pmid stringlengths 8 8 | pmcid stringlengths 8 11 ⌀ | source stringclasses 2
values | rank int64 1 9.78k | sections unknown | tokens int64 3 46.7k |
|---|---|---|---|---|---|
25621279 | PMC4286716 | pmc | 232 | {
"abstract": "In the last two decades, genetic and genomic studies have revealed the astonishing diversity and ubiquity of microorganisms. Emergence and expansion of the human microbiome project has reshaped our thinking about how microbes control host health—not only as pathogens, but also as symbionts. In coral reef environments, scientists have begun to examine the role that microorganisms play in coral life history. Herein, we review the current literature on coral-microbe interactions within the context of their role in evolution, development, and ecology. We ask the following questions, first posed by McFall-Ngai et al. ( 2013 ) in their review of animal evolution, with specific attention to how coral-microbial interactions may be affected under future environmental conditions: (1) How do corals and their microbiome affect each other's genomes? (2) How does coral development depend on microbial partners? (3) How is homeostasis maintained between corals and their microbial symbionts? (4) How can ecological approaches deepen our understanding of the multiple levels of coral-microbial interactions? Elucidating the role that microorganisms play in the structure and function of the holobiont is essential for understanding how corals maintain homeostasis and acclimate to changing environmental conditions.",
"conclusion": "Conclusions Scleractinian corals have survived dramatic environmental changes during their evolutionary history including changes in sea level, pH, and temperature—however current anthropogenically-induced changes are occurring at rates unprecedented in the geological record (Glynn, 1991 , 1993 ; Hoegh-Guldberg, 1999 ; Pandolfi et al., 2003 ; Hoegh-Guldberg et al., 2007 ; Mora, 2008 ). Whether extant corals will survive under current regimes of rapid change will depend upon their ability to adapt to new conditions on anthropogenic timescales. It is likely that coral-associated microbiota will be linked to the capacity of corals to ultimately adapt to new environmental conditions. In plant systems, it has become increasingly clear that crop improvement through microbiome optimization is preferable to plant trait selection (Coleman-Derr and Tringe, 2014 ). In a similar fashion coral-associated microbes may mediate the resilience of corals to stresses such as resistance to coral-specific pathogens (Reshef et al., 2006 ; Rosenberg et al., 2007 ) and perhaps to other stressors. Our understanding of coral physiology and interaction as a holobiont is now enhanced by the expanded scientific focus on the diverse members of the coral microbiota. Access to whole genome data from the major players in a single holobiont, in particular, organisms that may be vertically transmitted from parent colony to offspring, will enable researchers to develop new, readily testable hypotheses regarding metabolic complementation, genome expansion/reduction, coevolutionary patterns, and rates of HGT across all genomes in a holobiont. Such information will yield insight into the adaptations that have enabled coral holobionts to be so successful in nutrient poor environments. In order to motivate future work to refine our mechanistic understanding of microorganisms as coral partners through evolution, development, and in ecological interactions within the holobiont we have selected eight questions to highlight based on recent literature (Table 2 ). Continued research is essential to promote a detailed understanding of the mechanisms shaping the coral holobiont on ecological and evolutionary timescales as pressures emerge for reefs to acclimate and adapt to conditions of global change. Table 2 Questions for future research . • What are the mechanisms of nutritional and defensive mutualisms between holobiont members? This is a fundamental question to understand the ecology and physiology of the holobiont. The answers may be of biomedical interest as novel antimicrobial compounds or mechanisms that disrupt pathogen colonization or virulence without bactericidal or bacteriostatic activity (e.g., Krediet et al., 2013 ) may be involved. • What is the diversity of vertically-transmitted microorganisms in coral reproduction? • Do these populations represent obligate or facultative mutualisms? Novel cultivation-based approaches and genome sequencing or single-cell genomics to recover populations that resist cultivation are necessary. Such improvements will enable testing of hypotheses regarding coevolution and codiversification between corals and their microbiota. • What is the role of Symbiodinium in shaping microbiota acquisition in gametes, larvae, and adult coral colonies? • Do algae produce chemical signals that mediate allelopathic recognition or modulate colonization of the microbiota? • How do chemical signals produced by CCA biofilms promote settlement and/or metamorphosis of coral larvae and how is receipt of these signals coordinated to optimize recruitment? • What microbial taxa and activities are associated with coral cell-associated microbial aggregates (CAMAs) and what factors mediate their distribution in the coral polyp? • Do the enclosed microorganisms interact with the coral tissue as mutualists, parasites, or commensals? • Since several studies have documented lower levels of microbial colonization in Montipora spp. compared to other corals (Marquis et al., 2005 ; Work and Aeby, 2014 ), to what extent is antimicrobial activity specific to the coral species or their associated Symbiodinium ? • What role does the endolithic microbial community play in the homeostasis of the coral holobiont? A potentially transformative avenue for inquiry is exploring the relationship between bacterial activity and skeleton formation in juvenile corals as recently suggested (Sharp et al., 2012 ). • What are the molecular mechanisms of the host/holobiont stress response that allow proliferation of pathogen-like bacteria (e.g., Vibrio spp. , Alteromonas spp.)? Emerging coral holobiont transcriptomics studies will shed new light on how the coral holobiont responds to stresses and exists as a robust and resilient system when in a healthy state. 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.",
"introduction": "Introduction Multicellular hosts harbor communities of beneficial microbes (Bosch and McFall-Ngai, 2011 ). Compelling evidence illustrates how microorganisms have facilitated the origin and evolution of animals and are integral parts of all animal life (McFall-Ngai et al., 2013 ). The coral holobiont is comprised of the coral animal and its associated microorganisms consisting of bacteria, archaea, fungi, viruses, and protists including the dinoflagellate algae Symbiodinium (Rohwer et al., 2002 ). The coral holobiont is a dynamic system, whose members fluctuate depending on environmental conditions and daily requirements (Shashar et al., 1993 ; Tanner, 1996 ; Ainsworth et al., 2011 ). In their review of animal evolution, McFall-Ngai et al. ( 2013 ) proposed that our understanding of microbes' role in the evolution of animal partners could be increased by examining mutual interactions and reciprocal influences during development and genomic evolution. In this review, we follow their framework to evaluate the role coral-microbe associations have played in (1) genome evolution in both host and microbial partners, (2) shaping and driving coral development, (3) modulating holobiont homeostasis, and (4) defining the ecology of interactions in the coral holobiont. We aim to review the current literature on coral-microbe interactions in order to create a consolidated resource of knowledge in an emerging and quickly developing field. As the symbiosis between corals and Symbiodinium represents a long-standing field of investigation (Trench, 1983 , 1993 ; Weber and Medina, 2012 ) we will focus in this review on our emerging understanding of other microbial components of the holobiont that we collective refer to as the microbiota."
} | 2,024 |
37992160 | PMC10664981 | pmc | 233 | {
"abstract": "Unlike reef-building, scleractinian corals, Caribbean soft corals (octocorals) have not suffered marked declines in abundance associated with anthropogenic ocean warming. Both octocorals and reef-building scleractinians depend on a nutritional symbiosis with single-celled algae living within their tissues. In both groups, increased ocean temperatures can induce symbiont loss (bleaching) and coral death. Multiple heat waves from 2014 to 2016 resulted in widespread damage to reef ecosystems and provided an opportunity to examine the bleaching response of three Caribbean octocoral species. Symbiont densities declined during the heat waves but recovered quickly, and colony mortality was low. The dominant symbiont genotypes within a host generally did not change, and all colonies hosted symbiont species in the genus Breviolum. Their association with thermally tolerant symbionts likely contributes to the octocoral holobiont’s resistance to mortality and the resilience of their symbiont populations. The resistance and resilience of Caribbean octocorals offer clues for the future of coral reefs.",
"introduction": "INTRODUCTION Reef corals, which create the structural framework for some of the most biologically diverse ( 1 ) and economically important ( 2 ) ecosystems on the planet, are declining ( 3 ). This decline is due to a range of anthropogenic and natural perturbations, the foremost being increased ocean temperatures ( 4 ). Increased sea surface temperatures can lead to a breakdown of the nutritional mutualism between reef cnidarians and endosymbiotic algae in the family Symbiodiniaceae ( 4 ). Many cnidarians, including reef-building stony corals (scleractinians) and soft corals (octocorals), depend on their algal endosymbionts for nutrients and growth ( 5 ) and the loss of these algal symbionts, termed coral bleaching, can result in coral death. Over the past four decades, coral bleaching has become increasingly common and is now arguably the most important determinant of coral reef ecosystem dynamics. Most studies of bleaching have focused on scleractinians, as these ecosystem engineers play a crucial role in determining reef structure and have seen major bleaching-related population declines. Studies have shown that variation in scleractinian bleaching response is due in part to the genetic identity of the symbiont, with scleractinians that do not bleach often harboring thermally tolerant symbionts, such as Cladocopium thermophilum , or species within the genus Durusdinium ( 6 – 9 ). In contrast to their scleractinian relatives, Caribbean octocorals have not exhibited population declines in recent years; rather, these anthozoans are maintaining coral cover and at some sites, increasing in abundance despite rising oceanic heat content ( 10 – 12 ). The basis for the comparative success of this group in the face of anthropogenic ocean warming remains unclear, however. The ability of organisms to persist in the face of a stressor can be due to resistance (the ability to withstand disturbance) or resilience (the ability to recover after disturbance). Historically, bleaching among Caribbean octocorals has been rare, particularly in comparison to scleractinians ( 13 – 18 ), suggesting that these holobionts are likely exhibiting resistance, either at the colony or symbiont level ( 17 ). Notably, symbiont diversity among Caribbean octocorals is lower than that of scleractinian species with the vast majority of Caribbean octocorals harboring symbionts in the Breviolum B1/B184 lineage [nomenclature based on variation in the internal transcribed spacer region 2 ( ITS2 ) of nuclear ribosomal DNA ( rDNA ) and the chloroplast 23 S rDNA (cp- 23S rDNA ), respectively ( 19 – 21 )]. This association with symbiont species in the genus Breviolum may contribute to octocorals’ resistance to bleaching, as there is considerable genetic diversity within this lineage ( 22 – 24 ) and laboratory studies have shown that several of these symbiont species grow well at elevated temperatures ( 25 – 27 ). Furthermore, in the few reports of Caribbean octocoral bleaching before 2000 [1983 ( 13 ), 1987 ( 28 ), and 1998 ( 29 )], bleaching was generally restricted to octocoral hosts that did not typically harbor Breviolum B1 symbiont species [i.e., Plexaurella dichotoma ( 20 , 21 ), Plexaurella nutans ( 21 ), Eunicea sp. ( 19 ), Erythropodium caribaeorum ( 19 , 20 ), and Briareum asbestinum ( 15 )]. Thus, one possible explanation for the continued success of Caribbean octocorals is the prevalence of symbioses with symbiont species within the Breviolum B1 lineage, with bleaching-resistant hosts harboring more thermally tolerant symbiont species or genotypes. Yet, resilience is also likely to play a key role in the stability of octocoral populations. In addition to harboring potentially thermal tolerant symbionts, bleaching-resistant and -resilient octocorals may be able to shuffle symbiont types in response to thermal stress. This latter scenario is supported by observations of bleaching within B. asbestinum . When the octocoral B. asbestinum bleaches, the dominant symbiont, a species in the Breviolum B19 lineage, is replaced by a more thermally tolerant species in the Breviolum B1 lineage ( 15 , 30 , 31 ). This ability to shift symbiont partners during and after bleaching also suggests a role of resilience in the maintenance of octocoral communities. With the exception of the B. asbestinum studies already noted, the few analyses of symbiont types within bleached Caribbean octocorals have been short-term laboratory-based studies ( 17 , 18 , 32 ). There have been no field studies tracking the dynamics of bleaching across a marine heat wave among Caribbean octocorals. Symbiont types within octocorals during a bleaching event have been characterized in soft coral species from the Great Barrier Reef (GBR) and Guam ( 33 , 34 ), but those host species are morphologically, ecologically, and taxonomically distinct from the Caribbean species. Unlike Caribbean species, the majority of studied Pacific octocoral species harbor symbionts within Cladocopium or Durusdinium ( 21 , 33 , 34 ). As symbionts vary physiologically in their response to thermal stress both within and between species ( 25 – 27 , 35 , 36 ), it is important to examine the dynamics of coral-algal symbiosis and its ability to cope with the changing temperature at multiple levels of resolution. As bleaching susceptibility may depend on the response of both the host species ( 37 ) and the symbiont ( 7 ), approaches examining responses between and within specific host-symbiont pairs must be used. Although Caribbean octocoral bleaching has historically been rare, the frequency of widespread coral bleaching events is increasing ( 38 – 40 ), and octocoral bleaching, although it remains relatively uncommon, is also on the rise. In 2014–2017, an extended El Niño–Southern Oscillation event resulted in scleractinian bleaching on a global scale ( 41 ). Octocoral bleaching was observed in the Florida Keys during the late summer and autumn of 2014 and 2015. This provided a unique opportunity to examine octocoral response to thermal stress over time in a field-based setting. We surveyed bleaching among octocorals in 2014 and then followed the bleaching status and symbiont genotypes of individual colonies of three Caribbean octocoral species before, during, and after the 2015 bleaching event. We specifically asked (i) whether symbiont types changed during bleaching and recovery and (ii) how bleaching susceptibility, resistance, or resilience varied with symbiont genotype and host species. Octocorals have fared relatively well in the past several decades and their increased relative abundance on Caribbean reefs has been suggested to be a “new normal” ( 12 ), yet our understanding of their response to bleaching lags behind that of scleractinian corals. An in-depth understanding of the dynamics of octocoral-algal symbioses through bleaching events is important in assessing their continued success as sea surface temperatures continue to rise.",
"discussion": "DISCUSSION 2015 was the warmest year on record at the time of the study and, in contrast to other bleaching events, in both 2014 and 2015, widespread bleaching was reported among octocorals, including species that had rarely bleached in the past (tables S1 and S2). This was evident at our study sites where M. atlantica and P. dichotoma were visibly bleached in September 2015, and all three host species had significant declines in symbiont density. While bleaching had previously been reported in P. dichotoma and possibly M. atlantica ( 13 , 16 ), bleaching appears to be a relatively rare occurrence among most Caribbean octocorals, particularly as compared to bleaching in Caribbean scleractinians. Despite the loss of symbionts, mortality was low during the marine heat waves. All the tagged colonies survived between the September and November 2015 surveys (i.e., during and immediately following the height of bleaching), and symbiont density in colonies returned to prebleaching levels by November 2015 ( Fig. 3 ). In contrast to the low mortality rates that we observed among M. atlantica , Prada et al . ( 16 ) reported high mortality among Muricea muricata/atlantica colonies during the 2005 ocean warming event. Our observations of low mortality among colonies of M. atlantica suggest that the result reported by Prada et al . ( 16 ) might reflect mortality specific to M. muricata or possibly a location-specific mortality. Generally, our observations of low mortality across all studied species are consistent with other studies of Caribbean octocorals ( 13 , 16 ). Unlike the widespread bleaching we observed in 2014 and 2015, laboratory studies have found limited sensitivity to high temperatures among Caribbean gorgonians. McCauley et al . ( 18 ) reported relatively minor symbiont loss in Eunicea tourneforti and Pseudoplexaura crucis ranging from 26 to 35% following short-term (7 days) exposure to elevated temperatures. In a similar study, Goulet et al . ( 17 ) did not observe bleaching in Eunicea flexuosa , E. tourneforti , and Pseudoplexaura porosa after 5 days of temperature stress. The difference between these laboratory studies and our field observations of substantial symbiont loss is likely due to a combination of the sustained exposure to elevated temperatures seen in the field where corals experienced 8 to 13 DHWs, and potentially the host/symbiont species involved. Several researchers have found that colonies with diverse symbiont populations have lower stress tolerance ( 43 ) and have proposed that greater diversity within the symbiont community could increase the host’s susceptibility to stress. Given that M. elongata had high allelic and genotypic diversity (table S6), one might predict greater susceptibility to stress. However, although more M. elongata died, the mortality rate was not significantly different from the other host species (χ 2 = 5.794, df = 2, P = 0.055). In addition, M. elongata lost proportionally fewer symbionts during the bleaching event than the other two host species and mortality did not correlate with bleaching susceptibility. Roles of symbionts and hosts The coral holobiont is a complex consisting of the coral animal, algal symbionts, and other microbes. In an ecological context, the terms resistance and resilience refer to the responses of a population to a stressor. Resistant populations do not exhibit changes in their vital statistics (e.g., increases in mortality), whereas resilient populations will exhibit such change but then recover to pre-stressor norms. This seemingly simple dichotomy becomes complex when we consider organisms as holobionts. In the case of octocorals, there are populations of coral hosts on a reef, and then within each individual host, there are population(s) of symbionts. Using mortality as the measure of resistance, the colonies in our study were strongly resistant to thermal stress in 2014 and 2015. In contrast, the loss of symbionts (i.e., bleaching) suggests a lack of stress resistance at the level of the symbiont populations. However, symbiont populations were resilient as they recovered rapidly. The highest densities were seen in March 2016. That pattern is similar to that observed by Fitt et al . ( 44 ) in Orbicella (previously Montastrea ) faveolata and annularis after the 1995 bleaching in the Florida Keys. These authors speculated that the overshoot may be the result of increased symbiont growth rates in response to increased nutrient availability due to low symbiont numbers. Holobiont survival is often attributed to shuffling or switching to a more thermally tolerant symbiont, indicating that symbiont genotype likely plays a role in the holobiont’s resistance and/or resilience to thermal stress. Symbiont shuffling in response to elevated temperature (examined at the genus or species level) has been reported among scleractinians ( 6 , 8 ), but changes in symbiont type are not always seen in response to thermal stress in octocorals and scleractinians [see, e.g., ( 17 , 18 , 32 , 45 – 47 )]. Shuffling or switching of symbionts did not account for the holobiont’s resistance to thermal stress in M. atlantica and M. elongata but may have played a role in the response of P. dichotoma . Among the P. dichotoma colonies, 43% harbored symbionts within the genus Cladocopium in May 2015, in addition to the Breviolum B1 lineage ( Fig. 4 ). In September 2015, all colonies continued to harbor symbionts within Breviolum B1, but the proportion also harboring symbionts within the genus Cladocopium had dropped to 20% ( Fig. 4 ). Cladocopium populations apparently recovered by May 2016, with 48% of the colonies harboring Cladocopium in addition to Breviolum ( Fig. 4 ). Thus, Breviolum and Cladocopium populations differed in their resistance to the thermal stressor, but both were resilient, recovering to prebleaching levels by May 2016. Apart from the loss of symbionts within Cladocopium in some colonies of P. dichotoma , symbiont genera and species did not change over the course of 27 months of monitoring. This stability extended to the level of the symbiont genotype as most colonies maintained the same Breviolum symbiont genotypes across the 27 months (figs. S3 to S5 and table S6). Changes in symbiont genotype observed were not specific to a certain month or sampling point, suggesting that the changes were not a response to bleaching ( Fig. 6 and figs. S3 to S5). Other studies that have examined symbiont composition at the level of intraspecific genetic variation over the course of bleaching have also found stability within host colonies. Goulet and Coffroth ( 45 ) used DNA fingerprinting to assess genetic identity and found no change in symbiont DNA fingerprints within octocoral clones across 10 years. Coral bleaching occurred at those sites over that time span ( 48 ), but symbiont genotypes in octocorals remained constant, which is consistent with our findings. The ubiquity of Breviolum in Caribbean octocorals, the continued prevalence of Breviolum across a bleaching event, and the absence of symbiont shuffling during the bleaching event (in addition to the observed loss of Cladocopium among P. dichotoma colonies) point to a major role of Breviolum in determining holobiont susceptibility to thermal stress. Breviolum was traditionally characterized as a thermally sensitive genus ( 49 ), but those studies examined Breviolum minutum . There is considerable genetic diversity within the Breviolum B1 lineage ( 22 – 24 ) and other species of Breviolum are more thermally tolerant ( 17 , 18 , 25 ), with certain genotypes of Breviolum antillogorgium even exhibiting positive growth at 32°C ( 25 – 27 ). In our study, all three host species harbored symbiont species within the genus Breviolum , which may have contributed to the resilience of the symbiont populations and the overall resistance of the holobiont to thermal stress. Cluster analysis showed that symbionts within the colonies of each species could be partitioned into groups on the basis of the genotypic composition, but this patterning did not contribute to the variance in cell densities. As noted in the results, cell densities were not significantly different between clusters within a host species, suggesting the slightly different combinations of symbiont genotypes were equally resilient to the stress. That is, while host species differed in their response to thermal stress, among individuals within a host species, specific hosts were not associated with differences in the bleaching or recovery based on the symbiont genotype that they harbored ( Figs. 6 and 7 and fig. S7). However, it is important to note that the host also plays a role in holobiont responses to thermal stress ( 17 , 18 , 23 , 33 , 47 , 50 , 51 ). For instance, Kenkel et al . ( 52 ) found that inshore and offshore colonies of Porites astreoides were genetically differentiated and had symbionts that did not differ in ITS haplotype. Despite the similarity of the symbiont populations, there were differences in thermal tolerance between offshore and inshore colonies. While studies have found variation in thermal stress susceptibility not explained by symbiont genetic variation, note that many studies characterize symbiont diversity at either the genus or species level, identifying symbionts based on variation in symbiont nuclear 18 S \n rDNA , ITS2 , or cp- 23S rDNA . Yet, substantial physiological variation exists even within a symbiont species ( 25 – 27 , 35 , 36 ), and the use of genetic markers with within-species genotypic resolution is necessary to fully understand variation in holobiont responses. Studies that have resolved the symbiont genotype at the within-species level indicate that the host-symbiont combination is important ( 23 , 47 , 53 ). For example, Parkinson et al . ( 23 ) found Acropora palmata colonies all harboring the same symbiont genotype varied in their response to cold shock, and Kavousi et al . ( 47 ), also using high-resolution markers to characterize symbiont variation, showed that response to thermal stress varies among different host-symbiont combinations. Future research should continue to explore the response to stress by examining within-species differences among host-symbiont pairs. In addition, a clearer understanding of how the holobiont responds to thermal stress can be gained by examining other physiological parameters [see, e.g., ( 17 , 18 )] and changes in host and symbiont transcriptomics and expression of thermal genes ( 54 , 55 ). Last, other components of the holobiont also may contribute to the response to thermal stress. For example, studies have shown that thermally sensitive corals inoculated with the microbiome from thermal tolerant corals bleached less than control (thermally sensitive) individuals ( 56 ). We differentiated the dynamics of host and symbiont in response to a thermal stress event. Our findings support the conclusion that, in general, Caribbean octocorals exhibit both resistance and resilience in the face of marine heat waves. This is consistent with previous studies that have shown a lack of bleaching among many octocoral species; octocoral bleaching, when it does occur, occurs later and after a longer exposure time than that of scleractinian corals, and in general, there is low mortality among octocorals that do bleach ( 13 , 16 ). Furthermore, we demonstrated that within octocorals, symbiont genotype does not change across a bleaching event. Our results emphasize that in understanding the response to marine heat waves, it is critical to follow individual colonies across an event with long-term monitoring of both host and symbiont response and examine this response at least at the level of symbiont species (if not genotype) to identify potentially resilient species. Work such as this will be critical to understanding the dynamics and response of corals to heat waves and making predictions about the impact of these events on coral reef community structure."
} | 5,066 |
37938252 | PMC9723718 | pmc | 234 | {
"abstract": "Using the Mediterranean coral Balanophyllia europaea naturally growing along a pH gradient close to Panarea island (Italy) as a model, we explored the role of host-associated microbiomes in coral acclimatization to ocean acidification (OA). Coral samples were collected at three sites along the gradient, mimicking seawater conditions projected for 2100 under different IPCC (The Intergovernmental Panel on Climate Change) scenarios, and mucus, soft tissue and skeleton associated microbiomes were characterized by shotgun metagenomics. According to our findings, OA induced functional changes in the microbiomes genetic potential that could mitigate the sub-optimal environmental conditions at three levels: i. selection of bacteria genetically equipped with functions related to stress resistance; ii. shifts in microbial carbohydrate metabolism from energy production to maintenance of cell membranes and walls integrity; iii. gain of functions able to respond to variations in nitrogen needs at the holobiont level, such as genes devoted to organic nitrogen mobilization. We hence provided hypotheses about the functional role of the coral associated microbiome in favoring host acclimatation to OA, remarking on the importance of considering the crosstalk among all the components of the holobiont to unveil how and to what extent corals will maintain their functionality under forthcoming ocean conditions.",
"introduction": "Introduction Scleractinian corals live in close association with a diverse array of phylogenetically disparate microorganisms, including endocellular photoautotrophic dinoflagellate symbionts (belonging to the Symbiodiniaceae family) and complex communities of bacteria, archaea, viruses, and unicellular eukaryotes ( i.e ., microbiomes). Indeed, each coral anatomic compartment (e.g. surface mucus, soft tissue, and skeleton) constitutes a microhabitat characterized by specific conditions supporting different micro-ecosystems (reviewed in ref. [ 1 ]). The consortium of coexisting pluri- and unicellular organisms is termed “coral holobiont”, whose microbial counterpart is believed to maintain organismal function under varying environmental conditions [ 2 , 3 ]. Indeed, complex bacterial communities inhabiting coral mucus, tissue, and skeleton exert a crucial role in ensuring the health and survival of the coral as they provide their host with a variety of functions, such as assistance in recovering and recycling of nutrients (carbon, nitrogen and sulfur, but also vitamins and essential amino acids), protection against pathogens invasion, and production of chemicals that drive larval settlement (as suggested by ref. [ 4 – 9 ]). Besides being tightly related to host phylogeny, the composition and metabolism of coral microbiomes change temporally (across seasons and along coral lifespan), spatially (across the compartments defined by the coral anatomy), and in response to environmental variations [ 1 , 10 , 11 ]. Shifts in coral microbiome composition in response to environmental changes may affect the co-metabolic networks, possibly contributing to acclimatization of the coral holobiont. Indeed, microbial communities as a whole have the possibility to acclimatize faster to environmental changes than their metazoan host, thanks to their greater genetic diversity, shorter generational time, and remarkable metabolic potential [ 12 , 13 ]. Such propensity for a fast response to environmental changes has been investigated as possibly involved in the acclimatization and adaptation of the coral holobiont to climate change-related phenomena, such as ocean warming and ocean acidification (OA) [ 2 , 9 , 12 , 14 , 15 ]. Ocean acidity has increased worldwide by 25–30% (0.1 pH units) since the beginning of the nineteenth century, and it is expected to drop by a further 0.29 pH units by 2080–2100 [ 15 , 16 ]. OA is expected to alter the survival, growth, and reproduction of key components of marine ecosystems, especially calcifying species [ 17 ], both at microbial and multicellular levels [ 18 – 20 ]. Thus, it is of outmost importance, for projecting ecological processes in the forthcoming oceans, to understand how OA will affect the microbiome of important ecosystem forming organisms such as corals [ 21 ]. In this context, natural underwater CO 2 seeps represent a precious study system to understand how corals will respond to OA [ 22 , 23 ]. The Mediterranean Sea, which will likely be one of the most affected seas by climate change [ 24 ], hosts naturally acidified shallow sites with relatively stable underwater CO 2 emission at ambient temperature with no detection of toxic compounds, that have been recognized as fundamental study models for OA [ 25 – 27 ]. To date, few studies have explored coral microbiome variations in natural coral populations at CO 2 seeps [ 6 , 28 – 31 ], showing different microbiome responses depending on the host species. For instance, at natural CO 2 seeps in Papua New Guinea, the endolithic community associated with massive Porites spp. does not change with pH [ 6 , 32 ], while large shifts in tissue-associated bacterial communities were found in Acropora millepora and Porites cylindrica [ 29 ]. Indeed, as recently highlighted by Shore and colleagues [ 31 ], the response to decreasing pH of the microbiome associated to Porites corals seems to be species-specific and does not reflect a breakdown in bacteria-host symbiosis. In the Mediterranean coral Astroides calycularis growing at the Ischia CO 2 vents, the mucus-associated microbiome was more affected by acidification than soft tissue and skeleton, with a general increase in subdominant bacterial groups with OA, some of which may be involved in the nitrogen cycle [ 30 ]. The target species of the present study is the solitary Mediterranean coral Balanophyllia europaea that naturally lives along a pH gradient generated by an underwater volcanic crater located close to Panarea Island (Italy). B. europaea is a temperate, zooxanthellate, scleractinian coral, widespread in the Mediterranean Sea, where it thrives on rocky substrates at a depth of 0–50 m [ 33 ]. To date, the majority of the studies aimed at understanding the involvement of coral microbiomes in acclimatization to future acidified water conditions were focused on tropical and subtropical corals [ 6 , 34 – 36 ]. However, temperate species, such as B. europaea , might represent a model for more pronounced acclimatization capability, being exposed to a twice as high range of seasonal temperature fluctuations and intrinsically more capable to accommodate environmental variations (as suggested by [ 37 ]). Here, we focus on the bacterial component of the B. europaea holobiont and on the variations in the metabolic potential of the microbial communities residing in surface mucus, soft tissue, and skeleton to the decreasing pH. The CO 2 seep near Panarea Island (Italy) from which samples were taken is an underwater crater at 10 m depth releases persistent gaseous emissions (98–99% CO 2 without instrumentally detectable toxic compounds), resulting in a stable pH gradient at ambient temperature that has been characterized in detail [ 25 , 26 , 38 ]. Sampling sites along this gradient match mean pH values projected for 2100 under different IPCC scenarios [ 39 ]. Since a considerable amount of research showed an overall acclimatization of this model organisms to low pH conditions, principally through the homeostatic balance of several physiological parameters e.g. gross calcification rate, calcifying fluid pH, skeletal calcium carbonate polymorph, aragonite fiber thickness, skeletal hardness and organic matrix content [ 25 , 26 , 40 ], here we aim at exploring the possible role of the coral-associated microbiomes in this acclimatization process. Specifically, to verify our hypothesis, we combined 16S rRNA gene sequencing and shotgun metagenomics to highlight changes in genetic functions included in the microbial metagenome in coral holobionts acclimatized to different OA levels under natural conditions, i.e. in corals collected at different distances from the crater.",
"discussion": "Discussion Ocean acidification (OA) poses a massive threat to marine ecosystems due to its possible impact on calcifying organisms (reviewed by ref. [ 76 ]). In spite of the intensive efforts made lately devoted to exploring the effects of global changes on corals, concerning especially tropical species, our understanding on how their biological and physiological processes may change under OA is still limited (as reviewed by ref. [ 77 , 78 ]). To this concern, temperate species such as B. eurapaea targeted in our study, may represent a valuable model given its more pronounced acclimatization capability with respect to tropical species, being naturally able to accommodate wider seasonal environmental variations (as suggested by ref. [ 37 ]). In the present study, we highlighted genetic functions, included in the microbial metagenome of the anatomic compartments of B. europaea , changing along with OA, and ultimately identify possible bacteria-related acclimatization processes. In terms of phylogenetic composition, we observed variations at the family-level in the microbial community associated to the surface mucus of corals with increasing acidification, whereas significant shifts were not observed in soft tissue and skeleton samples, which are confirmed as ecologically distinct habitats, as previously highlighted [ 79 ]. This is in line with previous findings on microbiome variations induced by acidification in another temperate but non-zooxanthellate coral species ( i.e ., Astroides calycularis [ 30 ]). Our observation is also coherent with the fact that mucus niche is a “first line” defense layer, located at the interface between the coral itself and the surrounding environment, as suggested by Shnit-Orland and Kushmaro [ 80 ], and more recently confirmed by Pollock et al. [ 81 ], whose research on Australian corals pointed at the mucus microbiome as more environmentally responsive than the communities associated to tissue and skeleton. The phylogenetic shift associated to increasing acidification in the mucus microbial community is accompanied by changes at a functional level. For instance, coherently with recent studies showing modifications of the ion transport system in both tropical and temperate corals subjected to pH variations [ 82 , 83 ], transporters of small molecules (i.e., metals, short chain fatty acids, and lipids) underwent changes in prevalence in mucus – and in this case also skeleton - samples from corals living in acidified conditions. The mucus microbiome was the one in which a wider and more consistent gain of functions associated to stress response was observed. Among the stress-related functions gained by the mucus of corals living under highly acidified conditions we could find the mazE gene, a toxin-antitoxin system activated during adverse environmental conditions [ 84 ], a tyrosinase function involved in production of protective pigments during environmental stress [ 72 , 85 ], the pyrroloquinoline quinone (pqq) biosynthesis protein that synthesizes a redox cofactor for bacterial dehydrogenases under environmental stress conditions [ 86 ], a homocysteine S-methyltransferase (mmuM) previously found upregulated under osmotic stress in corals [ 87 ], the gene rhaB that is involved in the response to cell wall and membrane stress in bacteria [ 88 ], and the competence protein ComFA that has a role in DNA uptake during horizontal gene transfer [ 89 ], which represents an adaptive stress-response mechanism in different bacteria [ 90 ]. Moreover, some of the functions acquired with acidification are known to be involved specifically in the resistance to acidic stress in model bacteria; these functions include gluconate 2-dehydrogenase alpha chain [ 91 ] and 2-octaprenyl-3-methyl-6-methoxy-1,4-benzoquinol hydroxylase [ 92 ]. Finally, additional stress-response functions were gained also by the skeleton microbial community in highly acidified conditions, such as the transcriptional regulator nhaR (K03717), responsible for the osmotic induction of a promoter of the stress-inducible gene osmC [ 93 ]. Conversely, in the same compartment, we observed the loss of transporters for important coral osmolytes (arabinose and taurine) used by the metaorganisms to cope with environmental fluctuations [ 79 ]. Taken together, these observations point at a possible mechanism of selection of stress-adaptable microbiome components, which might contribute to the coral acclimatization process. Secondly, the microbiome of corals living in acidified conditions showed quantitative shifts in the pathways of carbohydrate metabolism and changes in processes involving metabolites necessary for maintaining protective cell structures, such as lipid membranes and cell walls. Our findings indicate that carbohydrate metabolism in coral microbiome under OA, in particular in the mucus compartment, was subjected to a shift from energy production to maintenance of cell membrane and wall integrity, with a decrease in direct sugar consumption and an increase in structural sugar biosynthesis pathways (Fig. 4 ). According to our data, the utilization of carbon sources in the mucus underwent a shift in favor of amino and nucleotide sugars metabolism, which are important precursors of the lipidic membranes and cell wall [ 94 , 95 ], to the detriment of energetic pathways, such as glycolysis/gluconeogenesis. Environmental changes, including lowering pH, might lead to cell membrane damages [ 96 ] or alteration in cell structural lipids [ 97 ], meaning that membrane bioenergetics and lipid physiology are closely related to the stress response [ 98 ]. Since a wide range of nucleotide sugars are required for lipopolysaccharide (LPS) biosynthesis [ 95 ], we can assume that changes in environmental conditions might influence the metabolism of nucleotide sugar production by provoking alterations in bacterial cells membranes. Moreover, nucleotide sugars are also essential for sucrose synthesis, with sucrose synthase enzyme having a dual role in producing both UDP-glucose, necessary for cell wall and glycoprotein biosynthesis, and ADP-glucose, necessary for starch biosynthesis [ 99 ]. Hence, the increasing trend of both nucleotide sugar pathway and sucrose metabolism that we observed under high acidification is coherent with the strong interconnection between these two pathways and the damages possibly induced by OA. These findings are also supported by the gain of a short chain fatty acid transporter and of an outer membrane protein involved in membrane repair (Blc) in the metagenome of skeleton samples collected under highly acidified conditions. Among their multiple roles, fatty acids are indeed important structural constituents of phospholipids, which are the building blocks of cell membranes [ 100 ]. In addition, the ability of Blc to bind several fatty acids and lysophospholipids (LPLs), key inner membrane intermediates of phospholipid metabolism, makes this membrane protein likely involved in cell envelope LPL transport in case of membrane damage [ 71 , 101 ]. Fig. 4 Proposed model of carbohydrate metabolism shift in coral microbiomes under acidification conditions, from direct energy production to structural maintenance pathways. KEGG pathways and KO entries showing acidification related modifications in relative abundance or prevalence, respectively, are reported in rectangles using the same color code of the coral compartment in which variations were observed (mucus, yellow; soft tissue, green; skeleton, gray). Upward and downward arrows indicate that the KEGG pathway increased or decreased in terms of relative abundance with the increasing acidification, respectively. Reported KO entries (*) were detected in all replicates from the highly acidified site while absent in samples from control sites. Cell functions hypothesized to be connected with the variations of KEGG pathways abundance and KO entries detection are reported in dashed circles. Dotted arrows represent the influence of acidification on carbohydrates utilization. Dashed arrows represent connections between observed increasing/decreasing pathways and cell functions. Abbreviations: LPS lipopolysaccharides; Ala Alanine; Asp Aspartate; Glu Glutamate. Amino sugars also have an important structural role as components of the prokaryotic cell walls, where they occur in peptidoglycan, LPS, and pseudopeptidoglycan [ 94 ]. For example, it has been shown in a model cyanobacterium that peptidoglycan incorporates L-alanine, D-alanine, D-glutamate and meso-diaminopimelate into peptide bridges, which are linked to polymers consisting of alternating amino sugar (acetyl-glucosamine and acetyl-muramate) monomers [ 95 ]. This is in line with the increasing relative abundance in the pathway of alanine, aspartate and glutamate metabolism within the amino acids metabolism that we observed with augmented OA, although we could not find a direct link between ocean acidification and cell wall modifications in the available literature. Supporting this possible change in peptidoglycan structure due to highly acidified conditions, we also observed the penicillin-binding protein B2 function, involved in the polymerization of peptidoglycan [ 102 ], appearing in mucus and increasing in prevalence in tissue metagenomes under high acidification conditions (Fig. 4 ). Finally, in our model of coral acclimatized to low pH, we observed an acidification-induced selection of functions related to Nitrogen (N) mobilization in the mucus metagenome. As OA alters microorganism biogeochemical environment, it is of particular relevance to understand whether and how it is able to affect N cycling in ecologically relevant benthic holobionts [ 103 ]. Our results suggest that organic N mobilization is promoted by acidification, especially in the mucus, through the gain of the cyanophycinase function. Cyanophycin is a water-insoluble storage biopolymer acting as N reservoir and synthesized by the enzyme cyanophycin synthetase [ 104 ]. Cyanophycinase, responsible for the release of the dipeptide β-aspartyl-arginine from cyanophycin and subsequent hydrolyzation to aspartate and arginine by an isoaspartyl dipeptidase [ 104 – 106 ], only appears in the microbiome associated to corals growing under highly acidified conditions. Accordingly, the appearance of a regulatory protein of arginine utilization in coral mucus growing in acidified sites supports our hypothesis, since arginine is a building block for cyanophycin [ 107 ]. On the contrary, N fixation did not show acidification-related modifications in the involved genetic functions, in our Mediterranean coral model, confirming the importance of this pathway for coral survival pointed out by previous studies using 15 N 2 tracer technique [ 108 ]. Growth and density of Symbiodiniaceae algal symbiont within the coral host is highly dependent on N availability, and N fixation performed by bacteria could contribute to the stability of the coral–algae symbiosis, in particular under sub-optimal scenarios [ 109 ]. It is tempting to hypothesize that, with the increased acidification, the N demand in either all or one among the components of the B. eurapaea holobiont ( i.e ., the prokaryotic community, the coral host, the symbiotic algae) might increase, and that the gain in terms of functions for N mobilization from storage polymers in the microbial community might be a coping strategy for the sub-optimal environmental conditions. Studies linking coral N metabolism and environmental variations, in particular heat and eutrophic stresses, have provided a wide array of different, sometimes contrasting, results, depending on the species. For instance, increased ammonia availability allows the maintenance of photosynthesis and calcification rates in the coral Turbinaria reniformis under thermal stress [ 110 ], whereas an excess of N of anthropogenic origin exacerbates the bleaching reaction to thermal stress in Acropora and Pocillopora [ 111 ]. Pogoreutz et al. [ 112 ] proposed a coral bleaching model in which an increased N fixation is synergic with ocean warming in determining the loss of control over the symbiosis with Symbiodiniaceae in Pocillopora model. However, the balance in nutrients exchange among bacteria, the coral host, and the zooxanthellae is deemed extremely complex, and it involves mechanisms of N limitation and phosphorous starvation to allow the host to exert control on the photosynthesis in the algal symbiont and maintain the symbiotic homeostasis under changing environmental conditions [ 112 ]. To the best of our knowledge, mechanistic studies linking N cycle and tolerance to OA in temperate corals are still unavailable, and our results highlight that any attempt at deepening our knowledge in this field needs to consider the N storage and mobilization pathways and, most importantly, take into account the crosstalk among all the components of the holobiont (coral, algae, and bacteria). Other peculiar changes involved in N metabolism observed in the mucus metagenome was loss of the urease gene (ureA). Urea has been proposed to represent an important metabolite for coral calcification through degradation by urease [ 113 ], which catalyzes the hydrolysis of urea to inorganic carbon and ammonia that are involved in the calcification process [ 114 ]. The net calcification rate of B. europaea is actually reduced with increasing acidification [ 25 ], thus it is possible that a loss in the urease activity might be involved in the process. In conclusion, the metagenomic changes observed in corals acclimatized to low pH suggest a functional shift able to mitigate the sub-optimal environmental conditions at three different levels. First, at mucus level, the low pH of surrounding water could exert a selective pressure on microbiome composition promoting the acquisition of bacteria genetically equipped for dealing with environmental stress, as demonstrated by the gain of functions related to stress resistance. Secondly, the carbohydrate metabolism of the coral microbiome, especially in the mucus compartment, is affected by acidification in ways that hint at a more efficient maintenance of cell protective structures, confirming the importance of membrane bioenergetics in connection to the response to acidification, as previously reported in different contexts [ 98 ]. Thirdly, acidification promotes the selection of genetic functions that can respond to variations in nitrogen needs at the holobiont level possibly in relation to its control over the algal symbiont. Our results point at the importance to consider the crosstalk among all the three components (coral host, symbiotic algae, and bacterial communities) of the holobiont to further unveil nitrogen-involving processes that allow photosynthetic corals to maintain their functionality under adverse environmental conditions. Our study expands the current knowledge on processes of coral acclimatization to OA and confirms that temperate corals represent a promising model of microbiome adaptation. When confirmed by future mechanistic studies, also microbiome transplants in controlled environment, the processes hypothesized in the present work represent an important step towards a holistic comprehension of the tripartite crosstalk between coral host, symbiotic algae and bacterial communities, as well as a deepened understanding on how this relationship changes under environmental variations allowing for the survival and health of these ecosystem forming holobionts in the forthcoming oceans."
} | 5,936 |
27489567 | PMC4971668 | pmc | 235 | {
"abstract": "Background Biohythane is a new and high-value transportation fuel present as a mixture of biomethane and biohydrogen. It has been produced from different organic matters using anaerobic digestion. Bioenergy can be recovered from waste activated sludge through methane production during anaerobic digestion, but energy yield is often insufficient to sludge disposal. Microbial electrolysis cell (MEC) is also a promising approach for bioenergy recovery and waste sludge disposal as higher energy efficiency and biogas production. The systematic understanding of microbial interactions and biohythane production in MEC is still limited. Here, we report biohythane production from waste sludge in biocathode microbial electrolysis cells and reveal syntrophic interactions in microbial communities based on high-throughput sequencing and quantitative PCR targeting 16S rRNA gene. Results The alkali-pretreated sludge fed MECs (AS-MEC) showed the highest biohythane production rate of 0.148 L·L −1 -reactor·day −1 , which is 40 and 80 % higher than raw sludge fed MECs (RS-MEC) and anaerobic digestion (open circuit MEC, RS-OCMEC). Current density, metabolite profiles, and hydrogen-methane ratio results all confirm that alkali-pretreatment and microbial electrolysis greatly enhanced sludge hydrolysis and biohythane production. Illumina Miseq sequencing of 16S rRNA gene amplicons indicates that anode biofilm was dominated by exoelectrogenic Geobacter , fermentative bacteria and hydrogen-producing bacteria in the AS-MEC. The cathode biofilm was dominated by fermentative Clostridium . The dominant archaeal populations on the cathodes of AS-MEC and RS-MEC were affiliated with hydrogenotrophic Methanobacterium (98 %, relative abundance) and Methanocorpusculum (77 %), respectively. Multiple pathways of gas production were observed in the same MEC reactor, including fermentative and electrolytic H 2 production, as well as hydrogenotrophic methanogenesis and electromethanogenesis. Real-time quantitative PCR analyses showed that higher amount of methanogens were enriched in AS-MEC than that in RS-MEC and RS-OCMEC, suggesting that alkali-pretreated sludge and MEC facilitated hydrogenotrophic methanogen enrichment. Conclusion This study proves for the first time that biohythane could be produced directly in biocathode MECs using waste sludge. MEC and alkali-pretreatment accelerated enrichment of hydrogenotrophic methanogen and hydrolysis of waste sludge. The results indicate syntrophic interactions among fermentative bacteria, exoelectrogenic bacteria and methanogenic archaea in MECs are critical for highly efficient conversion of complex organics into biohythane, demonstrating that MECs can be more competitive than conventional anaerobic digestion for biohythane production using carbohydrate-deficient substrates. Biohythane production from waste sludge by MEC provides a promising new way for practical application of microbial electrochemical technology. Electronic supplementary material The online version of this article (doi:10.1186/s13068-016-0579-x) contains supplementary material, which is available to authorized users.",
"conclusion": "Conclusion This study proved that biohythane could be produced directly in biocathode MECs using waste sludge. The highest biohythane production rate of 0.148 L·L −1 -reactor·day −1 was obtained in the alkali-pretreated sludge fed MECs (AS-MEC), which was 80 % higher than that in the anaerobic digestion. Real-time quantitative PCR and VFAs results demonstrated that MEC and alkali-pretreatment accelerated enrichment of hydrogenotrophic methanogen and hydrolysis of waste sludge that resulted in a higher biohythane production. The most predominant population on the anode of AS-MEC was affiliated to exoelectrogenic Geobacter , while biocathode was dominated by fermentative Clostridium . The majority of methanogenic archaea on the cathodes of AS-MEC belonged to hydrogenotrophic Methanobacterium . The community analyses implied that multiple syntrophic interactions between fermentative bacteria, exoelectrogenes and methanogenic archaea in MECs drive biohythane production from waste sludge. Compared to anaerobic digestion, biohythane production by MEC became more competitive using carbohydrate-deficient substrates, and provided a new approach for bioenergy production using waste sludge.",
"discussion": "Discussion MECs enhance the hydrolysis of waste sludge and biohythane production This study proves for the first time that biohythane could be produced directly in biocathode MECs using raw or alkali-pretreated waste sludge. MECs showed the highest biohythane production from alkali-pretreated sludge. Both MECs and conventional anaerobic digestion (open-circuit MEC) showed considerable biohythane production using raw sludge. No methane and hydrogen were detected in alkali-pretreated sludge fed open circuit MEC (AS-OCMEC) during a 9-d operation (Fig. 1 ). The community analyses indicated that archaeal community in RS-OCMEC dominated by an acetoclastic Methanosaeta [ 34 ], while the majority of dominant archaeal populations in MECs belonged to hydrogenotrophic methanogens ( Methanocorpusculum and Methanobacterium ) (Fig. 6 a). Obviously, the alkali-pretreatment suppressed acetoclastic methanogens in the raw waste sludge and facilitated the acidogenesis that provide the VFAs for exoelectrogen enrichment. Nevertheless, hydrogenotrophic methanogens or electromethanogens prevailed fast and contributed to biohythane production in MECs during a 9-d operation, not in AS-OCMEC. These results have showed that MEC has advantage of fast reaction velocity compared with anaerobic digestion as reported previously [ 10 , 22 , 25 , 35 ]. The biogas component of AS-MECs was almost consistent with commercial hythane [ 2 ]. Alkali-pretreatment played an important role in accelerating succedent decomposition of waste sludge, which enhanced biohythane production in AS-MECs (Fig. 1 ). The results showed that waste sludge is an appropriate substrate for biohythane production by MECs. In contrast with two-phase anaerobic digestion, biohythane production by MEC became more competitive using carbohydrate-deficient substrates. A recent study showed that high concentration methane of 95 % was produced from waste activated sludge using MEC at ambient temperature [ 24 ]. To optimize biohythane composition, organic loading rate, sludge retention time, temperature, substrate variety, cathode potential and system integration should be investigated in the future. The biohythane of a full-scale MEC reactor can be collected continuously using a gas storage tank before use in the industrial applications. The component of biohythane can be adjusted (5–20 % of hydrogen) using a gas blending systems to meet the end-use devices such as household appliances and vehicles, which approach is same as the hythane production. The acetic acid concentration in AS-MEC was two times higher than that in open-circuit AS-MEC by day 2 (Fig. 3 ), suggesting that microbial electrochemical system facilitated the acidification of alkaline pretreated waste sludge compared with conventional anaerobic digestion [ 10 ]. As alkaline pretreatment destroyed sludge flocs and accelerated organic matter’s hydrolysis, the acidogenesis in open-circuit AS-MEC was better than that in RS-MEC and RS-OCMEC [ 35 ]. However, no methane was detected in AS-OCMEC in 9 days, suggesting the majority of acetoclastic methanogens in the initial raw sludge were lysed certainly during the alkali-pretreatment. MEC also accelerated methanogen enrichment that resulted in a higher biohythane production rate. Propionic acid as a central intermediate often accumulated in the degradation of complex organic matters, especially in methanogenic environments. VFAs analyses showed that propionic acid accumulation (200–300 mg/L) present in close and open circuit AS-MEC after 9 days, suggesting that enriching propionate-oxidizing acetogenic bacteria in MECs may further enhance biohythane production from waste sludge. Biohythane provides a new perspective to view methanogenesis in hydrogen-producing MECs Hydrogen re-consumption by hydrogenotrophic methanogens in MECs has been a major challenge for hydrogen-producing MECs [ 36 , 37 ]. To achieve a high yield and high purity of H 2 in MECs, several methods including methanogen inhibitors (e.g., bromoethanesulfonate, lumazine), short hydraulic retention time, intermittent exposure to air and low temperatures have been used to depress methanogenesis [ 16 ]. The methanogens could be significantly repressed at the relatively low temperatures [ 16 , 37 ], suggesting that MEC should be operated at 15 °C considering both hydrogen production and methanogenesis inhibition. Hydrogenotrophic methanogens will prevail over time when hydrogen-producing MECs using waste sludge are operated above room temperature. Biohythane as mixture of biomethane and biohydrogen produced from organic waste could be directly used in internal combustion engines, which offered an alternative approach to solve troublesome methanogenesis in hydrogen-producing MECs. Multiple syntrophic interactions drive cascade utilization of waste sludge in MECs Syntrophy is an essential intermediary step in the anaerobic metabolism, especially for the complete conversion of complex polymers such as polysaccharides, proteins, nucleic acids, and lipids to methane [ 38 ]. Metabolic crossfeeding is an important process that can broadly shape microbial communities. Illumina Miseq sequencing and principal component analyses indicate that microbial community structures greatly distinguished from each other in samples obtained from different reactors (Figs. 4 , 5 ). Diverse trophic groups in MECs belonged to primary/secondary fermentative bacteria (proteolytic and saccharolytic bacteria, hydrogen-producing bacteria), acetogenic bacteria, exoelectrogenic bacteria and hydrogenotrophic methanogenic archaea according to the taxonomic identification [ 39 ]. The predominant populations in the anode biofilm of AS-MEC were affiliated with Geobacter (22 %), Alistipes (10 %), Spirochaeta (9 %), Proteiniphilum (6 %) and Petrimonas (3 %). The relative abundance of exoelectrogenic Geobacter was higher in AS-MEC than that in other MECs, which is consistent with the findings of higher current production because Geobacter is the most efficient exoelectrogen using acetate reported in literature. Alistipes can produce VFAs and hydrogen using protein and carbohydrates [ 40 ]. Spirochaeta as saccharolytic bacterium is responsible for decomposition of (poly) carbohydrates and production of acetate, carbon dioxide and hydrogen [ 41 ]. Proteiniphilum as proteolytic bacterium is capable of producing acetic and propionic acids using yeast extract, peptone and arginine [ 42 ], and its relative abundance increased with the order of RS-OCMEC, RS-MEC and AS-MEC. Petrimonas , an acidogenic bacterium, can degrade protein and carbohydrates, which was also reported in previous studies as a predominant genus in sludge fed MECs [ 11 , 43 ]. The majority of predominant genera in the cathode biofilm of AS-MEC belonged to putative hydrogen-producing Clostridium (15 %). The sequencing analyses indicated putative fermentative hydrogen-producing bacteria were enriched in both electrode biofilms, and hydrogen production on the electrodes was also proved by hydrogen microsensor measurements (Fig. 2 ). Archaeal community analyses indicated that the majority of methanogenic populations was affiliated with hydrogenotrophic Methanocorpusculum (relative abundance of 85 %) and Methanobacterium (98 %) in the anode and cathode biofilms of AS-MEC, respectively (Fig. 6 a). Methanobacterium capable of electromethanogenesis has been reported, which was the most predominant methanogen in the cathode biofilm of electromethanogenic MEC using inorganic carbon source [ 29 ]. The predominant populations in the biofilms proved that hydrogen production by fermentation and electrolytic process, hydrogenotrophic methanogenesis and electromethanogenesis occurred simultaneously in the single-chamber MECs. The microbial community structure reveals that different functional groups interacted synergistically in the MEC reactors to convert recalcitrant sludge into biohythane. The multiple levels of interactions in these syntrophic consortia include three groups. First metabolic crossfeeding occurred between fermentative and acetogenic bacteria and exoelectrogenic bacteria. Fermentative and acetogenic bacteria also partnered with methanogenic archaea. Real-time quantitative PCR results showed that the amount of methanogens was higher in AS-MEC than that in RS-MEC and RS-OCMEC (Fig. 6 b), suggesting that alkali-pretreatment and MEC facilitated hydrogenotrophic methanogen enrichment in the anode and cathode biofilms as hydrogen production. Compared to the cathode biofilm of AS-MEC, the anode biofilm enriched large amount of methanogens (Fig. 6 b), implying that third syntropic interaction may occur between methanogenic archaea and exoelectrogenic bacteria on the anode as reported previously [ 44 ]. However, putative interspecies electron transfer between Methanocorpusculum and Geobacter should be further proved based on co-culture test."
} | 3,331 |
31692992 | PMC6827597 | pmc | 236 | {
"abstract": "Abstract Neuromorphic systems consisting of artificial neurons and synapses can process complex information with high efficiency to overcome the bottleneck of von Neumann architecture. Artificial neurons are essentially required to possess functions such as leaky integrate‐and‐fire and output spike. However, previous reported artificial neurons typically have high operation voltage and large leakage current, leading to significant power consumption, which is contrary to the energy‐efficient biological model. Here, an oscillatory neuron based on Ag filamentary threshold switching memristor (TS) that has a low operation voltage (<0.6 V) with ultralow power consumption (<1.8 µW) is presented. It can trigger neuronal functions, including leaky integrate‐and‐fire and threshold‐driven spiking output, with high endurance (>10 8 cycles). Being connected to an external resistor or a resistive switching memristor (RS) as synaptic weight, the TS clearly demonstrates self‐oscillation behavior once the input pulse voltage exceeds the threshold voltage. Meanwhile, the oscillation frequency is proportional to the input pulse voltage and the conductance of RS synapse, which can be used to integrate the weighted sum current. As an energy‐efficient memristor‐based spiking neural network, this combination of TS oscillatory neuron with RS synapse is further evaluated for image recognition achieving an accuracy of 79.2 ± 2.4% for CIFAR‐10 subset."
} | 362 |
30197472 | null | s2 | 237 | {
"abstract": "The balletic motion of bird flocks, fish schools, and human crowds is believed to emerge from local interactions between individuals, in a process of self-organization. The key to explaining such collective behavior thus lies in understanding these local interactions. After decades of theoretical modeling, experiments using virtual crowds and analysis of real crowd data are enabling us to decipher the 'rules' governing these interactions. Based on such results, we build a dynamical model of how a pedestrian aligns their motion with that of a neighbor, and how these binary interactions are combined within a neighborhood in a crowd. Computer simulations of the model generate coherent motion at the global level and reproduce individual trajectories at the local level. This approach yields the first experiment-driven, bottom-up model of collective motion, providing a basis for understanding more complex patterns of crowd behavior in both everyday and emergency situations."
} | 245 |
24786096 | PMC4057678 | pmc | 239 | {
"abstract": "With the ever-increasing population of the world (expected to reach 9.6 billion by 2050), and altered life style, comes an increased demand for food, fuel and fiber. However, scarcity of land, water and energy accompanied by climate change means that to produce enough to meet the demands is getting increasingly challenging. Today we must use every avenue from science and technology available to address these challenges. The natural process of symbiotic nitrogen fixation, whereby plants such as legumes fix atmospheric nitrogen gas to ammonia, usable by plants can have a substantial impact as it is found in nature, has low environmental and economic costs and is broadly established. Here we look at the importance of symbiotic nitrogen fixation in the production of biofuel feedstocks; how this process can address major challenges, how improving nitrogen fixation is essential, and what we can do about it.",
"conclusion": "4. Conclusions Today growth of food and biofuel plants relies heavily on nitrogen fertiliser, production of which is dependent on fossil fuels. However, the real danger of depletion of existing fossil fuels, associated cost increases in the depletion period, and the environmental impact of their use is making it important to look for alternatives to synthetic fertilisers. Symbiotic nitrogen fixation by legume plants provides such an alternative. Legumes are unique in that they have the ability to form a symbiotic relationship with nitrogen-fixing bacteria (collectively called rhizobia), which are housed in special root organs called nodules (NB, there are other nitrogen-fixing symbioses with plants and the bacterium called “Frankia”, i.e ., with the non-legume casuarina or elm trees. This field of investigation has been highly descriptive, since the Frankia bacterium is difficult to handle microbiologically). Several strategies to transfer the process via inter-specific gene transfer to non-legumes like rice and corn are being pursued (reviewed by [ 73 ]). The fact that the more common mycorrhizal fungus symbiosis shares many common signaling elements (SymRK to CCaMK) with the rhizobial symbiosis gives hope that transfer of the additional elements through biotechnological methods may be possible. However, the transfer to a monocot non-legume may stretch the possibilities as the inherent phytohormone regulation of monocot and dicot is highly divergent ( c.f. , the differential auxin herbicide toxicity spectrum). Perhaps it would be easier to consider nodulation of non-legumes like potato or canola, (a) because they are dicot and (b) because other dicot non-legumes nodulate and fix nitrogen (e.g., Parasponia and the Frankia –nodulated plants). The most important target for genetic engineering is the Nod-factor receptor (NFR1), since Nod-factors as well as the Myc-factors, both being lipochito-oligosaccharides suggest a broader recognition spectrum. Several nodulation-associated transcription factors are also being targeted. Another more ambitious approach is to introduce the nitrogenase enzyme complex into plant cells via chloroplast transformation. The resulting “nitroplast” would harvest solar energy, but if inactivated by mutation for Photosystem II would not yield abundant oxygen. This concept was already developed by John Postgate, Ray Dixon and associates nearly 40 years ago; however, despite elegant biotechnological approaches, no significant advances have been achieved. Success in any of these methods will have a drastic impact on the way food and biofuel plants are grown. However, these methods are still some way away, and in the meantime we propose the use of non-food legumes for direct biofuel production and companion cropping. In addition to being a legume, it is important for a biofuel feedstock not to compete with food crops for land, water and labour. Pongamia pinnata is such a plant, as being drought- and salinity-tolerant; it can grow on marginal land not suitable for most food crops. Globally such land areas are abundant. However, more importantly, being a legume means that it does not require supplied nitrogen fertilisers, thereby increasing its sustainability.",
"introduction": "1. Introduction Two of humanity’s major needs are food and energy. With the human population rising at an alarming rate (1 billion increase from present 7.2 billion in the next 12 years (UN estimate)), food security, mainly due to increasing scarcity of land and water resources has become a major political and scientific concern. Ever since humans transitioned from hunter-gatherers to a stable agriculture-based society, improvement of crops has been a major goal. Early farmers observed that land became less productive when planted year after year and concluded that plants absorbed certain nutrients from the soil. In the 1730s, crop rotation was implemented in Europe as a method to improve productivity of major crops. However, it was not until the mid-1800s that the understanding of plant nutrition had advanced enough to realise the importance of added nitrogen and phosphorus to the soil. The German chemist Justus von Liebig was the first to promote the importance of ammonia and inorganic minerals to plant nutrition and developed the first commercial fertiliser by treating phosphate of lime in bone meal with sulphuric acid. Although this failed, because of not being properly absorbed by the crops, it started a trend in fertiliser development. The early fertilisers were mainly based on manures and the effects of many types of manure on plant growth were tested and used. The Rothamsted (UK) Research Station, started at this time by the British entrepreneur John Bennet Lawes and Joseph Gilbert (a student of von Liebig) is still involved in the study of the effect of organic and inorganic fertilisers on crop yield. However, it was not until the process of atmospheric nitrogen fixation was established, first by Henry Cavendish in 1784, that synthetic fertilisers became widespread. His process was soon replaced by a more efficient Haber-Bosch process, which revolutionised agriculture and won the inventors Fritz Haber and Carl Bosch a Nobel Prize in chemistry (separately in 1918 and 1932!). The process utilises molecular nitrogen (N 2 , available in abundance (78%) in the atmosphere) and methane (CH 4 ) in an economically sustainable, though environmentally expensive synthesis of ammonia (NH 3 ). The ammonia produced this way is used as a raw material by the modern chemical industry for the production of most of the commonly used fertilisers, such as nitrates. The Haber-Bosch process is one of only three ways in which inert atmospheric N 2 is converted to NH 3 , the other two being biological nitrogen fixation by prokaryotic microbes containing the nitrogenase enzyme complex, and geochemical conversion by lightening. Today crop producers world-wide rely heavily on synthetic fertilisers to enhance plant productivity; this trend seems likely to continue as a steadily rising population needs increased food mass and quality as well as renewable fuel. The problem with the “fertiliser scenario” is that plants absorb at any one time only a small percentage of this applied supplement. The majority of it (30% to 50%) [ 1 ] is wasted and runs off into waterways causing environmental pollution on a massive scale ( i.e ., the Mississippi River Delta). In many areas algal blooms and eutrophication are huge problems. In addition the nitrogen in the soil is broken down by soil bacteria, through a process called denitrification, to N 2 O (nitrous oxide), which reacts with oxygen to give rise to NO (nitric oxide), which in turn reacts with ozone (O 3 ). Natural sources of N 2 0 are soils (contributing 6.6 Tg N/year), oceans, rivers, and estuaries (contributing 5.4 Tg N/year). According to the US Environmental Protection Agency (EPA), N 2 O has about 294 times higher impact per unit mass (global warming potential) than carbon dioxide and therefore even in small quantities can contribute as GHG in a big way. Application of nitrogenous fertilisers accounts for the majority of N 2 O emissions. It has been suggested that for every 100 kg of fertiliser N added to the soil, on average 1.25 kg of N is emitted as N 2 O, which is equivalent in GHG effect to around 600 kg of CO 2 . The GHGs affect the temperature of the earth by absorbing and emitting radiation within the thermal infrared range (the “Greenhouse Effect”). Moreover, every step in the production, delivery and application of nitrogen fertiliser requires fossil fuels. Even though formation of fossil fuels is occurring naturally through anaerobic decomposition of buried plants and animals, they are considered non-renewable as they take millions of years to form in large quantities, and reserves are being depleted much faster than new ones are being formed. The world energy consumption has increased dramatically in recent times, growing at the rate of 2.3% each year [ 2 ]. Unfortunately an estimated 86% of this energy comes from burning of fossil fuels (36% from petroleum, 27% from coal and 23% from natural gas [ 2 ]. The current demand for oil from fossil fuels is around 85 million barrels per day (about 159 litres per barrel), which is expected to rise to around 106 million barrels per day by 2030 [ 3 ]. In addition, burning of fossil fuel is considered to be the largest source of GHG emissions due to human activity, with electricity production (coal combustion), transportation (petrol, diesel and aviation fuel) and industry (gas and coal) being the major culprits."
} | 2,386 |
39689059 | PMC11651616 | pmc | 240 | {
"abstract": "Social insects, such as ants and bees, are known for their highly efficient and structured colonies. Division of labour, in which each member of the colony has a specific role, is considered to be one major driver of their ecological success. However, empirical evidence has accumulated showing that many workers, sometimes more than half, remain idle in insect societies. Several hypotheses have been put forward to explain these patterns, but none provides a consensual explanation. Task specialisation exploits inter-individual variations, which are mainly influenced by genetic factors beyond the control of the colony. As a result, individuals may also differ in the efficiency with which they perform tasks. In this context, we aimed to test the hypothesis that colonies generate a large number of individuals in order to recruit only the most efficient to perform tasks, at the cost of producing and maintaining a fraction of workers that remain inactive. We developed a model to explore the conditions under which variations in the scaling of workers’ production and maintenance costs, along with activity costs, allow colonies to sustain a fraction of inactive workers. We sampled individual performances according to different random distributions in order to simulate the variability associated with worker efficiency. Our results show that the inactivity of part of the workforce can be beneficial for a wide range of parameters if it allows colonies to select the most efficient workers. In decentralised systems such as insect societies, we suggest that inactivity is a by-product of the random processes associated with the generation of individuals whose performance levels cannot be controlled.",
"introduction": "Introduction Colonies of eusocial insects like ants, corbiculate bees, and termites are renowned for their bustling activity. However, several studies reveal a surprising and yet common feature of these societies: a significant portion of the workers, sometimes more than half of them, remains inactive for extended periods of time [ 1 , 2 ]. Why would colonies invest resources in producing a large workforce if a substantial number of individuals remain inactive? Both theoretical and experimental work have explored the potential adaptive benefits of inactivity [ 3 – 7 ]. A frequently invoked hypothesis proposes that these inactive individuals act as a reserve workforce, readily available to respond to surges in colony needs [ 4 , 8 ]. This assumption is based on the fact that individuals with high response thresholds, previously inactive, are recruited when task demands increase or when workers previously engaged in those tasks are lost [ 3 ]. However, empirical support for this hypothesis is mixed. For example, in Temnothorax rugatulus ant colonies, removing the most active workers leads to the recruitment of previously inactive workers to compensate for this loss [ 4 ]. However, in bumble bee colonies, removing the most active individuals before an increased need for defence or thermoregulation did not activate idle workers [ 9 , 10 ]. Instead, already active bees increased their activity to compensate. Even if inactive workers can be recruited when the colony’s needs increase in certain circumstances, we can still question the original cause of inactivity. So far, no comprehensive hypothesis has been put forward to explain the widespread, and seemingly universal, occurrence of inactivity in social insect colonies. Identifying a common origin of inactivity across different taxa implies finding a mechanism shared by social insects that is related to the division of labour, with individual variations being key [ 11 , 12 ]. Here, we propose that inactivity could be a by-product of a process that selectively favours the best-performing workers when individual performance is determined randomly. The interplay between genetic and non-genetic factors shapes the behavioural diversity of workers. Environmental factors like food and temperature experienced by the brood during development can be partially controlled by nurses. For example, exposure of ant brood to different temperatures can influence worker sensitivity to thermal cues in adulthood [ 13 ]. On the other hand, the genetic background of workers, produced from fertilized eggs laid by the queen, remains beyond any control. These very mechanisms that foster a diverse workforce may also result in colonies producing workers with unpredictable efficiency. If worker efficiency varies, a potential optimal strategy from a colony perspective might involve producing a large number of individuals to increase the likelihood of obtaining highly efficient ones. This ensures that the most competent workers complete tasks efficiently, even if this entails costs by keeping others in an inactive state. We aim to test this hypothesis from a theoretical perspective, considering the costs and benefits associated with the presence of inactive individuals within colonies. Two types of costs are usually distinguished: production costs arise from the renewal of the workforce, while maintenance costs are incurred for workers’ operation [ 14 ]. The later differ depending on activity level, with inactive workers incurring lower costs compared to those actively engaged in task completion. In the leaf-cutter ant Atta sexdens , for example, ants engaged in digging expend 15% more energy than their inactive nestmates [ 15 ]. Active workers can also vary in their performance when engaged in their tasks, with some individuals being outperforming others. Foraging success and defensive behaviours, for instance, can vary significantly among workers in social insects [ 16 – 20 ]. In this context, we introduce a simple model to test the influence on colony performance of the ratio between maintenance costs and production costs (hereafter, coefficient δ), as well as of the ratio between maintenance costs of inactive individuals and those of active individuals (hereafter, coefficient β ), when workers’ performances are drawn from different random distributions.",
"discussion": "Results and discussion For each of five random distributions and different combinations of δ and β values, we determined which optimal fraction of active workers, chosen among the individuals with the highest performance and leaving the other workers inactive, maximized colony efficiency ( Fig 1 ). The notion of individual performance here indicates that workers vary in their sensitivity to task-associated stimuli (i.e., differences in response thresholds [ 11 ]) or in their effectiveness in completing tasks. The selection of the most efficient workers can be understood by the competition between workers, where only the most ‘motivated’ engage in tasks [ 21 ]. Our model shows that it can be beneficial for colonies to leave low-performing workers inactive when their maintenance costs are high relative to their production costs (i.e., high δ), if this allows the best-performing individuals to be recruited. Such strategy is also advantageous if the maintenance costs of active workers are higher than those of inactive individuals (i.e., low β ). In contrast, when the maintenance costs of active and inactive workers are comparable (i.e., high β ), colonies should minimise the fraction of inactive individuals. Under these conditions, maintaining inactive workers would only produce marginal savings, so it is worthwhile for the colonies to recruit all individuals, including those with low individual performance. Evaluating efficiency by the difference between performance and costs, rather than the ratio, yielded similar results ( S1 Fig ), confirming the robustness of our conclusions regardless of the calculation method. Interestingly, the different random distributions we explored gave rise to similar qualitative patterns, demonstrating the generality of our findings. The normal, uniform and bimodal distributions used in our simulations have the same mean ( μ = 0.5) for individual performance but differed in their standard deviations (SD = 0.19, SD = 0.29, SD = 0.39, respectively). Our results showed that increasing variance amplified the range of parameter values ( β , δ) where maintaining a pool of inactive workers is most profitable for colonies ( Fig 1 ). 10.1371/journal.pcbi.1012668.g001 Fig 1 Colony efficiency as a function of the proportion of active workers in colonies of 500 individuals. Left: Histogram of individual task performances for five different random distributions. Middle: Colony efficiency as a function of the fraction of active workers for values of δ = 2 and β = 0.2 (arbitrary choice of values to illustrate methods). δ corresponds to the ratio of maintenance costs to production costs and β to the ratio of maintenance costs for inactive to active individuals. For each fraction of active workers, colony efficiency is normalized by the value obtained when all individuals are active. The red dotted line indicates the fraction of active individuals maximizing colony efficiency for this combination of δ and β values. Right: Heat maps of the fraction of individuals that need to be active to maximize colony efficiency for different combinations of δ and β . For results with group size of 50 and 5,000 see S1 Fig . When the distribution of individual performances is left-skewed, with the majority of group members being high performers, colonies maximize efficiency by keeping all workers active. In situations of intraspecific competition, colonies might aim to direct worker production towards the most efficient individuals in order to out-compete rival groups. However, since colonies of the same species share the same mechanisms and constraints in worker production, they are equally influenced by chance in determining individual quality. Thus, no colony should be able to dominate intraspecific competition because of its ability to control the distribution of workers. Several studies have documented that individual worker performance often follows a right-skewed distribution [ 18 – 20 ]. Our simulations show that the highest fraction of inactive individuals maximizing colony efficiency was obtained with such right-skewed distribution ( Fig 1 ). Empirical data also suggests that, in ant colonies, maintenance costs typically exceed production costs (i.e., δ>1) [ 14 ]. This finding aligns with the value found in our simulations, where maintaining an inactive fraction of workers becomes advantageous. The higher the ratio of maintenance costs to production costs, the greater the benefit of keeping a large proportion of workers inactive. Refining our model to include different costs associated with morphological differences, variations in performance over an individual’s lifetime or an influence of group size on social dynamics would not change our main conclusions. Overall, the best strategy for colonies with inefficient workers is to keep them inactive, unless they are cheap to maintain and their cost of activity is low. Assuming that variations in colony size have no influence at the individual level, our analysis also showed that changes in the number of workers had no impact on the patterns observed ( S2 Fig ). However, groups cannot expand indefinitely, as this entails additional costs, such as an increased risk of infection or greater competition [ 22 ]. In multi-level societies, fission-fusion dynamics can mitigate the costs associated with belonging to a large group by splitting into smaller units [ 23 ], but social insect colonies generally remain cohesive (except when colonies are swarming to reproduce). Further analysis should assess how the societies with different social organisations and levels of cohesion deal with the presence of inactive individuals. Our conclusions are based on the assumption that colonies have no control over workers’ performance. This hypothesis is consistent with the occurrence of multiple mating in several species of social insects, such as in honey bees where queens can mate with up to 30 males, which is a means of increasing the breadth of distributions where workers are sampled and improving colony homeostasis [ 24 ]. However, to a certain extent, societies can develop strategies to minimise the impact of chance on the behavioural trajectory of individuals. For instance, around 30% of ant genera exhibit polymorphism with workers from distinct physical sub-castes tending to engage in different tasks [ 25 , 26 ]. These variations in worker body size are primarily attributed to epigenetic factors, such as larval nutrition or exposure to abiotic factors that are under the control of the workers [ 27 ]. In the ant Pheidole pallidula , for example, foragers exposed to competitors during foraging bouts direct brood development towards the production of soldiers to improve colony defence [ 28 ]. From our perspective, the evolution of polymorphism can be viewed as a mechanism for minimizing the role of chance in determining the behavioural destiny of the workforce. However, this reduces but does not eliminate the contribution of randomness, as stochasticity is likely to shape individual behaviour within each subcaste. Such polymorphic species would be ideal systems for testing our model. In these species, the ratio of maintenance to production costs (i.e., δ) is higher for major workers than for minors [ 29 ] and major workers tend to exhibit more inactivity than minor workers [ 30 ]. However, we currently lack data on how maintenance costs differ between active and inactive states within each caste (value of β), which complicates predictions about how activity levels should vary between castes according to our model. The presence of inactive workers does not necessarily jeopardize colonies because the remaining active individuals are sufficient to cover all needs. In ants, for instance, the energy costs expended by a forager on a foraging trip are far outweighed by the energy benefits gained from the retrieved food [ 31 , 32 ]. Even if there are variations between species, the energy returns from successful foraging trips can exceed the metabolic costs of its foraging trips by several orders of magnitude [ 33 ]. Therefore, a fraction of active individuals can readily cover the expenses of the colonies, making inactive individuals less of a burden. However, it is important to note that this strategy might not work in other collective systems. The inactivity of a portion of the group could be detrimental if the remaining active individuals cannot fulfil all requirements or if inactive elements directly hinder overall efficiency. For instance, in a sports team, inactive players can ruin collective performance, such as in tug-of-war or rowing, where success depends on everyone’s effort. Our results can explain the presence of inactive workers within social systems and shed light on why no adaptive hypothesis has been successful in providing a compelling justification for their widespread existence. One might wonder why colonies do not eliminate inactive workers to reduce their costs, even if minimal. In our opinion, colonies lack the ability to evaluate individuals’ performance on collective outcome. This might also be true for other systems lacking supervision or control mechanisms to assess individuals’ performance. In contrast, more centralized systems such as human corporations implement recruitment procedures based on an evaluation of individual skills and competences, thereby reducing the influence of random processes in selection. Our conclusion that inactive individuals can be maintained within colonies because there is no control over their performance echoes one explanation introduced to account for redundancy in collectives [ 34 ]. Redundancy—where groups maintain more members than are strictly needed for task completion—has been observed in a range of biological systems, from cellular assemblies to animal societies [ 35 – 37 ]. The challenge is to explain how groups avoid the invasion of defectors when individuals pay high costs but contribute minimally to the outcome. One hypothesis suggests that cooperative behaviours are still possible when individuals are unaware of others’ strategies [ 34 ]. The inability of group members to evaluate the efforts or contribution of others may therefore play an important, but often overlooked, role in shaping workload distribution in collective systems. In conclusion, our study shows that the presence of inactive individuals is not necessarily detrimental to the functioning of society. The strength of our simple model lies in introducing a new hypothesis: that a single mechanism can explain inactivity in social groups, something previous hypotheses have not successfully achieved. The presence of inactive individuals could then be co-opted at later stages to fulfil functions that were not the primary cause of their existence. For instance, these individuals could act as a reserve workforce that can be recruited to cover certain additional needs or to serve as living reservoirs to store food [ 1 ]. Overall, our results should encourage us to consider that any behaviour should not be regarded as adaptive, but that it may be the result of an uncontrollable random process, a dimension frequently underestimated in the study of social systems."
} | 4,326 |
37253015 | PMC10266060 | pmc | 241 | {
"abstract": "Significance Bio-inspired self-healing conductors may significantly extend the service life of electronic devices. The promising potentials are often challenged by slow and nonspontaneous reparation processes demanding external triggering conditions. We have introduced an innovative design of self-healing conductors with extremely low thresholds to restore electrical properties. The healing mechanism applies to different levels of structural damages from tiny microcracks to severe fractures. Our work advances electrically self-healing conductors to enable robust device performances in flexible and stretchable electronics.",
"conclusion": "Conclusion In summary, we have demonstrated the universal self-healing capability of a compliant conductor suitable for minor injuries and severe fractures under various stress conditions. In a scalable fabrication process, screen-printed liquid metal microcapsules are coated with a conductive Cu layer by electroless deposition to establish strong interfacial coupling. The unique healing mechanism is associated with the rapid and efficient rupture of embedded liquid metal microcapsules by the stress field of microcracks under bending, folding, and stretching conditions. The self-adaptive release of liquid metal at the damaged locations efficiently repairs the damaged conductive layer, which is extremely sensitive to tiny microcracks and also applicable to large fractures. The compliant conductor exhibits high electrical conductivity of ∼12,000 S/cm, an ultralow threshold to activate the self-healing processes, and ultrahigh deformability of up to 1,200% strain. The conductor is employed to create durable devices including an LED matrix display and a skin-attachable electronic patch. A generic design strategy is introduced here to improve the self-healing sensitivity and capability of compliant conductors, which may open up a wide range of applications in flexible and stretchable electronics.",
"discussion": "Results and Discussion Design and Preparation of Electrically Self-Healing Conductors. As schematically illustrated in Fig. 1 A , the electrically self-healing conductor exhibits a bilayer architecture comprising a conductive copper layer on top of liquid metal microcapsules. Cracks in the Cu layer are generated as the major degradation mechanism under stress conditions during long-term practical applications ( 20 ). Liquid metal microcapsules are punctured by the tip-stress field of the cracks along the propagation paths due to the strong interfacial coupling with the Cu layer ( 37 , 38 ). The released liquid metal recovers the electrical properties by reestablishing the lost conducting pathways in the fractured Cu conductor. A top-down, high-yield synthesis utilizes ultrasonication to create liquid metal/fatty acid core-shell microcapsules with controlled sizes ( SI Appendix , Fig. S1 ) ( 33 ). A viscous ink is formulated by mixing 2.8 µm-sized microcapsules and a styrene−isoprene−styrene (SIS) elastomer in toluene solution. In Fig. 1 B , screen printing is employed as a scalable approach to create arbitrary patterns as exemplified by the flower feature in metallic gray color. The printed composite is an electrical insulator consisting of isolated liquid metal microcapsules, as shown by scanning electronic microscopy (SEM) image in Fig. 1 C . In a chemical bath, a galvanic displacement reaction selectively generates Cu seeds over liquid metal microcapsules ( 39 , 40 ), thereby promoting the subsequent electroless deposition of the Cu layer. The printed liquid metal microcapsules therefore serve as the template for the bilayer conductor. The overall process allows the scalable fabrication of delicate conductive features with a resolution of 50 µm ( SI Appendix , Figs. S2 and S3 ). In Fig. 1 C , SEM image reveals the conformal deposition of the copper layer over liquid metal microcapsules to establish the strong interfacial interactions. Fig. 1. Design and preparation of the self-healing conductor. ( A ) Schematic illustration of the ultrasensitive self-healing mechanism. ( B ) Optical images showing a flower-shaped pattern of screen-printed liquid metal microcapsules before ( Left ) and after ( Right ) the electroless deposition process. (Scale bars: 1 cm.) ( C ) SEM images acquired before ( Left ) and after ( Right ) the electroless deposition process. (Scale bars: 5 µm.) ( D ) Normalized resistance of the self-healing conductor under stretch–relaxation cycles to varying tensile strains. ( E ) Optical microscopy images revealing the microstructure evolutions at different tensile strains for a copper film ( Top ), a self-healing conductor ( Middle ), and a liquid metal microcapsule composite ( Bottom ). (Scale bars: 100 µm.) ( F ) Quantitative structural analysis on the fractured Cu layer in the self-healing conductor. The electrically self-healing conductor has compliant mechanical properties ( SI Appendix , Fig. S4 ). In Fig. 1 D , a representative conductor is evaluated by cyclic tensile deformations, with the peak strain progressively increasing from 3 to 50%. The resistance exhibits steady increases upon stretching and almost full recovery after relaxation to demonstrate robust electrical properties. The behavior is in sharp contrast to the completely lost conductivity of regular copper films at a tiny strain of ~3%, as shown in SI Appendix , Fig. S5 . The conductor is immune to electrical failures by spontaneously repairing any damages in the conductive copper layer. The electrically self-healing capability is not the inherent property of the liquid metal microcapsules. As-printed microcapsules require a large activation strain of ~30% to release sufficient liquid metal for an interconnect conductive network ( 32 , 41 ), as exemplified in SI Appendix , Fig. S6 . The bilayer architecture effectively lowers the triggering threshold for electrical restorations. Microstructure characterizations provide insights into the healing mechanism. In Fig. 1 E , the copper film is mechanically brittle and readily fractured at 3% strain. In the electrically self-healing conductor, the damage in the copper layer follows a similar trend likely due to the compliant nature of liquid metal microcapsules. Due to the robust interfacial coupling, the liquid metal microcapsules are ruptured by the stress field of copper film microcracks to achieve the selective release of a large amount of healing agent at the damage site ( Fig. 1 E and SI Appendix , Fig. S7 ). A remarkable observation is the spontaneous filling of liquid metal into the cracks immediately following their formations according to Movie S1 . The unique phenomenon is responsible for the ultrasensitive reparation of minute electrical damages under low tensile strains. The instantaneous completion of the healing process within a few microseconds further enables the real-time restoration of the electrical properties as illustrated in SI Appendix , Fig. S8 . In comparison, as-printed liquid metal microcapsules are randomly ruptured to release the encased conductor upon tensile deformations ( Fig. 1 C and SI Appendix , Fig. S9 ). Conventional liquid metal–based conductors have demanding triggering conditions largely due to the random sintering process as an inefficient healing mechanism. In addition, the fragmentation process of the copper layer upon stretching has been quantitatively analyzed as shown in Fig. 1 F . The initial regime is dominated by the generation of microcracks with the Cu domain size ( 42 ). The crack width shows slow increases with the strain. The embedded microcapsules are ruptured along the crack propagations for the spontaneous release of the liquid metal. As the strain increases above ~25%, the tensile deformation is primarily accommodated by the expanded cracks due to the reduced stress transfer to the Cu layer ( 43 ). The inhomogeneous strain distribution promotes the selective mechanical sintering of additional liquid metal microcapsules to replenish the widened cracks ( 41 , 44 ). The strongly coupled conductive layer with liquid metal microcapsules is essential for the universal healing capability applicable to various structural damages through the self-adaptive release of the healing agent. Attractive Physical Properties. The as-prepared conductor has the characteristic reddish-orange color of copper, as shown in Fig. 2 A , Inset . A gradual color shift to silvery white is observed in response to large tensile deformations of up to 1,200% strain. In Fig. 2 A , the resistance of the electrically self-healing conductor exhibits stable and limited increases upon stretching. Specifically, the normalized resistance is 1.5 at 100% strain, 5.7 at 400% strain, and 38.1 at 1,200% strain. The resistance almost returns to the original value with a minor irreversible change of ~30% after releasing the tensile strain. In Fig. 2 B , the optical microscopy image reveals the continuous expansion of the cracks under the enormous tensile strains. The fractured copper layer is interconnected by soft liquid metal joints to retain electrical stability. In spite of tensile deformations and relaxations, the liquid metal is stabilized by native oxides to avoid any obvious spread ( 45 ), as shown in SI Appendix , Fig. S10 . The conductor is further examined by tensile fatigue tests with different strain levels ( Fig. 2 C and SI Appendix , Fig. S11 ). A relatively stable resistance is preserved during 1,000 stretch–relaxation cycles to 300% strain, as shown in Fig. 2 C , thereby confirming the electromechanical durability for long-term practical applications. In the initial cycles, the minor increase of the resistance at the relaxed state is likely associated with the irreversible elongation of the SIS substrate ( SI Appendix , Fig. S12 ). Apart from tensile deformations, the conductor is sufficiently robust to withstand decent compressive loads of up to 1.5 MPa. The liquid metal is filled into emerging cracks of the Cu layer without any noticeable overspill, as shown in SI Appendix , Fig. S13 . The electrically self-healing capability also applies to a wide range of temperatures from −15 °C to 80 °C ( SI Appendix , Fig. S14 ). In addition, the conductor has excellent environmental stability to allow long-term storage under ambient conditions ( SI Appendix , Fig. S15 ). Interestingly, the high conductivity and stretchability are negligibly affected by the storage in an aqueous environment, as shown in SI Appendix , Fig. S16 . The stable properties of the electrically self-healing conductor may expand the scope of the potential application settings. Fig. 2. Ultrastretchable self-healing conductors. ( A ) Normalized resistance versus tensile strain during loading–unloading to 1,200% strain. ( Inset ) Optical images of the self-healing conductor in the strain range from 0 to 1,200%. (Scale bars: 1 cm.) ( B ) Optical microscopy images of the self-healing conductor under different tensile strains. (Scale bars: 500 μm.) ( C ) Change in resistance over 1,000 stretch–relaxation cycles to 300% strains. ( Inset ) Normalized resistance at 301st to 305th and 701st to 705th cycles. Influences of the Copper Layers. The Cu layer is electrolessly deposited to provide the original conductivity. The Cu mass density increases with the deposition duration, as shown in SI Appendix , Fig. S17 . In Fig. 3 A , the nominal conductivity of the electrically self-healing conductor is positively correlated with the deposition time. Specifically, the conductivity value reaches 6.1 × 10 3 S/cm at 30 min, 1.2 × 10 4 S/cm at 40 min, and 2.5 × 10 4 S/cm at 60 min. The self-healing capability is explored by inducing extensive structural damages under uniaxial tensile deformation of 300% strain. The healed conductor at the relaxed condition exhibits enhanced conductivity for short electroless depositions and lowered conductivity for lengthy depositions ( Fig. 3 B ). A rough compensation is achieved at an intermediate deposition duration of 40 min, generating ~1.5 µm-thick copper layer over 25 µm-thick liquid metal microcapsules ( SI Appendix , Fig. S18 ). The condition is therefore adopted in most experiments with the optimal recovery rate. In Fig. 3 C , the healing capability is also correlated with the level of structural damage. The healed resistance initially drops by increasing the experienced tensile strain as released liquid metal overcompensates for lost conducting pathways. The healed resistance switches to a rising trend for experienced tensile deformations above 400% strain. The increase in the resistance is ascribed to the irreversible elongation of the elastomer substrate after large tensile strains ( SI Appendix , Fig. S19 ). The conductor demonstrates fairly stable resistance within 40% variations for tensile strains of up to 1,200%, representing universal healing capability to withstand large loadings and unexpected impacts in practical applications. The behavior contrasts sharply with liquid metal microcapsule composites exhibiting strain-dependent resistance ( SI Appendix , Fig. S20 ). Additionally, the Cu layer allows an alternative deposition method through thermal evaporation. In SI Appendix , Fig. S21 , a 250-nm-thick Cu layer on liquid metal microcapsules retains excellent self-healing behavior of electrical properties, demonstrating the general applicability of the bilayer architecture. Fig. 3. Influence of the electroless deposition condition on the self-healing conductor. ( A ) Conductivity and sheet resistance of the self-healing conductors based on different deposition durations. ( B ) Conductivity at pristine and healed conditions. The conductors undergo 300% tensile strain to induce extensive structural damage. ( C ) Normalized resistance of the healed conductor after experiencing different levels of tensile deformations. Influences of the Liquid Metal Microcapsule Composites. The liquid metal microcapsules are the healing agents for electrically self-healing conductors. The ultrasonication time is the effective knob to control the dimension of liquid metal microcapsules, as illustrated in SI Appendix , Fig. S22 . Large microcapsules of 11.6 ± 5.1 μm lead to poor printing quality with highly textured surfaces. Liquid metal microcapsules with reduced dimensions are well suited for generating dense, pinhole-free composite layers. In SI Appendix , Fig. S23 , the corresponding conductor based on ultrafine microcapsules of 0.9 ± 0.5 μm cannot fully recover the original electrical properties after tensile deformations, likely associated with the inverse correlation of the rupture pressure on the microcapsule diameter ( 46 ). Accordingly, the optimal choice is liquid metal microcapsules in 2.8 ± 1.8 μm. In addition, SIS binder concentration is also critical for liquid metal microcapsule composites. As illustrated in SI Appendix , Fig. S24 , additional liquid metal microcapsules are exposed on the composite layer by reducing the binder, thereby accelerating the deposition rate of the copper film. The corresponding conductor exhibits improved electromechanical properties by facilitating stress transfer onto liquid metal microcapsules ( 44 ). However, inks with insufficient binders tend to print low-quality features with excessive cracks. A weight ratio 2:50 between SIS binder and liquid metal microcapsules effectively balances printability and healing capability. Electrically Self-Healing Conductors for Flexible Electronics. The electrically self-healing conductor represents an attractive candidate material to increase the life span of flexible electronic devices. To evaluate the mechanical flexibility, the as-prepared conductor on the polyethylene terephthalate (PET) substrate is repetitively actuated between the flat and outward bending states using a motorized translational stage. The resistance versus the number of bending cycles is shown in Fig. 4 A . The conductor retains robust electrical properties over 5,000 cycles with different bending radii from 2 to 5 mm. In spite of some microcracks as shown in Fig. 4 B , the conductor retains the conducting pathways through the on-demand release of liquid metal at the damaged site. The healing capability, therefore, exhibits ultrahigh sensitivity to minor injuries at small bending strains of down to 0.25%. In contrast to conventional mechanisms demanding external cutting or large deformations for activation, the extremely low healing threshold observed here allows the conductor to operate stably under the bending conditions. In the absence of electrical restorations, a copper film exhibits continuously increased resistance during bending fatigue tests ( SI Appendix , Fig. S25 ). Additionally, the compliant conductor maintains stable resistance over 100 folding–unfolding cycles ( Fig. 4 C ). In Fig. 4 D , an optical microscopy image reveals that dense microcracks are concentrated at the crease. The exceptional resilience against mechanical folding is attributed to the electrically self-healing capability to repair the fractured copper conductive layer. As expected, the regular copper film is mechanically brittle and completely fails upon folding manipulation. Fig. 4. Implementation of self-healable conductors in flexible electronic devices. ( A ) Change in resistance of the self-healing conductor over 5,000 bending cycles to different radii. ( B ) Optical microscopy image captured after the cycling with a radius of 2.0 mm. ( C ) Change in resistance over 100 folding cycles. ( D ) Optical microscopy images revealing the microstructure of the crease on the cycled conductor. ( E ) Schematic showing the layer-by-layer architecture of the flexible LED matrix display. ( F ) Optical images of an as-prepared LED matrix to display different alphabets. (Scale bars: 1 cm.) ( G ) Mechanical manipulations of an LED matrix displaying a heart-shaped pattern. A multilayer matrix display is constructed in a layer-wise sequence by the heterogeneous interconnection of LED chips with the electrically self-healing conductor on a PET substrate, as schematically illustrated in Fig. 4 E . In Fig. 4 F , a representative LED display in the form of a 5 × 5 matrix allows the convenient display of variable words and graphic patterns after interfacing with an external controller ( Fig. 4 F and Movie S2 ). The device is sufficiently robust to retain a stable luminous pattern under various manipulations including bending, folding, and crumping, as shown in Fig. 4 G and Movie S3 . The exceptional durability is associated with the self-healing capability of the conductor to recover from accidental electrical damage. Skin-Attachable Electronics Using Electrically Self-Healing Conductors. Stretchable electronics allow conformal and intimate interactions with skins for wearable sensing and simulations. The electrically self-healing conductor is highly conductive, mechanically compliant, and durable, which represents a suitable candidate material for stretchable and wearable devices. As schematically illustrated in Fig. 5 A , a multifunctional electronic patch is composed of sensing electrodes for biopotential recording, stimulation electrodes for electrical therapy, and a heater for thermotherapy. Fig. 5 B shows an as-prepared electronic patch which is a soft and stretchy multilayer laminate in a thin layout. The electronic patch is manually mounted on the forearm using silicone gel adhesive to achieve conformal and intimate interactions with the skin. In Fig. 5 C , the contact impedance of the epidermal electrodes is competitive against commercial Ag/AgCl gel electrodes, which suggests the suitability for electrophysiological recording and simulation. The sensing electrodes readily capture surface electromyogram (EMG) signatures of muscle activations associated with hand open and close gestures, as shown in Fig. 5 D . The high signal-to-noise ratio of 25.4 dB is comparable to commercial Ag/AgCl gel electrodes (26.2 dB). In addition, the electrical stimulator on the soft patch makes use of monophasic square voltage pulses to trigger subconsciously neuromuscular response, which represents an effective approach for pain relief and physical rehabilitation ( 47 , 48 ). In Fig. 5 E , the threshold voltage for perception is lowered by increasing the stimulation frequency due to the reduced contact impedance ( 49 ). In the biopotential sensor, the voltage signal shows a strong correlation with the stimulation pulses to allow simultaneous monitoring of the treatment ( Fig. 5 F ). The large-area electroresistive heater allows controlled delivery of thermal stimulations. The heater exhibits fast temperature responses and durable heating performances ( Fig. 5 G and H ). The limited drifts in temperature upon stretching are ascribed to the negligible changes in the resistance by adopting the serpentine Peano curve design ( Fig. 5 I ). Infrared camera images of the heater reveal uniform surface temperature distributions at different tensile strains ( SI Appendix , Fig. S26 ). The conformal attachment of the electronic patch on the skin allows reliable heat transfer for thermal therapy ( Fig. 5 J ). The combined thermal and electrical simulations are a well-recognized approach for a synergetic therapeutic outcome ( 48 ). Notice that wearable heaters of compact dimensions are also easily prepared using this fabrication strategy ( SI Appendix , Fig. S27 ). In addition, the entire electronic patch is highly durable to withstand repetitive tensile deformations. All sensing and stimulation capability are well preserved after 1,000 stretch cycles to 60% strain, as shown in SI Appendix , Figs. S28–S30 . The reliable operation is ascribed to the self-healing conductor capable of recovering from stress-induced electrical damages. Notably, the soft liquid metal joints leave no obvious residues on the skin, likely due to the passivation by native oxides ( SI Appendix , Fig. S31 ). Fig. 5. Stretchable and wearable electronics based on the self-healing conductor. ( A ) Schematic illustration of the multifunctional electronic patch comprising a biopotential sensor, an electrical stimulator, and an electroresistive heater. ( B ) Optical image of an as-prepared electronic patch. ( C ) Skin contact impedance of a self-healing conductor electrode and a commercial Ag/AgCl gel electrode. ( D ) EMG signals of hand open and close gestures acquired by Ag/AgCl gel electrodes ( Top ) and self-healing conductor electrodes ( Bottom ). ( E ) Threshold voltage for perception as a function of the stimulation frequency. ( F ) EMG waveforms ( Top ) acquired during electrical stimulations with monophasic voltage pulses ( Bottom ). ( G ) Temperature profiles of the electroresistive heater under different applied voltages. ( H ) Temperature responses to on/off voltage cycles with an amplitude of 10 V. ( I ) Temperature and normalized resistance of the heaters under tensile deformations. ( J ) Infrared camera images showing the reliable heating performance of the electronic patch attached to the forearm."
} | 5,777 |
36405773 | PMC9667314 | pmc | 242 | {
"abstract": "Summary The misuse of petroleum-based plastics has resulted in serious environmental pollution and resource wastage. Biodegradable plastics can be used as green substitutes for traditional plastics. Here, we discuss the feasibility and technical bottlenecks in developing microbial cell factories for the production of biodegradable plastics from lignocellulosic wastes. First, we introduce the basic properties of the main biodegradable plastics on the market, including poly(lactic acid), poly(hydroxyalkanoate), and poly(butylene adipate-co-terephthalate). We then demonstrate the feasibility of synthesizing petroleum-based biodegradable plastic monomers from bio-based raw materials and propose strategies to further advance their commercial production through metabolic engineering and synthetic biology. We also analyze the main challenges facing the current development of bio-based biodegradable plastic biosynthesis technology. Finally, we discuss the current major lignocellulose bioconversion processes and explore way to further improve the utilization efficiency of the main carbohydrates in lignocellulosic hydrolysates by microorganisms, from the perspectives of sugar transport, sugar assimilation, and carbon catabolite inhibition.",
"conclusion": "Concluding remarks and future perspectives Conversion of lignocellulosic wastes into biodegradable plastics via microbial cell factories is an attractive biotechnological approach. Here, we briefly review related works in recent years and propose future research suggestions in this field. Currently, the main biodegradable plastics on the market include PHA, PLA, and PBAT. PHA and PLA are recognized as bio-based plastics and have relatively mature bio-based synthesis processes. The next step is to improve the production efficiency of lignocellulosic wastes as substrates. In addition, PHB occupies a mainstream position in the PHA market, but its mechanical properties are not satisfactory. Therefore, it is crucial to develop an industrial production process to obtain PHA with improved mechanical properties. On the other hand, researchers have also recently attempted to develop processes for generating PLA polymers directly from microbial cells, representing a new direction in this field. Through a biosynthetic pathway similar to PHA, PLA can be synthesized in 1 step, avoiding the subsequent polymerization process. However, the yield and molecular weight of PLA obtained through direct biosynthesis are currently low and cannot meet industrial requirements. However, further research is expected to provide a breakthrough in this issue. A commercial bio-based synthesis process for 1,4-butanediol, which is a monomer of the petroleum-based polymer PBAT, has been established using glucose as a substrate. However, there is a lack of an efficient 1,4-butanediol production process from pentose or lignocellulose, which limits the synthesis of PBS from lignocellulose wastes. Therefore, developing processes for the efficient synthesis of 1,4-butanediol using pentose or lignocellulose hydrolysates as substrates has become a focus of future research in this field. Relatively effective biosynthetic pathways have been developed for adipic acid, another monomer of PBAT, but the current titers and yields do not meet the requirements of industrialization. Metabolic analysis of synthetic pathways enables the discovery of key bottlenecks that limit the efficiency of the synthesis of these products, such as insufficient reducing power and carbon loss. Next, these bottlenecks need to be overcome through systems metabolic engineering strategies. Terephthalic acid, the monomer of PBAT, cannot be biosynthesized from carbohydrates using microbial fermentation methods. This may require the development of new synthetic pathways and the expansion of the substrate scope of existing enzymes. The efficient utilization of lignocellulose resources is another major challenge, including developing efficient lignocellulose bioconversion processes and improving the utilization efficiency of carbon sources in lignocellulose hydrolysates by microorganisms. The pretreatment and degradation of lignocellulose is key to limiting its biotransformation. The existing SHF and SSF processes rely on high-temperature and high-pressure chemical pretreatment methods, which incur additional costs and potential environmental and energy problems. With the gradual maturity of the CBP process, the participation of microbial cells makes the hydrolysis of lignocellulose greener and more sustainable, which may be a future development trend. Improving the utilization efficiency of carbon sources by microorganisms is a key factor in improving the entire fermentation process; however, CCR, which represses pentose utilization in the presence of glucose, has become the key limiting factor of the co-utilization of pentoses and hexoses in lignocellulose hydrolysates. In addition, the assimilation pathway and transport efficiency of nonpreferred carbon sources (mainly pentoses) by some microorganisms are low, which is another major challenge for lignocellulose utilization. These issues can be ameliorated by transporter engineering, metabolic engineering, and methods that alter the global regulation of microbial cells, requiring continued efforts by researchers. At present, the biodegradable polymer-production strains and the lignocellulosic hydrolysate-transformation strains are not consistent. More specifically, some microbial characteristics with industrial or metabolic advantages, such as acid resistance, high-temperature resistance, high osmolarity resistance, and high-efficiency synthetic reducing power, are also often distributed in different strains. 124 , 125 From the perspective of industrial applications, integrating these excellent characteristics into the same microorganism can significantly simplify the production process and reduce production costs. The rapid development of synthetic biology in recent years has made it possible to construct such super microbial cell factories with these features and properties in the same chassis in a modular fashion. We believe that, with further technological developments, the use of microbial cell factories to convert lignocellulosic wastes into biodegradable plastics will gradually replace the traditional petroleum-based plastics market, especially in the fields of food packaging and agricultural mulching films.",
"introduction": "Introduction Since its invention in the 19 th century, plastics have been widely used in all aspects of our lives because of their light weight, durability, and anti-corrosion properties. 1 The annual consumption of plastics per capita has been reported to exceed 140 kg. 1 However, although the large-scale use of plastics is convenient for our lives, it also causes serious environmental pollution. The misuse of plastics generates a large amount of plastic wastes, of which less than 10% can be recycled, and most of which is directly discharged into the natural environment. 2 Traditional plastics are stable in the environment for a long time, and the first plastic sample ever made has not degraded yet. 1 This means that the adverse effects of plastics on the environment remain over the long term. However, some plastics are degraded in the natural environment through physical or chemical processes to form microplastics that can enter animals and even human bodies through food chains. 3 A recent study reveals that microplastic components can indeed be detected in human blood. 4 Therefore, there is an urgent need for safe and environmentally friendly novel materials to replace traditional plastic to solve a series of environmental problems. Biodegradable plastics are excellent alternatives to conventional plastics, with material properties similar to those of plastics and the ability to be rapidly degraded by microorganisms into CO 2 , water, or usable compost. 5 However, compared to traditional plastics, the production costs of biodegradable plastics are still relatively high, which limits their further applications. 6 In addition, many biodegradable plastics are petroleum-based ones, which may cause problems such as resource consumption during the production processes. Even the current raw materials for bio-based plastics are mostly derived from food crops, 1 which may occupy valuable agricultural resources. Therefore, the current biodegradable plastics market urgently requires new, low-cost, and sustainable substrates. Waste-derived lignocellulosic biomass is an alternative to traditional petroleum-based feedstocks. The global annual amount of lignocellulosic biomass obtained from wastes such as straws and wood chips can reach 150 billion tonnes. 7 Lignocellulosic biomass wastes, whether they are directly discharged or incinerated, have adverse effects on the environment. Therefore, recycling of lignocellulosic biomass is a green process. From a cost perspective, waste-derived lignocellulosic biomass is more economical than the current major bio-based feedstocks and, therefore, has potential advantages in cost control. This substrate is also renewable and does not require additional resources. Production of biodegradable plastics from lignocellulosic biomass is challenging. On the one hand, lignocellulose has a very stable structure and is difficult to be degraded by green and efficient processes. 8 On the other hand, some petroleum-based biodegradable plastics cannot be efficiently synthesized from lignocellulose hydrolysates. However, with the development of metabolic engineering and synthetic biology techniques, exciting breakthroughs have been made in these areas. In recent years, efficient bio-based synthesis processes for several monomers of petroleum-based biodegradable plastics, such as 1,4-butanediol 9 and adipic acid, 10 have been successively established. In addition, the participation of microorganisms makes the degradation process of lignocellulose green and efficient. 11 Therefore, the use of microbial cell factories to produce biodegradable plastics from lignocellulosic biomass represents a promising green plastic approach. Here, we comprehensively demonstrate the promising prospects of using microbial cell factories to produce degradable plastics from lignocellulose. Through metabolic engineering and a rational design of synthetic biology, microbial cell factories can process lignocellulosic biomass and new green materials, thereby building a circular production system for biodegradable plastics ( Figure 1 ). In this system, energy and carbon are retained to the greatest extent possible, thereby revolutionizing the interaction between humans and nature in the production of plastics. Through a comprehensive review of current bio-based production processes for biodegradable plastics and lignocellulose bioconversion, we profile the essential issues and future development directions in this field. Figure 1 Resource recycling economic scheme based on bio-based biodegradable plastics The microbial cell factory converts agricultural waste-derived lignocellulosic biomass into biodegradable plastics in an environmentally friendly manner. Agricultural wastes are recycled in the form of novel materials. Waste biodegradable plastics are biodegraded into compost to support crops, recycling this carbon back into nature."
} | 2,839 |
31300639 | PMC6626051 | pmc | 243 | {
"abstract": "The global decline of coral reefs heightens the need to understand how corals respond to changing environmental conditions. Corals are metaorganisms, so-called holobionts, and restructuring of the associated bacterial community has been suggested as a means of holobiont adaptation. However, the potential for restructuring of bacterial communities across coral species in different environments has not been systematically investigated. Here we show that bacterial community structure responds in a coral host-specific manner upon cross-transplantation between reef sites with differing levels of anthropogenic impact. The coral Acropora hemprichii harbors a highly flexible microbiome that differs between each level of anthropogenic impact to which the corals had been transplanted. In contrast, the microbiome of the coral Pocillopora verrucosa remains remarkably stable. Interestingly, upon cross-transplantation to unaffected sites, we find that microbiomes become indistinguishable from back-transplanted controls, suggesting the ability of microbiomes to recover. It remains unclear whether differences to associate with bacteria flexibly reflects different holobiont adaptation mechanisms to respond to environmental change.",
"introduction": "Introduction Scleractinian corals live in close association with endosymbiotic dinoflagellates of the family Symbiodiniaceae and a diverse community of bacteria (among other microorganisms), collectively referred to as the microbiome 1 . The community of the coral host and its associated microbiome comprises a metaorganism and is referred to as the coral holobiont 2 . While most corals depend on Symbiodiniaceae to meet their energetic demands by the transfer of photosynthetically fixed carbon 3 , 4 , bacterial microbiome members fulfill a range of other functions including nitrogen fixation, sulfur cycling, and protection against pathogenic bacteria 5 – 9 . Accordingly, restructuring of the microbiome is proposed to contribute to coral holobiont plasticity and adaptation 1 , 10 – 12 . Recent studies have found differences in the degree to which coral microbiomes vary over environmental gradients or experimental treatments 13 – 15 . For example, microbiomes of Ctenactis echinata varied between different reef habitats to the degree that abundance of coral host species was associated with the presence/absence of specific bacteria 16 . Further, microbial diversity was shown to increase with depth in several coral species, possibly allowing corals to access a broader range of food sources 17 . In addition, some studies have found seasonal fluctuations in coral-associated microbiomes 16 , 18 , 19 and tide-related shifts on much shorter time scales 20 , while other corals maintain temporally stable microbiomes 21 . Unidirectional transplantation experiments of the coral species Acropora muricata 22 and Porites cylindrica 23 from a pristine site to impacted or modified sites further illustrate that microbiomes change under adverse environmental conditions. In contrast, a transplantation experiment between different thermal habitats showed that microbiomes of heat tolerant Acropora hyacinthus can be acquired by heat sensitive corals upon environmental transplantation over the course of 17 months 11 . Notably, these corals exhibited increased thermotolerance in a subsequent heat stress experiment, harboring a more robust and stable microbiome. At present, it is unclear whether the potential for microbiome restructuring is a conserved trait across coral species or whether species-specific differences exist. For instance, the coral Pocillopora verrucosa shows a globally conserved association with its main bacterial symbiont Endozoicomonas 14 that remains unchanged even under conditions of bleaching and mortality 24 . P . verrucosa further was shown to maintain a stable Symbiodiniaceae community during a cross-transplantation experiment over depth 25 and between seasons and reefs, while Porites lutea sampled under the same conditions had a highly flexible Symbiodiniaceae community 26 . Therefore, it appears that the ability of corals to associate with distinct microbial associates may depend on location, environmental setting, and coral host species. To assess potential differences in flexibility of bacterial association across coral species, we conducted a long-term large-scale reciprocal transplantation experiment using the coral species Acropora hemprichii and P . verrucosa . Based on previous studies, these coral genera were suspected to differ in the flexibility of their association with different microbial communities across environmental gradients 14 , 24 , 27 , 28 . The use of a reciprocal transplant design between reef sites subjected to different levels of anthropogenic impact allowed us to assess whether environmental differences align with distinct bacterial communities and whether the ability to adapt bacterial community composition differs between coral species. As coral microbiomes have previously been shown to recover from stress events, such as bleaching 29 or disease 30 , we were interested to elucidate whether coral microbiomes can recover from chronic pollution, i.e. return to a state that resembles conspecific microbiomes at unaffected sites upon transplantation of coral fragments from affected to pristine sites. Our study shows that the degree of bacterial community restructuring upon transplantation to reef sites with different levels of anthropogenic impact differs between host species. A . hemprichii harbors a highly flexible bacterial community that is characterized by many differentially abundant bacterial taxa between impacts. In contrast, the bacterial community of P . verrucosa remains remarkably stable between impacts. In addition, microbial communities recovered to their original states upon cross-transplantation to unaffected sites. Thus, distinct degrees of host-associated bacterial community restructuring exist, but their role in holobiont adaptation to environmental change is currently unknown.",
"discussion": "Discussion Changes in microbial community structure represent a potentially fast and flexible mechanism that may facilitate coral holobiont adaptation and broaden plasticity 1 , 31 , 32 . In this study we tested two coral species for their capacity to undergo microbiome restructuring in response to changing environmental conditions during a long-term reciprocal transplantation experiment. We found that the A . hemprichii microbiome is highly flexible and more variable, whereas the P . verrucosa microbiome is fairly stable and overall less variable in response to changing environmental conditions. These findings suggest that coral species exhibit different degrees of flexibility in holobiont structure and composition. A . hemprichii underwent strong microbiome restructuring when transplanted from unimpacted sites to impacted ones. This was characterized by increased bacterial diversity and decreased evenness (i.e., the loss of dominant taxa from the unimpacted site). An increase in bacterial diversity in coral microbiomes often accompanies the holobiont stress response as a result of emerging opportunistic taxa that are otherwise absent or suppressed 33 , 34 . Also, increased bacterial diversity has been repeatedly observed in diseased coral microbiomes 35 , 36 . Notably, the changes in bacterial communities of A . hemprichii at impacted sites were characterized by decreased relative abundances and potential loss of Endozoicomonadaceae, the bacterial family containing the enigmatic coral symbiont genus Endozoicomonas 37 , 38 . Moreover, bacterial communities of A . hemprichii fragments transplanted to impacted sites had higher abundances of bacterial families that have previously been characterized as opportunists and that have been associated with coral disease. For instance, taxa within the Rhodobacteraceae have been found on corals with white plague disease 35 and the family Flavobacteraceae also contains potentially pathogenic taxa 36 . Nonetheless, other bacterial families that were more abundant in A . hemprichii at the impacted sites (Erythrobacteraceae, Comamonadaceae, Oxalobacteraceae, Moraxellaceae) have also been isolated from healthy corals, which illustrates that shifts in microbial abundances do not exclusively or necessarily reflect a pathobiome, but rather an environmentally selected, putatively more beneficial microbiome 39 – 41 . Indeed, the notion of an environmentally explicit set of bacterial taxa that fill specific functional niches through changes in the microbiome of A . hemprichii 17 is supported by the high number of specific taxa found under different impacts in our LEfSe and indicspecies analyses. In contrast, the number of such environmentally explicit taxa in P . verrucosa was an order of magnitude lower (5 and 7 taxa in P . verrucosa vs. 60 and 62 taxa in A . hemprichii for the indicspecies and LEfSe analyses, respectively). This is also in line with the current notion that holobiont composition (or metaorganism structure for that matter) is not static, but rather dependent on age, development, sex, and environment, among other factors 42 . As such, consistent or close association of bacteria with their animal hosts is not a sensu stricto criterion for functional relevance 42 . Furthermore, the microbiome of P . verrucosa was remarkably consistent between sites, with only small changes between impact levels in bacterial diversity and evenness. The most abundant Endozoicomonas OTU in P . verrucosa dominated most samples and, in contrast to the loss of Endozoicomonadaceae in A . hemprichii at impacted sites, was consistent. Generally, there were no major abundance changes in bacterial taxa between impact levels in P . verrucosa , with the exception of the Simkaniaceae family, whose role we can only speculate on. Additional analyses excluding the Endozoicomonadaceae family further support the notion of a more stable and less variable microbiome of P . verrucosa compared to A . hemprichii . Whether dominance of a particular bacterial lineage promotes a generally more stable microbiome (inducing stability of less abundant members), or whether it is simply the indication of an inflexible host-microbial association, remains to be determined. The differences we observe in bacterial assemblages between A . hemprichii and P . verrucosa across different sites argue for different degrees of microbiome flexibility. Thus, these potentially represent differences in the underlying strategy employed by the two species to cope with environmental stress. Importantly, it has to be considered that corals are in general long-lived, sessile animals that are unable to escape changes in their environment. And because of their long generation times, evolutionary change is supposedly slow 43 . Strategies to cope with and survive rapid environmental change are therefore critical. One mechanism by which corals may adjust more rapidly to change may be through their association with different bacterial taxa, whereby selection occurs for the most advantageous and beneficial microbiome in a particular environment 1 , 10 . It is therefore striking to find in A . hemprichii a readily “responding” microbiome, in contrast to which the bacterial community of P . verrucosa seems rather “inert”. However, this finding is not entirely unexpected. Previous studies have shown highly stable Symbiodiniaceae 26 , 44 and bacterial 14 , 24 , 28 communities both in P . verrucosa and also its close relative Pocillopora acuta 45 , even under conditions of mortality. In comparison the microbial communities of Acropora are flexible and seem to align with environmental patterns 13 , 29 , 46 . In line with these studies, we found the bacterial microbiome of A . hemprichii to be highly variable between sites and impacts and also flexible upon reciprocal transplantation. In contrast, P . verrucosa harbored a far less variable bacterial microbiome that showed less flexibility upon transplantation. Consequent to these findings, we argue that coral species differ in their ability to associate flexibly with different bacterial assemblages. In analogy to strategies for coping with osmotic stress 47 , Acropora might be referred to as a “microbiome conformer” (showing microbial adaptation to the surrounding environment), whereas Pocillopora might be referred to as a “microbiome regulator” (showing microbial regulation that maintains a constant microbiome). Hence we propose the term “microbiome flexibility” to describe a coral species’ potential for dynamic microbiome restructuring in the face of environmental change. It remains to be investigated how pervasive these patterns of microbiome flexibility are across and within coral taxa, as illustrated by instances of reversed patterns of relatively high microbiome flexibility in Pocillopora 41 and relative low microbiome flexibility in Acropora 27 (although each of the studies tested only one species separately, rendering cross-species comparisons difficult). It further remains to be determined whether differences in microbiome flexibility represent distinct holobiont adaptation mechanisms to environmental change. High microbiome flexibility presumably supports holobiont adaptation to environmental change and follows a generalist strategy albeit with the risk of losing important associates/functions or acquiring pathogens. Conversely, low microbiome flexibility helps to maintain stable and robust relationships with conserved microbial functions reflecting a more specialized strategy at the expense of a putatively low capacity for microbiome adaptation and potential susceptibility to rapid environmental change. The degree of microbiome flexibility may be linked to life history strategy of the host 48 . A . hemprichii and P . verrucosa both have wide distribution ranges 49 , 50 and inhabit the shallow to mid reef slope 50 . However, the microbiome conformer A . hemprichii has a relatively slow growth rate 51 and, in contrast to Pocilloporids, Acroporids generally have a competitive strategy, with longer generation times 52 . Indeed, the microbiome regulator P . verrucosa shows a rather limited physiological plasticity and ecological niche space, being adapted to high-light wave-exposed environments near the reef crest, in particular in the Red Sea 44 . Pocilloporids generally favor an opportunistic colonization strategy that is characterized by fast growth, high reproduction rates, and relatively rapid generation times 53 . Given that a limited diversity in the diet of deep-sea corals has been linked to reduced diversity within the microbiome 48 , the mainly autotrophic lifestyle and low heterotrophic capacity of P . verrucosa may also partially explain its low microbiome flexibility 48 . However, it remains to be determined whether low microbiome flexibility is a cause or consequence of this strategy. The presence of varying degrees of microbiome flexibility among different coral taxa may also help to resolve the discrepancy regarding the absence of a consistent coral core microbiome. While the concept of a core microbiome may be debatable, especially in the light of the coral probiotic hypothesis 1 , 10 , the expectation is that at least some bacterial associates are intimately and tightly associated with a coral and not expendable 14 , 17 , 54 , 55 . A recent study 54 , after examining two coral species across a wide range of habitats, proposed seven distinct bacterial phylotypes as universal coral core microbiome members. Surprisingly, these core OTUs were nevertheless not ubiquitous, the threshold used to qualify as a core microbiome member being comparably low at ≥30% of samples. Our data suggest that attempts to identify a universal coral core microbiome might result in only a small number of bacterial taxa, if different coral species are compared that display such high or low microbiome flexibility. Potentially relevant here is that coral phylogeny is characterized by a deep phylogenetic split (between “complex” and “robust” clades) that dates back >245 mya 56 , 57 . This led to substantial genomic divergence 58 and consequences for host-microbe pairings that may be shaped by phylosymbiosis 59 . Such a separation might arguably underlie not only differences in microbial association, but also differences in microbiome flexibility, such as observed here. A final point arising from results of cross-transplantation and back-transplantation is that the bacterial communities of both coral species were similar to their local back-transplanted conspecifics after cross-transplantation to unimpacted sites. This finding may hold the promise of microbiome “recovery”. In other words, coral fragments transplanted from impacted to control sites did not, after 21 months, continue to share similarities with fragments of the same colonies that remained at the impacted sites. These findings suggest that stress-induced microbiome alterations may be reverted upon removal of chronic and long-term stressors, similar to the recovery observed after coral bleaching 29 or disease 30 . Following the notion that coral microbiomes contribute to coral health, our results indicate that reducing and removing sources of pollution and sedimentation may result in the reversal of microbiomes. Hence, anthropogenic pollution may not irreversibly disrupt microbiomes in supporting coral health 1 , 60 , 61 . Our results are in line with recent studies, which report that increases in coral disease caused by experimental nutrient enrichment were reversed 6–10 months after termination of the experimental treatment 62 . Taken together, our data create an additional incentive to reduce sources of anthropogenic pollution and sedimentation close to coral reefs, even if the corals on the target reefs already appear stressed and in poor condition. Changes in the microbiome of plant and animal hosts are increasingly being associated with the potential for acclimatization and adaptation in multicellular organisms. In particular, stony corals seem to be strongly reliant on their microbial associates, as highlighted by their obligate endosymbiosis with algal dinoflagellates. However, the potential for differences in microbiome flexibility across coral species had not until now been systematically investigated. Our study supports the notion that distinct degrees of microbiome flexibility exist, potentially reflecting different holobiont adaptation mechanisms to environmental change. Thus, bacterial community structures may respond in a host-specific manner (not “one-size-fits-all”), a situation which would hamper elucidation of a universal core microbiome. Importantly, altered microbial community structures of corals from impacted sites recovered, being indistinguishable from local conspecifics when transplanted back into an unimpacted control environment. This finding holds the promise of microbiome recovery, encouraging the reduction of anthropogenic pollution, even in reef areas where coral assemblages are already degraded."
} | 4,803 |
37862425 | PMC10588945 | pmc | 244 | {
"abstract": "Superhydrophobic (SH) surfaces have progressed rapidly in fundamental research over the past 20 years, but their practical applications lag far behind. In this perspective, we first present the findings of a survey on the current state of SH surfaces including fundamental research, patenting, and commercialization. On the basis of the survey and our experience, this perspective explores the challenges and strategies for commercialization and widespread practical applications of SH surfaces. The comprehensive performances, preparation methods, and application scenarios of SH surfaces are the major constraints. These challenges should be addressed simultaneously, and the actionable strategies are provided. We then highlight the standard test methods of the comprehensive performances including mechanical stability, impalement resistance, and weather resistance. Last, the prospects of SH surfaces in the future are discussed. We anticipate that SH surfaces may be widely commercialized and used in practical applications around the year 2035 through combination of the suggested strategies and input from both academia and industry.",
"introduction": "INTRODUCTION Ollivier first reported superhydrophobicity in 1907 ( 1 ), which has received great attention in academia and industry since ~2000, as Barthlott and Neinhuis revealed in 1997 the self-cleaning mechanism of lotus leaves ( 2 ). There are >27,000 documents including papers, patents, and books in the Web of Science about “superhydrophobic*” as of May 2023 with a fast-growing trend ( Fig. 1A ). Superhydrophobic (SH) surfaces feature high water contact angle (CA ≥ 150°) and low CA hysteresis (≤10°) or sliding angle (SA ≤ 10°). On the basis of the unique wettability of SH surfaces, researchers have proposed numerous potential applications. Meanwhile, it is generally believed in the industry that SH surfaces will have extensive applications in various fields. However, the widespread practical applications of SH surfaces lag considerably behind their vibrant fundamental research. In this perspective, we show challenges and strategies for commercialization and widespread practical applications of SH surfaces after a survey of the current state of SH surfaces including fundamental research, patenting, and commercialization. Fig. 1. Current state of SH surfaces. ( A ) Documents in the Web of Science about “superhydrophobic*” (topic) and some representative studies. ( B ) Patent data collected from CNIPA, https://www.freepatentsonline.com , and Web of Science about “superhydrophobic*” (topic)."
} | 639 |
34202857 | PMC8271930 | pmc | 245 | {
"abstract": "Electrospinning technology, which was previously known as a scientific interdisciplinary research approach, is now ready to move towards a practice-based interdisciplinary approach in a variety of fields, progressively. Electrospun nanofiber-applied products are made directly from a nonwoven fabric-based membranes prepared from polymeric liquids involving the application of sufficiently high voltages during electrospinning. Today, electrospun nanofiber-based materials are of remarkable interest across multiple fields of applications, such as in electronics, sensors, functional garments, sound proofing, filters, wound dressing and scaffolds. This article presents such a review for summarizing the current progress on the manufacturing scalability of electrospun nanofibers and the commercialization of electrospun nanofiber products by dedicated companies globally. Despite the clear potential and limitless possibilities for electrospun nanofiber applications, the uptake of electrospinning by the industry is still limited due to the challenges in the manufacturing and turning of electrospun nanofibers into physical products. The recent developments in the field of electrospinning, such as the prominent nonwoven technology, personal views and the potential path forward for the growth of commercially applied products based on electrospun nanofibers, are also highlighted.",
"conclusion": "6. Conclusions The review presented here collectively highlighted the development of the scaling-up method in electrospinning technologies, current progress on the manufacturing scalability of electrospun nanofibers materials and the commercialization of electrospun nanofiber-based products. We started with a brief opening related to the history of electrospinning, followed by a detailed literature review of its genesis principle and typical apparatus. The discussion narrowed into their revolution as a potent technology for the production of nanoscale dimension materials with diversified compositions, and their properties and applications were revealed. The potential for the commercialization applications of electrospun nanofiber-based materials opened up a new avenue towards economically extending and enlightening the values of electrospinning products. Through electrospinning nanotechnology, ultra-strong, durable and specific function-oriented nanofiber-based fabrics have been now tested for a number of consumer product applications, such as filtration, defense and protection garments, medical dressings, home furnishings, food packaging and, also, in cosmetics. As mentioned, nanotechnology has overcome the limitations of applying conventional methods to impart certain properties to electrospun nanofiber-based materials. Electrospinning techniques have facilitated the fabrication of nonwoven fibers in nano-sized scale dimensions, which now has become possible for highly effective modern performance products. Thus, there is no doubt that, in the next incoming years, nanotechnology through electrospinning will penetrate every area of the functional textiles industry for more innovative product applications. Yet, there are other concerns about the electrospinning technology that need to be addressed by the manufacturer, as well as the researchers. The issues are the hazardous vapors emitting from electrospinning solutions while forming nanofiber webs that need to be recovered or disposed of in an environmentally friendly manner. This will definitely involve additional equipment and cost. Another issue is the raised concern over possible health hazards to humans due to the inhalation of fibers, which needs to be taken into another, future study. From the authors’ point of view, the functionality of electrospun nanofibers has been greatly improved and proven to have abilities in various fields of application. However, the shelf-life issue of electrospun nanofibers fabrics prior to laundering and restoration is certainly limited, and insufficient work has been reported. The fabric should have both appropriate mechanical properties and the durability to withstand the physical and chemical stresses of laundering. The importance of the lifespan of electrospun nanofibers products should be thoroughly evaluated before actual industrial production begins. Finally, this review article offered the readers with interesting perspectives regarding applied products of electrospun nanofibers. We strongly believe that all the difficulties and constraints faced in the manufacturing, research and development, as well as processing, of electrospun nanofibers will be clearly resolved throughout advanced research exploration in the coming future.",
"introduction": "1. Introduction Nanotechnology is a fast-growing interdisciplinary technology that is concerned with the novel changes and drastic improvements in the properties of materials conjugated to their nanosized structures [ 1 ]. The production of electrospun nanofibers has contributed to a new generation of nonwoven fabric-based materials for useful applications in multidisciplinary research areas [ 2 , 3 , 4 , 5 ]. The emergence of driven research and the development of electrospinning technology and its electrospun nanofibers is denoted by the increasing publication of related electrospinning research works. Figure 1 presents a survey of the total publications with search topics of electrospun nanofibers between 2010 and June 2021. Quantitative data in the literary study surveys were supported from the Web of Science online search system. Electrospun nanofibers are derived from electrospinning technology, with exceptional characteristics such as a high surface-to-volume ratio, interconnected ultrafine fibrous structure, high tortuosity, high permeability and lightweight materials. Electrospun nanofibers with diameters as low as 100 nm have continuously attracted broad attention from worldwide researchers, who are ready to move towards a practice-based interdisciplinary approach in a variety of fields [ 6 , 7 , 8 , 9 , 10 ]. This path of inventions and discoveries of electrospinning technology was started in the early of 1930s by a German inventor, Anton Formhals, with the big idea of processing artificial silk-like fibers. At the beginning of the 20th century, silk was very popular as an expensive material and high-end textile. It was thus an interesting search to find a replacement material that was inexpensive, and Formhals was actually the beginning to finding a way to generate threads from a dissolved solid in an electrical field. His series of 22 patents, associated with the progress of the electrospinning process, however, could not compete against the commercial large-scale fiber-spinning techniques in the past decades; thus, this delayed their further development for years [ 11 ]. Though the technique of electrospinning is quite old, it has undergone rapid development in recent years after the breakthrough of the exceptional nanostructural characteristic of nanofibers produced from electrospinning in the late 1960s [ 12 , 13 ]. At present, electrospinning is widely used to produce one-dimensional (1D) nanostructures of polymer fibers and is known as the prominent processing technique to generate an interconnected fibrous web from solutions of different polymers and polymer blends that relies on a high-voltage environment. Electrospinning is a versatile technique and is evidenced by the accessibility to tailor electrospun fiber production into a rich variety of fiber morphologies. These electrospun nanofibers can be designed into various morphological structures, such as core shell hollow, porous, deposited in the aligned or a random orientation of nanofiber mesh, using a single or blended copolymer, with the incorporation of additive materials, subjectively [ 14 ]. The versatility of electrospinning technology has accelerated the progressive development of a wide range of nanofiber-based products, with tailored compositions, dimensions and morphologies to carry out various functionalities [ 15 , 16 , 17 ]. Figure 2 shows a schematic representation of an electrospinning setup. A standard laboratory electrospinning machine prerequisite was equipped with a high-voltage source, precision syringe pump and conductive collector, separated at an appropriate distance. The power supply was attached to each charged spinneret and, also, the grounded collector. Spinning solutions filled in the syringe were loaded onto the metal spinneret and subjected to high-voltage charges. The induced electrical forces to surpass the surface tension of the charged liquid followed by the transformation of the pendant drop shape of the liquid meniscus into a “Taylor cone” shape. The whipping and bending instability upon charging the liquid produces a jet of ultrafine fibers before reaching the collector as electrospun fibers [ 18 , 19 , 20 , 21 ]. Upon the electrospinning process, the deposited nanofiber mesh is usually collected as sheets of 10–30-mm-thickness, on top of a conductive substrate. The sheet of nanofibers mesh usually needs to be removed from the conductive substrate and be practicable to be applied in their next applications. The growing interest in electrospinning technology has paved the way to their advanced development by offering the controllable inclusion of the nanoparticles into these nanosized fibers through the blending of the nanoparticles-polymer solution prior to the spinning process. The simplicity, high efficiency, low cost and high reproducibility of nanofibers have categorized electrospinning as the potent fabrication technique of fibrous structured materials in nanometer sized [ 22 , 23 ]. In line with prior published studies, the exploration for practical applications of nanofibers primarily focused on filtration, functional textile, biomedical and electronic devices [ 11 , 24 , 25 , 26 ]. Researchers have published a number of review articles on electrospinning parameters, characterization, and electrospun nanofiber applications [ 14 , 18 , 22 , 25 , 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 ]. However, these past research has not extensively addressed the commercialization aspect of the fibers and their commercial nanofiber-based product. Hence, this article aims to present an overview for summarizing the current progress on the manufacturing scalability of electrospun nanofibers and the commercialization of electrospun nanofibers products by the dedicated companies globally. The scope of this review will cover the recent progress on the nanofiber fabrics manufacturing by electrospinning, the challenge faced in quality control and manufacturing scalability, the latest finding on nanofiber-based products and an outlook on what the future may hold for electrospun nanofibers as prospect resources are well highlighted in this review article."
} | 2,709 |
30923568 | PMC6423811 | pmc | 246 | {
"abstract": "Background Biological routes for utilizing both carbohydrates and lignin are important to reach the ultimate goal of bioconversion of full carbon in biomass into biofuels and biochemicals. Recent biotechnology advances have shown promises toward facilitating biological transformation of lignin into lipids. Natural and engineered Rhodococcus strains (e.g., R. opacus PD630 , R. jostii RHA1, and R. jostii RHA1 VanA − ) have been demonstrated to utilize lignin for lipid production, and co-culture of them can promote lipid production from lignin. Results In this study, a co-fermentation module of natural and engineered Rhodococcus strains with significant improved lignin degradation and/or lipid biosynthesis capacities was established, which enabled simultaneous conversion of glucose, lignin, and its derivatives into lipids. Although Rhodococci sp. showed preference to glucose over lignin, nearly half of the lignin was quickly depolymerized to monomers by these strains for cell growth and lipid synthesis after glucose was nearly consumed up. Profiles of metabolites produced by Rhodococcus strains growing on different carbon sources (e.g., glucose, alkali lignin, and dilute acid flowthrough-pretreated poplar wood slurry) confirmed lignin conversion during co-fermentation, and indicated novel metabolic capacities and unexplored metabolic pathways in these organisms. Proteome profiles suggested that lignin depolymerization by Rhodococci sp. involved multiple peroxidases with accessory oxidases. Besides the β-ketoadipate pathway, the phenylacetic acid (PAA) pathway was another potential route for the in vivo ring cleavage activity. In addition, deficiency of reducing power and cellular oxidative stress probably led to lower lipid production using lignin as the sole carbon source than that using glucose. Conclusions This work demonstrated a potential strategy for efficient bioconversion of both lignin and glucose into lipids by co-culture of multiple natural and engineered Rhodococcus strains. In addition, the involvement of PAA pathway in lignin degradation can help to further improve lignin utilization, and the combinatory proteomics and bioinformatics strategies used in this study can also be applied into other systems to reveal the metabolic and regulatory pathways for balanced cellular metabolism and to select genetic targets for efficient conversion of both lignin and carbohydrates into biofuels. Electronic supplementary material The online version of this article (10.1186/s13068-019-1395-x) contains supplementary material, which is available to authorized users.",
"conclusion": "Conclusions The development of effective lignin valorization is a key solution to improve the carbon efficiency of the entire process of biorefinery. In this study, the co-fermentation of lignin and carbohydrates by co-culture of wild-type and engineered Rhodococci for lipid production was investigated. Carbon utilization preference of glucose over lignin was observed. Lignin model compounds (vanillin, vanillic acid, and benzoic acid) were rapidly consumed as well, suggesting the bacterial lignin conversion may be limited by the comparably weak extracellular depolymerization activity to produce reactive lignin molecules as viable carbon sources for cell growth or energy storage. The low lipid production and up-regulation of β-oxidation of fatty acid degradation using lignin as the sole carbon source also showed the sign of lack of reducing power and energy. Nevertheless, 40.1% low-molecular weight lignin derived from dilute acid pretreated poplar wood was degraded, which could be supported by the extracellular peroxidases of Rhodococci with assistance of oxidases under strong oxidative conditions. However, the oxidative environment generated can be an alternative explanation of fatty acid degradation due to the competition for NADPH pool caused by the need of oxidative stress response system. It also suggested a combination of enzymatic and chemical reactions of lignin depolymerization. A strong in vivo aromatic degradation network provided a variety of paths, including both the β-ketoadipate pathway and the phenylacetic acid pathway to overcome the heterogeneity of lignin-derived aromatics. Collaboration of strains and synergistic pathways of sugar and aromatic metabolism were proposed, suggesting a promising basis for the design of the synergistic platform for the lignocellulosic biofuel production.",
"discussion": "Discussion Besides numerous studies on carbohydrate-based biofuel production, lignin valorization has attracted increasing attentions to obtain high carbon conversion of the entire process. Previous studies investigated various aromatic degradation pathways of possible lignin-derived monomers in wild-type or engineered strains to produce lipids, PHAs, or biochemicals from lignin model compounds or mixed biomass-derived lignin substrates [ 14 , 16 , 58 – 60 ]. However, the production capability from lignin of strains is not comparable with that of carbohydrates, and the knowledge on converting lignin into valuable products is limited. Herein, we studied the case of co-fermentation of lignin and carbohydrates derived from real lignocellulosic biomass by mixed wild-type and engineered Rhodococci for lipid production. Rhodococcus strains are known for promising potential of bioremediation and lignin valorization due to their great catabolic capabilities of utilizing broad range of compounds [ 32 ]. Indeed, the tested lignin model compounds were rapidly utilized simultaneously with glucose, despite of the lower consumption rate. During the fermentation using dilute acid-pretreated biomass slurry supplemented with alkali lignin, the concentration of lignin decreased by almost half within 48 h (Fig. 2 d), which is comparably more efficient than those of previous studies [ 3 , 15 , 24 ] in terms of higher depolymerization rate or shorter fermentation time. We proposed that mixed carbon source of glucose and lignin as well as co-culture of three Rhodococcus strains contributed to this result. The metabolism of glucose in R. jostii RHA1 and R. opacus PD630 goes through the glycolysis pathway, ED pathway, and PP pathway (Additional file 2 : Table S1) [ 45 , 61 ]. The supplementation of glucose to lignin may compensate the energy (in the form of ATP and NADH) for aromatic degradation pathways through the glycolysis pathway, while the ED pathway and the PP pathway generate NADPH for fatty acids synthesis [ 45 , 62 – 64 ]. Meanwhile, compared with single culture of each strain or co-culture of R. opacus PD630 and R. jostii RHA1 vanA − , the co-culture of three strains in this study showed the highest activity of lignin conversion (Fig. 2 a), which may be due to the synergy among R. opacus PD630, R. jostii RHA1, and R. jostii RHA1 vanA − . Although the in vivo aromatic degradation pathways of all three strains were similar (Figs. 4 , 5 , Additional file 1 : Figures S4, S5), the extracellular depolymerization of lignin was different. Despite the non-comparable cell density, the catalase–peroxidase of R. jostii RHA1 was up-regulated around twofold which was higher than the fold-change of that of R. opacus PD630 (Additional file 2 : Table S1). Also, many of the possible lignin-degrading related enzymes we detected were originated from R. jostii (Additional file 3 : Tables S1, S2). These observations may suggest that during co-fermentation, R. opacus PD630 had lower extracellular activity during lignin fermentation compared with R. jostii RHA1 and R. jostii RHA1 vanA − . Thus, the co-culture of R. opacus PD630 with other two strains may help R. opacus PD630 get access to lignin-derived products as carbon sources for cell growth and lipid production as discussed in our previous work [ 3 ]. However, more detailed work such as secretomic or genetics study is needed to confirm the synergic effect of co-culturing these strains. Fig. 4 Proposed pathways of lignin degradation in Rhodococci . The genes followed by E.C. number of encoded enzymes of the key pathways were provided in red or blue color referring to the genes of R. opacus PD630 or R. jostii RHA1 (or R. jostii RHA1 vanA − ), respectively. The numbers in the boxes, matched with the color of the boxes, were the level of protein abundance in four different samples: lysate samples of glucose fermentation (GL), secretome samples of glucose fermentation (GS), lysate samples of lignin fermentation (LL), and secretome samples of lignin fermentation (LS) \n Fig. 5 Proposed pathways of fatty acid metabolism in Rhodococci. The same layout was applied as in Fig. 4 To our knowledge, the peroxidase DypB is the only characterized lignin-degrading enzyme in R. jostii RHA1. An encapsulin protein was reported to be encoded by a gene located immediately downstream of dypB gene and packed DypB by binding its terminal targeting peptide. However, the location of DypB–encapsulin complex remains unresolved. The encapsulin nanocompartment has been observed only intracellularly in bacteria with some exceptions: the encapsulin-related linocins from B. linens and M. tuberculosis were detected extracellularly [ 53 – 55 ]. Our observation suggested the secretion of DypB–encapsulin complex by R. jostii RHA1, which needs further characterization. In previous studies, R. jostii RHA1 showed consistent activity of lignin degradation in the absence of hydrogen peroxide, suggesting that R. jostii RHA1 may possess the ability of H 2 O 2 generation similar to the fungal system to assist the activity of peroxidases or extracellular ligninases like laccases which uses O 2 for oxidation [ 13 , 65 – 67 ]. Herein, we demonstrated that sarcosine oxidase and the putrescine oxidase were up-regulated in vivo during lignin co-culture fermentation. However, the diffusion of H 2 O 2 across cell membrane by bacteria can be limited [ 68 , 69 ]. Our secretome results confirmed that they were secreted by Rhodococci as accessory enzymes for peroxidase activity. Besides these two oxidases, many other oxidases were identified in the secretome as well for H 2 O 2 generation (Additional file 4 : Table S3). The identification of glutathione peroxidase, catalase–peroxidase, and superoxide dismutase which act as the first line of defense also proved the in vitro presence of H 2 O 2 or other reactive oxygen species resulting in oxidative stress [ 51 , 70 ], while further quantification and characterization are needed to determine their involvement in extracellular lignin depolymerization. Notably, both catalase–peroxidase and superoxide dismutase were reported to modify phenolic lignin model compound or organosolv and kraft lignin, respectively [ 49 , 56 , 57 ], which suggested the potential synergy of various enzymes for lignin oxidation activity. On the other hand, H 2 O 2 and reactive oxidative/aromatic radicals (e.g., hydroxyl radical, veratryl alcohol cation radical) were proposed as strong oxidizers to initiate the attack on lignin through non-enzymatic reactions [ 65 , 71 , 72 ]. Further investigation is needed on the correlation among concentration of H 2 O 2 and reactive oxygen species and lignin depolymerization. Overall, our results suggested that Rhodococci mobilized a multi-peroxidases system for lignin depolymerization with the assistance of oxidases under strong oxidative condition. This also implied a synergistic system of both enzymatic and chemical reactions, which shared common with Pseudomonas putida A514 possessing a dye peroxidase-based enzymatic system along with a redox-cycling reaction for hydroxyl radical generation proposed by Lin et al. [ 73 ]. Nevertheless, the existence of other types of ligninases cannot be denied. Low laccase activity of R. jostii RHA1 was detected by Salvachúa et al. [ 24 ]. Herein, we identified a multicopper oxidase with three cupredoxin domains in the secretome sample (Additional file 4 : Table S2), which was predicted to be a bacterial endospore coat component CotA belonging to the laccase-like multicopper oxidase family based on the conserved domain hits within NCBI database. It may share similarity with the best-studied bacterial laccase CotA of Bacillus subtilis participating in the pigmentation of spores and providing protection of spore from UV light and H 2 O 2 [ 74 ], which need further detailed identification and characterization. Lignin-derived aromatic single-ring compounds such as vanillin, p -coumaric acid, ferulic acid, 4-hydroxybenzoate, etc. which yielded protocatechuate and entered the β-ketoadipate pathway were proposed and discussed in previous studies [ 3 , 58 , 64 ]. In our work, the catechol branch of the β-ketoadipate pathway was identified, suggesting other possible intermediates produced from lignin. It may be benzoate or its analog as the benzoate transporter (RHA1_RS14235) was up-regulated 4.5-fold ( p = 0.028, FDR = 0.114) [ 75 ], which was identified by GC–MS as well (Additional file 1 : Figure S1). Besides, 3-phenylpropionate, trans-cinnamate or their analogs can be lignin degradation products as well, and degraded by catalyzing the substituted hydroxyl groups in alternative peripheral pathway [ 76 ] followed by extradiol cleavage, which was in a good agreement with the up-regulation of 2-keto-4-pentenoate hydratase expressed by a mph D gene. Though the PAA pathway was not functionally characterized [ 33 ], it should play an important role in the in vivo catabolism of lignin-derived aromatics as a central aromatic degradation pathway in Rhodococci (Figs. 3 a, 4 ) [ 35 ]. Furthermore, it was predicted that Rhodococci possessed the homogentisate pathway as characterized in P. putida [ 33 , 77 ], and HmgB encoding a fumarylacetoacetate hydrolase within this pathway was up-regulated 1.4-fold during lignin fermentation. Despite it did not pass our test as a significant changed one, it still suggested another possible involved central aromatic degradation pathway since the central intermediate homogentisate can be provided by the activity of 4-hydroxyphenylpyruvate dioxygenase, which was up-regulated greatly in our results (Fig. 3 a). Generally, it suggested a sophisticated and multichannel aromatics catabolism network to support the conversion of complex and heterogenous components derived from lignin, which further implied that the bacterial extracellular lignin-degrading activity may not be efficient enough to provide available and adequate aromatics taken in by cells to compete with glucose utilization. Nevertheless, glucose was apparently the preferred carbon source of Rhodococci , which was consumed much faster than lignin (Fig. 2 d). Also, the presence of glucose resulted in higher TAG accumulation during the co-fermentation than that using lignin as the sole carbon source. R. opacus PD630 was reportedly significant capable of lipid production from glucose [ 78 ]. However, using lignin or lignin model compounds, less TAG was obtained by Rhodococci ; the titer was small as well [ 1 , 14 ]. To promote the efficiency of lignin conversion into valuable products, many attempts have been made, such as modification of lignin structure before fermentation, genetical engineering of promising strains, novel fungal–bacterial or enzymatic–bacterial systems, etc. [ 1 , 15 , 21 , 73 ]. In this study, the low lipid production from lignin may result from insufficient supply of reducing power and energy due to ineffective bacterial lignin catabolism system, and severe competing for NADPH between TAG accumulation and oxidative stress response. We demonstrated that the co-culture of Rhodococcus strains and co-fermentation of lignin with nutrient-rich substrates like glucose can be a new strategy for lignin valorization. Considering the cost of glucose is high, using lignocelluloses-pretreated hydrolysates is an alternative substrate containing both carbohydrates and lignin to produce lipid or other value-added commodity products. However, it requires more detailed studies to characterize the effect of its complex components on cell growth and carbon flux capacity to target products, followed by optimization of cultures and conditions to make a viable process."
