluation {#s2b} ---------------------------- Hospital records of consecutive patients who satisfied the entry criteria were scrutinized for baseline characteristics (age, sex, current smoking status, current alcohol intake, diabetes mellitus, hypertension and current statin use), laboratory findings relevant to rhabdomyolysis (CK, urine myoglobin, admission electrolyte panel including calcium and phosphorus, serial serum creatinine values, blood gases, plasma lactate and uric acid), investigations to identify foci of sepsis (chest X-ray, ultrasound abdomen and cultures of blood, urine, cerebrospinal fluid, etc) and final outcome (death or survival). The study cohort was divided into 3 groups. Group-I (sepsis with gram positive pathogens), group--II (sepsis with gram negative pathogens) and group-III (culture negative sepsis). Statistical analysis {#s2c} -------------------- Baseline characteristics of study patients were expressed in number (%) for discrete variables, and as mean±standard deviation for continuous variables. Laboratory findings (CK, urine myoglobin, serum sodium, potassium, bicarbonate, calcium and phosphorus, serial serum creatinine values, blood gases, plasma lactate and uric acid) in different groups were compared using one-way analysis of variance or Kruskal Wallis analysis of variance. Multiple comparison between groups were done using Turkey HSD (honesty significance difference) test. A p value of \<0.05 was considered statistically significant. All statistical analysis were done using SPSS version 15. Results {#s3} ======= 103 patients formed the study cohort. Baseline characteristics of the study patients are shown in [**Table 1**](#pone-0007182-t001){ref-type="table"}. Mean age of our study cohort was 55 years with a predominant male distribution (60%). 47% were smokers and 41% were alcoholics. Nearly two-thirds (64%) were suffering from diabetes mellitus and more than a third (43%) had hypertension. 30% of our study cohort were on statins. Among diabetic patients 35% were on statins. 10.1371/journal.pone.0007182.t001 ###### Baseline characteristics of study cohort (n = 103). {#pone-0007182-t001-1} Variable Group-I Group-II Group --III P value -------------------------- --------- ---------- ------------- --------- Age in years -- mean±S.D 58±6 55±8 55±9 0.5 Sex Male- no (%) 11(73) 18 (53) 33 (61) 0.39 Female- no (%) 4 (27) 16 (47) 21 (39) Smoking- no (%) 11 (73) 11 (32) 26 (48) 0.02 Alcoholism- no (%) 5 (33) 14 (41) 23 (43) 0.8 Diabetes- no (%) 10 (67) 22 (65) 34 (63) 0.9 Hypertension--no (%) 6 (40) 16 (47) 22 (41) 0.82 Statin use -no (%) 4 (27) 11 (32) 16 (30) 0.91 Group --I: Gram positive sepsis. Group-II: Gram negative sepsis. Group-III: Culture negative sepsis. The laboratory profile of the study cohort is shown in [**Table 2**](#pone-0007182-t002){ref-type="table"}. The mean CK was 7114 IU/L and urine myoglobin was positive in 45%. Mean values of serum electrolytes were: sodium (Na) −136 meq/L, potassium (K) −3.7 meq/L, bicarbonate −13.3 meq/L (was \<15 meq/l in 50%), calcium (Ca) −8.4 meq/L and phosphorus −4.5 meq/L. Mean serum creatinine on admission was 2.4 mg/dl and progressive increase in serum creatinine was observed in all groups. On admission the mean arterial Ph was 7.18 (46% had ph \<7.2), mean lactate −2.8 meq/L and mean corrected anion gap −16 meq/L. 27.2% had hypokalemia defined as serum potassium \<3 meq/l, 8% had hypernatremia defined as serum sodium \>150 meq/l, 2.9% had hypophosphatemia defined as serum phosphate \<2 mg/dl. None had severe hyponatremia defined as serum sodium \<120 meq/l. 81 of 103 (78.6%) of our patients had one or more additional causal factor for rhabdomyolysis. The frequency of additional risk factor for rhabdomyolysis in our study patients is shown in [**Table 3**](#pone-0007182-t003){ref-type="table"}. All patients received intensive care. 10.1371/journal.pone.0007182.t002 ###### Laboratory profile of the study cohort. {#pone-0007182-t002-2} Variable Group I Group II Group III p value ----------------------- ----------- ------------ ----------- --------- CK (IU/L) 8219±6022 8936±13109 5660±4131 0.18 Myoglobinuria 7 (47) 14 (41) 25 (46) 0.88 S.Na (meq/L) 132±8 136±7 137±9 0.15 S.K (meq/ 3.9±1.7 3.6±1 3.7±0.9 0.6 S.Bicarbonate (meq/L) 14±3.6 13±4.7 13±4.7 0.4 S.Ca (mg/dl) 8±0.7 8.4±0.7 8.3±0.8 0.7 S.Phosporus (mg/dl) 4.7±1.1 4.2±1.1 4.5±1.1 0.34 S.Uric acid (mg/dl) 7.8±1.1 7.1±1.6 7.2±1.7 0.35 Anion gap (meq/L) 16±2.7 15±4 16±3.7 0.43 S. creatinine (mg/dl) Admission 2.5±1.1 2.3±0.6 2.5±1 0.71 Dialysis day 4.2±1.8 5.3±3 5.1±2.8 0.47 Arterial Ph 7.14±0.15 7.2±0.15 7.17±0.14 0.52 Plasma lactate(meq/l) 2.8±0.6 2.8±0.9 2.7±0.8 0.86 10.1371/journal.pone.0007182.t003 ###### Frequency of additional risk factors for rhabdomyolysis. {#pone-0007182-t003-3} Risk factor Frequency -------------------- ----------- Chronic alcoholism 41% Statin intake 30% Hypokalemia 27.2% Hypernatremia 8% Hypophosphatemia 2.9% Analysis of variance revealed that there was no difference between the study groups (group-I to group-III) with respect to mean CK, frequency of myoglobinuria, admission creatinine, dialysis day creatinine, arterial Ph, plasma lactate, serum sodium, potassium, bicarbonate, calcium, phosphorus, uric acid and corrected anion gap (level of significance shown in [**Table 2**](#pone-0007182-t002){ref-type="table"}). Further, multiple comparisons between groups by Turkey HSD test on laboratory values mentioned above showed no statistically significant differences. Isolated pathogens and foci of sepsis among culture positive patients are shown in [**Table 4**](#pone-0007182-t004){ref-type="table"}. Lung was the commonest foci (38.8%). The cause of sepsis could be identified in 47.5% of cases. Gram negative sepsis was more frequent (33%) than gram positive sepsis (14.5%). All patients required dialysis. 61 of 103 (59%) patients died. 10.1371/journal.pone.0007182.t004 ###### Isolated pathogens and foci of sepsis among culture positive patients (n = 49). {#pone-0007182-t004-4} Foci of sepsis Group I Group II ----------------------- ------------------ ------------------------- Lung (n = 20) S. pneumoniae: 4 P. aerogenosa: 4 S. aureus: 2 Enterobacter species: 1 S. viridans: 1 K. pneumoniae: 5 E. coli: 3 Meninges (n = 10) S. aureus:1 E. coli: 2 S. pneumoniae: 2 P. aerogenosa: 3 P. mirabilis: 2 Urinary tract (n = 9) Nil P. aerogenosa: 3 K. pneumoniae: 2 E. coli: 3 A. baumanni: 1 Endocardium (n = 6) S. aureus: 3 E. coli: 2 S. viridans: 1 No foci (n = 4) S. pneumoniae: 1 P. aerogenosa: 1 E. coli: 2 S. pneumoniae: Streptococcus pneumoniae. S. aureus: Staphylococcus aureus. S. viridans: Streptococcus viridans. P. aerogenosa: Pseudomonas aerogenosa. K. pneumoniae: Klebsiella pneumoniae. E. coli: Escherichia coli. P. mirabilis: Proteus mirabilis. A. baumanni: Acinatobacter baumanni. Discussion {#s4} ========== Infections accounted for 5% of rhabdomyolysis in the literature 25 years ago [@pone.0007182-Gabow1], [@pone.0007182-Veenstra1]. Betrosian et al. [@pone.0007182-Betrosian1], using a strict entry criteria for rhabdomyolysis, compared the pattern of rhabdomyolysis due to bacterial sepsis (n = 35) with patients having sepsis without rhabdomyolysis (n = 122) and observed that infections with gram positive pathogens (especially Staphylococcus aureus and Streptococcus faecalis) more frequently caused rhabdomyolysis than gram negative pathogens with lung being the foci of sepsis in most patients (34%) followed by urosepsis, gall bladder infection, pancreatitis and catheter related infections. In contrast our study has shown the dominance of gram negative pathogens (especially Pseudomonas, E.coli and Klebsiella) over gram positive pathogens. Similar to their study lung was the most frequent (38.8%) foci of sepsis in our cohort. We did not observe gall bladder infection or pancreatitis probably due to selection bias as these patients are predominantly managed by surgeons in our study centers. Further, their study [@pone.0007182-Betrosian1] used a CK level more than 2500 IU/L as definition for rhabdomyolysis and excluded patients with acidosis defined as serum bicarbonate less than 15 meq/L in addition to alternative causes for rhabdomyolysis (trauma, surgery, dysnatremia, hypokalemia, hypophospatemia) since they were attempting to correlate plasma osmolality with levels of CK. We did not exclude patients with bicarbonate less than 15 meq/L, dysnatremia, hypokalemia and hypophosphatemia since these abnormalities commonly co-exist with sepsis and our objective was to identify all contributing factors to rhabdomyolysis in patients with sepsis. Furthermore the baseline characteristics of their study cohort [@pone.0007182-Betrosian1] with reference to co-morbidities like diabetes, hypertension, chronic alcohol intake, concomitant statin intake is not known. It should be noted that significant number of our patients were suffering from diabetes (64%) and hypertension (43%). Given this fact it is not surprising that 30% of our study patients were on statins which can independently cause rhabdomyolysis. Further our patients had co-existing causal factors for rhabdomyolysis like chronic alcoholism (41%), hypokalemia (27.2%), hypernatremia (8%) and hypophosphatemia (2.9%). The predominance of gram negative pathogens in our study can be partially explained by the high proportion of diabetic patients. The proposed mechanism of rhabdomyolysis due to sepsis are (i) direct muscle invasion by the pathogen, (ii) toxin generation, (iii) cytokine mediated muscle cell toxicity, and (iv) muscle ischemia due to shock [@pone.0007182-Allison1], [@pone.0007182-Gabow1], [@pone.0007182-Baracos1]--[@pone.0007182-Morton1]. Rhabdomyolysis due to direct muscle invasion and toxin generation is classical with Staphylococcus aureus [@pone.0007182-Wu1]. Streptococci, Salmonella and Staph. aureus have been demonstrated in muscle biopsies of patients with rhabdomyolysis while Legionella, a frequent pathogen associated with rhabdomyolysis has not been demonstrated in muscle biopsies indicating a probable distant mechanism [@pone.0007182-Allison1]. The mean CK and presenting biochemical investigations between our sub-groups showed no significant differences. This emphasizes the fact that despite different possible mechanisms adopted by microbes the pattern of final clinical presentation of rhabdomyolysis is similar. The pathogens identified in our study are identical to previous published literature except Acinetobacter which has not been previously reported. Bacterial pathogens associated with rhabdomyolysis are Legionella, Streptococcus pneumoniae, Staphylococcus aureus, Streptococcus viridans, Salmonella species, Staphylococcus epidermidis, Francisella tularensis, Streptococcus faecalis, Meningococci, Hemophilus influenza, E.coli, Pseudomonas, Klebsiella, Enterococcus faecalis and Bacteroides [@pone.0007182-Allison1], [@pone.0007182-Betrosian1], [@pone.0007182-Asao1], [@pone.0007182-Singh1], [@pone.0007182-Funada1], [@pone.0007182-vanDeuren1]. Lung was the most frequent (38.8%) foci of sepsis in our study. This is consistent with previous reports which show a close relation between lung sepsis and rhabdomyolysis [@pone.0007182-Betrosian1], [@pone.0007182-Caron1], [@pone.0007182-Garcia1]. As discussed earlier statins, chronic alcoholism, hypokalemia, hypernatremia and hypophosphatemia were additional causes which could have facilitated the occurrence of rhabdomyolysis in our study cohort. Combination of causal factors in patients with rhabdomyolysis is well known [@pone.0007182-Gabow1]. Statins are supposed to induce defective glycoprotein synthesis in muscle membrane, deficient chloride channel activation in muscle membrane and increased intracellular calcium leading to impaired membrane function, all leading to myocyte injury [@pone.0007182-Evans1]. Our study cohort is notable for the high proportion of diabetics and 35% of them were on statins. The possibility of statin intake precipitating rhabdomyolysis in diabetic patients with sepsis cannot be overlooked since diabetes is a known risk factor for statin induced myopathy [@pone.0007182-Thompson1]. Rhabdomyolysis due to acute alcoholism is well known [@pone.0007182-Allison1]. Rhabdomyolysis may also occur secondary to chronic alcohol intake [@pone.0007182-Qiu1]. The proposed mechanism for rhabdomyolysis due to chronic alcoholism are electrolyte disturbances like hypokalemia, hypophosphatemia and hypomagnesemia as evidenced by marked depletion of intracellular potassium, phosphorus and magnesium in skeletal muscles of chronic alcoholics and experimental animals fed with alcohol [@pone.0007182-Knochel1], [@pone.0007182-Ferguson1]. Potassium depletion can decrease glycogen synthesis within muscle and also impair potassium dependent vasodilatory response [@pone.0007182-Knochel2]. This combination of diminished glycogen reserve and impaired oxygen delivery may lead to ATP depletion producing muscle injury [@pone.0007182-Allison1]. Hypophosphatemia may also limit ATP production and produce an energy deficient state leading to muscle injury [@pone.0007182-Allison1]. The mortality in our study cohort was 59%. Since our study cohort had frequent co-morbid conditions the mortality observed cannot be interpolated to a relatively healthy population. All our study patients received hemodialysis probably due to the presence of multiple disorders and progressive raise in serum creatinine warranting aggressive management by the treating physician to avoid unfavorable outcomes. We preferred not to analyze the factors contributing to death since treatment of rhabdomyolysis varied between physicians in our study. Though alkalinization of urine and diuretics are frequently used to treat rhabdomyolysis their efficacy in preventing progression of renal failure is unknown. Recent evidence highlights the lack of efficacy and possible harm with bicarbonate and diuretics [@pone.0007182-Brown1]. Our study did not analyze the severity of illness using Acute Physiology and Chronic Health Evaluation (APACHE) score or a comparable score. This is an important limitation of the study since this information would have shed more light on disease severity on admission and during hospital stay. Since the study primarily looked at the etiology of bacterial sepsis and not the risk factors for poor outcome, lack of these details may not affect the primary objective of the study. In conclusion, our study observed that association between rhabdomyolysis and sepsis is relatively common and is associated with additional causal factors for rhabdomyolysis like statin use, chronic alcohol intake, hypokalemia, hypernatremia and hypophosphatemia. Lung was the most frequent septic foci and we observed that gram negative organisms were more frequent than gram-positive pathogens. Furthermore, our data questions the safety of continuing statins in diabetic and alcoholic patients with active sepsis. **Competing Interests:**The authors have declared that no competing interests exist. **Funding:**The authors have no support or funding to report. [^1]: Conceived and designed the experiments: AAK EB GPSS GA. Performed the experiments: AAK GPSS. Analyzed the data: AAK EB GPSS PS. Contributed reagents/materials/analysis tools: AAK GPSS. Wrote the paper: AAK EB GPSS GA.

This article is available from: <http://casesjournal.com/casesjournal/article/view/8122> Introduction {#s01} ============ LMMC (or lipoma of the conus medullaris) the most commonly fatty mass occurring along or within the spine. LMMC is a form of occult spinal dysraphism in which a subcutaneous fibrofatty mass traverses the lumbodorsal fascia, causes a spinal laminar defect, displaces the duramatter, and infiltrates and tethers the spinal cord \[[@bib-001]-[@bib-003]\]. LMMC is the commonest cause of congenital tethering and causes neurological deterioration due to the conus medullaris and root ischemia \[[@bib-003]-[@bib-005]\]. We have not found the entity of lateral lower thoracal LMMC and especially as large as this size mass at ages of our patient in the literature. Here we presented a 10-month-old girl patient harboring LMMC, which was under skin and extending to spinal canal in left lateral lower thoracal (Th) region without neurological deficit and urinary dysfunction. Case presentation {#s02} ================= The Turkish 10-month-year old girl who born with normal spontaneous vaginal delivery admitted to our out-patient clinic for progressively swelling mass. This mass was approximately 12 × 8 cm size over left lateral lower thoracal region under skin after birth for approximately 3 months. Her mother took periconceptual folic acid and there was no pathological finding in her gynecological controls. Vaccination history was satisfactory. Such this congenital anomaly learned from family history was not seen before. On examination, there was no remarkable finding in the patient, except cosmetic swelling in the left lateral side of the lower thoracal region. From the patient\'s history, no neurological retardation or abnormality were determined by her family and pediatrician. On our clinical examination, no neurological deficit was found and urinary dysfunction tests were negative. Magnetic resonance imaging (MRI) studies showed LMMC that was located at the level between left Th9-Th12 paraspinal region, including muscle fibers, expanding spinal foramina at these levels, extending to spinal canal and slightly deplacing the canal to right lateral, and the greatest diameters of LMMC were 5 × 3.5 cm and 12 × 3 cm at axial images and sagittal images in fat-suppressed sequences, respectively ([Figures 1](#fig-001){ref-type="fig"},[2](#fig-002){ref-type="fig"}). {#fig-001} {#fig-002} The patient was operated under general anesthesia in prone position, after subcutaneous dissections at left lateral lower thoracal region, we met fatty tissue which was extending to the 9, 10, 11, 12, and following and dissecting these pathologic fatty formations, we reached the spinal canal through the small dural defect at base and then micro-surgically excised these adhered lesions from spinal canal and the lipomyelomeningocele was corrected without any complication. The excision material that sent to pathology laboratory during surgery reported as LMMC by the laboratory. Discussion {#s03} ========== Related to LMMC, the mechanism was proposed by Mc Lone et.al. Cutaneous and neural ectoderm separation occurs after neural tube closed normally but in this mechanism of LMMC, this separation occurs before neural tube closing and paraxial mesenchymal tissue prevents closure of open neural tube and causes a segmental myeloschisis by effect of the lateral somatic mesoderm that is continuous to grow \[[@bib-006]\]. Marine-Padilla proposed an other theory. According to them, the paraxial mesoderm which was accepted as main stimulant to embryologic layers was insufficient \[[@bib-007]\]. Also the theory in which excessive neural tissue growth preventing neural cleft fusion in spina bifida region \[[@bib-008]\] and an autosomal inheritance were also suggested for LMMC \[[@bib-005]\]. LMMC are lipomas that are tightly attached to the dorsal surface of a neural placode and extend dorsally through a spina bifida to be continuous with the subcutaneous fat \[[@bib-001]-[@bib-003]\]. LMMC rate has been estimated to be 2.5 per 10000 births \[[@bib-009]\]. In general, lipomas located subcutaneously have benign behavior and cause no problem except cosmetical matter but according to their localization, sometimes may cause pressure symptoms. In our first examination, the lesion was subcutaneously located to left lateral lower thoracal region, palpated as not tough mass and causing no pressure symptoms in our patient. Despite the patient had no complaint except cosmetical matter but we got suspicious about any congenital pathology and asked radiologic imaginations which revealed the subcutaneous lesion extending spinal canal through thoracal fassia, lamina and than piamatter. The above findings were confirmed on surgery. Majority of such these lesions are considered to be subcutaneously located benign mass, especially for pediatric age group of patients. If the benign mass was located to subcutaneously patient's posterior region, even if lateral region, LMMC must be held in our mind for differential diagnosis of posterolateral located masses. Necessary diagnostic tools must be used and prompt referral to neurosurgery department for possible emerging complications is advantageous for patients with this type lesion. LMMC is the most common cause of tethered cord \[[@bib-003]-[@bib-005]\] but related to which neurological deterioration or urinary dysfunction were not found in our patient. Because the natural progression of LMMC is causing neurologic deficits, and/or urological disturbances, the prophylactic surgery is recommended for these lesions by some authors \[[@bib-001],[@bib-010]\]. Treatment of LMMC is like other lesions and we operated the patient just as other lesions, but localization and size of the lesion in our case was especially important because approach to this type lesion must be different in context of thoracal vs. abdominal complications, characteristics of spinal canal, and neurological complications When thoracal spinal surgery is considered, at least these features must be in mind. Rarely growth of spinal lipomas may be found on sequential imaging studies, and lipomas have occasionally recurred after resection and lipomas will typically enlarge with age in proportion to body growth so that significant weight gain, particularly if associated with obesity, may contribute to recurrence of spinal lipomas \[[@bib-011]\]. The lateral lower thoracic LMMC may be considered as a variant of LMMC, which is located subcutaneously lateral lower thoracal region and then extending to spinal canal where it may cause neurological pathologies at birth or with advancing age and/or weight gain. In our patient, symptoms or radiologic findings of tethered cord were not detected for 15 months follow-up. Conclusions {#s04} =========== Here we presented lateral lower thoracic LMMC, which is needed detailed researches for treatment, follow-up and results of this localization of LMMC and at present, these parameters may be modified to LMMC parameters of other vertebra localization. Consent ======= Written informed consent was obtained from the patient for publication of this case report and accompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal. Competing interests =================== The authors declare that they have no competing interests. Authors' contributions ====================== NU, HY examined the patient, searched the literature and helped to draft the manuscript. FSE treated the patient, searched the literature, drafted the manuscript critically revised the manuscript for important intellectual content and gave final approval of the version to be published. All authors read and approved the final manuscript. LMMC : Lipomyelomeningocele MRI : Magnetic resonance imaging Th : Thoracic

Stanozolol (5α-androstane-17α-methyl-17β-ol \[3,2-c\] pyrazole), a heterocyclic anabolic androgenic steroid was first synthesized by Clinton in 1959 \[[@B1]\]. It differs in structure with endogenous steroid hormones and most commonly used exogenous anabolic steroids \[[@B2]\]. Despite the restriction by the International Olympic Committee (IOC) since 1974, stanozolol is one of the most frequently misused synthetic anabolic steroids in sport. Several authors reported it's used in the treatment of osteoporosis and deficiency in protein synthesis \[[@B3]-[@B5]\]. It is also used in veterinary medicine to boost appetite, to weight gain, and for the treatment of certain types of anemia \[[@B6],[@B7]\]. In order to develop stability indicating methods, stress degradation study is a powerful tool used routinely in pharmaceutical development that lead to a quality stability data and to recognize the degradation pathways of the active drug ingredient and its products. These degradation studies are employed for monitoring the drug stability which provides evidence that in what manner drug quality varies with passage of time by the consequence of different environmental factor including temperature, light, humidity, etc. and effect of stomach and intestine pH. Various strategies and principles are provided by International Conference on Harmonization (ICH) and the FDA for conducting forced degradation studies \[[@B8]-[@B10]\]. Different analytical techniques such as gas chromatography--mass spectrometry (GC-MS), high performance liquid chromatography-tandem mass spectrometry (HPLC-MS/MS), high throughput ultra performance liquid chromatography-tandem mass spectrometry (UHPLC-ESI -MS/MS) have been described for the analysis of stanozolol in variety of biological samples \[[@B11]-[@B14]\]. However, there is no report for the development of stability-indicating TLC-densitometric method for the quantification of stanozolol in the presence of its degraded products so far. Currently, TLC-densitometric method getting considerable attention on account of its various benefits including simultaneous running number of samples, utilizing little amount of eluent in contrast to HPLC, reduced cost and analysis time, mobile phase of extreme pH and direct application of various type of samples (dirty, turbid or suspensions) can also be employed \[[@B15]\]. In continuation of our studies on the determination of organic compounds of biological interest and pharmaceutical importance \[[@B16]-[@B19]\], and taking ICH guidelines Q1A in consideration, present study describes a simple stability-indicating TLC densitometric method for the quantification of stanozolol in presence of its degraded products formed under the applied stress conditions. Experimental ============ Standards and chemicals ----------------------- Standard stanozolol was purchased from Sigma Aldrich (USA). Stanozolol® tablets (containing 5 mg stanozolol/tablet (S1) and 10 mg stanozolol/tablet (S2) manufactured by LA Pharma S.r.I) were procured from the local pharmacies of Karachi, Pakistan. Aluminum sheets precoated with silica gel (60F-254, 20 cm × 20 cm) were purchased from the Merck (Germany). Hydrochloric acid (HCl) and hydrogen peroxide (H~2~O~2~, 35% v/v) were purchased from the Fisher Scientific (UK) while sodium hydroxide was procured from the BioM Laboratories (Cerritos, USA). Deionized water was acquired from Millipore Milli Q Plus System (Bedford, USA). All additional analytical grade reagents and chemicals were procured from Merck (Germany). Chromatographic state and instrumentation ----------------------------------------- A CAMAG automatic TLC sampler (Linomat 5) was used for the spotting of all samples and standards, which were spotted in the form of bands of width 6 mm with a CAMAG 100 μL syringe on pre-coated silica gel TLC aluminum sheets. CAMAG Reprostar 3 was used for the video densitometry of the chromatoplate, and the integrated software of WinCATS was used for the analysis. Sample was applied at the constant rate of 0.1 μL/s and 9.1 mm space between the two bands was maintained. The eluent composed of petroleum ether and acetone (6:4, v/v). 20 cm × 10 cm CAMAG twin trough glass chamber was used for linear ascending development of TLC plate under unsaturated condition and 10 mL of organic solvent was consumed per run. The TLC plate was developed up to a distance of 80 mm. Chromatographic process was carried out at room temperature (25 ± 2°C) with relative humidity of 42 ± 5%. The developed TLC plate was dried with the help of air dryer for 5 min and subjected to the staining process using CAMAG Chromatogram Immersion Device. Phosphomolybdic acid (PMA) was used to visualize the analyte. The analyte complexed with PMA was then heated with CAMAG TLC Plate Heater 3 at 120--125°C for 3--4 minutes. CAMAG TLC Scanner 3 was used for scanning at λ~max~ 750 nm and operates in reflection absorbance mode by winCATS software. Each sample and standard level was spotted in triplicate, and during scanning baseline correction was employed. Evaluation was accomplished via peak areas using linear regression. Calibration curve of stanozolol ------------------------------- Stock solution of stanozolol (1 mg mL^-1^) was prepared in methanol. The stock solution was further diluted to obtain a series of working standards over the range of 200--1200 ng spot^-1^ for calibration curve. Prepared solutions were stored at 4°C until use. Each working standard was spotted in triplicate, total 18 tracks were spotted on each TLC plate. The spot volume of each standard level was 5 μL. The spotted plate was developed as mentioned in previous section. This practice was repeated six times to get an average standard calibration curve. Linear least- square regression was performed on the data of peak areas plotted versus the corresponding concentrations. Validation of method -------------------- The method validation was done as described by the ICH guidelines. The method sensitivity was estimated with respect to limit of detection (LOD), limit of quantification (LOQ) and correlation coefficient. Working solutions containing 200--1200 ng of stanozolol were spotted on TLC aluminum sheet. In order to evaluate LOD and LOQ, calibration curve was used and were evaluated by using following equation: LOD = 3.3 δ/S, LOQ = 10 δ /S where, S = the slope of the calibration curve, δ = the residual standard deviation of regression line or the Y-intercept of regression line. The LOD and LOQ were estimated as 3 and 10 times of the noise level, correspondingly. Moreover, both were tentatively determined by diluting the familiar concentration of stanozolol standard till the mean instrumental responses were nearly three and ten times of the standard deviation of the responses for six manifold analyses. The Intra- and inter-day variation for the estimation of stanozolol was evaluated for method precision. It was achieved by using three different concentration levels of 300, 500 and 700 ng spot^-1^. Repeated analyses were carried out in a same day for intra-day analysis while the same practice was repeated next day for inter-day analysis. Intra- and inter-day analyses were performed to check the repeatability and reproducibility of the method, respectively and results were statistically evaluated in terms of % R.S.D. In order to check the robustness, following parameters were intentionally changed within the range of ± 5% at 300, 500 and 700 ng/ spot concentration level; mobile phase composition, different types of TLC plates e.g., silica-gel glass plates and aluminum sheets of other supplier (Macherey-Nagels, Germany), time from spotting to chromatography, time from chromatography to staining, time for staining and time from staining to scanning. The accuracy of the method was assessed by performing recovery study of pre-analyzed samples, spiked with extra 50, 100 and 150% standard stanozolol. The analysis of spiked samples was repeated six times. Preparation and analysis of pharmaceutical products --------------------------------------------------- To estimate the stanazolol content in pharmaceutical products, the tablets (label claim: 5 mg/tablet, 10 mg/tablet) were crushed and amount equivalent to one tablet was extracted in 2 mL methanol. To confirm the complete extraction of the stanazolol, it was shaken at 40°C for 20 minutes in a thermomixer (Eppendorf® Thermomixer Comfort, UK) at 300 rpm. The resulting solution was filtered with glass syringe with disposable micropore filter tips (13 mm × 0.45 μm Teflon filter) and the filtrate was used for the drug content analysis. From the filtrate, 5 μL was spotted onto the plate and then development, staining and scanning were done as previously described. This was repeated in triplicate. Preparation of forced degradation products ------------------------------------------ Stress degradation studies of stanozolol were performed using parallel synthesizer (Smart Start Synthesizer, Chem Speed Ltd., Switzerland). Stock solution (1 mg mL^-1^) was used for forced degradation. After the completion of reaction, all the resulting solutions were preserved at −20°C prior to analysis. Average peak areas of active components were analyzed after triplicate analysis. For acidic hydrolysis, 0.1N, 1N and 5N HCl were used, for alkaline hydrolysis, 0.1N, 1N and 5N NaOH were used while for neutral hydrolysis Milli Q water was used. 3 mL of each concentration of acidic and alkaline solutions and Milli Q water were added into 3 mL (1 mg/mL) stock solutions of each. To study wet heating degradation, 3 mL (1 mg/mL) of stock solution was used. Oxidation was carried out by adding 3 mL of H~2~O~2~ (35% v/v and 5% v/v) in 3 mL stock solution of each set. All the resultant solutions were refluxed for two hours at 80°C in parallel synthesizer in the dark, in order to prohibit the possible degradative effects of light. 2 μL (1000 ng/spot) of 0.1N, 1N and 5N HCl treated solution, 0.1N, 1N and 5N NaOH treated solutions, and neutral hydrolysis solution while 1 μL (1000 ng/spot) of wet-heating degradation mixture and 2 μL (1000 ng/spot) from oxidation mixture solution were applied on TLC plate in triplicate and densitogram were developed. Dry heating degradation was conducted by taking standard stanozolol and heated in oven at 90°C for 6 hrs. Photochemical degradation was conducted by taking standard stanozolol and placed in direct sunlight for three days from 8 to 18 hrs at 30 ± 2°C. 1 mg of each treated standard was dissolved in 2 mL of methanol and 2 μL (1000 ng/spot) of each resultant solution was spotted on TLC plate in triplicate for chromatographic analysis. For oxidation reaction at room temperature, 3 mL stock solution of stanozolol was added with 3 mL of H~2~O~2~ (35% v/v) and the resultant solution was kept for 24 hours at room temperature. 2 μL (1000 ng/spot) of treated solution was applied on TLC plate in triplicate for chromatographic analysis. Isolation and characterization of acid degradation product ---------------------------------------------------------- In order to isolate the acid degradation product in bulk amount, 500 mg of stanozolol in methanol (1 mg/mL) was refluxed in 1N HCl for 3 hours and monitored by TLC analysis (petroleum ether: acetone; 6:4, *v/v*). At the end of reaction, the mixture was neutralized with sodium bicarbonate solution and extracted with dichloromethane. The reaction mixture was loaded on to a silica gel column, which on elution with hexane: ethyl acetate; 6.5:3.5, v/v afforded purified degradation product (110 mg; 22% yield) which was identified with the aid of 1D- and 2D-NMR spectroscopic techniques and ESI-QqTOF-MS/MS analysis. Results and discussion ====================== Method optimization ------------------- For the development of chromatographic method, the mobile phase composition was optimized with a view to develop stability-indicating assay method. TLC sheet was simultaneously spotted with standard stanozolol, pharmaceutical samples and degraded products and was developed in different solvent systems. Various mobile phases were tried to resolve stanozolol from its degraded products. *R*~f~ values and peak widths of stanozolol in different mobile phases are summarized in Table [1](#T1){ref-type="table"}. Suitable separation with best resolution was achieved with petroleum ether and acetone (6:4, v/v). This system showed sharp and symmetrical peaks of stanozolol and its acid degradation product under unsaturated condition with *R*~f~ value at 0.46 ± 0.02 and 0.55 ± 0.01 respectively, (Figure [1](#F1){ref-type="fig"}) with 1.5 resolution. Various derivatizing agents like ceric sulfate, vanillin, phosphomolybdic acid (PMA) and ceric ammonium sulfate were tried. However, derivatization with phosphomolybdic acid (PMA) gave significantly better result, the spots are visible, sharper and more colourful as compared to the plate sprayed with other derivatizing agents described above. {#F1} ###### ***R***~**f**~**values and peak widths of stanozolol in different mobile phases** **S. No.** **Compositions (v/v)** **Proportions** ***R***~**f**~ **Peak widths (cm)** ------------ -------------------------- ----------------- ---------------- ---------------------- 1 Petroleum ether-acetone 8:2 0.24 0.07 2 petroleum ether-acetone 7.5:2.5 0.29 0.09 3 Petroleum ether-acetone 6:4 0.46 0.06 4 Hexane-ethyl acetate 8:2 0.05 0.06 5 Chloroform-methanol 9:1 0.17 0.09 6 Dichloromethane-methanol 9.2:0.8 0.13 0.09 The linear regression data for the calibration curves (n = 6) showed a good linearity r = 0.9990 ± 0.0004, (Slope = 5.841 ± 0.755, Intercept = 1420.3 ± 36.4) over the concentration range of 200--1200 ng/spot with respect to peak area. In residual linearity test of stanozolol, the random pattern of residuals against respective standard levels showed a linear model of standard calibration curves Additional file [1](#S1){ref-type="supplementary-material"}: Figure S1). Method validation ----------------- The repeatability and reproducibility were showed in terms of percentage relative standard deviation (R.S.D. %) between two analyses and was found to be \< 2% (Table [2](#T2){ref-type="table"}). Similarly, intra- and inter-day variation of stanozolol at three different concentration levels 300, 500 and 700 ng/spot was found to be \< 2% in all the cases (Table [2](#T2){ref-type="table"}). ###### Precision and accuracy for quality control standard of stanazolol   **Analyst 1** **Analyst 2** ----- --------------- --------------- ------- -------------- ------ ------- --------------- ------ ------- 300 303.56 ± 5.00 1.65 98.81 306.63± 5.52 1.80 97.79 300.62± 1.452 0.48 99.79 500 503.33 ± 1.06 0.21 99.33 504.49± 3.38 0.67 99.10 502.63± 0.78 0.15 99.47 700 705.80 ± 1.50 0.21 99.17 703.85± 1.14 0.16 99.45 703.04± 0.71 0.10 99.56 ^\*^Values are mean of n = 3. The standard deviation of % yield of three standard levels (300, 500 and 700 ng) was estimated for each parameter for robustness analysis. Mean % R.S.D. was 2.4 for varying in mobile phase composition, 1.39 for the use of different TLC plates, 2.27 for varying time from spotting to chromatography, 1.26 for varying time from chromatography to staining, and 1.77 for varying time for staining and 2.1 for varying time from staining to scanning (Table [3](#T3){ref-type="table"}). RSD % is reasonably low (\< 2.5%) for all parameters which indicates the robustness of the developed method. The LOD for standard stanozolol with signal/noise ratio of 3:1 was found to be 1.6 ng per spot and LOQ with signal/noise ratio 10:1 was found to be 5.1 ng/spot. This showed the satisfactory sensitivity of the developed method. The spectra were compared at peak start, peak apex and peak end positions of the spot, to check the peak purity of stanozolol. Excellent correlation, *r*^2^ (start to middle) = 0.9996 and *r*^*2*^ (middle to end) = 0.9998 was observed by comparing the spectra of standard and corresponding peak in pharmaceutical samples. Spiking with 50, 100 and 150% of additional standard stanozolol afforded recovery of 92.2--95.2% as showed in Table [4](#T4){ref-type="table"}. This practice was repeated six times. Summary of all validation parameters is highlighted in Table [5](#T5){ref-type="table"}. ###### Robustness testing (n = 6) **Standard level** **Mobile phase composition** **Nature of TLC** **Time from spotting to chromatography** **Time from chromatography to staining** **Time for staining** **Time from staining to scanning** -------------------- ------------------------------ ------------------- ------------------------------------------ ------------------------------------------ ----------------------- ------------------------------------ --------------- ------ ---------------- ------ ---------------- ------ 300 303.7 ± 11.54 3.80 298.3 ± 1.54 0.51 311.29 ± 8.03 2.58 284.18 ± 6.81 2.39 300.63 ± 9.34 3.11 287.3 ± 10.06 3.5 500 505.01 ±12.44 2.46 500.7 ±13.58 2.71 498.51 ± 18.85 3.78 513.85 ± 6.91 1.34 516.18 ± 2.58 0.50 508.53 ± 12.96 2.54 700 705.23 ± 6.64 0.94 698.46 ± 6.64 0.95 717.03 ± 3.43 0.47 699.61 ± 0.51 0.07 714.95 ± 12.30 1.72 704.29 ± 3.33 0.47 ###### Recovery studies (n = 6) **Sample** **Standard spiked %** **Theoritical ng spot**^**-1**^ **Mean recovery %** **RSD %** ------------ ----------------------- --------------------------------- --------------------- -----------   50 351.5 92.2 5.00 S1 100 468.7 95.2 5.99   150 585.9 94.6 1.76 ###### Summary of validation parameters **Parameter** **Data of standard stanazolol** ------------------------------- --------------------------------- Linearity range 200-1200 ng/spot Correlation coefficient 0.9990 ± 0.0004 Limit of detection (LOD) 1.6 ng/spot Limit of quantification (LOQ) 5.1 ng/spot Y = mx + c Y = 5.841x + 1420.3 Slope ± SD 5.841 ± 0.755 Intercept ± SD 1420.3 ± 36.4 Intra-day analysis (n=3), 0.690 \% R.S.D. Inter-day analysis (n=3), 0.877 \% R.S.D. Robustness Robust Specificity Specific Analysis of commercial dosage form ---------------------------------- In chromatogram of tablet's extracted samples, a single spot (*R*~f~ value of 0.46 ± 0.02) was observed corresponding to stanazolol. No interference from the excipients and binders commonly present in the marketed tablet formulation was observed. Two different dose strengths, 5 and 10 mg were analysed through developed method and it showed drug content of 5.198 ± 0.16 mg and 10.126 ± 0.25 mg, respectively. The mean percent recovery of the proposed method is 103.96 for S1 and 101.26 for S2. The low R.S.D. % value indicated the suitability of this method for routine analysis of stanozolol in pharmaceutical products. Stability-indicating property ----------------------------- Stanazolol was subjected to various stress degradation conditions. The chromatogram of stanozolol treated with different strength of acid showed well-separated peak of stanozolol as well as one additional peak of degradant (*R*~f~ = 0.55) (Figure [1](#F1){ref-type="fig"}). Stanazolol showed 11.7%, 23.7% and 54.9% degradation under 0.1N HCl, 1N HCl and 5N HCl conditions, respectively. Very low degradation was also observed after exposing the stanozolol to stressed conditions of base, and neutral hydrolysis, wet and dry heating, sunlight and oxidation. However, peaks of degraded products were not observed. Results of the stress degradation studies are summarized in Table [6](#T6){ref-type="table"}. Acid degradation product was purified and characterized through various spectroscopic methods as 17, 17-dimethyl-l8-nor-5α-androst-13(14)-eno \[3,2c\] pyrazole which was earlier reported as a synthetic derivative of stanozolol \[[@B20]\]. The HRESI-MS (positive mode) showed peak \[M+H\]^+^ peak at *m/z* 311.2452 corresponding to C~21~H~31~N~2~ (calc. 311.2481), 18 amu less than stanozolol, indicating a dehydrated product. The ESI-MS/MS analysis of *m/z* 311 \[M+H\]^+^ showed base peak at *m/z* 81, corresponding to pyrazole ring (Additional file [1](#S1){ref-type="supplementary-material"}: Scheme S1). The product ion at *m/z* 95 is suggested to derive from the same protonated species as *m/z* 81 by a rearrangement of the bonds. Fragment due to retro Diel-Alder cleavage was observed at *m/z* 189 from fragments *m/z* 311. The CID-MS/MS spectra and proposed mechanistic fragmentation pathway are shown in Additional file [1](#S1){ref-type="supplementary-material"}: Figure S2 and Additional file [1](#S1){ref-type="supplementary-material"}: Scheme S1, respectively and the characteristic MS/MS fragments are summarized in Additional file [1](#S1){ref-type="supplementary-material"}: Table S2. Proposed mechanism for the acid degradation of stanozolol is shown in Scheme [1](#C1){ref-type="fig"}. ^13^C-NMR and ^1^H-NMR data of a degraded product is provided in Additional file [1](#S1){ref-type="supplementary-material"}: Figure S3 and Additional file [1](#S1){ref-type="supplementary-material"}: Table S1. {#C1} ###### Summary of stress degradation studies of stanozolol **Degradation conditions** **% Recovery** **% Degradation** ***R***~**f**~**of degraded products** ---------------------------- ---------------- ------------------- ---------------------------------------- Acidic hydrolysis ^a^       0.1N HCl 88.28 11.72 0.49, 0.55 1N HCl 76.25 23.75 0.49, 0.55 5N HCl 45.07 54.93 0.49, 0.55 Basic hydrolysis ^a^       0.1N NaOH 98.40 1.60 Not detected 1N NaOH 95.19 4.81 Not detected 5N NaOH 90.71 9.29 Not detected Neutral hydrolysis ^a^ 98.65 1.35 Not detected Wet heating ^a^ 99.59 0.41 Not detected Dry heating 98.32 1.68 Not detected Oxidation ^a^       5%v/v H~2~O~2~ 97.90 2.10 Not detected 35%v/v H~2~O~2~ 99.59 0.41 Not detected Oxidation at room temp 98.30 1.70 Not detected Photostability-daylight 98.62 1.38 Not detected ^a^Reflux in parallel synthesizer for two hours at 80°C. Conclusion ========== A new stability-indicating TLC method combined with densitometry has been developed and validated for the identification and quantification of stanozolol in its pharmaceutical dosage form. Satisfactory precision and accuracy, minimum cost, and quicker analysis are the core features of this method. The ICH guideline was followed for method validation. Reproducibilty and specificity of method was determined by statistical analysis. Moreover, this study demonstrates the degradation susceptibility of stanozolol to different stress conditions and thus helps in determining the changes in chemical and physical properties of the stanozolol with the passage of time. Current study may be expanded to degradation kinetics of stanozolol and for its assessment in urine, plasma, serum and other biological fluids. Competing interests =================== Authors declare that they have no competing interests. Authors' contributions ====================== SGM: Participated in the experimental designing and method optimization. QA: Performed the experiments and wrote the manuscript. WM and NK: Participated in experimental work and also involved in the useful discussion. All authors read and approved the final manuscript. Supplementary Material ====================== ###### Additional file 1: Figure S1 Standard calibration curve (A) and its residual linearity test (B). **Figure S2** CID-MS/MS spectra of acid-degradation product of stanozolol at the collision energy of 45 ev. **Figure S3** 13C-NMR and 1H-NMR (400 MHz, CDCl3) spectra of acid degradation product of stanozolol **Table S1** 13C-NMR and 1H-NMR (400 MHz, CDCl3) chemical shifts of acid degradation product of stanozolol (δ in ppm). **Table S2** Elemental composition of daughter ions of acid degradation product of stanozolol (m/z 311). **Scheme S1** Proposed CID-MS/MS fragmentation pathway of acid-degradation product of stanozolol. ###### Click here for file Acknowledgements ================ The authors are grateful to Mr. Arslan Ali and Miss. Madiha Goher, H.E.J. Research Institute of Chemistry, University of Karachi, for useful discussions and for the financial support from the institutional research grant.

INTRODUCTION ============ A lipoma is a benign, slow-growing, asymptomatic tumor of mesenchymal origin; lipomas consist of adipose tissue and can be sessile or pedunculated, as well as single or lobulated, typically surrounded by a fibrous capsule.[@B001]^,^[@B002] Although the etiology of lipoma remains unclear, causal factors include endocrine changes, trauma, and genetic makeup.[@B003] Although lipoma of the oral cavity is uncommon, lipomas of the head and neck account for 15-20% of all head and neck neoplasms.[@B004] Lipomas and 1-5% of all neoplasms of the oral cavity.[@B002] The diagnosis of lipoma is predominantly based on clinical findings, histopathological findings being conclusive in all cases. Imaging tests can be useful, especially when lipomas are located in the deeper fascia of the neck or face. Histologically, lipomas consist of well-differentiated adipose tissue surrounded by a connective tissue capsule.[@B005] Here, we report a case of bilateral lipoma of the lateral borders of the tongue treated with surgical excision. CASE REPORT =========== An 85-year-old female patient sought medical attention complaining of a "lump" in her tongue. The patient reported that the lump had been present for approximately 5 years. Oral examination revealed painless, endophytic nodules in the right and left lateral borders of the tongue. The lesion on the right was 1.0 cm in diameter, compared with 1.8 cm for the lesion on the left, in their longest axis. Both were soft, yellowish lesions with well-defined borders and no ulcerations ([Figure 1](#g01){ref-type="fig"}). {#g01} A presumptive diagnosis of lipoma was made, and the patient underwent surgical excision under local anesthesia. Macroscopic examination revealed yellowish nodules with a lobular surface ([Figure 2](#g02){ref-type="fig"}). Microscopic examination revealed the presence of mature adipose tissue interspersed with bands of dense connective tissue, confirming the diagnosis of lipoma. {#g02} In the case reported here, the histological features of the lipomas (i.e., predominance of simple lipomas), their clinical features (asymptomatic, encapsulated, round, yellowish/mucosa-like in color, smooth, and without ulceration), their anatomical location (the dorsum of the tongue), and their presentation (bilateral) were consistent with those reported in the literature, although the age and gender of the patient (an 85-year-old female) were not. At this writing, 2 years after the surgical excision of the lipomas, there were no signs of recurrence. DISCUSSION ========== Lipomas rarely affect the oral cavity, accounting for only approximately 5% of all intraoral neoplasms. The etiology of lipoma has yet to be definitively established but includes endocrine changes, hereditary factors, local trauma, and infection.[@B006] Histologically, lipomas are indistinguishable from normal adipose tissue.[@B005] Although some studies have reported that males are more often affected,[@B002]^,^[@B007] others have reported that lipomas are more common in females.[@B008]^,^[@B009] Individuals in the 30-70 year age bracket are the most affected, lipomas being rare in those under 20 years of age.[@B002]^,^[@B010] Regarding the intraoral commitment, the buccal mucosa, the vestibular fornix, the floor of the mouth, the tongue, the inner surface of the lips, the alveolar ridge, the palate, and the salivary glands, are the most commonly affected sites.[@B010] Lipomas can be double (bilateral) or, in rare cases, multiple. Multiple lipomas are generally related to lipomatosis, neurofibromatosis type 1 (formerly known as von Recklinghausen's disease), Gardner syndrome, and adiposis dolorosa (also known as Dercum's disease). The differential diagnosis of lipoma includes epidermoid cysts, lymphoepithelial cysts, ranulas, pleomorphic adenomas, and mucoepidermoid carcinomas.[@B011] In suspected cases of deep-seated lipoma, computed tomography can assist in ruling out vascular lesions and in identifying important anatomical structures that are adjacent to the lesions.[@B012] The histopathological features of lipomas are typical: they consist of well-differentiated adipose tissue surrounded by a connective tissue capsule. Although morphologically indistinguishable from normal fat cells, tumor cells are metabolically more active. The fibrous capsule may sometimes be absent or ruptured.[@B013] Conservative treatment of lipomas involves complete surgical excision by excisional biopsy, especially in cases of infiltrative lipoma.[@B014] Recurrence is rare when the lesion is completely excised.[@B002]^,^[@B003]^,^[@B006]^,^[@B010] As evidenced by the case reported here, lipoma of the oral cavity can be treated successfully through conservative treatment. Oliveira Junior EM, Alves CAF, Santos-Queiroz AG, Elias FM, Campos AC. Bilateral lingual lipoma: a case report. Autopsy Case Rep \[Internet\]. 2012;2(2):21-23. <http://dx.doi.org/10.4322/acr.2012.012> [^1]: Conflict of interest: None

Introduction {#Sec1} ============ Poisoning is an important cause of wildlife mortality, and in some instances has been responsible for extensive population declines (Green et al. [@CR8]). Primary exposure to poisons occurs when wildlife is intentionally poisoned for hunting (Ogada [@CR17]) or due to human-wildlife conflict (Venkataramanan et al. [@CR26]; Ragothaman and Chirukandoth [@CR22]). Accidental poisoning may occur by secondary exposure to poisons in the environment or via contaminated food sources. Examples include pesticide residues in Irrawady dolphins living in contaminated water bodies (Kannan et al. [@CR12]) and Sarus cranes consuming monocrotophos-treated grains (Pain et al. [@CR18]), respectively. In all such instances, non-target wildlife may also be unintentionally poisoned when they scavenge on carcasses containing high concentrations of poisons (Kalaivanan et al. [@CR11]; Molenaar et al. [@CR16]). Although the threats posed by intentional or accidental poisoning to wildlife in India have long been recognised (Spillett [@CR25]), statistics on the proportion of wildlife deaths attributable to poisoning are lacking. Poisons or chemicals detected in wildlife in India include organochlorine (OC) (Kannan et al. [@CR12]; Pathak [@CR19]), organophosphate (OP) (Pain et al. [@CR18]; Kalaivanan et al. [@CR11]) or carbamate pesticides (Venkataramanan et al. [@CR26]) and rodenticides (Ragothaman and Chirukandoth [@CR22]). Ragothaman and Chirukandoth (Ragothaman and Chirukandoth [@CR22]) listed 15 separate incidents of pesticide poisoning in wildlife between 2000 and 2010, including six instances (40%) of OP and four instances (27%) of carbamate pesticide poisoning. However, wildlife mortality due to poisoning may be under-reported in India, either because poisoning is not considered as a possible cause; carcasses are found in advanced stages of decomposition or cases are lost to follow up after submission of tissue samples to testing laboratories (personal observation). This is true in the case of wildlife mortality reporting from the south Indian state of Kerala with its rich biodiversity owing to the Western Ghats, a global biodiversity hotspot which runs across the entire length of the state (Gunawardene et al. [@CR9]). A search of the scientific literature on five databases (PubMed, Web of Science, Scopus, Ovid, ProQuest) and Google Scholar using the search terms "wildlife", "poison\*" and "Kerala" did not find any publications that identified the specific poison or chemical implicated in suspected poisoning events in the state. The four publications that mentioned wildlife poisoning in Kerala used generic terms such as 'rodenticides' (Jayahari and Jayson [@CR10]), merely identified poisoning as the potential or confirmed cause of death (Cheeran [@CR2]; Rohini et al. [@CR23]) or provided anecdotal evidence (e.g. poisoning of elephants by 'Folidol', an OP pesticide) (Spillett [@CR25]). Knowledge of the specific chemical compounds involved in wildlife mortality events is the first step in enabling government agencies to implement regulatory measures that can prevent incidents of poisoning or help populations that have suffered declines to recover (e.g. impact of the diclofenac ban on vulture recoveries) (Paudel et al. [@CR20]). To fill this knowledge gap, this report provides details of the specific poisons/chemical compounds detected in wildlife mortality events from Kerala over a 2-year period. Methods {#Sec2} ======= Periyar Tiger Reserve (PTR) in central Kerala has the services of a full-time wildlife veterinarian designated as Assistant Forest Veterinary Officer (AFVO). The AFVO deals with treatment of ill and injured wildlife, assists in human-wildlife conflict resolution and conducts necropsies and disease surveillance in PTR and adjoining wildlife reserves. For this report, all cases of wildlife mortality recorded between January 2011 and March 2013 at the AFVO's office were reviewed. Cases were identified where poisoning was considered as a differential diagnosis based on circumstantial evidence, necropsy lesions or the need to rule out poisoning as a cause of death when no lesions were identifiable in decomposed carcasses. Records of mortality events prior to and after this period were not immediately accessible and have not been included in the review. Results and discussion {#Sec3} ====================== Eleven separate wildlife mortality events recorded between January 2011 and March 2013 met the selection criteria mentioned above. In ten instances, post-mortem tissue samples were submitted in saturated salt solution for toxicological analyses to the Regional Chemical Examiners Laboratory (RCEL), Ernakulam, one of three government laboratories in Kerala providing toxicological testing services free of charge to government departments (Chemical Examiner's Laboratory Department [@CR3]) (Table [S1](#MOESM1){ref-type="media"}). Test results were reported as positive or negative for a specific poison after analyses of processed samples by thin layer chromatography (TLC) and a second method (gas chromatography--mass spectrometry or ultraviolet spectrophotometry). Detection of coloured spots with a retardation factor (*R*~f~) value equal to that of a reference compound during TLC confirmed presence of the poison. Specific poisons/chemicals were detected in three cases, although concentrations were not reported. In the other seven instances, negative test results were reported. In a fourth case where tissue samples were not submitted for analyses, poisoning was determined to be the cause of death based on lesions and physical presence of the chemical compound in gastric contents. These four cases are described in detail below.Imidacloprid in an Asian elephant A wild adult female Asian elephant (*Elephas maximus*) was found dead in a commercial cardamom plantation adjoining a forest reserve in Ayyappankovil range, Kattappana in November 2011. The carcass lay next to a demolished shed that had been used to store and prepare pesticides including imidacloprid (a systemic neonicotinoid pesticide), quinalphos and chlorpyriphos (OP pesticides). The elephant was suspected to have consumed water from a barrel which was used to dilute pesticides prior to spraying, whereby it may have been poisoned. Lesions observed included marked lingual and corneal cyanosis, diffuse pulmonary haemorrhage and hepatomegaly with subcapsular haemorrhages. The presence of large numbers of dead or moribund flies in the elephant's mouth and around the carcass (Fig. [1](#Fig1){ref-type="fig"}), as well as dead gastric bots (commonly of the oestrid fly *Cobboldia elephantis)* in the stomach pointed to the likelihood of pesticide poisoning. Imidacloprid was detected in hepatic tissue and gut contents submitted to the RCEL.Fig. 1Elephant (*Elephas maximus*) carcass found within a cardamom plantation, next to a demolished shed used to store pesticides. Masses of dead flies were found near the elephant carcass and inside the oral cavity (inset) Poisoning with neonicotinoid pesticides has been reported in humans (Cimino et al. [@CR4]), domestic animals (Caloni et al. [@CR1]) and small vertebrate wildlife such as birds and fish (Gibbons et al. [@CR6]). Exposure to pesticides, including imidacloprid, was suspected to be associated with the occurrence of congenital deformities and reproductive problems in chimpanzees and baboons in Uganda (Krief et al. [@CR13]). However, to the author's knowledge, this is the first report of mortality of a large wild mammalian species linked to imidacloprid consumption. This highlights the need for further studies on the use of neonicotinoid pesticides and their impacts on wildlife and the wider environment. Such studies, sorely lacking in the Indian context, are urgently needed given the accumulating evidence globally of their role in population declines of bees as well as non-target invertebrate species (Pisa et al. [@CR21]).2.Endosulfan in a gaur In August 2012, an adult male gaur (*Bos gaurus*) carcass was found in PTR in a forest patch on the border between Kerala and the adjoining state of Tamil Nadu, abutting a large commercial tea plantation. As the carcass was partially decomposed and lay close to the plantation where pesticide use was likely, samples of hepatic tissue and gut contents were submitted for toxicological analyses. Endosulfan, a toxic OC pesticide, was detected in all tissue samples. Given the state of decomposition of the carcass and absence of information on tissue concentrations of endosulfan, it was not possible to attribute death to pesticide exposure. It was considered more likely that long-term exposure occurred by consumption of pesticide-laden foliage or water around the plantation.3.Carbofuran in a Bonnet Macaque The carcass of an adult female bonnet macaque (*Macaca radiata*) was found on the terrace of a building in Kumily town, adjoining PTR, in October 2011. Necropsy lesions included severe ocular cyanosis and pulmonary congestion and marked splenomegaly. The stomach was filled with undigested food mixed with numerous dark purplish-blue granules (Fig. [2](#Fig2){ref-type="fig"}), tentatively identified as carbofuran, a carbamate pesticide widely available in Kerala and sold under various trade names (e.g. 'Furadan') (Cheeran [@CR2]; Ragothaman and Chirukandoth [@CR22]). Based on these findings, the macaque was suspected to have died of carbofuran poisoning after consuming pesticide-laced food. However, no tissue samples were submitted for toxicological analyses.Fig. 2**a** Ocular cyanosis in a bonnet macaque (*Macaca radiata*). **b** Dark purplish-blue granules of carbofuran mixed with gastric contents Substantial populations of bonnet macaques thrive in regions of Kerala adjoining wildlife habitat. They are generally viewed as pests and forest department officials often deal with complaints from the public when macaques steal food from local homes or restaurants, damage property or display aggression (Rohini et al. [@CR24]). Incidents of deliberate poisoning of macaques are thus not uncommon (Ragothaman and Chirukandoth [@CR22]).4.Warfarin in wild boars In September 2012, carcasses of four wild boars (*Sus scrofa*) of various ages were found in PTR. No distinct necropsy lesions were identifiable as all were actively decomposing. Poisoning was suspected since the carcasses were found over two days and near each other, close to office buildings and residential quarters. Analysis of hepatic tissue and gut contents detected warfarin, a first-generation anticoagulant rodenticide. It was unclear how the boars may have been exposed to warfarin. As they were known to frequent the area around the offices and quarters in search of food, malicious poisoning was considered likely, although accidental exposure (e.g. consumption from a carelessly discarded container while scavenging) could not be ruled out. Wild boars frequently cause extensive agricultural losses in Kerala (Rohini et al. [@CR24]) and retaliatory poisoning is commonly reported (Kalaivanan et al. [@CR11]). They also scavenge food waste from garbage dumps close to human habitation, posing a potential risk to people. In 2011, the Government of Kerala decided to allow farmers who had incurred losses to apply for permits to kill wild boar, although this decision has been contested (Gopakumar et al. [@CR7]). In all the cases reported here, lack of information on concentrations of detected poisons makes it difficult to conclusively establish poisoning as the cause of death. Despite this, the findings are significant as they form the first report of specific poisons identified in wildlife mortality events in Kerala. In doing so, they add to the literature on wildlife poisoning in India and globally. The report also reinforces the need for considering poisoning as a differential diagnosis in investigations of wildlife mortality events, irrespective of whether the primary cause of mortality can be established or not. This is particularly relevant given the proximity of wildlife habitats to human settlements and agricultural land in biodiversity-rich regions of the world such as India (Ragothaman and Chirukandoth [@CR22]). A thorough investigation of the immediate environment and circumstances of the mortality may raise suspicion of poisoning as a differential diagnosis. However, as is often the case in free-ranging wildlife mortalities, carcasses may be too decomposed at the time of detection for a necropsy to pinpoint the cause of death. Toxicological analyses can help to rule out poisoning or identify other causes for concern for wildlife, environmental and human health. Wildlife could even act as sentinels of environmental contamination, exemplified in this report by the detection of compounds such as endosulfan and imidacloprid. Wildlife researchers must make full use of the toxicological testing services provided free of charge by government laboratories like the RCEL in Kerala, and possibly other Indian states, to investigate the occurrence and impacts of poisoning. Finally, the study highlights how pesticides continue to be easily available and inadequately regulated in India. Their use in public responses to human-wildlife conflict in India has been reported previously (Madhusudan [@CR14]; Cheeran [@CR2]; Venkataramanan et al. [@CR26]). Since being banned in India in 2011 (Dhillon [@CR5]), endosulfan has continued to be available for agricultural use, and human exposures have also been reported (Menezes et al. [@CR15]), highlighting the lax implementation of existing regulations. The frequent involvement of such toxic compounds in animal and human deaths in India points to the need for strengthening legislation that regulates their availability and use without affecting their application in agricultural and allied sectors (Ragothaman and Chirukandoth [@CR22]). Doing so will have wide-ranging benefits, extending beyond wildlife health and conservation to human and environmental health. Electronic supplementary material ================================= {#Sec4} Supplementary Table 1(PDF 15.8 kb) I would like to acknowledge support and encouragement from officials of the Department of Forests and Wildlife, Government of Kerala, particularly the Principal Chief Conservator of Forests (PCCF) and Head of Forest Force, the PCCF and Chief Wildlife Warden, Mr. Sanjayan Kumar IFS, former Deputy Director, Periyar East Division, Thekkady, Mr. KV Subramaniam, former Field Director (Project Tiger), Kottayam and numerous Range Officers and supporting staff. I am grateful for the technical support and manpower provided to carry out fieldwork and diagnosis. I also thank Sekharan, Kaja, Ramadas and Unniettan for their constant dedication and support and great company while in the field. This paper developed from a poster presented at the 66th annual international conference of the Wildlife Disease Association. I am grateful for funding from the Wellcome Trust, UK which made it possible to present at the conference and to write this paper. Finally, I thank Sarah Hayes and an anonymous reviewer for invaluable feedback on this manuscript.

Three articles regarding the effects of (gene-modified) bone marrow mesenchymal stem cells transplantation for treatment of spinal cord injury and the migration process and distribution of transplanted stem cells observed by *in vivo* magnetic resonance imaging tracking, providing experimental evidence for spinal cord injury treatment with stem cell transplantation were reported in *Neural Regeneration Research*. We hope that our readers find these papers useful to their research. **Abstract** Bone marrow mesenchymal stem cells were isolated, purified and cultured *in vitro* by Percoll density gradient centrifugation combined with the cell adherence method. Passages 3--5 bone marrow mesenchymal stem cells were transplanted into rats with traumatic spinal cord injury *via* the caudal vein. Basso-Beattie-Bresnahan scores indicate that neurological function of experimental rats was significantly improved over transplantation time (1--5 weeks). Expressions of choline acetyltransferase, glutamic acid decarboxylase and synapsins in the damaged spinal cord of rats were significantly increased after transplantation, determined by immunofluorescence staining and laser confocal scanning microscopy. Bone marrow mesenchymal stem cells that had migrated into the damaged area of rats in the experimental group began to express choline acetyltransferase, glutamic acid decarboxylase and synapsins, 3 weeks after transplantation. The Basso-Beattie- Bresnahan scores positively correlated with expression of choline acetyltransferase and synapsins. Experimental findings indicate that intravenously transplanted bone marrow mesenchymal stem cells traverse into the damaged spinal cord of rats, promote expression of choline acetyltransferase, glutamic acid decarboxylase and synapsins, and improve nerve function in rats with spinal cord injury. {#F1} **Abstract** Rat bone marrow mesenchymal stem cells expressing brain-derived neurotrophic factor were successfully obtained using a gene transfection method, then intravenously transplanted into rats with spinal cord injury. At 1, 3, and 5 weeks after transplantation, the expression of brain-derived neurotrophic factor and neurofilament-200 was upregulated in the injured spinal cord, spinal cord injury was alleviated, and Basso-Beattie-Bresnahan scores of hindlimb motor function were significantly increased. This evidence suggested that intravenous transplantation of adenovirus- mediated brain-derived neurotrophic factor gene-modified rat bone marrow mesenchymal stem cells could play a dual role, simultaneously providing neural stem cells and neurotrophic factors. {#F2} **Abstract** Non-invasive tracing *in vivo* can be used to observe the migration and distribution of grafted stem cells, and can provide experimental evidence for treatment. This study utilized adenovirus-carrying enhanced green fluorescent protein (AD5/F35-eGFP) and superparamagnetic iron oxide (SPIO)-labeled bone marrow mesenchymal stem cells (BMSCs). BMSCs, double-labeled by AD5/F35-eGFP and SPIO, were transplanted into rats with spinal cord injury *via* the subarachnoid space. MRI tracing results demonstrated that BMSCs migrated to the injured spinal cord over time (T2 hypointensity signals). This result was verified by immunofluorescence. These results indicate that MRI can be utilized to trace *in vivo* the SPIO-labeled BMSCs after grafting. {#F3}

Introduction {#sec1} ============ Adolescence is a key developmental time period for the emergence of psychosocial difficulties, and mental health disorders (Thapar *et al.* [@ref59]; Blakemore & Mills, [@ref5]). To promote adaptive psychosocial functioning (PSF) during adolescence, appropriate conceptualisation and quantification of resilient functioning and its predictors is a crucial first step. In psychiatry, resilience refers to "*a dynamic process wherein individuals display positive adaptation despite experiences of significant adversity or trauma*" (Luthar & Cicchetti, [@ref32]). In the general population, it is well established that negative childhood experiences such as parental discord and/or lack of parental affection can have a negative impact on adolescent PSF (Egeland, [@ref16]; Trocmé *et al.* [@ref60]; van Harmelen, [@ref62]; Harpur *et al.* [@ref24]; Stoltenborgh *et al.* [@ref57]). Adolescent resilient PSF may therefore be seen as reflecting positive adaptation compared to others with similar experiences in the family environment. However, individuals with comparable experiences may not *appraise* their experiences in the same way (Rutter, [@ref49], [@ref51]). For instance, *perceived* levels of threat, rather than actual threat, determine later stress reactivity (van Wingen *et al.* [@ref65]). Therefore, including self-reported appraisal of childhood family experiences may contribute to a more valid and quantifiable measure of adolescent resilient functioning. Resilience captures positive adaptation across emotional, cognitive, behavioural and social domains of functioning (Masten, [@ref34]), and should be relevant to the environmental events and difficulties experienced (Luthar *et al.* [@ref33]). From this multidimensional perspective the presence of personal impairment or psychopathology does not necessarily preclude concurrent resilient functioning (Luthar *et al.* [@ref33]). For example, an adolescent can suffer considerable distress after a personal loss, but simultaneously continue to attend school and learn and see friends and can therefore be considered to be functioning 'resiliently' in those domains despite experiencing berievement. This multidimensional perspective indicates that a valid measure of adolescent resilient functioning in the general population should capture adaptive behaviour across a comprehensive range and level of psychosocial domains. Furthermore, resilient functioning is not a personality trait that is constant over time (Luthar & Cicchetti, [@ref32]; Rutter, [@ref51]; Cicchetti, [@ref11]; Masten, [@ref34]). Rather, resilient functioning waxes and wanes, possibly under the influence of protective factors such as family and friendship support (Rutter, [@ref49]; Afifi & Macmillan, [@ref1]; Cicchetti, [@ref11]; van Harmelen *et al.* [@ref64]). Therefore, having low resilience at some time does not preclude the presence of future resilience or vice versa. Consequently, it is important to study adolescent resilient functioning, and its influences, over time (Bonanno *et al.* [@ref6]). Understanding how adolescent resilient functioning varies over time and revealing how various factors influence such variation remains to be fully elucidated. Adolescent friendships and family support are important protective factors after early life stress (Rutter, [@ref49], [@ref51], van Harmelen *et al.* [@ref64]). Recently, we showed that adolescent family support reduces later depressive symptoms after differential levels of childhood family adversity, whereas adolescent friendship support reduced later depressive symptoms after childhood family adversity and/or peer victimisation (van Harmelen *et al.* [@ref64]). These findings support the stress-support matching hypothesis; support should match the type of adversity experienced in order to be most beneficial (Cohen & Wills, [@ref12]). This is also evidence for multidimensionality of resilient functioning and further evidences the value of developing a resilient index across domains of experiences. Such a measure has however yet to be reported. Thus, the concurrent and predictive role of adolescent friendships and family supports on adolescent psychosocial resilient functioning across multiple domains (whilst taking self-reported childhood family experiences into account) is unknown. Here, we investigate the relationship between adolescent family and friendship support on concurrent and prospective adolescent resilient PSF in a community sample (*N* = 2389) of healthy adolescents and young adults (ages 14--24) from the longitudinal Neuroscience in Psychiatry Network (NSPN; <http://www.NSPN.org>). We quantify a measure of resilient functioning by taking into account both functioning across multiple psychosocial domains (i.e. psychiatric symptoms, personality traits and mental wellbeing) and self-reported experiences of the family environment in childhood in a healthy population. This allows us to create a multidimensional index of functioning from which we can ascertain the degree to which an individual functions better or worse than expected given their family environment in early life. Finally, we test whether such functioning is associated specifically with family and friendship factors concurrently and prospectively 1 year later using path models. Methods {#sec2} ======= Sample {#sec2-1} ------ Participants in this report were part of the NSPN study cohort. NSPN is a multi-centre accelerated longitudinal community cohort study focusing on normative adolescent to young adult ('adolescent') development between the ages of 14 and 24. The NSPN cohort (*N* = 2389) completed a home questionnaire pack (HPQ) at baseline (Time 1), and \~1 year later \[median = 1 year, mean = 1.11 ([s.e.]{.smallcaps} = 0.01) year, min--max: 0.91--2.69 years\], *N* = 1674 individuals from the NSPN cohort completed the same HPQ at Time 2. For our cross-sectional analyses we had complete data on all measures used (online Supplementary Table S1) for *N* = 1890. This cross-sectional sample did not differ from the entire NSPN cohort (*N* = 2389) on age \[*t* (4055) = 0.02, *p* = 0.98\], gender (χ^2^ = 0.01, df = 1, *p* = 0.91), socio-economic status \[SES; index of multiple deprivation based on participant postcodes; *t* (4058) = 1.416, *p* = 0.16\], nor ethnicity (χ^2^ = 4.19, df = 5, *p* = 0.52). Overall, online Supplementary Table S1 shows that this sample (*N* = 1890) can be described as a healthy sample reporting low levels of psychopathological symptoms, behaviours and personality traits, and average mental wellbeing scores. For our longitudinal analyses we had complete data for *N* = 1093. This longitudinal sample was not different from the sample used in our cross-sectional analyses (*N* = 1890), nor the entire NSPN cohort (*N* = 2389) in terms of age \[*t* (2058 & 2218)\< −0.74, *p* \> 0.46\], and ethnicity distribution \[i.e. *N* = 1890: (χ^2^ = 2.73, df = 5, *p* = 0.74); *N* = 2389: (χ^2^ = 9.15, df = 5, *p* = 0.10)\]. However, there were slightly more females in the longitudinal sample (*N* = 1093; 57% females) when compared with the cross-sectional sample (*N* = 1890) and the NSPN cohort (*N* = 2389) (χ^2^ \> 4.16, df = 1, *p* \< 0.04), that both had 53% females. Finally, the longitudinal sample (*N* = 1093) had similar SES compared with the cross-sectional sample (*N* = 1890) *t* (2364) = −1.5, *p* = 0.13). However, the longitudinal sample (*N* = 1093) had lower SES compared with the NSPN cohort (*N* = 2389) (mean\'s 15.5 & 16.9, *t* (2237) = −2.82, *p* = 0.005). Measures {#sec2-2} -------- ### Psychosocial functioning (PSF) {#sec2-2-1} Negative family environments in early life form a risk factor for maladaptive-psychiatric symptomatology (van Harmelen *et al.* [@ref63]), -personality traits (Hart *et al.* [@ref25]; Rogosch & Cicchetti, [@ref46]) and reduced overall mental wellbeing (Hart *et al.* [@ref25]). Therefore, we focussed our measure of resilient functioning relative to these psychosocial domains to assess overall '*PSF*'[†](#fns01){ref-type="fn"}[^1^](#fn1){ref-type="fn"}. To do so, we included sum scores of all questionnaires (assessed both at times 1 and 2) that focussed on: Psychopathological symptoms: The mood and feelings questionnaire (Angold *et al.* [@ref2]), Revised Children\'s Manifest Anxiety Scale RCMAS self-report questionnaire (Reynolds & Richmond, [@ref45]), Short Leyton Obsessional Inventory (Bamber *et al.* [@ref3]), Kessler Psychological Distress scale (K10; Kessler *et al.* [@ref28]), behaviours checklist. Personality characteristics: The Antisocial Process Screening Device (Frick *et al.* [@ref21]), The Child and Adolescent Dispositions Scale (Lahey *et al.* [@ref31]), the inventory of Callous-unemotional traits (ICU) to measure callous and unemotional traits (Roose *et al.* [@ref47]), the Schizotypal Personality Questionnaire (SPQ) (Raine, [@ref43]) and the Barratt Impulsivity Scale (BIS) (Stanford *et al.* [@ref56]). Mental wellbeing: the Warwick-Edinburgh Mental Well Being Scale (WEMWBS) (Tennant *et al.* [@ref58]). More information about these measures is provided in the online Supplement. ### Childhood family experiences {#sec2-2-2} Appraisal of early life parenting behaviours were measured at baseline and time 2 with two self-report measures; the Alabama Parenting Questionnaire (APQ) and the measure of parenting styles (MOPS). *Measure of Parenting Style*: The MOPS is a 12-item self-report measure that assesses perceived parenting styles across three domains; indifference, overcontrol and abuse (Parker *et al.* [@ref39]). Participants were asked to rate both their mother\'s and father\'s parenting behaviour on 15 statements, on a 4-point scale. The full response range is 'not true at all', 'slightly true', 'moderately true', 'extremely true'. The 'abuse' scale consisted of five items, asking whether maternal/paternal behaviours were verbally abusive, unpredictable, physically violent, elicited feelings of danger or elicited feelings of lack of safety. The 'overly controlling' scale consisted of four items where maternal/paternal behaviour was overprotective, over controlling, critical, or made the participant feel guilty. Finally, the 'indifference' scale assessed six items of maternal/paternal behaviour where the parent was 'ignoring, uncaring, rejecting, uninterested in, would forget about, or would leave the participant on his/her own a lot. Sum scores to responses in these items were calculated with higher scores representing more abusive, over controlling or indifferent behaviour reported. Internal consistency was good for the maternal subscales (Cronbach\'s alpha maternal over control = 0.70, indifference = 0.86, abuse = 0.78). For paternal parenting, the internal consistency at baseline ranged from acceptable (Cronbach\'s alpha paternal over control = 0.65) to excellent (Cronbach\'s alphas paternal abuse = 0.88, paternal indifference = 0.93). *Alabama Parenting Questionnaire*: The APQ measures parenting practices. We used the nine-item short-form (Elgar *et al.* [@ref17]), and added the 'Corporal Punishment' (three items) and 'Involvement' scale (three items). Participants were asked to rate how typical each item occurred or used to occur in their family home on a 5-point scale ranging from '*never*', '*almost never*', '*sometimes*', '*often*' *to* '*always*'. We calculated sum scores for the five subscales: Positive Parenting, Inconsistent Discipline, Poor Supervision, Involvement, and Corporal Punishment, with higher scores reflecting higher frequency of the behaviour. Thus, high scores can indicate positive parenting (i.e. involvement, positive parenting) or negative parenting (i.e. inconsistent discipline, poor supervision, corporal punishment). Internal consistency at baseline was acceptable (inconsistent discipline & poor supervision: Cronbach\'s alpha \> 0.62) and good (positive parenting, involvement, Corporal Punishment Cronbach\'s alpha \> 0.71). Note that all results remained when the positive parenting scores (APQ positive parenting and APQ involvement) were removed from the analyses. ### Predictors of resilient functioning {#sec2-2-3} *Family Assessment Device (FAD)*: Adolescent family support was assessed at baseline and time 2 with the McMaster FAD-General Functioning Scale (FAD-GF; Epstein *et al.* [@ref18]), administered to adolescents. The FAD-GF is a 12-item self-report questionnaire where respondents rate statements such as 'we can express our feelings to each other' or 'there are lots of bad feelings in the family'. Responses ranged from 'Strongly Agree' to 'Strongly Disagree'. The FAD-GF yields an estimate of overall family functioning (Miller *et al.* [@ref36]). In our analyses, high scores reflect a positive family environment ('family support'). Internal consistency at baseline was very high (Cronbach\'s alpha = 0.92). *Cambridge Friendship Questionnaire (CFQ)*: Perceived quality of friendships at baseline and time 2 were assessed with the self-report CFQ (J. Memarzia *et al.* unpublished observations; van Harmelen *et al.* [@ref64]). The CFQ is an eight-item questionnaire assessing the number, availability and quality of friendships (e.g. 'Do you feel that your friends understand you?', 'Are you happy with the number of friends that you\'ve got at the moment', 'Can you confide in your friends?'). Higher scores indicate better perceived overall quality of friendships (i.e. 'Friendships'). The CFQ has good measurement invariance and external validity, and adequate test--retest reliability across 2-week intervals (Kappa = 0.80) (J. Memarzia *et al.* unpublished observations). Within NSPN, baseline internal consistency was good (Cronbach\'s alpha = 0.72). Stats and results {#sec2-3} ----------------- All analyses were conducted in R version 3.03 (Warm Puppy), using the packages Dplyr (Wickham & Romain, [@ref67]), Psych (Revelle, [@ref44]), Lavaan (Rosseel, [@ref48]) and ggplot2 (Wickham, [@ref66]). All data and code for the below analyses are available from (<http://www.annelauravanharmelen.com/data> & <https://figshare.com/authors/_/1376682>). To calculate a multi-modal composite score for PSF we conducted a principal component analysis (PCA) for PSF on standard-normally transformed individual total scores on the MFQ, RCMAS, S-LOI, K10, BCL, APSD, CADS, ICU, SPQ, BIS-11 and WEMBES. Similarly, we conducted a PCA, including standard-normally transformed sum scores for the MOPS the APQ subscales to create a composite score for childhood family experiences. From both analyses, we extracted individual scores for the first component to reflect individual current PSF and recalled childhood family experience scores. Next, we regressed the PSF component score against the childhood family experiences score, testing for possible linear, quadratic or cubic relationships. From the best-fitting regression we extracted the residual scores as these reflect a spectrum ranging from *risk to resilient* functioning: *the extent to which an individual has better, or worse, PSF outcomes than the average score expected given their childhood family experiences* (see for a similar approach Bowes *et al.* [@ref8]; Miller-Lewis *et al.* [@ref37]; Sapouna & Wolke, [@ref52]; Collishaw *et al.* [@ref13]). For parsimony, we will refer to these scores as '*resilient functioning*' with higher scores reflecting better (conditional) PSF outcomes. Next, we predicted resilient functioning from adolescent family and friendship support. Age, gender (coded 0-1, 1 being males) and socio-economic status (SES) were specified as covariates. Note that all results remained the same when these covariates were not included in the regressions. We examined these relations cross-sectionally at baseline in *N* = 1890 using multiple regression. Finally, we investigated whether the relationships between friendship and family support and our multidimensional measure of resilient functioning is dependent on the cross-sectional (i.e. simultaneous) timing of assessments, (potentially reflecting reporting bias) or whether these relationships also appeared over time. Therefore, we conducted longitudinal analyses using Structural Equation Modelling (SEM) in Lavaan (Rosseel, [@ref48]). We specified a full identified model that tested the relations and interrelations of baseline and later friendships, family support and resilient functioning. In this model, gender, age and SES were specified as covariates on friendships, family support and resilient functioning at baseline and follow-up. Results {#sec3} ======= Resilient functioning; functioning that is better than expected given one\'s childhood family experiences {#sec3-1} --------------------------------------------------------------------------------------------------------- A PCA for PSF (MFQ, RCMAS, S-LOI, K10, BCL, APSD, CADS, ICU, SPQ, BIS-11 and WEMBES) revealed a first component that explained 44% variance. Higher scores on the PSF factor suggest better PSF (see online Supplementary Table S2). The PCA for child family experiences revealed a first component that explained 37% variance in the MOPS and APQ subscales (online Supplementary Table S2 for loadings). The childhood family experiences principle component scores were inverted so that a higher score reflects more negative family experiences. We next regressed the childhood family experiences component score on the component score for PSF. This relationship could best be described as quadratic ([Fig. 1](#fig01){ref-type="fig"}) (Est = −0.76, [s.e.]{.smallcaps} = 0.03, *t* = −24.87, *p* \< 2 × 10^−16^, quadratric term: Est = 0.05, [s.e.]{.smallcaps} = 0.006, *t* = 7.32, *p* = 3.66 × 10^−13^, additional information in online Supplement). Next, individual residual scores were extracted from this relationship as these residuals reflect degree of risk to resilient functioning: *the extent to which an individual functioned better than expected* ('*high, or resilient*'; *green lines* [Fig. 1](#fig01){ref-type="fig"}), *or worse than expected* ('*low or risk*' *red lines* [Fig. 1](#fig01){ref-type="fig"}), *given their childhood family experiences.* Note that higher residual scores reflect *more* resilient functioning. Fig. 1.Relationship between PSF and Childhood family experiences in *N* = 1890. Association between adolescent friendships and family support and resilient functioning {#sec3-2} --------------------------------------------------------------------------------------- Adolescent friendships had a strong positive association with concurrent resilient functioning; more friendship support related to more resilient functioning \[*r* = 0.43, *t* (1834) = 20.57, *p* \< 2.2 × 10^−16^, [Fig. 2](#fig02){ref-type="fig"}*a*\]. Similarly, family support was positively associated with concurrent resilient functioning \[*r* = 0.23, *t* (1853) = 10.37, *p* \< 2.2 × 10^−16^, [Fig. 2](#fig02){ref-type="fig"}*b*\]. Fig. 2.The relationship between friendships (*a*) and family support (*b*) and baseline resilient functioning (*N* = 1890). Friendships and family support were correlated \]*r* = 0.39, *t* (1834) = 18.67, *p* \< 2.2 × 10^−16^\]. Therefore, we next investigated their unique relations with resilient functioning using multiple regression. We defined friendships, family support, gender, age and SES as predictors of resilient functioning. This analysis showed that friendships and family support were both positive predictors of resilient scores, with friendships being the strongest predictor ([Table 1](#tab01){ref-type="table"}). Furthermore, age and male gender, but not SES, were also associated with resilient functioning. Table 1.Predictors of resilient functioning at baseline (time 1)BaselineEstimateβ[s.e.]{.smallcaps}*tp* (\>\|*t*\|)Entire sampleFriendship0.210.410.0117.90\<2 × 10^−16^\*\*\*(*N* = 1890)Family0.020.060.012.810.00\*\*Age0.050.070.013.220.00\*\*Gender0.310.080.083.670.00\*\*\*SES0.00−0.010.00−0.400.69 Longitudinal predictors of resilient PSF {#sec3-3} ---------------------------------------- To investigate the relationship between friendships and family support at baseline (time 1) with resilient functioning at time 2 (\~1 year later) we recalculated resilient functioning scores in a subset of the sample that had complete data on all measures at both times (*N* = 1093; see online Supplement and Supplementary Table S4 for details). Resilient functioning at times 1 and 2 had a strong positive association (*r* = 0.66, *t* = 28.93, df = 1091, *p* \< 2.2 × 10^−16^), suggesting that resilient functioning is relatively stable over the course of 1 year in our sample. A path analysis showed that adolescent friendships and resilient functioning were significant *positive* predictors of psychosocial resilient functioning over the course of 1 year ([Table 2](#tab02){ref-type="table"} and [Fig. 3](#fig03){ref-type="fig"}). In contrast, adolescent family support was *negatively* associated with later psychosocial resilient functioning. Friendships at time 2 were only positively predicted by baseline resilient functioning; whereas family support at time 2 was positively predicted by baseline family *and* friendship support. Interestingly, family support at time 2 was negatively predicted by baseline resilient functioning. Note that as this path model is saturated, model fit is not informative, but included in the caption of [Table 2](#tab02){ref-type="table"} for completeness. Fig. 3.Significant paths in the Structural Equation Model. For reasons of parsimony we only depict significant positive (green) or negative (red) paths (unstandardised Estimates and [s.e.]{.smallcaps}). Thicker lines indicate stronger associations. Table 2.Predictors of later resilient functioning (N = 1093)Dependent variablePredictorsEstimate[s.e.]{.smallcaps}*z*-value*p* (\>\|*z*\|)Resilient functioning time 1Friendships time 10.2190.0239.4670.000Family time 10.0360.013.4940.000Age time 10.030.0211.4590.144Sex0.3780.1272.980.003SES0.0010.0050.1430.887Resilient functioning time 2Resilient functioning time 10.6530.03518.5630.000Friendships time 10.0410.0172.4350.015Family time 1−0.020.008−2.4220.015Age time 10.0190.0161.2320.218Sex0.0410.10.4090.682SES0.0050.0041.2860.199Family time 2Resilient functioning time 1−0.4590.114−4.0330.000Friendships time 10.2050.0812.5410.011Family time 10.690.04216.5160.000Age time 10.0190.0720.2690.788Sex−0.0390.451−0.0850.932SES−0.0160.018−0.8650.387Friendships time 2Resilient functioning time 10.1360.0771.7630.078Friendships time 10.5810.0599.7880.000Family time 10.0060.0210.2740.784Age time 1−0.0460.042−1.0950.274Sex−0.140.25−0.560.576SES−0.0070.009−0.840.401modelfitχ^2^(0) = 0, *p* = NA, CFI = 1, TLI = 1, RMSEA = 0(0--0) Discussion {#sec4} ========== Here we examine predictors of adolescent resilient functioning across a range of psychosocial domains ('PSF'; i.e. psychiatric symptoms, personality traits and mental wellbeing), whilst taking into account individual childhood family experiences. We create a measure of resilient PSF in three steps. First, we use a data-reduction technique (i.e. PCA) to establish individual composite scores for PSF, and childhood family experiences. Second, we regressed PSF on childhood family experiences. Third, we extract residual scores from this relationship as these reflect individual level of psychosocial resilient functioning: *the degree to which a participant is functioning better or worse than expected based on his/her childhood family experiences:* see for a similar approach (Bowes *et al.* [@ref8]; Miller-Lewis *et al.* [@ref37]; Sapouna & Wolke, [@ref52]; Collishaw *et al.* [@ref13]). We found that childhood family experiences have a significant association with PSF in our community sample of healthy adolescents (*N* = 1890). Specifically, recalling more negative family experiences was associated with worsening current PSF, supporting previous studies (Gilbert *et al.* [@ref23]; van Harmelen *et al.* [@ref63]). We then related adolescent friendship and family support with continuous risk to resilient PSF measure. We found that adolescent friendship support, but not adolescent family support, was positively related with immediate *and* later resilient PSF. Friendship and family support were both positive predictors of *immediate* resilient PSF. Notably, friendship support was a stronger predictor of immediate resilient functioning than family support, which is in line with the notion that adolescents are especially sensitive to their peer environment (Crone & Dahl, [@ref15]). Furthermore, we found that adolescent friendship support was also a positive predictor of *later* resilient PSF, which was apparent even after accounting for the effect of baseline resilient functioning and family support. These findings suggest that friendship support may be an important protective factor in adolescence. Our findings corroborate and extend those that showed that adolescent friendship support promotes subsequent resilient functioning in those exposed to negative childhood family environments (Collishaw *et al.* [@ref14]; Powers *et al.* [@ref42]; van Harmelen *et al.* [@ref64]). The exact mechanisms through which adolescent friendships increases resilient functioning are yet unknown. One potential explanation may be that our friendships score captures individual skills that promote social competence, such as social interaction and relationship building, and social competence could mediate the link between resilient functioning and friendship interactions. However, in our model, the relationship between baseline resilient functioning and later friendships was weak at best. This suggests that our interpretation that friendship promote resilient functioning over time is unlikely to be explained by the alternative notion that prior resilient functioning promotes better social competence (and friendships) and thereby later resilience function. Future studies should however test the specific role of social competence in the link between resilient functioning and subsequent friendships. Other explanations for the link between friendships and resilient functioning may come from studies that suggest that adolescent friendship support may increase resilient functioning through offering companionship (Cohen & Wills, [@ref12]) when these interactions are pro-social, as adolescent prosocial peer relationships, but not anti-social relationships, reduced later behavioural problems (Fergusson & Lynskey, [@ref19]; Fergusson *et al.* [@ref20]). Friendships may also increase resilient functioning is through increasing interpersonal skills (Buhrmester, [@ref10]), and through supporting social decision making skills (Jehn & Shah, [@ref27]). Additionally, adolescent friendships may reduce feelings of loneliness (Parker & Asher, [@ref40]), and dampen stress responses (Cohen & Wills, [@ref12]; Masten *et al.* [@ref35]). Furthermore, friendship support may increase resilient functioning through reducing negative experiences with peers (Pellegrini & Bartini, [@ref41]). Overall therefore these emotion--cognition mechanisms accruing via positive adolescent friendships may increase resilient functioning through updating negative self-cognitions. Negative self-cognitions are found in children that have low peer support; those that have experienced peer victimisation (Sinclair *et al.* [@ref54]), or report to be lonely (Vanhalst *et al.* [@ref61]). Negative self-cognitions colour individuals\' appraisal and behaviour in interpersonal situations and negatively influence individuals\' memories of these situations (Beck, [@ref4]). Negative self-cognitions mediate the link between very negative family environments and poor mental health (van Harmelen *et al.* [@ref63]). Adolescent friendship support may offer a unique opportunity to learn from positive peer experiences, which perhaps results in a more positive update of self-cognitions. Examining the potential mechanisms through which adolescent friendship support increases psychosocial resilient functioning is an important avenue for future research. The relationship between adolescent *family* support and resilient functioning across psychosocial domains appeared to be more complicated in our sample. Although family support had a *positive* relationship with immediate resilient PSF, family support was *negatively* related with *later* adolescent resilient functioning (when baseline resilient functioning and friendship support were taken into account). These findings are in line with findings that family support is not linked to positive adaptation in more severely maltreated children than those studied here (Cicchetti, [@ref11]). It may be that, in adolescence, family involvement is not adaptive, especially in the context of a negative family environment. In line with this idea, adolescent family support was not associated resilient functioning when peer relationships were taken into account (Fergusson & Lynskey, [@ref19]). Similarly, family support was not associated with teacher-reported mental health resilient functioning in young children with parental report of high cumulative family adversity (Miller-Lewis *et al.* [@ref37]). Although, family support was positively related with mental health resilient functioning if functioning was reported by parents in these children (Miller-Lewis *et al.* [@ref37]). Finally, our findings are in line with those that friendships, but not family support, are related with self-reported resilient functioning rates on a resilient functioning questionnaire in young adults with histories of child abuse (Howell & Miller-Graff, [@ref26]). However, our findings are in contrast with those that suggest that family support is predictive of childhood and early adolescent (ages 13--14) resilient functioning against depressive symptoms after child adversity (Bowes *et al.* [@ref8]; Sapouna & Wolke, [@ref52]). These findings also contrast our previous report in a different sample that adolescent family support at age 14 reduces adolescent depressive symptoms at age 17 after CFA (van Harmelen *et al.* [@ref64]). Taken together, whereas previous studies suggest that early adolescent family support may predict resilient functioning against later depression, our current findings suggest that adolescent family support is not related with later adolescent resilient functioning when resilient functioning is assessed across multiple psychosocial domains. Contrary to common concepts of resilient functioning where only the outcomes (e.g. absence of psychopathology, above average functioning) are taken into account \[e.g., see for an overview (Klika & Herrenkohl, [@ref30])\], we use an approach that allows individuals who have moderate outcomes in the face of very negative childhood family experiences to be included as 'resilient' (Bowes *et al.* [@ref7]; Miller-Lewis *et al.* [@ref37]; Sapouna & Wolke, [@ref52]; Collishaw *et al.* [@ref13]). This approach paints a more complete picture of adolescent PSF. A limitation of this approach is that taking the subjectivity of self-reported childhood family experiences into account when quantifying resilient functioning may be inherently biased: those that are highly resilient may report more positive childhood family experiences, whereas those that are less resilient may report more negative childhood family experiences. However, current psychopathology has not been found to bias self-report of child abuse and neglect (Spinhoven *et al.* [@ref55]). In fact, a previous work suggests that negative childhood experiences are more likely to be *underreported* rather than overreported (Brewin, [@ref9]). Finally, even if those with low psychosocial resilient functioning overreported negative family experiences, and those with high psychosocial resilient functioning over reported positive family experiences this would only lead to a *reduction* in power to find associations with resilient functioning. For these reasons, it is unlikely that this limitation would explain our current findings. Finally, an important limitation is that, on average, our sample reported only low levels of negative family experiences at best, and the childhood family experiences score explained only moderate variance *r* = 0.37% in the MOPS and APQ assessments. Future studies should investigate whether friendship support similarly predicts resilient PSF after more severe childhood family experiences including a sufficient sample of adolescents with manifest histories of physical and sexual maltreatment in childhood that are not studied in this investigation. In sum, we quantify resilient functioning by taking into account functioning across a range of psychosocial domains *and* individual childhood family experiences. We show that friendship support, but not family support, is an important positive predictor of both immediate and later resilient PSF in adolescence and early adulthood. Therefore, interventions that promote the skills needed to acquire and sustain adolescent affiliate friendships may be crucial in increasing adolescent resilient functioning. The notes appear after the main text. This work was supported by a strategic award from the Wellcome Trust to the University of Cambridge and University College London (095844/Z/11/Z), a Netherlands Organization for Scientific Research Rubicon grant (AlvH, NO 446-13-006), and a Royal Society Dorothy Hodgkin Fellowship (AlvH; No DH150176). Study data were collected and managed using REDCap electronic data capture tools hosted at the University of Cambridge. P.F is in receipt of a National Institute for Health Research (NIHR) Senior Investigator Award (NF-SI-0514-10157)and is in part supported by the NIHR Collaboration for Leadership in Applied Health Research and Care (CLAHRC) North Thames at Barts Health NHS Trust. E.T.B. is employed half-time by the University of Cambridge and half-time by GlaxoSmithKline; he holds stock in GlaxoSmithKline. The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript. We would like to thank the NSPN sample for their participation. This study was designed and analysed by AlvH. The corresponding author (AlvH) had full access to the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. All authors contributed to the interpretation of the data, the writing of the paper, and approved the final manuscript. As our aim is to capture resilient functioning in its broadest sense, we choose to incorporate the more stable personality traits in our measure of PSF. Note that our results remain the same when we repeated all analyses, whilst only including psychiatric symptoms and mental wellbeing in our PSF variable. Supplementary material {#sec5} ====================== For supplementary material accompanying this paper visit http://dx.doi.org/10.1017/S0033291717000836. ###### click here to view supplementary material None. The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.