} | 4,075 |
39898965 | PMC11827105 | pmc | 247 | {
"abstract": "Two-dimensional-material-based\nmemristor arrays hold\npromise for\ndata-centric applications such as artificial intelligence and big\ndata. However, accessing individual memristor cells and effectively\ncontrolling sneak current paths remain challenging. Here, we propose\na van der Waals engineering approach to create one-transistor-one-memristor\n(1T1M) cells by assembling the emerging two-dimensional ferroelectric\nCuCrP 2 S 6 with MoS 2 and h -BN. The memory cell exhibits high resistance tunability (10 6 ), low sneak current (120 fA), and low static power (12 fW).\nA neuromorphic array with greatly reduced crosstalk is experimentally\ndemonstrated. The nonvolatile resistance switching is driven by electric-field-induced\nferroelectric polarization reversal. This van der Waals engineering\napproach offers a universal solution for creating compact and energy-efficient\n2D in-memory computation systems for next-generation artificial neural\nnetworks."
} | 238 |
34012760 | PMC8111432 | pmc | 248 | {
"abstract": "Wearable triboelectric nanogenerators (TENGs) have recently attracted great interest because they can convert human biomechanical energy into sustainable electricity. However, there is a need for improvement regarding the output performance and the complex fabrication of TENG devices. Here, a triboelectric nanogenerator in single-electrode mode is fabricated by a simple strategy, which involves a sandwich structure of silicone rubber and silver-coated glass microspheres (S-TENG). The S-TENG exhibits a remarkable performance in harvesting human motion energy and as flexible tactile sensor. By optimizing the device parameters and operating conditions, the maximum open-circuit voltage and short-circuit current of the S-TENG can reach up to 370 V and 9.5 μA, respectively. The S-TENG with good stretchability (300%) can be produced in different shapes and placed on various parts of the body to harvest mechanical energy for charging capacitors and powering LED lights or scientific calculators. In addition, the good robustness of the S-TENG satisfies the needs of reliability for flexible tactile sensors in realizing human–machine interfaces. This work expands the potential application of S-TENGs from wearable electronics and smart sensing systems to real-time robotics control and virtual reality/augmented reality interactions.",
"conclusion": "Conclusion In summary, an easily manufactured, inexpensive and stretchable single-electrode mode S-TENG was designed and fabricated, of which the electrode was made of a conductive fabric. The top and bottom layers of the sandwich structure are silicone rubber and the middle layer are SCGMs both as conducting layer and frictional layer. The peak values of V OC and I SC of the S-TENG are nearly 200 V and 4.2 μA at a frequency of 2.5 Hz. Moreover, the device has good long-term stability with almost no degradation of the electrical output after 3000 cycles. In addition, the S-TENG can light up 235 LEDs and power a commercial calculator by padding the S-TENG with the hand. The S-TENG device can be made into various shapes and sizes, and can be freely placed on different parts of body to harvest human motion energy. It has been proven that the device is suitable for wearable energy harvesting. A large-scale device could be used to power portable electronic devices.",
"introduction": "Introduction Traditional batteries cannot provide a durable and reliable power supply for small portable electronic devices, personalized healthcare, and Internet-of-Things (IoT) devices [ 1 – 6 ]. Thanks to the progress in low-power technology, the power consumption of microelectronic devices has dropped to the level of micro- or nanowatts, which makes the use of environmentally friendly energy a good and practical strategy. Multiple sources of energy could be used, such as wind energy [ 7 ], solar energy [ 8 ], thermal energy [ 9 ], electromagnetic energy [ 10 ], and mechanical energy [ 11 ], among which mechanical energy is created almost everywhere. Mechanical energy has many obvious advantages over other energy forms, such as high energy density, wide distribution, and simple acquisition. Regarding this, it is desirable to develop wearable devices that convert mechanical energy from human body motion into electricity [ 12 ]. Triboelectric nanogenerators (TENGs), with a wide range of material choices and simple device structures, capture the energy of human motion in real time [ 13 ]. This form of energy conversion can not only provide sustainable power for electronic systems, but also provide reliable solutions for active sensing and human–computer interfaces [ 14 ]. A stretchable TENG with double-helix structure was previously designed. It consisted of silver-coated glass microspheres (SCGMs) and silicone rubber as stretchable conductive thread (SCT) and a silicone rubber-coated SCT as the other triboelectric thread [ 15 ]. This TENG can convert the biomechanical energy from human joint motions. The elastomer matrix guarantees that the TENG can be applied in stretchable electronic systems. The TENG generates an open-circuit voltage of 3.82 V and a short-circuit current of 65.8 nA. There are two more references focused on stretchable TENGs utilizing SCGMs to harvest biomechanical energy [ 16 – 17 ]. Zhang et al. invented a closed-structure TENG made of stretchable materials for harvesting human motion energy and monitoring [ 17 ]. It can produce an open-circuit voltage up to 150 V and an optimal instantaneous power density of 44.6 mW/m 2 . From the abovementioned references [ 15 – 16 ], it can be seen that the output performance remains to be improved despite the complex fabrication procedure of the TENG devices. More recently, Qian et al. proposed a nylon-regulated TENG in contact-separated working mode, whose open-circuit voltage and short-circuit current can reach up to 1.17 kV and 138 µA, respectively [ 16 ]. Although this is a tremendous advancement in output performance, the contact-separated working mode with double electrodes makes it difficult to connect the current conducting wires to moving objects when trying to harvest mechanical energy. Besides, the organic–inorganic composites prepared by embedding relatively hard SCGMs into a soft silicone rubber matrix reduce the stretchability of the TENG devices, which is adverse regarding wearables. Therefore, further investigations to enhance the stretchability of TENGs by innovative design are still required. In this work, we developed a single-electrode mode, stretchable triboelectric nanogenerator (S-TENG) using a simple strategy. The single-electrode mode enables the TENG to scavenge energy from the irregular mechanical motion of a free object, independent of electrode position and shape [ 18 – 19 ]. The S-TENG with sandwich structure consists of silicone rubber and SCGMs. The latter are either used as electrode or triboelectric layer. The S-TENG can charge commercial capacitors and power LED lights and scientific calculators and has three distinct advantages. These advantages are: (a) It can be made into multiple shapes and placed on various parts of the body to harvest mechanical energy, requires only a simple fabrication process, (b) it shows stable output and long working life, which provides sustainable electricity, and (c) due to the unique structural design of the device and the high elasticity of the silicone rubber, the S-TENG can be stretched easily to 300% to realize a conformal assembly in stretchable electronic systems. The distinct advantages of the S-TENG indicate broad application prospects in wearable electronics and smart sensing systems.",
"discussion": "Results and Discussion Figure 1a is the schematic of the structural design of the S-TENG. The device is composed of three layers, that is, the top layer and the bottom layer of silicone rubber and the middle layer of SCGMs. The scanning electron microscopy (SEM) image ( Figure 1b ) and the energy dispersive X-ray spectroscopy (EDS) measurement ( Figure 1c ) show that the SCGMs are evenly dispersed across the silicone rubber. Figure 1d presents the flow chart for the fabrication of the S-TENG. Silicone rubber and SCGMs are filled in a 3D-printed mold and cured. After that, the device is removed from the mold. Figure 1 (a) The structural design of the single-electrode mode TENG. (b) SEM and (c) EDS measurements of the SCGM surface. (d) Flow chart for the fabrication process of the S-TENG. (e) Photographs of the S-TENG (original, stretched, twisted and rolled). The S-TENG can be cut into different shapes for possible applications. The device can be stretched to 300% of the initial length (Video 1, Supporting Information File 1 ). Also, it can be rolled and twisted easily, as shown in Figure 1e . Figure 2a illustrates the electricity generation mechanism of the single-electrode mode S-TENG, which is based on a conjunction of contact triboelectrification and electrostatic induction [ 2 , 20 – 23 ]. In the initial state ( Figure 2a -I), the frictional layer and the S-TENG are in balance without potential difference. When the frictional layer contacts the silicone rubber ( Figure 2a -II), the positive charges in the frictional layer are equal to the negative charges in the silicone rubber. Once the surfaces of frictional layer and silicon rubber are separating ( Figure 2a -III), the negative charges on the silicone rubber surface drive the electrons of the SCGMs to flow to ground, generating a reverse triboelectric potential. When frictional layer and silicone rubber are entirely separated, the electrostatic equilibrium between the SCGMs and ground is re-established, no output signal can be observed ( Figure 2a -IV). When the frictional layer is close to the silicone rubber, the electrons will transfer from ground to the SCGMs and generate a positive potential through the triboelectric effect. Finally, the charge distributions of the two surfaces return to the initial stage. Figure 2 (a) Schematic of the working principle of the S-TENG. (b-I) Open-circuit voltage and (b-II) short-circuit current of the S-TENG at different motion frequencies and an applied force of 90 N. Magnified curves of (c-I) open-circuit voltage and (c-II) short-circuit current when tapping at a motion frequency of 2.5 Hz. In order to measure the electrical output of the S-TENG, a piece of 30 × 40 mm 2 was stuck onto an acrylic plate that was fixed on a linear motor. With the linear motor, displacement and motion frequency of the other triboelectric layer relative to the silicon rubber can be controlled. The open-circuit voltage ( V OC ) peaks remain unchanged when the frequency varies from 1 to 2.5 Hz ( Figure 2b -I). The short-circuit current ( I SC ) increases from 1 to 4.2 μA when the frequency goes from 1 to 2.5 Hz ( Figure 2b -II). The peak values of V OC and I SC go up to nearly 200 V and 4.2 μA, respectively, at 2.5 Hz, as shown in Figure 2c -I and 2c-II. Based on Maxwell’s displacement current, with the increasing number of contact/separation cycles during a unit of time, the charge movement rate between electrode and ground is increasing. Therefore, the I SC of the S-TENG can be increased at high frequencies. The size plays a crucial role regarding the electrical output of the S-TENG. The output of the S-TENG was studied while changing the size from 10 × 10 mm 2 to 80 × 80 mm 2 . As shown in Figure 3a and Figure 3b , under a force of 50 N, both V OC and I SC increase when the contact area increases. When the contact area is 80 × 80 mm 2 , V OC and I SC reach up to about 370 V and 9.5 μA, respectively. The S-TENG can be used as a large wearable device. With bigger contact area, more charges and, consequently, higher I SC values are generated. V OC and I SC both increase proportionally to the contact area from 10 × 10 mm 2 to 50 × 50 mm 2 . However, V OC and I SC of 80 × 80 mm 2 are not proportionally increased. This may be attributed to the fact that the large device collapses easily in a non-uniform manner. Specifically, the upper and lower surfaces may not contact or separate efficiently, which causes the actual contact area to be smaller than expected [ 17 ]. Figure 3 Open-circuit voltage and short-circuit current of the S-TENG (a, b) as functions of the contact area, (c, d) as function of the ratio between silicone rubber and silver-coated glass microspheres, and (e, f) as function of the applied force. Different devices were prepared by adjusting the ratio between silicone rubber and SCGMs (1:1, 1:1.5, 1:2, 1:2.5). As can be seen from Figure 3c and Figure 3d , when the mass ratio between silicone rubber and SCGMs is 1:1.5, V OC and I SC reach the largest values of 250 V and 6 μA, respectively, under a force of 150 N. As the content of SCGMs continues to increase, V OC and I SC gradually decrease. The larger amount of SCGMs causes less air in the same volume. Hence, there is less friction between silicone rubber and SCGMs. The applied force is another factor that impacts V OC and I SC . As shown in Figure 3e and Figure 3f , when the applied force is increased from 20 to 100 N, the V OC values increase from 85 to 225 V and the I SC values rise from 1 to 5.3 μA. The reason is that the stronger compressive force leads to an intensification of friction and the generation of more charges. A similar trend is observed for V OC , as expected. Because of the mismatch between AC and DC systems a full-wave rectifier circuit was introduced to the setup. Figure 4a shows the voltage of different capacitors (2.2, 4.7, 10, and 33 μF) as function of the charging time with the rectified S-TENG output. The voltage reaches a saturation value of 14 V for charging a 2.2 μF capacitor. Figure 4 (a) Voltage of different capacitors (2.2, 4.7, 10, and 33 μF) as function of the charging time with the rectified S-TENG output. The inset shows the circuit diagram of the charging system. (b) Cyclic stability of the S-TENG for about 3000 cycles. The inset shows enlarged vies of the middle and the last five cycles. (c) Voltage/current trends and (d) output power of the S-TENG under different load resistances. The maximum power is 308 μW under an external load resistance of 2 MΩ. (e) Photograph of the custom-made resistance test platform. (f) Sheet resistance under tensile strain from 0 to 300%. Reliability is a key parameter considering the practical application of the S-TENG. As depicted in Figure 4b , the I SC values of the device do not decline after 3000 cycles for 25 min, demonstrating the long-term stable operation of the S-TENG on the human body. The impedance matching experiment was designed to measure the powering capability of the S-TENG with different resistances. Figure 4c ,d shows the relationship between load resistance and I SC , V OC , and power of the S-TENG (30 × 40 mm 2 ). With the resistance in parallel changing from 1000 Ω to 1 GΩ, V OC increases from 0.6 to 147 V and I SC decreases from 5.6 to 0.45 μA. The output power (blue line) is calculated as P = U · I . The instantaneous power can achieve a peak value of 308 μW ( V OC = 109 V and I SC = 2.8 μA) with an external load resistance of 2 MΩ. As can be seen from Figure 4e and Figure 4f , the resistance linearly increases from 0.35 to 1.18 Ω/cm 2 when the tensile strain reaches 300%. In the process, the thickness of the SCGM layer becomes thinner and the resistance of the S-TENG increases linearly. When the S-TENG is further stretched, the slope of the curve falls again because the relative variation of the SCGM layer decreases. After releasing the strain, the resistance is recovered at 0.11 Ω/cm 2 resulting from a hysteresis in the rearrangement of the SCGMs. Although the curves between sheet resistance and tensile strain have different shapes, the resistance value before and after stretching is of the same order, exhibiting the excellent flexibility and mechanical robustness of the S-TENG. Applications of S-TENG Charging performance and monitoring human motion The electrical output of S-TENGs has been used to power small electronic components [ 17 ]. In Video 2 ( Supporting Information File 2 ) the S-TENG is connected to a linear motor. 235 LEDs connected in series can be lit up after rectification of the output. Because of the video frame rate, some LEDs that light up are not captured. Figure 5a and Video 3 ( Supporting Information File 3 ) show that a scientific calculator is powered on after padding the S-TENG for eight seconds. If the S-TENG device continued to generate electrical output, the calculator would work for a longer time when needed. Figure 5b and Video 4 ( Supporting Information File 4 ) show that LEDs showing the word “HENU” are lit when they are connected to the S-TENG. Figure 5 Optical image of the power supplied by padding the S-TENG with the hand for (a) a calculator and (b) LEDs showing “HENU”. The inset shows the circuit diagram of the rectifier. (c) Demonstration of the S-TENG to detect human motion; (I, II) the position of the S-TENG in a shoe, and (III) powered LEDs; (IV) current output of the S-TENG during the two different motions. (d) The S-TENG can be fixed at the waist of a human and can light up 235 LEDs. The S-TENG can be put into shoes to monitor human movement [ 24 ]. An S-TENG device with a diameter of 50 mm and a thickness of 3 mm is suitable to be worn regularly. As shown in Video 5 ( Supporting Information File 5 ), the primary movement of a human being is recorded. The current of the S-TENG for walking and jumping is 1.5 and 2 μA, respectively, as shown in Figure 5c . Two forward electrical signals can be captured every one second during walking. From the graph, the backward-signal I SC generated by the three exercise modes is higher than the forward-signal I SC . The air mixed with SCGMs of the S-TENG is squeezed during the process of stamping continuously. Thus, the S-TENG cannot be recovered to the original state. Different motions produce different signals. The I SC generated by jumping is higher than that generated by walking, due to the larger applied force under intense exercise conditions. The S-TENG device can be potentially applied to harvest energy from different human motions and yield motion statistics. These data can be used for the analysis of physical exertion and exercise intensity by using a micro-processing unit that includes analog-to-digital conversion and wireless transmission. From the number of steps a person takes each day, the number of calories burned through exercise can be estimated. A person can keep trying to lose weight using the device. Also, the S-TENG device can monitor the rehabilitation training of post-op patients who need to avoid excessive physical training that may be harmful to heart or lung. It can also be used to check heartbeat and breathing rate, which are low-intensity movements, for comprehensive health analysis and evaluation [ 17 ]. The S-TENG can be placed in different positions of the human body to harvest motion energy with stimulation from another triboelectric layer, such as clothing or the hand. An S-TENG with an area of 100 × 100 mm 2 was placed on the waist of a person. As shown in Figure 5d and Video 6 ( Supporting Information File 6 ), the S-TENG can light up 235 LEDs by the continuously padding the device in single-electrode contact-separation mode. The S-TENG also can be placed on elbow and knee joints and harvest body motion energy for wearable devices [ 25 ]. Sensing applications The S-TENG provides an effective power source for electronic devices. Another potential application for the S-TENG is as flexible tactile sensor that can serve as electronic skin for a more comfortable interactive experience between humans and external objects by sensing all kinds of information, such as size, shape, and texture [ 26 – 27 ]. The flexible tactile sensor can generate electrical signals in response to different mechanical stimuli for the self-supply with energy. An S-TENG with an area of 20 × 20 mm 2 was placed on each of five fingertips, as exhibited in Figure 6a . When the thumb touches index finger, middle finger, ring finger and small finger in turn, the resulting I SC is 1, 1.2, 0.8, and 0.3 μA, respectively, as depicted in Figure 6b . Figure 6 (a) The position of the S-TENGs on the fingers. (b) Current output of the thumb touching the other fingers. (c, d) Potential application of the S-TENG in robotic sensing. (e) The circuit diagram of the signal processing unit. A single-electrode S-TENG was installed on the finger of a robotic hand [ 26 ], as shown in Figure 6c ,d and Video 7 and Video 8 ( Supporting Information File 7 , Supporting Information File 8 ). An electrical signal is generated when the robot touches an object. When the output signal passes through the signal-processing circuit, the digital waveform can be obtained. The current level (high or low) indicates whether finger and object are in contact. Further data processing can realize basic contact perception. Figure 6e shows the circuit diagram of the signal-processing unit."
} | 5,031 |
38167379 | PMC10761713 | pmc | 249 | {
"abstract": "Memristor-integrated passive crossbar arrays (CAs) could potentially accelerate neural network (NN) computations, but studies on these devices are limited to software-based simulations owing to their poor reliability. Herein, we propose a self-rectifying memristor-based 1 kb CA as a hardware accelerator for NN computations. We conducted fully hardware-based single-layer NN classification tasks involving the Modified National Institute of Standards and Technology database using the developed passive CA, and achieved 100% classification accuracy for 1500 test sets. We also investigated the influences of the defect-tolerance capability of the CA, impact of the conductance range of the integrated memristors, and presence or absence of selection functionality in the integrated memristors on the image classification tasks. We offer valuable insights into the behavior and performance of CA devices under various conditions and provide evidence of the practicality of memristor-integrated passive CAs as hardware accelerators for NN applications.",
"introduction": "Introduction Artificial neural networks (ANN) are indispensable for a wide range of artificial intelligence (AI) applications, including real-world data processing, such as pattern recognition, classification, and predictive modeling 1 – 8 . However, the growing complexity of these applications requires advanced ANN architectures capable of processing vast amounts of data with high precision and low power consumption. The simultaneous development of software and hardware is crucial for accelerating NN computations. Various software-level approaches have been proposed to obtain lightweight semantic segmentation networks, including quantization, compression, and lightweight architecture design 9 – 14 . Quantization and compression are effective strategies, but result in a large loss of accuracy 9 – 12 . By contrast, the design of lightweight architecture, such as depth-wise separable convolution, can enhance computational efficiency without sacrificing accuracy 13 , 14 . Hardware accelerators such as graphic processing units (GPUs), field programmable gate arrays (FPGAs), and application-specific integrated circuits (ASICs) can speed up NN algorithms 15 – 18 . However, these devices are still limited by the memory wall in the von Neumann architecture and cannot meet the high-speed and low-energy-consumption requirements of ANNs. The process-in-memory (PIM) computing architectures have recently been proposed to overcome the limitations of conventional systems for advanced ANNs. These architectures have gained popularity as alternatives to conventional computing systems because of their highly advantageous system topology, which enables the concurrent implementation of data processing and storage in a single chip. When processing units are directly integrated into the memory of a PIM architecture, computations can be performed in the memory, thereby eliminating the need for data transfer between the memory and processing units, significantly reducing data movement overheads and enhancing energy efficiency. Moreover, the PIM strategy enables parallel data processing with reduced energy consumption and increased memory bandwidth, rendering it a promising solution for applications that require high speed, low power consumption, and complex ANN operations 19 – 23 . Indeed, PIM computing has achieved 10- and 100–1000-fold improvements over CPU and GPU accelerators in terms of speed and energy efficiency, respectively 24 . Memristive crossbar arrays (CAs) have emerged as a key component that serves as the memory unit in PIM architectures. These CAs enable fundamental computational operations, with vector matrix multiplication (VMM) as a crucial operation that exploits the principles of Ohm’s and Kirchhoff’s laws. Accelerating VMM operations using memristive CAs offers significant advantages in tasks that involve heavy matrix computations such as speech and image classification. The processing efficiency of these matrix-intensive tasks can be significantly enhanced by harnessing the power of memristors. A key benefit of memristors is their non-volatile nature, which profoundly impacts the overall computing system performance. The non-volatility of memristors eliminates the need for frequent data fetching and communication between the memory and processing units. This reduction in data movement minimizes latency and results in substantial energy savings. Consequently, memristor-based CAs could potentially improve the performance of computing systems, particularly in applications in which speed and energy efficiency are critical. Recent studies have utilized memristor CAs to implement NN applications. Various memristor devices, including magnetic random-access memory (MRAM), phase-change random-access memory (PRAM), and resistive random-access memory (RRAM), have been integrated into CAs to accelerate NN computations. In particular, RRAM has gained significant attention owing to its favorable characteristics, including high scalability, good analog-switching ability, excellent endurance, low power consumption, and high switching speed 25 – 35 . However, the integration of memristors into CAs is critically challenged by sneak currents from neighboring cells, which can lead to interference and inaccurate results. This issue is commonly addressed by inserting selection functionality into CAs. Various approaches, including the integration of transistors with memristors in a one-transistor one-memristor (1T1M) configureation 31 , 36 – 39 , and the incorporating of selector devices such as ovonic threshold switching 40 – 42 , mixed ionic-electronic conductors 43 , and field-assisted super-linear threshold switching devices 44 in a one-selector one-memristor (1S1M) configuration have been explored. However, integrating transistors in the 1T1M configuration can result in area overhead issues, and integrating selectors in the 1S1M configuration poses practical and compatibility challenges for memristors. Self-rectifying memristors (SRMs) have emerged as a promising solution to overcome these challenges. They possess inherent selection functionality, which allows them to effectively suppress sneak currents and enable accurate and efficient VMM operations in CAs. The need for additional transistors or selectors can be eliminated by leveraging the inherent rectifying behavior of SRMs, leading to fewer area overhead and compatibility issues and straightforward and efficient VMM operations. Most previous research implementing NN applications, particularly classification, has concentrated on simulating the performance and accuracy of CA computations, rather than showcasing practical computations within memristor-based CAs. Hardware-based implementations are highly practical and straightforward; however, the resolution of reliability issues associated with the CA and integrated memristors that constitute the hardware accelerator is challenging. Several conditions must be satisfied to achieve the practical implementation of this approach. First, the CA must have sufficient yield and functional control. Second, the memristor device must exhibit reliability in terms of non-volatility, uniform operating characteristics across all cells, selection ability, and operational repeatability. Developing a hardware-accelerating system that fulfils these conditions is extremely difficult, which explains the limited number of research results on hardware-based implementations for NN applications. Recent studies have shown that memristor-based NN computations can accelerate VMM operations for real-world data classification such as facial images and the Modified National Institute of Standards and Technology (MNIST) dataset. Hu et al. 36 . utilized a 128 × 64 memristor-based CA for MNIST dataset classification and achieved 89.9% classification accuracy using a 1T1M-based CA to prevent sneak currents. Kim et al. 45 . integrated TiN/Al/Ti/TiO 2-x /Al 2 O 3 /TiN stacked memristors into a 64 × 64 passive CA and demonstrated high recognition accuracy for MNIST image classification. However, no studies on fully hardware-based image classification implementations using SRM-based passive CAs have been demonstrated. Moreover, most studies on SRMs have not addressed reliability issues, such as non-volatility, lack of sufficient selectivity, and feasibility of pulse-based operations, which are crucial for practical CA circuit operations. We had previously demonstrated. 46 a highly reliable and energy-efficient SRM-based passive CA with excellent performance, and identified the feasibility of a VMM accelerator using this CA; however, the device did not achieve actual data processing, such as image classification. In this study, we propose a fully hardware-based ANN computation accelerator using a 1 kb passive CA integrated with a previously developed HfSiO x -based SRM. We successfully fabricated the 1 kb CA in 100% yield. The high selectivity, low device-to-device (D2D) variation, and robust non-volatility of the developed device enabled highly reliable VMM operations. Training and inferencing for image classification operations were also achieved with very low error rates. We then conducted experiments to validate the impact of key variables on the inferencing accuracy of an SRM-based hardware accelerator. First, we investigated the effect of the inclusion of 30% and 50% defective cells in the 1 kb CA on the inferencing accuracy in image classification operations. Second, we explored the impact of the memristor’s reading margin (on/off ratio) on the classification accuracy. Although previous researchers assumed that a large reading margin is essential to improve inferencing ability, we found that the reading margin of the memristive device does not significantly affect its inferencing accuracy. Finally, we emphasized the importance of incorporating selection functionality into CAs for the reliable operation of ANN computation accelerators. We comparatively investigated the impact of the absence of selection functionality on the accuracy of VMM operations by integrating a memristive device without selection functionality into an 8 × 8 CA. We found that selection functionality is essential for accurate VMM operations in a CA, and that the absence of selection functionality causes a significant degradation in NN computing accuracy.",
"discussion": "Discussion In this study, we propose a fully hardware-based image classification method using SRM devices integrated into a 1 kb CA. The successful demonstration of image classification tasks using the CA is attributed to the highly reliable characteristics of the integrated SRM. Although general memristive devices are regarded as immature devices for implementation in largely integrated devices owing to their stochastic nature, we successfully achieved reliable characteristics in our memristive device and integrated the SRM into a 1 kb CA without the rendering of a transistor. The most important characteristic of the SRM is its non-filamentary switching operation, which can ensure a narrow operational distribution for high yields in the CA. By excluding the initial electroforming process, we can achieve favorable cycle to cycle (C2C) and D2D characteristics; by contrast, CF-based memristive devices suffer from a large operational distribution. Whereas CF-based memristive devices generally show large reading-margin and robust retention characteristics, non-filamentary-type devices generally exhibit a low reading-margin and poor retention characteristics 48 , 54 , 55 . However, the weaknesses of non-filamentary-type memristive devices can be improved by modifying the materials and device structure. In our previous research, we achieved reliable retention characteristics even in the low-operating-current regime (<100 nA) by adopting functional layers (i.e., an oxygen reservoir and oxygen-diffusion barrier layers) in the SRM. In this study, we attempted to demonstrate that another weakness of non-filamentary-type memristive devices, that is, their low reading-margin, is not a crucial factor for image classification tasks. As confirmed by our experiments, variations in reading margin did not influence the inferencing accuracy in image classification tasks. It has been thought that the CAs are able to tolerate defects in performing VMM operations owing to their severely parallel structure. Thus, even if some defective points are present in the CA, the array can render the defective cells redundant and achieve the correct answer through other signal pathways. However, the experimental demonstration in this study showed that defective cells in the CA had a significant effect on the VMM operation and the accuracy of image classification tasks. Note that we set the defective cells in the CA to its HRS, assuming an open electrical circuit. However, if some defective cells in the CA are in an electrical short circuit, the image classification tasks would show an additional degradation in accuracy. Therefore, for reliable CA operation, the operational yield of individual memristive devices in the CA should be close to 100%. In this case, non-filamentary-type memristive devices may be a good candidate because of their high yield and reliability. In addition, considering the classification accuracy with terms of bit precision in our system, it is essential to clarify the difference between the nature of the proposed CA conventional digital architectures. While typical digital systems rely on bit precision, the proposed CA operates primarily on analog signals. This fundamental difference means that our approach does not directly engage with bit precision in the traditional sense. However, it is crucial to understand the steps taken to align the MNIST data set with the CA’s operational mode. MNIST’s input was reduced to 320 units and normalized pixel values were rounded off to 0 or 1 to simulate the CA’s input. Though digital, this process merely digitized input, not weights or CA procedures. The device array utilizes experimental weights adjusted to match its conductance and input current scales. This ensures the CA computes using genuine analog values. To elaborate on the learning process, the reduced input of 320 was chunked into groups of 10 and fed to a three-line array. Each segment has 32 inputs and 3 outputs. Multiplying by weights yields the handwriting classification result for each 10-part segment. The final input is estimated using a soft-max function after aggregating these scores. The calculated weights were converted to the target conductance and applied to the CA using the following formula: 2 \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${{{{{{\\rm{G}}}}}}}_{{{{{{\\rm{T}}}}}}}={{{{{\\rm{W}}}}}}\\times \\left({{{{{{\\rm{G}}}}}}}_{{{{{{\\rm{LRS}}}}}}}{{{{{\\rm{\\hbox{--}}}}}}}{{{{{{\\rm{G}}}}}}}_{{{{{{\\rm{HRS}}}}}}}\\right)$$\\end{document} G T = W × G LRS – G HRS Let, the conductance for training = G T , simulated synaptic weight = W, G LRS = LRS conductance of CA, and G HRS = HRS conductance of CA. Regarding the weights obtained through software, they are represented using the float 32 data type, offering sufficient precision for our needs. Given that the device responsible for applying conductance to the array can handle a range from 0.1 A to as precise as 0.1 pA, there should be confidence in the CA’s conductance resolution to represent and manipulate these synaptic weights accurately. Furthermore, in terms of the required precision for NN applications, it varies based on the specific task and network architecture. However, with our approach and the inherent precision of the CA’s design, we believe that the system is well-equipped to meet and potentially exceed the precision requirements commonly associated with NN applications. The memristive device in the CA must have selection functionality to ensure reliable VMM operations based on the CA. This functionality is necessary for both training and inferencing, as it enables correct and accurate operations using the CA. In this study, we comparatively investigated the operation of a CA integrated with a memristive device without selection functionality. Although our intention was to obtain randomly distributed resistance states for the HRS and LRS in the 8 × 8 CA (Fig. 6c ), the verification results (Fig. 6d ) differed significantly from the initially intended resistance state distribution. Previous research often employed 1T1M-structured unit cells to implement selection functionality in each cell in the CA. Given the increasing demand for a vast memory capacity in future computing technologies to process large amounts of data, a singular form of selection and memory can provide an advantage in terms of the dense integration of electronic devices. From this perspective, a non-filamentary switching-type SRM with high selectivity offers reliable performance and robust reproducibility. In summary, we proposed a 1 kb CA integrated with SRMs as a hardware accelerator for ANN applications. Image classification tasks using our hardware accelerator revealed a remarkable recognition accuracy of 100%, which is comparable with that of software-based simulations. Furthermore, the impact of non-ideal factors on the hardware accelerator was investigated. We highlighted the merits of hardware accelerators based on SRMs. Our accelerator demonstrated uniform operational characteristics across all devices in the CA, with extremely low defectiveness (resulting in high yields) during fabrication and high immunity to sneak currents from neighboring cells. We believe that non-filamentary-type-based SRMs can serve as a key device component in future ANN applications. We also provide valuable insights for addressing reliability issues associated with memristive devices. Through this study, the NN applications utilizing memristive crossbar arrays are compared in Table 1 . It is noteworthy that our study represents the first reported demonstration in the field of SRMs, utilizing a purely self-rectifying memristor-based passive crossbar array for hardware implementations of NN applications. While the energy efficiency of the proposed CA is currently lower than ref. 36 , 56 ., we anticipate that higher efficiency can be achieved by increasing the array density. Table 1 The comparison results of hardware implementation for neural network applications using memristive crossbar arrays This work Ref. 57 . Ref. 36 . Ref. 30 . Ref. 31 . Ref. 56 . Array configuration 1 R (SRC) 1S-1R 1T-1R 1T-1R 1T-1R 1T-1R Array density 32 × 32 (1 kb) 9 × 9 (<1 kb) 128 × 64 (8 kb) 54 × 108 (~5 kb) 128 × 16 (2 kb) 128 × 64 (8 kb) Array yield (%) 100 100 98.9 Not available 99 Not available Energy efficiency (TOPS/W) 4.35 Not available 115 1.37 11 77.4 Application MNIST classification Pattern classification MNIST classification Pattern classification MNIST classification MNIST classification Accuracy 100 100 89.9 94.6 96.19 92.3 Integrated device stack Ru/HfSiO y / Al 2 O 3 /HfSiO x /TiN Pt/Ru /TiO 2 /RuO 2 /Pt /HfO 2 /TiN Ta/HfO x /Pd Pd/ WO x /Au TiN/TaO x /HfO x /TiN Pt/TaO x /Ta Furthermore, we compare the energy efficiency of our hardware accelerator with traditional microelectronic chips such as GPUs, FPGAs, and ASICs in Table 2 . The results of the comparison show that our hardware accelerator achieves relatively high energy efficiency due to significantly lower power consumption compared to traditional microelectronic chips. For detailed calculations, refer to Supplementary Note 1 . Table 2 The comparison results with traditional microelectronic chips, which function as hardware accelerators for computation This work Ref. 58 . Ref. 59 . Ref. 60 . Ref. 61 . Ref. 62 . Ref. 63 . Hardware Memristive passive crossbar array GPU (Titan X) FPGA (Virtex 7) FPGA (Arria-10) FPGA (Zynq-7000) ASICs (YodaNN) ASICs (Google TPU) Energy efficiency 4.35 TOPS/W 98 GOPS/W 935 GOPS/W 30 GOPS/W 14.3 GOPS/W 8.6 TOPS/W 2.3 TOPS/W"
} | 5,056 |
25097866 | PMC4109225 | pmc | 250 | {
"abstract": "Microbial species have evolved diverse mechanisms for utilization of complex carbon sources. Proper combination of targeted species can affect bioenergy production from natural waste products. Here, we established a stable microbial consortium with Escherichia coli and Shewanella oneidensis in microbial fuel cells (MFCs) to produce bioenergy from an abundant natural energy source, in the form of the sarcocarp harvested from coconuts. This component is mostly discarded as waste. However, through its usage as a feedstock for MFCs to produce useful energy in this study, the sarcocarp can be utilized meaningfully. The monospecies S. oneidensis system was able to generate bioenergy in a short experimental time frame while the monospecies E. coli system generated significantly less bioenergy. A combination of E. coli and S. oneidensis in the ratio of 1 : 9 (v : v) significantly enhanced the experimental time frame and magnitude of bioenergy generation. The synergistic effect is suggested to arise from E. coli and S. oneidensis utilizing different nutrients as electron donors and effect of flavins secreted by S. oneidensis . Confocal images confirmed the presence of biofilms and point towards their importance in generating bioenergy in MFCs.",
"conclusion": "4. Conclusions In summary, we have demonstrated bioenergy generation in MFCs by employing a natural and abundant feedstock, coconut sarcocarp. The common EAB, S. oneidensis , and the non-EAB, E. coli , were employed and a possible synergy was suggested, based on the ratio of microbial species introduced to the system. This demonstration paves the way forward for exploration of alternative and natural energy sources using mixed species consortia.",
"introduction": "1. Introduction Unprecedented industrialization and the continued spurt in population growth have vastly depleted global natural energy sources. This has led to an acute need for alternative, clean, and renewable energy sources. In particular, extensive efforts have been invested into increasing efficiencies of solar cells [ 1 ], which has been envisioned as the next frontier in renewable energy. Another potential source of alternate energy lies in producing bioenergy from agricultural waste products via microbial activity. One potential approach to producing bioenergy from natural waste products is through microbial fuel cells (MFCs) which employ the extracellular electron transport (EET) functionalities of electrochemically active bacteria (EAB) to facilitate electron transport and thus produce electricity from diverse energy sources [ 2 ]. The free electrons and protons originate from microbial metabolism of organic components found in the media within an anaerobic anode chamber. Metabolism is achieved when electrons move along the cascading energy pathway of the electron transport chain, which releases energy for continued survival of the microorganism. These electrons are further transported by various EET mechanisms to the external terminal electron acceptors. A voltage is generated in the process of electrons moving across the external resistor towards the cathode. Protons diffuse simultaneously across the selective proton exchange membrane to the aerobic cathode chamber. In this compartment, oxygen is reduced by electrons and protons to produce water molecules in order to complete the charge balance. Although this technology has matured over time and shows promise in concurrent bioremediation and power generation [ 3 ], it has seen little commercial success. This is due to high material cost and low power performance that is partly caused by limitations in inferior charge transport at the inherently insulating microbe-electrode interface. Much effort has been invested in circumventing these bottlenecks. Recently, enhanced power output in MFCs has been demonstrated through chemical modification of the insulating interface junction across Escherichia coli cellular membrane [ 4 , 5 ] and genetic engineering of Pseudomonas aeruginosa to enhance endogenous secretion of pyocyanin mediators [ 6 ]. Further, small-scale stacked MFCs have been shown to power mobile devices using human urine as an energy source [ 7 ]. Better understanding of microbial species interactions employing EET processes has been proposed as a promising strategy to improve the performance of MFCs [ 8 , 9 ]. In this contribution, a synergistic microbial consortium was established and modified for bioenergy generation from a complex energy source, in the form of the coconut sarcocarp, which is defined as the fleshy part of the fruit. According to statistics from the Food and Agriculture Organization of the United Nations, ~54 million tonnes of coconut were produced in 2010 from mainly tropical coastal countries, of which a large amount is wasted [ 10 , 11 ]. However, coconuts are known to be rich sources of sugars, fats, oil, and carbohydrates with small beneficial concentrations of vitamins and salts [ 12 ]. Hence, coconuts which are considered as waste products can be potentially used as an alternative and natural energy source for tropical coastal countries. By choosing the model non-EAB ( E. coli ) and the EAB ( Shewanella oneidensis ), we demonstrate that modification of the ratio of bacterial strains introduced into the microbial consortium can significantly improve MFC performance.",
"discussion": "3. Results and Discussion 3.1. Electrical Performance Dual-chamber MFCs were employed to investigate the bioenergy generated as the coconut sarcocarp is broken down through microbial oxidation by the respective bacterial species. The average current densities generated over 72 h were recorded ( Figure 1 ). The consistently low current densities from all operated MFCs were attributed to high internal resistances within the bioelectrochemical devices, which impede charge movement. Further, the media in the anode and cathode chambers contained the sarcocarp slurry, which has limited conductivity. This can be averted through various forms of optimization, such as adopting different device architectures [ 14 ], apparatus components, or electrode engineering [ 15 ]. However, the focus of this study was to demonstrate facile bioenergy generation through the use of a natural, abundant, and readily available energy source, coconut sarcocarp, by employing common bacterial species. MFCs inoculated with E. coli generated an average maximum current density of ~0.015 μ A/cm 2 ( Figure 1 , black trace), whereas S. oneidensis MFCs generated ~0.05 μ A/cm 2 ( Figure 1 , red trace). The rapid decrease in average current density generated by the S. oneidensis MFCs after ~6 h is caused by the depletion of suitable energy sources available for S. oneidensis . This is because the single fed batch MFC system was employed in this study, which is in contrast to a continuous fed system, where the energy source can be renewed through a steady exchange of spent and fresh medium. The average current density generated by the S. oneidensis MFCs stabilized at a significantly lower current density of ~0.005 μ A/cm 2 up to 72 h. MFCs without any inoculum were also operated and negligible current density was generated ( Figure 1 , grey trace). This indicates that the observed current densities were driven by the microbial actions of E. coli and S. oneidensis mono- and cocultures on the sarcocarp. It has been well established that electrochemically active S. oneidensis has various forms of EET mechanisms, such as conducting outer membrane cytochromes [ 16 ], nanoappendages [ 17 ], and secretion of flavins [ 18 ], which act as charge transport mediators. These mechanisms are electrical conduits to transfer microbially released electrons to terminal electron acceptors. The ~3-fold difference in average maximum current density from monoculture systems is attributed to poorly evolved E. coli EET mechanisms which lack the diversity and effectiveness of EET mechanisms in S. oneidensis . Notably, significant bioenergy generation by S. oneidensis started from an early stage, while output from E. coli only started to increase later. This suggests that S. oneidensis and E. coli might utilize different energy sources present in the sarcocarp for bioenergy generation. It is further hypothesized that, in coculture MFCs containing E. coli and S. oneidensis , a possible synergistic effect involving flavins has been created. To test this hypothesis, coculture systems utilizing various ratios of E. coli and S. oneidensis were operated to investigate possible synergistic interactions. Interestingly, a 5 : 5 (50% : 50%, v : v) coculture system produced a maximum current density of ~0.045 μ A/cm 2 ( Figure 1 , blue trace). As compared to the monoculture systems ( Figure 1 , red trace for S. oneidensis , black trace for E. coli ), the 5 : 5 coculture system could generate a relatively sustainable and significant current density over 72 h. It is thus noteworthy to further probe the effect of different bacterial ratios on the extent of bioenergy generation. A 1 : 9 (v : v) E. coli and S. oneidensis system generated a maximum current density of ~0.055 μ A/cm 2 ( Figure 1 , orange trace), whereas a 9 : 1 (v : v) E. coli and S. oneidensis system generated a maximum current density of ~0.025 μ A/cm 2 ( Figure 1 , green trace). The ratio modification study suggests that introducing a higher concentration of S. oneidensis in the coculture systems allows for maximum exploitation of suitable energy sources for S. oneidensis . This strategy may minimize consumption of such energy sources by E. coli , which generates significantly lesser bioenergy, and facilitate generation of excess secreted flavins to enhance E. coli bioenergy generation at a later stage when the species uses suitable energy sources for itself. 3.2. Biofilm Characterization Further, the role of biofilms in the bioelectrochemical systems was elucidated by confocal microscopy characterization. Representative overlaid brightfield and confocal images were acquired from random strands of electrodes in respective MFCs. Biofilms were formed in all systems (Figures 2(a) and 2(b) ). To differentiate between each species, E. coli was tagged with red fluorescent protein (RFP), whereas S. oneidensis was tagged with green fluorescent protein (GFP). The RFP-tagged E. coli biofilm and GFP-tagged S. oneidensis biofilm were evident on the electrode fiber surfaces (Figures 2(a) and 2(b) ). The confocal images corroborate the importance of the biofilm in the electrical performances with specific bacterial strains. 3.3. Mechanistic Insights of the Functional Coculture System The following possible mechanisms occurring in the coculture system were proposed ( Figure 3 ). Various favourable nutrients (represented by blue and green dots) present in the sarcocarp can be broken down specifically by the independent microbial oxidative actions of non-EAB ( E. coli ) and EAB ( S. oneidensis ) in different stages of MFC operation to produce bioenergy. From the electrical data ( Figure 2 ), it is suggested that, for significant and sustained bioenergy production, the EAB should be introduced at a higher concentration. This is to restrict nutrient consumption by non-EAB. The EAB also breaks down its suitable energy source and secretes flavins, which can be utilized by non-EAB at a later stage to facilitate EET. Gradual decline in current densities is due to lack of available nutrients in the closed system ( Figure 1 )."
} | 2,880 |
37448593 | PMC10338049 | pmc | 252 | {
"abstract": "Spider silk is considered a promising next-generation biomaterial due to its exceptional toughness, coupled with its renewability and biodegradability. Contrary to the conventional view that spider silk is mainly composed of two types of silk proteins (spidroins), MaSp1 and MaSp2, multi-omics strategies are increasingly revealing that the inclusion of complex components confers the higher mechanical properties to the material. In this review, we focus on several recent findings that report essential components and mechanisms that are necessary to reproduce the properties of natural spider silk. First, we discuss the discovery of MaSp3, a newly identified spidroin that is a major component in the composition of spider silk, in addition to the previously understood MaSp1 and MaSp2. Moreover, the role of the Spider-silk Constituting Element (SpiCE), which is present in trace amounts but has been found to significantly increase the tensile strength of artificial spider silk, is explored. We also delve into the process of spidroin fibril formation through liquid-liquid phase separation (LLPS) that forms the hierarchical structure of spider silk. In addition, we review the correlation between amino acid sequences and mechanical properties such as toughness and supercontraction, as revealed by an analysis of 1,000 spiders. In conclusion, these recent findings contribute to the comprehensive understanding of the mechanisms that give spider silk its high mechanical properties and help to improve artificial spider silk production.",
"conclusion": "Conclusion Spider silk is a high-performance next-generation biomaterial with exceptional physical properties and biodegradability, making it a sustainable alternative to synthetic fibers, and a potential contributor to the Sustainable Development Goals (SDGs). Molecular mechanisms realizing the extraordinary properties of spider silk is beginning to be uncovered, with the advent of omics technologies. Firstly, the full list and stoichiometry of the components of dragline silk was revealed by long-read sequencing and quantitative proteomics, and a non-spidroin component SpiCE was identified as a minor but essential component for silk strength, which doubles the tensile strength of artificial spider silk in vitro . Secondly, the self-assembly process of spidroins involves LLPS, and the N/C terminal domains are essential for this purpose. Spider silk assembled via LLPS realizes the hierarchically bundled structure. Thirdly, through the study of 1000 spiders from all over the world, relationships between spidroin sequences and mechanical properties have been identified, and this data has been compiled in the publicly available Spider Silkome Database. Together, these recent findings lay the foundation for the industrial-scale application of spider silk.",
"introduction": "Introduction (Toughest Material: Spider Silk) Silk has been valued as a luxurious and versatile material for centuries, and the Silk Road was a major trade route for silk fabrics from the 2nd century BC to the 15th century. Silkworms produce silk that is not only used in clothing, but also has a variety of applications in fields such as medicine as a biologically-derived material. In the late 20th century, the oil industry saw significant growth, leading to the widespread adoption of synthetic fibers like nylon and polyester as substitutes for natural fabrics. However, synthetic fibers have environmental drawbacks, such as the release of carbon and microplastics, as they are made from oil are not biodegradable [ 1 , 2 ]. In an era where sustainable development goals (SDGs) are being set globally and a new industrial revolution driven by biotechnology is sought, protein materials designed using protein engineering have garnered renewed interest as the next generation of high-performance materials, that are highly functional, renewable, and biodegradable [ 3 ]. Many arthropods especially in the order Lepidoptera utilize silk, with a tensile strength of about 0.5 GPa in Bombycidae (silkworms) and 0.7 GPa in Psychidae (bagworms). Silk from orb weaving spiders often surpasses arthropod silk in strength, with a tensile strength of more than 1 GPa and a degree of extensibility (30%) comparable to that of nylon. This makes spider silk one of the toughest of all natural materials [ 4 – 6 ]. However, breeding spiders is challenging due to their carnivorous nature and tendency to engage in cannibalism, making the mass production of natural fibers difficult. In recent years, there have been promising developments in the effort to industrialize artificial spider silk, including the synthesis of spider silk through microorganisms [ 7 ] and synthesis of polypeptides that mimic its molecular structures [ 8 ]. While much is still unknown about the relationship between the arrangement and properties of spider silk, reproducing the exceptional properties of natural spider silk requires a deeper understanding of the molecular-level structural design principles and mechanisms of protein materials [ 9 – 12 ]. Recent advances in high-throughput molecular biology, including genomics, transcriptomics, and proteomics collectively referred to as multi-omics have made it possible to conduct analyses in this regard. This review article is an extended version of the Japanese article [ 13 ]."
} | 1,329 |
36589136 | PMC9799978 | pmc | 253 | {
"abstract": "Plant-plant interactions and coexistence can be directly mediated by symbiotic arbuscular mycorrhizal (AM) fungi through asymmetric resource exchange between the plant and fungal partners. However, little is known about the effects of AM fungal presence on resource allocation in mixed plant stands. Here, we examined how phosphorus (P), nitrogen (N) and carbon (C) resources were distributed between coexisting con- and heterospecific plant individuals in the presence or absence of AM fungus, using radio- and stable isotopes. Congeneric plant species, Panicum bisulcatum and P. maximum , inoculated or not with Rhizophagus irregularis , were grown in two different culture systems, mono- and mixed-species stands. Pots were subjected to different shading regimes to manipulate C sink-source strengths. In monocultures, P. maximum gained more mycorrhizal phosphorus uptake benefits than P.bisulcatum . However, in the mixed culture, the AM fungus appeared to preferentially transfer nutrients ( 33 P and 15 N) to P.bisulcatum compared to P. maximum . Further, we observed higher 13 C allocation to mycorrhiza by P.bisulcatum in mixed- compared to the mono-systems, which likely contributed to improved competitiveness in the mixed cultures of P.bisulcatum vs. P. maximum regardless of the shading regime. Our results suggest that the presence of mycorrhiza influenced competitiveness of the two Panicum species in mixed stands in favor of those with high quality partner, P. bisulcatum , which provided more C to the mycorrhizal networks. However, in mono-species systems where the AM fungus had no partner choice, even the lower quality partner (i.e., P.maximum ) could also have benefitted from the symbiosis. Future research should separate the various contributors (roots vs. common mycorrhizal network) and mechanisms of resource exchange in such a multifaceted interaction.",
"conclusion": "5 Conclusion The two host plants in our study supported their fungal partner in different ways. We found a disproportionate allocation of C resources from different plant species to their associated AM fungi, with more 13 C transferred from P.bisulcatum than from P.maximum to the mycorrhiza in both mono and mixed systems, suggesting that P.bisulcatum invests more C than P.maximum into the mycorrhizal symbiosis. The higher C allocation of P.bisulcatum to mycorrhiza suggests a high-quality (or high-intensity) interaction between P.bisulcatum and its fungal partner, leading to higher abundance of AM fungi in its roots and surrounding soil compared with P.maximum. \n Our results demonstrated the advantage of more mycorrhiza-dependent P.bisulcatum , when grown together with less mycorrhiza-dependent P.maximum , particularly in terms of uptake of recent nutrients (shoot 33 P and 15 N) under variable light conditions in the mixed system. In addition, P.maximum was negatively affected by the enhanced competitive ability of P.bisulcatum in the presence of AM fungus. These findings suggested that the effects of AM fungal presence in mixed system were closely related to the degree of the host plant dependency on mycorrhizal association. The fungus preferentially transferred more nutrients to the more mycorrhiza-dependent plants, which in turn provided more C and enhanced its ability to thrive even under shading, on the expense of the less mycorrhiza-dependent plant. In contrast, the effects of CMNs formed by a single fungal taxon on plant nutrient uptake in a mono system are mainly exploited by the plant partner, since there is obviously no other choice for the fungus. Overall, our results showed that the mycorrhizal symbiosis strongly affected plant species coexistence by enhancing differences in plant fitness through asymmetric resource allocation in favor of a higher quality host. On this basis, preferential allocation could enhance the success of plant species with greater mycorrhiza-dependency and/or more C provision to the AM fungus, when in plant communities. Future research is required to test the general validity of our observations (more plant and fungal species to be included) and identify the factors that may further condition asymmetries in resource exchange in plant–mycorrhizal interactions. Particularly, separation of root and AM fungal contribution should be achieved [although these two often intermingle in natural settings and not all AM fungi efficiently colonize root-free patches ( Smith et al., 2004 )] to improve mechanistic understanding of the systems and our capacity to predict outcomes of the competitive interactions. This could eventually lead to better predictions of the plant community spatial and temporal dynamics and mycorrhizal functioning under different and/or gradually changing environmental conditions.",
"introduction": "1 Introduction Understanding the factors that influence the coexistence of plant species in natural ecosystems is a central concept in plant community ecology ( Bengtsson et al., 1994 ; Bever et al., 2010 ; Wilson, 2011 ). According to the most widely accepted theory of plant coexistence in ecological communities, only species with sufficiently different resource requirements (e.g., nutrients, light, water) and traits (e.g., rooting depth, phenology) can coexist in the long run ( Gause, 1934 ; Connell, 1983 ; Aarssen, 1989 ; Dybzinski and Tilman, 2007 ). Because only a limited number of distinct niches are available in the natural environment, niche overlap can lead to negative intraspecific interactions, competition for available resources, and eventually a limited number of coexisting species ( MacArthur and Levins, 1967 ; Tilman, 1982 ; Connell, 1983 ; Schoener, 1983 ). In addition to abiotic factors, biotic factors can also influence plant coexistence, such as the presence of organisms that interact positively or negatively with plants. Among those, soil microorganisms are particularly important, in spite of being little visible but having a major impact on plant coexistence, interactions, and community composition ( van der Heijden et al., 2006 ; Vogelsang et al., 2006 ; Bever et al., 2010 ; Moora and Zobel, 2010 ). Plant-plant interactions and coexistence can be directly mediated by arbuscular mycorrhizal (AM) fungi ( Leake et al., 2004 ; Scheublin et al., 2007 ; Simard et al., 2012 ), which are ubiquitous plant root symbiotic partners in a variety of terrestrial ecosystems ( Spatafora et al., 2016 ; Brundrett and Tedersoo, 2018 ). The fungal symbiont relies fully on photosynthetic carbon (C) obtained from the plant roots; in return, it provides mineral nutrients to plants, particularly phosphorus (P) and nitrogen (N), taken outside of the rhizosphere and thus out of reach for the roots themselves ( Smith and Read, 2008 ; van der Heijden et al., 2008 ). The fungal partner also provides a number of non-nutritional benefits to their host, such as improving plant-water relations ( Augé et al., 2015 ), resistance to abiotic (e.g., salinity, heavy metals, drought) and biotic (e.g., pathogens, herbivores) stresses ( Kikuchi et al., 2016 ; Faghihinia et al., 2020 ; Faghihinia et al., 2021 ; Zai et al., 2021 ). Remarkably, AM fungi can colonize the roots of different plant species simultaneously and interconnect neighbouring or co-cultivated plants by forming so called common mycorrhizal networks (CMNs) in the soil ( Klironomos, 2000 ; Selosse et al., 2006 ; Jakobsen and Hammer, 2015 ). There is compelling evidence that CMNs distribute both nutritional (e.g., transfer of nutrients) and non-nutritional benefits (transfer of defense signals or allelochemicals) between coexisting plants ( Bever et al., 2010 ; Babikova et al., 2013 ; Song et al., 2014 ), which may eventually lead to over-yielding of plant communities ( Li et al., 2022 ). Interestingly, AM fungi could disproportionately affect the fitness of coexisting plants through asymmetric resource partitioning among plants ( Bever et al., 2010 ; Weremijewicz and Janos, 2013 ), which can be attributed to some extent to host preference in resource exchange with a certain partner ( Vandenkoornhuyse et al., 2003 ; Montesinos-Navarro et al., 2019 ). If this is fully reciprocated (which is predicted by a market theory), coexisting plants benefit from mycorrhizal symbiosis based on their corresponding C investments into their (shared) fungal partner. In other words, mycorrhiza may preferentially transfer more nutrients to the plants that provide more C and less nutrients to the plants that allocate less C to the mycobiont ( Bever et al., 2009 ; van der Heijden and Horton, 2009 ; Lekberg et al., 2010 ; Kiers et al., 2011 ). However, it seems that the cost-benefit relationships between AM fungi and plants are not always interlinked. The benefits of each individual plant species from AM fungi may vary with environmental contexts, e.g., due to differences in soil type and nutrient availability ( Konvalinková et al., 2015 ; Wang et al., 2016 ; Voříšková et al., 2019 ), plant and fungal identity ( Argüello et al., 2016 ; Wang et al., 2016 ), plant size and growth stage ( van der Heijden and Horton, 2009 ), and/or existing competition among plants belonging to the same or different species for other resources ( Scheublin et al., 2007 ). In such cases, some plant species may invest more C in mycorrhiza while the other plants derive most benefits from the shared mycorrhizal networks ( Walder et al., 2012 ). Nevertheless, the terms of resource exchange between plants and their shared CMNs and the influence of mycorrhizal fungi on the outcome of plant-plant interactions are not yet fully understood ( van der Heijden and Horton, 2009 ; Montesinos-Navarro et al., 2019 ; Davison et al., 2020 ). In general, the questions of which plant species benefits most from mycorrhiza and how this is physiologically organized are still difficult to answer or predict. The redistribution of symbiotic benefits and costs can also be attributed to the different ecological strategies of plants included in a community. Indeed, the exchanged resources such as C or P could be controlled by both plant and fungal partners ( Kiers and van der Heijden, 2006 ; Bever et al., 2009 ). On this basis, coexisting plants of different species or functional groups are expected to adopt different strategies under different environmental conditions when plants are incorporated into an existing mycorrhizal network ( Bever et al., 2010 ; Jakobsen and Hammer, 2015 ). There have been few empirical attempts to experimentally assess the investment in CMNs by individual plants of the same or different species or functional groups (e.g., mycorrhizal status, growth forms, photosynthetic pathways, etc.) ( Walder et al., 2012 ; Sepp et al., 2019 ; Davison et al., 2020 ). For example, Walder et al. (2012) assessed carbon investment and nutrient gains of C 3 \n Linum usitatissimum and C 4 \n Sorghum bicolor into and from their interconnected CMNs, respectively, formed by Rhizophagus intraradices or Funneliformis mosseae using 13 C, 15 N, and 33 P as tracers. They found that the C 4 plant invested more C in the shared CMNs and received (proportionally) less N and P in return than the C 3 plant, which in turn benefited more from the symbiosis by investing less photosynthate and receiving the greatest share of nutrients ( Walder et al., 2012 ). It should be noted that disproportionate cost-benefit ratios of different plant species in the same community can be confounded by plant size, with larger individuals often receiving a larger share of limited resources and even suppressing the growth of other individuals ( van der Heijden and Horton, 2009 ; Merrild et al., 2013 ; Jakobsen and Hammer, 2015 ). To avoid misinterpretation of the results of interactions between plants associated with AM fungi, phylogenetically close plant species with approximately the same size and growth rates are an ideal model for studying interactions between plants and the fungi as affected by the AM symbiosis in general and CMN formation in particular ( Řezáčová et al., 2018b ). To date, few ecophysiological studies using up to four individuals of plants or fungi have been conducted to decipher the underlying mechanisms of C-for-P exchange between plants and the mycorrhizal networks and the interactions between coexisting plants ( Nakano-Hylander and Olsson, 2007 ; Bever et al., 2009 ; Lekberg et al., 2010 ; Kiers et al., 2011 ; Walder et al., 2012 ; Merrild et al., 2013 ; Weremijewicz and Janos, 2013 ; Fellbaum et al., 2014 ; Řezáčová et al., 2018b ; Ingraffia et al., 2021 ). One experimental approach to study the influence of mycorrhizal fungi on coexisting plant interactions and resource exchange is to impose experimental shading in order to manipulate the strength of C source sink ( Kaschuk et al., 2009 ; Olsson et al., 2010 ; Konvalinková and Jansa, 2016 ; Lang et al., 2021 ). Shading duration and intensity could significantly regulate the exchange of nutrients for C and thus, the cost-benefit ratio of the symbiosis ( Konvalinková et al., 2015 ; Zheng et al., 2015 ). This is of particular concern because light limitation can occur to varying degrees and at different temporal scales in different ecosystems, e.g., at the regional level due to sudden changes in weather (e.g., cloudy weather, monsoons, and thunderstorms), at the local level due to canopy cover by neighboring plants, or even at the microscopic level due to the formation of microbial biofilms on plant leaves ( Konvalinková and Jansa, 2016 ). Experimental evidence showed that reduced investment in symbiosis by both the plant and the fungal partner in response to light limitation may result in reduced plant biomass, AM fungal root colonization rate, C allocation from the plant to the fungal partner, and P transfer from the AM fungi to the plant ( Olsson et al., 2010 ; Fellbaum et al., 2014 ; Füzy et al., 2014 ; Shi et al., 2014 ; Konvalinková et al., 2015 ). However, it is not yet entirely clear how the reciprocal resource exchange (C in return for N and/or P) between the plant and the fungus is modulated by the reduction of assimilate supply caused by light deficiency to one or all plant partners ( Weremijewicz and Janos, 2013 ; Konvalinková and Jansa, 2016 ). Here, we aimed to understand how the presence of a single AM fungus affected plant individuals in con- and heterospecific mixtures competing for shared soil resources in their mycorrhizosphere in response to experimental shading. We tested how congeneric model grasses Panicum bisulcatum (C 3 ) and Panicum maximum (C 4 ) inoculated or not with the mycorrhizal fungus Rhizophagus irregularis , and growing in separation or in a mixture, responded to different shading treatments in terms of their biomass production, mycorrhizal colonization, and P, N, and C exchanges. For this purpose, plants were grown side by side either in a “mono system” as a pair of individuals belonging to the same plant species (C 3 -C 3 and C 4 -C 4 ) or in a “mixed system” as a pair of individuals each belonging to a different plant species (C 3 -C 4 ), inoculated or not with the AM fungus. We assessed the carbon and nutrient investments of the plant and fungal partners by tracking stable and radio isotopes using 13 C, 15 N, and 33 P as tracers, and by manipulating light interception by the different plants individually. The roots were allowed to intermingle freely and labeling of soil nutrients was not confined to a root-free compartment to achieve a greater relevancy to the field setting.",
"discussion": "4 Discussion The observations made in our glasshouse experiment offer unique insights into P, N, and C fluxes between an AM fungus and two different but closely related plant species growing in a mixture under contrasting light conditions applied locally on one or both plants per pot. This experimental model is particularly suitable for testing re-arrangements of nutrient-for-C exchanges as it prevents confounding effects of plant size under the different light regimes. We constructed a model mono system consisting of a pair of individuals of either P.bisulcatum or P.maximum grasses growing in microcosms with or without mycorrhiza to set the baseline. We further studied the association between a heterospecific plant community and the AM fungus in a mixed system, accommodating those two different plant species within the same microcosm. We found that the two plant species in our study benefited differently from their associated mycorrhiza in the mono system, and that such inequalities were generally amplified and, in consequence, significantly affected resource use by the different plant species in the mixed system. 4.1 Mono system Our study showed that P.maximum , when alone, performed somewhat better than P.bisulcatum in terms of shoot, root and total biomass production [similar to previously published results, e.g., Řezáčová et al. (2018b) and Řezáčová et al. (2017) ] in both M and NM inoculation treatments and under all shading regimes. Accordingly, P.maximum had significantly higher total P and total N contents than P.bisulcatum in the mono system. In contrast, P.bisulcatum showed significantly higher biomass partitioning between shoot and roots compared to P.maximum . Interestingly, our data also showed that P.bisulcatum generally had higher uptake of recent nutrients (total 33 P, total 15 N) than P.maximum . These results, combined with the observations of higher 13 C allocation to AM fungi (indicating a higher quality host) and higher mycorrhizal abundance in the roots and rhizosphere soil of P.bisulcatum compared to P.maximum in the mono system, led us to conclude that the two plant species likely differed in their dependence on the fungal symbiont, with P.bisulcatum exhibiting higher level of mycotrophy (or mycorrhizal dependence, particularly with respect to mineral nutrient acquisition) than P.maximum. The differential levels of dependence of the plant species on mycorrhizal symbiosis for nutrition suggests that the benefits and costs derived from the symbiotic association may differ among the two plant species, consistent with previous research ( van der Heijden and Horton, 2009 ; Bever et al., 2010 ; Hempel et al., 2013 ; Jakobsen and Hammer, 2015 ), even in the case of phylogenetically such closely related (albeit physiologically quite different) plant species as studied here. The general perception is that mycorrhizal fungi discriminate between host plants that are interconnected via CMN and preferentially allocate more mineral nutrients to high-quality (i.e., more C rewarding) hosts ( Hammer et al., 2011 ; Kiers et al., 2011 ; Fellbaum et al., 2014 ). This may also be true from a phytocentric perspective: The more the plant depends on mycorrhiza for nutrient uptake, the more likely it is to provide ample C resources to the mycorrhizal network ( Jakobsen and Hammer, 2015 ). However, in our mono system, the more mycorrhiza-dependent plant, P.bisulcatum , in spite of providing more 13 C to its fungal symbiont actually received lower 33 P benefits than the plant with the lower mycorrhizal abundance (and dependence), P. maximum ( \n Figure 6 \n ), at least with respect to the 33 P in the leaves. In this case, the fungus probably derives more benefit from symbiosis with the mycorrhiza-dependent P. bisulcatum in plant-fungal association in the mono system. In contrast, in the monoculture of P.maximum , the fungus, as an obligate biotroph, only had access to a low C rewarding (low quality) host and had no choice but providing nutrients to such a host, even when the quality of the host further decreased by shading. In the latter case, it is probably the host plant that benefits more from the symbiosis. Thus, in agreement with Fellbaum et al. (2014) , we demonstrated that in the absence of choice for the fungus, the cost to benefit ratio of the mycorrhizal symbiosis shifts in favor of the less photosynthate-rewarding hosts ( \n Figure 6 \n ). Figure 6 Summarization of changes in shoot and root dry biomass, 33 P and, 15 N resource uptake into above- and belowground plant tissues in mycorrhizal and non-mycorrhizal P.bisulcatum (Bis) and P.maximum (Max) as well as 13 C allocation to AM fungi and abundance of mycorrhizal fungi in roots under full light and full shade regimes, from mono- to mixed-systems. “M” and “NM” refer to mycorrhizal and non-mycorrhizal status, respectively. Arrows indicate changes from one condition to another conditions. The values calculated for the transition from mono- to mixed systems are based on a theoretical model prediction, using the mono system as a baseline, and calculated response values from mono- to mixed-systems. AM fungal abundance in the roots is based on qPCR quantification of the AM fungal DNA in roots, whereas the 13 C allocation to AM fungi is based on the fatty acid analyses in soil. Plant dependence on mycorrhiza can also be altered by changes in environmental conditions such as light intensity ( Konvalinková and Jansa, 2016 ). Experimental shading has been shown to significantly affect C allocation to mycorrhiza and C-P interactions by altering the plant photosynthetic rates ( Kaschuk et al., 2009 ; Konvalinková et al., 2015 ; Zheng et al., 2015 ; Lang et al., 2021 ). Our results showed that in mycorrhizal P.maximum , 33 P partitioning between shoot and roots significantly increased, and total 33 P significantly decreased by shading in the mono system. Conversely, no significant effect of shading on 33 P partitioning and total 33 P uptake was observed in mycorrhizal P.bisulcatum . These findings suggest that the more mycorrhiza-dependent P.bisulcatum is likely to cope better with stress caused by changes in incoming light intensity than the less mycorrhiza-dependent (and more resources-demanding C4 species) P.maximum . In other words, mycorrhizal association could possibly attenuate the response of P.bisulcatum to the shading stress. In addition, previous studies have shown that photosynthate investment in mycorrhiza decreases as light intensity decreases, resulting in lower mycorrhizal root colonization below a certain light intensity/duration threshold ( Gorzelak et al., 2015 ; Konvalinková et al., 2015 ). In fact, to maintain the resource economics, plants are thought to invest more biomass in aboveground structures and less biomass in mycorrhizal network when light intensity is low ( Johnson, 2010 ). Accordingly, Konvalinková and Jansa (2016) have shown that plants do not deliver more C to mycorrhiza under intensive shading conditions extending over several weeks, compared to ample light conditions. However, we did not detect any changes in 13 C allocation to mycorrhizal networks in the studied plant species along the shading gradient in the mono system. This could in fact be due to the short duration of shading. We applied a short-term shading regime here (slightly less than one week) that may not have been sufficient to observe changes in mycorrhizal colonization and/or mycorrhizal C allocation. Similarly, Konvalinková et al. (2015) found no significant change in mycorrhizal colonization of roots under short-term shading (6 days), but colonization of roots by the AM fungi was significantly reduced after long-term shading (38 days) ( Konvalinková et al., 2015 ). Therefore, not only the intensity but also the duration of light shortage could influence the exchange of resources between plants and their associated mycorrhizal networks. 4.2 Mixed system Based on the results of plant-mycorrhizal interactions in mono systems, we could speculate that the outcome of plant-plant interactions in mixed system is likely to be antagonistic, due to contrasting rates of resources possibly provided by the mycorrhizal networks, as well as due to direct root competition for (limited) soil resources that may take place between different plant species. The two plants grown in a mixture showed different biomass production compared to those grown in monocultures, with the dry weight of the generally more productive but less mycorrhiza-dependent P.maximum reduced in the mixed culture compared to the monoculture ( \n Figure 6 \n ). In contrast, biomass production was promoted in the more mycorrhiza-dependent P.bisulcatum in mixed culture, suggesting that the presence of AM fungus could significantly alter plant productivity in favor of the more C rewarding host. In addition, in M pots under no shading in the mixed system compared with the respective mono systems, 33 P and 15 N increased by 73% and 89%, respectively, in the shoots of P.bisulcatum , whereas they both decreased by 28% in shoots of P.maximum . Moreover, 33 P increased by 38% in the roots of the fully lit pots with mycorrhiza ( \n Figure 6 \n ). Despite the increase in 13 C allocation to mycorrhiza (+103%) by P.bisulcatum , mycorrhizal abundance in roots and soil decreased by 67% and 36%, respectively, in the unshaded pots. This can be explained by the observation that C allocation to mycorrhiza by P.maximum decreased by as much as 76% in in the mixed system as compared to the mono system. It appears that more mycotrophic P.bisulcatum acts as a better competitor for the uptake of recent nutrients than P.maximum , and provides more C resources for mycorrhiza, if inoculated with the AM fungus and grown together, under no shading. P.bisulcatum provided even more C to mycorrhiza (+190% increase in 13 C allocation to AM fungi) when two AM-inoculated plants were shaded in the mixed system compared to the mono system, while P.maximum maintained the same strategy of decreasing C allocation to the mycorrhiza (-80%). P.maximum allocated more photosynthates to AM fungi than P.bisulcatum only when exposed to full light, whereas its competitor, P.bisulcatum , was suppressed by shading at the same time. Similar to our results, in a compartmented microcosm and using AM fungi-specific fatty acid C16:1ω5, Řezáčová et al. (2018b) found that P.bisulcatum preferentially fed the CMNs (consisting of five mycorrhizal fungal genera), and this contrasted to P.maximum , even at a high temperature, when these two plants were grown together in a mixed system, and where C 4 photosynthesis type ( P.maximum ) would be predicted to be a more beneficial trait for plant growth/C reserve accumulation ( Edwards and Still, 2008 ). Such differences in C inputs into the AM fungi by different plant species has also been reported in non-congeneric plant species growing in mixtures. For instance, Walder et al. (2012) found that in microcosms where two plant species were connected by the shared CMNs, the plant that invested less carbon in CMNs received relatively greater share of the nutritional (P and N) benefits from the CMNs under unshaded conditions. Thus, coexisting plant species may not benefit equally from mycorrhiza in terms of nutrient acquisition and biomass production ( Walder et al., 2012 ; Jakobsen and Hammer, 2015 ; Wang et al., 2016 ; Řezáčová et al., 2018a ). In fact, in the mixed systems, when plants competed for available resources, mycorrhizal fungi gain a “bargaining power” and are likely to transfer more nutrients to those plants that provide more C to the them ( Bücking et al., 2016 ). This could in consequence amplify inequalities among plant species in a community by providing additional nutrients and promoting the competitiveness of more C rewarding hosts/individuals ( van der Heijden and Horton, 2009 ; Booth and Hoeksema, 2010 ; Merrild et al., 2013 ; Weremijewicz and Janos, 2013 ) ( \n Figure 7 \n ). On the other hand, this could potentially promote the diversity in plant communities by suppressing dominants and promoting community evenness (see below). It should be noted, however, that the imbalanced outcome of competition between mycorrhizal plant species is not always directly related to C investments into mycorrhiza. Contribution of context-dependency of plant-mycorrhizal interactions such as nutrient availability, plant species, and fungal identity, should still be quantified ( van der Heijden and Horton, 2009 ; Gorzelak et al., 2015 ; Montesinos-Navarro et al., 2019 ). Further research is also needed to better understand the underlying the exact mechanisms of resource exchange in plant-mycorrhizal interactions, particularly when more co-occurring plant individuals are connected to the shared mycorrhizal networks. Figure 7 A conceptual model of interactions between a single AM fungal taxon and coexisting plants (A, B) , relevant to experimental results presented here. Benefits from a plant perspective were defined in terms of biomass and nutrient accumulation in aboveground plant tissues. In Scenario 1, where two individuals of highly-mycotrophic plant species A are connected via mycorrhizal networks, fungus receives relatively more C resources from plants and the abundance of mycorrhiza in the root/soil thus increases as compared to scenario 2. However, in return, the mycorrhiza may not equally benefit the plants by providing more N and P, compensating for the C investments. Thus, we assume that in this case the mycorrhiza benefits more than the plant from the symbiotic association. In scenario 2, where two individuals of a less-mycotrophic plant species B are interconnected via a mycorrhizal network (or colonized by two overlapping networks), the fungus receives less C resources from the plants compared to scenario 1, but as an obligate biotroph, has no choice but to provide nutrients to the plants. Thus, we assume that the plant is actually “in control” of the symbiosis in this case. In scenario 3, where coexisting plant species A and B are associated with the shared CMNs, more mycorrhiza-dependent (= more mycotrophic) plant A benefits more from the association with mycorrhiza compared to less mycorrhiza-dependent plant (B) . Plant (A) also receives more recent nutrients and shows higher competitiveness. Thus, AM fungi amplify inequalities among different individuals of plant species A and B by preferential rewarding of the different host plants. It has also been suggested that resource sharing through mycorrhizal networks may act as a fitness balancing mechanism that minimizes fitness differences among plant species, leading to improving plant coexistence ( Bever et al., 2010 ; Montesinos-Navarro et al., 2012 ; Bücking et al., 2016 ). This could be the case under natural conditions when multiple plant species and mycorrhizal fungi interact simultaneously in a complex network of many interactions and the symbiotic partners are hardly ever dependent on a single partner, particularly given the low host specificity in AM symbiosis ( van der Heijden et al., 1998 ; Bücking et al., 2016 ). However, in our experiment, a single fungal taxon formed CMNs, which may explain why no positive or even neutral plant-plant interactions were observed in our mixed system. Admittedly, we did not measure fitness here, but biomass production could serve as a crude proxy for plant fitness ( Younginger et al., 2017 ). Different fungal taxa demonstrably differ in providing resources to different host plants and also have different effects on plant responses to biotic and abiotic stresses ( Klironomos, 2000 ; Montesinos-Navarro et al., 2019 ). In addition, conspecific individuals tend to have greater niche overlap than heterospecific plant individuals, which could lead to greater competition for available resources and suppressing fitness/growth. Overall, although simplified and artificial experimental setups with low complexity may overlook environmental heterogeneity and other potentially contributing factors, such studies as presented here, using microcosms with plants interconnected or not by the same AM fungal network, are of particular interest because they could contribute to a better understanding of the processes occurring in mycorrhizal networks, especially when multi-isotope labeling is employed ( van der Heijden and Horton, 2009 )."
} | 8,036 |
37679429 | PMC10579422 | pmc | 254 | {
"abstract": "Propionate is a key intermediate in anaerobic digestion processes and often accumulates in association with perturbations, such as elevated levels of ammonia. Under such conditions, syntrophic ammonia-tolerant microorganisms play a key role in propionate degradation. Despite their importance, little is known about these syntrophic microorganisms and their cross-species interactions. Here, we present metagenomes and metatranscriptomic data for novel thermophilic and ammonia-tolerant syntrophic bacteria and the partner methanogens enriched in propionate-fed reactors. A metagenome for a novel bacterium for which we propose the provisional name ‘ Candidatus Thermosyntrophopropionicum ammoniitolerans’ was recovered, together with mapping of its highly expressed methylmalonyl-CoA pathway for syntrophic propionate degradation. Acetate was degraded by a novel thermophilic syntrophic acetate-oxidising candidate bacterium. Electron removal associated with syntrophic propionate and acetate oxidation was mediated by the hydrogen/formate-utilising methanogens Methanoculleus sp. and Methanothermobacter sp., with the latter observed to be critical for efficient propionate degradation. Similar dependence on Methanothermobacter was not seen for acetate degradation. Expression-based analyses indicated use of both H 2 and formate for electron transfer, including cross-species reciprocation with sulphuric compounds and microbial nanotube-mediated interspecies interactions. Batch cultivation demonstrated degradation rates of up to 0.16 g propionate L −1 day −1 at hydrogen partial pressure 4–30 Pa and available energy was around −20 mol −1 propionate. These observations outline the multiple syntrophic interactions required for propionate oxidation and represent a first step in increasing knowledge of acid accumulation in high-ammonia biogas production systems.",
"conclusion": "Conclusions Use of a long-term enrichment approach to increase the abundance of an ammonia-tolerant syntrophic propionate-degrading community made it possible to identify key species and their metabolic activities, and to distinguish activities potentially related to the syntrophic lifestyle. Two novel ammonia-tolerant and thermophilic syntrophic species were identified, and we propose the name ‘ Candidatus Thermosyntrophopropionicum ammoniitolerans’ for the SPOB. Batch cultivation, 16S rRNA gene analyses (sequencing and expression) and qPCR analysis indicated that Methanothermobacter could be crucial for syntrophic methanogenesis from propionate. Similar dependence for acetate degradation was not observed, indicating that the SAOB cooperated well with the Methanoculleus sp. also present in the syntrophic communities. Transcriptome data revealed activity related to sulphur metabolism, intercellular contact and molecular exchange by pili/flagellar appendages and nanotubes by the candidate SPOB, which can be crucial for efficient interdependent metabolism in a thermodynamically unfavourable environment. An additional bacterial species in the syntrophic community displayed activity for the reductive glycine pathway, but the wide substrate span of related bacterium and decreased abundance during cultivation without yeast extract suggest that this species is not directly involved in acid degradation. Thus, caution is needed when claiming that the reductive glycine pathway can be operated in the oxidative direction by SAOB to oxidise acetate. A deeper understanding of important syntrophic players and their mutualistic interactions under high-ammonia conditions is key for optimal design of anaerobic processes degrading protein-rich biomass. Future work should focus on identifying and characterising the functional interactions of ammonia-tolerant, thermophilic VFA-oxidising and methanogenic syntrophic communities as a model, which would be a milestone in metabolic modelling and systems biological approaches to anaerobic digester systems. With enhanced understanding of syntrophic synergy and coupled metabolic networks, industrial reactor operation can be steered to obtain higher efficiency and productivity of the methanogenic process.",
"introduction": "Introduction Through anaerobic digestion, waste is efficiently converted to a combustible gas, comprising methane (CH 4 ) and carbon dioxide (CO 2 ). This is one of the most sustainable options for renewable energy production when accounting for the additional benefits, such as abated greenhouse gas emissions, efficient waste management, recovery of nutrients and substitution of synthetic and mineral fertiliser when using the residue as bio-fertiliser [ 1 ]. The anaerobic degradation process involves a series of microbial degradation steps engaging different anaerobic microorganisms, often operating at near thermodynamic equilibrium [ 2 , 3 ]. In the process, complex organic materials are initially hydrolysed to sugars, amino acids and long-chain fatty acids that are further fermented to volatile fatty acids (VFA), alcohols and smaller amounts of hydrogen (H 2 ). In the following anaerobic oxidation step, VFA longer than acetate, such as propionate and butyrate, are degraded to acetate, H 2 and CO 2 , which in a terminal step are converted to CH 4 and CO 2 by methanogenic archaea. Although the anaerobic digestion process is generally operated in continuous/semi-continuous mode, disturbances can occur during operation, triggered, for instance, by a change in feeding composition/rate, temperature fluctuations, trace element deficiency or ammonia toxicity [ 4 ]. A pervasive consequence, particularly with ammonia-induced perturbations if the anaerobic digester is fed protein-rich substrates, is accumulation of propionate and acetate, which exposes the process disturbance [ 5 , 6 ]. At high ammonia levels, acetate accumulation is often caused by inactivation of acetate-cleaving methanogens, opening up an opportunity for development of an alternative acetate-degrading community involving a synergy between syntrophic acetate oxidisers (SAOB) and hydrogenotrophic methanogens (HM) [ 7 ]. The exact cause of propionate accumulation at high ammonia concentrations is poorly understood, but may relate to suppression of ammonia-sensitive propionate-degrading communities [ 6 , 8 , 9 ] and a temporal lapse in establishment of ammonia-tolerant propionate degraders [ 10 ]. Propionate is degraded through interactions between syntrophic propionate-oxidising bacteria (SPOB) and hydrogen- and/or formate-utilising HM. Acetate is subsequently degraded by aceticlastic methanogens in low ammonia conditions [ 11 , 12 ] or by SAOB and hydrogen- and/or formate-utilising HM in high-ammonia conditions [ 13 ]. The SPOB characterised to date are from the families Peptococcaceae and Syntrophobacteraceae , which all degrade propionate using the methylmalonyl-CoA pathway or the dismutating pathway (only Smithella ) [ 10 ]. SPOB candidates include members of the phylum Cloacimonadota found in both mesophilic and thermophilic biogas processes [ 14 , 15 ], ‘ Candidatus Propionivorax syntrophicum’ discovered in a mesophilic wastewater treatment plant [ 16 ] and the only known ammonia-tolerant SPOB, ‘ Candidatus Syntrophopropionicum ammoniitolerans’ identified from a mesophilic biogas process [ 13 ]. The majority of these SPOB originate from biogas reactors, clearly demonstrating that syntrophic propionate oxidation (SPO) is a distinct feature of the anaerobic digestion process. However, although high-ammonia anaerobic digestion has been widely studied from a process perspective, there are indications that several acid-degrading microorganisms with key roles in that process have not been identified, isolated or characterised [ 10 , 13 ]. Consequently, the reciprocal communal interactions within and between such communities, i.e. between SAOB, SPOB and HM, are currently underexplored. In particular, current understanding of SPO in thermophilic and high-ammonia biogas processes and the cross-species interactions enabling stepwise conversion of propionate to methane is limited. In the present study, a thermophilic SPOB community in high-ammonia conditions was enriched and its molecular exchange and interaction network with SAOB and HM were analysed. The microbial communities were explored through both 16S rRNA gene amplicon sequencing and whole metagenome sequencing, and their physiological activity during propionate and acetate degradation was characterised using metatranscriptomics. Chemical monitoring of batch trials of propionate- and acetate-supplemented cultures was performed to study the effects of intermediate product formation and consumption on propionate- and acetate-degradation kinetics.",
"discussion": "Results and discussion Reactor performance revealed temporal changes in propionate degradation rate The four propionate- and acetate-fed reactors used in the study produced biogas with an average methane content of 62–70% (Table S2 ). The pH was 8.1–8.3, resulting in an ammonia-nitrogen level of 0.7–0.9 g NH 3 L -1 . This free ammonia level is well above the threshold at which many microorganisms are inhibited, frequently causing reductions in overall methane production and accumulation of VFAs even in ammonia-adapted biogas processes [ 54 ]. In agreement with previous findings for thermophilic acetate-fed reactors [ 19 ], acetate content remained stable at around 0.7–0.9 g L −1 in the acetate-fed reactors RA1 and RA2. However, VFA degradation in the propionate-fed reactors was less stable. Propionate fluctuations (0.8–3.9 g L −1 ) were especially pronounced in reactor RP2 during the latter stages of operation (days 200–320, Fig. S1 ). As a result of changes in propionate level, acetate level in RP2 fluctuated from below detection to 3.5 g L −1 throughout the operating period. Other VFAs analysed were not detected above the detection limit of 0.2 g L −1 in the reactors. Comparing thermophilic reactor performance with that in previous mesophilic enrichment study [ 13 ], revealed less efficient propionate removal in thermophilic than in the mesophilic propionate reactors (Table S4 ). One reason for this may be the somewhat higher pH and associated higher ammonia levels in the thermophilic than in the mesophilic reactors. Batch assays and thermodynamics For reasons unclear, it was not possible to initiate propionate-degrading activity in batch assays by preparing culture media and inoculating with 5–20% (v/v) of culture from the continuously fed reactors, a procedure routinely applied with success for mesophilic propionate-degrading cultures (unpublished data) originating from analogous mesophilic reactor experiment [ 13 ]. Propionate and acetate-degradation rates were therefore analysed by adding these VFAs to batches consisting of undiluted cultures from the continuously fed reactors. The degradation rates of propionate and acetate, H 2 level and methane production rate were similar in the duplicate batches originating from the same reactor, but the degradation rate of propionate differed between the RP1 and RP2 communities (Fig. S2 ). The communities originating from RP1 (batches B01-B02) had the highest propionate-degradation rate and degraded the added propionate within 25 days. The rate of propionate degradation in the RP2 batches was considerably lower, and up to 100 days were required for complete propionate degradation. In batches B01-B02 propionate was consumed and methane formed according to the expected stoichiometry of 1.75 mol methane per mol propionate, whereas in batches B03-B04 the methane yield was slightly higher than expected (Table S3 ). Hydrogen levels were relatively similar in the propionate-fed batch cultures, foremost ranging between 3.5 and 12 Pa (Table S5 ), and were in line with levels previously reported for thermophilic acetate-oxidising cultures [ 55 , 56 ], but somewhat higher than values reported for mesophilic propionate-oxidising communities [ 20 , 57 ]. The underlying cause of the slower propionate degradation of the RP2 community compared with the RP1 community is unknown. However, the relatively similar acetate-degradation rates and H 2 partial pressures in all communities (Table S5 ) indicate that the slower propionate degradation in RP2 was not because of low activity of the acetate-degrading community or slow removal of H 2 in this reactor. Thus, the SPOB in RP2 were most likely inhibited by another factor. Furthermore, the relatively rapid propionate degradation in the RP1 batches (B01-B02) resulted in formation of 2–3 g acetate L −1 before the SAOB initiated acetate degradation (Fig. S2 ). The accumulated acetate and the calculated rate of propionate and acetate degradation in the present study also indicated that the SPOB community was able to degrade propionate at a rate that exceeded the acetate-degrading capacity of SAOB (Fig. S2 , Table S5 ). These results support previous observation of a peak in acetate concentration due to rapid propionate degradation in thermophilic conditions [ 58 ]. Similar elevations in acetate levels following propionate degradation have been observed in high-ammonia mesophilic reactors after VFA pulsing and in dairy manure digesters [ 59 , 60 ]. However, in several studies of mesophilic and low-ammonia processes with aceticlastic methanogens as the main acetate degrader, the acetate concentration has remained at low levels despite degradation of 1–4 g propionate L −1 [ 16 , 61 – 63 ]. One reason for the disparity between high- and low-ammonia conditions is possibly that acetate above a certain concentration may be required before initiation of acetate degradation by SAOB. Alternatively, less efficient HM activity at higher ammonia levels [ 64 ] increases the H 2 and/or formate levels, impeding the activity of SPOB and SAOB. This emphasises the need to support the activity of both syntrophic bacteria and HM in order to obtain a stable process [ 13 , 57 , 63 , 65 – 67 ]. In the continuously fed reactors in the present study, the relatively constant flow of acetate formed from the propionate was manageable by the SAOB community, in maintaining acetate level at <2.5 g L −1 (Fig. S1 ). This is important, since acetate at >4.8 g L −1 (80 mM) has been shown to severely restrict propionate oxidation [ 68 ]. The ∆G values calculated from measured parameters from the batch experiment varied somewhat between the different species, but for the SPOB the ∆G value for conversion of propionate to acetate and H 2 fluctuated around −20 mol −1 propionate. For acetate oxidation to CO 2 and H 2 by the SAOB, the ΔG value was −10 to −30 kJ mol −1 acetate. For HM, the ΔG value was similar to that of propionate in both series of batches, although during the later stages of propionate degradation of the batches with slow propionate degradation (B03-B04) the ΔG values for SPOB were more favourable than those for HM (Fig. S3A ). In the acetate fed batches, ΔG was consistently more favourable for HM than for SAOB. This agrees well with results previously obtained for mesophilic SAOB cultures, where the HM also obtained more energy than the SAOB (−20 and −10 kJ mol −1 , respectively) [ 20 ]. Due to the dual syntrophy in the propionate-fed systems, the outcome changed when energy distribution was evaluated per mole of propionate mineralised to CH 4 (Table S6 ). HM then gained most of the energy, as 1.75 moles of methane were generated for each mole of propionate mineralised. The dual syntrophy underpinning propionate degradation in these systems makes H 2 one of the central intermediates, as it can be produced by both SAOB and SPOB, and low H 2 levels are beneficial for both. For acetate, the interdependency is more complicated, i.e. the SPOB benefit from low levels, while the SAOB benefit from high levels. In the batch experiment, the average hydrogen levels were slightly lower in the propionate-fed batches than in the acetate-fed batches, irrespective of the acetate level (Figs. S3B, C ). To gain further thermodynamic insights into the microbe interplay in the dual syntrophy and to unravel why and when the hydrogen scavenger operates at lower hydrogen levels in syntrophy with SPOB than with SAOB, future studies should monitor growth of the syntrophs and the methanogen under a set of constant H 2 levels. Microbial community structure (16S rRNA gene) in enrichment reactors and batch assays Microbial community structure based on 16S rRNA gene amplicon sequencing most strikingly revealed enrichment of the family Pelotomaculaceae only in the propionate-fed reactors (RP1, RP2; 2–48%) and not in the acetate-fed reactors (RA1, RA2) (Fig. 1 ). Pelotomaculaceae harbours many known [ 10 ] and proposed SPOB [ 13 , 16 ]. Major families observed throughout the experimental period in both the propionate- and acetate-fed reactors were Acetomicrobiaceae (3–13%), Campylobacteraceae (20–50%), Ch115 (5–20%) and Thermacetogeniaceae (2–34%). (Fig. 1 ). These five families were equally dominant in all batch assays except for family Pelotomaculaceae , which was specifically higher in relative abundance in the assays inoculated from the propionate-fed continuous reactors (B01-B08, Fig. S4 ). The batch assays from the propionate-fed reactors prepared with acetate as growth substrate (B05-B08) showed declining relative abundance (<5%) of Pelotomaculaceae , whereas its relative abundance was higher (3–50%) in batches fed propionate (B01-B04, Fig. S4 ). The enrichment of Pelotomaculaceae in propionate-fed continuous reactors (RP1, RP2) and its consistent presence at high relative abundances in propionate-based batch assays (B01-B04) indicate that members of this family were involved in SPO under the high-ammonia thermophilic conditions in the present study. The 16S rRNA gene sequencing indicated presence of two methanogenic species belonging to the genera Methanoculleus and Methanothermobacter in both acetate- and propionate-fed communities (Fig. 1 , S4 ). These methanogenic genera have previously been suggested to be partners of a thermophilic Pelotomaculum sp. growing in low-ammonia conditions [ 69 ]. The absence of aceticlastic methanogens in the propionate-degrading community in the present study demonstrates the importance of the SAOB for acetate removal in high-ammonia conditions. Further detailed information of the sequencing result is given in Supplementary note 5 . Fig. 1 Microbial community structure resolving the exclusive presence of Pelotomaculaceae in the propionate-fed continuous reactors. Bubble plot showing percentage relative abundance (>2%) of microbial communities at family level in the acetate-fed (RA1, RA2) and propionate-fed (RP1, RP2) enrichment reactors. Metagenomic binning and metatranscriptomics-based functional analysis Nine good-quality MAGs were obtained and based on taxonomic annotation and genomic content indicating putative involvement in syntrophic interactions, four of these MAGs were chosen for detailed analyses (Table S7 ). For instance, the MAGs affiliating to Campylobacteraceae and Ch115, highly abundant in the enrichment cultures, were shown to lack several of the crucial genes required for SAO/SPO-activity and were not included in further analyses. The four MAGs of interest for syntrophic acid degradation and their functional activities and their pathways are described in detail below. Novel SPOB of family Pelotomaculaceae ‘ Candidatus Thermosyntrophopropionicum ammoniitolerans’ A high-quality MAG (MAG4, Table S7 ) classified to the family Pelotomaculaceae [phylum Bacillota _B, class Desulfotomaculia , order Desulfotomaculales ] was recovered in metagenomic sequencing of samples from the propionate-fed reactors. No MAG with similar classification was recovered from the acetate-fed reactors. In a phylogenetic assessment based on whole-genome sequencing, MAG4 clustered together with ‘ Ca . Propionivorax syntrophicum’, which further sub-clustered under SPOBs in the family Pelotomaculaceae (Fig. S5 ). Moreover, in an assessment based on 16S rRNA gene retrieval from MAGs, MAG4 showed a relationship to Pelotomaculum spp. and ‘ Ca . Syntrophopropionicum ammoniitolerans’ (Fig. S6 ). Comparison of MAG4 with available genomes of Pelotomaculaceae spp. and other known or proposed SPOBs revealed similarities below recommended cut-offs for delineating a new species (i.e. 70% dDDH, 95% ANI, 60% coverage) [ 70 , 71 ]. MAG4 had highest similarities with ‘ Ca . Propionivorax syntrophicum’ (dDDH of 42%, ANI of 90% and AAI of 89%). However, the genome assembly of ‘ Ca . Propionivorax syntrophicum’ is of low quality (74.7% completeness), lack the 16S rRNA gene sequence and is considerably smaller (2.0 Mbp) than MAG4 (3.2 Mbp), which obstruct a complete and accurate comparison. The taxonomic analysis against other SPOB revealed highest dDDH with Desulfofundulus thermobenzoicus (24%), AAI with P. thermopropionicum (71%) and ANI with Pelotomaculum schinkii and Pelotomaculum thermopropionicum (74%) (<45% coverage) (Fig. S7 ). These results indicate that this bacterium will form a novel genus when isolated and characterised and we propose the provisional name ‘ Candidatus Thermosyntrophopropionicum ammoniitolerans’. Methylmalonyl-CoA pathway MAG4 harboured and expressed a complete set of genes required for propionate degradation through the methylmalonyl-CoA (MMC) pathway (Figs. 2 , 3 , S8 ), which strongly suggest that this bacterium can perform syntrophic propionate oxidation in high-ammonia, thermophilic biogas systems. MAG4 expressed CoA-transferase and carboxyltransferase. This indicates that, as in the thermophilic and mesophilic SPOB P. thermopropionicum and P. schinkii [ 72 , 73 ], MAG4 coupled the two first endergonic steps, propionate activation (step P1 in Fig. 2 ) and propionyl-CoA carboxylation (P2), with the downstream and exergonic steps forming acetate (P11) and pyruvate (P9), respectively. Other genes encoding enzymes involved in the MMC pathway expressed by MAG4 include methylmalonyl-CoA epimerase, methylmalonyl-CoA mutase and succinate-CoA synthetase (P3-P5). To catalyse the energetically most unfavourable step in the MMC pathway, the oxidation of succinate to fumarate (P6), MAG4 expressed a gene encoding the membrane-bound succinate dehydrogenase/fumarate reductase complex, requiring reducing power via reverse electron transport. Genes encoding fumarate hydratase catalyses the conversion of fumarate to malate (P7) and malate dehydrogenase catalyses the conversion of malate into oxaloacetate (P8) were expressed. MAG4 also contained one Fe-S-containing hydrolyase which was annotated as fumarase/fumarate hydratase in P. thermopropionicum (BAF59538.1). The gene encoding pyruvate carboxylase in step P9 (Fig. 2 ), i . e. conversion of oxaloacetate to pyruvate, was not found in MAG4. However, the gene encoding methylmalonyl-CoA carboxyltransferase for step P2 (Figs. 2 , 3 ) was expressed and this enzyme has been found to catalyse the conversion of oxaloacetate into pyruvate (KEGG reaction: R00930). For conversion of pyruvate to acetyl-CoA (P10), MAG4 encoded pyruvate:ferredoxin flavodoxin oxidoreductase. Fig. 2 Metabolic reconstruction of syntrophic propionate and acetate oxidation and the interspecies hydrogen/formate transfer with hydrogenotrophic methanogens employed under thermophilic and high ammonia conditions. Visualisation of the molecular exchange anchored interplay and metabolic pathways employed by the multiple syntrophic bacteria and their methanogenic partner during syntrophic propionate degradation under thermophilic and high-ammonia conditions. The figure highlighted the cooperation of syntrophic propionate oxidising bacteria (SPOB, MAG4 ) via acetate assimilation by syntrophic acetate oxidising bacteria (SAOB, MAG9 ). These SPOB and SAOB further obligately establish formate or hydrogen pivoted syntrophic network to circumvent the reducing potential which is used by hydrogenotrophic methanogen ( MAG1 ) to reduce carbon dioxide and generate methane. Fig. 3 Gene expression profile of the methylmalonyl CoA pathway for propionate oxidation by the SPOB candidate. Metatranscriptomics expression profile of the methylmalonyl CoA (MMC) pathway of propionate metabolism (based on transcripts per million (TPM) counts) in propionate versus acetate batch assay (B01, B03 and B09) for the novel syntrophic propionate-oxidising bacteria (SPOB) ` Candidatus Thermosyntrophopropionicum ammoniitolerans´ ( MAG4 ). The numerical values with the enzyme name denote the step in the MMC pathway. The values on heatmap represented are the aggregated TPM counts of all copies and subunits for respective gene present and expressed in the metagenome assembled genome. In MAG4, most of the genes coding for the MMC pathway enzymes were found to be clustered together (except the genes for steps P1 and P6). The gene for propionate-CoA transferase (PCT) was not found in MAG4 (and is also absent in P. thermopropionicum ). Instead, MAG4 expressed genes for two other CoA-transferases (acyl/acetate transferase, glutaconate-CoA transferase) that are homologous to PCT and have also been suggested to activate propionate [ 72 , 74 , 75 ]. These CoA-transferases genes were clustered in an operon fashion and highly expressed compared with other flanking genes in MAG4. MAG4 also expressed MMC pathway-associated genes, viz . formate and sodium/solute (propionate) transporters, formate dehydrogenase, methylmalonyl-CoA decarboxylase etc . (Figs. 2 , 3 ). A gene encoding an uncharacterised protein likely involved in propionate catabolism (32% similarity to an uncharacterised protein (BAF60599.1) found in the P. thermopropionicum genome (AP009389.1)) was expressed by MAG4 (Fig. 3 ). This gene showed 100% similarity and query coverage (Blastx) to MmgE/PrpD family protein (NLW37044.1) belonging to a Peptococcaceae bacterium MAG (JAAYEO000000000.1). The PrpD family is involved in propionate oxidation to pyruvate in E. coli [ 76 ], but this protein has still not been annotated or characterised in other anaerobic bacteria. The expression of this gene by MAG4 indicate that this protein might be involved in propionate degradation by an as yet unknown mechanism. Hydrogen/formate production and energy conservation systems In SPO, hydrogenases or formate dehydrogenases catalyse electron transfer from NADH or reduced ferredoxin (Fd, generated from substrate oxidation) to the final electron acceptors H + and CO 2 [ 12 , 77 ]. MAG4 expressed genes for both hydrogenases and formate dehydrogenases (Fig. 3 ). More specifically, expression of genes for cytoplasmic [FeFe] electron-bifurcating- (HndAC) ([FeFe] group A3), membrane [Fe]-bound [NiFe] and iron-only hydrogenases ([FeFe] group A4) was revealed (Fig. S9 ). The latter has been shown to couple the endergonic formation of H 2 from NADH to its exergonic formation from Fd red [ 78 ]. In MAG4, the gene encoding Hnd was found to be flanked by expressed genes for formate dehydrogenase (FDH) and a formate transporter (FdhC) (Fig. 3 , S8 ). It has been speculated that two FDH are needed for syntrophic growth on propionate, one for fixing CO 2 by the reductive Wood-Ljungdahl pathway (the membrane-bound FDH1) and one for removal of reducing equivalents as format (the cytoplasmic FDH2) [ 79 , 80 ]. For P. thermopropionicum , four types of FDH have been found [ 73 ] and these can be differentially and independently up- or down-regulated [ 81 ]. MAG4 expressed four FDH types (Fig. 3 , S8 ), indicating the possibility that in MAG4, FDH could be utilising reducing equivalents (together with electron transfer/bifurcating flavoproteins and electron transport Rnf complex) to form H 2 [ 82 ]. Further, formate transporter (transmembrane FocA) could be assisting in mediation of H 2 /formate-dependent electron sharing or electron bifurcation [ 83 ] between MAG4 (as SPOB) (Fig. 3 , S9 ) and the HM, as also reported previously [ 16 ]. Genome and transcriptomic analysis of candidate SAOB For SAO, a distinct candidate (MAG9) was identified by genomic and transcriptomic analysis in both propionate- and acetate-degrading reactor communities. MAG9 was classified to genus DTU068 (95% dDDH similarity with place holder species sp001513545 in the phylum Firmicutes , class Moorellia , family Thermacetogeniaceae , Fig. S10 ). The taxonomic placement of MAG9 and the transcriptomic data in the propionate- and acetate-fed batch assays strongly indicate that this species represents a novel thermophilic SAOB, related to the mesophilic SAOB Syntrophaceticus schinkii (Figs. S11 , S12 ). However, the genome sequence of MAG9 had high contamination (~15%) thus MAG quality was not sufficiently high for proposal of a provisional name for this species. MAG9 harboured and expressed a complete set of genes for the Wood-Ljungdahl pathway (WLP) in both propionate and acetate cultures (Fig. 4 ). The genome revealed a cluster of several WLP genes (steps A3-A6, A12 and A9 in Fig. 2 ) but for acetate activation (A1-A2) the genes were located separately and the transcriptome data indicated that MAG9 activates acetate through the ATP-consuming acetate kinase. Acetate can potentially also be activated through an ATP-independent aldehyde ferredoxin oxidoreductase followed by oxidation of acetaldehyde to acetyl-CoA, as postulated to be used by Thermacetogenium phaeum to balance the overall ATP budget [ 84 ]. Even though MAG9 encoded aldehyde ferredoxin oxidoreductase, the low transcript level of the encoding gene compared with the gene for acetate kinase indicates that, as seen in S. schinkii [ 85 ] MAG9 consumes ATP in this first step and forms acetyl-CoA using phosphate acetyltransferase. For the carbonyl branch, CO-methylating acetyl-CoA synthetase and carbon monoxide dehydrogenase were expressed (A4, Figs. 2 , 4 ), whereas expression of corrinoid methyltransferases (A3), methylene tetrahydrofolate (THF) reductase (A5), methylene THF dehydrogenase/cyclohydrolase (A6, A7), formyl THF synthetase (A8) and format dehydrogenase (A9) indicated their importance in operation of the methyl branch (Figs. 2 , 4 ). Fig. 4 Gene expression profile of the Wood-Ljungdahl pathway by the SAOB candidate. Metatranscriptomics expression profile of the Wood-Ljungdahl pathway and other genes of relevance for acetate metabolism (based on transcripts per million (TPM) counts) in propionate versus acetate batch assay (B01, B03 and B09) for the novel syntrophic acetate-oxidising bacteria (SAOB) MAG9 and for MAG5 belonging to the genus Acetomicrobium . The numerical values with the enzyme name denote the step in the pathway. The values on heatmap represented are the aggregated TPM counts of all copies and subunits for respective gene present and expressed in the metagenome assembled genome. In the direction of acetate oxidation, the methylene THF reductase (A5) releases electrons at a redox potential too low to be used directly for NAD + reduction [ 86 ]. For T. phaeum , this has been proposed to be solved by electron transfer to a methyl-viologen-reducing hydrogenase subunit D (MvrD), followed by heterodisulphide reductase (HdrABC) and further to a quinone, which in turn is re-oxidised by formate dehydrogenase [ 84 ]. The gene expression seen for MvhD, HdrABC and NAD-quinone oxidoreductase, the four-iron-four-sulphur (4Fe-4S) cluster and 4Fe-4S ferredoxin by MAG9 indicate that a similar path is followed by this ammonia-tolerant SAOB (Fig. S12 ). Furthermore, expression of hydrogenase Fe-S, which was found to be encoded next to the genes for step A6 (Figs. 2 , 4 ), indicates importance of electron transport and proton translocation. However, since the pathway proposed for T. phaeum requires establishment of a proton gradient from ATP hydrolysis (reverse electron transport), MAG9 metabolism would not generate enough ATP to drive acetate activation. Thus, further research is needed to confirm the metabolic route used by this and other SAOB. Similarly to S. schinkii and T. phaeum , [ 84 , 85 ] MAG9 expressed hydrogenase EchCE, formate dehydrogenase and Ni-Fe hydrogenases in both acetate- and propionate-fed batches. MAG9 also expressed a complete set of genes for NADH-quinone oxidoreductase (Fig. S12 ). Additional active bacterium in the syntrophic consortia A MAG (MAG5) belonging to the genus Acetomicrobium (76% [sourmash] and with 84% [dDDH] similarity to GCA_012518015.1) was present and showed activity in both acetate- and propionate-degrading cultures. This species expressed genes encoding the reductive glycine pathway (rGlyP), including the glycine cleavage system, the glycine reductase complex, pyruvate synthase and associated proteins (Fig. 4 , S13 ). A detailed discussion regarding the role of MAG5 is given in Supplementary note 6 , with the conclusion that considering the wide range of substrates used by members of Acetomicrobium [ 87 – 89 ] and that continuing cultivation of the syntrophic community in the present study demonstrated decreased abundance of Acetomicrobium on omitting yeast extract in the growth media (data not shown), MAG5 most likely fermented compounds included in the yeast extract or grew oxidatively using cysteine as electron acceptor. Genome and transcriptomics analysis of methanogen The 16S rRNA gene sequencing analysis of the reactor microbial communities demonstrated higher abundance of Methanothermobacter _NA than Methanoculleus spp. (Fig. S14 ). The qPCR analyses demonstrated that Methanomicrobiales sp. decreased from 10 7-8 to 10 5 gene copies L −1 over the course of operation of the continuously fed reactors. Species belonging to Methanobacteriales were present at relatively stable levels over time, varying between 10 5 and 10 7 gene copies L −1 in all reactors (Table S8). However, in the batch assays, 16S rRNA gene amplicon sequencing analysis showed that Methanothermobacter _NA and ‘ Ca . Methanoculleus thermohydrogenotrophicum’ were often present in similar relative abundance in batches from RP1. In batches from RP2, ‘ Ca . Methanoculleus thermohydrogenotrophicum’ was the only dominant species (Fig. S15 ). This is in agreement with the qPCR results demonstrating higher abundance of Methanobacteriales in batches with faster propionate degradation (B01-B02, 10 6 gene copies L -1 ) than in batches with slower propionate degradation (B03-B04, 10 4 gene copies L −1 ) (Table S8 ). Methanomicrobiales sp. were present at 10 6 gene copies L −1 in all batch assays. Taxonomic profiling using the custom krakan2 [ 90 ] database (Supplementary note 4 ) with the metagenomics data revealed that ‘ Ca . Methanoculleus thermohydrogenotrophicum’ was the most dominant methanogen and that Methanothermobacter sp. was higher in relative abundance in reactor RP1 compared to RP2 (Fig. S16 ). Metatranscriptomics results were in agreement with the 16S rRNA gene sequencing results from batch assays (Fig. S15 ), showing a high number of 16S rRNA transcripts mapped to Methanoculleus and Methanothermobacter , with a particularly high number of reads mapping to Methanothermobacter (default kraken2 database) (Fig. S17 , Table S6 ) in the B01 having higher propionate degradation rate. Although Methanothermobacter was among the dominant methanogenic genera in the 16S rRNA gene sequencing, metagenomics and metatranscriptomics analyses (Figs. S14 – S17 ), we were unfortunately not able to recover a MAG belonging to the genus Methanothermobacter , making it difficult to reveal the activity of this methanogen in the syntrophic cultures. However, a high-quality MAG (MAG1) was recovered and it showed relationship to ‘ Ca . Methanoculleus thermohydrogenotrophicum’ (78% dDDH) (GCA_001512375.1) (Figs. S18 , S19 , Table S6 ). As reported for the ammonia-tolerant mesophilic Methanoculleus bourgensis [ 91 ], methanogenesis pathway genes were found to be clustered together in MAG1 (Fig. S20 ). Metatranscriptomic data revealed expression of genes by MAG1 coding the transferases, reductases and dehydrogenases needed for hydrogenotrophic metabolism, including methyl CoM-reductase and formylmethanofuran dehydrogenase (Fig. S20 ). Moreover, MAG1 expressed genes for the V-type ATP synthetase and alcohol dehydrogenases (Fig. S21 ). The latter has been found previously in Methanoculleus genomes, indicating a trait of using alcohol as electron donor [ 92 ]. As in other thermophilic Methanoculleus [ 69 ], the electron-bifurcating hydrogenases (coenzyme F 420 ) (FrhABDG) and formate dehydrogenases (FdhABD) were expressed by MAG1 (Fig. S21 ). MAG1 also expressed genes for hyp type (HypABCDE), which encode proteins for expression and maturation of hydrogenases (Fig. S21 ). For the metabolic process and biosynthesis, MAG1 encoded and expressed acetate/acyl-CoA ligase (Fig. S21 ), which uses ATP for activation of acetate to acetyl-CoA. Metatranscriptomics quantification indicated almost identical expression pattern in MAG1 in both the propionate- and acetate-fed batch experiments. Overall, expression of the methanogenesis pathway and associated genes in MAG1 did not appear to give any specific differences in propionate versus acetate treatments that could reveal its partnership with SPOB or SAOB (Figs. S20 , S21 ). Expression of other genes potentially related to a syntrophic lifestyle Low energy gain is a well-known challenge and bottleneck for the thermodynamically constrained syntrophic interactions in microbial communities. Hence, it is reasonable to believe that strategies to reduce energy investment in cell metabolism is important for the species involved. To shed light on how the thermophilic ammonia-tolerant syntrophic communities in this study acclimatised to energy scarcity, particular attention was paid to activities with potential to increase energy gain and facilitate interspecies interactions. For energy production, the candidate syntrophs, MAG4 and MAG9, both expressed the F 0 F 1 -type ATP synthase complex (Figs. S9 , S12 ), as reported for known SAOB ( S. schinkii and T. phaeum ) [ 84 , 85 ] and SPOB members of Peptococcaceae and Syntrophobacteraceae [ 10 ]. Both MAG4 and MAG9 express ATP synthase subunit C to a higher extent than the other subunits. Subunit C has been shown to be crucial for ion translocation that leverages the proton/sodium motive force across the cell membrane and prevents ion leakage [ 93 , 94 ]. It has also been reported that the number of protons translocated is proportional to the number of subunit C [ 95 ] and that regulation of ATP synthase operon is proportional to ATP generation [ 96 – 98 ]. This suggests that this type of proton translocation mechanism could also be involved in the bioenergetics of SAO and SPO communities and efficient energy conserving ATP synthesis near thermodynamic equilibrium [ 97 , 98 ]. Moreover, it has been suggested for the SPOB P. thermopropionicum , P. schinkii and S. fumaroxidans that presence of a reverse electron transfer mechanism, menaquinone loop and higher number of expressed genes encoding hydrogenases and formate dehydrogenase (and associated higher enzymatic and cellular activity) could provide more metabolic agility and flexibility in the case of varying hydrogen or formate consumption by the syntrophic methanogenic partner [ 72 , 98 , 99 ]. In MAG4 and MAG9, these genes are encoded as either more than one copy in the genome, or were highly expressed (or both), which further indicates that the proteins encoded by these genes play a critical role in the complex syntrophic interactions among acetate-/propionate-degrading and methanogenic communities at the thermodynamic borderline (Figs. 3 , 4 , S9 , S12 , Table S6 ). For initiation of syntrophic oxidation, the SPOB and SAOB need to transport the substrate across the cell membrane, which can be done actively using a transport system or through passive diffusion [ 10 ]. The candidate SPOB MAG4 expressed sodium/solute transporter and both MAG4 and the candidate SAOB MAG9 expressed MFS transporter proteins for putative propionate intake (Figs. S9 , S12 ). MFS transporters are broad-spectrum transport systems involved in uniport, symport or antiport of various cellular metabolites, sugars and organic acids [ 100 ] and have been reported to play a role in tolerance to high levels of acetate and propionate, for example in Acetobacter spp., E. coli and P. putida [ 101 , 102 ]. Moreover, a gene belonging to the oxalate/formate antiporter (OFA) family of MFS transporters was located in the operon together with CoA-transferases and showed higher expression by MAG4 in the culture with faster propionate oxidisation relative to the culture with slower propionate degradation (B01 vs. B03, Fig. S9 ). This suggests that MFS transporters maybe responsible for acetate/propionate or formate transport in MAG4. The role of sulphur compounds in metabolic cooperation The transcriptome data revealed activity related to sulphur metabolism by the candidate SPOB (Fig. S9 ), including Hdr and CoA-disulphide reductase ( cdr ). The Hdr gene complex has previously been found to assist in electron confurcation in S. fumaroxidans [ 99 ]. Moreover, MAG4 expressed genes for dissimilatory sulphite reductase ( dsrC ) and anaerobic sulphite reductase ( asrAB ), which are the key determinants for sulphur reduction-based energy conservation in sulphate-reducing bacteria [ 99 , 103 ]. In MAG4, these sulphite reductase genes are encoded next to expressed putative NADH:ubiquinone oxidoreductase ( nfC ) and a mvhD , which are likely involved in electron transport phosphorylation or hydrogenase activities. Other important genes involved in dissimilatory sulphate reduction (adenylylsulphate reductase ( aprAB ), sulphate adenylyltransferase ( sat ), pyrophosphatase ( ppaX ), ABC-type sulphate transporter) are also encoded by MAG4, but were not highly expressed under the conditions investigated here. However, presence of all the genes required for sulphate reduction in MAG4 strongly indicates that this bacterium has the ability to respire sulphate if available. The ability of MAG4 to perform sulphate reduction and the exact mechanisms involved warrant further investigation, since this would improve understanding of another thermophilic SPOB, P. thermopropionicum . It is suggested to be a sulphate/thiosulphate/sulphite reducer [ 104 ], even though it has been described as unable to utilise sulphate due to absence of aprB and dsrAB genes [ 105 , 106 ]. Similar sulphur metabolism potential as observed for MAG4 was observed for ‘ Ca . Propionivorax syntrophicum’, possibly as a step in a series of reactions for sulphate-reducing metabolism, and a complex of hdr , rnfC and dsr , which could provide reduced ferredoxin for H 2 /formate production and also for low-energy metabolism [ 16 , 99 ]. However, the cultivation medium in the present study contained no sulphate or sulphite, contradicting the suggestion that the enzymes are involved in metabolism of these sulphur compounds. Instead, the cultivation medium included Na 2 S, cysteine, yeast extract and sulphur-containing vitamins, i.e. biotin, thioctic acid and thiamine, which might give rise to hydrogen sulphide and other sulphur compounds [ 107 , 108 ]. Many bacteria can also produce different di- and trisulphides [ 109 , 110 ]. Furthermore, as discussed above, MAG5 was taxonomically related to species that can reduce cysteine to sulphide, indicating similar activity by MAG5 in the enrichment culture. MAG4 also expressed cysteine desulfurase ( icsS ), which is responsible for sulphur activation in the cysteine degradation pathway. However, it might also be involved in formation of amino acids, as observed for the SAOB Schnuerera ultunensis , in which this gene is associated with production of alanine and sulphane/persulphide sulphur intermediates from cysteine degradation [ 111 – 113 ]. Apart from cysteine serving as a reducing agent in the medium, it may also mediate the electron carrier in SPO and subsequent methanogenesis [ 114 ]. Protein trisulphides are of interest in the present context since they are involved in sulphur reduction machinery-based energy conservation. The DsrC associated protein trisulphide ( metacycM:DsrC-trisulphides ) can act as a key intermediate in the reversible redox reaction producing and consuming sulphite [ 103 , 104 ]. Expression of asrAB and dsrC is a key determinant for sulphur reduction-based energy conservation in sulphate-reducing organisms [ 103 ]. Considering that several sulphate-reducing organisms have been found to establish syntrophic interactions in environments where sulphur is absent or limited [ 115 ], sulphate-reducing metabolic potential of MAG4 is further indicated. The genes for the sulphate-reducing pathway, together with different hydrogenases, could possibly also be involved in low-energy metabolism rather than sulphate reduction [ 99 ]. Mobility and other features with potential to facilitate interspecies cooperation MAG4 and MAG9 contained six and 23 motility associated proteins, respectively (Figs. S22 , S23 ). One of the pilus-associated proteins in MAG4 is PilT (type IV pili) (Figs. S22 , S23 ), which has been found to be associated with twitching motility, cellular adhesion, pilus retraction and sequence-specific DNA uptake [ 116 , 117 ]. Further, MAG4, MAG5 and MAG9 all expressed TIGR00282 family metallophosphoesterase proteins which were similar to YmdB, characterised for its role in nanotube and biofilm formation and intercellular molecular exchange in Bacillus subtilis [ 118 – 121 ]. Different types of flagellar and pilus-related proteins are known to play a role in initiating cellular contact, biofilm formation and establishing syntrophy. For instance, P. thermopropionicum FliD is used to establish contact with the partner methanogen M. thermautotrophicus and to synchronise their metabolism [ 122 , 123 ]. Considering the absence of the FliD gene in MAG4, it is likely that if this candidate SPOB uses direct interspecies electron transfer (DIET), it is employing a mechanism somewhat different from those characterised previously. Further, MAG4 and MAG9 expressed genes for cysteine synthase/O-acetylserine sulfhydrolase (CysK) and stage 0 sporulation protein (Spo0A) (Fig. S23 ), which can be involved in biofilm formation, as suggested for Vibrio fischeri (CysK) [ 124 ] and B. subtilis and Clostridium difficile (Spo0A) [ 125 , 126 ]. This perhaps explains the absence of motility-related (flagella) proteins and strongly suggests that MAG4 and MAG9 use pilus appendages for physically establishing deep physical contact with each other and the syntrophic methanogenic partner when present in close proximity (PilT-mediated) or for biofilm formation or nanotube communication and intercellular molecular exchange [ 118 , 127 , 128 ]. This feature would resemble that in P. thermopropionicum , which is characterised for biofilm and nanowire formation and interspecies electron and hydrogen sharing when growing in syntrophy with Methanothermobacter [ 123 , 129 , 130 ]. Cross-cellular communication, signalling and quorum sensing is another important concept intrinsic to syntrophic associations. Quorum sensing and signalling mechanism-related genes were expressed in all MAGs (Fig. S23 ). These genes, together with other associated genes, e.g. for motility, signalling, biofilm formation have been shown to be involved in DIET, intercellular metabolite exchange and communications [ 131 ]. The exact mechanism of the cooperation (e.g. establishment of nanotubes, use of flagellar or pilar assemblies) used by the syntrophic bacteria and the methanogens warrants further investigation. Stress response Expression of genes related to stress response, viz . chaperones (DnaJ, DnaK, ClpB, different chaperonins), heat shock protein (Hsp20) and hyperosmotic response (GrpE), are important for stress tolerance [ 132 , 133 ], was seen for MAG1 MAG4 and MAG9 (Fig. S24 ). Ammonia tolerance and resistance is a physiological phenomenon rather than a genetic property. Several complex mechanisms, i.e. osmo-tolerance, ionic membrane transport, molecular chaperones. etc . impart physiological resistance to evade metabolic deterioration under ammonia stress [ 132 – 135 ]. These stress genes have been found to be upregulated under acetate/acetic acid stress in E. coli [ 101 ]. Expression of stress-related proteins in MAG1, MAG4 and MAG9 could be due to the thermophilic temperature and high ammonia concentration applied in this study (Fig. S24 ). The transcriptomic response of stress-related genes of SPOB has not been characterised, so the exact role of these stress-regulated genes in MAG4 requires further investigation. Several genes in the candidate SPOB (MAG4), SAOB (MAG9) and the HM (MAG1), as also discussed above, were found to be present and expressed in an operon-like fashion (e.g. MMC, WLP, HM pathway genes, CoA-transferases, hydrogenases/dehydrogenases). The clustering of genes in SPOB ( Pelotomaculaceae family, reviewed elsewhere [ 10 ]) has been proposed to enable energetically advantageous coordinated expression of series of genes, since it requires less transcriptional machinery [ 73 , 136 ]. Similar to SPOB, the present results also indicate that the coordinated expression of series of genes energetically beneficial to ammonia-tolerant SAOB and methanogens."
} | 12,422 |
28989639 | PMC5625591 | pmc | 255 | {
"abstract": "‘Amine-reactive’ multilayers of a nano-complex are introduced by exploiting the Michael addition reaction to adopt ‘internal’ super-oil-wettability under water with impeccable physical/chemical durability.",
"conclusion": "Conclusions In conclusion, the design of amine-‘reactive’ NC multilayers provided insight about the fundamentals of underwater superoleophilic and superoleophobic properties in detail. The ‘reactive’ multilayers of NC provided a facile avenue to control both the chemistry and topography of the polymer coatings, and this fresh design independently revalidated some key hypotheses that explained the extremes of oil-wetting properties in detail. Moreover, the approach of using covalently cross-linked NC multilayers resulted in the desired ‘internal’ underwater super-oil-wettability properties, which can withstand various severe physical and chemical insults. Furthermore, the LbL deposition process allowed a wide range of substrates to be endowed with desired underwater super-wetting properties for both protecting the substrates from oil-contamination, and for cleaning oil from the oil-contaminated bare substrates. Current findings are anticipated to provide an avenue to further extend the horizon of bio-inspired underwater super oil-wetting properties for designing/developing advanced and multifunctional materials in the interest of a wide range of relevant applications in practical settings.",
"introduction": "Introduction Under-water superoleophobic materials—those possessing both superoleophilic and superhydrophilic properties in air— can strongly repel liquid-oil droplets (with an advancing oil CA >150° and contact angle hysteresis <5°) under water. 1 – 3 Synthesis of such underwater superoleophobic coatings, which were inspired by fish scale features, 4 is of potential interest in various basic and applied contexts including anti-bio-fouling coatings, rapid oil/water separation, marine anti-fouling coatings, synthesis of efficient organic field-effect transistors, etc. \n 2 , 6 – 11 The general requirements for developing such coatings are quite unusual in comparison to the conventional fluorine-based superoleophobic property that displays extreme oil-repellency mostly in air. 12 , 13 The underwater anti-oil-wetting property is often explained using the Cassie–Baxter model, where the trapped liquid water layer within the material contributes to the heterogeneous wetting of liquid-oil on the coated substrate under water. 4 , 5 So, the fabrication of a hierarchical surface topography composed of high surface energy materials is hypothesized to adopt desired underwater superhydrophobic properties. Jiang and co-workers first introduced an artificial underwater superoleophobic coating by mimicking the topography of fish scales using polyacrylamide (PAM) hydrogel. 4 Thereafter, various materials were developed to obtain such underwater superoleophobic properties, 4 , 9 , 10 , 14 – 19 mostly based on polymeric hydrogel 4 , 9 , 10 , 18 – 20 and metal oxide 14 , 21 – 23 coatings, which are highly susceptible to erosion in harsh physical/chemical environments (such as extremes of pH, salt, etc. ). 24 However, such approaches are well recognized and widely practiced in the literature as they are known to reveal both the fundamentals and prospective applications of this anti-wetting property. Examples of durable coatings that can withstand severe chemical/physical treatments are still rare in the literature, 10 , 24 and further development is essential to synthesize highly robust underwater superoleophobic coatings. In the recent past, the pursuance of oil/water separation has emerged as an important research theme due to the growing risk of water pollution from frequent oil spill accidents and the continuous discharge of industrial (oil-contaminated) wastewater. 2 , 11 The underwater anti-oil-fouling property is recognised for its self-cleaning ability, which together with the anti-fouling property provides a highly efficient avenue to separate/collect water from aqueous-oil contamination (emulsified or non-emulsified). 7 , 16 , 19 – 21 Another underwater super-wetting property known as underwater superoleophilicity, 25 , 26 which completely soaks up oils (with a contact angle of 0°) 25 under water, recently appeared in efficient materials that both separate and collect oil from aqueous media 25 – 27 and clean oil-contaminated surfaces. It was recently observed that such oil-wetting properties 25 – 29 often (with rare exceptions 25 , 30 ) collapse in the absence of particular stimuli. 26 – 29 Examples of these various stimuli-responsive underwater superoleophilic materials are useful in specific proof-of-concept demonstrations, but a material with durable super-oil-wetting properties may further enlighten the path to several prospective and convenient applications in various practical settings including extremes of pH, salt, and other complex chemical scenarios. The work presented here is primarily motivated from past demonstrations, where several functional materials are fabricated by exploiting the robust Michael addition reaction 31 – 35 between primary amines and acrylates at ambient conditions including (1) synthesis of polymer coatings with various complex nanostructures, 33 , 35 (2) dendritic amplification of desired functional groups, 34 and (3) selective and three-dimensional (3D) functionalization of polymer microstructures. 31 This facile chemical approach is found to be appropriate for tailoring both the chemistry and the topography of the material. 31 – 35 Here, we introduce an amine-‘reactive’ multilayer of a polymeric nano-complex (NC) exploiting the 1,4-conjugate addition reaction between the amine and acrylate groups of branched poly(ethyleneimine) (BPEI, polymer) and dipentaerythritol penta-/hexa-acrylate (5Acl, small molecule), respectively ( Scheme 1A and B ), to synthesize (1) covalently cross-linked, (2) thicker, (3) chemically/physically durable, (4) optically transparent, and (5) substrate-independent polymeric coatings with ‘internal’ (including both surface and interior) underwater superoleophobic/superoleophilic properties, which is unprecedented in the literature. This current design has been further explored to reveal the fundamentals of the extremes of oil-wettability under water in detail. In the recent past, Wang and coworkers 29 hypothesized the role of a trapped air layer in achieving continuous and discontinuous three-phase contact lines (TPCLs), which eventually controlled the underwater wettability of oil droplets at the multi-phase interfaces. Here, in our present design, we have developed continuous (for superoleophobicity) and discontinuous (for superoleophilicity) TPCLs just by adopting the appropriate chemistry in the material through strategic and facile post-chemical modification of the ‘reactive’ multilayers of NC, to control the underwater oil-wettability on solid surfaces as shown in Scheme 1C . Scheme 1 (A) The 1,4-conjugate addition reaction between primary amine and acrylate groups. (B) The chemical structures of poly(ethylenimine) (BPEI) and dipentaerythritol penta-acrylate (5Acl). Mixing of these chemicals in ethanol provides amine-‘reactive’ nano-complexes. (C) A schematic illustrating the construction of the ‘reactive’ multilayer by covalent LbL deposition of the ‘reactive’ nano-complex and BPEI. (D and E) Post-chemical modification of the multilayers with primary amine-containing small molecules (glucamine and octadecylamine) providing underwater superoleophobicity/superoleophilicity.",
"discussion": "Discussion Here, we have developed ‘amine-reactive’ covalently cross-linked multilayers of a nano-complex (NC) via a facile 1,4-conjugate addition reaction between the acrylate and primary amine groups of dipentaerythritol pentaacrylate (5Acl, multifunctional small molecules) and branched poly (ethyleneimine) (BPEI, polymer), respectively, in ethanol at ambient conditions. The consecutive layer-by-layer (LbL) deposition of the reactive (due to the presence of residual acrylate groups) and growing NC provided a basis to tailor both the topography and chemistry of the multilayers, and eventually allowed decoration of the material with essential fundamentals ((1) hierarchical topography and (2) appropriate chemistry), to achieve extremes of oil-wettability properties under water. This precise control over both the topography and chemistry of the multilayers further allowed the fundamentals behind the properties of underwater superoleophilicity and superoleophobicity to be revalidated independently with appropriate experimental demonstrations. The existence of meta-stable trapped air, which is hypothesized to be essential for underwater superoleophilicity, was experimentally revealed using our synthesized material. Furthermore, the controlled change in the morphology of the material (by strategic control of the LbL deposition cycles) in combination with appropriate post chemical modification of the ‘reactive’ multilayer provided a basis to tailor the underwater oil-wettability, starting from the superoleophilicity to the superoleophobicity. Moreover, the fraction of contact area between the beaded oil droplet and the glucamine-modified NC multilayers under water controlled the adhesive interaction, and eventually provided both the adhesive and non-adhesive superoleophobic coatings under water. The covalently cross-linked and thick (2.2 μm) multilayer coatings that were strategically post-functionalized with amine-containing small molecules (glucamine and octadecylamine) yielded ‘internal’ superoleophilic and superoleophobic properties under water, where both the surface and interior of the multilayers were with the appropriate topography and chemistry to display robust super-oil-wetting properties under water. As a consequence, the underwater super-oil-wettability properties remained unperturbed even after severe physical damage, including removal of the top surface of the material, which destroyed both the surface topography and surface chemistry of the material. On the other hand, extensive covalent cross linkages through the 1,4-conjugate addition reaction between the acrylate and amine groups provided impeccable chemical stability to the material, and thus, the underwater superoleophilic and superoleophobic properties remained unaltered, even after continuous exposure to complex aqueous conditions including artificial sea water, extremes of pH and protein- and surfactant-contaminated aqueous media. Furthermore, the anti-wetting property remained intact at raised temperatures (including boiling temperature), and during and after the freezing process. Thus, the current design provided underwater superoleophobic/superoleophilic coatings with exceptional physical and chemical durability. This covalent and rapid LbL deposition process further allowed different types of substrates to be coated with durable super-oil-wetting properties under water including the surfaces of wood, plastic, metal etc. Moreover, the substrates with a complex geometry, including the interior of the glass tube and the fibrous cotton, could be decorated with desired oil-wetting properties under water. The glass tube that was coated with underwater superoleophobicity was further exploited in demonstrating the protection of the substrate from oil-contamination and guided oil-transport under water, whereas the cotton that was decorated with underwater superoleophilicity was found to be useful in the facile cleaning of oil-contaminated bare glass surfaces. Thus, both these super-oil-wetting properties provided complimentary approaches for self-cleaning (underwater superoleophobicity) and facile post-cleaning (underwater superoleophilicity) of oil-contaminations under water, which can also be seen as a leading approach to protect/prevent water pollution and for water purification."