All relevant data are within the paper and its Supporting Information files. Introduction {#sec001} ============ Na^+^/H^+^ exchangers are ubiquitous in nature as they ensure that organisms are capable of regulating their intracellular Na^+^ concentration, pH and volume \[[@pone.0182293.ref001]\]. The best studied Na^+^/H^+^ exchangers belong to the cation proton antiporter (CPA) superfamily, which includes both electrogenic and electroneutral members \[[@pone.0182293.ref002]\]. To this date, the crystal structures of four CPA Na^+^/H^+^ exchangers have been solved, beginning with the prototype of the family, NhaA from *E*. *coli* (EcNhaA) \[[@pone.0182293.ref003]\] and continuing in recent years with the structures of NapA from *T*. *thermophilus* \[[@pone.0182293.ref004]\], NhaP from *P*. *abyssi* \[[@pone.0182293.ref005]\] and NhaP1 from *M*. *jannaschii* \[[@pone.0182293.ref006]\]. We have previously shown that CPA Na^+^/H^+^ exchangers can be described by a simple kinetic model that explains their pH-dependent activity profile by competition of the two substrates, H^+^ and Na^+^, for a common binding site \[[@pone.0182293.ref007]\]. By comparison, less attention was given to the characterization of non-CPA Na^+^/H^+^ exchangers, which, despite having no homologues in humans, are present in prokaryotes and play roles in their Na^+^ or pH homeostasis. These include, but are not limited to, members of the NhaB \[[@pone.0182293.ref008]\], NhaC \[[@pone.0182293.ref009]\], or NhaD \[[@pone.0182293.ref010]\] families. Besides their value as model system for human homologues, investigation of prokaryotic Na^+^/H^+^ exchangers can serve two valuable roles--a better understanding of the general principles of the Na^+^/H^+^ exchange mechanism and, potentially, the development of novel antibacterial treatments, taking into account that the NhaA and NhaB exchangers have been shown to be involved in the virulence of *Y*. *pestis* \[[@pone.0182293.ref011]\]. In this respect, the absence of a NhaB homologue in humans may represent an advantage for the development of a side effect-free NhaB targeted antibiotic. *Klebsiella pneumoniae* is facultative anaerobic rod-shaped bacterium belonging to the Enterobacteriaceae family that is ubiquitously found in nature, either in the environment or on the mucosal surfaces of mammals, including humans \[[@pone.0182293.ref012]\]. In humans, *Klebsiella* infects the respiratory and urinary tracts \[[@pone.0182293.ref012]\] and is a major source of hospital-acquired infections, threatening especially patients which are immuno-compromised and neonates in intensive care units \[[@pone.0182293.ref013]\]. A cause of major concern is the fact that, over the years, *K*. *pneumoniae* has acquired resistance to carbanepems, the first-choice drugs used in the treatment of *K*. *pneumoniae* infections \[[@pone.0182293.ref014]\]. The prevalence of carbanepem-resistant *K*. *pneumoniae* strains is continuously increasing, with one of the countries most severely affected being Greece, where in 2014, more than 60% of *K*. *pneumoniae* isolates from hospital wards were carbanepem-resistant \[[@pone.0182293.ref015]\]. A recently published systematic review of the literature points to a 42% pooled mortality rate of patients infected with carbanepem-resistant *K*. *pneumoniae*, double than that for those infected with carbanepem-susceptible strains \[[@pone.0182293.ref016]\]. Furthermore, it has been reported that *K*. *pneumoniae* strains have also started to develop resistance to last resort treatments such as colistin (polymyxin) \[[@pone.0182293.ref016]\]. In an effort to raise awareness regarding antibiotic-resistant bacteria, the World Health Organization has named carbanepem-resistant Enterobacteriaceae as Priority 1 pathogens against which antibiotic therapies have to be designed \[[@pone.0182293.ref017]\]. The genome of *K*. *pneumoniae* encodes for four Na^+^/H^+^ exchangers, two belonging to the NhaA family (KpNhaA1 and KpNhaA2), one belonging to the NhaB family (KpNhaB) and one belonging to the NhaP family (KpNhaP2). Out of these, according to the Transporter Classification system TC \[[@pone.0182293.ref018]\], three exchangers (KpNhaA1, KpNhaA2 and KpNhaP2) belong to the CPA superfamily, while KpNhaB belongs to the ion transporter (IT) superfamily. In terms of physiological role, NhaA and NhaB exchangers have been shown to mediate electrogenic Na^+^ (or Li^+^) export from the cell \[[@pone.0182293.ref019]\], having either a 2:1 H:Na^+^ stoichiometry (NhaA) \[[@pone.0182293.ref020]\] or a 3:2 H^+^:Na^+^ stoichiometry (NhaB) \[[@pone.0182293.ref021]\]. The role of the electroneutral NhaP exchangers is to regulate intracellular pH via import of Na^+^ ions \[[@pone.0182293.ref022]\]. No studies have so far been published regarding the Na^+^/H^+^ exchangers of *K*. *pneumoniae*. A sequence alignment of the NhaA and NhaB exchangers investigated in this work with homologues from other species is shown in [Fig 1](#pone.0182293.g001){ref-type="fig"}. Extensive research on EcNhaA has revealed specific structural features and conserved residues that are critical to its function. Thus, TMs IV and XI ([Fig 1A](#pone.0182293.g001){ref-type="fig"}) are each interrupted by an unwound region \[[@pone.0182293.ref003]\], leading to the existence of short helices oriented either towards the cytoplasm (IVc, XIc) or towards the periplasm (IVp, XIp). The partial positive dipoles of helices IVc and XIp are stabilized by the presence of a negatively charged residue (Asp-133 in EcNhaA), which has been shown to be part of the substrate binding site \[[@pone.0182293.ref023]\] of EcNhaA. The partial negative dipoles of helices IVp and XIc are stabilized by Lys-300, which has recently been shown to be essential for the stability of the EcNhaA transporter \[[@pone.0182293.ref024]\]. Two neighbouring negatively charged residues, Asp-163 and Asp-164, are also part of the substrate binding site of EcNhaA \[[@pone.0182293.ref025]\], with mutations in these residues making the transporter inactive \[[@pone.0182293.ref026]\]. As it can be seen ([Fig 1A](#pone.0182293.g001){ref-type="fig"}), all these residues are conserved in the sequences of KpNhaA1 and KpNhaA2. ![Sequence alignment of *K*. *pneumoniae* Na^+^/H^+^ exchangers.\ A, Sequence alignment of NhaA exchangers from *K*. *pneumoniae* (KpNhaA1, KpNhaA2) and *E*. *coli* (EcNhaA). B, Sequence alignment of NhaB exchangers from *K*. *pneumoniae* (KpNhaB), *E*. *coli* (EcNhaB) and *V*. *alginolyticus* (VaNhaB). Horizontal lines denote transmembrane helices in EcNhaA (A) or VaNhaB (B), as determined in \[[@pone.0182293.ref003]\] and \[[@pone.0182293.ref027]\], respectively. Red asterisks show conserved charged residues shown to be essential for transport function or stability of the exchangers. Alignments were performed using Clustal Omega \[[@pone.0182293.ref028]\] and drawn using Jalview \[[@pone.0182293.ref029]\].](pone.0182293.g001){#pone.0182293.g001} Unlike NhaA or NhaP transporters, which have 12 (or 13) transmembrane helices (TMs) and were shown to belong to the same structural fold termed the "NhaA fold" \[[@pone.0182293.ref030]\], NhaB was shown to have only 9 TMs \[[@pone.0182293.ref027]\]. Additionally, no clear homologues for the functionally relevant residues described for NhaA exist in the sequence of NhaB exchangers ([Fig 1B](#pone.0182293.g001){ref-type="fig"}). A mutational study done on the only negatively charged residue that seems to be present close to the middle of a TM in NhaB was performed in NhaB of *V*. *alginolyticus*, where mutation of Asp-147 to either Gly, Thr, Met or Glu abolished Na^+^/H^+^ exchange activity, but, interestingly, not Na^+^/Na^+^ exchange activity of the transporter \[[@pone.0182293.ref031]\]. The purpose of this work was the functional characterization of the KpNhaB Na^+^/H^+^ exchanger by using solid-supported membrane (SSM)-based electrophysiology \[[@pone.0182293.ref032]\] as the main investigation technique. Additionally, we also characterized KpNhaA1 and KpNhaA2 and compared their transport activities with that of KpNhaB. Our study represents the first electrophysiological investigation of an NhaB Na^+^/H^+^ exchanger. We found that the competition-based kinetic model \[[@pone.0182293.ref007]\] that is valid for CPA Na^+^/H^+^ exchangers also describes the non-CPA KpNhaB transporter. Additionally, we found that, despite their high homology to EcNhaA, KpNhaA1 and KpNhaA2 have altered activity profiles. Overall, the profiles of the three investigated Na^+^/H^+^ exchangers from *K*. *pneumoniae* are complementary, indicating that they contribute to the survival of the bacterium under different conditions of salinity and pH. Results {#sec002} ======= Overexpression of KpNhaB, KpNhaA1 and KpNhaA2 in E. coli {#sec003} -------------------------------------------------------- The genes encoding for KpNhaB, KpNhaA1 and KpNhaA2, containing a C-terminal His-tag and cloned in either the pET21d or the pTrcHis2 TOPO expression vectors, were used to transform *E*. *coli* strains BL21(DE3) or KNabc. Expression of the target proteins in *E*. *coli* membranes was verified by performing SDS-PAGE followed by Western blot, using anti-His IgG as primary antibody ([Fig 2A](#pone.0182293.g002){ref-type="fig"}). Clear bands could be observed for all three proteins on the Western blot. As NhaA was shown to be a dimer \[[@pone.0182293.ref033]\], we assigned the higher mass bands observed for KpNhaA1 and KpNhaA2 as the dimer forms of the protein. The existence of dimer bands even in the presence of the harsh detergent SDS has previously been reported for CPA Na^+^/H^+^ exchangers \[[@pone.0182293.ref034]\]. No dimer band was observed for KpNhaB. In all three cases, the monomeric forms of the proteins migrated on the gel to masses lower than their calculated molecular weights (KpNhaB-- 57 kDa; KpNhaA1--43 kDa; KpNhaA2--41 kDa), which is typical for membrane proteins \[[@pone.0182293.ref035]\]. {#pone.0182293.g002} Following solubilization of the *E*. *coli* membranes and purification using immobilized metal affinity chromatography, proteins were subjected again to SDS-PAGE and visualized using Coomassie Blue staining of the gel ([Fig 2B](#pone.0182293.g002){ref-type="fig"}). Essentially only the same bands seen in [Fig 2A](#pone.0182293.g002){ref-type="fig"} were observed, indicating that proteins were purified to a high degree. Survival assays of E. coli KNabc expressing the K. pneumoniae transporters {#sec004} -------------------------------------------------------------------------- In order to check the functionality of the expressed transporters in *E*. *coli* 7and also to verify their role as Na^+^ export systems, we assayed the capability of the *K*. *pneumoniae* exchangers of rescuing the survival of the Na^+^/H^+^ exchanger-deficient *E*. *coli* KNabc strain under conditions of high salinity. Under the conditions tested ([Table 1](#pone.0182293.t001){ref-type="table"}), we observed that *E*. *coli* KNabc could only grow in medium where Na^+^ was replaced by K^+^ (LBK), while high concentrations of either Na^+^ or Li^+^ prevented the growth of the vector-transformed strain. When either the control NhaA from *H*. *pylori* (HpNhaA) or one of the *K*. *pneumoniae* Na^+^/H^+^ exchangers were expressed, survival of the cells was restored under high amounts of Li^+^ and Na^+^ at pH 7. However, at pH 8.3 only KpNhaA1 and HpNhaA could provide survival in presence of a high Li^+^ concentration, while none of the exchangers could provide survival under a high Na^+^ concentration. 10.1371/journal.pone.0182293.t001 ###### Survival of the Na^+^/H^+^ exchanger-deficient *E*. *coli* strain KNabc overexpressing *K*. *pneumoniae* Na^+^/H^+^ exchangers. {#pone.0182293.t001g} Exchanger LBK LBK pH 7 0.6 M NaCl LBK pH 7 0.1 M LiCl LBK pH 8.3 0.6 M NaCl LBK pH 8.3 0.1 M LiCl ------------- ----- --------------------- --------------------- ----------------------- ----------------------- **KpNhaB** ++ \+ ++ -- -- **KpNhaA1** +++ +++ +++ -- ++ **KpNhaA2** ++ \+ ++ -- -- **HpNhaA** +++ +++ +++ -- +++ **Vector** +++ -- -- -- -- Survival was assessed under different salt concentration and pH values. Experiments were repeated at least three times with virtually identical results. NhaA from *H*. *pylori* (HpNhaA) was used as a positive control, while empty pTrcHis2TOPO vector was used as a negative control. +++ = maximum number of colonies, ++ = moderate number of colonies, + = small number of colonies,-- = no growth detected. LBK = LB Broth containing K^+^ instead of Na^+^. Acridine orange dequenching assays {#sec005} ---------------------------------- In order to ascertain the functionality of the overexpressed proteins in *E*. *coli* membranes, acridine orange dequenching assays were performed by using everted *E*. *coli* membrane vesicles that were acidified by addition of Tris-D-lactate. All three transporters were active ([Fig 3A--3C](#pone.0182293.g003){ref-type="fig"}), as shown by the dequenching of the acridine orange fluorescence observed when Na^+^ was added to the outside of the vesicles. As EcNhaB, which has high homology to KpNhaB ([Fig 1B](#pone.0182293.g001){ref-type="fig"}), was reported to show a pH-independent activity \[[@pone.0182293.ref036], [@pone.0182293.ref037]\], we recorded the pH dependence of the dequenching for KpNhaB and found essentially no pH dependence in the pH range tested ([Fig 3D](#pone.0182293.g003){ref-type="fig"}). {#pone.0182293.g003} SSM-based electrophysiological measurements {#sec006} ------------------------------------------- Proteoliposomes containing the reconstituted *K*. *pneumoniae* Na^+^/H^+^ exchangers were investigated via SSM-based electrophysiology. Thus, the proteoliposomes were subjected to Na^+^ concentration jumps under conditions where the pH was the same inside and outside the proteoliposomes (symmetrical pH). In the case of all exchangers, transient currents of negative polarity were recorded ([Fig 4A--4C](#pone.0182293.g004){ref-type="fig"}). The negative polarity of these currents indicates the transport of positive charge out of the proteoliposomes (or net negative charge inside), and is in line with the expected electrogenicity of the Na^+^/H^+^ exchange. Concentration jumps of Li^+^ (data not shown) were also performed, yielding transient currents of the same polarity and shape as the ones recorded for Na^+^ concentration jumps. {#pone.0182293.g004} pH dependence of the transport activity {#sec007} --------------------------------------- As previously shown \[[@pone.0182293.ref038]\], the amplitude of the recorded transient currents is a good measure of the steady-state transport activity of the exchanger, in the case where the recorded transient currents show no pre steady-state component. This was true for nearly all of the recordings done on the *K*. *pneumoniae* Na^+^/H^+^ exchangers, with the exception of Na^+^ concentration jumps on KpNhaB performed at pH 9 and pH 9.5, where the transient currents showed a pre steady-state component ([Fig 4A](#pone.0182293.g004){ref-type="fig"}). As this component appears only at high pH, it is likely associated with the Na^+^ binding or translocation event in the transporter's reaction cycle, as was previously seen in the EcNhaA G338S mutant \[[@pone.0182293.ref038]\] or in the NhaP exchangers from *M*. *jannaschii* \[[@pone.0182293.ref022]\] and *P*. *abyssi* \[[@pone.0182293.ref039]\]. For these recordings, the transport currents were numerically reconstructed \[[@pone.0182293.ref040]\] and the stationary component of the reconstructed currents was determined ([S1 Fig](#pone.0182293.s001){ref-type="supplementary-material"}). The pH-dependent activity profile of KpNhaB determined by SSM-based electrophysiology is shown in [Fig 4D](#pone.0182293.g004){ref-type="fig"}. A high pH dependence of the transport activity was observed ([Fig 4D](#pone.0182293.g004){ref-type="fig"}), unlike the profile determined by acridine orange dequenching measurements ([Fig 3D](#pone.0182293.g003){ref-type="fig"}). The exchanger was down-regulated in the acidic, with the activity increasing with pH up to a maximum that is reached at pH 8--8.5, after which activity decreased rapidly, reaching almost zero at pH 9.5 ([Fig 4D](#pone.0182293.g004){ref-type="fig"}). By comparison, KpNhaA1 and KpNhaA2 ([Fig 4E and 4F](#pone.0182293.g004){ref-type="fig"}) showed the same down-regulation in the acidic, but the activity remained high even at higher pH. Na^+^ and Li^+^ dependence of the transport activity {#sec008} ---------------------------------------------------- At pH 8.5, increasing the value of the Na^+^ concentration jumps yielded a hyperbolic increase of the transporter-dependent currents, which was observed in all the investigated transporters ([Fig 5A--5F](#pone.0182293.g005){ref-type="fig"}). The determined values for the Na^+^ apparent affinities (K~m~ values) at pH 8.5 are similar for all three transporters, in the range of 7--14 mM Na^+^, and are also close to the value of 11 mM previously determined for EcNhaA at the same pH ([Table 2](#pone.0182293.t002){ref-type="table"}). {#pone.0182293.g005} 10.1371/journal.pone.0182293.t002 ###### Kinetic parameters of *K*. *pneumoniae* Na^+^/H^+^ exchangers. {#pone.0182293.t002g} Exchanger I~max~ (nA) $\mathbf{K}_{\mathbf{m}}^{\mathbf{N}\mathbf{a}}\ \left( \mathbf{p}\mathbf{H} \right)$ $\mathbf{K}_{\mathbf{m}}^{\mathbf{L}\mathbf{i}}\ \left( \mathbf{p}\mathbf{H} \right)$ pK $\mathbf{K}_{\mathbf{D}}^{\mathbf{N}\mathbf{a}}$ k~2~/k~1~ *n* *m* --------------------------------------------------- ------------- --------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------- --------------- -------------------------------------------------- ----------- ----- ----- ----- **KpNhaB** 6.1 ± 1 7 ± 1 (8.5) 74 ± 8 (7.0) 14 ± 4 (8.5) 8 3.6 23 1.3 1.6 **KpNhaA1** 3.3 ± 1 14 ± 4 (8.5) 210 ± 78 (7.5) 2 ± 0.2 (8.5) 9.2 1.6 13 1 1 **KpNhaA2** 4.3 ± 1 9 ± 2 (8.5) 46 ± 16 (7.5) 3 ± 0.5 (8.5) 8.4 2.6 100 1 1 **EcNhaA[^\*^](#t002fn002){ref-type="table-fn"}** 12 ± 2 11 ± 1 (8.5) 102 ± 7 (7.5) 7 (8) 8.8 3 7 1 1 Values were determined following the fit of the model to the experimental data. Hill coefficients indicating cooperative Na^+^ (*n*) and H^+^ (*m*) binding were introduced for KpNhaB and fixed as 1 for the NhaA transporters which displayed no cooperative behavior. I~max~ indicates the maximum transient current amplitude recorded for the transporter. ^\*^from Refs. \[[@pone.0182293.ref038], [@pone.0182293.ref041]\]. For EcNhaA, the I~max~ values were recorded for proteoliposomes reconstituted at a lipid to protein ratio of 5, while an LPR of 10 was used for the *K*. *pneumoniae* exchangers. When pH was lowered, however, the peak currents recorded for KpNhaB at pH 7 showed a sigmoidal dependence ([Fig 5D](#pone.0182293.g005){ref-type="fig"}), indicating cooperativity in the substrate binding. In this case, the data could be fitted by the use of a Hill equation, yielding a Hill coefficient *n* = 2. The dependence of the currents on the Na^+^ concentration remained hyperbolical for KpNhaA1 ([Fig 5E](#pone.0182293.g005){ref-type="fig"}) and KpNhaA2 ([Fig 5F](#pone.0182293.g005){ref-type="fig"}). In all investigated exchangers, lowering pH had the effect of decreasing the affinity for Na^+^ ([Fig 5](#pone.0182293.g005){ref-type="fig"} and [Table 2](#pone.0182293.t002){ref-type="table"}), indicating the presence of competition between Na^+^ and H^+^. We also performed Li^+^ concentration jumps at pH 8.5, given that most Na^+^/H^+^ exchangers are also capable of transporting Li^+^. KpNhaB displayed a twofold lower affinity for Li^+^ than for Na^+^ ([Fig 5G](#pone.0182293.g005){ref-type="fig"}). For KpNhaA1 ([Fig 5H](#pone.0182293.g005){ref-type="fig"}) and KpNhaA2 ([Fig 5I](#pone.0182293.g005){ref-type="fig"}), the affinities for Li^+^ were higher than the affinity for Na^+^, which is in line with the behavior previously seen in EcNhaA ([Table 2](#pone.0182293.t002){ref-type="table"}). Determination of kinetic parameters {#sec009} ----------------------------------- [Fig 6](#pone.0182293.g006){ref-type="fig"} shows the kinetic model describing the transport mechanism of Na^+^/H^+^ exchange, which is in accordance to the alternating access model proposed by Jardetzky \[[@pone.0182293.ref042]\]. In brief, the transporter can bind either of the substrates H^+^ and Na^+^ to the same binding site in either its inward- or outward-open conformations. This causes a conformational transition that exposes the bound substrate ion to the opposite side of the membrane, where it is released. One of the advantages of this simple model is that, in the absence of cooperative binding, the activity of each transporter can be described by a set of only 3 kinetic parameters ([Fig 6](#pone.0182293.g006){ref-type="fig"}), corresponding to the affinity of the transporter for H^+^ (pK), the affinity for Na^+^ ($K_{D}^{Na}$) and the ratio between the rates of H^+^ and Na^+^ transport (k~2~/k~1~). {#pone.0182293.g006} We previously used the competition-based kinetic model in order to characterize several CPA exchangers, including transporters from the NhaA and NhaP families \[[@pone.0182293.ref007]\]. The pH-dependent activity profile of KpNhaB ([Fig 4D](#pone.0182293.g004){ref-type="fig"}) as well as the existence of competition ([Fig 5D](#pone.0182293.g005){ref-type="fig"}) indicated that the model might also be appropriate for the characterization of this non-CPA exchanger. In order to account for the cooperativity observed for KpNhaB at pH 7 we included two additional kinetic parameters, which are the Hill coefficients *n* for the Na^+^ binding reaction and *m* for the H^+^ binding reaction, respectively ([Fig 6](#pone.0182293.g006){ref-type="fig"}). A kinetic analysis of each of the investigated exchangers using the competition-based model was performed. [Table 2](#pone.0182293.t002){ref-type="table"} shows the obtained kinetic parameters for the investigated exchangers, and the fits of the model to the experimental data are shown in [Fig 7](#pone.0182293.g007){ref-type="fig"}. The *m* and *n* Hill coefficients were fixed as 1 for the NhaA exchangers, where no cooperativity was apparent. {ref-type="table"}. Exp. = experimentally determined data points, fit = modeled curve following the fit of the kinetic model to the experimental data.](pone.0182293.g007){#pone.0182293.g007} As it can be seen ([Table 2](#pone.0182293.t002){ref-type="table"}), the three transporters differ in their pK values, with KpNhaB being the most acid-shifted and KpNhaA1 the most alkaline-shifted. The determined $K_{D}^{Na}$ values are similar, in the range of 1.6--3.6 mM Na^+^. Regarding k~2~/k~1~, a major aspect of obtaining a precise value is measuring the activity of the exchanger in a range where down-regulation at both acidic and alkaline pH can be observed. While this was the case for KpNhaB and KpNhaA1, where we obtained values for k~2~/k~1~ of 23 and 13, respectively, in the case of KpNhaA2 this value was less defined, most likely due to the weak down-regulation in the alkaline. Thus, although our fit gave a value of \~2000 to k~2~/k~1~ for KpNhaA2, we assigned it a k~2~/k~1~ value of 100, as values above 100 for this parameter yielded essentially identical pK and $K_{D}^{Na}$ values. Discussion {#sec010} ========== KpNhaB, KpNhaA1 and KpNhaA2 are high turnover, electrogenic Na^+^ export systems {#sec011} -------------------------------------------------------------------------------- EcNhaA and EcNhaB have previously been shown to play a vital role in the survival of *E*. *coli* under conditions of high salinity \[[@pone.0182293.ref019]\]. Indeed, *E*. *coli* strains deficient in EcNhaA and EcNhaB such as the EP432 strain \[[@pone.0182293.ref043]\] or the KNabc strain that we employed in this work \[[@pone.0182293.ref044]\] do not survive under conditions of high Na^+^ or Li^+^ concentration. As our results ([Table 1](#pone.0182293.t001){ref-type="table"}) show, expressing either of the electrogenic *K*. *pneumoniae* Na^+^/H^+^ exchangers in *E*. *coli* KNabc restored the salt resistance of this strain at neutral pH. Conversely, at pH 8/0.6 M Na^+^, none of the expressed Na^+^/H^+^ exchangers, including the control HpNhaA, were able to restore resistance. This is in line with the higher susceptibility of *E*. *coli* KNabc to salt stress compared to other Na^+^/H^+^ exchanger-deficient strains such as EP432 \[[@pone.0182293.ref045], [@pone.0182293.ref046]\], and most likely results from the fact that, besides EcNhaA and EcNhaB, *E*. *coli* KNabc is missing also the unspecific ChaA transporter. Thus, considering our results as well as the role of the NhaA and NhaB exchangers in other organisms \[[@pone.0182293.ref019], [@pone.0182293.ref045]\], we can conclude that the KpNhaB, KpNhaA1 and KpNhaA2 exchangers contribute to salt resistance in *K*. *pneumoniae*. This conclusion is additionally substantiated by the pH-dependent activity profiles of the exchangers ([Fig 4D--4F](#pone.0182293.g004){ref-type="fig"}), which are in line with the expected profile for an exchanger that has as a main role Na^+^ export \[for an in-depth discussion, see Refs. [@pone.0182293.ref007] and [@pone.0182293.ref022]\]. An analysis of the transient currents recorded using SSM-based electrophysiology following Na^+^ (or Li^+^) concentration jumps clearly shows that all investigated exchangers are electrogenic, carrying net positive charge out of the proteoliposomes (or net negative charge inside) when Na^+^ is applied from the outside. This conclusion is substantiated by the negative polarity of the recorded transients, the decay time constants that are Na^+^ concentration dependent ([Fig 5A--5C](#pone.0182293.g005){ref-type="fig"}) and by the increase of the decay time constants of the transient currents with the increase of the lipid to protein ratio (LPR, [S2 Fig](#pone.0182293.s002){ref-type="supplementary-material"}), as previously shown for EcNhaA by Zuber *et al*. \[[@pone.0182293.ref041]\]. In addition, based on the maximum amplitude of the transient currents recorded for all *K*. *pneumoniae* Na^+^/H^+^ exchangers ([Table 2](#pone.0182293.t002){ref-type="table"}), we can conclude that all of them are high turnover systems, as the recorded amplitudes are comparable with those previously recorded for EcNhaA ([Table 2](#pone.0182293.t002){ref-type="table"}), which has a turnover of more than 1000 ions/second \[[@pone.0182293.ref047]\]. KpNhaB can be described by the same competition-based mechanism as the NhaA exchangers {#sec012} -------------------------------------------------------------------------------------- Our previous research activity has led to a simple competition-based transport mechanism ([Fig 6](#pone.0182293.g006){ref-type="fig"}) that we have shown to be valid for Na^+^/H^+^ exchangers of the CPA superfamily \[[@pone.0182293.ref007]\]. So far, the transporters that were shown to follow this mechanism were the NhaA exchangers from *E*. *coli* \[[@pone.0182293.ref038]\], *H*. *pylori* \[[@pone.0182293.ref022]\] and *S*. *typhimurium* \[[@pone.0182293.ref048]\], the NhaP exchangers from *M*. *jannaschii* \[[@pone.0182293.ref022]\] and *P*. *abyssi* \[[@pone.0182293.ref039]\] and the NapA exchanger from *T*. *thermophilus* \[[@pone.0182293.ref049]\]. The mechanism explains the highly pH-dependent activity of Na^+^/H^+^ exchangers as a consequence of its transport mechanism, not necessitating additional pH-sensitive regions (so-called "pH sensors") other than the substrate binding site. As we expected, the characterization of KpNhaA1 and KpNhaA2 was fully in line with the competition-based mechanism. Thus, competition is readily apparent in both exchangers as shown by the reduction of the Na^+^ affinity when pH decreases ([Fig 5](#pone.0182293.g005){ref-type="fig"}). Interestingly, we found that KpNhaB can equally be described by the competition mechanism. Thus, competition can be observed in KpNhaB by the same decrease in Na^+^ affinity with the lowering of pH as observed for the NhaA exchangers ([Fig 5D](#pone.0182293.g005){ref-type="fig"}). An intrinsic property of this mechanism ([Fig 6](#pone.0182293.g006){ref-type="fig"}) is that transport is down-regulated at extreme pH: in the acidic, down-regulation is explained by the H^+^ out-competing the Na^+^ ions, while in the alkaline, it is explained by H^+^ depletion that reduces the overall availability of this substrate and hence, turnover. Indeed, the pH profile of KpNhaB ([Fig 4D](#pone.0182293.g004){ref-type="fig"}) clearly shows both alkaline and acidic down-regulation while for KpNhaA1 ([Fig 4E](#pone.0182293.g004){ref-type="fig"}) and KpNhaA2 ([Fig 4F](#pone.0182293.g004){ref-type="fig"}) only acidic down-regulation is observed, most probably because alkaline down-regulation is shifted out of the experimentally accessible pH-range. Our finding for KpNhaB is especially relevant taking into account that previous reports regarding the pH dependence of NhaB exchangers gave mixed conclusions \[[@pone.0182293.ref008], [@pone.0182293.ref036]\] and that NhaB belongs to a different transporter family than NhaA. In addition, NhaB has a distinct structure as far as conclusions from sequence analysis are predictive in this respect. We will treat these considerations in turn. The first investigated NhaB exchanger was NhaB from *E*. *coli*, which was identified more than 25 years ago \[[@pone.0182293.ref050]\]. One of the earliest reported properties of EcNhaB was that its transport activity seemed independent of pH in the range 6.5 to 8.5, as seen in acridine orange dequenching assays \[[@pone.0182293.ref036], [@pone.0182293.ref037]\]. On the other hand, the activity of the NhaB exchanger from *V*. *alginolyticus* (VaNhaB) was shown to be highly pH dependent using the same assay \[[@pone.0182293.ref051]\]. The acridine orange dequenching assays performed on KpNhaB ([Fig 3D](#pone.0182293.g003){ref-type="fig"}) also showed no pH dependence of the transport activity in the same pH range. However, our electrophysiological results clearly show that the activity of KpNhaB is highly pH dependent ([Fig 4D](#pone.0182293.g004){ref-type="fig"}). Therefore, the apparent "pH insensitivity" observed in the acridine orange dequenching assay is most probably due to the limited dynamic range of the dequenching assays, as previously reported for *H*. *pylori* NhaA \[[@pone.0182293.ref052]\]. Here it is interesting to note that KpNhaB, just as *H*. *pylori* NhaA \[[@pone.0182293.ref052]\], has an acid-shifted activity profile with respect to EcNhaA. Using only the acridine orange dequenching assay, KpNhaB might, therefore, wrongly be qualified as "pH insensitive", as HpNhaA was previously thought to be \[[@pone.0182293.ref052]\]. Thus, it is important that, when such considerations are made, a more sensitive assay than acridine orange dequenching is used. According to the TC system \[[@pone.0182293.ref018]\], NhaB exchangers are classified into the Ion Transporter (IT) superfamily, while NhaA and NhaP exchangers belong to the Cation Proton Antiporter (CPA) superfamily. Furthermore, a topological study of VaNhaB showed that NhaB exchangers possess 9 transmembrane helices (TMs), unlike the 12 or 13 TMs present in NhaA or NhaP exchangers \[[@pone.0182293.ref027]\]. Also, NhaB exchangers, while electrogenic as NhaA exchangers, do not seem to possess obvious conserved motifs as were shown to be required for substrate transport in NhaA such as the two adjacent Asp residues present in TM V of NhaA ([Fig 1A](#pone.0182293.g001){ref-type="fig"}). However, a definite answer to the question how NhaA and NhaB transporters compare in terms of functional groups requires a high resolution structure which is presently not available. Notwithstanding the structural dissimilarities between NhaA and NhaB type exchangers, we could establish that KpNhaB follows the same general mechanism as shown for the CPA exchangers. In particular, competition between H^+^ and Na^+^ obviously is a key element of the Na^+^/H^+^ exchange mechanism in these transporters and may be a general concept for all Na^+^/H^+^ exchangers. The major difference between the behavior of KpNhaB and the NhaA exchangers was the observed cooperativity of Na^+^ binding in KpNhaB ([Fig 5D](#pone.0182293.g005){ref-type="fig"}), which was not found for the NhaA exchangers. This is a clear indication that more than one Na^+^ ion is involved in the transport mechanism and indeed this agrees with the different stoichiometries of Na^+^/H^+^ exchange: while NhaA transports 1 Na^+^ ion for 2 H^+^ \[[@pone.0182293.ref020]\], NhaB exchanges 2 Na^+^ ions for 3 H^+^ \[[@pone.0182293.ref021]\]. Kinetic analysis of the investigated exchangers {#sec013} ----------------------------------------------- The pH and Na^+^ dependence of all three investigated Na^+^/H^+^ exchangers from *Klebsiella pneumoniae* could be described by a simple kinetic model requiring only 3 (for NhaA type transporters) and 5 (for the NhaB type transporter) kinetic parameters ([Fig 6](#pone.0182293.g006){ref-type="fig"}). The fits of the model to the experimental data obtained for the three *K*. *pneumoniae* Na^+^/H^+^ exchangers are shown in [Fig 7](#pone.0182293.g007){ref-type="fig"}. The good quality of the fits and the low number of kinetic parameters represent strong support for the competition-based mechanism in all three transporters. The obtained pK and $K_{D}^{Na}$ values for the *K*. *pneumoniae* Na^+^/H^+^ exchangers ([Table 2](#pone.0182293.t002){ref-type="table"}) are in line with previously determined kinetic parameters for Na^+^/H^+^ antiporters of other organisms. Thus, the Na^+^ affinity is virtually the same as that determined for the NhaA exchangers of *E*. *coli*, *H*. *pylori* and *S*. *typhimurium* \[[@pone.0182293.ref052]\]. The lowest pK value was determined for KpNhaB (8.0) and is the same as for *H*. *pylori* NhaA \[[@pone.0182293.ref052]\], while the highest was determined for KpNhaA1 (9.2) and is the same as for *S*. *typhimurium* NhaA \[[@pone.0182293.ref052]\]. The role of *K*. *pneumoniae* Na^+^/H^+^ exchangers in bacterial survival {#sec014} ------------------------------------------------------------------------- The pH-dependent activity profiles of the three investigated exchangers obtained under symmetrical pH conditions are compared in [Fig 8A](#pone.0182293.g008){ref-type="fig"}, and can provide, along with the determined kinetic parameters, the basis of understanding the role of these exchangers in *K*. *pneumoniae*. However, it has to be kept in mind that a full comparison would require knowledge of the expression rate of the three exchangers, which we did not have access to. {ref-type="fig"}. B, Activity profiles of *K*. *pneumoniae* Na^+^/H^+^ exchangers calculated using the determined kinetic parameters in [Table 2](#pone.0182293.t002){ref-type="table"} under physiological stress conditions and normalized to the transport rate determined at pH~out~ = 5.5. Parameters used are listed in the panel. Dashed lines in B denote pH 7 and 8.3, where the survival assays presented in [Table 1](#pone.0182293.t001){ref-type="table"} were performed. In A, *out* and *in* denote the exterior and interior of the proteoliposomes, respectively, while in B, *out* and *in* denote the periplasmic and cytoplasmic space, respectively.](pone.0182293.g008){#pone.0182293.g008} In *E*. *coli*, EcNhaB is the main Na^+^ export system at acidic and neutral pH \[[@pone.0182293.ref019]\], while EcNhaA is highly up-regulated under conditions of salt stress \[[@pone.0182293.ref053]\] and is more efficient at removing Na^+^ at alkaline pH compared to EcNhaB \[[@pone.0182293.ref019]\]. Based on the kinetic parameters determined for the *K*. *pneumoniae* exchangers we modeled ([Fig 8B](#pone.0182293.g008){ref-type="fig"}) the behavior of these transporters under physiological stress conditions that *K*. *pneumoniae* can encounter, such as the ones that we used in our survival assays presented in [Table 1](#pone.0182293.t001){ref-type="table"}. Unlike conditions used in the SSM experimental setup ([Fig 8A](#pone.0182293.g008){ref-type="fig"}), the physiological activity of the exchangers occurs under a negative-inside membrane potential and potentially at a different periplasmic pH and Na^+^ concentration compared to the cytoplasmic pH and Na^+^ concentration ([Fig 8B](#pone.0182293.g008){ref-type="fig"}). It can be observed in [Fig 8B](#pone.0182293.g008){ref-type="fig"} that all three transporters are capable of exporting Na^+^ against a high concentration gradient at acidic pH. At high values of the salinity, KpNhaB is hard pressed to export Na^+^ once the periplasmic pH reaches neutral; under those conditions the two NhaA exchangers can perform this role, with KpNhaA1 being capable of functioning better at higher periplasmic pH. Overall, the three exchangers have complementary activity profiles ([Fig 8B](#pone.0182293.g008){ref-type="fig"}). One more factor to consider is the behavior of the transporters when high amounts of Li^+^ are present. We have shown that KpNhaB has a comparatively lower affinity for Li^+^ ([Table 2](#pone.0182293.t002){ref-type="table"}), whereas both KpNhaA1 and KpNhaA2 have much higher Li^+^ affinities ([Table 2](#pone.0182293.t002){ref-type="table"}). Thus, it stands to reason that when the cell is faced with Li^+^ stress, the two NhaA exchangers are more adapted to ensure survival. The modeled activity profiles can also be excellently correlated with the experimental results obtained in our survival assays ([Table 1](#pone.0182293.t001){ref-type="table"}). Thus, at pH 7, in the presence of 600 mM Na^+^, all three exchangers are capable of ensuring survival of *E*. *col* KNabc, with the KpNhaA1-expressing cells surviving better. When pH is increased to 8.3, none of the expressed antiporters can restore survival of the strain in presence of 600 mM Na^+^, which fits very well the modeled activity in [Fig 8B](#pone.0182293.g008){ref-type="fig"}. An exception at pH 8.3 is represented by the capability of KpNhaA1 cells of surviving in 100 mM Li^+^, which can be explained by both the fact that this protein has the highest affinity for Li^+^ out of the exchangers investigated in this work and has also the most alkaline-shifted pK. A better understanding of the way in which *K*. *pneumoniae* adapts to various stress conditions is essential, considering the danger posed by *K*. *pneumoniae* infections and the fact that many strains are resistant to currently used antibiotics. In all, the existence of three Na^+^/H^+^ exchangers with complementary transport profiles in *K*. *pneumoniae* indicates that this bacterium is well prepared to survive salt stress at various pH values. Designing specific, high affinity inhibitors against these Na^+^/H^+^ exchange systems is a possible avenue to consider in the task of finding a treatment against carbanepem-resistant *K*. *pneumoniae* infections. Materials and methods {#sec015} ===================== Genetic constructs and bacterial strains {#sec016} ---------------------------------------- The genes encoding the *K*. *pneumoniae* exchangers KpNhaA1 (Strain MGH 78578, Uniprot accession number: A6T4F6), KpNhaA2 (Strain MGH 78578, Uniprot accession number: A6TJ58) and KpNhaB (Strain 342, Uniprot accession number: B5XQ77) were synthesized by Genscript (Piscataway, NJ, USA) in the pET-21d(+) vector (Merck Millipore, Billerica, MA, USA). The genes were optimized for *E*. *coli* expression and contained an additional C-terminal 6-His tag. These constructs were used for protein production in the BL21(DE3) *E*. *coli* strain. For expression in the Na^+^/H^+^ exchanger deficient strain KNabc \[[@pone.0182293.ref044]\], the genes were cloned into the pTrcHis2 TOPO expression vector (Life technologies, Darmstadt, Germany) using the NcoI and EcoRI restriction sites. HpNhaA cloned in the pTrcHis2 TOPO expression vector obtained as previously described \[[@pone.0182293.ref022]\] was used as a control. Survival assays {#sec017} --------------- Bacterial survival in presence of high concentrations of Na^+^ or Li^+^ was assessed essentially as previously described \[[@pone.0182293.ref023]\]. Briefly, *E*. *coli* KNabc cells, deficient in the expression of the Na^+^/H^+^ antiporters NhaA, NhaB and ChaA \[[@pone.0182293.ref044]\] were transformed with the KpNhaA1, KpNhaA2, KpNhaB, HpNhaA constructs or with the empty pTrcHis2 TOPO expression vector. Cells were grown to OD~600~ of 0.6--0.7 in modified Luria-Bertani medium in which NaCl was replaced by KCl (LBK). The medium was buffered with 50 mM MOPS and contained 50 μg/ml kanamycin, 36 μg/ml chloramphenicol and 100 μg/ml ampicillin. 2 μl samples of serial 10-fold dilutions of the cultures were spotted onto LBK agar (1.5%) plates containing 0.6 M NaCl or 0.1 M LiCl at either pH 7 or 8.3 and incubated for 48 h at 37 °C. LBK agar plates with no addition of NaCl or LiCl were used as a control. Acridine orange dequenching in everted membrane vesicles {#sec018} -------------------------------------------------------- Everted vesicles from *E*. *coli* KNabc transformed with the recombinant plasmids for KpNhaA1, KpNhaA2 and KpNhaB were prepared as described previously \[[@pone.0182293.ref054]\]. Na^+^/H^+^ antiport activity was assessed based on the measurement of Na^+^-induced changes in the ΔpH as measured by acridine orange, a fluorescent probe of ΔpH. Everted vesicles were resuspended in buffer containing 10 mM Tris (titrated to the pH 7 using HCl), 250 mM sucrose and 140 mM choline chloride. Total protein concentration was measured using the Bradford assay \[[@pone.0182293.ref055]\]. Fluorescence was measured using a Hitachi F4500 Fluorimeter (Hitachi High-Technologies Corporation, Tokyo, Japan) at excitation and emission wavelengths of 495 nm and 530 nm, respectively. Dequenching assays were performed in buffer containing 10 mM MES (titrated to the indicated pH using Tris), 145 mM choline chloride, 5 mM MgCl~2~, 2 μM acridine orange. 100 μg (total protein) of everted vesicles were added to 1 ml external buffer. Acidification of the vesicles was induced using 2.5 mM Tris-D-lactate (at the corresponding pH). After reaching steady-state fluorescence, dequenching was induced by adding 10 or 50 mM NaCl. Finally, the pH gradient was dissipated by addition of 8 mM NH~4~Cl. Dequenching was calculated as $Dequenching\ \% = \frac{F_{deq} - F_{min}}{F_{fin} - F_{min}} \bullet 100$ where *F*~deq~ is the steady-state level of fluorescence achieved after dequenching, *F*~min~ is the steady-state level of fluorescence after Tris-D-lactate addition, and before Na^+^ addition, and *F*~fin~ is the steady-state level of fluorescence after dissipation of the pH gradient. Overexpression, purification and reconstitution {#sec019} ----------------------------------------------- C-terminally His-tagged proteins were produced in *E*. *coli* BL21(DE3) cells and purified using immobilized Ni^2+^ affinity chromatography, as previously described \[[@pone.0182293.ref056]\]. Reconstitution of purified protein into proteoliposomes was performed using *E*. *coli polar* lipids extract (Avanti Polar Lipids, Alabaster, AL, USA) at a calculated LPR of 10 or 50, as previously described \[[@pone.0182293.ref038]\]. As previously shown \[[@pone.0182293.ref057]\], an LPR of 10 corresponts to a protein density of \~ 1000 protein particles/μm^2^. SSM-based electrophysiology {#sec020} --------------------------- SSM measurements were performed as described previously \[[@pone.0182293.ref038]\]. Briefly, 30 μl of proteoliposomes at a lipid concentration of 3.3 mg/ml were adsorbed to an octadecanethiol / phospholipid hybrid bilayer on a gold surface (sensor). Unless otherwise stated, most measurements were performed using LPR 10 proteoliposomes. Proteoliposomes were allowed to adsorb to the sensor for 1 h. Electrogenic transport was initiated by a rapid change of substrate ion concentration in a single solution exchange protocol: non-activating solution (0.5 s)--activating solution (0.5 s)--non-activating solution (0.5 s). Currents were amplified using a current amplifier set to a gain of 10^8^−10^9^ V/A and a rise time of 10 ms. Non-activating solutions contained 25 mM MES, 25 mM Hepes, 25 mM Tris, 5 mM MgCl~2~ and 300 mM KCl. Activating solutions contained 25 mM MES, 25 mM Hepes, 25 mM Tris, 5 mM MgCl~2~, x mM NaCl (or LiCl) and (300---x) mM KCl. All solutions were titrated to the desired pH using HCl or KOH. In most cases, the amplitude of the recorded transient currents following Na^+^ concentration jumps was used in order to quantify steady-state Na^+^/H^+^ exchange activity. An exception was made for currents recorded for KpNhaB at pH 9.0 and pH 9.5, which showed a substantial pre steady-state component in addition to the steady-state component. In these cases, the currents were reconstructed according to the procedure described by Mager et al. \[[@pone.0182293.ref038]\] that allowed the measurement of the stationary component of the reconstructed current. Kinetic analysis {#sec021} ---------------- The steady-state solution for the kinetic model was calculated as described previously by Mager et al. \[[@pone.0182293.ref038]\]. This calculation yields turnover or activity of the transporter at a given Na^+^ concentration and pH. For KpNhaB, the model was modified by the addition of the Hill coefficients *m* and *n* that denote cooperative H^+^ and Na^+^ binding, respectively. Simultaneous fits of the model to the experimentally determined pH and Na^+^ dependences yielded values for the kinetic parameters pK, $K_{D}^{Na}$, k~2~/k~1~ and, in the case of KpNhaB, also for *m* and *n*. For the NhaA exchangers, where no cooperativity was apparent, *m* and *n* were fixed to 1. Supporting information {#sec022} ====================== ###### Measured *vs* reconstructed currents for KpNhaB. Current traces recorded for KpNhaB at pH 9.0 and 9.5 were reconstructed in order to determine the stationary component of the reconstructed current as described by Tadini-Buoninsegni and Fendler \[[@pone.0182293.ref040]\]. Presented is a trace recorded for a 100 mM Na^+^ concentration jump at pH 9.5, where there is a significant pre steady-state component. (TIF) ###### Click here for additional data file. ###### Transient currents recorded for different values of the lipid to protein ratio (LPR). Current traces were recorded following 100 mM Na^+^ concentration jumps at pH 8.5 for KpNhaA1 (A), KpNhaA2 (B) and KpNhaB (C). For a better comparison of decay time constants at different LPR values, currents were normalized to their maximum amplitude. (TIF) ###### Click here for additional data file. We are grateful to Etana Padan for critically reading the manuscript and helpful discussions. [^1]: **Competing Interests:**The authors have declared that no competing interests exist.