} | 2,976 |
36132530 | PMC9418872 | pmc | 256 | {
"abstract": "Since the emergence of memristors (or memristive devices), how to integrate them into arrays has been widely investigated. After years of research, memristor crossbar arrays have been proposed and realized with potential applications in nonvolatile memory, logic and neuromorphic computing systems. Despite the promising prospects of memristor crossbar arrays, one of the main obstacles for their development is the so-called sneak-path current causing cross-talk interference between adjacent memory cells and thus may result in misinterpretation which greatly influences the operation of memristor crossbar arrays. Solving the sneak-path current issue, the power consumption of the array will immensely decrease, and the reliability and stability will simultaneously increase. In order to suppress the sneak-path current, various solutions have been provided. So far, some reviews have considered some of these solutions and established a sophisticated classification, including 1D1M, 1T1M, 1S1M (D: diode, M: memristor, T: transistor, S: selector), self-selective and self-rectifying memristors. Recently, a mass of studies have been additionally reported. This review thus attempts to provide a survey on these new findings, by highlighting the latest research progress realized for relieving the sneak-path issue. Here, we first present the concept of the sneak-path current issue and solutions proposed to solve it. Consequently, we select some typical and promising devices, and present their structures and properties in detail. Then, the latest research activities focusing on single-device structures are introduced taking into account the mechanisms underlying these devices. Finally, we summarize the properties and perspectives of these solutions.",
"introduction": "1. Introduction Information transfer between the central processing unit (CPU) and the memories in von Neumann systems inevitably imposes limits on the performance and scalability of the architecture and results in large additional power consumption. This problem becomes more severe for tasks needing vast vector matrix multiplication (VMM) computing, such as real-time image recognition, data classification, and natural language processing, where state-of-the-art von Neumann systems difficultly work to match the performance of an average human brain. 1 A potential candidate hardware neuromorphic network, which mimics the operations of the human brain, has recently aroused much attention. Among numerous solutions to realize the required functions, neuromorphic networks based on memristors appear extremely promising. One of the crucial obstacles for an efficient memristor crossbar array is the so-called sneak path current problem, which decreases the reliability of the array by importing error when programing/reading the resistance state of memristors. Vast research studies are dedicated to solve this sneak-path current issue. Some general reviews 2–5 dealing with fundamental mechanisms, materials and architectures of memristors have partially addressed the sneak-path current problem and its solutions. Generally speaking, most of the early studies focused on structures, including one transistor-one memristor (1T1M), one diode-one memristor (1D1M) and one selector-one memristor (1S1M). Besides the multiple device solutions mentioned above, single-device systems, including self-selective memristors and self-rectifying memristors, have also drawn large amounts of attention due to their simple structure. These solutions are not only classified based on their I – V characteristics, but also depending on the composition of the devices in a memory cell. For a specific category, the devices differ from each other in terms of their intrinsic physical mechanisms. In this review, we introduce the concept of the sneak-path current issue and basics associated with its solutions. Especially, typical and promising devices are presented in detail. Finally, we summarize the properties and perspectives of these solutions. 1.1 VMM based on memristor crossbar arrays Memristor crossbar arrays whose discrete conductance states stand for synaptic weights could accomplish efficient brain-inspired computation. Massive parallelism could be performed in an analog manner using their intrinsic physical laws. Fig. 1a shows a typical memristor crossbar array. Memristors are located at each cross point of top electrodes (rows) and bottom electrodes (columns). The total current out of every column is a summation of the current through each memristor on this column following Kirchhoff's current law, while the current through the memristor is the multiplication of input voltage and memristor conductance following Ohm's law. This column current follows the formula: . In the same way, the charges collected from each column of the crossbar are expressed as: , when input voltage pulses keep a constant amplitude of V and vary their widths ( t i ). Thus, vector matrix multiplication (VMM), which is the basis for parallel computation in artificial neural networks, could be implemented by memristor crossbar arrays. In this VMM process by memristor crossbar arrays, the value of a matrix cell is encoded as the analogue memristor conductance of the crossbar array, the input vector is encoded as different voltage pulse amplitudes (widths) to the rows of the crossbar, and the VMM outputs correspond to currents (charges) collected from columns of the memristor crossbar. 6–8 For the current domain VMM method, according to Ohm's law, a strict linear current–voltage ( I – V ) characteristic of the memristor is required so that voltage pulse amplitudes are easily encoded as input vectors for multiplication computing. On the contrary, the charge domain VMM method tolerates nonlinear current because the voltage pulse widths are encoded as input vectors with a fixed voltage pulse amplitude. Furthermore, the fixed amplitude immensely simplifies the peripheral circuits in the charge domain VMM method. Both analog approaches finish the VMM computing in a single step, regardless of the matrix size, attracting huge interest for implementing brain-inspired computation. 9 Fig. 1 Diagrams show (a) a crossbar array, (b) the sneak-path issue in the crossbar array, and (c) the equivalent circuit of the sneak-path issue. 1.2 Sneak path current issue During the analog VMM computing, the memristor conductances (resistances) in the crossbar array need to be duly updated. One of the crucial obstacles in the resistance programing and reading process is the so-called sneak path current problem. Fig. 1b and c show the case of sneak-path current in a 2 × 2 crossbar array. When we intend to apply a voltage between A1 and B1 lines to switch the resistance state of memristor one (M1), the blue path is the desired current path. However, current could also un-intentionally flow through the red path which is called the sneak path current. Not only does it lead to incorrect reading of the resistance state of memristors, but it also disturbs the precise resistance modulation of the array because M2, M3 and M4 memristors in series also experience the voltage. The sneak path currents also induce high energy consumption. Vast research studies are devoted to this urgent and significant task to eliminate or suppress the sneak path current issue in memristor crossbar arrays. 1.3 Programing and reading schemes in memristor crossbar arrays The sneak path current could be effectively suppressed by designing the bias scheme for the programing and reading process. As shown in Fig. 2 , the resistance of memristors sandwiched between word lines and bit lines is programed or read under two common types of write bias schemes: the V /2 method 10 and V /3 method. 11 In the V /2 scheme, the selected word line and selected bit line are applied full voltage ( V ) and 0 voltage, respectively. The unselected word lines and bit lines are applied half voltage ( V /2). As a result, the selected memristor is under V bias, half-selected memristors are under V /2 bias, and unselected memristors are under no bias. As for the V /3 bias scheme, the selected word line and selected bit line are applied full voltage ( V ) and 0 voltage, respectively. The unselected word lines are applied V /3, whereas the unselected bit lines are applied 2 V /3. Accordingly, the selected memristor is under V bias, half-selected memristors are under V /3 bias, and unselected memristors are under − V /3 bias. Note that more than one selected cell could be programed parallelly in the memristor crossbar array. The V /2 bias in the V /2 method and ± V /3 bias in the V /3 method inevitably contribute to energy consumption. The nonlinear I – V curves give a lower energy consumption than the linear one. For a specific array size and nonlinearity, the V /3 method is more energy efficient for small arrays; as the array size increases and the number of selected cells decreases, the V /2 method achieves greater energy efficiency. 12 Fig. 2 Schematics of the bias method (a) and method (b). These bias schemes are effective ways to update and obtain states of the memristor crossbar array. However, for realization of efficient states update in situation where voltage pulses is messaged and complicated, such as spike neural network, and for a lower energy consumption, device level to suppress sneak path currents issue for precise resistance modulation of the array is necessary."
} | 2,367 |
23914141 | null | s2 | 257 | {
"abstract": "Spider dragline silk is a proteinaceous fiber with impressive physical characteristics making it attractive for use in advanced materials. The fiber is composed of two proteins (spidroins MaSp1 and MaSp2), each of which contains a large central repeat array flanked by non-repetitive N- and C-terminal domains. The repeat arrays appear to be largely responsible for the tensile properties of the fiber, suggesting that the N- and C-terminal domains may be involved in self-assembly. We recently isolated the MaSp1 and MaSp2 N-terminal domains from "
} | 137 |
36437042 | PMC9875615 | pmc | 258 | {
"abstract": "Abstract A memristive crossbar array (MCA) is an ideal platform for emerging memory and neuromorphic hardware due to its high bitwise density capability. A charge trap memristor (CTM) is an attractive candidate for the memristor cell of the MCA, because the embodied rectifying characteristic frees it from the sneak current issue. Although the potential of the CTM devices has been suggested, their practical viability needs to be further proved. Here, a Pt/Ta 2 O 5 /Nb 2 O 5‐ \n \n x \n /Al 2 O 3‐ \n \n y \n /Ti CTM stack exhibiting high retention and array‐level uniformity is proposed, allowing a highly reliable selector‐less MCA. It shows high self‐rectifying and nonlinear current‐voltage characteristics below 1 µA of programming current with a continuous analog switching behavior. Also, its retention is longer than 10 5 s at 150 °C, suggesting the device is highly stable for non‐volatile analog applications. A plausible band diagram model is proposed based on the electronic spectroscopy results and conduction mechanism analysis. The self‐rectifying and nonlinear characteristics allow reducing the on‐chip training energy consumption by 71% for the MNIST dataset training task with an optimized programming scheme.",
"conclusion": "3 Conclusion In this work, we proposed the Pt/Ta 2 O 5 /Nb 2 O 5‐ \n \n x \n /Al 2 O 3‐ \n \n y \n /Ti CTM device exhibiting self‐rectifying, nonlinear, highly uniform, analog, and low power operation characteristics. Moreover, the device retention was secured, making it practically available for the crossbar array applications. We proposed an electronic band diagram model supported by in‐depth spectroscopy analysis, which helped to understand the mechanism of the memristive system. We integrated the cell on a 32 × 32 crossbar array and suggested its energy‐efficient neuromorphic computing operation using a dedicated biasing scheme synchronized with the device's I – V characteristics. This high potential of the CTM system can be widely applicable for various applications where energy consumption is the matter, such as an edge neuromorphic device.",
"introduction": "1 Introduction A memristive crossbar array (MCA) attracts intensive attention due to its various emerging applications, such as high‐density storage class memory or artificial synapse network for neuromorphic computing. [ \n \n 1 \n \n ] The significant advantage of the MCA is its high‐density capability; the cell requires only 4 F \n 2 ( F = minimum feature size) area, and it can be easily stackable either vertically or horizontally, which can multiply its areal density. Furthermore, its multi‐level programming capability can provide even higher bit‐storage density. Despite such a high potential, the MCA may suffer from a sneak current problem in both reading and programming due to the missing of the cell transistor. Inserting a two‐terminal selector in series with the memristor cell can be a viable solution, [ \n \n 2 \n \n ] but it requires a significant burden in material development and integration processing. In this regard, charge trap memristors (CTM) are promising memristive systems due to their inherent forming‐free and self‐rectifying characteristics. [ \n \n 3 \n , \n 4 \n , \n 5 \n , \n 6 \n , \n 7 \n , \n 8 \n , \n 9 \n , \n 10 \n , \n 11 \n \n ] Also, they show analog conductance change characteristics at low operation current range, enabling high‐density MCA applications. [ \n \n 12 \n , \n 13 \n \n ] The CTM mechanism is identical to the charge‐trap flash (CTF) memory, in which reversible charge trapping and de‐trapping to the charge trap layer alter the threshold voltage of the transistor. [ \n \n 14 \n , \n 15 \n , \n 16 \n \n ] Unlike the three‐terminal CTF, the two‐terminal CTM directly reads the conductance change through the charge trap layer, allowing higher density and simpler operation. [ \n \n 17 \n , \n 18 \n , \n 19 \n \n ] However, as the programming and reading operations share identical terminals, there is a contradiction between the stability of the trapped charges (i.e., retention) and the reliability of the programming process (i.e., writing or erasing). [ \n \n 20 \n \n ] The higher stability of the trapped charges can be achieved by reducing the tunneling probability of the trapped charges by forming thicker oxides. However, they make the charge trapping and de‐trapping process difficult, increasing the programming voltage or time. To achieve both high stability and programming reliability, it is crucial to understand the charge trapping and de‐trapping processes, as well as developing an optimized device stack. Meanwhile, nonlinear current–voltage ( I – V ) characteristics (curvature of the I – V curve at the forward bias) are also crucial for high‐density memory applications in the MCA. It was demonstrated that inserting a nonlinear selector device in series with the self‐rectifying CTM could reduce the total power consumption by suppressing the unnecessary power consumption from half‐selected cells. [ \n \n 7 \n \n ] If the nonlinear characteristic is embedded inherently, larger array integration would be possible, and the practicality of the CTM‐based MCA could be further secured. In this work, we propose a Pt/Ta 2 O 5 /Nb 2 O 5‐ \n \n x \n /Al 2 O 3‐ \n \n y \n /Ti (PTNAT) CTM device exhibiting > 5 × 10 4 of self‐rectifying ratio, < 1 µA of programming current, > 10 5 cycles of endurance, and > 2 × 10 5 seconds of retention at 150 °C, which is suitable for the high‐density memory applications. We propose a plausible CTM mechanism model accounting for the retention‐secured characteristics based on systematic experiments and electronic spectroscopy analysis results. Also, we show its high uniformity from a 32 × 32 array device and demonstrate its neuromorphic device operation. With an optimized weight update biasing scheme utilizing the self‐rectifying and nonlinear characteristics, the system can achieve high accuracy (≈91%) at the MNIST dataset recognition challenge with only 29% of energy consumption compared to the conventional programming scheme.",
"discussion": "2 Results and Discussion 2.1 Electrical Characteristics of the Pt/Ta 2 O 5 /Nb 2 O 5‐ \n \n x \n /Al 2 O 3‐ \n \n y \n /Ti Charge Trap Memristor Device \n Figure \n \n 1 a shows a cross‐sectional transmission electron microscopy (TEM) image of the PTNAT CTM structure, where the thicknesses are ≈10 nm for Ta 2 O 5 , ≈28 nm for Nb 2 O 5‐ \n \n x \n , and ≈8 nm for Al 2 O 3‐ \n \n y \n . All oxides are amorphous, confirmed by the fast‐Fourier transform of the TEM (insets in Figure 1a ) and XRD measurements (see Figure S1 , Supporting Information, for the XRD results). Figure 1b shows analog I – V characteristics of 25 µm 2 ‐area device by various positive set voltages from 6 to 10 V, followed by a fixed ‐10 V reset voltage. The device is electroforming‐free, and its pristine state is the high resistance state (HRS). Any intermediate resistance states between the HRS and the low resistance state (LRS) can be attained by either changing a compliance current ( I \n cc ) or the maximum set voltage. Figure 1c shows analog I – V curves, where the applied voltage was consecutively increased from 6 to 10 V without the reset process. The analog switching characteristic is attributed to a continuous charge trapping into the defect states of the Nb 2 O 5‐ \n \n x \n layer. As the amount of trapped charges increases, they form a higher internal negative field, lowering the Schottky barrier height (SBH) at the Ta 2 O 5 /Nb 2 O 5‐ \n \n x \n interface. Also, the I – V curves show a flat band voltage ( V \n fb ) at ≈1.3 V (gray dashed line). This can increase the nonlinearity in the forward bias, making it more beneficial for the energy‐efficient array operation. More details for the switching modeling and the array demonstration will be discussed in the following sections. Figure 1d plots the on/off ratio (= I \n LRS (V) / I \n HRS (V)) and the rectifying current ratio (= I \n LRS (V) / I \n LRS (‐V)) as a function of the voltage, obtained from 10 V/−10 V DC I – V sweep curve of Figure 1b . The maximum on/off ratio is ≈8 × 10 3 at 4.5 V. In the high voltages (>6 V, the set threshold voltage), the device set‐switched gradually, so the on/off ratio decreased accordingly. At 10 V, the device set‐switched completely to the LRS, so the on/off ratio collapsed to 1. The maximum rectifying current ratio is ≈5.7 × 10 4 at 7.5 V, and the rectifying ratio exceeds 10 4 over a wide voltage range exceeding ≈5.0 V at the LRS, suggesting the device may efficiently suppress the sneak currents during both reading and programming. To minimize the sneak currents, get a high on/off window, and have sufficient voltage margin with the set voltage (≈6 V), the optimal read voltage could be around 4 to 5 V. Figure 1e shows cumulative probability of the LRS current levels at various read voltages from 4 to 8 V. The dashed lines are the average current values obtained from the first voltage sweeps of 20 cells, confirming high cell‐to‐cell uniformity. Note that the device is electroforming‐free, which can realize higher uniformity than the electroformed devices. Figure 1f shows a cycling endurance up to 10 5 , where the set and reset pulse conditions were 13 V for 10 ms and −13 V for 10 ms, and the reading pulse was 4 V for 1 ms. This endurance is comparable with the endurance of the commercialized storage (compare NAND flash guarantees about 10 5 endurance), suggesting it is sufficient for non‐volatile memory applications. After ≈5 × 10 4 cycles, the LRS current was unstable, which is associated with the electron trapping at the Al 2 O 3‐ \n \n y \n tunneling oxide. In the NAND flash, it is well‐known that the program/erase (P/E) efficiency can be deteriorated over P/E cycles due to the oxide aging and unwanted additional electron trapping. [ \n \n 21 \n , \n 22 \n \n ] Similarly, the tunneling oxide may be aged over cycling, leading to unintended charge trapping and the LRS fluctuation. After 10 5 cycles, the HRS collapsed to the LRS, meaning the trapped charges could not escape from the traps. This is associated with the deterioration of the Nb 2 O 5‐ \n \n x \n deep trap sites, which form permanent fixed charges. Figure 1g shows long‐term potentiation (LTP) and long‐term depression (LTD) behaviors for the artificial synapse application. The inset shows the pulse conditions for LTP and LTD. The conductance was read by 4 V to prevent the unwanted conductance change during reading. Potentiation and depression pulse amplitudes were set to 11 and −9.5 V, which could achieve the full memory window. These voltages can readily offer a high rectifying ratio as shown in Figure 1d , so it can be used in the array operation, which is discussed later. Figure 1h shows 10 LTP and LTD cycles, confirming a high cycling uniformity. Figure 1i shows retention characteristics of selected eight conductance states at 125 °C (read at 2 V), verifying high stability up to 1000 s at the elaborated temperature except for some white noise from the measurement system. Figure 1 Pt/Ta 2 O 5 /Nb 2 O 5‐ \n \n x \n /Al 2 O 3‐ \n \n y \n /Ti (PTNAT) cell characteristics. a) The cross‐section transmission electron microscopy (TEM) image of the device and fast‐Fourier transform (FFT) image. b) The resistance switching I – V curves of the device with 1 µA compliance current ( I \n CC ) measured at different positive voltage sweeps (6–10 V) and −10 V fixed voltage sweeps. c) I – V curves with continuously increasing voltage sweeps (6–10 V) without reset. d) The on/off ratio and the rectifying ratio as a function of the applied voltages. e) The cumulative probabilities of the current levels of LRS in the positive voltage region, which were obtained from the first voltage sweeps in 20 cells. f) The endurance properties up to 10 5 cycles read at 4 V at room temperature. g) The average and standard deviation of both LTP and LTD for 70 pulses during ten cycles. h) 10 cycles of LTP and LTD operation of the device. i) The retention characteristics of 3‐bit intermediate states at 125 °C at 2 V. 2.2 Study on the Retention Characteristics of the Pt/Ta 2 O 5 /Nb 2 O 5‐ \n \n x \n /Al 2 O 3‐ \n \n y \n /Ti Device To investigate the high retention of the PTNAT device, we investigated I – V curves and retention characteristics of Pt/Nb 2 O 5‐ \n \n x \n /Ti (PNT), Pt/Nb 2 O 5‐ \n \n x \n /Al 2 O 3‐ \n \n y \n /Ti (PNAT), Pt/Ta 2 O 5 /Nb 2 O 5‐ \n \n x \n /Ti (PTNT), and PTNAT devices. The PNT device showed a narrow on/off ratio due to a high leakage current with a low set switching voltage of 3 V. (see Figure S2a , Supporting Information, for the PNT device I – V data.) Also, the device returned to the HRS shortly after the set switching, meaning the trapped charges were de‐trapped quickly as soon as the electric field was removed. When the Al 2 O 3‐ y \n layer was inserted at the bottom (i.e., PNAT device), the on/off ratio was drastically improved by suppressing the HRS current and the rectifying characteristic was revealed, as shown in Figure \n \n 2 a . Compared with the PNAT device, when the Ta 2 O 5 layer was inserted underneath the Pt (PTNT device), the HRS was still leaky, as shown in Figure 2b , meaning the upper Ta 2 O 5 layer was not suppressing the leakage current. Figure 2c shows the I – V curve of the PTNAT device with both the upper Ta 2 O 5 and the lower Al 2 O 3‐ \n \n y \n layers. Figure 2 Retention tendency comparison with 2‐layer and 3‐layer devices. The resistance switching I – V curves of the a) Pt/Nb 2 O 5‐ \n \n x \n /Al 2 O 3‐ \n \n y \n /Ti (PNAT), b) Pt/Ta 2 O 5 /Nb 2 O 5‐ \n \n x \n /Ti (PTNT), and c) Pt/Ta 2 O 5 /Nb 2 O 5‐ \n \n x \n /Al 2 O 3‐ \n \n y \n /Ti (PTNAT) device. Each I – V curve was measured with a 10 V positive voltage sweep and a −10 V negative voltage sweep with a 10 −6 A compliance current. The retention characteristics of d) PNAT (4 V at room temperature (RT)), e) PTNT (3 V at RT), and f) PTNAT (2 V at 150 °C). Schematic energy band diagram models of g) PNAT, h) PTNT, and i) PTNAT devices. The retention characteristics of the three devices were also distinguishable. In Figure 2d , the PNAT device showed improved stability than the PNT device. However, the LRS collapsed to the HRS, so the on/off ratio was reduced from ≈200 to ≈10 after ≈1000 s. It means the trapped charges are still unstable, so they may escape from the trap sites to the Pt electrode. Meanwhile, it suggests that the Al 2 O 3‐ \n \n y \n layer may stabilize the HRS. In the PTNT device, the retention failure trend was different compared to the PNT and PNAT devices; the HRS collapsed to the LRS, as shown in Figure 2e . It suggests that the upper Ta 2 O 5 layer may suppress the de‐trapping of charges and stabilize the LRS, although the HRS was unstable. Eventually, the PTNAT device showed stable LRS and HRS even at the harsh temperature, as shown in Figure 2f (150 °C, read at 2 V). The retention tendency could be understood with the schematic energy band diagram of each device. For the band structure modeling, reflection energy loss spectroscopy (REELS) and ultraviolet photoelectron spectroscopy (UPS) analysis were performed. (see Figure S3 , Supporting Information, for the REELS and UPS results of Al 2 O 3‐ \n \n y \n , Nb 2 O 5‐ \n \n x \n , and Ta 2 O 5 layers.) From the analysis results, the bandgap ( E \n g ) and the electron affinity ( χ ) values of each dielectric layer were obtained. The estimated E \n g values from the threshold energy of the REELS were 6.6 eV (Al 2 O 3‐ \n \n y \n ), 4.2 eV (Nb 2 O 5‐ \n \n x \n ), and 4.5 eV (Ta 2 O 5 ), respectively. The UPS analysis gave the χ of each layer to 2.29 eV (Al 2 O 3‐ \n \n y \n ), 4.03 eV (Nb 2 O 5‐ \n \n x \n ), and 3.44 eV (Ta 2 O 5 ). From the analysis results, the band diagram of each device can be suggested, as shown in Figure 2g–i . In all devices, two trap levels are present; a deep trap level around ≈1.1 eV below the conduction band and a shallow trap level around ≈0.22 eV below the conduction band, which are attributed to the oxygen vacancies of the amorphous Nb 2 O 5‐ \n \n x \n layer. [ \n \n 23 \n \n ] The Pt/Ta 2 O 5 /Al 2 O 3‐ \n \n y \n /Ti (PTAT) device missing the Nb 2 O 5‐ \n \n x \n layer did not show memory operation, confirming that the Nb 2 O 5‐ \n \n x \n layer is responsible for the charge trapping. (see Figure S2b , Supporting Information, for the PTAT device I – V data). Among the two trap levels, the deep trap level is responsible for the charge trap‐based memory operation. The shallow trap level acts as a stepping‐stone‐like energy state, affecting the charge trapping and de‐trapping process, which is elaborated in the modeling section. The retention failure mechanisms can be suggested in the given band diagram. In the PNAT device (Figure 2g ), the trapped charges can easily escape from the traps toward the Pt electrode as there is no blocking barrier at the top interface. The de‐trapping is possible via direct tunneling from the traps to the Pt electrode. In the PTNT device (Figure 2h ), the trap sites are spontaneously trapped over time due to the low activation energy for the electron tunneling process from the Ti bottom electrode to the Nb 2 O 5‐ \n \n x \n layer, while the upper Ta 2 O 5 layer suppresses the de‐trapping process. Therefore, an LRS failure was observed. In addition, the shallow traps at the Nb 2 O 5‐ \n \n x \n /Ti interface generated by the oxygen scavenging effect of the Ti electrode may help the charge trapping process via band‐to‐trap direct tunneling. Lastly, in the PTNAT device (Figure 2i ), both Ta 2 O 5 and Al 2 O 3‐ \n \n y \n dielectric layers stabilize the trapped charges and suppress spontaneous trapping even at high temperatures. 2.3 Conduction Mechanism Analysis and Band Diagram Modeling of Pt/Ta 2 O 5 /Nb 2 O 5‐ \n \n x \n /Al 2 O 3‐ \n \n y \n /Ti Device Conduction mechanisms of the PTNAT device were examined to confirm the proposed band diagram model and understand the detailed switching mechanism. Figure \n \n 3 a shows the LRS I – V curves measured at temperatures ranging from 40 to 70 °C. The results revealed that the LRS is highly temperature‐dependent, suggesting that the conduction is associated with the thermionic conduction mechanism. For more detailed analysis, the temperature‐dependent I – V curves at a high bias region (3.5–6.5 V) were plotted to the Schottky emission equation (ln( J / T \n 2 ) vs E \n 1/2 ), as shown in Figure 3b . In addition, multiple LRS curves programmed by different voltages from +6 to +9 V at fixed room temperature were plotted with the Schottky emission equation, as shown in Figure 3c . Both plots confirmed that the Schottky emission dominates the conduction of the device. [ \n \n 24 \n \n ] From the fitting, the optical dielectric constant ( ε \n op ) and SBH ( Φ \n B ) of the conduction‐limiting interface could be extracted at each temperature and each conductance state, as shown in Figure 3d and Figure 3e , respectively. For the fitting, the local electric field on each layer was calculated using the thickness and dielectric constant of the oxide layers. The thicknesses were 9.7 nm of Ta 2 O 5 , 27.7 nm of Nb 2 O 5‐ \n \n x \n , and 8.3 nm of Al 2 O 3‐ \n \n y \n from the TEM image in Figure 1a . The dielectric constants were 24 for Ta 2 O 5 , [ \n \n 25 \n \n ] 45 for Nb 2 O 5‐ \n \n x \n , [ \n \n 26 \n \n ] and 9 for Al 2 O 3‐ \n \n y \n , [ \n \n 27 \n \n ] which were taken from literature. Figure 3 Conduction mechanism fitting of PTNAT device. a) I – V curves of the LRS measured at temperatures ranging from 40 to 70 °C. b) Schottky emission form (ln( J / T \n 2 ) vs E \n 1/2 ) plot for the voltages ranging from 3.5 to 6.5 V in LRS with the calculated Ta 2 O 5 layer partial electric field. c) Analog LRS states Schottky emission fitting in +4.5 V read voltage and room temperature condition after +6 to +9 V different sweep voltages. d) High‐frequency permittivity ( ε \n op ) and Schottky barrier height ( Φ \n B ) extracted at each temperature condition I – V curves. e) ε \n op and Φ \n B values extracted at each analog state. Φ \n B decreased with a more programmed analog state (higher conductance state). After investigating all possible cases, we could conclude that the Schottky barrier at Ta 2 O 5 /Nb 2 O 5‐ \n \n x \n interface limited the overall conduction (see Figure S4 , Supporting Information, for the conduction mechanism fitting results limited by other layers.) The estimated ε \n op values ranged from 4.4 to 3.4 as the temperature increased from 40 to 70 °C, which coincided with the reference value ( ε \n op ≈ 4) of Ta 2 O 5 calculated by the square of its refractive index ( n ≈ 2). In addition, the estimated Φ \n B ranged from 0.588 to 0.591 eV at 40–70 °C, suggesting the band offset at Ta 2 O 5 /Nb 2 O 5‐ \n \n x \n interface. (The effective electron mass condition ( m \n e ≈ 0.3 m \n 0 ) of Ta 2 O 5 was used for the Φ \n B calculation. [ \n \n 28 \n \n ] ) This band offset also coincides with the χ difference of the Ta 2 O 5 /Nb 2 O 5‐ \n \n x \n interface measured from the UPS analysis, 4.03 eV (Nb 2 O 5‐ \n \n x \n ), and 3.44 eV (Ta 2 O 5 ). Similarly, the ε \n op and Φ \n B were also calculated from multiple conductance states. The estimated ε \n op was 3.82–4.08, consistent with the temperature‐dependent fitting results in Figure 3d . Interestingly, the Φ \n B were decreased from 0.72 eV (lower conductance state) to 0.60 eV (higher conductance state) as the set voltage increased, suggesting the Φ \n B decreased by further charge trapping. Such Φ \n B modulation behavior with respect to the charge trapping process can be plausibly explained by the following switching model. \n Figure \n \n 4 a–c shows the XPS depth profile results of the Ta 4 f , Nb 3 d , and Al 2 p core levels at each dielectric layer of the PTNAT device. The left panels show the raw XPS depth profile data (etch level from 1 to 57) to navigate the etch levels of interest (right panels) for each figure. In Figure 4a , the binding energy of two peaks have coincided with the reference Ta 2 O 5 phase peaks (black dashed lines at 26.8 eV for Ta 4 f \n 7/2 and 28.6 eV for Ta 4 f \n 5/2 ). [ \n \n 29 \n \n ] The core levels showed no specific changes with the etch level increase, meaning a stoichiometric Ta 2 O 5 layer was formed by the plasma‐enhanced atomic layer deposition (PEALD) process. Figure 4b shows Nb 3 d core levels with the reference peak positions are as follows: Nb 5+ (Nb 2 O 5 ) peaks at 210.0 and 207.3 eV (blue dashed line), Nb 4+ (NbO 2 ) peaks at 208.8 and 206.0 eV (red dashed line), and Nb 2+ (NbO) peaks at 206.8 and 204 eV (black dashed line). [ \n \n 30 \n \n ] The XPS results of the Nb 2 O 5‐ \n \n x \n layer suggested complicated composition of various sub‐oxide phases, meaning a significant amount of oxygen vacancies exist in the Nb 2 O 5‐ \n \n x \n charge trap layer. (see Figure S5 , Supporting Information, for the Nb 3 d XPS spectra deconvolution results of the PTNAT device.) At the Nb 2 O 5‐ \n \n x \n /Al 2 O 3‐ \n \n y \n interface, the peaks shifted toward the lower binding energy slightly, meaning the higher oxygen vacancies at the interface. Moreover, Al 2 p peaks also showed lower binding energy shifting (orange dashed line) characteristics at the Nb 2 O 5‐ \n \n x \n /Al 2 O 3‐ \n \n y \n interface compared with the bulk region (black dashed line), [ \n \n 31 \n \n ] as shown in Figure 4c . Both indicate that a high concentration of oxygen vacancies is formed at the Nb 2 O 5‐ \n \n x \n /Al 2 O 3‐ \n \n y \n interface. This may be originated from the sputtering damage during the Nb 2 O 5‐ \n \n x \n deposition kicking off the oxygens from the underlying Al 2 O 3‐ \n \n y \n layer. Eventually, high‐density trap levels were formed at the interface helping the trap‐assisted tunneling process. Figure 4 PTNAT XPS depth profile results and resistance switching mechanism of the device. Depth profiling results of a) Ta 4 f , b) Nb 3 d , and c) Al 2 p core levels of the Ta 2 O 5 (10 nm)/Nb 2 O 5‐ \n \n x \n (28 nm)/Al 2 O 3‐ \n \n y \n (8 nm)/Ti sample. The left panels show the raw XPS depth profile data. The right panels enlarge the square region. d) The schematic energy band diagram of the device for zero bias condition. Shallow and deep trap levels exist in the non‐stoichiometric Nb 2 O 5‐ \n \n x \n layer. e) The illustration of the resistance switching process with charge trapping and de‐trapping. In the HRS, deep traps are empty, and conduction is not fluent with the high band offset at the Ta 2 O 5 interface (I). When a positive set bias is applied on the Pt electrode, the electrons can be trapped in the Nb 2 O 5‐ \n \n x \n deep trap sites (II). With the induced negative space charge, the partial electric field across the Ta 2 O 5 layer increased (III), and LRS is achieved by effective Schottky barrier height lowering (IV). When a negative bias is applied (V, VI), the trapped electrons can be released to the Ti electrode, and the HRS is obtained. Based on the conduction mechanism study and REELS, UPS, and XPS results, we could establish a detailed switching model of the PTNAT device. Figure 4d shows an energy band diagram of the PTNAT device. The work functions ( Φ ) of the electrodes were assumed to 5.6 eV for Pt top electrode and 4.3 eV for Ti bottom electrode. The bandgap ( E \n g ) and the electron affinity ( χ ) values were estimated from REELS and UPS results in Figure S3 , Supporting Information, which are 4.5 and 3.44 eV for Ta 2 O 5 ; 4.2 and 4.03 eV for Nb 2 O 5‐ \n \n x \n ; and 6.6 and 2.29 eV for Al 2 O 3‐ \n \n y \n , respectively. As mentioned above, the Nb 2 O 5‐ \n \n x \n charge trap layer contains two trap levels (≈0.22 eV and ≈1.1 eV below the conduction band) associated with the oxygen vacancies. The deep trap level is responsible for the charge trapping. The shallow trap levels assist fluent carrier injection via Poole‐Frenkel (P‐F) emission. [ \n \n 32 \n \n ] Also, in the Al 2 O 3‐ \n \n y \n layer, oxygen vacancies were observed at the Nb 2 O 5‐ \n \n x \n /Al 2 O 3‐ \n \n y \n interface. These defects could enhance the band‐to‐trap tunneling through the Al 2 O 3‐ \n \n y \n layer, allowing the Al 2 O 3‐ \n \n y \n to act as a tunneling layer despite the high band offset at the Al 2 O 3‐ \n \n y \n /Ti interface. With the proposed band diagram, the analog charge trapping associated resistance switching characteristics can be well understood. Figure 4e shows schematic band diagrams during the set and reset process. In the pristine HRS state (panel i), the deep traps are empty, and the conduction is not fluent due to the high SBH at the Ta 2 O 5 /Nb 2 O 5‐ \n \n x \n interface. When a positive set bias (>6 V) is applied to the Pt top electrode, the injected electrons start to transit to the deep traps and act as negative space charges. They form an internal electric field and pull up the conduction band, resulting in two changes to the programming process; first, it lowers the SBH at the Ta 2 O 5 /Nb 2 O 5‐ \n \n x \n interface, making the conduction more fluent. Second, it also pulls up the shallow trap levels relative to the Fermi level of the Ti electrode and suppresses additional charge trapping. Therefore, the set switching process stops at the given set switching voltage, making an analog conductance change possible. Once the space charges are stored, the SBH at the Ta 2 O 5 /Nb 2 O 5‐ \n \n x \n interface can be decreased by an image force‐associated Schottky barrier lowering effect. [ \n \n 33 \n \n ] The amount of barrier lowering depends on the applied partial electric field across the dielectric film, which can be given as\n \n (1) \n Δ ϕ E = e 3 E / 4 π ε s + e α E 2 \n where α is a material constant given by \n \n (2) \n α = e ℏ 2 / 24 m ∗ ( k B T ) 2 \n \n ε \n s is a permittivity, h is the Plank constant, and m * is the effective electron mass. Thus, as more charges are trapped, the lower SBH is formed. These trapped charges are stable due to the high band offset at the Ta 2 O 5 /Nb 2 O 5‐ \n \n x \n and Nb 2 O 5‐ \n \n x \n /Al 2 O 3‐ \n \n y \n interfaces, as shown in Figure 2f . At a sufficient negative bias for reset, the trapped charges are de‐trapped by the reverse sequence of the trapping process from deep traps via the shallow traps to the electrode. 2.4 Demonstration of Energy‐Efficient Neuromorphic Computing We fabricated a 32 × 32 MCA embedding the PTNAT cell, and examined an energy‐efficient on‐chip training method to get the maximum benefits of the device. Figure \n \n 5 a shows a top‐view image of the integrated device taken by scanning electron microscope, where the line width of the crossbar is ≈5 µm. Figure 5b shows the I – V curves of randomly selected 80 cells (gray), and one representative curve (red), confirming high array‐level uniformity. The current sensing limit of the array testing platform was about ≈10 −8 A, so the lower current levels were not detectable. Nevertheless, the read currents at 4 V and 6 V were distinguishable with high uniformity, as shown in Figure 5c . Figure 5 32 × 32 MCA integration and energy‐efficient neuromorphic computing demonstration. a) A top‐view SEM image of the 32 × 32 MCA. b) I – V curves of randomly chosen 80 cells with the half voltage scheme. The inset shows the half voltage array measurement scheme. c) The cumulative probabilities of the current levels at the LRS (read voltages of 1.3, 4, and 6 V), and the HRS (read voltages of 4 and 6 V). d) The serial (left) and parallel (right) programming schemes for weight update. e) Suggested potentiation and depression scheme for parallel programming. f) The double‐layer perceptron neural network used for the MNIST dataset training simulation. It consists of 784 input neurons, 256 hidden neurons, and 10 output neurons. g) The MNIST data recognition rate by training epoch considering the experimental synaptic characteristics of the CTM device. The final accuracy was about 91%. h) Estimated energy required for training by serial and parallel programming in half voltage scheme and those in the suggested scheme. At on‐chip training, the energy consumption would be high if all cells in the array were updated individually. The self‐rectifying and nonlinear behaviors of the device can offer a way of energy‐efficient programming. It is reported that parallel programming could reduce energy consumption compared to the conventional serial programming method by decreasing the total number of programming events and energy consumption per event. [ \n \n 34 \n \n ] Figure 5d compares the conventional serial programming and proposed parallel programming schemes. In parallel programming, the portion of the half‐selected cells is considerable, so managing both unselected and half‐selected currents is crucial. In the proposed PTNAT device, both the unselected cell and the half‐selected cell currents can be suppressed due to self‐rectifying and nonlinear behaviors, allowing the parallel programming scheme to be viable. Figure 5e illustrates parallel programming biasing schemes of potentiation (left panel) and depression (right panel) operations. For potentiation, the selected word and bit lines are biased to 6 and −5 V, respectively, offering a net 11 V potential for programming, which is the identical potentiation condition of Figure 1g . Meanwhile, the unselected word and bit lines are biased to 0 and 4.7 V, resulting in 1.3 V of the half‐selected cell voltage and −4.7 V of the unselected cell voltage, respectively. As shown in Figure 1b , 1.3 V corresponds to the flat band voltage, allowing a minimum leakage current at the half‐selected cells. Similarly, for depression, the selected word and bit lines are biased to −5 and 4.5 V, while the unselected bit and word lines are biased to 0 and −1.3 V, respectively, making the half‐selected cell voltage to −3.7 or −4.5 V, and the unselected cell voltage to 1.3 V. The energy consumption of the CTM array as neuromorphic hardware was examined at a software‐based emulator embodying the CTM properties for the synapse device. The emulator can virtually expand the array size to the practical level and calculate the energy consumption accurately. The rationality of the methodology has been proven by multiple groups. [ \n \n 35 \n , \n 36 \n , \n 37 \n \n ] For the energy‐efficiency examination, a fully‐connected double‐layer perceptron (DLP) neural network (constituting 784 input neurons, 256 hidden neurons, and 10 output neurons) was employed, and the MNIST dataset training task was performed, as illustrated in Figure 5f . The DLP was trained by the backpropagation algorithm, and the weight change was obtained by the stochastic gradient descent [ \n \n 38 \n \n ] optimization method. The network was trained with 60 000 training datasets using a batch size of 50 for 10 epochs. For the simulation, a bi‐directional weight update system was adopted, where positive and negative weight values were represented by two CTM cells, [ \n \n 35 \n , \n 37 \n \n ] making a total of 407 060 synaptic cells in the network. For the weight update, oneshot update method using the analog characteristic of Figure 1g as a look‐up table was used (see Figure S6 , Supporting Information, for the details of the neural network simulation and Figure S7 , Supporting Information, for the fitting parameters). At each weight update, each cell consumes energy corresponding to its biasing condition, which can be expressed as follows:\n \n (3) \n E tot pgm = E selected + E half − selected + E unselected \n \n \n (4) \n E selected = ∑ cell G i j ini + G i j fin 2 ( V pgm ) 2 n pulse t pulse \n \n \n (5) \n E half − selected = ∑ cell G i j ini ( V half − selected ) 2 n pulse t pulse \n \n \n (6) \n E unselected = ∑ cell G i j ini ( V unselected ) 2 n pulse t pulse \n where G i j ini and G i j fin are conductance values before and after weight update, V \n pgm , V \n half − selected , and V \n unselected are node voltages of the cells, n \n pulse is the number of potentiation pulses, and t \n pulse is the pulse time. Figure 5g shows the estimated MNIST recognition rate of the device for the training and test dataset for 10 epochs of simulation. Our simulation showed ≈91% accuracy, which is lower than the ideal network accuracy of ≈97% possible in the given DLP network. This is due to the non‐ideal linearity characteristics and the conductance variation of the analog states. [ \n \n 39 \n \n ] Li, C. et al. proposed that almost ideal accuracy can be achieved on a 1T1R memristive array, suggesting that linearity improvement is crucial for better accuracy. [ \n \n 40 \n \n ] Although our device is not ideal yet, the estimated accuracy is comparable to other hardware‐based simulation performances ranging from 70% to 94%. [ \n \n 41 \n , \n 42 \n , \n 43 \n , \n 44 \n , \n 45 \n \n ] Figure 5h summarizes the energy consumption for serial and parallel programming of the conventional half voltage scheme and those of the suggested voltage scheme of Figure 5e . At the conventional half voltage scheme and serial programming, the total energy consumption was calculated to 0.51 J, which can be reduced to 0.31 J (39% reduction) by introducing the parallel programming scheme. At the suggested biasing scheme, the total energy consumption decreased drastically from 0.66 to 0.19 J (71% reduction). This energy reduction is attributed to the increased number of half‐selected cells (right panel of Figure 5d ) biased to 1.3 V, suppressing the leakage current."
} | 8,847 |
38075822 | PMC10701864 | pmc | 259 | {
"abstract": "It is not uncommon for metals to corrode, causing the\nproperties\nof the material to be affected. Superhydrophobic materials have made\neffective advances in metal corrosion protection because they can\neffectively insulate liquids from being trapped on metal surfaces.\nIn this study, self-assembled films were formed using octadecanethiol\n(ODT) modification to obtain superhydrophobic as well as superoleophilic\nbifunctional materials. With a water contact angle (WCA) of 156°,\nthe material surface exhibits excellent self-cleaning properties.\nIt is also stable in highly corrosive environments. The good hydrophobicity\nof the material is due to the more tightly arranged conical structure\nand the ODT coatings of the treated copper mesh surface. The Cassie–Baxter\nequation calculations showed that the total exposed area of water\ndroplets in air (91.35%) is significantly higher than the area in\ncontact with metal surfaces. This work provides a new strategy for\nthe design of self-assembled surface-modified superhydrophobic materials\nwith excellent performance and stable properties by controlling the\nchemical composition and morphology of the material surface. The materials\nare prepared by avoiding cumbersome steps and the use of unusual materials\nand instrumentation, which allows our designs to greatly reduce the\neconomic costs of time, labor, and raw materials, and to facilitate\nlarge-scale industrial preparation and application. The prepared superhydrophobic\nand superoleophilic synergistic surfaces have excellent self-cleaning\nproperties, wetting stability, anti-corrosive properties, oil–water\nseparation properties, coagulation properties, and durability and\nhave a wide range of applications in the fields of anti-corrosion\nand seawater desalination.",
"conclusion": "4 Conclusions Superhydrophobic materials\nhave great application space and development\npotential in production and life; for example, the transparent coating\nof the finishing touches in life can reduce the retention of dust\nand stains, so that the architectural decoration maintains the aesthetic\nproperties. For metal materials, to carry out some processing to make\nthem more wear-resistant, reduce surface stains and water droplets\nbrought about by metal corrosion and other problems, and extend the\nservice life of metal material products, in the construction industry,\naerospace, ships, and industrial equipment has large application prospects.\nThe treatment of metal materials is divided into the treatment of\nthe surface layer of metal materials and the direct treatment of the\nbody of the material, in which only the surface layer of the material\nwill not change the characteristics of the body of the material, when\nthe failure of the surface coating material, you can still be the\noriginal treatment of the material, because the body of the material\nhas not changed, so the recycling of the material as well as the follow-up\nof the more advanced application of the material to provide a convenience\nin addition to the body of the material. The other is to carry out\na series of modifications to the material body so that the metal material\nloses the characteristics of the original pure metal, the preparation\nof the metal material as a whole shows hydrophobic qualities, the\noverall stability will be more stable, and it will not appear after\nthe coating failure of the material surface directly back to the original\nshape of the embarrassment. In summary, the superhydrophobic\noleophilic copper mesh was successfully\nprepared by using a self-assembly method. The differences in the oxidation\ntime of the copper mesh heated in CuSO 4 solution and in\nthe modification time of the modifier ODT/ethanol solution were investigated\nseparately to optimize the experimental results and to select the\noptimal experimental conditions. Observation of the surface of the\ncopper mesh after droplets rolled off from the modified mesh before\nand after the modification revealed that the modified mesh had a mirror\neffect, further demonstrating the self-cleaning properties of the\ncopper mesh. The stability of the material in acidic and alkaline\nenvironments was evaluated by recording the values of the contact\nangles at different pH levels, which all resulted in contact angles\nof >150°, also demonstrating the acidic and alkaline resistance\ncharacteristics of the material. The material was then tested for\nstability, and a NaCl solution was prepared to simulate the copper\nmesh within a seawater environment. Tafel curves were plotted, and\nit was found that the I corr values for\nthe copper mesh were 3 orders of magnitude smaller than those before\nthe reaction, indicating better corrosion resistance of the modified\nsurface. We chose to record and test the efficiency of the separation\nusing a solution of hexane and trichloromethane mixed with water,\nand the experimental results found that the copper mesh had a good\nhigh separation efficiency and recyclability.",
"introduction": "1 Introduction Copper (Cu) is a common\nengineering material widely used in industry. 1 However, in some corrosive environments, such\nas seawater, acid rain, 2 and prolonged\ncontact with biological liquids, 3 copper\nmay suffer from corrosion that affects its serviceability. 4 , 5 Traditional anticorrosion methods, including paints and corrosion\ninhibitors, 6 , 7 are harmful to human health and the environment.\nTo solve this problem, superhydrophobic surfaces have been prepared\non metal substrates to improve their corrosion resistance. 8 − 10 According to Young’s equation, solid surfaces with\nwater\ncontact angle (WCA) θ < 90° are hydrophilic; while solid\nsurfaces with WCA θ > 90° are hydrophobic, and superhydrophobic\nsurfaces have WCA θ > 150°. 11 There are two main traditional ways to obtain superhydrophobic surfaces: 12 increasing surface roughness 13 and decreasing surface energy. 14 According to these two principles, the main methods to prepare superhydrophobic\nsurfaces are etching, 15 , 16 hydrothermal, 17 anodization, 18 electrochemical\ndeposition, 19 − 21 milling method, 22 sol–gel\nmethod, 23 and in situ growth. 24 , 25 However, there are many methods proposed in the research with complicated\nsteps and experimental setups, such as ultrasonic etching technique\nto obtain copper mesh surfaces with parabolic-like rough structure\nby Qiao 26 et al., or limited by certain\nsubstrates and expensive raw materials, such as the sol–gel\nmethod for polyethylene glycol ethylene terephthalate (PET) filters\nusing ethyl orthosilicate (TEOS) as the precursor and perfluorodecyltriethoxysilane\n(PFDTES) as the modifier adopted by Dong 27 et al. Yuan 28 et al. used the backside\nof a fresh bamboo leaf as the original template, and by combining\nthe template on a copper foil, a regular multiscale layered structure\nwas prepared on the copper foil; after further treatment with stearic\nacid, the sample had superhydrophobic properties with a WCA of 160°.\nHowever, the corrosion resistance of this sample was not tested. Jinsong 11 et al. reported the preparation of superhydrophobic\nsurfaces on copper substrates by jet electrodeposition. However, this\nmethod is time-consuming and difficult to control precisely. A one-step\nelectrodeposition method has been used by Liu 29 et al. to fabricate cuprous oxide on copper substrates to obtain\nsuperhydrophobic surfaces, which is easy to operate; however, the\nsurface stability and the durability of the modified layer tend to\nbe too poor because water droplets can easily penetrate the rough\nstructure with the increase of immersion time. It is difficult to\ncontrol the uniformity and stability of superhydrophobic surfaces\nby focusing only on the roughness of the material surface. Therefore,\nit is of interest to create a thin and uniform film on the surface\nand then use it as a superhydrophobic material. In addition, due to\nthe thin and uniform self-assembled monolayer (SAM) film, it is often\nused as a model surface in research applications, 30 which can be designed as a material surface with different\nproperties according to the experimental requirements, so 1-octadecanethiol\n(ODT) are often adsorbed on the surface of copper substrates to form\nself-assembled monolayers used in superhydrophobic materials. Talesh\nBahrami 31 et al. first oxidized the copper\nmesh with alkali solution and then modified it with a 1-octadecanethiol\nsolution to prepare a self-assembled monolayer on the surface of the\ncopper mesh. However, strong bases are often harmful to the environment.\nIn this work, we used a simple heating oxidation method to soak the\noxidized copper mesh in ODT ethanol solution to produce a SAM superhydrophobic\nsurface and demonstrated its excellent corrosion resistance and stability\nthrough subsequent tests. The increase of oily effluents due\nto the large amount of petroleum\nextracts used in industrial production has made oil–water separation\na global task. 32 For superhydrophilic and\nsuperoleophilic mesh, oil-prewetted mesh is superoleophilic and superhydrophobic\nin oil to remove water; 33 according to\nthis property, we prepared copper mesh that can also be used for oil–water\nseparation, 34 which has important research\nvalue. During the separation of various oil–water mixtures,\nthe obtained copper mesh shows both extremely high separation efficiency\nand hydrostatic resistance. In addition, the excellent self-cleaning\nability and excellent stability enable copper mesh to be used in extremely\nharsh environments. The preparation process of superhydrophobic copper\nmesh is shown in Scheme 1 . Scheme 1 Schematic Illustration of the Superhydrophobic Surface",
"discussion": "3 Results and Discussion 3.1 Surface Morphology and Composition Analysis 3.1.1 SEM Analysis In order to further\ncharacterize the copper mesh morphology, SEM tests were performed\non the bare sample, treated sample, and modified sample, and the SEM\nimages are shown in Figure 1 . Figure 1 a,d\nshows the bare copper mesh morphology. The diameter of the copper\nwire in the copper mesh is about 450 μm. It can be seen that\nthe surface of the copper mesh is smooth. As shown in Figure 1 d, there is no other shaped\nmaterial on its surface. A study of the oxidized copper mesh (in Figure 1 b) revealed some\nrough material generated on its surface, and a local magnification\nof it (×5000k) revealed a prismatic conelike structure on its\nsurface, as shown in Figure 1 e. The particles on the surfaces are not uniform in size,\nranging from about 1 to 7 μm. The modified conical structure\nis Cu 2 O, which is consistent with the XRD analysis in Figure 3 , indicating the formation of a Cu 2 O material on the surface\nafter modification. The comparison of Figure 1 f with 1 e at the same\nmagnification shows that the surface structure of the material modified\nby the ODT/ethanol solution becomes small and compact at the microscopic\nlevel instead of the large prismatic structure of the material surface\nafter the second treatment step. It was found by XPS characterization\nthat the S element existed on the surface as shown in Figure 1 e,f. Figure 1 Images of SEM on different\nsamples: (a) bare mesh, (b) treated\ncopper mesh, (c) modified copper mesh, (d) enlarged view of the bare\nmesh, (e) enlarged view of the untreated copper mesh, and (f) enlarged\nview of the modified copper mesh. Figure 2 EDS spectra of the modified copper mesh surface. Figure 3 XRD spectra of the bare sample (a, black line), treated\nsample\n(b, red line), and modified sample (c, blue line). 3.1.2 EDS Analysis Element mappings were\nperformed in order to further analyze the composition of the modified\ncopper mesh. Figure 2 shows the main elemental composition of Cu and O. The distribution\nof each element on the surface of the material is relatively uniform.\nAccording to the EDS analysis, the content of Cu was 42.31% and the\ncontent of O was 33.91%. In addition, the modified copper mesh also\ncontained 23.22% C and 0.56% S, which was due to the introduction\nof ODT. 3.1.3 XRD Analysis In order to characterize\nthe phase structure and composition of the bare sample, treated sample,\nand modified sample, an X-ray diffraction (XRD) test was performed.\nThe results of XRD patterns are shown in Figure 3 . Figure 3 a shows the XRD of the bare copper mesh. It can be\nseen from the figure that there are three distinct diffraction peaks,\n2θ = 43.3, 50.5, and 74.3° corresponding to the (111),\n(200), and (220) crystal planes of pure copper, respectively (JCPDS\nno. 85-1326). Figure 3 b shows an XRD image of the oxidized copper mesh. The data analysis\nreveals that in addition to the diffraction peaks of pure copper,\nthere are also diffraction peaks of cuprous oxide. (110), (111), (200),\n(220), and (311) of Cu 2 O correspond to 2θ = 29.6,\n36.5, 42.4, 61.5, and 73.6°, respectively (JCPDS no. 75-1531). Figure 3 c shows the XRD image\nof the modified sample. The data analysis reveals that there are also\ndiffraction peaks of pure copper and cuprous oxide. At 2θ =\n29.6, 36.5, 42.4, 61.5, and 73.6° (JCPDS no. 75-1531), there\nare diffraction peaks corresponding to the (110), (111), (200), (220),\nand (311) crystal planes of Cu 2 O, respectively. It shows\nthat the phases contained in the treated sample are Cu and Cu 2 O. Upon comparison of the intensity of the peaks of the XRD\nimage of the material surface before and after the modification, it\ncan be seen that the intensity of each peak becomes lower after the\nmodification, indicating that the ODT modification had an effect on\nthe material surface. 3.1.4 XPS Analysis The chemical composition\nand surface element valence states of the copper mesh before and after\nmodification were further analyzed by XPS, as shown in Figure 4 . Figure 4 a shows the full spectrum of the copper mesh\nbefore and after modification. The spectrum clearly shows four peaks\nof Cu 2p, C 1s, S 2p, and O 2s, which proves that the copper mesh\nafter modification contains four elements of Cu 2p, C 1s, S 2p, and\nO 2s. The results are consistent with Figure 2 element mapping. In the spectrum of Cu 2p,\nthere are two distinct peaks at 932.1 and 951.9 eV, corresponding\nto Cu 2p 3/2 and Cu 2p 1/2 , respectively. Figure 4 c shows the peaks\nat binding energies of 284.2 and 284.7 eV corresponding to –CH 2 and –CH 3 . It indicates the molecular existence\nof –CH 2 – on the surface of the superhydrophobic\nmaterial obtained after immersion in octadecanethiol/ethanol solution.\nIn Figure 4 d, there\nremains a peak at 162.4 eV corresponding to S 2p. It can finally be\ndemonstrated that the material surface has changed after heating and\noxidation and soaking in an ODT/ethanol solution, combined with the\nprevious elemental analysis of the copper mesh surface by EDS. Figure 4 XPS spectral\nanalysis: (a) total energy spectrum of the copper\nnetwork before and after modification, (b) 2p orbital spectrum of\nCu, (c) 1s orbital spectrum of C, (d) 2p orbital spectrum of S, and\n(e) 2s orbital spectrum of O. 3.2 Performance Analysis 3.2.1 Wettability Figure 5 shows the images and measured contact angles\n(CA) of droplets falling on the surface of the copper mesh before\nand after different treatments, respectively, where (a–c) shows\nthe water contact angle (WCA) of the copper mesh subjected to different\ntreatments, and the green box in the upper right corner shows the\nsurface visualization of the copper mesh. (d) Shows the contact angle\nof an oil droplet (OCA) falling on the surface of the modified copper\nmesh. Figure 5 a shows\nthe water droplet diagram on the surface of the pure copper mesh with\na measured WCA of 107°, indicating that the surface of the copper\nmesh itself is hydrophobic, and then the copper mesh was heated in\nCuSO 4 solution to undergo an oxidation reaction to obtain\nproduct (b), which has a WCA of only 8°, indicating that the\nsurface of the material has become hydrophilic. Finally, the copper\nmesh was modified with ODT solution, and after the reaction in the\nprevious step, a superhydrophobic surface with a WCA of 156°\nwas obtained (c), and the contact angle with n -hexane\nwas almost 0° (d). The comparison of the WCA data shows that\nthe material surface changes from superhydrophilic to superhydrophobic\nin one step, indicating that ODT that is loaded on the Cu network\nafter the reaction is an important factor for its superhydrophobic\nperformance. Figure 5 (a) WCA on bare copper mesh surfaces, (b) WCA on oxidized\ncopper\nmesh surfaces, (c) WCA on superhydrophobic copper mesh surfaces, and\n(d) OCA with hexane on superhydrophobic copper mesh surfaces. Cassie and Baxter suggested that a rough, inhomogeneous\nsolid surface\ncan be conceived as a composite surface. In general, the formula for\ndescribing a composite surface is 35 , 36 where θ* is the contact angle of gas–liquid\non the surface, θ is the contact angle of the smooth surface, f 1 represents the area fraction of the solid\nat the composite interface, and f 2 is\nthe area fraction between air and water on the same surface. Both f 1 and f 2 are less\nthan 100% and they have the relationship f 2 = 1 – f 1 , so the equation can\nbe transformed into a new equation θ* = 156° and θ = 107°\nwhen the water droplets are on the modified mesh surface. The equation\ncan be used to obtain f 1 = 8.65%, i.e.,\n91.35% of the area of the droplet in air and 8.65% of the surface\nof the rest of the droplet in contact with the solid surface. This\nmeans that the surface area of water droplets is 91.35% of the total\nexposed area in air, and the area of water droplets in air is significantly\nhigher than the area in direct contact with the metal material. The\ncalculated area percentage also shows that the copper mesh has good\nhydrophobicity after the ODT modification. 3.2.2 Self-cleaning Ability In order\nto test the self-cleaning ability of the superhydrophobic material,\nwe performed superhydrophobic tests, and the results are shown in Figure 6 . Finely ground dust\nwas sprinkled on the surface of the pure copper mesh and the superhydrophobic\noleophilic copper mesh, respectively, and then droplets of water were\nslowly dropped onto the surface of the copper mesh at a certain height\nabove the mesh to observe the surface of the mesh after the droplets\nrolled off from these two different types of mesh. As shown in Figure 6 a, pictures of dust\nspilled on a pure copper mesh are shown, and it can be seen that when\nthe water droplet falls on the surface of the copper mesh, the droplet\nand the dust mix to remain on the surface of the mesh, as the water\ndroplet gradually falls, eventually the surface of the pure copper\nmesh remains stained, indicating that the pure copper mesh has no\nself-cleaning ability. Figure 6 b shows the superhydrophobic oleophilic copper mesh during\nthe test, in these three pictures, it can be observed that the drops\nof water as they roll down from the surface of the material will take\naway the dust on the surface of the material at the same time, as\nlong as there are drops of liquid rolling down, all remain clean,\nfinally as the drops of liquid completely roll down, the surface of\nthe superhydrophobic oleophilic copper mesh is in a completely clean\nstate, in line with the characteristics of superhydrophobic materials, 37 showing good self-cleaning. The surface of the\nsuperhydrophobic copper mesh was completely clean, as the droplets\nrolled off. Overall, the modified copper mesh is self-cleaning compared\nto the pure copper mesh. Figure 6 Self-cleaning test: (a) bare copper mesh and\n(b) superhydrophobic\ncopper mesh. 3.2.3 Mirror Effect The phenomenon that\nthe modified surface shows a silver color is called the mirror effect.\nThis is due to the fact that the surface is not in direct contact\nwith water, with a layer of gas spacing in between, 38 , 39 and the refraction of light causes the surface color to be seen\nby the naked eye to be different from the original copper mesh. The\nrelatively thin air prevents wetting between the liquid and the solid\nsurface and significantly reduces the actual contact area between\nthe liquid and the substrate. A protective air cushion is formed on\nthe surface, which ultimately inhibits corrosion effectively. 40 Thus, good superhydrophobicity and corrosion\nprotection can also be inferred from the mirror effect. The mirror\neffect was tested, and the results are shown in Figure 7 . A beaker was filled with distilled water,\nand from Figure 7 a,b,\nit can be seen that the surface of the water in contact with the copper\nmesh is clearly below the horizontal and has the effect of moving\ndownward, while there is a clear difference in the color of the mesh\nabove and below the surface of the water, with the mesh above the\nsurface of the water being black and the mesh appearing below the\nsurface of the water showing a bright silver color. Figure 7 Pictures of the mirror\neffect on the modified surface. Pictures\nof the process of gradually placing a black copper mesh underwater\nusing tweezers (a,b); picture of copper mesh pressed completely underwater\n(c). 3.2.4 Chemical Stability As it is unlikely\nthat the everyday use of the material will be in a completely neutral\nenvironment, it was necessary to test the durability of the experimentally\nprepared copper mesh at different pH levels. Therefore, in order to\nanalyze the chemical stability of the materials under different acidic\nand alkaline conditions, we tested the WCA values of the superhydrophobic\nlipophilic products under different pH conditions, and the results\nare shown in Figure 8 . The experiments were carried out by taking pH values of 1, 3, 5,\n7, 9, 11, and 13. The products were immersed in the corresponding\nacid–base solution for 24 h, removed, and cleaned, and the\nWCA of the surface was measured. Figure 8 WCA of the material surfaces at different\npH levels. In Figure 8 , from\nthe change in the WCA with the pH of the solution, it can be seen\nthat the contact angle between the surface of the experimentally prepared\ncopper mesh and the water always remains above 150°, possessing\na stable corrosion resistance in different acidic and alkaline environments,\neffectively widening the environmental conditions to which the material\ncan be applied. This resistance to acid and alkali is also a testament\nto the mirror phenomenon observed between the surface of the material,\nwhere the presence of an air layer between the metal surface and the\nsolution prevents the acid and alkali solutions from coming into direct\ncontact with the surface of the material, thus giving the surface\nexcellent chemical stability. 3.2.5 Corrosion Resistance In order to\ntest the corrosion resistance of the materials, we simulated a seawater\nenvironment and tested the Tafel polarization curves of pure copper\nand its products in 3.5 wt % NaCl solution, and the results are displayed\nin Figure 9 . The corrosion\nresistance of the material was characterized by the corrosion potential\n( E corr ) as well as the current density\n( I corr ). The E corr and I corr of the pure copper mesh were\n−0.157 V and 3.519 × 10 –7 A/cm 2 , while those of the modified copper mesh were −0.264 V and\n3.769 × 10 –10 A/cm 2 , respectively.\nThe corrosion current represents how quickly the material is corroded;\nthe higher the corrosion current, the faster the material corrodes.\nThe I corr of the superhydrophobic material\nis 3 orders of magnitude smaller than that of the pure copper mesh.\nIn the electrochemical preparation of copper-based superhydrophobic\nstructures by Zhang et al., the self-corrosion current density of\nsuperhydrophobic-treated copper in 3.5% NaCl solution was reduced\nby about 2 orders of magnitude compared to that of untreated copper\n(from 5.3 × 10 –6 to 4.2 × 10 –8 A/cm 2 ). 19 Jiang et al. successfully\nformed agglomerated micro-nanostructures on modified copper coatings\nto obtain superhydrophobic surfaces using copper stearate. Corrosion\ntests were carried out on bare copper and superhydrophobic coatings\nin 3.5 wt % NaCl solution. The corrosion current density of the superhydrophobic\ncoating was the lowest at 4.28 × 10 –7 A/cm 2 compared to the corrosion current density of bare copper\n(2.27 × 10 –6 A/cm 2 ). 14 Jia et al. used a 0.1 mm flat-bottom sharp knife\nto mill the surface of a Cu substrate in a CNC engraving machine to\nconstruct the microstructure of rectangular bumps. A superhydrophobic\nsurface was obtained after the stearic acid modification. Its test\ndata showed that the I corr of the superhydrophobic\ncopper samples was generally reduced by an order of magnitude compared\nto the Cu substrate. 22 Compared with the\nprevious research results, the changes in I corr were more pronounced in the present study. According to the above\nresults, the superhydrophobic surface prepared by the present method\nhas better corrosion resistance. Figure 9 Potentiodynamic Tafel polarization curves\nof the bare (a, black\nline) and modified (b, red line) sample in 3.5 wt % NaCl aqueous solution. 3.2.6 Oil–Water Separation Efficiency In order to test the oil–water separation efficiency, we\nconducted an oil–water separation experiment, and the results\nare shown in Figure 10 . The superhydrophobic copper mesh was folded into the shape of an\nopen rectangle. The length, width, and height of the folded rectangle\nwere about 2, 2, and 1 cm, respectively. As shown in Figure 10 a, as the superhydrophobic\noleophilic copper mesh was placed into the Petri dish, the orange-colored\noil near the mesh began to disappear gradually from the copper mesh\nand the oily reagent disappeared rapidly. As can be seen in Figure 10 c, there are only\na few drops of oil on the water surface. From Figure 10 d, you can see that the copper mesh has\nabsorbed the oil from the water surface, and you can see that the\nfinal water surface is in a clean state. In order to make the experimental\nconclusions more concrete, in this experiment, we chose to mix heavy\noil and light oil with water separately, and then discuss their separation\nefficiency. The experimental setup is shown in Figure 10 g, where the height of the copper mesh to\nwithstand water pressure is at least 15 cm. Experiments have proved\nthat the modified copper mesh has good resistance to water pressure. Figure 10 Images\nof the different stages of oil droplet absorption by the\ncopper mesh prepared for the oil–water separation test (a–f).\nImage of the experimental setup for the oil–water separation\ntest (g). 3.2.7 Cyclic Stability Performance To\nfurther test the oil–water separation properties, we tested\nthe efficiency of different mesh sizes of copper mesh for oil–water\nseparation in hexane and trichloromethane. Stained n -hexane and trichloromethane with Sudan III and mixed with water.\nThe oil–water separation efficiency was recorded using 60,\n150, and 250 mesh copper nets, respectively. One point to note is\nthat during the test, as hexane is less dense than water and tends\nto float on the water surface, it is important to keep the experimental\napparatus tilted throughout the measurement process as the droplets\nfall with regard to penetration of the oil droplets. Also, given the\nvolatile nature of organic matter, the whole experimental process\nshould be kept efficient and the results should take into account\nthe errors caused by volatilization of organic matter and the possible\nfall of trace amounts of the oil–water mixture. In Figure 11 a, the efficiency\nof the oil–water separation in hexane tested with three different\nmesh sizes of copper mesh can be seen to remain above 90% as the number\nof cycles increases, and the separation efficiencies all remain within\na reasonable margin of error. Figure 11 b shows the separation efficiency of the copper mesh\nfor trichloromethane, and the bar graph shows that the overall situation\nis more balanced, compared to (a), which shows that the overall efficiency\nfor separating trichloromethane is more stable, probably due to the\nvolatility of hexane which can easily cause fluctuations in values. Figure 11 Three\ndifferent mesh sizes of copper mesh tested for the efficiency\nof oil–water separation in an n -hexane solution\n(a). Three different mesh sizes of copper mesh to test the efficiency\nof oil–water separation in trichloromethane solution (b)."
} | 7,055 |
26714176 | PMC4694616 | pmc | 260 | {
"abstract": "We report a methodology for enhancing the mass transfer at the anode electrode of sediment microbial fuel cells (SMFCs), by employing a fabric baffle to create a separate water-layer for installing the anode electrode in sediment. The maximum power in an SMFC with the anode installed in the separate water-layer (SMFC-wFB) was improved by factor of 6.6 compared to an SMFC having the anode embedded in the sediment (SMFC-woFB). The maximum current density in the SMFC-wFB was also 3.9 times higher (220.46 mA/m 2 ) than for the SMFC-woFB. We found that the increased performance in the SMFC-wFB was due to the improved mass transfer rate of organic matter obtained by employing the water-layer during anode installation in the sediment layer. Acetate injection tests revealed that the SMFC-wFB could be applied to natural water bodies in which there is frequent organic contamination, based on the acetate flux from the cathode to the anode.",
"conclusion": "Conclusion The maximum current density (220.46 mA/m 2 ) and maximum power density (69.14 mW/m 2 ) of an SMFC having a separate water layer (SMFC-wFB) in the anode sediment were 3.9 and 6.6 times higher than the respective values for an SMFC having no water layer (SMFC-woFB). It was found that the organics transfer rate in the separate water layer was much more facile compare to that for the sediment layer; as a result, the anode kinetic activity in the water layer was enhanced and the performance in the SMFC-wFB substantially increased. It is expected that this water layer in the anode structure for SMFCs can be an efficient way to facilitate the substrate transfer to an anode embedded in sediment, leading to further increases in the power output in SMFCs.",
"introduction": "Introduction Sediment microbial fuel cells (SMFCs) are being considered for use as a power source for aquatic water quality sensors such as pH, temperature, and dissolved oxygen sensors [ 1 , 2 , 3 ]. The greatest benefit in using SMFCs is that they utilize organic matter that is distributed in a natural aquatic environment as the fuel source for generating electricity [ 4 , 5 , 6 ]. Typically, for SMFC installation, the cathode electrode is exposed to the oxygen-rich aqueous phase, and the anode electrode is embedded in the organic-rich sediment without a membrane [ 7 , 8 , 9 ]. Electricity can then be produced based on the electron production from electrochemically active bacteria (EAB) or sulfate-reducing bacteria that use organic matter in the sediment as the electron donor [ 10 ]. However, it has been reported that the organic content in sediment is as low as 0.4% to 2.2% [ 11 ] and thus the anodes in SMFCs likely suffer from mass transfer limitations [ 12 ]. To overcome this inherent constraint, Rezaei et al. (2007) proposed the use of chitin and cellulose as an assistant substrate; this approach improved the power density in SMFCs by approximately 40 times, though further study is required regarding the installation and fabrication of anode electrodes before this approach becomes practical [ 13 ]. Shantaram et al. (2005) previously demonstrated the potential for utilising a manganese alloy as an electron donor, a so-called sacrificial anode, which led to a significant increase in the power output [ 14 ]. However, there is controversy over the classification of this abiotic anodic reaction-based fuel cell, which includes corrosion of the manganese alloy, as a microbial fuel cell. In more recent attempts to enhance the redox reactions of sulphate and sulphide, an intriguing concept for reforming the anode using anthraquinone-1,6-disulfonic acid was suggested by Lowy et al. (2008); they confirmed that a power density of 100 mW/m 2 to 110 mW/m 2 can be achieved, but the durability of the functional group on the anode electrode was not ensured [ 15 , 16 ]. Furthermore An et al. (2013) proposed a simpler and easier approach for utilising microbial physiological characteristics to enhance both the current density and working voltage for a single SMFC [ 17 , 18 ]. In their study, a maximum power density of 14.5 mW/m 2 was observed at a sediment depth of 10 cm, which was 2.2 times higher than could be obtained at a sediment depth of 2 cm. However, this method requires the use of preliminary tests to determine the optimal anode depth for the SMFCs. The mass transport of dissolved organics in sediment generally occurs by diffusion, with the diffusion rate of dissolved organics in sediment being much slower than in water phase. The physicochemical properties of the sediment that decrease the mass transport rate of dissolved organics include porosity, tortuosity, pore size, etc. Nonetheless, the effect of organic transfer rate in sediment on the performance of SMFCs has yet to be reported in literature. We believe that employing a separate water-layer during installation of the SMFC anode could facilitate the organic flux to the anode electrode, and subsequently improve the power output by increasing the anode kinetic activity of the MFCs. In this work, we demonstrate an “electrode-spacing method” which is to create a separate water layer surrounding the anode electrodes by using a fabric baffle inside the sediment, to enhance the organic transfer rate at the anode electrode. We also investigate the behavior of organic transfer in two different phases, sediment and water phase. We find that the water-layered anode structure for SMFCs (SMFC-wFB) was remarkably effective in terms of power and current increases, such that it could be an efficient way to facilitate the substrate transfer to an anode embedded in sediment, leading to an increase in the overall power output in SMFCs.",
"discussion": "Results and Discussion Higher performance in SMFC-wFB than -woFB \n Fig 2 presents the OCV and CCV developments in SMFC-wFB and-woFB which were installed with sediment-wAc (see Materials and Methods ). At day 2, the OCV stabilized at 0.82 ± 0.02 V in SMFC-woFB and 0.81 ± 0.01 V in SMFC-wFB. After the MFCs were changed to closed circuit mode by employing a 5 kΩ external resistance, the current in the MFCs stabilized at 0.07 mA for 0.35 V (SMFC-woFB) and 0.14 mA for 0.72 V (SMFC-wFB) ( Fig 2 ). 10.1371/journal.pone.0145430.g002 Fig 2 Voltage measurements of SMFC-woFB and SMFC-wFB under closed circuit mode at a 5 knd SMFC-wFBls CHas CHteas analyzed by conver The arrow indicates when the operation mode was changed to CCV. To confirm whether or not the performance of SMFC-wFB was superior to SMFC-woFB through all external resistors, we performed polarization tests on the MFCs. As shown in Fig 3A , the maximum power and current density were higher in SMFC-wFB (69.1 mW/m 2 for 220 mA/m 2 ), as compared to those in SMFC-woFB (10.5 mW/m 2 for 57.1 mA/m 2 ). The overvoltage for the current of 0.2 mA was as small as 0.09 V in SMFC-wFB, whereas SMFC-woFB had high overvoltage of 0.53 V at the identical current (see Fig 3A ). There could be two possible reasons for the different performances in SMFC-wFB and SMFC-woFB. First, it could be due to the difference in the internal resistance of the MFCs; and second, it could be due to the difference in the anode kinetics of the MFCs because all operational conditions for the cathode in the SMFCs were identical. As seen in Table 1 , the ohmic resistance between the anode and cathode of SMFC-woFB was observed to be 207 ± 0.7 Ω whereas the resistance for SMFC-wFB was slightly higher at 228 ± 0.1 ΩwaThe ohmic resistance between the cathode and top sediment layer surface in the MFCs was similar at 35.8 Ω. The 1 cm water layer in the middle of the sediment could be responsible for the slightly higher ohmic resistance observed in the SMFC-wFB (see Fig 1 ). The higher ohmic resistance in SMFC-wFB does not support the argument that the higher performance in SMFC-wFB might be due to lower ohimc resistance in SMFC-wFB than in -woFB. 10.1371/journal.pone.0145430.g003 Fig 3 (A) Performance curves for the SMFC-woFB and SMFC-wFB installed with sediment-wAc; (B) electrode potential plots for anode and cathode monitored during the polarization tests for the MFCs. 10.1371/journal.pone.0145430.t001 Table 1 Internal resistances of SMFCs obtained from electrochemical impedance spectroscopy: internal resistances between the water-sediment interface and anodes and between the cathodes and the water-sediment interface. Internal resistance (Ω) Water-sediment interface to anode Cathode to water-sediment interface Total internal resistance SMFC-woFB 171 ± 0.3 35.8 ± 0.03 207 ± 0.7 SMFC-wFB 192 ± 0.7 35.8 ± 0.01 228 ± 0.1 To investigate whether the better performance in SMFC-wFB was due to the higher anodic kinetic activity, the evolution of electrode potentials in SMFC-wFB and -woFB were monitored during the polarization tests for the MFCs (see Fig 3B ). In the figure, the anode and cathode open circuit potentials of the MFCs were similar at -0.45 and 0.37 V, respectively; the anode and cathode overpotential in the both types of SMFCs then increased by lowering the external resistors. The anode overpotential at the current of ~0.2 mA in SMFC-woFB was substantial at ~-0.48 V, whereas SMFC-wFB had small anode overpotential of -0.09 V at the same current. In comparison, the cathode overpotential in SMFC-wFB was slightly smaller at -0.02 V than the anode overpotential in SMFC-wFB, and the cathode overpotential of SMFC-woFB was much smaller at -0.09V compared to the anode overpotential in the MFC. It was observed that 90.7% of the total overvoltage for the current of 0.2 mA in SMFC-woFB was responsible for the anode overpotential of 0.48 V. These results clearly indicate that the anode kinetic activity in SMFC-wFB was much more facile compared to that in SMFC-woFB. The initial acetate concentration of the reservoir water used to create the water layer for the SMFC-wFB was lower than the detection limit of HPLC (see Materials and Methods ), but after 2 days of operation the liquid sample collected from the same water layer had a 3.1 mM acetate concentration. It is posited here that acetate is being eluted from the sediment through the fabric baffle. However, the COD for 3.1 mM acetate (5.5 mg COD / g liquid) is 10.7 times lower compared to that contained in the sediment (57 mg / g sediment), which means that though SMFC-woFB had more substrate it did not produce a higher performance than SMFC-wFB. The higher performance in the SMFC-wFB, which had a much lower COD, implies that there was a much more facile mass transfer in the water layer than that in the sediment layer. To confirm this interpretation, we injected acetate in the cathode solution and monitored the current evolution of both SMFC-woFB and -wFB installed in the sediment-woAc (see Materials & Methods for more details). The diverse organics such as acetate, propionate, lactate, pyruvate and butyrate in sediment is an electron source for electrochemically active bacterial (EAB) or sulfate-reducing bacteria (SRB) which are directly or indirectly involved in electricity generation of SMFCs [ 18 , 22 , 23 ]. These bacteria are widely distributed in most sedimentary environments [ 24 , 25 , 26 ]. For this reason, we believed that EAB (or SRB) could oxidize the acetate that was added to sediment-woAc, which was discussed in the following section. Substrate diffusion rate in SMFC-woFB and -wFB \n Fig 4A presents the evolution of OCV and CCV (at 2 kΩ) in SMFC-wFB and -woFB after the injection of acetate (30mM) into the catholyte. The currents in SMFC-wFB and -woFB stabilized at 0.28 mA and 0.24 mA, respectively, similar to the initial current after a couple of hours. However, after 12 h, the current in SMFC-wFB gradually increased and then was saturated at 0.296 mA at 6 h prior to that of SMFC-woFB. In comparison, the current in SMFC-woFB started increasing 2 h later than that of SMFC-wFB and became saturated at 0.261 mA after 8h from the acetate injection. As presented in Fig 4 , during the increase in the current of the MFCs the anode potential decreased from -0.35 V to -0.38 V in SMFC-wFB and -0.22 V to -0.30 V in SMFC-woFB. In contrast, the change of the cathode potential (cathode overpotential) in SMFC-wFB and -woFB was relatively negligible at 0.21 V and 0.23 V. From these results, it is clear that the acetate in SMFC-wFB reached the anode earlier than that of SMFC-woFB, indicating that the acetate diffusion rate in the 1 cm of the water layer for the anode of the SMFC-wFB was much more facile when compared to an identical thickness of sediment layer (i.e., 1 cm). 10.1371/journal.pone.0145430.g004 Fig 4 Aceate injection tests for the SMFC-woFB and SMFC-wFB constructed with sediement-woAc; OCV and CCV evolution (2 kΩ) in the MFCs (A), development of cathode potential (B) and anode potential (C) in the MFCs; the arrows with the letters “A” and “B” mark the acetate injection points and the resulting current response points respectively. However, the current increment of SMFC-wFB by the acetate injection to the cathode could be considered to be negligible. To access the time taken for acetate diffusion from the cathode to the water layer, we performed a separated acetate injection test using sediment and a fabric baffle with no anode and cathode electrode (see Fig 1C for experimental details). The initial acetate concentration of the water layer was lower than the detection limit of HPLC even after a couple of hours operation ( Fig 5 ). Six hours after acetate injection to the center of the sediment-FB the acetate concentration in the water layer between the fabric baffle and the lower sediment started increasing, and then was saturated at 3 mM in another 6 h (Figs 1C and 5 ). From this result, it can be found that the current response to the acetate injection, shown in Fig 4 , was 6h later than acetate arrival to the water layer, suspecting that acetate diffusion rate via the 1 cm of the water layer in SMFC-wFB could be very slow when considering current response time. However, the late current response at the earlier acetate arrival to the water layer could be attributed to number of EAB activity or their population density on the anode electrode. 10.1371/journal.pone.0145430.g005 Fig 5 Change of acetate concentration in the water-layer after 30 mM acetate injection to the cathode of the SMFC-FB installed with sediment-woAc. To investigate the acetate diffusion rate, SMFC-wFB installed in sediment-woAc (but not SMFC-wFB constructed with sediment-wAc) was used for acetate injection test. Accordingly, we suspected the most EAB on the anode of the SMFC-wFB, installed in the sediment-woAc, might be not favorable to acetate utilization [ 23 ]. It is well known that some EAB such as Geobacter spp. can use acetate as an electron donor [ 15 , 16 ]. Thus, it was believed that a sulfide-sulfur-sulfate cycle [ 27 , 28 ] which is another pathway for electricity generation of SMFCs could be involved in the anode of SMFC-wFB installed in the sediment-woAc. Sulfate reducing bacteria (SRB) and sulfate-oxidizing bacteria (SOB) have a key role in the cycle; sulfides from SRB donate electrons to electrodes and are oxidized to elemental sulfur on the anode electrode, and then are further oxidized into sulfate by SOB. These interpretations stated above could be supported with the distinct increase in the current by the acetate injection to the SMFC-wFB installed in the sediment-wAc (see Fig 6 ). 10.1371/journal.pone.0145430.g006 Fig 6 Change of current density and acetate concentration on SMFC-FB by injection of acetate. Arrow indicates acetate injection into the catholyte at the water-sediment interface in SMFC-wFB. No acetate injected SMFC-wFB was also operated as a control set. As seen in Fig 6 , on injection of 30 mM acetate to the catholyte of the SMFC-wFB installed in the sediment-wAc the current significantly increased from 1.05 to 1.19 mA, 10 times higher than the SMFC-wFB constructed in sediment-woAc (also see Fig 4A ). The small and sluggish current response to the acetate injection in the SMFC-wFB installed in the sediment-woAc could be due to a small population of EAB that directly use acetate. These results support the hypothesis that the organic transfer rate via the separate water layer in the SMFC-wFB is much more facile compared to the continuous sediment layer in the—woFB analogue. Finally, to propose the physical explanation about how the water-layer makes the difference in the anode kinetic, Fig 7 was added that molecular diffusion is described by Fick's first law, which explained that solute movement rate is proportional to the spatial concentration gradient, given, in one dimension, by\n J x = − D m × ( δ C / δ x ) (1) \nwhere: J x is the molecular diffusive flux in the x-direction, D m is the molecular diffusion coefficient and ∂C/∂x is the tracer concentration gradient in the x-direction. In case of solutes in water, molecular diffusion value typically range between 0.5 to 2.0 /10 9 · m 2 /s and are determined empirically [ 29 ]. However, the coefficient of molecular diffusion through sediment is different from that through free fluid. The solute must flow around the sediment particles, creating a longer flow path and thereby effectively reducing the coefficient. The molecular diffusion coeffcient through sediment (D′ m ), has been explored theoretically [ 30 ], as well as empirically by relating it to sediment tortuosity that is related to sediment porosity (θ), by several studies [ 31 ]. This leads to the general expression for molecular diffusion in sediments\n D ′ m = β × D m (2) \nwhere: β represents an empirical expression for tortuosity as a function of sediment porosity and is described by Iversen and Jrgenson [ 32 ] as\n β = 1 / ( 1 + n ( 1 + θ ) ) (3) \nwhere: n is the sediment type constant (n = 3 for clay-silts and n = 2 for sands). However, n = 3 was used by O'Connor and Harvey [ 33 ] throughout their analysis, regardless of sediment type. The sediment porosities (θ) measured are slightly higher than the theoretical value of 0.37 for randomly placed spheres [ 34 ]. Therefore, diffusion coefficient of material around anode area in SMFC-woFB and—wFB could be only one important parameter that determines the substrates transfer rate to anode because other parameters for molecular diffusive flux (i.e. sediment tortuosity, tracer concentration gradient) were same due to all operational conditions in the SMFCs were identical; as a results, we could expect SMFC-wFB which has 1 + n (1 + θ) times higher diffusion coefficient than SMFC-woFB has enhanced mass transfer rate for kinetically favorable tor anodphiles and electrophiles. 10.1371/journal.pone.0145430.g007 Fig 7 Physical explanations of difference in the molecular diffusion rate in water and sediment-layer on the anode described by Fick’s first law."
} | 4,701 |
38371809 | PMC10870754 | pmc | 262 | {
"abstract": "Superhydrophobic surfaces have been studied extensively\nover the\npast 25 years. However, many industries interested in the application\nof hydrophobic properties are yet to find a suitable solution to their\nneeds. This paper looks at the rapid functionalization of nanoparticles\nand the fabrication of superhydrophobic surfaces with contact angles\n> 170°. This was achieved by simply mixing commercial products\nand applying the new formulation with scalable techniques. First,\ninexpensive and nontoxic superhydrophobic nanoparticles were made\nby functionalizing nanoparticles with fatty acids in under an hour.\nA similar methodology was then used to functionalize a commercial\npolymer coating to express superhydrophobic properties on it by lowering\nthe coating’s surface energy. The coating was then applied\nto a surface by the spray technique to allow for the formation of\nhierarchical surface structures. By combining the low surface energy\nwith the necessary roughness, the surface was able to express superhydrophobic\nproperties. Both the particles and the surfaces then underwent characterization\nand functional testing, which, among other things, allowed for clear\ndifferentiation between the functionalization properties of the zinc\noxide (ZnO) and the silica (SiO 2 ) nanoparticles. This paper\nshows that suitable superhydrophobic solutions may be found by simple\nadditions to already optimized commercial products.",
"conclusion": "4 Conclusions This paper has shown a\nquick and simple route to an inexpensive,\nnontoxic, and accessible superhydrophobic surface. This was obtained\nby functionalizing a commercial polyurethane product with readily\navailable fatty acids and ZnO nanoparticles using a spray technique.\nSurfaces were shown to have excellent superhydrophobic properties,\nwith some x̅ WCAs exceeding 170°, as well as antistaining\nand self-cleaning properties. We have also explained the importance\nof using a polar nanoparticle when attempting to functionalize particles\nquickly and with minimal steps. While we were able to functionalize\na commercial polyurethane product,\nthe same formulation was not compatible with a commercial epoxy resin\nproduct, despite evidence of epoxy resins being functionalized in\nthe literature under similar conditions. The failure of the epoxy\nresin solution may be due to the specific formulation of the polymer\nor may be due to an additive in the commercial formula interfering\nwith one of the reagents being used in the functionalization process.\nDespite this, this concept has been shown to be promising, yet more\nwork is still required before the product can overcome all the limitations\nlisted by Manoharan and Bhattacharya 39 and\nprove to be commercially viable. If these limitations can be overcome,\nthis concept could produce superhydrophobic surfaces rapidly through\nspray deposition at room temperature.",
"introduction": "1 Introduction Superhydrophobic surfaces,\na surface with a water contact angle\n(WCA) > 150°, may have applications as water repellent surfaces,\nself-cleaning surfaces, anti-icing surfaces, antibiofouling, etc. 1 − 3 Since the 1950s, multiple drug-resistant pathogens have begun to\nemerge, predominantly due to our over-reliance on antibiotics. 4 A continued arms race between antibiotics and\npathogens led the World Health Organization to declare drug-resistant\npathogens as one of the top global public health threats facing humanity. 5 Superhydrophobic surfaces have been shown to\nreduce the spread of pathogens through the prevention of colonization\nof the surface, reducing the need for antibiotic use. 6 While these properties are desired by the textile and health\nindustries for obvious reasons, self-cleaning and anti-icing properties\nare also required by the energy industry. As we continue to move away\nfrom fossil fuels, the efficiency and cost of renewable energy production\nmust be maximized. Self-cleaning and anti-icing surfaces may be part\nof the solution as they could significantly increase the efficiency\nand reduce the maintenance costs of isolated solar cells and wind\nturbines. 7 − 9 The investigation of superhydrophobic surfaces\npredominantly began\nwith Barthlott and Neinhuis’s report on the lotus leaf in 1997. 10 Since then, many have studied and fabricated\nsurfaces that are water-repellent and often biomimetically inspired. 11 − 13 As with superhydrophobic surfaces in nature, the properties of fabricated\nsuperhydrophobic surfaces are influenced by both their surface topography\nand surface free energy. 14 An example of\nthis is the lotus plant, which achieves superhydrophobic properties\nand self-cleaning properties by combing a low-energy waxy layer called\nthe cuticle with a rough surface caused by topological microstructures. 15 Both top-down and bottom-up strategies\nhave been investigated when\nconsidering the fabrication techniques for superhydrophobic surfaces.\nPlasma etching, 16 , 17 photolithography, 18 , 19 and electron beam lithography 20 , 21 are all excellent examples\nof top-down techniques that have been used to produce superhydrophobic\nsurfaces. While these techniques have excellent tunability and accuracy,\nthey are expensive and lack the scalability required when considering\nlarge surface areas. 22 , 23 The use of these techniques to\ncreate templates and molds has also been explored, but there are still\nquestions in regard to the scalability of these techniques and how\ndurable the templates and molds would need to be so as to be cost-effective. 24 , 25 Chemical vapor deposition (CVD) is a bottom-up technique that has\nbeen shown to produce viable superhydrophobic surfaces, but it again\nsuffers from the same issues as top-down methods concerning costs\nand scalability. 26 , 27 However, there are some bottom-up\nmethodologies that have shown promise for economic viability. Both self-assembly and the functionalization of nanoparticles have\nbeen used as the basis for bottom-up fabrication techniques. 28 − 30 Traditionally, many of the superhydrophobic surfaces produced through\nbottom-up fabrication techniques have relied on the presence of fluorine\nto achieve low surface energy. 31 These\ncompounds have more recently been found to cause harm to the environment\nand human health, leading to industries looking to move away from\ncomposites containing fluorine. 32 − 34 On the back of a report last\nyear by Brunn et al., 35 a proposed restriction\nof around 10,000 per- and polyfluoroalkyl substances was put forward\nto the European Chemicals Agency by authorities in Denmark, Germany,\nthe Netherlands, Norway, and Sweden. 36 However,\nlong-chain fatty acids have been identified by a number of reviews\nas having the potential to replace fluorinated compounds. 37 − 39 As we look to move away from our reliance on fluorinated compounds\nfor superhydrophobic properties, there are several approaches that\ncan be considered. Agrawal et al., 40 Zhu\net al., 41 and Heale et al. 42 all fabricated superhydrophobic surfaces through a combination\nof nanoparticles and fatty acids, achieving x̅ WCAs of between\n142 and 161°. Others combined their low-surface-energy compound\nwith an additional base polymer in the hope of fabricating a more\ndurable superhydrophobic surface. Zhi et al. 43 explored both polyurethane and epoxy resin, with Li et al. 44 also using epoxy resin, whereas Thasma Subramanian\net al. 45 used an acrylic resin. All three\nused spray application techniques to deposit their coatings on the\ndesired surface, achieving WCAs of between 159 and 169°. While there have been considerable developments regarding superhydrophobic\nsurfaces over the past decade, there has been minimal commercial impact.\nManoharan and Bhattacharya 39 listed a number\nof limitations that may be holding the field back, including stability,\ndurability, cost, health, environmental risks, etc. One way to remove\nsome of these limitations is to attempt to functionalize a commercial\nproduct rather than starting from scratch. Jafari and Farzaneh 46 took a novel approach and were able to modify\na commercial latex with CaCO 3 and stearic acid to achieve\na WCA of ∼158° using spray deposition. Zhuang et al. 47 produced robust superhydrophobic surfaces with\na WCA of ∼160° and a sliding angle < 1° by functionalizing\na commercial epoxy resin with polydimethylsiloxane through aerosol-assisted\nCVD. While the combination of a latex polymer and a spray deposition\ntechnique is an easily scalable technique, latex polymers lack the\ndurability of more robust polymers. By comparison, the combination\nof an epoxy resin polymer and a CVD technique produces a robust product,\nbut CVD techniques lack the straightforward scalability of spray deposition\ntechniques. This paper will look to advance on these concepts by outlining\nan approach capable of achieving an inexpensive, nontoxic, and accessible\nsuperhydrophobic surface using readily available commercial products.\nThis will be done by first outlining a rapid method for the functionalization\nof nanoparticles with fatty acids before modifying a commercial polyurethane\ncoating and a commercial epoxy resin with nontoxic fatty acids and\nnanoparticles. The coatings were deposited onto glass microscope slides\nusing a spray technique before undergoing characterization and functional\ntesting.",
"discussion": "3 Results and Discussion For the coatings,\na number of variables were considered, each of\nwhich had an impact on the surfaces’ wetting properties or\ndurability. Both ZnO and SiO 2 particles were investigated\nas possible scaffolds, with both palmitic acid and stearic acid being\ninvestigated as functionalization agents. Palmitic acid and stearic\nacid were used due to their innate low surface energy, which is required\nif a surface is to repel water. Other variables that also impacted\nthe surface properties were the fatty acid-to-polyurethane ratios\nand the spray distance. Formulations were produced with the minimal\namount of solvent that would allow them to spray without clogging\nthe spray gun. Functionalized particles were produced by functionalizing\n1 g of\neither ZnO or SiO 2 particles in 1.5 g of fatty acid dissolved\nin acetone solution for 20 min. Centrifugation was then used to remove\nthe functionalized particles from the solution before they were washed\n3 times with acetone. Once dry, the functionalized particles were\nadded to double-sided tape so their wetting properties could be analyzed.\nWhile these initial results showed that there was little difference\nin the wetting properties of particles functionalized with stearic\nacid as opposed to palmitic acid, they did show a significant difference\nwhen comparing the ZnO and SiO 2 particles. While\nprefunctionalized or pretreated SiO 2 has previously\nbeen functionalized with fatty acids, 42 attempts to functionalize pure SiO 2 particles were unsuccessful.\nThis is unsurprising when considering the proposed functionalization\nmechanism. As fatty acids are long-chain carboxylic acids, hydrogen\ncan be removed from the terminal hydroxyl group when dissolved in\nacetone. It is this resulting polar end of the molecule that is then\ncapable of reacting with the particle surfaces. There are a number\nof factors that may influence the interactions between the fatty acid\nand the SiO 2 particles. First, SiO 2 is less\npolar than ZnO, which could contribute to the reduced interaction\nbetween the dissolved fatty acid and the particles. There may also\nbe a hydration layer on the surface of the particles interfering with\nthe SiO 2 and fatty acid interactions. Although this can\nbe removed by treating the particles prior to functionalization, this\nwould require an additional step. ZnO is more polar by nature and\ncould easily be functionalized with the fatty acids in a relatively\nshort period of time, resulting in an x̅ WCA of ∼173°. The functionalization of ZnO with the fatty acid and the nonfunctionalization\nof SiO 2 were supported by FT-IR analysis ( Figure 1 ). The fatty acids have a notable\nC=O peak at 1699 cm –1 and a large C–O\npeak at 1297 cm –1 . 49 , 50 These peaks\nare of particular note as once the fatty acid becomes bound to the\nsurface through the proposed mechanism, the oxygen atoms go into resonance,\neliminating the C=O and C–O peaks. These peaks are instead\nreplaced with a symmetric COO – stretch at 1399 cm –1 and an asymmetric COO – stretch\nat 1540 cm –1 that occur due to the resonance. 51 While these resonance peaks can be observed\nin the spectrum for the functionalized ZnO as expected, they were\nnot observed in the postfunctionalization SiO 2 FT-IR. In\nfact, the combination of the spectrum for the postfunctionalized SiO 2 and their wetting properties suggests that the fatty acid\nwas not bound to the particles and that all the fatty acid was removed\nduring the wash step. Figure 1 FT-IR analysis of stearic acid (top), functionalized ZnO\n(middle),\nand functionalized SiO 2 . The next step was to formulate the mixture for\ncoating. This was\ndone by first dissolving the stearic acid in 50 mL of acetone using\nheat and stirring before adding the particles, followed by 10 g of\npremixed commercial polyurethane or epoxy resin polymer (3 part hardener\nand 7 part base). While the amounts of solvent and polymer were kept\nconsistent, the amount of metal oxide particles and stearic acid was\nvaried to allow for different component ratios. The solvent was then\nadded to the suspension to give a final volume of 50 mL. Coatings\nwere then applied to a glass microscope slide using dip-coat\nor spray methods. Immediately prior to application, the suspensions\nwere sufficiently shaken and sonicated to give a uniform suspension\n(∼5 min). The dip-coat application method simply required a\nmicroscope slide to be dipped into the uniform suspension before being\nremoved ∼2 s later. The spray coat method used a compressed\nair spray gun set at 3 bar, with the aperture opened in the minimal\namount that still allowed for the suspension to freely pass through\nand not clog the gun. Coated samples were then left to air-dry for\n24 h before being tested. Despite the use of sonication and\nshaking, the solvent and solid\ncomponents of the suspension containing the epoxy resin remained firmly\nseparated, with a uniform suspension being unachievable as the solid\nphase kept crashing out of the solvent. This meant that we could proceed\nwith only the solution containing the polyurethane. Once the polyurethane\ncoatings were dry, they were analyzed and compared. As can be seen\nin Figure 2 , the spray\ncoating method produced surfaces that were more consistently rough\nwhen compared to those of the dip coating methods. However, when investigating\ndifferent spray distances, it was found that due to the volatility\nof the acetone and the inconsistent environmental conditions (the\ntemperature of the lab could not be controlled and resulted in variations\nof the temperature by up to 20 °C), the ideal spray distance\nvaried day to day. The impact of this was that if a surface was coated\nfrom too close a distance, the surface lacked roughness caused by\nthe hierarchical buildup of the coating, resulting in surfaces with\ngood durability but reduced WCAs. On the other hand, if the surface\nwas coated from too far away, the acetone evaporated away from the\nsuspension in the air, effectively meaning a dry powder was being\ndeposited on the surface. This resulted in surfaces with good WCAs\nbut poor durability. Overall, the ideal spray distance fell in the\nrange of 30–60 cm from the surface. Figure 2 SEM images of polyurethane\ncoating containing stearic acid, applied\nvia dip coating (top left). Polyurethane coating containing stearic\nacid, applied via spray coating (top right). Polyurethane coating\ncontaining ZnO and stearic acid, applied via dip coating (middle left).\nPolyurethane coating containing ZnO and stearic acid, applied via\nspray coating (middle right). Polyurethane coating containing SiO 2 and stearic acid, applied via spray coating (bottom). When comparing coatings of different polyurethane\nto stearic acid\nratios, a balance needs to be found between superhydrophobicity and\ndurability. Coatings of different ratios were formulated by mixing\n10 g of polyurethane with 0.25–3 g of stearic acid. Tape tests\nunsurprisingly showed that the 40:1 (polyurethane/stearic acid) coating\nwas the most durable. This was expected as the polyurethane is harder\nthan stearic acid; therefore, the greater the fraction of polyurethane,\nthe greater the expected durability of the surface. This trend continued\nwith testing, which also showed that the coating with a 10:3 ratio\nwas the least durable, although any surface with a polyurethane content\nof less than 5:1 had poor durability. From investigations using 0.25\ng fractions of stearic acid, it was discovered that the coatings reached\ntheir optimal hydrophobic properties at a 10:1 ratio. These surfaces\nalso maintained reasonable durability, with them being resistant to\ntape testing and somewhat scratch-resistant. While surfaces with lower\nstearic acid content had improved scratch resistance, there was also\na noticeable drop-off in WCA for each 0.25 g reduction of stearic\nacid. FT-IR analysis of the coatings provided some intriguing\nand unexpected\nresults. As could be seen when analyzing the ZnO particles functionalized\nwith fatty acids, the binding between the ZnO and the fatty acid results\nin a resonance structure that produces a distinctive symmetric COO – stretch at 1399 cm –1 and an asymmetric\nCOO – stretch at 1540 cm –1 . 51 However, when analyzing the coating’s\nspectra, neither of these peaks could be observed. As there is very\nlittle difference between the spectra of the pure polyurethane, the\npolyurethane mixed with stearic acid, and the polyurethane mixed with\nstearic acid and ZnO ( Figure S1 ), it is\npossible that the peaks associated with the polyurethane were of such\nhigh intensity that they were masking other peaks. Functional\ntesting was then carried out on coatings made up of\n10 g of polyurethane, 1.5 g of stearic acid, and 1 g of ZnO particles\nor 0.25 g of SiO 2 particles. Although 0.25 g of ZnO was\nsufficient, a higher concentration was used as there is no noticeable\nimpact on the surfaces’ wetting properties, but it does have\ninnate antimicrobial properties that SiO 2 does not have.\nThis may increase the surfaces’ viability for antibiofouling\napplications, but the testing of the surfaces’ antibiofouling\nproperties is yet to be carried out. Standard drop shape analysis\nwas used to show that while the pure polyurethane had hydrophobic\nproperties (x̅ WCA of ∼95°), both the dip- and spray-coated\nsurfaces had superhydrophobic properties. However, while the dip-coated\nsurfaces’ x̅ WCAs did not exceed 154°, some of the\nspray-coated surfaces’ x̅ WCAs exceeded 170° ( Figure 3 ). When analyzing\nboth the IR analysis and the scanning electron microscopy (SEM) imaging,\nthere was little difference between surface-containing particles and\nthose without. This trend continued when looking at the x̅ WCAs\nof the sprayed surfaces, suggesting that the presence of the nanoparticles\nmay not be required. Figure 3 WCA of polyurethane coating containing stearic acid, applied\nvia\ndip coating (top left). Polyurethane coating containing stearic acid,\napplied via spray coating (top right). Polyurethane coating containing\nZnO and stearic acid, applied via dip coating (middle left). Polyurethane\ncoating containing ZnO and stearic acid, applied via spray coating\n(middle right). Polyurethane coating containing SiO 2 and\nstearic acid, applied via spray coating (bottom). When the x̅ WCAs obtained were compared to\nthose found in\nthe literature, they were found to be very high. Looking across five\nreviews, most surfaces failed to achieve water x̅ WCAs ≥\n170°, with even fewer materials achieving a ≥170°\nx̅ WCA while not being reliant on a fluorinated compound. 52 − 56 While Sharifi et al. 57 were able to produce\na fluorine-free superhydrophobic surface with a WCA ≥ 170°,\ntheir method required a TiO 2 feedstock suspension to be\nsuspension plasma sprayed onto a grit-blasted substrate. By comparison,\nour spray coating could be formulated from scratch in a one-pot method\nand sprayed onto a substrate in under an hour. When specifically comparing\nthe surface to superhydrophobic polyurethane surfaces from some of\nthe more recent publications, the surface had a greater x̅ WCA\nthan that of the other reported surfaces. 58 − 60 This was despite\nbeing fluorene, vastly reducing the environmental impact of the surface\ncompared to that of the equivalent fluorene-containing compounds in\nthe literature. Another property to be determined was the coating’s\nsurface\nrolling angle. However, there is very little consensus on how this\nmeasurement should be carried out or what qualifies as a measurement.\nFor instance, neither Qi et al. nor Penna et al. reported the height\nat which their water droplet was dropped from or the volume of their\ndroplet, both of which we found significantly impacted rolling angle\nmeasurements. 61 , 62 Another issue is at what point\nshould a measurement be taken or be valid. Is it when a droplet moves\nat all or when the droplet clears the surface? Due to these discrepancies,\nthe rolling angle was clearly defined as the angle of tilt a surface\nneeded to be at, to which a 13 μL water droplet dropped from\n10 mm above the surface would be unable to stay on the surface it\nwas applied to, with the implementation of this method being shown\nin Figure 4 . Testing\nwas used to determine that the rolling angle of the spray coatings\ncontaining no nanoparticles was 3° and that containing ZnO was\n2°, while the rolling angle of the SiO 2 surface was\n1°. However, the rolling angle for the dip-coated surfaces was\ndetermined to be in excess of 10°, the maximum angle tested. Figure 4 Demonstration\nof how a rolling angle measurement is taken. The\nstage is set to a designated angle, at which point a 13 μL water\ndroplet is applied to the surface using a syringe dispenser raised\n10 mm above the surface. If the water droplet was unable to stay on\nthe surface, the current angle was deemed the surface’s rolling\nangle. The final analysis of the surfaces’ wetting\nproperties was\nan inspection of the surfaces’ CAH. As CAH is a measure of\na surface’s ability to keep a droplet in place, it does not\nspecifically determine if a surface is superhydrophobic. 63 Surfaces with a low CAH value are better described\nas either superhydrophobic or have the potential to be superhydrophobic.\nThis is because surfaces with low surface energy adhere poorly to\nwater molecules, regardless of their surface roughness. Superhydrophobic\nsurfaces are reliant on having both low surface energy and surface\nroughness. This means that if a surface has a low CAH value and a\nlow WCA value, it must be relatively flat. 64 However, if a surface of the same chemical composition is roughed,\nthen it should see an increase in its WCA value. In theory,\nthe CAH is a surface’s ability to keep a droplet\nin place, and as such, the results should follow the same trend as\nthe rolling angle analysis. 63 The CAH was\ndetermined using a method by Huhtamäki et al. 48 where the difference between the ACA and the RCA measurements\nis deemed to be the hysteresis. The ACA was measured as a water droplet\nwas dispensed and expanded onto the surface being tested, with the\nRCA then measured as the droplet was being retracted off the surface.\nCAH measurements were taken for the two samples that had previously\nexhibited the best wetting properties, resulting in a CAH value of\n2.05 for the coating containing ZnO and a value of 5.58 for the value\ncontaining SiO 2 . These CAH results were slightly\nsurprising as the initial expectation\nwas that they would match the rolling angle results. While initially\nsurprising, when the methodology of the CAH testing is considered,\nalong with the two surfaces’ roughness ( Figure 2 ), it is possible to form a hypothesis about\nwhat is happening. As CAH measures a surface’s ability to keep\na droplet in place, with increasing values corresponding to increased\nadhesion, surfaces with a lower surface energy should have a low CAH\nvalue. Like the WCA and rolling angle measurements, good CAH values\nare reliant on low surface energy. However, while WCA and rolling\nangle analysis allow for the surface to simply repel the water, CAH\nanalysis forces the water onto the surface. This means that unlike\nWCA and rolling angle measurements, which see a positive correlation\nbetween surface roughness and their superhydrophobic measurements,\nCAH measurements may decline when a surface becomes too rough. This\nadded force may cause surface structures to penetrate the droplet’s\nsurface, partially pinning it and reducing the water’s ability\nto freely move across the surface. The next stage was to test\nthe surfaces’ resistance to staining.\nStain testing was performed by placing a sample at an 80° angle,\nthen using a Pasteur pipet to apply a drop of a staining liquid to\neach surface. The stains used were crystal violet (20 ppm), red wine,\nand instant coffee. When the stain testing was carried out, the surfaces\nwere able to completely repel all 3 liquids, resulting in no staining.\nBesides that, when samples were placed at a 10° angle and covered\nin loose dirt (glitter), the samples showed self-cleaning properties.\nThis was done by simply rolling water droplets down the surface and\nthrough the dirt, which was removed by the rolling droplet. Finally, the durability of the surfaces was tested by using a tape\ntest. This test simply required the firm application of 10 strips\nof Scotch Magic Tape to the surfaces and their removal. After testing\nall the sprayed surfaces, all coatings maintained their wetting properties\nadmirably with a ≤2° change in the x̅ WCA of the\ncoating containing ZnO (171–169°), SiO 2 (172–170°),\nor no nanoparticles (170–169°). The surfaces also appeared\nto be durable as little to none of the coating was removed by the\nScotch Magic Tape."
} | 6,493 |
23646825 | null | s2 | 264 | {
"abstract": "Flagelliform spider silk is the most extensible silk fiber produced by orb weaver spiders, though not as strong as the dragline silk of the spider. The motifs found in the core of the Nephila clavipes flagelliform Flag protein are GGX, spacer, and GPGGX. Flag does not contain the polyalanine motif known to provide the strength of dragline silk. To investigate the source of flagelliform fiber strength, four recombinant proteins were produced containing variations of the three core motifs of the Nephila clavipes flagelliform Flag protein that produces this type of fiber. The as-spun fibers were processed in 80% aqueous isopropanol using a standardized process for all four fiber types, which produced improved mechanical properties. Mechanical testing of the recombinant proteins determined that the GGX motif contributes extensibility and the spacer motif contributes strength to the recombinant fibers. Recombinant protein fibers containing the spacer motif were stronger than the proteins constructed without the spacer that contained only the GGX motif or the combination of the GGX and GPGGX motifs. The mechanical and structural X-ray diffraction analysis of the recombinant fibers provide data that suggests a functional role of the spacer motif that produces tensile strength, though the spacer motif is not clearly defined structurally. These results indicate that the spacer is likely a primary contributor of strength, with the GGX motif supplying mobility to the protein network of native N. clavipes flagelliform silk fibers."
} | 386 |
27531136 | PMC4987665 | pmc | 265 | {
"abstract": "During the 20 th century, seawater temperatures have significantly increased, leading to profound alterations in biogeochemical cycles and ecosystem processes. Elevated temperatures have also caused massive bleaching (symbiont/pigment loss) of autotrophic symbioses, such as in coral-dinoflagellate association. As symbionts provide most nutrients to the host, their expulsion during bleaching induces host starvation. However, with the exception of carbon, the nutritional impact of bleaching on corals is still unknown, due to the poorly understood requirements in inorganic nutrients during stress. We therefore assessed the uptake rates of nitrogen and phosphate by five coral species maintained under normal and thermal stress conditions. Our results showed that nitrogen acquisition rates were significantly reduced during thermal stress, while phosphorus uptake rates were significantly increased in most species, suggesting a key role of this nutrient. Additional experiments showed that during thermal stress, phosphorus was required to maintain symbiont density and photosynthetic rates, as well as to enhance the translocation and retention of carbon within the host tissue. These findings shed new light on the interactions existing between corals and inorganic nutrients during thermal stress, and highlight the importance of phosphorus for symbiont health.",
"discussion": "Discussion Our analysis of the DIN and DIP uptake rates of four scleractinian and one soft tropical coral species shows that coral holobionts regulate the uptake of nutrients and points to higher DIP uptake rates under thermal stress. In addition, corals maintained under DIP enrichment avoided bleaching and maintained high rates of photosynthesis, conversely to control, non-enriched corals, which experienced both a decrease in symbiont density and in the acquisition of inorganic carbon. All together, these observations strongly suggest that phosphorus is an essential nutrient for the symbiosis during stress events. Conversely, nitrogen uptake rates significantly decreased under thermal stress, either due to coral bleaching and/or to avoid excess nitrogen in the coral tissue. Our results highlight the importance of nitrogen and phosphorus availability in the coral response to thermal stress. Under control conditions, DIP and DIN uptake rates normalized to symbiont cell or holobiont biomass, as well as the ratios of DIN:DIP uptake, presented a threefold variation among the coral species tested, potentially reflecting different nutritional requirements and ecological trade-offs in the allocation of nitrogen and phosphorus amongst macromolecules associated with diverse functions 33 . Furthermore, we show that H. fuscescens exhibited higher DIN uptake rates during its pulsating activity, as previously demonstrated for the jellyfish Cassiopeia sp 34 . Therefore, rhythmic pulsations not only improve H. fuscescens photosynthesis by preventing seawater re-filtration 31 , but also enhanced its acquisition of essential nutrients. All scleractinian coral species bleached in response to thermal stress with a reduction in rates of net photosynthesis for P. cactus and P. damicornis . Despite this loss of symbionts and/or chlorophyll, rates of phosphate acquisition per protein content or symbiont cell were significantly enhanced in four out of the five species investigated, and for both clades C1 and D1. Since the same trend was observed with S. pistillata in symbiosis with another Symbiodinium clade (clade A) 35 , increased phosphate uptake seems to be a general response of corals to thermal stress, independently of the clade genotype or on the polyp activity. Although it might only be due to a temperature dependency of the phosphorus transporters at the cell membranes 36 , our results tend to confirm the fact that phosphorus is actively needed in corals to offset the negative effect of thermal stress 26 and that corals underpin an active control mechanism for phosphorus uptake and allocation, as observed in plants 37 . Indeed, our second experiment, in which DIP enrichment of the seawater prevented bleaching and increased the photosynthetic efficiency and the translocation of photosynthates during thermal stress, highlights the importance of phosphorus to maintain the physiological functions in corals under thermal stress. Wiedenmann et al. 26 also demonstrated that limited phosphorus availability induces a shift from phospholipids to sulpholipids in symbiont membranes, and increases the susceptibility of corals to temperature-induced bleaching 38 . In addition to being a structural membrane component, phosphorus takes part in the synthesis of nucleic acids and other energy generation processes 39 . It is also involved in cellular signalling via thylakoid protein phosphorylation 40 . It has to be noticed that although corals are able to increase their rates of phosphorus acquisition under warming conditions, a minimal concentration of inorganic phosphorus, which remains to be determined, is required to avoid bleaching. In contrast to phosphorus, three coral species ( P. cactus , P. damicornis and G. fascicularis) out of the five investigated showed a significant decrease in nitrogen uptake rates normalized to protein content. Nitrogen uptake rates normalized to symbiont cell, however, remained unchanged between 25 °C and 30 °C, except for nitrate in P. cactus. A similar finding was observed with the scleractinian coral S. pistillata associated with clade A1 symbionts 35 , which significantly decreased both nitrate and ammonium uptake rates down to zero, at 33 °C. These results suggest that the reduction in nitrogen uptake is a general response of corals to thermal stress, independent of the clade genotype. Whether this reduction is a consequence of bleaching (and reduced symbiont activity), or is due to a physiological control or feedback aimed at decreasing the amount of nitrogen entering the symbiotic association, requires further investigation. Both hypotheses have pros and cons. The first hypothesis is in agreement with studies showing that nitrogen uptake is mostly driven by the symbionts 8 9 41 42 . In H. fuscescens , which did not experience any bleaching or decrease in photosynthetic activity, due to its pulsating activity and/or its association with the thermally resistant clade D1 43 44 45 , no decrease in nitrogen uptake rate was indeed observed. A decrease in nitrogen uptake rates in the scleractinian coral species may lead to nitrogen limitation, which in turn may impair protein repair 23 , weaken the photosynthetic capacities, and decrease carbon fixation, as already noted in mulberry leafs 46 . This may explain why a supplementation of seawater with ammonium or nitrate was shown to increase the resilience of coral species to stress-induced bleaching 21 22 47 . On the other hand, the second hypothesis is in agreement with studies showing an increased bleaching susceptibility of heat-stressed corals in presence of high nitrogen and low phosphorus concentrations 26 48 49 50 . High nitrogen availability for symbionts during thermal stress tends to favour symbiont growth and reduce the amount of photosynthates transferred to the host 14 , together with inducing a higher phosphorus limitation 26 . This could explain the reduction in nitrogen uptake rates by the symbiotic association during thermal stress, in order to reduce the amount of nitrogen available to symbionts. Overall, the decrease in the N:P uptake ratios during thermal stress, which is consistent for most coral species, may imply that the corals internally experience an increase in their N:P ratio, maybe due to reallocation of nitrogen that was stored in tissue reserves 10 . The relevance of the N:P ratio for terrestrial and marine symbiotic associations has recently been discussed through a meta-analysis 50 which showed that a decoupling in mutualism performance occurs whenever phototrophs benefit from nutrient enrichment at the expense of their heterotrophic partners. It has also been demonstrated in recent studies 26 51 , in which changes in seawater nutrient ratios induced changes in carbon acquisition, translocation and allocation within the symbiosis. The benefits that the symbiotic association retrieves from a low N:P ratio condition are highlighted in our second experiment, in which the supplementation of seawater with 2 μM phosphate, and the subsequent decrease in the N:P ratio, prevented bleaching, increased the photosynthetic capacities, the carbon translocation, as well as the carbon retained into animal biomass during thermal stress. Understanding how corals respond to alterations in seawater N:P ratio, or how they adjust their internal N:P ratio in response to environmental changes, is an important task for coral reef research today. Our results clearly show that corals are phosphorus-limited during thermal stress and that they are able to increase their acquisition of phosphorus independently of the Symbiodinium clade involved in the symbiotic association. Further investigations, which will trace the fate of nitrogen and phosphorus within the symbiosis at elevated temperatures, are however needed to bring detailed insights into the processes involved in the nutrient acquisition and allocation by corals under thermal stress."
} | 2,335 |
30966413 | PMC6415204 | pmc | 266 | {
"abstract": "A large amount of research has been devoted to developing novel superhydrophobic coatings. However, it is still a great challenge to pursuean environmentally friendly method that leads to superhydrophobic coatings. Herein, we demonstrate for the first time, an environmentally friendly method for the preparation of conductive superhydrophobic coatings with sandwich-like structures by using aminoethylaminopropyl polydimethylsiloxane modified waterborne polyurethane (SiWPU) and N -octadecylamine functionalized multi-wall carbon nanotubes. These environmentally friendly coatings with the sheet resistance of 1.1 ± 0.1 kΩ/sq exhibit a high apparent contact angle of 158.1° ± 2° and a low sliding angle below 1°. The influence of the surface texture before and after heat treatment on the wetting properties is discussed. In addition, the coatings can be electrically heated by 3~113 °C with a voltage of 12~72 V, and thus, can be used for deicing. Furthermore, the resulting coatings demonstrate good performance of wear resistance and ultraviolet resistance, which will have broad application potential in harsh environments.",
"conclusion": "4. Conclusions It is reported for the first time that an environmentally friendly approach to fabricate conductive superhydrophobic coatings with sandwich-like structures of “SiWPU-MWCNTs-SiWPU” has been completed. This process is simple, safe, and avoids toxic solvent, fluoride material, and pollution. The functionalized multi-walled carbon nanotubes grafted with N -octadecylamine (MWCNTs-ODA) gain good dispersibility and stability in ethanol and create both microscale and nanoscale roughness, while the waterborne polyurethane modified AEAPS acts as a low surface energy material and a binder. The coating’s feature contact angle reaches 158.1° ± 2°, and the sliding angle is below 1°. Meanwhile, the coating shows good conductivity, electric heating characteristics, wear resistance, and UV resistance. Therefore, the multifunctional coating will have the potential for a wide application in the cold, outdoor environment and in devices that heat with electricity.",
"introduction": "1. Introduction Recently, surfaces with superhydrophobic properties have attracted extensive interest due to their potential applications, including self-cleaning, anti-icing, anti-bacteria, and anti-adhesion [ 1 , 2 , 3 , 4 , 5 , 6 ]. A variety of techniques, such as the etching method, electrospinning method, chemical vapor deposition method, self-assembly method, sol-gel method, electrodeposition method, hydrothermal method, and phase separation method, have been developed to fabricate superhydrophobic surfaces with different structures [ 7 , 8 , 9 , 10 , 11 , 12 , 13 ]. However, during the preparation process of these superhydrophobic coatings or surfaces, harmful organic solvents (such as chloroform, tetrahydrofuran, dimethyl formamide, acetone, and toluene) were used to disperse the nanoparticles or dissolve the organic components [ 14 , 15 , 16 ]. In addition, in some superhydrophobic surfaces, fluoride-rich materials have been adopted to enhance the hydrophobic effects [ 17 , 18 ]. These less eco-friendly materials not only contaminate the environment and do harm to the health of producers and users, but also increase the treatment cost of the pollutants and the production cost. Although the theoretical study and preparation methods of superhydrophobic surfaces have made great progress, it is still a great challenge for the researchers to fabricate superhydrophobic surfaces without the use of organic components and fluoride-rich materials. So, it is necessary to develop a novel technology system to eliminate the pollutants fundamentally. Until now, the reported preparation methods of environmentally friendly superhydrophobic surfaces are mainly the electrochemical deposition method [ 19 ], the electrochemical etching method [ 20 ], femtosecond laser irradiation [ 21 , 22 ], the hot pressing method [ 23 , 24 ], and spraying polymer nanocomposites and waterborne suspensions [ 25 ]. Among these techniques, the spray method is attractive since it is fast, can be highly automatized on an industrial scale, and is without any restriction on substrate categories, which make it most likely to realize industrialization. It seems paradoxical that hydrophobic coatings are fabricated from waterborne polymer emulsion, because it was generally believed that hydrophobic polymer could not dissolve into water. But the fact is that researchers are able to use these hydrophobic polymers (e.g., fluorinated polyacrylate and organic silicon polyurethane) with carboxyl groups simultaneously. After ionization of the carboxyl groups, this kind of polymer is similar to polymeric surfactants, which can form in the micelle with the carboxyl arranged outward and the hydrophobic chain segment wrapped inward. This form of oil in water exists stably in the aqueous media. During drying, with the evaporation of water, the emulsion was gradually demulsified. Meanwhile, the micelle structure was damaged, and the hydrophobic chain segment was enriched toward the two ends of air so that a hydrophobic surface was ultimately generated. Therefore, this type of waterborne polymer emulsion provides a new way of fabricating environmentally friendly superhydrophobic coatings. Schutzius et al. [ 26 ] fabricated the superhydrophobic composite coatings on a variety of substrates by spraying water-based polyolefin-exfoliated graphite nanoplatelets dispersion. Aslanidou et al. [ 27 ] prepared the superhydrophobic protective coatings on silk by spraying water-soluble siloxane emulsion enriched with silica nanoparticles, without the use of any organic solvent. Similarly, Chatzigrigoriouet et al. [ 28 ] fabricated the superhydrophobic coatings by the same way. Raoet et al. [ 29 ] obtained waterborne self-healing superhydrophobic coatings by mixing fluoroalkyl silane (FAS)-loaded microcapsules, photocatalytic TiO 2 nanoparticles, and FAS-modified SiO 2 nanoparticles with waterborne polysiloxane resins. Chen et al. [ 30 ] also prepared all water-based self-repairing, superhydrophobic coatings by the same way. Mates et al. [ 25 ] demonstrated a water-based superhydrophobic coating on nonwoven and cellulosic substrates by spraying bentonite nanoclay and aqueous fluoroacrylic copolymer dispersions. Milionis et al. [ 31 ] presented a simple, one-step, water-based spray coating process to obtain superhydrophobic and superoleophobic coatings on metals comprising hydrophilic silica nanoparticles and fluoroacrylic polymer. Although these methods avoided the use of organic solvents, some of them still used fluoride-rich materials, which would bring about new contamination. Moreover, in these methods, nanoparticles and polymer waterborne dispersions were directly mixed together and thus led to poor dispersion of nanoparticles in polymer dispersions. As well, the addition of nanoparticles could lead to a sharp increase in viscosity of the composition system so as to easily block the spray head. We report herein, for the first time, an environmentally friendly approach to fabricate conductive superhydrophobic coatings with sandwich-like structures. Firstly, aminoethylaminopropyl polydimethylsiloxane modified waterborne polyurethane (SiWPU) dispersion was successively sprayed onto glass slides and thoroughly dried. Then, ethanol dispersion of multiwalled carbon nanotubes (MWCNTs) was successively sprayed onto the SiWPU coatings and dried out completely. Next, SiWPU dispersion was sprayed onto the MWCNTs coatings and thoroughly dried. Finally, through heat treatment of the composite coatings, superhydrophobic coatings were obtained. So far, it has not been reported that waterborne polyurethane modified with silicone was used to fabricate superhydrophobic coatings. To improve the dispersibility and stability of the MWCNTs in ethanol, N -octadecylamine with a long aliphatic chain was firmly grafted to the surface of the MWCNTs. In the process of the fabrication of the superhydrophobic coatings, only ethanol and water were used as dispersion media, and all materials used were fluoride-free. So, this novel technology eliminated the pollution from headstream. Compared with the polymer/MWCNTs composite coatings fabricated by physical mixing method, in this system, the MWCNTs were dispersed evenly, and thus, the coating showed great conductivity and good electric heating characteristics. In addition, this conductive superhydrophobic coating is wear resistant and ultraviolet (UV)-resistant, so it has good prospects of application.",
"discussion": "3. Result and Discussion Herein, conductive superhydrophobic coatings were fabricated by using N -octadecylamine grafted onto the MWCNTs and AEAPS-modified waterborne polyurethane. To improve the stability and dispersion of the raw MWCNTs in ethanol, MWCNTs-ODA was fabricated. As shown in Scheme 1 , N -octadecylamine grafted onto the MWNCTs was firstly obtained via a similar process. FTIR and XPS were used to monitor the process ( Figure 1 a,b). As can be seen in the infrared spectra, the raw MWCNTs show peaks with quite low intensity at 3466 cm −1 , which can be ascribed to the O–H vibration, because the MWCNTs hold water and oxygen easily in air [ 32 ]. For the spectrum of the MWCNTs-COOH, the band at 3435 cm −1 is strengthened and a new band appears at 1717 cm −1 , assigning to the C=O stretching vibration, indicating that a carboxyl group has been introduced on the surface of the MWCNTs after the nitric acid treatment [ 33 ]. In the spectrum of the MWCNTs-ODA, the prominent peaks at 2916 and 2843 cm −1 correspond to the CH 3 – and –CH 2 – bond vibrations in the N -octadecylamine, which shows that the N -octadecylamine is successfully grafted onto the MWCNTs. Figure 1 a also shows that the prominent C=O peak of the MWCNTs-ODA red shifts from 1717 to 1703 cm −1 , which is attributed to the formation of a hydrogen bond between the C=O and N–H groups [ 34 ]. Meanwhile, the fact that O intensity at 532.6 eV of the MWCNTs-COOH is much higher than that of the raw MWCNTs, which indicates that the raw MWCNTs have been oxidized by the concentrated HNO 3 , and the appearance of obvious N signals at 400.1 eV with a lower-binding-energy shoulder at 399.8 eV indicates that the N -octadecylamine grafted onto the MWCNTs have been successfully obtained according to the process shown in Scheme 1 . TGA measurement has provided further evidence for the grafting of the N -octadecylamine to the MWCNTs. Figure 1 c shows their TGA curves. The raw MWCNTs have only a 0.9% weight loss in the range between 50 and 600 °C. However, the MWCNTs-COOH has a 4.73% weight loss. The 3.83 wt % weight loss can be assigned to the decomposition of the grafted carboxyl groups. For the TGA curve of the MWCNTs-ODA, there is a 11.87% weight loss in the range of 50–600 °C, and a 7.14% weight loss can be used to appropriately estimate the mass percentage of the N -octadecylamine grafted onto the MWCNTs [ 32 ]. In Figure 1 d, for WPU and SiWPU, N–H stretching vibration peaks in the ureido are at 3324 cm −1 , C=O stretching vibration peaks in the ureido are at 1716 cm −1 , and N–H deformation vibration peaks in the ureido are at 1530 cm −1 . The appearance of the above three absorption peaks indicates that there exist carbamic acid ester groups. At 2270 cm −1 , both do not show –CNO’s characteristic absorption peaks, indicating that –CNO is completely reacted. Around 803 cm −1 , SiWPU shows that the characteristic absorption peak of CH 3 symmetrical deformation vibration belongs to Si–CH 3 , indicating that AEAPS has successfully grafted onto the WPU chain segment [ 35 ]. Usually, the dispersibility and the stability of nanoparticles in an organic solvent determine whether to generate the uniform and stable nano dispersion or not, and even determine whether to form the homogeneous and well-dispersed nano coating on the surface of the substrate. In the present study, we explored the dispersibility and the stability of the MWCNTs in ethanol before and after modification by TEM and Turbiscan Lab, respectively. From Figure 2 a, it shows that opaque agglomerates formed by the raw MWCNTs appear. In its high-resolution images, a single MWCNT even cannot be observed. The appearance of agglomerates is mainly due to the quantum-size effect of the MWCNTs. Moreover, as one-dimensional materials, MWCNTs have a large aspect ratio, and thus, the numerous MWCNTs are easily mingled and tangled together. So, it is quite difficult for these agglomerates to disperse because the two effects integrated with each other. However, after being modified by N -octadecylamine, the dispersibility of the MWCNTs in ethanol has been improved obviously, and the tube diameter of the MWCNTs becomes clearly bigger (the inset of Figure 2 b). This is because long alkyl chains are formed on the surface of the MWCNTs grafted by octdecylamine, which improves their solubility in ethanol so as to improve their dispersibility. The stability of the MWCNTs before and after modification can be tested by a universal stability analyzer. The mass fraction of test dispersions of the raw MWCNTs and the MWCNTs-ODA in ethanol is 0.02%, and the two samples are respectively sonicated for 1 min before testing. It is set to scan once 1 h, and there are a total 24 times of scanning. The stability of the MWCNTs in ethanol is measured with the transmittance of a near-infrared light pulse (λ = 800 nm). Figure 2 c,d show the transmittance curves of the MWCNTs and the modified MWCNTs in ethanol dispersions at different times, respectively. It is known from the results that the settling rate of the raw MWCNTs in ethanol is very fast, and the transmittance of the dispersion is increased by over 66.7% after 23 h, while the settling rate of the MWCNTs-ODA is very slow and the transmittance is increased by only 0.2% after 23 h. Figure 2 e,f show the mean transmittance of the raw MWCNTs and MWCNTs-ODA every hour. Figure 2 shows that within 4 h the settling rate of the raw MWCNTs is quite fast, but after 10 h, the transmittance of the dispersion tends to be stable. Whereas the settling rate of the MWCNTs-ODA is faster within 2 h, after 3 h, the transmittance of the dispersion tends to be stable. This is due to the lack of active functional groups on the surface of the raw MWCNTs, which thus leads to poor solubility in all kinds of solvents. After being modified by N -octadecylamine, long chain alkyl groups are generated on the surface of the raw MWCNTs, which greatly improve their solubility in ethanol so as to form stable dispersion. The conductive superhydrophobic coating with sandwich-like structures of “SiWPU-MWCNTs-SiWPU” is fabricated via the spraying method and further heat treatment, as is shown in Figure 3 a. From scanning electron microscope (SEM) images, it can be known that a large number of irregular micrometer-scale papillae are formed on the surface of the composite coatings sprayed by the SiWPU emulsion, the MWCNTs ethanol dispersion, and the SiWPU emulsion successively. The sizes of these papillae are between 13.1 and 35.7 μm ( Figure 3 c). The formation of them is due to the rapid volatilization of water in the drying process. When the SiWPU emulsion is sprayed on the MWCNTs coating, both are mixed physically to form a viscous SiWPU/MWCNTs mixture. During drying, with the volatilization of water, the concentrations of SiWPU and MWCNTs in the droplets are increased rapidly and their viscosity is sharply increased. When these poor fluidity droplets contact the polyurethane coating at the bottom, they cannot spread out rapidly on the surface. In the meantime, the countless, highly viscous droplets are aggregated and adhered to one another. With the complete volatilization of water, the micrometer-scale papillae are gradually formed. In high-magnification images, it can be seen that a few MWCNTs are distributed on the surface (as shown in Figure 3 e), while a majority of the MWCNTs are wrapped into the papillae. Although the micrometer-scale papillae are produced on the surface, the composite coating fails to reach superhydrophobicity due to the small roughness of surface. The contact angle of these surfaces is 129.4°, and the droplets are still fastened to the surface, which shows the Wenzel state. Jiang Lei et al. [ 36 ] found that there are nanoscale hierarchical structures in the micrometer-scale papillae on the lotus leaf surface by observing the microstructure of the lotus leaf surface. The micro- and nanoscale hierarchical structure is responsible for the superhydrophobicity of the lotus leaf surface. From the observation of the lotus leaf microstructure, it can be learned that to fabricate the surface with superhydrophobicity, these micrometer-scale papillae also need to have nanoscale hierarchical structures on them. In order to cause the MWCNTs to bulge out within the micrometer-scale papillae, the composite coating is heated over the viscous flow temperature point of the polyurethane (about 150~160 °C) and kept at a certain time to make polyurethane resin fuse and flow toward the interior of the skeleton of the MWCNTs so that a nanoscale structure tangled with the MWCNTs is formed on the micrometer-scale papillae ( Figure 3 f). Moreover, it can be observed from SEM (as shown in Figure 3 b) that while the polyurethane resin is fused to flow downward, it can wrap the MWCNTs partially. After heat treatment, the microstructure of the composite coating surface resembles the lotus leaf surface, which is covered with SiWPU and octdecylamine with low surface energy, showing the superhydrophobicity. The water contact angle reaches 158.1° ± 2°, and the sliding angle is below 1°. Based on the Cassie equation [ 37 ], if the solid surface has only one rough structure, the apparent contact angle (θ*) only depends on the solid phase fraction. From Figure 3 d, it is known that after heat treatment, a large number of micron-scale cavities are generated between the micron papillae on the surface of the composite coating. And nano-scale cavities are formed on every papilla ( Figure 4 a). In these micron-/nano-scale cavities, a lot of air is embedded, and thus, the composite coating has extremely low solid fraction. Meanwhile, the rough surface covered with SiWPU and ODA has low surface energy so that the composite coating shows the superhydrophobic property. In addition, it is also observed that the MWCNTs is wrapped by SiWPU, and the wrapping thickness is about 20 nm. Figure 4 c,d are the sectional SEM images of the composite coating before and after heat treatment. From Figure 4 c, it is known that the coating thickness before heat treatment is about 50~70 μm, and the polyurethane coating at the bottom can be clearly observed. Because the superficial polyurethane coating is mixed with the MWCNTs physically, the clear interface layer cannot be seen. After heat treatment, because the superficial polyurethane layer flows toward the interior of the MWCNTs coating, the thickness of the whole becomes thinner, and the thickness is about 30~50 μm. In addition, it can also be seen that in the composite coating after heat treatment, the interface at the bottom of the polyurethane resin layer disappears. This is because during heating the polyurethane resin is fused, spread out again, and combined to the substrate closely. This fusion and recombination are favorable to increasing the binding force of the composite coating with the substrate. It is generally known that a lot of air is embedded into the roughness structure of the superhydrophobic surface and generates an air layer. When exposed to supercooled water, the superhydrophobic surface will make the water droplets bounce so as to delay or even prevent the formation of snow, ice, and frost on the surface. However, once ice comes, the superhydrophobic surface usually loses its superhydrophobic properties due to the damage of the micro-nanometer hierarchical structure on the surface. Therefore, if we endow the superhydrophobic coatings with conductivity and utilize electric heating characteristics to remove ice, the above problems are readily solved. According to the literature [ 38 ], carbon nanotubes (CNTs) are an ideal conductive filler, and their recombination with polymer resins can obviously increase the conductivity of the composite material. The test results of the sheet resistance of superhydrophobic coatings under different humidities are shown in Figure 5 a. It is known that with the increase of relative humidity from 50% to 90%, the sheet resistance values are changed within a fairly small range, about 1.1 ± 0.1 kΩ/sq, indicating stable conductivity. Furthermore, when the conductive superhydrophobic coatings are connected in series to the circuit, the light-emitting diode (LED) was lightened under the low voltage of only 6 V, as shown in Video S1 . Usually, polymer-carbon nanotube composite materials fabricated by the physical mixing method possess poor conductivity, because of high viscosity results from the large, specific surface area of the MWCNTs, which reduce the conductivity of the composite material remarkably. Compared with the physical mixing method, the preparation method in this paper effectively avoids this problem by step-by-step spraying of the MWCNTs dispersion and resin emulsion. The connection between the MWCNTs is closer, and it is easier to form a 3D f conductive circuit so that the fabricated coatings exhibit excellent conductivity. The electric heating characteristic of the conductive superhydrophobic coating has been studied by the following method. First, precisely two pure copper alligator clips are put on the two sites at a distance of 2 cm on the composite surface. Then, a wire is used to connect the two alligator clips, an adjustable transformer, and the power source in series to form a direct current circuit. Next, the adjustable transformer is used to exert the different applied voltages on the composite coating while the surface temperatures are detected by a thermal infrared camera at different times. Figure 5 b shows the results of the temperature variation of the composite coating surface with the heating time at different applied voltages. As shown, within a certain period, the temperature of the composite coating is increased exponentially while the applied voltage is increased linearly. And under a certain applied voltage, as the heating time goes on, the temperature of the composite coating is increased first, and then tends to return to a certain value, which corresponds to Joule’s law Q = ( V 2 / R ) t . That is to say, for the constant sheet resistance, the quantity of heat is closely related to the voltages on the two ends and the heating time. When the applied voltage is increased to 72 V and kept for 5 min, the surface temperature of the composite coating will reach 113 °C, showing good electric heating characteristics. Moreover, the composite coating maintains its superhydrophobic property, no matter how variable the surface temperature. The contact angle falls slightly. When the surface temperature of the composite coating is over the boiling point of the water droplets, the droplets are rapidly evaporated, and the coatings are ultimately kept in the dry state. However, the applied voltage must be controlled below the viscous flow temperature point of the polyurethane resin, otherwise the coatings are damaged easily. These superhydrophobic coatings with low adhesion and electric heating characteristics have a promising future in the field of ice resistance. The mechanical durability of the conductive superhydrophobic coating is evaluated by the wear resistant test. The wear resistance testing method is shown in the inset of Figure 6 a. The results show that within 10 cycles, the coating maintains its superhydrophobic property, showing excellent mechanical durability. This is because the micro-nanometer hierarchical structure is covered and reinforced by polyurethane resin, and the areas between microparticles of the surface and between the MWCNTs of surface are filled firmly with the polyurethane resin. However, after 10 cycles of friction, the micro-nanometer hierarchical structure is damaged and the superhydrophobic coating loses its superhydrophobic property accordingly. Figure 6 b shows the CA and SA changes of the as-prepared superhydrophobic coatings with UV irradiation time. It was found that the contact angle and the sliding angle show no clear change after 50 h UV exposure, indicating the excellent resistance of the superhydrophobic coating to UV light."
} | 6,171 |
20160978 | null | s2 | 267 | {
"abstract": "We present a facile and inexpensive approach to superhydrophobic polymer coatings. The method involves the in-situ polymerization of common monomers in the presence of a porogenic solvent to afford superhydrophobic surfaces with the desired combination of micro- and nano-scale roughness. The method is applicable to a variety of substrates and is not limited to small areas or flat surfaces. The polymerized material can be ground into a superhydrophobic powder, which, once applied to a surface, renders it superhydrophobic. The morphology of the porous polymer structure can be efficiently controlled by composition of the polymerization mixture, while surface chemistry can be adjusted by photografting. Morphology control is used to reduce the globule size of the porous architecture from micro down to nanoscale thereby affording a transparent material. The influence of both surface chemistry as well as the length scale of surface roughness on the superhydrophobicity is discussed."