{#sp1 .24} {#sp2 .25} {#sp3 .26} {#sp4 .27} {#sp5 .28} {#sp6 .29}

{#sp1 .53} {#sp2 .54}

1. Introduction: {#sec1-microorganisms-06-00001} ================ *Propionibacterium acnes* is a Gram-positive anaerobic bacterium and member of the 'cutaneous' group of human propionibacteria on the skin, although it can also be isolated from the oral cavity as well as the genitourinary and gastrointestinal tracts \[[@B1-microorganisms-06-00001]\]. The organism is an opportunistic pathogen associated most notably with the common skin condition acne vulgaris after which it is named \[[@B2-microorganisms-06-00001]\]. Acne is an inflammatory disease of the pilosebaceous follicle affecting the face, chest and back, primarily during adrenarche ([Figure 1](#microorganisms-06-00001-f001){ref-type="fig"}). Over the last 20 years, however, there has been an increasing recognition that we may have significantly underestimated the role of this bacterium in other human infections and clinical conditions. From the scientific literature, we now have a large body of evidence associating this bacterium with infections of indwelling medical devices, especially shoulder prostheses, as well as sarcoidosis, prostate cancer, back pain and fatal granuloma after soft tissue trauma \[[@B3-microorganisms-06-00001],[@B4-microorganisms-06-00001],[@B5-microorganisms-06-00001],[@B6-microorganisms-06-00001],[@B7-microorganisms-06-00001]\]; further work is, however, required to prove whether the bacterium actually plays a role in the development of some of these conditions, especially lumbar disc herniation and prostate disease. Within the last decade, we have also seen significant advances in our understanding of the intraspecies phylogeny of *P. acnes*. Distinct phylogroups have been discovered, and specific strains or sequence types (ST) associated with human health or disease revealed \[[@B8-microorganisms-06-00001],[@B9-microorganisms-06-00001],[@B10-microorganisms-06-00001],[@B11-microorganisms-06-00001],[@B12-microorganisms-06-00001]\]. In this review we will initially focus on how DNA sequence typing schemes based on *recA* and *tly* loci were used to reveal the basic underlying phylogenetic structure of *P. acnes*. Specific issues or caveats with *recA* and *tly* sequence typing will then be highlighted, alongside a detailed description of newer, high resolution, molecular typing methods that have been developed to characterize the bacterium and better detail its phylogeny. 2. Introduction of *recA* and *tly* Gene Sequencing for Phylogenetic Analysis of *P. acnes* {#sec2-microorganisms-06-00001} =========================================================================================== Gene sequence analysis of *P. acnes* based on the non-ribosomal housekeeping (HK) gene *recA* (1047 bp; PPA1012) and the putative haemolysin/FtsJ-like methyltransferase gene *tly* (777 bp; PPA1396) was first described in 2005 as part of a larger study investigating the phylogenetic relationship between the two serotypes of *P. acnes*, known as types I and II \[[@B8-microorganisms-06-00001]\]. Analysis of these genetic loci revealed that types I and II represented highly distinct phylogenetic groupings or divisions within the species. It also revealed further phylogenetic sub-divisions within the type I clade, designated types IB and IC, which displayed atypical reactions with monoclonal antibodies (mAb) originally developed for the identification of type I (QUBPa1) and type II (QUBPa2) strains \[[@B8-microorganisms-06-00001],[@B11-microorganisms-06-00001],[@B13-microorganisms-06-00001]\]. Both antibodies, which target antigenically variable cell surface adhesins that bind dermatan sulphate (DsA1; DsA2) (QUBPa1) and a glycolipid-containing antigen (QUBPa2), show no significant reaction with the type IB lineage, but do react with type IC strains \[[@B8-microorganisms-06-00001],[@B11-microorganisms-06-00001],[@B14-microorganisms-06-00001]\]; other type I strains, designated type IA, react with QUBPa1 but not QUBPa2. Recently, the DsA1 protein has also been shown to bind human fibrinogen \[[@B15-microorganisms-06-00001]\]. Interestingly, just five months after the description of the *recA* and *tly* sequencing results, Cohen et al. \[[@B5-microorganisms-06-00001]\] described three distinct pulsed field gel electrophoresis (PFGE) groups of *P. acnes* isolated from cancerous prostate tissue. We now know these pulsogroups actually correspond to the *recA* and *tly* types IA, IB and II \[[@B11-microorganisms-06-00001],[@B12-microorganisms-06-00001]\]; in the same study, sequencing of the transcarboxylase 12S gene could differentiate pulsogroups I (type IA) and II (type IB) from group III (type II), but could not further subtype the type I clade. In 2008, application of *recA* gene sequencing was also instrumental in the identification of a new phylogenetic division within *P. acnes*, which was designated *recA* type III \[[@B9-microorganisms-06-00001]\]. Strains of the type III clade, which were originally isolated from spine intervertebral disc material and now also human skin, have the capacity to form a filamentous-like cell morphology not observed with types I and II, and are also non-reactive with QUBPa1 and QUBPa2. 3. Choice of *recA* and *tly* Loci for Phylogenetic Analysis {#sec3-microorganisms-06-00001} ============================================================ The *recA* gene was chosen by McDowell et al. \[[@B8-microorganisms-06-00001]\] since it is a valuable phylogenetic marker in bacterial systematics, and classifications based on *recA* have proved to be robust and consistent with those obtained by using rRNA genes \[[@B16-microorganisms-06-00001]\]. Furthermore, protein-encoding HK genes can often provide greater phylogenetic resolution of very closely related organisms compared to the 16S rRNA gene due to a higher molecular clock speed, thereby generating a larger pool of informative sites for phylogenetic reconstructions \[[@B17-microorganisms-06-00001]\]. Consistent with this, in the case of types I and II only one type-specific polymorphism at position 827 was identified within the 16S gene of the strains selected for analysis, versus 10 type-specific differences in *recA* \[[@B8-microorganisms-06-00001]\]. The *recA* gene has also been successfully used for inter- and intra-species differentiation of organisms of the *Burkholderia cepacia* complex, *Agrobacterium* spp. and for typing of *Vibrio cholerae* isolates \[[@B18-microorganisms-06-00001],[@B19-microorganisms-06-00001],[@B20-microorganisms-06-00001],[@B21-microorganisms-06-00001]\]. In the case of the *tly* gene, it was selected for analysis since it had a putative function as a cell surface/extracellular haemolysin/cytotoxin. In contrast to HK gene loci, such genes are often under strong positive selection for non-synonymous mutations and recombinational events by the host immune system leading to rapid diversification and increased discriminatory power for finer scale typing \[[@B22-microorganisms-06-00001]\]; such 'hyper-variable' genes are often used to investigate very recent patterns of descent (short term epidemiology) and used as a complement to Multilocus Sequence Typing (MLST) analysis, as well as providing valuable information on the evolution of virulence loci. In the case of the *tly* gene, however, no clear evidence for diversifying selection exists upon analysis of multiple allele sequences, and this locus appears to have co-evolved with HK genes potentially indicating an important role in commensal existence for the bacterium \[[@B11-microorganisms-06-00001],[@B12-microorganisms-06-00001]\]. The observation that the Tly protein family now needs to be redefined as an RNA-binding FtsJ-like methyltransferase involved in ribosomal biogenesis may be important in this context, indicating a dual function as a HK gene alongside a role as a haemolysin \[[@B12-microorganisms-06-00001],[@B23-microorganisms-06-00001]\]. The *tly* gene is also present in other human propionibacteria, including *P. avidum*, *P. humerusii* and the recently described species *P. namnetense* \[[@B24-microorganisms-06-00001]\]. 4. Adoption of *recA* and *tly* Gene Sequencing as a Typing Tool for *P. acnes* {#sec4-microorganisms-06-00001} =============================================================================== Prior to the description of *recA* and *tly* sequencing analysis of *P. acnes*, typing methods for the bacterium focused on fermentation, phage and serological reactions, DNA macrorestriction profiling by PFGE and Random Amplification of Polymorphic DNA analysis \[[@B25-microorganisms-06-00001],[@B26-microorganisms-06-00001],[@B27-microorganisms-06-00001],[@B28-microorganisms-06-00001],[@B29-microorganisms-06-00001]\]. Soon after publication, researchers started to adopt *recA* and *tly*-based sequencing as a moderate resolution typing method, with the genes being utilized individually for single locus sequence typing (SLST), or concatenated for multilocus analysis. These genes have been used as typing tools to investigate whether the different phylogroups of *P. acnes* are associated with specific human infections or clinical conditions, including prosthetic joint and dental infections, sarcoidosis, prostate cancer and lumbar disc herniations highlighted earlier \[[@B30-microorganisms-06-00001],[@B31-microorganisms-06-00001],[@B32-microorganisms-06-00001],[@B33-microorganisms-06-00001],[@B34-microorganisms-06-00001],[@B35-microorganisms-06-00001],[@B36-microorganisms-06-00001]\]. These sequence-based typing methods have also been used to genetically characterize *P. acnes* strains isolated from healthy human stomach mucosa \[[@B37-microorganisms-06-00001]\]. Such studies started to provide initial evidence that differences may indeed exist in the association of specific phylogroups with disease, including strains of type IB and II with prosthetic joint infections and type I and II with prostate cancer. In addition to disease associations, characterization of *P. acnes* strains based on their *recA* and *tly* phylogenies has provided a framework onto which intraspecies genomic and proteomic characteristics can be mapped \[[@B38-microorganisms-06-00001],[@B39-microorganisms-06-00001]\], as well as a broad range of putative virulence, biochemical and immunological properties \[[@B8-microorganisms-06-00001],[@B9-microorganisms-06-00001],[@B13-microorganisms-06-00001],[@B31-microorganisms-06-00001],[@B32-microorganisms-06-00001],[@B40-microorganisms-06-00001]\]. A good example is the production of DsA1 and DsA2 adhesins by strains of type IA and IC, but not IB, II or III, and the differential production of CAMP-factors, in particular the abundant production of CAMP factor 1 by type IB and II strains, but not type IA \[[@B13-microorganisms-06-00001],[@B39-microorganisms-06-00001]\]. Collectively, such studies started to reveal differences in pathogenic potential between phylogroups and their association with human disease. 5. *recA* and *tly* Gene Sequencing to Characterise *P. acnes* Type Zappe Strains {#sec5-microorganisms-06-00001} ================================================================================= *recA* and *tly* gene sequence analysis has even been used to characterize strains of the grapevine-associated *P. acnes* lineage known as *P. acnes* type Zappe \[[@B41-microorganisms-06-00001]\]. This new type of *P. acnes* appears to have adapted to an endophytic lifestyle, and is believed to have arisen from a human-to-plant interkingdom bacterial transfer approximately 7500 years ago during the domestication of grapevines \[[@B41-microorganisms-06-00001]\]. Interestingly, although *P. acnes* type Zappe forms a distinct phylogenetic cluster from *P. acnes* upon 16S rDNA analysis, it groups with *P. acnes* type I strains based on *recA* and *tly* sequences. These discordant gene trees are likely to result from incomplete lineage sorting due to recent diversification. Furthermore, unlike *P. acnes*, the *recA* gene in *P. acnes* type Zappe appears to have lost its function during this unusual host switching event, possibly reflecting endosymbiosis, leading to the accumulation of a large number of non-synonymous mutations; in comparison, the *tly* and 16S rRNA genes of *P. acnes* type Zappe are much more conserved. These gene sequence characteristics, along with the detection of *P. acnes* type Zappe within plant tissues (bark, pith and fibres of Xylem vessels), provides evidence to support the endophyte nature of this bacterium and tight symbiosis with the plant. 6. Taxonomic Legacy {#sec6-microorganisms-06-00001} =================== When the *recA* and *tly* typing methods were first described, our understanding of the underlying population genetic structure of *P. acnes* was poor. The identification of distinct genetic divisions and sub-divisions therefore provided a platform for more in depth study of the phylogenetic and taxonomic heterogeneity of *P. acnes*, and how this related to human health and disease. This has ultimately led to the recent proposal of the type I, II and III phylogroups as distinct subspecies known as *P. acnes* subsp. *acnes*, *P. acnes* subsp. *defendens* and *P. acnes* subsp. *elongatum*, respectively based on phylogenetic, genomic, phenotypic differences, as well as associations with different types of infections and clinical conditions including acne, progressive macular hypomelanosis (PMH), and prosthetic joint and soft tissue infections \[[@B42-microorganisms-06-00001],[@B43-microorganisms-06-00001]\]. The cutaneous propionibacteria have now also been recently proposed as a novel genera, known as *Cutibacterium*, with *P. acnes* renamed as *Cutibacterium acnes* \[[@B44-microorganisms-06-00001]\]. 7. *recA* and *tly* Alleles Provide Hints of Conjugal Transfer of Large DNA Segments {#sec7-microorganisms-06-00001} ==================================================================================== The subsequent development of more recent molecular typing methods for *P. acnes* based on multilocus sequence typing (MLST) schemes (discussed later) combined with the growing number of available whole genome sequences, has provided newer, high resolution methods for detailed scrutiny of the population genetic structure of the bacterium. This has revealed much greater levels of phylogenetic structure within the type I phylogroup, enabling additional differentiation of strains into types IA~1~ and IA~2~, alongside the type IB and IC groups, and identification of clonal complexes (CC) and singleton strains \[[@B10-microorganisms-06-00001],[@B11-microorganisms-06-00001],[@B12-microorganisms-06-00001],[@B43-microorganisms-06-00001],[@B45-microorganisms-06-00001],[@B46-microorganisms-06-00001]\]. These high-resolution MLST and whole genome sequencing (WGS) methods have, however, also revealed caveats with *recA* and *tly* sequencing, since the combination of *recA* and *tly* allele sequences originally used to identify the type IB phylogroup have now also been found in a subset of clearly defined type IA~1~ strains from CC4 (MLST~8~ nomenclature), all strains from the IA~2~ lineage, and the aberrant type IA~1~ singleton strain SK187 which displays a mosaic of genes found in other type I lineages, including type IB \[[@B10-microorganisms-06-00001],[@B14-microorganisms-06-00001],[@B38-microorganisms-06-00001],[@B45-microorganisms-06-00001]\]. This is illustrated in [Figure 2](#microorganisms-06-00001-f002){ref-type="fig"}A, which demonstrates phylogenetic clustering of strains crossing these different genetic distances based on concatenated *recA* and *tly* gene sequences (1824 bp) (the same clustering of strains is seen with single *recA* and *tly* trees; [Figure S1](#app1-microorganisms-06-00001){ref-type="app"}), yet clear, non-overlapping clustering of these phylogroups when high resolution MLST~8~ analysis is applied ([Figure 2](#microorganisms-06-00001-f002){ref-type="fig"}B). These results help to explain why a sub-set of strains presumptively identified as type IB based on *recA* and *tly* analysis displayed reaction with the type IA mAb QUBPa1 \[[@B11-microorganisms-06-00001],[@B14-microorganisms-06-00001]\]. *P. acnes* has a clonal population structure and is in linkage disequilibrium \[[@B11-microorganisms-06-00001],[@B12-microorganisms-06-00001]\]. The organism has a high degree of sequence conservation, although recombination events are still evident in the history of the bacterium. Due to this, *P. acnes* is a candidate organism for the development of phylogenetically accurate SLST schemes. Interestingly, phylogenetic network analyses reveals only the *tly* sequences conforming to a very clear tree-like structure, although the Phi test for recombination is not statistically significant for either loci ([Figure 3](#microorganisms-06-00001-f003){ref-type="fig"}). Furthermore, phylogenetic trees based on the analysis of the individual genes are congruent ([Figure S1](#app1-microorganisms-06-00001){ref-type="app"}). The discovery of linked alleles across different phylogenetic lineages does, however, suggest conjugal transfer and homologous recombination of very large DNA segments; these genes comprise a genomic fragment that stretches to at least 420 Kbp (KPA171202 coordinates 1095280-1515117). This scenario would help to explain why *recA* and *tly* sequences generate congruent phylogenies. Further support also comes from the identification of *camp5* (KPA171202 coordinates 1305348-1306193) and *gms* (KPA171202 coordinates 1503535-1504578) alleles that are also shared between CC4 and types IA~2~ and IB, and located within the same segment \[[@B10-microorganisms-06-00001]\]. Collectively, this would tentatively suggest that multiple conjugation events of unusually large chromosomal replacements have shaped the genome dynamics of the type I clade; such large chromosomal replacements have also been observed with *Staphylococcus aureus* and *Streptococcus agalactiae* \[[@B47-microorganisms-06-00001],[@B48-microorganisms-06-00001]\]. While such events are assumed to be rare due to the large size of the fragments, they may have had a dramatic effect on the evolution of the bacterium. More detailed sequence-based analyses will, however, be required to confirm this possibility. Alongside this, *P. acnes* has a flexible gene pool with the presence and absence of different island-like genomic regions with aberrant G+C content and flanking insertion sequences that encode putative virulence factors and traits potentially associated with fitness and niche adaptation \[[@B38-microorganisms-06-00001],[@B49-microorganisms-06-00001],[@B50-microorganisms-06-00001]\]. Rates of recombination also appear to differ throughout the population, and the association of alleles appears less significant when distinct phylogroup populations are considered \[[@B11-microorganisms-06-00001],[@B12-microorganisms-06-00001]\]. Reduced rates of recombination between the different subsp. of *P. acnes* may indicate ecological differences within the body since members of the same habitat are more likely to undergo recombination events (sympatric speciation) \[[@B51-microorganisms-06-00001]\]. Against this background, it is interesting that only type II strains have an active CRISPR/Cas system, making type I and III strains potentially more susceptible to horizontal gene transfer (HGT) of fitness and virulence traits \[[@B46-microorganisms-06-00001],[@B52-microorganisms-06-00001]\]. 8. Newer Molecular Typing Methodologies for *Propionibacterium acnes* {#sec8-microorganisms-06-00001} ===================================================================== As the *recA* and *tly* loci cannot reliably identify type IB strains from those related to the type IA~1~ sub-clade corresponding to CC4, or strains of type IA~2~, their application as a typing tool has limitations; the methods are still valid, however, for the identification of almost all other type IA~1~ strains, as well as those from the type IC, II and III phylogroups. Since the *recA* and *tly* sequencing approaches were originally reported over 10 years ago, a wide range of newer DNA-based typing methods have been described with distinct advantages and disadvantages to one another; most of these are still superior to the original *recA* and *tly* typing methods in terms of its moderate resolving power. These key methods include: 8.1. MLST Schemes {#sec8dot1-microorganisms-06-00001} ----------------- MLST is based on indexing genetic variation within multiple HK genes, normally seven \[[@B55-microorganisms-06-00001]\]. Unique alleles for each loci are given an arbitrary number, and the different combinations of these alleles, known as the allelic profile, assigned a ST number. The allelic profile therefore does not take into consideration the number of nucleotide differences between alleles, and this lack of weighting in regard to sequence divergence helps to correct for HGT events which can distort phylogenetic signals \[[@B55-microorganisms-06-00001]\]. In addition to concatenated DNA sequence analysis which provides minimal information of ancestry or patterns of descent, MLST enables isolates to also be grouped or clustered based on sharing a defined number of alleles (usually ≥6/7) with at least one other member of the group, thus generating non-overlapping CCs and predicting founding genotypes. This type of analysis, which is performed using an algorithm such as eBURST, provides the opportunity to infer appropriate patterns of evolutionary descent between isolates over short timescales \[[@B56-microorganisms-06-00001]\]. Two independent MLST schemes and associated, publically available, databases for *P. acnes* have now been developed based on completely different sets of protein-encoding gene loci \[[@B10-microorganisms-06-00001],[@B11-microorganisms-06-00001]\]. These schemes provide high resolution typing of the bacterium and generate phylogenies which are essentially congruent with those based on WGS analysis. One scheme highlighted earlier in the review, and based on the analysis of eight loci (MLST~8~; 4253 bp), was originally developed at the University of Warwick and then modified by researchers at Queen's University, Belfast to include the *camp2* and *tly* genes \[[@B11-microorganisms-06-00001]\] ([Figure 2](#microorganisms-06-00001-f002){ref-type="fig"}B); the database for this scheme can be found at *Propionibacterium acnes* MLST Databases \[[@B54-microorganisms-06-00001]\]. A second scheme, which was based on nine loci and corresponding amplification and sequencing primers developed at the University of Bath, was fully adopted by researchers at Orebro University (4536 bp) \[[@B57-microorganisms-06-00001]\]. Researchers at Aarhus University also utilized this method but replaced the *cob* gene with *recA* (known as MLST~9~; 4233 bp) and also developed a database for the scheme which can be found at Multi Locus Sequence Typing \[[@B53-microorganisms-06-00001]\]. Both MLST~8~ and MLST~9~ schemes are generally concordant in how they cluster strains into different CCs, but some differences do exist in the resolution of particular strains within these CCs. These CC equivalencies and specific strain differences have been detailed elsewhere \[[@B11-microorganisms-06-00001]\]. At the time of writing, the MLST~8~ database comprises 127 distinct STs with nine CCs superimposed on a background of 19 singletons in sequence space ([Figure 4](#microorganisms-06-00001-f004){ref-type="fig"}). The MLST~9~ database comprises 125 STs. Being a DNA-sequencing based method, MLST offers portability over gel-based typing approaches, and is an extremely valuable and stable system for global epidemiology and genetic population studies. Indeed, using these methods we and others have shown that *P. acnes* has an epidemic population structure, with highly successful clonal lineages that are globally disseminated \[e.g., ST1 and ST3 (type IA~1~), ST2 (type IA~2~) and ST5 (type IB) based on MLST~8~ analysis\] within the human population \[[@B11-microorganisms-06-00001],[@B12-microorganisms-06-00001]\] ([Figure 5](#microorganisms-06-00001-f005){ref-type="fig"}); some of these epidemic lineages appear to be permanent members of the skin microbiota and are associated with acne \[[@B11-microorganisms-06-00001],[@B12-microorganisms-06-00001]\]. Despite all its advantages, the key downsides of MLST are its labor and time consuming nature as well as expense, especially when analyzing large numbers of isolates. In an attempt to circumvent these problems, and to streamline the MLST workflow, it has been shown that only four of the eight MLST~8~ gene loci (*aroE*, *guaA*, *camp2*, *tly*) need to be sequenced to correctly predict phylogroup, CC and in most cases ST from the full MLST profiles available in the database \[[@B12-microorganisms-06-00001]\]. This is only possible because of the clonal nature of *P. acnes*, and the relatively limited number of genotypes that are found. 8.2. Multiplex PCR {#sec8dot2-microorganisms-06-00001} ------------------ A set of three separate PCR assays for phylogroup typing of *P. acnes* was originally described in 2006 by Shannon et al. \[[@B58-microorganisms-06-00001]\]; this method was only capable of resolving type IA, IB and II strains, providing no data on type IA~1~, IA~2~, IC or type III status. More recently, a multiplex touchdown PCR typing method for *P. acnes* isolates that provides unambiguous identification of all the main phylogroups (type IA~1~, IA~2~, IB, IC, II and III) in a single reaction has been described \[[@B59-microorganisms-06-00001]\]. The method utilizes six primer sets that target the 16S rRNA gene to confirm species identify (all isolates), ATPase (types IA~1~, IA~2~, and IC), *sodA* (types IA~2~ and IB), *atpD* (type II), and *recA* (type III) HK genes, as well as a Fic family toxin gene (type IC); the HK loci on which the multiplex assay was developed are from the MLST~8~ scheme where multiple alleles of each gene are available for analysis and identification of type-specific single nucleotide polymorphisms (SNPs) for primer design. Phylogroup identification is based on the visual pattern of reaction with the different primer sets. The key advantage of this 6-plex PCR assay is that it provides a rapid, high-throughput, and technically undemanding typing method for epidemiological and phylogenetic investigations. The multiplex assay also provides a simple way to identify potentially novel taxa due to atypical or non-typeable PCR patterns that could arise, for example, when new alleles or allele combinations are encountered in new STs. A limitation of the multiplex method is that it only resolves isolates to the phylogroup level; if very high resolution typing is required then other methods, such as MLST or WGS, should be carried out. These latter methods, however, only provide significantly more information for type IA~1~ and type II phylogroups due to their greater genetic heterogeneity and deeper level of phylogenetic structure relative to types IA~2~, IB, IC, and III. In the latter case, the phylogroups represent relatively tight phylogenetic clusters (only one CC for each phylogroup) with a limited number of STs, some of which are highly dominant in the population and widely disseminated ([Figure 4](#microorganisms-06-00001-f004){ref-type="fig"} and [Figure 5](#microorganisms-06-00001-f005){ref-type="fig"}). The multiplex assay has now been described for the typing of *P. acnes* isolates from patients with acne \[[@B60-microorganisms-06-00001],[@B61-microorganisms-06-00001],[@B62-microorganisms-06-00001]\], PMH \[[@B63-microorganisms-06-00001]\], wound exudates and abscesses \[[@B62-microorganisms-06-00001]\], microdiscectomy tissue \[[@B64-microorganisms-06-00001]\], central nervous system and prosthetic joint infections \[[@B65-microorganisms-06-00001]\], and other clinical conditions \[[@B60-microorganisms-06-00001]\]. It has also been used for high throughput typing of *P. acnes* isolates from the pre-operative skin and surgical wound of patients undergoing spinal operations; this study was a large randomized controlled trial investigating the effects of povidone-iodine-alcohol (PVI) and chlorhexidine gluconate-alcohol on surgical wound contamination compared to PVI alone \[[@B66-microorganisms-06-00001]\]. The validation of the multiplex method by a number of independent groups has led to the conclusion that it is a robust and valid method for *P. acnes* typing \[[@B62-microorganisms-06-00001]\]. 8.3. Ribotyping {#sec8dot3-microorganisms-06-00001} --------------- A ribotyping scheme for *P. acnes*, based on analysis of the 16S rRNA gene (\~1450 bp; positions 29-to-1483 bp), has been described by Fitz Gibbon et al. \[[@B46-microorganisms-06-00001]\]; novel ribotypes (RT) in this SLST scheme were assigned based on at least one unique SNP. Upon Sanger sequencing of amplicons obtained from genomic DNA pooled from pilosebaceous units ('pores') on the nose of acne and control subjects, more than 11K RTs were identified, although only a very small number appear abundant in the skin with the minor RTs representing singletons. Compared to RT1, the most abundant of the RTs, all other defined RTs have ≥99% sequence identify which is in keeping with the highly conserved nature of the 16S rRNA locus at the intraspecies level. Despite this, the limited number of SNPs available can differentiate the main type I, II and III phylogroups from one another, as well as type IA~1~ from IA~2~ strains which have unique RTs in keeping with their distinct phylogenies ([Table 1](#microorganisms-06-00001-t001){ref-type="table"}). Within the type IA~1~ clade, strains corresponding to CC4 by MLST~8~ analysis (equivalent to CC31 by MLST~9~) can also be resolved, but a limitation of the method is that strains from CC1 and CC3 by MLST~8~ analysis (equivalent to CC18 and CC3, respectively by MLST~9~), as well as type IB and IC strains, cannot be separated as they share RT1 and RT5 ([Table 1](#microorganisms-06-00001-t001){ref-type="table"}). As the ribotyping approach has specific strain resolution limitations, this is the main disadvantage to its use. As it is only based on one locus, the data is also not amenable to allelic profile clustering. It does, however, significantly reduce Sanger sequencing costs compared to MLST analysis of isolates while retaining the advantage of DNA sequence portability and stability between laboratories. Currently, there is no public database for the *P. acnes* ribotyping scheme. 8.4. High Resolution SLST (HR-SLST) {#sec8dot4-microorganisms-06-00001} ----------------------------------- The development of a high resolution SLST (HR-SLST) scheme for *P. acnes* is clearly desirable since it would reduce the expense and workflow involved in MLST analysis, alongside its ability to directly analyze complex biological samples for *P. acnes* population structure. While *P. acnes* is an appropriate choice for the development of such a method, the low level of genetic variability found amongst the HK genes of this clonal organism does make such a task difficult. To address this issue, Scholz et al. \[[@B67-microorganisms-06-00001]\] utilized novel python scripts to 'mine' the available whole genome sequences of all known *P. acnes* phylogroups, and identify sufficiently variable target loci on which a novel SLST scheme could be developed. This computational analysis pinpointed a 483--497 bp sequence which differentiated strains to a level of resolution approaching that obtained by MLST ([Figure 6](#microorganisms-06-00001-f006){ref-type="fig"}, [Table S1](#app1-microorganisms-06-00001){ref-type="app"}); although the pattern of phylogroup clustering within the type I clade varies somewhat from that obtained by MLST and WGS analysis. This sequence is located immediately upstream of the CAMP factor 1 gene (PPA1340), and partially overlaps with a number of hypothetical genes \[[@B67-microorganisms-06-00001]\]. The success of this locus suggests that recombination events are extremely limited in its history, and this is confirmed by phylogenetic network analysis which shows a tree-like structure, and a non-significant Phi test for recombination ([Figure 7](#microorganisms-06-00001-f007){ref-type="fig"}). The SLST method uses a letter and number-based nomenclature for ST assignment and a database for the scheme can be found at SLST for *Cutibacterium acnes* (formerly *Propionibacterium acnes*) \[[@B68-microorganisms-06-00001]\]. In addition to isolate analysis, the SLST method has also been used to examine the *P. acnes* community structure within skin swab samples taken from various body sites of a healthy individual \[[@B67-microorganisms-06-00001]\] and patients with PMH \[[@B69-microorganisms-06-00001]\] via Roche 454 pyrosequencing, which provides reads sufficiently long to cover the target sequence; this metagenomic approach enables mixed strain populations of the bacterium to be analyzed. The method can, however, be adapted to run on other next generation sequencing (NGS) platforms, such as Illumina using pair-end reads (2 × 300 bp). Similar to ribotyping, the SLST approach will reduce Sanger sequencing costs compared to MLST, and is a portable and stable system, although as it is only based on one locus the data is also not amenable to allelic profile clustering. 8.5. Multiple Locus Variable Number of Tandem Repeat (VNTR) Analysis (MLVA) {#sec8dot5-microorganisms-06-00001} --------------------------------------------------------------------------- MLVA is a typing method for bacteria that is based on the variable number of tandem repeats (VNTR) present in multiple loci. These hypervariable loci are short sequences of repetitive DNA that arise due to slipped strand mispairing during DNA replication, and frequently vary between strains from the same species. An MLVA scheme for *P. acnes* based on the analysis of 13 VNTRs (MLVA~13~) has been described by Hauck et al. \[[@B70-microorganisms-06-00001]\]. These included 10 large VNTRs with repeat units ranging from 14--31 bp, and three small VNTRs with 6 bp size repeats. Like MLST, MLVA generates an allelic profile which in this case corresponds to the number of repeats at each locus, and a specific MLVA type number. The selected VNTR markers in *P. acnes* are present in genes that include putative HtaA domain and adhesin proteins, fibrinogen- and penicillin-binding proteins and hypotheticals, as well as an intergenic region \[[@B70-microorganisms-06-00001]\]. With the MLVA~13~ scheme, clustering of *P. acnes* isolates was found to be in good agreement with phylogenies based on WGS analysis, although a small number of discrepancies were observed. An advantage of this method is that it is relatively cheap, fast and easy to perform but portability can be a significant issue when amplicons are sized as banding patterns by conventional electrophoresis \[[@B71-microorganisms-06-00001]\]; this can be greatly improved by capillary electrophoresis on an automated DNA sequencing instrument. Another potential issue with this method is the stability of the VNTR markers, although the small VNTRs used in this scheme appear to be stable upon repeat testing \[[@B70-microorganisms-06-00001]\]. To date, the *P. acnes* MLVA scheme is yet to be utilized by the wider research community. The *P. acnes* MLVA database is scheduled to be released shortly at MLVAbank for Microbes Genotyping \[[@B72-microorganisms-06-00001]\]. 8.6. Whole Genome Sequencing {#sec8dot6-microorganisms-06-00001} ---------------------------- High resolution MLST of *P. acnes* generates phylogenies that, in general, match those obtained by WGS analysis. When analyzing a large number of isolates, however, it is an expensive typing methodology, especially with Sanger sequencing costs which are generally in the range of £3-to-£5 per read depending on the service provider. As the costs of NGS have continued to fall, and are comparable per isolate to MLST, rapid and high quality WGS analysis of bacterial strains is now increasing common for epidemiological investigations \[[@B73-microorganisms-06-00001],[@B74-microorganisms-06-00001]\]. Unlike traditional MLST which is normally based on seven loci, WGS provides the opportunity to develop scaled up core-genome MLST (cgMLST) schemes, or a whole-genome MLST (wgMLST) approach, where all the equivalent loci between isolates are compared (core + accessory); this is based on de novo assembly of sequences to a reference genome \[[@B75-microorganisms-06-00001]\]. As an alternate to the gene-by-gene approach, the phylogenetic relationships between isolates can also be based on the analysis of differences in genome core SNPs, which offers fine resolution but requires a reference sequence; this method is based on mapping short reads to a reference genome. To date, over 120 *P. acnes* genomes have been completed or are at the draft stage \[[@B76-microorganisms-06-00001]\], and both core SNP and cgMLST analysis (based on the concatenated sequence of 76 HK genes) have been applied to this dataset to create whole genome trees \[[@B38-microorganisms-06-00001],[@B43-microorganisms-06-00001],[@B45-microorganisms-06-00001],[@B46-microorganisms-06-00001]\]. WGS is the ultimate typing tool providing information on the full gene complement and genomic rearrangements. The gene-by-gene approach does, however, enable a hierarchical approach to typing where the number of genes analyzed is selected on the level of resolution required. Ultimately, WGS will likely replace other current molecular methods of *P. acnes* typing as costs continue to decline and bench top NGS sequencers become commonplace; the genome is also relatively small for sequencing (2.5 Mbp). Furthermore, various online web servers now provide software that will generate MLST profiles from whole genome data, such as Bacterial Isolate Genome Sequence Database (BIGSdb) \[[@B77-microorganisms-06-00001],[@B78-microorganisms-06-00001]\] and the Centre for Genomic Epidemiology \[[@B79-microorganisms-06-00001]\]. A key issue to address moving forward, however, is agreement on the optimal number of genes to use for WGS typing of *P. acnes* while maintaining epidemiological concordance; this would help to standardize interlaboratory comparisons. 9. Typing Algorithm {#sec9-microorganisms-06-00001} =================== The level of resolution required for a typing scheme depends on the epidemiological questions being addressed. For *P. acnes* we currently use the algorithm highlighted in [Figure 8](#microorganisms-06-00001-f008){ref-type="fig"} as our normal approach to typing of the bacterium. For culture-based investigations, multiple colonies (*n* = 15--20) from a sample are rapidly pre-screened by multiplex PCR as a first line typing method, before a further subset of these isolates (*n* = 5) are analyzed by high resolution MLST/ SLST if required. This approach not only helps to estimate the degree of phylogroup heterogeneity within a sample, but also maximizes the number of genetically diverse isolates selected for additional high resolution typing while keeping costs down. For non-culture-based typing, we utilize HR-SLST for direct metagenomic analysis based on the Illumina MiSeq platform. This approach enables the relative abundance of different pylogroups and SLST genotypes within a sample to be determined, but is more technically demanding. 10. Conclusions {#sec10-microorganisms-06-00001} =============== In conclusion, the introduction of *recA* and *tly* gene sequence typing of *P. acnes* over 10 years ago proved an important first step in dissecting the underlying phylogenetic structure of this bacterium and the association of different genetic groupings with human health and disease. Since then, many newer molecular-based typing methodologies for *P. acnes* have been introduced. In particular, multiplex PCR and MLST/ HR-SLST, either individually or combined, now represent popular choices for typing, providing valuable and accurate information of the genetic population structure of the bacterium within clinical samples. A.M. is funded under the European Union Regional Development Fund (ERDF) EU Sustainable Competitiveness Programme for N. Ireland & the N. Ireland Public Health Agency (HSC R&D). The following are available online at <http://www.mdpi.com/2076-2607/6/1/1/s1>. ###### Click here for additional data file. The author declares no conflict of interest. {#microorganisms-06-00001-f001} ![Minimum evolution phylogenetic trees of concatenated *recA* and *tly* sequences (1824 bp) from isolates selected to represent all known phylogroups (**A**), and concatenated gene sequences (4235 bp) from all current STs in the MLST~8~ database (**B**); the latter is updated from \[[@B43-microorganisms-06-00001]\] with STs. Bootstrapping statistics were performed using 500 data sets, and only bootstrap values ≥70% are shown. Clonal complexes (CC) are indicated. An overlapping cluster of strains from CC4 (type IA~1~), CC2 (type IA~2~) and CC5 (type IB) can clearly be seen based on analysis of *recA* and *tly* sequences (**A**), but on MLST~8~ analysis strains from these different phylogroups form separate, distinct clusters with high bootstrap values (**B**).](microorganisms-06-00001-g002){#microorganisms-06-00001-f002} ![Split decomposition analysis of all current *recA* (**A**) and *tly* (**B**) allele sequences. *recA* allele sequences (753 bp) were taken from the MLST~9~ database \[[@B53-microorganisms-06-00001]\] and *tly* allele sequences (777 bp) from the MLST~8~ database \[[@B54-microorganisms-06-00001]\]; the latter is updated from \[[@B12-microorganisms-06-00001]\] with new allele sequences. No evidence of statistically significant recombination was identified in either genes using the Phi test (*p* = 0.135 for *recA*; *p* = 0.735 for *tly*), although some interconnected pathways were present.](microorganisms-06-00001-g003){#microorganisms-06-00001-f003} {#microorganisms-06-00001-f004} {#microorganisms-06-00001-f005} ![Minimum evolution tree of the current 112 STs from the 483--497 bp HR-SLST target locus. Sequences are from the SLST database at SLST for *Cutibacterium acnes* (formerly *Propionibacterium acnes*) \[[@B68-microorganisms-06-00001]\]. Bootstrapping statistics were performed using 500 data sets, and only bootstrap values ≥70% are shown.](microorganisms-06-00001-g006){#microorganisms-06-00001-f006} ![Split decomposition analysis of all current SLST allele sequences (483--497 bp) from the HR-SLST database \[[@B68-microorganisms-06-00001]\]. No evidence of statistically significant recombination was identified using the Phi test (*p* = 0.399). Updated from \[[@B59-microorganisms-06-00001]\] with new allele STs.](microorganisms-06-00001-g007){#microorganisms-06-00001-f007} {#microorganisms-06-00001-f008} microorganisms-06-00001-t001_Table 1 ###### Correlation between ribotypes, phylogroups and MLST CCs. MLST~8~ MLST~9~ ----------- ------------ ---------------- ---------------- Ribotype Phylogroup Clonal Complex Clonal Complex 1, 5, 532 IA~1~ CC1 CC18 1, 4, 5 IA~1~ CC3 CC3 8 IA~1~ CC4 CC31 3, 16 IA~2~ CC2 CC28 1 IB CC5 CC36 5 IC CC107 Singletons 2, 6 II CC6/CC72 CC60 9 III CC77 ND