} | 247 |
34729468 | PMC8543385 | pmc | 268 | {
"abstract": "Bioelectrochemical systems (BES) represent a wide range of different biofilm-based bioreactors that includes microbial fuel cells (MFCs), microbial electrolysis cells (MECs) and microbial desalination cells (MDCs). The first described bioelectrical bioreactor is the Microbial Fuel Cell and with the exception of MDCs, it is the only type of BES that actually produces harvestable amounts of electricity, rather than requiring an electrical input to function. For these reasons, this review article, with previously unpublished supporting data, focusses primarily on MFCs. Of relevance is the architecture of these bioreactors, the type of membrane they employ (if any) for separating the chambers along with the size, as well as the geometry and material composition of the electrodes which support biofilms. Finally, the structure, properties and growth rate of the microbial biofilms colonising anodic electrodes, are of critical importance for rendering these devices, functional living ‘engines’ for a wide range of applications.",
"introduction": "1 Introduction The original example of all bioelectrochemical systems (BES) is the microbial fuel cell (MFC), first reported by Potter in 1911 [ 1 ]. MFCs (and MFC derived microbial desalination cells; MDC) are the only type of BES that produce electricity thanks to bioelectrochemical activity of bacteria forming a biofilm on the electrode surface. An MFC consists of two half-cells, i.e. the anode and cathode, usually separated by a semi-permeable membrane material. At initial sterile conditions, and for the same electrode material in both half-cells, there is no potential difference across the circuit. Following colonisation of one of the chambers by a bacterial community, that chamber becomes a negatively charged anode. The cathode usually consists of an oxidising agent (e.g. oxygen from free air), that completes the reaction, and closes the circuit. Closing the circuit, usually by applying an appropriate load resistor allows electrons to flow causing charge to be transferred, releasing the energy produced in the MFC. The main components of the MFC are shown in Fig. 1 a–e, illustrating examples of various designs and configurations. Microbes in the anode compartment are capable of utilising suitable organic substrates allowing them to grow and metabolise. This produces reducing power within the cell that can be tapped and transferred to the anode electrode by a number of mechanisms. These mechanisms are described in detail in Section 1.4. Fig. 1 Examples of membrane-based MFC designs: a) cuboid, double chamber MFC, b) cylindrical double chamber MFC, c) spherical double chamber MFC, d) cuboid MFC with open to air cathode, e) H-type, double chamber MFC. In all designs, “A” and “C” indicate anode and cathode respectively. Inputs and outputs (when these are used for continuous flow) are marked with arrows. Fig. 1 1.1 Architecture of MFCs MFC can be differentiated according to size: macro-, meso- or micro scale. One of the largest MFC systems to be reported is a modularised MFC with a total volume of 1000 L, operated successfully over 12 months [ 2 ]. This system did not consist of a single large MFC of 1000 L, but of 50 stacked MFC modules instead, each with a volume of 20 L. One of the smallest MFCs to be reported is a microfluidic 1.5 μL anode chamber with a 4 μL cathode chamber [ 3 ]. The largest (20 L) and the smallest (totalling 5.5 μL) give more than a 7-log fold difference in size. Therefore, scale and size of the anodic compartment is by far the biggest difference reported in MFCs and is therefore likely to have the largest effect on the formation and behaviour of the biofilms that colonise electrode surfaces. It appears that small scale MFCs (<20 ml) are more energy dense than larger volume systems [ 4 , 5 ]. Moreover, highest power outputs are thought to be maintained from the use of highly permeable, or perfusible electrodes [ 6 ]. What is lesser known is whether or not MFCs containing diverse mixed communities can also show long term functional and ecological stability. This will be discussed in Section 3.3 . However, microfluidic MFCs benefit from unique properties such as – laminar flow, surface tension, capillary forces, fluid-to-surface and fluid-to-fluid interfacial tension inherited from having microfluidic geometry [ 7 ]. The microfluidic scale further eliminates the external influence of inertial forces on the fluidic channel ( Fig. 2 ). Microfluidic MFCs can also be membraned-based (M + MMFC) or membraneless (M−MMFC), however M−MMFC are more common in microfluidic scales, as laminar flow streams create separate layers in common channels. This feature eliminates the requirement for a physical membrane or separation for the anode and cathode. Construction examples range from traditional to photolithographic; photolithographic methods are common in sub-microlitre construction. Conductive polymers such as polyanilines (PANI) or polypyrrols (PPY) [ 8 ] in addition to graphene and traditional carbon materials employed in electrode construction can alleviate geometric constrains [ 9 , 10 ]. For a comprehensive review regarding microfluidic MFC, then readers are referred to Ref. [ 11 ]. Fig. 2 Schematic design of M-MFC: a) top view, b) cross section. Fig. 2 The structure of the stereotypical MFCs as shown in Fig. 1 consists of two chambers separated by a membrane that allows ion exchange [ 12 ]. Membranes used for MFC architecture are expensive and prone to fouling, as well as present material integrity challenges in long term use. To overcome the drawbacks of commercial membranes, alternative materials have been investigated as MFC separators. Ceramics such as earthenware, terracotta or clayware are some of the most commonly used because of their low cost, natural availability, robustness for long-term processes and their low maintenance requirements that facilitates their use in commercial applications [ 13 ] ( Fig. 3 ). On the contrary, polymeric proton exchange membrane (PEM) have the obvious disadvantage that they are not recyclable, and the material is chemically inert and slow to break down or biodegrade in the environment [ 14 ]. Furthermore, they are relatively expensive compared to ceramic membranes [ 15 ]. Fig. 3 Non-polymeric membranes e.g. terracotta MFC with a multi-MFC modular stack. Fig. 3 There are of course those MFC architectures that are truly membraneless, in the sense that there is no separating material between the anode and cathode electrodes. Instead, polarity difference occurs due to redox gradients that are the result of heterogeneous conditions; the classic example describing this is the Winogradsky column and heterogenous conditions are achieved when the cathode electrode is partially in the bulk and partially exposed to air [ 16 , 17 ]. Potential applications include in situ maritime/environmental and weather telemetry instruments [ 18 , 19 ]. This is a well-covered topic and so will not be extensively discussed here. A highly diverse range of microorganisms have been found to be capable of forming biofilms on electrodes, both anodes and cathodes. Species that interact with the anode have been referred to as anodophiles [ 20 ], exoelectrogens [ 21 ], electricigens [ 22 ], electrochemically active microorganisms [ 23 ], anode-respiring [ 24 ] or electrogenic [ 25 ]. Whether or not this terminology accurately describes the real purpose of these microbes in natural ecosystems is open to debate but for the purposes of this discussion, it would suffice to say that in MFCs, the generated power is a direct function of microbial colonisation. 1.2 Mechanisms of electron transfer 1.2.1 Synthetic (exogenous) mediators Allen and Bennetto were the first researchers to utilise synthetic soluble redox mediators as a means to harvesting the reducing power of living bacteria [ 26 ]. It should be noted that any microbial cell within the anodic chamber, whether attached to the electrode or in planktonic suspension may contribute to power generation providing it is permeable to the mediator molecules. One of the earliest mediators used was the redox dye methylene blue, first prepared in 1876 by the German chemist Heinrich Caro [ 27 ]. According to Arup [ 28 ], it was Neisser and Wechsberg in 1900, who first suggested that methylene blue was a useful medium for judging the bacterial contents of milk. Soon after, the methylene blue reduction test was used in the field of dairy manufacturing. The test was based on the fact that the blue colour imparted to milk by the addition of methylene blue, disappears depending on a number of factors, most important of which are the bacterial content of the milk, its growth rate and the operating temperature. If all controlling factors are kept constant other than the bacterial content of the milk, the time required for the colour to disappear will be determined by the number of bacteria. Some of the best artificial mediators known for use in MFCs, in addition to methylene blue are: thionin, neutral red, 2,6-dichlorophenol, indophenol, safranine-O, phenothiazine, and benzyl viologen [ 29 , 30 ] ( Fig. 4 a). Fig. 4 Three main mechanisms for anodic electron transfer from cell reducing power (NADH/NADPH) to the anode electrode in MFC through (a) soluble mediator, (b) direct contact to an outer membrane cytochrome, (c) direct contact to a membrane cytochrome via conductive pili. Fig. 4 1.2.2 Natural (endogenous) mediators Habermann and Pommer [ 31 ] described MFCs that used sulphate reducing bacteria (SRB) to generate hydrogen sulphide that was active at the anode, being oxidised back to sulphate (or other oxyanions of sulphur). Shewanella oneidensis MR-1 secretes flavins (FMN, FAD and riboflavin) in the concentration range of 100–500 nM after 1 week of operation [ 32 ], while phenazines, are also excreted by Pseudomonas aeruginosa [ 33 ]. With the exception of SRB that can reduce sulphate to sulphide as an important part of their central metabolism, the production rate of soluble mediators like FMN and FAD is likely to be slow and they may only produce significant concentrations in batch culture in the stationary phase when many cells are lysing or if the anode environment is poised at the redox level, which is suitable for accelerated mediator generation [ 34 ] ( Fig. 4 a). 1.2.3 Direct electron transfer Up until now, only a few genera and species have been shown to be electrochemically active by means of direct conductive mechanisms; these include Shewanella , Geobacter, Rhodoferax ferrireducens, Pelotomaculum thermopropionicum, Geothrix and Geoalkalibacter . Of these, Geobacter and Shewanella have been studied the most. The first group to report this phenomenon were using Shewanella [ 35 ], that was later named as a cable bacterium [ 36 ]. A summary of important components in the electron transport mechanism from cells to the anode in Geobacter MFCs has been provided by Lovley [ 37 ]. Metabolism of electron rich (reduced) substrate such as acetate or lactate drives the production of reduced NADH from NAD + within the cell. In order for the cell to maintain reducing power, it must re-oxidise NAD + by abstracting electrons by using dehydrogenase and the cytochrome system consisting of quinone/menaquinone pool, periplasmic proteins MacA, PpcA, and outer membrane proteins, OmcE and OmcS. Together these are able to transport the electrons by a series of redox reactions spanning the inner cytoplasmic membrane across the periplasmic space until the electron is conducted across the outer membrane to the anode electrode via the outer membrane cytochromes OmcE and OmcS ( Fig. 4 b). Further transfer of electrons, even within multilayers of cells may also occur, via a dense network of appendages with metal-like conductivity called bacterial nanowires [ 36 ]. The transfer may occur cell by cell over distances of more than 1 cm, until electrons are donated to the electrodes [ 38 ] ( Fig. 4 c). Comparison of the electrode respiring capacity of wild type Shewanella decolorationis S12 and an outer membrane cytochrome-C (OMC)-deficient mutant [ 39 ] showed that the mutant had a much-reduced capacity at producing current, but not zero, probably due to the secretion of flavin molecules, suggesting that some species may use all three transport mechanisms shown in Fig. 4 . The majority of species studied to date are Gram negative organisms. However, Thermincola potens strain JR, is a Gram-positive isolate obtained from the anode surface of a microbial fuel cell [ 40 ]. Despite careful study this species produced no evidence of any soluble redox-active components being secreted into the surrounding medium. Confocal microscopy revealed highly stratified biofilms in which the cells contacting the electrode surface were disproportionately viable relative to the rest of the biofilm. Furthermore, there was no correlation between biofilm thickness and power production, suggesting that cells in contact with the electrode were primarily responsible for current generation. 1.3 Electrodes In general, the higher the macro-scale (geometric) surface area of electrode ( Fig. 5 a), the higher the potential area for accommodating a microbial colony, that would potentially result in higher power output. However, felt, veil or foam electrode materials possess both micro and nano-scale geometries, including pores. Research shows that power output is both electrode area and pore diameter dependent [ 41 , 42 ] and in the case of Shewannella , power is optimal at 5∼7 μm of biofilm thickness [ 41 ]. Recent research shows that controlled micro and nano porous configurations increased current density [ 43 ]. Electrode materials and geometry differ depending on MFC volume; common materials include carbon, carbon composites and mixtures in the form of blocks, rods, brushes, felt, cloth and veil. In constructing microscale electrodes, metal or carbon conductive material is coated thinly by placement or deposition to give a large surface area [ 44 ]. In larger scale electrodes, the use of carbon veil as electrode material proves advantageous due to (i) sufficient microchannels, that allow high permeability enabling nutrient transfer through advective transport via perfusion; (ii) thread continuity, that results in low resistance in comparison to carbon felt or other carbon material discontinuity of strands ( Fig. 5 b). Fig. 5 MFC carbon electrode structure: a) plain carbon electrodes of same geometric macro size compared and b) carbon veil and carbon felt (or mat) compared. SA:V represents the surface area to volume ratio. Fig. 5"
} | 3,664 |
30210278 | PMC6123369 | pmc | 269 | {
"abstract": "We present a massively-parallel scalable multi-purpose neuromorphic engine. All existing neuromorphic hardware systems suffer from Liebig’s law (that the performance of the system is limited by the component in shortest supply) as they have fixed numbers of dedicated neurons and synapses for specific types of plasticity. For any application, it is always the availability of one of these components that limits the size of the model, leaving the others unused. To overcome this problem, our engine adopts a unique novel architecture: an array of identical components, each of which can be configured as a leaky-integrate-and-fire (LIF) neuron, a learning-synapse, or an axon with trainable delay. Spike timing dependent plasticity (STDP) and spike timing dependent delay plasticity (STDDP) are the two supported learning rules. All the parameters are stored in the SRAMs such that runtime reconfiguration is supported. As a proof of concept, we have implemented a prototype system with 16 neural engines, each of which consists of 32768 (32k) components, yielding half a million components, on an entry level FPGA (Altera Cyclone V). We verified the prototype system with measurement results. To demonstrate that our neuromorphic engine is a high performance and scalable digital design, we implemented it using TSMC 28nm HPC technology. Place and route results using Cadence Innovus with a clock frequency of 2.5 GHz show that this engine achieves an excellent area efficiency of 1.68 μm 2 per component: 256k (2 18 ) components in a silicon area of 650 μm × 680 μm (∼0.44 mm 2 , the utilization of the silicon area is 98.7%). The power consumption of this engine is 37 mW, yielding a power efficiency of 0.92 pJ per synaptic operation (SOP).",
"introduction": "Introduction Neurobiological processing systems can easily outperform the most up-to-date computers at robustly accomplishing real-world tasks such as sensory-motor tasks. It still remains largely unknown how biological brains can achieve this with slow, stochastic, and heterogeneous computing elements ( Wang, 2013 ). In the late 1980’s, Caver Mead introduced neuromorphic engineering – a multidisciplinary approach to develop a new generation of computing technologies, building sensory and processing systems using very large scale integration (VLSI) circuits inspired by principles of the biological nervous system. Since the first silicon neuron proposed by Mahowald and Douglas (1991) , significant progress has been made and various designs based on analog, digital and mixed-signal VLSI have been developed ( Wang, 2013 ). Examples include the Neurogrid project which emulates one million neurons connected by six billion synapses ( Boahen, 2006 ), the BrainScaleS project, a wafer-scale neural network, which contains 384 analog network chips for a total of 40M synapses and 200K neurons ( Schemmel et al., 2008 , 2010 ), the SpiNNaker project ( Furber et al., 2014 ), which uses ARM processors to run software neural models and a 48-node SpiNNaker board is capable of simulating 250,000 neurons and 80 million synapses in real time, the IBM TrueNorth chip ( Merolla et al., 2014 ) that is capable of running one million leaky-integrate-and-fire (LIF) neurons in real time. The HiAER-IFAT system has five FPGAs and four custom analog neuromorphic integrated circuits, yielding 262k neurons and 262M synapses. The full-size HiAER-IFAT network has four boards, each of which has one IFAT module, serving 1M neurons and 1G synapses. It is generally accepted that learning in the brain arises from synaptic modifications. Hence, plastic synapses, which can adapt their gain according to one or more adaptation rules, play a vital role in neural systems. The STDP algorithm ( Gerstner et al., 1996 ; Magee, 1997 ; Markram et al., 1997 ; Bi and Poo, 1998 ), is one of the adaptation rules observed in biology. It modulates the weight of a synapse based on the relative timing between the pre-synaptic spike and the post-synaptic spike. Due to its importance, many hardware implementations of this STDP learning rule have been proposed ( Chicca et al., 2003 ; Bofill-i-petit and Murray, 2004 ; Indiveri et al., 2006 ; Häfliger, 2007 ; Koickal et al., 2007 ; Mitra et al., 2009 ; Giulioni et al., 2012 ). Besides weight adaptation, some observations suggest that the propagation delays of neural spikes, as they are transmitted from one neuron to another, may be adaptive ( Stanford, 1987 ). Axonal delays seem to play an important role in the formation of neuronal groups and memory ( Izhikevich, 2006 ). In our previous work ( Wang et al., 2011b , 2012 ), a delay adaptation algorithm, STDDP, inspired by STDP was developed to fine-tune delays that had been programmed into the network. We have successfully implemented the STDDP learning rule on hardware ( Wang et al., 2011a , b , 2012 , 2013a ). However, all these neuromorphic hardware systems are subject to Liebig’s law, because they have a fixed number of neurons and a fixed number of synapses of each specific type of plasticity that was implemented. In most cases, these dedicated components are all hardwired, e.g., one neuron’s learning synapses can’t be used by other neurons. In almost all cases, for any implemented architecture, there is one type of component that will be used up first, leaving the others unused. To address this, we have designed a system where each component can be reconfigured to perform different roles. We have previously presented a hardware implementation of a synaptic plasticity adaptor that is capable of performing either STDP or STDDP ( Wang R. et al., 2014 ; Wang et al., 2015 ). However, that system does not support online reconfiguration: recompiling of the design is required to switch functions. More importantly, this adaptor can only perform learning rules and it can’t be used to implement neurons. Here, we report its follow up work: a massively-parallel scalable multi-purpose neuromorphic engine.",
"discussion": "Discussion and Future Work Our work fills in a vacant niche along the spectrum of neuromorphic platform solutions between integrated specialized hardware for diverse spiking neural network components and software emulation on general purpose hardware, e.g., CPUs and GPUs. This is a previously unexplored point of trade-off that inspires theoretic insights into algorithmic designs of spiking neural networks and warrants hardware investigations. To the author,s acknowledgment, our neural engine is the first and only hardware neuromorphic implementation that is capable of working as a LIF neuron, a learning synapse, and an axon. In the SpiNNaker ( Furber et al., 2014 ) project, ARM processors run software neural models and are programmable. In theory, it should be capable of trading off the number of the neurons and the synapses. However, it is a software implementation, which performs numerical simulations, and hence is not comparable to our hardware implementation that emulates the LIF neurons and plastic synapses on silicon directly. To compare our work with state-of-the-art neuromorphic implementations, a performance and specification summary of them is provided in Table 6 [adapted and completed from Frenkel et al. (2018) ]. On the left are the mixed-signal designs ( Schemmel et al., 2010 ; Benjamin et al., 2014 ; Park et al., 2014 ; Qiao et al., 2015 ; Mayr et al., 2016 ; Moradi et al., 2018 ), digital designs ( Seo et al., 2011 ; Merolla et al., 2014 ; Davies et al., 2018 ; Frenkel et al., 2018 ) together with this work on are on the right. Toward large-scale spiking neuromorphic platforms, the key figures of merit are density, flexibility, synaptic plasticity, and energy consumed per SOP. Table 6 Comparison of with the state of the art of spiking neuromorphic circuits, adapted and completed from ( Frenkel et al., 2018 ). HICANN NeuroGrid ROLLS DYNAPs IFAT Mayr et al., 2016 Seo et al., 2011 TrueNorth Loihi Odin This work Implementation Mixed-signal Mixed-signal Mixed-signal Mixed-signal Mixed-signal Mixed-signal Digital Digital Digital Digital Digital Technology 180 nm 180 nm 180 nm 28 nm 90 nm 28 nm 45 nm 28 nm 14 nm 28 nm 28 nm Area of a neurosynaptic core [mm 2 ] 26.3 168 51.4 7.5 0.31 0.36 0.8/1.15 0.095 0.4 0.086 0.44 Neurons per core 512 64k 256 256 2k 64 256 256 max. 1024 256 256k † Synaptic weight storage 4-bit SRAM Off-chip Capacitor 12-bit CAM Off-chip 4-bit SRAM 1-bit/4-bit SRAM 1-bit SRAM 1- to 9-bit SRAM (3+l)-bit SRAM 3-bit SRAM Embedded online learning STDP No SDSP No No SDSP Probabilistic STDP No Programmable SDSP STDP STDDP Synapse per core 112k – 128k 16k – 8k 64k 64k 1M to 114k (1-to 9-bit) 64k 256k † Neuron model Adaptive exponential IF Adaptive quadratic IF Adaptive LIF Adaptive LIF 2-compartment LIF Adaptive exponential IF Configurable LIF Configurable LIF Adaptive LIF Phenomenological 2-compartment LIF Time constant Accelerated Biological Biological Biological Biological Bio to accel Biological Biological N/A Bio to accel Bio to accel Neuron core density [neur/mm 2 ] ∗ 19.5 390 5 34 6.5k 178 827/575 2.6k Max. 640 3.0k 582k † Synapse core density [syn/mm 2 ] ∗ 4.3k – 2.5k 2.1k – 22.2k 207k/144k 673k 640k to 71k 741k 582k † Supply voltage 1.8 V 3.0 V 1.8 V 1.3 V-1.8 V 1.2 V 0.75 V, 1.0 V 0.53 V–1.0 V 0.7 V–1.05 V 0.5 V–1.25 V 0.55 V–1.0 V 0.9 V Energy per SOP N/A 941 pJ 77 fJ 134 fJ–417 fJ 22 pJ >850 pJ° N/A 26p at 0.775 V >23.6 pJ at 0.75 ‡ 9.8 pJ at 0.55 V 0.92 pJ † Assuming all the components have been configured for neurons or synapses. ∗ Neuron/synapse core density of digital designs is normalized to a 28 nm technology node. Normalization is not applied to mixed-signal designs as the analog parts require redesign to compensate for performance degradation during technology scaling. °Represents a lower bound, estimated by considering the synaptic weights at half their dy ‡ Represents a lower bound as it includes only the contribution of the synaptic operation, without taking into account the cost of neuron update and learning engine update. Our future work will focus on scaling up the network that we have presented here. As a fully digital implementation, the neural engine is a scalable design. The number of physical components, i.e., the ones that can be activated simultaneously, will increase linearly with the number of available logic gates, which are usually the bottleneck for high performance FPGA designs. But in our system, the design of the physical component costs only a few logic gates and plenty of resources are left for additional physical components or other systems modules. High-end FPGAs, such as Xilinx’s Virtex UltraScale+ (XCVU09P of a Xilinx VCU118 kit) have ∼400 M bits on-chip SRAMs. For a system running at 400 MHz, we can achieve 4k TM components with one physical engine (the time slot keeps the same, 100 clock cycles) for a sub-millisecond time resolution. Each engine has 16 physical components and thus 16 × 4k = 64k TM components using 512k bits SRAMs. We can implement 400 M/512k = 800 neural engines, yielding 50 M TM components. Based on the above calculations and analysis, we can conclude that it is practical to scale the proposed system up to a system with 50 M TM components on a commercial off-the-shelf high-end FPGA. Another future improvement will be to accommodate more complicated learning rules. Since this paper is only for proof of concept, we made an arbitrary design choice: for both the STDP and STDDP learning rule, the time window (ramp) needs to become “inactive” before it can be started again. This means the first spike dictates when the learning window begins and ends. The same spike train might have disparate learning behaviors with a potentially small modification in the length of the learning window. This could affect the robustness and efficiency of the learning system, since a small perturbation in a minor parameter might result in different outcomes. Multiple methodologies have been developed in addressing such learning “conflicts,” one of which is a resource-based STDP ( Hunzinger et al., 2012 ). These learning rules are hardware friendly, as they introduced few additional state variables. Due to its unique architecture, the proposed neural engine can be easily adjusted to such learning rules."
} | 3,065 |
34945866 | PMC8700644 | pmc | 270 | {
"abstract": "Reservoir computing is a machine learning method that solves tasks using the response of a dynamical system to a certain input. As the training scheme only involves optimising the weights of the responses of the dynamical system, this method is particularly suited for hardware implementation. Furthermore, the inherent memory of dynamical systems which are suitable for use as reservoirs mean that this method has the potential to perform well on time series prediction tasks, as well as other tasks with time dependence. However, reservoir computing still requires extensive task-dependent parameter optimisation in order to achieve good performance. We demonstrate that by including a time-delayed version of the input for various time series prediction tasks, good performance can be achieved with an unoptimised reservoir. Furthermore, we show that by including the appropriate time-delayed input, one unaltered reservoir can perform well on six different time series prediction tasks at a very low computational expense. Our approach is of particular relevance to hardware implemented reservoirs, as one does not necessarily have access to pertinent optimisation parameters in physical systems but the inclusion of an additional input is generally possible.",
"introduction": "1. Introduction Reservoir computing (RC) is a machine learning method that is particularly suited to solving dynamical tasks [ 1 ]. It was introduced as a way of using recurrent networks for machine learning but circumventing the costly training of the network weights [ 2 ]. The main principle underpinning reservoir computing is that the reservoir projects the inputs into a sufficiently high dimensional phase space such that it suffices to linearly sample the response of the reservoir in order to approximate the desired target for a given task. For this to work, the reservoir must fulfil certain criteria: the response to sufficiently different inputs must be linearly separable, the reservoir must be capable of performing nonlinear transforms, and the reservoir must have the fading memory property [ 2 ]. However, even when these criteria are fulfilled, the performance depends greatly on the dynamics of the reservoir. Hence, in the past two decades a lot of research in the reservoir computing community has focused on the optimisation of the reservoir parameters [ 3 , 4 , 5 , 6 , 7 , 8 , 9 ]. Furthermore, the optimisation of the reservoir is a task-specific problem [ 1 , 10 , 11 , 12 ] and a universal reservoir, which performs well in a range of tasks, remains elusive. In a recent paper [ 13 ], the authors aim to eliminate the issue of hyperparameter optimisation altogether by removing the reservoir. Their approach essentially takes the well-known nonlinear vector autoregression (NVAR) method, uses a less parsimonious approach to filling the feature vector, and adds Tikhonov regularisation. However, the method of [ 13 ] trades the optimisation of the reservoir hyperparameters for the optimisation of the feature vector elements and it cannot be asserted that the latter is generally less costly. Furthermore, one of the main factors driving research into reservoir computing forward is the possibility for hardware implementation [ 14 , 15 , 16 , 17 , 18 , 19 ], which is impractical when the reservoir is absent. In this contribution we demonstrate a new approach that reduces the need for hyperparameter optimisation and is well suited to boosting the performance of physically implemented reservoir computers. Specifically, we show that, by adding a time-delayed version of the input for a given task, the performance of an unoptimised reservoir can be greatly improved. We demonstrate this by using one unaltered reservoir to perform six different time series prediction tasks. In each case the only optimisation parameters are the delay and input strength of the additional delayed input. The aim of this work is not to achieve the best possible performance, but rather to demonstrate that reasonable performance can be achieved for various tasks using the same reservoir and at a very low computational cost. Using time-delayed input is a common approach for adding memory to feedforward networks [ 20 , 21 , 22 , 23 ] and is the basis of statistical forecasting methods [ 21 , 24 ]. However, despite the simplicity of this idea, to the best of our knowledge, time-delayed inputs have not been widely used to optimise the performance of reservoir computers. This may be because the focus has been on constructing reservoirs that have the necessary memory to perform a given task [ 1 ]. One study in which time-delayed inputs have been used to improve the performance of a time series prediction task is [ 25 ]. However, in [ 25 ], the manner in which the time-delayed input was constructed assumed that the memory requirements of the task monotonically decrease with increasing steps into the past and did not allow for the input scaling of the delayed input to be varied as a free parameter. Our results are of particular relevance to the hardware implementation of reservoir computing, because in physical systems one does not always have access to the relevant hyperparameters necessary for optimisation of the task-dependent performance but it should always be possible to add an additional input.",
"discussion": "4. Discussion We have shown that, for various time series prediction tasks, including a delayed version of the input can lead to a substantial improvement in the performance of a reservoir. We have demonstrated this using a simple map describing a semiconductor optical amplifier nonlinearity and a ring-like coupling realised via memory cells. With this approach we were able to use one unaltered reservoir to perform well on six different tasks, each with different memory and nonlinear transform requirements. The performance boost due to the delayed input is achieved over a wide range of the reservoir parameters and was also demonstrated for a time continuous system, indicating that our approach is applicable to a wide range of reservoirs. Our results are significant for a number of reasons. Firstly, we have demonstrated that computationally expensive hyperparameter optimisation can be circumvented by tuning only two input parameters. By including an additional delayed input, reasonable performance can be achieved using an unoptimised reservoir. Nevertheless, we note that, depending on the requirements for a given task, additional hyperparameter optimisation may be necessary. Secondly, to the best of our knowledge, this is the first demonstration of an identical reservoir performing well on such a large range of tasks. Thirdly, the simplicity of our approach means that it is well suited to be applied on physical reservoirs. This study has raised several questions surrounding delay-based reservoir optimisation that require further investigation. For example, the optimal delayed-input parameters are task dependent and how these relate to a given task is not fully understood. The NARMA10 results presented in this study indicate that the optimal delayed-input parameters are related both to the reservoir and requirements of the task. This means that it may be possible to not only use reservoir computing for real-world time series prediction tasks, but also to gain insights into the dynamical systems being investigated. For example, in tasks such as El Niño prediction where the underlying dynamical system is very complex and the relevant physical processes are not fully understood [ 46 ]. Here, investigations surrounding delay-based input could provide critical insight into the involved timescales. Furthermore, the minimum requirements for a reservoir to yield good performance on a range of tasks by only tuning the delayed input parameters remain to be determined. A natural extension of our proposed approach is to include multiple delayed input terms. This would bring the reservoir computing approach closer to classical statistical forecasting methods such as NVAR and could lead to a further improved performance, especially for tasks involving multiple disparate timescales. However, possible performance improvement with added input terms must be weighed against the associated increase in the computational cost as each added input adds two new optimisation parameters."
} | 2,076 |
34499428 | PMC8564445 | pmc | 272 | {
"abstract": "Abstract Remarkable progress has been made in surface icephobicity in the recent years. The mainstream standpoint of the reported antiicing surfaces yet only considers the ice–substrate interface and its adjacent regions being of static nature. In reality, the local structures and the overall properties of ice–substrate interfaces evolve with time, temperature and various external stimuli. Understanding the dynamic properties of the icing interface is crucial for shedding new light on the design of new anti‐icing surfaces to meet challenges of harsh conditions including extremely low temperature and/or long working time. This article surveys the state‐of‐the‐art anti‐icing surfaces and dissects their dynamic changes of the chemical/physical states at icing interface. According to the focused critical ice–substrate contacting locations, namely the most important ice–substrate interface and the adjacent regions in the substrate and in the ice, the available anti‐icing surfaces are for the first time re‐assessed by taking the dynamic evolution into account. Subsequently, the recent works in the preparation of dynamic anti‐icing surfaces (DAIS) that consider time‐evolving properties, with their potentials in practical applications, and the challenges confronted are summarized and discussed, aiming for providing a thorough review of the promising concept of DAIS for guiding the future icephobic materials designs.",
"conclusion": "5 Conclusions and Perspectives Unwanted icing is one of the long‐lasting challenges accompanying mankind in the whole civilization history. It is unfortunate that the most common ways utilized today to combat unwanted icing are still the traditional deicing methodologies involving direct mechanical forces, energy‐intensive thermal treatments or costly chemical handling. Despite that the passive anti‐icing concept has been proposed for more than a decade, automatic ice removal on application scale or at any industrial technology readiness level is still unavailable. Nevertheless, there are significant successes in the anti‐icing materials related researches, especially on the superlow ice adhesion surfaces that enable natural forces for effective ice removal. Although challenges such as durability, anticontamination, functioning at extremely low temperature and harsh conditions need to be addressed, the upscale of such anti‐icing surfaces from laboratory does promise a bright future. In concluding this review, we summarize the dynamic design principles of anti‐icing surfaces through enabling dynamic changes in the chemical/physical states of the ice/substrate/ice–substrate interface with tailored functions by the three DAIS categories, namely, surfaces with dynamic substrates, dynamic interfaces, and dynamic ice. Owing to the abilities of response to stimuli, DAIS with dynamic substrates possess highly appealing potentials. This type of anti‐icing surfaces can utilize various stimuli like temperature, magnetic field, and light for enabling surfaces icephobicity, although not all external stimuli are available in harsh icing situations. For making more practical DAIS in this category, taking the advantages of the intrinsic properties of water and ice for triggering responses of the substrates is the most feasible approach, given that water and ice are always the existing component in the anti‐icing operations. It can be envisioned that novel DIAS with substrates responding sensitively to the interfacial water/ice in cold temperature can provide sensible possibilities in application. DAIS with dynamic interfaces mainly focus on the ice–substrates interaction. To enable dynamic evolution of the interface, the following two pathways are recommended. The first is to enable dynamic change at the interface states, either by generating new interlayers to replace the ice–substrate interfaces or by melting the interface from rigid solid to lubricating liquid. The second is to enable dynamic evolution of the interactions between ice and substrates. This will require better understanding on the fundamental interactions between ice and solid. New interface layers and lubricating layers no doubt have reduced ice–substrates interactions. However, the practical approach of utilizing the intrinsic properties of ice and solids for lowering ice–substrates interactions is still missing. Further investigations that can accelerate such evolutions and apply them for effectively weakening ice adhesion are required. The current DAIS with dynamic ice indicates that ice evolution after formation is not independent from surface properties. Both structures and chemical components of the surfaces can be predesigned for affecting ice propagation, growth and following evolution. It should be noted that both ice growth/propagation inhibitors and programming ice‐free zone cannot make surfaces free of ice. These methods provide alternative choices in specific situations when ice‐free surfaces in confined regions and limited time are required. Dynamic ice melting using solar illumination can be a better choice for outdoor large‐scale anti‐icing applications. However, more studies on the improvement of solar absorption efficiency of these surfaces, especially with ice/snow covering, are still highly essential before these surfaces are ready for practical applications. The DAIS also offers alternative pathways for solving the durability problem of icephobic surfaces. Generally, anti‐icing surfaces are designed by modifying the contacts between water/ice and surfaces, for instance, using micro/nanostructured surfaces for depressing water–surface interaction and designing surfaces with low surface energy for weak ice–surface interactions. Such surfaces have poor durability owing to the degradation of the surface topography in usage, because surface nanostructures and chemicals degradation are almost unavoidable in practice. The icephobicity of SAIS commonly suffers from such irreversible decay. There is an unavoidable trade‐off in the performance of SAIS, namely, choosing hard materials for the increase the durability by sacrificing of a decrease in icephobicity or using soft materials for the increase the icephobicity by sacrificing of a decrease in durability. The DAIS pinpoints the importance of evolutions after ice formation in anti‐icing surfaces design. One promising durable icephobicity could be achieved by integrating tough materials/structures with large capacity of abrasion resistance. Such tough materials/structures can provide long‐term durability, while the dynamic components in the surface can weaken ice–substrate interactions and provide excellent icephobicity. More works focus on enabling dynamic functions into robust surfaces and on extending the life spans of dynamic components can promote practical applications of the design of durable icephobic surfaces. There is still a long way for well‐established DAIS approaching large‐scale applications. First, DAIS such as stimuli‐responsive surfaces and photothermal trap surfaces integrate expensive functional components, which significantly increase the cost of large‐scale fabrication. Although alternative methods, such as utilizing PTSLIPS, [ \n \n 36 \n \n ] is relatively low‐cost with abundant materials available, the scale‐up performance still needs further exploration. Furthermore, there is currently no large‐scale demonstration of DAIS available. Extensive studies focusing on scaling up fabrication will speed up transformation of DAIS from laboratory products to practical application. Given the abundant research results on anti‐icing materials, the understanding of the icing problem also needs to be updated. Regarding ice formation on a surface, it is crucial to realize the dynamic evolution of the interaction between the ice and its substrates. For the practical usage, the surfaces that can maintain long time‐span dynamic properties and can respond to external stimuli are of promising potentials. For instance, ice adhesion on the liquid layer generators and other surfaces that secrete interfacial lubricant can constantly decrease to a negligible level with time. As such, automatic ice removal can be realized even in the outdoor environment if the functionality life‐time of such surfaces is further enhanced. Several anti‐icing surfaces have combined external energy inputs in design. It is fair to say that energy input from solar illumination is one of the most favorable and sustainable choices. Such surfaces still have relatively low power density or suffer from critical environmental limitation such as extremely low temperature, which deserves continuous research effort in the future research. Last but not least, the rational structure–property–function relationship in the current anti‐icing surfaces is still missing. It is certainly difficult to establish this relationship given that approaches to realize surface icephobicity vary greatly as well as a wide variety of materials have been selected for surface fabrication. More theoretical researches on DAIS and new approaches should be promoted, including atomistic modeling and simulation, multiscale approaches, and even machine learning. Based on the current fast evolving status of anti‐icing research, the society can hold an optimistic attitude on unwanted icing and expect practical passive dynamics anti‐icing surfaces in a very near future.",
"introduction": "1 Introduction Icing is one of the most common natural phenomena that greatly impact human activities. Undesired ice formation and accumulation can result in numerous safety problems to aircraft, power grid, transmission line, roadway, marine vessel, renewable energy infrastructure, and many others. [ \n \n 1 \n , \n 2 \n , \n 3 \n , \n 4 \n , \n 5 \n , \n 6 \n , \n 7 \n , \n 8 \n \n ] Traditional methods used for dealing with icing problems, for instance, mechanical deicing, thermal or chemical treatments, are often highly costly and at the same time low‐efficient. [ \n \n 9 \n , \n 10 \n \n ] As such, enormous interests have been aroused in deploying surfaces that can control icing and mitigate its related damages. The so‐called icephobic or anti‐icing surfaces with properties like repelling incoming water droplets, delaying ice nucleation, repressing ice growth and weakening ice adhesion are designed for anti‐icing purpose. [ \n \n 11 \n , \n 12 \n , \n 13 \n , \n 14 \n \n ] From the early lotus‐leaf inspired superhydrophobic surfaces (SHS) fabricated for repelling water droplets and delaying ice nucleation to the recent omniphobic pitcher‐plants‐inspired slippery liquid‐infused porous surfaces (SLIPS) developed for multiple anti‐icing, [ \n \n 11 \n , \n 12 \n , \n 15 \n , \n 16 \n , \n 17 \n , \n 18 \n , \n 19 \n , \n 20 \n , \n 21 \n , \n 22 \n , \n 23 \n , \n 24 \n \n ] there are currently a colorful spectrum of anti‐icing surfaces reported in the literature showing great potentials with practical low ice adhesion strength 0.2–10 kPa (generally, icephobic surfaces are defined as τ \n ice < 100 kPa, and the passive removal of ice requires much lower value τ \n ice < 10 kPa) and easy achievable large‐scale ice remove capacity. [ \n \n 14 \n , \n 25 \n , \n 26 \n , \n 27 \n \n ] \n Despite the remarkable progress already made in surface icephobicity in the recent years, the anti‐icing surfaces are generally designed from a static perspective, for instance, texturing the surfaces structures, tuning the modulus of substrates, and modifying the surfaces energy without considering the evolution of properties. [ \n \n 11 \n , \n 12 \n , \n 13 \n , \n 14 \n , \n 15 \n , \n 16 \n , \n 17 \n , \n 18 \n , \n 19 \n , \n 20 \n , \n 21 \n , \n 22 \n , \n 23 \n , \n 24 \n \n ] One of the key characteristics of these surfaces is that the ice–substrate contact area after ice formation is regarded as “constant.” The main focus of surface design lies on the states before ice formation but not the dynamic changes after icing. For instance, the anti‐icing performance of SHS including repelling incoming water and delaying ice nucleation has been widely discussed, [ \n \n 11 \n , \n 12 \n , \n 17 \n , \n 18 \n , \n 19 \n \n ] with attentions only on the phenomena before ice formation. However, once ice forms on surfaces, especially in icing/deicing cycles, the anti‐icing performance of the SHS decays very fast because the surfaces asperities can be easily damaged. [ \n \n 18 \n \n ] Another commonly used anti‐icing surface SLIPS with lubricant layer atop that enable the surfaces with excellent repellence to any immiscible materials also focus on the surface activities before icing. After ice formation on SLIPS, the lubricants can be exhausted easily due to ice removal. [ \n \n 24 \n \n ] Generally, the static anti‐icing surfaces (SAIS) have disadvantages, for example, inapplicability at extremely low temperature, fragility to surfaces damage and surfaces degradation, and inadaptability to environmental changes. [ \n \n 16 \n , \n 17 \n , \n 18 \n , \n 28 \n , \n 29 \n , \n 30 \n , \n 31 \n , \n 32 \n , \n 33 \n \n ] \n Very recently, we witness a shift in anti‐icing surface design principles from being of static nature, namely, from no consideration on the changes at the ice–substrate contacting areas after ice formation, to the focus on enabling dynamic changes of the chemical/physical states of the ice/substrate/ice–substrate interface for enhanced anti‐icing performances. [ \n \n 34 \n , \n 35 \n , \n 36 \n , \n 37 \n , \n 38 \n , \n 39 \n , \n 40 \n \n ] These emerging anti‐icing surfaces can be regarded as dynamic anti‐icing surfaces (DAIS), thanks to the integrated evolving properties that can mitigate the interactions between ice and substrate even after ice formation. Extraordinarily, certain effective approaches in the traditional active deicing methodologies were featured in the design of DAIS. Unlike the direct and abrupt application of external mechanical, chemical, and electrical energy as driving force to remove ice, DAIS integrate the active strategies into the evolution of the ice–substrate contacting regions to reduce the resistance of removing ice. For instance, integrating dynamic antifreeze‐secreting capacity into superhydrophobic surfaces to introduce continuous assistance for ice removal even after ice formation. [ \n \n 34 \n \n ] This strategy inherits the advantages of superhydrophobic surfaces in preventing ice forming and triggers evolution of properties at icing region after ice formation for easy ice removal. New anti‐icing surfaces with photothermal traps can actively melt interfacial ice, which utilize solar energy in deicing directly. [ \n \n 38 \n \n ] Importantly, DAIS exhibit improved durability, wider temperature tolerability, and better environment adaptivity, and thus have gained increasing interests in the anti‐icing research and application fields. [ \n \n 14 \n , \n 35 \n , \n 36 \n , \n 37 \n , \n 38 \n , \n 39 \n , \n 40 \n \n ] \n This review aims to provide a thorough survey on the newest development of DAIS. Focusing on the most relevant ice–substrate interfacial regions ( Figure \n 1 a ), and their spontaneous/stimuli‐responsive changes in the chemical/physical states impacting ice adhesion during and after ice formation. The state‐of‐the‐art anti‐icing surfaces are reassessed and classified into three categories, namely, surfaces with dynamic substrate, dynamic interface, and dynamic ice changes, as shown in Figure 1b . Surfaces with dynamic properties in the substrate generally include functional structures that respond to internal and external stimuli, and thus modify the substrate properties and enhance anti‐icing performances. [ \n \n 36 \n , \n 37 \n , \n 41 \n , \n 42 \n , \n 43 \n , \n 44 \n \n ] Surfaces with dynamic properties at the ice–substrate interfaces provide the possibility of altering interface interactions for lowering ice adhesion. [ \n \n 14 \n , \n 35 \n , \n 39 \n , \n 40 \n , \n 45 \n , \n 46 \n , \n 47 \n \n ] Surfaces with dynamic ice are able to direct ice growth, propagation, and even ice melting, which can mitigate ice accumulation and assist ice removal on the surfaces. [ \n \n 13 \n , \n 38 \n , \n 48 \n , \n 49 \n , \n 50 \n , \n 51 \n , \n 52 \n , \n 53 \n , \n 54 \n \n ] One can see that DAIS is introduced in this review with the purpose of promoting rethinking of the design principles of new anti‐icing surfaces rather than recategorizing of the known anti‐icing materials. It should be emphasized that the traditional deicing methodologies are still actively in use despite their obvious shortcomings. Under the DAIS concept, the focus is to utilize the evolution of the ice–substrates and its adjacent region in materials design, but not to distinguish active or passive deicing. The following sections detailed the categories of DAIS surfaces. Figure 1 DAIS. a) The three most important regions close to the ice–substrate interface determine anti‐icing performance of a surface. b) DAIS targeting the three ice–substrate interfacial regions. The dynamic substrates include substrates that can respond to the internal/external conditions, namely those by tuning the surface state and affecting the ice formation/adhesion on the top. The dynamic interfaces cover the surfaces that can introduce dynamic evolution in the chemical/physical states of the ice–substrate interface after ice formation, thus facilitating easy ice removal. The dynamic ice encompasses the surfaces that can tailor ice growth, propagation or even melt ice for the purpose of mitigating ice accumulation."
} | 4,359 |
31572647 | PMC6760469 | pmc | 274 | {
"abstract": "Abstract Engineering surface wettability is of great importance in academic research and practical applications. The exploration of hydrogel‐based natural surfaces with superior properties has revealed new design principles of surface superwettability. Gels are composed of a cross‐linked polymer network that traps numerous solvents through weak interactions. The natural fluidity of the trapped solvents confers the liquid‐like property to gel surfaces, making them significantly different from solid surfaces. Bioinspired gel surfaces have shown promising applications in diverse fields. This work aims to summarize the fundamental understanding and emerging applications of bioinspired gel surfaces with superwettability and special adhesion. First, several typical hydrogel‐based natural surfaces with superwettability and special adhesion are briefly introduced, followed by highlighting the unique properties and design principles of gel‐based surfaces. Then, the superwettability and emerging applications of bioinspired gel surfaces, including liquid/liquid separation, antiadhesion of organisms and solids, and fabrication of thin polymer films, are presented in detail. Finally, an outlook on the future development of these novel gel surfaces is also provided.",
"conclusion": "6 Conclusion and Perspective In this review, we summarize recent progress and emerging applications related to bioinspired gel surfaces with superwettability and special adhesion. Biological hydrogel‐based surfaces, including carp scales, filefish skins, seaweed surfaces, the toe of the tree fog, Nepenthes pitcher, and articular cartilage, show superwettability and special adhesion, as well as outstanding performance in their native environments. It has been suggested that the achievement of these unique properties is attributed to the unique liquid‐like gel surfaces, which can be engineered by designing polymers, solvents, and surface structures. By using such a design principle, artificial gel surfaces with superwettability and special adhesion have been developed and exhibit unprecedented performance in diverse applications, such as liquid/liquid separation, antibiofouling, antisolid adhesion, antifriction, and the synthesis of functional thin polymer films. Among these, we suggest a new principle for the design and development of next‐generation advanced materials with superior surfaces for practical applications. Although considerable progress has been made in this field, the investigation of bioinspired gel surfaces with superwettability and special adhesion is still in its infancy, and numerous challenges in both fundamental research and practical applications must be addressed with great efforts. First, the exploration of novel natural livings with functional surfaces, especially those spending most of their lifetime in wet, rainy, or aqueous habitats, is pivotal to enable the discovery of unique mechanisms and design principles for advanced gel surfaces. To achieve this goal, the issues encountered for current gel surfaces, such as the biocompatibility of gels when used in medical devices, 64 , 65 have solutions with much higher possibilities. Second, only a limited number of polymers, solvents, and surface structures have been used to fabricate gel surfaces. There are abundant polymers (supramolecular polymers, conducting polymers, biomacromolecules, etc.), solvents (metallic liquids, magnetic liquids, liquid crystals, etc.), and surface structures (gradient structures, asymmetric structures, multiscale structures, etc.), which could be utilized to design and develop new gel surfaces. Therefore, there are many opportunities to promote the development of advanced gel surfaces by engineering the polymers, solvents, and/or surface structures. Additionally, intensive efforts should be exerted toward the exploitation of novel materials and advanced technologies for the fabrication of bioinspired gel surfaces in a facile, efficient, cost‐effective, and scalable way. Synthetic gel surfaces with multiple functions, such as biocompatibility, environmentally friendly, and long‐term stability/durability under harsh conditions, remain a challenge. The combination of solid surfaces with water, air, and oil can create a total of 64 wetting states. 1 However, as shown in Figure 2 , only several wetting states toward water and oil have been investigated for gel surfaces to date. Therefore, the superwettability of gel surfaces is far from well understood. We should attribute great efforts to explore the possible wetting behaviors of water, oil, and gas 170 on gel surfaces under various conditions, such as in air, underwater, or under organic liquids. Bioinspired gel surfaces with superwettability and special adhesion hold potential to overcome the challenges faced in diverse applications. For instance, scalable fabrication of functional composite thin films with highly ordered structures 171 challenges the traditional methods, e.g., interfacial polymerization at the oil/water interface. 172 The superspreading of liquids on immersed gel surfaces provides a promising way to address such problem by utilizing the spreading‐induced shear forces and formed thin liquid layers. Although numerous efforts have been devoted to improving the performance of implants, 173 , 174 , 175 , 176 such as the blood vessel prosthesis, nearthrosis, and contact lens, none of them can compare favorably with natural ones due to the lack of some essential features, for example, biocompatibility, durability, and self‐repair ability. 177 , 178 In this case, gel‐based surfaces may be useful because their similar framework to the extracellular matrix of organs offers the opportunity to coat implants to minimize the chance of rejection. 179 , 180 , 181 Therefore, numerous opportunities are available to settle critical issues in areas ranging from chemistry and materials science to biology. Along with these opportunities, we envision the development of new concepts and ideas that will reshape our understanding of bioinspired gel surfaces, which will significantly benefit our healthcare and daily life in the near future.",
"introduction": "1 Introduction Natural evolution and election has led to the development of numerous mysterious living organisms endowed exquisite surfaces with unique properties, especially superwettability. 1 , 2 , 3 , 4 , 5 As the first and well‐known example, lotus leaf has been widely studied because of its self‐cleaning feature, which has been demonstrated to be attributed to its superhydrophobicity and low water adhesion in air. 6 , 7 To date, extensive studies on biological surfaces, such as rose petals, 8 rice leaves, 6 and gecko feet, 9 have revealed that the superwettability arises from the synergy of the surface chemical composition and multiscale structures. Following the above principle, scientists have established a superwettability system, motivating the explosive development of functional solid surfaces with exceptional functions for extraordinary applications from agriculture and industry to our daily life. 4 , 10 , 11 , 12 , 13 , 14 In recent decades, solid surfaces with superwettability have been a focus of research. Although significant progress has been made, solid surfaces still present some severe limitations in practical applications. For example, solid surfaces, which are usually rigid, are rarely used when there is a need for soft and deformable properties. Additionally, solid surfaces suffer from the loss of superwettability due to the irreversible damage of their surface micro/nanostructures or surface chemical compositions under severe conditions. 15 Thus, new design principles and engineering strategies are urgently needed to develop artificial surfaces with superwettability and special adhesion to address the above challenges. Organisms living in their natural environments, such as fishes, 16 seaweed, 17 and tree frogs, 18 employ soft organic materials, especially hydrogels, which mainly consist of proteins and polysaccharides, as their surficial materials. In contrast to solid materials, hydrogels are composed of a water phase entrapped in a three‐dimensional (3D) cross‐linked polymer network through weak interactions (for example, hydrogen bonds and van der Waals forces). 19 , 20 , 21 Accordingly, hydrogels behave like a solid that can maintain their shapes. Moreover, hydrogels are wet and soft and can be considered to be in a quasi‐liquid phase, in stark contrast to solid materials, which are usually dry and hard. The liquid phase trapped inside the cross‐linked polymer networks can maintain their mobility to some extent, causing the hydrogels to have liquid‐like surfaces. This unique property makes gel surfaces excellent candidates for the development of new artificial surfaces with superwettability and special adhesion to address the issues encountered by solid surfaces. As expected, bioinspired gel surfaces have emerged as promising materials in diverse fields. For example, hydrogel surfaces inspired by fish scales show underwater superoleophobicity and have been used for highly efficient water/oil separation. 22 Organogel surfaces inspired by the Nepenthes pitcher have extremely low friction with immiscible liquids, organisms, and solids, showing promising applications as self‐cleaning coatings. 23 \n In this review, we summarize recent progress in bioinspired gel surfaces with superwettability and their emerging applications. First, we will introduce several representative hydrogel‐based natural surfaces with superwettability and special adhesion, and then highlight the unique liquid‐like properties, design principles, and fabrication strategies of bioinspired gel surfaces. Subsequently, several established types of gel surfaces with superwettability will be discussed in detail. Next, the emerging applications of bioinspired gel surfaces, including liquid/liquid separation, antibiofouling, antisolid adhesion, antifriction, and fabrication of functional thin polymer films, will be introduced in detail. Finally, we conclude this review by presenting the challenges and perspectives related to these functional gel surfaces."
} | 2,550 |
38349619 | PMC10906085 | pmc | 275 | {
"abstract": "Ionic memristor devices are crucial for efficient artificial\nneural\nnetwork computations in neuromorphic hardware. They excel in multi-bit\nimplementation but face challenges like device reliability and sneak\ncurrents in crossbar array architecture (CAA). Interface-type ionic\nmemristors offer low variation, self-rectification, and no forming\nprocess, making them suitable for CAA. However, they suffer from slow\nweight updates and poor retention and endurance. To address these\nissues, the study demonstrated an alkali ion self-rectifying memristor\nwith an alkali metal reservoir formed by a bottom electrode design.\nBy adopting Li metal as the adhesion layer of the bottom electrode,\nan alkali ion reservoir was formed at the bottom of the memristor\nlayer by diffusion occurring during the atomic layer deposition process\nfor the Na:TiO 2 memristor layer. In addition, Al dopant\nwas used to improve the retention characteristics by suppressing the\ndiffusion of alkali cations. In the memristor device with optimized\nAl doping, retention characteristics of more than 20 h at 125 °C,\nendurance characteristics of more than 5.5 × 10 5 ,\nand high linearity/symmetry of weight update characteristics were\nachieved. In reliability tests on 100 randomly selected devices from\na 32 × 32 CAA device, device-to-device and cycle-to-cycle variations\nshowed low variation values within 81% and 8%, respectively.",
"conclusion": "Conclusions Memristor devices are essential circuit\nelements for the implementation\nof neuromorphic hardware, but reliability issues due to sneak current\nand large cycle-to-cycle and device-to-device variations in the crossbar\narray structure have been long-standing obstacles. In this study,\nwe developed an alkali ion-based self-rectifying memristor device\nand the fabrication method using the diffusion and doping effect of\nthe adhesion layer. The device exhibited relatively high reliability\nin terms of cycle-to-cycle and device-to-device variation and sneak\ncurrent issues. As the name “interface type” implies,\nthe resistance change event occurs at the interface, and the control\nof the interface is very important. When we consider that the ALD\nprocess is generally performed at 150 °C ∼ 300 °C\nexcept for amorphous material deposition, interdiffusion occurring\nat the interface with the BE is a very important factor that should\nbe considered for interface memristors. We derived the formation of\nalkali ion reservoir layers at the top and bottom interfaces by using\nthe diffusion effect of an adhesion layer and obtained a self-rectifying\nmemristive characteristic by the electrochemical migration of alkali\ncations. Since the alkali cations have high mobility in the solid,\nit exhibits faster switching speed and higher reliability than the\noxygen anion-based interface type memristor.",
"introduction": "Introduction The resistive switching (RS) characteristics\nresulting from the\nstate change of materials, including ionic memristor, ferroelectric\ntunnel junction (FTJ), phase-change memory, and magnetic tunnel junction\n(MTJ), have been considered promising candidates as next-generation\nnonvolatile memories (NVMs). 1 − 10 They have in common the characteristic to express nonvolatile multi-bit\n(or analog) resistance states, depending on the magnitude, duration\ntime, and number of programming pulses, used in a two-terminal structure,\nenabling them to be utilized as synapse devices in neuromorphic hardware.\nThe matrix-vector multiplier using memristors with a crossbar-array\narchitecture (CAA) is the foremost scheme for chip implementation\n(analog accelerator) of artificial neural networks (ANNs). 11 − 14 Among the promising memristors, the ionic memristor is particularly\nsuited for neuromorphic and analog computation in terms of switching\nspeed, simple structure (metal/insulator/metal), availability of various\nmaterials (transition metal oxides, TMOs), and multi-bit capability\n(larger on/off ratio). 13 − 16 However, long-standing issues such as the sneak current issue in\nCAA and reliability issues (such as device-to-device and cycle-to-cycle\nvariation) caused by the RS mechanism “defect control”\nhave hindered neuromorphic hardware implementation. 17 − 19 The RS\ncharacteristics of ionic memristors observed in TMOs are\nclassified into filament-type and interface-type. 6 , 20 In\nthe case of filament-type RS, it is mainly observed in TMO thin films\nwith high resistance. When a high voltage is applied with a compliance\ncurrent, a conductive path (called filament) consisting of a large\namount of oxygen vacancy inside of the oxide film can be formed by\nsoft breakdown, and then a part of the conductive filament is reversibly\nbroken/recovered by an appropriate programming voltage to implement\nthe RS characteristics. 6 , 19 , 20 On the other hand, the interface-type RS is a characteristic observed\nin relatively lower resistance TMO films, which induces a change in\nthe interface resistance between the electrode and the oxide by redistribution\nof oxygen vacancies in the TMO film when the programming voltage is\napplied. 6 , 19 , 20 Therefore,\na filament-type memristor can be considered a small memristor in an\ninsulating thin film and has a parallel structure of a capacitor and\na memristor. Filament-type memristors have the disadvantage of requiring\nan essential filament-forming process, but they generally exhibit\nfaster switching speeds and stable retention and endurance characteristics\ncompared to interface-type memristors. 17 − 20 And, a hallmark of the RS behavior\nof filament-type memristors is the abrupt resistance change characteristic\nduring the break/recovery of local conducting filaments. This characteristic\nis due to the fact that the resistance change behavior occurs within\na very small filament, resulting in a large amount of current passing\nthrough the narrow filament and a fast switching speed. In the past,\nfilament-type memristors have been mainstream for NVM (resistive RAM)\napplications that require only 1-bit implementation, 21 − 23 but recently, interface-type memristor devices that exhibit gradual\nswitching characteristics have attracted attention in neuromorphic\nhardware applications that require multi-bit implementation. 1 , 10 , 11 , 13 , 14 Since filament-type memristors have difficulty\ncontrolling filament formation and ensuring uniformity, the device-to-device\nand cycle-to-cycle variation is larger than that of interface-type\nmemristors. In addition, because they exhibit an Ohmic conduction\nregime in the low-resistance state (LRS), a cell selector such as\na transistor or ovonic threshold switching (OTS) device is essential\nto avoid errors caused by the sneak current in the CAA. 24 − 26 Filament-type memristors, which typically demand operating currents\nof the mA range for programming, require a relatively large channel\nof transistors for a cell selector, which is disadvantageous in terms\nof integration. In addition, when OTS and memristors are stacked in\na series structure, the compatibility issue in the operating voltage\nand current of the OTS and memristor devices still hinders reliability.\nAbrupt overvoltage or overcurrent delivered to the memristor layer\nduring switching of the OTS element causes damage such as hard breakdown\nor irreversible RS. On the other hand, interface-type memristors utilize\ninterface resistance, so self-rectifying RS can be implemented by\nasymmetric interface configuration, which can implement selector-less\nCAA. However, the interface-type memristor is relatively inferior\nto the filament-type memristor in terms of weight update (switching)\nspeed, retention, and endurance (deterioration by oxygen vacancy clustering),\nand these properties should be improved. 27 − 31 One of the solutions for this is to use alkali metal\ncations with high mobility in solids, such as Li and Na, instead of\noxygen anion. 10 , 32 Highly mobile alkali cations\nare expected to offer improvement in switching speed, and simultaneity,\nit is disadvantageous for retention characteristics. Therefore, in\norder to improve the retention characteristics, we need to consider\nintroducing a diffusion barrier to prevent the redistribution of alkali\nions in the oxide matrix by the program voltage from returning to\nequilibrium or a reservoir layer that can safely store alkali cations. The introduction of alkali metals in the fabrication process for\non-chip analog accelerators may pose compatibility issues with CMOS\nprocesses. The CAA is formed in the back end of line (BEOL) process.\nAt this stage, the insertion of an alkali metal layer on top of the\ncompleted CMOS devices in the front end of line (FEOL) could lead\nto diffusion concerns into the transistor (selector) components beneath\neach synapse cell. However, unlike in active arrays, the passive array\nwith self-rectifying memristors eliminates the need for transistors\n(selectors) beneath each synapse cell, thereby mitigating potential\nissues with CMOS devices by diffusion. Additionally, introducing a\ndiffusion barrier can minimize the spread of alkali ions. In the case\nof cross-contamination issues in the process, another consideration\ncould be combining the drive chip and the synapse array devices, manufactured\nseparately, through chip bonding. We expect such methods to circumvent\ncompatibility issues with the CMOS process. In general, the\nredox phenomenon of the electrodes at the interface\nof the memristor oxide layer and the electrodes can also contribute\nand help to induce RS, 33 − 35 but in order to avoid the redox effect with the electrodes\nand consider only the ion migration effect inside of the TMO thin\nfilm (or to ensure reliability), novel metals such as Au and Pt (not\neasily redoxable) are used. However, novel metals require an adhesion\nlayer due to their weak adhesion to the oxide (insulating substrate),\nand the adhesion layer diffuses during the device fabrication process,\ncausing the oxide layer to be unintentionally doped with the adhesion\nlayer material. 36 In this study, we propose\na strategy for fabricating an alkali ion-based memristor device by\nutilizing the doping effect of the adhesion layer. Since lithium is\nrapidly oxidized and forms stable Li x O\nwhen exposed to the atmosphere, a high-temperature process is required\nto dope the TMO (memristor) film. However, when the Li metal is used\nas an adhesion layer of the BE, it can be safely protected from oxidation\nby the Au (BE) layer until the oxide film (memristor layer) is deposited,\nand at the same time, it can be effectively doped locally by diffusion\nduring the memristor deposition process (supplied to the oxide layer).\nIn this case, the asymmetric interface of the TE and BE is formed,\nwhich facilitates the implementation of a self-rectifying memristor\nwith high selectivity, and the diffused Li ions form a Li reservoir\nlayer at the bottom interface, which can be expected to improve the\nendurance characteristics similar to previous reports such as bilayer\noxide memristor devices (active layer/oxygen reservoir layer). 37 , 38 We have previously reported highly reliable memristor characteristics\nwith in-situ ALD-grown Na:TiO 2 memristors\nusing NaOH aqueous solution as a homemade reactant. 10 In this study, we adopted the Na:TiO 2 as the\nmemristor layer to maximize the self-rectifying characteristics. (See Figure S1 ). In addition, we performed a systematic\nexperiment on the memristor characteristics according to the amount\nof Al doping to suppress the diffusion of alkali cations in the TiO 2 matrix (for the retention characteristic). As a result, with\nthe increase of Al doping amount, the retention characteristics were\nimproved, but simultaneously, the operating current and switching\nspeed decreased. We also observed that Al doping changes the linearity\nand symmetry of the long-term potentiation/depression (LTP/LTD) curve\nunder fixed weight update pulse conditions. Although there was a trade-off\nbetween switching speed and retention characteristics, the controllability\nof retention characteristics, switching speed, and linearity and symmetry\nof weight update by the Al doping can be useful control factors to\nimplement device characteristics required for the neuromorphic hardware.\nAs a result, retention characteristics were improved up to 20 h at\n125 °C and +1 V reading conditions, and no degradation was observed\nin the 550,000 cycles endurance test.",
"discussion": "Results and Discussion Figure 1 a represents\nthe atomic layer deposition (ALD) supercycle configured to grow Al,Na:TiO 2 in this experiment. The Ti cycles in a supercycle consisted\nof n cycles of [titanium isopropoxide (TTIP) pulse\n1s → Ar purge 5s → reactant pulse 1s → Ar purge\n5s] for Na:TiO 2 and 1 cycle of [trimethylaluminum (TMA)\npulse 1s → Ar purge 5s → reactant pulse 1s →\nAr purge 5s] for Al doping. Al cycles were organized in the ratio\nof 1:0, 83:1, 41:1, and 27:1, and the numbers of Al cycles were 0,\n3, 6, and 9 within a total of 250 cycles to complete four different\nAl,Na:TiO 2 memristors of equal thickness. (Hereafter, the\nfour samples are referred to as Al 0, Al 3, Al 6, and Al 9.) Aqueous\nNaOH solution was used as a homemade reactant in all the processes\nto induce in-situ Na doping during the film growth\nprocess. 10 Figure 1 b shows a schematic (left) and a scanning\ntransmission electron microscope (STEM) image (right) of the completed\ndevice structure, which is fabricated as a Au(80 nm)/Li(5 nm)/Al,Na:TiO 2 (18 nm)/Au(40 nm)/Li(5 nm) structure on a SiO 2 (300 nm)/p-Si substrate. Here, the Li adhesion layer underneath\nthe BE is diffused out to the top of the BE and doped on the TiO 2 bottom interface in a 250 °C ALD process of 2 h. In\nthe bright field STEM image in Figure 1 b (light elements are shown as bright, and heavy elements\nare shown as dark), the Li adhesion layer between the Au BE and the\nSiO 2 substrate is identified. In addition, a large amount\nof bright-colored excess Li or Na elements was also identified at\nthe top and bottom interfaces of the TiO 2 layer. In the\nmemristor device, these two regions serve as an alkali metal reservoir,\nand the TiO 2 layer acts as a diffusion barrier, which provides\nstable storage for Li ions redistributed by the program voltage (accumulated\nat the interface) to improve the retention characteristics and endurance\ncharacteristics. Figure 1 c is a schematic illustration of the doping process of the Li adhesion\nlayer in the Al,Na:TiO 2 growth process and the defect species\nthat can be induced. As shown in the figure, the three doped elements,\nLi, Al, and Na, can occupy the Ti site (substitutional defects: Li Ti ’’’ , Al Ti ’ , Na Ti ’’’ ) or interstitial sites (interstitial\ndefects: Na i • , Li i • , Al i ••• ) in the TiO 2 matrix. Broken charge neutrality by those defects can also\ninduce oxygen vacancies ( V O •• ) and\nTi defect occupying interstitial sites ( Ti i •••• ) to compensate for charge valence. 39 In\nthe case of substitutional defects ( Li Ti ’’’ , Al Ti ’ , Na Ti ’’’ ), they form a negative charge defect that can accommodate (or trap)\nelectrons in TiO 2 , as marked by the blue dashed line in Figure 1 c, which acts as\na p-type defect and shifts the Fermi level to the valence band maximum\n(VBM) in the band gap. 40 − 45 On the other hand, the positive charge defect species, Li, Al, Na,\nand Ti occupying interstitial sites and the O vacancy (marked by green\ndashed line) such as Na i • , Li i • , Al i ••• , Ti i •••• , and V O •• , can donate electrons as an\nn-type defect and shift the Fermi level to the conduction band minimum\n(CBM). 40 − 45 This study considers the simplified eight types of defects listed\nabove. Figure 1 (a) Schematic representation of the Al,Na:TiO 2 in-situ ALD process. One supercycle consists of n Ti cycles composed of TTIP and homemade reactant\nand 1 Al cycle composed of TMA and homemade reactant. The\namount of Al doping was controlled by adjusting the n Ti :1 Al ratio. (b) Schematic diagram (left)\nand cross-sectional bright field STEM image (right) of the self-rectifying\nalkali-ion memristor device. The device with Au (80 nm)/Li (5 nm)/Al,Na:TiO 2 (18 nm)/Au (40 nm)/Li (5 nm) structure was fabricated on\nSiO 2 (300 nm)/p-Si substrate. The Li adhesion layer and\nLi Na reservoir are marked in the device schematic and bright field\nSTEM image as black and red arrows, respectively. (c) Schematic representation\nof in-situ ALD growth (left) and schematic band diagram\n(right) of Al,Na:TiO 2 . From the Li adhesion layer, Li diffuses\nover the bottom electrode and dopes into the bottom of the Al,Na:TiO 2 film. The eight defect species that can be considered in\nthis process are indicated by dotted boxes, p-type defects ( Li Ti ’’’ , Al Ti ’ , Na Ti ’’’ ), and n-type defects ( Na i • , Li i • , Al i ••• , Ti i •••• , V O •• ) indicated by blue and green\ndotted boxes, respectively. The p-type and n-type defects form energy\nlevels at the bottom and top within the TiO 2 bandgap, respectively,\nand shift the Fermi level in the CBM or VBM direction. The concentration\nof each defect species affects the direction of the Fermi level shift\ndepending on the dominant species. Figures 2 a–d\nshow the dynamic secondary ion mass spectrometry (SIMS) analysis results\nof Al 0, Al 3, Al 6, and Al 9 thin films measured without TE. First,\nas shown in the figure, the Li and Na cation concentration is high\nat the bottom interface between the TiO 2 layer and the\nAu BE (marked by yellow boxes). This results from the Li adhesion\nlayer under the BE being doped into the TiO 2 film (bottom\ninterface region) by diffusion during the 250 °C ALD process\nfor TiO 2 growth. 36 Interestingly,\nin all films, the Ti and O concentrations at the bottom interface\nare relatively higher and lower, respectively, than inside the film.\nAnd on the other hand, in the case of the top interface, the Ti:O\nratio is lower than the film inside. Therefore, while the concentration\nof Ti increases from the surface to the bottom interface, the concentration\nof O decreases. First, the high Ti concentration and low oxygen concentration\nat the bottom interface (indicated by the red arrows in the figure)\ncan be interpreted as the result of Li in the adhesion layer diffusing\ninto the bottom interface and occupying Ti sites ( Li Ti ’’’ ), leading to n-type\ndefects Ti i •••• and V O •• . 43 In addition, it was also observed that Na ions were concentrated\ntogether with Li at the bottom interface. As shown in Figure 2 e, at the initial stage of in-situ Na:TiO 2 ALD growth using NaOH aqueous\nsolution, the diffused Li results in the presence of three surface\nspecies (which can be considered), Ti-OH, Li-OH, and O-Li. In this\ncase, Na doping with desolvation of Na-H 2 O cluster is expected\nto be relatively promoted by the Li-OH and O-Li surface, which has\na larger partial charge than Ti-OH. The electronegativities of Li,\nTi, and O are 0.98, 1.54, and 3.44, respectively. The desolvation\nof Na ions is a dominant factor for Na doping (to form chemical bonds\non the surface), because Na ions in the reactant are introduced into\nthe chamber in a solvated form by water molecules. Therefore, Li-OH\nand Li-O with a large partial charge on the surface are expected to\npromote Na doping. As shown in Figures 2 a–d, both the top and bottom interfaces are\nrich in Na and Li, but the concentration distribution of Ti and oxygen\nvacancies is different at the top and bottom interfaces. The difference\nin Ti:O ratio at the top and bottom interfaces can be understood that\nthe bottom interface is due to the participation of Li and Na in the\nALD growth, while the Na- and Li-rich at the top interface is caused\nby the exposure of the TiO 2 thin film to the atmosphere\nafter the process, rather than during the ALD process. Various molecules\nare known to be adsorbed on the (n-type) surface by the excess electrons\nin TiO 2 . 46 − 50 In this study, we can expect oxygen absorption when taken out from\na chamber of an Ar environment and exposed to the atmosphere after\nthe film deposition is completed. From the excess oxygen detected\non the surface, we can consider the adsorbed molecule species such\nas superoxo (O 2 – ), peroxo (O 2 2– ), OOH, etc. , or diffusion into\nthe TiO 2 surface and the resulting formation of a negatively\ncharged surface. 46 − 50 And these excess oxygens on the surface could lead to the diffusion\nof Na and Li cations to the surface. Figure S2a shows the XPS measurement results of Li 1s, and Li was detected\non the surface of all thin film samples, in agreement with the SIMS\nanalysis results. The compositional analysis of Ti, O, and Li based\non the XPS results of the Al 0 sample showed that the at% of each\nelement was 26.70 at%, 62.03 at%, and 11.27 at%, respectively. (Na\nwas excluded from the quantitative analysis because it overlapped\nwith the Ti Auger peak) Assuming the TiO 2 composition of\nthe thin film, this resulted in an approximately 1:1 excess of O and\nLi. Consequently, alkali cation reservoirs were already formed at\nboth the top and bottom interfaces before the top electrode deposition.\nDue to the compositional gradient of Ti and O, the top interface formed\na robust and high Schottky barrier with relatively excess oxygen,\nwhile the Schottky barrier at the bottom interface has a lower height\nand smaller depletion width by the excess Ti and O vacancies. In the\ncase of Al, it was found that the amount of Al doping gradually increased\nwith the increase of Al cycles in Al 0, Al 3, Al 6, and Al 9 films\n( Figure S2b shows the results of Al 2p\nmeasurements, which are in agreement with the SIMS results), and the\namount of Al doping did not affect the distribution of Li, Na, Ti,\nand O in the film. Since Al was introduced as a single cycle, it can\nbe confirmed that the AlO layer is not formed in a laminate structure\nbut is diffused and uniformly distributed in the film. Figure 2 Dynamic SIMS depth profile\nresults of (a) Al 0, (b) Al 3, (c) Al\n6, and (d) Al 9/Au (40 nm)/Li (5 nm) samples. The yellow box in the\nfigure indicates the bottom interfacial region of the Al,Na:TiO 2 layer with the Au BE. (e) Schematic representation of three\nsimplified surface species Ti-OH, Li-OH, and O-Li considering results\nby diffused Li. (f) O 1s and (g) Ti 2p XPS results of Al 0, Al 3,\nAl 6, and Al 9 samples. The dotted line and red arrow in the figures\nindicate the binding energy of the main peak in the Al 0 sample and\nthe direction of the peak shift, respectively. (h) Schematic representation\nof the native oxygen vacancy present in TiO 2 and (i) the\npassivation effect and additional oxygen vacancies created by Al doping\n(top) and the resulting band diagram (bottom). (Defects caused by\nalkali ions are excluded.) Figures 2 f and g\nare the O 1s and Ti 2p XPS measurement results. Table 1 summarizes the atomic percent of the four\nelements (Li, Al, Ti, O) and the ratio of oxygen vacancy peak intensity\nobtained from the XPS results of each sample. (Na is excluded due\nto overlap with Na 1s and Ti Auger peaks.) First, it is noted that\nthe oxygen vacancy peak (labeled as O v ) exhibits a slight increase with the rise in aluminum (Al) doping,\na consequence of the Schottky defect. While it varies depending on\ndoping concentration and environment, it is generally reported that\nthe Al Ti ’ defect, where\nthe Al 3+ substitutes the Ti 4+ accompanied by\nan oxygen vacancy, is more dominant than the Al i •••• defect that occupies the interstitial site. 41 , 43 , 45 On the one hand, despite the increase in\nthe n-type O v defect, both the O 1s and\nTi 2p XPS peaks, as illustrated in Figures 2 f and g, undergo a red shift of approximately\n0.10 eV, indicated by red arrows. This shift is attributed to the\nAl doping effect, which, akin to p-type doping, moves the Fermi level\ntoward the conduction band minimum (CBM). Figures 2 f and i provide simplified representations\nof the defect types and band diagrams before and after Al doping.\nThe native defect O v present in undoped\nTiO 2 generates a shallow trap state in the bandgap and\ntwo electrons, as shown in Figure 2 f. In thermal equilibrium, these two electrons are\nunable to function as carriers. However, when a strong electric field\nis applied to the 18-nm-thick film, mechanisms like trap-to-trap hopping\nor the Poole–Frenkel effect can mobilize them, allowing their\nparticipation in electrical conduction. When Ti neighboring an oxygen\nvacancy is substituted by Al, it can passivate the oxygen vacancies\nas the green dotted box. While Al doping creates new oxygen vacancies\ndue to Schottky defects as the orange dotted box, it can also create\na deep trap level in the bandgap that can accommodate the electron\nas the blue dotted box. Thus, Al doping acts as a p-type dopant by\npassivating oxygen vacancies and trapping electrons. This reduces\nthe number of electrons available for transport in the film, increasing\nthe resistivity, and shifts the Fermi level, increasing the band offset.\nTherefore, the red shift observed in the XPS results is a reflection\nof the above case. Table 1 Atomic Percent of Al 0, Al 3, Al 6,\nand Al 9 Films as Derived from XPS Data, along with the Intensity\nRatios of the Oxygen Vacancy Peaks Determined through O 1s Peak Fitting Li 1s (at %) Al 2p (at %) Ti 2P (at %) O 1s (at %) O v /(O L +O v ) (peak area) Al 0 11.44 1.14 26.27 61.15 0.126 Al 3 11.12 1.35 26.34 61.19 0.140 Al 6 11.63 1.65 25.73 60.99 0.145 Al 9 10.81 1.76 26.20 61.23 0.151 Author: Figure S3a shows\nthe X-ray diffraction\n(XRD) results of Al 0, Al 3, Al 6, and Al 9 thin films, where diffraction\npeaks of TiO 2 films were not observed. In a previous study\nusing Pt (BE)/Ti (adhesion), an anatase phase was observed in the\nsame ALD process. 10 However, in this study\nusing Au (BE)/Li (adhesion) as the BE, the crystallinity was very\nweak, suggesting that the diffusion of Li hindered the crystallization\nof TiO 2 . From the Raman spectra in Figure S3b , very faint peaks at 140, 240, 340, and 432 cm –1 (indicated by red arrows) were observed in the Al\n0 sample. Assuming the rutile phase, the peak observed at 140 cm –1 is the B 1g peak, while the other three\npeaks (240, 340, and 432 cm –1 ) are observed in lithiated\nTiO 2 (Li x TiO 2 , 0.3\n< x < 0.5) (or lithium titanate) and are related\nto the vibrations of TiO 6 , LiO 6 , and LiO 4 (F 2g , F 2g , E g ), respectively. 51 , 52 The Al 0 Raman spectrum reveals that Li is doped in the TiO 2 layer and has a weak crystalline or weak ordered structure.\nThese soft peaks were not observed in Al-doped thin films, which is\ninterpreted as a result of Al doping increasing the structural disorder. 53 Figure S3c shows\nthe fast-Fourier-transform (FFT) pattern and high-resolution transmission\nelectron microscopy (HRTEM) image of the TiO 2 layer (the\ncenter of the film) of the Al 0 sample, where a rutile phase was observed.\nIn Al 3, Al 6, and Al 9 films, the same low crystallinity rutile phase\nwas observed as in Al 0, but the change in crystallinity with Al doping\nwas difficult to determine from the FFT pattern. In the case of diffraction\nspots of the (130) planes of the rutile (indicated by the yellow solid\ncircle), diffraction spots with larger d -spacing\nwere observed in the 90° rotation direction (indicated by the\nred circle). Their d -spacing was 1.45 and 1.52 Å,\nrespectively, showing an increase in d -spacing of\nabout 4.83%. (This change was also observed in the (210) plane.) This\nis interpreted as a result of Li cations occupying the interstitial\nsites of the rutile, causing lattice expansion and distortion ( a and b lattice parameter changes). Figure 3 shows the\nRS characteristics of Al 0, Al 3, Al 6, and Al 9 memristor devices\nmeasured for 10 cycles each (arrows and numbers indicate the sweep\nsequence), which shows the gradual resistance change characteristics\nwithout any forming process. First, the Al 0 device shows gradual\nresistance change characteristics with self-rectifying characteristics.\nThis is due to the asymmetric top and bottom interfaces caused by\nthe composition gradient of Ti and O, as shown in the SIMS analysis.\nIn the case of the bottom interface, since Li is supplied during the\nthin film deposition process, it forms a Li–O bond (or Na–O)\nand occupies a Ti site, as shown in Figure 2 . On the other hand, in the case of the top\ninterface, it can be expected that the alkali cations have occupied\nthe interstitial site by the electrochemical migration after thin\nfilm deposition and formed an alkali metal reservoir layer. Considering\nthis pristine state, the migration of the alkali cations in the bottom\nreservoir is relatively difficult compared to the alkali cations of\nthe top interface. Therefore, in the pristine state, the device starts\nwith a high-resistance state (HRS). Figure 3 e is a schematic illustration of the RS mechanism,\nwhere the numbers in the figure correspond to the states of the sweep\nsequence number shown in the I – V curve. Both the top and bottom interfaces participate in the RS\nevent, representing the process of changing the resistance of the\ntop and bottom interfaces by migration of alkali ions (this has been\nconfirmed by dynamic SIMS and Schottky barrier measurement in our\nprevious work). 1 , 10 , 54 As shown 1 → 2 in Figure 3 e, the alkali cation migrates to the bottom interface\nunder positive voltage, reducing the Schottky barrier height and the\ndepletion width of the bottom interface. In this case, the device\nswitches to a low resistance state (LRS), and then when a negative\nvoltage is applied (Sequence #3), the device exhibits HRS with an\nincreased Schottky barrier of the top interface (reverse biased) by\nthe reduced alkali cations, and the alkali cation moves back to the\ntop interface, inducing LRS switching (Sequence #4). 10 The top and bottom interfaces of the pristine state are\nin an equilibrium state with respect to chemical composition and charge\nvalence. As described above, the redistribution of the alkali cation\nby the programming voltage breaks the charge valence at the top and\nbottom interfaces, and a force is exerted to return to equilibrium.\nThis impairs retention in the interface-type memristor. In the device\nof this study, abundant negative charge defect species at the bottom\ninterface, as shown in Figures 1 and 2 , could make it easy to hold\ncations Li + and Na + . (Role of alkali cation\nreservoir) And the middle of TiO 2 layer acts as a diffusion\nbarrier, Al dopant in TiO 2 acts to increase the diffusion\nbarrier, which can improve the retention characteristics. Figures 3 a–d show\nthat the device loses its self-rectifying characteristic (asymmetry\nof the DC curve) with an increase of Al dopant. This behavior is caused\nby the formation of the n-type defect states V O •• accompanied by the p-type defect Ai i ’ , as shown in the band diagram on the right\nside of each DC curve, which facilitates the electron injection from\nthe top electrode to the conduction band of TiO 2 . On the\nother hand, the current level at a positive voltage decreased with\nthe increase of Al dopant, which means an increase of interface resistance\nat the bottom interface. As confirmed by the XPS data, the Fermi level\nshifted (redshift) toward the VBM with the increase of Al doping,\nwhich acts as a factor in increasing the Schottky barrier height at\nthe bottom interface. In addition, the p-type defect Ai i ’ serves to trap the n-type carriers\nand reduce the carrier concentration in TiO 2 . We cannot\nexclude the structural factor that the increase in structural disorder\ncaused by Al doping (as shown in Figure S3 ) could also contribute to the increase in resistance. As a result,\nAl doping increases the resistance of the memristor and weakens its\nself-rectifying characteristics. The selectivity values using 1/3\nbiasing scheme “ I@V op /I@–1/3 V op ” for each device Al 0, 3, 6, 9, extracted from self-rectifying\nasymmetric curves, were 2 × 10 4 , 1 × 10 4 , 2 × 10 3 , and 3 × 10 1 , respectively.\nHere, the measured current value of the noise level at –1/3V op was used for the I@–1/3V op , and it is expected that a higher\nselectivity will be observed in actual large-scale arrays. Low operating\ncurrent is advantageous in terms of power consumption, but on the\nother hand, it also reduces the sensing margin, which is a trade-off\ndilemma for memristors that need to implement multi-bit. This is an\nimportant issue for neuromorphic hardware, which emphasizes energy-efficient\ncomputing. Consequently, the capacity to regulate the operating current\nby means of aluminum (Al) doping allows for the optimization of the\naforementioned trade-off factors. Figure 3 f presents the results of a three-dimensional\ngraph depicting the depth profiles of O, Ti, Li, and Na from the surface\nto the bottom electrode interface for the Al 0 sample, based on SIMS\ndata. As illustrated, O and Ti exhibit compositional gradients, which\ncontribute to the self-rectifying characteristics. The distribution\nof alkali ions shows higher concentrations at the top and bottom interface\nregions of the thin film compared to the middle, indicating stability\nat these locations in the pristine state and suggesting the middle\nregion acts as a barrier. The anticipated band diagram under this\ncondition is represented on the left in Figure 3 g, corresponding to the initial state or\nHRS@positive voltage. Upon application of a positive programming voltage,\nNa and Li dopants migrate toward the bottom interface, as depicted\nin the band diagram on the right in Figure 3 g, resulting in a reduction in both the height\nof the Schottky barrier and the depletion width at the bottom interface\nwhich induce LRS@positive voltage. Figure 3 Resistive switching properties (10 cycles\nof DC sweep curves) of\n(a) Al 0, (b) Al 3, (c) Al 6, and (d) Al 9 devices fabricated by 5\num × 5 um crosspoint structure and schematic band diagram (right)\nof each devices. Above and below the band schematic are labeled n-type\ndefects (red) and p-type defects (blue) species, respectively. The\nn-type (red) and p-type (blue) defect species are labeled above and\nbelow the band diagram. (e) A schematic representation of the RS mechanism\nof the self-rectifying alkai-ion memristor. Here, both the top and\nbottom interfaces change their interface resistance by the redistribution\nof alkali ions. (f) Three-dimensional compositional distribution graph\nfor elements O, Ti, Li, and Na from the surface to the bottom interface,\nconstructed based on SIMS data of the Al 0 sample. (g) Expected band\ndiagrams for each state: HRS@positive voltage (left); LRS@positive\nvoltage (right). Figure 4 shows the\nretention characteristics of Al 0, Al 3, Al 6, and Al 9 memristor\ndevices, which were measured with a read voltage of +1 V at 125 °C\nand read interval of 2 s for 20 h under relatively harsh conditions\n(to accelerate degradation) after entirely switching to HRS and LRS\nusing DC sweep. As shown in the figure, an improvement in retention\ncharacteristics was clearly observed with increasing Al doping level,\nbut at the same time, the on/off ratio with operating current decreased.\nIn general, it is common for the resistance of LRS to increase gradually\nwith time, but in the case of the LRS curve of Al 0, it was observed\nthat the resistance decreased from 100 s to 2,000 s and then increased\ncontinuously. It is expected that highly mobile alkali cations, which\nare responsible for the resistance change in this study, exhibit unstable\nbehavior in the 125 °C environment. Interestingly, the retention\nproperties became increasingly stable as the amount of Al dopant increased.\nThis is understood to be a consequence of Al doping hindering the\ndiffusion of redistributed alkali ions back to their equilibrium state,\nas by the programming voltage. This phenomenon occurs in the same\ncontext as the observed decrease in the on/off ratio. Several factors\ninfluencing the diffusion properties of Li in solids are notable,\nincluding defect type, defect concentration, migration barrier, and\nstructural symmetry. As shown in the XPS results, Al doping increases\nthe oxygen vacancy and simultaneously passivates the oxygen vacancy,\nwhich would act as a factor to facilitate the diffusion of alkali\nions from the defect perspective (lowering the migration barrier).\nOn the contrary, an increase in structural disorder observed in Figure S3b acts as a factor hindering the migration\nof alkali ions. Therefore, based on the analytical data obtained in\nthis study, it is reasonable to understand that the stability of retention\nproperties with Al doping observed in Figure 4 is due to an increase in structural disorder.\nFor more details, We measured Warburg impedance to investigate the\nLi (or Na) diffusion coefficient in Al 0, Al 3, Al 6, and Al 9 devices,\nand the measured Warburg plots (ω –1/2 vs Z Im ) are given in Figure 4 e. Due to the small electrode\narea (5 μm × 5 μm) and thin film thickness (18 nm),\ndiffusional impedance was observed, 55 and\nthe Warburg coefficient ( A W ) was extracted in the high-frequency region. Utilizing the\nfitted slopes in conjunction with the Warburg coefficient–diffusivity\nrelationship (see Eq. S1 in the Supporting\nInformation for details), we calculated the Li (or Na) diffusion coefficients\nfor the Al 0, Al 3, Al 6, and Al 9 devices, which are presented in Table 2 . The derived values\nfell within the acceptable range (from 10 –6 to 10 –15 cm 2 /s in Rutile TiO 2 ) found\nin the literature. 56 − 58 (Note that this analysis focuses on the relative\nvalues between devices rather than the absolute values.) As shown\nin Table 2 , the measured\ndiffusion coefficient decreased with increased aluminum dopant. In\ndevices from Al 0 (undoped) to Al 3 (3 cycles doping), there is a\nsignificant decrease in the diffusion coefficient, followed by relatively\nsmaller reductions for Al 6 and Al 9. This trend is analogous to the\nchanges in structural disorder observed in Raman spectra. In TiO 2 , the migration of Li ions is known to be strongly dependent\non the density (Ti–O distance) and the crystal structure (symmetry,\nmigration path, and the size of the migration channel). Therefore,\nthe TiO 2 rutile phase is recognized for having highly anisotropic\nLi ion migration paths. In the [110] directions, the zigzag path results\nin low mobility (∼10 –15 cm 2 /s),\nwhile along the c -axis, the straight migration path\nleads to significantly higher Li ion mobility (∼10 –6 cm 2 /s). 56 Therefore, we infer\nthat the structural disorder by Al doping reduces the diffusivity\nof Li (or Na) ions and contributes to improved retention characteristics.\nOn the other hand, such a reduction in diffusivity adversely affects\nthe weight update speed. To compare the weight update speeds, we measured\nthe conductance change of Al 0, Al 3, Al 6, and Al 9 devices according\nto pulse width, as presented in Figure 4 f. The amplitude was fixed at +5 V, and five different\npulse widths (5 μs, 10 μs, 20 μs, 30 μs, 50\nμs) were used for weight updates. The pulse width-dependent\naverage change in conductance was obtained by utilizing the slope\nof the conductance vs pulse number plot (potentiation). As shown in Figure 4 f, although the average\nconductance change rate of each device increases with the pulse width,\nit decreases significantly with the increase in Al doping amount.\nTherefore, while a decrease in diffusivity is effective in improving\nretention characteristics, it also reduces the weight update speed.\nHowever, this feature could be advantageous for implementing multi-bit\noperations if appropriately used. In this study, we cannot completely\nrule out the influence of oxygen ion migration in the observed resistance\nchange characteristics. However, considering the measured diffusion\ncoefficient values, these are approximately 10–1000 times higher\nthan the range reported in the literature for oxygen ion mobility\nin Rutile TiO 2 at room temperature. 59 − 63 Consequently, under the applied RS voltage, the redistribution\nof defects within the thin film is likely dominated by the migration\nof Li (or Na) rather than oxygen migration. As previously mentioned,\nin filament-type devices, current flows through a very narrow conducting\npath, leading to Joule heating, which thermally assists the migration\nof oxygen (both current-driven and electric field-driven), making\nit relatively easier. 64 , 65 (Ion migration strongly depends\non temperature.) However, thermal assistance is significantly less\nfor the interface-type devices in our study, considering the current\npassing through the entire interface during switching (below 10 μA).\nTherefore, it can be anticipated that the migration of Li, which has\nhigher mobility, becomes the dominant factor in the resistive switching\ncharacteristics rather than the less mobile oxygen ions. Figure 4 Retention properties\nof (a) Al 0, (b) Al 3, (c) Al 6, and (d) Al\n9 devices, measured by +1 V every 2 s at 125 °C, after fully\nswitching to LRS and HRS using ±3 V DC sweep. (e) Warburg plots\n(ω –1/2 vs Z im ) measured at room temperature for Al 0, Al 3, Al 6, and Al\n9. The A W values marked\non each plot are the Warburg coefficients obtained from the slopes.\n(f) Pulse width-dependent average change in conductance for Al 0,\nAl 3, Al 6, and Al 9 at 5 μs, 10 μs, 20 μs, 30 μs,\nand 50 μs. The average change in conductance was measured by\nemploying pulses of varying widths (5 μs, 10 μs, 20 μs,\n30 μs, and 50 μs) over 100 weight update cycles. For each\ncycle, conductance was measured using a read voltage of +1 V for potentiation,\nand the slope of 100 cycles of potentiation was used to determine\nthe average conductance change. Table 2 Diffusion Coefficients at Room Temperature\nDerived from Warburg Impedance Measurements for Al 0, Al 3, Al 6,\nand Al 9 Devices Al 0 Al 3 Al\n6 Al 9 D Li (or Na) (cm 2 /s) 1.68 × 10 –12 8.96 × 10 –13 7.65 × 10 –13 6.96 × 10 –13 Figures 5 a–d\nshow the results of weight update characteristics using 256 + (−)\n5 V, 5 us pulses for long-term potentiation (LTP) (long-term depression,\nLTD) curves using a read voltage of +1 V for Al 0, Al 3, Al 6, and\nAl 9 memristor devices. The left of each figure represents the LTP/LTD\ncurves repeated for 6 cycles in a single device (cycle-to-cycle variation),\nwhile the right shows the results of a polar plot (pulse number vs\nconductance) for the LTP/LTD curve obtained from one cycle in 20 different\ndevices (device-to-device variation). As the Al dopant increases,\nthe resistance change rate decreases for a constant programming pulse\nsize, i.e., the switching speed decreases. The resistance change rate\nmeans the amount of ions that are redistributed, which depends on\nthe magnitude of the write voltage, the amount of current, and the\ndiffusivity of the ions that cause the resistance change. Therefore,\nAl dopant decreases the weight update rate because it decreases the\namount of current passing through the memristor and the diffusivity\nof Li + and Na + cations. And it also affects\nthe linearity and symmetry of the LTP/LTD curve, as shown in the figure,\nwhich means that Al dopant can control the linearity and symmetry\nof the LTP/LTD curve. In the polar plot, the blue and red curves represent\nLTP and LTD, respectively. (The device-to-device variation illustrated\nby violin plots is depicted in Figure S4 .) The symmetry between each semicircle indicates the symmetry of\nthe weight update, and the size distribution of each semicircle demonstrates\nthe device-to-device variation. As the doping amount of Aluminum increases,\nthe available conductance range decreases, but it is also possible\nto reduce the cycle-to-cycle and device-to-device variations. Considering\nthe stochastic weight update characteristics due to ion migration,\nit can be anticipated that a higher switching ratio may also lead\nto a large fluctuation in weight update, which degrades reliability.\nConsequently, Al dopant loses in switching speed but gains in linearity,\nsymmetry, retention, and reliability. Therefore, Al doping can be\nused as a control factor in the design of memristor devices. Figure 5 Six cycles\nof LTP/LTD curves obtained from 1 device (left) and\npolar plots (pulse number vs conductance) for the LTP/LTD curves obtained\nfrom 20 different devices (right) of (a) Al 0, (b) Al 3, (c) Al 6,\nand (d) Al 9 devices. 256 weight updates were performed using ±5\nV amplitude and 5 μs width of pulse train and read using +1\nV at each step. Simulation results of 28 × 28 (e) digit and (f)\nfashion MNIST classification of Al 0, Al 3, Al 6, and Al 9 devices\nderived based on the long-term potentiation and depression curves\nobtained in (a)–(d). Figures 5 e and f\nshow the results of 28 × 28 (pixel) digit MNIST and 28 ×\n28 (pixel) fashion MNIST classification simulations based on the linearity\nand symmetry of Al 0, Al 3, Al 6, and Al 9 memristor devices. In 28\n× 28 pixel handwritten digit MNIST and fashion MNIST image recognition,\n28 × 28 pre-neurons corresponding to the 28 × 28 pixel images\nact as the input layer of our multilayer perceptron. The fully connected\n300 hidden neurons in a single hidden layer and 10 output neurons\ncorrespond to 10 handwritten digits and clothing data sets. Each neural\nnetwork was trained for up to 40 epochs, with each epoch training\nand testing 10,000 images at random. During the forward propagation\nstage, the activation of each neuron is propagated to the next neurons\nvia synaptic weights and transferred through a nonlinear function.\nThe sigmoid function serves as the mathematical activation unit in\na neural network. Python was used to implement the pattern recognition\nprocessing described above. Recognition rates for 10 handwritten digits\nand 10 types of clothing images are shown for training based on 60,000\ntraining data sets and 10,000 test data sets, respectively. The nonlinearity\nand asymmetry of the LTP/LTD and the accuracy (%) obtained by each\ndevice for Digit and Fashion MNIST are summarized in Table S1 . As is well-known, the linearity and symmetry of\nthe weight update are the most important factors determining the accuracy\nof neuromorphic hardware. 13 As a result,\nthe Al 6 device with the highest symmetry and linearity has the highest\nrecognition rate. Figures 6 a and b\nshow an optical microscope image of a 32 × 32 crossbar array\ndevice fabricated based on an Al 6 device and a field emission scanning\nelectron microscope (FE-SEM) image of the three cells enlarged, respectively.\nAs shown in the SEM image, the area where the TEs and BEs are crossed\nis 10 μm × 10 μm, which is four times larger than\nthe unit device in Figure 4 (5 μm × 5 μm). The thickness was reduced\nby 30% compared to the unit device through thickness optimization,\nand the DC sweep range was limited to ±2 V to avoid hard breakdown.\nThe measured DC switching curves of 100 devices randomly selected\nfrom 1,024 devices in a 32 × 32 array are shown in Figure 6 c. First, the figure shows\nthat the stable RS characteristics were obtained without interference\nby sneak current from the self-rectifying characteristic (high selectivity).\nThe base current was about 1.5 orders of magnitude higher than that\nof the unit device, which was confirmed by leakage within the switch\nmatrix device. Figure 6 c inset shows the I – V curves\nof the three devices that failed by hard breakdown out of 100 devices.\nTherefore, 97% of the 100 devices worked well without failure, and\nto check the device-to-device variation, the cumulative probability\nof each current level of HRS and LRS under the +1 V read condition\nis shown for the 97 working devices. The relative standard deviation\nfor each resistance state (HRS and LRS) of the 97 devices in the crossbar\narrays (not unit devices) was measured to be 76.7% and 81.7%, respectively.\nThis variation is expected to be due to the uniformity issue of the\nTEs and BEs deposited by the thermal evaporator (compared to the ALD\nprocess of the memristor layer). Figure 6 e shows the endurance characteristics measured\nunder the conditions of programming pulse width 100 μs, ±5\nV, DC read +1 V. A longer pulse width was used compared to the LTP/LTD\nmeasurement condition in Figure 5 to better distinguish between the two resistance states\nof R1 and R2. As a result, degradation was not observed up to 550,000\ncycles. The cumulative probability obtained from the measured endurance\ndata is shown in Figure 6 f to confirm the cycle-to-cycle variation. The relative standard\ndeviations for the R1 and R2 resistance states were measured to be\n8% and 4%, respectively, indicating a very low cycle-to-cycle variation.\nIn order to compare the performance of the self-rectifying alkali\nion memristor devices proposed in this study, we listed the device\nstructures with device performance indicators such as endurance, retention,\nswitching speed, and test scale characteristics from previous reports\nof oxygen ion-based interface-type memristor devices in Table S2 . The comparison table shows that the\nself-rectifying alkali-ion memristor of the Au/Li/Al,Na:TiO 2 /Au/Li device proposed in this study exhibits relatively improved\nRS characteristics. Figure 6 (a) Optical microscope image of a 32 × 32 crossbar\narray device\nfabricated using Al 6. (b) SEM image of three magnified device cells\nin a 32 × 32 crossbar array device. (c) ±2 V DC sweep curves\nmeasured on randomly 100 selected devices. The inset is the DC curves\nof the three devices that experienced hard breakdown among the 100\ndevices. (d) A cumulative probability plot of HRS and LRS extracted\nat read voltage +1 V from 97 devices, excluding the 3 failed devices.\n(e) Endurance data measured 5.5 × 10 5 times using\nswitching condition pulse amplitude ±5 V, width 100 μs,\nand read condition +1 V (DC). (f) Cumulative probability plot extracted\nfrom endurance data for cycle-to-cycle variation."
} | 12,464 |
36223455 | PMC9555773 | pmc | 276 | {
"abstract": "Spider silks are among the toughest known materials and thus provide models for renewable, biodegradable, and sustainable biopolymers. However, the entirety of their diversity still remains elusive, and silks that exceed the performance limits of industrial fibers are constantly being found. We obtained transcriptome assemblies from 1098 species of spiders to comprehensively catalog silk gene sequences and measured the mechanical, thermal, structural, and hydration properties of the dragline silks of 446 species. The combination of these silk protein genotype-phenotype data revealed essential contributions of multicomponent structures with major ampullate spidroin 1 to 3 paralogs in high-performance dragline silks and numerous amino acid motifs contributing to each of the measured properties. We hope that our global sampling, comprehensive testing, integrated analysis, and open data will provide a solid starting point for future biomaterial designs.",
"introduction": "INTRODUCTION Modern genomics combined with advanced bioinformatics methodologies allow us to understand much more about complex living systems than was ever previously possible. In the realm of human biology, for instance, recent developments have given us the ability to pinpoint the genes influencing diseases such as cancers. One area where these novel technologies can be anticipated to exert a huge impact but have thus far remained underused is the study of structural biomaterials. Spider silk is a prime example of an extended phenotype, whose extraordinary mechanical properties are governed by the underlying composition and structure of protein building blocks called spidroins. All spiders use silk for various critical purposes, including foraging, locomotion, nesting, mating, egg protection, and communication ( 1 ). Different types of threads are used for diverse purposes, each produced in specific glands in the abdomen ( 2 ). For example, orb-weaving spiders use up to seven different types of silks, named after the gland that produces these threads. Major ampullate silk is the toughest silk used as draglines and as frames of orb webs, minor ampullate silk is used as scaffold during orb web weaving, piriform silk adheres the frame of the orb web to wood or other substrates, and capture thread of the orb web is composed of flagelliform silk backbone and aggregate glue. Aciniform silk is used for prey wrapping and sometimes for decorations of the web, and tubiform (or cylindrical) silk is used to make an egg sac. While spiders are successful predators and are often associated with orb webs, orb-weaving spiders of superfamily Araneoidea only comprise about 25% of spider species. A more ancestral clade of spiders such as those belonging to the infraorder Mygalomorphae is comprised mostly of ground-wandering spiders that produce sheet and maze webs for prey capture. Wandering hunters and abandoned silk capture webs make up a more modern clade of spiders in the retrolateral tibial apophysis (RTA) clade; this group comprises as much as 50% of all spider species ( 3 ). Therefore, spiders have diversified, selected, and specialized various uses of silk adapting to their ecological needs. Such extraordinary plasticity and university of silk and silk proteins is an ideal target to model the link between sequence and its physical property to fully understand the underlying design principles to apply the wide range of physical properties as biomaterials. Spider silks are renowned for their diverse and impressive mechanical properties, frequently displaying a combination of high tensile strength, extensibility, and exceptional toughness that is unmatched industrially. Hence, the processing-property space that these silks occupy makes them a unique source of inspiration for protein biopolymer materials with low embodied energy and high performance ( 4 – 6 ). However, this property space has yet to be fully explored, defined, and exploited. Silk fiber diversity scales rapidly, as spiders produce multiple types of silk, each of which are composed of specific proteins known as spidroins, whose mostly monophyletic origins ( 7 ) endow them with specific mechanical properties ( 2 , 8 ). One type of spider silk protein, major ampullate spidroin (MaSp; which is often included in dragline threads), has received substantial academic and industrial attention, as this silk typically shows strength and toughness comparable to those of synthetic high-performance fibers, with an approximately 1-GPa breaking strength, a 30% breaking strain, and a toughness of 130 to 200 MJ/m 3 ( 9 – 11 ). However, there are lesser-known taxa and species of spiders, suggesting that the limits of silk properties are yet to be defined ( 12 ). On the other hand, a unique property known as supercontraction, where the dragline silk shrinks in length by up to 60% when wetted, is often considered undesirable industrially, and expectations are high for protein engineering methods to reduce such property by modifying the primary sequence. Hence, a comprehensive, coordinated global effort combining taxonomy, genomics, and materiomics is required to first understand and then unlock the true potential of these materials ( 13 ). The diversity of spidroin sequences has been explored for decades. Pioneering work by Gatesy et al . ( 14 ) identified and analyzed spidroin sequences from several spider lineages, including basal spider groups, thus enabling a glimpse into the complex evolution of spidroin sequences. Subsequently, there have been a large number of studies that have explored the subject of spidroin sequence diversity and evolution, including focused studies on various spidroin paralogs ( 15 – 29 ), and those from more phylogenetic perspective ( 7 , 30 – 34 ), predominantly based on the conserved terminal sequences. On the other hand, the mechanical properties of silk fibers are governed largely through the repetitive regions that dominate the silk protein sequence, and the study on the diversity of spidroin repetitive regions, particularly in the more evolutionarily divergent taxa, has been limited to date. Thus, there is still an unmet need to map out the evolutionary design space of silk sequences and mechanical performance. This is especially relevant in light of the recent major breakthroughs in the field of spider phylogenomics ( 3 , 35 , 36 ). Undoubtedly, part of the reason for the scarcity of data on spidroin repetitive sequences has been the serious technical challenges faced when attempting to sequence highly repetitive low-complexity sequences such as found in silk proteins (compounded by the presence of multiple paralogs in the case of spider silk proteins). Recent advances in sequencing methods ( 37 ), however, have made such initiatives possible, as we present in this work. To address this need, we sequenced the silk genes of more than 1000 spider species encompassing the entire order Araneae using de novo transcriptome sequencing and assembly, alongside the comprehensive measurement of the material properties of their dragline silk fibers.",
"discussion": "RESULTS AND DISCUSSION Expanding the repertoire of silk genes The transcriptomes of 1774 individual spiders were sequenced, which included 1098 species belonging to 441 genera and 76 families, globally sampled from four continents. Redundant sampling was performed for certain species to observe locality or sex differences in spidroin expression ( 22 , 38 ) and sequence variations within species. After the curation of the assembled transcripts, a total of 11,155 putative spidroin genes were identified ( Fig. 1 and data file S1). All of the data are openly accessible from the Spider Silkome Database ( https://spider-silkome.org ). Fig. 1. Overview of the taxonomic distribution of spidroins and physical properties of dragline silks. Left: The phylogenetic tree of spider families constructed from the transcriptome data obtained from 1000 spiders in this work. The Araneoidea superfamily and the RTA clade are highlighted in red and blue, respectively. Family names in red represent those without previous report of spidroins in the NCBI Protein Database. Family names marked with orange circle represent those without previous transcriptome data. Total number of species sequenced in this work for each family, as well as species-level decomposition of unreported spidroin and transcriptome, are shown to the right of the family names. As this table shows, the vast majority of species reported in this work is previously unreported for their spidroin sequences or transcriptome. Middle: Heatmap of the conservation level of spidroin types within the spider families. For example, MaSp3 of family Araneidae has a value of around 0.5, as can be seen from the color code shown in the bottom left corner, which indicates that around 50% of the 191 species studied in this work contains MaSp3. The orb-weaving spiders in the superfamily Araneoidea (highlighted in pink) have greater diversity of spidroin types, and the RTA clade (highlighted in light blue) lost the capture web silks Flag and AgSp. MaSp sequence subtypes are not well differentiated in the RTA clade, where MiSp, ampullate spidroin (AmSp), and MaSp are more conserved than MaSp1 and MaSp2. Right: Distribution of physical properties among the spider families. Mirrored with the diversity of spidroins, orb-weaving Araneoidea spiders tend to have higher performance than other clades. The present study greatly expands the number and diversity of known spidroin sequences; we report sequences from 58 spider families not previously represented in public database, including members of basal taxa (Mesothelae, Mygalomorphae, Synspermiata, and allied groups), Araneoidea (which comprises the ecribellate orb weavers), and previously poorly sampled but extremely diverse groups such as the RTA clade and other taxa. At the time of writing this manuscript, spidroin sequences in the National Center for Biotechnology Information (NCBI) Protein database come from only 52 species in 18 families, and 23% of these sequence is derived from the single genus Trichonephila , and majority (73%) of the registered sequences are of major/minor ampullate spidroins (MiSps). In Fig. 1 , family names colored in red indicate those with species where spidroin sequence is previously unreported, and family names with orange circles in front indicate those without previously reported transcriptome data. As the number of species indicates to the right of the family names, the vast majority of species reported in this work is previously unreported for spidroins and transcriptome data. Within the “haplogyne” spider groups (Synspermiata and allied groups), we obtained sequences from nine previously unexplored families, including the first aciniform spidroin (AcSp), pyriform spidroin (PySp), and cribellar spidroin (CrSp) from these taxa. These proteins are consistent with more specialized silk types tuned to distinct biological functions in contrast to the undifferentiated spidroins identified from more ancestral Mesothelae and Mygalomorphae. The most extensive sampling was conducted within Araneoidea, including family Araneidae, where we identified previously-unidentified spidroin sequences from the major subdivisions within the family ( 39 ), and likewise from underrepresented web-building families such as Tetragnathidae and Linyphiidae. Our results showed that the greatest diversity of spidroin types existed within the araneoid taxa, and spidroins associated with the capture spiral and aggregate glue of orb webs [flagelliform spidroin (Flag) and aggregate spidroin (AgSp)] were conserved only within the superfamily. Enrichment of the diversity of paralogs of the MaSp dragline gene was also observed in the group, and clear distinctions were possible among the different ampullate sequences (MaSp and MiSp) in terms of both terminal domain and repetitive sequences [for instance, MaSp2 is characterized by the presence of glutamine (Q)–containing dipeptide motifs in diglutamine (QQ)/proline-glutamine (PQ)/serine-glutamine (SQ)] ( 40 ). This is in contrast to Synspermiata and RTA clade sequences, where it is often difficult to distinguish between MaSp and MiSp types. The existence of a third type of MaSp (MaSp3), including the nephilid variant MaSp3B, appears to be specific to Araneidae (see below) ( 41 , 42 ). The RTA clade accounts for approximately half of all spider biodiversity, yet silk sequences from these mostly non–web-building groups have thus far received little attention. Our sampling identified a wide range of spidroin types from the RTA clade. To illustrate, we have identified the first spidroin sequences originating from jumping spiders (Salticidae), which has the highest species diversity among all spider families, with multiple representatives of MaSp, MiSp, AcSp, and cyllindrical spidroin (CySp), as well as unclassified spidroins from 63 different genera. We also extensively sampled spider groups situated between the araneoid and RTA clades [the so-called Uloboridae, Deinopidae, Oecobiidae, and Hersiliidae (UDOH) grade] and obtained the first reported spidroin sequences for Nicodamidae, Oecobiidae, and Hersilidae. Insights from sequence analysis: Some highlights The sequencing and annotation of the huge number and high diversity of spidroin genes from diverse spider taxa enable a deeper look into the more poorly resolved spidroin classes than previously possible. Here, we provide some examples of analyses made possible by access to such an extensive spidroin sequence database. \nCribellar spidroins: Highly conserved through evolution\n From analysis of data from the most basal spider group (suborder Mesothelae: family Liphistiidae), we identified several new spidroin sequences that include the N-terminal domain region. We found that these sequences bear a close similarity with cribellar spidroins (CrSps), recently identified as a main constituent of the nonsticky capture threads of cribellate spiders ( Fig. 2A ). In addition, on the basis of analysis of sequences from the C-terminal side, we identified CrSp sequences from eight new families that encompass a wide phylogenetic spread ( Fig. 1 ). Notably, we obtained the CrSp sequences from Hickmania troglodytes (Austrochilidae), a basal araneomorph species, in addition to representatives from Eresidae, Deinopoidae, and Oecobiidae, and a number of families from the diverse RTA clade. Analysis of the core repetitive regions of CrSp sequences showed a high degree of conservation of the amino acid composition even among widely separated groups ( Fig. 2B ). The most notable feature of these repetitive sequences is the high abundance of charged residues (around 20%) and particularly of negatively charged glutamate (E) residues that occur as clusters interspersed throughout the sequence along with a relatively high proportion of hydrophobic amino acids leucine, isoleucine, valine, and phenylalanine (L, I, V, and F, respectively; collectively around 25%), a combination unique to CrSp sequences and not seen in other spidroin types. Fig. 2. Some insights from analysis of large spidroin dataset. ( A ) Spidroin N-terminal domains obtained from basal Mesothelae bear close resemblance to CrSp sequences. H.k. , Heptathela kimurai (Liphistiidae); H.y. , Heptathela yanbaruensis (Liphistiidae); R.n. , Ryuthela nishihirai (Liphistiidae); S. sp., Stegodyphus sp. (Eresidae); O.s. , Octonoba sybotides (Uloboridae). ( B and C ) Analysis of residue composition in spidroin repetitive regions, with residue types colored according to the legend. (B) Conservation of amino acid abundance in CrSp repetitive sequences across spider taxa. H.t. , H. troglodytes ; D. sp., Deinopis sp.; M.o. , Miagrammopes orientalis ; N.a. , Nurscia albofasciata ; C.h. , Callobius hokkaido . (C) Conservation of amino acid abundance in Flag repetitive sequence among araneoid species. E.a. , E. affinis ; N.r. , Nesticodes rufipes ; C.b. , Coleosoma blandum ; D.p. , Doenitzius peniculus ; L.m. , Lepthyphantes minutus ; U.o. , Ummelatia osakaensis ; W.c. , Weintrauboa contortripes ; Z.h. , Zygiella hiramatsui ; N.l. , Nephilingis livida ; C.d. , Caerostris darwini ; C.y. , Cyrtarachne yunoharuensis ; G.k. , Gasteracantha kuhli ; T.e. , Tetragnatha extensa ; L.s. , Leucauge subgemmea ; M. sp., Mesida sp. \nFlag: One framework, diverse compositions\n Flagelliform silk refers to the stretchable silk fibers produced by araneoid spiders (superfamily Araneoidea) and known particularly as making up the prey capture spirals of orb-weaver spiders. The sequence of the constituent Flag had previously been reported from only two families (Araneidae and Theridiidae), with the core repetitive sequences only available from Araneidae. Here, we have considerably expanded the availability of Flag sequences by including previously unrepresented core repetitive and terminal sequences from the web-building families Theridiidae, Linyphiidae, Pimoidae, and Tetragnathidae ( Fig. 1 ). Figure 2C shows the amino acid composition of Flag repetitive regions from a number of species from different families, wherein a diversity in the abundance of amino acid residues is clearly apparent. The most divergent repetitive sequences were found in Theridiidae, at the base of the araneoid clade, which also showed a larger number of residues represented compared to the more derived families. Some Flag repeat sequences from Theridiidae showed a marked resemblance to CrSp in terms of amino acid composition [as exemplified by Episinus affinis in Fig. 2C ; compare with Fig. 2B ]; this might reflect the close evolutionary link between Flag and CrSp, as previous studies have suggested ( 21 , 26 ). The Flag repetitive regions from other theridiid species tend to have a more reduced set of residues, with an abundance of proline and glycine residues. The species-rich Linyphiidae, predominantly sheet web builders, also exhibited somewhat divergent Flag sequences that feature short repeating motifs enriched in glycine (G), proline (P), asparagine (N), and serine (S). In contrast, Flag repeat sequences from the canonical orb-weaving families Araneidae and Tetragnathidae showed the most compositionally simplified Flag sequences, converging on a design that features a hyperabundance of glycine (G) residues (sometimes exceeding 50%) as well as proline (P) and/or serine (S) residues. It might be hypothesized that different araneoid spider groups have adapted the Flag repetitive sequences to fulfill different prey capture strategies; for instance, spiders that build orb webs designed to catch insects in flight (e.g., Araneidae and Tetragnathidae), where fiber extensibility is most important, correlate with the highest proportion of glycine in the repetitive regions. Spider silkome: An integrated database of sequences and material properties Along with the spidroin sequence data, dragline silk fibers were collected from selected spider species, which were then subjected to a comprehensive array of analyses to obtain the individual profiles across 12 index parameters, including mechanical performance (toughness, Young’s modulus, tensile strength, and strain at break), morphological and structural properties [fiber diameter, birefringence, and degree of crystallinity based on wide-angle x-ray scattering (WAXS) analysis], thermal degradation profiles (onset temperature for 1, 5, and 10% weight loss), and hydration properties (fiber water content and degree of maximum supercontraction), for the reeled dragline silk of 446 spider species ( Fig. 3 and fig. S1). Spiders belonging to Araneoidea show particularly diverse uses of threads ( 43 ), and the majority of the dragline samples included in this project was obtained from this superfamily, because the relatively large body size and copious fiber production of these species facilitate extended fiber collection. Fig. 3. Overview of the physical properties of 446 spider silk samples. ( A ) Pearson correlation heatmap of the physical properties of dragline silk fibers measured in this work. Toughness is not only correlated with tensile strength and strain at break but also correlated with Young’s modulus. Supercontraction is correlated with strain at break. ( B ) Scatter plot of toughness versus strain at break (with spot size proportional to tensile strength). The collected samples represent an almost continuous spectrum of toughness from <0.01 to >0.40 GJ/m 3 . Spots are colored according to broad phylogenetic grouping: Araneoidea (red) includes the orb-weaving spiders and tends to show a relatively high toughness distribution relative to wandering species (such as the RTA clade, indicated in light blue). (C) Screenshots of the Spider Silkome Database ( https://spider-silkome.org ), a fully searchable, public repository of all spidroin sequences and material property data generated from the 1000 spider silkome project (the main page and individual profile data for Trichonephila clavata are shown). Together, these data represent the largest collection obtained to date linking genotype to phenotype for a particular type of protein biopolymer (Spider Silkome Database; Fig. 3C ), a fully searchable platform with integrated Basic Local Alignment Search Tool (BLAST) search capability. All sequence data are also available from DNA Data Bank of Japan (data files S1 and S2). Study on the sequence to property linkage of spider silk has been a challenge, since the source of variability is threefold: interspecific, intraspecific, and intraindividual ( 44 ). Varying protocols for silking and mechanical property measurement also complicate meta-analysis, for which the silking strain rate and humidity is known to have significant effects ( 45 ). Our data are entirely obtained under a single standardized protocol and realize comprehensive comparisons. We therefore first observed the distribution of mechanical properties by families and genera. The mechanical property data obtained in this project represent an almost continuous spectrum of toughness reaching up to 0.45 GJ/m 3 , a strain at break up to 60%, and a tensile strength up to 3 GPa ( Fig. 3B and figs. S1 and S2); thus, this dataset seems promising for ascertaining relationships between the amino acid sequences of silk proteins and the physical properties of draglines across the spider phylogeny [see also Craig et al . ( 46 )]. Toughness is highly correlated with the tensile strength and strain at break, as expected from its definition. Notably, the correlation between tensile strength and strain at break is low, indicating that the strength and elasticity of silk are independent factors ( Fig. 3A and table S1). Birefringence reflects the degree of molecular orientation of silk protein chains and is a good predictor of tensile strength; crystallinity is a similar predictor for strain at break. Silk diameter is correlated with strain at break and supercontraction, but the latter probably represents a pseudo-correlation with Sparassidae and Araneidae silks, which tend to exhibit large diameters and high supercontraction. Overall, web-weaving spiders, or those belonging to the superfamily Araneoidea, tend to express superior mechanical, physical, structural, thermal, and water-based properties relative to basal spider groups ( Fig. 3B and fig. S1). Diversity in the mechanical properties was also the largest in the family Araneidae, mirrored by the high variability in the repetitive region sequences of MaSp-type spidroins (fig. S3), whose diversity nearly covers the entire variability within the 1000 spiders encompassing 76 families. We conducted variable selection to probe structure-function associations in dragline silks based on the mean differences in the physical properties of the silks according to taxonomic categories and spidroin types (figs. S4 to S6). Briefly, the different ampullate-like spidroin sequences found across the different spider taxa were classified according to conserved patterns within repetitive domains; this led to the categorization into 20 sequence groups, which comprised seven MiSp subtypes, seven MaSp1 subtypes, four MaSp2 subtypes, and two MaSp3 subtypes, including MaSpN. We then analyzed the contributions of the different groups to the different physical properties of the corresponding dragline fibers. For instance, the silks of spiders from the genus Argiope and family Araneidae showed significantly higher toughness (mean differences of +0.068 and +0.039 GJ/m 3 , respectively) and expressed unique spidroins, including MaSp3 (group 19), MaSp2 (group 11), and MaSp1 (group 17), resulting in mean differences in silk toughness of +0.041, +0.031, and +0.035 GJ/m 3 , respectively ( Fig. 4A and fig. S7A). This suggests that the possession of MaSp3 (group 19) resulted in an increase in toughness of at least 0.041 GJ/m 3 , corresponding to an increase of approximately 32% relative to the overall average of 0.127 GJ/m 3 . However, this was most likely as combined effect of Araneidae-type MaSps, including MaSp2 (group 11) and MaSp1 (group 17), coinciding with the existence of MaSp3 (group 19). A similar significant superiority of Araneidae dragline fibers was observed in terms of strain at break, crystallinity, diameter, thermal degradation temperature, and supercontraction. Strain at break and supercontraction were the only properties for which the possession of the MaSp2 subtype was a greater determinant than belonging to family Araneidae, as tensile strength increased 3.7% in association with MaSp2 (group 13) and supercontraction increased 15.7, 15.8, 14.3, and 11.0% in association with MaSp2 (group 14), MaSp2 (group 13), MaSp2 (group 11), and MaSp1 (group 17), respectively ( Fig. 4B and fig. S7B). The significant contribution of MaSp2 to spider dragline supercontraction and elasticity was in line with previous suggestions regarding the different roles of MaSp1 and MaSp2 ( 47 , 48 ), but one spidroin subtype, MaSp2 (group 15), conversely influenced supercontraction (−8.3%; see fig. S7B). A close inspection of the repetitive motifs of MaSp2 (group 15) revealed longer polyalanine regions. Accordingly, the average β sheet region length (typically the polyalanine region but defined as stretches of multiple A, S, and V for more than five amino acid residues, as these amino acids tend to substitute for polyalanine) was negatively correlated with supercontraction (−0.508 for MaSp1 and −0.306 for MaSp2 β sheet regions). Furthermore, the correlation was higher when both the amorphous region and the polyalanine lengths were taken together in the ratio (figs. S8 and S9). The average amorphous to β sheet region length ratios for all repeats within the spidroins of interest were 0.526 for MaSp1 and 0.394 for MaSp2. Therefore, the proportion of amorphous regions within the spidroin is the key factor contributing to supercontraction. The contribution of the relaxation of orientation in the amorphous region of spidroins to supercontraction was suggested in previous works ( 49 , 50 ) and was confirmed by the analysis of our comprehensive dataset. Considering the effects of the amorphous and crystalline regions on the measured physical properties, as described above, the repetitive sequences, rather than the terminal domains, can be considered to play the main roles in determining these physical and mechanical properties. Shrinkage of artificial spider silk threads and textiles due to supercontraction is often considered an undesirable property for industrial use, and these findings may contribute in designing primary sequences, avoiding supercontraction while preserving toughness of the material. Fig. 4. Linking sequences to the physical properties of dragline silk. The different ampullate-like spidroin sequences found across the different spider taxa were classified according to conserved patterns within repetitive domains; this led to the categorization into 20 sequence groups, which comprised seven MiSp subtypes, seven MaSp1 subtypes, four MaSp2 subtypes, and two MaSp3 subtypes (figs. S3 to S5). MaSp groups most strongly contributing to the physical properties were selected through statistical screening (see Materials and Methods). ( A ) Toughness distribution among different spider families, as correlated with the presence or absence of selected MaSp subtypes: MaSp3 (group 19), MaSp2 (group 17), and MaSp1 (group 17). ( B ) Supercontraction distribution among different spider families, compared with the presence (+) or absence (−) of specific MaSp subtypes. Four MaSp2 groups (groups 14, 13, 11, and 17) showed higher average supercontraction than Araneidae. ( C ) Scatterplot of physical properties (toughness or supercontraction) as a function of the average abundance per repeat (%) of certain amino acid motifs. See data file S4 for comprehensive screening of amino acid sequence motifs contributing to the physical properties. Abundance of motifs was normalized by the number of repetitive sequences within a spidroin fragment, and this normalized abundance was correlated with the physical properties to screen for highly contributing motifs. Spot color denotes the spider family, and Pearson correlation values are shown in the top right corners. Here, AGQG motif in MaSp1 is positively correlated with supercontraction, and AAAAAAAA motif of MaSp2 is negatively correlated. Likewise, YGQGG motif in MaSp1 is positively correlated with toughness. To further extract the sequence features contributing to the physical properties of spider silk, we screened the amino acid motifs correlated with the measured properties (data file S4), and the main findings are summarized in Table 1 . Confirming the above analysis of categorical variable selection according to gene class and taxonomy, the degree of supercontraction was strongly negatively correlated with the frequency of the appearance of polyalanine sequences and was correlated with short (one– to four–amino acid) motifs corresponding to amorphous regions such as G, GG, and AGQG ( Fig. 4C and data file S4). Likewise, strain at break was negatively correlated with polyalanine prefixed with Ser in MaSp2 and positively correlated with MaSp1/2 amorphous regions including Pro, which presumably adds to the elasticity of this region ( 51 ). Concerning tensile strength, the inclusion of Ala in the amorphous region of MaSp1 and Pro in that of MaSp2 had a negative effect, while the inclusion of Ser in the amorphous region of MaSp1 had a positive influence. The GYGQGG motif in MaSp1 was most strongly correlated with both tensile strength ( r = 0.377) and strain at break ( r = 0.416) and was consequently also correlated with toughness [YGQGG was ranked 1 ( r = 0.547), and GYGQGG was ranked 2 ( r = 0.531)] ( Fig. 4C ). The Tyr residues in the amorphous regions of MaSp1 may play a critical role in intermolecular chain packing in the spider dragline, similar to the intermolecular interactions suggested from the structural analysis of silkworm silk ( 52 ) . The inclusion of Pro in the MaSp2 amorphous region, along with the SY and SV motifs in MaSp1, was negatively correlated with toughness. The presence of GGS after the polyalanine region in MaSp1 was positively correlated with toughness. Confirmation of the contribution of these motifs to the physical properties using recombinant properties would be a future direction to fully understand the primary sequence designs, leading to the extraordinary mechanical properties of spider silk. Table 1. Feature extraction summary. Amino acid sequence features of the underlying MaSp repetitive domains that have positive and negative effects on the different physical properties of spider dragline silks are presented. Poly-Ala, polyalanine. \n Positive effect \n \n Negative effect \n Toughness MaSp1-GYGQGG P, SQGP in MaSp2 MaSp1–poly-Ala ending with GGS SY, SV in MaSp1 MaSp1-GGGQ Tensile strength MaSp1-GYGQGG MaSp2-PQ MaSp1-SS before poly-Ala Lacking S in GQG motif in MaSp1 MaSp1-QGGS A before GQG motif in MaSp1 Strain at break MaSp1-GYGQGG ASA before poly-Ala QGP, PGA in MaSp1 Young’s modulus PA in MaSp2 Q in MaSp2 GL in MaSp1 and MaSp2 MaSp1-GGQ MaSp1-GQ Crystallinity PA, N, A, GA in MaSp2 GT in MaSp1 MaSp1-GQ MaSp1-GGQ Birefringence SS, N, GQQ in MaSp2 MaSp1-GQGGAGAA TGG in MaSp1 Diameter MaSp1-GAAAAAAG MaSp2-PSGPGS MaSp1-AAGGAGQG MaSp2-SQG MaSp2-PQG MaSp2-AAGGY MaSp1-QS N % water loss MaSp2-PGGYGP MaSp1-SQGAG MaSp2 poly-Ala V in MaSp2 GT in MaSp1 Water content MaSp1-GSG MaSp2-QQGPG MaSp2-GAS MaSp1-PGAA A in MaSp1 and MaSp2 Supercontraction MaSp2 presence Poly-Ala in MaSp1 and MaSp2 MaSp1-AGQG MaSp1-GLG Together, our findings provide a thorough mechanistic evaluation of the pathways of spidroin evolution. First, the physical properties of spider dragline silk have significantly diversified and specialized with the deployment of orb webs related to Araneoidea species ( 43 ), and this is mirrored by the diversification of MaSp paralogs, as previously suggested through meta-analysis of silk mechanics and sequence motifs ( 46 ) . We propose that MaSp1 is specialized to increase fiber strength, while MaSp2 is specialized to increase fiber elasticity, and the combination of these paralogs results in the high toughness of dragline silk. Furthermore, species requiring extraordinary fiber toughness have evolved to produce a third paralog, MaSp3, whose presence was clearly shown to be one of the strongest determinants of high toughness in our analysis. The full complexity of the proteome composition of dragline silk is beginning to be elucidated. However, MaSp3 was shown to be the major component of Nephilinae and Araneus dragline silks, and the complexity of these silks extends beyond the composition of spidroins ( 42 ), involving other essential components referred to as spider silk-constituting elements (SpiCE), which has been shown to double the tensile strength of an artificial spider silk–based film in vitro ( 41 ) . Elasticity and supercontraction are related properties of dragline silk that are likely linked to the sequence features of MaSp2, in which the ratio of amorphous to β sheet regions plays critical roles. Similarly, the compositions of several amino acid motifs in the amorphous regions of MaSp1 were shown to be highly correlated with the toughness of dragline silk; these sequence-level design elements derived from the comprehensive analysis of 1000 spiders provide a foundation for the design and production of artificial spider silks. Many of these designs may also be applicable to other protein-based and polymeric materials. In this study, we have provided a comprehensive dataset encompassing the genotypes and phenotypes (including the mechanotypes) of spider silks and identified the design elements responsible for the extraordinary mechanical and physical performances of these silks. Silk proteins have convergently evolved in various lineages ( 53 ), but the sequence motifs ( 54 ), amino acid composition ( 55 ), and the trade-off between tensile strength and elasticity as a function of ratio between amorphous and crystalline regions ( 56 ) have been shown to have a certain degree of shared characteristics, something supported by our spider silk data. Therefore, these data will serve as a framework for the future analysis of silk proteins and other structural proteins as biomaterials. Similar data-driven approaches encompassing protein materials excelling in properties other than toughness, such as elastomers and adhesive proteins, could also accelerate our understanding on the genetic design principles of the biomaterials. Methods including computational modeling and simulation that allow the prediction of the outcomes of molecular interactions between the multiple components of these biomaterials, such as multiple MaSp-type spidroins and SpiCE proteins, would be an important future direction. We focused on the silk mechanics in this work, but the 1000 spider transcriptome data should also facilitate arachnid and arthropod phylogenomics."
} | 9,043 |
30913545 | null | s2 | 277 | {
"abstract": "Biofilms are communities of sessile microbes that are bound to each other by a matrix made of biopolymers and proteins. Spatial structure is present in biofilms on many lengthscales. These range from the nanometer scale of molecular motifs to the hundred-micron scale of multicellular aggregates. Spatial structure is a physical property that impacts the biology of biofilms in many ways. The molecular structure of matrix components controls their interaction with each other (thereby impacting biofilm mechanics) and with diffusing molecules such as antibiotics and immune factors (thereby impacting antibiotic tolerance and evasion of the immune system). The size and structure of multicellular aggregates, combined with microbial consumption of growth substrate, give rise to differentiated microenvironments with different patterns of metabolism and gene expression. Spatial association of more than one species can benefit one or both species, while distances between species can both determine and result from the transport of diffusible factors between species. Thus, a widespread theme in the biological importance of spatial structure in biofilms is the effect of structure on transport. We survey what is known about this and other effects of spatial structure in biofilms, from molecules up to multispecies ecosystems. We conclude with an overview of what experimental approaches have been developed to control spatial structure in biofilms and how these and other experiments can be complemented with computational work."
} | 383 |
26510159 | PMC4624983 | pmc | 278 | {
"abstract": "Tropical reef-building coral stress levels will intensify with the predicted rising atmospheric CO 2 resulting in ocean temperature and acidification increase. Most studies to date have focused on the destabilization of coral-dinoflagellate symbioses due to warming oceans, or declining calcification due to ocean acidification. In our study, pH and temperature conditions consistent with the end-of-century scenarios of the Intergovernmental Panel on Climate Change (IPCC) caused major changes in photosynthesis and respiration, in addition to decreased calcification rates in the coral Acropora millepora . Population density of symbiotic dinoflagellates ( Symbiodinium ) under high levels of ocean acidification and temperature (Representative Concentration Pathway, RCP8.5) decreased to half of that found under present day conditions, with photosynthetic and respiratory rates also being reduced by 40%. These physiological changes were accompanied by evidence for gene regulation of calcium and bicarbonate transporters along with components of the organic matrix. Metatranscriptomic RNA-Seq data analyses showed an overall down regulation of metabolic transcripts, and an increased abundance of transcripts involved in circadian clock control, controlling the damage of oxidative stress, calcium signaling/homeostasis, cytoskeletal interactions, transcription regulation, DNA repair, Wnt signaling and apoptosis/immunity/ toxins. We suggest that increased maintenance costs under ocean acidification and warming, and diversion of cellular ATP to pH homeostasis, oxidative stress response, UPR and DNA repair, along with metabolic suppression, may underpin why Acroporid species tend not to thrive under future environmental stress. Our study highlights the potential increased energy demand when the coral holobiont is exposed to high levels of ocean warming and acidification.",
"conclusion": "Conclusions Our results at both the phenotypic and global gene expression level indicate that Acroporid corals are highly sensitive to predicted future increase in ocean temperature and OA, and may fail to thrive and show decreased calcification. This phenotypic response is perhaps not a direct result of reduced calcification as this process can be sustained at high pCO 2 levels [ 29 , 41 , 43 ], but instead a by-product of increased maintenance cost under new environmental conditions of high temperature and pCO 2 , and diversion of cellular ATP to pH homeostasis, oxidative stress response, UPR and DNA repair. This, together with metabolic suppression, can result in decreased amounts of energy available for calcification and other cellular functions, which may in the longer term lead to increased levels of coral mortality. In order to elucidate mechanisms for the large intraspecific variability of coral sensitivity to both thermal and ocean acidification stress, there is a need for future long-term studies of a range of coral species exposed to modulations in temperature and pCO 2 both in isolation and as a combined effect. In addition there is a need for studies to investigate the adaptation or acclimatization potential of reef building corals to future ocean temperature and acidification conditions.",
"introduction": "Introduction Changes in atmospheric CO 2 are likely to fundamentally alter ocean ecosystems through their influence on sea temperature and carbonate ion chemistry [ 1 , 2 ]. Coral reef ecosystems are among the major oceanic systems that are likely to be detrimentally affected by global warming and ocean acidification (OA) [ 3 , 4 ]. These highly productive and biologically diverse ecosystems provide important goods and services to more than 450 million people in coastal communities around the world [ 5 ]. Local and global stressors are affecting corals and reef communities, and threaten their long-term survival [ 6 ]. Rapidly warming oceans threaten reef-building corals as anomalously high temperatures lead to the breakdown of symbiosis between the coral host and its symbiotic dinoflagellates, a phenomenon called coral “bleaching”. Over the past few decades, mass coral bleaching events have increased both in frequency and intensity [ 7 – 9 ]. The result of these events is usually high coral mortality and for colonies that survive, decreased colony growth and depressed reproductive output is common [ 10 , 11 ]. There are differences in bleaching susceptibility and severity among coral species and it has been shown that fast-growing branching genera such as Acropora are more susceptible to severe bleaching [ 8 , 12 ]. High oceanic uptake of increased atmospheric CO 2 resulting in OA also poses threats to a range of marine organisms, as it can potentially affect rates of calcium carbonate deposition due to the reduction in the saturation state of carbonate forms such as aragonite, calcite and magnesium calcite [ 13 ]. Most of the experimental studies on the effects of OA on marine calcifiers have been conducted in tank or mesocosm experiments [ 4 ], with a few studies conducted in the field showing changes in coral species composition or shifts from hard to soft coral dominance [ 14 , 15 ]. Studies have shown that reef calcifiers, such as corals and calcifying algae, will have decreased rates of calcification under future OA conditions [ 16 , 17 ]. To date, studies on the effect of OA on the rate of calcification of marine calcifiers have tended to dominate the literature [ 16 ], with a strong negative correlation between calcification rates and OA [ 4 , 16 , 18 , 19 ]. Given the importance of variables such as pH and the carbonate ion concentration, it is perhaps not surprising that a large range of physiological processes appear to be influenced by OA in marine organisms [ 20 – 22 ] and there is evidence that OA can affect symbiont population density and depress metabolism, processes that lead to the biological deposition of calcium carbonate in reef-building corals, before effects on calcification rates are apparent [ 23 ]. To a large extent, most previous studies have examined ocean warming and ocean acidification in isolation of each other with notable exceptions [ 24 ]. Considering that future changes in atmospheric CO 2 concentrations will affect both ocean temperature and chemistry at the same time, it is important to study the combined effects of temperature change and OA. Recently there has been an increase in reports on the combined effect of increased temperature and pCO 2 on marine calcifiers [ 24 – 27 ]. Overall it is clear that responses are variable, can be nonlinear and there are intra- and inter-specific variances for reef-building corals, where the combined effect of high temperatures and high pCO 2 levels mostly leads to decreases in calcification rates [ 24 – 27 ]. Previous studies have mainly focused on the effect of future changes in ocean temperature and chemistry on calcification, while many other physiological processes, such as photosynthesis, energy metabolism, changes to cell membrane physiology, reproduction, overall fitness and energy costs associated with acclimation to environmental conditions have received less attention [ 23 ]. In addition, there is a lack of knowledge about the transcriptional regulation of specific molecular pathways involved in changes in calcification, bleaching and stress response observed at the phenotype level, when exposed to changes in ocean temperature and chemistry. A few studies have investigated changes in global gene expression in response to OA in juvenile [ 28 ] and adult corals [ 23 , 29 ], and there is knowledge on the effect of increases in temperatures on both larval and adult coral transcriptomes [ 30 – 34 ]. To date, however, there is no information on the effect of different future scenarios of ocean warming and chemistry on the physiology of reef-building corals, specifically in terms of photosynthesis and respiration rates, Symbiodinium density and pigment concentrations, host protein and lipid levels, and calcification rates, coupled with the underlying molecular mechanisms for changes observed at the phenotype level. The aim of our study was to contribute to this knowledge gap and investigate how changes in both temperature and pCO 2 affect the physiology of the coral holobiont, and how the changes are reflected in the metatranscriptome. Branching Acroporid corals are important reef builders that create most of the habitat complexity in the Indo-Pacific, and have been found to be highly sensitive to both thermal and OA stress [ 8 , 12 , 18 ]. Based on these biological features we chose Acropora millepora as the model species for this study. In this study we aimed to measure changes in selected physiological processes in the coral holobiont, which highlight important biological functions such as photosynthesis (oxygen evolution rates, Symbiodinium pigment concentrations), respiration, coral-algal symbiosis ( Symbiodinium population densities), energy storage potential (lipid and protein) and calcification. These selected physiological processes can provide an indication of overall coral holobiont health status and calcification potential of the organism when exposed to environmental stress.",
"discussion": "Discussion Changes in selected physiological processes of the phenotype Our study shows that exposure of the reef-building coral, A . millepora , to predicted future pH and temperature conditions under the business-as-usual projection RCP8.5, results in corals with reduced Symbiodinium populations and chlorophyll a levels, indicating mild coral bleaching, these results support previous findings on the compounding effect of the interaction between increased temperature and OA stress, and where destabilization of the coral-algal symbiosis occurs at lower thermal thresholds [ 19 , 20 , 35 ]. We also report major changes in photosynthesis and respiration, in addition to decreased calcification ( Fig 1 ). Although our study cannot shed light on the adaptation or acclimatization potential of reef building corals to future ocean temperature and acidification conditions, we provide information about potential impacts for branching corals under such conditions. Our evidence for decreasing rates of gross photosynthesis per surface area and Symbiodinium cell, compounded by reduced Symbiodinium populations, indicates disruption of photophysiological processes in Symbiodinium and may lead to a reduction in photoassimilates translocated to the host coral. Such results can have serious impacts on the ability of the host to recovery from bleaching. Of considerable interest, was the observation that there was a 2-fold downturn in dark respiration per coral surface area (cm -2 ) which suggested reduced growth rates and/or metabolism under these environmental conditions. These changes are likely to have long-term negative effects on host growth and fecundity, with the prospect of increased susceptibility to disease and mortality, especially if Symbiodinium populations fail to recover rapidly [ 36 ]. Interestingly, there was an increase in photosynthesis for PI corals compared to their conspecifics under PD conditions ( Fig 1C and 1D ), which perhaps reflects that these corals are more suited to ocean conditions seen 100 years ago, and is consistent with mesocosm studies finding reef communities performing better under pre-industrial conditions [ 24 ]. In our study, however, increased productivity in PI corals did not translate into higher tissue growth or calcification rates. This may be a consequence of the short time frame of the experiment (5 weeks) and it is possible that this benefit may have been observed if a longer experimental incubation was carried out. Regardless, implications of these results need to be taken into account when speculating on the potential of reef builders to acclimatize or acclimate to future conditions (as stressed by [ 24 ]). Despite variability among reef calcifiers, overall calcification rates have been predicted to decrease by the end of the century, as a result of changes in both pCO 2 and temperature [ 4 , 37 – 39 ]. Our results show that corals exposed to conditions like that of RCP8.5 had negative calcification rates and further emphasize the extent of challenges that reef building corals are likely to face in warmer and more acidic future oceans. Corals have the ability to maintain cell pH homeostasis, as it is critical to a range of cellular functions [ 40 ] and can up-regulate the pH at the calcification site, even under highly acidified conditions [ 41 – 43 ]. These pH regulation processes require elevated maintenance costs [ 4 , 22 , 23 , 44 ]. In our study, under both higher pCO 2 and temperature, our results do not reflect a decline in energy reserves (total lipids and proteins); in fact, we saw an increase in lipids for both RCP4.5 and RCP8.5 corals, and an increase in proteins under RCP4.5 conditions ( Fig 1 ). However, it is important to note that our measure of total lipids did not separate the coral fraction from the Symbiodinium fraction, and it may be that Symbiodinium plays an important role in determining lipid levels in Acropora millepora under different environmental conditions, as it has been shown in other coral species [ 45 ]. Our results showing an increase in lipids fit well with observations that under long term exposure to OA, some coral species are able to survive despite losing their skeleton and changing into large solitary polyps with three times the biomass of coral colonies under controlled conditions [ 46 ]. Although we cannot relate our results to changes in acidification alone, changes observed in the present study also correlate well with changes in energy reserves for A . millepora [ 47 ], where energy reserves were not metabolized as a function of increased acidification or increases in both temperature and acidification. Schoepf et al [ 47 ] had similar results in that calcification declined under high OA and temperature, while lipid concentrations increased. It was suggested that maintenance of energy reserves such as lipid concentrations under environmentally stressful conditions may enable corals to maintain their reproductive output [ 48 ]. This may explain the results in our study, as our A . millepora corals were exposed to the experimental conditions just before their annual spawning event. Given that exposure to increased temperatures can lead to photosynthetic dysfunction in Symbiodinium [ 36 ] and that inhibited photosynthesis in can influence sensitivity to OA and host cell recovery from cellular acidosis [ 49 ], our exploration of Symbiodinium pigment profiles in addition to measuring photosynthesis provided further insight into the photosynthetic capacity of the coral holobiont. Corals can adjust photosynthetic pigment concentrations to maximize irradiance levels available for the photosynthetic endosymbiont [ 50 ]. Despite a reduction of the endosymbiont population in RCP8.5 corals, the chlorophyll a level per cell did not change in comparison to PD and PI corals, while in RCP4.5 corals chlorophyll a concentrations per cell increased ( Fig 2 ). In addition, RCP4.5 corals also increased their concentrations of accessory photosynthetic pigments per cell ( Fig 2 ). This may reflect the ability of RCP4.5 corals to adjust their photosynthetic unit under changed environmental conditions, to maximize light utilization and potentially higher productivity, which was also observed at the per cell level in RCP4.5 corals ( Fig 1D ). RCP8.5 corals did increase accessory pigments levels, which in this case may reflect the changed light environment within these corals as the endosymbiont population was reduced by half and accessory pigments were needed to absorb excess light energy [ 51 ]. High light exposure has been also been shown to induce carotenoid production in corals [ 52 ]. Xanthophyll de-epoxidation, i . e . the conversion of diadinoxanthin (Ddn) to diatoxanthin (Dtn) upon light absorption, can alleviate high light stress through dissipation of absorbed radiation to heat [ 53 – 55 ]. Only RCP4.5 corals had greater de-epoxidation (Dtn/Dtn+Ddn) compared to PD corals. A correlation between an increase in de-epoxidation and greater non photochemical quenching has been shown [ 56 ]. This is consistent with a possible greater need for non-photochemical quenching in RCP4.5 corals under changed environmental conditions [ 57 ]. The observation that RCP8.5 corals did not change their de-epoxidation levels compared to PD corals may reflect that photosynthesis is not functioning adequately in these coral-algal complexes, supporting the downturn in productivity ( Fig 1C and 1D ) and metabolic suppression ( Fig 1E ). Energy dissipation is essential for organisms exposed to environmental stress, where an increased degree of excessive light absorption is accompanied by an increased need for energy dissipation [ 57 ]. The lack of an up-regulation of xanthophyll de-epoxidation in this case, may point to potential oxidative stress and/or damage present within the coral holobiont. Global gene expression response: Pre-industrial scenario Our analysis of the global metatranscriptome changes in A . millepora allowed us to identify potential molecular pathways and candidate genes involved in the holobiont response of selected physiological processes seen at the phenotype level (Figs 1 and 2 ), and to predict changes during past and future ocean warming and chemistry conditions. Although the coral holobiont is composed of the coral host, Symbiodinium populations and other symbiont partners (bacteria, viruses, fungi and others) [ 58 – 60 ] our global gene expression response was dominated by coral and secondly Symbiodinium genes, while genes from the microbial community in the coral holobiont were a small proportion. This limits our study’s ability to determine important changes that may be occurring in the coral holobiont microbial partners which may have affected the changes that we saw in physiological processes at the phenotype level. We also did not measure changes in microbial communities in our coral branches under experimental treatments and acknowledge the importance of measuring fluxes in microbial communities in future studies to explore what role they may play. Over half of the modifications to the metatranscriptome in corals exposed to PI temperature and pCO 2 conditions were involved in metabolic processes, such as generation of precursor metabolites and energy, and protein metabolic processes ( Fig 4A , S1 Table ). Our pattern of coral host metabolic transcripts involved in generation of precursor energy pathways such as glycolysis, being up regulated while metabolic transcripts for glycolysis, Calvin cycle and photorespiration of Symbiodinium origin were down regulated ( S1 Table ) implies a complex array of metabolic pathways in host and endosymbiont underpinning an increase in photosynthesis seen at the phenotype level and warrants further investigation to understand the underlying mechanisms. Photosynthesis was also a major target of transcriptional modulation, where Symbiodinium transcripts for psbA, psbB, psbC, psbD1 and petB were up-regulated ( Fig 4A , S1 Table ), reflecting increases in photosynthesis seen at the physiological level ( Fig 1 ) and suggesting greater energy production potential for PI corals as compared to their PD conspecifics. Our Acropora millepora colonies contained the C3 clade of Symbiodinium . In a study that compared the transcriptomes of four Symbiodinium clades (A, B, C and D) [ 61 ], found that photosynthesis related transcripts were highly conserved and shared among all four clades and therefore supporting the notion that they are unique for dinoflagellates and have not changed through evolutionary time. The observed increase in cell cycle processes, including mitotic cell cycle, and DNA duplex unwinding during replication and M phase ( Fig 4 , S1 File ), may reflect an increase in growth either in the endosymbiont population and/or coral host. In addition, the fact that PI conditions resulted in the largest transcriptional regulation of Symbiodinium genes further points to changes in the endosymbiont population. These changes in metabolism, photosynthesis and cell cycle processes seen at the transcriptional level, reiterate the fact that past environmental conditions may in fact be more beneficial for A . millepora , perhaps suggesting that adaption to present-day conditions has not yet occurred. Global gene expression response: reduced emission scenario (RCP4.5) In response to environmental perturbations in ocean temperature and chemistry simulated by the emissions scenario (RCP4.5), a large part of the biological function of changes to the metatranscriptome belonged to metabolic and biosynthetic processes ( Fig 4B , S1 Table ). These findings may point to a greater energy potential for RCP4.5 corals as compared to PD corals, supporting the increase in photosynthesis per algal cell seen at the phenotype level ( Fig 1D ). This may be due to an increase in temperature and pCO 2 , both of which have, in the past, been shown to increase productivity [ 20 ]. Although not seen in this study due to the short time frame, Acroporid corals are likely to be stressed and show signs of bleaching and reduced calcification under an RCP4.5 scenario, especially after longer exposure to these conditions and during and/or after the summer [ 24 ]. A range of Ribosomal proteins were highly expressed under RCP4.5 conditions ( Table 3 , S1 Table ) and may suggest that cell growth is occurring in response to environmental change. Ribosome biogenesis regulation is a key element in controlling cell growth as ribosomes are required for growth and also because ribosome biogenesis consumes a large proportion of cellular energy. In fact, in a growing cell, up to 95% of total transcription can be attributed to ribosome biogenesis [ 62 ]. This is consistent with gene enrichment in cell proliferation and growth, and cell cycle biological processes ( Fig 4B ). The increase in reproductive processes as compared to PD corals is likely a reflection of this experiment occurring just prior to the annual spawning event, and RCP4.5 corals being exposed to higher temperatures. It has been shown that when corals are exposed to higher temperatures during the late period of gametogenesis, spawning can occur earlier in the spawning month [ 63 ], which may explain why in our study, there was an increase in genes related to reproductive events in RCP4.5 corals. Global gene expression response: business as usual scenario (RCP8.5) Metabolism Under the business as usual scenario, RCP8.5, the A . millepora holobiont showed signs of a stress response at the phenotype level through reductions in endosymbiont population density and chlorophyll a concentrations, reduced productivity and decreased respiration rates that may point to potential metabolic depression ( Fig 1 ). The cellular stress response was also apparent on the global gene expression level. Gene enrichment analysis showed that RCP8.5 corals had the smallest proportion of metabolic biological processes compared to corals exposed to RCP4.5 or PI conditions ( Fig 4 ). This is also supported by an overall down-regulation of transcripts involved in glucose metabolism, TCA and oxidative phosphorylation, indicating reduced oxidative metabolism and capacity to generate ATP and NADPH, as well as a reduction in coral sugar and glucose transporters ( Fig 3B , S3 Table ). Also, biosynthetic processes were less represented as compared to corals under PI and RCP4.5 conditions, perhaps reflecting a decrease in energy-requiring pathways of metabolism. In addition, there were fewer transcripts dedicated to reproduction, embryogenesis and morphogenesis as compared to RCP4.5 corals, which may again point to a more negative energy balance in stressed RCP8.5 corals. Metabolic depression in this study could also be a result of cellular acidosis as our RCP8.5 corals had decreased photosynthesis and reductions in Symbiodinium populations ( Fig 1 ) which may point to photosynthetic inhibition, and it has been shown that Symbiodinium photosynthetic activity is tightly coupled to the ability of the host cell to recover from cellular acidosis after high pCO 2 exposure [ 49 ]. It is common for organisms to depress their metabolic rates as a result of environmental stress, to temporarily deal with adverse conditions [ 64 ]. Metabolic depression seen in this study, however, is particularly alarming if it occurs during prolonged time frames, as corals are unable to return to their normal resting metabolic rate [ 22 ]. Circadian clock The circadian clock is a core mechanism in nearly all organisms which controls and synchronizes physiological processes to the day/night cycle. As such, circadian rhythms are important in controlling the metabolism of an organism (see review, [ 65 ]). It has been shown that disruptions to circadian rhythms, whether environmental or genetic, can have profound effects on metabolism and lead to metabolic diseases in mammals [ 66 ]. There are components of the circadian clock that can sense alterations to the cell’s metabolism, and it has been suggested that metabolism is not only affected by the circadian clock, but can also act as a modulator of the circadian clock [ 65 ]. In corals, like in many organisms, metabolic gene expression is linked to circadian rhythms [ 67 ]. Given the dramatic change in metabolism in terms of decreased respiration rates ( Fig 1 ) and down regulation of coral transcripts involved in oxidative metabolism, indicating potential metabolic depression ( S3 Table ) seen in our RCP8.5 corals, it is interesting to note that elements of the coral circadian clock machinery are up-regulated ( Fig 3A , S3 Table ), and NPAS3 has been shown to be involved in the circadian regulation of gluconeogenesis in mammals [ 68 , 69 ]. This may suggest that the circadian clock in corals is intrinsically linked, perhaps to controlling metabolic events (such as metabolic depression) during environmental perturbations. This is an avenue that needs to be explored in future studies. Increasing cellular stress In response to environmental stressors of increased temperature and OA, our measurements showing declines in Symbiodinium densities and pigment concentrations indicated a breakdown in the coral-algal symbiosis. Overall, our metatranscriptomic analysis points to a cellular stress response which is a universally conserved mechanism for protecting macromolecules within cells from potential damage resulting from abiotic stress [ 70 ]. It has been suggested that in Stylophora pistillata exposed to heat stress, the cellular stress response is to engage in Endoplasmic Reticulum (ER)-unfolded protein response (UPR) and ER associated degradation (ERAD) [ 33 ]. On the other hand, analysis of global gene expression in adult Pocillopora damicornis exposed to extreme CO 2 -driven pH declines of ≤ 7.4, showed a lack of cellular stress and changes were seen in transcripts associated with the transport of calcium and carbonate ions, organic matrix, photosynthesis and increased energy metabolism [ 29 ]. Our adult corals, which were exposed to both an increase in temperature and a CO 2 -driven decrease in seawater pH, did show a cellular stress response which was similar to that seen in heat-stressed S . pistillata [ 33 ], perhaps supporting the notion that exposure to OA can lower the thermal threshold for corals to experience thermal stress [ 19 , 20 , 35 ]. Our results point to up-regulation of processes involved in oxidative stress, DNA repair, Ubiquitin-mediated proteolysis, apoptosis and toxins ( S1 File , Fig 3 , S3 Table ). We also saw modulations in calcium signaling/homeostasis and cytoskeletal interactions ( Fig 3 , S3 Table ), which have previously been shown to be part of the coral holobiont stress response [ 23 , 30 , 33 ]. Nuclear factor (NF)-κB signaling can regulate physiological processes involved in innate immune response, cell death and inflammation (see review [ 71 ]). The co-expression of transcripts involved in NF-κB signaling ( e . g . Traf3) with transcripts encoding for toxins and cytolysis, suggests that this network of genes may be involved in innate immunity in our stressed coral holobionts. Stress-induced breakdown of the coral-algal association can result in shifts in microbial communities and increased susceptibility to coral pathogens [ 72 , 73 ]. This may be why there is an increase in pathways potentially involved in innate immunity in the stressed RCP8.5 corals. There were 14 genes enriched in the Ubiquitin-mediated proteolysis pathway ( Table 3 , S3 Table ), which is involved in protein degradation and turnover [ 74 ]. Eight of the Ubiquitin-mediated proteolysis enriched transcripts found in this study (UBLE1B, UBE2D_E, UBE2J1, UBE2W, HERC2, CYC4, F-box and Apc2), were the same as what Maor-Landow et al [ 33 ] found in their heat-stressed corals. It has been suggested that protein ubiquitination and an increase in ubiquitin expression levels is a sign of cellular heat stress [ 75 ]. Potential protein degradation found in this study may be one of the cellular markers for both temperature and ocean acidification stress in corals. We also observed a down-regulation in Notch signaling, which was also seen in heat-stressed S . pistillata [ 33 ]. The observation of up-regulation in a range of coral and Symbiodinium transcripts encoding oxidative stress DNA damage/repair, molecular chaperones, ER stress and apoptosis proteins under RCP8.5 conditions ( S3 Table ), further point to a cellular stress response [ 23 , 30 , 33 ]. These changes in the adult coral gene expression profiles also further supports the notion that the coral stress response is fairly similar across a range of environmental stressors such as temperature [ 30 – 32 ], darkness [ 76 ] and low pH [ 23 ]. Modification of calcification Many genes involved in calcification in corals are likely to be taxonomically restricted and unique to corals [ 28 ]. As our transcriptomic results only refer to transcripts with hits to known proteins, we are unable to compare potential genes involved in the calcification process [ 28 ] which have no hits to the Swissprot database. At the phenotype level, we observed a decrease in biomineralization ( Fig 1 ) under the RCP8.5 scenario. Interestingly, we saw an increase in LRP5 and other components of the canonical Wnt signaling pathway ( Fig 3A ). In human skeletons, LRP5 signaling has been shown to regulate bone mass [ 77 ], and it may be that in corals, this pathway can play a role in calcification, which could be an avenue for future work. Global gene expression patterns showed a complex pattern of up-regulation of coral T-type calcium channel and up-regulation of other calcium transporters, while there was a down-regulation in bicarbonate transport, carbonic anhydrases and transcripts potentially implicated in the organic matrix ( S3 Table ). T-type calcium channels have been suggested to be associated with bone development in vertebrates [ 78 , 79 ], and the fact that up-regulation of these was found in RCP8.5 corals in this study, and in the high CO 2 treatment in [ 28 ], may imply that these calcium transporters are part of the coral response to acidification stress, although further studies should elucidate their exact roles. Up-regulation of transcripts for a voltage-gated Ca 2+ channel was also found in P . damicornis adult corals in response to ocean acidification conditions [ 29 ]. In our study, we had an up-regulation of Ca 2+ channels of coral origin, other than those found in [ 28 ], suggesting that an increase in both temperature and pCO 2 , as manipulated in this study, has either a direct effect (transport of calcium to the extracellular medium) and/or an indirect effect on calcification, through changes in central calcium signaling pathways ( Fig 3 ). Carbonic anhydrases catalyze the inter conversion of HCO 3 \n - and CO 2 , and are involved in pH regulation. In corals, they are also involved in biomineralization and carbon exchange between host and endosymbiont [ 80 – 82 ]. Similar to results seen in [ 28 ], we report down-regulation for one bicarbonate transporter (SLC4a4), and down-regulation of six coral carbonic anhydrases; while there was up-regulation of one Symbiodinium carbonic anhydrase, which differs to results seen in [ 29 ] where an up-regulation of several bicarbonate transporters and two carbonic anhydrases was seen under acidification levels comparable with our study. The differences between our results and those seen in [ 29 ] are likely to be due to difference in coral species and experimental conditions used, especially highlighting that our study measured the effect of combined temperature and OA stress. Our study also highlights a down-regulation in a few coral transcripts (Bmp6 and SVEP1) that may encode for skeletal organic matrix proteins. In our study, modulations to ion transport and potential matrix proteins support our findings at the phenotype level, where there was a reduction in biomineralization—a change that was not measured in both [ 28 , 29 ], which highlights the necessity to measure phenotypic alterations so that modulations at the global transcriptomic level can help us understand the molecular processes underlying the phenotypic change. Cell-wide responses to RCP8.5 Based on phenotype-level changes and global gene expression levels, we propose a model of cellular events occurring in the A . millepora holobiont in response to future ocean temperature and chemistry conditions as predicted by the RCP8.5 scenario ( Fig 5 ). This is an attempt to highlight how the proposed coral heat stress mechanisms of URP and ERAD [ 30 , 33 ] may fit with the coral specific acidosis response in terms of acid base regulation [ 23 , 28 , 29 ], metabolic depression [ 23 , 28 ], and novel responses of DNA damage/repair and circadian clock modifications observed in this study. Evidence of acid base regulation changes and ion transport in both coral host and endosymbiont ( Fig 3 , S3 Table ) indicate that cellular homeostasis may not have been reached under new environmental conditions, as proton transport out of the cell is overall down-regulated both by the host and as a result of decreased photosynthesis [ 49 ]. As a result, damage or disruption to either endosymbiont cell and/or both host cell/ mitochondrion and nucleus may occur. The result is an increase in oxidative stress and DNA damage, as indicated by an increase in oxidative stress and DNA damage/repair transcripts ( Fig 3 , S3 Table ). This can, in turn, lead to changes in calcium signaling/homeostasis [ 23 , 32 , 36 ], which has an effect on modifications in the extracellular matrix, cytoskeletal interactions, cell signaling (including transcription regulation and RNA processing) and immunity/cell death through the NF-κB pathway ( Fig 3A , S3 Table ). Additionally, evidence of up-regulation of the ubiquitin-mediated proteolysis pathway and ER stress ( Fig 3 , Table 3 , S3 Table ) suggest that cellular energy may be used for the UPR response [ 30 , 33 ], and this may lead to cell cycle arrest and control cell proliferation/death. The DNA damage/repair balance may also regulate cell proliferation and lead to cell cycle arrest, as well as regulate other signaling events [ 83 ] ( Fig 3 , S3 Table ). Disruption in endosymbiont and/or host can also lead to metabolic suppression as seen by down-regulation of metabolic transcripts involved in energy metabolism in both the coral host and Symbiodinium ( Fig 3 , S3 Table ), and evidence at the phenotype level ( Fig 1 ). The up-regulation of circadian clock components ( Fig 3A , S3 Table ) may point to a role that the coral clock can play in regulating metabolic pathways in a stressed coral host (in this case, suppressing metabolism). Since, in corals, the genes involved in stress response and cell protection are tightly linked to the circadian clock [ 67 ], we suggest here that in our stressed coral, the increase in oxidative stress and changes in metabolic transcripts create a feedback loop in which the circadian clock genes are modulated, which in turn affects coral metabolic pathways ( Fig 5 ). 10.1371/journal.pone.0139223.g005 Fig 5 A proposed model of cellular events occurring in response to future Representative Concentration Pathway RCP8.5 scenario with increased temperature and high CO 2 conditions (for gene list see S3 Table ). These changes lead to compromised health in Acropora millepora (reduction in symbiont cells, decreased photosynthesis and respiration and reduced calcification). The schematic depicts an endodermal cell which contains the Symbiodinium cell. Cellular events depicted here, especially acid base regulation at the cell membrane are also likely to occur in other cell types which do not contain symbiont cells. Exposure to increased temperature and seawater carbonate chemistry leads to changes in acid base regulation and ion transport at the cell membrane both in coral host and in the symbiont cell. Despite regulatory changes at the cell membrane cellular homeostasis may not be reached and acidosis may occur. The result may be an increase in reactive oxygen species due to a disruption (*) in the symbiont cell (S) and /or in the coral host mitochondrion (M), which may also produce reactive nitrogen species. In addition DNA damage may increase and not equal repair mechanisms in the cell nucleus (N) and / or M and S, which can lead to cell cycle arrest and /or cell death. Oxidative stress and endoplasmic reticulum (ER) stress, due to mis-folded protein accumulation and increased ubiquitin mediated proteolysis, may also lead to cell cycle arrest and eventually cell death, in addition to affecting cell calcium homeostasis/signaling, which in turn may lead to modifications in the extracellular matrix, cytoskeletal interactions and cell signaling. Oxidative stress can also affect immunity and /or cell death through the NF-κB pathway. In addition to the disruption in both S and M leading to suppression of metabolism, the coral circadian clock can also be influenced by oxidative stress and/or change in cell metabolism and in turn affect coral host metabolic pathways, which in this case is metabolic depression. Black arrows indicate up regulation and white arrows indicate down regulation. Conclusions Our results at both the phenotypic and global gene expression level indicate that Acroporid corals are highly sensitive to predicted future increase in ocean temperature and OA, and may fail to thrive and show decreased calcification. This phenotypic response is perhaps not a direct result of reduced calcification as this process can be sustained at high pCO 2 levels [ 29 , 41 , 43 ], but instead a by-product of increased maintenance cost under new environmental conditions of high temperature and pCO 2 , and diversion of cellular ATP to pH homeostasis, oxidative stress response, UPR and DNA repair. This, together with metabolic suppression, can result in decreased amounts of energy available for calcification and other cellular functions, which may in the longer term lead to increased levels of coral mortality. In order to elucidate mechanisms for the large intraspecific variability of coral sensitivity to both thermal and ocean acidification stress, there is a need for future long-term studies of a range of coral species exposed to modulations in temperature and pCO 2 both in isolation and as a combined effect. In addition there is a need for studies to investigate the adaptation or acclimatization potential of reef building corals to future ocean temperature and acidification conditions."
} | 9,988 |
25895940 | null | s2 | 279 | {
"abstract": "Nearly all bacteria form biofilms as a strategy for survival and persistence. Biofilms are associated with biotic and abiotic surfaces and are composed of aggregates of cells that are encased by a self-produced or acquired extracellular matrix. Vibrio cholerae has been studied as a model organism for understanding biofilm formation in environmental pathogens, as it spends much of its life cycle outside of the human host in the aquatic environment. Given the important role of biofilm formation in the V. cholerae life cycle, the molecular mechanisms underlying this process and the signals that trigger biofilm assembly or dispersal have been areas of intense investigation over the past 20 years. In this Review, we discuss V. cholerae surface attachment, various matrix components and the regulatory networks controlling biofilm formation."
} | 211 |
35425436 | PMC8981094 | pmc | 280 | {
"abstract": "In this study, a simple method without any additional chemical modification is proposed to fabricate underoil superhydrophobic surfaces with micro- and nano-hierarchical structures using a nanosecond laser system. The fabricated surfaces exhibited extreme superhydrophobicity and underoil superhydrophobicity with high contact angles of 153.8 ± 1.5° and 161.3 ± 1.1°, respectively. The results show that even after 20 abrasion cycles, the fabricated surfaces retained water repellency and self-cleaning performance under oil, while the superhydrophobicity in air was not resistant to wear. In addition, the fabricated brass meshes can also be used to separate oil in an oil–water mixture based on the prewetting induced underoil superhydrophobicity after being damaged. The separation efficiency was as high as 97.8%, which made them more appropriate for the oil–water separation than those based on superhydrophobicity. The proposed fabrication method is suitable for large-scale and mass production and provides a new avenue and possibility for further development of robust functional interface materials.",
"conclusion": "4. Conclusions This study presents a simple fabrication method of underoil superhydrophobic surfaces based on laser-ablated technology, which does not require any chemical modification. The proposed method is suitable for large-scale and mass production of underoil superhydrophobic materials. The sandpaper abrasion test is performed to verify the durability of the underoil superhydrophobic surfaces fabricated by the proposed fabrication method. Test results demonstrate that the fabricated surfaces have a more durable water-repellent property in oil than in air. Moreover, based on the underoil superhydrophobicity, the abraded surfaces (meshes) can maintain steady self-cleaning property and oil–water separation. This work can provide helpful guidance for the fabrication of underoil superhydrophobic surfaces, which could be useful for further research of wettability, self-cleaning and oil–water separation in practical applications.",
"introduction": "1. Introduction Superhydrophobic surfaces with a water contact angle larger than 150° and a sliding angle smaller than 10° can be found in many animals and plants in nature. 1,2 Recently, they have attracted much attention owing to their potential application in the fields of self-cleaning, anti-icing, corrosion resistance, oil/water separation and microfluidics. 3–8 Despite the success of the fabrication of superhydrophobic surfaces without any additional chemical modification, 9–11 mechanical wear is usually revealed as increased sticking of water, especially for the hydrophilic metal materials, resulting in loss of the non-wetting properties owing to the exposure of hydrophilic materials. 12,13 In response to this challenge, materials with superhydrophobic properties under oil, which are characterized by high water contact angle and low sliding angle under oil, have been introduced. They have potential applications in many fields, including self-cleaning, anti-fouling and oil/water separation. 14,15 Underoil superhydrophobic materials have the advantage of the oil-favoring properties that allow oil absorption near the material's surface, which minimizes the contact between the solid surface and water in an oil/water/solid three-phase system, so that the substrate shows extreme water-repellent behavior when submerged into the oil. Thus far, there have been fewer studies on the preparation and application of underoil superhydrophobic materials. Tian et al. 16 prepared an underwater superoleophobic and underoil superhydrophobic surface using the combination of re-entrant surface curvatures and delicately matched surface chemistry. They have pointed out that metastable states are necessary to achieve unusual wetting phenomenon. Lu et al. 17 used the coating method to create a robust self-cleaning surface that could function when exposed to oil. Although the self-cleaning and oil–water separation performances of underoil superhydrophobic materials have been verified, 18–20 their advantage of the underoil superhydrophobic property compared with the superhydrophobic property in the air, has not been profoundly studied, and there has not been sufficient basis for selection of underoil superhydrophobic materials. Moreover, the current fabrication methods of underoil superhydrophobic materials are complex and costly. Thus, it is necessary to develop a simple fabrication method of underoil superhydrophobic materials. To address this challenge, this work proposes a simple fabrication method of underoil superhydrophobic brass surfaces with self-cleaning properties in the air and oil based on laser-ablated technology. The proposed method does not require any chemical modification and is suitable for large-scale and mass production. The durability of surfaces has been evaluated using the mechanical abrasion test, and the results show that the fabricated surfaces can remain excellent underoil water-repellent and self-cleaning properties and high separation efficiency even after abrasion. The proposed method opens a new avenue for the development of durable functional materials with self-cleaning and oil–water separation properties.",
"discussion": "3. Results and discussion 3.1 Surface topography and chemical composition The surface morphology of oxide coatings obtained at different powers and intervals of the scan line was characterized by the SEM. The SEM and cross-sectional images of the as-fabricated surfaces at 6 W, 12 W, and 18 W are presented in Fig. 2 . After the laser-ablated process, the three as-fabricated surfaces were rough. The SEM images in Fig. 2(a1–c1) show that the micro-flower-like structures with the size of 10 μm were irregularly distributed on the surface, and the higher power led to the formation of larger microstructures on the surface. The high magnification SEM images show that the nano-villus-like structures could be observed on the micro-flower-like structures with the increase in power, as shown in Fig. 2(a2–c2) . The nano-villus-like structures were formed due to the melting and re-solidification of the materials in the laser-ablated process. The sample's cross-sectional view shown in Fig. 2(a3–c3) indicates that the surface obtained at high power was more textured than that obtained at low power. Fig. 2 SEM and cross-sectional images of as-fabricated surfaces at (a) 6 W, (b) 12 W and (c) 18 W. The XRD and EDS were used to characterize the chemical composition of the brass surfaces fabricated by the laser-ablated technology. Fig. 3a shows the XRD patterns of the as-fabricated surfaces obtained at different powers. The characteristic peaks of the surfaces at 31.17°, 42.22°, 43.42°, 49.32°, 63.27°, 72.27°, 79.72° and 95.52°, which were obtained at different powers were in good agreement with the polished brass surface, indicating that there was no new peak for the surfaces fabricated using the nanosecond laser. The reason for this could be that species formed by the laser-ablated process were amorphous and could not be detected by the XRD. Moreover, element O was observed in the EDS analysis results of the as-fabricated surfaces, as shown in Fig. 3b–d , and its content increased with the power. The results indicated that the oxide coating was successfully formed on the brass substrates. Fig. 3 (a) XRD patterns and (b–d) EDS spectra of the surfaces fabricated with different powers. 3.2 Wettability analysis As shown in Fig. 4a and b , under the power of 12 W, the laser-ablated surface had the superhydrophobicity and oleophilicity in the air with a water contact angle of 153.8 ± 1.5° and an oil contact angle of 13.4 ± 0.5°, respectively. Fig. 4c shows the image of underoil water droplets with an approximately spherical shape on the surface and a water contact angle of 161.3 ± 1.1°, indicating an excellent underoil superhydrophobicity. The reason for this could be that the oil wetted the microstructure due to the oleophilicity of the surface when it was placed in an oil environment. The formed oil film on the microstructure prevented the water droplet from directly contacting the surface, thus ensuring the surface's superhydrophobicity under oil. Moreover, the surface had an ultra-low water adhesion under oil, as shown in Fig. 4d . Fig. 4e illustrates the rolling process of a water droplet under the gravitation effect, where it can be seen that the water droplet could roll off easily from the 1°-tilted surface under oil. The superhydrophobic property of the surface was tested under the different types of oils, including kerosene, soybean oil, silicone oil, dodecane, hexadecane, and chloroform ( Fig. 4f ), and all of the water contact angles of the as-fabricated surface were greater than 150°. The wettability of the brass surfaces obtained with different powers was measured under oil, as shown in Fig. 5 . For the polished brass surface (the power was zero), the water droplet under the oil exhibited weak hydrophobicity with the contact angle of 138.8 ± 2.7°. As the power increased from 3 W to 18 W, the underoil water contact angle of the surface increased slightly from 159.3 ± 1.1° to 162.4 ± 0.9° while the sliding angle decreased, indicating that under the oil, the surface's wettability varied from the hydrophobicity to superhydrophobicity. Fig. 4 (a and b) Water and oil contact angles of the laser-ablated surface in the air. (c) Water contact angle of the surface under oil. (d) Low adhesion behavior of the water droplet to the surface under oil. (e) Rolling process of a water droplet on the 1°-tilted surface under oil. (f) Water contact angle of the surface under the different types of oils. Fig. 5 Underoil water-wettability of the brass surfaces obtained with different powers. 3.3 Mechanical stability analysis of underoil superhydrophobic surface Surface durability is a very important factor, which indicates the possibility of widespread application of special wettability surfaces. 12,23–28 The micro- and nano-hierarchical structures are mechanically weak and can be easily destroyed by a physical impact and friction, resulting in the removal of a substantial amounts of material. 12 We described interaction of the underoil water-repellent surfaces in the abrasion test, and characterized the structure and wettability of the surfaces fabricated under the power of 12 W. During the test, a weight of 100 g was applied on the surfaces with the size of 2 cm × 3.5 cm. After 20 abrasion cycles, the SEM results revealed that the repeated abrasion with the sandpaper led to scratching, and a large amount of material was removed from on the surface ( Fig. 6a ). As shown in Fig. 6b , severe physical damage had a slight effect on the underoil water contact angle, and the contact angle decreased from 161.3 ± 1.1° to 159.5 ± 1.9°. Correspondingly, the underoil water sliding angle changed from 0.9 ± 0.3° to 18.1 ± 2.3°. The results showed that the abraded surfaces maintained an excellent water-repellent property when exposed to the oil as presented in Movie S1. † Although the hierarchical structures could be damaged and the hydrophilic bulk materials could be exposed after the exposure of the surface in the abrasion test, the water-repellent behavior of the surface remained stable to a certain extent, which was due to the oil absorption nature of the bulk materials and the change in the micro- and nano-hierarchical morphology that was sufficient to maintain the underoil water-repellent property, as shown in Fig. 6c and d . However, as shown in Fig. 6e , for the wettability of the abraded surfaces in air, the contact angle and the sliding angle changed significantly by decreasing from 153.8 ± 1.5° to 137.1 ± 4.7° and increasing from 2.6 ± 0.7° to 90°, respectively. The water droplet adhered to the abraded surface without sliding in Wenzel model, 29 as shown in Fig. 6f and Movie S2. † Fig. 6 (a) SEM image of the underoil water-repellent surface after 20 abrasion cycles. (b) Underoil water contact angle and sliding angle of the surface after different abrasion cycles. (c) Exposed hydrophilic bulk materials after the abrasion test. (d) Underoil water-repellent property of the abraded surface in Cassie model. (e) Water contact angle and sliding angle of the surface after different abrasion cycles in air. (f) Water droplet adhered to the abraded surface in Wenzel model. 3.4 Self-cleaning property and its mechanical stability analysis The loss of superwetting properties of interface materials due to the damage of the micro- and nano-hierarchical structures can affect their function and thus can immensely impact their applications. Recently, underoil superhydrophobic surfaces and meshes have been developed for achieving self-cleaning property and oil–water separation. Weak water-repellent property has been one of the main issues as well. We subsequently studied the self-cleaning property of the surfaces fabricated by the laser-ablated process. The underoil superhydrophobic surfaces showed a good self-cleaning property in both air and oil. As presented in Movies S3 and S4, † water droplets could roll off easily from the slightly sloped surface and take away the Fe powders when exposed to both air and oil. After 20 abrasion cycles, the self-cleaning property of the abraded surfaces was demonstrated. As shown in Fig. 7a , when the abraded surface was under oil, the Fe powders on the surface could be easily removed by water droplets because of the excellent water-repellent property of the surface ( Fig. 6c ). The results showed that the surfaces maintained an excellent self-cleaning property when being abraded, even after 20 abrasion cycles of their exposure to the oil. However, when the self-cleaning test was performed in air, Fe powders could not be easily removed by water droplets, as shown in Fig. 7b . The droplet with Fe powders, namely, droplets 1, 2, 6 and 9, easily adhered to the surface. Only when another droplet was merged with the adhered droplet, for instance, when droplets 1 and 3, droplets 6 and 7 were merged, the Fe powders could be rejected by a large merged droplet, namely, droplets 4 and 8, respectively. The reason for this phenomenon was that the abraded surface had the adhesion property for droplets, as shown in Fig. 6f . The results show that the underoil superhydrophobic surfaces had a more durable self-cleaning performance. Fig. 7 Self-cleaning performances of the abraded surface in both (a) oil and (b) air. 3.5 Oil/water separation and mechanical stability of underoil superhydrophobic meshes In addition to the self-cleaning property, the oil–water separation performance of the laser-ablated brass mesh was investigated. In the sandpaper abrasion test, a weight of 100 g was applied to the meshes with the size of 5 cm × 5 cm. The SEM images of the mesh fabricated under a power of 12 W before and after 10 abrasion cycles are presented in Fig. 8a–d . For the abraded mesh, a certain amount of material was removed from the wire of the mesh. However, the abraded mesh maintained an excellent water-repellent property when being exposed to the oil. As shown in Fig. 8e , the contact angle of the abraded mesh decreased from 152.4 ± 1.9° to 151.1 ± 2.8°, and the underoil water sliding angle changed from 1.8 ± 1.6° to 4.9 ± 1.5°. For the wettability of the abraded mesh in air, the contact angle and sliding angle changed significantly, decreasing from 151.9 ± 2.1° to 149.1 ± 3.2° and increasing from 3.7 ± 1.9° to 53.7 ± 5.1°, respectively, as shown in Fig. 8f . The results demonstrated that the abraded mesh lost its excellent superhydrophobic property in the air. Fig. 8 SEM images of the mesh (a and b) before and (c and d) after 10 abrasion cycles. Contact angle ( ) and sliding angle ( ) of the mesh before and after abrasion (e) under oil and (f) in the air. The oil–water separation was performed using homemade devices, driven solely by gravity without applying any external pressure. Before the sandpaper abrasion test, when the mixture of water and chloroform was poured onto the mesh prewetted with chloroform, the water dyed red was accumulated on the mesh, but the chloroform quickly penetrated through the prewetted mesh and felt into the collection container, as shown in Movie S5. † The separation efficiency was 98.1%, which calculated by separation efficiency = liquid collection (g)/liquid (g) × 100%, where the liquid was chloroform. After 10 abrasion cycles, the oil–water separation process of the abraded mesh was demonstrated based on the underoil superhydrophobicity of the mesh, and the separation efficiency was as high as 97.8%, indicating that the abraded mesh maintained an excellent oil–water separation property, as shown in Fig. 9a . Moreover, as expected, the abraded mesh could maintain high separation efficiency even after 10 successive separations based on the underoil superhydrophobicity, and the separation efficiency was above 97%, as shown in Fig. S3. † However, by using the superhydrophobicity in the chloroform–water separation process, the oil–water separation of the abraded mesh failed, as shown in Fig. 9b . Therefore, the mesh based on the prewetting induced underoil superhydrophobicity was more appropriate for the oil–water separation than that based on the superhydrophobicity. Fig. 9 Separation of the mixture of water and chloroform using the abraded mesh based on the (a) underoil superhydrophobicity and (b) superhydrophobicity."
} | 4,377 |
30828497 | PMC6396749 | pmc | 281 | {
"abstract": "It is well understood that heat stress causes bleaching in corals. Much work has focused on the way heat stress disrupts corals’ symbiotic relationship with endosymbiotic algal dinoflagellate, Symbiodiniaceae, a process called bleaching. However, the damage to the coral tissue that occurs during the bleaching process and, importantly, the factors that contribute to subsequent recovery, are not well understood. I hypothesize that the host tissue damage created by heat stress initiates cascades of wound healing factors that maintain epithelial integrity. These factors may be found to contribute to the coral’s potential capacity to recover. In this study, I present evidence that heat stress causes damage to the coral host tissue and that collagen is present in the gastrodermis of heat-stressed corals. I found that, during the early stages of bleaching, an important transcription factor for wound healing, Grainyhead, is expressed throughout the gastrodermis, where the cellular and tissue rearrangements occur. Lastly, using phylogenetics, I found that cnidarian Grainyhead proteins evolved three distinct groups and that evolution of this protein family likely happened within each taxonomic group. These findings have important implications for our study of coral resiliency in the face of climate change.",
"conclusion": "Conclusions In this study, I show that heat stress has a physical effect on the tissue and the cells of the coral A. hyacinthus, possibly triggering wound-healing factors similar to those in other organisms. I demonstrated the physical effects by using histology to show that tissue integrity is compromised and that collagen is expressed throughout the gastrodermis of heat stressed corals. Additionally, these experiments revealed that GRH , a known transcription factor for epithelial integrity in other model organisms, was expressed throughout the gastrodermis of heat stressed corals. In fact, cnidarians possess three distinct groups of GRH proteins, suggesting the role of GRH in corals is potentially significant and worthy of further study. Significant progress has been made in understanding the disruptive effects of bleaching on coral tissue. Increasing our understanding of the heat stress recovery process of coral tissue allows us to identify corals with the potential to be more resilient. This knowledge will help us support the coral reef ecosystem, an environment that is crucially important for facilitating biodiversity, ocean health, and human health.",
"introduction": "Introduction Corals (phylum: Cnidaria) are critical ecosystem builders that are important for promoting marine biodiversity, economic development, and human health ( Hughes et al., 2003 ). Reef building corals consists of polyps that secrete calcium carbonate, over which tissue and colonial polyps form. Within the endodermal epithelium, called the gastrodermis, many coral cells contain symbiotic dinoflagellates called Symbiodinium (recently renamed Symbiodiniaceae ( LaJeunesse et al., 2018 )). Through this critical partnership, the Symbiodiniaceae provides nutrients for the coral host, and in turn Symbiodiniaceae uses the wastes of the coral ( Gates, Baghdasarian & Muscatine, 1992 ; Weis, 2008 ). During disturbance events such as heat stress, corals can “bleach”, disassociating from the Symbiodiniaceae partner, and the coral’s normally brown-pigmented tissue appears white as the skeleton shows through the translucent tissue. The phenomena of bleaching is highly variable, and different levels of bleaching can occur in different species of coral, as well as, under different conditions such as variable light intensities, salinity changes, and temperatures ( Downs et al., 2009a ; Downs et al., 2009b ; Baker & Cunning, 2015 ; Brown & Dunne, 2015 ; Bieri et al., 2016 ). However, the signaling mechanisms leading to heat stress induced bleaching can occur very quickly, with the activation of stress response genes being upregulated within 150 min of heat stress ( Traylor-Knowles et al., 2017 ). This indicates that the mechanisms for promoting bleaching are active well before the bleaching becomes visibly detectable ( Traylor-Knowles et al., 2017 ). The cellular mechanisms that are activated include degradation of the symbiont within the coral host cell, coral host cell apoptosis, coral host cell necrosis, exocytosis of the symbiont from the host cell, and detachment of the host cell with the symbiont still within it ( Gates, Baghdasarian & Muscatine, 1992 ; Weis, 2008 ). In this research article, I am defining a wound as a disruption of the epithelial integrity. I propose that the cellular damage of heat stress is, on a molecular level, akin to an epithelial wound, activating wound-healing pathways that could potentially help the coral recover from the damage. In many previous gene and protein expression studies on heat stress in corals, factors known to be involved in wound healing have been identified, including many different collagens ( Desalvo et al., 2008 ; Moya et al., 2012 ; Barshis et al., 2013 ; Bay et al., 2013 ; Kenkel et al., 2013 ; Maor-Landaw et al., 2014 ; Seneca & Palumbi, 2015 ; Rose, Seneca & Palumbi, 2015 ; Ricaurte et al., 2016 ). Collagen production is a hallmark of wound healing, and, in many organisms, is typically increased at the site of a wound during the late wound-healing phase ( Diegelmann & Evans, 2004 ; Deonarine et al., 2007 ). In reaction to heat stress, both increases and decreases in gene expression of collagen were found ( Table 1 ). For example, in the coral Montastraea faveolata , expression of collagen precursors was downregulated in response to heat stress, but in the coral Acropora hyacinthus , collagen expression was upregulated ( Barshis et al., 2013 ; Desalvo et al., 2008 ). This variation in expression could be due to the different species of corals, the different types of collagens (there are more than 30 in the Acropora digitifera genome alone), and the different types of heat stress exposures that were performed in each study. Despite this variation, it is evident from these previous studies that collagens are reacting to heat stress in corals. Based on this observation, I hypothesize that heat stress creates damage to the host tissue, which in turn initiates cascades of wound healing factors that maintain epithelial integrity in response to the heat damage. One such cascade involves the Grainyhead transcription factor pathway. 10.7717/peerj.6510/table-1 Table 1 Collagen gene and protein expression patterns from previous coral and anemone heat stress studies. Collagen type Experiment Organism Citation ↓ procollagen type I, alpha 2 Heat stress and bleaching, microarray Montastraea faveolata Desalvo et al. (2008) ↑ collagen type IV Chronic heat stress, qPCR transcriptome Porites astreoides Kenkel et al. (2013) ↑ collagen alpha-1(I) chain, ↑ mini collagens Lab heat stress, transcriptome Acropora hyacinthus Barshis et al. (2013) ↓ collagen Heat stress, microarray Stylophora pistillata Maor-Landaw et al. (2014) ↓ collagen Heat stress, transcriptome Acropora hyacinthus Seneca & Palumbi (2015) ↑ collagen Temperature acclimation, qPCR Acropora millepora Bay et al. (2013) ↓ collagen Bleached versus unbleached, proteomics Acropora palmata Ricaurte et al. (2016) ↑ collagen Transcriptional module heavily weighted for collagens, negatively correlated with bleaching outcomes Acropora hyacinthus Rose, Seneca & Palumbi (2015) ↓ Collagen alpha-1, ↓ Collagen alpha-2, ↓ Collagen-like Heat and UV stress, microarray Anemonia viridis Moya et al. (2012) Grainyhead (GRH) is a transcription factor which functions as a critical regulator of genes, including transglutaminase, dopa decarboxylase, and others that are crucial for tissue remodeling as has been shown in studies involving mice, Xenopus , and Drosophila ( Mace, Pearson & McGinnis, 2005 ; Harden, 2005 ; Ting et al., 2005 ). In mice, GRH is involved in the development and maintenance of epithelial integrity ( Harden, 2005 ; Ting et al., 2005 ). The mouse GRH is required during embryogenesis, where it is expressed exclusively in the developing ectodermal epithelium ( Ting et al., 2005 ). Additionally, mouse GRH-like 2 is necessary for the expression of important adheren and tight junction genes in many different types of epithelia including the surface ectoderm and gut endoderm ( Werth et al., 2010 ). Likewise, in Xenopus , XGRH1 has been implicated in the development of the epidermis ( Tao et al., 2005 ). In morpholino studies, knockdown of XGRH1 led to loss of surface structures and pigmentation as well as neck and eye defects associated with epidermal instability ( Tao et al., 2005 ). In Drosophila, GRH maintains the tension of the cuticle and induces cuticle development and repair following injury ( Mace, Pearson & McGinnis, 2005 ; Moussian & Uv, 2005 ). In cnidarians, GRH has been was bioinformatically characterized in Nematostella vectensis ( Traylor-Knowles et al., 2010 ). However, little is understood about the phylogenetic relationship of GRH among cnidarians, and even less is understood about its function. In this study, I conducted a series of experiments to determine whether heat stress causes an epithelial disruption similar to a wound, thus activating wound-healing pathways. To test if coral epithelial integrity was compromised during a short-term heat stress, I used histology to identify cellular architectural changes. I then used in situ hybridization to examine spatial expression of GRH and found that GRH is expressed in the gastrodermis of the coral tissue, the same epithelium where bleaching occurs. I also examined the phylogenetics of cnidarian GRH and discovered that within the cnidarians tested, GRH had three distinct clades that are primarily driven by taxonomic grouping. This diversity of GRH within cnidarians could indicate a wider repertoire of functional significance.",
"discussion": "Results and Discussion Coral tissue and cell integrity are damaged by heat In this study I found evidence for heat stress causing a cellular pathology of necrosis and degradation, and the initiation of wound healing factors including collagen production ( Figs. 1B , 2 and 3 ). The samples examined in this study had an average visual bleaching score of 2.3 out of 5 ( Seneca & Palumbi, 2015 ; Rose, Seneca & Palumbi, 2015 ), indicating that colonies were partially bleached ( Fig. S1 ). Slides of heat stressed coral stained with H & E had degradation and atrophy of the gastrodermis and the epidermis, along with shrunk and/or necrotic Symbiodiniaceae. Overall cell staining was very light, and cell walls were disrupted, damaged, and misshapen ( Fig. 1B ). Heat stress causes Symbiodiniaceae to produce large amounts of reactive oxygen species (ROS), which can overwhelm the cellular environment, causing Symbiodiniaceae to be released or degraded by many different cellular mechanisms ( Nielsen, Petrou & Gates, 2018 ). Previously, in the coral Acropora aspera , cellular aspects of bleaching including apoptosis were observed in corals exposed to heat stress early in the bleaching response, indicating that the coral host was reacting to the heat stress long before the bleaching event occurred ( Ainsworth et al., 2008 ). These cellular mechanisms cause a breakdown of the gastrodermis, which can leave a coral more vulnerable to pathogen invasion ( Mydlarz et al., 2008 ; Palmer, Bythell & Willis, 2010 ). 10.7717/peerj.6510/fig-1 Figure 1 Histological serial cross sections of heat-stressed and control coral tissues, stained with H & E in the coral A. hyacinthus . (A) Sagittal cross section through control coral tissue, not exposed to heat-stress conditions. In the controls, tissue epithelia and mesoglea are intact with normal architecture and strong staining. Symbiodiniaceae are present within coral gastrodermal cells, and appear to be healthy and undamaged. Spirocytes are present and are not discharged or released from the epidermis. (B) Sagittal cross section through heat-stressed coral tissue. In the heat-stressed samples, epithelia and mesoglea are damaged. Few Symbiodiniaceae are present, with the exception of some that are necrotic. Swollen mucocytes are also present. Tissue staining is not as strong as compared with the control, indicating loss of proteins, vacuolation, and pycnosis of nuclei. Abbreviations: SY, Symbiodiniaceae ; SP, Spirocyte; SU, symbiont containing gastrodermal cell; MU, mucocytes. 10.7717/peerj.6510/fig-2 Figure 2 Masson’s trichrome-stained sections for collagen expression in control coral samples. Control samples along the edge of a gastrovascular cavity shows intact tissue layers, and clear staining. Collagen (blue) is present within the mesoglea, and intact Symbiodiniaceae are present in gastrodermal cells. 10.7717/peerj.6510/fig-3 Figure 3 Masson’s trichrome-stained sections for collagen expression in heat-stressed coral samples. Heat-stressed samples have more diffuse collagen present (blue stain) through the gastrodermis. Symbiodiniaceae are showing signs of necrosis or are not present. Tissue structure is more damaged than what was observed in the control. Black arrows denote areas where collagen staining is expanded through the tissue area. Collagen as a measurement of damage after a heat stress event The common theme between previous transcriptome and microarray studies on heat stress in corals is that collagen, no matter what type, is reacting to heat stress. However, in corals the spatial location of where collagen is expressed after heat stress is not well understood. To address this, I used Masson’s trichrome to stain for collagen in heat stressed and control coral samples. In the control samples, collagen was present as part of the mesoglea ( Fig. 2 ). However, in the heat stressed samples, areas of the tissue beyond the mesoglea had collagens present, including the epithelia ( Figs. 2 and 3 ). The gastrodermis, which houses the Symbiodiniaceae, had collagen staining present as well ( Fig. 3 ). Early production and deposition of collagen after an acute heat stress event may be an important survival trait for corals that are subjected to high temperatures. The broad staining of collagen fibers in heat stress samples indicates that tissue rearrangement was occurring in response to heat stress. This is particularly surprising given that in the previous transcriptomic study on these same coral genotypes, collagen was down regulated in response to the same heat stress protocol ( Seneca & Palumbi, 2015 ). This difference may be due to the type of collagen that was transcriptionally active in ( Seneca & Palumbi, 2015 ), as well as, the types of collagens that are stained with Masson’s trichrome. This evidence indicates that the cellular damage caused by heat stress happens very quickly and early. Based on this observation, I hypothesize that collagen deposition may be an important mechanism that protects the coral in high temperature. In the future, it will be important to measure the rate and abundance of collagen that is deposited when a coral is recovering from a heat stress event, as this process could be an important recovery mechanism. Cnidarian grainyhead is found in three distinct clades, and is expressed in the gastrodermis of heat-stressed corals Of the invertebrates that have previously been investigated only cnidarians have been found to have more than one GRH present in their genome ( Traylor-Knowles et al., 2010 ). With the exception of Hydra , there has been an expansion of the GRH proteins within cnidarian linages, where 2–4 different paralogs of this protein are found ( Fig. 3 ). The functional significance of this diversification is not understood, but it is possible that the function of this protein family is beyond wound healing. In mammals, GRH possesses many alternative splice sites, which enables GRH to produce many different types of protein products, thus increasing its overall ability to affect downstream wound healing genes ( Miles, Dworkin & Darido, 2017 ). In this study we show that anthozoans have several different paralogs of GRH, which could possess many different alternative splice sites, potentially expanding their functional roles. Future studies on the alternative splices sites within cnidarian GRH may present interesting insights into the wide breadth of possible functional roles within cnidarians. To determine the evolutionary relationship of different cnidarian GRH proteins, a Neighbor-Joining phylogenetic analysis was used. The analysis involved 22 amino acid sequences and a total of 44 amino acid positions were analyzed in the final dataset ( Fig. 4 ). I found that the different cnidarian GRH paralogs cluster into three different clades. Within clade 1, the GRHs concentrated into subclades according to taxa. Scleractinia is in one subclade, Actinaria in another, and Hydra was the most derived. Within clade 2 a similar pattern was observed, where Scleractinia and Actinaria formed separate subclades. Within group 3, resolution of Scleractinia and Actinaria was not as clear, with the protein sequence GRH A. fenenstrafer_23.121, being the most derived within the subclade. With the exception of group 3, the placement of the different GRH proteins indicates that the sequence evolution of this protein likely occurred within each taxonomic group, and may coincide with the diversification of specific traits of that taxonomic group. 10.7717/peerj.6510/fig-4 Figure 4 Neighbor-Joining tree of the evolutionary relatedness of Grainyhead proteins in cnidarians. The evolutionary history was inferred using the Neighbor-Joining method ( Saitou & Nei, 1987 ). The optimal tree with the sum of branch length = 5.13670293 is shown. The percentage of replicate trees in which the associated taxa clustered together in the bootstrap test (2,000 replicates) is shown next to the branches ( Felsenstein, 1985 ). The tree is drawn to scale, with branch lengths in the same units as those of the evolutionary distances used to infer the phylogenetic tree. The evolutionary distances were computed using the Poisson correction method ( Zuckerkandl & Pauling, 1965 ) and are in the units of the number of amino acid substitutions per site. The analysis involved 22 amino acid sequences. All positions with less than 95% site coverage were eliminated. That is, fewer than 5% alignment gaps, missing data, and ambiguous bases were allowed at any position. There were a total of 44 positions in the final dataset. Evolutionary analyses were conducted in MEGA7 ( Kumar, Stecher & Tamura, 2016 ). Scleractinia is in red italic writing, while all other cnidarians are in black bold face. Sequences used in this study are found in Table S2 . Due to GRH’s conserved evolutionary role in maintaining the integrity of the epithelium and that several paralogs for this gene are found in corals, I next examined whether this gene was expressed in heat stressed coral tissue. GRH is a master transcription factor for wound healing and targets the activation of specific genes important to reestablishing the epithelium after a wound event ( Mace, Pearson & McGinnis, 2005 ; Harden, 2005 ; Ting et al., 2005 ; Traylor-Knowles et al., 2010 ; Wang & Samakovlis, 2012 ). During wound healing, GRH is expressed in surface-lining epithelia of Drosophila and mice, and the cuticle secreted by the epithelium in Drosophila ( Mace, Pearson & McGinnis, 2005 ; Ting et al., 2005 ). In cnidarians, little is understood about the role of GRH in epithelial integrity, but it is hypothesized that its role could be similar to what has been documented in Drosophila and mice wound healing ( Mace, Pearson & McGinnis, 2005 ; Harden, 2005 ; Ting et al., 2005 ). After heat stress, GRH is expressed throughout the gastrodermis of the coral ( Figs. 5A and 5B ). The expression is specific to the mesoglea adjacent to the gastrodermal cells, as well as, the cytoplasm of gastrodermal cells, which surround the Symbiodiniaceae ( Fig. 5B ). Staining was not found in the epidermis or within the nematocytes and spirocytes. In other organisms, the expression of GRH during wound healing is generally in the epidermis ( Harden, 2005 ; Mace, Pearson & McGinnis, 2005 ; Tao et al., 2005 ; Ting et al., 2005 ; Wang & Samakovlis, 2012 ). This difference in expression may be due to the type of tissue damage, as well as the complexity of the tissue. For example, most studies on wound healing in Dropsophila and mice have been on subcutaneous, mechanical wounds, rather than wounds that were caused by heat ( Harden, 2005 ; Mace, Pearson & McGinnis, 2005 ; Tao et al., 2005 ; Ting et al., 2005 ; Wang & Samakovlis, 2012 ). 10.7717/peerj.6510/fig-5 Figure 5 Grainyhead spatial expression in response to heat stress in corals. The following samples were done on serial sections from the same coral colonies. (A) The GRH sense control probe had no staining present. (B) Staining for the GRH anti-sense probe was found throughout the mesoglea and gastrodermis. Expression was not present within the spirocytes, but was found surrounding the Symbiodiniaceae , and the extracellular matrix of the gastrodermis layer. Black arrows point to purple staining of GRH antisense probe. Abbreviations: SY, Symbiodiniaceae ; SP, Spirocyte"
} | 5,341 |
31798400 | PMC6868054 | pmc | 283 | {
"abstract": "The massively parallel nature of biological information processing plays an important role due to its superiority in comparison to human-engineered computing devices. In particular, it may hold the key to overcoming the von Neumann bottleneck that limits contemporary computer architectures. Physical-model neuromorphic devices seek to replicate not only this inherent parallelism, but also aspects of its microscopic dynamics in analog circuits emulating neurons and synapses. However, these machines require network models that are not only adept at solving particular tasks, but that can also cope with the inherent imperfections of analog substrates. We present a spiking network model that performs Bayesian inference through sampling on the BrainScaleS neuromorphic platform, where we use it for generative and discriminative computations on visual data. By illustrating its functionality on this platform, we implicitly demonstrate its robustness to various substrate-specific distortive effects, as well as its accelerated capability for computation. These results showcase the advantages of brain-inspired physical computation and provide important building blocks for large-scale neuromorphic applications.",
"conclusion": "4.3. Conclusion In this work we showed how sampling-based Bayesian inference using hierarchical spiking networks can be robustly implemented on a physical model system despite inherent variability and imperfections. Underlying neuron and synapse dynamics are deterministic and close to their biological archetypes, but with much shorter time constants, hence the intrinsic acceleration factor of 10 4 with respect to biology. The entire architecture—sampling network plus background random network—was fully deterministic and entirely contained on the neuromorphic substrate, with external communication used only to represent input patterns and labels. Considering the deterministic nature of neurons in vitro (Mainen and Sejnowski, 1995 ; Reinagel and Reid, 2002 ; Toups et al., 2012 ), such an architecture also represents a plausible model for neural sampling in cortex (Jordan et al., 2017 ; Dold et al., 2019 ). We demonstrated sampling from arbitrary Boltzmann distributions over binary random variables, as well as generative and discriminative properties of networks trained with visual data. The framework can be extended to sampling from arbitrary probability distributions over binary random variables, as it was shown in software simulations (Probst et al., 2015 ). For such networks, the two abovementioned computational tasks (pattern completion and classification) happen simultaneously, as they both require the calculation of conditional distributions, which is carried out implicitly by the network dynamics. Both during learning and for the subsequent inference tasks, the setup benefitted significantly from the fast-intrinsic dynamics of the substrate, achieving a net speedup of 100–210 compared to biology. We view these results as a contribution to the nascent but expanding field of applications for biologically inspired physical-model systems. They demonstrate the feasibility of such devices to solve problems in machine learning, as well as studying biological phenomena. Importantly, they explicitly address the search for robust computational models that are able to harness the strengths of these systems, most importantly their speed and energy efficiency. The proposed architecture scales naturally to substrates with more neuronal real-estate and can be used for a wide array of tasks that can be mapped to a Bayesian formulation, such as constraint satisfaction problems (Jonke et al., 2016 ; Fonseca Guerra and Furber, 2017 ), prediction of temporal sequences (Sutskever and Hinton, 2007 ), movement planning (Taylor and Hinton, 2009 ; Alemi et al., 2015 ), simulation of solid-state systems (Edwards and Anderson, 1975 ), and quantum many-body problems (Carleo and Troyer, 2017 ; Czischek et al., 2018 ).",
"introduction": "1. Introduction The aggressive pursuit of Moore's law in conventional computing architectures is slowly but surely nearing its end (Waldrop, 2016 ), with difficult-to-overcome physical effects, such as heat production and quantum uncertainty, representing the main limiting factors. The so-called von Neumann bottleneck between processing and memory units represents the main cause, as it effectively limits the speed of these largely serial computation devices. The most promising solutions come in the form of massively parallel devices, many of which are based on brain-inspired computing paradigms (Indiveri et al., 2011 ; Furber, 2016 ), each with its own advantages and drawbacks. Among the various approaches to such neuromorphic computing, one class of devices is dedicated to the physical emulation of cortical circuits; not only do they instantiate neurons and synapses that operate in parallel and independently of each other, but these units are actually represented by distinct circuits that emulate the dynamics of their biological archetypes (Mead, 1990 ; Indiveri et al., 2006 ; Jo et al., 2010 ; Schemmel et al., 2010 ; Pfeil et al., 2013 ; Qiao et al., 2015 ; Chang et al., 2016 ; Wunderlich et al., 2019 ). Some important advantages of this approach lie in their reduced power consumption and enhanced speed compared to conventional simulations of biological neuronal networks, which represent direct payoffs of replacing the resource-intensive numerical calculation of neuro-synaptic dynamics with the physics of the devices themselves. However, such computation with analog dynamics, without the convenience of binarization, as used in digital devices, has a downside of its own: variability in the manufacturing process (fixed pattern noise) and temporal noise both lead to reduced controllability of the circuit dynamics. Additionally, one relinquishes much of the freedom permitted by conventional algorithms and simulations, as one is confined by the dynamics and parameter ranges cast into the silicon substrate. The main challenge of exploiting these systems, therefore, lies in designing performance network models using the available components while maintaining a degree of robustness toward the substrate-induced distortions. Just like for the devices themselves, inspiration for such models often comes from neuroscience, as the brain needs to meet similar demands. With accumulating experimental evidence (Berkes et al., 2011 ; Pouget et al., 2013 ; Haefner et al., 2016 ; Orbán et al., 2016 ), the view of the brain itself as an analytical computation device has shifted. The stochastic nature of neural activity in vivo is being increasingly regarded as an explicit computational resource rather than a nuisance that needs to be dealt with by sophisticated error-correcting mechanisms or by averaging over populations. Under the assumption that stochastic brain dynamics reflect an ongoing process of Bayesian inference in continuous time, the output variability of single neurons can be interpreted as a representation of uncertainty. Theories of neural sampling (Buesing et al., 2011 ; Hennequin et al., 2014 ; Aitchison and Lengyel, 2016 ; Petrovici et al., 2016 ; Kutschireiter et al., 2017 ) provide an analytical framework for embedding this type of computation in spiking neural networks. In this paper we describe the realization of neural sampling with networks of leaky integrate-and-fire neurons (Petrovici et al., 2016 ) on the BrainScaleS accelerated neuromorphic platform (Schemmel et al., 2010 ). With appropriate training, the variability of the analog components can be naturally compensated and incorporated into a functional network structure, while the network's ongoing dynamics make explicit use of the analog substrate's intrinsic acceleration for Bayesian inference (section 2.3). We demonstrate sampling from low-dimensional target probability distributions with randomly chosen parameters (section 3.1) as well as inference in high-dimensional spaces constrained by real-world data, by solving associated classification and constraint satisfaction problems (pattern completion, section 3.2). All network components are fully contained on the neuromorphic substrate, with external inputs only used for sensory evidence (visual data). Our work thereby contributes to the search for novel paradigms of information processing that can directly benefit from the features of neuro-inspired physical model systems.",
"discussion": "4. Discussion This article presents the first scalable demonstration of sampling-based probabilistic inference with spiking networks on a highly accelerated analog neuromorphic substrate. We trained fully connected spiking networks to sample from target distributions and hierarchical spiking networks as discriminative and generative models of higher-dimensional input data. Despite the inherent variability of the analog substrate, we were able to achieve performance levels comparable to those of software simulations in several benchmark tasks, while maintaining a significant overall acceleration factor compared to systems that operate in biological real time. Importantly, by co-embedding the generation of stochasticity within the same substrate, we demonstrated the viability of a fully embedded neural sampling model with significantly reduced demands on off-substrate I/O bandwidth. Having a fully embedded implementation allows the runtime of the experiments to scale as O (1) with the size of the emulated network; this is inherent to the nature of physical emulation, for which wall-clock runtime only depends on the emulated time in the biological reference frame. In the following sections, we address the limitations of our study, point out links to related work and discuss its implications within the greater context of computational neuroscience and bio-inspired AI. 4.1. Limitations and Constraints The most notable limitation imposed by the current commissioning state of the BrainScaleS system was on the size of the emulated SSNs. At the time of writing, due to limited software flexibility, system assembly and substrate yield, the usable hardware real-estate was reduced to a patchy and non-contiguous area, thereby strongly limiting the maximum connectivity between different locations within this area. In order to limit synapse loss to small values (below 2 %), we restricted ourselves to using a small but contiguous functioning area of the wafer, which in turn limited the maximum size of our SSNs and noise-generating RNs. Ongoing improvements in post-production and assembly, as well as in the mapping and routing software, are expected to enhance on-wafer connectivity and thereby automatically increase the size of emulable networks, as the architecture of our SSNs scales naturally to such an increase in hardware resources. To a lesser extent, the sampling accuracy was also affected by the limited precision of hardware parameter control. The writing of analog parameters exhibits significant trial-to-trial variability; in any given trial, this leads to a heterogeneous substrate, which is known to reduce the sampling accuracy (Probst et al., 2015 ). Most of this variability is compensated during learning, but the 4 bit resolution of the synaptic weights and the imperfect symmetry in the effective weight matrix due to analog variability of the synaptic circuits ultimately limit the ability of the SSN to approximate target distributions. This leads to the “jumping” behavior of the D KL ( p ∥ p * ) in the final stages of learning ( Figure 4A ). In smaller networks, synaptic weight resolution is a critical performance modifier (Petrovici et al., 2017b ). However, the penalty imposed by a limited synaptic weight resolution is known to decrease for larger deep networks with more and larger hidden layers, both spiking and non-spiking (Courbariaux et al., 2015 ; Petrovici et al., 2017a ). Furthermore, the successor system (BrainScaleS-2, Aamir et al., 2016 ) is designed with a 6-bit weight resolution. In the setup we used shared bias neurons for several neurons in the sampling network. This helped us save hardware resources, thus allowing the emulation of larger functional networks. Such bias neuron sharing is expected to introduce some small amount of temporal correlations between the sampling neurons. However, this effect was too small to observe in our experiments for several reasons. First, the high firing rate of the bias neurons helped smooth out the bias voltage induced into the sampling neurons. Second, the different delays and spike timing jitter on the hardware reduces such cross-correlations. Third, other dominant limitations overshadow the effect of shared bias neurons. In any case, the used training procedure inherently compensates for excess cross-correlations, thus effectively removing any distortions to the target distribution that this effect might introduce (Bytschok et al., 2017 ; Dold et al., 2019 ). In the current setup, our SSNs displayed limited mixing abilities. During guided dreaming, images from one of the learned classes were more difficult to generate ( Figure 7 ). Restricted mixing due to deep modes in the energy landscape carved out by contrastive learning is a well-known problem for classical Boltzmann machines, which is usually alleviated by computationally costly annealing techniques (Desjardins et al., 2010 ; Salakhutdinov, 2010 ; Bengio et al., 2013 ). However, the fully-commissioned BrainScaleS system will feature embedded short-term synaptic plasticity (Schemmel et al., 2010 ), which has been shown to promote mixing in spiking networks (Leng et al., 2018 ) while operating purely locally, at the level of individual synapses. Currently, the execution speed of emulation runs is dominated by the I/O overhead, which in turn is mostly spent on setting up the experiment. This leads to the classification (section 3.2) of one image taking 2.4 to 3.9 ms, whereas the pure network runtime is merely 50 μs. A streamlining of the software layer that performs this setup is expected to significantly reduce this discrepancy. The synaptic learning rule was local and Hebbian, but updates were calculated on a host computer using an iterative in-the-loop training procedure, which required repeated stopping, evaluation and restart of the emulation, thereby reducing the nominal acceleration factor of 10 4 by two orders of magnitude. By utilizing on-chip plasticity, as available, for example, on the BrainScaleS-2 successor system (Friedmann et al., 2017 ; Wunderlich et al., 2019 ), this laborious procedure becomes obsolete and the accelerated nature of the substrate can be exploited to its fullest extent. 4.2. Relation to Other Work This study builds upon a series of theoretical and experimental studies of sampling-based probabilistic inference using the dynamics of biological neurons. The inclusion of refractory times was first considered in Buesing et al. ( 2011 ). An extension to networks of leaky integrate-and-fire neurons and a theoretical framework for their dynamics and statistics followed in Petrovici et al. ( 2013 ) and Petrovici et al. ( 2016 ). The compensation of shared-input correlations through inhibitory feedback and learning was discussed in Bytschok et al. ( 2017 ), Jordan et al. ( 2017 ), and Dold et al. ( 2019 ), inspired by the early study of asynchronous irregular firing in Brunel ( 2000 ) and by preceding correlation studies in theoretical (Tetzlaff et al., 2012 ) and experimental (Pfeil et al., 2016 ) work. Previous small-scale studies of sampling on accelerated mixed-signal neuromorphic hardware include (Petrovici et al., 2015 , 2017a , b ). An implementation of sampling with spiking neurons and its application to the MNIST dataset was shown in Pedroni et al. ( 2016 ) using the fully digital, real-time TrueNorth neuromorphic chip (Merolla et al., 2014 ). We stress two important differences between (Pedroni et al., 2016 ) and this work. First, the nature of the neuromorphic substrate: the TrueNorth system is fully digital and calculates neuronal state updates numerically, in contrast to the physical-model paradigm instantiated by BrainScaleS. In this sense, TrueNorth emulations are significantly closer to classical computer simulations on parallel machines: updates of dynamical variables are precise and robustness to variability is not an issue; however TrueNorth typically runs in biological real time (Merolla et al., 2014 ; Akopyan et al., 2015 ), which is 10,000 times slower than BrainScaleS. Second, the nature of neuron dynamics: the neuron model used in (Pedroni et al., 2016 ) is an intrinsically stochastic unit that sums its weighted inputs, thus remaining very close to classical Gibbs sampling and Boltzmann machines, while our approach considers multiple additional aspects of its biological archetype (exponential synaptic kernels, leaky membranes, deterministic firing, stochasticity through synaptic background, shared-input correlations etc.). Moreover, our approach uses fewer hardware neuron units to represent a sampling unit, enabling a more parsimonious utilization of the neuromorphic substrate. 4.3. Conclusion In this work we showed how sampling-based Bayesian inference using hierarchical spiking networks can be robustly implemented on a physical model system despite inherent variability and imperfections. Underlying neuron and synapse dynamics are deterministic and close to their biological archetypes, but with much shorter time constants, hence the intrinsic acceleration factor of 10 4 with respect to biology. The entire architecture—sampling network plus background random network—was fully deterministic and entirely contained on the neuromorphic substrate, with external communication used only to represent input patterns and labels. Considering the deterministic nature of neurons in vitro (Mainen and Sejnowski, 1995 ; Reinagel and Reid, 2002 ; Toups et al., 2012 ), such an architecture also represents a plausible model for neural sampling in cortex (Jordan et al., 2017 ; Dold et al., 2019 ). We demonstrated sampling from arbitrary Boltzmann distributions over binary random variables, as well as generative and discriminative properties of networks trained with visual data. The framework can be extended to sampling from arbitrary probability distributions over binary random variables, as it was shown in software simulations (Probst et al., 2015 ). For such networks, the two abovementioned computational tasks (pattern completion and classification) happen simultaneously, as they both require the calculation of conditional distributions, which is carried out implicitly by the network dynamics. Both during learning and for the subsequent inference tasks, the setup benefitted significantly from the fast-intrinsic dynamics of the substrate, achieving a net speedup of 100–210 compared to biology. We view these results as a contribution to the nascent but expanding field of applications for biologically inspired physical-model systems. They demonstrate the feasibility of such devices to solve problems in machine learning, as well as studying biological phenomena. Importantly, they explicitly address the search for robust computational models that are able to harness the strengths of these systems, most importantly their speed and energy efficiency. The proposed architecture scales naturally to substrates with more neuronal real-estate and can be used for a wide array of tasks that can be mapped to a Bayesian formulation, such as constraint satisfaction problems (Jonke et al., 2016 ; Fonseca Guerra and Furber, 2017 ), prediction of temporal sequences (Sutskever and Hinton, 2007 ), movement planning (Taylor and Hinton, 2009 ; Alemi et al., 2015 ), simulation of solid-state systems (Edwards and Anderson, 1975 ), and quantum many-body problems (Carleo and Troyer, 2017 ; Czischek et al., 2018 )."
} | 4,986 |
26925210 | PMC4763989 | pmc | 284 | {
"abstract": "Bio-orthogonal non-canonical amino acid tagging allows time-resolved proteomic analysis of quorum sensing in Vibrio harveyi .",
"introduction": "Introduction Bacteria assess their cell numbers and the species complexity of the community of neighboring cells using a chemical communication process called quorum sensing. Quorum sensing relies on the production, release, accumulation and group-wide detection of signal molecules called autoinducers. Quorum sensing controls genes underpinning collective behaviors including bioluminescence, secretion of virulence factors, and biofilm formation. 1 – 3 The model quorum-sensing bacterium Vibrio harveyi integrates population-density information encoded in three autoinducers AI-1, CAI-1, and AI-2, which function as intraspecies, intragenus, and interspecies communication signals, respectively. 4 – 6 \n V. harveyi detects the three autoinducers using the cognate membrane-bound receptors LuxN, CqsS, and LuxPQ, respectively. 7 – 9 At low cell density (LCD), autoinducer concentrations are low, and the unliganded receptors act as kinases, funneling phosphate to the phosphorelay protein LuxU. 10 LuxU transfers the phosphoryl group to the response regulator protein LuxO, which activates transcription of genes encoding five homologous quorum regulatory small RNAs ( qrr sRNAs). 11 , 12 The Qrr sRNAs post-transcriptionally activate production of the transcription factor AphA and repress production of the transcription factor LuxR. AphA and LuxR are the two master quorum-sensing regulators that promote global changes in gene expression in response to population density changes. 12 – 15 At high cell density (HCD), autoinducer binding to the cognate receptors switches the receptors from kinases to phosphatases, removing phosphate from LuxU and, indirectly, from LuxO. Dephosphorylated LuxO is inactive so transcription of the qrr sRNA genes ceases. This event results in production of LuxR and repression of AphA. 12 Thus, the circuitry ensures that AphA is made at LCD, and it controls the regulon required for life as an individual, whereas LuxR is made at HCD, and it directs the program underpinning collective behaviors. Previous microarray studies examined the transcriptomic response during quorum-sensing transitions. That work showed that AphA and LuxR control over 150 and 600 genes, respectively and ∼70 of these genes are regulated by both transcription factors. 15 Both AphA and LuxR act as activators and as repressors, and thus the precise pattern of quorum-sensing target gene expression is exquisitely sensitive to fluctuating levels of AphA and LuxR as cells transition between LCD and HCD modes. Developing a comparable understanding of the quorum-sensing-controlled proteome requires measurement of dynamic changes in protein abundance throughout the transition from individual to collective behavior. In this work, we used the bio-orthogonal non-canonical amino acid tagging (BONCAT) method to track the proteome-wide quorum-sensing response in V. harveyi with temporal precision. BONCAT enabled us to identify 176 proteins that are regulated during the transition from individual to collective behavior; 90 of these proteins are in addition to those identified in earlier studies. We show that a broad range of protein functional groups, including those involved in metabolism, transport, and virulence, change during the transition to group behavior. We demonstrate how particular temporal patterns of protein production are linked to particular tiers of the regulatory cascade by comparing the proteomic profiles of the regulon controlled by the post-transcriptional Qrr sRNAs to the regulon controlled by the transcriptional regulator LuxR. Using this approach, we, for example, determined that the V. harveyi type VI secretion system is LuxR-regulated.",
"discussion": "Discussion and conclusions Global transcriptomic studies of V. harveyi have uncovered a continuum of changes in gene expression during the transition from LCD to HCD. As V. harveyi responds to changes in concentrations of autoinducers, shifts in the levels of the regulatory components AphA, LuxR, and the Qrr sRNAs occur, which in turn alter the expression of the downstream genes in the quorum-sensing regulon. Here we used the BONCAT method to measure changes in the quorum-sensing-regulated proteome during the transition from LCD to HCD, with a time-resolution of 10 min. We found correlated changes in production of the LuxCDABE enzymes and in the intensity of bioluminescence produced by the culture, and we observed regulation of the core regulatory components AphA, LuxR, and LuxO. Notably, the increase in LuxO upon induction of quorum sensing occurred at the level of the protein, but not the mRNA, consistent with the hypothesis that the luxO mRNA is regulated by sequestration by the Qrr sRNAs. 25 \n The time resolution of the BONCAT method allowed us to identify proteins whose rates of synthesis were altered during the early, intermediate, and late stages of the LCD to HCD transition. The proteins found to be regulated within the first 20 min of autoinducer treatment included seven of the 20 known Qrr sRNA targets along with 19 other proteins not previously associated with Qrr regulation. No known Qrr targets were regulated at later times. In contrast, changes in the known LuxR targets occurred between 30 and 90 min following induction. Notably, proteins in the TSSS were up-regulated between 40 and 50 min following autoinducer treatment, suggesting LuxR regulation of type VI secretion in V. harveyi ; this conclusion was confirmed by electrophoretic mobility shift assays. Several LuxR-regulated genes exhibited changes in protein production only very late in the BONCAT experiment, which suggests either that they are responsive to accumulating LuxR levels, that they are regulated by another transcription factor downstream of LuxR, or that they are co-regulated by other factors. We found quorum-sensing-dependent changes in 176 proteins that span a broad range of functional groups, including those related to iron homeostasis, molecular transport, metabolism, and chemotaxis. Ninety of these proteins are newly associated with quorum sensing in V. harveyi , and expand what is known about the roles that quorum sensing plays in these processes. 13 , 32 The remaining 86 proteins are members of the previously established quorum-sensing, AphA, and/or LuxR regulons. Interestingly, nearly 200 other proteins from these regulons were identified by BONCAT but were not significantly up- or down-regulated. For example, the quorum-sensing regulon, which was defined by differences in gene expression between a mutant V. harveyi strain locked at LCD and a strain locked at HCD, contains 365 regulated genes as determined by microarray analysis. 15 We quantified protein expression levels of 127 (35%) of these genes, 45 (35%) of which were significantly regulated. The differences between the genetic and proteomic results may arise, at least in part, from differences in regulation at the levels of mRNA and protein, or from differences in the growth media used in the two experiments (rich (LM) medium in the genetic study vs. minimal (AB) medium here). 13 , 15 Furthermore, we would not expect the rapid addition of saturating amounts of AI-1 to a V. harveyi culture to reproduce precisely the effects of genetically locking the strain into either the LCD or the HCD state. Determining how environmental conditions affect the quorum-sensing response will be important to the development of a full understanding of bacterial communication in complex natural environments. The BONCAT method has allowed us to identify a diverse set of proteins that respond to the induction of quorum sensing in V. harveyi . The method facilitates monitoring of changes in protein synthesis on a time scale of minutes, and enables correlation of those changes with the underlying temporal pattern of regulation of the quorum-sensing response. The approach described here should prove useful in studies of a wide variety of time-dependent cellular processes."
} | 2,028 |
24574952 | PMC3922083 | pmc | 285 | {
"abstract": "Restricted Boltzmann Machines (RBMs) and Deep Belief Networks have been demonstrated to perform efficiently in a variety of applications, such as dimensionality reduction, feature learning, and classification. Their implementation on neuromorphic hardware platforms emulating large-scale networks of spiking neurons can have significant advantages from the perspectives of scalability, power dissipation and real-time interfacing with the environment. However, the traditional RBM architecture and the commonly used training algorithm known as Contrastive Divergence (CD) are based on discrete updates and exact arithmetics which do not directly map onto a dynamical neural substrate. Here, we present an event-driven variation of CD to train a RBM constructed with Integrate & Fire (I&F) neurons, that is constrained by the limitations of existing and near future neuromorphic hardware platforms. Our strategy is based on neural sampling, which allows us to synthesize a spiking neural network that samples from a target Boltzmann distribution. The recurrent activity of the network replaces the discrete steps of the CD algorithm, while Spike Time Dependent Plasticity (STDP) carries out the weight updates in an online, asynchronous fashion. We demonstrate our approach by training an RBM composed of leaky I&F neurons with STDP synapses to learn a generative model of the MNIST hand-written digit dataset, and by testing it in recognition, generation and cue integration tasks. Our results contribute to a machine learning-driven approach for synthesizing networks of spiking neurons capable of carrying out practical, high-level functionality.",
"introduction": "1. Introduction Machine learning algorithms based on stochastic neural network models such as RBMs and deep networks are currently the state-of-the-art in several practical tasks (Hinton and Salakhutdinov, 2006 ; Bengio, 2009 ). The training of these models requires significant computational resources, and is often carried out using power-hungry hardware such as large clusters (Le et al., 2011 ) or graphics processing units (Bergstra et al., 2010 ). Their implementation in dedicated hardware platforms can therefore be very appealing from the perspectives of power dissipation and of scalability. Neuromorphic Very Large Scale Integration (VLSI) systems exploit the physics of the device to emulate very densely the performance of biological neurons in a real-time fashion, while dissipating very low power (Mead, 1989 ; Indiveri et al., 2011 ). The distributed structure of RBMs suggests that neuromorphic VLSI circuits and systems can become ideal candidates for such a platform. Furthermore, the communication between neuromorphic components is often mediated using asynchronous address-events (Deiss et al., 1998 ) enabling them to be interfaced with event-based sensors (Liu and Delbruck, 2010 ; Neftci et al., 2013 ; O'Connor et al., 2013 ) for embedded applications, and to be implemented in a very scalable fashion (Silver et al., 2007 ; Joshi et al., 2010 ; Schemmel et al., 2010 ). Currently, RBMs and the algorithms used to train them are designed to operate efficiently on digital processors, using batch, discrete-time, iterative updates based on exact arithmetic calculations. However, unlike digital processors, neuromorphic systems compute through the continuous-time dynamics of their components, which are typically Integrate & Fire (I&F) neurons (Indiveri et al., 2011 ), rendering the transfer of such algorithms on such platforms a non-trivial task. We propose here a method to construct RBMs using I&F neuron models and to train them using an online, event-driven adaptation of the (CD) algorithm. We take inspiration from computational neuroscience to identify an efficient neural mechanism for sampling from the underlying probability distribution of the RBM. Neuroscientists argue that brains deal with uncertainty in their environments by encoding and combining probabilities optimally (Doya et al., 2006 ), and that such computations are at the core of cognitive function (Griffiths et al., 2010 ). While many mechanistic theories of how the brain might achieve this exist, a recent neural sampling theory postulates that the spiking activity of the neurons encodes samples of an underlying probability distribution (Fiser et al., ( 2010 ). The advantage for a neural substrate in using such a strategy over the alternative one, in which neurons encode probabilities, is that it requires exponentially fewer neurons. Furthermore, abstract model neurons consistent with the behavior of biological neurons can implement Markov Chain Monte Carlo (MCMC) sampling (Buesing et al., 2011 ), and RBMs sampled in this way can be efficiently trained using CD, with almost no loss in performance (Pedroni et al., 2013 ). We identify the conditions under which a dynamical system consisting of I&F neurons performs neural sampling. These conditions are compatible with neuromorphic implementations of I&F neurons (Indiveri et al., 2011 ), suggesting that they can achieve similar performance. The calibration procedure necessary for configuring the parameters of the spiking neural network is based on firing rate measurements, and so is easy to realize in software and in hardware platforms. In standard CD, weight updates are computed on the basis of alternating, feed-forward propagation of activities (Hinton, 2002 ). In a neuromorphic implementation, this translates to reprogramming the network connections and resetting its state variables at every step of the training. As a consequence, it requires two distinct dynamical systems: one for normal operation (i.e., testing), the other for training, which is highly impractical. To overcome this problem, we train the neural RBMs using an online adaptation of CD. We exploit the recurrent structure of the network to mimic the discrete “construction” and “reconstruction” steps of CD in a spike-driven fashion, and Spike Time Dependent Plasticity (STDP) to carry out the weight updates. Each sample (spike) of each random variable (neuron) causes synaptic weights to be updated. We show that, over longer periods, these microscopic updates behave like a macroscopic CD weight update. Compared to standard CD, no additional connectivity programming overhead is required during the training steps, and both testing and training take place in the same dynamical system. Because RBMs are generative models, they can act simultaneously as classifiers, content-addressable memories, and carry out probabilistic inference. We demonstrate these features in a MNIST hand-written digit task (LeCun et al., 1998 ), using an RBM network consisting of one layer of 824 “visible” neurons and one layer of 500 “hidden” neurons. The spiking neural network was able to learn a generative model capable of recognition performances with accuracies up to 91.9%, which is close to the performance obtained using standard CD and Gibbs sampling, 93.6%.",
"discussion": "4. Discussion Neuromorphic systems are promising alternatives for large-scale implementations of RBMs and deep networks, but the common procedure used to train such networks, (CD), involves iterative, discrete-time updates that do not straightforwardly map on a neural substrate. We solve this problem in the context of the RBM with a spiking neural network model that uses the recurrent network dynamics to compute these updates in a continuous-time fashion. We argue that the recurrent activity coupled with STDP dynamics implements an event-driven variant of CD. Event-driven CD enables the system to learn on-line, while being able to carry out functionally relevant tasks such as recognition, data generation and cue integration. The CD algorithm can be used to learn the parameters of probability distributions other than the Boltzmann distribution (even those without any symmetry assumptions). Our choice for the RBM, whose underlying probability distribution is a special case of the Boltzmann distribution, is motivated by the following facts: They are universal approximators of discrete distributions (Le Roux and Bengio, 2008 ); the conditions under which a spiking neural circuit can naturally perform MCMC sampling of a Boltzmann distribution were previously studied (Merolla et al., 2010 ; Buesing et al., 2011 ); and RBMs form the building blocks of many deep learning models such as DBNs, which achieve state-of-the-art performance in many machine learning tasks (Bengio, 2009 ). The ability to implement RBMs with spiking neurons and train then using event-based CD paves the way toward on-line training of DBNs of spiking neurons (Hinton et al., 2006 ). We chose the MNIST handwritten digit task as a benchmark for testing our model. When the RBM was trained with standard CD, it could recognize up to 926 out of 1000 of out-of-training samples. The MNIST handwritten digit recognition task was previously shown in a digital neuromorphic chip (Arthur et al., 2012 ), which performed at 89% accuracy, and in a software simulated visual cortex model (Eliasmith et al., 2012 ). However, both implementations were configured using weights trained off-line. A recent article showed the mapping of off-line trained DBNs onto spiking neural network (O'Connor et al., 2013 ). Their results demonstrated hand-written digit recognition using neuromorphic event-based sensors as a source of input spikes. Their performance reached up to 94.1% using leaky I&F neurons. The use of an additional layer explains to a large extent their better performance compared to ours (91.9%). Our work extends (O'Connor et al., 2013 ) with on-line training that is based on synaptic plasticity, testing its robustness to finite weight precision, and providing an interpretation of spiking activity in terms of neural sampling. To achieve the computations necessary for sampling from the RBM, we have used a neural sampling framework (Fiser et al., 2010 ), where each spike is interpreted as a sample of an underlying probability distribution. Buesing et al. proved that abstract neuron models consistent with the behavior of biological spiking neurons can perform MCMC, and have applied it to a basic learning task in a fully visible Boltzmann Machine. We extended the neural sampling framework in three ways: First, we identified the conditions under which a dynamical system consisting of I&F neurons can perform neural sampling; Second, we verified that the sampling of RBMs was robust to finite-precision parameters; Third, we demonstrated learning in a Boltzmann Machine with hidden units using STDP synapses. In neural sampling, neurons behave stochastically. This behavior can be achieved in I&F neurons using noisy input currents, created by a Poisson spike train. Spike trains with Poisson-like statistics can be generated with no additional source of noise, for example by the following mechanisms: balanced excitatory and inhibitory connections (van Vreeswijk and Sompolinsky, 1996 ), finite-size effects in a large network, and neural mismatch (Amit and Brunel, 1997 ). The latter mechanism is particularly appealing, because it benefits from fabrication mismatch and operating noise inherent to neuromorphic implementations (Chicca and Fusi, 2001 ). Other groups have also proposed to use I&F neuron models for computing the Boltzmann distribution. (Merolla et al., 2010 ) have shown that noisy I&F neurons' activation function is approximately a sigmoid as required by the Boltzmann machine, and have devised a scheme whereby a global inhibitory rhythm drives the network to generate samples of the Boltzmann distribution. O'Connor et al. ( 2013 ) have demonstrated a deep belief network of I&F neurons that was trained off-line, using standard CD and tested it using the MNIST database. Independently and simultaneously to this work, Petrovici et al. ( 2013 ) demonstrated that conductance-based I&F neurons in a noisy environment are compatible with neural sampling as described in Buesing et al. ( 2011 ). Similarly, Petrovici et al. find that the choice of non-rectangular PSPs and the approximations made by the I&F neurons are not critical to the performance of the neural sampler. Our work extends all of those above by providing an online, STDP-based learning rule to train RBMs sampled using I&F neurons. 4.1. Applicability to neuromorphic hardware Neuromorphic systems are sensible to fabrication mismatch and operating noise. Fortunately, the mismatch in the synaptic weights and the activation function parameters γ and β are not an issue if the biases and the weights are learned, and the functionality of the RBM is robust to small variations in the weights caused by discretization. These two findings are encouraging for neuromorphic implementations of RBMs. However, at least two conceptual problems of the presented RBM architecture must be solved in order to implement such systems on a larger scale. First, the symmetry condition required by the RBM does not necessarily hold. In a neuromorphic device, the symmetry condition is impossible to guarantee if the synapse weights are stored locally at each neuron. Sharing one synapse circuit per pair of neurons can solve this problem. This may be impractical due to the very large number of synapse circuits in the network, but may be less problematic when using Resistive Random-Access Memorys (RRAMs) (also called memristors ) crossbar arrays to emulate synapses (Kuzum et al., 2011 ; Cruz-Albrecht et al., 2013 ; Serrano-Gotarredona et al., 2013 ). RRAM are a new class of nanoscale devices whose current-voltage relationship depends on the history of other electrical quantities (Strukov et al., 2008 ), and so act like programmable resistors. Because they can conduct currents in both directions, one RRAM circuit can be shared between a pair of neurons. A second problem is the number of recurrent connections. Even our RBM of modest dimensions involved almost two million synapses, which is impractical in terms of bandwidth and weight storage. Even if a very high number of weights are zero, the connections between each pair of neurons must exist in order for a synapse to learn such weights. One possible solution is to impose sparse connectivity between the layers (Murray and Kreutz-Delgado, 2007 ; Tang and Eliasmith, 2010 ) and implement synaptic connectivity in a scalable hierarchical address-event routing architecture (Joshi et al., 2010 ; Park et al., 2012 ). 4.2. Outlook: a custom learning rule our method combines I&F neurons that perform neural sampling and the CD rule. although we showed that this leads to a functional model, we do not know whether event-driven CD is optimal in any sense. This is partly due to the fact that CD k is an approximate rule (Hinton, 2002 ), and it is still not entirely understood why it performs so well, despite extensive work in studying its convergence properties (Carreira-Perpinan and Hinton, 2005 ). furthermore, the distribution sampled by the I&F neuron does not exactly correspond to the Boltzmann distribution and the average weight updates in event-driven CD differ from those of standard CD, because in the latter they are carried out at the end of the reconstruction step. A very attractive alternative is to derive a custom synaptic plasticity rule that minimizes some functionally relevant quantity (such as Kullback-Leibler divergence or Contrastive Divergence), given the encoding of the information in the I&F neuron (Deneve, 2008 ; Brea et al., 2013 ). A similar idea was recently pursued in Brea et al. ( 2013 ), where the authors derived a triplet-based synaptic learning rule that minimizes an upper bound of the Kullback–Leibler divergence between the model and the data distributions. Interestingly, their rule had a similar global signal that modulates the learning rule, as in event-driven CD, although the nature of this resemblance remains to be explored. Such custom learning rules can be very beneficial in guiding the design of on-chip plasticity in neuromorphic VLSI and RRAM nanotechnologies, and will be the focus of future research. 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."
} | 4,077 |
27117333 | PMC4846871 | pmc | 287 | {
"abstract": "Coral reef success is largely dependent on the symbiosis between coral hosts and dinoflagellate symbionts belonging to the genus Symbiodinium . Elevated temperatures can result in the expulsion of Symbiodinium or loss of their photosynthetic pigments and is known as coral bleaching. It has been postulated that the expression of light-harvesting protein complexes (LHCs), which bind chlorophylls (chl) and carotenoids, are important in photobleaching. This study explored the effect a sixteen-day thermal stress (increasing daily from 25–34 °C) on integral LHC (chlorophyll a -chlorophyll c 2 -peridinin protein complex ( acpPC )) gene expression in Symbiodinium within the coral Acropora aspera. Thermal stress leads to a decrease in Symbiodinium photosynthetic efficiency by day eight, while symbiont density was significantly lower on day sixteen. Over this time period, the gene expression of five Symbiodinium \n acpPC genes was quantified. Three acpPC genes exhibited up-regulated expression when corals were exposed to temperatures above 31.5 °C ( acpPCSym_1:1 , day sixteen; acpPCSym_15 , day twelve; and acpPCSym_18 , day ten and day sixteen). In contrast, the expression of acpPCSym_5:1 and acpPCSym_10:1 was unchanged throughout the experiment. Interestingly, the three acpPC genes with increased expression cluster together in a phylogenetic analysis of light-harvesting complexes.",
"discussion": "Discussion This study investigated the effect of increased temperatures on the expression of five acpPC genes in Symbiodinium under prolonged thermal stress of A. aspera and is the first to examine the expression of acpPC genes from different LHC Clades. The sixteen-day thermal regime was selected to enable sampling at temperatures leading up to, and inclusive of, a bleaching event and is the first experiment to investigate differential expression of integral antenna proteins in Symbiodinium within a coral host under thermal stress. Quantitative PCR was used to quantify the expression of five acpPC genes that are dispersed though out three clades of the chl a / c lineage of the LHC phylogeny ( Fig. 4 ). Over the course of the experiment temperature significantly effected Symbiodinium density and physiology. Symbiont cells decreased to approximately half the density in thermally stressed corals compared to control corals ( Fig. 2a ) as has been found in variety of other studies 4 35 38 . In addition chl a and chl c levels were elevated over the course of the experiment ( Fig. 2b,c ) in a manner seen before in this species (Gierz and Leggat, unpublished data) 35 . However, a statistical difference was only observed on day sixteen in chl c ( Fig. 2 c), this is consistent with other studies where an increase in chl pigments were observed 35 . In corals, heat-related increases in chl a have previously been recorded at low symbiont densities 39 40 , though in other experiments Symbiodinium pigmentation may be unchanged or decreased 41 42 . Increases in chl pigments have been attributed to repackaging of chls in the chloroplast membrane, with evidence that specific pigment-protein complexes may absorb more light at specific wavelengths 43 . In phytoplankton, chl a -specific absorption of different pigment-protein complexes from the same organism can be highly variable 43 . Therefore, it is possible that the increases in Symbiodinium pigments observed in heated corals may be attributed to alterations in the type of pigment-protein complexes expressed under thermal stress. Imaging-pulse amplitude modulated (PAM) fluorometry analysis demonstrated that Symbiodinium cells exposed to elevated temperatures exhibited decreased photosynthetic efficiency ( Fig. 3a ), this is consistent with previous studies demonstrating the response of cells to elevated temperatures. Decreases in dark-adapted yield occurred throughout the experiment despite small changes in symbiont density, F v /F m levels of ~0.00 were recorded on day sixteen despite cell density being approximately five hundred thousand per cm 2 , indicating cells were incapable of photosynthesis at the end of the stress period. Increased NPQ response in cells at days eight, ten and twelve ( Fig. 3b ) illustrates that the cells were dissipating excess light energy. However, this NPQ response was not present on day sixteen of thermal stress indicating that the symbionts had passed a threshold where photosynthetic processes were no longer functioning. Together, the photosynthetic efficiency results, Symbiodinium densities and changes to pigment levels, demonstrate that in this experiment Symbiodinium were subjected to the full range of temperatures that are seen in a bleaching event, with responses from initial thermal stress through to Symbiodinium expulsion. As such it is reasonable to conclude that acpPC expression patterns are representative of what would be seen in a natural bleaching event. Expression of acpPC genes was found to vary in Symbiodinium cells throughout the experiment. Functionally little is known about the diversity of LHCs, for example whether complexes only associate with specific photosystems, are some more efficient at light capture or energy transfer, are others favoured for photoprotection or do some display increased stability under high temperatures. Characterisation of Symbiodinium \n acpPCs has shown that there is large diversity within the gene super-family 21 . Analysis of the Symbiodinium genome has provided more of an insight into the diversification of the LHC family 23 , reinforcing theories on gene duplication and deletion events leading to the current structure of the Symbiodinium genome. The complexity observed in the integral LHC family has been attributed to multiple rounds of intra- and inter-genic gene duplication events 20 23 . A large gene super-family encodes integral LHCs, and a significant level of sequence similarity has been detected between the protein complexes 21 23 . Phylogenetic analysis of LHCs and LHC-like protein super-families indicate that the ancestor is most likely a central group of two-helix stress-enhanced proteins that had previously evolved from a gene-duplication event of the high-light induced proteins of cyanobacteria 44 45 . However, based upon sequence divergence, it is reasonable to assume that different clades of acpPC may have different functions. As such, the differences in expression between those acpPC ( Fig. 5a–e ) from Clade 1 and 2 versus Clade 3b ( Fig. 4 ), may be indicative of functional roles, with Clade 1 and 2 possibly being involved in stress response while those of Clade 3b are constitutively expressed under the conditions used here. Some ways in which acpPC may functionally vary is in the binding of varied pigment ratios, specificity for association to photosystems and response to stress events. For example it has been found that a variety of acpPC transcripts are missing key chl and pigment binging residues 21 . In addition it is not clear to which photosystems different acpPCs bind. In green plants, ten highly conserved genes encoding chl a / b binding proteins have been identified, associated with photosystem I (PSI) are four pigment - protein complexes (encoded by genes Lhca1, Lhca2, Lhca3 and Lhca4 ), and associated with PSII are six pigment – protein complexes (encoded by genes Lhcb1, Lhcb2 , Lhcb3 , Lhcb4 , Lhcb5 , Lhcb6 ) 20 . This can be contrasted to Symbiodinium where there is high sequence diversity coupled with high copy number, and as yet it is not clear which proteins bind to PSI or PSII 21 23 . It has been suggested that this sequence diversity allows for functional diversity such as, stress response 21 , attachment/dissociation 32 and enhanced photoprotection 46 . As such, it will only be with the linkage of more transcriptome and genome studies, and the analysis of chl and accessory pigments binding residues, linked to functional studies, that we will be able to elucidate the reason for the expansion of this gene family in dinoflagellates. Core photosystem genes, psaA and psbA have previously been investigated in Symbiodinium under thermal stress. Decreases of psaA and psbA are hypothesised to significantly impair the mechanisms associated with coping with thermal stress 47 . In this study, the expression of the psbA gene, which encodes the core PSII D1 protein, was also quantified ( Fig. 5f ). Over the course of the experiment psbA expression increased on days eight and ten ( Fig. 5f ) although expression in treatment samples was not statistically significant. However, on day sixteen, psbA expression decreased ( Fig. 5f ), potentially to reduce light absorption to limit the amount of energy captured under stress conditions as a photoprotective mechanism. As in previous studies, investigating transcript abundance in Symbiodinium very small changes in gene expression were observed in this study. In the five acpPC genes quantified, the largest observed change was a 2.44 fold increase ( acpPCSym_18 ) on day ten of the thermal stress experiment. In Symbiodinium in hospite , these small changes in transcripts have been observed previously 33 34 35 36 and it is postulated that regulation is most likely post-translational and not at the transcriptional level 33 47 48 . This study exploited a bleaching experiment to investigate the effect of thermal stress on photosynthetic genes. Quantitative PCR was used to determine the expression of five integral LHC genes. Three LHC genes ( acpPCSym_1:1 , acpPCSym_15 and acpPCSym_18 ) were found to have increased expression over the duration of the experiment and interestingly, grouped in Clade 1 and Clade 2 of the LHC phylogeny. Additionally two LHC genes ( acpPCSym_5:1 and acpPCSym_10:1 ) grouped with Clade 3b did not exhibit differences in expression. Though transcriptional changes were detected, expression changes observed were less than 2.5 fold throughout the experiment. This is consistent with previous studies where small-scale changes in gene expression were also observed. Given that we currently do not know how the diverse range of LHCs are associated with the photosystems, both PSII and PSI, their specific functional roles, (e.g., light harvesting efficiency versus photoprotection), or the importance of Symbiodinium photosynthesis to the survival of corals, it is imperative that future research focuses on the specific roles of LHCs."
} | 2,625 |
36033624 | PMC9399768 | pmc | 289 | {
"abstract": "Spiking Neural Networks (SNNs) have shown capabilities of achieving high accuracy under unsupervised settings and low operational power/energy due to their bio-plausible computations. Previous studies identified that DRAM-based off-chip memory accesses dominate the energy consumption of SNN processing. However, state-of-the-art works do not optimize the DRAM energy-per-access, thereby hindering the SNN-based systems from achieving further energy efficiency gains. To substantially reduce the DRAM energy-per-access, an effective solution is to decrease the DRAM supply voltage, but it may lead to errors in DRAM cells (i.e., so-called approximate DRAM ). Toward this, we propose EnforceSNN , a novel design framework that provides a solution for resilient and energy-efficient SNN inference using reduced-voltage DRAM for embedded systems. The key mechanisms of our EnforceSNN are: (1) employing quantized weights to reduce the DRAM access energy; (2) devising an efficient DRAM mapping policy to minimize the DRAM energy-per-access; (3) analyzing the SNN error tolerance to understand its accuracy profile considering different bit error rate (BER) values; (4) leveraging the information for developing an efficient fault-aware training (FAT) that considers different BER values and bit error locations in DRAM to improve the SNN error tolerance; and (5) developing an algorithm to select the SNN model that offers good trade-offs among accuracy, memory, and energy consumption. The experimental results show that our EnforceSNN maintains the accuracy (i.e., no accuracy loss for BER ≤ 10 −3 ) as compared to the baseline SNN with accurate DRAM while achieving up to 84.9% of DRAM energy saving and up to 4.1x speed-up of DRAM data throughput across different network sizes.",
"conclusion": "7. Conclusion We propose a novel EnforceSNN framework to achieve resilient and energy-efficient SNN inference considering reduced-voltage-based approximate DRAM, through weight quantization, error-aware DRAM mapping, SNN error-tolerance analysis, efficient error-aware SNN training, and effective SNN model selection. Our EnforceSNN achieves no accuracy loss for BER ≤ 10 −3 with minimum retraining costs as compared to the baseline SNN with accurate DRAM while achieving up to 84.9% of DRAM energy saving and up to 4.1x speed-up of DRAM data throughput. In this manner, our study may enable efficient SNN inference for energy–constrained embedded devices like the Edge-AI.",
"introduction": "1. Introduction Spiking neural networks (SNNs) have demonstrated the potential of obtaining high accuracy under unsupervised settings and low operational energy due to their bio-plausible spike-based computations (Putra and Shafique, 2020 ). A larger SNN model is usually favorable as it offers higher accuracy than the smaller ones, as shown by our experimental results in Figure 1A . Here, the 1MB-sized network achieves only 75%, while the 200MB-sized network achieves 92% accuracy for the MNIST dataset. This MNIST dataset is a set of training and test images for handwritten digits 0–9 (Lecun et al., 1998 ). On the other hand, most of the SNN hardware platforms have relatively small on-chip memory, e.g., less than 100MB (Roy et al., 2017 ; Sen et al., 2017 ; Frenkel et al., 2019a , b ). Therefore, running an SNN model with a larger size than the on-chip memory of SNN hardware platforms will require intensive access to the off-chip memory. Previous studies show that single access to the off-chip memory (i.e., DRAM) incurs significantly higher energy consumption than single access to the on-chip memory (i.e., SRAM) (Sze et al., 2017 ; Putra et al., 2021b ). Moreover, previous study also identified that memory access dominate the energy consumption of SNN processing, incurring 50–75% of the total system energy across different SNN hardware platforms, as shown in Figure 1B . The reason is that DRAM access energy is significantly higher than other SNN operations (e.g., neuron operations) (Krithivasan et al., 2019 ). This problem is even more critical for AI applications with stringent constraints (e.g., low-cost embedded devices with a small on-chip memory size) (Shafique et al., 2021 ) since it leads to even more intensive DRAM accesses. Consequently, this problem hinders SNN-based embedded systems from obtaining further efficiency gains. Figure 1 (A) Accuracy profiles of small-sized and large-sized SNN models on the MNIST dataset which are obtained from our experiments. (B) Breakdown of the energy consumption of SNN processing on different SNN hardware platforms, i.e., PEASE (Roy et al., 2017 ), SNNAP (Sen et al., 2017 ), and TrueNorth (Akopyan et al., 2015 ); adapted from studies in Krithivasan et al. ( 2019 ). (C) Bit error rate (BER) of approximate DRAM and its respective supply voltage V supply ; adapted from studies in Chang et al. ( 2017 ). (D) The estimated DRAM energy savings achieved by our technique when combined with the weight pruning across different rates of network connectivity (i.e., synaptic connections) for a network with 4900 excitatory neurons. The results are obtained from experiments using the LPDDR3-1600 4Gb DRAM configuration and the DRAMPower simulator (Chandrasekar, 2014 ). Targeted Research Problem: \n How can we substantially decrease the DRAM access energy for the SNN inference, while maintaining accuracy . The solution to this problem will enable efficient SNN inference for energy–constrained embedded devices and their applications for Edge-AI and Smart CPS. Edge-AI is the system that runs Artificial Intelligence (AI) algorithms on resource- and energy-constrained computing devices at the edge of the network, i.e., close to the source of data (Shi et al., 2016 ; Satyanarayanan, 2017 ; Yu et al., 2018 ; Chen and Ran, 2019 ; Liu et al., 2019 ; Cao et al., 2020 ). Meanwhile, Smart CPS (Cyber-Physical System) is the system that includes the interacting networks of computational components (e.g., computation and storage devices), physical components (e.g., sensors and actuators), and human users (Chattopadhyay et al., 2017 ; Griffor et al., 2017 ; Kriebel et al., 2018 ; Shafique et al., 2018 ). 1.1. State-of-the-art and their limitations To decrease the energy consumption of SNN inference, state-of-the-art works have developed different techniques, which can be loosely classified as the following. Reduction of the SNN operations through approximate neuron operations (Sen et al., 2017 ), weight pruning (Rathi et al., 2019 ), and neuron removal (Putra and Shafique, 2020 ). These techniques decrease the number of DRAM accesses for the corresponding model parameters. Quantization by reducing the range of representable values for SNN parameters (e.g., weights) (Rathi et al., 2019 ; Putra and Shafique, 2020 ; Putra and Shafique, 2021a ). These techniques reduce the amount of SNN parameters (e.g., weights) to be stored in and fetched from DRAM. Limitations: These state-of-the-art works mainly aim at reducing the number of DRAM accesses, but do not optimize the DRAM energy-per-access and do not employ approximations in DRAM that provide an additional knob for obtaining high energy efficiency. Therefore, optimization gains offered by these works are sub-optimal, hindering the SNN inference systems from achieving the full potential of DRAM energy savings. Therefore, the effective optimization should jointly minimize the DRAM energy-per-access and the number of DRAM accesses, by leveraging the approximation in DRAM to expose the full energy-saving potential, while overcoming the negative impact of the approximation-induced errors (i.e., bit-flips in DRAM cells). Figure 1C shows the approximation-induced error rates in DRAM. To address these limitations, we employ approximate DRAM (i.e., DRAM with reduced supply voltage) with efficient DRAM data mapping policy and fault-aware training to substantially reduce the DRAM access energy in SNN inference systems while preserving their accuracy . Moreover, our proposed technique can also be combined with state-of-the-art techniques to further improve the energy efficiency of SNN inference. For example, Figure 1D shows the estimated DRAM energy savings achieved by our technique when combined with the weight pruning. To highlight the potential of reduced-voltage approximate DRAM, we perform an experimental case study in the following section. 1.2. Motivational case study and key challenges In the case study, we aim at studying (1) the dynamics of DRAM bitline voltage ( V bitline ) for both the accurate and approximate DRAM settings, and (2) the DRAM access energy for different access conditions (including a row buffer hit, miss, or conflict). Note, V bitline is defined as the voltage measured in each DRAM bitline when a DRAM supply voltage ( V supply ) is applied, as shown in Figures 2A , 4C . Further details on the dynamics of V bitline are provided in Section 2.2.2. For DRAM access conditions, a row buffer hit means that the requested data has been loaded in the DRAM row buffer, thus the data can be accessed without additional DRAM operations. Meanwhile, a row buffers miss or conflict needs to open the requested DRAM row before the data can be loaded into the row buffer and then accessed. Further information on the DRAM access conditions is discussed in Section 2.2.1. Figure 2 (A) The dynamics of V bitline under different V supply values. (B) DRAM access energy for a row buffer hit, a row buffer miss, and a row buffer conflict, under different V supply values. For the experimental setup, we employ the DRAM circuit model from the study of Chang et al. ( 2017 ) and the SPICE simulator to study the dynamics of V bitline . The accurate DRAM operates at 1.35V of the supply voltage ( V supply ), while the approximate one operates at 1.025V. Further details on the experimental setup are discussed in Section 5. Furthermore, we consider the LPDDR3-1600 4Gb DRAM configuration as it is representative of the low-power DRAM types for embedded systems. We employ the DRAMPower simulator to estimate the DRAM access energy because it has been validated against real measurements (Chandrasekar, 2014 ) and has been widely used in the computer architecture communities. Figure 2 presents the experimental results, from which we make the following key observations. The V bitline decreases as the V supply decreases, hence forcing the DRAM cells to operate under lower reliability as the weak cells may fail to hold the correct bits. Weak cells are DRAM cells that fail when the DRAM parameters (e.g., voltage, timing) are reduced (Chang et al., 2017 ; Kim et al., 2018 ). The reduced-voltage DRAM decreases the DRAM energy-per-access across different access conditions, i.e., by up to 42% of energy reduction for each access. The row buffer hit has lower energy consumption than the row buffer miss or conflict. Moreover, row buffer hit also incurs less latency than the row buffer miss or conflict (Putra et al., 2020 ; Putra et al., 2021b ). Therefore, the row buffer hit should be exploited to optimize the DRAM latency and energy. Although employing the approximate DRAM can substantially decrease the DRAM energy-per-access, it also decreases the DRAM reliability since the bit errors increase when the V supply decreases, as shown in Figure 1C . These errors may degrade the accuracy of SNN inference since they can change the weight values in DRAM, which then deteriorates the neuron behavior. Associated Research Challenge: \n How to achieve low DRAM access energy for SNN inference using approximate DRAM, while minimizing their negative impact on the accuracy . 1.3. Our novel contributions To overcome the above research challenges, we propose the EnforceSNN framework , which enables resilient and energy-efficient SNNs considering approximate DRAMs (i.e., reduced-voltage DRAMs) for embedded systems. Based on the best of our knowledge, it is the first effort that employs approximate DRAM for improving the energy efficiency of SNN inference, while enhancing their error tolerance against bit errors in DRAM. Our EnforceSNN framework employs the following key steps. Employing weight quantization to reduce the memory footprint for SNN weights and the number of DRAM accesses for SNN inference, thereby optimizing the DRAM access energy. Devising an efficient DRAM data mapping to maximize row buffer hits for optimizing the DRAM energy-per-access while considering BER in DRAM. Analyzing the SNN error tolerance to understand the SNN accuracy profile under different DRAM supply voltage and different BER values. Improving the SNN error tolerance by developing and employing efficient fault-aware training (FAT) that considers SNN accuracy profile and bit error locations in DRAM. Devising an algorithm to select the SNN model that offers good trade-offs among accuracy, memory, and energy consumption from the given model candidates using the proposed reward function. Key Results: We evaluate the EnforceSNN framework for (1) classification accuracy using PyTorch-based simulations (Hazan et al., 2018 ) on a multi-GPU machine considering the MNIST and Fashion MNIST datasets 1 , and (2) DRAM access energy using DRAMPower (Chandrasekar, 2014 ). We perform an epoch of unsupervised learning (60K experiments) for each retraining process considering each combination of the SNN model, workload, and training BER; then perform inference (10K experiments) for each combination of the SNN model, workload, and testing BER. The experimental results indicate that our EnforceSNN reduces the DRAM access energy by up to 84.9% and improves the speed-up up to 4.1x while maintaining the accuracy (no accuracy loss) across different network sizes for BER ≤ 10 −3 .",
"discussion": "6. Results and discussion 6.1. Improvements of the SNN error tolerance Figure 11 shows the accuracy of the baseline model and the EnforceSNN-improved model with accurate and approximate DRAM across different BER values and precision levels, i.e., FP32 and FxP8 (“signed Q 1.6” and “unsigned Q 1.7”), network sizes, and workloads (i.e., the MNIST and Fashion MNIST datasets). In general, we observe that the baseline model with approximate DRAM achieves lower accuracy than the baseline model with accurate DRAM, and the accuracy decreases as the BER increases. These trends are observed across different weight precision levels, network sizes, and datasets. The reason is that the weights are changed (i.e., flipped) if they are stored in the faulty DRAM cells, and these weights are not trained to adapt to such bit flips. Therefore, the corresponding neuron behavior deteriorates from the expected behavior, hence decreasing accuracy. On the other hand, the EnforceSNN-improved model with approximate DRAM improves accuracy over the baseline model with accurate and approximate DRAM, across different BER values, network sizes, and datasets, as shown in ❶. We also observe that the EnforceSNN-improved model with approximate DRAM improves the accuracy over the baseline model with accurate and approximate DRAM, even in the high error rate case (i.e., BER = 10 −2 ), as shown in ❷ for FP32 and ❸ for FxP8 weight precision levels. The reason is that our EnforceSNN incorporates the error profiles from the approximate DRAM across different BER values in the training process, which makes the SNN model adaptive to the presence of DRAM errors, thereby improving the SNN error tolerance. For the MNIST dataset, a high error rate (i.e., BER = 10 −2 ) typically decreases the accuracy of the SNN-FxP8 more than the SNN-FP32, as shown in ❹. The reason is that the MNIST dataset has a narrow weight distribution in each class to represent its digit features, hence bit errors may change the weight values significantly in the FxP8 precision than the FP32 due to its shorter bit-width. As a result, the corresponding neuron behavior deteriorates from its ideal behavior, hence degrading accuracy. For the Fashion MNIST dataset, the SNN-FxP8 may achieve higher accuracy than the SNN-FP32 in some cases, as shown in ❺. The potential reason is the following. The Fashion MNIST dataset has relatively more complex features than the MNIST dataset, hence having a wider weight distribution in each class to represent its various features which may overlap with features from other classes (i.e., non-unique features). Then, the quantization removes these non-unique features by eliminating the less significant bits of the trained weights (i.e., like the denoising effect), and the retraining makes the quantized weights adaptive to bit flips, thereby leading to higher accuracy than the non-quantized ones. Furthermore, we also observe that the accuracy of the SNN-FxP8 starts showing notable degradation at a high error rate (i.e., BER = 10 −2 ). For quantized models, in general, the “unsigned Q 1.7” and “signed Q 1.6” formats have similar trends and comparable accuracy as they represent similar weight values which differ only in the least significant fractional bit, thereby leading to similar neuron behavior and accuracy. These formats may have notable accuracy differences for some cases, such as after the retraining process, as shown by ❻. The possible reason is that these formats have different bit positions for the sign, integer, and fraction, thereby making the DRAM errors affect different weight bits and lead to different learning qualities during the respective fault-aware training. Figure 11 The accuracy of the baseline model with accurate and approximate DRAM, as well as the EnforceSNN-improved model with approximate DRAM for (A) MNIST and (B) Fashion MNIST datasets, across different precision levels, different BER values, and different network sizes. In summary, our EnforceSNN maintains accuracy (i.e., no accuracy loss) as compared to the baseline with accurate DRAM when BER ≤ 10 −3 across different datasets. Meanwhile, for higher BER values (i.e., 10 −3 < BER ≤ 10 −2 ), our EnforceSNN still achieves higher accuracy than the baseline with accurate DRAM across different datasets. Therefore, these results show that our EnforceSNN framework effectively improves the SNN error tolerance against DRAM errors with minimum retraining efforts . 6.2. DRAM access energy savings and throughput improvements Figure 12A shows the normalized energy consumption of the DRAM accesses for an inference (i.e., inferring one input sample) required by the baseline model and the EnforceSNN-improved model with accurate and approximate DRAM, across different V supply values, precision levels, network sizes, and workloads. We observe that different network sizes show similar normalized DRAM access energy, hence we only show a single figure representing the experimental results for all network sizes. For accurate DRAM cases across different network sizes, the baseline model achieves 75% DRAM energy saving when it employs the quantization technique, while our EnforceSNN-improved model achieves 75.1% DRAM energy saving due to the quantization and the proposed DRAM mapping policy, as shown in ❼. Meanwhile, the difference in these DRAM energy savings comes from the DRAM mapping policy. That is, our EnforceSNN optimizes the DRAM energy-per-access by maximizing the row buffer hits and the multi-bank burst feature, thereby having fewer row buffer conflicts than the baseline which only exploits the single-bank burst feature. For the FP32 precision across different network sizes, employing the approximate DRAM in the baseline model reduces the DRAM energy savings by up to 39.2% as compared to employing the accurate DRAM. Meanwhile, employing the approximate DRAM in the EnforceSNN-improved model reduces the DRAM energy savings by up to 39.5% as compared to employing the accurate DRAM, as shown in ❽. These energy savings come from the reduced DRAM energy-per-access due to the reduction of operational V supply . Moreover, the difference in energy savings between the baseline and our EnforceSNN also comes from the DRAM mapping policy. For the FxP8 precision (i.e., “signed Q 1.6” and “unsigned Q 1.7”) across different network sizes, employing the approximate DRAM in the baseline model reduces the DRAM energy savings by up to 84.8% over employing the accurate one. Meanwhile, employing the approximate DRAM in the EnforceSNN-improved model reduces the DRAM energy savings by up to 84.9% over employing the accurate one, as shown in ❾. These energy savings come from the reduced weight precision and the reduced DRAM energy-per-access due to V supply reduction. Moreover, the difference in energy savings between the baseline and our EnforceSNN also comes from the DRAM mapping policy. Furthermore, we also observe that our EnforceSNN-improved model obtains 4.1x throughput speed-up over the baseline model across different V supply values, workloads, and network sizes; refer to ❿ in Figure 12B . It is achieved through (1) the quantization technique which reduces the number of DRAM accesses, and (2) our proposed DRAM mapping policy which optimizes the DRAM latency-per-access by maximizing the row buffer hits and the multi-bank burst features. The results also show that the “unsigned Q 1.7” and “signed Q 1.6” achieve comparable DRAM access energy savings and throughput improvements since they employ the same bitwidth of weights, thereby having similar DRAM access behavior. Figure 12 (A) The normalized DRAM access energy for an inference incurred by the baseline model and the EnforceSNN-improved model with accurate and approximate DRAM, and (B) the normalized speed-up of DRAM data throughput for an inference achieved by our EnforceSNN-improved model over the baseline model, across different V supply values, different workloads (datasets), and different network sizes (N-900, N-1600, N-2500, and N-3600). These results are applicable for all network sizes. They are also applicable for both the MNIST and Fashion MNIST datasets, as these workloads have similar DRAM access energy, due to the same number of weights and number of DRAM accesses for an inference. In summary, the results in Figure 12 indicate that our EnforceSNN framework substantially reduces the DRAM access energy by employing the reduced-voltage approximate DRAM and our efficient DRAM mapping policy, while effectively improving the DRAM data throughput mainly due to the quantization . 6.3. Model selection under design trade-offs Figures 13 , 14 show the results of the accuracy-memory-energy trade-offs for the MNIST and Fashion MNIST datasets, respectively. In this evaluation, the quantized models consider the FxP8 precision in “signed Q 1.6” format. For the given SNN model candidates, we observe that the models that incur small memory size typically employ FxP8 precision, as shown in Figure 13A for the MNIST and Figure 14A for the Fashion MNIST. Considering that the accuracy of the FxP8-based models is comparable to the FP32-based models, we narrow down the candidates to only the FxP8-based models. Figure 13 The trade-offs among accuracy, memory footprint, and energy consumption for the MNIST. (A) Accuracy profiles of SNN models. (B–G) Reward profiles of SNN models. The network sizes represent the memory sizes and the BER values represent the energy savings from approximate DRAM. Figure 14 The trade-offs among accuracy, memory, and energy consumption for the Fashion MNIST. (A) Accuracy profiles of SNN models. (B–G) Reward profiles of SNN models. The network sizes represent the memory sizes, and the BER values represent the energy savings from approximate DRAM. To analyze the design trade-offs, we explore the impact of different μ and ε values on the rewards. For instance, if we consider that the accuracy should have a higher priority than the memory and energy consumption, we set μ and ε low (e.g., μ = 0 and ε = 0). Meanwhile, if we consider that the memory should have a higher priority than the energy consumption, we set μ higher than ε (e.g., μ = 10 and ε = 0). For both cases, the highest reward is achieved by the EnforceSNN-improved N-1600 FxP8 for the MNIST and the EnforceSNN-improved N-2500 FxP8 for the Fashion MNIST under 10 −5 error rate; refer to 🅓 for μ = 0 and ε = 0 cases, and refer to 🅔 for μ = 10 and ε = 0 cases. The reason is that these models employ our efficient FAT technique to improve their error tolerance, thereby leading to high accuracy under a high error rate. We also observe that having μ higher than ε makes the high rewards shift toward smaller models, as shown by 🅔 in Figure 13C . The reason is that a higher μ makes the small m norm have a smaller impact on the reward reduction than the large m norm , thereby maintaining the high reward values. If the energy consumption should have a higher priority than the memory footprint, we set μ lower than ε (e.g., μ = 0 and ε = 10). The highest reward is achieved by the EnforceSNN-improved N-1600 FxP8 under 10 −5 error rate for the MNIST and the EnforceSNN-improved N-2500 FxP8 under 10 −4 error rate for the Fashion MNIST. In this case, we observe that high rewards are shifted toward models with smaller energy consumption (represented by higher BER); refer to 🅕 in Figures 13D , 14D . The reason is that a higher ε makes the small E norm have a smaller impact on the reward reduction than the large E norm , thereby maintaining the high reward values. Furthermore, if the memory and energy consumption should have a higher priority than the accuracy, we set μ and ε high (e.g., μ = 10 and ε = 10). The highest reward is achieved by the EnforceSNN-improved N-1600 FxP8 under 10 −5 error rate for the MNIST and the EnforceSNN-improved N-2500 FxP8 under 10 −4 error rate for the Fashion MNIST. In this case, high rewards are shifted toward models with smaller memory and energy consumption (represented by high BER), but their overall rewards decrease as the values of μ and ε increase; refer to 🅖 in Figures 13E , 14E . The reason is that higher μ and ε jointly make the m norm and E norm decrease the reward. It means that if we want to significantly reduce the memory footprint and energy consumption, we have to accept more accuracy degradation. In summary, the results in Figures 13 , 14 show that our EnforceSNN framework has an effective algorithm to trade off the accuracy, memory footprint, and energy consumption of the given SNN models , thereby providing good applicability for diverse embedded applications with their respective constraints. 6.4. Optimization of the retraining costs The conventional FAT for neural networks usually injects errors at an incremental rate during the retraining process from the minimum value to the maximum one for avoiding accuracy collapse (Koppula et al., 2019 ). Therefore, in this study, the conventional FAT considers BER = {10 −8 , 10 −7 , 10 −6 , …, 10 −2 }, while our efficient FAT in EnforceSNN only considers BER = {10 −4 , 10 −3 , 10 −2 } in the retraining process. Retraining Speed-ups: The conventional FAT with FxP8 (cFAT8) obtains speed-up over the one with FP32 (cFAT32) by up to 1.16x and 1.14x for the MNIST and the Fashion MNIST respectively, since the cFAT8 employs quantized weights, thereby leading to a faster error injection and learning process. Meanwhile, our efficient FAT with FP32 (eFAT32) obtains a 2.33x speed-up over the cFAT32, since our eFAT32 has fewer iterations of the retraining process. Furthermore, we also observe that our efficient FAT with FxP8 (eFAT8) obtains more speed-up, i.e., by up to 2.71x for the MNIST and 2.65x for the Fashion MNIST as shown by 🅗 in Figure 15A , since our eFAT8 employs quantized weights in addition to fewer iterations of the retraining process. Figure 15 (A) The retraining speed-ups across different network sizes (i.e., N-900, N-1600, N-2500, and N-3600), and (B) the retraining energy for the MNIST, which are normalized to the conventional FAT with FP32 for a 900-neuron network. The results for the Fashion MNIST show similar trends to the MNIST since these workloads have similar DRAM access latency and energy due to the same number of weights and number of DRAM accesses for a training phase. Here, FxP8 represents both the “signed Q 1.6” and “unsigned Q 1.7” formats. Retraining Energy Savings: The cFAT8 achieves energy saving over the cFAT32 by up to 13.9% for the MNIST and 12% for the Fashion MNIST since the cFAT8 employs quantized weights which incur lower energy consumption during the error injection and learning process. Meanwhile, our eFAT32 achieves energy saving over the cFAT32 by 57.1%, as the eFAT32 performs fewer iterations of the retraining process as compared to the cFAT32. Our eFAT8 achieves further energy saving, i.e., by up to 63.1% for the MNIST and by up to 62.3% for the Fashion MNIST as shown by 🅘 in Figure 15B , since it employs quantized weights in addition to fewer iterations of the retraining process, thereby leading to a higher energy saving. Carbon Emission Reduction: The retraining process also poses additional challenges that correspond to environmental concerns, i.e., carbon emission. Recent studies have highlighted that the carbon emission from neural network training should be minimized to prevent the increasing rates of natural disasters (Strubell et al., 2019 , 2020 ). To estimate the carbon emission of neural network training, the study of Strubell et al. ( 2019 ) proposed Equation 5 and Equation 6. In these equations, CO 2 e denotes the estimated carbon ( CO 2 ) emission during the training, which is a function of the total power during the training ( p t ). Meanwhile, t is the training duration, p c is the average power from all CPUs, p r is the average power from all main memories (DRAMs), p g is the average power from a GPU and g is the number of GPUs. \n (5) \n C O 2 e = 0 . 954 · p t \n \n (6) \n p t = 1 . 58 · t ( p c + p r + g · p g ) 1000 \n These equations indicate that if we assume p c , p r , p g , and g are the same for different FAT techniques, then the difference will come from the training duration t . Therefore, our efficient FAT in EnforceSNN employs fewer iterations of the retraining process than the conventional FAT, thereby producing less carbon emission. Moreover, our EnforceSNN also reduces the operational power of the main memory (DRAM) through the reduced-voltage approximation approach, thereby further reducing the emission. In summary, our EnforceSNN framework effectively offers speed-up of retraining time, reduction of retraining energy, and less carbon emission than the conventional FAT technique , thereby making it more friendly for our environments. 6.5. Further discussion Previous studies that exploit the reduced-voltage DRAM concept mainly aim at improving the energy efficiency of mobile systems (Haj-Yahya et al., 2020 ), personal computing systems (Nabavi Larimi et al., 2021 ; Fabrício Filho et al., 2022 ), and server systems (David et al., 2011 ; Deng et al., 2011 , 2012a , b ; Nabavi Larimi et al., 2021 ). This concept is also employed for minimizing the energy consumption of deep neural networks (DNNs) (Koppula et al., 2019 ). Since SNNs have different data representation, computation models, and learning rules as compared to DNNs, our EnforceSNN provides a different framework with different techniques that are crafted specifically for improving the resilience and the energy efficiency of SNNs. Furthermore, the reduced-voltage DRAM is also used to generate noise (i.e., from DRAM errors) for obfuscating the intellectual property (IP) against security threats, such as IP stealing (Xu et al., 2020 ). Our EnforceSNN framework can be put in the approximate computing field, especially in the context of the approximation for main memory through voltage scaling (Venkataramani et al., 2015 ; Mittal, 2016 ; Xu et al., 2016 ). Therefore, some of the techniques in our EnforceSNN are suitable for different domains outside SNNs: (1) DRAM voltage reduction for optimizing the DRAM access energy, (2) quantization for reducing the memory footprint, and (3) error-aware DRAM data mapping policy for minimizing the negative impact of DRAM errors on the data. These techniques are applicable for error-tolerant applications, such as image/video processing (e.g., data compression) and data analytic applications (e.g., data clustering).\n\n6.5. Further discussion Previous studies that exploit the reduced-voltage DRAM concept mainly aim at improving the energy efficiency of mobile systems (Haj-Yahya et al., 2020 ), personal computing systems (Nabavi Larimi et al., 2021 ; Fabrício Filho et al., 2022 ), and server systems (David et al., 2011 ; Deng et al., 2011 , 2012a , b ; Nabavi Larimi et al., 2021 ). This concept is also employed for minimizing the energy consumption of deep neural networks (DNNs) (Koppula et al., 2019 ). Since SNNs have different data representation, computation models, and learning rules as compared to DNNs, our EnforceSNN provides a different framework with different techniques that are crafted specifically for improving the resilience and the energy efficiency of SNNs. Furthermore, the reduced-voltage DRAM is also used to generate noise (i.e., from DRAM errors) for obfuscating the intellectual property (IP) against security threats, such as IP stealing (Xu et al., 2020 ). Our EnforceSNN framework can be put in the approximate computing field, especially in the context of the approximation for main memory through voltage scaling (Venkataramani et al., 2015 ; Mittal, 2016 ; Xu et al., 2016 ). Therefore, some of the techniques in our EnforceSNN are suitable for different domains outside SNNs: (1) DRAM voltage reduction for optimizing the DRAM access energy, (2) quantization for reducing the memory footprint, and (3) error-aware DRAM data mapping policy for minimizing the negative impact of DRAM errors on the data. These techniques are applicable for error-tolerant applications, such as image/video processing (e.g., data compression) and data analytic applications (e.g., data clustering)."