**Core tip**: Diet, as well as intestinal microbiota, is a key regulator of intestinal permeability, the alteration of which is central in the derangement of the gut-liver axis. In patients with non-alcoholic fatty liver disease (NAFLD), intestinal permeability is increased, promoting translocation of bacteria-derived products into the portal circulation and increasing hepatic exposure to injurious substances that stimulate hepatic inflammation and fibrosis. In animal models, high-fat diet or high-fructose intake has been associated with increased gut permeability. The aim of this study was to detect diet-induced modification of intestinal permeability in non-diabetic patients with NAFLD undergoing a Mediterranean diet or a low-fat diet. INTRODUCTION ============ In western countries, non-alcoholic fatty liver disease (NAFLD) is the most common chronic liver disease and is closely associated with an unhealthy lifestyle. Patients with NAFLD are reported to engage in an excessive consumption of total energy, refined carbohydrates (including fructose), saturated fats and cholesterol with an insufficient intake of polyunsaturated fats, fibres and antioxidants (vitamin C and vitamin E)\[[@B1],[@B2]\]. Because there is no pharmacotherapy currently available for patients with NAFLD, lifestyle modifications remain the cornerstone management\[[@B3],[@B4]\]. In overweight/obese patients with NAFLD, calorie restriction drives weight loss, reducion of liver fat, and histological improvement of non-alcoholic steatohepatitis (NASH)\[[@B5],[@B6]\] , but there are conflicting data on which hypocaloric dietary plan should be adopted according to the macronutrient composition\[[@B7],[@B8]\]. Low-fat diets, popular in the treatment of obesity, are reported to be more effective than low-carbohydrate diets in reducing total cholesterol and low-density lipoprotein (LDL) cholesterol concentrations\[[@B9]\], whereas low-carbohydrate diets can be more effective than low-fat diets in increasing high-density lipoprotein (HDL) cholesterol concentrations and reducing triglyceride levels and transaminase levels through a greater reduction in the concentration of 24-h circulating insulin under both isocaloric and hypocaloric conditions\[[@B10]-[@B13]\]. The Mediterranean diet, first described in the 1960s by Ancel Keys, is a low-carbohydrate nutritional model inspired by the traditional diets of countries along the Mediterranean Sea Basin, an area of origin for the *Olea Europaea* plant\[[@B14],[@B15]\]. European guidelines for management of NAFLD suggest the Mediterranean diet as a reference for macronutrient composition\[[@B16]\]. Diet, as well as intestinal microbiota, is a key regulator of intestinal permeability, the alteration of which is central in the derangement of the gut-liver axis\[[@B17]\]. In patients with NAFLD, intestinal permeability is increased, promoting translocation of bacteria-derived products into the portal circulation and increasing hepatic exposure to injurious substances that stimulate hepatic inflammation and fibrosis\[[@B18]\]. In animal models, adaptation of a high-fat diet or high-fructose intake has been associated with increased gut permeability and metabolic endotoxiemia\[[@B19]-[@B21]\]. Although diet can significantly influence intestinal permeability, clinical studies investigating the effects of dietary interventions on the intestinal permeability of NAFLD patients are lacking. The aim of this study was to detect diet-induced modification of intestinal permeability in non-diabetic patients with NAFLD undergoing a Mediterranean diet or a low-fat diet in a crossover comparison. Secondary outcomes were modifications in transaminase levels, body weight and waist circumference. MATERIALS AND METHODS ===================== Patients -------- This single-centre prospective dietary intervention open-label trial was conducted in the outpatient clinic for Gastroenterology and Liver Disease at the Catholic University of the Sacred Heart of Rome and was approved by the local Ethics Committee. Every patient provided written informed consent before enrolment. Adult patients with a histologically confirmed diagnosis of NAFLD and increased levels of alanine aminotransferase (ALT) \> 1.5 time the upper normal limit (30 IU/L in males and 19 IU/L in females) were included\[[@B22]\]. We chose to include only patients with elevated transaminase levels at baseline because one of the secondary end-points of the study was a reduction in transaminases. All other causes of liver damage were ruled out. In particular, all patients had negative serological markers for hepatitis B and hepatitis C, an alcoholic consumption of \< 30 g of ethanol for men and \< 20 g for women, were not taking any hepatotoxic drug, did not have evidence of metabolic disease (Wilson\'s disease, hereditary haemochromatosis) or autoimmune disease (autoimmune hepatitis, primitive biliary cirrhosis, primitive sclerosing cholangitis). The main exclusion criteria were the presence of liver cirrhosis, diabetes, body mass index (BMI) ≥ 35 kg/m² or previous bariatric surgery, because these conditions already have well-established dietary recommendations. Participants had to have stable body weight (variation \< 5% within the preceding 3-mo period) and maintain their usual level of physical activity for the duration of the intervention. The participants who were already taking medications had to be on a stable regimen for at least 6 months before study enrolment. The dose of these medications had to remain stable during the study. Participants were allowed to start a new medication only if it was medically necessary. No consumption of supplements (including vitamin E) was allowed during the study. Design of the study ------------------- Patients underwent a dietary treatment via a crossover design lasting 48 wk: 16-wk of a Mediterranean diet (W1-W16), 16-wk of a free wash-out diet (W17-W32) and 16-wk of a low-fat diet (W33-W48). Both the Mediterranean diet (40% carbohydrates, 40% fat with \< 10% saturated fat, 20% protein) and low-fat diet (62% carbohydrates, 18% fat, 20% protein) were caloric-equivalent (1400 Kcal) and were processed by the dietary service of our Hospital in accordance with international standards (Supplementary Tables 1-3). We chose hypocaloric diets as opposed to isocaloric diets because the beneficial effect of caloric restriction on NAFLD has been well-documented in the literature, and it would have been unethical to not have provided patients with an indication recommended by all guidelines. Before starting each diet and at the end of each dietary period, patients underwent a medical examination, provided and performed electrocardiogram and blood tests, including blood count, ALT, aspartate aminotransferase (AST), gamma glutamyl transpeptidase (GGT), alkaline phosphatase, bilirubin, blood glucose, basal insulinaemia, total cholesterol, HDL cholesterol, triglycerides, albumin, creatinine, sodium, potassium, and urine test. All blood tests were determined by standard methods in a certified clinical chemistry laboratory. Anthropometric measurements were directly measured by the same dietitian at every visit: weight (light clothing, shoes off) to the nearest 0.1 kg, height (standing tall, feet hip-width apart and head level) to the nearest 1 cm, waist and hip circumference (tape lying flat and level, taut but not tight) to the nearest 1 cm. All patients were re-evaluated bimonthly by the same dietitian for compliance to the prescribed diet and physical activity level. Dietary history was carried out by means of a 24-h recall\[[@B23]\], food frequency questionnaire (including alcohol intake)\[[@B24]\], and 10-point Mediterranean-diet Score\[[@B25]\], with subsequent calculation of caloric-nutritional intake. Physical activity was assessed at both baseline and at the end of each dietary period using the International Physical Activity Questionnaire (IPAQ) long form\[[@B26]\]. Evaluation of intestinal permeability ------------------------------------- Intestinal permeability was evaluated at baseline and at the end of each diet period using the validated method of urinary excretion of chromium-51 ethylene diamine tetraacetate excretion testing (51Cr-EDTA). Many methods for measuring intestinal permeability in the clinical setting have been validated (*e.g*., lactulose/mannitol, sucralose, sucrose), but none has proven to be superior to others nor has gained a place in everyday clinical practice\[[@B17]\]. In our study we chose to use the 51Cr-EDTA test because it is poorly influenced by bacterial degradation in cases of small intestine bacterial overgrowth, sometimes associated with NAFLD, and it is easily repeatable and less time-consuming for the health workers compared to oligosaccharides-based tests\[[@B27]\] . After an overnight fast, patients drank 1.85 MBq 51CrEDTA (Amersham Health, England) in 10 mL water, and urine was collected for the next 24 h. Two 3-mL samples of the collected urine were assessed with a gamma counter (LKB-Wallac 1282 Compugamma, Turku, Finland), as previously described\[[@B28]\]. 51Cr-EDTA clearance was calculated using the following formula: \[(Mean urinary counts × Urinary volume) × (Standard counts × 50)\]^-1^. Results were expressed as percentages of the ingested dose. In normal subjects, 1%-3% of an orally administered dose of 51Cr-EDTA is absorbed by the gastrointestinal tract. Statistical analysis -------------------- The data were analysed using a descriptive statistical method (Statistical Package for Social Sciences 17.0), and the results were expressed as the means and standard deviations. Student's *t*-tests were used to determine statistically significant differences (*P* \< 0.05) between groups. RESULTS ======= Fifty-four patients with NAFLD underwent liver biopsy at our centre during the two-year timeframe (Figure [1](#F1){ref-type="fig"}). Twenty-four patients were not eligible for the study because of one or more exclusion criteria (16 patients for ALT levels \< 1.5 time the upper normal limit, 5 patients for diabetes, 3 patients for BMI ≥ 35 kg/m²). Ten patients refused to participate in the study (5 patients because of lack of motivation to start a diet program and 5 patients because of working reasons). Twenty patients were enrolled and started the Mediterranean diet. Eighteen patients completed the 16-wk Mediterranean diet (W1-W16), whereas 2 patients dropped out due to poor compliance to the diet. After 16 weeks of a free wash-out diet (W17-W32), 4 patients refused to continue the study. 14 patients started the 16 wk of the low-fat diet (W33-48), but 2 patients dropped out due to poor compliance to the diet. Twelve patients completed the entire protocol. According to the results of IPAQ long form, no patient significantly changed his or her level of physical activity during the study. {#F1} Demographics ------------ Clinical and laboratory features of the cohort are summarized in Table [1](#T1){ref-type="table"}. Overall, 18 (90%) of the patients enrolled were male, with a mean age of 42.7 years. Thirteen (65%) patients had hypertension, 5 (25%) had hypercholesterolemia, and 13 (65%) met diagnostic criteria for metabolic syndrome according to ATP III\[[@B29]\] (Table [1](#T1){ref-type="table"}). Based on liver biopsy results, 12 (60%) presented with a Kleiner score of 3-4 and 8 (40%) with a Kleiner score ≥ 5, indicative of NASH\[[@B30]\]. Mean BMI was 30.9 ± 3.6, mean waist circumference was 107.2 ± 9.0 cm, mean hip circumference was 107.5 ± 6.7 cm, mean waist/hip ratio was 0.99 ± 0.05, mean systolic blood pressure was 129.0 ± 14.3 mmHg, and mean diastolic blood pressure was 84.5 ± 11.4 mmHg. Mean fasting blood glucose was 93.8±13.1 mg/dL, mean HOMA-IR score\[[@B31]\] was 5.2, mean total cholesterol was 186.5 ± 45.1 mg/dL, mean LDL cholesterol was 112.9 ± 44.1 mg/dL, mean HDL cholesterol was 41.7 ± 6.3 mg/dL, mean triglycerides were 139.0 ± 90.4 mg/dL, mean ALT levels were 80.5 ± 41.6 IU/L, mean AST levels were 39.8 ± 15.1 IU/L, and mean GGT levels were 102.3 ± 123.5 IU/L. ###### Effects on clinical, metabolic and intestinal permeability paramethers after 16 wk of Mediterranean diet or low-fat diet in patients with non-alcoholic fatty liver disease **Basal** **After Mediterranean diet** ***P* value (≥ 0.05 NS)** **After wash-out** **After low-fat diet** ***P* value (≥ 0.05 NS)** --------------------------------- --------------- ------------------------------ --------------------------- -------------------- ------------------------ --------------------------- Weight (kg) 91.3 ± 10.9 86.0 ± 12.4 0.003 86.3 ± 8.4 85.7 ± 9.4 NS Body mass index 30.9 ± 3.6 29.3 ± 4.1 NS 29.7 ± 2.9 29.3 ± 3.2 NS Waist (cm) 107.2 ± 9.0 99.3 ± 11.5 0.001 102.0 ± 11.5 96.8 ± 11.5 NS Hip (cm) 107.5 ± 6.7 103.9 ±6.9 0.001 103.7 ± 6.3 101.7 ± 11.1 NS Waist/hip ratio 0.99 ± 0.05 0.96 ± 0.08 NS 0.97 ± 0.04 0.95 ± 0.06 NS Systolic blood pressure (mmHg) 129.0 ± 14.3 130.5 ± 20.3 NS 130.0 ± 8.9 122.5 ± 17.5 NS Diastolic blood pressure (mmHg) 84.5 ± 11.4 81.0 ± 14.7 NS 79.2 ± 9.2 75.8 ± 14.3 NS Glycemia (mg/dL) 93.8 ± 13.1 92.9 ± 11.8 NS 95.5 ± 9.0 96.2 ± 7.3 NS HOMA-IR 5.2 ± 2.5 3.5 ± 2.1 NS 4.3 ± 1.2 3.8 ± 1.5 NS Total Cholesterol (mg/dL) 186.5 ± 45.1 176.9 ± 42.9 NS 186.3 ± 65.7 177.0 ± 50.5 NS LDL cholesterol (mg/dL) 112.9 ± 44.1 113.7 ± 35.2 NS 119.7 ± 55.8 109.5 ± 44.6 NS HDL cholesterol (mg/dL) 41.7 ± 6.3 42.1 ± 9.3 NS 44.7 ± 4.5 44.7 ± 4.7 NS Tryglicerid (mg/dL) 139.0 ± 90.4 105.4 ± 45.4 NS 110.5 ± 48.7 105.8 ± 36.1 NS ALT (UI/L) 80.5 ± 41.6 52.2 ± 32.3 0.0001 56.2 ± 33.8 58.3 ± 38.7 NS AST (UI/L) 39.8 ± 15,1 33.4 ± 15.7 0.01 26.5 ± 10.4 29.0 ± 11.3 NS GGT (UI/L) 102.3 ± 123.5 89.6 ± 129.5 NS 100.7 ± 128.3 70.7 ± 75.3 NS Albumin (g/dL) 4.6 ± 0.3 4.5 ± 0.3 NS 4.6 ± 0.4 4.5 ± 0.4 NS Emoglobin (g/dL) 15.1 ± 1.1 14.9 ± 1.1 NS 15.4 ± 1.3 15.3 ± 1.3 NS Creatinin (mg/dL) 0.9 ± 0.1 0.9 ± 0.1 NS 0.9 ± 0.1 0.9 ± 0.1 NS W1: Week 1; W16: Week 16; W32: Week 32; W48: Week 48; HOMA-IR: Homeostatic model assessment - insulin resistance; LDL: Low density lipoprotein; HDL: High density lipoprotein; ALT: Alanine transaminase; AST: Aspartate transaminase; GGT: Gamma glutamyl trasferase. Nutritional assessment ---------------------- At the baseline nutritional assessment, patients ingested a mean of 1798 kcal per day, divided as follows: 53.2% carbohydrates, 18.3% proteins and 28.4% lipids, of which 11.4 g were saturated fat, 4.5 g polyunsaturated fats, 29.7 g monounsaturated fatty acids and 248.2 g cholesterol (Table [2](#T2){ref-type="table"}). Regarding micronutrients, patients ingested a mean of only 16.9 g of total daily fibre, 88.8 mcg of vitamin A, 68.3 g of vitamin C, 7.7 g of vitamin E and 9.5 g of total iron. Regarding the daily nutrient distribution, 22% of patients usually did not have breakfast; 45% were not used to having a full lunch and 75% did not have a mid-morning and afternoon snack. The results of the food frequency questionnaire administered at baseline are shown in the supplementary appendix. According to the 10-point Mediterranean-diet Score, 45% of patients obtained a score ≤ 3, indicative of a food pattern poorly adherent to the Mediterranean diet. ###### Nutritional assessment **Basal (W1)** **After mediterranean diet (W16)** **After wash-out (W32)** **After low-fat diet (W48)** --------------------------------- ---------------- ------------------------------------ -------------------------- ------------------------------ Macronutrients kcal per day 1798 1421 1853 1414 Carbohydrates (%) 53.2 40.5 55.0 62.1 Proteins (%) 18.3 21.1 16.5 20.2 Lipids (%) 28.4 38.4 28.5 17.7 Saturated fat (g) 11.4 7.2 9.0 5.5 Polyunsaturated fats (g) 4.5 3.5 4.2 2.6 Monounsaturated fatty acids (g) 29.7 27.2 25.0 14.4 Cholesterol (g) 248.2 157.5 258.2 153.3 Micronutrients Total daily fiber (g) 16.9 28.2 16.8 21.2 Vitamin A (mcg) 88.8 47.6 86.1 33.4 Vitamin C (g) 68.3 143.3 72.6 119.3 Vitamin E (g) 7.7 7.8 6.7 3.3 Total iron (g) 9.5 15.5 10.7 12.7 W1: Week 1; W16: Week 16; W32: Week 32; W48: Week 48. Effects of nutritional intervention ----------------------------------- An assessment of adherence to the two diet periods is shown in Table [2](#T2){ref-type="table"}. A total of 18 patients out of 20 were judged to be adherent to the Mediterranean diet, and 12 out of 14 patients were judged to be adherent to the low-fat diet. At the end of 16 wk of Mediterranean diet (W1-W16), were observed a significant reduction in mean body weight (-5.3 ± 4.1 kg, *P* = 0.003), mean waist circumference (-7.9 ± 4.9 cm, *P* = 0.001), mean hip circumference (-8.2 ± 8.6 cm, *P* = 0.001), and mean transaminase levels (ALT -28.3 ± 11.9 IU/L, *P* = 0.0001; AST -6.4 ± 56.3 IU/L, *P* = 0.01). (Table [1](#T1){ref-type="table"} and Figure [2](#F2){ref-type="fig"}). {#F2} At the end of 16 wk of the free wash-out diet (W16-W32), no significant variations were observed in anthropometric or laboratory parameters compared to results observed at the end of the Mediterranean diet period. At the end of the 16 wk of the low-fat diet (W32-W48), no significant variations were observed in any of the anthropometric or laboratory parameters analysed (Table [1](#T1){ref-type="table"} and Figure [2](#F2){ref-type="fig"}). Analysing only patients who completed both diets (Figure [3](#F3){ref-type="fig"}), we observed that after 16-weeks of the Mediterranean diet, compared to 16-wk of the low-fat diet in the same patients, there was a significant reduction in mean body weight (-7.8 ± 2.3 kg *vs* 0.6 ± 2.4 kg, *P* = 0.004), mean waist circumference (-11.2 ± 2.6 cm *vs* 4.6 ± 7.6 cm, *P* = 0.04), and mean ALT levels (-28.5 ± 9.5 IU/L *vs* 2.2 ± 5.5 IU/L, *P* = 0.04). {#F3} No adverse events related to either the Mediterranean diet or low-fat diet were recorded. Intestinal permeability ----------------------- Seventy percent of patients had intestinal permeability alteration at baseline with a mean percentage retention of 51Cr-EDTA of 5.4% ± 4.7% (Figure [4](#F4){ref-type="fig"}). No significant changes in intestinal permeability were observed at the end of the Mediterranean diet period (6.0% ± 3.3%), after the wash-out period (7.2% ± 5.4%) or at the end of the low-fat diet period (7.3% ± 4.3%). {#F4} DISCUSSION ========== Our crossover study demonstrated a significant reduction in body weight, waist circumference and transaminase levels after 16 wk of Mediterranean diet and maintenance of these results following a low-fat diet period. The Mediterranean diet is a low-carbohydrate high-unsaturated fat nutritional model that appears to be an ideal diet for improving insulin sensitivity in NAFLD patients\[[@B32]\]. In a study conducted in Israel with 259 diabetic obese patients, a Mediterranean diet was superior to two types of low-fat diets for reducing serum transaminase levels at 6 and 12 mo\[[@B33]\]. A meta-analysis demonstrated that omega-3 fatty acids, found in the Mediterranean diet, were beneficial for reducing hepatic steatosis\[[@B34]\]. In a small pilot study conducted in Australia with 12 patients with NAFLD, 6 wk of *ad libitum* Mediterranean diet were more effective than a low-fat diet for improving insulin sensitivity and reducing the fat content in the liver, measured by magnetic resonance spectroscopy, even without a change in weight or serum transaminase levels\[[@B35]\]. A randomized trial of the Mediterranean diet *vs* a low-fat diet (MEDINA) is currently ongoing in Australia and New Zealand, with improvement of insulin sensitivity set as the primary end-point\[[@B36]\]. The Prevención con Dieta Mediterránea (PREDIMED) trial, conducted in Spain with over 7000 persons at high risk for cardiovascular disease, showed that a Mediterranean diet supplemented with extra-virgin olive oil or nuts reduced the incidence of major cardiovascular events over a control diet\[[@B37]\]. In a subsequent sub-analysis, it was observed that patients with lower adherence to the Mediterranean diet had higher values of serum transaminases\[[@B38]\]. In our study, we confirmed that 16 weeks of a Mediterranean diet was effective for improving serum transaminase levels in patients with NAFLD. Recently, a randomized 12-wk interventional study conducted in Australia with *ad libitum* isocaloric diets demonstrated that both a Mediterranean and a low-fat diet reduced hepatic steatosis and transaminases by a similar degree, but the Mediterranean diet also improved Framingham risk scores\[[@B39]\]. The choice of a crossover design for this study was based on our attempt to control confounding factors in the best possible way, as each subject served as their own control. However, it is very likely that there was an effect of the sequence of the diets, because the patients arrived at the end of a free wash-out diet (W32) with the same clinical characteristics that they had reached at the end of the Mediterranean diet (W16). In other words, despite a planned wash-out period, patients did not start the low-fat diet in the same condition in which they started the Mediterranean diet. We can therefore state that, in our study, 16 weeks of a low-fat diet was sufficient to maintain the benefits obtained after the Mediterranean diet period. It is well-known that elevated concentrations of fructose favour pro-inflammatory microbiota, thereby producing endotoxins and suppressing production of short-chain fatty acids that are essential for intestinal barrier function\[[@B40]\]. Pro-inflammatory microbiota and their products recruit macrophages and bind to toll-like receptors, leading to the release of cytokines, such as tumour necrosis factor-α, causing mucosal inflammation\[[@B41]\]. Subsequently, inflammation decreases expression of tight junction proteins, resulting in higher permeability of the gut barrier\[[@B42]\]. Diet-induced increases in blood lipopolysaccharide levels are known as metabolic endotoxaemia and play an important role in the activation of Toll-like receptor-mediated low-grade liver inflammation, which are associated with NAFLD and NASH\[[@B43]\]. Current evidence from animal studies suggests that a high-fat diet or a high-fructose diet can induce metabolic endotoxaemia by altering the intestinal tight junction proteins, mainly zonula occludens-1 and occludin\[[@B44]-[@B47]\]. Our group has provided evidence that increased intestinal permeability in NAFLD patients is caused by disruption of intestinal tight junctions, as documented by decreased expression of zonula occludens 1 in the intestinal mucosa\[[@B48]\]. In NAFLD adolescents, postprandial endotoxin levels were increased compared to healthy subjects in response to fructose, but not glucose, beverages (consumed with meals) in a 24-hour feeding challenge\[[@B49]\]. On the other hand, there are currently no data concerning diet modulation of intestinal permeability in patients with NAFLD, and it is thought that a healthy diet can reduce intestinal permeability in patients with NAFLD by restoring the integrity of tight junctions. The Mediterranean diet contains a high intake of mono- and polyunsaturated fatty acids, fibres, polyphenols, antioxidants and phytochemicals; many of these components promote short-chain fatty acid-producing gut bacteria and have significant prebiotic effects\[[@B50]\]. As such, we hypothesized a therapeutic effect of the Mediterranean diet for reducing impaired intestinal permeability in patients with NAFLD. Under conditions of enhanced intestinal permeability, an increased fraction of orally administered 51Cr-EDTA crosses the intestinal epithelium along the paracellular pathway, enters the blood stream and is rapidly cleared through glomerular filtration\[[@B27]\]. Arslan et al\[[@B51]\] found that 51Cr-EDTA excretion is a good indicator of the severity of gut-mucosal inflammation and loss of tight junctions. In our study, the majority of patients presented at baseline, as expected, with high intestinal permeability evaluated according to 51Cr-EDTA, but neither 16 wk of a Mediterranean diet nor 16 wk of a low-fat diet were sufficient to modulate it. In our view, the modulation of intestinal permeability in humans is much more difficult to obtain than in animal models. For example, probiotics cannot modify migraine-associated intestinal permeability changes\[[@B52]\] and a recent Finnish study demonstrated that nor probiotics nor long chain polyunsaturated fatty acid supplements can counteract intestinal permeability changes in pregnancy\[[@B53]\]. On the other hand, a potential improvement of intestinal permeability (assessed by an indirect test) induced by higher dietary fiber intake was observed after a longer dietary intervention\[[@B54]\]. We ask ourselves whether our result could be modified with a different duration of the observation and of the diet program. A limitation of our study was the difficulty with recruitment because many patients refused to enrol for personal reasons. This limitation reflects the difficulty that patients experience adhering to a lifestyle intervention, which is far more compelling than taking medication. Larger longer-term studies are needed to confirm the durability of the benefits of a Mediterranean diet. In conclusion, the Mediterranean diet is a safe and effective strategy for treating overweight, visceral obesity and serum transaminase in patients with NAFLD, thereby confirming that it should be the first step in a treatment program for patient with NAFLD. If the Mediterranean diet can improve intestinal permeability in patients with NAFLD, it deserves further investigation. ARTICLE HIGHLIGHTS ================== Research background ------------------- In patients with non-alcoholic fatty liver disease (NAFLD), intestinal permeability is increased, promoting translocation of bacteria-derived products into the portal circulation and increasing hepatic exposure to injurious substances that stimulate hepatic inflammation and fibrosis. In animal models, adaptation of a high-fat diet or high-fructose intake has been associated with increased gut permeability and metabolic endotoxiemia Research motivation ------------------- Although diet can significantly influence intestinal permeability, clinical studies investigating the effects of dietary interventions on the intestinal permeability of NAFLD patients are lacking. The Mediterranean diet contains a high intake of mono- and polyunsaturated fatty acids, fibres, polyphenols, antioxidants and phytochemicals; many of these components promote short-chain fatty acid-producing gut bacteria and have significant prebiotic effects. As such, we hypothesized a therapeutic effect of the Mediterranean diet for reducing impaired intestinal permeability in patients with NAFLD. Research objectives ------------------- Aim of the study is to detect diet-induced modification of intestinal permeability in patients with NAFLD undergoing a Mediterranean diet or a low-fat diet. Research methods ---------------- Patients underwent a dietary treatment via a crossover design lasting 48 wk: 16-wk of a Mediterranean diet, 16-wk of a free wash-out diet and 16-wk of a low-fat diet. Intestinal permeability was evaluated at baseline and at the end of each diet period using the validated method of urinary excretion of chromium-51 ethylene diamine tetraacetate excretion testing (51Cr-EDTA). 51Cr-EDTA excretion is a good indicator of the severity of gut-mucosal inflammation and loss of tight junctions. Research results ---------------- Patients with NAFLD experienced a significant reduction in body weight, waist circumference and transaminase levels after 16 wk of a Mediterranean diet. Seventy percent of patients had intestinal permeability alteration at baseline, but no significant changes in intestinal permeability were observed at the end of the Mediterranean diet period or at the end of the low-fat diet period. Research conclusions -------------------- The Mediterranean diet is a safe and effective strategy for treating overweight, visceral obesity and serum transaminase in patients with NAFLD, thereby confirming that it should be the first step in a treatment program. If the Mediterranean diet can improve intestinal permeability in patients with NAFLD, it deserves further investigation. Research perspectives --------------------- In our view, the modulation of intestinal permeability in humans is much more difficult to obtain than in animal models. We ask ourselves whether our result could be modified with a different duration of the observation and of the diet program. Manuscript source: Invited manuscript Specialty type: Gastroenterology and hepatology Country of origin: Italy Peer-review report classification Grade A (Excellent): 0 Grade B (Very good): B Grade C (Good): C Grade D (Fair): 0 Grade E (Poor): 0 Institutional review board statement: The study was reviewed and approved by the Ethics Committee of Catholic University of Sacred Heart (Rome, Italy). Informed consent statement: All study participants provided informed written consent prior to study enrollment. Conflict-of-interest statement: There are no conflicts of interest to report. STROBE statement: The authors have checked the manuscript according to STROBE checklist. Peer-review started: November 5, 2018 First decision: December 20, 2018 Article in press: January 18, 2019 P- Reviewer: Gong ZJ, Seta WK S- Editor: Ma RY L- Editor: A E- Editor: Yin SY [^1]: Author contributions: Grieco A was the guarantor and designed the study. Biolato M, Marrone G, Cefalo C and Racco S participated in the acquisition, analysis, and interpretation of the data, and drafted the initial manuscript. Manca F and Miggiano GA designed the diets and managed all the nutritional aspects of the study. Valenza V performed intestinal permeability tests. Gasbarrini A and Miele L revised the article critically for important intellectual content. Supported by the Italian Ministry of Education, University and Research (MIUR), scientific research programs of relevant national interest year 2010-2011, No. 2010C4JJWB. Corresponding author: Marco Biolato, MD, PhD, Staff Physician, Liver Transplant Medicine, Fondazione Policlinico Universitario A. Gemelli IRCCS, Largo A. Gemelli 8, Rome 00168, Italy. <marco.biolato@policlinicogemelli.it> Telephone: +39-6-30154469

Angiomyomatous hamartoma (AMH) of the lymph node is characterized by partial replacement of normal nodal parenchyma by disorganized blood vessels and smooth muscle cells with or without adipose tissue within a fibrous stroma. Inguinal and femoral lymph nodes are commonly involved \[[@b1-jptm-49-2-156]-[@b3-jptm-49-2-156]\], while popliteal lymph node involvement is uncommon \[[@b4-jptm-49-2-156],[@b5-jptm-49-2-156]\]. We report a rare case of AMH of the popliteal lymph node in a young patient with a clinical diagnosis of Baker's cyst. CASE REPORT =========== An 18-year-old male presented with pain and swelling in the left popliteal fossa lasting 3 years. The swelling slowly increased in size. There was no history of any trauma, fever, tuberculosis, weight loss, or chronic illness. A clinical diagnosis of Baker's cyst was made without any imaging studies. The lesion was operated on under local anesthesia. The swelling persisted, however, and physical examination revealed a 1.5 cm-sized, mildly tender, non-reducible, firm mass in the left popliteal fossa. An magnetic resonance imaging (MRI) scan revealed a 1.5×1.0-cm soft tissue mass close to the popliteal blood vessels without encasing them ([Fig. 1A](#f1-jptm-49-2-156){ref-type="fig"}, [B](#f1-jptm-49-2-156){ref-type="fig"}). No cyst was seen. The radiological differential diagnoses offered were benign neoplasm and pseudotumor. The left popliteal fossa was explored again and a firm, soft tissue mass loosely adherent to the popliteal blood vessels was identified and excised. The postoperative period was uneventful. Hematoxylin and eosin--stained sections revealed a lymph node with partial replacement of the parenchyma from the hilum to the cortex by fibrous tissue containing several irregular blood vessels of varying sizes, interspersed with spindle cells and smooth muscle cells ([Fig. 2A](#f2-jptm-49-2-156){ref-type="fig"}, [B](#f2-jptm-49-2-156){ref-type="fig"}). Mature adipose tissue infiltration was seen in a small area near the hilum of the lymph node ([Fig. 2C](#f2-jptm-49-2-156){ref-type="fig"}). The capsule was thickened. The subcapsular and medullary sinuses were obliterated. Cortical lymphoid tissue showed variable atrophy. Immunohistochemistry with a primary antibody against smooth muscle actin (SMA; 1:400, Thermo Scientific, Waltham, MA, USA) demonstrated smooth muscle cells in the blood vessel walls and in the stromal tissue ([Fig. 2D](#f2-jptm-49-2-156){ref-type="fig"}, [E](#f2-jptm-49-2-156){ref-type="fig"}). The rich vascularity of the lesion was highlighted by CD34 antibodies (1:100, Diagnostic BioSystems, Pleasanton, CA, USA) ([Fig. 2F](#f2-jptm-49-2-156){ref-type="fig"}). A diagnosis of AMH of the lymph node was made. DISCUSSION ========== Angiomyomatous hamartoma of the lymph node was first described by Chan *et al*. \[[@b1-jptm-49-2-156]\] in 1992 as a distinctive vascular hamartomatous lesion that primarily occurs in inguinal and femoral lymph nodes \[[@b2-jptm-49-2-156],[@b3-jptm-49-2-156]\]. Occasional cases have been reported in popliteal and cervical lymph nodes \[[@b4-jptm-49-2-156],[@b5-jptm-49-2-156]\]. Patients may present with painless or painful swelling \[[@b4-jptm-49-2-156]\]. Painful lesions in the popliteal fossa require careful evaluation because a number of non-neoplastic and neoplastic lesions can mimic this entity \[[@b6-jptm-49-2-156]\]. Baker's cyst results from herniation of the synovial membrane through the posterior capsule of the knee joint or by an escape of synovial fluid through an anatomic bursa next to the semimembranosus or gastrocnemius muscle. Baker's cysts are usually diagnosed by physical and radiological examination between the semimembranosus and medial head of the gastrocnemius \[[@b7-jptm-49-2-156]\]. Sometimes it is difficult to differentiate a Baker's cyst from other causes of posterior knee pain, and the differential clinical diagnosis of a Baker's cyst can include arterial aneurysm or tortuous blood vessels in the popliteal fossa, deep vessel thrombosis, adipose tissue or tumor \[[@b8-jptm-49-2-156]\]. The best imaging for evaluation of a popliteal cyst is MRI as it helps in localizing the cyst and any other internal derangements \[[@b9-jptm-49-2-156]\]. In the index case the initial misdiagnosis of Baker's cyst was due to the presence of painful swelling in the popliteal fossa and the diagnosis was based on physical examination alone. No radiological investigation was done prior to the initial surgery. To the best of our knowledge, to date only twenty-nine cases of AMH of the lymph node have been reported in the literature \[[@b3-jptm-49-2-156]\]. This is the third documented case in a popliteal lymph node. Mauro *et al*. \[[@b4-jptm-49-2-156]\] reported a case of AMH of the popliteal lymph node which presented with pain in the posterior knee in a 41- year-old male. The second case was reported by Prusac *et al*. \[[@b5-jptm-49-2-156]\] in a 14-year-old boy who presented with a right popliteal mass and right leg edema. In both cases, MRI scans revealed mass lesions with heterogeneous signal intensity. The index case also showed similar radiologic features. The radiologic diagnosis of AMH of the lymph node is usually a hemangioma or a tumor \[[@b4-jptm-49-2-156],[@b5-jptm-49-2-156]\]. In our case, the MRI scan findings suggested a benign tumor or pseudotumor. Microscopically, AMH is characterized by disorganized blood vessels and smooth muscle cells in a collagenous stroma with or without adipose tissue. Our case showed a small area of adipose tissue near the hilum of the lymph node. Histopathological differential diagnosis of AMH includes lymphangiomyomatosis, leiomyomatosis, and angiomyolipoma of the lymph node \[[@b10-jptm-49-2-156]\]. Nodal lymphangiomyomatosis occurs exclusively in women and is histologically characterized by the presence of smooth muscle cells forming fascicles and sheets around anastomosing ectatic vascular spaces, resulting in a pericytomatous pattern. Histologically smooth muscle cells are plumper, with lighter/clear cytoplasm, and sclerosis is absent. Nodal leiomyomatosis resembles leiomyoma and is made up of a proliferation of compact bundles of smooth muscle cells with an insignificant vascular component. Angiomyolipoma of the lymph node shows an epithelioid appearance, hypercellularity, pleomorphism, prominent perivascular arrangement and positivity for melanoma associated antigen human melanoma black 45. Our index case did not show ec-static vessels, pleomorphism or hemangiopericytoma like patterns. Adipose cells are known to occur as a significant component in a number of vasoproliferative lesions, such as nodal hemangioma, intramuscular hemangioma (infiltrative angiolipoma) and angiolipomatous hamartoma associated with Castleman's disease. It has been suggested that all cases of AMH of the lymph node with a significant adipose tissue component should be termed angiomyolipomatous hamartoma \[[@b10-jptm-49-2-156]\]. The pathogenesis of this lesion is unclear. A possible explanation is that AMH represents a vascular and smooth muscle proliferative response to chronic impairment of nodal lymphatic flow or to previous nodal inflammation \[[@b3-jptm-49-2-156]\]. Lesions in the popliteal fossa should be evaluated carefully, especially when associated with pain. The treatment depends on the type of lesion. Radiological investigation is mandatory for proper characterization. Histopathological examination will confirm this unusual benign entity that is managed surgically. **Conflicts of Interest** No potential conflict of interest relevant to this article was reported. {#f1-jptm-49-2-156} {#f2-jptm-49-2-156}

1. Introduction {#sec1} =============== Patients with end-stage renal disease (ESRD) often suffer from neurological complications, thereby contributing to morbidity and mortality \[[@B1]--[@B3]\]. Peripheral neuropathy is the most commonly reported neurological complication associated with chronic renal failure \[[@B4]\], with an incidence rate of more than 60% in patients on dialysis. It can affect sensory, motor, and cranial nerves and is characterized by axonal degeneration and demyelination \[[@B5]\]. Furthermore, uremic neuropathy is typically a distal, symmetric, and predominantly axonal type which affects the legs more than the arms \[[@B1], [@B2]\]. Lindblom and Tegner reported abnormalities in thermal sensation in 30% of patients with ESRD and concluded that small fiber neuropathy may be a distinct entity \[[@B6]\]. Small nerve fibers were traditionally thought to be invisible as they could not be detected in routine nerve conduction studies, leading to underestimation of small fiber neuropathy and physicians\' overlook. Skin biopsy for detecting altered interepidermal nerve fiber density remains the gold standard in diagnosing small fiber neuropathy, but it is still an invasive approach \[[@B7], [@B8]\]. Quantitative sensory testing (QST) is a noninvasive, psychophysical examination of small fiber functions through assessment of thresholds to thermal and cold signals \[[@B9]\]. Furthermore, QST has been shown to be useful in diagnosis of small fiber neuropathy \[[@B10]--[@B12]\], with a reported diagnostic sensitivity ranging from 60% to 85% \[[@B13]--[@B16]\]. Small fiber neuropathy is characterized by the presence of abnormal thermal thresholds in the distal limbs, linking to decreased quality of life in affected individuals, and is often neglected in clinical practice because of a paucity of readily available diagnostic methods \[[@B17]\]. Nutritional aspects and microvascular dysregulation were previously advocated as the potential causes of small fiber neuropathy \[[@B11], [@B18], [@B19]\]. In particular, arterial stiffness and malnutrition are common among patients undergoing hemodialysis (HD) and play their vital roles in prognostic significance \[[@B20]--[@B24]\]. The geriatric nutritional risk index (GNRI) is a useful and accurate indicator for the assessment of nutritional status in maintenance HD patients \[[@B25], [@B26]\]. Several studies have suggested that GNRI and arterial stiffness are associated with cognitive impairment, frailty, and inflammation \[[@B27]--[@B31]\]. However, the relationships of malnutrition risk and arterial stiffness with small fiber neuropathy remain not clearly understood and have never been investigated. Therefore, the aim of this study is to examine the associations of malnutrition risk and arterial stiffness, assessed by GNRI and brachial-ankle pulse wave velocity (baPWV), with small fiber neuropathy in patients receiving HD. 2. Materials and Methods {#sec2} ======================== 2.1. Study Patients {#sec2.1} ------------------- In the present study, we enrolled all patients (*n* = 170) undergoing thrice weekly maintenance HD treatment for more than 3 months at a dialysis unit of a regional hospital in Taiwan in April 2014. Patients who refused to undergo QST or baPWV examinations (*n* = 15) and had bilateral below knee amputations (*n* = 3) were excluded from the study. This study was approved by the Institutional Review Board of Kaohsiung Medical University Hospital. All study participants provided their written informed consent. The methods were carried out in accordance with the approved guidelines. 2.2. Measurement of QST for Assessment of Small Fiber Neuropathy {#sec2.2} ---------------------------------------------------------------- A Medoc TSA-II Neurosensory Analyzer was used for the QST \[[@B32]\]. The patients were seated comfortably in a quiet room with an ambient temperature of 24--25°C. The testing areas of the QST were the dorsolateral border of the feet for the lower limbs and the hypothenar eminence of the thumbs for the upper limbs. Cold and warm sensations were then tested after the test had been carefully explained to the patients, including the need to react promptly to any change in temperature. The test was performed by placing a 30 mm × 30 mm thermode on the testing areas of the skin. Four threshold temperature values for each testing area were recorded and averaged for analysis. If the sensation was identified incorrectly, this was also recorded. Based on values obtained from a control population, the mean threshold temperature ± 2 standard deviation (SD) for each testing area was considered as the upper (or lower) limit of normal. Small fiber neuropathy was defined as the abnormalities of cold or warm threshold in one of the testing areas (either hands or feet) using QST \[[@B9], [@B11]\]. 2.3. Measurement of baPWV for Assessment of Arterial Stiffness {#sec2.3} -------------------------------------------------------------- The baPWV was measured 10--30 minutes before HD session using an automatic waveform analyzer (VP-1000, Colin, Komaki, Japan) for each patient. The baPWV value was calculated as the transmission distance divided by the transmission time. The highest of bilateral baPWV values was used as the representative value for analysis. Arterial stiffness was defined as baPWV \> 1400 cm/s \[[@B33]\]. 2.4. Calculation of the GNRI {#sec2.4} ---------------------------- The GNRI was calculated according to baseline serum albumin level and body weight as follows: GNRI = \[14.89 × albumin (g/dL)\] + \[41.7 × (body weight/ideal body weight)\] \[[@B34]\]. The ideal body weight in the present study was defined as the value calculated from the height and a body mass index (BMI) of 22 \[[@B35]\]. If the patient\'s body weight was greater than the ideal body weight, body weight/ideal body weight was set to 1 \[[@B27]\]. 2.5. Demographic, Medical, and Laboratory Data {#sec2.5} ---------------------------------------------- Each patient\'s overnight fasting blood samples were obtained within 1 month of study enrollment for laboratory tests using an autoanalyzer (Roche Diagnostics GmbH, D-68298 Mannheim COBAS Integra 400). Kt/V was evaluated based on the Daugirdas formula to assess the efficiency of HD treatment \[[@B36]\]. Demographic and medical information, including age, gender, smoking history (ever *vs*. never), and comorbid conditions, were obtained from study patients\' medical records and interviews. 2.6. The Primary and Secondary Outcomes {#sec2.6} --------------------------------------- The primary outcome of this study was to elucidate the effects of diabetes and arterial stiffness on thermal threshold in QST among HD patients. The secondary outcome was to identify the determinants of abnormal QST threshold as well as the roles of nutritional risk and arterial stiffness in small fiber neuropathy. 2.7. Statistical Analysis {#sec2.7} ------------------------- Statistical analysis was performed using SPSS for Windows version 19.0 (SPSS Inc., Chicago, IL, USA). Data are expressed as percentage, mean ± SD, or median (25^th^--75^th^ percentile) for the duration of dialysis and serum triglycerides. Between-group differences were analyzed using the chi-square test for categorical variables and the independent *t*-test for continuous variables. Multivariate forward logistic regression analysis was performed to identify the factors associated with abnormal QST threshold, with adjustment for age, gender, duration of dialysis, smoking history, diabetes mellitus, baPWV \> 1400 cm/s, GNRI, hemoglobin, calcium-phosphorus (Ca × P) product, and Kt/V. A difference was considered significant for the *p* value \< 0.05. 3. Results {#sec3} ========== A total of 152 patients (78 men and 74 women, mean age 60.7 ± 10.7 years) were included in the present study. The mean baPWV was 1925 ± 526 cm/s, and the mean GNRI was 103.4 ± 9.0. 3.1. Comparison of QST Threshold between Diabetic and Nondiabetic Patients {#sec3.1} -------------------------------------------------------------------------- As diabetes mellitus (DM) is one of the major risk factors of neurologic complications, the comparison of baseline characteristics and thermal threshold on QST between diabetic and nondiabetic patients is summarized in [Table 1](#tab1){ref-type="table"}. Diabetic patients were more likely to be older in age and had shorter duration of dialysis, higher baPWV, lower cold threshold, and higher warm threshold among the QST of both hands and feet. Diabetic patients significantly had higher prevalence of abnormal cold and warm threshold on QST, except for warm threshold of feet, when compared to nondiabetic patients. 3.2. Comparison of QST Threshold between Patients with baPWV ≤ 1400 cm/s and \>1400 cm/s {#sec3.2} ---------------------------------------------------------------------------------------- Among QST results, patients with baPWV \> 1400 cm/s had significantly lower cold threshold of the right foot (*p* = 0.005) and significantly higher warm threshold of the right (*p* = 0.002) and left feet (*p* = 0.049) in comparison to QST in patients with baPWV ≤ 1400 cm/s ([Figure 1](#fig1){ref-type="fig"}). 3.3. Determinants of Abnormal QST Threshold {#sec3.3} ------------------------------------------- The study patients were then stratified into two groups according to normal or abnormal QST threshold. The comparison of baseline characteristics between patients with normal or abnormal QST threshold of the hands is shown in [Table 2](#tab2){ref-type="table"}. The prevalence of abnormal cold and warm sensation in the hands was 73.0% and 88.8%, respectively. Compared to patients with normal cold sensation, those with abnormal cold sensation in the hands had older age, shorter duration of dialysis, higher prevalence of DM, and higher baPWV. Moreover, compared to patients with normal warm sensation, patients with abnormal warm sensation in the hands had higher prevalence of male gender. The comparison of baseline characteristics between patients with normal or abnormal QST threshold of the feet was demonstrated in [Table 3](#tab3){ref-type="table"}. The prevalence of abnormal cold and warm sensation in the feet was 84.9% and 89.5%, respectively. Compared to patients with normal cold sensation, those with abnormal cold sensation in the feet had older age, higher prevalence of DM, higher baPWV, and lower GNRI. Furthermore, compared to patients with normal warm sensation, patients with abnormal warm sensation in the feet had higher Ca × P product. [Table 4](#tab4){ref-type="table"} shows the determinants of abnormal QST threshold in our study patients using multivariate forward logistic regression analysis. In the multivariate analysis (adjusted for age, gender, duration of dialysis, smoking history, DM, baPWV \> 1400 cm/s, GNRI, hemoglobin, Ca × P product, and Kt/V), old age (per 1 year; odds ratio (OR), 1.083; 95% confidence interval (CI), 1.042 to 1.126; *p* \< 0.001) was independently associated with abnormal cold threshold of the hands. Old age (per 1 year; OR, 1.081; 95% CI 1.026 to 1.139; *p* = 0.003) and male gender (OR, 4.450; 95% CI, 1.250 to 15.836; *p* = 0.021) were significantly associated with abnormal warm threshold of the hands. Furthermore, DM (OR, 3.996; 95% CI, 1.351 to 11.819; *p* = 0.012) and low GNRI (per 1 unit; OR, 0.935; 95% CI, 0.887 to 0.985; *p* = 0.012) were significantly associated with abnormal cold threshold of the feet. Finally, baPWV \> 1400 cm/s (OR, 5.479; 95% CI, 1.312 to 22.870; *p* = 0.020) and high Ca × P product (per 1 mg^2^/dL^2^; OR, 1.071; 95% CI, 1.013 to 1.132; *p* = 0.015) were independently associated with abnormal warm threshold of the feet. 4. Discussion {#sec4} ============= In this study, we found that the prevalence of abnormal QST threshold and baPWV were significantly higher in diabetic HD patients. Furthermore, we examined factors associated with small fiber neuropathy. In multivariate analysis, older age was associated with abnormal cold and warm threshold of the hands; DM and lower GNRI were significantly associated with abnormal cold threshold of the feet, whereas baPWV \> 1400 cm/s and higher Ca × P product were significantly associated with abnormal warm threshold of the feet in HD patients. The first important finding of this study is that patients undergoing HD with baPWV \> 1400 cm/s had significant higher warm threshold and lower cold threshold in the feet on QST. In addition, patients with abnormal thermal threshold in the hands and abnormal cold threshold in the feet had significantly higher baPWV level than those with normal QST threshold. This suggests that arterial stiffness might be an important factor contributed to abnormalities of small fiber function in HD patients. Our findings are in line with previous studies showing an independent association between pulse wave velocity and peripheral neuropathy in patients with DM \[[@B37]--[@B39]\]. Elevated baPWV evinces structural changes of the arterial wall, medial smooth muscle calcification, and breaks in elastic fibers in ESRD \[[@B40]\]. Increased arterial stiffness might lead to damage of the microcirculation, such as vasa nervorum, through increasing the transmission of detrimental pulsatile pressure waves \[[@B41]\]. Furthermore, emerging evidence suggests independent association of arterial stiffness with galectin-3 among HD patients \[[@B42]\]. Galectin-3 has been recently considered as a key molecule in the neural functions and nerve regeneration, and inhibition of galectin-3 could attenuate neuropathic pain after peripheral nerve injury \[[@B43], [@B44]\]. Our results may bring additional insight into the role of arterial stiffness in small fiber neuropathy in maintenance HD patients. Another important finding of this study is that a low GNRI was associated with an abnormal QST threshold. The pathogenesis of small fiber neuropathy in patients with ESRD is complex and remains not fully understood. Malnutrition is associated with increased level of tumor necrosis factor-*α* (TNF-*α*) in patients undergoing dialysis \[[@B45]\]. TNF-*α* has been shown to act a central role in the development of inflammatory demyelination \[[@B46]\]. Moreover, uremic toxins, combined with oxidative stress-related free radical activity and hyperkalemia, have been reported to cause motor, sensory, and autonomic nerve damage, making these as potentially causative factors in the development of uremic neuropathy \[[@B4]\]. Therefore, malnutrition risk might be involved in the pathogenesis of small fiber neuropathy in this patient population. Abnormalities of homeostasis among calcium, phosphate, and parathyroid hormone are quite common in patients with ESRD, and increased Ca × P product promotes vascular calcification \[[@B47]\]. Recent studies demonstrated that impaired intracellular calcium balance is implicated in the development of diabetic polyneuropathy \[[@B48]\], while calcimimetic can deter the progression of neuropathy by ameliorating inflammation, apoptosis, and autophagy through increased expression of the calcium-sensing receptors \[[@B49]\]. In this study, we found that a higher level of Ca × P product was associated with an abnormal QST threshold, suggesting that vascular calcification and homeostasis of calcium and phosphate may have their roles mediating functional abnormalities in small nerve fibers in patients undergoing HD. Maintenance HD patients often have certain neurological complications, including small fiber neuropathy, which can be linked to frailty and decreased quality of life. To the best of our knowledge, this is the first study to investigate the associations of GNRI and baPWV with small fiber neuropathy in HD patients. Nonetheless, there are several limitations in the present study. First, the number of study patients is relatively small. Second, this study was cross-sectional in design; therefore, the causal relationship and long-term clinical outcomes could not be confirmed. Further prospective studies and more participants are warranted to examine the impacts of nutrition and vascular factors as well as their involving roles in small fiber neuropathy. Third, confirmation of small fiber neuropathy using skin biopsy was lacking. Although skin biopsy is able to detect altered interepidermal nerve fiber density, it remains an invasive approach. 5. Conclusion {#sec5} ============= Our results demonstrated that small fiber neuropathy was associated with a lower GNRI and baPWV \> 1400 cm/s. Furthermore, older age, DM, and a higher level of Ca × P product were associated with an abnormal QST threshold. Physicians should devote more attention toward maintenance HD patients with malnutrition risk and arterial stiffness to early diagnose small fiber neuropathy, prevent serious neurological complications, and improve quality of life. The research presented in this article is supported by the grants from Kaohsiung Municipal Siaogang Hospital (kmhk-102-003), Kaohsiung Medical University, Kaohsiung, Taiwan. Data Availability ================= The data supporting the findings of the present study are available within the article or are available from the corresponding author upon reasonable request. Conflicts of Interest ===================== The authors declare no conflicts of interest. Authors\' Contributions ======================= The research idea and study design were from M.C.K., J.C.H., and J.M.C.; data acquisition was performed by P.Y.W., H.C.M., J.C.H., and S.C.C.; data analysis/interpretation was performed by M.C.K., J.C.H., P.Y.W., S.C.C., and Y.W.C.; statistical analysis was performed by J.C.H., S.C.C., J.M.C., and H.C.C.; supervision or mentorship was done by J.C.H., J.M.C., and H.C.C. All authors contributed important intellectual content during manuscript drafting or revision and approved the final version of the manuscript. {#fig1} ###### Comparison of cold and warm threshold and baPWV between diabetic and nondiabetic patients. Variables Diabetic (*n* = 72) Nondiabetic (*n* = 80) *p* ------------------------------------------ --------------------- ------------------------ --------- Age (year) 62.9 ± 9.1 58.8 ± 11.7 0.018 Male gender (%) 55.6 47.5 0.321 Duration of dialysis (year) 3.5 (1.2--7.6) 9.1 (4.9--13.1) \<0.001 baPWV (cm/s) 2091 ± 545 1774 ± 461 \<0.001 Body mass index (kg/m^2^) 24.7 ± 3.2 23.6 ± 4.3 0.074 GNRI 104.2 ± 7.4 102.7 ± 10.2 0.289 Cold threshold of the left hand (°C) 26.4 ± 3.2 27.4 ± 3.2 0.043 Warm threshold of the left hand (°C) 41.6 ± 4.9 38.7 ± 3.5 \<0.001 Cold threshold of the right hand (°C) 25.0 ± 3.2 26.7 ± 3.0 0.001 Warm threshold of the right hand (°C) 43.9 ± 4.5 40.8 ± 4.2 \<0.001 Cold threshold of the left foot (°C) 21.8 ± 2.7 23.9 ± 3.0 \<0.001 Warm threshold of the left foot (°C) 47.8 ± 3.2 45.3 ± 2.7 \<0.001 Cold threshold of the right foot (°C) 21.8 ± 2.7 23.7 ± 3.2 \<0.001 Warm threshold of the right foot (°C) 47.8 ± 2.6 45.8 ± 2.8 \<0.001 Abnormal cold threshold of the hands (%) 81.9 65.0 0.019 Abnormal warm threshold of the hands (%) 95.8 82.5 0.009 Abnormal cold threshold of the feet (%) 93.1 77.5 0.008 Abnormal warm threshold of the feet (%) 94.4 85.0 0.058 Abbreviations: baPWV: brachial-ankle pulse wave velocity; GNRI: geriatric nutritional risk index. ###### Comparison of baseline characteristics between patients with normal or abnormal QST threshold of the hands. Parameters Cold threshold Warm threshold ------------------------------- ----------------- ----------------- --------- ------------------ ----------------- ------- Age (year) 54.8 ± 10.8 62.9 ± 9.8 \<0.001 53.7 ± 8.3 61.6 ± 10.7 0.004 Male gender (%) 48.8 52.3 0.704 23.5 54.8 0.015 Duration of dialysis (year) 8.8 (4.7--13.3) 6.1 (2.0--10.3) 0.023 10.7 (5.1--13.3) 6.3 (2.3--10.4) 0.066 Smoking history (%) 34.1 39.6 0.536 17.6 40.7 0.109 Diabetes mellitus (%) 31.7 53.2 0.019 17.6 51.1 0.009 Coronary artery disease (%) 2.4 11.7 0.114 0 10.4 0.369 baPWV (cm/s) 1766 ± 406 1982 ± 553 0.027 1672 ± 367 1956 ± 535 0.041 Body mass index (kg/m^2^) 24.4 ± 4.6 23.9 ± 3.5 0.495 23.6 ± 3.6 24.1 ± 3.9 0.603 GNRI 104.9 ± 11.0 102.8 ± 8.1 0.217 101.9 ± 9.9 103.6 ± 8.9 0.480 Laboratory parameters  Albumin (g/dL) 3.9 ± 0.4 3.9 ± 0.3 0.155 3.8 ± 0.4 3.9 ± 0.3 0.552  Triglycerides (mg/dL) 125 (94--273) 139 (91--217) 0.340 149 (90--249) 131 (91--219) 0.707  Total cholesterol (mg/dL) 185.3 ± 34.6 179.7 ± 41.1 0.439 190.9 ± 40.0 180.0 ± 39.4 0.284  Hemoglobin (g/dL) 10.7 ± 1.0 10.6 ± 1.3 0.828 10.5 ± 1.3 10.7 ± 1.2 0.685  Total calcium (mg/dL) 9.5 ± 1.1 9.3 ± 1.0 0.362 9.0 ± 1.2 9.4 ± 1.0 0.241  Phosphorus (mg/dL) 4.6 ± 1.2 4.5 ± 1.0 0.846 4.7 ± 1.6 4.5 ± 1.0 0.490  Ca × P product (mg^2^/dL^2^) 43.2 ± 11.7 42.2 ± 11.1 0.624 43.3 ± 15.4 42.4 ± 10.7 0.755 Kt/V (Daugirdas) 1.6 ± 0.3 1.6 ± 0.3 0.695 1.7 ± 0.3 1.6 ± 0.3 0.070 Abbreviations: baPWV: brachial-ankle pulse wave velocity; GNRI: geriatric nutritional risk index; Ca × P product: calcium-phosphorus product. ###### Comparison of baseline characteristics between patients with normal or abnormal QST threshold of the feet. Parameters Cold threshold Warm threshold ------------------------------- ----------------- ----------------- ------- ----------------- ----------------- ------- Age (year) 55.6 ± 10.5 61.7 ± 10.5 0.011 56.6 ± 12.7 61.2 ± 10.4 0.100 Male gender (%) 65.2 48.8 0.148 23.5 54.8 0.242 Duration of dialysis (year) 9.0 (3.4--11.8) 6.5 (2.3--10.8) 0.614 9.9 (2.9--12.6) 6.4 (2.4--10.6) 0.218 Smoking history (%) 47.8 36.4 0.300 31.3 39.0 0.548 Diabetes mellitus (%) 21.7 51.9 0.008 25.0 50.0 0.068 Coronary artery disease (%) 0 10.9 0.130 0 10.3 0.364 baPWV (cm/s) 1691 ± 352 1966 ± 541 0.023 1876 ± 711 1931 ± 502 0.694 Body mass index (kg/m^2^) 25.9 ± 6.1 23.7 ± 3.2 0.011 24.3 ± 3.5 24.0 ± 3.9 0.797 GNRI 107.9 ± 12.6 102.6 ± 7.9 0.008 103.7 ± 9.3 103.3 ± 9.0 0.873 Laboratory parameters  Albumin (g/dL) 3.9 ± 0.3 3.9 ± 0.3 0.242 3.9 ± 0.3 3.9 ± 0.3 0.919  Triglycerides (mg/dL) 125 (87--304) 139 (92--219) 0.309 108.5 (88--229) 139 (91.3--219) 0.928  Total cholesterol (mg/dL) 182.4 ± 27.3 181.1 ± 41.3 0.885 171.7 ± 34.8 182.4 ± 39.9 0.307  Hemoglobin (g/dL) 10.9 ± 1.2 10.6 ± 1.2 0.208 10.9 ± 1.5 10.6 ± 1.2 0.435  Total calcium (mg/dL) 9.3 ± 1.1 9.3 ± 1.0 0.844 9.1 ± 1.1 9.3 ± 1.0 0.371  Phosphorus (mg/dL) 4.7 ± 1.4 4.5 ± 1.0 0.625 4.1 ± 1.0 4.6 ± 1.1 0.058  Ca × P product (mg^2^/dL^2^) 43.1 ± 13.0 42.3 ± 11.0 0.753 37.0 ± 9.6 43.1 ± 11.3 0.041 Kt/V (Daugirdas) 1.6 ± 0.3 1.6 ± 0.3 0.557 1.6 ± 0.3 1.6 ± 0.3 0.857 Abbreviations: baPWV: brachial-ankle pulse wave velocity; GNRI: geriatric nutritional risk index; Ca × P product: calcium-phosphorus product. ###### Determinants of abnormal QST threshold using multivariate forward logistic regression analysis. Parameters Multivariate (forward) -------------------------------------- ------------------------ --------- Abnormal cold threshold of the hands  Age (per 1 year) 1.083 (1.042--1.126) \<0.001 Abnormal warm threshold of the hands  Age (per 1 year) 1.081 (1.026--1.139) 0.003  Male gender (*vs.* female) 4.450 (1.250--15.836) 0.021 Abnormal cold threshold of the feet  Diabetes mellitus 3.996 (1.351--11.819) 0.012  GNRI (per 1 unit) 0.935 (0.887--0.985) 0.012 Abnormal warm threshold of the feet  baPWV \> 1400 cm/s 5.479 (1.312--22.870) 0.020  Ca × P product (per 1 mg^2^/dL^2^) 1.071 (1.013--1.132) 0.015 Values expressed as odds ratio (OR) and 95% confidence interval (CI). Adjusted for age, gender, duration of dialysis, smoking history, diabetes mellitus, baPWV \> 1400 cm/s, GNRI, hemoglobin, Ca × P product, and Kt/V. [^1]: Academic Editor: Alexandra Scholze