} | 8,533 |
34824906 | PMC8590396 | pmc | 290 | {
"abstract": "Background Corals, which form the foundation of biodiverse reef ecosystems, are under threat from warming oceans. Reefs provide essential ecological services, including food, income from tourism, nutrient cycling, waste removal, and the absorption of wave energy to mitigate erosion. Here, we studied the coral thermal stress response using network methods to analyze transcriptomic and polar metabolomic data generated from the Hawaiian rice coral Montipora capitata . Coral nubbins were exposed to ambient or thermal stress conditions over a 5-week period, coinciding with a mass spawning event of this species. The major goal of our study was to expand the inventory of thermal stress-related genes and metabolites present in M. capitata and to study gene-metabolite interactions. These interactions provide the foundation for functional or genetic analysis of key coral genes as well as provide potentially diagnostic markers of pre-bleaching stress. A secondary goal of our study was to analyze the accumulation of sex hormones prior to and during mass spawning to understand how thermal stress may impact reproductive success in M. capitata . Methods M. capitata was exposed to thermal stress during its spawning cycle over the course of 5 weeks, during which time transcriptomic and polar metabolomic data were collected. We analyzed these data streams individually, and then integrated both data sets using MAGI (Metabolite Annotation and Gene Integration) to investigate molecular transitions and biochemical reactions. Results Our results reveal the complexity of the thermal stress phenome in M. capitata , which includes many genes involved in redox regulation, biomineralization, and reproduction. The size and number of modules in the gene co-expression networks expanded from the initial stress response to the onset of bleaching. The later stages involved the suppression of metabolite transport by the coral host, including a variety of sodium-coupled transporters and a putative ammonium transporter, possibly as a response to reduction in algal productivity. The gene-metabolite integration data suggest that thermal treatment results in the activation of animal redox stress pathways involved in quenching molecular oxygen to prevent an overabundance of reactive oxygen species. Lastly, evidence that thermal stress affects reproductive activity was provided by the downregulation of CYP-like genes and the irregular production of sex hormones during the mass spawning cycle. Overall, redox regulation and metabolite transport are key components of the coral animal thermal stress phenome. Mass spawning was highly attenuated under thermal stress, suggesting that global climate change may negatively impact reproductive behavior in this species.",
"conclusion": "Conclusions This work contributes to our understanding of the coral response to thermal stress and the potential effects that a warming ocean will have on the reproductive health of these organisms. The early thermal stress response of M. capitata involves the downregulation of growth and DNA replication and the upregulation of signaling and the immune response. Later stages show downregulation of metabolite transport and biomineralization, as well as an upregulation of transcriptional regulators. Activation of animal redox stress pathways potentially as a mechanism for the detoxification of reactive oxygen species was found to be a major outcome of thermal stress. Whereas there was a noticeable increase in sex hormones ( e.g ., progesterone) in our samples prior to a natural mass spawning event, the release of egg-sperm bundles by M. capitata was highly attenuated in June 2019 (DB, HMP unpublished data), suggesting that thermal stress may negatively impact the reproductive behavior in this species. Significant effort will be needed to modify this polygenic trait in coral holobionts to boost resilience to thermal stress in the long term. Nonetheless, we have identified several novel genes that are promising candidates for functional analysis using the recently developed CRISPR/Cas9 tools for corals ( Cleves et al., 2018 ; Cleves et al., 2020 ). It is important to remember that the algal symbionts of corals play a key role in holobiont biology and stress response vis-à-vis symbiotic nutrient cycling ( Rädecker et al., 2021 ). Therefore, future gene-metabolite interaction analyses need to address in situ algal gene expression to address the integration of the host-symbiont response to thermal stress.",
"introduction": "Introduction Coral reefs are vitally important natural resources because they are home to about one-quarter of all marine biodiversity ( Reaka-Kudla, 1997 ) and support an estimated one-half to one billion people living in coastal communities by providing food, income from tourism, and coastal protection ( Woodhead et al., 2019 ). Since their radiation in the Middle Triassic period ~240 million years ago (Ma) ( Veron, 1995 ), stony corals have survived five mass extinction events ( Jackson, 2008 ). Their long-term survival underscores the inherent resilience of these holobionts in particular when considering the nutrient-poor marine environments in which they have thrived ( Frankowiak et al., 2016 ). The coral holobiont (meta-organism) is comprised of the cnidarian animal host, algal symbionts, fungi, microbial aggregates, and viruses. Under ambient conditions, the algal cells can provide up to 100% of host energy needs in the form of lipids, carbohydrates, and amino acids, as well as excess O 2 ( Falkowski et al., 1984 ). In return, excess nitrogen and inorganic waste from the coral animal, namely water, ammonium, and CO 2 , are recycled by the algae, fueling cell metabolism ( Yonge & Nicholls, 1931 ). Environmental shifts can lead to destabilization of the symbiosis (dysbiosis) between the coral animal and its partners because symbionts experience photo-oxidative stress and reduce provision of photosynthetic products. The coral animals may then expel their symbionts in the phenomenon known as “coral bleaching” ( Muscatine & Porter, 1977 ). The target of our study, the hermaphroditic, broadcast spawning Hawaiian coral Montipora capitata ( Fig. 1A ), is a robust species that resists bleaching, even under conditions causing mortality in more susceptible species ( Jokiel & Brown, 2004 ). The basis of bleaching resistance in M. capitata is yet to be fully explained but is most likely due to heterotrophic feeding ( Grottoli, Rodrigues & Palardy, 2006 ). 10.7717/peerj.12335/fig-1 Figure 1 Analysis of the rice coral Montipora capitata . (A) M. capitata photographed in waters near the Hawaiʻi Institute of Marine Biology (HIMB) in O‘ahu, HI. Photo credit: Debashish Bhattacharya. (B) Color scores and their standard errors for the ambient (green line) and high temperature (red line) treated M. capitata nubbins that were cultured in tanks at HIMB. Low color scores indicate the bleaching phenotype in coral holobionts. The omics data sampling points are shown with the white lines at T1 (5/22/19), T3 (6/03/19), and T5 (6/07/19) (for details, see Williams et al. (2021) ). The date of the New Moon in June 2019 is also shown. We subjected M. capitata nubbins (coral fragments) to thermal stress over a 5-week period, during which time transcriptomic and polar metabolomic data were collected at three different time points ( Fig. 1B ). The period of sampling (late May to early June 2019) coincided with the first of three annual mass-spawning events for M. capitata in the region. Therefore, genes and metabolites involved in coral reproduction were expected to be present in the RNA-seq and polar metabolomics data. We studied genes of both known and unknown function ( i.e ., ‘dark’) and investigated the temporal dynamics and biological shifts that sustain the coral animal under heat stress. Dark genes are either novel or too highly diverged (BLASTP e -value cut-off ≤ 1e −5 against the nonredundant NCBI database) to identify putative homologs in existing data, although some may encode a known domain associated with novel sequence ( Cleves et al., 2020 ). For example, 33% of dinoflagellate algal genes lack an annotation, but 1.4% of these unknown proteins contain a known domain ( Stephens et al., 2018 ). In our study, differentially expressed genes (DEGs) were filtered to only include reads which mapped to predicted M. capitata protein-coding genes ( Shumaker et al., 2019 ): i.e ., excluding algal RNA-seq reads. The animal data were integrated using networks to investigate molecular transitions in the coral. Network analysis can be a powerful framework for studying the structure of complex biological systems ( Williams et al., 2021 ) with nodes representing units at all levels of the biological hierarchy and edges, interactions between them, including transcriptional control, biochemical interaction, energy flow, and species interactions. Usage of DEGs allowed us to focus on the most consequential gene expression changes. Modules containing known genes with known functions were used to investigate their roles in the thermal stress response, as well as to identify dark genes which provide interesting potential candidates for future gene knockout or knockdown experiments.",
"discussion": "Discussion Coral reefs are under worldwide threat from warming oceans and local human-caused stressors such as over-fishing, the discharge of pollutants, and uncontrolled development ( National Academies of Sciences, Engineering, & Medicine, 2019 ). In response, many advances have been made in identifying individual gene and metabolite markers of coral thermal stress ( National Academies of Sciences, Engineering, & Medicine, 2019 ), but little has been done to link these two omics data sources. This is explained by the complexity of holobiont metabolomic interactions, combined with the massive number of dark genes and dark metabolites in corals for which currently no function, and therefore no causal relationship exists ( Williams et al., 2021 ). In addition, because metabolites are shared among holobiont members, obscuring metabolite origin, it is challenging to make biologically meaningful predictions from these data alone. For this reason, we used MAGI to find links between polar metabolite accumulation and gene expression. This approach provides a foundation for studying non-model systems by exploiting the consensus between metabolite identification and gene annotation to generate metabolite-gene associations ( Erbilgin et al., 2019 ). The MAGI analysis revealed the heightened response of the coral animal to redox stress, including the scavenging of excess molecular oxygen. The rate of metabolism at higher temperatures increases and can lead to physiological hyperoxia. Under elevated temperatures, oxygen absorbs excitation energy and becomes active in the form of superoxide radicals and hydrogen peroxide ( Lesser, 1997 ). These ROS, which are likely to be key contributors to coral thermal stress ( Cziesielski et al., 2018 ; Cleves et al., 2020 ), derive their reactivity from the unpaired electron. Hence, the enrichment of oxidoreductases is an expected outcome. Their catalysis solely involves the transfer of electrons; therefore, we postulate that corals utilize oxidoreductases to maintain redox homeostasis, remove excess molecular oxygen, and thereby, limit the production of ROS. In addition, we find evidence that progesterone metabolism may be implicated in the unsynchronized mass spawning events that have occurred at the study site in recent years ( Fogarty & Marhaver, 2019 ). Progesterone, a sex steroid, can be produced multiple ways, but usually involves β -hydroxylation reactions catalyzed by CYP enzymes ( Lu et al., 2020 ). Many examples of CYP enzymes metabolizing progesterones occur in metazoans ( Baker, 2001 ), such as CYP1A in humans ( Lu et al., 2020 ) and CPY 17 dehydrogenase (CYP17) in scleractinian corals ( Rougée, Richmond & Collier, 2015 ; Blomquist et al., 2006 ). There is evidence that sex steroids may regulate scleractinian reproduction ( Tarrant, Atkinson & Atkinson, 1999 ). CYP17 converts progesterone to androgens and Rougée, Richmond & Collier (2015) found that in the absence of thermal stress the enzymatic activity of CYP17 remained consistent over the lunar cycle in the brooding coral Pocillopora damicornis ( Twan et al., 2006 ). Twan, Hwang & Chang, 2003 found that the production of androgens increased prior to spawning in Euphylia ancora . The dysregulation of coral spawning due to environmental stress has been reported ( Fogarty & Marhaver, 2019 ) and occurred during the first mass spawning event for M. capitata around O‘ahu, HI in June 2019. Therefore, our results indicate that thermal stress, among other functions, affects the production of hormones contributing to reproductive activity. One of the most notable findings of the co-expression network analysis is that they are dominated by downregulated metabolite transport genes. The suppression of metabolite transport by the coral host may potentially be a response to reduction in algal productivity. More likely, it indicates redox stress, resulting from the animal host and/or algal symbionts, which leads to the generation of reactive species due to dysfunction in electron transport (see below; Roberty, Furla & Plumier, 2016 ). The inhibition of organic carbon production by the algae, precipitated by prolonged thermal stress ( Hillyer et al., 2017 ), can ultimately lead to their expulsion, resulting in bleaching ( Slavov et al., 2016 ). That is, in addition to a role in host processes, the coral animal may be dampening algal proliferation by reducing access to nutrients needed for growth such as ammonium, as demonstrated in the cnidarian model Aiptasia under the symbiotic stage ( Cui et al., 2019 ). This hypothesis conflicts with the findings of Fernandes de Barros Marangoni et al. (2020) who found that ammonium enrichment reduced thermal stress in the coral Stylophora pistillata and supported symbiont stability. This aspect may be less important for Hawaiian M. capitata that can meet 100% of its energy needs through heterotrophic feeding during periods of bleaching ( Grottoli, Rodrigues & Palardy, 2006 ). Our study provides important advances in the areas described above, however, three aspects of the results deserve further discussion. The first is the gene-metabolite interaction analysis of the phenylalanine-4-hydroxylase pathway in which BH 4 was unexpectedly absent in the MAGI results. Some plausible explanations for this result are as follows. In the kinetic model, P4H requires BH 4 , Phe, and O 2 to be bound, in that order ( Volner, Zoidakis & Abu-Omar, 2003 ). BH 4 binds first, converting the enzyme to its inactive form, E i , until sufficient Phe is present in plasma, at which point Phe binds and converts P4H to its activated form, E a ( Xia, Gray & Shiman, 1994 ); BH 4 bound to P4H would not have been detected in our analysis. Given that BH 4 is involved in other cellular functions it is possible that its levels might be depleted under heat stress, despite upregulation of the P4H pathway. This is relevant when considering the stoichiometry of the reaction, specifically, the number of BH 4 molecules needed as cofactors depends on cellular conditions. Higher pH and temperature may require more than one BH 4 to reduce two iron atoms ( Fitzpatrick, 2003 ), further reducing the number of free BH 4 molecules available for detection. It is also possible that another tetrahydropterin was used as an electron donor because BH 4 is primarily used to combat oxidative stress ( Kraft et al., 2019 ), potentially limiting its supply during high temperature stress. Existing data demonstrate the likely involvement of P4H in the symbiotic relationship between Hydra viridissima and its photosymbiont Chlorella sp. A99 ( Hamada et al., 2018 ). The second aspect is the impact of thermal stress on the coral reproductive cycle. Inspection of Fig. 5 shows that sex steroid accumulation is generally reduced under thermal stress, however, at T5, recover to near ambient and wild sample levels for several compounds ( e.g ., estrone, androstenedione, testosterone). This suggests that M. capitata may be able to partially acclimate to warming waters vis-à-vis sex steroid production, although these preliminary results need validation. More broadly, our results demonstrate that thermal stress impacts the production of hormones linked to reproductive activity. It is likely that the negative impact of environmental stress on coral mass spawning events will become more prevalent as oceans become warmer. Despite this not being the original intent of our study, the data we have generated provides valuable insights into how thermal stress disturbs the reproductive cycle of broadcast spawning corals. The consequences of this disturbance may have profound impacts not only on the health of existing reef ecosystems, but also on the ability of coral reefs to recover and recolonize an area after a major bleaching event or any environmental disturbance. The combined impact of thermal stress and mass spawning were addressed in our study, and it is possible that their interactions make it more difficult to interpret thermal stress impacts in isolation. Peak bleaching occurs in Hawaiian M. capitata in the month of October when the water temperatures are at their highest ( Cunning, Ritson-Williams & Gates, 2016 ). However, as our study in 2019 demonstrated ( Williams et al., 2021 ), local warming events can occur during mass spawning periods and will impact coral reproduction (current data). Therefore, rather than being weakened by the co-occurrence of warming and spawning in our study corals, we consider our data to be important for understanding how these combined stresses may impact future coral health as local warming events, like those we encountered, become more frequent. Finally, it should be noted that although the interaction between the coral and its algal endosymbionts represents the cornerstone of reef ecosystems, we chose to target the host animal response to thermal stress in this study. Whereas the metabolomic data analyzed here is derived from the whole holobiont ( i.e ., coral, algae, and other microbiome components), the RNA-seq data only captured transcripts from the eukaryotic component ( i.e ., coral and algae). The integration of the algal data was hindered by the lack of reference genomes for the endosymbionts of M. capitata and the likely presence of cryptic eukaryotic components of the holobiont that might contribute to non-animal RNA-seq data ( Kwong et al., 2019 ). A recent paper demonstrated that Symbiodiniaceae genomes are highly diverged, even between species in a single genus ( Symbiodinium ; González-Pech et al., 2021 ) and that multiple algal symbionts from different genera may reside in a single coral nubbin. Furthermore, metabolomes of the host and symbiont are not affected by variation in the abundance of the two algal symbionts that dominate Hawaiian M. capitata colonies ( i.e ., Durusdinium glynnii and Cladocopium spp.) ( Matthews et al., 2020 ). For these reasons, we concluded that the host response to thermal stress, reflecting the holobiont contribution, was the best target for this poorly characterized coral model. The results presented here provide a foundation upon which questions regarding coral-algal interactions during stress can be addressed in future studies."
} | 4,887 |
31690886 | PMC6976677 | pmc | 291 | {
"abstract": "Reef-building corals harbour an astonishing diversity of microorganisms, including endosymbiotic microalgae, bacteria, archaea, and fungi. The metabolic interactions within this symbiotic consortium are fundamental to the ecological success of corals and the unique productivity of coral reef ecosystems. Over the last two decades, scientific efforts have been primarily channelled into dissecting the symbioses occurring in coral tissues. Although easily accessible, this compartment is only 2–3 mm thick, whereas the underlying calcium carbonate skeleton occupies the vast internal volume of corals. Far from being devoid of life, the skeleton harbours a wide array of algae, endolithic fungi, heterotrophic bacteria, and other boring eukaryotes, often forming distinct bands visible to the bare eye. Some of the critical functions of these endolithic microorganisms in coral health, such as nutrient cycling and metabolite transfer, which could enable the survival of corals during thermal stress, have long been demonstrated. In addition, some of these microorganisms can dissolve calcium carbonate, weakening the coral skeleton and therefore may play a major role in reef erosion. Yet, experimental data are wanting due to methodological limitations. Recent technological and conceptual advances now allow us to tease apart the complex physical, ecological, and chemical interactions at the heart of coral endolithic microbial communities. These new capabilities have resulted in an excellent body of research and provide an exciting outlook to further address the functional microbial ecology of the “overlooked” coral skeleton."
} | 408 |
32482859 | PMC7306781 | pmc | 292 | {
"abstract": "Significance All plants and animals are associated with communities of viruses and microbes that interact via a suite of metabolites. These components play critical roles in the success of these assemblages; however, the role of individual components (i.e., bacteria, viruses, metabolites) and how these govern ecological interactions between macroorganisms is not understood. This study investigates the role of holobiont components in coral–turf algal interactions. The data demonstrate that an emergent microbiome and metabolome form at the interface between coral and turf algae in competitive interactions. Machine learning analyses show that this emergent community predicts the outcome of these interactions. These results provide insight into rules of community assembly in microbiomes and the roles of holobiont components in mediating ecological interactions.",
"conclusion": "Conclusion Statement Overall, this study demonstrates that there are differences in both the surface-associated microbial community and the total holobionts of coral and turf algae, and that when these organisms interact, there is an emergent interface community. We hypothesize that this emergent community is driven by the coral microbiome feeding on the energy-rich exudates released by the adjacent turf algae, a phenomenon we term the algae feeding hypothesis. The data also show that specific bacterial groups such as Bacteroidetes and Firmicutes play a role in determining the competitive outcome of coral–turf algae interaction events. However, what this role is remains an open question and will require further investigation. In sum, we emphasize the role of host-associated microbial communities in ecological processes and highlight that the holobiont plays an important part in determining the outcome of coral–turf algal interactions and overall reef health.",
"discussion": "Discussion The data presented herein illustrate that there are significant differences in the size, abundance and community composition of microbes across in situ coral–turf algal interfaces ( Fig. 1 ). These differences show there is an emergent community that forms at the interface between coral and turf algae, which is characterized by larger and more numerous bacterial cells, a higher proportion of Bacteroidetes, a lower proportion of Firmicutes, an enrichment in genes involved in bacterial cell growth and division, and an increase in the potentially proapoptotic compound ceramide 18:1/16:0. Microbial and Viral Abundances. The highest viral abundance was in the coral holobiont. The increased viral abundance in the coral-surface holobiont may be due to the bacteriophage adherence to mucus (BAM) dynamics described in Barr et al. ( 42 ). BAM dynamics imply that bacterial viruses (bacteriophage or phage) adhere to mucus glycoproteins through noncovalent interactions with capsid proteins. Corals may use these mucus-attached phages to defend against invading bacterial pathogens. The combination of the differences in viral and bacterial abundances leads to a significant difference in the VMR across the interface, where coral holobionts have the lowest microbial load but the highest VMR ( Fig. 1 D ). Furthermore, metagenomic analyses demonstrated higher levels of prophage in algal samples ( SI Appendix , Fig. S20 ), where VMR was the lowest ( Fig. 1 D ). Taken together, these data suggest a trend toward a decrease in lytic activity and an increase in viral lysogeny at higher microbial concentrations ( Fig. 1 E and SI Appendix , Fig. S20 ). A similar trend has been seen in other environments ( 43 , 44 ), including the water column of tropical coral reefs, and has been described as Piggyback-the-Winner dynamics ( 43 ). Piggyback-the-Winner posits that when VMRs are low, such as in interface and turf algal samples, there are more bacterial cells harboring lysogenic phage. This means that the bacterial assemblages at the interface and turf algae harbor a higher proportion of phage-encoded genes, which has been linked to increased pathogenicity of the overall community ( 45 – 50 ). These microbe-phage dynamics may be another mechanism at play in the complex interactions of coral and algal holobionts. Microbial Biomass and Energetics. The results also demonstrate that there are significantly larger microbial cells at the interface between coral and turf algae ( Fig. 1 F ). This change in cell size coupled to the cell concentrations leads to a higher predicted metabolic power output at the coral–algal interface ( Fig. 1 G ). This trend of higher power output has also been reported using calorimetry in controlled laboratory experiments ( 22 ). A higher microbial power output at the coral–algal interface means that the microbial populations here are using energy at a faster rate and are dissipating more of that energy as heat ( 22 , 51 ). The increase in metabolic rate at the interface may be responsible for the reported decreased oxygen levels at the interface ( 22 , 52 – 55 ). Understanding the direct links between bacterial taxa, cell size, power output, and biological oxygen demand may provide a more complete conceptual model of the way bacterial metabolism is involved in competitive interactions between benthic macroorganisms. Emergent Microbiome and Metabolome at the Coral–Turf Algal Interface. Metabolomic samples showed clear differences between coral and noncoral holobionts. The interface exhibited a unique chemical signature, however, the metabolites driving differences at the interface were mostly unknown compounds. One known compound (level 2 according to the metabolomics standards initiative) was the potentially proapoptotic molecule, ceramide 18:1/16:0 ( Fig. 3 A ). Other bioactive lipids and proapoptotic inflammatory modulators have previously been shown to play a role in the coral holobiont ( 56 – 58 ), suggesting that nonself-recognition among some of the oldest extant holobionts (i.e., corals) involves bioactive lipids identical to those in highly derived taxa like humans. The data here further strengthen the hypothesis that major players of the immune response evolved during the pre-Cambrian era ( 59 ). Furthermore, turf algal metabolites were found to have more negative nominal oxidation states of carbon and higher ∆G of carbon oxidation ( Fig. 3 C and D ), suggesting that the biochemicals in the turf algal holobiont are more reduced and, thus, more energy rich. It may be the combination of naive coral microbes being exposed to high-energy turf algal compounds at the interface, which leads to the increase in size and power output of the bacterial cells at the interface ( Fig. 1 F and G ). The feeding of coral microbes on turf algal metabolites at the interface may also be in part responsible for the decrease in oxygen levels previously observed at the coral–algal interface ( 22 , 53 ). Thus, we propose the “algae feeding hypothesis” where reduced, high-energy turf algae exudates feed interface and coral-associated microbial communities, often to the detriment of the coral animal. The metagenomic and metabolomic data show that there are specific bacterial taxa, functional genes, and metabolites that distinguish coral, turf algae, and interfaces ( Fig. 2 ), as well as winning and losing interactions ( Fig. 4 and SI Appendix , Tables S4 and S5 ). Furthermore, these data indicate that the interface is not merely a mix of coral and turf algal holobionts, but rather has its own emergent signature ( SI Appendix , Figs. S2–S6 ), which is more similar to the turf algal holobiont than the holobiont of coral ( Fig. 2 ). Specifically, members of the Bacteroidetes clade are overrepresented at the interaction interface, while the phylum Firmicutes is underrepresented at the interface ( Fig. 5 and SI Appendix , Fig. S7 ). A similar trend is observed in coral samples, where Bacteroidetes are enriched in losing corals and Firmicutes are depleted in losing coral samples ( SI Appendix , Fig. S7 ). Thus, the data demonstrate that the Bacteroidetes-to-Firmicutes ratio is a significant predictor of whether a coral will win or lose in a competitive interaction with algae. The ratio of Bacteroidetes to Firmicutes is also a significant predictor of health status in other systems such as the human gut where this ratio has been linked to obesity ( 60 ) and in the human lung where it is linked to disease states in cystic fibrosis patients ( 61 ). The Bacteroidetes-to-Firmicutes ratio was a significant predictor of the amount of cell division genes and total microbial biomass in these holobionts ( Fig. 5 D ). Studies in mice and humans have shown that the change in the Bacteroidetes-to-Firmicutes ratio can have significant impacts on energy output and biomass of microbial communities, with Bacteroidetes having an increased capacity to harvest energy from reduced compounds ( 62 ). Given that the abundance of Bacteroidetes and the Bacteroidetes-to-Firmicutes ratio was higher at the interface and in losing coral samples, we propose a working model ( Fig. 6 ) whereby the reduced metabolites released by turf algae select for an increase in Bacteroidetes relative to Firmicutes, which, in turn, leads to a faster growing bacterial community with larger cells and higher energy use rate. These fast-growing microbes can outcompete the corals for resources such as oxygen, which weaken the coral and lead to eventual algal overgrowth. This link between the energy content of algal metabolites, bacterial taxonomic composition, community metabolism, and coral health provides interesting insight into the ways that different components of the holobiont affect the outcome of ecological interactions and eventually shape entire community assemblages. Fig. 6. Working model of how different components of coral and turf algal holobionts mediate ecological interactions on the reef. Viruses are most abundant in the coral epibiont while bacteria are most abundant at the interface. This leads to the highest VMR in the coral holobiont and the lowest VMR in the turf algal holobiont. When coupled to the increased abundance of prophage found in the turf algal metagenomes this suggests increased lysogeny in the turf algal holobiont and increased lytic activity in the coral holobiont. Bacterial cell size, microbial metabolic power output, bacterial cell division, and the Bacteroidetes-to-Firmicutes ratio is highest at the interface. The potentially proapoptotic metabolite ceramide 18:1/16:0 is found only in the coral and interface samples but is most abundant at the interface. Despite the current progress in the field, it is worth noting that environmental multiomics still has a long way to go. Metabolomics databases are sparsely populated in regard to environmental metabolites making annotation difficult and leaving the majority of compounds unclassified. As this work and others (e.g., ref. 63 ) have demonstrated the need to consider all components of the holobiont, it is clear that new methods and increased sequencing efforts will be needed to provide the amount of microbial coverage necessary to properly describe the roles of the less abundant components of the holobiont such as archaea and viruses. Thus, we highlight that future work is needed to provide more robust analyses of the coral and algal holobionts and their associated metabolites."
} | 2,831 |
23762518 | PMC3678486 | pmc | 293 | {
"abstract": "Spatially intimate symbioses, such as those between scleractinian corals and unicellular algae belonging to the genus Symbiodinium , can potentially adapt to changes in the environment by altering the taxonomic composition of their endosymbiont communities. We quantified the spatial relationship between the cumulative frequency of thermal stress anomalies (TSAs) and the taxonomic composition of Symbiodinium in the corals Montipora capitata , Porites lobata , and Porites compressa across the Hawaiian archipelago. Specifically, we investigated whether thermally tolerant clade D Symbiodinium was in greater abundance in corals from sites with high frequencies of TSAs. We recovered 2305 Symbiodinium ITS2 sequences from 242 coral colonies in lagoonal reef habitats at Pearl and Hermes Atoll, French Frigate Shoals, and Kaneohe Bay, Oahu in 2007. Sequences were grouped into 26 operational taxonomic units (OTUs) with 12 OTUs associated with Montipora and 21 with Porites . Both coral genera associated with Symbiodinium in clade C, and these co-occurred with clade D in M. capitata and clade G in P. lobata . The latter represents the first report of clade G Symbiodinium in P. lobata . In M. capitata (but not Porites spp.), there was a significant correlation between the presence of Symbiodinium in clade D and a thermal history characterized by high cumulative frequency of TSAs. The endogenous community composition of Symbiodinium and an association with clade D symbionts after long-term thermal disturbance appear strongly dependent on the taxa of the coral host.",
"introduction": "Introduction Recent global warming has contributed to changes in species ranges and community composition (Walther et al. 2002 ; Perry et al. 2005 ; Hoegh-Guldberg and Bruno 2010 ). For corals, the warming of the oceans temperature has resulted in an increase in the frequency and magnitude of bleaching events (Hoegh-Guldberg 1999 ). Bleaching is the paling of the external coloration of corals that reflects the breakdown of their obligate endosymbiosis with dinoflagellates in the genus Symbiodinium (Hoegh-Guldberg and Smith 1989 ). Bleaching often precedes the death of corals, and widespread bleaching events have driven mass coral mortality in some regions of the world (Hoegh-Guldberg 1999 ). The rapidly changing ocean environment has potentially dire consequences in the near future for reef ecosystems and the IUCN recently estimated that one third of reef corals are under an elevated threat of extinction (Carpenter et al. 2008 ). A better understanding of how corals could adapt and survive through changing ocean conditions is critical to developing predictions of species composition in future reef ecosystems. The genus Symbiodinium is genetically diverse comprising nine evolutionary lineages referred to as clades A–I (Pochon and Gates 2010 ). The taxonomic composition of endosymbiotic Symbiodinium in corals is broadly recognized as an important factor that contributes to the environmental threshold of the host coral (Baker 2003 ; Berkelmans and van Oppen 2006 ; Stat et al. 2006 ). For example, corals such as Acropora and Pocillopora spp. that harbor clade D Symbiodinium show a higher thermal tolerance and resistance to bleaching than conspecifics with symbiotic communities dominated by clade C (Rowan 2004 ; Berkelmans and van Oppen 2006 ; but see Abrego et al. 2008 ). In addition, there have also been reports of symbiont community shifts in corals to clade D on reefs that have recently experienced bleaching and high ocean temperatures (Baker et al. 2004 ; Rowan 2004 ; Berkelmans and van Oppen 2006 ; Jones et al. 2008 ). These observations point toward the potential importance of Symbiodinium clade D in corals' adaptive response to changes in the environment. However, depressed growth rates in juvenile corals associated with clade D Symbiodinium , as compared with conspecifics in symbiosis with clade C (Little et al. 2004 ), have raised questions about the long-term benefits and/or ecological implications of hosting different Symbiodinium strains (Stat et al. 2008a ; Cantin et al. 2009 ; Mieog et al. 2009 ; Jones and Berkelmans 2010 ; Ortiz et al. 2013 ). While some studies have shown that the abundance of Symbiodinium clade D in corals increases during thermal stress and during recovery following bleaching (Jones et al. 2008 ; LaJeunesse et al. 2009 ), others have shown that the symbiont community in corals do not change under such conditions (Thornhill et al. 2006 ; LaJeunesse et al. 2007 ; Costa et al. 2008 ; Stat et al. 2009a ). These inconsistencies point to the importance of the magnitude and duration of the stress and host-specific responses, as factors that shape the Symbiodinium communities in corals during and following bleaching (Goulet 2006 ; Stat and Gates 2011 ). It has also been shown that while the Symbiodinium in corals can become dominated by clade D under stress, reversion back to the original population in the absence of stress occurs in subsequent years, a feature indicating that chronic temperature stress is required to maintain symbioses dominated by clade D (Thornhill et al. 2005 ). More recently, the effects of thermal stress on Symbiodinium communities in corals over longer periods of time using remote-sensing satellite data (i.e., tens of years) as opposed to shorter periods such as a bleaching event (i.e., 1–2 years) is an alternative approach to investigating how ocean temperature influences the community composition of Symbiodinium . Oliver and Palumbi ( 2009 ) used remote-sensing information on ocean sea surface temperature for 1998–2006 from NOAA's Pathfinder v5 satellite data and investigated whether the number of degree heating weeks (DHWs) correlated with a greater abundance of Symbiodinium clade D in Acroporid corals from American Samoa, Fiji, Palmyra Atoll, and the Philippines. Interestingly, they showed that while Fiji yielded the greatest number of DHWs from the regions investigated, clade D was absent, and was only found in American Samoa, an area that had experienced threefold less DHWs. In American Samoa though, clade D abundance was correlated with higher ocean temperatures. The authors' interpreted spatial differences in the correlation between clade D and the history of ocean thermal stress to other factors; notably, local environmental conditions linked to a region. Cooper et al. ( 2011 ) identified water clarity and sediment type as one local condition influencing the distribution of clade D in Acropora from the Great Barrier Reef, showing that sea surface temperature anomalies did not explain the abundance and distribution of Symbiodinium clade D alone. In a global assessment by Selig et al. ( 2010 ) on the frequency of thermal stress anomalies (TSAs) using NOAA's Pathfinder v5 dataset, the Hawaiian archipelago was shown to have some of the lowest frequencies and shortest durations of thermal stress events in the Pacific, although it experienced relatively high magnitudes, between 1985 and 2005. Coral reef ecosystems in the Hawaiian archipelago are dominated by five coral species, and of these corals, Porites lobata , Porites compressa , and Montipora capitata are among the most widespread and abundant (Fenner 2005 ; pers. obs.). Montipora in the Pacific and in some areas of Hawaii associates with Symbiodinium in clades C and D, however, the latter is extremely rare in Porites in the Pacific and has only ever been reported in two colonies from Palau (Fabricius et al. 2004 ; LaJeunesse et al. 2004a ; Stat et al. 2011 ; Franklin et al. 2012 ). The aim of this study was to determine whether a higher frequency of cumulative TSAs is correlated with a higher occurrence of Symbiodinium clade D in Porites and Montipora across the Hawaiian archipelago.",
"discussion": "Discussion Association of Symbiodinium clade D with Montipora and Porites The algal endosymbiont, Symbiodinium clade D associates with corals in the genus Montipora but not Porites in Hawaii. The occurrence of clade D in Montipora correlates with an area that has experienced the highest recorded frequency of TSAs for lagoonal habitats in Hawaii. This spatial distribution suggests that thermal stress may influence the distribution of clade D in this coral. While the cause behind the correlation between clade D and ocean temperature stress remains unclear, a shift in response to recent environmental conditions, notably the frequency of TSAs in the region, is one plausible explanation. A community shift of coral endosymbionts toward a population dominated by clade D in response to elevated ocean temperature stress is consistent with reports from other corals in the Pacific, including Acropora and Pocillopora , and is consistent with regional reports of an increase in the abundance of clade D on reefs that have recently experienced thermal stress (Baker et al. 2004 ; Jones et al. 2008 ; LaJeunesse et al. 2008 ). Clade D can also be common in corals located in an environment characterized by relatively high ocean temperatures compared with most other regions where corals are found, such as the Persian Gulf where the temperature reaches 33°C (Mostafavi et al. 2007 ; LaJeunesse et al. 2010 ; Stat and Gates 2011 ). As the ocean temperature of Hawaiian reefs do not fall into this category and remain under 30°C (NOAA National Weather Service), and the occurrence of clade D in Montipora is extremely rare globally (Franklin et al. 2012 ), it is unlikely that this association is the result of long-term local adaptation to challenging ocean temperatures, but rather a response to recent TSAs. Furthermore, the annual incidence of TSAs in the years leading up to sampling implies that chronic temperature stress in the absence of bleaching can provide a competitive advantage for clade D Symbiodinium to persist in these corals in Kaneohe Bay. This is consistent with the observation on the abundance of clade D increasing in corals during ocean warming prior to bleaching (LaJeunesse et al. 2008 ). It is also possible that anthropogenic impacts, like pollution and sedimentation, have contributed to the occurrence of clade D Symbiodinium in Montipora at Kaneohe Bay (Cooper et al. 2011 ; Stat and Gates 2011 ). Oahu is the most populated island in Hawaii, and Kaneohe Bay specifically has been exposed to high levels of pollution (Hunter et al. 1995 ). In contrast, French Frigate Shoals and Pearl and Hermes are located within the Paphānaumokuākea Marine National Monument, a protected marine environment that is arguably one of the least impacted coral reef ecosystems in the world. Future work will focus on the occurrence of clade D on the island of Oahu, and investigate whether TSA or other anthropogenic impacts like pollution, or their synergism, accounts for the higher abundance of clade D in Kaneohe Bay. Even though clade D sequences were found associated with Porites ( n = 2), the extremely low abundance (0.2% of sequences) could suggest that they represent surface contaminants, although it is impossible to rule out the possibility that they are low abundant endosymbionts (Mieog et al. 2007 ; Silverstein et al. 2012 ). In the Pacific, clade D is rare in Porites and has only been identified in two colonies from Palau (Fabricius et al. 2004 ), even though genotyping of its Symbiodinium community extends to numerous regions including the southern, central, and northern Great Barrier reef, Johnston Atoll, Japan, Guam, Hawaii, and American Samoa (LaJeunesse et al. 2003 , 2004a , b ; Apprill and Gates 2007 ; Stat et al. 2008b ; Barshis et al. 2010 ; Pochon et al. 2010 ; Franklin et al. 2012 ). While clade D in Porites from Palau originated from colonies in a chronically warm environment, most Porites in the study from that location (and others) harbored clade C. Why some corals show flexibility in their symbioses and a shift toward clade D under certain environmental conditions and others do not (Thornhill et al. 2006 ; LaJeunesse et al. 2007 , 2008 ; Costa et al. 2008 ; Jones et al. 2008 ; Stat et al. 2009a ; McGinley et al. 2012 ) remains unclear. The corals used in this study are highly abundant in Hawaii and both occupy similar environments and acquire their symbionts via maternal transmission; however, they show very different affinities for clade D. One explanation may lie in the dependency of the host for their endosymbiotic community. Porites and Montipora show differences in the dependency for their endosymbiont population, especially during periods of thermal stress (Grottoli et al. 2006 ; Rodrigues and Grottoli 2007 ). Montipora capitata shifts from autotrophy to heterotrophy during episodes of thermal stress and bleaching. In contrast, P. compressa and P. lobata are highly autotrophic, do not make a significant transition to heterotrophy, and thus rely on their endosymbiotic population for nutrients during recovery. The switch to heterotrophy in Montipora supports a less specific association in Montipora and a lower dependency on their endosymbionts, which may partly explain the observed community shift from their dominant C31 symbiont to clade D in areas of high thermal history in Hawaii. However, if the switch to clade D Symbiodinium allows corals to adapt to environmental change and increases their thermal tolerance (i.e., symbiont dependence; Jones and Berkelmans 2010 ) then the concurrent switch to heterotrophy (i.e., symbiont independence) during such conditions is somewhat of a paradox. One explanation is that clade D may provide the host with a reduced amount of nutrients, but enough to supplement the amount acquired through host heterotrophy under periods of stress and collectively equating to the amount needed to sustain the host. This is consistent with the opportunistic nature of clade D and reports of less carbon that is translocated to the host by clade D compared with clade C Symbiodinium (Cantin et al. 2009 ). The corals' tolerance or susceptibility to changes in the environment is therefore a culmination of numerous factors that includes but is not limited to (a) the dynamics of host–symbiont assemblages; (b) the differential survival of symbionts under varying conditions; (c) the contributions of various symbionts to the host; and (d) the dependence of the host for their Symbiodinium community and the ability to make a transition to heterotrophy. Symbiodinium diversity inferred using OTUs We applied a sequence similarity threshold to group Symbiodinium sequences into OTU's to assess diversity. As in other taxa, some OTU groups may represent a species cluster or functional group, while others may combine species or represent subspecies. This inconsistency reflects the biological diversity of organisms and the lack of a uniform genetic divergence that delimits species boundaries. Also, grouping sequences into OTUs based on sequence similarity does not overcome all the problems associated with PCR artifacts and intragenomic variation (Thornhill et al. 2007 ; Stat et al. 2011 ). However, these caveats are not limited to Symbiodinium and are common across taxa, and this method is a widely utilized approach for analyzing cloned amplicons from environmental populations of prokaryotes, basal eukaryotes (e.g., Landeweert et al. 2003 ; Bjorbækmo et al. 2010 ; Brazelton et al. 2010 ), and more recently to diversity studies using next-generation sequencing in eukaryotes (e.g., Blaalid et al. 2012 ). Statistical parsimony networks were constructed using representative sequences from each OTU for the Symbiodinium clades that were identified in this study (A, C, D, and G). As expected, the relationship among OTUs is similar to the phylogenies constructed using ITS2 types identified using the dominant band in DGGE fingerprints (Pochon et al. 2007 ; LaJeunesse et al. 2008 ), but with a reduction in complexity. Furthermore, a comparison of the OTUs and their evolutionary relationship compared with the “species clusters” in ITS2 networks identified by Correa and Baker ( 2009 ) using a different method are very similar. This study utilizes more sequence data to infer OTUs as an outcome of increased diversity discovered since the Correa and Baker analysis in 2009 and the incorporation of cloning data that increases the likelihood of detecting low abundant symbionts and/or intragenomic variation. This added sequence data likely contributes to the differences in the number of OTU's identified in this study (A:6, C:41, D:2) and the number of “species clusters” that were inferred by Correa and Baker ( 2009 ; A:7, C:23, D:1). As the diversity of Symbiodinium observed increases, coupled with the amount of genetic data that will likely flood future analysis due to next-generation sequencing platforms, cluster-based approaches to infer Symbiodinium diversity will be a necessity. Symbiodinium diversity in Montipora and Porites As with the majority of corals in the Pacific, Porites and Montipora predominantly associate with clade C Symbiodinium , while clade D is occasionally found in Montipora (LaJeunesse 2005 ; Stat et al. 2009b , 2011 ; Franklin et al. 2012 ; this study). Furthermore, Porites and Montipora show specificity with Symbiodinium strains within clade C. In Porites , endosymbionts belonging to the C15 or the C15-like symbiont cluster are found throughout the Pacific (LaJeunesse et al. 2003 , 2004a , b ; LaJeunesse 2005 ; Stat et al. 2008b , 2009b ; Barshis et al. 2010 ; but see Wicks et al. 2010 ). This ubiquitous distribution of a specific host–symbiont association over a large biogeographic area infers a long-standing association that has developed over evolutionary timescales. Interestingly, Montipora predominantly associates with C31 in Hawaii, but in the Great Barrier Reef it associates with C31, and C15 – the symbiont found in Porites throughout the Pacific (LaJeunesse et al. 2004a ; Stat et al. 2008b ). Why C15 associates with Porites but not Montipora in Hawaii remains unknown. In addition, novel Symbiodinium OTUs in Porites that were not found in Montipora (C15.21–C15.29, Fig. 3 ) form a monophyletic group with C15 at the root. This implies that an intimate association between C15 and Porites in the remote Hawaiian Islands is providing the opportunity for the radiation of new symbiont lineages in the C15 cluster that is specific to this host. A paradox that exists in the specificity of coral–algal symbioses becomes evident when extending the observed interactions beyond clade C. Coral hosts belonging to a variety of genera show specificity to unique Symbiodinium types or lineages within clade C (e.g., Montipora and C31, Pocillopora and C42; LaJeunesse et al. 2004a ). The same host genera, however, are also found in unions with different Symbiodinium clades, specifically clades A and D (LaJeunesse et al. 2007 ; Stat et al. 2009b ). Therefore, while there is apparent specificity among closely related symbionts within clade C, the barrier to specificity breaks down among clades. The Symbiodinium associations found in this study also support these observations. In addition to the specificity between clade C Symbiodinium and the hosts Montipora and Porites , clade D was found associated with Montipora , while clades A, D, and G were found associated with Porites . While clade A (and clade B) can be the dominant symbiont in Porites from the Caribbean (Thornhill et al. 2006 ; Finney et al. 2010 ), its occurrence in Porites in the Pacific is extremely rare. As only a single clade A sequence was recovered, the occurrence of this Symbiodinium lineage in Porites likely represents a surface contaminant or low abundant endosymbiont, like clade D in Porites . The association between Porites and clade G Symbiodinium at French Frigate Shoals is a very interesting observation. This lineage of Symbiodinium is usually found in Foraminifera, sponges and soft corals (van Oppen et al. 2005 ; Pochon et al. 2007 ; Granados et al. 2008 ; Hill et al. 2011 ), and has only been found to associate with single colonies of the corals Coeloseris and Montastraea in the Indian Ocean (LaJeunesse et al. 2010 ). Interestingly, Porites are often bioeroded by sponges that associate with clade G Symbiodinium , and the shared interaction with clade G by these hosts may reflect this three-way interaction (Sammarco and Risk 1990 ; Granados et al. 2008 ). This study presents evidence for the differential association of the algal endosymbiont Symbiodinium clade D and two dominant corals in Hawaii. While clade D can occur as the dominant symbiont in Montipora , it is nearly absent in Porites . Furthermore, the distribution of clade D correlates with the region that has experienced the greatest history of thermal stress, providing additional evidence for the observation of this Symbiodinium lineage in areas where ocean conditions are challenging for corals or have recently experienced ocean warming. This study also adds to the accumulating evidence for the interaction of multiple Symbiodinium clades with hosts that have been perceived as forming specific symbioses, and calls for a reassessment of what defines specificity in coral–algal symbioses (Silverstein et al. 2012 ). The difference in abundance and distribution of clade D, and the presence of multiple clade lineages in addition to the dominant-specific symbiont of these corals (i.e., C15 for Porites in Hawaii, and C31 for Montipora ) highlight the biological complexity of these unions. The dependence of the host for their endosymbiotic community, the contribution of different Symbiodinium clades and subclades to their host, and how changes in the environment effect these interactions will be a focus of future research investigating the adaptive potential of corals."
} | 5,512 |
19623250 | PMC2708352 | pmc | 294 | {
"abstract": "Coral bleaching, during which corals lose their symbiotic dinoflagellates, typically corresponds with periods of intense heat stress, and appears to be increasing in frequency and geographic extent as the climate warms. A fundamental question in coral reef ecology is whether chronic local stress reduces coral resistance and resilience from episodic stress such as bleaching, or alternatively promotes acclimatization, potentially increasing resistance and resilience. Here we show that following a major bleaching event, Montastraea faveolata coral growth rates at sites with higher local anthropogenic stressors remained suppressed for at least 8 years, while coral growth rates at sites with lower stress recovered in 2–3 years. Instead of promoting acclimatization, our data indicate that background stress reduces coral fitness and resilience to episodic events. We also suggest that reducing chronic stress through local coral reef management efforts may increase coral resilience to global climate change.",
"introduction": "Introduction Ecological studies have demonstrated that stressors currently affecting coral reefs include, among others, coral diseases [1] , over-fishing [2] , and a combination of pollution and sedimentation from coastal development [3] . These chronic stressors are often associated with the gradual loss of coral cover and overgrowth by fleshy algae. However, abrupt and severe episodic events, such as coral bleaching, may also be responsible for coral reef degradation [4] . An outstanding issue is whether the combination of multiple stressors reduces coral resistance or resilience to episodic events such as bleaching [5] – [7] , or alternatively whether acclimatization to stressful conditions can increase coral resistance—the ability of corals to withstand future stress [8] , [9] . Bleaching is a generalized term for the loss of symbiotic dinoflagellate zooxanthellae or their pigments in scleractinian corals and is typically associated with sustained, unusually warm water temperatures [10] . Several studies have found that bleaching reduces skeletal growth in corals [11] – [13] . Ocean acidification may also reduce the ability of corals to calcify as normal; a recent study of corals from the Great Barrier Reef attributes a 14.2% decrease in calcification since 1990 to a combination of acidification and warming [14] . Here we define resistance as an individual coral's ability to continue normal skeletal growth even under stress (whether chronic or episodic), and resilience as a coral's ability to recover to normal growth rates after a stressful event ( Fig. 1a–c ). To test our hypothesis that chronic local stress reduces coral resistance and resilience to bleaching, we focus on coral growth before and after the 1998 mass-bleaching event [15] from four sites on the Mesoamerican Reef ( Fig. 2 , Table 1 ) with relatively high and low chronic local stress. 10.1371/journal.pone.0006324.g001 Figure 1 X-radiographs of various coral cores showing the different types of growth behavior discussed. (A) Coral without the 1998 growth suppression, indicating resistance to bleaching in 1998. (B) Coral with the 1998 growth suppression, recognized by the bright high-density band, but with a quick return to pre-1998 extension rates, indicating resilience after bleaching. (C) Coral with the 1998 growth suppression and continuing depressed extension rates after 1998, indicating a lack of both resistance and resilience to bleaching. (D) A coral with relatively high average extension rate. (E) A coral with relatively low average extension rate. (F) A coral with a partial mortality scar on the left (noted by white arrow), coincident with the 1998 growth anomaly. 10.1371/journal.pone.0006324.g002 Figure 2 Map of the Mesoamerican Reef with locations of coral collections as black circles. Dark grey denotes coral, light grey denotes land areas. T1, T2 = Turneffe Atoll (4 cores from T1, 13 from T2), S = Sapodilla Cayes (44 cores), U = Utila (17 cores), C = Cayos Cochinos (14 cores). 10.1371/journal.pone.0006324.t001 Table 1 Coral core collection site locations. Site Dive Site Name Coordinates Total Cores 1998 Stress Band Partial Mortality in 1998 Previous Stress Bands Turneffe 1 Dog Flea Caye 17°29′59″N, 87°45′30″W 17 71% (12) 6% (1) 2 Turneffe 2 Harry Jones 17°18′25″N, 87°48′04″W Sapodilla Frank's Caye, NE buoy 16°07′45″N, 88°14′59″W 44 100% (44) 16% (7) 0 Utila Diamond Caye 16°03′52″N, 86°57′30″W 17 100% (17) 12% (2) 1 Cayos Cochinos Pelican Point, Peli 2 15°58′41″N, 86°29′06″W 14 100% (14) 21% (3) 0 \n Total \n \n 92 \n \n 95% (87) \n \n 14% (13) \n \n 3 \n Location names with dive site name or nearby caye, coordinates, number of cores from each site, along with growth anomalies in 1998 and earlier. Table lists the total number of cores which were drilled and slabbed along the growth axis, the percentage and number of these that have dense stress bands associated with the 1998 event, the percentage and number that contained a partial mortality scar, and any previous individual stress bands.",
"discussion": "Results and Discussion At all four sites, nearly every coral displays a visually prominent stress band in 1998 [11] ( Fig. 1b ), indicated by an increase in skeletal density and decrease in extension and calcification rates more than four standard deviations outside the mean chronology. Indeed, 95% of coral cores show a marked reduction in extension rates that persists for two years or more following the 1998 event ( Fig. 3 , Table 1 ). Additional rare stress bands (denser than the long-term mean by at least 1.5 standard deviations) are seen in three individual coral cores in 1950, 1965 and 1995. However, no other coral cores displayed significant changes in growth during these years, nor did we observe any partial mortality scars before 1998. Although the 1998 bleaching event affected the entire Mesoamerican reef system, there are spatial differences in its intensity revealed in our coral cores. We compared the number of coral cores with partial mortality scars ( Fig. 1f ) and 1998 stress banding between sites using a permutation test. Turneffe Atoll, the site with the lowest level of local stress, is also the only site without ubiquitous stress banding (p<0.05) ( Fig. 1a ), and has the lowest, but non-significant, frequency of partial mortality scars ( Table 2 ). The lack of stress banding in some coral cores at Turneffe Atoll suggests that at least some of those corals may have resisted bleaching. By comparison, the universal occurrence of stress banding at all other sites supports the hypothesis that high chronic stress decreases coral resistance to bleaching. In contrast to the modest variability in resistance to the 1998 bleaching event, there are large differences in the resilience of corals between sites. At Sapodilla and Utila where stress indices are high, coral growth still did not recover completely by the time of collection more than eight years following the 1998 bleaching event, even controlling for long-term decreasing trends at these sites (GLM: Sapodilla, F (2,50) = 20.70, p<0.001, R 2 = 0.45; Utila, F (2,50) = 18.26, p<0.001, R 2 = 0.42, Table S1 ). Such a long recovery period is unprecedented in the literature, with most studies reporting growth suppression due to bleaching on the order of one year [11] – [13] and the longest growth suppression reported for four years [22] . In comparison, corals from lower-stress sites (Turneffe and Cayos Cochinos) recovered to pre-bleaching extension rates in about three years ( Fig. 3 , Table S1 ). However, at sites with high local stress, corals have been unable to recover to pre-disturbance growth rates. The between-site differences in both the level of impact of the 1998 bleaching event and the subsequent recovery time are not easily explained by between-site differences in the strength of the 1998 bleaching event. We tested the hypothesis that our sites experienced differences in heat stress in 1998 by calculating the degree-heating-weeks (DHW) [23] for our four study areas from 7-day composite night time sea surface temperature data [supplementary information Table S1 ]. Our findings indicate that during 1998, heat stress was higher at Cayos Cochinos and the Sapodilla Cayes (6.84 and 5.54 maximum DHW, respectively), compared to Utila and Turneffe Atoll (3.34 and 2.27 maximum DHW, respectively). However, while lower temperature stress may help explain the lower stress banding at Turneffe it cannot explain the lack of resilience at Utila or the higher resilience in Cayos Cochinos (which experienced the highest heat stress). We also examined the possibility that a hurricane strike may have had different effects across the Mesoamerican Reef. Category 5 hurricane Mitch (October 21–29, 1998) produced extreme runoff over most of the southern portion of the Mesoamerican Reef and reduced water clarity for several weeks [24] . However, we have found no geochemical signature associated with runoff from Mitch, even in coral cores analyzed at extremely high resolution using laser ablation [25] . The lack of signal indicates the corals stopped calcifying due to the bleaching event (August 1998) prior to the hurricane. While poor water quality could explain the subsequent low resiliency of corals at our southern sites, it is notable that corals at the most southerly site, Cayos Cochinos, recover to pre-1998 growth rates just as rapidly as corals at Turneffe where the runoff impact of the hurricane was low. In addition, earlier hurricane strikes have left almost no record in our coral growth rate data suggesting that their overall impact has been low on Montastraea faveolata . We conclude that the large differences in chronic stress between our sites are responsible for differences in coral resilience following exposure to the 1998 bleaching event. To date the 1998 bleaching event remains the most significant bleaching event recorded on the Mesoamerican reef, as the 2005 event was significantly less severe than in other parts of the Caribbean [26] . Our data do not support the hypothesis that exposure to stress might help coral colonies acclimatize and therefore resist bleaching. Instead, it is clear that coral colonies experiencing higher local stress before 1998 were more severely affected by bleaching and recovered more slowly than those exposed to lower chronic stress. Possibly, the acclimatization hypothesis is only applicable for the same stressor or for lower levels of stress than Sapodilla and Utila experience, and the multi-species coral community may exhibit acclimatization patterns different from the individual coral colony response measured in this study. For example, repetitive bleaching may increase a coral's ability to withstand future heat stress [8] , [9] , but other local stressors such as increased sedimentation may depress a coral's energy reserves [27] , making it less likely to survive or recover from a bleaching event [28] . Even if acclimatization can occur in some cases, the differential responses of M. faveolata across various stress regimes indicate that local conservation efforts that reduce stress, such as reducing runoff by replanting mangroves at the coast or protecting an area from overfishing, could have significant impacts on the ability of corals to withstand the effects of climate change. Future research could investigate whether this interaction between local and global stressors extends to other coral species."
} | 2,872 |
25489283 | PMC4256962 | pmc | 298 | {
"abstract": "In this paper, a neuromorphic crossbar circuit with binary memristors is proposed for speech recognition. The binary memristors which are based on filamentary-switching mechanism can be found more popularly and are easy to be fabricated than analog memristors that are rare in materials and need a more complicated fabrication process. Thus, we develop a neuromorphic crossbar circuit using filamentary-switching binary memristors not using interface-switching analog memristors. The proposed binary memristor crossbar can recognize five vowels with 4-bit 64 input channels. The proposed crossbar is tested by 2,500 speech samples and verified to be able to recognize 89.2% of the tested samples. From the statistical simulation, the recognition rate of the binary memristor crossbar is estimated to be degraded very little from 89.2% to 80%, though the percentage variation in memristance is increased very much from 0% to 15%. In contrast, the analog memristor crossbar loses its recognition rate significantly from 96% to 9% for the same percentage variation in memristance.",
"conclusion": "Conclusions In this paper, the binary memristor crossbar circuit was proposed for neuromorphic application of speech recognition. Compared with analog memristors that are rare in available materials and need a complicated fabrication process, binary memristors which are based on the filamentary-switching mechanism are found more popularly and easy to be fabricated. Thus, we developed the neuromorphic crossbar circuit using filamentary-switching binary memristors instead of interface-switching analog memristors. The proposed binary memristor crossbar could recognize five vowels with 64 input channels and a 4-bit resolution. The proposed crossbar array was tested by 2,500 speech samples and verified to be able to recognize 89.2% of the total tested samples. Moreover, the recognition rate of the binary memristor crossbar is degraded very little only from 89.2% to 80%, even though the percentage statistical variation in memristance is increased from 0% to 15%. In contrast, the analog memristor crossbar is degraded significantly from 96% to 9% with the same percentage variation in memristance.",
"discussion": "Results and discussion In this work, the memristor-CMOS hybrid circuits were simulated by Cadence Spectre software. Here, memristors were modeled by Verilog-A [ 20 , 21 ], and CMOS SPICE parameters were obtained from Samsung's 0.13-μm CMOS technology. The training and recalling process of the memristor crossbar array are shown in Figure 8 a. In this paper, we used 100 samples for training a crossbar array to learn the vowel ‘a’. Similarly, we used 400 samples for the crossbar array to learn four vowels: ‘i’, ‘u’, ‘e’, and ‘o’. By the training process, we can find the best memristance values of the crossbar array for maximizing the recognition rate of five vowels: ‘a’, ‘i’, ‘u’, ‘e’, and ‘o’ [ 18 ]. The memristance values that are found by the training process were written to the crossbar array circuit by the V DD /3 write scheme that is known better in mitigating the half-selected cell problem compared to the V DD /2 write scheme [ 22 ]. Figure 8 Training and recalling process of binary memristor crossbar and human cochlea simulation by MATLAB. (a) Training and recalling of the binary memristor crossbar for recognizing five vowels: ‘a’, ‘i’, ‘u’, ‘e’, and ‘o’, and (b) the function of the human cochlea that is simulated by MATLAB software. For the training process, we have to convert the original speech signal to a 4-bit 64-channel digitized signal. In a biological system, the cochlea in the human ear can perform this conversion function. In this paper, we used MATLAB software that performs the same conversion function with the human cochlea. The cochlea function that is simulated by MATLAB software is shown in Figure 8 b. The function of the cochlea can be modeled by preprocessing, framing, windowing, discrete Fourier transforming (DFT), band-pass filtering, and digitization [ 23 ]. For the digitization process, 64 outputs from 64 band-pass filters are converted to 4-bit binary signals and they are delivered to the rows of the memristor crossbar array. For the band-pass filtering, the nonlinear frequency scale which is known as the mel scale is used [ 23 ]. In the mel scale, the frequency scale is linear up to 1,000 Hz and is logarithmic when the input voice has a higher frequency than 1,000 Hz [ 23 ].Figure 9 shows the simulation results for the recognition rate of the proposed binary memristor crossbar circuit. In this case, we tested 2,500 input voices for recognizing five different vowels. Each vowel is tested by 500 different voices. The average recognition rate of five different vowels is estimated to be around 89.2%. Among the five vowels, the recognition rate of ‘u’ is the highest at 95.2% while the vowel ‘e’ has the lowest recognition rate, as low as 84%. Figure 9 The simulated recognition rate of binary memristor crossbar for recognizing five vowel: ‘a’, ‘i’, ‘u’, ‘e’, and ‘o.’ Here, the number of tested voices is 2,500. Figure 10 a shows the statistical variation of memristance in HRS and LRS with the standard deviation (= σ ) of 10%. The statistical variation was obtained by Monte Carlo simulation that was also provided by Cadence software. This statistical simulation is very important because real memristors are susceptible to process variation. To analyze how tolerant the proposed binary memristor crossbar is against the memristance variation, we tested various cases of memristance variation from 0% to 15%. In Figure 10 b, we compared the proposed binary memristor crossbar circuit with the analog memristor crossbar one increasing the percentage variation in memristance from 0% to 15%. Figure 10 Statistical distribution of memristance and comparison of recognition rate between analog and binary memristor crossbar. (a) Statistical distribution of memristance with the standard deviation as much as 10%, and (b) comparison of the recognition rate between the analog memristor crossbar and binary memristor crossbar with varying percentage variation in memristance from 0% to 15%. When the memristance variation is as low as 0%, the recognition rate of the analog memristor array is higher by 6.8% than the binary memristor array. This is due to the fact that the proposed binary memristor crossbar has a 4-bit resolution; thus, it loses some amount of accuracy compared to the analog memristor crossbar. As the percentage of variation in memristance is increased, the recognition rate of analog memristor crossbar becomes degraded very rapidly. For example, when the percentage variation in memristance becomes 5%, the recognition rate of the analog crossbar is decreased from 96% to 23%. On the contrary, the binary memristor crossbar can keep almost the same amount of recognition rate for five vowels. For a percentage variation as severe as 15%, the analog crossbar shows a recognition rate as low as 9%. However, the binary crossbar still keeps the recognition rate as high as 80%, indicating that it is only degraded by 9.2% compared to the percentage variation of 0%. This strong tolerance of the binary memristor crossbar is due to the fact that the accuracy of the information stored in binary memristors can be little affected by the percentage variation in memristance. Memristance of LRS can still be much smaller and cannot become larger than that of HRS, even though the percentage variation in LRS is very large. This is the reason why the binary memristor crossbar can maintain the recognition rate over 80% regardless of the percentage variation in memristance."
} | 1,913 |
36412724 | PMC9680393 | pmc | 299 | {
"abstract": "Nature has proven to be a valuable resource in inspiring the development of novel technologies. The field of biomimetics emerged centuries ago as scientists sought to understand the fundamental science behind the extraordinary properties of organisms in nature and applied the new science to mimic a desired property using various materials. Through evolution, living organisms have developed specialized surface coatings and chemistries with extraordinary properties such as the superhydrophobicity, which has been exploited to maintain structural integrity and for survival in harsh environments. The Lotus leaf is one of many examples which has inspired the fabrication of superhydrophobic surfaces. In this review, the fundamental science, supported by rigorous derivations from a thermodynamic perspective, is presented to explain the origin of superhydrophobicity. Based on theory, the interplay between surface morphology and chemistry is shown to influence surface wetting properties of materials. Various fabrication techniques to create superhydrophobic surfaces are also presented along with the corresponding advantages and/or disadvantages. Recent advances in the characterization techniques used to quantify the superhydrophobicity of surfaces is presented with respect to accuracy and sensitivity of the measurements. Challenges associated with the fabrication and characterization of superhydrophobic surfaces are also discussed.",
"conclusion": "5. Conclusions The field of biomimetics is increasingly gaining popularity as scientists continually discover how nature has evolved organisms to overcome challenges. However, challenges still exist in mimicking the desired properties. In the case of the lotus leaf, numerous fabrication techniques have been proposed to create SH surfaces on metals, polymers, fabrics, glasses, and ceramics. As researchers continue to explore novel fabrication techniques to create SH surfaces, special consideration should be given to: (i) the scalability of the technique; (ii) the durability of the resulting SH surfaces; (iii) the potential environmental impact of the chemicals used in the process; and (iv) the cost of the instruments and materials used in the process to make the technique commercially viable. As for the characterization of the wetting properties of SH surfaces, optical-based techniques will likely remain relevant in future studies because of the simplicity, availability, and ease of operation of the technique. However, force-based techniques provide higher sensitivities and may be required to truly differentiate the wetting properties of SH surfaces with similar WCAs, SAs, and CAH. Although the fundamental science behind superhydrophobicity and the roles that surface chemistry and surface morphology play in tuning superhydrophobic properties are well understood, further research is still needed to fully understand their origins and their relative contributions to CAH.",
"introduction": "1. Introduction The wetting of solid surfaces with various liquids is ubiquitous in our daily lives, from water drops sliding down a window on a rainy day to beading up on a rose petal or even completely bouncing off the wing of a butterfly. In nature, living organisms have evolved intricate surface chemistries and morphologies resulting in superior properties [ 1 ] required for survival. For example: some desert beetles use their structured fused wings to collect water from wind-driven morning fog [ 2 ]; the wings of butterflies and dragonflies are water-repellent, which allow them to survive in constant rainfall environments [ 3 ]; the water repellency of water strider ( Gerridae ) legs allows the insect to walk on water [ 4 ]; and emperor penguin ( Aptenodytes forsteri ) feathers trap air and provide insulation against the harsh weather in Antarctica [ 5 ]. In several of these examples, the surface relies on superhydrophobicity and/or water repellency (see Figure 1 [ 6 , 7 , 8 ]), i.e., surfaces with a static water contact angle (WCA), θ , of more than 150° and a sliding angle (SA) of less than 10° [ 9 , 10 ]. Superhydrophobic (SH) surfaces typically possess hierarchical micro- and nano-roughness and consist of low surface (or interfacial) energy materials or coatings. Inspired by the novel properties imbued by SH surfaces, researchers have proposed attractive applications of such surfaces in engineering technologies. Some of these biomimetic applications are in oil–water separation [ 11 , 12 ], corrosion and fouling resistance [ 13 ], anti-icing [ 14 ], drag reduction [ 15 ], self-cleaning [ 16 , 17 ], and antibacterial surfaces [ 18 ]. Over the years, there has been a growing interest in this field, which is evidenced by the increasing number of publications on this subject as depicted in Figure 2 . Although the fabrication of SH surfaces can be traced back to 1907 in work by Ollivier [ 19 ], the concept of superhydrophobicity did not gain significant attention until this phenomenon was described in lotus leaves ( Nelumbo nucifera ) [ 20 ]. The fundamental science behind superhydrophobicity is generally well understood, however, researchers are still actively exploring novel fabrication techniques that are adaptable to metal, polymer, and ceramic surfaces and that can also increase the durability of SH surfaces or coatings against mechanical or chemical degradation [ 21 , 22 , 23 ]. Polymers have specifically played an important role; SH surfaces have either been fabricated directly from inherently hydrophobic polymers or polymers have been used to create SH coatings on various materials. The main thrust of this review article is to provide a detailed understanding of the origin of superhydrophobicity, from the derivation of critical equations from a thermodynamic perspective to appreciating how surface chemistry and surface topography influence the performance of SH surfaces. Section 3 provides an overview of recent, and the most common, methods of fabricating SH surfaces along with their advantages and disadvantages. Section 4 provides an overview of the various techniques used to characterize SH surfaces. Based on these recent developments, we conclude with a summary of the main limitations that impede the development and use of SH surfaces in practical applications."
} | 1,568 |
28174567 | PMC5258690 | pmc | 300 | {
"abstract": "Climate change-related coral bleaching, i.e., the visible loss of zooxanthellae from the coral host, is increasing in frequency and extent and presents a major threat to coral reefs globally. Coral bleaching has been proposed to involve accelerating light stress of their microalgal endosymbionts via a positive feedback loop of photodamage, symbiont expulsion and excess in vivo light exposure. To test this hypothesis, we used light and O 2 microsensors to characterize in vivo light exposure and photosynthesis of Symbiodinium during a thermal stress experiment. We created tissue areas with different densities of Symbiodinium cells in order to understand the optical properties and light microenvironment of corals during bleaching. Our results showed that in bleached Pocillopora damicornis corals, Symbiodinium light exposure was up to fivefold enhanced relative to healthy corals, and the relationship between symbiont loss and light enhancement was well-described by a power-law function. Cell-specific rates of Symbiodinium gross photosynthesis and light respiration were enhanced in bleached P. damicornis compared to healthy corals, while areal rates of net photosynthesis decreased. Symbiodinium light exposure in Favites sp. revealed the presence of low light microniches in bleached coral tissues, suggesting that light scattering in thick coral tissues can enable photoprotection of cryptic symbionts. Our study provides evidence for the acceleration of in vivo light exposure during coral bleaching but this optical feedback mechanism differs between coral hosts. Enhanced photosynthesis in relation to accelerating light exposure shows that coral microscale optics exerts a key role on coral photophysiology and the subsequent degree of radiative stress during coral bleaching.",
"introduction": "Introduction Solar radiation governs coral photophysiology and ultimately drives the productivity and growth of coral reefs ( Falkowski et al., 1990 ). Light stimulates photosynthesis of coral microalgal endosymbionts ( Symbiodinium spp.), generating O 2 and carbohydrates that are exported to the host fueling coral animal metabolism ( Muscatine et al., 1981 ). However, excess light enhances Symbiodinium photodamage ( Warner et al., 1999 ; Takahashi et al., 2004 ) and, in combination with anomalous seawater temperatures, can induce the breakdown of the coral-algal symbiosis known as coral bleaching ( Brown, 1997 ; Hoegh-Guldberg, 1999 ). Coral bleaching events are regarded as a major threat to the future of coral reefs ( Ainsworth et al., 2016 ) and hence the physiological mechanisms triggering coral bleaching have been a major research focus for decades ( Weis, 2008 ). Coral bleaching susceptibility is affected by a combination of factors that act on different spatial and temporal scales, including coral thermal history ( Brown et al., 2002 ; Hughes et al., 2003 ), Symbiodinium genotype ( Sampayo et al., 2008 ), as well as biochemical pathways and tissue properties of the coral host species ( Baird et al., 2009 ). At the cellular scale, coral bleaching involves enhanced thermal and radiative exposure of Symbiodinium cells, resulting in photodamage and the subsequent generation of reactive oxygen species (ROS) that induce the breakdown of the symbiosis ( Lesser, 1996 ; Hoegh-Guldberg, 1999 ; Weis, 2008 ). The in vivo light and temperature exposure of Symbiodinium within the host tissue ultimately controls whether Symbiodinium undergoes photodamage, and it is thus important to resolve the optical and thermal microenvironment of coral hosts ( Enriquez et al., 2005 ; Jimenez et al., 2008 ; Wangpraseurt et al., 2012 ; Swain et al., 2016 ). Application of light microsensors has shown that the in vivo light exposure within coral tissues can be enhanced over the incident downwelling irradiance ( Kühl et al., 1995 ; Wangpraseurt et al., 2012 , 2014a , b ; Brodersen et al., 2014 ). Such irradiance enhancement is modulated by the unique optical properties of coral tissue and skeleton ( Enriquez et al., 2005 ; Teran et al., 2010 ; Kahng et al., 2012 ; Marcelino et al., 2013 ; Wangpraseurt et al., 2016a ) and can improve photosynthesis under low light conditions ( Brodersen et al., 2014 ; Wangpraseurt et al., 2014a ) or lead to light stress under high irradiance ( Marcelino et al., 2013 ; Swain et al., 2016 ). The loss of Symbiodinium spp. cells from corals under environmental stress has been hypothesized to further increase irradiance exposure in hospite due to decreased shading by photopigments and increased backscattered light from the coral skeleton ( Enriquez et al., 2005 ; Teran et al., 2010 ; Swain et al., 2016 ). According to this so-called optical feedback hypothesis ( Enriquez et al., 2005 ; Swain et al., 2016 ), skeleton backscattering can further stimulate symbiont loss inducing an accelerating cycle, where symbiont loss promotes light enhancement and vice versa. Several studies have speculated on the relevance of such a feedback loop exacerbating coral bleaching, arguing that the light microenvironment during a stress event could serve as a key factor in determining the severity of a bleaching event ( Hoegh-Guldberg, 1999 ; Franklin et al., 2006 ; Hoogenboom et al., 2006 ; Abrego et al., 2008 ; Weis, 2008 ; Baird et al., 2009 ). However, experimental proof of such a mechanism has been lacking. Here, we provide the first direct measurements of the in vivo light environment of Symbiodinium during coral bleaching. We performed a thermal stress experiment and monitored changes in the in vivo light environment using light microsensors in two coral species with contrasting optical properties in concert with O 2 microsensor-based measurements of gross photosynthesis, net photosynthesis and light respiration. We provide evidence for the acceleration of in vivo light exposure upon coral bleaching and show that such light enhancement differs between coral hosts. This highlights the importance of skeleton and tissue optics for coral photophysiology and stress responses.",
"discussion": "Discussion Microscale light measurements revealed an average E 0 (PAR)-enhancement of ∼2 for P. damicornis , with maximal values of up to 2.6 within bleached polyp tissues ( Figure 5 ). Such measured enhancement in P. damicornis can at first approximation be explained by simple light scattering events within the coral skeleton ( Enriquez et al., 2005 ; Wangpraseurt et al., 2014a ). As Symbiodinium cell density becomes reduced ( Figure 1 ), more light penetrates toward the skeleton ( Figure 3A ) and an increased fraction of this light is backscattered ( Enriquez et al., 2005 ). For a flat isotropically scattering skeleton with a reflectance, R , the average path length of upwelling photons is twice that of photons in the incident collimated light beam traversing a thin layer of coral tissue (ignoring tissue scattering) ( Kühl and Jørgensen, 1994 ). Here, the contribution of the coral skeleton reflectance on the scalar irradiance can be calculated as ( Kortüm, 2012 ): E o = (1 + 2R)E d , where E d is the incident downwelling irradiance. The flux absorbed by Symbiodinium can thus be estimated as Φ abs = + = (1 + 2R) , where is the absorption of the incident beam, i.e., = A × E d , where A is the absorption cross-section of Symbiodinium ( Enriquez et al., 2005 ). Diffuse reflectance of P. damicornis skeletons in our study was on average R = 0.5 (data not shown), which is comparable to previous measurements ( R = 0.36, Swain et al., 2016 ). As such, our theoretical calculations would predict an E 0 (PAR)-enhancement of 2, a value consistent with our direct measurements for the thin bleached coenosarc tissues of P. damicornis ( Figure 5A ). In contrast to the consistent theoretical and measured E 0 (PAR)-enhancement values for P. damicornis coenosarc tissues, we observed an E 0 (PAR) enhancement of >2 for polyp tissues ( Figures 5B,C ). E 0 (PAR) enhancement of >2 suggests an additional scattering contribution from the coral tissue ( Wangpraseurt et al., 2014a ) and/or from the concave corallite architecture ( Ow and Todd, 2010 ). Indeed, measurements for bare skeletons (Supplementary Figure S1 ) revealed a higher E 0 within the corallite than over the coenosteum, suggesting that the small concave corallite for P. damicornis (width and length ∼1 mm; Nothdurft and Webb, 2007 ) effectively homogenizes and redirects the incident radiation, through multiple reflections from the skeletal walls toward the center of the corallite, whereas light escapes more easily as diffuse reflectance from the rather flat coenosteum. Given the thin tissue in P. damicornis , multiple scattering by such millimeter-sized skeletal structures likely contributes to the observed higher E 0 (PAR)-enhancement of Symbiodinium in polyp vs. coenosarc tissues in the intact living coral ( Figure 4A ). However, the E 0 (PAR)-enhancement observed in intact bleached P. damicornis was still higher than for bare skeletons indicating additional light enhancement through tissue scattering. It is possible that diffusely backscattered photons from the skeleton are trapped by the tissue (see detailed discussion in Wangpraseurt et al., 2014a ) and/or that the scattering coefficient is higher for coral tissue than for coral skeleton, as is the case for Favites sp. ( Wangpraseurt et al., 2016a ). The optical feedback hypothesis predicts an inverse power-law relationship between Symbiodinium cell density and in vivo light exposure, i.e., the rate of light field ‘amplification’ increases as symbiont cell density decreases ( Teran et al., 2010 ; Swain et al., 2016 ). For a small cup-shaped polyp, with low tissue absorption and scattering, the greatest effect of corallite architecture on light scattering is expected in close proximity to the skeletal surface at the center of the corallite ( Teran et al., 2010 ; Wangpraseurt et al., 2016a ). Our measurements for P. damicornis support these predictions since the highest rate of light enhancement was measured within aboral polyp tissues ( R 2 = 0.45, Figure 5C ). Scattering by (sub)millimeter-sized skeletal surface elements in such a thin-tissued coral with small polyps thus play a key role in the measured light enhancement dynamics upon bleaching. Coral bleaching also involves a change in the spectral exposure of the remaining symbionts ( Figure 4 ). In a normally pigmented healthy coral, blue light (400–450 nm) is rapidly attenuated vertically within the coral tissue due to peak light absorption by Chl a , leaving symbionts in aboral tissue layers exposed to a green to red shifted light spectrum ( Figure 3 ; Szabó et al., 2014 ). As coral bleaching progresses, symbionts in aboral tissues are progressively exposed to greater amounts of blue light ( Figure 4 ). It is important to note that such spectral changes could not be captured solely based on surface scalar irradiance or diffuse reflectance measurements ( Figure 4 ), highlighting the relevance of depth resolved measurement of in vivo scalar irradiance. Minor changes in the spectral composition of light can affect Symbiodinium photosynthesis in vivo ( Wangpraseurt et al., 2014c ) and it will be interesting in the future to study the effect of changing spectral exposure on symbiont photosynthesis along a vertical gradient within the tissue ( Lichtenberg et al., 2016 ). Our data demonstrated the presence of species-specific patterns of light modulation, whereby light gradients were alleviated in polyp tissues of P. damicornis but remained present in polyp tissues of Favites sp. after bleaching ( Figures 3A–D ). Favites sp. are characterized by thick, light scattering tissues ( Wangpraseurt et al., 2014a , 2016a ) with host pigments such as GFP ( Salih et al., 2000 ; Lyndby et al., 2016 ) and mycosporine-like amino acids ( Dunlap and Shick, 1998 ; Lesser and Farrell, 2004 ) that often remain present during bleaching ( Smith et al., 2013 ). Although, P. damicornis does also have a GFP-like pigment ( Takabayashi and Hoegh-Guldberg, 1995 ), the GFP-like pigments in Favites sp. are arranged in a chromatophore system that strongly enhances light scattering ( Lyndby et al., 2016 ; Wangpraseurt et al., 2016b ) facilitating a steep light attenuation along the enhanced optical path within thick scattering coral tissue ( Wangpraseurt et al., 2012 ; Lyndby et al., 2016 ). Our finding of low light levels in the aboral polyp tissues therefore suggests that tissue background scattering and absorption effectively attenuate light even in bleached tissues of Favites sp. Additionally, differences between P. damicornis and Favites sp. in skeleton optical properties are likely ( Marcelino et al., 2013 ; Swain et al., 2016 ; Wangpraseurt et al., 2016a ), which would further affect the fraction of light ‘amplification’ by the skeleton ( Marcelino et al., 2013 ). Thus our results indicate that symbionts in P. damicornis will be more severely affected by ‘optical feedback’ than symbionts in Favites sp. during coral bleaching. The moderate light environment in aboral polyp tissues of bleached Favites sp. could facilitate photoprotection of remaining cryptic Symbiodinium sp. ( Silverstein et al., 2012 ), which could be a key determinant for symbiont repopulation and subsequent coral recovery from bleaching ( Rowan et al., 1997 ). However, it is likely that a variety of structural design solutions exist that provide photoprotection on different spatial scales (see also Yost et al., 2013 ). For a branching coral, such as P. damicornis , light attenuates along the branch due to shading by neighboring branches ( Kaniewska et al., 2011 ). The magnitude of light attenuation along a single branch is similar to the light attenuation observed within the polyp tissues of Favites sp. (compare Kaniewska et al., 2011 and Wangpraseurt et al., 2012 ). Thus the presence of moderate light microenvironments is expected for P. damicornis due to its colony-level architectural complexity, which was however, not assessed in the present study. We visually observed that corals bleached first over the coenosarc tissue area, while symbionts remained within the tentacles of P. damicornis despite the enhanced in vivo light environment in polyp tissues. It was not possible to measure the light microenvironment within the highly contractile tentacle tissue and it remains unknown, whether a more moderate light microenvironment remained within the tentacles facilitating resistance to bleaching. Additionally, it is possible that bleaching was alleviated for tentacle tissues because of tentacle movement leading to enhanced gas exchange ( Shashar et al., 1993 ) and possibly reducing the build-up of high O 2 levels alleviating oxidative stress. We observed increased symbiont cell-specific rates of gross O 2 evolution and light respiration in bleached relative to healthy P. damicornis , although F v /F m values and areal photosynthetic rates (gross and net) decreased ( Figure 6 ). Numerous studies have reported lowered F v /F m values ( Warner et al., 1999 ; Jones et al., 2000 ; Supplementary Figure S2 ) and reduced net photosynthesis of Symbiodinium in response to thermal and/or light stress ( Iglesias-Prieto and Trench, 1994 ; Warner et al., 1996 , 1999 ), but few studies have measured gross O 2 evolution independent of light respiration ( Kühl et al., 1995 ; Abrego et al., 2008 ; Schrameyer et al., 2014 ). A lowering of F v /F m is often interpreted as a sign of photoinhibition of Symbiodinium and can result from the formation of non-functional PS II reaction centers, i.e., a downregulation of PSII efficiency that still allows for O 2 evolution to occur ( Hill et al., 2004 ). The high cell-specific gross photosynthetic rate ( Figure 6D ) could be explained as a downregulation of PSII efficiency counteracting the rapidly increasing in vivo irradiance ( Figure 4 ). It is also possible that the observed dynamics reflect photoacclimation of Symbiodinium , as high light acclimation involves an increase in the functional absorption cross section of PSII that leads to a lowering of F v /F m ( Hennige et al., 2008 ; Suggett et al., 2009 ). Additionally, it is possible that bleaching affected the symbiont community, potentially favoring more efficient photosynthesisers under high temperature ( Roth, 2014 ; Sampayo et al., 2016 ). The relative increase of cell-specific photosynthesis for bleached relative to healthy corals will depend on the quantity of incident solar radiation. Assuming a scenario where incident radiation saturates photosynthesis in all algal cell layers of a healthy coral, the in vivo light enhancement due to bleaching would not allow for a further enhancement of photosynthesis. Likewise, a faster increase in temperature would lead to more severe photoinhibtion and ultimately reduced photosynthetic rates. Additionally, our data showed that cell density (per surface area) declined to a greater extent than chlorophyll content (per g host tissue biomass). The reason for this mismatch remains unknown but could be related to reduced tissue thickness, which would affect chlorophyll but not cell density data, and/or potentially reflect the limited capacity of the HPLC to detect the Chl a signals in our microsamples. Normalization of gross photosynthesis data per chlorophyll a content of the corals showed that gross photosynthesis was only enhanced for polyp tissues (by about 1.6-fold) of bleached vs. healthy P. damicornis corals (data not shown). The enhancement of cell-specific photosynthesis during coral bleaching should thus be interpreted with caution and needs further investigation. Light respiration significantly increased during the bleaching experiment, accounting for about 35% of gross photosynthesis in healthy tissues up to about 150% of gross photosynthesis in bleached tissues ( Figure 6 ). Enhanced, apparently light-driven, O 2 consumption is a sign of increased electron flow through alternative electron pathways, of which the Mehler reaction is especially prevalent in Symbiodinium ( Roberty et al., 2014 ). While an upregulation of the Mehler reaction has been interpreted as a photoprotective mechanism to decrease excitation pressure on PSII under high light ( Roberty et al., 2014 ), the Mehler reaction generates ROS that can readily lead to oxidative stress, if antioxidants such as superoxide dismutase and ascorbate peroxidase are not upregulated as well ( Krueger et al., 2014 ; Roberty et al., 2014 ). Together, the enhanced metabolic activity likely reflects a response of Symbiodinium to the strongly accelerated in vivo light environment ( Figure 5 ) showing that coral microscale optics exert a key role in coral photophysiology. Climate change-related coral bleaching is arguably the prime cause of global reef decline and better predictions of coral bleaching susceptibility rest on understanding the mechanisms at play ( Hoegh-Guldberg, 1999 ; Hoegh-Guldberg et al., 2007 ; Weis, 2008 ). The optical feedback hypothesis ( Enriquez et al., 2005 ), whereby Symbiodinium undergoes light exposure ‘amplification’ during coral bleaching ( Enriquez et al., 2005 ; Swain et al., 2016 ) has lacked experimental evidence in terms of direct measures of the tissue light field before and during coral bleaching. Here, we give the first experimental proof that the in vivo scalar irradiance in corals is enhanced during coral bleaching due to changes in the balance between absorption and scattering upon loss of Symbiodinium . Light amplification differs between coral hosts, as tissue light gradients and optical shelter remained in the massive coral Favites sp. but were alleviated in P. damicornis . The finding of optical shelter in bleached Favites sp. tissues implies a photoprotective microenvironment that could sustain coral resilience by facilitating the repopulation from cryptic Symbiodinium after stress. Our study shows that coral bleaching can in fact enhance cell-specific photosynthetic rates of remaining Symbodinium , which is interpreted as a response to the accelerated in vivo light exposure during bleaching. We conclude that coral microscale optics have a fundamental role in shaping the radiative exposure of Symbiodinium and thus photophysiological stress responses during coral bleaching."
} | 5,134 |
21698298 | PMC3115960 | pmc | 301 | {
"abstract": "Plant-pollinator mutualistic networks are asymmetric in their interactions: specialist plants are pollinated by generalist animals, while generalist plants are pollinated by a broad range involving specialists and generalists. It has been suggested that this asymmetric –or disassortative– assemblage could play an important role in determining the observed equal susceptibility of specialist and generalist plants under habitat destruction. At the core of the analysis of the phenomenon lies the observation that specialist plants, otherwise candidates to extinction, could cope with the disruption thanks to their interaction with a few generalist pollinators. We present a theoretical framework that supports this thesis. We analyze a dynamical model of a system of mutualistic plants and pollinators, subject to the destruction of their habitat. We analyze and compare two families of interaction topologies, ranging from highly assortative to highly disassortative ones, as well as real pollination networks. We found that several features observed in natural systems are predicted by the mathematical model. First, there is a tendency to increase the asymmetry of the network as a result of the extinctions. Second, an entropy measure of the differential susceptibility to extinction of specialist and generalist species show that they tend to balance when the network is disassortative. Finally, the disappearance of links in the network, as a result of extinctions, shows that specialist plants preserve more connections than the corresponding plants in an assortative system, enabling them to resist the disruption.",
"introduction": "Introduction Habitat destruction is the major cause of species extinctions and a main driving force behind current biodiversity loss [1] – [3] . In the context of habitat fragmentation, one of the most actively studied processes is animal-mediated pollination, which is crucial for the sexual reproduction of flowering plants. The strength of the effect of fragmentation on pollination and on plant reproductive success shows a highly significant correlation, suggesting that one of the most important causes of reproductive impairment in fragmented habitats may be pollination limitation [4] . In the mutualistic interaction between plants and pollinators, plant species are typically considered generalists when pollinated by several or many animal species of different taxa, and specialists if pollinated by one or a few taxonomically related pollinators [5] – [8] . Most plant-pollinator mutualistic networks have shown to be highly asymmetrical in their topologies, with specialist plants being pollinated mostly by generalist pollinators, whereas generalists are pollinated by both specialists and generalists pollinators [9] , [10] . Some ecological consequences of the asymmetry of the plant-pollinator mutualistic network have been studied. Using mathematical models it has been shown that the asymmetry of plant-pollinator networks differs from random networks in their response to habitat destruction. Networks with topologies present in real communities start to decay sooner than random communities, but persist for higher destruction levels. When the destruction level is above a given threshold the whole community collapses [11] . Besides, theoretical studies have suggested that habitat destruction would affect preferentially specialised plants, because they would not be able to counterbalance the loss of their few specific mutualist partners with other alternative pollinators [5] , [7] , [12] . Generalist plants, instead, should be more adaptable to the changes imposed by fragmentation on their pollinator assemblages because the absence of one or some of their pollinators could be compensated by other pollinators from their wide assemblages [13] . Contrary to these theoretical expectations, no significant difference was found in the mean effect on specialist and generalist plant species, both being equally negatively affected by habitat fragmentation [14] , [15] . One explanation for the equal susceptibility of specialist and generalist plants to habitat destruction lies precisely on the asymmetric interaction. Because specialist plants interact mainly with generalist pollinators, they would be able to keep their few pollinators in fragmented habitats, and thus their reproduction would not be so severely impaired as previously thought. Generalist plants, which interact with both generalist and specialist pollinators, would tend to loose their specialist pollinator fraction from their assemblages and retain their generalist pollinators. Thus, a decrease in the remaining generalist pollinators population would therefore have equal effects on the two groups of plants [16] . Mathematical models differ from verbal theories in giving a precise connection between assumptions and conclusions. They are a key tool needed to illuminate how the network architecture influences species extinction or persistence [17] . In this work we constructed assortative and disassortative networks and analyzed the effect of habitat destruction in each case, focusing on the relative effect on specialist and generalist species. We found that the way in which species are interconnected determines in a great deal who gets extinct, and in which way the perturbation affects the balance of specialization. In accordance with the theory proposed by Ashworth et al. [16] , we observed that in asymmetric (disassortative) networks, generalist plants loose their connections with specialist pollinators, but specialist plants loose by far much less connections than specialist ones in the symmetric (assortative) networks. Our results support the idea that network asymmetry explains the equal susceptibility of generalist and specialist plants to habitat disturbance.",
"discussion": "Discussion Mathematical models of plant-pollinator interaction networks have given many insights into the effect of habitat fragmentation on ecological communities [11] , [17] . One of the main characteristics of the topology of plant-pollinator interaction networks is their asymmetry: specialist plants are mainly pollinated by generalist pollinators whereas generalist plants are pollinated by both specialist and generalist pollinators [9] , [10] . Such asymmetric interaction could be the reason why specialist and generalist plant species show similar response to habitat fragmentation, as argued in [16] . The main aim of this work has been to test this hypothesis while giving it a theoretical framework. To this goal, we have constructed symmetric and asymmetric networks of plant-pollinator interactions ( Fig. 2 ). We have calculated the degree of asymmetry of such networks, as well as real ones, expressed by the measurements of their assortativity ( Fig. 3 ). We then analyzed the extinction pattern of these networks as a function of the disturbance ( Fig. 4 ). We have also analyzed the assortativity and density of the networks resulting from different degrees of habitat destruction ( Fig. 5 ). We have introduced entropy as a measure of the differential effect of habitat fragmentation on generalist and specialist species ( Fig. 6 ). Most importantly we have found that both the connectivity and the degree of habitat fragmentation are factors that increase the pattern of equal susceptibility of generalist and specialist plant species to habitat destruction ( Fig. 7 and 8 ). A deeper analysis of the pattern of species extinction in symmetric and asymmetric networks shows that, in asymmetric (disassortative) networks, generalist plants loose their connections with specialist pollinators, but specialist plant loose by far much less connections than specialist plants in the symmetric (assortative) networks ( Fig. 10 ). Therefore, and in accordance with Ashworth [16] , our results suggest that network asymmetry explains the equal susceptibility of generalist and specialist plants to habitat disturbance. Our approach is similar than the one from Fortuna [11] in that it does not include obligatory interactions on plants nor pollinators. We have assumed a community of facultative species in which the absence of their interacting partner does not implies species extinction. Obligatory interaction such as the one present in self-incompatible plants, may have a role on the pattern of species extinctions [4] . We did not include other complex features in our model such as temporal variation in the association plant-pollinator [27] or spatial effects [28] . These processes can have a role in the response to habitat destruction and deserve further investigation. Our aim was to capture, with the simplest model, the effect of asymmetry on the pattern of extinction in response to habitat destruction."
} | 2,192 |
35558469 | PMC9088758 | pmc | 302 | {
"abstract": "In recent years, there have been great achievements in superhydrophobic coatings. However, there are still some barriers restricting superhydrophobic coatings in practical applications, such as widely used organic solvents and poor oleophobicity. In this study, we proposed a method for fabricating absolutely waterborne superamphiphobic coatings in two steps. Firstly, we synthesized the waterborne SiO 2 sol using methyltriethoxysilane, and then the SiO 2 sol was modified in an aqueous system with a fluorocarbon surfactant. The results showed that the coating had contact angles of 160°, 153° and 150° and sliding angles of 1°, 4.7° and 6.3° with respect to water, soybean oil and hexadecane. Moreover, the coating could withstand 300 °C heating and immersion in various corrosive solutions for several hours. Furthermore, it is worth mentioning that the waterborne coating showed excellent performances in antifouling, self-cleaning, and damp-proof fields.",
"conclusion": "4 Conclusion In conclusion, we have demonstrated a simple, low cost, environmentally friendly method to fabricate absolutely waterborne superamphiphobic coatings using methyltriethoxysilane, fluorinated alkyl silane and a fluorocarbon surfactant. The coating showed an excellent superamphiphobic ability with contact angles of 160°, 153° and 150° and sliding angles of 1°, 4.7° and 6.3° with respect to water, soybean oil and hexadecane. The coating could be applied on various substrates, such as glass, copper foam and polyester fiber, through spraying or dip-coating. Furthermore, after immersing in acid (pH = 1), alkali (pH = 13) and salt (1 mol L −1 NaCl), and after treating at 200 °C for 2 h, the coating still maintained its superamphiphobic ability. Additionally, this superamphiphobic coating exhibited prospective performance for anti-fouling, self-cleaning and damp-proof applications.",
"introduction": "1 Introduction The demand for water repellent and antifouling properties has been increasing gradually owing to severer and severer environmental problems. 1–3 Inspired by the lotus leaf, superhydrophobic surfaces, having a greater water contact angle (CA) than 150° and a lower sliding angle (SA) than 10°, were proposed in an attempt to satisfy the aforementioned demand. To our delight, superhydrophobic surfaces have tremendous potential applications in self-cleaning, 4–7 anti-icing, 8–13 drag reduction, 14–16 water–oil separation 17–22 and atmospheric water capture. 23–25 However, the superhydrophobic surfaces tend to lose their effects when exposed to external environments for a long time due to the presence of various organic contaminants, to a great extent. As a consequence, the question of how to improve the durability of superhydrophobic surfaces after long-term exposure to the outside has drawn the attention of researchers. Superamphiphobic surfaces, with both superhydrophobicity and superoleophobicity, are very desirable due to the fact that they are promising for solving the problems above. Nevertheless, it is much harder to fabricate superamphiphobic surfaces, which possess larger roughness and lower surface free energy. Several methods have been proposed to prepare superamphiphobic surfaces, such as functionalization with nanoparticles, 26 etching, 27,28 lithography, 29 template-assisted synthesis 5 and sol–gel routes. 30 Recently, Susarrey-Arce and his coworkers 31 developed a method to fabricate arrays of microstructures via the reactive ion etching of a silicon wafer, which was covered by a patterned photoresist layer. Regrettably, it seems that with this method it was difficult to fabricate superamphiphobic surfaces on a large scale because of the complex process and costly equipment. Among various methods, wet-chemical processes by dip-coating or spraying are the focus of concern due to their great convenience for large-scale fabricating with a low reliance on equipment. To the best of our knowledge, most of the solvents that are used in the wet-chemical systems are organic, including ethanol, 32 acetone, 33 dimethylformamide (DMF), 33 and tetrahydrofuran (THF), 34 which are considered to be harmful to both the environment and human health. Consequently, it is impossible to prepare superamphiphobic coatings consisting of an organic solvent on a large scale. Water, known as a green and cost-effective solvent, is of great interest to scientists around the world. In recent years, Bayer and co-workers 35 have reported the formation of a superamphiphobic surface through spraying an aqueous solution consisting of spherical and porous silicon dioxide nanopowder, hydrophilic fumed silica and waterborne acrylic fluorochemical dispersion on high temperature metallic substrates. However, most of the reported waterborne superamphiphobic coatings have an inaccessible preparation cost, restricting their further practical application. Herein, we developed a simple, low-cost, environmentally friendly method to fabricate waterborne superamphiphobic coatings using methyltriethoxysilane, fluorinated alkyl silane and a fluorocarbon surfactant as the principal raw materials. Firstly, waterborne SiO 2 sol was synthesized via a hydrolysis process of methyltriethoxysilane in water. The following step was the modification of the SiO 2 sol through fluorinated alkyl silane in water with a fluorocarbon surfactant. Subsequently, the superamphiphobic coatings were prepared by spraying or dip-coating on glass, polyester fiber, copper foam and some other substrates. The coating showed an excellent superamphiphobic ability with contact angles of 160°, 153° and 150° and sliding angles of 1°, 4.7° and 6.3° with respect to water, soybean oil and hexadecane. Furthermore, the coating could be coated on various substrate surfaces, such as glass, copper foam and polyester fiber, it performed very well in terms of chemical durability and thermal stability and it has potential for practical use in antifouling, self-cleaning and damp-proof fields.",
"discussion": "3 Results and discussions 3.1 Preparation of the waterborne coatings The waterborne superamphiphobic coatings were prepared via a modified reaction between the SiO 2 sol and 1 H ,1 H ,2 H ,2 H -perfluorodecyltrichlorosilane (PFDTES), and the schematic for this is shown in Fig. 1a . First of all, SiO 2 sol was synthesized using methyltriethoxysilane (MTES). The MTES was hydrolyzed in an aqueous system and sodium dodecyl benzene sulfonate (SDBS) was used to promote the solubility of MTES in water. The finally prepared SiO 2 sol is shown in Fig. S1. † It is apparent from the optical photograph that the synthesized SiO 2 sol was milky. In addition, we found that the hydrolysis time for methyltriethoxysilane was an important factor for controlling the SiO 2 sol morphology. Fig. S2 † shows the TEM images of the SiO 2 sol with different hydrolysis times from 12 h to 60 h. As the TEM images show, the diameter of the SiO 2 particles was about 30 nm regardless of the hydrolysis time, but a difference lay in the extent of the polycondensation between the SiO 2 particles. When the reaction was allowed to occur for 12 h, the extent of the polycondensation of the SiO 2 particles was relatively low and there was a large gap between the particles. The opposite was the case when the hydrolysis time was 60 h. Based on the above results, the determined hydrolysis time was 36 h, which was considered to be the optimum time for preparing the SiO 2 sol, and all of the samples mentioned later were fabricated using SiO 2 sol that was reacted for 36 h unless noted otherwise. Subsequently, the as-prepared SiO 2 sol and the ForaperLe 323 (FL323) (which is a kind of fluorocarbon surfactant consisting of a waterborne acrylic fluoropolymer) were well-dispersed in an alkaline aqueous system. Tetraethoxysilane (TEOS) and PFDTES were added in the following step. After constant stirring for 24 h, the preparation process was finished. Changes to the chemical bonding during fabrication are shown in Fig. S3. † The inset optical image in Fig. 1b is a photograph of the as-prepared waterborne paint, indicating that the solution was semi-transparent. The dynamic light scattering (DLS) measurements shown in Fig. 1b revealed that the average diameter of the waterborne paint was approximately 660 nm. The most positive aspect was that the waterborne superamphiphobic paint could be prepared on a large scale (Fig. S4 † ) with an acceptable cost, which can be seen as a breakthrough in the application fields of waterborne superamphiphobic coatings. Fig. 1c shows the FTIR spectra of the as-prepared SiO 2 sol and coating, indicating that the SiO 2 sol was successfully modified by PFDTES in the aqueous system. For the SiO 2 sol, the asymmetric stretching vibration of Si–O–Si gave strong absorption peaks at 1033 cm −1 and 1120 cm −1 . The absorption peak at 782 cm −1 can be assigned to the presence of Si–C bonds, while the strong bands at 1409 cm −1 , 2935 cm −1 and 2973 cm −1 can be attributed to the asymmetrical stretching of the –CH 3 groups. After being modified, the FTIR spectra sketch of SiO 2 had a significant change. Apart from the absorption peaks attributed to SiO 2 that were mentioned before, the modified SiO 2 had a new absorption peak at 1725 cm −1 , which can be assigned to the symmetric stretching vibration of C \n \n\n<svg xmlns=\"http://www.w3.org/2000/svg\" version=\"1.0\" width=\"13.200000pt\" height=\"16.000000pt\" viewBox=\"0 0 13.200000 16.000000\" preserveAspectRatio=\"xMidYMid meet\"><metadata>\nCreated by potrace 1.16, written by Peter Selinger 2001-2019\n</metadata><g transform=\"translate(1.000000,15.000000) scale(0.017500,-0.017500)\" fill=\"currentColor\" stroke=\"none\"><path d=\"M0 440 l0 -40 320 0 320 0 0 40 0 40 -320 0 -320 0 0 -40z M0 280 l0 -40 320 0 320 0 0 40 0 40 -320 0 -320 0 0 -40z\"/></g></svg>\n\n O in the fluorocarbon surfactant. As we know, the characteristic peaks of the various stretching vibrations of –CF 2 and –CF 3 are at 1121, 1145, 1198, and 1241 cm −1 , which are very close to the asymmetric stretching vibrations of Si–O–Si at 1033 cm −1 and 1120 cm −1 . As a consequence, the width of the absorption peak was enlarged distinctly. Fig. 1 Preparation of the waterborne superamphiphobic coating. (a) Schematic of the fabrication of the superamphiphobic coating. (b) Size distribution in the waterborne superamphiphobic paint (inset: photograph of the waterborne superamphiphobic paint in a transparent glass bottle). (c) FTIR spectra of the SiO 2 sol and the waterborne superamphiphobic coating (all the samples were dried into a powder at 80 °C to test FTIR). 3.2 Effect of the hydrolysis time for methyltriethoxysilane on the coating As mentioned before, the hydrolysis time of methyltriethoxysilane had an important effect on the SiO 2 sol morphology, which is considered to be one of the principal factors in preparing superamphiphobic surfaces. As Fig. 2a shows, with the changing hydrolysis time, the coatings were all super-repellent to water (surface tension: 72.1 mN m −1 ) and soybean oil (35.7 mN m −1 ), but only a few coatings were super-repellent to hexadecane (27.5 mN m −1 ). When the hydrolysis time for methyltriethoxysilane was 36 h, the coating had a maximum contact angle (150°) and minimum sliding angle (6.3°) to hexadecane, indicating that this coating had the best superamphiphobicity. The reason for this lies in the final morphology of the coatings, and scanning electron microscopy (SEM) images are shown in Fig. 2b–d . Compared to the other two coatings (where the hydrolysis time of methyltriethoxysilane is 12 h ( Fig. 2b ) and 60 h ( Fig. 2d )), the coating prepared using the SiO 2 sol that was hydrolyzed for 36 h ( Fig. 2c ) had a more compact and uniform morphology at the micro-scale. Furthermore, the network structure in Fig. 2c had fewer microholes than the two other coatings did. A similar case appeared when the magnification times were enlarged at the nano-scale. The coating with the SiO 2 that had reacted for 36 h had more roughness to its structure and the nanoparticles in this sample were denser than they were in the two other coatings. The dense and uniform micro-nano structure resulted in better repellency of hexadecane. The fracture surface morphology of the superamphiphobic coating is presented in Fig. S5 † and the upside-down circular truncated cone structure (which was called the “reentrant structure”) could be seen clearly, resulting in the superamphiphobicity of our prepared coating. Fig. 2 (a) Contact angles and sliding angles of water, oil and hexadecane on the waterborne coating with different hydrolysis times of methyltriethoxysilane. SEM images of the waterborne coatings with hydrolysis times of methyltriethoxysilane of (b) 12 h, (c) 36 h and (d) 60 h. 3.3 Applicability of the waterborne coatings on various substrates In order to test the superamphiphobicity on various substrates, three typical materials were selected: flat (glass), porous (copper foam) and soft (polyester fiber) substrates. As showed in the left position of Fig. 3a–c , all three substrates revealed excellent superhydrophobicity and there was no distinct change in the original state of the substrates after being coated. For example, the coated glass was translucent and we could see the badge beneath the substrates clearly. The water and oil contact angles and sliding angles are presented in Fig. S6. † For the contact angle values, these were about 160° and 153° for water and oil, respectively. For the sliding angle values, the values for glass and copper foam were all below 10° for both water and oil. However, for fiber, the sliding angle was relatively large (15° and 37° for water and oil, respectively) due to the coarser texture of the fiber resulting in strong adhesion between the substrate and the liquid drop. When the volume of tested liquid (water and oil) was set at 25 μL, the sliding angle values decreased sharply to 7° and 16° for water and oil, respectively. We also chose some commercial fabrics to test and both showed an excellent superamphiphobicity (Fig. S7 † ). Based on the results above, we believe that the waterborne coating has an outstanding applicability on various substrates (for more details, see Video S1 † ). Fig. 3 Photographs (left, inset: oil contact angle images), SEM images (middle) and impacting test results (right, with a volume of 5 μL and a height of 15 mm between the water droplet and the substrate) on (a) glass, (b) copper foam and (c) polyester fiber substrates. The middle position of Fig. 3a–c shows the SEM images of the different substrates (the original morphologies of the copper foam and polyester fiber are listed in Fig. S8 † ) after being coated with the waterborne paints, showing the textured surface morphologies for each one. The right position shows the high speed camera images of the bouncing water droplets on the various substrates (the height between the water drop and the substrate was 15 mm, and the volume of water was 5 μL). The water droplets could completely leave the surface without wetting, penetrating or damaging it for both the hard and soft substrates, showing that water has an extremely low adhesion with the various substrates. Based on the facts above, we believe that the coatings could stabilize on different substrates through the spraying or dip-coating method. 36,37 3.4 Thermal stability It was unknown whether the as-prepared coatings could remain superamphiphobic under high temperature. Therefore, the water and oil contact angles and sliding angles under temperatures ranging from 100 °C to 300 °C were investigated to evaluate the thermal stability of the coating. As shown in Fig. 4a , when the temperature was below 200 °C, the coating still remained superamphiphobic and there was no visible change in the water and oil contact angles and sliding angles. However, the coating became oleophilic and superoleophilic when the temperature increased to 250 °C and 300 °C, respectively, while it remained as superhydrophobic as before. However, the coating was still superamphiphobic after 150 °C heating for 48 h (Fig. S9 † ). Fig. 4 Thermal stability of the coating. (a) Change in the contact angles and sliding angles (water and oil) with the temperature increasing from 100 °C to 300 °C for 2 h (inset: water and oil contact angle images). (b) Optical photographs of the coating on the glass substrate after 300 °C heating and after fluorination (inset: oil contact angle images). SEM images before 300 °C heating (c) and after heating (d) (inset table: content of F and Si elements). It was interesting that the coating treated at 300 °C (which was superhydrophobic and superoleophilic) returned to being superamphiphobic after being deposited in a layer of PFDTES (namely the fluorination mentioned in Fig. 4b ). Based on the results above, we believed that the coating becoming superoleophilic was attributed to the degradation of the substance containing the F element. SEM images and energy dispersive spectrometry (EDS) were measured to prove this point. As shown in Fig. 4c and d , the structure of the as-prepared coating and the coating treated at 300 °C had little change, but the content of the F and C elements decreased sharply, while there was no obvious change in the content of Si after treating at 300 °C (the whole EDS is listed in Fig. S10 † ), indicating that the fluorocarbon surfactant degraded during the heating process. 3.5 Chemical durability In addition to high temperature, the superamphiphobicity of coatings tends to recede when they are exposed to corrosive environments such as strong acid, base and high salt spray. Herein, we put the coated glasses into different pH values from 1 to 13 and into 1 mol L −1 NaCl solutions and immersed them for 2 h to test the chemical durability of the coatings. Fig. S11a † shows digital photographs of the coatings that were immersed in different corrosive solutions and after immersing for 2 h, revealing that the surfaces were all clean and there were no residual liquids on the coated glasses. More importantly, the coated glasses could keep clean even when immersed in oil after 2 h of immersion in different corrosive solutions (Fig. S11b † ). The water and oil contact angle and sliding angle values are listed in Fig. 5a and b . As the results show, the contact angle values were all close to 160° and 150° and the sliding angle values were still below 10° and 15° for water and oil, respectively. In other words, the coating could withstand different corrosive environments for a certain time. Fig. 5 Change in the contact angles and sliding angles for water (a) and oil (b) in different corrosive solutions after immersing for 2 h. 3.6 Applications of the waterborne coating Antifouling and self-cleaning properties As we know, superhydrophobic surfaces can always remain clean under various conditions because of self-cleaning effects. Nevertheless, whether the surfaces can still remain clean under the severe conditions contained in oil fouling has rarely been reported. We chose a glove and dyed water and oil to test the antifouling properties. The index finger of the glove was treated with the coating and the middle finger was untreated, as shown in Fig. S12a. † Then the two fingers were immersed in the dyed oil simultaneously for several minutes. After being taken out, the index finger didn’t have any oil stains on it while the middle finger was wetted by the dyed oil completely. Even after bending several times, the index finger could still remain completely clean. Based on the results above, we are convinced that the waterborne coating is able to be qualified in the antifouling field under most conditions. As for the self-cleaning properties test, ordinary sludge powder and sludge powder containing oil fouling were used to simulate a complex environment. As shown in Fig. S12b, † the as-prepared coating exhibits excellent self-cleaning properties and water (dyed blue) can roll-off from the surface without a hitch, and as a result of which, the sludge powder can be taken away (for more details, see Video S2 † ). When the sludge powder containing oil fouling (a mixture of edible oil and sludge powder) was used, the mixture could also slide from the surfaces (Video S3 † ). In the end, there was no residue on the surfaces, indicating that the coatings have excellent self-cleaning properties under various complicated conditions. Damp-proof testing on polyester fiber Fibers tend to go mouldy when they are in a humid environment, and as a result of which, damp-proof properties are one of the most primary demands for various fibers. It is universally acknowledged that superhydrophobic surfaces have an excellent performance in the water-proof field. However, it remains in doubt whether all superhydrophobic surfaces are capable in humid environments. Fig. 6a shows the schematic of the damp-proof test, in which two glass tubes and a humidifier were used. The samples were fixed between the two glass tubes (detailed data of the samples and glass tubes are listed in Fig. S13 † ) and the dampness was brought in from below and allowed to get out at the top. The mass of the sample was recorded every half hour and an ordinary fiber was introduced for comparison. Here, φ is defined as the unit area mass of growth: where m is the current mass, m 0 is the initial mass and S is the area of the sample. As shown in Fig. 6b , the φ value of the untreated fiber had a rapid increase from 0 to 1.5 h, while no discernible changes were observed in the treated fiber. Other treated and untreated fibers showed little obvious growth in the φ value after 1.5 h, indicating that the ordinary fiber was wetted entirely by the dampness, whereas there was little effect on the superhydrophobic fiber ( Fig. 6c and d ). More importantly, the treated fiber had equal breathability as the untreated did (Video S4 † ). Fig. 6 Damp-proof test on polyester fiber. (a) Schematic of the damp-proof test. (b) Unit area mass change of the treated and untreated fibers at a certain humidity. Photograph of the untreated fiber (c) and treated fiber (d)."
} | 5,547 |
35609948 | null | s2 | 303 | {
"abstract": "Bacteria orchestrate collective behaviors and accomplish feats that would be unsuccessful if carried out by a lone bacterium. Processes undertaken by groups of bacteria include bioluminescence, biofilm formation, virulence factor production, and release of public goods that are shared by the community. Collective behaviors are controlled by signal transduction networks that integrate sensory information and transduce the information internally. Here, we discuss network features and mechanisms that, even in the face of dramatically changing environments, drive precise execution of bacterial group behaviors. We focus on representative quorum-sensing and second-messenger cyclic dimeric GMP (c-di-GMP) signal relays. We highlight ligand specificity versus sensitivity, how small-molecule ligands drive discrimination of kin versus nonkin, signal integration mechanisms, single-input sensory systems versus coincidence detectors, and tuning of input-output dynamics via feedback regulation. We summarize how different features of signal transduction systems allow groups of bacteria to successfully interpret and collectively react to dynamically changing environments."
} | 293 |
23029158 | PMC3460938 | pmc | 304 | {
"abstract": "Background The stability of cooperative interactions among different species can be compromised by cheating. In the plant-mycorrhizal fungi symbiosis, a single mycorrhizal network may interact with many plants, providing the opportunity for individual plants to cheat by obtaining nutrients from the fungi without donating carbon. Here we determine whether kin selection may favour plant investment in the mycorrhizal network, reducing the incentive to cheat when relatives interact with a single network. Methodology/Principal Findings We show that mycorrhizal network size and root colonization were greater when Ambrosia artemisiifolia L . was grown with siblings compared to strangers. Soil fungal abundance was positively correlated with group leaf nitrogen, and increased root colonization was associated with a reduced number of pathogen-induced root lesions, indicating greater benefit to plants grown with siblings. Conclusions/Significance Plants can benefit their relatives through investment in mycorrhizal fungi, and kin selection in plants could promote the persistence of the mycorrhizal symbiosis.",
"introduction": "Introduction Many organisms cooperate even though they have the opportunity to cheat. The interaction between plants and mycorrhizal fungi is considered a mutualism because the fungus provides water, nutrients and pathogen defense to the plant in return for carbohydrates. Though most mycorrhizal fungi are obligate symbionts, dependent on plant carbon for growth [1] , plants may be obligate or facultative in their association with mycorrhizal fungi [2] . Moreover, mycorrhizal fungi may span the gradient from mutualism to parasitism. Cooperation, conflict, and cheating have all been observed to occur between fungi and plants [3] , [4] . The symbiosis is considered by economic models to be a biological market where there is a trade relationship between plant and fungi, each of which specializes on acquiring certain resources [5] – [7] . Models show that a mutualism can be stable through a trade relationship [5] , [6] . Plants tend to associate more with mycorrhizas when soil nutrients ( e.g. \n [8] , [9] ) or plant tissue phosphorus (P) concentration [10] are low, which supports a simple prediction from the biological market models. Recent experimental evidence indicates that, given a choice, plant and fungal partners can also choose to trade with more cooperative partners, thus promoting a stable mutualism where neither partner is in control of the other [11] . When many plants are connected to a common mycorrhizal network (CMN), tragedy of the commons theory models the mycorrhizal symbiosis as a social good, i.e. , a common good that is a shared resource created and/or maintained by the group [12] . For mycorrhizas, the CMN may be maintained by a group of plants and provides a common resource for that group. The size of the fungal network depends on plant carbohydrate contributions and thus, more soil colonization by fungal hyphae implies more investment by the plant partner [13] . Therefore, the value of the mycorrhizal network as a social good should depend on the summed carbon donations from host plants. Because attached plants will acquire more nutrients from larger networks with greater surface area and increased soil exploration, plants benefit each other by investing in the same fungal partner. However, as individuals pay a cost to participate in the symbiosis, this creates a conflict. If individuals can escape paying the cost while still reaping the benefits from their partner, there is strong incentive to cheat [14] . In the mycorrhizal symbiosis, several plants may be attached to a CMN and many fungal genets or species can simultaneously colonize a single plant. If either the fungus or plant do not identify cheaters and invoke sanctions, the symbiosis is open to non-cooperators since individuals may attach themselves to the mutualism without donating their fair share, ultimately leading to a tragedy of the commons [12] , [15] – [17] . A majority of research has concentrated on the potential role of sanctions against cheaters [18] – [20] . However, kin selection among plants offers an alternate incentive for cooperation between mutualists [21] , [22] because for a plant, investing carbon in the mycorrhizal network linked to close relatives could increase one's indirect fitness and may remove cheaters from the population [23] preventing a tragedy of the commons [12] . Plants frequently live in dense communities where relatedness may be high, providing the opportunity for kin selection [24] , [25] . Kin selection acts more strongly if individuals only demonstrate altruism toward relatives [26] , which then favours the evolution of kin recognition. Kin recognition has been demonstrated in several species of plants [27] – [30] . Though the mechanism is as yet unknown, root exudates have been demonstrated to convey a signal [27] . Kin recognition is also manifested as phenotypic plasticity in resource-gathering structures in response to relatedness of the plant group. In Cakile edentula , for example, allocation to fine roots was lower among individuals in sibling groups [28] relative to groups of non-related individuals. Because fine roots are the sites of nutrient and water absorption, this response suggests that competition for these resources was reduced among siblings ( i.e., kin). However, these studies demonstrating kin recognition have been done using non-mycorrhizal plants, and it is possible that the presence of a symbiont could influence interactions among kin. Although researchers have considered the importance of plant neighbourhood on mycorrhizas, these studies have focused on the benefits of fungal [31] and plant diversity [32] – [36] . In the only study that has tested whether the genetic relatedness of neighbours influenced plant interactions with mycorrhizas, Ronsheim & Anderson (2001) found that in the presence of soil fungi, biomass of individuals grown with clones or plants from the same population was greater than individuals grown with plants from a different population [37] . Their study addressed the question of local adaptation to soil fungal communities and they demonstrated benefits of growing with plants from the same population. However, no study has yet measured kin recognition in mycorrhizal plants or tested whether relatedness of a plant population affects mycorrhizal fungal growth. When mycorrhizas are present, greater cooperation among groups of siblings could be manifested through an increase in the CMN. Such an increase could result in greater total nutrient acquisition for the group [38] or reduce the likelihood of pathogen attack [39] , which should enhance the fitness of groups of siblings relative to groups of strangers. We examined whether the association between Glomus intraradices and pairs of Ambrosia artemisiifolia L. (common ragweed) seedlings depended on the relatedness of the two plants. G. intraradices colonizes plant roots aggressively [40] , suggesting that young plants may experience kin selection through mycorrhizas. Because arbuscules are the sites of nutrient exchange and an increase in root colonization by arbuscules indicates a well-established mutualism [41] , we predicted that plant kin selection would favour the colonization of arbuscules in sibling pairs. To determine whether related seedlings benefited from a potentially enhanced mycorrhizal association, we measured plant growth as well as susceptibility to pathogen attack by measuring the frequency of lesions on roots. Since an increase in mycorrhizal association in young seedlings may promote a well-developed CMN later in life, we carried out a second experiment to investigate whether plant relatedness and P level affected the symbiosis at the juvenile stage, when the CMN has had time to develop. Hyphae from spores of the same isolate of G. intraradices readily fuse together [42] , increasing the likelihood of a CMN forming. We predicted that kin selection would favour siblings to donate more carbon to the fungal partner, resulting in greater mycorrhizal association in groups of siblings than in groups of strangers. We also predicted that plants would promote mycorrhizal colonization in lower P environments, where the symbiosis could facilitate plant nutrient acquisition, regardless of the relatedness of the group. We examined whether an enhanced CMN, quantified as the length of the extraradical mycorrhizal hyphae, benefitted plants by measuring the relationship between CMN size and plant growth, as well as between CMN size and leaf nitrogen (N). We present results that show the mycorrhizal association meets two predictions supported by kin selection theory: plants grown in siblings groups had more mycorrhizal colonization and growth than when they are grown in stranger groups, and the increased mycorrhizal association benefits the plants. Seedlings grown with siblings had more arbuscules and root hyphae and a reduced proportion of lesions on the roots. Juveniles had longer soil hyphae when grown with siblings, suggesting a more developed CMN, and this was correlated with increased leaf N. We also found that stranger groups had longer soil hyphae in low P, but soil hyphal length and growth was promoted in sibling groups regardless of P level. Alternative hypotheses for these results were explored but these hypotheses were not supported.",
"discussion": "Discussion/Conclusion We provide the first evidence there is plant kin recognition, i.e. , plasticity to relatedness of neighbours, in the mycorrhizal symbiosis, and that siblings can benefit each other through increased mycorrhizal association. Though no evidence of kin recognition was found in the plants themselves, mycorrhizal colonization and growth may be considered an extended phenotype that responds to the host environmental conditions, including the relatedness of the plant group. In young seedlings, arbuscule and root hyphal colonization responded to relatedness, and pairs of siblings had fewer root lesions than strangers. Juvenile plant investment in the mycorrhizal network depended on the social environment and the nutrient conditions, which translated into a nutritional benefit for plant groups with more soil hyphae. Interestingly, we also found increased fungal colonization in low P, as predicted by the biological market model. The mycorrhizal response to siblings is supported by kin selection theory. In the presence of likely cheaters, i.e. , strangers, mycorrhizal colonization and growth were lower, whereas in the absence of likely cheaters, i.e., in solitary or sibling groups, mycorrhizal colonization and growth was greater. Although we found this pattern in both seedling and juvenile experiments, the mycorrhizal structures that responded were different. In seedlings, we found more arbuscules and root hyphae in siblings than in stranger pairs. Arbuscules, the sites of nutrient exchange, are relatively short-lived (4–10 days) [1] and thus the level of root colonization could easily change over a plant's lifetime. In juveniles we found more soil hyphal colonization in groups of siblings compared to strangers. Early in life, the net benefit of associating with mycorrhizas is lower compared to later on because the seedling is donating carbon to the fungal partner that could otherwise be used for its own growth and defence [4] . However, higher root colonization at the seedling stage can have benefits for nutrient uptake at the juvenile and adult stages [45] , which could translate into increased final fitness. This benefit would be even greater if plants were colonizing a CMN connected with related individuals, potentially increasing their inclusive fitness. Our findings from both experiments support this idea since sibling pairs had greater arbuscular colonization than strangers, and at a later life-stage, groups of siblings had increased soil hyphae. Greater soil hyphal length in juvenile sibling groups implies that the plants growing with siblings actively increased their investment in the mycorrhizal association. Consistent with predictions from the social good model, siblings appeared to contribute more to the symbiosis compared to strangers by supporting increased fungal growth in the soil. Plants have the ability to control their carbohydrate donations to fungi, preferentially allocating carbon to more beneficial fungal partners over more parasitic ones [11] , leading to increased fungal fitness [13] , so it is also possible that they could preferentially allocate to a CMN attached to siblings versus one attached to strangers. Similar to previous research [44] , we found no trade-offs between fungal traits ( Table S8 ), supporting the argument that soil hyphae is an indicator of plant contribution to fungal growth. The larger network size in groups of related plants implies that the fungus benefits from plant kin selection. Thus, the plant neighbourhood may be a key influence on the fitness of the fungal partner. It might be argued that the increased mycorrhizal association in sibling groups is evidence that the fungal partner can more effectively exploit genetically similar groups. In this parasitism hypothesis, finding more arbuscules in seedlings and more soil hyphae in juveniles can be interpreted as fungal success in sibling groups. Evidence against this parasitism hypothesis would be the observation that plants benefit from increased fungal colonization. We measured two potential short-term benefits that can specifically be attributed to mycorrhizas. First, we found fewer lesions with seedlings associating with mycorrhizas, with sibling pairs having significantly fewer than strangers. This decrease in general lesion number indicates an overall protective effect of mycorrhizas on young seedling roots, suggesting that there are early benefits for siblings who increase their association with mycorrhizal fungi at the seedling stage. The lesions observed on the roots from our seedling study could have come from various sources including fungal pathogens, parasites and root nematodes. However, mycorrhizal fungi are known to benefit plants by protecting them against root lesions through a variety of mechanisms, including competition between pathogens and AM-fungi (reviewed in [1] ). The second observation against the parasitism hypothesis is that our data suggests inoculated pots of juvenile plants had higher total leaf N, a result that is consistent with the generally positive effects of soil hyphal length on plant nutrient status [44] . N and P acquisition are often correlated and N is typically the most important limiting nutrient for plant growth [46] , and pollen and seed production [47] . Therefore, juvenile plants in sibling groups may have had improved nutrient acquisition ability through an extended mycorrhizal network resulting from their increased investment. Thus for both seedlings and juveniles, there are short-term benefits to having greater mycorrhizal association which could result in higher survival and fecundity for plants grown with siblings. This is further evidence supporting the argument for kin selection acting on the ragweed-mycorrhizal symbiosis. Our results suggest that juvenile siblings invested carbon in mycorrhizas even at high P, when the mutualism is likely less necessary for P uptake. Despite a common prediction that plants will have higher association with mycorrhizal fungi in low P [5] , we found that only strangers had this response. In contrast, siblings and solitary plants maintained consistently higher levels of soil hyphae across P levels. A high level of investment in mycorrhizas, despite high P, could provide multiple benefits including bet hedging against future demand for nutrients, increased water acquisition, and pathogen defense [1] , all of which could increase the chances of survival and, therefore, final fitness. These benefits could increase one's indirect fitness when attached to the same CMN as relatives. We were able to reject our alternative hypotheses about the causes of mycorrhizal and plant benefit differences across social environments. Previous studies of plant recognition have found phenotypic plasticity to neighbours in nutrient acquisition traits, including fine roots [27] – [30] . Consequently, one alternative hypothesis is that changes in plant morphology induced by kin recognition caused the differences found in mycorrhizas. However, in neither experiment were there shifts in biomass allocation or aboveground morphological changes in response to social environment. Therefore, plant morphological responses to social environment were not confounded with responses seen in the fungal partner. The only trait showing any social environment interactions was log aboveground biomass in juveniles. Here, the differences among families in solitary vs. shared effects and in kin vs. stranger effects (social environment × family, Table 3 , Fig S4 ) were the consequence of more variance among families in stranger than kin or solitary conditions. In the seedling study, we found no effect of family on fungal structures typically associated with strength of the mutualism, arbuscules (P<0.8706) and hyphae (P<0.7885), allowing us to reject the hypothesis that some plant genotypes may have higher specificity for a given fungus. There were no differences in soil hyphal length between the four genotypes of juvenile plants either ( Fig S5 ). Finding a lack of effect of family on mycorrhizal structures expected to be associated with a stronger symbiosis in both seedling and juvenile studies indicates that the increased colonization in siblings was not due to a particular family having stronger associations with the fungal genotype used in either experiment. We also investigated whether the differences in soil mycorrhizas were the result of soil hyphae being correlated with biomass of the root sample used for fungal quantification, coupled with systematic differences in root biomass between social environments. Post hoc analysis revealed no relationship between root sample mass and soil hyphal length ( Fig S6 ). Above- and belowground biomasses were strongly positively correlated with each other but not with any of the fungal traits. Root hyphal colonization and arbuscular colonization were negatively correlated (P<0.0278). No other fungal traits were correlated ( Table S8 ). Previous research in Arabidopsis thaliana has demonstrated that the mechanism for plant kin recognition involves root exudates [27] . We hypothesize that ragweed also uses root exudates to recognize the identity of surrounding plants. If ragweed recognizes that it is growing near siblings and it is also attached to a mycorrhizal fungal partner, it may altruistically donate more carbon to the fungal partner. Kin selection would favour this increased donation since the benefits that could be provided to neighbouring kin would increase the focal individual's inclusive fitness. Alternatively, if a focal individual recognized its neighbours as strangers, it could avoid costly contributions to the CMN that would benefit non-relatives and provide no inclusive fitness rewards. In conclusion, mycorrhizal colonization and growth was highest in sibling groups, supporting predictions from social good theory that kin selection can stabilize a mutualism [12] . Though a previous study provided evidence that plants benefit from population level specificity to soil fungal communities [37] , here we demonstrate that the mycorrhizal symbiosis is also affected by plant kin recognition. Low nutrient availability is known to favour mycorrhizal colonization [48] . However, our results indicate that plant neighborhood may determine the extent of this nutrient effect, since sibling plants invested more in the mycorrhizal network regardless of P level. Moreover, the effect of social environment on soil hyphae was much greater than the effect of increased P. Thus, even in high P where mutualism break down is predicted, plant kin selection may allow fungal populations to persist. Though these results were found in greenhouse studies, natural population structure created through limited seed dispersal can also generate proximity among siblings [49] , suggesting that kin recognition could be an important mechanism that reinforces the ancient mutualism between plants and fungi."
} | 5,121 |
31282031 | PMC6852117 | pmc | 305 | {
"abstract": "Abstract Evolutionary rescue of populations depends on their ability to produce phenotypic variation that is heritable and adaptive. DNA mutations are the best understood mechanisms to create phenotypic variation, but other, less well‐studied mechanisms exist. Marine benthic foundation species provide opportunities to study these mechanisms because many are dominated by isogenic stands produced through asexual reproduction. For example, Caribbean acroporid corals are long lived and reproduce asexually via breakage of branches. Fragmentation is often the dominant mode of local population maintenance. Thus, large genets with many ramets (colonies) are common. Here, we observed phenotypic variation in stress responses within genets following the coral bleaching events in 2014 and 2015 caused by high water temperatures. This was not due to genetic variation in their symbiotic dinoflagellates ( Symbiodinium “ fitti ”) because each genet of this coral species typically harbours a single strain of S. “ fitti ”. Characterization of the microbiome via 16S tag sequencing correlated the abundance of only two microbiome members ( Tepidiphilus , Endozoicomonas ) with a bleaching response. Epigenetic changes were significantly correlated with the host's genetic background, the location of the sampled polyps within the colonies (e.g., branch vs. base of colony), and differences in the colonies’ condition during the bleaching event. We conclude that long‐term microenvironmental differences led to changes in the way the ramets methylated their genomes, contributing to the differential bleaching response. However, most of the variation in differential bleaching response among clonemates of Acropora palmata remains unexplained. This research provides novel data and hypotheses to help understand intragenet variability in stress phenotypes of sessile marine species.",
"introduction": "1 INTRODUCTION Acclimatization is a nongenetic process by which an individual can increase its stress tolerance after exposure to a stressor, such as temperature anomalies (see Palumbi, Barshis, Traylor‐Knowles, & Bay, 2014 for a coral example). Acclimatization can lead to phenotypic variability in stress response among clonemates. Nonmutation‐based mechanisms resulting in phenotypic variability in isogenic lines include stochastic gene expression, errors in protein synthesis, protein promiscuity and epigenetic modifications (reviewed by Payne & Wagner, 2019 ). Of these, only epigenetic mutations have been studied as a mechanism for acclimatization in marine foundation fauna (reviewed by Eirin‐Lopez & Putnam, 2019 ). Interestingly, nonmutation‐based changes to phenotypes can prolong survival of a genet until such phenotypes become permanent (Yanagida et al., 2015 ), although these processes are not well understood. Understanding of all the mechanisms that produce phenotypic variability is essential to estimate the evolvability of threatened marine species (Payne & Wagner, 2019 ). The large‐scale bleaching event during 2014 and 2015 within the Florida Keys provided an unprecedented opportunity to understand the role of acclimatization and phenotypic variability in framework‐building foundation species of shallow Caribbean coral reefs. Because reef‐building corals harbour intracellular algal symbionts (family Symbiodiniaceae), discerning the relative contribution of host and symbiont to holobiont acclimatization can be difficult. However, the Caribbean elkhorn coral, Acropora palmata , has an uncomplicated symbiosis: it associates with just one symbiont species ( Symbiodinium “ fitti ”) and most colonies (~70%) harbour only one strain of S. “ fitti ” over space and time (Baums, Devlin‐Durante, & LaJeunesse, 2014 ) similar to other cnidarian–Symbiodiniaceae associations (e.g., Andras, Kirk, & Drew Harvell, 2011 ; Thornhill, Xiang, Fitt, & Santos, 2009 ). Thus, A. palmata is an excellent model to discern the coral host's acclimatization response and phenotypic variability to heat stress. Like many reef‐building corals, A. palmata frequently reproduces via fragmentation (an asexual process), sometimes forming large, monoclonal stands (Baums, Miller, & Hellberg, 2006 ; Foster, Baums, & Mumby, 2007 ; Pinzón, Reyes‐Bonilla, Baums, & LaJeunesse, 2012 ; Williams, Miller, & Baums, 2014 ). These colonies represent iterations of the same host–symbiont combination (i.e., they are isogenic replicates), experiencing similar environmental conditions. Initial surveys of A. palmata during the 2014 and 2015 bleaching events documented a range of bleaching responses. This response varied between reefs but also within single, monoclonal stands of A. palmata (see Figure 2 ). Thus, coral clonemates exhibited different bleaching susceptibilities despite data showing that they share identical (clonal) S. “ fitti ” symbiont communities, begging the question as to what mechanisms account for such phenotypic variability. Thus, we explored whether differences in the microbiome other than dinoflagellates (e.g., prokaryotes), and/or micro‐environmental differences, such as shading or exposure to water movement, induced epigenetic changes in the host genome that could explain differences in bleaching susceptibilities among ramets of the same genet. The answer may inform our understanding of how reefs might survive climate change and has implications when choosing genets for restoration. Corals associate with a diverse set of prokaryotes (Pollock et al., 2018 ). The microbiome plays important roles in the functioning of the holobiont, including coral nutrition, element cycling and disease responses (Peixoto, Rosado, Leite, Rosado, & Bourne, 2017 ). This community can shift in response to stressors such as heat, low pH or disease (Muller, Bartles, & Baums, 2018 ; Thurber et al., 2009 ). We are beginning to understand the specific function of some members of the microbiome such as an endosymbiotic cyanobacterium that fixes nitrogen in Montastraea cavernosa (Lesser et al., 2007 ) and the disease‐causing Vibrio corallyticus (Kimes et al., 2012 ). Other species, such as Endozoicomonas sp., show patterns of association with coral hosts that suggest an important role, but that role is not yet well understood (Neave et al., 2017 ; Pollock et al., 2018 ). DNA methylation is the most highly studied mechanism in epigenetics and is often used to elucidate how a phenotype is modified without altering the genetic code. DNA methylation occurs at the cytosine bases of eukaryotic DNA, which are converted to 5‐methylcytosine by DNA methyltransferase (DNMT) enzymes. Additional epigenetic changes include histone modifications, chromatin remodelling and gene regulatory mechanisms involving small noncoding RNAs (Danchin et al., 2011 ). Epigenetic modifications can rapidly produce new phenotypes in response to a change in the environment without mutations in the underlying genetic sequence (Finnegan, 2002 ; Richards, 2008 ). There are limited data on methylation mechanisms in invertebrates. Current evidence has shown that invertebrate genomes are far less methylated than vertebrate genomes (Gavery & Roberts, 2010 , 2013 ; Lyko et al., 2010 ; Olson & Roberts, 2014 ; Rivière, 2014 ; Suzuki, Kerr, Sousa, & Bird, 2007 ). Additionally, DNA methylation is predominately found in gene bodies in which highly expressed housekeeping genes are hypermethylated and regulated and/or inducible genes are hypomethylated (Dimond & Roberts, 2016 ; Elango, Hunt, Goodisman, & Soojin, 2009 ; Gavery & Roberts, 2010 ; Hunt, Brisson, Soojin, & Goodisman, 2010 ; Sarda, Zeng, Hunt, & Soojin, 2012 ). Hypermethylation of those genes essential for biological function is thought to imply that they are “protected” from plasticity in transcriptional opportunities. Such plasticity would be inherently lethal in housekeeping genes (Roberts & Gavery, 2012 ), and thereby gene body methylation (GBM) in corals is correlated with stable and active transcription (Dixon, Liao, Bay, & Matz, 2018 ; Liew, Zoccola, et al., 2018 ). Methylated genes in the anemone Aiptasia show a significant reduction of spurious transcription and transcriptional noise (Li et al., 2018 ). In contrast, the inducible gene's limited methylation may facilitate, albeit passively, specific transcriptional opportunities including increasing sequence mutations, access to alternative transcription start sites and exon skipping (Roberts & Gavery, 2012 ). There is evidence for a direct relationship between DNA methylation and phenotypic plasticity, as seen in the determination of caste structure in both honeybees and ants (Kucharski, Maleszka, Foret, & Maleszka, 2008 ), maternally inherited epigenetic patterns that influence the expression of the agouti gene in mice (Wolff, Kodell, Moore, & Cooney, 1998 ), and the differential methylation of the human NR3C1 gene in newborns depending on prenatal maternal mood (Oberlander et al., 2008 ). Scleractinian corals can display strong differences in their DNA methylation in response to stress, demonstrating that de novo DNA methylation may be a driving mechanism for phenotypic plasticity in acclimatization (Putnam, Davidson, & Gates, 2016 ). Here, we sampled six different genets from four different reef sites in Florida (Figure 1 ) and sampled each genet six times (either across the same colony six times, or different ramets of the same genet for a total of six samples). Samples were taken from the upward‐facing side of branches or the bases/trunks of colonies 6 weeks after colonies had experienced a thermal stress event that resulted in differential bleaching. We determined how many S. “ fitti ” strains were present in each sample and sequenced the 16S gene to characterize variability in the prokaryotic community. We then applied a reduced sequencing technique sensitive to the methylation status of cytosine called MethylRad (Wang et al., 2015 ) to identify sites that were differentially methylated between reefs, genets, position within the colony and peak bleaching status. Figure 1 \n Acropora palmata samples were obtained from four reef sites within the Florida Reef tract, Grecian Rocks, Sand Island, Molasses and Key Largo Dry Rocks. Distance between Grecian Rocks and Key Largo Dry Rocks is about 20 km. Maps created in Google Earth Pro [Colour figure can be viewed at http://wileyonlinelibrary.com ]",
"discussion": "4 DISCUSSION Evolutionary rescue of populations depends on a species’ ability to produce phenotypic variation that is heritable and adaptive. DNA mutations are the best understood mechanisms to create phenotypic variation, but other, less well‐studied mechanisms exist. Environmental conditions frequently change over the long lifespan of some reef‐building coral genets (Devlin‐Durante, Miller, Caribbean Acropora Research Group, Precht, & Baums, 2016 ). Phenotypic plasticity in response to environmental variation is a common trait in corals; however, the mechanisms by which corals achieve this plasticity are not well understood. This can be partly attributed to the difficulty of separating host response from those of algal symbionts and other members of the microbiome. Acropora palmata colonies showed variable responses to increased water temperatures, yet the identity of the associated symbiont strain did not explain this plasticity. In addition, photographic survey results showed little evidence of short‐term acclimatization. Interestingly, the abundance of two species of prokaryotes, Tepidiphilus and Endozoicomonas , differed among microbiome samples depending on their bleaching response and this supports findings that Endozoicomonas plays an important role in the functioning of the coral holobiont (Pollock et al., 2018 ). Ultradeep sequencing of host and symbionts may reveal the occurrence of somatic mutations in hosts (and/or symbionts) correlated with the bleaching phenotype (Van Oppen, Souter, Howells, Heyward, & Berkelmans, 2011 ) but nonmutation‐based mechanisms may also play a role (Goldsmith & Tawfik, 2009 ; Payne & Wagner, 2019 ), including detection‐based epigenetic modifications (sensu Shea, Pen, & Uller, 2011 ). Here we, for the first time, applied a new technique to assay methylation status in wild corals and have found that coral genets show large and consistent differences in the way they methylate their DNA even when growing on the same reef, consistent with the hypothesis that some portion of their methylation patterns was inherited (Dimond, Gamblewood, & Roberts, 2017 ; Liew, Howells, et al., 2018 ). Furthermore, there was significant variation in methylation state that was correlated with the location of the polyps within colonies and the bleaching response of the colonies. Changes in epigenetic marks based on the detection of environmental cues provides an avenue to effect phenotypic plasticity (Shea et al., 2011 ). Detailed data on systematic variation in temperature, flow and light regimes depending on polyp location within a colony and colony morphology are accumulating (Edmunds & Burgess, 2018 ; Ong, King, Kaandorp, Mullins, & Caley, 2017 ; Stocking, Rippe, & Reidenbach, 2016 ). Remarkably, there was some correspondence between genes previously shown to be differentially expressed, and those that were differentially methylated, indicating that methylation differences may translate into gene expression differences. These results are novel because they reveal a potential pathway by which these long‐lived corals can modify their phenotypes in response to the environment. Because these corals can produce large monoclonal stands, modifications of genome methylation produce mosaics of phenotypes despite low genotypic diversity. Whole genome bisulphite sequencing analysis should be performed to further investigate these patterns. Future work on this and other marine foundation species may significantly advance our understanding of the mechanism determining evolvability of threatened species. 4.1 Microbial assemblages and symbiotic dinoflagellates are minimally correlated with bleaching condition Corals live in an intimate symbiosis with algae in the family Symbiodiniaceae (LaJeunesse et al., 2018 ). While some coral species can host several species of algae, the elkhorn coral, A. palmata , usually hosts Symbiodinium “ fitti ” and furthermore ramets of the same coral genet often retain the same symbiont strain (Baums et al., 2014 ). This was also the case here. Genets differed in the S. “ fitti ” strain they hosted but within each genet, there was remarkable homogeneity. These findings allow us to discount genetic diversity of S. “ fitti ”, at least at the strain level, as an explanation for differential bleaching responses. However, variable densities of symbiotic dinoflagellates within the host tissues may have influenced bleaching susceptibility (Kemp et al., 2014 ; Stimson et al., 2002 ). We do not have prebleaching measurements of symbiont density in tissues, but this may have contributed to variable bleaching within a colony. Genetic diversity of S. “ fitti ” below the strain level may also influence the bleaching response. A. palmata colonies host large populations of S. “ fitti ” that are the result of countless cell divisions. Somatic mutations may accumulate and lead to functional variation among cells of the same S. “ fitti ” strain, as defined by a shared multilocus microsatellite genotype profile. Experimental evolution of Symbiodiniaceae cultures has recently been shown to result in functional divergence (Chakravarti & van Oppen, 2018 ). Ultradeep sequencing of S. “ fitti ” isolated from replicate samples representing large A. palmata genets would probably detect such somatic mutations (Wang et al., 2019 ), but assigning function to these mutations in symbionts associated with wild coral colonies remains a daunting challenge. Prokaryotic members of the coral microbiome play important roles in coral nutrition, element cycling and disease responses (Peixoto et al., 2017 ). We thus speculated that the differential bleaching response within and between colonies might be attributable to differences in the microbiome species composition. However, there were only two bacterial genera in which variance attributable to the bleaching condition of the coral exceeded 5%: Tepidiphilus ( σ \n 2 = 12.1%) and the coral symbiont Endozoicomonas ( σ \n 2 = 6.8%). Each taxon was represented by one strain only. There are three proposed functions of Endozoicomonas in the host coral, including maintaining the structure of the host microbiome, nutrient acquisition and provision, and a role in host health and/or disease (Neave et al., 2017 ). Evidence for a role in coral health was demonstrated through the altered abundances of Endozoicomonas in multiple scleractinian corals in relation to seawater pH (Webster et al., 2016 ), sedimentation and wastewater runoff (Ziegler et al., 2016 ), and the occurrence of diseased lesions (Meyer, Paul, & Teplitski, 2014 ). Differences in the microbiome that pre‐dated or immediately followed the bleaching event were obscured, because we were only able to sample tissues once they had recovered from bleaching. Previous studies looking at stress events in corals found a shift of the coral holobiont to a more potentially pathogenic state with disease‐associated bacteria and fungi (Thurber et al., 2009 ), including the fungus Ascomycota (Thurber et al., 2009 ) and the bacterial taxa Vibrionales (Thurber et al., 2009 ; Zaneveld et al., 2016 ) and Oscillatoriales (Zaneveld et al., 2016 ). A shift to a pathogenic state was not observed in our colonies, indicating that whatever impacts bleaching may have had on the microbiome, those effects were difficult to detect 6 weeks post‐stress using a standard 16S microbiome analysis approach. Future metagenome sequencing may reveal genetic diversity among prokaryotes that shared the same 16S sequencing tag and help to explain some of the unaccounted‐for variance among ramets of the same genet. 4.2 Variation in methylation patterns by genet It is important to distinguish inheritance of epigenetic marks in the soma versus the germline (Shea et al., 2011 ). During cell proliferation, somatic tissues inherit epigenetic marks from progenitor cells to, for example, give an epithelial cell its identity. Similarly, polyps within a colony and colonies belonging to the same genet share epigenetic signatures via somatic inheritance. However, when comparing genome methylation patterns between A. palmata genets, we found large differences even when the genets lived on the same reef (Figure 2 ). This suggests that at least some portion of genome methylation was inherited, otherwise a shared environment post‐fertilization should lead to shared methylation patterns among genets. Evidence for inherited germline methylation patterns in corals is accumulating (Dimond et al., 2017 ). Widespread depletion of CpG dinucleotides was observed in Acropora millepora and is a signature for historical germline DNA methylation (Dixon, Bay, & Matz, 2014 ). A recent study in Platygyra daedalea for the first time demonstrated intergenerational inheritance of DNA methylation patterns in corals, from parent to sperm, and evidence for maternal and paternal effects in larvae from reciprocal crosses (Liew, Howells, et al., 2018 ). In most animals with early germline segregation, epigenetic marks are reset during meiosis, and the mechanisms of inheritance of epigenetic marks in coral embryos are unknown (reviewed by Eirin‐Lopez & Putnam, 2019 ). Symbiont–host interactions may also influence host genome methylation patterns. Even genets that grew near each other on the same reef hosted a different strain of S. “ fitti ” while ramets of the same genet usually shared an S. “ fitti ” strain. Although previously undocumented, S. “ fitti ” strains may differentially alter the host methylome. Supporting evidence for this comes from genet by genet comparison in a GO enrichment analysis, which showed that the category “modification by symbiont of host morphology or physiology and modulation by symbiont of host cellular processes” was enriched. Differentially methylated genes in this category included the Homeodomain‐interaction protein kinase 2, eIF‐2‐alpha kinase GCN2 (three separate methylation sites within this gene), Gag‐Pol polyprotein (two separate methylation sites within this gene) and TNFAIP3‐interating protein 1. Genotype/genotype interactions between Symbiodiniaceae strains and host genets and their effects on host methylomes deserve further study (reviewed by Parkinson & Baums, 2014 ). 4.3 Methylation patterns vary between locations within the colony Some of the variance in genome methylation within genets was attributable to long‐term microenvironmental conditions between polyp locations within a colony rather than eukaryotic or bacterial symbiont community composition. Complex skeletal morphologies and varying tissue layer thickness create a variety of intracolonial light microniches, but in general the top of a branch will experience significantly higher solar irradiance than the base or trunk of a colony (Kaniewska et al., 2011 ; Wangpraseurt, Larkum, Ralph, & Kühl, 2012 ; Warner & Berry‐Lowe, 2006 ), and therefore polyps on the tops of branches have an increased need to avoid the damaging effects of excess light energy and the resulting oxidative stress. In A. globiceps , Symbiodiniaceae densities were consistent between internal and external branches but varied with depths (greater densities at lower depths) (Ladrière et al., 2013 ). The host coral can regulate Symbiodiniaceae densities through nutrient limitation or through digesting or expelling the excess symbiotic algae to maintain relatively low and consistent densities (Dunn, Bythell, Tissier, Burnett, & Thomason, 2002 ; Falkowski, Dubinsky, Muscatine, & McCloskey, 1993 ; Muscatine et al., 1998 ) as one option to reduce oxidative stress in areas or times of higher light exposure (Fitt, McFarland, Warner, & Chilcoat, 2000 ). The various mechanisms employed by the host, including epigenetic changes as suggested here, help the colony avoid bleaching even in the high‐light exposed polyp locations under nonstressful temperature conditions. Abnormally high temperatures accompanied by high irradiance can cause a breakdown in the coral–symbiotic algae symbiosis, resulting in expulsion of the algae, a process referred to as bleaching. We observed a higher incidence of bleaching in samples from branch regions (58%) versus those collected from the base (30%) in this study. In addition, there is also strong evidence for a division of labour between coral branch tips and bases in their gene expression (Hemond, Kaluziak, & Vollmer, 2014 ). Interestingly, galactosylceramide sulfotransferase was significantly differentially methylated between polyps sampled from branches versus other locations within colonies such as the base or trunk (Table S10 ). This result was obtained with both likelihood ratio T‐test and QL F ‐test statistics. Galactosylceramide sulfotransferase is involved in sphingolipid metabolism and was also differentially expressed by colony position, being upregulated in branch tips in A. palmata and A. cervicornis (Hemond et al., 2014 ). Sphingolipid metabolism may be involved in the regulation of algal symbionts. In anemones, the sphingosine rheostat can regulate the balance between stability and dysfunction in the cnidarian–dinoflagellate partnership (Detournay & Weis, 2011 ). However, under stressful temperatures, parts of the colonies that were exposed to high irradiation were unable to avoid bleaching even with the epigenetic modifications driven by the internal light gradient. Symbiodiniaceae in shallow corals must dissipate four times more light energy than is needed for photosynthesis on a bright summer day (Gorbunov, Kolber, Lesser, & Falkowski, 2001 ). This excess light energy absorbed by chlorophyll can be dissipated through heat loss, re‐emitted as fluorescence, or decayed via the chlorophyll triplet state that produces reactive oxygen species as a byproduct. Here, we identified differentially methylated sites by polyp location that were overrepresented in the GO categories of AMP biosynthetic and metabolic processes. The site is in the gene adenylosuccinate lyase, which catalyses two key steps in AMP synthesis. cAMP induces gene transcription through activation of cAMP‐dependent protein kinase (PKA) and subsequently activation of transcription factors including CREB (cAMP response element binding proteins)/ATF transcription factor family members such as CREM and ATF1 via phosphorylation by PKA. In a differential gene expression analysis of A. palmata fragments kept in complete darkness for 9 days compared to controls, two annotated genes were identified, cAMP‐responsive element modulator and cyclic AMP‐dependent transcription factor ATF‐4 (DeSalvo, Estrada, Sunagawa, & Medina, 2012 ). The ATF‐4 transcription factor responds to oxidative stress and amino acid starvation (Harding et al., 2003 ). The coral skeleton serves as an efficient light‐capturing device and colony and polyp morphology determine the light levels experienced by the intracellular symbionts (Enríquez, Méndez, & Iglesias‐Prieto, 2005 ; Swain et al., 2018 ). Symbiodiniaceae can also maximize light absorption and utilization by increasing photosynthetic pigments and photosynthetic efficiency in corals acclimatized to low light (Falkowski & Dubinsky, 1981 ). Among the enriched GO terms between branch and nonbranch polyp locations was the category protein‐chromophore linkage. The differentially methylated gene was in cyrptochrome‐1. Cryptochromes are flavoproteins that are sensitive to blue light. They regulate the circadian clock in plants and animals. Eight core circadian genes have been identified: Casein kinase 1e (CK1e), Cryptochrome1 (Cry1), Cryptochrome2 (Cry2), Period1 (Per1), Period2 (Per2), Period3 (Per3), Clock and BMAL1 (brain and muscle ARNT‐like protein, Arntl, MOP3). Cryptochrome1 and 2 have been previously reported to display diurnal patterns of transcription in corals, with higher expression found in the light phase than in the dark (Hoadley, Szmant, & Pyott, 2011 ; Levy et al., 2007 , 2011 ). Cry1 in A. millepora is not under control of an endogenous clock whereas Cry2 is. Both have higher expression during the day (Brady, Snyder, & Vize, 2011 ). Circadian clock genes affect a large number of downstream processes and thus serve as important nodes in transcriptional networks (Dunlap, 1999 ) and in the regulation of post‐translational modifications (Gallego & Virshup, 2007 ; Staiger & Koster, 2011 ). Differential methylation of these genes may thus be an effective means to alter the transcription of several downstream pathways in response to differential light levels within colonies. Future research related to the transcription and differential methylation of these circadian clock genes in corals is warranted. We unexpectedly found that the GO terms for the regulation of viral genome replication, viral life cycle and viral processes were enriched in the comparison between branch and nonbranch locations. We are not aware of any data indicating that viral load differs within colonies. Because this GO term was enriched across genets and reefs, we would expect viral loads to differ systematically between branch and nonbranch locations and this hypothesis deserves future testing. Interestingly, the bacterial communities did differ between the tips and the bases of colonies, suggesting that the branches may harbour a specialized microbiome. There is contradicting previous evidence with respect to within‐colony variation of the prokaryotic community in corals. In a previous study on A. palmata , no detectable community‐level differences were found among the prokaryotic microbiota of the uppermost, underside and base of A. palmata ( R \n 2 = 0.20, p = 0.51) (Kemp et al., 2015 ). In contrast, considerable within‐colony variation of bacterial assemblages was found in Orbicella annularis between the tops and the sides (Daniels et al., 2011 ). O. annularis also harbours several species of Symbiodiniaceae, and hence further research is required to understand what factors drive within‐coral‐colony diversity of the prokaryotic community. 4.4 Methylation patterns vary with bleaching history Methylation variation within genets was also attributable, to some extent, to whether tissues had recently bleached. The one significant methylation site was located within a genomic region that does not have a gene prediction or annotation. The closest gene annotation, at a little over 10,000 bp away, is Transposon TX1 uncharacterized 149‐kDa protein as found in Stylophora pistillata . By the alteration of splicing and polyadenylation patterns or through functioning as enhancers or promoters, transposable elements can exercise control over neighbouring genes (Girard & Freeling, 1999 ). Transposable elements are significantly differentially expressed in response to heat stress in corals (DeSalvo et al., 2010 ; Traylor‐Knowles, Rose, Sheets, & Palumbi, 2017 ) and in plants (Ito et al., 2011 ; Pecinka et al., 2010 ). Overall, less of the variation in methylation was explained by previous bleaching, suggesting that methylation changes may be effective in changing coral transcription in response to longer term differences in the light environment (e.g., between tips and bases) rather than more acute temperature stressors. Future research is required to correlate gene expression and methylation over a range of stress exposures and stress severity. High fragmentation rates and acute stress events necessitate that A. palmata polyps acclimatize to changes in environmental conditions. Our data suggest that acclimatization may be partially achieved via differential methylation. We suggest that differential genome methylation may be one of the mechanisms by which corals achieve their remarkable phenotypic plasticity in their natural environment. In a transplant experiment in A. millepora , GBM changed subtly, but much less than transcription (Dixon et al., 2018 ). Dixon et al. also found strong associations between gene body methylation and fitness, although gene body methylation was not directly correlated to transcription, resulting in the authors’ questioning what mechanism connects gene body methylation to phenotype and fitness (Dixon et al., 2018 ). Yet, methylation is known to affect transcription factor binding both negatively and positively, and thus alter transcriptional regulation bidirectionally, making correlation to gene expression complicated (Yin et al., 2017 ). A significant amount of the intragenet variation in phenotypic stress response observed here remains to be explained. Similarly, large residual variances in methylation that could not be attributed to any of the studied factors were observed in Porites porites (Dimond et al., 2017 ). Epimutations are stochastic events that result in random additions and losses of epigenetic marks (reviewed by Johannes & Schmitz, 2019 ), some of which are heritable. Distinguishing stochastic versus directed changes in epigenetic marks in corals will require careful experimentation. Plant researchers have made use of mutation accumulation lines grown under controlled conditions to make the distinction (e.g., Becker et al., 2011 ). Further, ultradeep sequencing of host and symbionts may reveal somatic mutations correlated with the bleaching phenotype (Van Oppen et al., 2011 ), although mechanisms other than detection‐based DNA methylation changes may also play a role (Goldsmith & Tawfik, 2009 ; Payne & Wagner, 2019 ), including selection‐based changes in epigenetic marks (Shea et al., 2011 ). Future work on this and other marine foundation species may significantly advance our understanding of these mechanisms in determining the evolvability of threatened species."
} | 8,039 |
37200786 | PMC10187575 | pmc | 306 | {
"abstract": "SUMMARY Research background This study provides insight into the use of a designed microbial community to produce biohydrogen in simple, single-chamber microbial electrolysis cells (MECs). The ability of MECs to stably produce biohydrogen relies heavily on the setup and microorganisms working inside the system. Despite having the most straightforward configuration and effectively avoiding costly membranes, single-chamber MECs are prone to competing metabolic pathways. We present in this study one possible way of avoiding this problem using characteristically defined, designed microbial consortium. Here, we compare the performance of MECs inoculated with a designed consortium to MECs operating with a naturally occurring soil consortium. Experimental approach We adapted a cost-effective and simple single-chamber MEC design. The MEC was gastight, 100 mL in volume, and equipped with continuous monitoring for electrical output using a digital multimeter. Microorganisms were sourced from Indonesian environmental samples, either as denitrifying bacterial isolates grouped as a designed consortium or natural soil microbiome used in its entirety. The designed consortium consisted of five species from the Pseudomonas and Acinetobacter genera. The headspace gas profile was monitored periodically with a gas chromatograph. At the end of the culture, the composition of the natural soil consortium was characterized by next generation sequencing and the growth of the bacteria on the surface of the anodes by field emission scanning electron microscopy. Results and conclusions We found that MEC using a designed consortium presented a better H 2 production profile, with the ability of the system to maintain headspace H 2 concentration relatively stable for a long time after reaching stationary growth period. In contrast, MECs inoculated with soil microbiome exhibited a strong decline in headspace H 2 profile within the same time frame. Novelty and scientific contribution This work utilizes a designed, denitrifying bacterial consortium isolated from Indonesian environmental samples that can survive in a nitrate-rich environment. Here we propose using a designed consortium as a biological approach to avoid methanogenesis in MECs, as a simple and environmentally friendly alternative to current chemical/physical methods. Our findings offer an alternative solution to avoid the problem of H 2 loss in single-chamber MECs along with optimizing biohydrogen production through bioelectrochemical routes.",
"conclusion": "CONCLUSIONS Designed consortium, preselected for its ability to grow in a nitrate-rich environment and carry out the denitrification process, performed better than native soil consortium for biohydrogen production. H 2 profile was sustained for a longer period without signs of transformation to methane, a familiar yet undesired phenomenon in single-chamber microbial electrolysis cells (MECs). The single-chamber configuration of MECs presents advantages over multi-chamber configurations thanks to its simplicity. Here we exploit this configuration to produce biohydrogen in a simple laboratory-scale setup. The results presented in the study suggest that reducing microbiome complexities in the inoculum may be beneficial to avoid undesired transformative pathways in MECs. This effect is evident when using preselected or ‘designed’ communities for specific characteristics. In this study, we demonstrate the avoidance of methanogenesis by co-culturing denitrifying bacteria in MECs with prior understanding of the inhibitory effect of denitrification on methanogenic bacteria. Further studies are needed to better understand the biological aspect of this phenomenon by utilizing more powerful analytical tools to explore the complexity of the two consortia better. It would also be interesting to optimize the designed microbiome's performance for biohydrogen generation using other consortium formulations and different growth media, \ni.e. wastewater as nitrate-rich growth medium.",
"introduction": "INTRODUCTION Bioelectrochemical systems (BES), more widely known as their derivatives microbial fuel cell (MFC) and microbial electrolysis cell (MEC), are electrochemical cells that utilize microorganisms to carry out reduction/oxidation reactions. Microorganisms responsible for the process can be referred to as electroactive bacteria, exoelectrogens, or anode/cathode-respiring bacteria. These organisms are collectively called electroactive bacteria because of their unique ability to transport electrons through biological membranes either from or to the environment ( 1 ). BES as a technology platform has been studied only recently, within the past two decades. In this field, the research focus has varied among optimization of the operational conditions of the system, classical study of electroactive microorganisms, or the design of the platform itself. BES has been studied for many applications, including wastewater treatment, fuel gas production as H 2 and CH \n4 , nutrient removal and recovery, chemical synthesis, desalination and bioremediation ( 2 – 5 ). Despite being coined as the future of clean energy, the majority (around 95%) of produced H 2 is obtained from fossil fuels through chemical conversion routes ( 6 ). The majority of H 2 is produced through thermal processes of natural gas or biomass, i.e. steam reforming and gasification. Alternatively, H 2 can be obtained through water-splitting methods like electrolysis or photolysis of H 2 O ( 7 ). The research focus of the H 2 production is now on increasing process efficiency and better economics ( 8 , 9 ). However, to meet the demand for a cleaner H 2 production method, bioprocesses have emerged with alternative processes like fermentation and bioelectrolysis (MEC) to generate biohydrogen as end-product with advantages of moderate operational parameters, lower energy requirements and better environmental footprints than fossil resources ( \n 4 , 7 ). When using MEC to produce H 2 , single-chamber configuration was proposed as a solution to avoid the higher cost incurred by the use of membranes found in two- or multi-chamber BES, as well as to reduce resistance due to the presence of a physical barrier between compartments ( 10 ). In a single-chamber configuration, both anodes and cathodes are located in the same space. Another advantage of the lack of membrane is reduced energy loss and higher energy recovery efficiency ( 4 ). However, since there are no practical barriers like in the multi-chamber configuration, difficulties can be met in the production of several end-products due to purity issues and product transformation to other unwanted metabolites. For example, in MEC operated under anaerobic conditions, the occurrence of methanogenesis greatly hinders effective biohydrogen production ( 10 \n ). Methanogens are responsible for this phenomenon. These microbes are obligate anaerobic microorganisms able to produce CH 4 out of H 2 or carbon substrate ( 11 ). High methanogenic activity is one of the most commonly reported causes of failure for MEC ( 11 – 14 ), along with the fact that most large MECs use wastewater, which may play a role in their low performance ( 15 ). As a result, there is an obvious need to improve H 2 recovery in MECs. Methanogenesis and denitrification have a very complex relationship. Previous studies have examined their interactions in natural and synthetic environments ( 16 , 17 ). Overall, methanogenic bacteria were inhibited by the activity of denitrifying bacteria. The inhibitory effect of denitrification on methanogenesis opens the door to exploiting denitrifying bacteria as a control method to suppress the growth of methanogenic bacteria in MEC. Traditionally, bioelectrochemical cells rely on microbe-rich inocula to fulfil their goals, most notably using digested sludge since complex microbial communities perform better in this setting ( 18 , 19 ). The use of a designed consortium is a developing research topic in metabolic engineering. In the bioelectrochemical field, designed consortia were used previously to study interspecies electron transfer mechanisms in biogas digestors ( 20 ) as well as to demonstrate the synergistic effect of two species ( Pseudomonas aeruginosa PA14 and Enterobacter aerogenes ) on electricity generation ( 21 ). Previously, co-culturing Shewanella oneidensis with Escherichia coli in MFC resulted in higher electrical output with the synergistic effect forming in a short time ( \n 22 ). A recent study has shown positive interaction between Geobacter sulfurreducens and Ethanoligenes harbinense in a co-culture for H 2 production in a single-chamber MEC, although methanogenesis is not discussed there ( 23 ). He et al. ( 24 ) also discussed that metabolic engineering approaches like co-culturing bacteria capable of metabolizing CH 4 along with electroactive bacteria may be the future alternative method of suppressing methanogenesis. This shows that metabolic engineering at the community level is under active research for MECs. In the past years, our research group has identified nine native microbes from 19 isolates found in local environmental soil and water samples that are spread among two genera: Acinetobacter and Pseudomonas ( 25 ). High-throughput screenings of these microbes suggest varying abilities for denitrification and exoelectrogenic activity. This study proposes the use of a designed consortium instead of an uncharacterized microbiome commonly used in bioelectrochemical cells. For this research, we would like to develop functional communities out of our denitrifying isolates to enhance biohydrogen production in MECs by avoiding the transformation of H 2 in other competing metabolic pathways, notably its transformation to CH 4 by methanogenesis. We expect that by reducing the complexity of the inoculum, headspace H 2 concentration can be sustained despite working in a simple single-chamber MEC setup.",
"discussion": "RESULTS AND DISCUSSION Biohydrogen production Biohydrogen profile obtained using the system is shown in Fig. 2 . At the beginning of the operation, the setup was run without external voltage. This period was designated as a preparatory period for the inoculum, during which no H 2 was produced (0–22 h) due to the lack of energy available to overcome the thermodynamic barrier of H 2 generation in MECs. Once the system was run with external voltage, both setups produced H 2 exponentially . This is clear if we examine the 0–30 h period. MECs inoculated with native soil bacteria exhibited a decline after peaking at 35% of headspace H 2 concentration at 51 h of operation. The decrease in the headspace concentration when using soil inoculum started prior to medium replenishment. It continued even after the medium was replenished, unlike with designed consortium, which responded to medium replenishment with a slight increase in H 2 concentration. On the other hand, headspace H \n2 concentration with the designed microbial consortium stabilized at ~43% at the end of the observation (270–520 h), slightly under its peak value of 47% at 167 h. The two microbial consortia started to differ significantly in H 2 concentration at 99 h of the operation (p<0.05) and continued to do so until the end despite exhibiting similar profiles in the period leading up to this point (0–60 h). Fig. 2 H 2 composition in the headspace. The system was run without an external voltage supply in the first 22 h in the anode preparation stage. Medium replenishment at 10% working volume was carried out after the first current drop below 0.01 mA The decrease of H 2 concentration in the headspace is likely to be attributed to a transformation into methane (methanogenesis) that is common in this type of reactor configuration. Methanogenesis was often found as a cause of failure to obtain biohydrogen in MECs. In the original study that inspired our setup, total conversion of H 2 to CH 4 occurred, leading to undetected quantities of H 2 in the headspace at the end of the culture ( 30 ). Our work has managed to maintain H 2 at a higher level throughout, ~45% with the designed consortium and ~18% with the native soil consortium over the observed culture period. Other works tried to avoid methanogenesis by physical or chemical means like adding antibiotics/inhibitors ( 35 , 36 ), intermittent oxygenation ( \n37 ) or ultraviolet irradiation ( 38 ). Each of these methods has its own benefits and limitations. For example, the addition of antibiotics in the culture medium poses a risk of a potential spill of the resistance trait over to the environment if care is lacking. Aside from H 2 , several other components of headspace gas were also detected ( Fig. 3 ). These components are H 2 , N 2 in the air, CO 2 and N 2 O. Our method has a limitation in separating air into its molecular components of N 2 and O 2 . Hence N 2 is referred to in the results as ‘N 2 in the air’. Additionally, O 2 is practically absent from the system due to the vacuum-flush cycle and the addition of oxygen-scavenging species in the media. Fig. 3 Headspace gas profile from the two microbiomes: a) all gases with designed consortium and b) all gases with soil consortium. Trace amounts of gases from: c) designed and d) soil consortia. Data are expressed as relative abundance over total detected gas concentration N 2 O is an intermediary metabolite in the denitrification process ( 39 ). The existence of this gas in the headspace suggests that nitrate-reducing activity was present in the system. The presence of N 2 in the system is expected since the gas is used in the beginning to purge O 2 out of the system. Hence, it cannot be used as a marker for a complete denitrification process despite being the end product of the pathway. Methane was interestingly absent from detection in this study. A possible explanation could be the transformation of methane into other metabolites like CO 2 in anaerobic methane oxidation, which could be the mechanism behind the significant jump in CO 2 concentration at the end of the cycle with soil MEC. The presence of CO 2 in MECs is otherwise normal since the breakdown of organic matter/substrate in the anode often results in CO 2 release ( 4 ). Methanogenic bacteria and denitrifying bacteria can interact in multiple metabolic pathways in the environment. For example, denitrifying anaerobic methane oxidation (DAMO) can be found in nature ( 40 ), which presents an opportunity for the same process to occur in the setup, leading to the transformation of produced methane to carbon dioxide. DAMO is currently of interest as a competing pathway to reduce methane in MECs ( \n 24 ). In the future, further characterization of the metabolic processes in the system is needed to confirm the presence of this interaction. To validate the H 2 profile obtained in this study for the designed consortium, which resembles a regular growth curve of batch culture, we chose a common growth model to fit the data. Natural soil consortium possesses a different H 2 profile, likely due to the consumption of H 2 , which does not suit the chosen growth model well. In batch culture, bacterial growth rate followed these well-known stages: lag, acceleration, exponential, slowing down, stationary and death phases. Growth models may take a linear form like Monod or non-linear forms, such as Gompertz and logistic models ( 41 ). The Gompertz model was selected for this study, relying on simple information related to H 2 evolution in the system. Meanwhile, Monod was unsuitable since it requires additional information related to substrate consumption. Nonetheless, the H 2 production in MEC with designed consortium can be described well using Gompertz model as presented in \nFig. 4 . The modified Gompertz model was first formulated by Zwietering et al. ( 42 ) and adjusted to describe H 2 in newer studies ( 34 ). Fig. 4 Hydrogen profile model fit for the designed consortium based on Gompertz growth model; data presented after external voltage were supplied to the system. Dotted lines represent the model's confidence band (CI=95%) To consider the period of preparation before running the MEC, this period was excluded from the model (0–22 h). The correlation coefficient (R 2 ) of the model fit was 0.973. Using the model, several parameters were obtained: r max (mg/(L·h)), γ max (mg/L) and λ (h). To determine the rate of H 2 generation r max can be used. γ max corresponds to the maximum H 2 concentration, while λ is related to the lag phase after initialization of the system. For this study, model fit values for r max , γ max , and λ were 0.247 mg/(L·h), 8.605 mg/L and 20.98 h, respectively. A confidence band based on a confidence interval of 95% was used to graphically present the true location of the curve ( \nFig. 4 ). The fit of the model, assessed from its correlation coefficient, corresponds well to more than 20 values presented by Wang and Wan ( 34 ) in the range of 0.90–1.0 despite coming from a different setup than the batch fermentation method that is traditionally used to produce biohydrogen. This result suggests that H 2 production in MEC can be modelled like a traditional batch fermentation process, given that the H 2 profile matches those of ordinary batch growth/processes. Microbiome characterization Characterization of the soil community is available in Fig. 5 . Soil bacteria were detected based on metagenomics approach. This approach was chosen since it can provide a more expansive overview of genetic materials present in environmental samples, including from microbes that may be difficult to isolate and preserve in laboratory settings. The sensitivity of the method provides insight into the complexity of soil microorganisms. Fig. 5 Comparison of microorganisms in the microbiomes in: a) soil consortium, along with relative abundance up to genus level, and b) composition of the designed consortium, presented in a phylogenetic tree (constructed from aligned sequences using neighbor-joining method and 1000 bootstrap value) The MEC inoculated with soil contained mostly bacteria of the class Gammaproteobacteria (48.54%) that includes both Pseudomonas and Acinetobacter genera. Pseudomonas and Acinetobacter are classified as Gram-negative bacteria, much like the other genera dominating natural soil consortia. The medium used in this study contains high nitrate concentration, requiring bacteria to possess the necessary adaptive ability to survive. For designed consortium, the bacteria were preselected based on their capability to survive in a nitrate-rich environment as well as to metabolize nitrate using the denitrification pathway ( 25 ). The use of denitrifying bacteria as competitors to methanogenesis is based on the idea that metabolites released from denitrification may act as inhibitors to methanogens in soil samples ( 17 ). Differences in H 2 content in the headspace can be attributed to the microbiome inside the system ( Fig. 2 ). Soil microbiome was dominated by several genera, in descending order: Pseudomonas , Brucella , Achromobacter , Bordetella , Klebsiella , Lachnoclostridium 5, Stenotrophomonas , Clostridium sensu stricto 18, Lactobacillus and Acinetobacter . On the other hand, the designed consortium consists only of several species belonging to two genera, Pseudomonas and Acinetobacter , which are also present among the top ten genera in the soil microbiome. This is in agreement with the fact that the two genera were originally isolated from soil samples in Indonesia. Hence, we drastically reduced the complexity of the community by reducing a rich source of microbiome to only two genera. Future adjustments of the composition of the designed consortium could include well-documented electroactive bacteria like \nGeobacter to further facilitate electron transfer and H 2 production ( 43 , 44 ). Electrode characterization Exoelectrogenic microorganisms may transfer electrons to their environment by physical or chemical means. Physical mechanisms include the presence of structural nanowires, a term for electrically conductive pili ( 45 ). Chemically, electron transfer may occur through mediators secreted by the bacteria, i.e. pyocyanin by Pseudomonas aeruginosa ( 46 ). In MECs, electroactive species present in the medium can spontaneously attach themselves to form biofilms on the surface of the anodes ( 47 ). In this study, the physical states of anodes post-operation were analysed by SEM imaging. The anodes were treated prior to imaging following a modified approach from existing literature ( 31 ) to prevent structural degradation of biofilms, as presented in \nFig. 6 . Chemical methods utilized to fix the biomass prior to imaging involve repeated washing, which may degrade the extracellular matrix of biofilms present on the surface (circled in yellow). However, it is a more straightforward method than cryotreatment, and suitable for surface imaging ( 31 ). Fig. 6 Anode surface characterization using field emission scanning electron microscopy (FESEM) in: a and b) plain anode material, c and d) soil microbiome, and e and f) designed consortium. Red circle=intact structure; yellow circle=degraded structure The presence of biofilms on the anodes suggests that in both consortia physical transfer of electrons is possible. Additionally, given that Pseudomonas comprise the majority of the designed consortium and a major fraction of the soil consortium, it is also possible that pyocyanin-secreting species are present, hence allowing mediator-based electron transfer. In future studies, it would be interesting to analyse and compare the composition of microbiomes found on the surface of the anodes with microbes suspended freely in the medium. A similar imaging approach was used in literature ( 23 ) where they managed to show nanowires used for direct interspecies electron transfer. This approach is interesting to use for consortium with distinct morphological differences. In our case, the bacteria were morphologically similar."
} | 5,535 |
34684778 | PMC8538766 | pmc | 307 | {
"abstract": "Biohydrometallurgy recovers metals through microbially mediated processes and has been traditionally applied for the extraction of base metals from low-grade sulfidic ores. New investigations explore its potential for other types of critical resources, such as rare earth elements. In recent times, the interest in rare earth elements (REEs) is growing due to of their applications in novel technologies and green economy. The use of biohydrometallurgy for extracting resources from waste streams is also gaining attention to support innovative mining and promote a circular economy. The increase in wastes containing REEs turns them into a valuable alternative source. Most REE ores and industrial residues do not contain sulfides, and bioleaching processes use autotrophic or heterotrophic microorganisms to generate acids that dissolve the metals. This review gathers information towards the recycling of REE-bearing wastes (fluorescent lamp powder, spent cracking catalysts, e-wastes, etc.) using a more sustainable and environmentally friendly technology that reduces the impact on the environment.",
"conclusion": "4. Conclusions The demand for REEs is growing due to their unique properties, and REE extraction is becoming an important issue. Biohydrometallurgy could contribute by alleviating challenges related to the scarcity of economic ore resources and an almost monopolistic market. Novel studies on biohydrometallurgy offer the possibility of facilitating the extraction of REEs from waste, increasing the number of commodities of critical materials. Bioleaching for the recovery of rare earth metals from industrial wastes can be carried out by autotrophic and heterotrophic microorganisms. Several mechanisms are involved in the mobilization of REEs: organic acids, enzymes, bacterial attachment, siderophores, etc. Moreover, the development of biotechnological strategies for the treatment of solid wastes might contribute to a sustainable economy, maximizing the number of resources and minimizing the harmful impact on the environment. Bioleaching of REEs is in its infancy, but the development of a global market and environmental policies, as well as the appearance of new drivers such as synthetic biology and digital revolution, could influence the evolution of biohydrometallurgy.",
"introduction": "1. Introduction Rare earth elements (REEs) are strategic metals that facilitate the transition from the current economy based on fossil fuel to an efficient circular economy based on renewable energy. These metals are often needed in small quantities; however, they are essential for fabricating a large amount of technologically smart products for electronic, optical, and magnetic applications [ 1 ]. Most of rare earths are common elements in the Earth’s crust, and some of them are even more abundant than other metals, such as copper or lead. Despite their moderate abundance, rare earth elements are scarcely concentrated in mineral deposits, hampering their extractive metallurgy, which is complex and demands economic solutions. The production of REEs is growing exponentially since their discovery in the 18th century, with a significant rise over time from 1000 t in 1930 to 133,600 t in 2010 [ 2 ]. The increasing demand for REEs has led an escalating production as well. REE resources are mostly present in oxidic form, mainly as rare earth oxides, phosphates, carbonates, and silicates, due to their strong affinity for oxygen. Recent estimates indicate that 100 Mt of rare earth oxides are accessible in more than thirty countries all over the world. More than 200 REE-bearing mineral ores have been identified; nevertheless, only three of them are considered mineral ores for economic extraction: bastnasite ((Ce,La)(CO 3 )F), monazite ((Ce,La,Nd,Th)PO 4 ), and xenotime (YPO 4 ) [ 3 ]. Consequently, the primary sources are disseminated worldwide but are confined mainly in China, Australia, and USA. Moreover, REEs are also present in industrial wastes in great amounts, and industrial wastes have been considered as a potential resource for these critical metals [ 4 , 5 , 6 ]. Phosphogypsum is one of the most remarkable REE-bearing wastes generated during the wet phosphoric acid process of fertilizer production. Red mud residues from the digestion of bauxites in the Bayer process are also rich in valuable rare earth metals and their recovery can be economically valuable. Additionally, some post-consumer wastes can be recycled due to their significant quantities of REE, such as magnets (38%), lamp phosphors (32%), and metal alloys (13%). These materials comprise more than 80% of the REE market. Modern fluorescent lamps typically retain more than 20% ( w / w ) REE (Ce, Eu, La, Tb, and Y) [ 7 ]. After ore and/or industrial waste concentration processing, rare earth metals are dissolved selectively from raw materials using acid (H 2 SO 4 , HCl, HNO 3 , H 3 PO 4 ) or alkaline (Na 2 CO 3 , NaHCO 3 ) reagents under high temperatures, and this could pose an environmental problem. Actinides, such as uranium and thorium, with similar chemical properties to REEs, are often co-dissolved during hydrometallurgical processes, leading to a complex methodology [ 8 , 9 , 10 ]. Biohydrometallurgy has been successfully applied at the industrial level for the recovery of metals such as uranium, copper, and gold [ 11 , 12 ]. Biohydrometallurgical processes are usually applied to materials that would not be feasible to mine or treat using conventional chemical methods and would be considered residues. Consequently, these technologies could play a fundamental role in the treatment of REE-bearing wastes since they offer an alternative to physicochemical-based methods. Furthermore, the bioleaching of REEs would be involved in the development of more cost-effective, less energy demanding, and less polluting metal extraction processes than pyro- and hydrometallurgical processes. These biotechnological processes take place through interactions between microorganisms and metal-bearing ores that dissolve valuable metals. REE recovery from solid resources has been investigated with a wide range of microorganisms, both autotrophic and heterotrophic, and using both pure and mixed microbial cultures [ 13 , 14 , 15 ]. This review provides an insight into the global situation of REEs and the potential application of microorganisms in the extraction of REEs from industrial residues."
} | 1,598 |
35641608 | PMC9156742 | pmc | 308 | {
"abstract": "Resistive switching devices have been regarded as a promising candidate of multi-bit memristors for synaptic applications. The key functionality of the memristors is to realize multiple non-volatile conductance states with high precision. However, the variation of device conductance inevitably causes the state-overlap issue, limiting the number of available states. The insufficient number of states and the resultant inaccurate weight quantization are bottlenecks in developing practical memristors. Herein, we demonstrate a resistive switching device based on Pt/LaAlO 3 /SrTiO 3 (Pt/LAO/STO) heterostructures, which is suitable for multi-level memristive applications. By redistributing the surface oxygen vacancies, we precisely control the tunneling of two-dimensional electron gas (2DEG) through the ultrathin LAO barrier, achieving multiple and tunable conductance states (over 27) in a non-volatile way. To further improve the multi-level switching performance, we propose a variance-aware weight quantization (VAQ) method. Our simulation studies verify that the VAQ effectively reduces the state-overlap issue of the resistive switching device. We also find that the VAQ states can better represent the normal-like data distribution and, thus, significantly improve the computing accuracy of the device. Our results provide valuable insight into developing high-precision multi-bit memristors based on complex oxide heterostructures for neuromorphic applications.",
"conclusion": "Conclusion In conclusion, we demonstrate a novel memristive device based on the Pt/LAO/STO heterostructures. The voltage-driven migration of the surface oxygen vacancies enables controlling the tunneling of 2DEG in the Pt/LAO/STO junctions, achieving multiple non-volatile conductance states, over 27, with high reliability. The multi-level switching capability of the 2DEG memristor allowed us to explore an advanced weight quantization strategy, the VAQ. Our simulation studies verify that the VAQ can effectively reduce the state-overlap issue of the 2DEG memristors. In addition, since the VAQ states better represent the normal-like data distribution, the VAQ can significantly improve the computing accuracy of the 2DEG memristors. Although we demonstrated only the limited number of VAQ states for performance verification, we believe that the 2DEG memristors can implement a larger number of VAQ states and thereby achieve higher performance for practical applications. Lastly, we address that the VAQ states can be simply implemented by pre-defining the optimal writing voltages, without additional integration with other circuits. This implies that the VAQ method is readily applicable to other conventional resistive switching devices. Therefore, our results will provide a stepping stone for developing high-performance multi-bit memristors based on complex oxide heterostructures as well as other novel materials.",
"introduction": "Introduction Resistive switching devices are one of the leading candidates for memristors for synaptic applications 1 – 4 . In recent years, research effort has been focused on their capability of multi-level non-volatile switching, aiming for a high-level in-memory computation 5 – 11 . The ultimate goal of the memristors is an analog operation with a nearly infinite number of conductance states, imitating the analog operation of biological synapses. Considering the mechanism of resistive switching, in principle, most of conventional memristors should be able to stabilize a tremendous number of conductance states with a separation of a conductance quantum G 0 = 2 e 2 / h , where e and h represents the unit charge and Plank’s constant, respectively 12 – 14 . However, because the uncontrolled resistive switching mechanisms and the defect-induced charge trapping phenomena inevitably cause the random fluctuation of output current, the weight values of the memristors always show non-negligible variation 14 , 15 . Thus, the conductance states of the memristors are often quantized so that the output signal with variation can be rounded to the nearest state 16 . In this way, the state-overlap issue can be circumvented, but the number of available states (i.e., the number of the representable weight values) severely decreases. At this stage, increasing the on/off conductance ratio of the devices is required to maximize the number of conductance states. The minimization of current fluctuation is also essential to fully take advantage of the limited conductance range. Additionally, it is worth considering how to define the conductance states of the memristors. In most of the previous research, the multiple conductance states of memristors are defined by uniformly dividing the available conductance range. That is probably because the uniform states can be simply programmed by linearly-incremental forming voltages. However, based on the knowledge of neural network training, the actual weight values as well as intermediate data during training, such as activations, commonly have a normal-like distribution 17 , 18 . This implies that the small-weight-regime contributes to the result of the training more dominantly than the high-weight-regime. Therefore, the conventional uniform configuration of the conductance states may not be optimal for achieving high accuracy of the memristor-based computations. The non-uniform conductance states, configured considering the data characteristics, will better represent the distribution of the weights and the activations. Therefore, to improve the fundamental performance of memristors, we focus on two issues: (1) to build an advanced resistive switching device that can realize a larger number of conductance states and (2) to appropriately define its conductance states for higher quantization accuracy. Herein, we demonstrate a resistive switching device based on Pt/LaAlO 3 /SrTiO 3 (Pt/LAO/STO) heterostructure, which is suitable for multi-level memristive applications. By redistributing the surface oxygen vacancies, which create defect states in the band gap of LAO, we precisely control the tunneling of two-dimensional electron gas (2DEG) through the ultrathin LAO barrier in a non-volatile way. The 2DEG-based memristive device, namely the 2DEG memristor, achieves multiple conductance states (in excess of 27 states) with high reliability. To further improve the multi-level switching performance of the device, we propose the variance-aware weight quantization (VAQ) method. Our simulation studies verify that the VAQ can effectively reduce the state-overlap issue of the 2DEG memristors. Furthermore, we find that the VAQ states can better represent the normal-like data distribution and, hence, provides greater accuracy in image classification processes. These results will offer a significant step toward developing practical multi-bit memristors based on complex oxide heterostructures for neuromorphic applications.",
"discussion": "Results and discussion Resistive switching devices based on Pt/LAO/STO heterostructures The LAO/STO heterointerfaces have emerged as a new playground for exploring emergent electronic properties. The polarity discontinuity at the LaO + /TiO 2 0 heterointerface generates an electric field pointing away from the bottom interface to the top surface in the LAO/STO heterostructure 19 , 20 . The built-in field is necessarily compensated by the formation of 2DEG at the bottom LAO/STO interface 21 . The oxide interface with the 2DEG was found to be highly conducting. Moreover, the 2DEG has shown many interesting physical properties distinct from conventional semiconductor heterostructures 22 , 23 , and thus offers possibilities for device applications 24 – 26 . We design a resistive switching device based on the 2DEG in the LAO/STO heterostructures. Figure 1 a shows the vertical device configuration of a Pt/LAO/STO heterostructure. The highly-conducting 2DEG at the LAO/STO interface serves as a reliable bottom electrode in this device structure. When a positive bias voltage is applied to the top Pt electrode the 2DEG tunnels through the insulating LAO layer, resulting in a vertical current. We employ oxygen vacancy point defects to modulate the tunneling conductance. Since the oxygen vacancies in the LAO form intermediate energy levels within the bandgap 27 , they can serve as hopping sites for electrons. Thus, the distribution of the oxygen vacancies in LAO determines the effective tunneling probability of the 2DEG across the LAO barrier. Notably, an as-grown LAO thin film has most of its oxygen vacancies at the top surface due to the internal built-in field 20 , 28 . By redistributing the surface oxygen vacancies, we can form a conducting region, so-called the conducting filament, near the Pt/LAO interface and control the effective conductance across the Pt/LAO/STO junction (see more details in Fig. S1 , Supporting Information ). Note that, unlike the conventional metal/oxide/metal structures, the surface oxygen vacancies are formed as a counterpart of the buried 2DEG in the Pt/LAO/STO heterostructure 19 . The internal electric field originated from the polarity discontinuity at the LAO/STO interface determines how many electrons are to be accumulated at the bottom interface and similarly how many oxygen vacancies are to be formed at the top surface (Fig. S2 , Supporting Information ). Thus, by controlling the thickness of the LAO and thereby the internal electric field, we can reproducibly control the densities of both 2DEG and surface oxygen vacancies. This makes the 2DEG heterostructures unique and suitable for resistive switching device applications. Figure 1 Resistive switching device based on a Pt/LAO/STO heterostructure. ( a ) Schematic depicting the mechanism for the resistive switching in the oxide heterostructure. The spatial distribution of oxygen vacancies determines the tunneling probability of the 2DEG between the LAO/STO interface and the top Pt electrode. ( b ) Thickness-dependent evolution of the in situ RHEED intensity oscillation during the PLD deposition of LAO thin films. The insets show the RHEED patterns before and after the film growth. ( c ) AFM topography image measured on the surface of a thermally-treated STO (001) substrate. ( d ) AFM topography image measured on the surface of an as-grown LAO thin film. ( e ) HAADF-STEM image of the LAO/STO heterostructure. ( f ) Intensity of a line profile along (001) obtained from the STEM image. ( g ) XRD θ –2 θ scan of the LAO/STO heterostructure. To build the 2DEG memristor, we synthesized a LAO thin film on a TiO 2 -terminated (001) STO substrate by pulsed laser deposition (PLD) with in situ monitoring of reflection high-energy electron diffraction (RHEED). Figure 1 b shows the oscillation and the patterns of RHEED, indicating the layer-by-layer growth of the single-crystalline LAO thin film. After the growth of the film, the LAO/STO heterostructure was slowly cooled down to room temperature without oxygen gas injection or post-annealing, so that the oxygen vacancies are not fully removed. The Pt electrodes were subsequently fabricated on the top surface of the LAO thin film through a conventional lift-off process. A commercial Ag paste is used for the bottom contact. We confirmed that the Ag contact does not involve the resistive switching mechanism in the Pt/LAO/STO heterostructures (Sect. 3 , Supporting Information ). Further details of the sample fabrication are found in the “ Methods ” section. Figure 1 c,d show the atomic force microscopy (AFM) images measured on the surface of a thermally-treated STO (001) substrate and an as-grown LAO thin film, respectively. The surface of the as-grown LAO film is atomically flat and smooth, indicating the high quality of the film. The step-and-terrace structure on the LAO surface, which is almost identical to that on the STO substrate, implies that the layer-by-layer growth mode is well preserved throughout the deposition process. The high-angle annular dark field scanning transmission electron microscopy (HAADF-STEM) image taken from the same sample (Fig. 1 e) also indicates the high quality of the LAO/STO heterostructure. The line profile along (001) from the STEM image (Fig. 1 f) confirms that the thickness of the LAO thin film is exactly 12 unit-cells, as we designed, and the atomic intermixing at the interface is minimal. Figure 1 g shows the out-of-plane θ –2 θ X-ray diffraction (XRD) pattern around (002) STO peak. Only a single peak representing the (002) reflection of the LAO is found, ensuring the epitaxial nature of the single-crystalline LAO thin film. All of these structural analyses confirm the high crystallinity and the well-defined heterointerface of the LAO/STO heterostructure, regardless of its oxygen deficiency. We examined the electrical switching characteristics of the Pt/LAO/STO heterostructure. Figure 2 a shows the representative I–V curve of the device. As indicated by the pinched hysteresis loop, the device exhibits a bipolar resistance switching behavior. The I–V characteristics show that the positive voltage results in the off-switching (i.e., decreasing the tunneling conductance), while the negative voltage results in the on-switching. This switching polarity supports our hypothetical resistive switching mechanism, which is based on the surface oxygen vacancies. The initial conductance of pristine Pt/LAO/STO heterostructures is confirmed to be quite low (Fig. S5 , Supporting Information ). When a negative voltage is applied to the bottom 2DEG interface, the electropositive oxygen vacancies migrate from the top surface of the LAO toward the bottom 2DEG interface. Since the oxygen vacancies provide the hopping sites for electrons, the downward migration of the oxygen vacancies can be considered as the forming process of a conducting filament, that is the on-switching. On the other hand, when a positive voltage is applied, the oxygen vacancies move away from the interface, resulting in the off-switching. The asymmetry of the hysteresis is attributed to different band offsets at the top Pt/LAO and the bottom LAO/STO interfaces. Moreover, since the charge screening lengths are different at each interface, the effect of the applied electric field naturally depends on the direction of the field. Despite the asymmetry, we found that the conductivity of the 2DEG memristor can be switched by both positive and negative voltage pulses, revealing potential applications for analog memristors (Fig. S6 , Supporting Information ). Figure 2 Multi-level switching behavior of the 2DEG memristor. ( a ) I–V characteristics of the 2DEG memristor. The arrows represent the directions of the voltage sweep. Note that the initially applied negative voltage switches the device to the on-state (step (1)), while the subsequent positive voltage switches the device back to the off-state (step (2)). ( b ) Sequentially programmed conductance states, showing the representative 27 states. ( c ) The standard deviation (red squares) and the averaged conductance values (blue squares) at all individual conductance states. ( d ) Current power spectral density of output currents at the representative 3 different states and the off-state. The gradual change of I v at the high voltage regime in Fig. 2 a gives us a hint that we can effectively implement multiple conductance states. Figure 2 b shows the multiple conductance states of the same device, programmed by different writing voltage V write . We first fully turned off the device by applying the V write of + 9 V and then gradually turned the device on by applying incremental V write . For each conductance state, I v is measured at a reading voltage V read of + 0.5 V. We could implement 27 discrete conductance states with the on/off ratio of 2.84 × 10 3 % at the V write ranged from − 3.00 to − 4.25 V. In this conductance range, the switching characteristics are quite reliable and reproducible. The retention and the endurance properties are additionally described in Sect. 6 of Supporting Information . Notably, the overall change of the conductance value is desirably linear. In artificial neural networks, considering the required linear relationship between the input signals and the weight change for the network training, the linear dependence of conductance on V write enables more efficient and accurate training (Sect. 7 in Supporting Information ). Therefore, the high linearity of the conductance change, without additional doping 29 or multilayer stacking 30 , makes this 2DEG memristor a promising candidate for synaptic applications. In fact, a larger number of conductance states (up to 52) were achieved in the same device with a broader range of V write . However, when the V write increased over − 4.25 V, the conductance value was found to be not highly reproducible. Thus, for the following study, we consider only this linear and reliable conductance regime (the 27 conductance states). Besides the remarkable resistive switching characteristics, it should be noted that the variance of the output current at each conductance state is not ideally small. The relatively large variance of output signal has been a nuisance not only for our device but for most of the newly proposed resistive switching devices as well 10 , 31 , 32 . We quantify the variance of the output signals from our device by calculating the averaged conductance value and the standard deviation at all individual states (Fig. 2 c). The standard deviation clearly increases with the conductance of the device. The standard deviation values at the states 23–27 are comparable with the conductance difference between the neighboring states, indicating that the fluctuation during the read operation can occasionally degrade the precision of the 2DEG memristor. This large conductance variance is inevitable at the high-conductance regime (Sect. 8 , Supporting Information ). To further clarify the noise characteristics, we measured the current power spectral density (PSD) S I ( f ) of the 2DEG memristor. Figure 2 d shows the PSD spectra at 4 distinct conductance states. After setting each conductance state, the PSD spectrum of the I v was measured at + 1 V. All the PSD spectra show a typical 1/ f behavior, indicating the presence of charge traps with a wide range of time constants. It is also clearly seen that the fluctuation of I v becomes stronger as the device conductance increases. Therefore, to fully take advantage of the resistive switching properties, we confront this noise and the resultant state-overlap issue. Variance-aware quantization In principle, the state-overlap issue can be simply resolved if we selectively use only the conductance states whose distributions do not overlap with each other at all. Thus, in the case of previous memristors, the weight values are assigned to uniformly- and coarsely-defined conductance states. This conventional method limits the number of available states, degrading the fundamental performance of the memristors. Therefore, we propose the VAQ, a non-uniform quantization method designed to address the overlap issue. The non-uniform quantization has been used in different fields like image processing 33 , 34 , signal processing 35 , 36 , and deep learning 37 , 38 . Figure 3 a schematically depicts the state configuration for the conventional uniform quantization and the VAQ. The conductance states are non-uniformly defined for the VAQ, such that the conductance distributions hardly overlap across any two neighboring states. Figure 3 Variance-aware weight quantization for the 2DEG memristors. ( a ) Schematics showing the conventional uniform quantization and the variance-aware quantization methods. ( b ) Conductance histogram of uniformly-separated 12 conductance states. The black lines represent the normal distribution fitting curves. ( c ) Simulated heatmap of measurement error, calculated based on the uniformly-separated states. ( d ) Conductance histogram of the nonuniformly-separated 12 conductance states. The separation between each state is set small in the low current regime, while it is expanded as the current increases. ( e ) Simulated heatmap of measurement error, calculated based on the nonuniformly-separated states. Note that the error is significantly reduced by the variance-aware quantization method. As a proof of concept, we select 12 non-uniform VAQ states out of the total 27 states (Fig. S10 , Supporting Information ) and then find the same number of uniformly-separated states (Fig. 3 b,d) within the 27 states. We do not use the entire 27 conductance states but only a subset of them, so that we can define the same number of states in both the uniform and the non-uniform quantization schemes. Although these 12 states do not represent the full performance of our 2DEG memristors, they can directly reveal the impact of the VAQ as compared to the conventional uniform quantization within the given on/off ratio. To estimate the quantization errors depending on the state configuration, we measured 80 conductance values at each state and assigned them to the nearest conductance state. The heatmaps of the incorrectly quantized weight values when using the uniform states and the non-uniform states are given in Fig. 3 c,e, respectively. The horizontal and the vertical axis of the heatmaps represent the intended weight values and the actually-quantized weight values, respectively. These heatmaps clearly show that the VAQ with the non-uniform states reduces the quantization errors effectively. Because the VAQ conductance states are configured to have minimal overlaps between the data distributions, the measured weight values are less likely assigned to the incorrect neighboring states, achieving lower quantization errors. The difference between the two quantization schemes is particularly significant at states 9–11, where the uniformly-separated states have largely overlapped distributions. This result empirically proves that the conductance states defined in a variance-aware manner exploit the available conductance range without suffering from the distribution overlap issue. Evaluation of the VAQ: matrix–matrix multiplication To verify the advantage of the VAQ, we perform matrix–matrix multiplication using the conductance states of the 2DEG memristor defined by the uniform quantization and the VAQ. Figure 4 a schematically shows how we extract the experimental data from a convolutional neural network (CNN) 39 . To demonstrate the VAQ performance under realistic settings, we use the data collected from the actual ResNet20 training on CIFAR-10 dataset (Sect. S10 , Supporting Information ) 40 , 41 . The histogram of the collected input activations and the weight values are given by Fig. 4 b,c, respectively. They have a normal-like distribution as expected. We multiply these two matrices to obtain the output activation. To identify the impact of different quantization schemes, we calculate the output activations using the quantized input activation and weight values, and then compare them to the ground-truth. The output activation calculated using the non-quantized input activation and weight values is considered as the ground-truth. Figure 4 Variance-aware weight quantization for image classification problems. ( a ) An illustration of basic matrix operations for neural network training. The output activation matrix (10 × 32) is computed by multiplying the weight matrix (10 × 64) by the input activation matrix (64 × 32). We collected the data from a neural network designed for image classification tasks (ResNet20 output layer). ( b ) Histograms of the input data (i.e., activations from the previous layer) at the ResNet-20 output layer during training on CIFAR-10 dataset. ( c ) Histograms of the weight values at the same ResNet-20 output layer during training on CIFAR-10. ( d ) Histograms of the number of correct/wrong element-wise uniform quantization of the output matrix. We first get the ground-truth output matrix by quantizing the product of the two floating-point input matrices. We consider the quantization is correct if the quantized output element is the same as the corresponding ground-truth element. ( e ) Histograms of the number of correct/wrong element-wise variance-aware quantization of the output matrix. ( f ) Mean Absolute Error (MAE) of the output matrix. The error is calculated for 12 states separately. Figure 4 d,e show the number of the correctly- and wrongly-quantized elements of the output activations using the uniform quantization and the VAQ, respectively. Note that the uniformly-configured states do not appropriately represent the original data. Due to the strong overlap in the small-value regime, most of the quantized elements are assigned to the incorrect neighbor states. Even though this error rate is somewhat exaggerated due to the small number of available states (i.e., 12), it is clear that the quantization error is severe in this conventional quantization scheme. On the other hand, the VAQ method remarkably reduces the error. This comparison indicates that the VAQ can improve the quantization accuracy by resolving the state-overlap issue. The advantage of the VAQ can also be revealed by examining the mean absolute error (MAE). The MAE for each state, calculated using the different state configurations, are given in Fig. 4 f. While the uniform quantization yields high errors especially in the small-value regime, the VAQ method markedly reduces the errors. The VAQ slightly increases the errors in the large-value regime because it has fewer states for it. Nonetheless, the overall errors are sufficiently small to be ignored, as compared to that in the uniform quantization method. Evaluation of the VAQ: a heavily re-used convolution filter We point out another important advantage of the VAQ approach. Notably, the non-uniformly-configured conductance states for the VAQ can better represent the practical data distribution. In general, the weight values as well as the intermediate data, generated in neural network training, have normal-like distributions (Fig. S11 , Supporting Information ). The conventional uniformly-configured conductance states can well represent the large weight values, that take up only a small portion of the entire model parameters, while having significant quantization errors for the many small weight values (Fig. S12 , Supporting Information ). On the contrary, the non-uniform VAQ conductance states are suitable to represent such normal-like data distributions. Because a larger number of states is assigned to the small weight values than to the large weight values, such a non-uniform state configuration is advantageous for representing the normal-like data distribution. Therefore, the VAQ will allow the 2DEG memristors to achieve the higher classification and regression performances. To directly demonstrate this advantage, we perform convolution operations using different state configurations for the uniform quantization and the VAQ. Figure 5 a shows a training image, sample #1888 from Fashion-MNIST dataset 42 . The inset schematically shows the calculation of the output activation matrix from the input training image. The detailed simulation procedure is described in Sect. 13 of Supporting Information . The top panel of Fig. 5 b shows the numerical input values for each pixel of the image. Note that most of the normalized data are lied between 0.2 and 0.6. This implies that the realistic data are not uniformly distributed over the entire range. To simulate the convolution operation of our 2DEG memristors, we quantized the non-uniformly distributed input data to 12 states before applying the convolution filter. The middle panel of Fig. 5 b shows the quantized input data using the conventional uniform quantization method. It is clearly shown that the uniform quantization leads to loss of a significant amount of information in the input data. A majority of the input activations are quantized into smaller state values than their actual values due to the limited number of states for the small data values. On the other hand, the quantization error is effectively reduced by the VAQ. The bottom panel of Fig. 5 b shows the quantized input data using the VAQ method. Unlike the uniform quantization, the finer-grained states for the small data values represent the input data more accurately. Particularly, while the uniform quantization zeroes out many small input data, the VAQ keeps them in non-zero states mitigating the loss of information. Figure 5 Variance-aware weight quantization for the convolution operation. ( a ) The training sample image (image #1888) of Fashion-MNIST dataset. The image consists of 28 × 28 gray-colored pixels. We apply a 3 × 3 convolution filter (stride of 1 × 1) to the input image and compare the output. ( b ) Normalized original input pixel values (top), the input pixel values quantized using the uniform states (middle), and the input pixel values quantized using the variance-aware states (bottom). ( c ) The original output of the convolution operations without quantization. ( d ) The output obtained by applying the uniform quantization. ( e ) The output obtained by applying the VAQ. As a reference, the original output data of the convolution operation without any quantization process is given in Fig. 5 c. Since we apply a 3 × 3 convolution filter on the image with a stride of 1 × 1, the input image matrix of size 28 × 28 provides an output activation matrix of size 26 × 26. Figure 5 d,e present the quantized convolution outputs computed based on the uniform quantization and the VAQ method, respectively. The uniform quantization fails to precisely represent the small output values making the overall image speckled and noisy, while the VAQ provides a comparable output image to the original output. These simulation results clearly show that the VAQ can effectively reduce the quantization errors in the convolution operation and, hence, is advantageous in terms of accuracy. Note that we have not considered any specific activation function in this simulation. If an activation function was used, such as sigmoid or hyperbolic tangent, the output activations would be rescaled to a range of 0–1. That is, the data distribution can be shifted to the small value regime, and the uniform quantization likely loses more information compared to the VAQ. Likewise, if rectified linear unit (ReLU) was used 43 , the magnitude of the overall data flow would be significantly reduced since all the negative values are zeroed out. Regardless of the type of activation function, therefore, the VAQ is expected to result in smaller quantization errors than those of the uniform quantization."
} | 7,702 |
26905285 | PMC4764913 | pmc | 312 | {
"abstract": "Triboelectric nanogenerators (TENGs) have emerged as a potential solution for mechanical energy harvesting over conventional mechanisms such as piezoelectric and electromagnetic, due to easy fabrication, high efficiency and wider choice of materials. Traditional fabrication techniques used to realize TENGs involve plasma etching, soft lithography and nanoparticle deposition for higher performance. But lack of truly scalable fabrication processes still remains a critical challenge and bottleneck in the path of bringing TENGs to commercial production. In this paper, we demonstrate fabrication of large scale triboelectric nanogenerator (LS-TENG) using roll-to-roll ultraviolet embossing to pattern polyethylene terephthalate sheets. These LS-TENGs can be used to harvest energy from human motion and vehicle motion from embedded devices in floors and roads, respectively. LS-TENG generated a power density of 62.5 mW m −2 . Using roll-to-roll processing technique, we also demonstrate a large scale triboelectric pressure sensor array with pressure detection sensitivity of 1.33 V kPa −1 . The large scale pressure sensor array has applications in self-powered motion tracking, posture monitoring and electronic skin applications. This work demonstrates scalable fabrication of TENGs and self-powered pressure sensor arrays, which will lead to extremely low cost and bring them closer to commercial production."
} | 353 |
37311855 | PMC10264419 | pmc | 314 | {
"abstract": "Memristive devices have been explored as electronic synaptic devices to mimic biological synapses for developing hardware-based neuromorphic computing systems. However, typical oxide memristive devices suffered from abrupt switching between high and low resistance states, which limits access to achieve various conductance states for analog synaptic devices. Here, we proposed an oxide/suboxide hafnium oxide bilayer memristive device by altering oxygen stoichiometry to demonstrate analog filamentary switching behavior. The bilayer device with Ti/HfO 2 /HfO 2−x (oxygen-deficient)/Pt structure exhibited analog conductance states under a low voltage operation through controlling filament geometry as well as superior retention and endurance characteristics thanks to the robust nature of filament. A narrow cycle-to-cycle and device-to-device distribution were also demonstrated by the filament confinement in a limited region. The different concentrations of oxygen vacancies at each layer played a significant role in switching phenomena, as confirmed through X-ray photoelectron spectroscopy analysis. The analog weight update characteristics were found to strongly depend on the various conditions of voltage pulse parameters including its amplitude, width, and interval time. In particular, linear and symmetric weight updates for accurate learning and pattern recognition could be achieved by adopting incremental step pulse programming (ISPP) operation scheme which rendered a high-resolution dynamic range with linear and symmetry weight updates as a consequence of precisely controlled filament geometry. A two-layer perceptron neural network simulation with HfO 2 /HfO 2−x synapses provided an 80% recognition accuracy for handwritten digits. The development of oxide/suboxide hafnium oxide memristive devices has the capacity to drive forward the development of efficient neuromorphic computing systems.",
"conclusion": "Conclusion The highly linear, symmetric, and wide dynamic range of conductance change characteristics were demonstrated by optimizing update operation conditions to control the filament formation in memristive Ti/HfO 2 /oxygen-deficient-HfO 2−x /Pt devices, which contains oxide/suboxide bilayer switching oxide with different oxygen stoichiometry. The XPS analysis confirmed the different concentrations of oxygen vacancies in each layer which assists in controlled filament formation and improved device stability, particularly for gradual conductance change for analog synapse applications. The bilayer devices showed an excellent low voltage operation with analog conductance modulation. The synaptic characteristics of the device could be further improved using various pulse conditions. The conductance change for synaptic weight update was optimized to be highly gradual, linear, and symmetric with a wide dynamic range and good endurance by reducing pulse steps in ISPP operations for potentiation and depression behaviors. Furthermore, the neural network learning simulation results by using a three-layer MLP neural network on the MNIST handwritten digits dataset exhibited the learning accuracy of ~ 80% on average. These results demonstrate the high performance of synaptic device composed of the oxide/suboxide homojunction hafnium oxide layer by optimizing operation conditions for future hardware-based neuromorphic computing systems.",
"introduction": "Introduction In an ever-evolving world of Internet of Things (IoT) and big data era, the abundant acquisition of digital data, often highly unstructured, needs fast and efficient on-chip data processing for timely decision making. Despite the progress made in conventional Si-based CMOS technology, it still struggles with high computation and energy consumption for data-intensive tasks 1 , 2 . One of the major concerns is CMOS architecture design where memory and processing units are two separate entities. In addition, the sequential operation processing in such CMOS processors demands constant data transfer between memory and processing unit which caps a limit on computing speed, known as the von Neumann bottleneck 3 . Therefore, a fundamental change in computing hardware and architecture is essential to address the growing demand for data-centric computing operations such as artificial intelligence and machine learning. The human brain nervous system, consisting of synapses and neurons, is an excellent computing system that processes information in a parallel, event-driven, and distributed structure. Thus, an energy-efficient, high operation speed neuromorphic system is considered as an alternative to the conventional von Neumann system 4 . Neuromorphic computing systems have a highly interconnected network that has inherited salient features of the human brain for parallel and fault-tolerant information processing by mimicking the functionalities of synapses and neurons 5 , 6 . These synapses can perform both processing and storing information at the same location, thereby reducing energy consumption. Various two- and three-terminal nanoscale electronic memory devices have been extensively studied to emulate the synaptic plasticity behavior, the pattern of learning and forgetting characteristics in the brain 7 . Owing to the simple structure and feasibility of three-dimensional (3D) vertical integration, two-terminal memory devices such as ferroelectric random access memory (FeRAM), phase change random access memory (PCRAM), resistive random access memory (RRAM) and magnetic random access memory (MRAM) have been widely investigated for the application to artificial synapse 8 – 11 . Among these, RRAM devices, also known as memristive devices, work on the principle of voltage-induced resistance state modulation in a non-volatile manner, and are considered to be suitable candidates for artificial synaptic devices due to low power operation, high scalability, and good CMOS compatibility. A variety of switching materials, including oxides, nitrides, perovskites, and organic materials have been explored for developing RRAM devices 12 – 18 . Primarily, transition metal oxide-based memory devices such as TiO 2 , NiO x , HfO 2 , TaO x , ZnO have been widely studied 19 – 23 . One of the major challenges in these devices is non-uniformity in switching parameters which arises from the stochastic nature of filament growth dynamics in the switching layer 10 . The critical requirement of an ideal synaptic device is the modulation of a large number of analog conductance states with linear and symmetric change or weight updates to minimize error and achieve high learning accuracy in a neural network 1 . However, in reality, memristive devices show mostly digital switching with non-linear and asymmetric weight updates when they operate by the formation and rupture of conducting filament. Also, memristive devices operating with resistance change originated by redistributed defects suffer from non-uniformity and unreliable switching characteristics due to uncontrolled ion transport through the defects in the switching layer. This leads to temporal (cycle-to-cycle) and spatial (device-to-device) variations in these synaptic devices which adversely affect the computing performances 10 . In memristive devices using filament for conductance modulation, therefore, various approaches to address the randomness of filament formation have been adopted such as impurity doping 24 , metal nanoparticle incorporation 25 , and inserting a bilayer 26 to confine filaments. In recent times, HfO 2 -based memristors have been explored for synaptic application owing to their scalability and compatibility with the current CMOS technology 24 – 27 . The switching mechanism in HfO 2 memristors is broadly attributed to the formation and rupture of nanoscale conductive filaments consisting of oxygen vacancies through redox reaction, ion migration or nucleation process. Furthermore, bilayer structured HfO 2 memristors have been reported with improved electrical performances compared to single-layer oxide devices. For example, Ye et al. reported lower operation voltage and improved uniformity of HfO 2 /TiO 2 bilayer structured memristors compared to single-layer devices 28 . Kim et al. exhibited gradual conductance change in HfO x /AlO y devices and demonstrated biological synaptic characteristics even at elevated temperatures confirming the stability and sustainability of neuromorphic chips 29 . In addition, a stable two-level resistive switching feature was reported on Pt/HfO 2 /HfO 2−x /TiN devices which was explained based on the migration of oxygen ions by utilizing TiN electrode as an oxygen reservoir 30 . Besides filamentary switching memristors, interface-type switching devices have also been reported to show highly uniform switching characteristics due to homogeneous change of resistance state through the interface reaction. Thus, the conductance of the device could gradually change, avoiding any abrupt change and requirement of electroforming process, which is one of the important properties of artificial synapses 31 , 32 . Hansen et al. has reported area-dependent switching in oxide heterojunction memristors using Al 2 O 3 /Nb x O y double-barrier layer with uniform current distribution for high and low resistance states; yet the device retention time significantly affected, which is critical to guarantee pattern classification accuracy 33 . Kunwar et al. demonstrated versatile synaptic functions with an excellent uniformity through interface-controlled Au/Nb-doped SrTiO 3 Schottky structure with reliable retention 34 . A two-terminal charge trapped memristor based on Pt/Ta 2 O 5 /Nb 2 O 5-x /Al 2 O 3-y /Ti device has been reported exhibiting highly self-rectifying and nonlinear characteristics with a long retention time achieving a good pattern recognition challenge 35 . In the memristors with resistance change using the formation of filament consisting of oxygen vacancies, the sequential steps of creation of oxygen vacancies during an initial forming process, and its drift and diffusion in the oxide layer lead to switching in the device. Thus, considering the improved memory and synaptic properties in heterojunction bilayer HfO 2 devices with different oxide materials, it is of great interest to explore the switching property and synaptic behavior of the device in a homojunction bilayer HfO 2 structure. Compared with the heterojunction bilayer devices composed of different oxide layers, the homojunction bilayer devices employing the same constituent oxide layers are expected to have more reliable and controllable switching behaviors because the creation and redistribution of oxygen vacancies would occur within the homojunction layers. Typically, creation of oxygen vacancies during an initial forming process, and its drift and diffusion in the oxide layer leads to switching in the device. However, in the devices with a stoichiometric oxide layer, it is difficult to control the electroforming process due to a low vacancy concentration at the initial point and thus requires a large voltage for electrical breakdown which may lead to large conductance change. Therefore, to ease the need for creating oxygen vacancies, an oxide/suboxide structure can be considered for reliable switching performances 36 . Thus, in this work, we developed a homojunction oxide/suboxide HfO x bilayer memristive device, i.e., Ti/HfO 2 /HfO 2−x /Pt, to emulate various synaptic functions. The oxide/suboxide HfO x bilayers are deposited using atomic layer deposition (ALD) system under different deposition temperature to control the stoichiometry of oxide layers. The top HfO 2 and bottom HfO 2−x layers served as a stoichiometric resistive (oxygen vacancy-poor) layer and a relatively conductive (oxygen-deficient or oxygen vacancy-rich) layer, respectively. Therefore, the concentration gradient of oxygen vacancies plays a significant role in the device operation of oxide bilayer device. The fabricated synaptic device could operate at low voltage with stable memory characteristics by limiting the location of filament formation and rupture in the upper layer of HfO 2 switching matrix. In particular, the tunable pulse operations such as incremental step pulse programming (ISPP) were employed by varying pulse amplitude, step, and width, then enhanced synaptic characteristics such as linear and symmetric weight update along with basic synaptic learning functions of biological brains were demonstrated. These results may pave the way for designing an effective artificial synaptic device for memory and neuromorphic computing.",
"discussion": "Results and discussion Figure 1 a depicts the schematic diagram of the device with oxide and suboxide hafnium oxide bilayer with different stoichiometry, i.e., Ti/HfO 2 /HfO 2−x /Pt. From the cross-sectional high-resolution TEM micrographs in Fig. 1 b, the switching oxide bilayer is found to be the mixture of crystalline and amorphous phases with total thickness of ~ 6 nm. In order to make a clear distinction of HfO 2 /HfO 2−x bilayer structure, it was also deposited on the flat SiO 2 substrate as shown in Fig. S1 (supplementary information). The bilayer has a uniform and flat interface with top Ti layer. In addition, a crystalline grain is also observed, which is well coincident with our previous reports 37 . Notably, the interface between HfO 2 and HfO 2−x is not clearly observed in the micrograph unlike the typical heterojunction bilayer structures. It should be noted that the stoichiometry, and oxygen vacancy concentration of a stoichiometric HfO 2 and oxygen-deficient HfO 2−x layer, and the resistive switching characteristics with those layers are quite different, as will be further discussed later. Nevertheless, the microstructure in the bilayer turned out to be homogeneous, which is beneficial to induce the creation and redistribution of oxygen vacancies for the resistive switching. Figure 1 ( a ) Schematic diagram of Ti/HfO 2 /HfO 2−x /Pt memristive device and ( b ) cross-sectional high-resolution TEM micrograph of the bilayer device. The typical current–voltage ( I − V ) characteristic curves of HfO 2 /HfO 2−x bilayer memristive device were obtained using DC voltage sweep measurements, as shown in Fig. 2 a. The device was in high resistance in its pristine state and required a soft dielectric breakdown with a limit to compliance current, called electroforming process. The device underwent the initial electroforming process by applying a voltage sweep up to + 2.4 V with a compliance current of 1 mA as shown in the inset of Fig. 2 a. The voltage scanning rate was 50 mV/s. An abrupt change in current to maximum value was observed at ~ 2.3 V transitioning to a low resistance state (LRS). Again, the device was retraced back to its high resistance state (HRS) by applying a voltage of − 2.5 V. After the electroforming process, the device could be reversibly altered between LRS and HRS states consistently, exhibiting a typical bipolar resistive set and reset switching behaviors. Figure 2 b shows cycle-to-cycle variation of switching voltages for 300 sweeping cycles, whose I − V sweep curves are shown in Fig. S2 (supplementary information). The set voltage for switching from HRS to LRS is found to be within 0.87 to 1.5 V and reset voltage for switching from LRS to HRS is nearly − 0.62 to − 1.1 V. It indicates a narrow distribution of switching voltages, which is plotted as cumulative probability distribution of switching voltage in Fig. 2 c. The statistical analysis of switching voltages revealed that the coefficient of variation for set and reset voltage (standard deviation (σ) to mean values (µ)) is found to be 11.8% and 13.22%, respectively. Additionally, the device-to-device variation of switching voltages was also depicted in Fig. 2 d. The set and reset voltage values measured at 30 randomly selected devices remain within a range from 0.58 to 1.45 V and − 0.53 to − 1.5 V, respectively, for most of the devices. This verifies a low voltage operation, good reliability and reproducibility of switching performance of the devices. Figure 2 e shows the retention property in both HRS and LRS states. There is no significant degradation in the conductance of the states were observed for more than 10 4 s thanks to the robust nature of filamentary switching. The stability of the device was also verified by performing an endurance test under pulse measurement at a read voltage of 0.5 V with a 60 µs pulse indicating more than 4000 pulse cycles, as shown in Fig. S3 . Figure 2 ( a ) Typical I – V characteristic curves of the memristive device at the maximum voltage sweep range of ± 2 V with voltage scanning rate of 50 mV/s (inset: I − V curve for initial electroforming process), ( b ) cycle-to-cycle variation of set and reset voltage for 300 sweeps, ( c ) cumulative probability distribution of set and reset voltage indicating a narrow distribution of parameters, ( d ) device-to-device variation of switching voltages, and ( e ) retention characteristics of LRS and HRS states for more than 10 4 s. In Fig. 2 , the memristive device exhibits only two conductance states with a quick transition between HRS and LRS due to filamentary switching behavior when DC voltage sweep measurements were performed with the voltage scanning rate of 50 mV/s. However, a progressive conductance change with multilevel resistance state modulation needs to be achieved to demonstrate the synaptic characteristics. In view of this, it is essential that the present devices enable to fine tune the various conductance states by controlling switching operation parameters such as limiting the compliance current or controlling set and reset voltage conditions. In the present devices, the analog switching could be achieved with multiple conductance states by controlling the voltage sweep scanning rate. Note that the analog conductance modulation also requires a pre-forming process to initiate the switching. As shown in Fig. 3 a, the device can be tuned to achieve various analog conductance states by continuously increasing the amplitudes of set and reset voltages. To achieve the gradual switching, a lower voltage scanning rate of 5 mV/s was applied to the device without any restriction on compliance current. The device exhibits distinguishable increasing conductance states as the amplitude of voltage is increased from 0.6 to 0.85 V, indicating that the device could be able to mimic the potentiation process of biological synapses as will be confirmed also by voltage pulse operations. Similarly, a decrease in current is observed as the magnitude of negative bias voltage is increased which demonstrates the potential to replicate the depression of the biological synapse. Furthermore, it is clear that compared to positive bias sweep, a greater number of conductance states can be achieved by controlling reset stop voltages. This is an obvious effect of filamentary switching behavior that causes a prominent increase in current due to generation of higher concentration of oxygen vacancy and thus restricts further increasing in current which limits the number of states at the positive bias sweep. On the other hand, due to gradual nature of filament rupture during negative bias sweep, a large number of states can be observed. The retention performance of these different conductance states was also examined with a pulse width of 100 µs at a read voltage of 0.5 V, as shown in Fig. 3 b. The LRS state at + 1.2 V and four different HRS states at negative biases show no degradation of current, indicating reliability of the device. The superior retention properties come from the use of resistance change by robust filamentary switching compared to non-filamentary resistance change. However, it can be observed that the device shows more variation as we apply higher negative reset voltage to turn to high resistance. This is obvious because of low current measurement which is consequence of random motion of electrical charge carriers and diffusion of oxygen ions in the device. Additionally, the stochastic nature of the switching process may also introduce variability in the time required for the filaments to rupture, further complicating the measurement and control of the device's electrical properties. Nevertheless, the analog switching behavior with gradual conductance modulation of multilevel states and stable retentivity of each state makes it suitable for demonstrating reliable synaptic plasticity. The results of multilevel conductance states simply by reducing DC voltage scanning rate indicate that the conductance states could be modulated in an analog manner by controlling the voltage application conditions, especially in pulse applications as demonstrated in following results. Figure 3 ( a ) I − V characteristic curves of the memristive device as varying maximum sweep voltage with the reduced voltage scanning rate of 5 mV/s, ( b ) retention of multilevel conductance states, and multiple conductance change in the pulse operation for ( c ) set and ( d ) reset switching process. For artificial synapse device application with synaptic behaviors of the present memristive devices, it is essential to demonstrate the synaptic weight update characteristics with respect to pulse voltage rather than DC voltage sweep. We have examined the conductance change, corresponding to synaptic weight update, in response to the application of both positive and negative voltage pulses of 10 µs width and pulse interval time of 1 µs at a readout voltage of 0.2 V. As shown in Fig. 3 c and d, the conductance of each state increases (decreases) as the amplitude of positive (negative) voltage pulse increases along with the pulse number. There is a quite similar behavior between DC sweeping mode and pulse mode in positive bias condition, where the increase in conductance for different amplitude of voltage is not significant. It is assigned to filamentary nature of current conduction in the devices. Due to availability of a large amount of oxygen vacancies, the current increases rapidly for the very first pulse; thus, subsequent increase in current is limited. However, due to gradual rupture of filament in negative bias, current decreases sequentially as depending systematically on the voltage amplitude and number of pulses. Obviously, the conductance decreases more significantly at the first few pulses with the higher voltage amplitude and consequently saturates. On the other hand, at the lower amplitude, e.g., − 1.9 and − 2.0 V, the conductance is found to decrease gradually. These characteristics verify the possibility of fine conductance tuning to improve synaptic weight update by optimizing pulse application conditions as discussed in the Fig. 8 , 9 , 10 , 11 . The physical origin of switching characteristics in this bilayer device may come from different concentrations of oxygen vacancies in each layer, which contribute to the gradual change of filament geometry for analog synaptic weight update characteristics. To understand the composition and chemical bonding state of the bilayer structure, XPS core level and depth analyses were performed on the devices. All the peaks were fitted with Gaussian–Lorentzian (G-L) functions after a Shirley background subtraction. Figure 4 illustrates the core level spectra of Hf 4f peak in the HfO 2 /HfO 2−x bilayers on Pt bottom electrode with Ti adhesion layer at different etching times from top oxide layer to bottom Pt layer. From the XPS depth profile of the bilayer structure, the positions of 24 and 30 s are within the stoichiometric upper HfO 2 layer and the ones of 48 and 60 s are within an oxygen-deficient lower HfO 2−x layer as shown in Fig. 6 . The Hf 4f spectra are deconvoluted into two spin–orbit split peaks of binding energies 18.25 eV and 19.94 eV corresponding to Hf 4f 7/2 and Hf 4f 5/2 peaks, respectively, which are assigned to Hf–O bonding (Hf 4+ ) from stoichiometric HfO 2 layer 38 . Additionally, a doublet peak located at lower binding energies of 15.29 eV and 16.80 eV is attributed to low chemical valence states of Hf n+ -O (n < 4) 39 . These two weaker peaks indicate the suboxides of hafnium valence states in the HfO 2−x layer, which suggests the presence of abundant oxygen vacancies. The content of two strong Hf peaks are roughly calculated by evaluating the area under each peak and it comes out that the percentage of Hf 4+ and Hf n+ is 96.58% and 3.42% respectively at 24 s etching time. Also, it is clear that the percentage of suboxide Hf n+ increases with etching time from 3.42 to 22.7% hinting that the concentration of oxygen vacancies is higher at the lower HfO 2−x layer, due to its lower deposition temperature in ALD process. A complete Hf spectra deconvolution starting from the etching time of 0 to 90 s have been shown in Fig. S4 (supplementary information). Figure 4 XPS spectra of Hf 4f peak evolution with etching time of ( a ) 24, ( b ) 30, ( c ) 48, and ( d ) 60 s from the top surface in bilayer HfO 2 /HfO 2−x structure. Figure 5 represents the O 1s core level spectra of the bilayer at different etching times. The O 1s spectra are deconvoluted into two peaks of binding energies at 531.5 eV and 532.3 eV which are assigned to lattice oxygen and non-lattice oxygen with oxygen vacancies, respectively. A minimal shift of O 1s strong peak to lower binding energy with increasing depth from the top surface is observed which may be related to the change in film composition. Alternatively, the peak shift is thought to be modulated by the relative amount of metal ions bonded to oxygens 40 . A higher amount of non-lattice oxygen content is observed to exist in the lower suboxide layer (HfO 2−x ) than in the upper stoichiometric HfO 2 layer. The percentage of non-lattice oxygen ions increases to 21.41% at the lower oxygen-deficient HfO 2−x layer after 60 s of etching time, while it is 13.25% at the upper stoichiometric HfO 2 layer. A complete O 1s spectra deconvolution from 0 to 90 s of etching have also been presented in Fig. S5 (supplementary information). The O 1s profiles disclose consistently the graded oxygen concentration in the bilayer structure as same with Hf 4f profiles. Figure 5 XPS spectra of O 1s peak evolution with etching time of ( a ) 24, ( b ) 30, ( c ) 48, and ( d ) 60 s from the top surface in bilayer HfO 2 /HfO 2−x structure (inset table: percentage of area under each peak of two binding energies). Figure 6 shows the composition depth profile obtained from XPS analysis. The upper layer at shorter etching time has the stoichiometry close to HfO 2 , while the lower layer region at longer etching time has reduced oxygen content. In addition, the oxygen content keeps decreasing in the lower oxygen-deficient HfO 2−x layers. From the Hf 4f and O 1s core level spectra and the composition depth profile, it is conclusive that the bilayer structure is composed of the upper stoichiometric HfO 2 and the lower oxygen-deficient HfO 2−x layers. Figure 6 XPS depth profile of bilayer HfO 2 /HfO 2−x structure. Generally, the concentration gradient of oxygen vacancies plays a major role in the filament-based resistive switching device. The XPS results also confirmed the presence of abundant oxygen vacancies in as-prepared devices. However, a mere increase in oxygen vacancy concentration is not an effective way for stable switching performance, but its effective distribution is key to uniform switching characteristics. Because the present device has the homojunction bilayer structure of HfO 2 /HfO 2−x with different oxygen content in each layer, as confirmed from XPS analysis, its resistive switching by filament formation would be determined by the redistribution within the bilayer as schematically illustrated in Fig. 7 . The as-fabricated device in its pristine state has random oxygen vacancies more in the bottom HfO 2−x layer than top HfO 2 layer (Fig. 7 a). During the electroforming process upon applying a positive bias voltage to the top Ti electrode, a few amounts of oxygen vacancies are more generated and the pre-existing randomly distributed oxygen vacancies in the lower suboxide (HfO 2−x ) layer tend to align to participate in the conductive filament formation under the external electric field (Fig. 7 b). This results in the transition from an initial HRS with 10 9 Ω to LRS with 2000 Ω soon after filament formation indicating a possible migration of oxygen vacancies from the lower suboxide layer to upper oxide (HfO 2 ) layer. In contrast to the single layer device where oxygen vacancies are generated inside the single layer through the process of oxygen migration from oxide to top electrode (e.g., Ti), the present bilayer device enables the redistribution of oxygen vacancies through exchanging between the two layers. Therefore, it would have a more stable resistance change, particularly for gradual change in an analog manner. Generally, a wider filament will grow in the lower suboxide layer compared to the upper oxide layer due to the availability of higher oxygen vacancies in the lower oxygen-deficient suboxide layer, and a confined filament will grow inside the upper oxide layer. 41 This minimization of filament growth location in the upper oxide layer brings more stable and reproducible switching performances in the bilayer device compared to the single layer devices. A negative bias takes the device to reset process by rupturing filament partially at the upper oxide layer while the major part of filament remains intact in the lower suboxide layer thanks to its higher oxygen vacancy concentration (Fig. 7 c). During reset process, the conductive filament starts rupturing from the tip of the top electrode as a result of the recombination between oxygen ions and oxygen vacancies, which leaves an insulating gap. Subsequently, when a positive bias is applied, the device sets to LRS again by reconnecting the filament. However, the set voltage required to achieve LRS is lower than initial forming voltage due to reconnection of only a fraction of pre-existing filament (Fig. 7 d). Therefore, the utilization of an oxide/suboxide bilayer structure leads to confinement of filament location which narrows the distribution of set and reset voltages (Fig. 2 b and d). The use of the redistribution of oxygen vacancies within the bilayer structure provides the potential of precise change of filament geometry, which is crucial for realizing analog resistance change to mimic analog synaptic weight update as shown in following results. Figure 7 Schematic illustration of resistive switching mechanism in Ti/HfO 2 /HfO 2−x /Pt memristive devices. In the next section, we have demonstrated the various synaptic functions in terms of potentiation and depression in Ti/HfO 2 /HfO 2−x /Pt memristive device at different voltage pulse application conditions. As the memristive conductance closely follows with biological synaptic weight, the change of conductance states corresponds to the synaptic weight update for the learning process with long term potentiation (LTP) and long-term depression (LTD) of the biological synapse. As shown in Fig. 8 , the synaptic weight of the device depends on various stimulation parameters of input pulse including its amplitude, width and interval time between pulses. To emulate the biological synaptic characteristics, several other works have also considered various programming schemes such as use of incremental pulse amplitude and pulse width 31 , 42 – 44 , addition of a heating spike before set and reset pulses 45 , current pulse mode 46 and device processing approaches such as N 2 annealing treatment 47 . Figure 8 a shows a monotonously increasing conductance (potentiation) with the number of pulses by applying identical square pulse waveform of amplitude of 2 V with a width of 10 µs and pulse interval of 1 µs. A constant readout voltage of 0.5 V is applied to read the conductance after each potentiation pulse. Applying a higher amplitude pulse induces the migration of more oxygen vacancies leading to an increased conductance. Similarly, a gradual decrease in conductance (depression) is observed with opposite pulse spikes of same conditions as shown in Fig. 8 b. Furthermore, it is observed that the conductance increases or decreases steadily with increasing the pulse amplitude, which is also consistent with the results of DC sweep measurement by varying amplitude of sweep voltages. The potentiation and depression behavior can be explained based on the continuous modulation of conductance in a confined region of filament with its dimensional change by a small amount of oxygen vacancies. The increase of pulse amplitude controls the dimension of filament which conceives to potentiation and depression behavior with variable update ranges. Figure 8 Conductance modulation characteristics under various pulse schemes for potentiation and depression as varying ( a ) and ( b ) pulse amplitude, ( c ) and ( d ) pulse width, and ( e ) and ( f ) pulse interval time. Figure 8 c and d show pulse width-dependent synaptic weight modulation at a fixed pulse amplitude of ± 2 V, pulse interval of 1 µs and readout at 0.5 V. The conductance of the device increases with the pulse number as the pulse width is increased from 20 to 80 µs. It is also noticed that a larger pulse width requires a smaller number of pulses to induce conductance increment and vice-versa. This could be due to that applying longer pulse provides oxygen vacancies with more sufficient time to migrate for the conductance change. In particular for the depression behavior, Fig. 8 d shows the gradual decrease of conductance for a shorter pulse width while a sharp decrease for longer pulse width. For example, 12 pulses are required to reset the conductance for a pulse width of 20 µs whereas it needs only 4 pulses for pulse width of 80 µs. Also, the conductance decrease at pulse width of 20 µs is quite abrupt, compared with the gradual decrease at 10 µs pulse width shown in Fig. 8 b, disclosing the strong dependence also on the pulse width. The dependence of synaptic weight update on pulse interval time was also examined at the condition of a constant pulse amplitude of 1 V and − 1.1 V with a pulse width of 20 µs. As shown in Fig. 8 e and f, the conductance change is found to decrease for both potentiation and depression behaviors as the pulse interval time is increased from 1 to 100 µs. It is noticeable that the conductance changes are rather significant at the first few pulses and then saturate with the increase of pulse number for the pulse interval time up to 50 µs. However, the conductance change is fairly gradual with almost identical increments with respect to the pulse number for a larger pulse interval time of 100 µs. The results showing the dependence of conductance change on pulse interval time indicate that it takes time for the filament geometry to be stabilized. Thus, the longer interval time results in more gradual and stabilized conductance changes. It is rational that the conductance change depends on pulse interval time between adjacent pulses because it involves the generation, recombination, and migration of oxygen vacancies to adjust the filament geometry. It implies that some unstable species undergo time-dependent transient dynamics in either generation, recombination, or migration. For example, unstable oxygen vacancies would disappear over time by recombining with oxygen ions, thus they do not contribute to the conductance change in the condition of longer pulse interval time. Therefore, the oxygen vacancies generation, recombination, and migration driven by electric field and the energy required for conductance change in the memristive device are well affected by the pulse conditions such as pulse amplitude, pulse width and pulse interval time. In the artificial synapse device application, the linear and symmetric synaptic weight update characteristics as well as its excellent endurance properties are crucial for energy-efficient and accurate learning operations 48 , 49 . In general, the dynamics of conductance change becomes non-linear and asymmetric in nature due to abrupt set and reset transition in filamentary synaptic devices. For instance, the repeated measurements of potentiation and depression behaviors read at 0.5 V using identical pulse amplitude of ± 1.5 V with a width of 10 µs and interval of 100 µs disclose an excellent endurance during repeated cycles of more than 3000 pulses without significant degradation as shown in Fig. S6 . To calculate the linearity of weight update during the potentiation and depression processes, the conductance change of potentiation and depression with the number of pulses can be modeled by the following equations in MATLAB R2023a software 50 , 1 \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$G_{{{\\text{potentiation}}}} = {\\text{B}}\\left\\{ {1 - exp\\left( { - \\frac{P}{A}} \\right)} \\right\\} + G_{{{\\text{min}}}}$$\\end{document} G potentiation = B 1 - e x p - P A + G min 2 \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$G_{{{\\text{depression}}}} = - {\\text{B}}\\left\\{ {1 - exp\\left( { - \\frac{P}{A}} \\right)} \\right\\} + G_{{{\\text{max}}}}$$\\end{document} G depression = - B 1 - e x p - P A + G max 3 \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${\\text{B}} = { }\\frac{{G_{{{\\text{max}}}} - G_{{{\\text{min}}}} }}{{1 - exp\\left( { - \\frac{{P_{max} }}{A}} \\right)}}$$\\end{document} B = G max - G min 1 - e x p - P max A where \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$G_{{{\\text{potentiation}}}}$$\\end{document} G potentiation and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$G_{{{\\text{depression}}}}$$\\end{document} G depression are the conductance for potentiation and depression, respectively. \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$G_{{{\\text{max}}}} , G_{{{\\text{min}}}}$$\\end{document} G max , G min and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$P_{max}$$\\end{document} P max are directly extracted from the experimental data, which represents the maximum conductance, minimum conductance and the maximum pulse number required to switch the device between the minimum and maximum conductance states. A is the parameter that controls the nonlinear behavior of weight update. B is a function of A that fits the functions within the range of \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$G_{{{\\text{max}}}} , G_{{{\\text{min}}}}$$\\end{document} G max , G min and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$P_{max}$$\\end{document} P max . The non-linearity values for potentiation and depression are calculated to be 4.88 and 3.7 respectively for identical pulse scheme as shown in Fig. S6 (a). The synaptic weight update is fairly non-linear and asymmetric. The potentiation current increases abruptly and then saturate more quickly than depression current following an obvious nature of filamentary switching as shown in Fig. S6 (b). To improve the linearity and symmetry of weight update for potentiation and depression, various pulse operation conditions were employed. Figure 9 shows the potentiation and depression behavior as varying the pulse width conditions. The device shows an improved linearity and symmetry of weight update when operates with variable pulse width as compared to previous case of constant pulse width. In this measurement, the potentiation and depression pulse amplitude are kept constant at + 1.2 and − 1.3 V, respectively, with the constant pulse interval of 10 µs. It is also observed that the potentiation conductance increases and depression conductance decreases systematically with the increase of pulse width, which follows with behavior shown in Fig. 8 c and d. The non-linearity values for potentiation and depression are found to be 3.7 and 2.03 respectively as shown in Fig. 9 a. Although the linearity and symmetry seem to be improved, increasing pulse width could not sufficiently increase the conductance dynamic range as shown in Fig. 9 b, which is also an essential condition for improving the pattern recognition accuracy in neuromorphic computing with high resolution. Therefore, to address the dynamic range with high resolution, linearity, and symmetry of synaptic weight update, an incremental step pulse programming (ISPP) operation scheme was adopted, which has been applied to programming in non-volatile flash memory 51 . Non-identical pulse operation programming achieves faster switching times, improved endurance, and enhanced scalability in memory devices. However, such a complex programming scheme requires non-trivial design efforts from the peripheral circuit's perspective because it requires precise calibration of the pulse amplitudes and pulse widths for each memory cell based on its current conductance state before applying the programming pulses. The calibration process requires additional circuitry which can increase the complexity and cost of the memory device. Figure 10 shows the potentiation and depression behaviors as applying non-identical pulses whose amplitude increases after every 10 pulses in steps of 100 mV from 0.9 to 1.6 V with a constant pulse width of 20 µs. The synaptic weight in the potentiation and depression can be modulated by up to 4 times of initial resistance value. Figure 10 a shows non-linearity curve with normalized conductance modulation for non-identical pulse operation. Although the non-identical ISPP operation scheme improved non-linearity of potentiation to 0.5, it fails to improve depression non-linearity (4.88). However, an abrupt change of conductance during update is still observed due to filamentary nature of conductance change, as shown in Fig. 10 b. An enlarged view of the abrupt transition was shown in Fig. 10 c for better representation. This abrupt change could be eliminated further by properly controlling step voltage and much improved gradual weight update could be achieved, as shown in Fig. 11 . An optimized pulse schemes using, non-identical pulses with a pulse width of 20 µs and pulse interval of 50 µs in reduced steps of 50 mV from 0.6 to 1 V for potentiation and step voltage of 20 mV from − 1.02 to − 1.18 V for depression was adopted. This resulted in a significant improvement in non-linearity parameters for both potentiation and depression to 2.4 and 1.55 respectively, as shown in Fig. 11 a. As illustrated in Fig. 11 b, an almost complete linear and symmetric weight update could be achieved through precisely controlled change of filament geometry. Figure 9 ( a ) Non-linearity of synaptic weight update from Ti/HfO 2 /HfO 2−x /Pt bilayer memristor under different pulse widths. ( b ) Repetitive conductance modulation by varying pulse width from 20 to 60 µs with a pulse amplitude of + 1.2 and − 1.3 V for potentiation and depression, respectively, with a pulse interval time of 10 µs. Figure 10 ( a ) Non-linearity of synaptic weight update from Ti/HfO 2 /HfO 2−x /Pt bilayer memristor by using non-identical pulse spikes under incremental step pulse programming (ISPP) operation. ( b ) Conductance modulation by using non-identical pulse spikes ISPP operation scheme with a constant step voltage of 100 mV. ( c ) An enlarged view of ( b ). Figure 11 ( a ) Improved non-linearity of synaptic weight update of bilayer memristor device by using ISPP mode with optimized pulse condition. ( b ) Conductance modulation by using non-identical pulse spikes under ISPP operation scheme with optimized step voltage. Some of the recent published state of the art work based on HfO 2 bilayer synaptic devices is compared with the present work and summarized in Table 1 . Table 1 Performance comparison of switching parameters of various HfO x bilayer synaptic devices. Device Set/reset voltage Endurance Retention (sec) On/Off ratio Non-linearity References Au/HfO x /HfO 2 /Pt + 0.2 V/− 0.18 V 50 10 4 > 10 NA Ref. 52 Pt/HfO 2 /HfO x /TiN − 1.6 V/+ 1.1 V 10 4 10 5 10 3 NA Ref. 53 Ti/HfO x /AlO y /TiN + 1 V/− 1.4 V 10 4 ~ 10 1.06/5.43 Ref. 54 ITO/ZnO/HfO x /ITO + 1.8 V/− 2 V 10 3 10 4 ~ 10 2 3.19/2.4 Ref. 55 TiN/Ti/HfO x /TaO y /HfO x /Au − 5 V/+ 5 V 10 6 10 5 50 5.3/11.9 Ref. 56 Ti/HfO 2 /HfO 2−x /Pt + 1.1 V/− 0.84 V (DC measurement) 300 > 10 4 > 10 2 2.4/1.55 This work + 1 V/− 1.2 V (pulse measurement) 4 × 10 3 > 10 4 ~ 20 In order to evaluate the potential applicability and performance of the present bilayer synaptic device for neuromorphic hardware implementation, the pattern recognition accuracy was examined by the simulation using an artificial neural network system, NeuroSim + , with the normalized conductance values extracted from the potentiation/depression cycle of bilayer synaptic device. More specifically, a three-layer multilayer perceptron (MLP) neural network was used which consists of 400 neurons as input layers, 100 hidden neurons and 10 output neurons corresponding to 10 classes of digits (0–9) as shown in Fig. 12 a. A cropped 20 × 20 pixels image from the MNIST data set was used here as an input for the image recognition task and each pixel from the image corresponds to one neuron of the input layer. Figure 12 b shows the learning accuracy of the device acquired from the MLP as a function of the number of training epochs. For each epoch, 60,000 training data sets and 10,000 test data sets were used to evaluate the recognition accuracy. The MLP neural network simulator uses a stochastic gradient descent method to update weights which calculates the error from current weight values and then propagates in backward to adjust the weight so as to minimize the prediction error 57 . We also considered various non-ideal parameters including nonlinearity, conductance level and variation during training. The baseline reference was obtained by changing the non-linearity value of both potentiation and depression to 1, while maintaining all other inputs such as conductance values, pulse number and pulse voltage conditions. The simulation results show a learning accuracy of ~ 80% which could be further enhanced by improving the switching dynamic range and/or sufficient accessible conductance states (precision) of weight values 58 . Although voltage controlled analog switching of the present device could help to access various resistance states, yet it is widely acknowledged that an ON/OFF ratio of 10 or above is required for better differentiation between the potentiation and depression states, which is important for accurate synaptic weight updates. In an array level integration with memristors, a lower ON/OFF ratio suffers from scalability issues as the minimum read margin cannot be satisfied in a large array. Therefore, it needs to be further pursued to improve the ratio with material optimization, or process parameter optimization such as, deposition parameters, annealing conditions. Figure 12 ( a ) Schematic of three-layer perceptron based neural network used for MNIST pattern recognition and ( b ) learning accuracy rate with training epochs using stochastic gradient descent (SGD) algorithm. Energy consumption of electronic synaptic devices in each operation is one of key parameters to have a promising application in imitating synapses. The reported energy consumption varies between ranges of 0.06 fJ to 800 pJ per synaptic event 53 , 59 – 62 . By considering the pulse width time, response current, and pulse amplitude, the energy consumption of our device was calculated to be 1.37 × 10 –10 J (137 pJ) per synaptic event for a 20 µs pulse at a reading voltage of 0.5 V. Nevertheless, the amount energy required per synaptic operation is also affected by the pulse amplitude and pulse width, which could be optimized furthermore for achieving an ultra-low switching operation."
} | 12,450 |
31961452 | PMC7317464 | pmc | 315 | {
"abstract": "Abstract Increasing ocean temperatures have widespread consequences for coral reefs, one of which is coral bleaching. We analyzed a global network of associations between coral species and Symbiodiniaceae for resistance to temperature stress and robustness to perturbations. Null networks were created by changing either the physiological parameters of the nodes or the structures of the networks. We developed a bleaching model in which each link, association, is given a weight based on temperature thresholds for specific host–symbiont pairs and links are removed as temperature increases. Resistance to temperature stress was determined from the response of the networks to the bleaching model. Ecological robustness, defined by how much perturbation is needed to decrease the number of nodes by 50%, was determined for multiple removal models that considered traits of the hosts, symbionts, and their associations. Network resistance to bleaching and robustness to perturbations differed from the null networks and varied across spatial scales, supporting that thermal tolerances, local association patterns, and environment play an important role in network persistence. Networks were more robust to attacks on associations than to attacks on species. Although the global network was fairly robust to random link removals, when links are removed according to the bleaching model, robustness decreases by about 20%. Specific environmental attacks, in the form of increasing temperatures, destabilize the global network of coral species and Symbiodiniaceae. On a global scale, the network was more robust to removals of links with susceptible Symbiodiniaceae than it was to removals of links with susceptible hosts. Thus, the symbionts convey more stability to the symbiosis than the hosts when the system is under an environmental attack. However, our results also provide evidence that the environment of the networks affects robustness to link perturbations. Our work shows that ecological resistance and robustness can be assessed through network analysis that considers specific biological traits and functional weaknesses. The global network of associations between corals and Symbiodiniaceae and its distribution of thermal tolerances are non‐random, and the evolution of this architecture has led to higher sensitivity to environmental perturbations.",
"introduction": "Introduction The resistance of coral reefs to changing environmental conditions is a central theme of coral ecology as reefs continue to decline worldwide. Coral bleaching, the breakdown of the association between the coral host and its endosymbiotic algae, is a considerable force behind the deterioration of coral reefs (Hughes et al. 2018 ). Bleaching responses vary across species, individuals, and stress events (Loya et al. 2001 , Baker 2003 ). Environmental factors drive bleaching patterns on a large scale (Nakamura and Van Woesik 2001 ), but the variation in bleaching response is attributed to the complex associations among coral hosts and their symbiotic algae, dinoflagellates in the family Symbiodiniaceae (LaJeunesse et al. 2008 , 2018 ). Corals hosting specific symbionts (Rowan 1998 , LaJeunesse 2001 , Glynn et al. 2001 , Toller et al. 2001 , van Oppen et al. 2001 ), multiple symbiont types (Loh et al. 2001 , Baker 2003 ), and diverse background symbiont populations (LaJeunesse 2002 , Quigley et al. 2014 , Ziegler et al. 2017 ) may be more resistant to thermal stress. These complex symbiotic associations can be analyzed as a network of coral species interacting with members of the family Symbiodiniaceae. Network science conceptualizes complex systems as components, represented by nodes, connected to other components, by their interactions. Complex systems from social systems to ecological systems have been found to have heterogeneous or close to scale‐free structure, where the distribution of node connections, the degree distribution, follows a power law in which some nodes have a lot of links (the hubs) and most nodes have only a few links (Holme 2019 ). The structure and topology of complex systems determines their ability to withstand perturbation (Albert et al. 2000 , Cohen et al. 2001 , Allesina and Pascual 2007 ). Unlike random networks, scale‐free and heterogeneous networks are robust to random failures (node removals) and highly susceptible to targeted “attacks” on hubs (Albert et al. 2000 ). Susceptibility to attacks due to network structure has been found in food webs (Solé and Montoya 2001 , Dunne et al. 2002 ) and mutualistic networks of plants and their pollinators (Bascompte and Jordano 2007 ). Typically, attacks on ecological networks are modeled as species extinctions, i.e., node removals. While the adaptive rewiring of links has been found to aggravate the effects of species loss (Gilljam et al. 2015 ), few studies have modeled attacks as interaction extinctions, or link removals (Valiente‐Banuet et al. 2015 ). With the advent of ecoinformatics databases (e.g., GeoSymbio; Franklin et al. 2012 ) and finer resolution sequencing technologies, network analysis has been used to better understand associations among corals and Symbiodiniaceae. Fabina et al. ( 2012 ) analyzed a selected, well‐sampled portion of the GeoSymbio database and found the network to be sparse and significantly nested, key attributes thought to support network stability (Saavedra et al. 2011 , Rohr et al. 2014 ). Fabina et al. ( 2013 ) modeled species loss effects on the robustness of the coral–symbiont network in Moorea, French Polynesia, and found that when Symbiodiniaceae nodes (designated ITS2 types) were removed based on clade‐level (currently viewed as genus level; LaJeunesse et al. 2018 ) thermal tolerance, the network was less robust than if Symbiodiniaceae were removed based on nutritional benefit. Network robustness also decreased when generalist (high‐degree nodes) and dominant symbiont types were removed (Fabina et al. 2013 ). Similarly, Ziegler et al. ( 2017 ) found that an abundance of rare background symbionts increased the robustness of a network of coral hosts and symbionts in the Red Sea, Sea of Oman, and the Persian/Arabian Gulf. These initial analyses provide insight into the coral–symbiont network’s robustness to perturbations. However, two key aspects are yet to be explored. First, there is evidence for within‐genus differences in symbiont thermal tolerance (Ladner et al. 2012 , Swain et al. 2017 ). Second, the removal conditions set by Fabina et al. ( 2013 ) ignore the contribution of the host to the coral’s ability to withstand thermal stress, when there is significant evidence supporting a combined host and symbiont, a holobiont, centric view of coral thermal tolerance (cf. Berkelmans and van Oppen 2006 , Baird et al. 2008 , Wooldridge 2014 ). Thus, a measure of combined resistance to temperature stress that incorporates more physiological and environmental data is needed for a better understanding of the coral–symbiont network. Although meta‐analyses of organismal physiology are common (Swain et al. 2016 , 2017 ), network analyses of ecological stability have yet to regularly incorporate these data when exploring impacts of climate change on ecosystems. Coral bleaching can be considered as a perturbation on the network, so that measures of network robustness (Dunne et al. 2002 , Yang et al. 2017 ) can be used to predict the association's stability under environmental stress. Our study further distinguishes resistance from robustness as a system‐specific measure. Resistance and robustness metrics are defined below in more detail; qualitatively, resistance measures performance of the network when under an environmental attack, while robustness measures network fragility when nodes or links are removed. As the likelihood of annual coral bleaching events continues to increase in the coming decades (Hughes et al. 2018 ), a global approach to understanding the complex network and fragile relationship of coral species and their algal symbionts is needed from the cellular to the ecosystem level (Suggett and Smith 2019 ). We developed a network model for coral bleaching that uses exposure to elevated sea surface temperatures (NOAA Coral Reef Watch) and assigns thermal tolerances to both hosts (Swain et al. 2016 ) and symbionts (Swain et al. 2017 ) to determine the point of ecophysiological breakdown of the symbiosis under temperature stress on a global scale. The network of coral species and their associated Symbiodiniaceae was created using the GeoSymbio database (Franklin et al. 2012 ). We define the network’s resistance as the rise in temperature (°C) that increases the percentage of hosts bleached from 10% to 90% normalized by the maximum possible temperature excursion. We use the accepted metric of ecological robustness, i.e., the percentage of nodes (or in our case, also links) that must be removed to decrease the nodes remaining to 50% (Dunne et al. 2002 ) to quantify the effect of various network perturbations. Our model of coral bleaching, metric for network resistance to temperature stress, and ecological robustness metric allow us to answer the following questions: (1) How does network structure and the distribution of thermal tolerances affect resistance to temperature stress? (2) How does spatial scale, and thus local association patterns and environment impact network resistance to temperature stress and robustness to perturbations? (3) Is the network more robust to interaction or species removals? (4) Does environment, hosts, or symbionts convey more robustness to the network? We demonstrate that a network approach allows us to determine patterns of resistance to temperature stress and ecosystem robustness on global, ocean‐basin, and subregional scales for coral species and their symbiotic algae.",
"discussion": "Discussion We developed a novel model to simulate bleaching on the global network of coral and Symbiodiniaceae associations using specific ecophysiological attributes. Our results indicate that the global network of coral species and Symbiodiniaceae associations is susceptible to perturbations that specifically take into account physiological and environmental data and that this susceptibility is, in part, due to the structure of the associations. As ecosystems continue to be threatened by climate change, modeling environmental stress on ecological networks that incorporate ecophysiological data will prove to be a powerful tool for understanding their stability. The networks studied here are of course limited by the data used to create them and do not account for possible temporal shifts in associations (Glynn et al. 2001 , Jones et al. 2008 , Sampayo et al. 2016 ) or greater symbiont diversity and host‐specificity that is masked by the use of ITS2 phylotypes (Thornhill et al. 2014 , Hume et al. 2019 ). In our bleaching model, temperature acts as a press perturbation that puts stress on the system’s associations until they break. Our “stress‐test” approach is an effective first pass at modeling coral bleaching to understand resistance and robustness as what happens leading up to the collapse of the coral–symbiont network. Heterogeneous structure of the global network decreases its resistance to temperature stress Our results show that when the connections of the natural global network are randomized, and thus no longer follow the truncated power‐law distribution of having a few hubs and many lowly connected nodes, these homogenous networks become more resistant to temperature stress under the bleaching model (Fig. 3 a, RBNDC simulation). Even when the associations are just shuffled with the natural degree distribution conserved, the network’s resistance to temperature stress increases (Fig. 3 a, RBDC simulation). This suggests that specific associations, not just the overall structure of the network increase susceptibility to link removals. Although these results are not seen for every location, the result that resistance varies across locations and spatial scales suggests that structure affects resistance. If we consider the bleaching model to be a targeted attack on this symbiosis network based on its ecophysiological properties, these results complement current theory of the resistance of heterogeneous networks to attacks (Albert et al. 2000 , Solé and Montoya 2001 ). However, our bleaching model is a novel, targeted attack type. Previously, attacks have mostly been modeled as species extinctions. The bleaching model attacks the links of the network, the associations of the ecosystem, on an environmental front. When the network’s links are randomized, thus shifting the overall structure to become homogeneous (the RBNDC null network), the stress is more evenly distributed across the network resulting in a more resistant system. The natural networks are a mixed landscape of strong and weak contributors that are connected in such a way that makes them susceptible to temperature stress. The natural distribution of thermal tolerances makes the system less resistant When the distribution of a coral–symbiont network’s tolerances is changed to that of a random uniform distribution, the network becomes more resistant to temperature stress. This is seen at all spatial scales (Fig. 3 ). We adapted thermal tolerances from two recent meta‐analyses (Swain et al. 2016 , 2017 ) that are to date the most comprehensive ranking of thermal tolerances for coral species and their symbiotic algae. Large comparative experiments and theoretical work like Swain et al. ( 2016 , 2017 ) are needed to drive the predictive power of network analyses that incorporate ecophysiological data. Network analyses and modeling provide the analytical toolbox for investigating large data sets like GeoSymbio, but as our results have shown, the ecophysiological data are an important part of the networks’ dynamics under stress. Associations among coral species and their algal symbionts create non‐uniform patterns of thermal tolerance that make the complex system more sensitive to environmental perturbations. Corals have been shown to acclimate to increasing temperatures (Marshall and Baird 2000 , Maynard et al. 2008 ). However, our results show that the overall distribution of thermal tolerances will have to shift to increase future bleaching resistance. Spatial scale and local environment affect network resistance and robustness Only the global and Pacific Ocean networks are less resistant than the null networks that changed the structure of associations (RBDC and RBNDC). However, resistance and robustness of the natural networks vary with location and scale supporting the notion that resistance is a function of network structure since the number of nodes and links, as well as the connectance, of the networks varies across scales (Table 1 ; Appendix S1 and S2 ). Hughes et al. ( 2018 ) found that the western Atlantic had two to three times more bleaching events from 1980 to 2015 than the Pacific, Indian, or Australasia regions. The Western Atlantic also experienced regular bleaching sooner than the other regions (Hughes et al. 2018 ). Their findings corroborate our results showing that the Caribbean Sea was the least resistant network of the ocean‐basin networks (Fig. 3 ). Thus, network analysis and our bleaching model can serve as a predictor of the bleaching resistance of coral reefs on a global scale. Elevated ocean temperature is the primary cause of mass bleaching and coral die‐offs (Jokiel and Coles 1990 , Fitt et al. 2001 ) and sea surface temperature is a reliable predictor of coral bleaching (Heron et al. 2016 ). The difference in resistance across scale may also be a function of the T \n MMM(2005) values attached to each host based on the subregion scale of its sampling location providing support for the influence of environment on coral bleaching. Additionally, the global network is more robust to removing links by susceptibility than to removing links by the bleaching model that incorporates temperature. Therefore, including environmental experience in the removal model decreases the system’s robustness to an environmental perturbation. However, other environmental factors like irradiance levels (Lesser et al. 1990 ) and water flow (Carpenter and Patterson 2007 , Carpenter et al. 2010 ) affect the occurrence and severity of coral bleaching. Our bleaching model is the first of its kind to model an environmental perturbation as a breakdown of an interaction on a physiological level on a network. However, it is only a first pass at modeling coral bleaching on a network, as the use of more and finer resolution environmental parameters may add more predictive power. The network is more robust to link removals, unless they are a targeted environmental stressor Link removals have mostly been ignored in studies of ecological robustness. In the case of coral bleaching, link removal is the most appropriate perturbation type for understanding the complex system under temperature stress. Across scales, networks of corals and their symbionts are more robust to link removals than to node removals. The links of the coral–symbiont network convey more stability to the network than individual nodes. However, the robustness of the global, Pacific Ocean, and Indian Ocean networks to link removals according to the bleaching model is less than that of the networks’ robustness to random removals (Fig. 4 ). These networks are vulnerable to a targeted environmental attack: bleaching. On ecological timescales, associations and interactions may play a more important role in ecosystem stability, as more often than not, environmental stressors will affect interaction patterns before eradicating entire species . Climate change is known to affect species associations and interactions by causing range shifts, behavioral changes, and impacting physiological performance (Doney et al. 2012 ). All of these happen on much shorter time scales than extinction. As organisms respond to climate change, the structure of ecological networks will first change by losing or shifting links, not by first losing nodes. Initial studies of ecological robustness may have underestimated system stability by only modeling node removals. The associations of the global coral–symbiont network lend the system stability until the links are under attack by a specific environmental stressor. Symbiodiniaceae increase the robustness of the coral–symbiont network Although the main bleaching model averaged the influence of host and symbiont on the coral holobiont’s thermal tolerance, the results of the different removal models allow us to determine which partner conveys more stability to the network. The global and ocean‐basin networks are more robust to the removal of links with susceptible symbionts than to removals according to the bleaching model, susceptible averaged tolerances, and susceptible hosts (Fig. 4 ). These results would suggest that Symbiodiniaceae convey higher levels of stability to the network in areas where it is needed, i.e., the more highly connected, generalist symbionts have higher thermal tolerances. However, the highly generalist Symbiodiniaceae only have average thermal tolerances (Appendix S1 : Table S2). Given the distribution of thermal tolerances and the degree distribution (Fig. 1 ), it is more likely that enough specialist symbionts have lower thermal tolerances than the key generalist symbionts. From a network science perspective, the coral hosts are the weak partner in this symbiosis when the coral holobiont is under attack by an environmental stressor. The possibility and limitations of applying “network triage” on coral reefs Our results indicate that the associations of a complex ecological system convey more stability to the system than the individual species. This is an important finding when considering how to model the stability of ecosystems under global climate change. Even more so, it is an important finding for understanding how to manage and conserve threatened ecosystems, like coral reefs (National Academies of Sciences, Engineering, and Medicine 2019 ). Protecting key associations may be more important than protecting individual species. In the context of the global coral reef ecosystem, preserving key associations for network stability starts with a better understanding of coral–symbiont interactions on an ecophysiological level. Future network studies should narrow in on region‐scale collections that sample multiple individuals within a given taxon in a wide range of environmental conditions to allow mapping of the network on a finer scale with more node‐specific information. Within species and individuals, symbiont community composition follows gradients of environmental irradiance (Rowan and Knowlton 1995 , Rowan et al. 1997 ) and temperature (Oliver and Palumbi 2009 , Baumann et al. 2018 ). The GeoSymbio database used to map the networks for this study may not capture local variation of coral–symbiont associations well, and thus local network collections would be useful to coral reef managers in determining reef resistance. Local variability could be mapped through targeted collections to determine where the network of coral–symbiont associations is strongest and where more conservation efforts should be focused. Once the use of ITS2 type profiles is more widely adopted (c.f. Hume et al. 2019 ), additional network analyses including the higher levels of host specificity and genetic variation should be explored. Network analysis may be useful to managers of coral reef ecosystems in planning restoration efforts and managing other stressors that impact reefs. It can provide insight into what species are most at risk. We can consider a coral species to be very susceptible to bleaching if it appears within the first 100 bleached nodes when the bleaching model is run in multiple simulations. Fig. 5 shows the number of host node occurrences in the first 100 bleached host nodes of 13 coral families and where those host nodes were sampled. According to our bleaching model, the most susceptible coral families include Acroporidae, Pocilloporidae, Poritidae, Favidae, Agaricidae, and Siderastreidae (Fig. 5 ). Species from these families should be targeted for future studies of network stability. Our resistance model could help decide whether reintroducing or engineering certain coral‐algal network associations (van Oppen et al. 2015 , National Academies of Sciences, Engineering, and Medicine 2019 ) would restabilize a collapsing network. Or it could also elucidate areas where management has neglected key associations that should be conserved. In theory, the distribution of thermal tolerances on a reef or the structure of coral–symbiont interaction patterns could be reconstructed to stabilize the network: a “network triage” approach to coral reef conservation. However, since our results show that environmental experience plays a large role in network robustness, the best conservation strategy is one that also tackles climate change. Figure 5 Occurrences of coral host family in the set of host nodes that are always in the first 100 bleached host nodes when the bleaching model is run on the global network of coral–symbiont associations. GBR, Great Barrier Reef. The global network of coral species’ associations with Symbiodiniaceae and associated thermal thresholds is nonrandom, and this architecture leads to a higher sensitivity to environmental perturbations. Associations, not species, stabilize an ecosystem when it is perturbed, unless those associations are susceptible to a certain stressor, as is the case for coral reefs. Our novel resistance metric can be adapted for different environmental stressors and ecosystems. Resistance is just the first step in a complex response to environmental change. Resilience can be defined as a complex system’s ability to adjust activity when faced with disturbances or stress and recover to a functional state of persistence. Network models of resilience (Gao et al. 2016 ) are a rapidly developing field. The coral–symbiont network provides a prime test‐bed for resilience metrics, as symbiont populations within corals can shift during and after bleaching events (Glynn et al. 2001 , Jones et al. 2008 , Sampayo et al. 2016 ) and the change in interaction patterns may be an adaptive action to increase thermal tolerance (Buddemeier and Fautin 1993 , Baker 2003 ). Future network analyses of resilience combined with our study of network resistance and robustness of global coral‐Symbiodiniaceae associations provide a new trajectory for the conservation of coral reefs under attack by climate change."
} | 6,192 |
38517972 | PMC10959402 | pmc | 320 | {
"abstract": "Neuro-symbolic artificial intelligence has garnered considerable attention amid increasing industry demands for high-performance neural networks that are interpretable and adaptable to previously unknown problem domains with minimal reconfiguration. However, implementing neuro-symbolic hardware is challenging due to the complexity in symbolic knowledge representation and calculation. We experimentally demonstrated a memristor-based neuro-fuzzy hardware based on TiN/TaO x /HfO x /TiN chips that is superior to its silicon-based counterpart in terms of throughput and energy efficiency by using array topological structure for knowledge representation and physical laws for computing. Intrinsic memristor variability is fully exploited to increase robustness in knowledge representation. A hybrid in situ training strategy is proposed for error minimizing in training. The hardware adapts easier to a previously unknown environment, achieving ~6.6 times faster convergence and ~6 times lower error than deep learning. The hardware energy efficiency is over two orders of magnitude greater than field-programmable gate arrays. This research greatly extends the capability of memristor-based neuromorphic computing systems in artificial intelligence.",
"introduction": "INTRODUCTION The objective of neuro-symbolic artificial intelligence is to leverage the computational strength of symbolic knowledge representation in conjunction with the adaptive learning abilities of deep neural networks (DNNs). Deep learning methods use massive amounts of data to extract information or knowledge ( 1 ). However, labeling vast amounts of data is time-consuming, resource-intensive, and expensive. In some fields, including medicine ( 2 ) and robotics ( 3 ), it is difficult to obtain enormous amounts of data. Their inability to be interpreted may violate commercial and medical ethics ( 4 , 5 ). Neuro-symbolic computing have the potential to enhance the interpretability, generalization, and robustness of deep learning systems by effectively combining perception, reasoning, and learning. Furthermore, neuro-symbolic approaches have the capability to integrate prior symbolic knowledge into the training aim of DNNs, hence addressing the issue of insufficient supervision from annotated instances ( 6 ). As a result, neuro-symbolic methods are widely used in domain-specific areas where a balance between performance, interpretability and robustness is preferred, e.g., automatic control ( 7 ) and data analyzing ( 8 ). The memristor is a two-terminal electronic component sometimes referred to as a “memory resistor” ( 9 – 11 ). It operates by leveraging physical principles to conduct computational tasks directly at the location where information is stored. This characteristic effectively eliminates the necessity of data transfer between the memory and computation. Memristors, integrated within a crossbar architecture, have been effectively used in feed-forward fully connected neural networks ( 12 – 23 ). These networks have demonstrated notable benefits in terms of power consumption and inference delay when compared to their complementary metal oxide semiconductor (CMOS)–based equivalents ( 24 – 26 ). However, experimental demonstrations of neuro-symbolic computing using memristors are yet to be achieved. In conventional CMOS-based hardware, the implementation of neuro-sybolic system is limited by the complexity in representing knowledge in symbolic form, which imposes a notable increase in storage, computing time and power consumption ( 27 ). Moreover, the intrinsic variability of memristors usually leads to performance degradation, especially in large arrays. Thus, two main challenges should be solved in implementation of memristive neuro-symbolic hardware: (i) an energy-efficient and compact realization of symbolic knowledge representation and (ii) utilization or depression of intrinsic memristor variability. Here, we report experimental implementation of neuro-fuzzy computing in one-transistor–one-resistor (1T1R) crossbar array and its applications in edge detection, nonlinear system identification, and robotic navigation. The symbolic knowledge in the form of fuzzy logic and rules are represented directly using the topological structure of memristive crossbar array. Intrinsic memristor variability is found to be a source of stochastic uncertainty that enhances the robustness in knowledge representation and improves system performance, challenging the conventional perception of memristive non-idealities as negative factors. Furthermore, this study proposes a hybrid in situ training technique within the framework of software-hardware co-optimization. The objective of this strategy is to address the issue of accumulated error in computing by using crossbar arrays, hence ensuring low error in computational processes. The memristor neuro-fuzzy hardware effectively capitalizes on the computing capability of matrix-vector multiplication (MVM) and the high parallelism offered by crossbar arrays. This results in an energy efficiency of 2.61 tera-operations per second per watt (TOP s −1 W −1 ) at integer 8-bit (INT8) (INT8), surpassing that of state-of-the-art field-programmable gate array (FPGA) by more than two orders of magnitude. In the domain of robotic navigation, the neuro-fuzzy hardware demonstrates superior adaptability to previously unknown environments, resulting in a convergence rate around 6.6 times faster and an error rate approximately six times lower compared to deep learning approaches. Consequently, the application scope of memristor-based neuromorphic computing systems in the field of artificial intelligence is notably broadened.",
"discussion": "DISCUSSION Our experimental demonstration validated neuro-fuzzy learning using memristive crossbar arrays in unsupervised, supervised, and transfer learning tasks. Our work provided a highly feasible solution to symbolic knowledge representation based on memristive crossbar arrays, revealed as a guidance for implementing other knowledge-based technique, e.g., metacognition, knowledge graph representation or incremental concept learning. Our findings challenged the conventional perception of memristive non-idealities as negative factors and suggested their potential in improving system performance. The proposed hardware-software co-optimization technique is a typical research paradigm that can be applied to various analog computing disciplines. Moreover, our findings are notable from the cognitive science perspective because all natural intelligent systems are hybrid, performing mental operations on both the symbolic and sub-symbolic levels ( 35 – 37 ). Therefore, this work represents an intriguing way toward the realization of future artificial general intelligence hardware."