Introduction {#Sec1} ============ Atomic nuclei---the finite quantum many-body systems consisting of protons and neutrons (known collectively as nucleons)---exhibit shell structure, in analogy to the electronic shell structure of atoms. Atoms with filled electron shells---known as the noble gases---are particularly stable chemically. The filling of the nuclear shells, on the other hand, leads to the magic-number nuclei. The nuclear magic numbers, as we know in stable and naturally-occurring nuclei, consist of two different series of numbers. The first series---2, 8, 20---is attributed to the harmonic-oscillator (HO) potential, while the second one---28, 50, 82 and 126---is due to the spin--orbit (SO) coupling force (see Fig. [1](#Fig1){ref-type="fig"}). It was the introduction of this phenomenological SO force---a force that depends on the intrinsic spin of a nucleon and its orbital angular momentum, and the so-called *j*--*j* coupling scheme that helped explain^[@CR1],[@CR2]^ completely the magic numbers, and won Goeppert-Mayer and Jensen the Nobel Prize. However, the microscopic origins of the SO splitting have remained unresolved due to the difficulty to describe the structure of atomic nuclei from ab initio nuclear theories^[@CR3]--[@CR5]^ with two- (NN) and three-nucleon forces (3NFs). Although the theoretical study^[@CR6]^ of the SO splitting of the 1*p*~1/2~ and 1*p*~3/2~ single-particle states in ^15^N has suggested possible roles of two-body SO and tensor forces, as well as three-body forces, the discovery of a prevalent SO-type magic number 6 is expected to offer unprecedented opportunities to understand its origins.Fig. 1Nuclear shell structure. The left diagram is the shell structure for a harmonic-oscillator potential plus a small orbital angular momentum (*l*^2^) term. The right diagram shows the splitting of the single-particle orbitals by an additional spin--orbit coupling force In her Nobel lecture, Goeppert-Mayer had mentioned the magic numbers 6 and 14---which she described as hardly noticeable---but surmised that the energy gap between the 1*p*~1/2~ and 1*p*~3/2~ orbitals due to the SO force is fairly small^[@CR7]^. That the *j*--*j* coupling scheme appears to fail in the *p*-shell light nuclei was discussed and attributed to their tendency to form clusters of nucleons^[@CR8]^. Experimental and theoretical studies in recent decades, however, have hinted at the possible existence of the magic number 6 in some semimagic unstable nuclei, each of which has a HO-type magic number of the opposite type of nucleons. For instance, possible subshell closures have been suggested in ^8^He^[@CR9]--[@CR11]^, ^14^O^[@CR12]^ and ^14^C^[@CR12]--[@CR14]^. Whether the subshell closure at the proton number *Z* = 6 is predominantly driven by the shell closure at the neutron number *N* = 8 in ^14^C or persists in other carbon isotopes is of fundamental importance. The isotopic chain of carbon---with six protons and consisting of thirteen particle-bound nuclei---provides an important platform to study the SO splitting of the 1*p*~1/2~ and 1*p*~3/2~ orbitals. Like other lighter isotopes, the isotopes of carbon are known to exhibit both clustering^[@CR15]--[@CR17]^ and single-particle behaviours. Although the second excited *J*^*π*^ = 0^+^ state in ^12^C---the famous Hoyle state and important doorway state that helps produce ^12^C in stars---is well understood as a triple-alpha state, it seems that the effect of the alpha-cluster-breaking 1*p*~3/2~ subshell closure is important to reproduce the ground-state binding energy^[@CR18]^. For even--even neutron-rich carbon isotopes, theoretical calculations using the anti-symmetrized molecular dynamics (AMD)^[@CR19]^, shell model^[@CR20],[@CR21]^, as well as the ab initio no-core shell model calculation^[@CR22]^ with NN + 3NFs have predicted near-constant proton distributions, a widening gap between proton 1*p*~1/2~ and 1*p*~3/2~ single-particle orbits, and a remarkably low proton occupancy in the 1*p*~1/2~ orbit, respectively. Gupta et al. ^[@CR23]^, on the other hand, have suggested the possible existence of closed-shell core nuclei in ^15,17,19^C on the basis of potential energy surfaces employing the cluster-core model. Experimentally, small *B*(*E*2) values comparable to that of ^16^O were reported from the lifetime measurements of the first excited 2^+^ $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left( {2_1^ + } \right)$$\end{document}$ states in ^16,18^C^[@CR24]--[@CR26]^. The small *B*(*E*2) values indicate small proton contributions to the transitions, and together with the theoretical predictions may imply the existence of a proton-subshell closure. Although still not well established, the size of a nucleus, which can be defined as the root-mean-square (rms) radius of its nucleon distribution, is expected to provide important insights on the evolution of the magic numbers. Recently, an unexpectedly large proton rms radius (denoted simply as proton radius hereafter) was reported^[@CR27]^, and suggested as a possible counter-evidence for the double shell closure in ^52^Ca^[@CR28]^. Attempts to identify any emergence of non-traditional magic numbers based on the analysis of the systematics of the experimental proton radii have been reported^[@CR12],[@CR29]^. For the 4 \< *Z* \< 10 region, the lack of experimental data on the proton radii of neutron-rich nuclei due to the experimental and theoretical limitations of the isotope-shift method has hindered systematic analysis of the radii behaviour. Such systematic analysis has become possible very recently owing to the development of an alternative method to extract the proton radii of neutron-rich nuclei from the charge-changing cross-section measurements. Here we present experimental evidence for a prevalent *Z* = 6 subshell closure in ^13--20^C, based on a systematic study of proton radii obtained from our recent experiments as well as the existing nuclear charge radii^[@CR12]^, electric quadrupole transition rates *B*(*E*2) between the $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2_1^ +$$\end{document}$ and ground $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left( {0_{{\mathrm{gs}}}^ + } \right)$$\end{document}$ states of even--even nuclei^[@CR30]^, and atomic-mass data^[@CR31]^. We show, by performing coupled-cluster calculations, that the observations are supported by the ab initio nuclear model that employs the nuclear forces derived from the effective field theory of the quantum chromodynamics. Results {#Sec2} ======= Experimental details {#Sec3} -------------------- The charge-changing cross section (denoted as *σ*~CC~) of a projectile nucleus on a nuclear/proton target is defined as the total cross section of all processes that change the proton number of the projectile nucleus. Applying this method, we have determined the proton radii of ^14^Be^[@CR32]^, ^12--17^B^[@CR33]^ and ^12--19^C^[@CR34],[@CR35]^ from the *σ*~CC~ measurements at GSI, Darmstadt, using secondary beams at around 900 MeV per nucleon. In addition, we have also measured *σ*~CC~'s for ^12--18^C on a ^12^C target with secondary beams at around 45 MeV per nucleon at the exotic nuclei (EN) beam line^[@CR36]^ at RCNP, Osaka University. To extract proton radii from both low-energy data and high-energy data, we have devised a global parameter set for use in the Glauber-model calculations. The Glauber model thus formulated was applied to the *σ*~CC~ data at both energies to determine the proton radii. A summary on the experiment at RCNP and the Glauber-model analysis is given in Methods. More details can be found in ref. ^[@CR37]^ Charge-changing cross sections and proton radii {#Sec4} ----------------------------------------------- For simplicity, we show only the results for ^17--19^C in Table [1](#Tab1){ref-type="table"}; for results on ^12--16^C, see ref. ^[@CR37]^. *R*~p~'s are the proton radii extracted using the Glauber model formulated in ref. ^[@CR37]^ The values for ^17,18^C are the weighted mean extracted using *σ*~CC~'s at the two energies, while the one for ^19^C was extracted using the high-energy data. In determining the proton radii, we have assumed harmonic-oscillator (HO)-type distributions for the protons in the Glauber calculations. The uncertainties shown in the brackets include the statistical uncertainties, the experimental systematic uncertainties, and the uncertainties attributed to the choice of functional shapes, that is HO or Woods--Saxon, assumed in the calculations.Table 1Cross sections and proton radii*E*~CC~ (*A* MeV)*σ*~CC~ (mb)*E*~CC~ (*A* MeV)*σ*~CC~ (mb)*R*~p~ (fm)^17^C46.31000(16)979754(7)2.43(4)^18^C42.81023(31)895747(7)2.42(5)^19^C895749(9)2.43(4)Measured charge-changing cross sections (*σ*~CC~) for ^17--19^C and the corresponding secondary-beam energies (*E*~CC~). The subscript CC denotes the charge-changing reaction. The data in the fourth and fifth columns are from ref. ^[@CR34]^ *R*~p~'s in the sixth column are the proton radii extracted from the *σ*~CC~'s in the third and fifth columns To get an overview of the isotopic dependence, we compare the proton radii of the carbon isotopes with those of the neighbouring beryllium, boron and oxygen isotopes. Figure [2](#Fig2){ref-type="fig"} shows the proton radii for carbon, beryllium, boron and oxygen isotopes. The red-filled and black-filled circles are the data for ^12--19^C, beryllium and boron isotopes extracted in this and our previous work^[@CR32]--[@CR34],[@CR37]^. For comparison, the proton radii determined with the electron-scattering and isotope-shift methods^[@CR12]^ are also shown in Fig. [2](#Fig2){ref-type="fig"} (open diamonds). Our *R*~p~'s for ^12--14^C are in good agreement with the electron-scattering data. In addition, we performed theoretical calculations. The small symbols connected with dashed and dotted lines shown in the figure are the results from the AMD^[@CR19]^ and relativistic mean field (RMF)^[@CR38]^ calculations, respectively. The blue-solid and blue-dash-dotted lines are the results (taken from ref.^[@CR34]^) of the ab initio coupled-cluster (CC) calculations with NNLO~sat~^[@CR39]^ and the NN-only interaction NNLO~opt~^[@CR40]^, respectively. The AMD calculations reproduce the trends of all isotopes qualitatively but overestimate the proton radii for carbon and beryllium isotopes. The RMF calculations, on the other hand, reproduce most of the proton radii of carbon and oxygen isotopes but underestimate the one of ^12^C. Overall, the CC calculations with the NNLO~sat~ interactions reproduce the proton radii for ^13--18^C very well. The calculations without 3NFs underestimate the radii by about 10%, thus suggesting the importance of 3NFs.Fig. 2Proton radii. Results are shown for carbon, beryllium, boron and oxygen isotopes. The red-filled and black-filled circles are, respectively, the proton radii from this and our recent work^[@CR32]--[@CR34],[@CR37]^. The open diamonds are the data from electron-scattering and isotope-shift methods^[@CR12]^. The error bars for the red-filled circles include the statistical and experimental systematic uncertainties, as well as the uncertainties due to the choice of density distributions. The error bars for other experimental data are taken from the literature. The small symbols connected with dashed and dotted lines are the predictions from the AMD^[@CR19]^ and RMF^[@CR38]^ models, respectively. The small blue symbols with solid and dash-dotted lines are the results from the ab initio coupled-cluster calculations with NNLO~sat~^[@CR39]^ and the NN-only interaction NNLO~opt~^[@CR40]^ It is interesting to note that *R*~p~'s are almost constant throughout the isotopic chain from ^12^C to ^19^C, fluctuating by less than 5%. Whereas this trend is similar to the one observed/predicted in the proton-closed shell oxygen isotopes, it is in contrast to those in the beryllium and boron isotopic chains, where the proton radii change by as much as 10% (for berylliums) or more (for borons). It is also worth noting that most theoretical calculations shown predict almost constant proton radii in carbon and oxygen isotopes. The large fluctuations observed in Be and B isotopes can be attributed to the development of cluster structures, whereas the almost constant *R*~p~'s for ^12--19^C observed in the present work may indicate an inert proton core, that is 1*p*~3/2~ proton-subshell closure. Systematics of nuclear observables {#Sec5} ---------------------------------- Examining the *Z* dependence of the proton rms radii along the *N* = 8 isotonic chain, Angeli et al. have pointed out^[@CR12],[@CR29]^ a characteristic change of slope (existence of a kink), a feature closely associated with shell closure, at *Z* = 6. Here, by combining our data with the recent data^[@CR32]--[@CR34],[@CR37]^, as well as the data from ref. ^[@CR12]^, we plot the experimental *R*~p~'s against proton number. To eliminate the smooth mass number dependence of the proton rms radii, we normalised all *R*~p~'s by the following mass-dependent rms radii^[@CR41]^:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_{\mathrm{p}}^{{\mathrm{cal}}} = \sqrt {3{\mathrm{/}}5} \left( {1.15 + 1.80A^{ - 2/3} - 1.20A^{ - 4/3}} \right)A^{1/3}{\mathrm{fm}}.$$\end{document}$$ Figure [3a](#Fig3){ref-type="fig"} shows the evolution of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_{\mathrm{p}}{\mathrm{/}}R_{\mathrm{p}}^{{\mathrm{cal}}}$$\end{document}$ with proton number up to *Z* = 22 and for isotonic chains up to *N* = 28. Each isotonic chain is connected by a solid line. For simplicity, only the symbols for *N* = 3--16 are shown in the legend in Fig. [3c](#Fig3){ref-type="fig"}; the data for *N* = 6--13 isotones are displayed in colours for clarity. For nuclides with more than one experimental value, we have adopted the weighted mean values. The discontinuities observed at *Z* = 10 and *Z* = 18 are due to the lack of experimental data in the proton-rich region. Note the increase/change in the slope at the traditional magic numbers *Z* = 8 and 20. Marked kinks, similar to those observed at *Z* = 20, 28, 50 and 82^[@CR29]^, are observed at *Z* = 6 for isotonic chains from *N* = 7 to *N* = 13, indicating a possible major structural change, for example emergence of a subshell closure, at *Z* = 6.Fig. 3Systematics of nuclear observables. Evolution of **a** $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_{\mathrm{p}}{\mathrm{/}}R_{\mathrm{p}}^{{\mathrm{cal}}}$$\end{document}$, **b** *B*(*E*2) and **c** *e*~p~ − *e*~p+1~ with proton number up to *Z* = 22 and for isotonic chains up to *N* = 28. Vertical dotted and thin-dashed lines denote positions of the traditional proton magic numbers and *Z* = 6, respectively. The error bars for data in **a** are evaluated using the errors, while the ones in **b** are the errors from the literature. For clarity, the error bars in **c**, some of which are slightly larger than the symbols, are not shown. **d** Two-dimensional lego plot of **c** The possible emergence of a proton-subshell closure at *Z* = 6 in neutron-rich even--even carbon isotopes is also supported by the small *B*(*E*2) values observed in ^14--20^C^[@CR25],[@CR26],[@CR30]^. Figure [3b](#Fig3){ref-type="fig"} shows the systematics of *B*(*E*2) values in Weisskopf unit (W.u.) for even--even nuclei up to *Z* = 22. All data are available in ref. ^[@CR30]^ Nuclei with shell closures manifest themselves as minima. Besides the traditional magic number *Z* = 8, clear minima with *B*(*E*2) values smaller than 3 W.u. are observed at *Z* = 6 for *N* = 8, 10, 12 and 14 isotones. To further examine the possible subshell closure at *Z* = 6, we consider the second derivative of binding energies defined as follows^[@CR42]^:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{\Delta }}_{\mathrm{p}}^{(3)}(N,Z) \equiv ( - 1)^Z\left[ {S_{\mathrm{p}}(N,Z) - S_{\mathrm{p}}(N,Z + 1)} \right]{\mathrm{/}}2,$$\end{document}$$where *S*~p~(*N*, *Z*) is the one-proton separation energy. In the absence of many-body correlations such as pairings, *S*~p~(*N*, *Z*) resembles the single-particle energy, and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2{\mathrm{\Delta }}_{\mathrm{p}}^{(3)}(N,Z)$$\end{document}$ yields the proton single-particle energy-level spacing or shell gap between the last occupied (*e*~p~) and first unoccupied proton orbitals (*e*~p+1~) in the nucleus with *Z* protons (and *N* neutrons). To eliminate the effect of proton--proton (*p*--*p*) pairing, we subtract out the *p*--*p* pairing energies using the empirical formula: Δ~p~ = 12*A*^--1/2^ MeV. Figure [3c](#Fig3){ref-type="fig"} shows the systematics of *e*~p~ − *e*~p+1~ (=$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2{\mathrm{\Delta }}_{\mathrm{p}}^{(3)}$$\end{document}$ (*N*, *Z*) − 2Δ~p~) for even-*Z* nuclides. All data were evaluated from the experimental binding energies^[@CR31]^. Here, we have omitted odd-*Z* nuclides to avoid odd--even staggering effects. The cusps observed at *Z* = *N* for all isotonic chains are due to the Wigner effect^[@CR43]^. Apart from the *Z* = *N* nuclides, sizable gaps (\>2 MeV) are also observed at *Z* = 6 for *N* = 7--14, and at *Z* = 8 for *N* = 8--10 and 12--16. For clarity, we show the corresponding two-dimensional lego plot in Fig. [3d](#Fig3){ref-type="fig"}. By requiring a magic nucleus to fulfil at least two signatures in Fig. [3a--c](#Fig3){ref-type="fig"}, we conclude that we have observed a prominent proton-subshell closure at *Z* = 6 in ^13--20^C. Although the empirical $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2{\mathrm{\Delta }}_{\mathrm{p}}^{(3)}$$\end{document}$ for ^12^C is large (\~14 MeV), applying the prescription from ref. ^[@CR44]^, we obtain about 10.7 MeV for the total *p*--*p* and *p*--*n* pairing energy. This estimated large pairing energy indicates possible significant many-body correlations such as cluster correlations. We note that ^12^C is known to be an intermediate-coupling nucleus lying in the middle of the *j*--*j* coupling and *L*--*S* coupling limits^[@CR45]^. The core is largely broken with only about 40% of the nominal (1*p*~3/2~)^8^ closed-shell component, and the occupation number of nucleons in the 1p~1/2~ shell is as much as 1.5 from shell model calculations using the Cohen--Kurath interactions^[@CR46]^. Discussion {#Sec6} ========== It is surprising that the systematics of the proton radii, *B*(*E*2) values and the empirical proton-subshell gaps for most of the carbon isotopes are comparable to those for proton-closed shell oxygen isotopes. To understand the observed ground-state properties, that are the proton radii and subshell gap of the carbon isotopes, we performed ab initio CC calculations on ^14,15^C using various state-of-the-art chiral interactions. We employed the CC method in the singles-and-doubles approximation with perturbative triples corrections \[Λ-CCSD(T)\]^[@CR47]^ to compute the ground-state binding energies and proton radii for the closed-(sub)shell ^14^C. To compute ^15^C (1/2^+^), we used the particle-attached equation-of-motion CC (EOM-CC) method^[@CR48]^, and included up to three-particle-two-hole (3p--2h) and two-particle-three-hole (2p--3h) corrections as recently developed in ref. ^[@CR49]^ Figure [4](#Fig4){ref-type="fig"} shows the binding energies as functions of the proton radii for (a) ^14^C and (b) ^15^C. The coloured bands are the experimental values; the binding energies (red-horizontal lines) are taken from ref. ^[@CR31]^, while proton radii are from ref. ^[@CR37]^ (orange bands) and the electron-scattering data^[@CR12]^ (green band). The filled black symbols are CC predictions with the NN + 3NF chiral interactions from ref. ^[@CR50]^ labelled 2.0/2.0 (EM)(black square), 2.0/2.0 (PWA)(black downward-pointing triangle), 1.8/2.0 (EM)(black circle), 2.2/2.0 (EM)(black diamond), 2.8/2.0 (EM)(black triangle), and NNLO~sat~^[@CR39]^ (black star). Here, the NN interactions are the next-to-next-to-next-to-leading order (N^3^LO) chiral interaction from ref. ^[@CR51]^, evolved to lower cutoffs (1.8/2.0/2.2/2.8 fm^--1^) via the similarity-renormalisation-group (SRG) method^[@CR52]^, while the 3NF is taken at NNLO with a cutoff of 2.0 fm^--1^ and adjusted to the triton binding energy and ^4^He charge radius. The error bars are the estimated theoretical uncertainties due to truncations of the employed method and model space. For details on the CC method and error estimation, see refs. ^[@CR5],[@CR53]^. Note that the error bars for the binding energies are smaller than the symbols. Depending on the NN cutoff, the calculated binding energy correlates strongly with the calculated proton radius. In addition, we performed the CC calculations with chiral effective interactions without 3NFs, that are the NN-only EM interactions with NN cutoffs at 1.8 (white circle), 2.0 (white square), 2.2 (white diamond) and 2.8 fm^--1^ (white triangle), and the NN-only part of the chiral interaction NNLO~sat~ (white downward-pointing triangle). Overall, most calculations that include 3NFs reproduce the experimental proton radii well. For the binding energies, the calculations with the EM(1.8/2.0) and NNLO~sat~ interactions reproduce both data very well. It is important to note that without 3NFs the calculated proton radii are about 9--15% (18%) smaller, while the ground states are overbound by as much as about 24% (26%) for ^14^C (^15^C). These results highlight the importance of comparing both experimental observables to examine the employed interactions.Fig. 4Binding energies versus proton radii. The results for **a** ^14^C and **b** ^15^C are shown. The coloured bands and red-horizontal lines are the experimental values. The green band represents the proton radius from the electron scattering. The filled black symbols are the CC predictions with SRG-evolved NN + 3NF chiral effective interactions at different NN/3NF cutoffs and NNLO~sat~, whereas the open symbols are the predictions with the NN-only EM and NNLO~sat~ interactions. The error bars are the estimated theoretical uncertainties due to truncations of the employed method and model space^[@CR53]^. See text for details The importance of the Fujita--Miyazawa type^[@CR54]^ or the chiral NNLO 3NFs^[@CR55],[@CR56]^ in reproducing the binding energies and the drip lines of nitrogen and oxygen isotopes have been suggested in recent theoretical studies. Here, to shed light on the role of 3NFs on the observed subshell gap, that is the SO splitting in the carbon isotopes, we investigate the evolution of one-proton separation energies for carbon and oxygen isotopes. In Fig. [5](#Fig5){ref-type="fig"}, the horizontal bars represent the experimental one-proton addition ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\epsilon _{ + {\mathrm{p}}}$$\end{document}$) and removal ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\epsilon _{ - {\mathrm{p}}}$$\end{document}$) energies for (a) carbon and (b) oxygen isotopes deduced from one-proton separation energies, that are binding energies of boron to fluorine isotopes, and the excitation energies of the lowest 3/2^−^ states in the odd--even nitrogen isotopes. The dotted bars indicate the adopted values for the observed excited states in ^19,21^N, which have been tentatively assigned as 3/2^−[@CR57]^. Other experimental data are taken from refs. ^[@CR31],[@CR58],[@CR59]^. For comparison, we show the one-proton addition and removal energies (blue symbols) calculated using the shell model with the YSOX interaction^[@CR60]^, which was constructed from a monopole-based universal interaction (*V*~MU~). Because the phenomenological effective two-body interactions were determined by fitting experimental data, they are expected to partially include the three-nucleon effect and thus can reproduce relatively well the ground-state energies, drip lines, energy levels, as well as the electric and spin properties of carbon and oxygen isotopes. As shown in Fig. [5](#Fig5){ref-type="fig"}, the shell model calculations reproduce the $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\epsilon _{ \pm {\mathrm{p}}}$$\end{document}$'s for carbon and oxygen isotopes very well.Fig. 5Shell evolution. Empirical one-proton addition ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\epsilon _{ + {\mathrm{p}}}$$\end{document}$) and removal ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\epsilon _{ - {\mathrm{p}}}$$\end{document}$) energies (horizontal bars) for **a** carbon, and **b** oxygen isotopes deduced from one-proton separation energies and the excitation energies of the lowest 3/2^−^ states in the odd--even nitrogen isotopes. The dotted bars indicate the adopted values for the observed excited states in ^19,21^N, which have been tentatively assigned as 3/2^−[@CR57]^. Other experimental data are taken from refs. ^[@CR31],[@CR58],[@CR59]^. The blue symbols are the shell model calculations using the YSOX interactions^[@CR60]^. Results of the CC calculations with and without 3NFs are shown by the red-solid and red-dashed lines, respectively As mentioned earlier, in the absence of many-body correlations, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\epsilon _{ \pm {\mathrm{p}}}$$\end{document}$ resemble the proton single-particle energies, and the gap between them can be taken as the (sub)shell gap. In the following, we consider ^14,15^C and the closed-shell ^14,16,22^O isotopes in more detail. We computed their ground-state binding energies and those of their neighbouring isotones ^13,14^B, ^13,15,16,21^N and ^15,17,23^F. We applied the Λ-CCSD(T) and the particle-attached/removed EOM-CC methods to compute the binding energies for the closed-(sub)shell and open-shell nuclei, respectively. The ground-state binding energies of ^14^B (2^−^) and ^16^N (2^−^) were computed using the EOM-CC method with reference to ^14^C and ^16^O employing the charge-exchange EOM-CC technique^[@CR61]^. Results of the CC calculations on ^14,15^C and ^14,16,22^O with and without 3NFs are shown by the red-solid and red-dashed lines, respectively. Here, we have opted for EM(1.8/2.0 fm^−1^), which yield the smallest chi-square value for the calculated and experimental binding energies considered, as the NN+3NF interactions. For the NN-only interaction, we show the calculations with EM(2.8 fm^−1^). The calculated $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\epsilon _{ - {\mathrm{p}}}$$\end{document}$(3/2^−^) for ^22^O with EM(2.8 fm^−1^) (and other NN-only interactions) has an unrealistic positive value, and is thus omitted. We found the norms of the wave functions for the one-particle (1p) 1/2^−^ and one-hole (1h) 3/2^−^ states of ^14^C, and the two corresponding 1p and 1h states of ^15^C (2^−^ states in ^14^B and ^16^N) to be almost 90%. The calculations suggest that these states can be accurately interpreted by having dominant single-particle structure, and that the gaps between these 1p--1h states resemble the proton-subshell gaps. It is obvious from the figure that the calculations with the NN + 3NF interactions reproduce the experimental $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\epsilon _{ \pm {\mathrm{p}}}$$\end{document}$ for ^14,15^C and ^14,16,22^O very well. Overall, the calculations without 3NFs predict overbound proton states, and in the case of ^14,15^C, much reduced subshell gaps. These results suggest that ^14^C is a doubly-magic nucleus, and ^15^C a proton-closed shell nucleus. Our results show that the phenomenon of large spin--orbit splitting is indeed universal in atomic nuclei, and the magic number 6 is as prominent as other classical SO-originated magic numbers such as 28. Although we have shown only results for ^14,15^C, we expect further systematic and detailed theoretical analyses on other carbon isotopes, in particular ab initio calculations using realistic and/or chiral interactions, to provide quantitative insights on the neutron-number dependence of the SO splitting and its origin. It will be interesting to understand also the origins of the diverse structures in ^12^C. Finally, we would like to point out that an inert ^14^C core, built on the *N* = 8 closed shell, has been postulated to explain several experimental data for ^15,16^C. For instance, a ^14^C + *n* model was successfully applied^[@CR62]^ to explain the consistency between the measured *g*-factor and the single-particle-model prediction (the Schmidt value) of the excited 5/2^+^ state in ^15^C. Wiedeking et al. ^[@CR25]^, on the other hand, have explained the small *B*(*E*2) value in ^16^C assuming a ^14^C + *n* + *n* model in the shell-model calculation. In terms of spectroscopy studies using transfer reactions, the results from the ^14^C(*d*, *p*)^15^C^[@CR63]^ and ^15^C(*d*, *p*)^16^C^[@CR64]^ measurements are also consistent with the picture of a stable ^14^C core. On the proton side, a possible consolidation of the 1*p*~3/2~ proton-subshell closure when moving from ^12^C to ^14^C was reported decades ago from the measurements of the proton pick-up (*d*,^3^He) reaction on ^12,13,14^C targets^[@CR65]^, consistent with shell model predictions. An attempt to study the ground-state configurations with protons outside the 1*p*~3/2~ orbital in ^14,15^C has also been reported^[@CR66]^ very recently. To further investigate the proton-subshell closure in the neutron-rich carbon isotopes, more experiments using one-proton transfer and/or knockout reactions induced by radioactive boron, carbon and nitrogen beams at facilities such as ATLAS, FAIR, FRIB, RCNP, RIBF and SPIRAL2 are anticipated. Methods {#Sec7} ======= Experiment and data analysis {#Sec8} ---------------------------- Secondary ^12--18^C beams were produced, in separate runs, by projectile fragmentation of ^22^Ne^10+^ ions at 80 MeV per nucleon incident on a ^9^Be (production) target with thickness ranging from 1.0 to 5.0 mm. The carbon beam of interest was selected by setting the appropriate particle magnetic rigidities using the RCNP EN fragment separator. The carbon beam thus produced was transported to the experimental area, and directed onto a 450-mg cm^−2^-thick natural carbon (reaction) target. The incident beam was identified by the measurements of energy loss in a 320-mm-thick silicon detector, and the time of flight (TOF) between the production and reaction targets. The TOF was determined from the timing information obtained with a 100-μm-thick plastic scintillation detector placed before the reaction target and the radio-frequency signal from the accelerator. Particles exiting the reaction target were detected by a multisampling ionisation chamber (MUSIC), consisting of eight anodes and nine cathodes, before being stopped in a 7-cm-thick NaI(Tl) scintillation detector. The outgoing particles were identified using the energy-loss and total-energy information obtained with the MUSIC and NaI(Tl) detectors. Data acquisition was performed using the software package babirlDAQ^[@CR67]^. The charge-changing cross sections were measured using the transmission method taking into account the geometrical acceptance of the MUSIC and NaI(Tl) detectors. In the present transmission method, the numbers of incident carbon beam and outgoing carbon particles, including lighter carbon isotopes, were identified and counted. Proton radii and Glauber-model analysis {#Sec9} --------------------------------------- The point-proton root-mean-square radius is defined as follows:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_{\mathrm{p}} \equiv \left\langle {r_{\mathrm{p}}^2} \right\rangle ^{1/2} = \left( {{\int} {\rho _{\mathrm{p}}({\bf{r}})r^2\mathrm{d}{\bf{r}}} } \right)^{1/2},$$\end{document}$$where *ρ*~p~(**r**) is the proton density distribution, **r** is the radial vector, and *r* is the radius. To extract the proton radii from the measured charge-changing cross sections, we performed reaction calculations using the recently formulated Glauber model^[@CR37]^ within the optical-limit approximation. We assumed that charge-changing cross section depends only on the proton density distribution in the carbon projectile. By adopting a simple one-parameter HO or a two-parameter Woods--Saxon (WS) density distribution for the protons, we determined the parameter(s) so as to reproduce the experimental data. *R*~p~ is then calculated by substituting the obtained proton density distribution into Eq. ([2](#Equ2){ref-type=""}). The difference (about 0.5%) between the *R*~p~ values determined with different functional forms was taken as the systematic uncertainty. The HO-type and WS-type density distributions are given by:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{l}\rho _{\mathrm{p}}^{{\mathrm{HO}}}\left( {Z,R_{{\mathrm{HO}}},r} \right) = \rho _{\mathrm{0}}^{{\mathrm{HO}}}{\kern 1pt} {\mathrm{exp}}\left[ { - \left( {\frac{r}{{R_{{\mathrm{HO}}}}}} \right)^2} \right]\left[ {1 + \frac{{Z - 2}}{3}\left( {\frac{r}{{R_{{\mathrm{HO}}}}}} \right)^2} \right],\\ \rho _{\mathrm{p}}^{{\mathrm{WS}}}\left( {R_{{\mathrm{WS}}},a,r} \right) = \rho _{\mathrm{0}}^{{\mathrm{WS}}}\left[ {1 + {\mathrm{exp}}\frac{{\left( {r - R_{{\mathrm{WS}}}} \right)}}{a}} \right]^{ - 1}\end{array}$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _{\mathrm{0}}^{{\mathrm{HO}}}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _{\mathrm{0}}^{{\mathrm{WS}}}$$\end{document}$ are the central densities, which are uniquely determined by the conservation of proton number (*Z*). *R*~HO~ is the HO width parameter, while the parameters *R*~WS~ and *a* are the half-density radius and diffuseness, respectively. Data availability {#Sec10} ----------------- The data that support the findings of this study are available from the corresponding author upon reasonable request. Electronic supplementary material ================================= {#Sec11} Peer Review File **Electronic supplementary material** **Supplementary Information** accompanies this paper at 10.1038/s41467-018-04024-y. **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. We thank T. Shima, H. Toki, K. Ogata and H. Horiuchi for discussion, K. Hebeler for providing matrix elements in Jacobi coordinates for 3NFs at NNLO, and C. Yuan for comments. H.J.O. and I.T. thank A. Tohsaki and his spouse, D.T.T. and T.T.N. acknowledge RCNP Visiting Young Scientist Support Program, D.T.T. and T.H.H. thank Nishimura International Scholarship Foundation and Matsuda Yosahichi Memorial Foreign Student Scholarship, respectively, for support. This work was supported in part by Hirose International Scholarship Foundation, the JSPS-VAST Bilateral Joint Research Project, Grand-in-Aid for Scientific Research Nos. 20244030, 20740163 and 23224008 from Japan Monbukagakusho, the Office of Nuclear Physics, U.S. Department of Energy, under grants DE-FG02-96ER40963, DE-SC0008499 (NUCLEI SciDAC collaboration), the Field Work Proposal ERKBP57 at Oak Ridge National Laboratory (ORNL), and the Vietnam government under the Program of Development in Physics by 2020. Computer time was provided by the Innovative and Novel Computational Impact on Theory and Experiment (INCITE) program. This research used resources of RCNP Accelerator Facility and the Oak Ridge Leadership Computing Facility located at ORNL, which is supported by the Office of Science of the Department of Energy under Contract No. DE-AC05-00OR22725. H.J.O. initiated the project, performed systematic analysis, prepared the figures and wrote the manuscript. D.T.T. performed data analysis and Glauber-model calculations. G.H., T.D.M. and G.R.J. performed the CC calculations. T.S. and T.O. performed the shell model calculations. Y.K.-E. and L.S.G. performed the AMD and RMF calculations, respectively. D.T.T., H.J.O., N.A., S.T., I.T., T.T.N., Y.A., P.Y.C., M.F., H.G., M.N.H., T.H., T.H.H., E.I., A.I., R.K., T.K., L.H.K., W.P.L., K.M., M.M., S.M., D.N., N.D.N., D.N., A.O., P.P.R., H.S., C.S., J.T., M.T., R.W. and T.Y. performed the experiments. All authors discussed and commented on the manuscript. Competing interests {#FPar1} =================== The authors declare no competing interests.

Introduction ============ Synthetic biological systems *in vitro* have recently attracted attention, holding promise as model systems for reducing biological complexity and for the development of new hybrid functional materials (see recent review by [@b6]; [@b5]). *In vitro* transcription could be used to design DNA-based devices and switches ([@b4]; [@b3]; [@b10]), and to produce RNA molecules with regulatory and catalytic capabilities ([@b7]) that could ultimately interface with nanoscale DNA nanomechanical systems (reviewed by [@b19]). Incorporating the complete transcription/translation reaction expands the scope of *in vitro* synthetic systems, for example to evolve novel proteins ([@b24]; [@b18]), synthesize protein nano-assemblies ([@b2]), explore synthetic gene circuits ([@b15]; [@b9]; [@b8]), and eventually develop artificial cells ([@b20]; [@b16]; [@b14]; [@b6]; [@b11]; [@b22]; [@b13]). Synthetic systems based on cell-free gene expression would benefit from appropriate materials platforms, and a surface with immobilized genes is an innate platform to regulate biosynthetic reactions ([@b21]), separate them spatially, and cascade their action by diffusion ([@b1]). On a surface, we can explore primitive forms of regulation, as compared with those used in living cells and ones who rely on physical constraints, geometry, and symmetry breaking, which would be difficult to implement in bulk solution or inside a lipid vesicle ([@b24]; [@b23]; [@b16]; [@b6]; [@b22]; [@b13]). However, progress has been so far hindered by our inability to control the immobilization of genes on a surface and to efficiently express them. Here, we used a unique biochip platform ([@b1]) to introduce dense gene brushes as a very primitive emulation of the cell nucleus. Inert polymer brushes are abundant and useful in everyday complex fluid systems and we recall their basic feature ([@b12]). Consider linear polymers attached to a surface. Their first tendency is to form short brushes with random-walk configurations in order to maximize the entropy. A second one is to form sparse tall brushes in order to be wet by solvent. These tendencies contradict if the distance between attachment sites is much smaller than the polymer radius of gyration. Hence the polymer free energy is a balance of two costs, namely stretching, which reduces their entropy, and overlapping with neighboring chains, which reduces the favorable contact with the solvent ([@b12]). As a result, at tight packing, the polymers adopt a configuration that is extended beyond their solution end-to-end distance. The assemblies presented here are made of end-attached linear double-stranded DNA (dsDNA) polymers coding for entire genes, whose lengths are a few thousand base pairs (bp) with controlled densities, ranging from sparse configurations to dense ones, with a local DNA concentration similar to that of *Escherichia coli*. An unexplored area in materials science, these synthetic gene brushes open the possibility to encode a biochemical function into the material: the physical properties of the brush may affect the biochemistry of biosynthetic reactions, and gene products could, in turn, change brush properties. Here, the first part of this possible feedback loop is explored by a 'structure--function\' study that seeks to unravel the affect of two-dimensional gene crowding on biosynthetic reactions. We recently observed a nonlinear relationship between proteins expressed from immobilized genes in a T7 RNA polymerase (RNAP)/wheat-germ extract, where prokaryotic transcription is *a priori* uncoupled from the eukaryotic translation machinery ([@b1]). This initial observation, lacking the most natural comparison between gene expression on a surface and that in solution, had stimulated a more rigorous search for an underlying mechanism of the pattern observed. For a basic understanding of the effect of gene localization on expression, it is imperative to identify the machinery components most sensitive to DNA surface configuration. It is important to elucidate the interplay between surface transcription and the concomitant translation, which can occur either from nascent transcripts or from fully released ones, depending on the particular expression system used and on the biophysical constraints on the surface. Here, we assembled gene brushes encoding for *firefly luciferase* reporter under control of the T7 promoter (*T7-luc* construct) and then carried out coupled transcription/translation in *E. coli* cell extracts as well as T7 transcription in a buffer on surfaces as a function of the brush density, conformation, and orientation, and in comparison with solution. We found that protein synthesis is significantly affected by gene brush structure, exhibiting simple linear gene-to-protein relationship for stretched genes and nonlinear, sigmoidal behavior for dilute relaxed conformations. We observed attenuation at dense gene packing only in the transcription/translation reactions and not in the transcription-only reactions, most likely due to molecular crowding and sensitivity of the translation machinery to the orientation of the grafted genes. The transcription-only assays imply that RNAP is indeed the primary sensor of brush properties. Biosynthetic kinetics and lifetime measurements allow us to deduce a significant enrichment of product accumulation near the surface and their spatial gradients. Results and discussion ====================== Gene brush assembly ------------------- We used a biochip platform for cell-free gene expression based on a new single-step photolithographic biocompatible monolayer ([@b1]) (termed 'daisy\'). A daisy molecule is comprised of a polyethyleneglycol backbone with a silicon-binding group at one end and an amine-protecting group at the other ([@b1]). Daisy molecules readily coat silicon dioxide, forming a smooth monolayer with amines every ∼1.6 × 1.6 nm^2^ that become available for chemical binding upon UV irradiation of the surface. Following photo deprotection of amines, we chemically bind biotin and then immobilize streptavidin-conjugated linear dsDNAs (SA-DNA) on the surface-bound biotins. The spatial pattern of dsDNAs on the surface can be determined by projecting a photo-mask: flux-dependent partial deprotection of amines through the mask converts image grayscale into surface density, as shown, for example, in [Figure 1A](#f1){ref-type="fig"}. The dsDNA brush is maintained in an aqueous solution to avoid collapse of the structure. Thus, by controlling the UV flux, we can prepare brushes with spatially varying surface densities. We verify the process of dsDNA photolithography by labeling dsDNA independently both at its surface proximal (green) and solution distal (red) ends ([Figure 1A](#f1){ref-type="fig"}). For studying the characteristics of protein biosynthesis as a function of gene density, we developed a quantitative DNA grafting assay that would allow us to more accurately quantify the density of grafted DNA. In this assay, rather than varying the UV flux ([@b1]) ([Figure 1A](#f1){ref-type="fig"}), we exposed daisy-coated chips uniformly (without a mask) and at the maximal UV flux required for deprotection of all surface amines. Following binding of biotin to the exposed amines, SA-DNAs at variable solution concentrations were incubated for attachment on the surfaces, which were then extensively washed. To accurately evaluate the correlation between solution concentration and equilibrium DNA surface density, we used radioactively-labeled dsDNAs, 420, 2160, and 2900 bp long with contour lengths of 140, 720, 960 nm, and an estimated relaxed end-to-end distance of 85, 240, 295 nm, respectively ([@b25]) ([Figure 1B](#f1){ref-type="fig"}). The two long dsDNAs are composed of several persistence lengths (*l*~p~*=*44 nm; [@b25]), whereas the short DNA fragment consists of a maximum of two persistence lengths. The mean distances between dsDNAs at equilibrium, for the three lengths, decreased monotonously with their solution concentrations, from 150, 235, 250 nm at dilute surfaces to 15, 30, 36 nm at maximal packing ([Figure 1B](#f1){ref-type="fig"}). Correspondingly, the maximal density of dsDNAs on the surface was 4000, 1070, 735 dsDNAs per μm^2^. A direct measurement of brush height and structure is experimentally challenging and beyond the scope of this work. Nonetheless, we note the following: (a) SA-DNA binds specifically to the surface through the SA--biotin bond, rather than nonspecifically at multiple points along the DNA polymer. This is evident from control experiments where we achieved specific end-attachment ratios of 50--200 as measured with respect to SA-DNA adsorbed to unexposed daisy surfaces, or to DNA not conjugated to SA that were adsorbed to fully exposed daisy surfaces. Realizing that the DNAs are specifically end-anchored to the surface and that they pack maximally at distances that are smaller than the persistence length, the rest of the polymer must extend out into solution to minimize its free energy ([@b12]). (b) The assembly of dsDNAs on the surface slows down considerably after a few hours (data not shown) supporting the notion that a tightly packed DNA arrangement had been formed on the surface, hampering the ability of additional SA-DNA from reaching the surface. Altogether, our brush density data necessarily supports a picture in which the dsDNAs at high density must acquire an extended conformation. This interpretation does not necessarily invoke complete extension of the DNA polymer chain, but rather a more extended conformation than the mushroom-like relaxed conformation, resulting in a brush height that is larger than the dsDNA\'s end-to-end distance. Therefore, the DNA brush schemes in the remaining of the figures are intended as a qualitative guide to depict the change in DNA conformation upon increase in density. The structure of these gene brushes is most likely intermediate between long flexible polymers and rod-like molecules. Sigmoidal response of protein expression from surface-bound genes ----------------------------------------------------------------- We studied the effect of gene localization on expression by immobilizing the *T7-luc* construct on surfaces followed by bathing the surfaces with an *in vitro* transcription/translation reaction mix. The T7 promoter was oriented close to the surface, separated by a 60 bp spacer (∼20 nm), with synthesis of mRNA directed outward into the solution ([Figure 2A](#f2){ref-type="fig"}). The aqueous solution was at a height of about 1.5 mm above the chip in an open reaction chamber that allowed us to collect aliquots of luciferase expressed on the surface and released into solution after a few hours of reaction. For comparison, we expressed luciferase from genes in solution, while maintaining the *same* gene number (in moles) and reaction volume as on the surface ([Figure 2A](#f2){ref-type="fig"}). The data showed the following features: (a) expression in solution followed a simple gene/protein linear relationship, whereas on a chip, the curve was nonlinear with a power law of 2.5±0.5; (b) the expression attenuated at high gene number on the chip; (c) saturation of expression on the chip occurred at 30-fold fewer genes than in solution. Orientation of surface-bound genes affects expression ----------------------------------------------------- The marked differences between biosynthesis on the surface and in solution, done at identical conditions except for gene localization, hint that brush structure is important for the expression efficiency of genes. Guided by the crowded nature of the brush, we searched for the origin of these distinct characteristics in the orientation of the genes with respect to the surface, reasoning that (a) if the surface-bound DNAs do not form a brush, then no effect of promoter orientation and distance would occur, and (b) if DNA brush does forms, then placing the promoter at the top of the brush rather than at the bottom may improve its accessibility to the machinery. To test these hypotheses, we assembled gene brushes once with their streptavidin conjugated at the promoter proximal end, and alternatively at the promoter distal end, resulting in 'bottom\' and 'top\' configurations, respectively ([Figure 2B](#f2){ref-type="fig"}). We carried out the same *firefly luciferase* expression experiment from 'bottom\' and 'top\' at varying gene number. Indeed, we found that top-oriented genes yielded a significant increase in the expression compared with bottom-oriented ones for the entire gene number range ([Figure 2B](#f2){ref-type="fig"}). This result is consistent with experiments done with DNA immobilized on beads ([@b17]) and provide strong evidence to support DNA extension at high gene densities. A structure--function relationship of protein expression from gene brushes -------------------------------------------------------------------------- The details of DNA brush build up ([Figure 1B](#f1){ref-type="fig"}) have strongly implied, concomitantly with an increase in gene density, a gradual change of dsDNA structure, from a sparse and relaxed conformation to an extended dense brush. To isolate the effect of dsDNA structure on expression, we therefore varied the gene number while fixing the overall dsDNA density to the highest value. This was achieved by adding 'dummy DNA\' of equal length to the genes but without promoter and coding regions. We prepared nine dsDNA brushes with equal overall dsDNA (gene and dummy), but each with a different ratio of bottom-oriented genes to dummy DNA. Hence, for all gene and dummy combinations, the total amount of DNA applied on the surface was identical, resulting in a similar conformation of the dsDNA ([Figure 2](#f2){ref-type="fig"}, scheme). The expression of luciferase from the mixed gene/dummy brushes showed a simple linear gene/protein relationship, in contrast to the sigmoidal behavior observed in the gene-only brushes ([Figure 2C](#f2){ref-type="fig"}). As expected, the gene-only and gene/dummy curves merged at maximal gene number where their content is basically identical and composed only of genes. We verified that dummy DNA did not affect transcription/translation efficiency in solution (not shown) and we hence postulate that its significant effect on expression from the surface is due to its modulation of DNA structure. Further insight is gained by replotting the utility of the genes ([Figure 2D](#f2){ref-type="fig"}), defined by the number of proteins synthesized divided by the number of genes, as a function of the mean intergene distance derived from the dsDNA surface densities ([Figure 1](#f1){ref-type="fig"}), rather than total expression as a function of gene number ([Figure 2C](#f2){ref-type="fig"}). The utility of genes in the gene/dummy systems is essentially constant, independent of distance. The crossover points between the gene-only and gene/dummy plots identified three gene-number regimes ([Figure 2](#f2){ref-type="fig"}, I, II, III): at low gene densities (regime I), where the mean distance between adjacent genes varies from 300 to 150 nm, the gene/dummy configuration was significantly more efficient than gene-only configuration. At the intermediate regime II (150--60 nm), the gene-only configuration became increasingly favorable as the intergene distance reached the 60 nm value. From that point onward (regime III; 60--30 nm), the expression efficiency of the gene-only configuration attenuated to the constant value displayed by the gene/dummy system. We hypothesized that, in the dilute regime, gene-only configuration assumes a relaxed, nonoverlapping conformation, which is less favorable for the machinery than extended genes ([Figure 2C and D](#f2){ref-type="fig"}, regime I). Upon decreasing the distance from 300 to 150 nm (regime I), the genes begin to overlap and experience mutual steric repulsion and hence assume an extended conformation. It is well established that the height of long polymer brushes extends nonlinearly with polymer surface density ([@b12]). As dsDNAs extend, they are likely to be more accessible to the biochemical machinery than the 'solution-like\' conformation in the dilute regime. In regime II, with intergene distances of 60--150 nm, the conformations of gene-only and those of genes/dummy are similarly stretched; yet, the actual density of dsDNA is higher in the latter. At high density, the machinery must be impeded from efficiently propagating down the brush, which is consistent with gene-only conformations being more efficient than those of genes/dummy in this regime. Indeed, in regime III, the densities of the two systems become progressively comparable as the intergene distance approaches the 30 nm limit and their expression efficiencies match. The 60 nm peak in expression marks a balance between the improvement due to extension at moderate density and the impediment due to crowding at high density. In contrast, the gene/dummy system has a fixed density and structure for the entire gene number range and hence the efficiency of expression per gene remains constant. To elucidate the role played by T7 RNAP in sensing the properties of the brush, we carried out transcription-only assays in an optimized buffer, rather than in the extract, and repeated the gene-only and gene/dummy experiments ([Figure 2E and F](#f2){ref-type="fig"}; [Supplementary Figure S2](#S1){ref-type="supplementary-material"}). The salient features observed in protein synthesis were recapitulated in these transcription-only assays, suggesting that RNAP is the primary sensor of the DNA brush. Importantly and in contrast to transcription/translation, these transcription-only experiments did not show any attenuation at high gene number in any of the configurations tested, consistent with the notion that the attenuation observed in the transcription/translation ([Figure 2C](#f2){ref-type="fig"}) is a manifestation of impediment of ribosome penetration into the brush at high densities. We note that, in all experiments comparing 'top\', 'bottom\', and solution, the 'bottom\' always saturated at lower values. This could possibly stem from nucleotide depletion within the brush. However, we disfavor this scenario, as nucleotide depletion should affect the rate of synthesis during the elongation step of transcription, which takes place along the entire length of the DNA. Therefore, any gradients of nucleotide concentrations are expected to alter RNA synthesis rates, whether initiating at the 'top\' and terminating at the 'bottom\' or vice versa. It is more likely that the initiation step of transcription, namely promoter binding, rather than the elongation step, should be affected by the localization of the promoter with respect to the surface. We therefore suggest that the marked differences between 'bottom\' and 'top\' configurations are due to changes in the local concentration of RNAP near the surface and at the interface with the solution, respectively. Although changes in salt concentrations may also affect binding of RNAP to the promoter, we suggest that salt gradients do not play a major role in determining transcription levels in the brush because of the following reasons: (a) the screening length is ∼2 nm under the experimental conditions; (b) the distance of the promoter from the surface is no less than 30 nm; (c) the distance between dense DNAs is also ∼30 nm. Surface-induced enrichment of biosynthetic products and concentration gradients ------------------------------------------------------------------------------- The above analysis stemmed from a comparison between expression curves on the surface at different DNA densities, but neglected to account for the marked differences between expression in *solution* and on the surface ([Figure 2A](#f2){ref-type="fig"}). We ruled out the possibility that these differences resulted from biochemical effects on the rates of translation per mRNA ([Supplementary Figure S1, c](#S1){ref-type="supplementary-material"}) and we have also shown that translation could occur from freely diffusing mRNA, uncoupled from the transcription process (not shown). Therefore, to fully appreciate the difference between solution and surface expression, we must consider the spatial distribution of both transcription and translation in the solution surrounding the brush. Unlike the DNA, the mRNA and protein expressed on the chip are not physically immobilized. Yet, as we now show by a simple calculation, they must be confined to a thin diffuse layer next to the surface. On the chip, mRNA molecules originate from the 'nucleus\' of localized genes. Although they are free to diffuse, their escape from the surface is limited by mRNA degradation in the cell extract ([Supplementary Figure S1](#S1){ref-type="supplementary-material"}). The transcription reaction can therefore be modeled as a one-dimensional reaction-diffusion equation with a localized source at *z*=0, combined with degradation. The steady-state solution is an exponential decaying concentration profile ([Supplementary information](#S1){ref-type="supplementary-material"}): where ν~tx~ is the transcription rate, *σ*=990 genes μm^−2^ is the gene density, *τ*=20 min is the mRNA lifetime, *D*~mRNA~≈6.5 μm^2^ s^−1^ is its diffusion constant and is the decay length ([Figure 3A](#f3){ref-type="fig"}). Consequently, although the *total* mRNA synthesis on the chip was 1/2 than that in solution ([Supplementary Figure S1, a](#S1){ref-type="supplementary-material"}), the *local* mRNA concentration in the vicinity of the surface was actually significantly higher relative to the corresponding solution expression by a factor (ν~tx~^chip^/ν~tx~^sol^)(*L*/λ)≈8, where *L*=1.5 mm is the chamber height ([Supplementary information](#S1){ref-type="supplementary-material"}). Given the same transcription rates and diffusion constant, faster degradation would lead to a sharper gradient. Translation follows transcription not only from nascent mRNAs in the gene brush, but also from fully transcribed diffusing ones. Hence, the level of protein translation is proportional to the local mRNA concentration and to the translation rate. Unlike the mRNA, luciferase is not degraded in the cell extract and hence its synthesis cannot reach a steady-state concentration profile (this may not be the case for other proteins). Therefore, we solved for the time-dependent accumulation of luciferase from the brush ([Supplementary information](#S1){ref-type="supplementary-material"}). At short times, the profile of protein concentration follows the mRNA exponential profile confined within 90 μm from the surface ([Figure 3B](#f3){ref-type="fig"}). With time, protein diffusion becomes significant and the protein profile resembles a Gaussian distribution centered at *z*=0 with a typical width of . After *t*=2*τ*, most of the luciferase in our on-chip expression reaction would be found concentrated at a layer of *h*~luc~=140 μm above the DNA brush and, as in the case of mRNA, its local concentration is higher than the solution expression experiment. The analysis of mRNA and protein concentration gradients points to a more realistic comparison between solution and surface expression, by taking into account the localization of genes and their protein products and rescaling their gene and protein numbers ([Figure 2A](#f2){ref-type="fig"}) to *effective concentration* according to the ratios of *L*/*h*~brush~ and *L*/*h*~luc~, respectively, with *h*~brush~≈500 nm ([Figure 3C](#f3){ref-type="fig"}). Presenting the expression profile of solution and chip each in their own corresponding effective volumes provides a new perception of on-chip expression, one that is highly efficient and localized compared with solution. Summary ======= The above analysis presents, in our view, the hallmark of expression from gene brushes: a heterogeneous system with localized, highly concentrated, and oriented DNA molecules, immersed in a solution of freely diffusing protein machinery. Protein biosynthesis is highly sensitive to gene density, conformation, and orientation. Gene expression products are localized in the vicinity of the brush due to the interplay of degradation, synthesis, and diffusion rates. Gene brushes could be used to deepen our understanding of other DNA-based reactions where confinement and crowding are expected to play a role, such as recombination, restriction, and methylation. In turn, fundamental understanding of gene brushes could be used to develop new synthetic systems based on controllable gene expression units on surfaces. This localized product enrichment, free of physical boundaries, opens up the possibility to switch and cascade biochemical processes that are based on weak interactions between macromolecules and are thus highly affected by solution concentration. Materials and methods ===================== Experimental methods -------------------- Protocols for surface modification, UV photolithography, biotinylation of deprotected surface amines, synthesis of biotinylated DNA radioactively and nonradioactively labeled, streptavidin-DNA conjugation, luciferase assay are all described in the Supporting information and partly by [@b1]). Radioactive DNA density quantitation ------------------------------------ Surfaces bound with radioactive DNA were thoroughly rinsed, air-dried, and exposed to phosphorous screen (FUJI), which was then scanned by a phosphorimager (FLA-5100, FUJI). Known amounts of radioactive DNA were spotted on the same surface *without washing* to derive the average specific activity of each DNA. The amount of moles of DNA was derived by dividing the radiation in each spot by the specific activity of that DNA. Parallel RNA and protein synthesis on a chip in the extract ----------------------------------------------------------- Daisy-coated silicon dioxide substrates (18 × 18 mm^2^) were placed in a custom-made Teflon reaction chamber (details in [@b1]) with nine separated regions each containing 10 μl of cell-free reaction mix. Cell-free coupled transcription/translation reactions were carried out using the RTS100 Cell-Free *E. coli* Extract System (Roche) according to the manufacturer protocol. This extract is optimized with minimal nuclease activity and is hence suitable for the use of linear DNA templates. Aliquots were taken from each region on the chip to measure luciferase luminescence. Transcription rates were measured by supplementing the cell-free *E. coli* extract with 0.2 μM ^32^P-α-rUTP. The reaction was stopped by the addition of 0.2% SDS/20 mM EDTA final concentrations and placed immediately on ice. A total of 1 μl of this stopped reaction was spotted on a polyethyleneimine (PEI) cellulose thin layer chromatography (TLC) plate (Merck). The RNA was resolved from its precursor nucleotides by developing in 0.3 M potassium phosphate, pH=7.0. TLC plates were dried and their radioactive spots were imaged as above. RNA concentration was quantified assuming 0.5 mM endogeneous rNTP. mRNA synthesis in a minimal transcription-only reaction ------------------------------------------------------- A transcription-only reaction was constructed by mixing 5 × transcription buffer (Promega) with 10 mM DTT (Promega), rNTP mix (0.5 mM rATP, rGTP, and rCTP and 0.25 mM rUTP), 3 U μl^−1^ recombinant RNasin ribonuclease inhibitor (Promega), and 2--20 U μl^−1^ T7 RNAP. mRNA production was measured by supplementing the reaction with 0.2 μM ^32^P-α-rUTP. The reaction was stopped and processed as described above. RNA degradation rate -------------------- Radioactive-labeled luciferase RNA was transcribed from grafted genes using 3 U μl^−1^ t7-RNA polymerase (Roche) in 1 × transcription buffer (Roche) supplemented with 200 μM of rNTPs (each of the four) and 0.2 μM ^32^P-α-rUTP. The reaction was incubated at 30°C for 200 min. A 5 μl aliquot (2 pmol) was removed from the chip into 20 μl *E. coli* cell extract and further incubated at 30°C. After 2, 5, 15, 40, and 200 min, 3 μl samples were collected, stopped, and quantified as described above. The theoretical methods for solving the reaction-diffusion equations for transcription/translation are described in the Supporting information. Supplementary Material {#S1} ====================== ###### Supplementary information We would like to thank S Safran and T Tlusty for useful discussions. This work was supported by grants from the Israel Science Foundation and the Minerva Foundation. {ref-type="supplementary-material"}). Streptavidin-conjugated dsDNA (SA-DNA, brown) are then immobilized with the mask grayscale translated into surface density of dsDNA imaged by dual fluorescent tags, FITC (green) covalently attached to the SA, and cy5 (red) covalently attached to the DNA at its surface-distal end. Scale bar is 50 μm. (**B**) Assembly of uniform linear dsDNA brushes. The packing of linear dsDNA polymer brushes following uniform deprotection of daisy-coated chips was quantified by radioactive labeling ([Supplementary information](#S1){ref-type="supplementary-material"}) of 420 (green), 2160 (red), and 2900 (blue) bp DNA. Each radioactive spot in a 3 × 3 array contained a different amount of DNA (shown above the graph) as imaged by phosphorimaging densitometry (Methods and materials). The signal from the adsorbed DNA was converted to number of molecules using a calibration curve deduced from spots of known amounts of radioactive DNA on the surface (left). The equilibrium mean distance between dsDNAs was dictated by their solution concentration (main plot).](msb200820-f1){#f1} {#f2} {ref-type="supplementary-material"}) and the measured transcription and translation rates. (**B**) The luciferase profile does not reach steady state and is shown for *t*=30, 55, and 80 min, relative to solution (chip, red; solution, black). (**C**) Replotting [Figure 2A](#f2){ref-type="fig"} after rescaling gene and protein numbers by brush and diffuse layer heights, h~brush~=500 nm and h~luc~=140 μm.](msb200820-f3){#f3}

Introduction ============ Childhood cataract is the most common form of treatable blindness in children \[[@r1]\]. It accounts for more than 1 million blind children in Asia and about 10% of childhood blindness worldwide \[[@r2]\]. Non-syndromic congenital cataracts have an estimated incidence of 1--6 per 10,000 live births \[[@r3],[@r4]\]. The prevalence of childhood cataract varies between different population, geographic area, socioeconomic condition and other factors. Etiologic factors include genetic, metabolic, systemic disorders, intrauterine infections, and trauma. Approximately 50% of the childhood cataracts are genetic and at least one-third may have the familial basis. It is both clinically and genetically diverse as demonstrated by several investigators suggesting involvement of possible modifying factors \[[@r2]\]. Hereditary cataracts typically follow Mendelian inheritance with majority being autosomal dominant type with complete penetrance and may be either static or progressive. Cataracts are classified based on morphology, location, or time of onset. Childhood cataracts have been mapped to 36 genetic loci and mutations in 22 genes have been identified toward the pathogenesis of various forms of congenital and developmental cataract including 12 loci for autosomal recessive cataract \[[@r5]-[@r8]\]. Biochemically up to 90% of the water soluble proteins in the vertebrate lens belong to α-, β-, and γ-crystallins. Being a part of small heat shock protein the α-crystallins form high-molecular aggregates and function as molecular chaperones. The β/γ-crystallin superfamily exhibits a characteristic Greek key motif, in a quadruple organization showing two in the NH~2-~ and two in the COOH-terminal domain. Evolutionary analysis has demonstrated the relationship of crystallins to other stress proteins and is expressed in other tissues of the body as well \[[@r9]\]. The human gamma-crystallins (*CRYG*) gene cluster comprise six genes including two pseudogenes. The spectrum of mutations in *CRYG* gene leading to diverse cataract phenotypes is on the increase \[[@r10],[@r11]\]. The aim of the present study therefore was to screen families with inherited cataracts to document known as well as novel mutations if any in human genes coding for alpha-A crystallin (*CRYAA*), beta-B2 crystallin (*CRYBB2*), gamma-crystallins (*CRYG*), gap junction alpha-3 and alpha-8 (*GJA3*, *GJA8*), and transcription factor paired box gene 6 (*PAX6*). Methods ======= Cases of childhood cataract were registered through Kasturba Hospital (KH), Manipal, India. Clinical details of the proband were recorded in all the cases. Ophthalmic investigation included slit lamp examination with dilated pupils, visual acuity testing, intraocular pressure measurement, and fundus examination done by a senior ophthalmologist (V.P.). In familial cases, identification of cataract phenotype and detailed examination of the affected as well as available unaffected family members were performed. A detailed pedigree of the kindred was ascertained by interviewing the parents or any available family member. Clinical details of the patients who previously had cataract extraction were obtained through medical records. Cases presenting conditions such as unilateral, congenital rubella, systemic disorders, traumatic, syndromic, and other known causes were excluded for further study. Seeking of informed consent from all participants and parents of the probands was in accordance with the Declaration of Helsinki and Institutional Ethical Committee of Manipal University. Blood samples were collected from available affected/unaffected members of the family. PCR and single stranded conformational polymorphism (SSCP) analysis ------------------------------------------------------------------- Genomic DNA was extracted from peripheral blood leukocytes using phenol chloroform method \[[@r12]\]. Exons and exon-intron boundaries of *CRYAA*, *CRYBB2*, *CRYGA* \>*D, GJA3, GJA8*, and *PAX6* were PCR amplified as per earlier reports on *CRYAA* \[[@r13]\], *CRYBB2, GJA8* \[[@r14]\], *CRYGA\>D* \[[@r15]\], *GJA3* \[[@r16]\] and *PAX6* \[[@r17]\]. The PCR reactions were performed at 94 °C for 4 min followed by 29 cycles at 94 °C for 30 s, at optimum annealing temperature for 45 s, at 72 °C extension for 45 s, and final extension for 10 min. Each reaction mix (25 µl) contained 50 ng of genomic DNA, 1× PCR buffer, 1.5 mM MgCl~2~, 200 µM dNTPs, 10 pmol each of sense and antisense primers ([Table 1](#t1){ref-type="table"}) and 1U of Taq DNA polymerase (Fermentas, Glen Burnie, MD). Thermal cycling was performed at suitable conditions using a VERITI 96 well thermal cycler (Applied Biosystems, Foster City, CA). ###### List of primers used for PCR amplification of lens specific genes. **Primer** **Exon** **Primer sequence 5'-\>3'** **Product size (bp)** ------------- ---------- --------------------------------- ----------------------- *CRYAA-1F* 1 CTCCAGGTCCCCGTGGTACCA 254 *CRYAA-1R*   GCGAGGAGAGGCCAGCACCAC   *CRYAA-2F* 2 CTGTCTCTGCCAACCCCAGCAG 223 *CRYAA-2R*   CCCCTGTCCCACCTCTCAGTGCC   *CRYAA-3F* 3 GGCAGCTTCTCTGGCATGGGG 312 *CRYAA-3R*   GGGGAGCCAGCCGAGGCAATG   *CRYBB2-1F* 1 TCTGTGGGCATTTGCTGACCC 292 *CRYBB2-1R*   GCTAACAGCATTGAAGTCTCTGCCC   *CRYBB2-2F* 2 GACCCCACAGCTCTGGGACAGTC 393 *CRYBB2-2R*   GGAGGGACTTTCAGTATCAGCTCCAAC   *CRYBB2-3F* 3 CACGGCTGCTTATAGCCAGAGCC 449 *CRYBB2-3R*   TCTATCTGACTGCAAAGCATGAATTATCTCC   *CRYBB2-4F* 4 GCTTTGGGCACAGCGATGTTCTG 744 *CRYBB2-4R*   GGCCCCTTCCTGGTCCCCA   *CRYBB2-5F* 5 AGTGGTCATAGACACGTAGTGGGTGCAC 706 *CRYBB2-5R*   CTGTTCCCAAACTTAGGGACACACGC   *CRYBB2-6F* 6 CCCCTCGTTCACCCTCCCATCA 506 *CRYBB2-6R*   CACTGTGTCCAAGGTCACACAGCTAAGC   *CRYGA-2F* 1&2 AGGTCCCTTTTGTGTTGTTTTTGCC 462 *CRYGA-2R*   CATGAGGAATTATACGGCAGGATTGG   *CRYGA-3F* 3 CAGACCAGCTCGCACAAGTTAAGG 353 *CRYGA-3R*   AAGAGCCACTTAGTGCAGGGAACACAAC   *CRYGB-2F* 1&2 TGCAAATCCCCTACTCACCAAAATGG 518 *CRYGB-2R*   AAAAAGATGGAAGGCAAAGACAGAGCC   *CRYGB-3F* 3 TTTGTTTACTCTTGCGTTTTCTGTCTGCC 410 *CRYGB-3R*   GAAAGAAAGACAGGGCTCTACTAGTGCC   *CRYGC-2F* 1&2 TGCATAAAATCCCCTTACCGCTGAG 522 *CRYGC-2R*   ACTCTGGCGGCATGATGGAAATC   *CRYGC-3F* 3 AGACTCATTTGCTTTTTTCCATCCTTCTTTC 407 *CRYGC-3R*   GAAAGAATGACAGAAGTCAGCAATTGCC   *CRYGD-2F* 1&2 GCAGCCCCACCCGCTCA 599 *CRYGD-2R*   GGGTAATACTTTGCTTATGTGGGGAG   *CRYGD-3F* 3 TGCTTTTCTTCTCTTTTTATTTCTGGGTCC 400 *CRYGD-3R*   AGTAAAGAAAGACACAAGCAAATCAGTGCC   *GJA8-L1F* 1 CGGGGCCTTCTTTGTTCTCTAGTCC 877 *GJA8-L1R*   AGGCCCAGGTGGCTCAACTCC   *GJA8-L2F* 1 CAGCCGGTGGCCCTGCC 907 *GJA8-L2R*   GTTGCCTGGAGTGCACTGCCC   *GJA3-1aF* 1 CTGCGATGCCTGTCCTGTGG 539 *GJA3-1aR*   TTGTCCTGCGGTGGCTCCTT   *GJA3-1bF* 1 CGCCCACCCTCATCTACCT 549 *GJA3-1bR*   GTGGGAACCCGATGGCAAC   *GJA3-1cF* 1 AGCTCAAGCAGGGCGTGACC 542 *GJA3-1cR*   CAAGGGCGGCTGGTGCATCT   *GJA3-1dF* 1 CCCCGGCGCTCAAGGCTTAC 545 *GJA3-1dR*   AACCCTTGTCCCCGCCACCC   *PAX6-1F* 1 CTCATTTCCCGCTCTGGTTC 300 *PAX6-1R*   AAGAGTGTGGGTGAGGAAGT   *PAX6-2F* 2 CACACTCTTTATCTCTCACTCTCCAGCC 300 *PAX6-2R*   AATAAAGCGAGAAAGAAGCGGAC   *PAX6-3F* 3 TCAGAGAGCCCATCGACGTAT 300 *PAX6-3R*   CTGTTTGTGGGTTTTGAGCC   *PAX6-4F* 4 TTGGGAGTTCAGGCCTACCT 153 *PAX6-4R*   GAAGTCCCAGAAAGACCAGA   *PAX6-5F* 5 CCTCTTCACTCTGCTCTCTT 257 *PAX6-5R*   ATGAAGAGAGGGCGTTGAGA   *PAX6-5aF* 5a TGAAAGTATCATCATATTTGTAG 237 *PAX6-5aR*   GGGAAGTGGACAGAAAACCA   *PAX6-6F* 6 GTGGTTTTCTGTCCACTTCC 299 *PAX6-6R*   AGGAGAGAGCATTGGGCTTA   *PAX6-7F* 7 CAGGAGACACTACCATTTGG 252 *PAX6-7R*   ATGCACATATGGAGAGCTGC   *PAX6-8F* 8 GGGAATGTTTTGGTGAGGCT 371 *PAX6-8R*   CAAAGGGCCCTGGCTAAATT   *PAX6-9F* 9 GTAGTTCTGGCACAATATGG 206 *PAX6-9R*   GTACTCTGTACAAGCACCTC   *PAX6-10F* 10 GTAGACACAGTGCTAACCTG 243 *PAX6-10R*   CCCGGAGCAAACAGGTTTAA   *PAX6-11F* 11 TTAAACCTGTTTGCTCCGGG 208 *PAX6-11R*   TTATGCAGGCCACCACCAGC   *PAX6-12F* 12 GCTGTGTGATGTGTTCCTCA 300 *PAX6-12R*   TGCAGCCTGCAGAAACAGTG   *PAX6-13F* 13 CATGTCTGTTTCTCAAAGGGA 957 *PAX6-13R*   GAACAATTAACTTTTGCTGGCC   One affected member per family was screened for mutations in *CRYAA*, *CRYBB2*, *CRYGA* \>*D*, *GJA3*, *GJA8*, and *PAX6* by SSCP analysis of PCR products. PCR products were purified by QIA quick purification kit (Qiagen, Hilden, Germany). The amplicons were mixed with SSCP denaturation solution (95% formamide, 10 mM NaOH, 0.25% bromophenol blue, 0.25% Xylene Cyanol- all reagents from Sigma-Aldrich, St. Louis, MO), denatured at 95 °C for 5 min and snap cooled on ice. Denatured samples were subjected to SSCP using DCode^[TM]{.smallcaps}^ Universal Mutation Detection System (Bio**-**Rad Laboratories, Hercules, CA). The denatured samples were then electrophoresed at 400 V for 16 h at RT as well as at 12 °C on 8%--10% polyacrylamide gels. The gels were silver stained for visualization of bands \[[@r18]\]. Samples showing variation in mobility as band shifts were further processed for sequencing. Restriction Fragment Length Polymorphism (RFLP) ----------------------------------------------- RFLP analysis was performed using AluI restriction enzyme (New England Biolabs Ltd., NEB, Hitchin, Herts, UK) as per the manufacturer's instructions in 10 µl volumes. The products were resolved on 2% agarose gel. The presence or absence of the AluI restriction site was checked in other family members and in 100 unrelated controls. DNA sequencing and structure prediction --------------------------------------- Samples showing mobility shift were subjected to direct sequencing using Big Dye terminator chemistry on an ABI 3130 genetic analyzer (Applied Biosystems). Sequencing reaction was run through a program which included 25 cycles of denaturation (96 °C for 10 s), annealing (50 °C for 5 s), and extension (60 °C for 4 min). Sequence data were analyzed by standard software and alignments done by [ClustalW](http://www.ebi.ac.uk/clustalw/) or [BLAST](http://www.ncbi.nih.gov/BLAST/) (GenBank [NM_006891](http://www.ncbi.nlm.nih.gov/nuccore/NM_006891)). The prediction of protein structure was made by using [PDB deep view](http://www.pdb.org/) or [PyMOL](http://www.pymol.org/) programs. Results ======= Clinical findings ----------------- The study was performed on patients with familial nonsyndromic bilateral childhood cataract. During the period (March 2004 to April 2009) forty eight subjects from 17 families and 148 isolated cases with childhood cataract were evaluated. Among these, zonular cataract was most frequent (46%), followed by total (13%) posterior sub capsular (10%), and nuclear cataracts (8%). Blue dot, sutural, and membranous cataracts were also recorded. In family C-35 ([Figure 1A](#f1){ref-type="fig"}) the proband (III:1) underwent a surgery for cataract removal at the age of 11 years. Right eye showed the phenotype as anterior polar with coronary cataract and left eye with only anterior polar cataract with a progressive loss of vision. Other ocular examination showed normal ocular movements with clear cornea. Visual acuity was found to be 6/60 (RE) and 6/36 (LE). The proband's younger sister (III:2) of age 9 years presented anterior polar cataract of progressive nature in both eyes. Visual acuity was 6/24 in both the eyes. The proband's father (II:4) showed anterior polar cataract in both eyes. The youngest sister (III:3) of the proband and the mother when examined for visual acuity have normal vision with clear cornea. A typical example of a slit lamp image of patient III:1 (proband in family CAT-35) is shown in [Figure 1B](#f1){ref-type="fig"}. {#f1} Mutation analysis ----------------- All exons and intron/exon boundaries and flanking sequences of the candidate genes, *CRYAA*, *CRYBB2*, *CRYGA*\>*D*, *GJA3*, *GJA8*, and *PAX6* were screened by PCR-SSCP, RFLP, and DNA sequencing methods for detection of mutations. Consensus exon/intron boundaries in *CRYAA*, *CRYBB2*, *CRYGA*\>*D, GJA3, GJA8*, and *PAX6* were verified by gene-specific primers designed to anneal to intronic sequence flanking exon boundaries. We have identified a sequence alteration in exon 2 of *CRYGD* in a proband (C-35) diagnosed to have anterior polar and coronary cataract using SSCP ([Figure 2](#f2){ref-type="fig"}). The samples showing the differential migration patterns were subjected to DNA sequencing. Sequence analysis of the three exons and immediate flanking regions in three affected family members (II:4, III:1, and III:2) using specific primers detected a heterozygous C\>A transversion in exon 2 resulting in a missense substitution of arginine to serine at codon 77 (R77S) which was not present in unaffected family member (II:3, II:5, and III:3; [Figure 3](#f3){ref-type="fig"}). The observed sequence variation was also confirmed through sequencing with the reverse primer in all the family members. {#f2} {#f3} This single-nucleotide change created an additional Alu1 restriction site in exon 2 of *CRYGD*. This Alu1 site (229AgkCT) co-segregated with affected individuals (II:4, III:1, and III:2) heterozygous for the R77S alteration (299, 196, 69, and 35 bp), but not in unaffected individuals (II:3, II:5; and III:3; 495, 69, and 35 bp) thus confirming the observed sequence variation and its segregation ([Figure 4](#f4){ref-type="fig"}). We excluded this variation R77S of *CRYGD* as a single-nucleotide polymorphism in a panel of 100 normal unrelated subjects of the same ethnicity. Alignment of amino acid sequences of *CRYGD* as per the [Entrez Protein](http://ncbi.nlm.nih.gov/entrez/) database revealed that arginine at codon position 77 is phylogenetically conserved across species ([Figure 5](#f5){ref-type="fig"}). {#f4} {#f5} Taken overall, the co-segregation of R77S was seen only in affected members of the pedigree (C-35) and its absence in 200 normal chromosomes strongly suggest that the non-conservative R77S substitution might be a causative mutation rather than a benign polymorphism to be associated with the disease. Sequencing of *CRYGD* gene in three affected (II:4, III:1, and III:2) and three unaffected individuals (II:3, II:5, and III:3) of C-35 family members showed four SNPs of which three ([NG_008039.1](http://www.ncbi.nlm.nih.gov/nuccore/NG_008039.1):g.5277T\>C; Y17Y; [rs2242074](http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=2242074), [NG_008039.1](http://www.ncbi.nlm.nih.gov/nuccore/NG_008039.1):g.7677A\>G; R95R; [rs2305430](http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=2305430), and [NG_008039.1](http://www.ncbi.nlm.nih.gov/nuccore/NG_008039.1):g.7929T\>C; [rs2305429](http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=2305429); 5′UTR) were reported earlier and a novel SNP E18F ([NG_008039.1](http://www.ncbi.nlm.nih.gov/nuccore/NG_008039.1):g.5278G\>A) was also observed. The inheritance of these five variations as a haplogroup in parents was analyzed ([Table 2](#t2){ref-type="table"}). It was found that the T-G-A-G-C haplogroup segregated with affected individuals showing transmission through paternal lineage. ###### Haplotype analysis of SNPs/mutation found in *CRYGD* gene in C-35 family members. ** ** **Genotype** ---------------------------------------------------------------------------------------- -------------- ---- ---- ---- ---- [NG_008039.1](http://www.ncbi.nlm.nih.gov/nuccore/190358502)
:g.5277T\>C; Y17Y CT CC CT CT CC [NG_008039.1](http://www.ncbi.nlm.nih.gov/nuccore/190358502)
:g.5278G\>A; E18F GG GA GG GG GA [NG_008039.1](http://www.ncbi.nlm.nih.gov/nuccore/190358502)
:g.5455C\>A; R77S CA CC CA CA CC [NG_008039.1](http://www.ncbi.nlm.nih.gov/nuccore/190358502)
:g.7677A\>G; R95R GG GA GG GG GA [NG_008039.1](http://www.ncbi.nlm.nih.gov/nuccore/190358502)
:g.7929T\>C;\-\-\-\-\-\-- CC CT CC CC CT Haplotypes involving the alleles at the SNP loci listed above, derived from pedigree of the family ([Figure 1A](#f1){ref-type="fig"}). Haplotype analysis shows the inheritance of T-G-A-G-C haplotype segregating with affected individuals. SNP loci are given in the order in which they occur in the reference sequence of the genomic contig. Prediction of mutational change on protein properties ----------------------------------------------------- Both normal and mutant proteins were analyzed for their structure. The R77S (in the processed protein) is situated in second Greek key motif in the linker region as the last amino acid before start of the next beta sheet. The isoelectric point (pI) was found to be almost same for both wild type (7.0) and mutant (6.58) proteins. Molecular weight of the mutant (20,669 Da) protein was similar to that of wild-type protein (20,738 Da). There was an increase in hydrophobicity at the mutant site and its neighborhood ([Figure 6](#f6){ref-type="fig"}). {#f6} The prediction of structural differences between wild-type and mutant proteins was performed using [PyMOL](http://www.pymol.org/) tool. In the 3D model of human CRYGD protein (PDB code 1HK0), it was observed that the arginine side chain being longer protrudes out and might interact with the glutamic acid at position 46 forming ionic bonds. In the mutant form serine having a shorter side chain and being surrounded by valine and serine do not maintain the strong bonding with glu46 which may unfold the protein from its original folding. Arginine has well spread electron density enabling for high solubility. Moreover, the increased surface area in the arginine may facilitate better interaction with solvents, thereby reducing the solubility of mutant protein with substitution of serine without much change in kinase activity. Discussion ========== Genes reported to cause cataract-specific mutations include those of the crystallins, cytoskeletal proteins, membrane proteins, transcription factors, glucosaminyl transferase 2 chromatin modifying protein-4B, and transmembrane protein 114 (TMEM114) \[[@r19]\]. Crystallin specific mutations account for 16.6% of inherited pediatric cataract in south India \[[@r20]\]. Among the reported genes, crystallins are of special interest because it encodes the major proportion of water soluble structural proteins of the lens fiber cells. In vertebrates, the α-, β-, and γ- crystallins are ubiquitous lens proteins. γ- Crystallin amounts for 25% of the total crystallins in the human lens. Both *CRYGC* and *CRYGD* are expressed at high concentrations in the fiber cells of the human embryonic lens which subsequently form the lens nucleus. The precise cellular micro-architecture and homeostasis are critical in maintaining the transparency of the lens. Alterations that impairs the proper solubility of lens proteins can lead to progressive protein aggregation that might act as light scattering centers in the lens \[[@r11]\]. Human γD-crystallin is a monomeric eye lens protein that must remain soluble throughout life for lens transparency. It is composed of two highly homologous beta-sheet domains which interact through interdomain side chain contacts forming two structurally distinct regions, a central hydrophobic cluster and peripheral residues. The specificity of domain interface interactions is likely important for preventing incorrect associations in the high protein concentrations of the lens nucleus \[[@r21]\]. Features of the interface between the two domains conserved among γ-crystallins are a central six-residue hydrophobic cluster, and two pairs of interacting residues flanking the cluster. In human γD-crystallin these pairs are Gln54/Gln143 and Arg79/Met147. It is suggested that these residues stabilize the native state by shielding the central hydrophobic cluster from solvent. In aged and cataractous lenses, glutamine and methionine side chains are among the residues covalently damaged. Such damage may generate partially unfolded, aggregation-prone conformations of human γD-crystallin that could be significant in cataract \[[@r22]\]. Many of the identified mutations in *CRYGD* seems to have been associated with autosomal dominant cataract phenotypes of congenital or juvenile nature. The eleven mutations identified so far occur as missense \[[@r10],[@r15],[@r23]-[@r34]\], nonsense \[[@r15],[@r34]-[@r36]\], and as frameshift \[[@r37]\] to have been associated with diverse phenotypes ([Table 3](#t3){ref-type="table"}). Three of the truncation mutations and a frameshift mutation seem to be associated exclusively with nuclear phenotype and all are congenital \[[@r15],[@r34]-[@r36]\]. The phenotype described here (C-35) is an anterior polar and coronary and the age at onset being within the first decade of life. As this mutation (R77S) involves the substitution of a highly basic and polar charged Arg for a less polar Ser residue it probably may not cause any major conformational change, as suggested earlier \[[@r32]\]. Along with the R77S mutation, four other variations as SNPs haplogroup show the transmission of T-G-A-G-C block from affected father to two affected children. The significance of this mutation, along with the prevalence of any specific haplotype, remains to be confirmed through analysis from a large number of familial cataracts. Predictions on structural changes in the mutant form of γD-crystallin using PyMOL reveals the shorter Ser side chain to establish weaker bonding with Glu46 that might result in unfolding. Furthermore, substitution with serine reduces the surface area of interaction with solvents thus hampering solubility of the mutant form as also revealed by increase in hydrophobicity ([Figure 6](#f6){ref-type="fig"}). This may therefore be considered as the causative mutation underlying the cataract phenotype in the family (C-35) investigated. Interestingly four of the *CRYGD* mutant alleles reported earlier also involves the residue Arg at different codon positions viz., 15, 37, 59, and 140 . In our study, by sequence alignment, the observed variation involves an arginine residue at position 77 which is highly conserved in γD-crystallin across various other species. Among cataract specific *CRYGD* mutations, the Arg residue has been replaced either by cysteine \[[@r10],[@r23]\], serine \[[@r28],[@r29],[@r32]\], Histidine \[[@r38]\], or for a stop codon \[[@r36]\]. Similar to the substitution in this study, the R37S mutation has been reported earlier to result in phenotypes with protein crystals in a Czech \[[@r28]\] and in a Chinese study \[[@r29]\]. Recently, in a Chinese family a corolliform cataract was found to be associated with an R15S allele of *CRYGD* \[[@r32]\]. It has been suggested that an increase in hydrophobicity and a putative phosphorylation site-mediated protein aggregation as the probable cause of opacification \[[@r32]\]. Both R15S and R37S results in congenital cataracts and fall within the I Greek key motif, while R77S falls within II Greek key motif and in association with a juvenile form of cataract in the C-35 family. Cataract-specific mutations involving the Arg residue has also been frequently reported in *CRYAA* \[[@r34],[@r35]\]. This suggests that Arg residues are more critical toward maintaining the structural and functional integrity of proteins. ###### Mutation spectrum of human *CRYGD* and cataract phenotypes in different childhood cataract families. **Exon** **Nucleotide** **Amino acid** **Inheritance** **Phenotype** **Ethnicity** **Ref** ---------- ---------------- ---------------------- ----------------- ----------------------------------------- ------------------------------------------- ------------ Ex2 c.43C\>T p.Arg15Cys (R15C) AD Punctate cataract, juvenile progressive Caucasian \[[@r10]\] Ex2 c.43C\>T p.Arg15Cys (R15C) AD Coralliform/nuclear Chinese \[[@r23]\] Ex2 c.43C\>A p.Arg15Ser (R15S) AD Coralliform Chinese \[[@r32]\] Ex2 c.70C\>A p.Pro24Thr (P24T) AD Lamellar Indian \[[@r15]\] Ex2 c.70C\>A p.Pro24Thr (P24T) AD Cerulean Moroccan \[[@r24]\] Ex2 c.70C\>A p.Pro24Thr (P24T) AD Coral-shaped, coralliform Caucasian \[[@r25]\] Ex2 c.70C\>A p.Pro24Thr (P24T) AD Flaky, silica-like nuclear cataract Australian pedigrees of European ancestry \[[@r26]\] Ex2 c.70C\>A p.Pro24Thr (P24T) AD Fasciculiform Chinese \[[@r27]\] Ex2 c.70C\>A p.Pro24Thr (P24T) AD Coralliform Chinese \[[@r32]\] Ex2 c.70C\>A p.Pro24Thr (P24T) AD Cerulean and Coralliform Saudi Arabian \[[@r33]\] Ex2 c.109C\>A p.Arg37Ser (R37S) AD with protein crystallization Czech boy \[[@r28]\] Ex2 c.109C\>A p.Arg37Ser (R37S) AD Nuclear golden crystal Chinese \[[@r29]\] Ex2 c.168C\>G p.Tyr56Stop (Y56X) AD Nuclear Brazilian \[[@r34]\] Ex2 c.176G\>A p.Arg59His (R59H) AD Aculeiform Macedonian \[[@r38]\] Ex2 c.181G\>T p.Gly61Cys (G61C) AD Coralliform Chinese \[[@r30]\] Ex2 c.229C\>A p.Arg77Ser (R77S) AD Anterior polar, Coronary Indian This study Ex3 c.320A\>C p.Glu107Ala (E107A) AD Nuclear Hispanic \[[@r31]\] Ex3 c.403C\>A p.Tyr134Stop (Y134X) AD No data Danish \[[@r35]\] Ex3 c.418C\>T p.Arg140Stop (R140X) AD Nuclear Indian \[[@r36]\] Ex3 c.470G\>A p.Trp157Stop (W157X) AD Nuclear Indian \[[@r15]\] Ex3 c.494delG p.Gly165fs AD Nuclear Chinese \[[@r37]\] In the 16 other families studied no putative mutation could be observed in the candidate genes screened which therefore makes it rather unlikely that the selected genes are involved in the cataract-forming process in these families. It prompts screening of other known candidate genes. This demonstrates that cataract need not arise only through point mutations but might be influenced also by many other factors, which may include unidentified modifier genes and other sequence variations. Detailed information about such factors and their precise role should enable one to understand the pathophysiology of cataracts and the biology of the lens in general. We are grateful to the family members for their participation in this study. This study was financially supported by Department of Biotechnology, Government of India under Indo-German Co-operation (BT/IN/FRG/JRS/2003-'04) and TIFAC-CORE in Pharmacogenomics, Department of Science and Technology, Government of India.

Introduction {#section5-1947603513515483} ============ Osteochondral lesions of the knee represent a large spectrum of injuries that range from softening and fibrillation of cartilage to complete loss of cartilage and bone. Techniques such as abrasion chondroplasty, drilling, or microfracture promote lesion infilling with fibrocartilaginous scar tissue. This tissue, which is predominantly composed of type I collagen, has poorer mechanical properties than the native hyaline cartilage.^[@bibr1-1947603513515483]^ Newer salvage techniques such as autologous osteochondral transplant, autologous chondrocyte implantation, and particulated juvenile articular cartilage have been developed in an attempt to repair the lesion with hyaline cartilage. This study describes the treatment performed on a 44-year-old woman who had previously undergone 2 meniscectomies and was still experiencing knee pain. *DeNovo* NT (Zimmer, Warsaw, IN) is a scaffold-free tissue, which can be used as a single-stage procedure for the repair of articular cartilage lesions and is typically sealed in place with fibrin glue. The tissue is procured from donors 13 years or younger since higher levels of type II collagen and proteoglycan production are typically seen in cartilage donors from this age group compared with adults.^[@bibr2-1947603513515483]^ To the best of our knowledge there have been no reported cases of equilibrium partitioning of an ionic contrast agent via micro--computed tomography (EPIC-µCT) analysis or histological findings from particulated juvenile articular cartilage allograft tissue treatment. Case Report {#section6-1947603513515483} =========== We detail a case of an active patient who initially presented to our clinic at the age of 44 years. She had complaints of left medial side knee pain and had already undergone 2 partial medial meniscectomies that had not resulted in any pain relief and so prompted our evaluation. The preoperative workup included an MRI and a long-leg frontal hip--knee--ankle alignment radiograph. She was noted to have varus malalignment and a full-thickness chondral lesion of the medial femoral condyle (MFC) measuring 22 mm wide by 36 mm long ([**Fig. 1A**](#fig1-1947603513515483){ref-type="fig"}). She was counseled on all surgical and nonsurgical options and elected to undergo a joint preserving salvage surgery. A high tibial valgus producing open wedge osteotomy was performed. At the same setting, the calcified subchondral layer was debrided and the chondral lesion was filled with particulated juvenile articular cartilage allograft (*DeNovo* NT) and sealed in place with fibrin glue. {#fig1-1947603513515483} The patient had improved pain after the procedure as her preoperative visual analogue scale (VAS) score was 8/10 and she had an improvement to 3/10. A routine follow-up MRI was performed at 7 months postoperation to evaluate healing which showed a residual lesion of 8 mm wide by 21 mm long on the medial aspect of the initial lesion ([**Fig. 1B**](#fig1-1947603513515483){ref-type="fig"}). The patient returned to the clinic 13 months after the procedure with complaints of worsening of medial side knee pain with her VAS pain score increased to 5/10 to 6/10 with activity. She was also experiencing pain at the site of the hardware. Another MRI was obtained that showed no change in lesion size from the previous MRI. In an attempt to continue with the salvage effort, the patient elected to proceed with the removal of hardware followed by an osteochondral autologous transfer to address the residual lesion in the MFC. At the time of surgery, we found the residual lesion to be 6 to 10 mm wide by 28 mm long, indicating a 72% surface area reduction compared with the original lesion. The area of the *DeNovo* NT tissue that had healed showed stable cartilage, which had an appearance and turgor comparable to the native surrounding cartilage. The area that had a residual osteochondral lesion was directly opposite the site where the previous partial menisectomy had been performed. Two 10 mm and one 6 mm diameter autologous osteochondral plugs were harvested from the lateral trochlea and used in a mosaicplasty fashion to fill the lesion ([**Fig. 1C**](#fig1-1947603513515483){ref-type="fig"}). Postoperatively, the patient's pain was decreased after the autologous osteochondral transfer, but she continued to have pain with moderate impact activities. A course of viscosupplementation injections also failed to improve her activity-related pain. At this time it was felt that joint preservation options had been exhausted. At approximately 28 months after her index procedure, the patient underwent a medial unicompartmental knee arthroplasty. At the time of unicompartmental knee arthroplasty, the area of the autologous osteochondral plugs, which was easily defined from the surrounding cartilage, appeared well healed and the cartilage appeared normal on gross examination ([**Fig. 1D**](#fig1-1947603513515483){ref-type="fig"}, photograph). The area of the *DeNovo* NT tissue also had good turgor, stability, and cartilage coverage. There was full-thickness cartilage loss on the MFC posterior to the area of previous cartilage repair. Samples of the tissue from the MFC, which included both areas of *DeNovo* NT tissue and autologous osteochondral plugs, were harvested. The tissue was subsequently fixed in 10% neutral buffered formalin at 4 °C for 48 hours. Once EPIC-µCT scanning was complete, the tissue was decalcified in Cal Ex II (Fisher Scientific, Waltham, MA) for 14 days and then divided into 4 segments ([**Fig. 1D**](#fig1-1947603513515483){ref-type="fig"}, µCT image) before paraffin embedding. The µCT images showed a heterogeneity of signal attenuation within the *DeNovo* NT tissue with areas of high attenuation being indicative of low proteoglycan content ([**Fig. 2A**](#fig2-1947603513515483){ref-type="fig"}). Histologically, the *DeNovo* NT tissue in segments 1 and 2 (i.e., in the vicinity of the plug tissue) stained positively for proteoglycans throughout and appeared hyaline-like under polarized light. In segments 3 and 4, furthest from the autologous plug tissue, the cartilage was metachromatic for Safranin O staining and had more disorganized areas of collagen under polarized light, suggesting fibrocartilage ([**Fig. 2B**](#fig2-1947603513515483){ref-type="fig"}). Four blinded reviewers were asked to grade both *DeNovo* NT and autologous tissue in the 4 segments using a modified OsScore index^[@bibr3-1947603513515483]^ for histological parameters (maximum score of 9 as mineralization could not be assessed because of tissue decalcification). The results of this grading are shown in [**Figures 2C**](#fig2-1947603513515483){ref-type="fig"} and [**2D**](#fig2-1947603513515483){ref-type="fig"}. There was an indication that the *DeNovo* NT tissue graft in segments 1 and 2, close to the autologous cartilage plugs, better maintained its hyaline nature. ![(**A**) Micro--computed tomography (µCT) image (representing segments 1 and 4) shows heterogeneity of signal attenuation in *DeNovo* NT tissue (blue/green = low attenuation corresponding to high proteoglycan \[PG\] content, red = high attenuation corresponding to low PG content). Tissue was submerged in 30% Hexabrix 320 contrast agent (Covidien Pharmaceuticals, Hazelwood, MO) in phosphate-buffered saline (PBS), for 24 hours at room temperature and scanned at *E* = 45 kVp, *I* = 177 µA, 300 ms integration, 36 mm field of view, 1024 × 1024 pixel matrix (36 µm voxel size) with µCT40 (Scanco Medical, Brüttisellen, Switzerland). Following automatic image reconstruction to grayscale 2-dimensional (2D) tomograms, 3D grayscale stacks containing linear attenuation data were rotated to match the histological sectioning plane. Pseudocolor scales were applied to demonstrate variations in linear attenuation coefficient, previously shown to be inversely proportional to PG content.^[@bibr4-1947603513515483]^ (**B**) Sagittal sections (5 µm thick) stained for hematoxylin and eosin (H&E), Safranin O (segment 1 top, segment 4 bottom). Scale bar = 1 mm. Deparaffinized sections were mounted unstained in Permount (Fisher Scientific, Waltham, MA) and viewed under polarized light.^[@bibr5-1947603513515483]^ (**C**) Graph of modified OsScores^[@bibr3-1947603513515483]^ of histological parameters in both *DeNovo* NT and autologous plug tissue for the four segments. A maximum score of 9 possible as mineralization could not be assessed because of tissue decalcification. Key: Tissue morphology: hyaline = 3, hyaline/fibrocartilage = 2, fibrocartilage = 1, fibrous tissue = 0 (H&E, polarized light). Matrix staining: near normal = 1, abnormal =0 (Saf 0). Surface architecture: near normal = 2, moderately irregular = 1, very irregular = 0 (H&E). Extent of chondrocyte clustering: none = 1, ≤25% = 0.5, \>25% = 0 (H&E). Presence of vascularization: absent = 1, present = 0 (H&E). Integration of calcified cartilage and underlying bone: good = 1, poor = 0 (polarized light). Adapted from Roberts *et al*^[@bibr3-1947603513515483]^). (**D**) Table of individual parameters constituting OsScore index for plug tissue and *DeNovo* NT tissue segments 1 to 4. Results are displayed as means ± standard deviation and show interobserver variability as there was only one sample.](10.1177_1947603513515483-fig2){#fig2-1947603513515483} Discussion {#section7-1947603513515483} ========== The purpose of this study was to evaluate the medium-term histologic outcome of *DeNovo* NT graft, used for the repair of cartilage in the knee. Despite showing good healing at 7 months postsurgery by MRI, and the patient experiencing reduced pain at 7 and 13 months, the excised cartilage tissue was heterogeneous, with some regions of hyaline-like cartilage and some regions of fibrocartilage. Further surgical intervention was associated with the persistence of a residual chondral lesion opposite the area of the previous meniscectomy, which was resurfaced with the autologous plugs. This suggests that an intact meniscus may be an important factor in the success of this procedure. A direct comparison of the *DeNovo* NT tissue to the autologous plug tissue was not entirely appropriate as the plugs were only implanted for 15 months whereas the *DeNovo* NT was implanted for more than 2 years. It does, however, appear that the later mosaicplasty may have helped maintain hyaline-like cartilage of the *DeNovo* NT tissue in its vicinity. Limitations of this study include a paucity of excised *DeNovo* NT tissue samples. Collection of data from a larger study population with longer follow-up times would provide more information on long-term efficacy. *DeNovo* NT tissue has been commercially available since 2007 and, as of March 2011, has been implanted in more than 2,200 patients in the United States.^[@bibr6-1947603513515483]^ Preliminary reports have shown promise. In a cohort of 4 patients reporting 24 months after *DeNovo* NT implantation, MRIs showed lesion infilling and pain indices were improved^[@bibr7-1947603513515483]^ and in a case report of a *De Novo* NT Cartilage implantation in the talus,^[@bibr8-1947603513515483]^ the patient was no longer experiencing ankle pain at 2 years postoperatively. In this case, the *DeNovo* NT tissue provided a stable, if heterogeneous, tissue on the lateral side of the osteochondral lesion. The differences between MRI and follow-up histology highlight the difficulties in assessment of cartilage repair surgery. This juvenile tissue graft is still offered to patients who meet the indications for cartilage repair and the multicenter *DeNovo* NT Longitudinal Data Collection (LDC) Knee Study ending in 2018 should provide a wealth of information on treatment outcomes. **Acknowledgments and Funding:** The author(s) received no financial support for the research, authorship, and/or publication of this article. **Declaration of Conflicting Interests:** The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Dr. Labib is a principal investigator on a *DeNovo* NT outcome study. All analysis for this study was done by nonconflicted authors. **Ethical Approval:** This study was approved by the Emory University Institutional Review Board.

1. Introduction {#sec1-vaccines-07-00191} =============== Gonorrhoea is a sexually transmitted disease estimated to cause at least 78 million new cases worldwide per year \[[@B1-vaccines-07-00191]\]. The causal organism, *Neisseria gonorrhoeae,* is listed as a high priority pathogen for research into novel treatments by the WHO \[[@B2-vaccines-07-00191]\] due to its ability to rapidly develop resistance to antibiotics \[[@B3-vaccines-07-00191]\]. Isolates with resistance to the last recommended treatment combining ceftriaxone with azithromycin have been detected in several individuals from the United Kingdom \[[@B4-vaccines-07-00191],[@B5-vaccines-07-00191]\] and Australia \[[@B6-vaccines-07-00191]\] shortly after overseas travel, emphasising the potential for global spread of intractable or even incurable *N. gonorrhoeae.* Infections are typically self-limiting and restricted to mucosal sites, however, untreated infection of females can lead to pelvic inflammatory disease, infertility and ectopic pregnancies \[[@B7-vaccines-07-00191]\]. Sexually transmitted infections including *N. gonorrhoeae* have also been linked with a significantly greater risk of contracting HIV \[[@B8-vaccines-07-00191]\]. Infection with *N. gonorrhoeae* results in an initial neutrophilic inflammatory response at the site of infection \[[@B9-vaccines-07-00191],[@B10-vaccines-07-00191]\] and limited, short-lived humoral responses \[[@B11-vaccines-07-00191]\]. Experimental infection of human subjects \[[@B12-vaccines-07-00191]\] and longitudinal studies of high risk individuals \[[@B13-vaccines-07-00191],[@B14-vaccines-07-00191]\] show that acquisition and subsequent clearance of an infection does not protect against further infections with *N. gonorrhoeae*. Humoral responses to highly variable but abundant cell wall components such as lipooligosaccharides (LOS) and pili, as well as cellular responses that are skewed towards T helper (Th) 17 and regulatory T (T~reg~) cells, are reported to be elevated after recent infection (reviewed in \[[@B15-vaccines-07-00191]\]). Recent studies in mice suggest that skewing mucosal responses away from Th17/T~reg~ axis towards Th1/cellular immunity leads to improved clearance of *N. gonorrhoeae* \[[@B16-vaccines-07-00191],[@B17-vaccines-07-00191]\]. Multiple pre-clinical vaccine candidate antigens have been identified as having promising activity in mouse models (reviewed in \[[@B18-vaccines-07-00191]\]); of note, a peptide mimic that targets a highly conserved epitope of LOS shows particular potential but has not yet been tested in humans (reviewed in \[[@B19-vaccines-07-00191]\]). To date, those vaccines tested in humans have largely failed to stimulate protection from infection \[[@B20-vaccines-07-00191],[@B21-vaccines-07-00191]\]. Detection of serum bactericidal antibody (SBA) is a key correlate of vaccine-induced protective immunity for *N. meningitidis* \[[@B22-vaccines-07-00191]\], a close relative of *N. gonorrhoeae* \[[@B23-vaccines-07-00191]\], and considered likely to be important for protection against *N. gonorrhoeae*. The SBA assay measures killing of bacteria after exposure to an exogenous source of complement and defined dilutions of immune antibodies, which leads to cell lysis by activation of the classical complement pathway \[[@B24-vaccines-07-00191]\]. Normal human serum (NHS) is often used in SBA assays as a source of complement, and a key component of assay validation is confirmation that addition of NHS alone does not initiate killing of bacteria. This is particularly important because *N. gonorroheae* clinical isolates show variable sensitivity to NHS, with those strains associated with disseminated disease more likely to be serum-resistant \[[@B25-vaccines-07-00191],[@B26-vaccines-07-00191]\], as are those freshly isolated from clinical specimens \[[@B27-vaccines-07-00191]\]. The conventional methodology for detection of SBA or serum sensitivity of gonococci or meningococci uses enumeration by colony counting, which is labour intensive, requires an overnight incubation step and relies on large quantities of agar plates \[[@B28-vaccines-07-00191]\]. Use of a commercially available reagent for detecting bacterial ATP in microtitre plates using a luminescent readout as a surrogate for bacterial viability/colony forming units (CFU) has been described for measurement of SBA to *N. meningitidis* serogroups A and W, as well as several other pathogenic bacteria \[[@B29-vaccines-07-00191],[@B30-vaccines-07-00191]\]. We describe the utility of this approach to measure serum sensitivity and SBA to gonococci. 2. Materials and Methods {#sec2-vaccines-07-00191} ======================== 2.1. Culture of Bacteria {#sec2dot1-vaccines-07-00191} ------------------------ *N. gonorrhoeae* strains FA1090 (ATCC 700825), MS11 (ATCC BAA-1833) and P9-17 \[[@B27-vaccines-07-00191]\] were used in this study. Fresh cultures of bacteria were prepared from frozen stocks by streaking onto gonococcal agar (GCA) consisting of gonococcal (GC) agar base supplemented with 5 g/L bovine dried haemoglobin and 1% v/v IsoVitaleX (BD Biosciences, Franklin Lakes, NJ, USA). Bacteria were grown overnight at 37 °C, 5% CO~2~, then streaked onto fresh GCA and cultured for 6 h to produce mid-log cultures. In experiments using cytidine-5′-monophospho-N-acetylneuraminic acid (CMP-NANA, Sigma C8271), bacteria from overnight cultures were used to inoculate GC-proteose peptone broth (GCB) supplemented with 1% v/v IsoVitaleX with/without 4 µg/mL CMP-NANA \[[@B31-vaccines-07-00191]\] and incubated at 37 °C with shaking for approximately 3 h until an A~600~ of 0.45 (\~2 × 10^8^ CFU/mL) was reached. 2.2. Human and Murine Sera {#sec2dot2-vaccines-07-00191} -------------------------- Normal human serum (NHS) from healthy human donors was used as a source of complement for serum sensitivity and SBA assays. Peripheral blood was obtained by venipuncture under The University of Auckland Human Participants Ethics Committee approval, reference 021200. All donors gave informed written consent prior to blood sampling. Blood was allowed to clot for a maximum of 30 min, serum collected after centrifugation (1250 × g, 20 min, 4 °C) and aliquots stored at −80 °C. NHS was diluted to 16.7% v/v for use as a complement source, which is within the recommended range of 10%--20% v/v for gonococcal bactericidal assays \[[@B32-vaccines-07-00191]\]. NHS was heat-inactivated at 56 °C for 30 min before use (hiNHS) as a complement inactivated control. Anti-sera to gonococcal outer membrane vesicles (OMV) were raised in mice. Preparation of detergent-extracted *N. gonorroheae* OMV extracts from P9-17, FA1090 and MS11 was on the basis of the methodology described for isolation of *N. meningitidis* OMVs for human vaccination \[[@B33-vaccines-07-00191]\], with minor adaptations \[[@B34-vaccines-07-00191]\]. *N. gonorrhoeae* was grown overnight in GCB at 37 °C, 5% CO~2~ with gentle shaking, followed by incubation at 40 °C for 2 h. Briefly, bacteria were harvested by centrifugation and the pellet treated twice for 30 min with 0.1 M Tris-HCl pH 8.6, 10 mM EDTA and 0.5% w/v sodium deoxycholate. The supernatants underwent ultracentrifugation (100,000 × g, 2 h, 4 °C), then the pelleted material was suspended in PBS, filter sterilized and stored at 4 °C. The protein content of the OMV extracts was quantified with a Pierce BCA Protein Assay Kit (Thermo Fisher Scientific, Auckland, New Zealand). Groups of three female specific pathogen free CD1 mice aged 5--6 weeks were immunised via the intra-peritoneal route with 4 µg of *N. gonorrhoeae* OMV extract adsorbed to alum adjuvant (Alu-Gel-S, 2%, Serva) on days 0, 14 and 28. Baseline blood samples were collected prior to vaccination and immune serum was collected 7 days after the final immunisation. Blood was collected into microvette 500 Z-gel tubes (Sarstedt), and serum was harvested after 1 hour and stored at −80 °C. Immune sera were pooled for the SBA and were heat-inactivated before use as a source of anti-gonococcal immune antibodies in the SBA assay. Animal work was approved by The University of Auckland Animal Ethics Committee (protocol 1816) and was conducted in accordance with the university's Code of Ethical Conduct and the Animal Welfare Act 1999. Mice were sourced from the Vernon Jansen Unit (University of Auckland, Auckland, New Zealand), monitored daily by qualified staff and suffered no adverse effects from these manipulations. 2.3. Comparison of CFU and ATP Readouts {#sec2dot3-vaccines-07-00191} --------------------------------------- *N. gonorrhoeae* was grown to mid-log on GCA and suspended in Dulbecco's modification of PBS (PBSB) at a UV A~260~ of 1 (\~4.2 × 10^9^ CFU/mL), further diluted to 4 × 10^8^ CFU/mL, then serial diluted in PBSB and incubated for 1 h at 37 °C, 5% CO~2~ in round-bottomed microtitre plates before measurement of ATP by luminescence or determination of CFU. The CFU of the starting inoculum was to confirm that bacteria numbers remained unaltered after 1 h in PBSB. For enumeration of colonies, 15 µL of the initial suspension was diluted, spread onto GCA without haemoglobin in triplicate using a sterile loop and incubated overnight at 37 °C, 5% CO~2~. Quantification of bacteria by measurement of ATP was determined with BacTiter-Glo (Promega) prepared as per the manufacturer's instructions. After the 1 h incubation step, the microtitre plate was centrifuged at RT for 10 min at 3220 × g, supernatant was removed from each well and the pelleted bacteria was re-suspended in PBSB. Half of this suspension was transferred to a 96-well white flat bottom microtitre plate (Greiner Bio-One, Kremsmünster, Austria) and an equal volume of BacTiter-Glo reagent added to each well. The reaction was incubated for 5--10 min at RT on an orbital shaker before the luminescence signal was read using a Victor X Light 2030 Luminescence Reader (PerkinElmer, Waltham, MA, USA). Each plate was read three consecutive times and the average counts per second (CPS) for each reaction recorded. The background signal from a blank well containing PBSB was subtracted from each reading. 2.4. Measurement of Serum Sensitivity {#sec2dot4-vaccines-07-00191} ------------------------------------- *N. gonorrhoeae* was grown to mid-log on GCA and suspended in Dulbecco's modification of PBS (PBSB) as detailed above. Reaction mixes containing \~1 × 10^3^ CFU *N. gonorrhoeae* and 16.7% v/v NHS or heat-inactivated NHS control for every donor were adjusted to a final volume of 0.1 mL in PBSB-hiFBS (PBSB-1% v/v heat-inactivated foetal bovine serum (Thermo Scientific, Waltham, MA, USA)). The reaction mix was incubated for 1 h at 37 °C, 5% CO~2~ in round-bottomed microtitre plates and bacteria enumerated as per [Section 2.3](#sec2dot3-vaccines-07-00191){ref-type="sec"}. Reaction mixes for ATP-based measurement of bacterial killing were identical to the CFU-based method, except that \~5 × 10^3^ CFU bacteria were used in the reaction mix for this assay; downstream processing was as per [Section 2.3](#sec2dot3-vaccines-07-00191){ref-type="sec"}. Serum sensitivity was calculated as a percentage of CFU or CPS signal in the presence of NHS compared to hiNHS. Values were reported as zero when bacteria numbers increased in the presence of NHS. 2.5. Quantification of SBA {#sec2dot5-vaccines-07-00191} -------------------------- Measurement of SBA by CFU was determined as previously described \[[@B35-vaccines-07-00191]\], with minor modifications for quantification of bacterial ATP by luminescence. *N. gonorrhoeae* was grown to mid-log on GCA and suspended in Dulbecco's modification of PBS (PBSB). Reaction mixes containing \~1 × 10^3^ CFU *N. gonorrhoeae* (CFU) or \~5 × 10^3^ CFU (ATP), 16.7% v/v NHS and 10% v/v diluted heat-inactivated murine anti-sera were adjusted to a final volume of 0.1 ml in PBSB-hiFBS. The murine anti-sera was diluted to a starting titre of either 1/16 or 1/32, then diluted twofold in PBSB to 1/2048. NHS- and hiNHS-only assay control reactions were included in every assay. The reaction mix was incubated for 1 hour at 37 °C, 5% CO~2~ in round-bottomed microtitre plates, and bacteria were enumerated as per [Section 2.3](#sec2dot3-vaccines-07-00191){ref-type="sec"}. Bacterial survival was calculated by expressing test CFU or CPS values as a proportion of the maximum CFU or CPS value (i.e., the NHS-only reaction). Survival values calculated as \>100% were reported as 100%. The titre was defined as the reciprocal of the interpolated serum dilution that killed 50% of the bacteria in comparison to the NHS only reaction. Titres were log-transformed for plotting and determination of exact SBA titres. These were calculated using Opsotiter3, an Excel-based data processing program from The University of Alabama at Birmingham reference library (UABRF) and licenced from UABRF \[[@B36-vaccines-07-00191],[@B37-vaccines-07-00191],[@B38-vaccines-07-00191]\]. 2.6. Statistical Analysis {#sec2dot6-vaccines-07-00191} ------------------------- All data were from three independent biological repeats. The results from each biological repeat are either shown individually on a single plot or combined to display mean ± standard deviation. All statistical analyses were carried out in GraphPad Prism Version 7.03 (GraphPad Software, Inc., San Diego, CA, USA); the tests applied are indicated in the figure legends. 3. Results {#sec3-vaccines-07-00191} ========== 3.1. Comparison of Luminescent and CFU Readouts {#sec3dot1-vaccines-07-00191} ----------------------------------------------- To determine whether BacTiter-Glo is a viable choice to quantify *N. gonorrhoeae* survival, initial experiments compared ATP readings from three *N. gonorrhoeae* isolates across a broad range of CFU (\~200--10^7^ CFU/well). The ATP (luminescent) signal rose with increased *N. gonorrhoeae* CFU across the range tested ([Figure 1](#vaccines-07-00191-f001){ref-type="fig"}). There was a high correlation between CFU and ATP levels for all strains across the full range of concentrations tested (*r* \> 0.9), with the most reliable readouts in the range of 2 × 10^3^--1 × 10^6^ CFU/well for each of the three *N. gonorrhoeae* strains tested ([Figure 1](#vaccines-07-00191-f001){ref-type="fig"}). All subsequent experiments used \~5 × 10^3^ CFU bacteria/well for the ATP assay to ensure a strong signal above background. 3.2. Use of Bioluminescent Assay to Screen for NHS Sensitivity {#sec3dot2-vaccines-07-00191} -------------------------------------------------------------- Sera from 10 NHS donors were screened for bactericidal activity against *N. gonorrhoeae* P9-17, MS11 and FA1090 using both CFU enumeration and ATP readouts in parallel. Reported levels of bacterial killing were either lower or equivalent in the ATP assay relative to enumeration of CFU in all three strains ([Figure 2](#vaccines-07-00191-f002){ref-type="fig"}). The P9-17 strain had low levels of serum-mediated killing (0%--30%) after exposure to sera from all 10 donors and both methods produced similar results ([Figure 2](#vaccines-07-00191-f002){ref-type="fig"}). In contrast, serum sensitivity results from the two assays diverged significantly for the MS11 strain (*p* \< 0.05 for all donors) and the FA1090 strain (*p* \< 0.05 for 5/10 donors). Of particular note, serum from donors 3 and 4 elicited little or no bactericidal activity in the ATP assay, but between 60%--90% killing after enumeration of CFU from both *N. gonorrhoeae* FA1090 and MS11. On the basis of these results, the P9-17 and MS11 strains had the best profiles for use in ATP-based SBA assays, representing resistant and susceptible strains, respectively. Unless specified otherwise, serum from donor 2 was used as a source of complement for all subsequent assays, in combination with the *N. gonorrhoeae* P9-17 strain. 3.3. Quantification of SBA by Detection of ATP {#sec3dot3-vaccines-07-00191} ---------------------------------------------- Assays to determine the SBA of murine anti-sera raised to *N. gonorrhoeae* P9-17 OMV extracts against *N. gonorrhoeae* P9-17 were carried out using the CFU or ATP method in parallel to directly compare the outcome. Each individual test used the same stock of bacteria and reaction mix. Bacteria were re-suspended in a volume of 100 µL buffer and 50 µL of this suspension used for detection of ATP in initial feasibility tests for this assay ([Figure 1](#vaccines-07-00191-f001){ref-type="fig"}). However further testing showed that comparable results were obtained from bacteria re-suspended in 30 µL buffer, with 25 µL of this suspension used for detection of ATP ([Figure 3](#vaccines-07-00191-f003){ref-type="fig"}). This adjustment meant that a larger proportion of bacteria from the reaction mix were used for reading out luminescence (\>80% vs. 50%) and yielded similar results (*r* = 0.98). There were no significant differences between any paired sets of data, but standard deviations were reduced and detection of bacterial viability at serum dilutions of 1/32 and 1/64 was improved with the smaller re-suspension volume. This modification reduced the use of a costly reagent without affecting the data obtained, and therefore was adopted for all further assays. A comparison of the range of raw values obtained with the ATP method versus colony counting readouts is shown in [Figure 4](#vaccines-07-00191-f004){ref-type="fig"}A. Detection of ATP via luminescence gave readouts ranging from an average of 1552 CPS at a serum dilution of 1/32 through to 6756 at a dilution of 1/2048. Enumeration by colony count gave values of 11 to 126 CFU at the same dilutions. Control wells comprising bacteria incubated in NHS alone had average values of 6755 CPS and 126 CFU, respectively. There was a degree of variability in the range of raw values, particularly for the ATP assay ([Figure 4](#vaccines-07-00191-f004){ref-type="fig"}A), but importantly readouts of bacterial survival remained consistent between assays ([Figure 4](#vaccines-07-00191-f004){ref-type="fig"}B). The average SBA titre obtained with the CFU assay was higher than the ATP assay (223 vs. 134) and the average maximum killing was also increased in the CFU assay relative to the ATP assay (92% vs. 78%) ([Figure 4](#vaccines-07-00191-f004){ref-type="fig"}B). However, a direct comparison of quantification of SBA titres by CFU and ATP showed a strong correlation between the two methods (*r* = 0.97, *p* \< 0.001). Although bacterial survival was higher at all dilutions using the ATP method, a significant difference between the assays was only evident at 1/64, *p* \< 0.05 ([Figure 4](#vaccines-07-00191-f004){ref-type="fig"}B). The ATP assay was performed with approximately five-fold more bacteria, which may partially account for the trend towards a lower titre and high bacterial survival. Antisera to OMV preparations from the other *N. gonorrhoeae* isolates were tested for cross-reactive SBA using the *N. gonorrhoeae* P9-17 ATP SBA assay; the anti-P9-17 OMV SBA data shown in [Figure 4](#vaccines-07-00191-f004){ref-type="fig"} were included as a point of comparison. A twofold titration starting at 1/16 showed no killing activity with sera against FA1090 OMVs (bacterial survival \>80% at all dilutions tested), whereas anti-sera against MS11 OMVs had moderate cross-reactive bactericidal activity against *N. gonorrhoeae* P9-17 (endpoint titre = 56, maximum killing = 67%) ([Figure 5](#vaccines-07-00191-f005){ref-type="fig"}). Sialylation of lipooligosaccharides (LOS) is linked with the resistance of *N. gonorrhoeae* to serum-mediated killing by preventing activation of the complement cascade \[[@B31-vaccines-07-00191],[@B39-vaccines-07-00191]\]; this is a critical variable to consider when determining potential bactericidal activity of anti-gonococcal immune sera. The FA1090 and MS11 strains, both of which showed a degree of sensitivity to sera from donors 1 and 2 ([Figure 2](#vaccines-07-00191-f002){ref-type="fig"}), were chemically sialylated with 4 µg/mL CMP-NANA to verify that reduced bacterial killing could be detected using quantification of ATP as a readout. Treatment with CMP-NANA prevented killing of *N. gonorrhoeae* MS11 ([Figure 6](#vaccines-07-00191-f006){ref-type="fig"}A; *p* \< 0.05 for donor 1 and *p* \< 0.01 for donor 2) and reduced NHS mediated killing of *N. gonorrhoeae* FA1090 by approximately 50% ([Figure 6](#vaccines-07-00191-f006){ref-type="fig"}B; *p* = 0.203 for donor 1 and *p* \< 0.05 for donor 2). A reduction in SBA activity after sialylation of the P9-17 strain was also detected using the ATP assay (53% vs. 14.5% killing at a serum dilution of 1/64, *p* \< 0.01; [Figure 6](#vaccines-07-00191-f006){ref-type="fig"}C). 4. Discussion {#sec4-vaccines-07-00191} ============= Detection and quantification of *N. gonorrhoeae* ATP using a commercially available luminescent substrate showed utility as an alternative approach to manual enumeration of CFU for measuring SBA to *N. gonorrhoeae*. In contrast, screening of NHS from multiple donors using the ATP method under-estimated serum sensitivity of the FA1090 and MS11 strains relative to the CFU method, suggesting that it is unsuited to this particular assay. These results indicate that detection of bacterial viability by ATP is most appropriate for assays where rapid killing of bacteria is expected, such as SBA, where killing is initiated by rapid activation of the classical complement pathway, leading to formation of C5b9 and lysis of gram-negative bacteria, a process that occurs within minutes of exposure \[[@B40-vaccines-07-00191]\]. To the best of our knowledge, this is the first report detailing the use of an ATP assay to measure SBA to *N. gonorrhoeae*. In line with an earlier report describing the utility of this reagent for other bacteria \[[@B29-vaccines-07-00191]\], use of this method to measure SBA offers multiple advantages, including a substantially reduced time to acquire results (3 h vs. \>24 h), a sizeable decrease in hands-on time by laboratory personnel and the capacity to test large sets of samples in a single day. The cost of reagents and consumables for the ATP assay is lower than the conventional plate-based method. This assay could also be applied to other tests that rely on measuring *N. gonorrhoeae* viability, such as screening of antimicrobials, to streamline testing and identification of new drugs. The *N. gonorrhoeae* strains used in this study were chosen because they are frequently used for identification or testing of novel vaccine candidate antigens \[[@B35-vaccines-07-00191],[@B41-vaccines-07-00191],[@B42-vaccines-07-00191]\], for understanding gonococcal virulence \[[@B43-vaccines-07-00191]\] and for use as challenge strains for murine \[[@B44-vaccines-07-00191],[@B45-vaccines-07-00191],[@B46-vaccines-07-00191]\] or human models of infection \[[@B9-vaccines-07-00191],[@B10-vaccines-07-00191]\]. Detection of bacteria using CFU versus ATP readouts produced a similar outcome with all three strains, suggesting that comparable results are likely to be obtained with other *N. gonorrhoeae* isolates. Studying clinically relevant strains will be important for determining potential strain-dependent variability of SBA in humans. The P9-17 strain was the most tractable strain in this study---it showed little or no serum sensitivity to NHS from multiple donors when measured by the CFU or ATP methods. In comparison the MS11 and FA1090 strains, both showed sensitivity to many or all of these sera under the conditions used, with higher levels of killing observed when bacterial survival was quantified by CFU. These results emphasise that NHS donors need to be screened carefully and/or that a smaller quantity of NHS should be used as a source of complement for assays using these strains. The FA1090 strain showed a surprising degree of serum susceptibility, given it is widely reported as a serum-resistant strain \[[@B10-vaccines-07-00191]\]. However, serum resistance can be lost with in vitro passaging \[[@B27-vaccines-07-00191],[@B47-vaccines-07-00191]\], which is the most likely explanation for this unexpected result. Alternatively, a portion of our NHS donors may have cross-reactive bactericidal antibodies to some strains of *N. gonorrhoeae*. Of note, a large proportion of New Zealand's population (\~1 million) received an OMV-based vaccine to *N. meningitidis* serogroup B (MenB) from 2004 to 2008 \[[@B48-vaccines-07-00191]\], and some of these individuals may have entered our pool of NHS donors. The NHS used in this study was not pre-treated to eliminate cross-reactive antibodies \[[@B28-vaccines-07-00191]\], but is an important consideration for future studies. Alternatively, it was demonstrated that serum sensitivity of both these strains could be reduced by sialylation with CMP-NANA, with the caveat that this is likely to inhibit SBA mediated killing. Reduced susceptibility to SBA-mediated killing was shown with *N. gonorrhoeae* P9-17 sialylated with CMP-NANA relative to untreated bacteria. Determination of bacterial survival by the ATP method was sufficient to show development of higher SBA titres to the P9-17 strain with anti-sera raised to homologous OMVs, compared to an intermediate or no detectable cross-reactive SBA with anti-sera against MS11 or FA1090 OMVs, respectively. The homologous anti-P9-17 SBA response reported here is similar to a previous study, which indicated that vaccination of mice with sodium deoxycholate-extracted OMVs from P9-17 stimulated a SBA titre of 256 when delivered in alum adjuvant, compared to a titre of \>1000 with OMVs prepared without detergent \[[@B35-vaccines-07-00191]\]. Some limitations of the ATP approach were observed. Reliable quantification of *N. gonorrhoeae* required the preparation of a reaction mix containing a higher number of bacteria (\~5 × 10^3^ CFU) to ensure luminescence readouts were well above background. This was partially because a portion of the re-suspended bacteria had to be transferred into another plate and combined with an equal volume of substrate for detection of ATP by luminescence. Results obtained with this higher number of bacteria were compared to a well-established CFU-based SBA that used \~1 × 10^3^ CFU \[[@B35-vaccines-07-00191]\]. The reduced serum sensitivity and antibody-mediated bacterial killing detected with the ATP assay was likely due to the addition of approximately five-fold more bacteria. Use of higher numbers of bacteria may have been responsible for the variation in serum-sensitivity between the ATP and CFU readouts, potentially by reducing or delaying non-specific serum mediated killing. Therefore, in its current format, the ATP assay under-estimates serum sensitivity of *N. gonorrhoeae* and should not be used in place of the traditional CFU-based method. However, despite the addition of higher numbers of bacteria for the ATP assay, detection of bacterial survival after incubation with bactericidal anti-sera was not significantly altered when compared to the conventional CFU-based method. Successful detection of SBA in sera from human subjects infected or recovering from infection with gonorrhoea has been reported using up to \~10^7^ CFU without a substantial loss of sensitivity \[[@B49-vaccines-07-00191]\]. Another significant difference between the standard CFU method and measurement of ATP is in processing time. *N. gonorrhoeae* and anti-sera are typically combined for 30--60 minutes prior to plating for CFU \[[@B28-vaccines-07-00191]\]. Plating for determination of CFU can be time-consuming, particularly if a large-scale assay is performed. Until transfer to agar plates and incubation, the bacteria remain in sub-optimal conditions for survival and continue to be exposed to the active components of the reaction. In contrast, the bacteria are pelleted and lysed for immediate quantification of ATP, a process that takes approximately 15 min, and therefore there is no potential for additional killing to occur. These differences in processing may partially explain the enhanced bacterial survival observed when serum sensitivity was compared using both ATP and CFU readouts. It has recently been observed that MenB vaccines reduce rates of gonorrhoea at a population level \[[@B50-vaccines-07-00191],[@B51-vaccines-07-00191],[@B52-vaccines-07-00191],[@B53-vaccines-07-00191]\], which now provides the impetus to retrospectively examine historic resources, such as samples from trials with New Zealand's MeNZB vaccine \[[@B54-vaccines-07-00191]\], as well as materials generated from more recent vaccine trials, to understand possible mechanisms of protection. This assay can be used to aid the global effort to develop a vaccine for gonorrhoea by facilitating testing of sera for possible cross-reactive SBA conferred by MenB OMV vaccines or for testing novel vaccine candidate antigens for bactericidal effects. 5. Conclusions {#sec5-vaccines-07-00191} ============== Measurement of *N. gonorrhoeae* survival by detection of ATP in a luminescent assay produced comparable results to manual plating and quantification of bacterial CFU for determination of SBA. Luminescent assays have the advantage of being high throughput, with a fast turnaround, offering the prospect of being able to screen large banks of samples for bactericidal activity relatively quickly. This approach could also be applied to contemporary clinical isolates of *N. gonorrhoeae*, including those with known anti-microbial resistance profiles, for rapid testing of new drugs. The development of new assays to study *N. gonorrhoeae* is a high priority because multi-drug resistant strains are a global concern and there is an urgent need to identify alternative treatments such as novel antibiotics or a vaccine. Conceptualization, M.C., F.J.R.; methodology, F.C., C.J.O.H, M.C., F.J.R.; validation, F.C.; formal analysis, F.C., F.J.R.; resources, M.C., F.J.R.; data curation, F.C., F.J.R.; writing---original draft preparation, F.C., F.J.R.; writing---review and editing, F.C., C.J.O.H., M.C., and F.J.R.; visualization, F.C., F.J.R.; supervision, M.C., F.R.; project administration, F.J.R.; funding acquisition, F.J.R. This study was supported by the Royal Society of New Zealand Te Apārangi Catalyst: Seeding Fund \[grant number 17-UOA-059-CSG\]. GlaxoSmithKline provide support for a PhD student of M.C. The remaining authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results. {#vaccines-07-00191-f001} {#vaccines-07-00191-f002} {#vaccines-07-00191-f003} {#vaccines-07-00191-f004} {#vaccines-07-00191-f005} {#vaccines-07-00191-f006}

Introduction ============ Anthropogenic debris is accumulating at a rapid rate in coastal and oceanic ecosystems worldwide ([@ref-7]). The most significant contributor to these growing marine litter deposits is plastic ([@ref-21]), a material widely used by humans ([@ref-11]). In the marine environment, plastic is initially buoyant, and easily dispersed over long distances via wave action and wind ([@ref-8]). Floating plastic is detrimental to both ecosystem nutrient cycling and marine wildlife, as it can absorb and secrete chemicals ([@ref-28]; [@ref-21]). Larger plastic pieces (macroplastic \> 5 mm) are degraded by UV light and wave action into microplastics (\<5 mm) ([@ref-16]; [@ref-11]). Plastic pollution, especially microplastic, is often confused for food by marine organisms ([@ref-2]), and absorbed toxins in plastic pollution can bioaccumulate in higher trophic levels ([@ref-36]; [@ref-12]). High concentrations of floating plastic debris have been reported in many areas of the ocean ([@ref-6]; [@ref-10]), in particular at the center of oceanic gyres (i.e., the great "garbage patches" [@ref-3]). The South Pacific gyre is dominated by microplastic particles (plastic \< 5 mm in diameter) ([@ref-11]). The islands of the South Pacific are sinks and sources for plastic in the marine ecosystem. Uninhabited islands in the area accumulate plastic debris at alarming rates, acting as sinks for plastic from the South Pacific gyre ([@ref-23]). Most of the inhabited islands in this region are also a source of plastic entering the marine environment, as their landfills are generally uncontrolled tipping locations on or near the coast ([@ref-29]; [@ref-18]). In French Polynesia, which has the most technologically advanced waste management center of the whole Pacific region, it is unclear if the amount of plastic pollution entering the marine environment has increased as funding for their landfill has shifted from the French government to local municipalities since 2010 ([@ref-29]; [@ref-18]). Coral reefs, a common coastal ecosystem of South Pacific Islands, are productive and biologically diverse ([@ref-30]; [@ref-38]). Although coral reefs cover 0.2% of the ocean floor, they contain around one third of all described marine species ([@ref-34]), and millions of people in the South Pacific depend on coral reefs for food ([@ref-5]). Coral reefs are threatened by anthropological effects on a local and global scale, including sedimentation, increasing temperatures from global climate change, and changes in sea water chemistry ([@ref-43]; [@ref-14]; [@ref-20]; [@ref-1]). Macroplastics such as fishing nets are known source of coral degradation ([@ref-9]), and microplastics contaminate the reefs around Australia, including the Great Barrier Reef ([@ref-35]; [@ref-17]). It is unclear how the magnitude of plastic pollution on the fringing reef of a South Pacific Island would compare to Australian coastal waters. Our knowledge of the biological impact of plastic pollution on coral is limited. The ingestion of microplastic by scleractinian (reef-building) corals has been observed in a laboratory setting ([@ref-17]), but never *in situ*. The ingestion of plastic by a large-bodied, non-calcifying coral ([@ref-25]; [@ref-22]), *Discosoma nummiforme* was examined in this study. Corallimorphians including *D. nummiforme* are more resilient to temperature and rising CO~2~ levels than scleractinian corals ([@ref-27]; [@ref-22]; [@ref-31]; [@ref-41]), and may ingest plastic particles at a slower rate than scleractinian corals. Categorizing the distribution of plastic pollution on the South Pacific island of Mo'orea, French Polynesia, as well as understanding the interaction between plastic pollution and resilient corals like corallimorphs will help us understand the magnitude of the threat that plastic pollution poses to coral reef ecosystems in the South Pacific region. This work specifically aimed to (1) understand the extent of the current macroplastic pollution problem on the South Pacific island of Mo'orea, French Polynesia (2) determine the concentration of microplastic particles in the water column of the fringing reef of Mo'orea and (3) evaluate if plastic particles are ingested by the corallimorph *Discosoma nummiforme.* {#fig-1} Methods ======= Island-wide field survey of macroplastic ---------------------------------------- Mo'orea (17 30′S, 149 50′W), French Polynesia, has a fringing coral reef that encircles the whole island ([Fig. 1](#fig-1){ref-type="fig"}). The three main populated centers on the island are Afareaitu (2012 census pop. 3,455), Haapiti (pop. 4,062) and Pao Pao (population 4,580) ([@ref-4]; see [Fig. 1](#fig-1){ref-type="fig"}). Plastic surveys were conducted over the course of six weeks, from October 10 to November 20 2016, on beaches around Mo'orea ([Fig. 1](#fig-1){ref-type="fig"}). Additional plastic sampling was conducted at the two largest public beaches on the island: Plage Public de Ta'ahiamanu (17 29′32.7″S 149 51′00.8″W), and Temae Beach (17 29′54.5′S, 149 45′42.9′W), The laboratory study was also conducted on the island, at the Richard B. Gump Research Station (17 29′25.1″S 149 49′35.5″W, [Fig. 1](#fig-1){ref-type="fig"}). Macroplastic (plastic \> 5 mm) abundance was surveyed on Mo'orea's perimeter road kilometer markers (called PK markers) ([Fig. 1](#fig-1){ref-type="fig"}). On the northern side of the island, the waterfront (typically beaches or river outlets) at every PK marker of the perimeter road were surveyed, and around the remainder of the island, the beaches were surveyed at every 3rd PK marker, due to accessibility issues (the waterfront was not accessible from the road at every PK marker, especially on the southern side and northeastern corner of the island). The PK markers were used to randomize the types and location of sites categorized. Upon reaching a site at the PK marker, the site was first categorized by anthropogenic "site type": residential, hotel, natural or public beach site (Of the 26 sites visited, ten were classified as residential sites, four as hotels, four as public beaches and eight as natural sites, see [Fig. 2](#fig-2){ref-type="fig"}). The presence or absence of the abundant macro algae *Turbinaria ornata* was then recoded. This macroalgae was recorded in the study because pilot data indicated that this algae forms mats on the surface of the water, in which plastic is often entangled. The residential and natural beaches were categorized as either non-*T. ornata* or *T. ornata* beaches, but the five river outlet sites and the four hotel sites surveyed were counted as separate categories in this analysis, as water movement in river outlets and hotel clean-up efforts affected *T. ornata* presence at these sites. Once the beach was categorized by anthropogenic site type and presence/absence of *T. ornata*, a timed five-minute trash pickup was conducted. Two researchers, beginning at exactly the PK marker, walked along the beach in opposite directions, collecting all plastic pieces within 5 m of shore, for the survey. At each site, after the number of plastic particles collected on the beach in the interval was recorded, the percent of plastic of the total pollution found at the site was estimated. Percent plastic of all pollution and total number of plastic pieces collected in the surveys were quantified separately because they were not correlated (linear regression *p* \> 0.5). Additionally, the distance in km between the nearest population center (Afareaitu, Haapiti, or Pao Pao, depending on which was closest to the site) and the beach site was recorded. {#fig-2} Microplastic survey ------------------- A plankton net (mouth size 0.07 m^2^; mesh size 0.05 mm) was used to collect water samples at Plage Publique de Ta'ahiamanu, to test for the presence of microplastic in the water column. A total of six 3-m plankton tows were conducted at the surface of the water in the intertidal zone of the beach, at randomly chosen intervals (of approximately 5 m) parallel to shore. The total area of water surveyed was calculated using the formula: *A* = *l*∗*w*, where *l* is the diameter of the net (0.3 m) and h is 18 m (3 m × 6 trials). The area of the water surface, as opposed to a volume of ocean water, was calculated to homogenize the results with similar studies on coastal plastic. Microplastics were identified under a light microscope at the Gump Research Station. Each piece discovered was measured and photographed, and the number of plastics was divided by the surveyed area to determine microplastic concentration. Corallimorph plastic ingestion experiment ----------------------------------------- The plastic used in the study was collected from Temae beach sand ([Fig. 1](#fig-1){ref-type="fig"}) and isolated from the bath product "LAINO Exfoliating shower gel". The naturally collected plastic found in the sand was used in the study because of its negative buoyancy; 5 mm pieces of plastic were taken to Gump Station, and smashed with a hammer until they matched the size of observable microplastic in the water column (0.2--1 mm diameter). The polyethelene plastic beads in the shower gel were isolated by water filtration in a 0.005 mm plankton sieve. All of the isolated beads were a uniform size of 0.2 mm, green in color, and positively buoyant. Plastic color was used to differentiate the buoyancy of the plastic, as the positively buoyant plastic beads from the shower gel were green, while the negatively buoyant plastic collected at Temae were blue. A total of 44 corallimorph polyps were collected from the fringing reef in a depth range of 1--3 m of water at Plage Publique de Ta'ahiamanu ([Fig. 1](#fig-1){ref-type="fig"}) Corallimorphs were identified to a species level following the descriptions of Fautin and the Mo'orea Biocode database as *Discosoma nummiforme* ([@ref-32]; [@ref-13]). The disconnected polyps (average diameter 5 mm) remained attached to coral rubble rocks for the duration of the experiment. At the Gump Station, the polyps were given one week to adjust to laboratory conditions in a large flow tank (28 °C unfiltered ocean water), then thirty-four were placed in a separate experimental tank (28 cm \*30 cm \*8 cm) where they were exposed to plastic. Ten organisms were maintained as control. In the experimental tank, 0.5 mL of both the lab-isolated and of the naturally collected plastic (1 mL total plastic) haphazardly placed on and around the corallimorph polyps. The water flow in the tank containing the corallimorphs was stopped during this procedure. Flow reduction of and the high plasitc concentration were used to ensure the most ideal conditions for plastic consumption. After 84 (*n* = 19) or 108 (*n* = 15) hours, the polyps were dissected, and number and color of plastic present in the organisms' tissues were recorded. Statistical analysis -------------------- Non-parametric tests were used as the number of samples per site type were not equal. Differences in amount of macroplastic present on beaches among the different beach classifications were examined for significance using a Kruskal--Wallace test, and a post-hoc Kruskal--Nemenyi test. To test for differences in macroplastic abundance in the presence of *Turbinaria*, a Kruskal--Wallace test was used. To test if distance from population centers on the island was predictive of macroplastic concentrations on sampled beaches a linear regression was used. For the microplastic feeding trial data, a Wilcoxon Rank Sum Test was used to examine the significance of the differences in natural-caught negatively buoyant plastic vs isolated, buoyant plastic over the time periods. This test was also used to examine the significance of the different hourly rates of consumption. All statistical tests were conducted in R ([@ref-33]). Results ------- Island-wide field survey of macroplastic ---------------------------------------- Plastic was found on every beach surveyed on the island. Other than plastic, the most common pollutants were glass, metal and cloth. Public beaches, natural areas and residential areas had similar amounts of plastic (45 ± 5 SD). These three site types had ten times higher levels of average number of plastic pieces than hotel beaches (4.75 ± 1 SD). The differences of means in plastic amount collected varied by site type (Kruskal--Wallace, Chi-sq = 9.6, 3 df, *p* \< 0.05, [Fig. 2A](#fig-2){ref-type="fig"}). In the post-hoc analysis, the mean of plastic collection on hotel beaches was lower than natural and public sites (*p* ≤ 0.05), but the mean amount of plastic collected on hotel beaches was not significantly lower than residential beaches (Posthoc Kruskal--Nemenyi, *p* \> 0.1). Of the 26 sites, 17 were natural beaches, of which 7 had *Turbinaria*, and 10 had no *Turbinaria.* 29% more plastic was present on beaches that had *Turbinaria* present than on clean beaches (Posthoc Kruskal--Nemenyi test, *p* \< 0.05). When compared to the remaining sites (river outlets and hotel beaches), mean plastic amount was higher on beaches with *Turbinaria* present. (Kruskal--Wallace, Chi-sq = 15.0, 3 df, *p* \< 0.01, [Fig. 2B](#fig-2){ref-type="fig"}). The percentage of plastic on the beaches ranged from 20% to 100% of the total pollution present, with an average of 68% plastic pollution. The number of plastic pieces, as well as the percentage of plastic of all waste found on the beaches, increased with distance from population centers, but not significantly (linear regression for both, *p* \> 0.05, [Fig. 3](#fig-3){ref-type="fig"}). {#fig-3} Microplastic concentration -------------------------- Microplastic was found in the water column at the collection site, with a total of four pieces found in the six tows. One of the four small pieces of plastic was assumed to be, based on color and texture, from a larger piece of plastic collected in the tow. Overall, the six tows contained 0.74 pieces of microplastic m^−2^ surface area. Exposing corallimorphs to microplastic -------------------------------------- Of the 34 corallimorph polyps exposed to plastic, 19 (55%) polyps ingested one or more plastic particles during the experiment. The number of plastic particles ingested by individual polyps varied from zero to eight. The average amount of plastic consumed in the shorter time trial was 0.7 pieces/polyp and in the longer time trial 1.5 pieces/polyp. The mean amount of total plastic ingested did not vary with treatment time (Wilcoxon rank sum test, *p* \> 0.05, [Fig. 4](#fig-4){ref-type="fig"}). Although the total amount of plastic did not vary over the treatments, the amount of the positively buoyant plastic (the green micro-bead plastic) consumed increased over the separate time trials. In the first treatment time of 84 h, an average of 0.7 blue plastic particles (BPP)/polyp were consumed by the corallimorphs and 0 green plastic particles (GPP)/polyp were consumed. In the longer trial of 108 h, an average of 0.9 BPP/polyp and 0.5 GPP/polyp were consumed. The GPP consumed varied significantly over the separate time trials. (Wilcox rank sum test, *p* \< 0.01, see [Fig. 4](#fig-4){ref-type="fig"}.) Additionally, a yellow plastic particle was found in a polyp in the first treatment, presumably present in the tissue before the experiment (see [Fig. 4](#fig-4){ref-type="fig"}). The ten control corallimorph polyps consumed zero plastic particles, experimentally introduced or otherwise. {#fig-4} Discussion ========== Plastic accumulated most on natural beaches, and the amount of plastic and percentage of plastic debris increased insignificantly with distance from major population centers. This distribution of plastic on the island is somewhat surprising when compared to the literature, which generally find larger litter loads near urban areas ([@ref-15]; [@ref-37]). While natural areas had a higher than expected amount of plastic, the large volume of plastic on the public beaches of Mo'orea supports previous literature on coastal plastic debris. A recent review article on coastal pollution categorized over 60% of coastal pollution as from sources of "shoreline and recreational activities" that are prevalent on public beaches ([@ref-40]). Finally, little scientific literature exists on hotel management of plastic debris, but the low amount of plastic discovered on hotel beaches is likely explained by hotel staff working to maintain a clean, white and sandy beach, as expected by hotel guests (D Venuit, pers. comm., 2016). The beaches contaminated with *Turbinaria ornata* had a significantly higher amount of plastic than beaches without it. In the study, the plastic pieces collected were often caught in the thallus of this algae. The comingling of marine organisms and plastic is not a new phenomenon, as invertebrates and algae have been associated with plastic ([@ref-42]). Retention of plastic by *T. ornata* mats floating at the water's surface may in fact account for the low presence of plastic found on beaches near population centers, as floating *T. ornata* mats can travel long distances ([@ref-26]), and the currents of Mo'orea change rapidly ([@ref-19]). Further investigation into the association between plastic pollution and *T. ornata* is necessary to fully understand their association. Surface waters of the intertidal zone of Mo'orea are contaminated with small plastics. This study adds to the evidence in the literature that suggests microplastics are the most abundant type of debris in all marine environments ([@ref-39]). The concentrations of microplastic is much higher in oceanic gyres than it is on Mo'orea; as high as 334 pieces m^−2^ in the Northeast Pacific ([@ref-28]), and 396 pieces m^−2^ in the center of the South Pacific gyre ([@ref-11]). The concentration of plastic found in Mo'orea's intertidal zone, however, is on the same order of magnitude as studies conducted outside of oceanic gyres of different geographic areas. For instance, the concentration in the Caribbean Sea (1.414 pieces m^−2^), and the concentration in the Gulf of Maine (1.534 pieces m^−2^) are both are slightly higher, but of a consistent magnitude when compared to the found concentration of 0.74 pieces m^−2^ in Mo'orea ([@ref-24]). This equivalent magnitude of plastic on Mo'orea and the North America and Carribean coast is unexpected, as the concentration of plastic off the coast of a highly developed area should have a higher magnitude of plastic than an island in the South Pacific. Off the coast of Australia researchers found that plastic accumulates in waters near population centers ([@ref-35]). Papeete, the capital of neighboring island Tahiti, is a relatively densely populated city, and likely has a large amount of plastic use, that may contribute to the plastic on Mo'orea. Any future study of the intertidal zone should include a larger portion of Mo'orea's and even Tahiti's coastline, covering km rather than meters of water, to further investigate microplastic accumulation in the region. This study demonstrated, for the first time, the ingestion of microplastic by the corallimorphia *D. nummiforme*. The corallimorph polyps in the study ingested plastic particles at a slower rate than the scleractinian coral from [@ref-17] paper (55% polyps inundated in 108 h, versus 21% polpys in 12 h). Although plastic consumption occurred at a slower rate, plastic presence in both studies caused considerable mucus formation that may represent an additional energy expense associated with microplastic contamination (pers. obs.; [@ref-17]). In both studies, plastic particles were found within the mesenterial tissue upon dissection, which may impede the digestion of these organisms ([@ref-17]). In the experiment, the negatively buoyant plastic was ingested more often by these benthic organisms. The ingestion of the positively buoyant plastic by the corallimorph polyps is more surprising, especially because it only occurred in the longer treatment time (108 h). It has been previously hypothesized that corallimorphs prey on zooplankton or absorb dissolved organic material at an increased rate to survive temperature changes ([@ref-22]), so it is possible that under the induced stressful conditions (no water flow), the polyps altered their feeding rate and consumed more plastic. It is also possible that the green microbeads began to sink over time, and as they entered the benthic environment the polyps consumed them as apparent food. The high concentration of plastic and low flow in the experiment were ideal conditions for plastic consumption; future studies more similar to the actual reef environment are necessary for predicting rates of plastic consumption by corallimorphs *in situ*. These studies are essential especially because in this study, the presence of yellow plastic in the stomach cavity of *D. nummiforme* demonstrates that plastic is being ingested by corallimorphs on the reef of Mo'orea. In conclusion, plastic pollution is prevalent on the beaches and reef of Mo'orea, as it was found on every beach visited during the study. Macroplastic was found in significant higher amounts on beaches with the algae *T. ornata,* and future study is warranted to understand if plastic pollution is transported around the island by this alga. Microplastic pollution was found in the water column of the fringing reef, and in the stomach tissue of the prevalent reef organism *D. nummiforme*. Under laboratory conditions, both buoyant and non-buoyant plastic were ingested by this corallimorphia. Further study into the distribution and biological consequences of plastic pollution on Mo'orea, and neighboring islands, is necessary to understand and combat this ongoing problem. Supplemental Information ======================== 10.7717/peerj.3733/supp-1 ###### Raw code for figures ###### Click here for additional data file. 10.7717/peerj.3733/supp-2 ###### Raw data ###### Click here for additional data file. The author thanks Dr. Jonathon Stillman, Dr. Justin Brashares, Dr. Cindy Looy, Dr. Patrick O'Grady (UC Berkeley) and three anonymous reviewers for their valuable comments on the manuscript; Mr. Eric Armstrong, Mr. Ignacio Escalante Meza, Dr. Natalie Stauffer-Olsen and the Gump Station staff for their assistance with field logistics; and Mr. Eric Witte, Ms. Charlotte Runzel and Ms. Jacey Van Wert for the assistance sampling corallimorphs. Additional Information and Declarations ======================================= The author declares that they have no competing interests. [Elizabeth J. Connors](#author-1){ref-type="contrib"} conceived and designed the experiments, performed the experiments, analyzed the data, contributed reagents/materials/analysis tools, wrote the paper, prepared figures and/or tables, reviewed drafts of the paper. The following information was supplied regarding data availability: The raw data has been supplied as a [Supplementary File](#supplemental-information){ref-type="supplementary-material"}.