} | 1,691 |
36825067 | PMC9941211 | pmc | 321 | {
"abstract": "Spider silk proteins (spidroins) are a remarkable class of biomaterials that exhibit a unique combination of high-value attributes and can be processed into numerous morphologies for targeted applications in diverse fields. Recombinant production of spidroins represents the most promising route towards establishing the industrial production of the material, however, recombinant spider silk production suffers from fundamental difficulties that includes low titers, plasmid instability, and translational inefficiencies. In this work, we sought to gain a deeper understanding of upstream bottlenecks that exist in the field through the production of a panel of systematically varied spidroin sequences in multiple E. coli strains. A restriction on basal expression and specific genetic mutations related to stress responses were identified as primary factors that facilitated higher titers of the recombinant silk constructs. Using these findings, a novel strain of E. coli was created that produces recombinant silk constructs at levels 4–33 times higher than standard BL21(DE3). However, these findings did not extend to a similar recombinant protein, an elastin-like peptide. It was found that the recombinant silk proteins, but not the elastin-like peptide, exert toxicity on the E. coli host system, possibly through their high degree of intrinsic disorder. Along with strain engineering, a bioprocess design that utilizes longer culturing times and attenuated induction was found to raise recombinant silk titers by seven-fold and mitigate toxicity. Targeted alteration to the primary sequence of the recombinant silk constructs was also found to mitigate toxicity. These findings identify multiple points of focus for future work seeking to further optimize the recombinant production of silk proteins and is the first work to identify the intrinsic disorder and subsequent toxicity of certain spidroin constructs as a primary factor related to the difficulties of production.",
"conclusion": "4 Conclusion We have produced a panel of recombinant spidroin constructs in ten strains of E. coli , and results show that both a restriction on basal expression and genetic mutations on multiple genes related to stress responses facilitated increased titers and plasmid maintenance. These attributes were combined to create a novel strain of E. coli that produces recombinant spidroins at levels 4–33 times higher levels than standard BL21. Interestingly, this novel strain offered no benefit for the production of an ELP with structural and compositional similarities to the recombinant spidroins. It was found that expression of the spidroins, but not the ELP, exerted toxicity on the E. coli host system. To our knowledge, our results are the first to suggest that this toxicity, which may result from the intrinsic disorder of the spidroin product, is a primary factor causing low titers of recombinant spidroins. Counterintuitively, a bioprocess design that excludes the addition of inducer and uses longer culturing times was found to mitigate toxicity and increase spidroin titers in strain BL21 by 5–7 times. Targeted alteration of the spidroin primary sequence using cDNA of natural spidroin terminal domains was also found to mitigate toxicity and increase final OD600 by 110%. Mutations on multiple genes related to stress response systems in E. coli such as yggW, yedY, yedW, yedY, speC, speB, uspC, hchA, loiP, mltC, envZ, ompR, yhgF, hupB were identified as unique attributes in the strain that produced the highest levels of spidroins and represent genes previously unknown to affect spidroin production. In conclusion, we have put forth a new hypothesis related to the persistent difficulty in producing high titers of certain recombinant spidroins, and we have outlined multiple points of focus for the future optimization of spidroin production that includes basal expression, promoter strength, induction strength, construct design, and specific genomic alterations.",
"introduction": "1 Introduction Spider silk proteins (spidroins) are of intense interest to engineers and researchers due to their high value material properties and utility in diverse applications. Orb-weaving spiders produce up to seven different types of silk, with dragline (major ampullate) silk serving as a safety line and framework of the web. Dragline silk fibers are five times stronger by weight than steel and three times tougher than the top-quality man-made fiber Kevlar ( Hardy et al., 2008 ; Fink and Zha, 2018 ). Additionally, silk is biodegradable and biocompatible, and silk proteins can be processed into numerous morphologies including coatings, hydrogels, and tissue scaffolds ( Hardy et al., 2008 ; Fink and Zha, 2018 ; Gosline et al., 1999 ). As such, the applications of silk proteins range from next-generation body armor to optofluidic devices and even coatings for food preservation ( Tsioris et al., 2010 ; Gould, 2002 ; Marelli et al., 2016 ). While silk protein from the Bombyx mori silkworm is farmed at scale for the textiles industry, dragline silk cannot be readily obtained through farming, as spiders are territorial and cannibalistic ( Tokareva et al., 2013 ). Thus, researchers have used recombinant production to obtain proteins that mimic or directly copy the sequences of natural dragline spidroins. Recombinant production currently represents the most promising method for producing dragline spidroins at scale while also presenting the ability to rationally design protein sequences with targeted properties 8-10. . The unique properties of dragline spidroins arise from specific peptide motifs, chemical interactions, and hierarchical organization that are highly conserved among orb-weaving spiders. Natural dragline fibers are composed of two proteins, Major Ampullate Spidroin 1 (MaSp1) and Major Ampullate Spidroin 2 (MaSp2) in a ratio of approximately three MaSp1 for every two MaSp2 ( Heim et al., 2009 ; Sarkar et al., 2019 ). Major ampullate spidroins are generally quite large at 250–350 kDa and take the form of a segmented copolymer with small non-repetitive N and C-terminal domains that flank a large repetitive core domain. The repetitive domain of dragline spidroins represents approximately 90% of the total protein, with repeating units that are typically 33–45 amino acids long ( Heim et al., 2009 ; Sarkar et al., 2019 ; Gatesy et al., 2001 ). The repeat unit of MaSp1 is characterized by a tandem alanine repeat (A n , where n ∼6–9) adjacent to a glycine-rich region that contains GGX motifs, where X often represents tyrosine (Y), glutamine (Q), or leucine (L). The repeat unit of MaSp2 also contains tandem alanine repeats, but its glycine-rich region is high in proline (P) and contains GPGXX and GGX motifs, where X often represents Q, Y, L, G, or serine (S) ( Teulé et al., 2009 ; Guerette et al., 1996 ; Malay et al., 2017 ). In both MaSp1 and MaSp2, the tandem alanine segments assemble into beta-sheet nanocrystals, and the glycine-rich regions form an amorphous matrix during fiber spinning. The interplay between these crystalline and amorphous domains endows spider silk with many of its unique properties, including a combination of high tensile strength and toughness ( Sarkar et al., 2019 ; Tokareva et al., 2014 ). Recombinant silk has been produced in a diverse set of host organisms, including bacteria, yeast, mammalian cells, insect cells, transgenic plants, and transgenic animals ( Wohlrab et al., 2014 ; Chung et al., 2012 ). Common practice in the field is to create synthetic spider silk genes that combine spidroin amino acid motifs (GGX (A) n , etc.), in ways that mimic the repetitive core of a natural dragline sequence. This is due to the difficulty in obtaining exact copies of full-length dragline spidroin genes by PCR, as they are long, repetitive, and have a high GC content ( Wohlrab et al., 2014 ; Chung et al., 2012 ). Recombinantly produced dragline spidroins generally have anywhere from 2 to 196 repeats of a relatively short “monomer” segment (typically around 35 amino acids) and may or may not include non-repetitive terminal domains. Most efforts to produce recombinant spidroins have suffered from low titers, preventing the production and utilization of artificial spider silk at a commercial scale ( Whittall et al., 2021 ; Edlund et al., 2018 ). Additionally, expressing recombinant spidroins in bacteria is often plagued by a high degree of plasmid instability, inclusion body formation, low solubility of the spidroin constructs, and transcriptional and translational errors ( Yang et al., 2016 ; Chung et al., 2012 ). These issues, particularly the low titers, correlate with recombinant spidroin size, which further limits the production of useful materials, as increasing spidroin size has been shown to increase the mechanical properties of resultant fibers ( Whittall et al., 2021 ; Wohlrab et al., 2014 ; Chung et al., 2012 ; Edlund et al., 2018 ; Bowen et al., 2018 ). However, two recent works have shown that high titers of recombinant spidroins are possible. Using an E. coli host system, Yang et al. achieved a titer of 3.6 g/L for 200 kDa dragline spidroin in a bioreactor kept at 16 °C. The researchers also employed a secondary plasmid to upregulate glycyl-tRNA supply ( Yang et al., 2016 ). Furthermore, Schmuck et al. documented a titer of 14.5 g/L for a small recombinant spidroin using an E. coli host system in a bioreactor. This 33 kDa recombinant spidroin only contained two monomer repeats in its primary sequence, but it could be spun into fibers that exhibited mechanical properties similar to much larger recombinant spidroins ( Schmuck et al., 2021 ). Despite these promising efforts to increase titer, the relatively large body of work surrounding recombinant silk production has not been able to fully understand the challenges in expressing dragline spidroin. Yang et al. demonstrated that glycyl-tRNA upregulation can facilitate higher titers of large, glycine-rich dragline spidroins. However, their work also showed that decreasing the culture temperature from 30 °C to 16 °C, independent of glycyl-tRNA upregulation, was sufficient to increase titers nearly an order of magnitude from 0.36 g/L to 2.9 g/L through an unknown mechanism ( Yang et al., 2016 ). Moreover, Bhattacharyya et al. recently showed that while increasing tRNAGly and tRNAPro was helpful for producing certain spidroin sequences, it could be non-beneficial or even toxic to the cells when other spidroin sequences were expressed ( Bhattacharyya et al., 2021 ). These findings indicate that mechanisms aside from translational difficulties need to be addressed. Work related to strain optimization of viable industrial hosts and hypotheses related to upstream production bottlenecks are sparse within the recombinant silk field. Likewise, little work outside of increasing spidroin length has been done to understand how alterations to the primary sequence affects upstream expression outcomes such as cell growth, plasmid maintenance, and titer. A thorough analysis of the E. coli platform using multiple strains and spidroin primary sequences can move the field forward, since many studies in this space have only used a “default” E. coli BL21 strain ( Whittall et al., 2021 ; Schmuck et al., 2021 ; Yang et al., 2016 ; Teulé et al., 2009 ; Bowen et al., 2018 ; Bhattacharyya et al., 2021 ; Cai et al., 2020 ; Wei et al., 2020 ; Andersson et al., 2017 ; Xia et al., 2010 ; Fredriksson et al., 2009 ; Humenik et al., 2014 ). As a long-standing host choice for synthetic biology, the E. coli platform has become highly diversified with numerous distinct strains. Several E. coli strains present characteristics that can elucidate additional cellular mechanisms to target for increasing recombinant spidroin titers. In this study, we sought to uncover a deeper understanding of how to construct microbial strains for optimal spidroin production. We also aimed to understand how the design of a recombinant spidroin construct affects production efficiency. To that end, we studied the expression of a panel of dragline spidroins that varied systematically in size and primary sequence in ten E. coli strains to characterize the relationships between upstream outcomes, host characteristics, and primary spidroin sequence. Our results suggest that the expression of MaSp2-mimetic recombinant spidroins exerts toxicity on E. coli , as shown through a negative effect on cell growth and plasmid maintenance, but that these effects could be attenuated with targeted primary sequence alteration. For further mechanistic insight, we extended our work towards expressing elastin-like peptide (ELP), a repetitive protein of interest for tissue engineering and drug delivery applications. Recombinant ELPs share several similarities with dragline spidroins, including a polymeric structure, a high glycine and proline content, and an ability to self-assemble when triggered by external stimuli ( Schmuck et al., 2021 ; Sarkar et al., 2019 ; Varanko et al., 2020 ). The ELP, however, exerted no observable toxicity on host cells during expression. We hypothesized that this is due to a higher intrinsic disorder for our spidroin constructs compared to the ELP. Eliminating basal expression, or refraining from gene induction, was found to play a key role in obtaining high titers of toxic recombinant spidroin constructs but had no beneficial effect for ELP production. Additionally, we identified multiple genes related to stress response as new targets for future strain optimization that aims to increase titers of toxic recombinant proteins. Using these findings, we developed a novel strain of E. coli that combines both a restriction on basal expression and unique genetic mutations to facilitate high spidroin titer and plasmid maintenance without sacrificing the ability to use a strong promoter or exert control over gene induction. This novel strain offered no benefit for ELP production, indicating that future work should address production bottlenecks that may be unique to recombinant spidroin constructs.",
"discussion": "3 Results and discussion 3.1 Design of recombinant spidroin constructs and screening of ten E. coli strains To screen recombinant spidroin expression in different E. coli strains, nine commercially available strains were used in this work ( Supplementary Table 1 ). Some strains were chosen based on factors previously shown or hypothesized to affect recombinant silk production, such as codon usage and inclusion body formation. These strains include RosettaGami B, which has upregulation of seven tRNAs for rare codons including those for glycine and proline, as well as BLR, which has a recA − mutation that may facilitate increased stability of plasmids containing repetitive sequences ( Yedahalli et al., 2016 ; Huemmerich et al., 2004 ; Karla et al., 2005 ). The strain SoluBL21 has been developed through directed evolution to produce soluble protein when its ancestral strain, BL21(DE3), does not yield detectable soluble product. Likewise, strain pGro7 expresses a chaperone protein that prevents inclusion body formation and promotes soluble production ( SoluBL21 Competent E. coli, 2022 ; Nishihara et al., 1998 ). The remaining strains were chosen to potentially identify unknown targets for the optimization of recombinant spidroin expression. This includes strains HMS174 and MG1655, which unlike other strains tested, are from the K-12 E. coli lineage instead of the B lineage ( Hausjell et al., 2018 ; Leal‐Egaña et al., 2012 ; Bachmann, 1972 ). Strain pLysS restricts basal expression, while strain Origami B contains mutations that change the cytoplasmic environment and cellular stress responses through alterations to the thiol-redox equilibrium, glutathione metabolism, and oxidative stress response ( Studier and Moffatt, 1986 ; Kong and Guo, 2014 ; Lobstein et al., 2016 ; Kunert et al., 1990 ). All strains used in this study were DE3 lysogens and proteins were expressed from the pET19b vector under control of the T7 promoter. We additionally studied a tenth hybrid strain, SoluBL21-pLysS, which we developed by combining features from pLysS and SoluBL21. Four different de novo designed spidroin constructs were expressed in these E. coli stains, with titer, plasmid maintenance, and OD600 measured as expression outcomes. The primary sequences and polymeric structure of the spidroin constructs are depicted in Table 1 . To assess the effect of protein size, recombinant spidroins were designed to have either four or sixteen identical monomer units in tandem (referred to as 4mers and 16mers, respectively). To assess the effect of modulating primary sequence, two different monomer units of 35 amino acids were designed, with one containing a segment of five tandem alanine residues (A5) and the other a segment of ten tandem alanine residues (A10). The remaining amino acids in the monomer sequences consists of multiple GPGQQ motifs (four for the A5 monomer and three for the A10 monomer) and single GPGEQ and GPGSG motifs. Both monomer units were designed based on naturally occurring primary sequences found in the MaSp2 dragline spidroin of orb-weaving spiders ( Teulé et al., 2009 ; Guerette et al., 1996 ; Malay et al., 2017 ). Modulating the tandem alanine length and total construct length were chosen as focal points to study the effect of construct design on expression outcomes, as these are primary factors implicated in determining the material properties of recombinant silks ( Bowen et al., 2018 ; Zhao et al., 2021 ; Bratzel and Buehler, 2012 ; Heidebrecht et al., 2015 ). All constructs expressed have an identical starting sequence that is present on the pET-19b expression vector, which contains a 10x histidine tag for purification followed by an enterokinase cleavage sequence (cleavage of the 10x histidine tag using the enterokinase site was not employed in this work). Table 1 Primary sequence of recombinant spidroins. Table 1 Spidroin Primary Sequence # of Repeats Molecular Weight (kDa) A5 4mer MGHHHHHHHHHHSSGHIDDDDKHMLEHMPG n = 4 16.1 (GPGQQ AAAAA GPGQQGPGQQGPGQQGPGEQGPGSG)n TSGS A5 16mer MGHHHHHHHHHHSSGHIDDDDKHMLEHMPG n = 16 52.7 (GPGQQ AAAAA GPGQQGPGQQGPGQQGPGEQGPGSG) n TSGS A10 4mer MGHHHHHHHHHHSSGHIDDDDKHMLEHMPG n = 4 15.6 (GPGQQ AAAAAAAAAA GPGQQGPGQQGPGEQGPGSG) n TSGS A10 16mer MGHHHHHHHHHHSSGHIDDDDKHMLEHMPG n = 16 50.9 (GPGQQ AAAAAAAAAA GPGQQGPGQQGPGEQGPGSG) n TSGS 3.2 Recombinant silk titers in ten E. coli strains Expression of the A5 and A10 constructs took place in LB media at 37 °C for 4 h. Initial findings showed that pLysS and SoluBL21 had higher production levels for the smaller recombinant spidroins (A5 4mer and A10 4mer). These strains yielded approximately 80–100 mg/L of A5 and A10 4mer protein, producing at levels several times above the other strains ( Fig. 1 a). Considering these results, the pLysS plasmid from the pLysS strain was extracted and transformed into SoluBL21 to form a novel hybrid strain, SoluBL21-pLysS, to investigate whether the advantages of the parent strains could be synergistic. The hybrid strain was able to produce the small spidroins at 201 (±6) and 189 (±10) mg/L for the A5 4mer and A10 4mer, respectively. These titers are approximately 70 mg/L higher than the reported shake flask titer for a recombinant spidroin that was recently produced at the highest titer ever reported for a recombinant spidroin in a bioreactor ( Schmuck et al., 2021 ; Andersson et al., 2017 ). Interestingly, these titers are approximately twice that of either parent strain. Moreover, when compared to BL21, which is a typical strain used for spidroin expression, these titers represent a 13X increase for the A5 4mer and a 33X increase for the A10 4mer ( Fig. 1 a). Fig. 1 (a) Soluble titer for the small recombinant spidroins, A5 4mer (black) and A10 4mer (gray), for ten strains of E. coli in LB media. Error bars represent standard deviations from the mean values of three replicates. (b) Soluble titer for the large recombinant spidroins, A5 16mer (black with white dots) and A10 16mer (white with black dots), in ten strains of E. coli in LB media. Error bars represent standard deviations from the mean values of three replicates. Fig. 1 The titers for both the A5 and A10 16mers were lower than that of the 4mers for nearly all strains, which is consistent with previous findings showing an inverse relationship between yield and spidroin length ( Xia et al., 2010 ). Despite displaying some of the highest titers for the 16mers, at 11–15 mg/L, strains pLysS and SoluBL21 showed no appreciable advantage over BL21, pGro7, or BLR, which all yielded similar titers ( Fig. 1 b). Interestingly, the SoluBL21-pLysS hybrid strain outperformed both of its parent strains for producing the 16mers, with titers of 53 (±4) and 49 (±3) mg/L for the A5 16mer and A10 16mer, respectively. This is approximately a four-fold increase in 16mer titer versus the parent strains. Several strains, including RosettaGami, HMS174, MG1655, and Origami B were barely capable of producing detectable levels of the 16mers, as shown by titers of 5 mg/L or less. The poor performance of RosettaGami B for both the 4mers and 16mers is of particular interest, as a leading hypothesis in the field relates the low titers of silk proteins to translational difficulties caused by the overabundance of a select few amino acids in spidroins, mainly glycine ( Yang et al., 2016 ; Bowen et al., 2018 ; Xia et al., 2010 ). Recent work has shown that upregulating glycyl-tRNA has favorable effects on the production of spidroins, however, the upregulation of glycine and proline tRNAs inherent in RosettaGami appeared to have no positive effect on titer ( Yang et al., 2016 ; Bowen et al., 2018 ; Xia et al., 2010 ). Moreover, we observed no translation truncation of the A5 and A10 spidroins in SDS PAGE analysis ( Fig. S1 ). Translational truncation, or early termination of the nascent recombinant spidroin from the ribosomal complex, has been observed by multiple researchers and is indicative of translational bottlenecks in silk production ( Bhattacharyya et al., 2021 ; Xia et al., 2010 ; Fahnestock and Irwin, 1997 ). These findings support the hypothesis that translational difficulties alone cannot fully explain the difficulties in obtaining high titers of recombinant silk. 3.3 Recombinant ELP production in BL21, pLysS, SoluBL21, and SoluBL21-pLysS To investigate if the increased spidroin titers achieved with the hybrid SoluBL21-pLysS strain could be extended to other repetitive, structural proteins, an elastin-like peptide (ELP) was produced in strains BL21, pLysS, SoluBL21, and SoluBL21-pLysS. The recombinant ELP, A4Y1, was chosen for production in these four strains based on a balance between similarity and difference when compared to the A5 4mer primary sequence ( Table S2 ). The A5 4mer and the A4Y1 ELP both have a 4mer polymeric structure, along with a near identical molecular weight and glycine, proline, and alanine contents. Furthermore, both recombinant spidroins and ELPs are known to self-assemble into supramolecular materials when triggered by external stimuli ( Schmuck et al., 2021 ; Sarkar et al., 2019 ; Varanko et al., 2020 ). However, key differences are that the A5 4mer has tandem alanines (A n ) while the ELP has alanine residues distributed throughout the construct. Additionally, the A5 4mer has a high amount of glutamine (21%), which the ELP lacks entirely, and the ELP has a high percentage of valine (16%) and some tyrosine (3%), both of which are missing from the A5 4mer. Fig. 2 a shows that strains pLysS, SoluBL21, and SoluBL21-pLysS offered no advantage over BL21 for the titer of the ELP construct. The soluble titers for these three strains were similar, at ∼240 mg/L. SoluBL21 performed the worst out of the four strains with a titer of 196 (±8) mg/L. In all cases, the A5, A10, and ELP constructs were expressed primarily in the soluble fraction of the lysate, with only negligible amounts found in the pellet (<2 mg/L for all strains). Notably, the titer for the ELP in BL21 is over 15 times higher than for the A5 4mer under identical expression conditions. This finding, as well as the observation that the hybrid strain offered no advantage, were both unexpected when considering the high degree of similarity between the A5 4mer and the A4Y1 ELP. As peptides of identical molecular weight and similar polymeric structure, a possible explanation for these outcomes may lie in structural differences that result from primary sequence variation, the mainly the disparity between valine and glutamine content. Fig. 2 (a) Titer of the A4Y1 ELP in BL21, pLysS, SoluBL21, and SoluBL21-pLysS (SB21-pLysS) strains. Error bars represent standard deviations from the mean values of three replicates. (b) Plasmid maintenance of the ELP construct in the strains BL21, pLysS, SoluBL21, and SoluBL21-pLysS (SB21-pLysS) at the end of a 4-h expression in LB media. Error bars represent standard deviations from the mean values of three replicates. Fig. 2 3.4 Plasmid maintenance during expression of silk and ELP constructs During a recombinant protein expression, cells can potentially lose the plasmid vector that was transformed into them. Plasmid loss is indicative of excessive metabolic burden, which may stem from repetitive recombinant DNA sequences or toxic recombinant protein products and is exacerbated by depletion of antibiotic selection factors ( Yang et al., 2016 ; Dumon-Seignovert et al., 2004 ; Corchero and Villaverde, 1998 ). Plasmid-free cells can continue to divide, leading to a substantial decrease in the overall number of cells in a culture that are producing recombinant protein. A high level of plasmid maintenance is a critical factor for achieving high titers, particularly for high-density cell cultivation in bioreactors ( Yang et al., 2016 ). Fig. 3 shows the plasmid maintenance of the A5 and A10 constructs in the ten E. coli strains, and Fig. 2 b shows the plasmid maintenance of the ELP construct in BL21, pLysS, SoluBL21, and SoluBL21-pLysS. Both figures represent the plasmid maintenance at the end of a 4-h expression in LB media. Fig. 4 shows that plasmid maintenance decreased in 75% of cases when a strain transitioned from a 4mer to a 16mer expression (ex. pLysS A5 4mer compared to pLysS A5 16mer), while increasing in only 10% of cases. Instability of recombinant silk plasmids has been previously reported, but to our knowledge this is the first work to observe a correlation between plasmid instability and number of genetic repeats across multiple spidroin designs and E. coli strains ( Yang et al., 2016 ; Chung et al., 2012 ; Edlund et al., 2018 ). Although plasmid maintenance of the 16mers decreased substantially for pLysS compared to the 4mers, the hybrid strain maintained the ability of SoluBL21 to maintain these 16mer vectors at 85% or above. Fig. 3 Plasmid maintenance of the A5 4mer (blue), A5 16mer (gray), A10 4mer (orange), and A10 16mer (yellow) in the ten E. coli strains at the end of a 4-h expression in LB media. Error bars represent standard deviations from the mean values of three replicates. Fig. 3 Fig. 4 OD600 of cultures directly before and after a 4-h expression of the four different silk constructs. All cultures were induced for protein expression at an OD600 of 0.6–0.8, the final OD600 at the end of the 4-h expression ranged from 1.58 to 3.86. Error bars represent standard deviations from the mean values of three replicates. Fig. 4 This data suggests that a high titer of recombinant spidroin required high plasmid maintenance, though high plasmid maintenance does not necessarily lead to high titer in every case. The strains that yielded the highest titers of the 4mer proteins, pLysS, SoluBL21, and SoluBL21-pLysS, all exhibit a plasmid maintenance of 90% or higher. Likewise, the strain that yielded the highest titers for the 16mers, SoluBL21-pLysS, exhibited one of the highest overall plasmid maintenance levels for the 16mers. However, several strains, namely MG1655, BLR, and HMS174, exhibited a high level of plasmid maintenance in some cases (upwards of 90%) but relatively low titers for all constructs. This indicates that factors outside of the maintenance of spidroin vectors are at least partially responsible for low titers. For the BL21, RosettaGami, pGro7, and Origami strains, there was nearly a complete loss of the plasmid during spidroin expression. In contrast, Fig. 2 b shows that strain BL21 exhibited a plasmid maintenance of 43% during the ELP expression, which is over 14 times higher than maintenance during the expression of the A5 4mer or any other spidroin. This is in accordance with the 15 times higher level of production that BL21 was able to achieve for the ELP versus the A5 4mer. The strains pLysS, SoluBL21, and SoluBL21-pLysS exhibited levels of plasmid maintenance for the ELP at or near 100%, similar to their behavior during spidroin expressions. 3.5 Cell growth during expression of spidroin and ELP constructs We investigated cell growth over a 4-h expression of the four different spidroin constructs ( Fig. 4 ). While all cultures were induced for protein expression at an OD600 of 0.6–0.8, the final OD600 at the end of the 4-h expression was highly variable among the strains, ranging from 1.58 to 3.86. The high-producing strains, pLysS, SoluBL21, and SoluBL21-pLysS, showed final OD600s that were at or near the median of the dataset obtained for the ten strains (median of 2.07). The OD600 at the end of a 4-h expression did not show an obvious relationship to spidroin titer, as strains with poor titers showed both higher and lower final OD600s than pLysS, SoluBL21, and SoluBL21-pLysS. Notably, strains BL21, pGro7, and Origami B, which grew the most during spidroin expression by routinely reaching final OD600s of above 3, were also the strains that showed the lowest levels of plasmid maintenance in addition to low titers. This is likely due to the degradation of ampicillin in the media, which allows non-plasmid bearing cells that are potentially fitter to proliferate. This phenomenon is particularly applicable in cases where the recombinant construct is harmful or toxic to the cells ( Dumon-Seignovert et al., 2004 ; Corchero and Villaverde, 1998 ). This problem is further exacerbated by using ampicillin over other antibiotics as the selection pressure, since the product of the beta-lactamase gene conferring resistance to ampicillin is secreted, with studies showing that rapid plasmid loss and growth of non-plasmid bearing cells can be difficult to prevent, even in cases where additional ampicillin is added to the culture ( Dumon-Seignovert et al., 2004 ; Corchero and Villaverde, 1998 ; Sieben et al., 2016 ; Rosano and Ceccarelli, 2014 ). Thus, the increased growth rates of strains that have lost the vector and produce little silk protein may suggest that expression of the A5 and A10 spidroins exerts toxicity on the cells. The possibility that expressing A5 and A10 spidroins causes host cell toxicity was further supported by observations made during the plasmid maintenance assay, in which 0.1 ml of a 10,000x culture dilution (generated through serial dilutions) was plated for colony counting. For most strains observed, this procedure resulted in several hundred single colonies on LB agar plates. However, the strains that showed moderate to high levels of plasmid maintenance but low titers and inhibited growth after induction (RosettaGami, BLR, HMS174, and MG1655) displayed a lack of colony forming units on LB plates using this protocol. Compared to other strains at the same OD600, cultures of RosettaGami, BLR, HMS174, and MG1655 required a 100x (instead of a 10,000x) dilution of a culture sample to obtain enough isolated colonies for the plasmid maintenance assay (minimum of 50 colonies required). This lack of colony-forming-units after recombinant protein induction is a documented effect of toxic protein expression in cases where the vector is still maintained ( Dumon-Seignovert et al., 2004 ; Rosano and Ceccarelli, 2014 ; Onodera et al., 1996 ; Saïda et al., 2006 ; Kwon et al., 2015 ). Support for a toxicity effect from the spidroins, but not necessarily the ELP , can be seen when the final OD600 for cells expressing ELP vs A5 4mer. ( Fig. S2 ). The final OD600 at the end of a 4-h expression for the ELP was higher for pLysS, SoluBL21, and SoluBL21-pLysS (+0.22, 0.72, 0.575, respectively), even with a higher titer of ELP than A5 4mer. There was a decrease in the final OD600 of −0.64 for BL21 during ELP expression when compared to A5 4mer. However, this can likely be attributed to the large difference in titers between the two proteins, with 15 times more ELP than A5 4mer made by BL21. Protein overexpression at a high level, independent of the toxicity of the construct, is associated with decreases in cell growth ( James et al., 2021a ). 3.6 Toxicity of recombinant spidroin constructs To further study the potential toxicity of the A5 and A10 spidroins when compared to the ELP, growth under a variety of conditions was measured for strain SoluBL21-pLysS. If a recombinant product is toxic to the host, then growth during production will be decreased or completely inhibited, leading to an earlier stationary phase at a lower OD600 when compared to uninduced or empty vector controls ( Dumon-Seignovert et al., 2004 ; Rosano and Ceccarelli, 2014 ; Onodera et al., 1996 ; Saïda et al., 2006 ; Kwon et al., 2015 ). Fig. 5 shows the OD600 of SoluBL21-pLysS in LB media for 10.25 h under a variety of conditions and starting from the point of inoculation (2% v/v inoculum from an overnight culture). The uninduced cultures and empty vector controls reach a late exponential or early stationary phase after 10.25 h and at an OD600 of approximately 3.72–4.18. The cultures induced for ELP production follow these curves closely in both rate of growth and overall growth, as there is no significant decrease in the slope of the curve after induction, and the final OD600 of 3.42 is only an 8% decrease from that of the induced empty vector control (3.72). These findings support the conclusion that ELP expression does not significantly affect cell health and fitness, even at titers above 200 mg/L. However, when expressing an A5 or A10 spidroin, there is a significant decrease in the slope of the growth curve directly after induction at 1.75 h. Then, a stationary phase is reached up to 5 h sooner and the final OD600 decreases on average by −61% and −57% when compared to the uninduced or induced empty vector controls, respectively. When expressing the A10 4mer, there is potentially a death phase (consistent decrease in OD600) that begins approximately 4 h after induction. Fig. 5 OD600 of SoluBL21-pLysS in LB media for 10.25 h under a variety of conditions. T = 0 indicates time of inoculation (2% v/v inoculum from an overnight culture). For induced cultures, induction occurred with 1 mM IPTG at 1.75 h post inoculation. Growth curves for all silk constructs and the ELP under induced (triangles, I) and uninduced (circles, U) conditions were generated. Growth curves for the strain harboring the empty expression vector (Empty) that lacks a recombinant gene were also generated with and without induction. Data points on the curve represent the mean OD600 for that time point from three replicates. Fig. 5 3.7 Disorder of recombinant spidroins as a primary factor for toxicity and low titers Our data on growth kinetics, plasmid maintenance, and expression levels among the strains, suggests a toxicity of the A5 and A10 spidroin constructs that can be addressed with proper strain engineering to obtain higher titers. As this is in sharp contrast to the ELP, we performed additional work to understand why the expression outcomes between A5 4mer and ELP, which have similar polymeric structure and amino acid composition, were profoundly different. An important observation throughout this work was the aberrant mobility of purified spidroins through an SDS-PAGE gel ( Fig. S1 ). The A5 4mer and ELP both have molecular weights of 16 kDa, but the A5 4mer shows an apparent weight of 38 kDa on an SDS PAGE, whereas the ELP appears at 14 kDa, which is close to its true molecular weight. A high degree of aberrant mobility has been shown to be a unique characteristic of disordered proteins and is observed for the three other spidroins as well ( Fig. S1 ) ( Pedersen et al., 2020 ; Tan et al., 2021 ). Analysis of amino acid content indeed suggests that the spidroins may be more disordered than the ELP. Amino acids can be ranked in terms of their propensity to promote structural order or disorder (from most order-promoting to least order-promoting): W, F, Y, I, M, L, V, N, C, T, A, G, R, D, H, Q, K, S, E, P ( Campen et al., 2008 ). While both the spidroin and ELP constructs have similar glycine and proline content (∼33% and ∼15%, respectively), the spidroins also contain serine, glutamate, and a high proportion of glutamine (more than 20% for the A5 constructs), all of which are lacking in the ELP sequence ( Table S2 ). Furthermore, the ELP sequence contains a higher proportion of amino acids that promote structural order, including valine and tyrosine (both are absent from the A5 4mer). Computational tools predict a much greater likelihood of disorder in our spidroins versus the ELP ( Figs. S3 and S4 ). The IUPRED2A disorder predictor shows near 100% probability that all residues within the A5 4mer sequence are in regions of structural disorder. In contrast, the ELP sequence fluctuates around approximately 50% disorder probability throughout, with the probability of disorder never rising above 75% ( Fig. S3 ) ( Mészáros et al., 2018 ). The DisMeta computational disorder calculator suggests the entire length of the A5 and A10 spidroin constructs as having a 90–100% chance of being disordered ( Huang et al., 2014 ). In contrast, the ELP contains large regions that have much lower probability of disorder, including sections with disorder probability of 10% or less ( Fig. S4 ). Additionally, FTIR and secondary structure analysis show that the A5 4mer contains a 37% (±11) random coil content, compared to 0% (±0) for the ELP ( Fig. 6 . Raw, reproduced, and deconvoluted spectra can be seen in Fig. S5 ), where random coils can be associated with structural disorder and conformational flexibility in proteins ( James et al., 2021b ; Choi et al., 2011 ; Marsh and Forman-Kay, 2010 ). Fig. 6 Secondary structure analysis derived from FTIR spectra for the A5 4mer and ELP proteins. Error bars represent standard deviations from the mean values of three replicates. Fig. 6 We hypothesize that the disordered nature of the spidroin constructs is a main factor underlying their toxicity, low titers, and low plasmid maintenance when compared to the ELP. Overexpression of disordered proteins has been shown to exert toxicity in both D. melanogaster and C. elegans in a dose-dependent manner ( Pedersen et al., 2020 ; Vavouri et al., 2009 ). Production of disordered recombinant proteins has also been shown to yield negligible titers and cause toxicity in E. coli and yeast species such as S. cerevisiae ( Treusch and Lindquist, 2012 ; Liang et al., 2010 ; Hwang et al., 2012 ; Guo et al., 2021 ). It is hypothesized that this toxicity may result from promiscuous and harmful binding interactions by disordered proteins within the intracellular milieu ( Pedersen et al., 2020 ; Vavouri et al., 2009 ; Treusch and Lindquist, 2012 ; Liang et al., 2010 ; Hwang et al., 2012 ; Guo et al., 2021 ). To our knowledge, this is the first work to show toxicity resulting from protein disorder as a primary factor causing unfavorable outcomes during recombinant spidroin expression. This is of particular interest to the field, as many recombinant MaSp2-mimetic spidroins produced by other groups are similar to our A5 and A10 constructs in that they are high in disorder-promoting amino acids (proline, glutamine, glycine, serine) and low in order-promoting amino acids with low titers consistently reported ( Table S3 ) . 3.8 Addressing spidroin toxicity with strain engineering and experimental design Our hybrid SoluBL21-pLysS strain showed substantially improved recombinant spidroin expression compared to other E. coli strains ( Fig. 1 ). To understand the underlying cellular mechanisms, a series of experiments examining basal expression were performed. Basal expression refers to the expression of a recombinant gene without induction, which can cause plasmid loss and subsequent low titers if the gene product is toxic. Strain pLysS (and SoluBL21-pLysS by default) is designed to exert tight control over basal expression through the action of the pLysS vector. The product of the pLysS vector is T7 lysozyme, which inhibits action of T7 polymerase and prevents basal expression of recombinant genes placed on pET vectors. Upon addition of IPTG T7 polymerase concentration increases and overcomes the inhibition of pLysS ( Rosano and Ceccarelli, 2014 ). Fig. 7 a presents experimental verification that basal expression was strongly restricted in strains pLysS and SoluBL21-pLysS, showing either an absence or low levels (<7 mg/L) of A5 4mer protein expression from an overnight culture grown without IPTG. Basal expression was also attenuated in strain SoluBL21, although slightly less so than for the aforementioned strains (11 (±3) mg/L from an overnight culture without IPTG). Strain BL21 exhibited copious basal expression, with overnight cultures (18 h) lacking any IPTG induction yielding an average A5 4mer titer of 102 (±11) mg/L. Cultures of BL21 inoculated from an overnight culture and grown to OD600 of 0.6–0.8 also showed basal expression at 8 (±1) mg/L. Moreover, when the strain was put through the same 4-h expression protocol as previously used, save for the addition of IPTG, the titer was 83 (±9) mg/L. These basal expression titers for the A5 4mer in BL21 are significant in that they are approximately equal to that of the pLysS and SoluBL21 strains, however, they are only half of what the high-performance SoluBL21-pLysS strain yielded. Fig. 7 (a) Basal Expression of A5 4mer in strains Bl21, pLysS, SoluBl21, and SoluBL21. Basal expression from a culture grown overnight (18 h) after inoculation from a single colony (blue bar). Basal expression from a culture inoculated from an overnight liquid culture (2% v/v inoculum), grown to OD600 of 0.6–0.8 and harvested (orange bar). The orange bar thus represents basal expression directly before what would have been the induction point for a typical expression culture. Basal expression during a 4-h expression without addition of IPTG (gray bar). The gray bar shows basal expression from a culture inoculated from an overnight liquid culture (2% v/v inoculum), grown to OD600 of 0.6–0.8, and allowed to grow for four more hours without addition of IPTG. Error bars represent standard deviations from the mean values of three replicates. (b) A5 4mer Basal Expression Growth Curve for strain BL21. OD600 over 10.25 h for strain BL21 with the A5 4mer or empty vector under induced (I) or uninduced (U) conditions. Cultures were inoculated with 2% v/v inoculum from an overnight culture. Induction (if applicable) occurred with 1 mM IPTG at 2.75 h total culture time when the OD600 was 0.6–0.8. Empty vector corresponds to strain BL21 transformed with the pET19b vector that has had the silk gene excised. Data points on the curve represent the mean OD600 for that time point from three replicates. Fig. 7 Growth curves for strain BL21 give further insight into the effect of basal spidroin expression. Fig. 7 b shows OD600 over 10.25 h for the strain under a variety of conditions, including growth with and without inducer and empty vector controls. This data shows that basal expression of the A5 4mer construct had almost no negative effect on the growth of the BL21 strain when compared to controls, even though 83 (±9) mg/L of A5 4mer protein had accumulated intracellularly by 6.75 h into the experiment. Interestingly, the induced strain showed no significant deviation in growth from the uninduced curves, however the titer at 6.75 h total culture time (4 h post induction with 1 mM IPTG) was only 15 (±8) mg/L. Thus, counterintuitively, the BL21 strain produces approximately 5.5 times more silk protein when it is not induced with IPTG. Fig. S6 shows that this outcome is likely a result of changes in plasmid maintenance in response to induction. The plasmid maintenance at 2.75 h (directly before induction) for both uninduced and induced are roughly the same (50%), albeit low due to the basal expression of a toxic gene ( Dumon-Seignovert et al., 2004 ; Sieben et al., 2016 ; Kwon et al., 2015 ). However, after induction with 1 mM IPTG, the plasmid maintenance of the induced culture decreases to just 3% after 4 h while the uninduced maintains the vector at 26%. This finding suggests that the leakiness, and likely strength, of a promoter plays an integral role in the successful production of toxic recombinant spidroins. In uninduced conditions, the BL21 strain is already expressing the recombinant silk gene, meaning that subsequent additions of IPTG may overwhelm cellular machinery beyond its ability to handle toxic protein expression, causing severe plasmid loss and low titers. Likewise, since the pLysS, SoluBL21, and SoluBL21-pLysS strains do not exhibit much basal expression they exhibit high plasmid maintenance and are likely more tolerant of the increase in promoter activity caused by IPTG ( Yang et al., 2016 ; Dumon-Seignovert et al., 2004 ; Corchero and Villaverde, 1998 ; Kwon et al., 2015 ). In this context, weaker promoters, or a strong promoter in combination with a restriction on basal expression, may be easily implemented tools for optimizing spidroin expression. The BL21 strain and strong promoters, mainly the bacteriophage T7 promoter, are commonplace in the recombinant silk field ( Table S3 ), yet they may be a reason for the fundamental problems related to low titers, biomass accumulation, and plasmid stability ( Huemmerich et al., 2004 ; Scheibel, 2004 ; Lewis et al., 1996 ; Prince et al., 1995 ; Mulinti et al., 2022 ; Malay et al., 2020 ; Teulé et al., 2007 ). Much in the same way that temperature downregulation mitigates toxicity of a recombinant construct and increases spidroin titer in E. coli ( Schmuck et al., 2021 ; Yang et al., 2016 ), tightly regulating basal expression through strain or plasmid engineering can further enhance bacterial production of recombinant spidroins ( Schmuck et al., 2021 ; Yang et al., 2016 ; Sørensen and Mortensen, 2005 ; Kawai et al., 2019 ; Mujacic et al., 1999 ). In cases where this is not feasible, forgoing an inducer and using longer culturing times may yield more favorable outcomes for strains or vector systems exhibiting basal expression of a recombinant spidroin or toxic protein. Furthermore, these methods may yield beneficial economic outcomes in scenarios where the relatively high cost of IPTG outweighs the cost of longer culturing times ( Edlund et al., 2018 ). Notwithstanding, restricting basal expression alone does not fully explain why the hybrid strain SoluBL21-pLysS performed better than either of its parent strains. The SoluBL21 strain was developed partly for toxic protein expression. Along with having a restriction of basal expression, sequencing of the strain reveals key mutations that differentiate it from BL21. We identified mutations on 47 genes in SoluBL21. Many of these mutations occur on uncharacterized proteins or on genes related to mannitol or glycerol metabolism, which are unlikely to affect silk production in LB media to our knowledge. However, there are 14 genes with mutations that are directly involved in stress responses in E. coli . These genes include yggW, yedY, yedW, yedY, speC, speB, uspC, hchA, loiP, mltC, envZ, ompR, yhgF, hupB . These genes play roles in stress responses pertaining to heat, reactive oxygen species, cell surface damage, salt changes, acid exposure, carbonyls, osmotic changes, putrescine production, nutrient starvation, ethanol exposure, radiation, and the SOS response ( Table S4 ) ( Nonaka et al., 2006 ; Urano et al., 2015 , 2017 ; Gennaris et al., 2015 ; Tabor and Tabor, 1985 ; Kurihara et al., 2005 ; Schneider et al., 2013 ; Hafner et al., 1979 ; Gustavsson et al., 2002 ; Mujacic et al., 2004 ; Subedi et al., 2011 ; Mujacic and Baneyx, 2006 , 2007 ; Weber et al., 2006 ; Huang et al., 2008 ; Hagiwara et al., 2003 ; Pomposiello et al., 2003 ; Tokishita et al., 1992 ; Chakraborty et al., 2017 ; Barbieri et al., 2013 ; Byrne et al., 2014 ; Goshima et al., 1990 ; Stojkova et al., 2019 ; Maslowska et al., 2019 ). Several of the mutations in SoluBL21 occur on genes responsible for extensive and broad stress pathway signaling within E. coli. This includes the envZ/ompR two-component system, in which a mutation on the envZ gene causes constant phosphorylation of the ompR transcriptional regulator. This results in a decreased repression of several stress response pathways, including those related to osmotic and acid stress ( Tokishita et al., 1992 ; Chakraborty et al., 2017 ; Barbieri et al., 2013 ). Furthermore, mutations in the speC and speB genes may alter putrescine production pathways, with putrescine production representing a fundamental way that several organisms, including E. coli , respond to a myriad of harmful conditions ( Tabor and Tabor, 1985 ; Kurihara et al., 2005 ; Schneider et al., 2013 ; Hafner et al., 1979 ). Likewise, the uspC (Universal stress protein C) gene is mutated, potentially promoting a more favorable cell phenotype in response to silk protein production, as uspC is induced by a diversity of stress factors that includes nutrient starvation (of multiple types), oxidative stress, DNA damage, radiation, heat shock, and ethanol exposure ( Gustavsson et al., 2002 ). There are also mutations in the DNA-binding region of hupB , a transcriptional factor that controls expression of 8% of the entire genome in regions where the genes are associated with adaptations to harsh environments, including the SOS response system, and oxidative and radiative stress systems ( Goshima et al., 1990 ; Stojkova et al., 2019 ). It stands to reason that these mutations promote a cell phenotype that is more tolerant to the expression of toxic and disordered proteins, and results in the favorable characteristics observed for strains SoluBL21 and SoluBL21-pLysS. Additionally, identification of these mutations provides previously unknown targets for continued strain optimization and supports the conclusion that strains which harbor targeted adjustments to their stress response genotype are necessary for improving recombinant spidroin production. 3.9 Addressing spidroin toxicity with targeted protein design Recent work in the recombinant silk field has initiated a deeper exploration into the non-repetitive terminal domains that are present in natural spidroins. These terminal domains, typically 100–200 amino acids in length, can promote solubility and controlled self-assembly of recombinant spidroins ( Andersson et al., 2017 ). Moreover, the highest reported titer for a recombinant silk protein in a bacterial system used a construct that included terminal domains flanking a repetitive core domain ( Schmuck et al., 2021 ). Previous work supports the idea that the terminal domains can dimerize and promote micelle formation among silk proteins, with the repetitive and potentially disordered core domain enveloped by more ordered terminal domains ( Römer and Scheibel, 2008 ; Schwarze et al., 2013 ). We hypothesized that this phenomenon may shield a recombinant host strain from the toxic effects of a disordered protein sequence by sequestering it into a compact micelle. Thus, we explored the use of terminal domains on increasing titers and decreasing toxicity in our system. We inserted cDNA copies of the terminal regions of L. hesperus (western black widow) MaSp1 dragline silk at either end of the A10 4mer gene to form the A10 4mer BWT construct ( Fig. S7 ). The 43 kDa A10 4mer BWT protein was readily produced and purified from the SoluBL21-pLysS strain at 96 (±12) mg/L, a titer that is approximately twice that of the similarly sized (50 kDa) A10 16mer which also contains an identical repeat sequence. Fig. 8 shows the growth of strain SoluBL21-pLysS over 10.25 h during induced expression of the A10 4mer and A10 4mer BWT constructs, with uninduced cultures serving as controls. Including the black widow termini increases the final OD600 versus the original A10 4mer by 110%. Unlike the original A10 4mer expressions, during expression of A10 4mer BWT, the strain exhibits a later stationary, and no death phase is observed over the 10.25 h. FTIR data on the constructs shows that there is a substantial shift in protein structure when the terminal domains are included, with A10 4mer BWT demonstrating an increase in alpha helices and a decrease in beta turns ( Fig. S8 ). Decreased disorder of A10 4mer BWT versus A10 4mer is supported by observation of more normal mobility in SDS PAGE, where the 43 kDA A10 4mer BWT shows an apparent molecular weight of 48 kDa while the 15 kDa A10 4mer shows an apparent weight of 27 kDa ( Figs. S1 and S9 ). Furthermore, computational disorder predictions for the A10 4mer BWT show large regions where individual residues have less than 10% probability of disorder, while the original A10 4mer construct is shown to have 90–100% probability of disorder at all residues ( Fig. S4 ). These findings support the hypothesis that including the terminal regions either decreases structural disorder and/or shields the disordered core domain from the host organism, facilitating increased biomass accumulation while maintaining a high titer. Thus, modulating primary sequence with spidroin terminal domains is another strategy that can be used to decrease toxicity and potentially increase titers of recombinant spidroins, while likely maintaining the construct's ability to exhibit silk-mimetic material properties ( Schmuck et al., 2021 ; Andersson et al., 2017 ). More broadly, these findings may provide valuable insight for producing toxic or disordered proteins in general. Particularly in situations where a toxic recombinant protein sequence cannot be altered due to requirements of a specific application, the fusion of flanking spidroin terminal domain sequences (which could be subsequently removed through the inclusion of cleavage sequences) provides a tool to mitigate toxicity. Fig. 8 Cell growth during expression of the A10 4mer (solid line – with induction (I), triangles – without induction (U), circles) and A10 4mer BWT (dashed line – with induction (I), triangles – without induction (U), circles) constructs with and without induction. For the induced curves, induction occurred with the addition of 1 mM IPTG at 1.75 h at an OD600 of 0.6–0.8. Data points on the curve represent the mean OD600 for that time point from three replicates. Fig. 8"
} | 13,850 |
35731877 | PMC9217086 | pmc | 322 | {
"abstract": "Biological synapses store multiple memories on top of each other in a palimpsest fashion and at different time scales. Palimpsest consolidation is facilitated by the interaction of hidden biochemical processes governing synaptic efficacy during varying lifetimes. This arrangement allows idle memories to be temporarily overwritten without being forgotten, while previously unseen memories are used in the short term. While embedded artificial intelligence can greatly benefit from this functionality, a practical demonstration in hardware is missing. Here, we show how the intrinsic properties of metal-oxide volatile memristors emulate the processes supporting biological palimpsest consolidation. Our memristive synapses exhibit an expanded doubled capacity and protect a consolidated memory while up to hundreds of uncorrelated short-term memories temporarily overwrite it, without requiring specialized instructions. We further demonstrate this technology in the context of visual working memory. This showcases how emerging memory technologies can efficiently expand the capabilities of artificial intelligence hardware toward more generalized learning memories.",
"introduction": "INTRODUCTION While neural networks in the cerebral cortex use an estimated 10 13 to 10 14 synapses to facilitate a plethora of cognitive abilities ( 1 , 2 ), their engineered counterparts require equivalent numbers of trainable parameters for a far narrower application spectrum ( 3 , 4 ). One candidate for explaining this discrepancy in learning capacity between biological and artificial intelligence (AI) suggests that synapses are able to consolidate multiple memories that can be revealed at different time scales—much like a palimpsest ( 5 ). Synapses can remember long-term plasticity events, namely, long-term potentiation (LTP) and long-term depression (LTD), while expressing altered states in the short term ( 6 ). This temporal partition enables the brain to use the same resources for multiple computation processes. The adoption of this flexibility by neuromorphic hardware is therefore a critical milestone toward the integration of AI in a wider range of on-the-edge, continuously-on learning systems. Palimpsest storage is realized biologically via the bidirectional interaction of hidden biochemical processes affecting the manifestation of synaptic efficacy at different time scales ( 5 ) after each memory modification. These processes are characterized by their own degrees of plasticity (i.e., learning rates) and lifetimes (i.e., “forgetting time constants”). Sparsely presented memories induce fast changes in synaptic efficacy, but these quickly decay to reveal older but more persistent memories that have successfully affected less plastic but more long-lasting processes. The coexistence of these processes allows synapses to be both plastic in the short term, enabling incoming memories to be written easily, and rigid in the long term, thus preserving old memories of validated significance. The flexibility promised by dynamic memory consolidation has naturally attracted the attention of AI hardware design and, particularly, that of memristive technologies, which have already showcased their potential in numerous neuromorphic applications ( 7 – 12 ). Memristor-based artificial synapses have demonstrated core plasticity functionality in the form of LTP/LTD. These implementations show how plasticity changes can become more pronounced in an analog regime when stimulation events are applied successively. These synaptic designs are largely based on phase-change memory (PCM) materials, which experience conductance changes when stimulation pulses are applied on them to emulate potentiation and depression. These designs achieve synaptic emulation, both by using standalone memristors ( 12 – 14 ) or by integrating them in more complex circuitry ( 15 , 16 ). While these studies have demonstrated the abilities of memristors to facilitate learning in artificial neural networks (ANNs), they have not considered how learned memories can be protected from continuous synaptic modifications—a crucial requirement for efficient online learning. Both PCM- and resistive random-access memory (RRAM)–based memristors have been used to implement metaplasticity, i.e., tuning of the learning rate ( 17 , 18 ), on complementary metal-oxide semiconductor–based artificial synapses in spiking neural networks ( 16 , 19 ). Metaplasticity has been studied extensively because of its potential for protecting consolidated memories via tunable learning rates. In a similar vein, nonvolatile RRAM synapses use explicitly modulated bias voltage to tune their switching (i.e., learning) rate ( 20 – 24 ). However, these studies have not been evaluated in the context of dynamic memory consolidation for two reasons. First, their implementation of variable learning rates occurs from appropriately tuned stimulation variation, implying that the need for plasticity rate changes is known a priori. Thus, they cannot operate in an online learning environment where the need for consolidation is usually unknown in real time. In addition, although these synaptic models showcase both LTP and LTD, they focus only on manipulating learning rates unidirectionally. This means that plasticity rates vary only within the context of stronger or lesser potentiation/depression independently, for instance, an already potentiated synapse experiencing lower plasticity rates toward further potentiation. Nevertheless, for metaplasticity to function properly, it is also imperative for a synapse to mitigate for catastrophic forgetting and protect its learned state against modifications in either direction concurrently ( 18 ). The protection of memory states has also been studied in the context of passive memory lifetime. Volatile RRAM has demonstrated short-term memory (STM) to long-term memory (LTM) transition where repeated presentations of the same memory induce longer changes in synaptic states, albeit being irreversible and unidirectional ( 10 , 25 – 27 ). This means that they have only worked in the context of LTP, where successive potentiation leads to memory states that are available for longer time windows. While this serves as a strong foundation toward using the time dynamics of volatile memristors, these implementations also ignore the protection of consolidated states when opposing synaptic modifications occur, and hence, the issue of catastrophic forgetting remains unresolved. Last, simulated ANNs based on metaplasticity principles have demonstrated an increase in specialized learning capacity ( 28 ). While the authors comment that palimpsest capabilities can expand capacity toward uncorrelated memories, now, neither commercially available nor emerging memristive technologies have exhibited the necessary properties required for dynamic memory consolidation. In this work, we built upon the studies that are mentioned above to bridge synaptic plasticity with automatic consolidation and memory protection, irrespective of plasticity direction. The characteristics of RRAM volatility ( 29 , 30 ) are exploited to emulate the function of the hidden biochemical processes that enable palimpsest consolidation. We harness the bidirectional volatile and nonvolatile responses of RRAM devices to practically realize two consolidation time scales in one device, effectively storing competing binary states in a single synapse with doubled STM and LTM capacity. Then, we upscale this principle to consolidate memory traces at variable consolidation intensity. Our technology can protect a strong memory in its long-term storage while allowing multiple short-term signals to take over the STM fleetingly and then quickly decay. Simple metaplasticity is also realized as a natural subset behavior when a given memory is consolidated consistently. Last, we show how this palimpsest memory can operate in a visual system where it also boasts unsupervised denoising abilities. Our memory system is unique in a number of key attributes. First, its expanded capacity is independent of the correlation between memories presented to it and even performs under fully destructive interference. Moreover, it can automatically sacrifice palimpsest capabilities for even stronger memory protection. These features unfold naturally as a result of single memristive device properties and do not require special biasing regimes or otherwise increased operational complexity.",
"discussion": "DISCUSSION In this work, we focused on binary synapses, which are known to support adequate learning in mathematical models ( 35 ) and deep-learning algorithms ( 28 ). The weight of the synapse is a binarized version of its resistive state, and the interplay between intense bidirectional volatility and small nonvolatile residues underlies its palimpsest capability. This concept can be naturally extended to synaptic weights of higher resolution. Also, we would like to point out that, beyond the possible extension to higher efficacy resolution, further improvement toward the wider adoption of this technology can be realized in two key areas. First, systematic studies for improving the fabrication uniformity of memristive synapses would be of high interest, particularly in the scope of large-scale hardware demonstration. Second, while we have prioritized conceptual clarity of results over writing speed and energy efficiency, these parameters are crucial milestones before the integration of this technology in real-world online learning applications. For these reasons, it is clear that future implementations of this work are by no means limited to our selected TiO 2 technology. As long as a candidate technology exhibits bidirectional volatility, it could then be evaluated according to application-specific needs, e.g., write speed, retention time, energy efficiency, etc. Nevertheless, the scope of this study has been strictly focused to the conceptual derisking of palimpsest synapses, which boast very interesting properties in several areas. We note that unidirectional volatility is already sufficient to support the transition from STM to LTM ( 25 – 27 ). However, this work differentiates that consolidated memories are also protected from competing memory signals, something that was overlooked by previous studies. Moreover, our synapses’ absolute capacity is effectively doubled, and palimpsest functionality has thoroughly been evaluated both in hardware and simulation demonstrations (see table S3 for a detailed comparison with memristive synaptic implementations). These features can be directly attributed to the bidirectional nature of our RRAM technology. It should be noted that bidirectional volatility in these devices has already been characterized for observation windows of up to 2 min ( 29 ), which could practically extend the memory lifetime of our synapses. Another remark about this palimpsest memory is that the contents of the memory are in general imperfect reflections of the desired memory. This is not unusual per se since neuro-inspired systems work on the basis of imperfect information typically by default (classifiers sort noisy inputs into neat classes), but, in palimpsest memories, we have the additional factor of LTM-STM relations to consider. The capabilities of this technology can be interpreted in several distinct ways. First, the palimpsest network can be evaluated in its capacity to recall multiple memories concurrently. While acceptable recall accuracy levels are relative with respect to application-specific needs, the absolute capacity of the network is tied to the number of available time scales in the memristive synapses. Here, nonconsolidated memories can only access the short-term network slot, and thus consecutive STMs interfere destructively with each other. Nevertheless, the correlation statistics of incoming memory streams play a decisive role in the degradation of old signals. To that extent, applications that can afford more noisy recollections are also able to recall a consolidated LTM and multiple random STMs with a mean 50% correlation simultaneously, as illustrated in Fig. 3A . While simple metaplasticity can also suffice for generalizing over multiple highly correlated memories [see ( 28 )], the advantage of our technology arises from the ability to remember consolidated states even when the attention of working memory falls on uncorrelated signals. An intuitive representation of at least two palimpsest memories coexisting in the system can be seen in Fig. 4C at snapshots T ∈ [0.3–3] s and T ∈ [10.1–11] s. In that scenario, memory degradation after recovery is much weaker. Further expanding the consolidation capacity and the initial signal overlap of the network will require manipulating the switching and relaxation dynamics of the memristive synapses such that they operate more flexibly in a proportionally larger number of time scales. This investigation in material science and a resulting more complex electrochemical device structure are certainly of great interest. Moreover, this technology bears some interesting similarities to how real estate is used for multiple storage in visual working memory systems ( 34 , 36 – 38 ). Our results within the context of a visual working memory encapsulate best its relevant capabilities. As it is evident from Fig. 4 , palimpsest operation may not necessarily need to expand absolute memory capacity to provide computational advantages. Contrarily, it can be enabled using a neural network flexibly without suffering the cost of forgetting older but consolidated signals. This flexibility and LTM reconstructive ability can enhance the performance of in-memory computing ( 12 , 39 , 40 ) where systems are required to adapt quickly to incoming stimuli and is thus of direct relevance to neuro-inspired applications. In these scenarios, systems benefiting from palimpsest functionality shift their resources on spontaneous online tasks while retaining a core consolidated functionality. As shown in Fig. 3 (B and D) , this can occur for hundreds of uncorrelated short-term signals without explicit needs for reinforcing the consolidated counterpart. Last but not the least, the short-lived span of overlapping memories resembles short-term attention mechanisms, which have recently shown promise toward more complex AI algorithms ( 41 ). Attention mechanisms can also be implemented using the high-capacity STM familiarity filters that are exhibited here (a familiarity filter is a memory that recognizes when a memory input is present inside the memory even if it no longer has enough information to reconstruct the memory). Illustrated in Fig. 3A , at least 50 uncorrelated memories can pass the familiarity filter simultaneously. Our memory also implements unsupervised (LTM) memory reconstruction in hardware, supporting previously linked theories of consolidation ( 5 , 18 ) and optimal recall in the CA3 area of the hippocampus ( 42 ). This partition of memory storage is an advantageous adaptation since only information that is relevant to a specific cognitive task is needed for undergoing the said task. The dual temporal capacity that is exhibited by our devices resembles the bistable switching that is known to govern synaptic plasticity ( 31 ). Specifically, the accumulation of nonvolatile residues after LTP/LTD can be thought of as an equivalent mechanism to calcium/calmodulin-dependent protein kinase II, which is considered to be a primary molecular memory mechanism ( 43 , 44 )."
} | 3,885 |
39528499 | PMC11554781 | pmc | 323 | {
"abstract": "Human skin is essential for perception, encompassing haptic, thermal, proprioceptive, and pain-sensing functions through ion movement. Additionally, it is mechanically resilient and self-healing for protection. Inspired by these unique properties, researchers have attempted to develop stretchable, self-healing sensors based on ion dynamics. However, most self-healing sensors reported to date suffer from low fracture strength and toughness. In this work, we present an ion-based self-healing electronic skin with exceptionally high fracture strength and toughness. We enhanced self-healing polymers and ionic conductors by introducing two types of orthogonal dynamic crosslinking bonds: dynamic aromatic disulfide bonds and 2-ureido-4-pyrimidone moieties. These dynamic bonds provide autonomous self-healing and high mechanical toughness even in the presence of ionic liquids. As a result, our self-healing polymer and self-healing ionic conductor exhibit remarkable stretchability (700%, 850%), fracture strength (34 MPa, 30 MPa), and toughness (78.5 MJ/m 3 , 87.3 MJ/m 3 ), the highest values reported among self-healing ionic conductors to date. Using our materials, we developed various fully self-healing sensors and a soft gripper capable of autonomously recovering from mechanical damage. By integrating these components, we created a comprehensive self-healing electronic skin suitable for soft robotics applications.",
"introduction": "Introduction The human skin forms an integral part of the human body. The skin is not merely an external protective barrier; it is equipped with a highly sophisticated somatosensory system based on ion movements 1 – 5 . This system endows the skin with the ability to detect and interpret various stimuli, including temperature, pressure, and pain. Furthermore, the skin possesses remarkable mechanical robustness and even self-healing capability, which can autonomously restore the skin’s inherent functions when damage is applied. Researchers have increasingly focused on the development of stretchable, self-healing sensors that mimic the remarkable properties of human skin 6 – 9 . These stretchable and self-healing sensing materials are fabricated by incorporating either electrically conductive or ionically conductive components into a self-healing polymer matrix. Self-healing capabilities could enhance the durability and lifetime of the sensors, while the electrical characteristics of the conductive fillers enable the detection of various stimuli. Among the numerous materials for self-healing sensors, self-healing ionogels are excellent candidates 10 – 16 . Self-healing ionogels-that is, self-healing polymer networks swollen with ionic liquids (ILs)—have been extensively studied as ionic conductors for soft self-healing electronics due to their humidity insensitiveness, non-volatility, mechanical stretchability, and excellent electrochemical properties. They have been successfully used as a sensing layer for self-healing mechanical sensors, including pressure, strain, and shear sensors, as well as temperature sensors. Additionally, they have been utilized as electrodes in stretchable light-emitting capacitors (LEC). Nevertheless, if ionic conductors possess weak mechanical properties, they may experience permanent dimensional changes (plastic deformation) during repeated sensor operations. Such changes significantly impede accurate sensing. Therefore, improving the mechanical robustness of self-healing ionic conductors is very important for electronic skin. However, reported self-healing ionic conductors have low mechanical properties for the following reasons. The intrinsic self-healing capability of self-healing polymers is typically achieved by incorporating dynamic supramolecular interactions, such as hydrogen bonding, π- π interactions, ionic interactions, and metal-ligand coordination, into low T g polymer matrices 17 , 18 . Due to their composition of dynamic bonds with weak bindings, most self-healing polymers exhibit low mechanical properties. Fortunately, several studies have been reported to design and synthesize tough, self-healing polymers trying to solve the trade-off relationship between self-healing capabilities and the mechanical properties of materials 19 – 23 . However, the self-healing ionic conductors designed for stretchable self-healing sensors exhibit significantly weaker mechanical properties when compared to their original self-healing polymers. This is attributed to the disruptive nature of ionic liquid on the dynamic bonds. For these reasons, no studies have developed an approach for simultaneously realizing stretchable, mechanically tough, and self-healing ionic conductors for stretchable self-healing sensors to the best of our knowledge. Here, we report a design of a self-healing polymer and ionic conductor, possessing exceptionally ultra-high fracture strength and toughness, for stretchable self-healing sensors. The designed polymer contains two types of dynamic bonds in its polycaprolactone (PCL) main polymer chain, which is known to be biocompatible: dynamic aromatic disulfide bonds (DS) and 2-ureido-4-pyrimidone (UPy) moieties, having the role of autonomous self-healing and high mechanical toughness, respectively. DS was selected to enable efficient self-healing at room temperature through main chain shuffling. Among various self-healing moieties, DS moieties are particularly appealing due to their capability of facilitating relatively faster self-healing at room temperature through efficient disulfide metathesis 24 – 26 . Next, UPy units were introduced as additional dynamic bonds in polymer design to enhance the mechanical properties. The UPy moieties impart good elasticity and fracture toughness by crosslinking the polymer chains through quadruple hydrogen bonding 27 , 28 . Significantly, due to the formation of strong bonds between UPy units, even when ionic liquids are integrated into the polymer matrix, the robust bonding between UPy units is not broken. Consequently, these UPy moieties ensure that ionic conductors maintain their robust mechanical properties in comparison to their original polymers. However, if UPy bonds are solely introduced as dynamic bonds in the polymer chain, self-healing cannot be achieved in polymer and ionogel due to the slow bond exchanges of UPy (Fig. 1a ). In contrast, if DS bonds are solely employed as dynamic bonds, the self-healing polymer (SHP) exhibit weak mechanical properties due to the low bonding strength between DS units (Fig. 1b ). Moreover, when ionic liquid is added to SHP, the ionic liquid disrupts the polymer interaction resulting in significantly weakened mechanical toughness (Fig. 1b ). Only when both dynamic bonds are used together, ionic conductor can achieve ultra-high toughness and self-healing property simultaneously (Fig. 1c ). Interestingly, all components of PCL, DS, and UPy in our polymer design are easily aligned and further aggregated during stretching. As a result, this can enhance the fracture strength through strain-induced aggregation. As a result, our self-healing ionic conductor (SHIC) shows high stretchability (850%), high fracture strength (30 MPa), and high toughness (87.3 MJ/m 3 ). To the best of our knowledge, our SHIC shows the highest mechanical properties among reported self-healing ionogels. Fig. 1 Design of tough, self-healable ionic conductor. a Schematic illustration of a polymer featuring UPy-urea as the sole dynamic bond. When using UPy-urea as the sole dynamic bond in a polymer, the strong hydrogen bonds between UPy-urea units result in high mechanical properties. However, due to the strong hydrogen bonds, the chain mobility is significantly reduced, greatly diminishing the polymer’s self-healing ability. b Schematic illustration of a polymer featuring aromatic disulfide as the sole dynamic bond. When using aromatic disulfide as the sole dynamic bond in a polymer, the polymer enables efficient self-healing at room temperature through disulfide exchange. However, when an ionic liquid is added to the polymer as an ionic conductor, it disrupts the interaction between polymer chains, resulting in significantly lower mechanical properties compared to the polymer alone. c Schematic illustration of a polymer featuring UPy-urea and aromatic disulfide as the dynamic bonds. In the case of disulfide exchange, self-healing occurs through the breaking and reforming of dynamic covalent bonds, which makes the impact of chain rigidity on self-healing efficiency relatively low. Additionally, due to the strong quadruple hydrogen bonding between UPy-urea units, the ionic liquid cannot disrupt these strong bonds, allowing the ionic conductor to maintain its high mechanical properties even with the addition of the ionic liquid. Consequently, when using both UPy-urea and aromatic disulfide as the dynamic bonds in a polymer, it is possible to design an ionic conductor that possesses high mechanical properties and enables self-healing at room temperature. Using our SHP and SHIC, we uniquely demonstrate tough, self-healing electronic skin and a soft gripper for somatosensitive soft robots. Due to their excellent mechanical, sensing, and self-healing properties, these soft robots can operate effectively for extended periods, even in dynamic environments.",
"discussion": "Discussion By incorporating two types of dynamic bonds into the polymer system, we have developed an elastomer and ionic conductor that possesses both excellent mechanical properties and efficient self-healing capability at room temperature. The self-healing capability was imparted by introducing an aromatic disulfide moiety, while a UPy moiety was added to maintain strong mechanical properties even with the incorporation of an ionic liquid. Only when both dynamic bonds are used together in the proper ratio can the ionic conductor achieve ultra-high toughness and self-healing properties simultaneously. We have developed various sensors for fully self-healable electronic skin and actuators, which can be applied to self-healing somatosensitive soft robotic systems. We believe that our material design strategy will significantly advance the progress of self-healing soft electronics."
} | 2,555 |
39747916 | PMC11695864 | pmc | 324 | {
"abstract": "Recent experimental studies in the awake brain have identified a rule for synaptic plasticity that is instrumental for the instantaneous creation of memory traces in area CA1 of the mammalian brain: Behavioral Time scale Synaptic Plasticity. This one-shot learning rule differs in five essential aspects from previously considered plasticity mechanisms. We introduce a transparent model for the core function of this learning rule and establish a theory that enables a principled understanding of the system of memory traces that it creates. Theoretical predictions and numerical simulations show that our model is able to create a functionally powerful content-addressable memory without the need for high-resolution synaptic weights. Furthermore, it reproduces the repulsion effect of human memory, whereby traces for similar memory items are pulled apart to enable differential downstream processing. Altogether, our results create a link between synaptic plasticity in area CA1 of the hippocampus and its network function. They also provide a promising approach for implementing content-addressable memory with on-chip learning capability in highly energy-efficient crossbar arrays of memristors.",
"introduction": "Introduction The brain has to solve a really challenging algorithmic problem when it stores episodic memories: It has to allocate on-the-fly neurons that form a trace or tag for a new experience that can subsequently be reactivated with partial cues. In other words, it needs a mechanism for creating a content-addressable memory (CAM) through one-shot learning. Recent experimental work has elucidated an essential ingredient of that: Behavioral time scale synaptic plasticity (BTSP) 1 – 3 . BTSP differs strongly from plasticity rules such as the Hebb rule and STDP (spike-timing-dependent plasticity) that have previously been considered in efforts to model the creation of memory traces in the brain. This is especially salient because in contrast to most experimental data on synaptic plasticity 4 , BTSP has been validated through experiments in awake and behaving animals. Other converging experimental data suggest that area CA1 of the hippocampus, where BTSP has been demonstrated, is a brain area that is central for the creation of memory traces for episodic and conjunctive memories 5 . BTSP differs in the following aspects from traditionally studied plasticity rules: BTSP does not depend on the firing of the postsynaptic neuron. Instead, it is gated by synaptic input from another brain area, the entorhinal cortex (EC). These gating signals appear to be largely stochastic 6 . BTSP does not require dozens of repetitions of a protocol for the induction of synaptic plasticity, but is effective in a single or few trials, i.e., it provides a mechanism for one-shot learning 1 . The direction in which BTSP changes a synaptic weight depends primarily on the preceding weight value 3 . BTSP acts on the time scale of seconds, a time scale that is suitable for creating episodic and other forms of conjunctive memories that integrate temporally dispersed information. A mathematical model for BTSP and its impact on the induction of place cells was provided by ref. 3 . There, and in the subsequent modelling studies 7 , 8 for the induction of place cells, BTSP was modelled through a differential equation with continuous time and weights but without noise. However, the experimental data for BTSP, see Fig. 3 C in ref. 3 , indicate a substantial amount of trial-to-trial variability of weight changes. Hence, the question arises whether the essence of experimentally observed BTSP can also be modelled by a simple stochastic rule for BTSP with binary weights. The first rule for BTSP with binary weights had been proposed in a preliminary version 9 of this work. Subsequently, a similar rule was also proposed as a model for new experimental data for BTSP in area CA3 10 . We will examine here the system of memory traces in area CA1 that is created by the rule 9 and its stochastic variant. Since these rules were chosen to be simple, this analysis can be carried out not only through numerical simulations but also analytically. In this way, one can even analyze the resulting memory capacity at the scale of the brain. We find that BTSP creates through one-shot learning a high-capacity CAM with just binary weights. This is remarkable since the most commonly studied CAM model, the Hopfield network (HFN), requires a number of weight values that grows linearly with the number of memory items (one often refers to this case as “continuous weights\"). HFNs have much lower memory capacity when continuous weights are rounded to binary weights 11 , and we are not aware of online learning rules for HFNs with binary weights. The brain employs a sparse coding regime, where relatively few neurons are simultaneously active. This sparse coding regime contributes to its astoundingly low energy consumption and is, therefore, also desirable for neuromorphic hardware. Consequently, we focus our investigation on CAM for memory items that are represented through brain-like sparse activity, including the case of overlapping memory items. Additionally, the human brain can not only create different memory traces for similar memory items, but memory traces for similar memory items exhibit a repulsion effect 12 – 14 . This effect, which could so far not be reproduced by HFNs or other learning rules, pulls memory traces for similar memory items actively apart, thereby enabling differential downstream processing. We show that BTSP can reproduce this repulsion effect of memory traces in the human brain. An especially interesting facet of experimental data on BTSP is its dependence on stochastic input signals from area EC that cause plateau potentials in CA1 neurons 6 . We show through theoretical analysis and numerical simulations that the probability of the occurrence of such a plateau potential is a critical parameter for the performance of resulting memory systems. It has no analogue in previous memory models or rules for synaptic plasticity. Our analysis suggests that the experimentally observed value of this parameter is close to optimal for the quality of the memory system that BTSP creates, and that the several seconds long time window of plasticity that is opened by a plateau potential plays a central role in bringing it into this range. The creation of memory traces through BTSP can be seen as an expansion of another well-known biological algorithm that also relies on stochasticity, the fly algorithm 15 . The fly algorithm also employs binary weights, but these are assumed to exist a-priori; they are not learnt. In contrast, BTSP produces a random projection from input patterns to memory traces through learning. In other words, the random projection that BTSP creates is custom-made for a particular ensemble of input patterns and can be continually extended through one-shot learning. We demonstrate that this difference from the fly algorithm, and from random projections in general, enables BTSP to create attractors around stored memory traces that support stable recall from partial cues. In addition, BTSP manages to overcome two bottlenecks that the fly algorithm faces when one wants to apply it to larger neural systems: It neither requires a projection of input patterns into a 40 times larger neural population, nor a global winner-take-all competition over the population of neurons into which input patterns are projected. These two requirements are met by the relatively small olfactory system of the fly but not by the memory system of the mammalian brain. The number of neurons in area CA1, the area into which memory items are projected from area CA3, is not 40 times larger but at most 1.5 times larger than area CA3, both in the rodent and in the human brain (see Table 3-1 16 ). Furthermore, area CA1 consists of 0.39 to 14 million neurons in the mammalian brain 16 , and the assumption of a winner-take-all computation is more problematic for such a large system. Hence, evolution had to invent a different method for storing large numbers of memory traces in the mammalian hippocampus: BTSP. Since BTSP is a local synaptic plasticity rule that requires in its simplified form only two weight values, it is especially suited for on-chip learning in neuromorphic hardware, in particular for online creation and expansion of CAM in crossbar arrays of memristors with just two required resistance states. Hence, this BTSP-based CAM paradigm provides a a new approach for creating highly energy-efficient large-scale implementations of CAMs for brain-like sparse patterns. This approach can drastically expand the memory capacity of already existing implementations of CAM in crossbar arrays of memristors that were based on the HFN paradigm 17 – 19 . We review experimental data on BTSP in the next section and show that a simple plasticity rule with binary weights provides a good approximation to these data for the case that the arrival times of synaptic inputs are statistically independent from the arrival times of plateau potentials. We then address the main new parameter of this rule, the probability of a stochastic gating signal, and elucidate its functional impact. In the subsequent section, we compare the properties of the system of memory traces created through BTSP with the memory system created through a random projection, which is the main component of the fly algorithm. Subsequently, we present theory and numerical simulation results for the recall of memory traces for overlapping memory items for BTSP. We then compare the properties of the CAMs that are created through BTSP with feedback connections with properties of CAMs that are implemented by HFNs, both for the case of binary and continuous-valued synaptic weights. Finally, we show that BTSP reproduces the repulsion effect of the human brain and elucidate parameters of BTSP that are critical for this.",
"discussion": "Discussion We have created and examined a simple rule for BTSP, the recently discovered one-shot learning rule in the brain 1 – 3 , see Fig. 1 D and E. Whereas previous models for BTSP aimed at reproducing experimental data from mice during navigation in mazes 7 , 8 , we have focused here on functional implications of the simple BTSP rule for the creation of memory systems for large numbers of generic sparsely encoded memory items. More detailed BTSP data and rules with continuous weights can be reproduced by this simple rule when multiple synaptic connections are taken into account, see Fig. 1 F and G. Importantly, this BTSP rule is sufficiently simple so that its functional impact can be studied analytically. One of the most surprising and intriguing features of BTSP is that it does not rely on postsynaptic neural activity, like STDP, Hebbian rules, and virtually all other synaptic plasticity rules that have been studied in the context of memory models or elsewhere in theoretical neuroscience. Instead, it relies on external stochastic gating signals that open the gate for synaptic plasticity in a neuron for several seconds. A generic functional advantage of this mechanism is that it spreads the “memory load\" uniformly over all neurons in the network, thereby avoiding degradation of previously generated memory traces, see Fig. 5 E. Because of the reliance of BTSP on stochastic gating signals, the probability of the occurrence of these gating signals, i.e., the parameter f q , is an essential parameter for BTSP that has no analogue in previously considered learning rules. We have analyzed the functional impact of f q , both theoretically and through numerical simulations, see Figs. 2 C, D, E, and 6 C, and Fig. S8 . Our results suggest that the experimentally determined value f q = 0.005 can be seen as the sweet spot for combining several functionally attractive features of BTSP. We have shown that the resulting predictions agree very well with the results of numerical simulations. It also reproduces the primary functional impact of BTSP that had been reported in ref. 1 : Instantaneous creation of place fields in a CA1 neuron as a result of a plateau potential (Fig. 1 D). Furthermore, our BTSP rule reproduces an effect that was highlighted in the study 1 as the hallmark of its plasticity window on the time scale of seconds, rather than the time scale of milliseconds that is characteristic for STDP: A substantial increase in the spatial extension of the place field in the case of higher running speed during plasticity induction, see Fig. 1 E. In other words, the CA1 neuron is induced to fire not only for the current spatial location at the onset of a plateau potential, but also for locations that were traversed several seconds before and after that onset. If one views the spatial locations that are traversed during this seconds-long plasticity window as frames of an episodic memory, one sees that BTSP can create memory traces for episodes that are several seconds long. Another functional impact of the seconds-long plasticity window of BTSP is that it makes the traces for such memory traces in area CA1 sufficiently large: It enables synaptic plasticity during an episode in sufficiently many CA1 neurons (see Fig. 2 D for a theoretical analysis), in spite of the experimentally found very low rate of plateau potentials 0.0005 per CA1 neuron per second 6 . In other words, it brings the probability f q that a presynaptic input occurs within the plasticity window that is opened by a plateau potential into a range that is especially suited for creating functionally useful memory traces. Furthermore, online learning with BTSP also creates well-working CAM with binary weights if the memory items are overlapping. Hence, this CAM can also be used to store episodic memories or conjunctive memories that have common components. Finally, we have shown in Fig. 6 that BTSP reproduces the repulsion effect of the human brain that pulls memory traces for similar memory items apart, as documented in refs. 12 – 14 , 32 . This effect can be seen as a vital component of higher cognitive function since it enables differential processing of experiences that differ only in a few but possibly salient features. However, the repulsion effect could not be reproduced with previously considered learning rules. On a more abstract level, our results show that the stochastic gating of BTSP can be seen as a method for porting some functionally attractive properties of the fly algorithm 15 into a learning system. More precisely, BTSP embeds these advantages of stochasticity into a learning mechanism that tailors the resulting “random projection\" to the particular ensemble of memory items that are to be stored. We have shown that this makes BTSP superior to the fly algorithm and random projections as methods for neural coding of input patterns because BTSP induces powerful attractor properties for the memory traces that it creates, see Figs. 3 , 4 , 5 . Another biological mechanism for one-shot creation of memory traces in a biological organism had been described in Imam and Cleland (2020) 33 for the mammalian olfactory bulb. This work has in common with our approach that it also provides one-shot insertion of memory items into a neural network that supports recall with noisy cues. Furthermore, they consider a spiking neural network implementation. Spike-based implementations of HFNs, where the timing of spikes conveys analog information, had already been introduced by ref. 34 . In principle, this model 34 also allowed one-shot learning, because the standard learning rule for HFNs is a one-shot learning rule. But both the HFN model and Imam and Cleland (2020) require that synaptic weights can assume a large number of values, whereas our BTSP-based model requires only binary weight values. Another difference to Imam and Cleland’s model 33 is that in their approach new neurons have to be added to the networks when new memory items are to be learnt. In contrast, our memory network can have a fixed architecture. Additionally, in terms of learning rules, two different mechanisms are employed in their model: A simplified STDP rule for synaptic weights and a less standard rule for modifying the blocking period for firing that is caused by a spike of the inhibitory GC neuron in the postsynaptic MC cell. Since they add new GC neurons whenever a new memory item is to be learnt, their model is likely to avoid the degradation of earlier memory traces that typically occurs with Hebbian-like plasticity rules for one-shot learning. In contrast, the stochastic nature of gating signals for BTSP, in conjunction with sparse coding, reduces the chance that previously created memory traces are overwritten, see our explanations for Fig. 5 E. Our BTSP rule is, to the best of our knowledge, the first online learning rule that is able to create a functionally powerful CAM with binary synaptic weights. Furthermore, updates of the CAM can be carried out through a simple one-shot learning rule that is within reach for on-chip learning. Hence, the creation of memory systems through BTSP provides an attractive alternative to HFNs for modelling in computational neuroscience and for creating CAMs in neuromorphic hardware. HFNs require weights whose number of distinguishable values grows linearly with the number of memory items that are stored. Results of numerical simulations in Fig. S9 confirm theoretical predictions from ref. 11 that the memory capacity of HFNs drops drastically if weights are constrained to binary values. But both the number of different values that a synaptic weight in a biological neural network 35 , and the number of distinguishable resistance states that a memristor can assume are severely limited. Furthermore, memristors whose resistance just has to assume a small number of values are substantially easier to fabricate and take less space on the chip. Therefore, CAMs that require just binary synapses are especially attractive from this perspective. Previous neuromorphic implementations of CAM were based on the HFN paradigm, and could therefore store only a very small number of memories 17 – 19 . Hence, CAMs based on BTSP are likely to substantially advance the state of the art for neuromorphic CAM. A substantially more ambitious goal would be to create CAMs with superlinear, or even exponential capacity in neuromorphic hardware. Modern Hopfield networks and dense associative memories 36 , 37 as well as bipartite Hopfield networks 38 provide interesting theoretical models for that. However, a closer look shows that the latter is not able to store externally given memory items, and that the former requires, like the classical HFN, synaptic weights that can assume a very large number of distinguishable values. Furthermore, the memory capacity of modern Hopfield networks and dense associative memory networks remains linear in the network size according to ref. 37 , unless one moves to networks with higher order interactions among neurons, i.e., if one goes beyond pairwise interaction of neurons via synapses. Obviously, this will require a completely new technology in order to create an energy-efficient neuromorphic implementation of such a network. This work suggests an alternative research direction, where new insight from the brain paves the road for the design of substantially more energy-efficient content-addressable memory systems. These require just binary weights, and can be updated through one-shot on-chip synaptic plasticity."
} | 4,889 |
33222325 | PMC8243963 | pmc | 326 | {
"abstract": "Abstract Coral bleaching is the single largest global threat to coral reefs worldwide. Integrating the diverse body of work on coral bleaching is critical to understanding and combating this global problem. Yet investigating the drivers, patterns, and processes of coral bleaching poses a major challenge. A recent review of published experiments revealed a wide range of experimental variables used across studies. Such a wide range of approaches enhances discovery, but without full transparency in the experimental and analytical methods used, can also make comparisons among studies challenging. To increase comparability but not stifle innovation, we propose a common framework for coral bleaching experiments that includes consideration of coral provenance, experimental conditions, and husbandry. For example, reporting the number of genets used, collection site conditions, the experimental temperature offset(s) from the maximum monthly mean (MMM) of the collection site, experimental light conditions, flow, and the feeding regime will greatly facilitate comparability across studies. Similarly, quantifying common response variables of endosymbiont (Symbiodiniaceae) and holobiont phenotypes (i.e., color, chlorophyll, endosymbiont cell density, mortality, and skeletal growth) could further facilitate cross‐study comparisons. While no single bleaching experiment can provide the data necessary to determine global coral responses of all corals to current and future ocean warming, linking studies through a common framework as outlined here, would help increase comparability among experiments, facilitate synthetic insights into the causes and underlying mechanisms of coral bleaching, and reveal unique bleaching responses among genets, species, and regions. Such a collaborative framework that fosters transparency in methods used would strengthen comparisons among studies that can help inform coral reef management and facilitate conservation strategies to mitigate coral bleaching worldwide.",
"conclusion": "Conclusions The common framework for coral bleaching experiments outlined in this paper provides some insights and suggestions that could help increase comparability among coral bleaching experiments. We recognize that studies are driven by specific research questions that may differ in scope or have requirements that are outside the framework parameters outlined here. Nevertheless, it is our hope that the common framework discussed here will encourage researchers to consider measuring and reporting more of the physicochemical conditions and variables (Table 1 ), better appreciate the value of reporting all of the relevant metadata (Table 2 ), and perhaps incorporate new analytical techniques or approaches in their research (see Appendix S1 ). The broad adoption of a common framework for coral bleaching experiments would increase the comparability of studies and enhance collaboration, which would have the net effect of increasing the efficacy and creativity of coral bleaching research. As coral reefs continue to change globally, every effort we can make to accelerate the pace of discovery will bring us that much closer to innovative solutions for protecting and restoring coral reefs.",
"introduction": "Introduction Temperature stress from ocean warming due to climate change is now the single largest threat to coral reefs globally (Veron et al. 2009 , Cantin et al. 2010 , Frieler et al. 2012 , Hughes et al. 2018 ). Reef ecosystems are experiencing unprecedented declines in coral colony abundance, coral diversity, and reef growth as a result of temperature‐induced coral bleaching, a phenomenon that is becoming more frequent and severe (e.g., Hoegh‐Guldberg et al. 2007 , Eakin et al. 2009 , Veron et al. 2009 , Hoegh‐Guldberg 2011 ). By the end of this century, tropical seawater temperatures are expected to rise by 1°C–3°C (IPCC 2013 ), and severe bleaching is expected to occur annually in some regions by 2030 and globally by 2055 (van Hooidonk et al. 2014 ). Coral bleaching is the visual manifestation of the breakdown in the symbiosis between the coral host and its endosymbiotic dinoflagellates (family Symbiodiniaceae; LaJeunesse et al. 2018 ) whereby the coral loses its endosymbiotic algae or pigments resulting in a pale or “bleached” appearance. Bleaching results in decreased coral health, growth, and reproductive output, as well as increased coral susceptibility to disease and mortality (e.g., Brown 1997 , Hoegh‐Guldberg 1999 , Omori et al. 1999 , Buddemeier et al. 2004 , Jokiel 2004 , Maynard et al. 2015 ). Despite the wide impact of bleaching events, the magnitude and extent of bleaching can vary substantially across scales, ranging from the individual colony to the ocean basin (e.g., Rowan et al. 1997 , Fitt et al. 2000 , Loya et al. 2001 , Grottoli et al. 2006 , 2014 , Palumbi et al. 2014 , Muller et al. 2018 , Morikawa and Palumbi 2019 ). Although it is well documented that temperature and irradiance are key drivers of coral bleaching, the processes causing broad variation in bleaching susceptibility and recovery across reefs, corals, and colonies are not fully resolved. Manipulative experiments remain a critical tool for elucidating the underlying mechanisms and responses of corals to thermal stress (McLachlan et al. 2020 ). However, few studies conduct detailed comparisons of results across data sets because it is not always straightforward to ascertain whether the variation in bleaching and recovery responses are due to (1) differences in experimental design (e.g., differences in light, baseline temperature, rate of temperature increase, experimental duration, etc.), (2) differences in bleaching and recovery measurements, (3) differences in coral biology, or (4) some combination of these differences. A detailed review of coral bleaching experiments by McLachlan et al. ( 2020 ) revealed that many important details about how experiments are designed and executed are sometimes missing from published papers, making comparisons between studies challenging. For example, knowing experimental heating temperature, heating duration, and lighting conditions are essential for cross‐study comparisons because all three variables can influence coral bleaching responses. In addition, some bleaching studies use a heat‐hold or heat‐pulse strategy of heating that mimics daily heat stress over a mid‐day low tide (Oliver and Palumbi 2011 ), whereas others mimic the onset and duration of a natural reef‐wide bleaching event with gradual increases in temperature and prolonged temperature exposure (Rodrigues and Grottoli 2007 ). Whether corals are exposed to pulse or gradual exposure may influence responses (Mayfield et al. 2013 b \n ). Therefore, clear reporting of experimental details and results is necessary for meaningful comparisons among studies (Gerstner et al. 2017 ) and for reliably identifying patterns in coral bleaching and recovery across species, habitats, reefs, and regions. One way to increase comparability and transparency among ongoing and future coral bleaching studies is to develop a common framework for reporting the conditions and results of coral bleaching experiments, while neither being overly prescriptive nor diminishing scientific innovation. A common framework for coral bleaching should include consideration of coral provenance, experimental conditions, and husbandry. Similar approaches have been successful in advancing other fields (e.g., ocean acidification research; Riebesell et al. 2010 , Cornwall and Hurd 2015 ), while also allowing for the rapid development of creative approaches to understanding underlying mechanisms. Doing so for experimental coral bleaching research will markedly improve our ability to detect important trends, identify species vulnerabilities and tolerances, and help coral researchers and managers devise solutions for coral persistence over the coming decades (Warner et al. 2016 ). The state of coral bleaching experimental design and methods Prior to the 1970s, the phenomenon of coral bleaching was relatively unknown. In 1971, coral bleaching was reported on a Hawaiian nearshore reef adjacent to a power plant that discharged warm water (Jokiel and Coles 1974 ). The first experimental research connecting coral bleaching with high‐temperature stress followed (Jokiel and Coles 1977 ). One of the first records of large‐scale heat‐induced coral bleaching was in Panama, which was attributed to a thermal anomaly associated with the 1982–1983 El Niño event at that time (Glynn 1983 ). Since then, experimental research on coral bleaching has accelerated, with at least 243 peer‐reviewed journal articles published since 1990, two‐thirds of which were published in the last 10 years alone (McLachlan et al. 2020 ). Manipulative experiments have been, and remain, critical for elucidating the triggers and responses of the coral holobiont to thermal stress and assessing their subsequent recovery. Research to date reveals that bleaching susceptibility and recovery vary among coral species, populations, seasons, reef habitats, and genetically distinct individuals (i.e., genets, Box 1) as well as among corals harboring similar or different algal endosymbionts or bacteria (e.g., Rowan et al. 1997 , Fitt et al. 2000 , Loya et al. 2001 , Grottoli et al. 2006 , 2014 , Palumbi et al. 2014 , Ziegler et al. 2017 , Muller et al. 2018 , Morikawa and Palumbi 2019 , Voolstra et al. 2020 ). Yet, it is unclear how much of the variation in bleaching responses is a consequence of biological differences in bleaching among coral holobionts, differences in experimental conditions (e.g., differences in light, baseline temperature, rate of temperature increase, experimental duration, flow, etc.), or methodologically inherent biases in how coral bleaching is measured (McLachlan et al. 2020 ). We know that the scientific understanding of coral bleaching relies heavily on experimental outcomes from three coral species ( Pocillopora damicornis , Stylophora pistillata , and Acropora millepora ), that experimental conditions are sometimes not reported (e.g., missing information on water flow, experimental location, heating rate), and that measurements of bleaching phenotypes are weighted heavily by responses of the endosymbiotic algae (McLachlan et al. 2020 ). Thus, direct comparisons among studies can be challenging. While experimental methods ultimately depend on the research question, this paper outlines a strategy for providing a common framework for coral bleaching experiments to enhance cross‐comparisons and strengthen coral bleaching meta‐analyses. The details were developed by 27 coral research scientists from 21 institutions, spanning research expertise in biological, geological, physical, and computational disciplines, who participated in the first Coral Bleaching Research Coordination Network (CBRCN) workshop at The Ohio State University in May of 2019. Box 1. Glossary of Terms \n Ambient temperature: temperature at time of collection. \n Baseline temperature: temperature from which heat‐stress offset is calculated (typically MMM). \n MMM: maximum monthly mean (i.e., average daily temperature of the hottest month of the year for the previous several years). \n Genets † : formed by sexual reproduction. All colonies and tissue that can trace their ancestry back to the same fertilization event belong to the same genet (Appendix S1 : Fig. S1). \n Genotype † : the genetic makeup of a sample for a given (set of) genetic marker(s). When enough markers are assayed, a sample can be assigned to a genet based on its genotype. \n Ramets † : physically independent modules arising from colony fragmentation or other asexual means of dispersion. A genet can have one or many ramets. Ramets can be experimentally generated nubbins, naturally occurring fragments, or attached colonies (Appendix S1 : Fig. S1). \n Phenotype: the set of observable characteristics of an individual resulting from the interaction of its genotype with the environment. \n Water flow rate: volumetric water flow rate per unit time (L/s −1 ). In a tank, this would be the fluid output from the exhaust of the pump or tank outflow in flow‐through systems. \n Water turnover time: time required to replace the entire volume of water in a tank(s), assuming the tank is continuously well mixed. Calculated by dividing the tank volume by the flow rate. \n Water flow velocity: motion of water relative to sessile coral (cm/s −1 ). \n † Baums et al. ( 2019 ). \n Experiments were separated into three temporally defined categories: (1) short‐term and acute (0–7 d of thermal stress), (2) moderate duration (8–30 d of thermal stress), and (3) long‐term and chronic (>31 d of thermal stress) experiments. The methods used and the experiments conducted within each category are clearly different from each other (McLachlan et al. 2020 ) and thus were considered separately. A summary of the common framework for coral bleaching experiments in each category is given in Table 1 (see details in the Proposed Common Framework section). Our summary is not intended to be prescriptive, but instead should be considered as a heuristic guide to help facilitate and strengthen comparisons among studies. One common finding that emerged from discussions of all three experimental categories was to provide guidance on the number of replicates in experiments. This topic will be discussed first as it applies to all experimental categories. In addition, we find that including measurements for common coral response variables in coral bleaching experiments would further enhance cross‐study comparisons by providing common physiological reference points across studies. A list of potential response variables is provided at the end of Table 1 . A brief review of common methods for measuring each listed variable is provided in Appendix S1 . A full discussion of the proposed common framework is detailed below. Table 1 Framework for coral bleaching experimental methods and coral response variables. Variable Appendix section Suggested target or range Acute and short‐term experiments(<7 d at BST) Moderate duration experiments (7–30 d at BST) Long‐term experiments (>30 d at BST) No. genets S1.1 5 minimum; >5 if possible 5 minimum; >5 if possible ≥5 No. replicate tanks per treatment Minimum two tanks per treatment ANOVA design, minimum of 3 tanks per treatment factor; regression design, gradient study with >3 treatment levels; avoid pseudo‐replication Avoid pseudo‐replication Acclimatization to experimental tanks Typically none 7–12 d following fragmentation and mounting 7–12 d following fragmentation and mounting Control temperature S1.2 Ambient temperature at collection site at time of collection Ambient temperature at collection site at time of collection Ambient temperature at collection site during the experimental period Baseline temperature S1.2 MMM and/or rapid temperature profiles corresponding to in situ temperature patterns if appropriate MMM MMM Bleaching stress temperature above local MMM Typically +3 to +9°C; increase temperature from MMM until death is observed, then set target temperature lower; if the goal is to observe phenotypic variability, expose corals to several temperatures to find the temperature at which half of the corals bleach; stress exposure should happen at the same time of day; temperature stress duration should be standardized within experiments +1 to +4°C depending on local ecological relevance and species, may need to be higher in extreme environments +1°C or more depending on local ecological relevance and species Temperature ramp‐up rate None recommended as it will depend on temperature stress duration; heating rates should be adjusted to take the same time across treatment temperatures 0.1–1°C/d Mimics increase in temperature rate observed during previous bleaching events at that site Temperature modulation Temperature ramp‐up to static elevated temperature, followed by recovery at baseline temperatures; profiles can be run once or multiple times May be static or diurnally modulated; choice of modulation should be the same in treatments and controls Static or diurnal for indoor experiments; diurnal and seasonal for outdoor experiments Control conditions At ambient temperature; exact same conditions as treatment, except for temperature At ambient temperature; exact same conditions as treatment, except for temperature At ambient temperature; exact same conditions as treatment, except for temperature; mimics natural conditions Light S1.3 Ideally, static light conditions for short‐term thermal exposures (with no light at night) or possibly diurnal variability if over several days; light levels match natural light conditions; minimum 250–500 µmol photons·m −2 ·s −1 \n Ideally, diurnal variability with 80% of maximum PAR light at collection site; minimum 250–500 µmol photons·m −2 ·s −1 \n Indoor tanks, diurnal variability (with moonlight cycles); outdoor tanks, apply shade to mimic PAR at collection depth; minimum 250 µmol photons·m −2 ·s −1 \n Flow Flow system S1.4 Flow‐through Report pump rate in liters pumped per hour 2–20 cm/s 2–20 cm/s, mimic natural conditions Closed Report pump rate in liters pumped per hour, tank volume Record flow rates, pump size, tank volume, and try to base the flow rate on in situ data Record flow rates, pump size, tank volume Tank volume turnover S1.4 Flow‐through 100% within 3–6 h 1–4 times per day 1–4 times per day Closed 100% within 3–6 h Case‐dependent and depends on system biomass Case‐dependent and depends on system biomass Feeding S1.5 None typically Minimum once per week to satiation; report feeding amount, rate, and food type Minimum once per week to satiation; ideally feed up to three times per week; report feeding amount, rate, and food type; mimic food availability in nature Seawater S1.6 Filtered or unfiltered; natural or artificial Filtered or unfiltered; natural or artificial Filtered or unfiltered; natural or artificial Post heat‐stress monitoring Hours to a few days (longer than the stress duration); this doubles the number of fragments needed If possible, immediate (0.2–1 month) and long‐term monitoring (>1 month) depending on the question 0.2–3 months depending on the question Other environmental conditions Salinity S1.7 Nutrients S1.8 pH S1.9 Dissolved oxygen S1.10 Coral bleaching responses Bleaching phenotype Image analysis of color S2.1a Chlorophyll concentration S2.1b Symbiodiniaceae cell density S2.1c Holobiont phenotype Mortality S2.2a Skeletal growth S2.2b Other Active chlorophyll fluorescence † \n S2.3a Symbiodiniaceae identity S2.3b Notes BST, bleaching stress temperature; MMM, maximum monthly mean (i.e., men temperature of the warmest month; ANOVA, analysis of variance; PAR, photosynthetically active radiation. A review of commonly used methods is summarized in Appendix S1. Glossary of terms is given in Box 1. † E.g., PAM fluorometry. John Wiley & Sons, Ltd"
} | 4,753 |
39247511 | PMC11378032 | pmc | 327 | {
"abstract": "Slippery liquid-infused porous surfaces (SLIPSs) inspired by Nepenthes have attracted much attention owing to their potential application in various cutting-edge fields. However, the performance of SLIPSs is impeded by surface damage and lubricant depletion, thereby limiting their further application. Herein, a UV-responsive slippery surface (SMEMG) was fabricated by introducing the UV-responsive functional group coumarin into the polymer side chain through random copolymerization, followed by crosslinking, curing and impregnation with vegetable oil. The self-healing ability and lubricant self-replenishing performance of the SMEMG were investigated. The results show that upon exposure to UV light, the damaged surface substrate can be repaired through a reversible photodimerization reaction between coumarin groups. Meanwhile, the lubricant oil within the bulk of the SMEMG substrate can be extruded to the surface during the photodimerization reaction, facilitating the recovery of surface wettability. The SMEMG exhibited excellent self-cleaning and anti-algal properties as well as durability in a harsh environment, demonstrating its promising application in marine anti-fouling.",
"conclusion": "4 Conclusions In this paper, a UV-responsive slippery surface (SMEMG) was constructed by introducing the coumarin group into the side chain of a polymer. Both polymer substrates MEMG@ED and the slippery surface SMEMG exhibited excellent self-healing properties. Upon exposure of UV light, the broken MEMG@ED and SMEMG could be completely repaired while maintaining their mechanical and slippery properties. The intelligent control of a droplet movement on the SMEMG could be realized by controlling the release of lubricant under UV irradiation due to its exceptional lubricant self-replenishing properties. Furthermore, the outstanding anti-algal adhesion performance demonstrated that the SMEMG has potential in the field of marine anti-fouling.",
"introduction": "1 Introduction Researchers have gained a profound understanding of the biological world and have utilized biomimetics to replicate diverse functional surfaces such as anti-wetting, 1–3 self-cleaning, 4–6 and anti-fouling 7 surfaces. The study of Nepenthes has revealed a unique slippery surface in its prey-trapping pitcher organs, which has led to the discovery of slippery liquid-infused porous surfaces (SLIPSs). 8,9 SLIPSs have excellent liquid-repellent ability, as well as extremely low contact angle hysteresis and self-repairing capabilities. 10 Consequently, SLIPSs have demonstrated significant advantages and practical applications in marine anti-fouling, 11–13 anti-icing, 14–16 self-cleaning, 17,18 and smart liquid manipulation. 19–21 By introducing self-healing polymers in a solid substrate, the durability of SLIPSs can be effectively improved. However, due to the inherent fluid characteristics of lubricant, they will inevitably experience loss in actual application due to external forces, such as contact, wiping, water flow impact, and the carrying of organic solvents, leading to a decline in wettability. 22 Therefore, how to timely add lubricant to the porous solid substrate, which completes the self-replenishing of SLIPSs, is crucial for enhancing their applicability. Li et al. 23 ingeniously utilized the structural characteristics of the earthworm skin to develop a dual-layer self-replenishing SLIPS comprising surface nanostructured membranes and square microwells that store lubricating fluid at the bottom, thereby ensuring long-lasting exceptional lubrication even in extreme conditions such as cooling, heating, and continuous droplet impact. Additionally, Zhang's group 24 introduced a novel self-replenishing SLIPS featuring primary microgrooves and secondary microcavities to address soft tissue adhesion on electrosurgical electrodes during minimally invasive surgery by storing lubricants within cavities for extended periods. Sun et al. 25 also reported embedding a large hole network into an inverse opal nano-porous structure as a reservoir for storing and supplying lubricants in self-replenishing applications, proposing that this achievement can be effectively applied in optical sensing, fluid transport, medical self-cleaning, and other fields. While there are several similar studies available, it should be noted that the self-healing function of these materials primarily relies on capillary forces within specially designed SLIPS structures to transport lubricants where they are needed. 26–31 However, such self-replenishing SLIPSs encounter challenges that pose difficulties in resolution through intelligent human control. The manipulation of a material's properties through light radiation has attracted considerable attention due to its ease of operation and remote controllability. 32–34 In our previous work, we fabricated an intelligent UV-responsive slippery surface using a porous substrate and silicon oil, which could achieve self-replenishment through conformational transformation of the azobenzene groups under UV light irradiation. 35 Herein, a UV-driven self-replenishing slippery surface (SMEMG) was fabricated based on vegetable oil as the lubricant and coumarin-modified polyacrylate as the substrate. Based on the reversible photodimerization and photodepolymerization between coumarin's functional groups under UV irradiation, the polymer substrates of the SMEMG showed excellent self-healing property. Notably, photodimerization of the coumarin groups could enhance the crosslinking degree of the polymer network, resulting in the extrusion of VO within the bulk of SMEMG to its surface and thus realizing a self-replenishing of the lubricant. Furthermore, the SMEMG exhibited excellent self-cleaning and anti-algal properties as well as durability under high temperature, high-speed shear force, and a dynamic flow environment, making it a promising material for application in marine anti-fouling.",
"discussion": "3 Results and discussion 3.1 Fabrication and characterization of the SMEMG As shown in Scheme 1 , HML was first modified with 3-bromopropene to synthesize 7-allyloxy-4-methylcoumarin. Then the copolymer MEMG was prepared via a random free radical polymerization of MHML, EHA, MMA, and GMA. Subsequently, the MEMG, ED as the cross-linker, and VO as the lubricant oil were evenly mixed and then cured at 70 °C for 5 h to fabricate the UV-responsive slippery surface (SMEMG). The chemical composition of MHML, MEMG, and the substrate MEMG@ED were identified by FT-IR spectroscopy. As shown in Fig. 1a , the adsorption peak for –OH stretching vibrations was observed at 3218 cm −1 in the FT-IR spectrum of HML, meanwhile, it had disappeared in the FT-IR spectrum of MHML. In addition, the broad brand at around 3000–2700 cm −1 was assigned to the C–H stretching vibrations in the alkyl and aromatic ring, whereby the intensity of the C–H absorption peak on MHML was significantly higher than that on HML. Besides, signals for C \n \n\n<svg xmlns=\"http://www.w3.org/2000/svg\" version=\"1.0\" width=\"13.200000pt\" height=\"16.000000pt\" viewBox=\"0 0 13.200000 16.000000\" preserveAspectRatio=\"xMidYMid meet\"><metadata>\nCreated by potrace 1.16, written by Peter Selinger 2001-2019\n</metadata><g transform=\"translate(1.000000,15.000000) scale(0.017500,-0.017500)\" fill=\"currentColor\" stroke=\"none\"><path d=\"M0 440 l0 -40 320 0 320 0 0 40 0 40 -320 0 -320 0 0 -40z M0 280 l0 -40 320 0 320 0 0 40 0 40 -320 0 -320 0 0 -40z\"/></g></svg>\n\n O and C C on the coumarin ring appeared in the structure of HML at 1686 and 1609 cm −1 , and could also be found in the FT-IR spectrum of MHML slightly shifted at 1731 and 1615 cm −1 . These results demonstrated that MHML was successfully synthesized. As shown in Fig. 1b , there was an absorption peak at 1613 cm −1 in the backbone of MEMG, suggesting that the C C bond still existed in the structure of MEMG. However, the absorption peak intensity of C C on MEMG was decreased compared to that on MHML. After the polymerization of MHML, 2-EHA, GMA, and MMA, most of the C C bonds disappeared through the process of polymerization, but a small amount of C C bonds on the coumarin ring of MHML did not participate in the polymerization. Moreover, the intensity of the C–H stretching vibration peak on MEMG at around 3000–2700 cm −1 was further enhanced compared to that on MHML, and the characteristic absorption peak of the epoxy group was discovered at around 908 cm −1 . These results indicated that the copolymer MEMG was successfully fabricated. Furthermore, the MEMG@ED showed a new broad band at 3349 cm −1 , which was the overlapping peak of –OH and N–H regenerated by the reaction of the epoxy group on MEMG and the amino group on ED. Besides, the absorption peak strength of the epoxy group at 908 cm −1 was also significantly reduced after the crosslinking, indicating the successful preparation of MEMG@ED. Scheme 1 Preparation procedure for the SMEMG. (a) Synthesis of 7-allyloxy-4-methylcoumarin (MHML); (b) fabrication of P(MHML- r -EHA- r -MMA- r -GMA) (MEMG); (c) fabrication of the UV-responsive slippery surface (SMEMG). Fig. 1 (a) FT-IR spectra of HML and MHML; (b) FT-IR spectra of MEMG and MEMG@ED before and after 365 nm UV irradiation for 3 h; (c) SEM image of SMEMG; inset shows the SEM image at high magnification; (d) 2D AFM image of the SMEMG over a scope of 5 μm × 5 μm. In this system, vegetable oil was employed as the lubricant to infuse into the substrate via blending and then forming the slippery surface SMEMG. Hence, the chemical affinity between the VO and MEMG@ED is crucial and could greatly affect the stability of the slippery materials. As shown in Fig. 2a , the surface of MEMG@ED exhibited an ultra-low initial VO contact angle at 0 s. Then, the VO completely spread on the surface of the substrate within 15 s, meanwhile forming a stable lubricant layer on the SMEMG. The results certified that there was a good affinity between the VO and MEMG@ED, which is conducive to the preparation of a highly stable slippery surface. Then the surface morphology and roughness were investigated. The SMEMG presented a defect-free smooth surface and the surface roughness was 0.12 nm, as shown in Fig. 1c and d , indicating the formation of a slippery surface with ideal smoothness. The sliding behavior of a droplet on the SMEMG was then recorded by taking digital pictures. As shown in Fig. 2b , the water droplet (7 μL) flexibly slid down the surface of the SMEMG and left no trace, suggesting the SMEMG possessed excellent slippery performance. Fig. 2 (a) Photographs of VO (3 μL) spreading on the surface of MEMG@ED. (b) Photographs showing the dynamic mobility of a water droplet (7 μL) on the SMEMG with a low tilting angle (5°). 3.2 Self-healing performance As presented in Fig. 3a , both the MEMG@ED and SMEMG exhibited good transparency, whereby the characters below could be clearly seen through the substrate film and slippery surface. After the infusion of VO into the substrate, the transparency of the SMEMG slightly declined compared to that of MEMG@ED, but it still showed favorable light transmittance owing to the good chemical affinity between the polyacrylate-based substrate and vegetable oil. Further, the lubricant oil could fill the destroyed area by capillary force and then realize the healing of the dewetting property once the SLIPS materials is damaged by external force. However, it was difficult to repair the damaged position once the skeleton substrate was broken, resulting in the loss of the slippery performance. Here, the self-healing ability of MEMG@ED was detected by cyclic “cutting–healing” tests. As shown in Fig. 3b , the broken interfaces of the substrate film were completely repaired after irradiating under UV light with a wavelength of 365 nm for 3 h and there was no obvious cutting mark on the damaged area, as shown in the optical microscopy photo. Besides, the repaired MEMG@ED presented good mechanical properties, and no tear marks appeared at the repaired fracture after tensile treatment. Fig. 3 (a) Digital photographs of MEMG@ED and SMEMG; (b) self-healing process of MEMG@ED; (c) self-healing process of a heart-shaped SMEMG; (d) optical microscopy images of a heart-shaped SMEMG, both MEMG@ED and the half sample of the SMEMG were dyed with Sudan I. To better illustrate the self-healing performance of the SMEMG, dyed and undyed samples were cut into a heart shape and then contacted with the broken interfaces without external pressure. As shown in Fig. 3c , after treatment under UV light for 3 h, the two fragments with different colors were joined in a new heart-shaped film. Similarly, the repaired SMEMG showed good mechanical strength. After stretching, the jointed SMEMG remained intact. The corresponding optical microscopy images also demonstrated the excellent self-healing ability of the SMEMG and the good mechanical property of the healed SMEMG ( Fig. 3d ). Besides, the dewetting properties of the SMEMG before and after healing were investigated. Here, the slippery surface was coated on the surface of an aluminum sheet. As shown in Fig. 4a , a water droplet (20 μL) easily slid down the surface of the SMEMG in 2.1 s. However, the sliding droplets would pin at the scratch after the surface was damaged by a scalpel ( Fig. 4b ). Nevertheless, after the scratched SMEMG was healed under UV light irradiation ( Fig. 4c ), the droplet re-slid down the surface ( Fig. 4d ), indicating the recovery of the liquid repellency ability of the SMEMG. The optical microscopy images also demonstrated the healing process of the damaged SMEMG coating ( Fig. 4e, f, and h ). All these results suggested that both the substrate and VO-infused slippery surface in this system have superb self-healing performance. The coumarin groups in the side chain of MEMG can undergo photodimerization at UV light wavelengths >300 nm, where the ethylenic group of the coumarin molecules undergo [2 + 2] cycloaddition to form a cyclobutane ring. 37–39 The evidence for this reaction is provided in Fig. 1b , where the peak at 1613 cm −1 , corresponding to the C C bond within the MEMG@ED spectra, exhibited a marked reduction after 3 h exposure to 365 nm UV light, confirming the occurrence of photodimerization. As a result, a new crosslinking network was formed to repair the damaged samples ( Fig. 4g ). Notably, the damaged matrix was preferentially irradiated under UV light at a wavelength of 254 nm for 30 min to completely photodepolymerize the coumarin groups before healing under 365 nm UV light. Fig. 4 Motion of a water droplet (20 μL) on the original SMEMG (a), SMEMG damaged via scalpel scratching, (b) and a self-healed SMEMG surface (d) under 365 nm UV light irradiation for 3 h; (c) a self-healed SMEMG surface under 365 nm UV light irradiation for 3 h. Optical microscopy images of original SMEMG (e), damaged SMEMG (f) and self-healed SMEMG surface (h). (g) Self-healing mechanism of damaged SMEMG under 365 nm UV light irradiation. The water droplet was dyed with methyl orange. 3.3 Self-cleaning and stability properties In practical environments, many materials are inclined to be contaminated by complicated liquids and fine particulate matters due to their high surface energy and structural defects, thus greatly weakening the function of such materials. Here, some compound liquids in daily life were used as model pollutants to evaluate the surface self-cleaning performance of the SMEMG. As shown in Fig. 5 , droplets of coffee and milk could flexibly slide down the surface of the SMEMG without any residuals. Even the sauce droplet with a complicated chemical composition and the honey droplet with ultra-high viscosity could be completely removed from the SMEMG. Whereas, these droplets were hard to slide down the surface of glass, and distinct sliding traces could be observed on the glass after liquid collecting. The defect-free surface with relatively low surface energy endowed the SMEMG with prominent anti-fouling performance. Subsequently, three typical solid particles were employed to assess the self-cleaning property of the SMEMG. As presented in Fig. S1, † a quantity of powders was firmly adhered to the surface of the SMEMG. With the water droplets sliding on the surface, the hydrophilic powders (CuCl 2 ·2H 2 O) were rapidly dissolved in the water and then carried away as the droplets slid down the surface. Differently, the hydrophobic particles (SiO 2 and sandy soil) floated on the liquids and then slid down with the water droplets. Benefiting from the molecular-scale smooth surface topography and good liquid repellency of SLIPS materials, these particles could not be trapped into the substrate. Therefore, the SMEMG presented a superb self-cleaning performance. Fig. 5 Snapshots showing the sliding motion of coffee, sauce, milk, and honey droplets (∼20 μL) on the SMEMG and glass. Durability is a very important factor in practical applications. However, the lubricant layer on the surface of SLIPS materials is really unstable due to the limited interaction between the substrate and lubricant oil. Hence, accelerated lubricant loss experiments were implemented here to detect the stability of the SMEMG. As shown in Fig. 6a , no obvious change was observed in the CA and sample mass tests after treatment at 80 °C for 15 days, indicating the SMEMG had remarkable thermal stability owing to the low volatility and high decomposition temperature of the VO. In addition, the SMEMG retained a good dewetting performance under high-speed shear force or vertical placement for 15 days ( Fig. 6b and d ). Moreover, the CA and mass loss of the SMEMG started to show slight increases after immersing in flowing water for 60 h. The excellent physical durability of the SMEMG was from the good compatibility between the MEMG@ED and VO. In summary, based on the synergistic effect between the excellent stability and self-healing ability of the SMEMG, the longevity of the slippery surface could be significantly prolonged, thus promoting the practical use of the SLIPS materials. Fig. 6 Durability of the SMEMG. Variation of the CA and Δm of the SMEMG during the stability test. (a) SMEMG after heating at 80 °C for 1–15 days; (b) SMEMG rotated at 0–5000 rpm for 1 min; (c) SMEMG after immersing in water with a spin rate of 30 rpm for 80 h; (d) SMEMG placed in the vertical position for 1–15 days. 3.4 Self-replenishing of the lubricant oil Although the SMEMG exhibited prominent stability under a harsh environment, exhaustion of the surface lubricant layer for SLIPS materials is inevitable when serving in a dynamic water flow given enough time, thus resulting in the reduction of the anti-fouling performance. Interestingly, there was a strong absorption peak in the fluorescence spectra of HML, MHML, and MEMG at the wavelength range of 340–480 nm, as presented in Fig. S2. † In addition, the SMEMG showed a remarkable fluorescence phenomenon under UV light irradiation at the wavelength of 365 nm (see inset in Fig. S2c † ). The results indicated that the MHML was successfully bonded into the side chain of MEMG and endowed the surface of the SMEMG with an extraordinary fluorescence effect. Based on this, we found that the SMEMG was able to respond to UV illumination at the wavelength of 365 nm, then extruding the lubricants in the bulk of substrate to the damaged surface, thus rebuilding a new lubricant layer on the surface and hence recovering the surface wettability. As shown in Fig. 7a , original SMEMG exhibited a good dewetting performance, whereby a water droplet could easily slide along the surface of the SMEMG within 2.1 s. Whereas the slippery ability of the SMEMG was immediately lost when the surface lubricant layer of the SMEMG was wiped off. As shown in Fig. 7b , the water droplet firmly stuck on the swabbed SMEMG and could not slide on the surface even after 10 s. Notably, the droplet would slide down the surface after being irradiated under UV light for 3 h in 2.2 s, thus demonstrating the re-construction of a new lubricant layer on the destroyed SMEMG. However, due to the superb chemical affinity between the vegetable oil and alkyl polyacrylate (MEMG), the oil-releasing velocity was limited. This ultra-high compatibility will promote the formation of an organogel, then restricting the extrusion of the lubricant oil. However, the low affinity will lead to the rapid loss of the lubricant. Fig. 7 Dynamic control of the mobility of a water droplet (20 μL) on the SMEMG. (a) Process of a water droplet slipping off the SMEMG at the initial state. (b) Process of a water droplet pinned on the SMEMG after the surface lubricant oil was swabbed. (c) Process of a water droplet re-sliding down the SMEMG after UV irradiation for 3 h. In the initial state, the rough structure of the substrate was covered by a layer of lubricant oil, and the surface of the SMEMG was ultra-smooth with an RMS of 0.12 nm ( Fig. 8a ). Hence, the droplet could easily slide down the surface. After the lubricant layer was exhausted, the rough structure of the substrate became exposed at ambient environment. The roughness of the surface thus increased to 27.5 nm ( Fig. 8b ). The exposed rough structure would greatly enhance the hysteresis angle of a droplet, then restricting the sliding motion of the water droplet. Afterwards, based on the UV-responsive ability of coumarin groups hanging on the side chain of MEMG, the VO could be released to the damaged surface and could then reform a new liquid film on the surface. The release of the VO was corroborated by the XPS data detailed in Table S1. † The content of the C element in VO was higher than that in the substrate, while the content of the O element was lower than that in the substrate. Also, the contents of C element and O element on the surface of the SMEMG formed by the infusion of VO into the substrate were between those of the VO and the MEMG@ED. When the surface lubricant oil of the SMEMG was lost and then self-replenished, the formed R -SMEMG surface element composition was basically the same as the initial SMEMG surface element composition. Therefore, it is considered that the VO inside the bulk of substrate self-filled the material surface in response to the UV irradiation, forming a new lubricant layer. As shown in Fig. 8d , the roughness of the recovered SMEMG was decreased to 0.16 nm. Hence, the hysteresis angle of a sliding droplet would be highly declined, and then the water droplet could move along the surface. Fig. 8 Schematic of the consumption and rebuilding process of the lubricant layer on the surface of the SMEMG. (a) Distribution of the lubricant oil on the original SMEMG and the corresponding 3D AFM image, RMS: 0.12 nm; (b) consumption of the surface lubricant layer of the SMEMG and the corresponding 3D AFM image, RMS: 27.5 nm; (c) reservoir lubricant oil in the bulk of the substrate was released to the surface of the destroyed SMEMG under UV light irradiation at 365 nm wavelength; (d) rebuilding of a new lubricant layer on the destroyed SMEMG and the corresponding 3D AFM image, RMS: 0.16 nm. The mechanism of the lubricant oil self-replenishing in this system is illustrated in Fig. 9 . Under UV light irradiation at a wavelength of 365 nm, the adjacent coumarin groups in the side chain of the polymer would undergo photodimerization, thus enhancing the crosslinking degree of the network, and then advancing the extrusion of VO in the bulk of the SMEMG to the surface to construct a new lubricant layer, thereby promoting the recovery of the surface dewetting performance. Notably, the swabbed SMEMG was first treated under UV light at 254 nm for 30 min before 365 nm UV light treatment to facilitate the photodepolymerization of the coumarin groups. To further prove the influence of the coumarin groups on the self-replenishing performance of the SMEMG, the UV-responsive capacity of the SEMG was detected as the control test (Fig. S3 † ). As shown in the result, the original SEMG possessed good slippery performance. A water droplet could be removed away from the surface of the SEMG within 2.2 s, leaving no trace. Then, the droplet pinned on the surface when the liquid layer of the SEMG was lost. However, the water droplet remained pinned on the surface of the SEMG after the damaged SEMG was illuminated under UV light for 8 h. Emphatically, the fabrication of the SEMG was the same as that of the SMEMG but without the introduction of MHML. In addition, no obvious absorption peak was observed in the fluorescence spectra of EMG (Fig. S2d † ), verifying that the SEMG without MHML was unable to respond to the UV light irradiation. Therefore, it seems that the coumarin groups on MHML endowed the SMEMG with a superb UV-responsive capacity. Hereafter, serial swabbing-UV illumination tests were performed to evaluate the lubricant oil-releasing durability. As shown in Fig. 10 , during the 10 cycles of the swabbing and releasing process, the CA and water sliding velocity of the SMEMG just showed a slight fluctuation. After 10 cycles, the CA of the SMEMG was maintained at 66.41° and the water droplet sliding velocity remained at 11.02 mm s −1 , suggesting the VO could be repeatedly replenished to the surface and then the dewetting performance of the SMEMG could be recovered. The excellent durability of the lubricant oil-releasing capability benefited from the good compatibility between the VO and MEMG@ED and will highly advance the actual application of SLIPS materials. Fig. 9 Lubricant oil self-replenishing mechanism of the SMEMG. Fig. 10 CA (a) and sliding velocity (b) changes on the SMEMG as a function of repeated swabbing and UV light irradiation at 365 nm wavelength for 3 h; inset images in (a) show the CA images and in (b) the water droplet sliding motion on the SMEMG. 3.5 Anti-algal adhesion performance In this work, the anti-algal adhesion performance of the SMEMG surface was investigated using diatom and chlorella as biological fouling models. As shown in Fig. 11a , after soaking in diatom solution for 24 h and simple washing, there was almost no diatoms attached on the surface of the glass sheet, while a certain amount of diatoms were observed on the MEMG@ED surface. This could be attributed to the hard shells of the diatoms, which tend to adhere to softer material surfaces. Meanwhile, a large number of chlorella was found adhered to the surface of the glass sheet ( Fig. 11b ), probably because chlorella is soft and easy to deform, making it easier to adhere to the hard surface. In addition, no attached diatoms or chlorella were found on the surface of the SMEMG. Due to the excellent stability of the SMEMG, the lubricating layer remained securely anchored on the substrate surface in flowing water. The smooth surface structures and flowing water prevented algae from adhering. However, prolonged soaking eventually led to a loss of the surface lubrication layer on the SMEMG, resulting in significant algae attachment on the lubrication-lacking surface (S-SMEGMA). Nevertheless, due to its self-replenishing capability with lubricant oil, under UV light exposure, SMEMG released the internal lubricant onto its surface forming a new lubrication layer that restored its anti-algal adhesion performance; thus, preventing adherence to newly formed R -SMEMG surfaces. Fig. 11 Anti-algal tests of different samples. (a) Anti-diatom test; (b) anti-chlorella test. S -SMEMG refers to the SMEMG without a lubrication layer and R -SMEMG refers to the lubricant oil self-replenished slippery surface based on a damaged SMEMG."
} | 6,906 |
34272239 | PMC8284889 | pmc | 328 | {
"abstract": "Memristor with topotactic phase transition demonstrates controllable analog switching and implements neural network pruning.",
"introduction": "INTRODUCTION The growth of computing power in digital hardware, including central processing unit and graphics processing unit, has driven the rapid development of artificial intelligence. This, in turn, raises higher and higher demand on the hardware performance, even exceeding the pace of Moore’s law. One of the key bottlenecks arises from the physical separation of memory and computing units in the widely adopted von Neumann architecture, which leads to a grand challenge of memory wall problem. Inspired by neurobiological systems, neuromorphic computing has emerged as a promising computing paradigm with the feature of massively parallel computation in memory to break the so-called von Neumann bottleneck ( 1 , 2 ). Various nonvolatile memories (NVMs), such as resistive random-access memory (RRAM) ( 3 , 4 ) and phase-change memory (PCM) ( 5 ), have been extensively studied as artificial synapses and neurons to build prototype artificial intelligence chips ( 6 – 8 ). Different from digital memory applications, here, reproducible analog switching characteristics (e.g., multilevel conductance states, weight update linearity and symmetry, and low variability) are desired to meet the requirement of high computing accuracy and energy efficiency ( 9 , 10 ). Unfortunately, those existing emerging NVMs still suffer from nonideal device characteristics (fig. S1), which are one of the main challenges for the hardware implementation of large-scale neuromorphic computing systems. For example, conventional filament-type RRAM relies on the random oxygen vacancy ( V o ) migration in the amorphous switching oxides, leading to intrinsically large device variations, while the absence of local Joule heating effect in interface-type RRAM usually results in poor retention and low speed ( 11 ). For PCM, it typically shows asymmetric switching due to the abrupt quench process in the crystalline-to-amorphous phase transition and also suffers from the conductance drift issue ( 12 ). It is noted that those imperfect device characteristics originate from their intrinsic working mechanisms and hence are difficult to be eliminated by simply optimizing their device structures ( 13 , 14 ). In addition, so far, these devices are mainly limited to mimic the functionalities of an individual neuron or synapse (synaptic plasticity, neuronal firing, etc.), while the biomimicry of many important network-level properties, such as neural network pruning that is critical for cognitive learning in biology, has not been explored yet. Therefore, for future high-performance neuromorphic computing, innovations in materials and devices with new working mechanisms are highly desired to yield more controllable analog switching characteristics and further construct more bioplausible neural networks ( 15 , 16 ). In search of new materials and structures for low-variability analog switching memristors, here, we propose a novel synapse, namely, topotactic phase transition RAM (TPT-RAM), using brownmillerite (BM) oxides [such as SrCoO 2.5 (SCO) ( 17 , 18 ) and SrFeO 2.5 ( 19 – 21 )] as the resistive switching oxide. We chose SCO as an exemplary material whose unique crystal structure formed by alternating stacks of oxygen octahedra and oxygen tetrahedra provides the favorable conditions to achieve uniform analog switching: (i) The highly ordered one-dimensional oxygen vacancy channels (OVCs) provide predefined freeway for the migration of oxygen ions to induce phase transition and resistive switching ( 22 ). Compared with other methods intended to confine the ion migration, such as metal doping ( 23 ) and dislocation engineering ( 15 ), the highly ordered and atomically precise OVCs in BM oxides are more uniform and easier to manipulate without additional ex situ processes ( 24 ). (ii) The multivalent cobalt ions change reversibly between BM and perovskite (PV) structures on the basis of the adjustment of oxygen stoichiometry without losing the intrinsic lattice architecture ( 25 – 27 ), which can yield gradual switching. (iii) High-quality, stable BM oxide as the resistive switching layer ensures excellent retention at multilevel conductance states and also enhanced endurance. In this work, to implement low-power neuromorphic computing, we designed and fabricated SCO-based TPT-RAM with tunable OVCs as memristive synapses. Such TPT-RAM provided an excellent platform to thoroughly study the topotactic phase transition–associated switching mechanism by correlating electrical and structural characterizations with comprehensive atomic-device modeling and simulations ( 28 – 30 ), which, however, is difficult to perform for conventional RRAM with amorphous oxide like HfO 2 ( 31 ). Experimentally, we demonstrated that the high-speed and uniform analog TPT-RAM can be achieved by manipulating OVCs through the top and bottom electrodes (BEs). Furthermore, inspired by the selective stabilization of developing synapses in biological neural networks, we implemented the online training of a sparse neural network through automatic pruning, realizing a substantial reduction of both network size and power consumption.",
"discussion": "DISCUSSION To sum up, we demonstrated TPT-RAM as a new type of memristive synapse relying on the topotactic phase transition in SCO. The unique oxygen migrations along the highly ordered OVCs led to excellent analog switching characteristics with a much reduced cycle-to-cycle variability of ~0.9% and a device-to-device variability of ~4.9%, the low operation voltage of 0.8 V, and a fast speed below 100 ns. DFT calculations and KMC simulations further confirmed the resistive switching mechanism consistency with the measured device electrical characteristics. These results demonstrated the significance of controlling the ion migration paths to improve the uniformity of RRAM, which is beneficial to guide the optimization of future neuromorphic devices. For future integration with silicon transistors to build functional synaptic arrays based on TPT-RAM, new techniques such as remote epitaxy and sacrificial layer–assisted film transfer could be adopted ( 48 , 49 ). Furthermore, the SCO-based synapse exhibited a unique diffusive nonvolatile dual mode, which was used to mimic the developing synapses of the human brain and implement neural network pruning during the online training, reducing up to 82.5% redundant synapses and improving the MNIST recognition accuracy to 99%. Our work points out a new direction to design and explore bioplausible analog switching memristive synapses for high-performance neuromorphic computing."
} | 1,679 |
27853136 | PMC5473601 | pmc | 329 | {
"abstract": "Mimicking the multifunctional bacterial type IV pili (T4Ps) nanofibres provides an important avenue towards the development of new functional nanostructured biomaterials. Yet, the development of T4Ps-based applications is limited by the inability to form these nanofibres in vitro from their pilin monomers. Here, to overcome this limitation, we followed a reductionist approach and designed a self-assembling pilin-based 20-mer peptide, derived from the presumably bioelectronic pilin of Geobacter sulfurreducens . The designed 20-mer, which spans sequences from both the polymerization domain and the functionality region of the pilin, self-assembled into ordered nanofibres. Investigation of the 20-mer revealed that shorter sequences which correspond to the polymerization domain form a supramolecular β-sheet, contrary to their helical configuration in the native T4P core, due to alternative molecular recognition. In contrast, the sequence derived from the functionality region maintains a native-like, helical conformation. This study presents a new family of self-assembling peptides which form T4P-like nanostructures.",
"discussion": "Discussion In this work, we have shown that T4P-like nanostructures can be obtained by using peptide self-assembly as a strategy for their formation. With this general strategy, the inability to assemble pilin monomers into T4P-like nanofibres in vitro can be circumvented. Furthermore, as in the design of other bio-inspired nanostructures, the established synthesis procedures and commercial availability of peptides highlight them as the building block of choice for the formation of T4P-like nanostructures. The investigated 20-mer peptide is a minimized form of the GS pilin subunit, encompassing two distinct sequences from the evolutionary conserved polymerization domain and the functionality-related variable region of the protein. The reductionist approach employed in this study showed that the N-terminal segment of the 20-mer, which corresponds to the GS pilin N-terminal polymerization domain α1-N, adopts a β-type conformation. While this segment, as a part of the conserved α1-N domain, natively adopts an α-helical conformation in the GS pilin and in pilin proteins generally, homologous sequences can form α-helices or β-strands in a variety of other proteins ( Supplementary Table 4 ). This suggests that the investigated α1-N sequences can be conformationally permissive and that the adoption of a particular conformation is a context-dependent event. Specifically, the membrane environment pertinent to pilin translation and polymerization in vivo is a factor likely to promote the helical conformation of α1-N, yet outside of this environment, a β-type conformation may arise. In the investigated system, the β-strand conformation is indeed adopted by the α1-N derived sequences. The ensuing supramolecular β-sheet configuration, which forms in the process of self-assembly, is a property shared by the different investigated assemblies. The totality of the data, and particularly the high-resolution structure of the α1-N 5-mer, indicates that the β-sheet interaction propagates along the morphological long axis of the assemblies, leading in all cases to their elongated shape. However, the width and height of the different assemblies varies, with considerable differences between the 4-mer nanofibres and nanoribbons, the 5-mer microcrystals, and the nanofibres formed by the longer peptides. Both the 4-mer and 5-mer peptides form wider assemblies that can form due to stable interactions between individual β-sheets in at least one plane perpendicular to their propagation direction. This is considerably more pronounced in the case of the 5-mer since Glu5 upholds multiple stabilizing interactions in both axes perpendicular to the β-sheet direction, as evident from the crystallographic data. In contrast, the longer peptides form fibres with low nanometric width and height. This difference can be regarded as the result less stable interfaces in the axes perpendicular to the axis of β-sheet propagation. Additionally, non-specific interactions between hydrophobic side-chains, which may protrude from the fibres as a part of the β-sheet arrangement, could lead to fibre bundling and clustering and thus prevent additional ordered growth perpendicularly to the β-sheet propagation axis. The supramolecular β-sheet configuration can be considered as the result of alternative molecular recognition between α1-N sequences, as compared with the configuration of the respective domain in native T4P core. While the existing structural model for T4P suggests an architecture based on spiraling helix bundles 20 , our data raises the possibility that α1-N sequences may in fact form supramolecular β-sheets in the core of T4Ps in vivo . This may be plausible since the existing approach for the structural elucidation of T4Ps utilizes the fitting of a high-resolution structure of the pilin monomer into a lower-resolution cryo-electron microscopy envelope of the intact nanofibre 64 ; while this approach is powerful, it is limited in providing atomic resolution data on the interface between monomers 65 , and does not enable the atomistic study of intact T4Ps. Furthermore, taking into account that the mechanism of T4Ps polymerization is still not fully understood, it may therefore be possible that in the assembled state, the pilin subunit α1-N domain diverges somewhat from its monomeric structure. In contrast to the N-terminal α1-N-derived segment of the 20-mer, the C-terminal α1-C-derived segment does not self-assemble and folds into a helical conformation, resembling its native conformation. In line with the role of α1-N in the polymerization of native T4P, and likely due to the higher hydrophobicity of α1-N as compared with α1-C, peptides derived from the former present a clear propensity to self-assemble, as opposed to peptides derived from the latter. Therefore, to obtain a short pilin-derived building block that can both self-assemble and display a native-like conformation, α1-N and α1-C derived segments can be conjoined in a single peptide ( Fig. 5 ). Further study of the obtained nanofibres may therefore reveal functionalities similar to those of the native GS T4P and can potentially lead to the development of new peptide-based bioelectronic materials. Our strategy may also be used in the design of other T4P-derived building blocks for mimetic nanostructured biomaterials, where specific segments with functional importance in other pilins are conjoined with an assembly-driving pilin-derived sequence."
} | 1,649 |
24113297 | null | s2 | 330 | {
"abstract": "Spider silk is a biomaterial with impressive mechanical properties, resulting in various potential applications. Recent research has focused on producing synthetic spider silk fibers with the same mechanical properties as the native fibers. For this study, three proteins based on the Argiope aurantia Major ampullate Spidroin 2 consensus repeat sequence were expressed, purified and spun into fibers. A number of post-spin draw conditions were tested to determine the effect of each condition on the mechanical properties of the fiber. In all cases, post-spin stretching improved the mechanical properties of the fibers. Aqueous isopropanol was the most effective solution for increasing extensibility, while other solutions worked best for each fiber type for increasing tensile strength. The strain values of the stretched fibers correlated with the length of the proline-rich protein sequence. Structural analysis, including X-ray diffraction and Raman spectroscopy, showed surprisingly little change in the initial as-spun fibers compared with the post-spin stretched fibers."
} | 270 |
32271573 | null | s2 | 331 | {
"abstract": "Many species of common bacteria communicate and coordinate group behaviors, including toxin production and surface fouling, through a process known as quorum sensing (QS). In Gram-negative bacteria, QS is regulated by "
} | 54 |
32198368 | PMC7083931 | pmc | 332 | {
"abstract": "The close replication of synaptic functions is an important objective for achieving a highly realistic memristor-based cognitive computation. The emulation of neurobiological learning rules may allow the development of neuromorphic systems that continuously learn without supervision. In this work, the Bienenstock-Cooper-Munro learning rule, as a typical case of spike-rate-dependent plasticity, is mimicked using a generalized triplet-spike-timing-dependent plasticity scheme in a WO 3−x memristive synapse. It demonstrates both presynaptic and postsynaptic activities and remedies the absence of the enhanced depression effect in the depression region, allowing a better description of the biological counterpart. The threshold sliding effect of Bienenstock-Cooper-Munro rule is realized using a history-dependent property of the second-order memristor. Rate-based orientation selectivity is demonstrated in a simulated feedforward memristive network with this generalized Bienenstock-Cooper-Munro framework. These findings provide a feasible approach for mimicking Bienenstock-Cooper-Munro learning rules in memristors, and support the applications of spatiotemporal coding and learning using memristive networks.",
"introduction": "Introduction Brain-inspired neuromorphic computing systems are attracting strong interest because of their massive parallelism, high energy efficiency, good error tolerance, and good ability to implement cognitive functions 1 – 6 . Hardware implementations of neuromorphic computing can take advantage of novel nanodevices that emulate the biological synapses with inherent learning functions 7 – 13 . The two-terminal memristor is widely recognized as a promising technology with which to mimic the biological synapse because of its functional resemblance to the biological counterpart 14 – 19 . The biorealistic realization of synaptic plasticity in the memristor is considered to be an important step toward realizing an artificial synapse with high accuracy. There have been many efforts to demonstrate basic synaptic learning functions using single and paired spikes, for example, long-term/short-term plasticity, spike-timing-dependent plasticity (STDP), and paired-pulse facilitation (PPF)/depression 20 – 27 . In fact, the stimulation mode of a spike train that contains plentiful spikes is a more general case than the single spike or paired spikes, and is produced by a neuron receiving multiple spikes from other connected neurons 28 . The information contained in a spike train allows specific advanced plasticity within a synapse that is referred to as spike-rate-dependent plasticity (SRDP) 29 . The Bienenstock-Cooper-Munro (BCM) learning rule is an important type of SRDP beyond the Hebbian learning rule and describes history-dependent synaptic modification. In the BCM framework, the high/low spike rate of a train can result in the potentiation/depression of the synaptic weight depending on whether the spike rate is higher than a threshold ( θ ) 30 – 33 . For the memristor-based artificial synapse, several groups have demonstrated BCM rules using rate-based presynaptic spikes, which have led to advances in the field 34 – 36 . These results show that the absolute change in the synaptic weight (i.e., the conductance change of the memristor, | Δ G c |) has a monotonic dependence on the spike rates in both the depression region (Δ G c < 0) and potentiation region (Δ G c > 0). However, such a monotonic change is different from the original BCM rule in neurobiology; that is, there should exist non-monotonic behavior (i.e., an enhanced depression effect (EDE)) in the depression region 30 – 33 , 37 , 38 . Additionally, previous memristor studies lack the following essential features: first, the lack of a multiplicative term between presynaptic and postsynaptic activities, and second the short-term modification 34 – 36 . This also marks a significant inconsistency with the biological BCM learning. According to a theoretical model of Pfister et al., it is expected that the use of triplet-STDP, instead of common rate-based presynaptic spikes, allows this issue to be solved 39 , 40 . Furthermore, the BCM rule can be generalized by the long-term triplet-STDP, thereby allowing higher-order spatiotemporal recognition in the visual cortex, for example, rate-based orientation selectivity 39 . Triplet-STDP means that a third spike, either presynaptic or postsynaptic, is introduced into the standard pair-STDP 33 , 40 – 42 . Importantly, in addition to the paired term contribution in pair spikes, a previous spike (presynaptic or postsynaptic) also causes the contribution of a triplet term in the triplet-STDP 33 , 41 , 42 . The relationship between the paired term and triplet term contributions provides the multiplicative correlations between presynaptic and postsynaptic activities, which is an essential requisite for BCM learning. There are two types of triplet-STDP in neuroscience: the first-spike-dominating model and last-spike-dominating model proposed by Froemke et al. and Wang et al., respectively 41 , 42 . Progress has been made in emulating these two types of triplet-STDP using first-order and second-order memristors 36 , 43 – 45 . However, the generalization from triplet-STDP to the BCM learning rule has not yet been experimentally demonstrated in memristors. Additionally, high-order spatiotemporal recognition that relies on generalized BCM learning rules has rarely been reported. The present work presents the demonstration of generalized BCM learning rules using the last-spike-dominating triplet-STDP in a WO 3−x -based second-order memristor. The second-order memristor has physical behavior similar to Ca 2+ dynamics in the bioneural network, which allows the emulation of rate-based plasticity naturally 16 , 34 , 46 . The EDE, which was typically missing in previous studies, is achieved using a long-term triplet-STDP scheme. Our experimental results are highly consistent with the mathematical model of the BCM framework in a biological system. Additionally, rate-based orientation selectivity is demonstrated on the basis of such a generalized triplet-STDP-based BCM learning rule, showing its strong potential in high-order spatiotemporal recognition.",
"discussion": "Discussion We demonstrated a generalized triplet-STDP-based BCM learning rule using a WO 3−x -based second-order memristor. Compared with the BCM rules realized by common rate-based presynaptic spikes, the EDE region missing in previous studies was found in our experimental data. A typical threshold sliding effect that depended on the learning history was also obtained. Furthermore, rate-based orientation selectivity was demonstrated in a feedforward network based on the generalized BCM framework in our memristors by simulation, which indicated its potential feasibility for high-order spatiotemporal patterns. It is noted that there are still certain limitations to a full implementation of the BCM learning at the synaptic level using memristors. For instance, the device physics and signal design may bring differences from the biological synapse, such as the spike-timing region, LTP/LTD window, and specific biological features. Further studies are still required to solve these above limitations toward a fully bio-mimetic BCM rule. It is believed that our study makes a progress towards the biorealistic mimicking of BCM learning rules in memristive synapses and paves the way for the application of memristors to spatiotemporal patterns in the future."
} | 1,876 |
38628862 | PMC11020090 | pmc | 334 | {
"abstract": "Plants engage in a variety of interactions, including sharing nutrients through common mycorrhizal networks (CMNs), which are facilitated by arbuscular mycorrhizal fungi (AMF). These networks can promote the establishment, growth, and distribution of limited nutrients that are important for plant growth, which in turn benefits the entire network of plants. Interactions between plants and microbes in the rhizosphere are complex and can either be socialist or capitalist in nature, and the knowledge of these interactions is equally important for the progress of sustainable agricultural practice. In the socialist network, resources are distributed more evenly, providing benefits for all connected plants, such as symbiosis. For example, direct or indirect transfer of nutrients to plants, direct stimulation of growth through phytohormones, antagonism toward pathogenic microorganisms, and mitigation of stresses. For the capitalist network, AMF would be privately controlled for the profit of certain groups of plants, hence increasing competition between connected plants. Such plant interactions invading by microbes act as saprophytic and cause necrotrophy in the colonizing plants. In the first case, an excess of the nutritional resources may be donated to the receiver plants by direct transfer. In the second case, an unequal distribution of resources occurs, which certainly favor individual groups and increases competition between interactions. This largely depends on which of these responses is predominant (“socialist” or “capitalist”) at the moment plants are connected. Therefore, some plant species might benefit from CMNs more than others, depending on the fungal species and plant species involved in the association. Nevertheless, benefits and disadvantages from the interactions between the connected plants are hard to distinguish in nature once most of the plants are colonized simultaneously by multiple fungal species, each with its own cost-benefits. Classifying plant–microbe interactions based on their habitat specificity, such as their presence on leaf surfaces (phyllospheric), within plant tissues (endophytic), on root surfaces (rhizospheric), or as surface-dwelling organisms (epiphytic), helps to highlight the dense and intricate connections between plants and microbes that occur both above and below ground. In these complex relationships, microbes often engage in mutualistic interactions where both parties derive mutual benefits, exemplifying the socialistic or capitalistic nature of these interactions. This review discusses the ubiquity, functioning, and management interventions of different types of plant–plant and plant–microbe interactions in CMNs, and how they promote plant growth and address environmental challenges for sustainable agriculture.",
"conclusion": "Conclusion and future directions Plant–plant and plant–microbe interactions are extremely complex. More research is required to fully understand these interactions and to clarify how they can be used in agriculture for things like nutrient acquisition, improving disease resistance, and stress tolerance. To reveal the dynamic microbial colonization functions, advanced characterization techniques and large-scale experimental approaches are required. Numerous microbes have still not been fully characterized at the physiological and molecular levels. A fundamental question is whether CMNs have a greater (positive or negative) impact on plant performance and other ecosystem services than effects that result from mycorrhizal fungi alone. The absolute characterization of molecules facilitating beneficial microbes and inducing resistance against pathogens is a significant challenge because plant root exudates are made up of thousands of different substances. Numerous studies are still required to fully understand the variety, makeup, purposes, and mechanisms underlying the exchange of VOCs. We know very little about how AMF distributes infochemicals and nutrient resources within their CMNs or how plants compete with one another for the limited nutrient resources that are available for their CMNs. It is now possible to efficiently monitor microbial species that interact with plants due to the development of biotechnological tools. These may help us achieve our objective of maintaining the agricultural ecosystem’s sustainability. The benefits of CMNs within agroecosystems are, of course, not limited to the supply of plant nutrients. We identified a number of other important ecological functions of CMNs in soil. These include recycling of nutrients, prevention of nutrient losses, contribution to soil structure, food for other organisms, and mycorrhizal fungal networks acting as hyphal highways for bacterial dispersion. One of the most pressing concerns in agriculture is soil “health” and structure. CMNs maintain soil quality and health via three aspects: soil structure, plant physiology, and ecological interactions. AMF deposit glomalin between the outer hyphal walls and adjacent soil particles to form micro-aggregates and further macro-aggregates, thus forming the backbone for soil aggregation. CMNs hold huge significance for our planet and society and thus play an essential role in the formation and maintenance of global ecosystems. They also have great potential for exploitation to facilitate a variety of sustainability programs in agriculture, conservation, and restoration, particularly relevant in the context of global climate change and the depletion of natural resources. It is clear that CMNs are an essential component of ecosystem biodiversity and also deliver, through their roles in plant nutrition and protection, significant ecosystem services that have the potential to play an important role in sustainability agendas. However, significant knowledge gaps remain covering the multitude of interactions between plants, fungi, people, and the environment. This editorial provides an overview of the relevance and potential roles of mycorrhizal fungi toward achieving global goals in sustainability, conservation, and their significance within society, and highlights key directions for future research.",
"introduction": "Introduction The exploration of various interactions between members of the same or distinct kingdoms is of paramount importance for ecological stability, nutrient cycling, and the efficient management of an ecosystem. Most land plants are associated with mycorrhizal fungi for their nutritional demand, development, and increased resistance to stress ( Compant et al., 2019 ; Bhatt et al., 2020 ). The history of interactions between plant–plant and plant–microbes is as old as plant colonization on Earth. In both natural and agricultural ecosystems, the invaded interactions may be positive or negative, depending on the mode of interest ( Kuebbing and Nuñez, 2015 ). As such, plants make the obligatory interactions necessary for their existence by suppressing the growth of others or by sharing resources for the benefit of each other. The improvement of seedlings ( Seiwa et al., 2020 ), impact on the plant and microorganism community ( Kadowaki et al., 2018 ; Teste et al., 2020 ), activation of plant defense responses ( Babikova et al., 2013 ; Song et al., 2014 ), and interplant nutrition ( Bücking et al., 2016 ; He et al., 2019 ; Fang et al., 2021 ) are among the most specific responses regulated by such interactions. Subsequently, the release of volatile compounds and the transfer of essential mineral nutrients to plants via mycorrhizal fungi are good examples of positive plant–plant and plant–microbe interactions, respectively ( Alaux et al., 2021 ; Ohsaki et al., 2022 ). Mycorrhizal fungi improve plant nutrient uptake and receive plant carbohydrates, interacting for the net benefit of both parties ( Bücking et al., 2016 ; Gilbert and Johnson, 2017 ). On the flip side, the mechanism of allelopathy and necrotrophy in the colonizing plants caused by saprophytic microbes is an example of negative interactions ( Dicke and Baldwin, 2010 ; Friedman, 2017 ). Most of these interactions affect the survival and behavior of connected plants and potentially influence competitiveness patterns at local and regional scales ( Bücking et al., 2016 ). Therefore, deciphering these interactions would greatly improve our understanding of how these interactions affect terrestrial ecosystems and the potential for feedback on global change. AMF are ancient fungal organisms that engage in mutualistic symbiotic relationships with the vast majority of land plant species for nutritional exchange ( Figure 1 ). AMF improve plant nutrition by accessing nutrient sources that are otherwise inaccessible to roots ( Wipf et al., 2019 ; Andrino et al., 2021 ). The great majority of AM fungi build extensive colonization networks with numerous neighboring plants for their carbon or nutritional supplies through CMNs, which are not host-specific ( Rhodes, 2017 ; Simard, 2018 ; He et al., 2019 ; Bacha et al., 2023 ). CMNs play a crucial role in plant–plant interactions by generating warning signals and activating defense information ( Song et al., 2015 ; Gilbert and Johnson, 2017 ; Oelmüller, 2019 ). Thus, evidence supports the multifunctionality of CMNs, which are involved in different types of AM interactions across ecosystems ( Gilbert and Johnson, 2015 ). Figure 1 Impacts of plant integration into CMN and mycorrhizal colonization. Plants integrated into CMNs can experience enhanced nutrient acquisition and absorption, improving their overall growth and survival. Mycorrhizal colonization plays a vital role in increasing a plant’s resistance to various environmental stresses and toxic substances. CMNs facilitate interplant communication, enabling plants to exchange signals and respond to changes in the environment, leading to improved ecosystem resilience and biodiversity. Similarly, CMNs can actively participate in improving plant resistance and tolerance to abiotic stress ( Plouznikoff et al., 2016 ; Bacha et al., 2023 ). Recent research indicates that CMNs influence the survival, fitness, behavior, and competitiveness of numerous fungal and plant species that interact and “communicate” via these networks. CMNs enable the fungus to establish connections with several trade partners, ensuring a consistent carbon supply for the fungus ( Fellbaum et al., 2014 ). This is particularly important when one host plant becomes unable to transmit resources to its fungal partner owing to disease, herbivore damage, or premature senescence ( Figure 2 ). From an ecosystem perspective, exploring the possible interactions of plant–plant and plant–microbes mediated by CMNs for nutritional strategies is a critical component for enhancing ecosystem services effectively. To date, numerous studies have been conducted to reveal these interaction processes; however, many paradoxical results still exist, and the debate about these issues has never ceased so far. Thus, it is of paramount importance to shed light on the most recent advances in literature and highlight the potential research question gaps for guiding the upcoming studies in the specified areas. Figure 2 Schematic diagram of interplant signaling via CMNs in plant–plant interaction. In this mutualistic association, the roots of both plants are colonized by hyphal threads extending from the fungal colonization within their roots. Through this interconnected network, plants transmit signals, which can be chemical compounds like hormones or volatile organic compounds (VOCs), in response to stimuli or environmental changes ( Rasheed et al., 2023 ). These signals travel through the CMNs via the mycorrhizal hyphal threads, spanning the soil and allowing for communication between the plants. The signals move from one plant to another through the interconnected hyphal threads. The roots of the receiving plant detect and receive these transmitted signals through specific receptors present in their root systems ( Abdul Malik et al., 2020 ). These receptors recognize the signals emitted by the sending plant and initiate a response. The response may involve various physiological, biochemical, or molecular changes, such as alterations in growth patterns, nutrient uptake, defense mechanisms, or adaptive behaviors. This interaction between the plants creates a feedback loop where both plants can generate signals that are transmitted through the CMNs, allowing for reciprocal communication and response. This process underscores the vital role of the CMNs and mycorrhizal fungi in facilitating interplant signaling and promoting plant–plant interactions in natural environments."
} | 3,162 |
36234583 | PMC9565409 | pmc | 335 | {
"abstract": "Convolutional neural networks (CNNs) have been widely used in image recognition and processing tasks. Memristor-based CNNs accumulate the advantages of emerging memristive devices, such as nanometer critical dimensions, low power consumption, and functional similarity to biological synapses. Most studies on memristor-based CNNs use either software models of memristors for simulation analysis or full hardware CNN realization. Here, we propose a hybrid CNN, consisting of a hardware fixed pre-trained and explainable feature extractor and a trainable software classifier. The hardware part was realized on passive crossbar arrays of memristors based on nanocomposite (Co-Fe-B) x (LiNbO 3 ) 100−x structures. The constructed 2-kernel CNN was able to classify the binarized Fashion-MNIST dataset with ~ 84% accuracy. The performance of the hybrid CNN is comparable to the other reported memristor-based systems, while the number of trainable parameters for the hybrid CNN is substantially lower. Moreover, the hybrid CNN is robust to the variations in the memristive characteristics: dispersion of 20% leads to only a 3% accuracy decrease. The obtained results pave the way for the efficient and reliable realization of neural networks based on partially unreliable analog elements.",
"conclusion": "4. Conclusions In summary, we fabricated and studied the crossbar arrays of nanocomposite-based (Co-Fe-B) x (LiNbO 3 ) 100−x memristors. Memristors in a single crossbar array demonstrate negligible c2c variations, while d2d variations are more pronounced, which was attributed to the considerable impact of the crossbar busses’ resistances to the total resistance of the structure. Using the nanocomposite-based crossbar arrays, we implemented a hybrid CNN, consisting of a hardware feature extractor with one/two kernels and a software classifier. The two-kernel CNN was able to classify the binarized Fashion-MNIST dataset with an accuracy of ~84%. The performance of the hybrid CNN is comparable to the full software and full hardware (memristive) systems, while the number of trainable parameters for the hybrid CNN is substantially lower. Moreover, the hybrid CNN is shown to be robust to the variations in memristive characteristics. The obtained results raise hope that enhanced performance may be achieved for any given image classification task in the future, if some expended set of fixed kernels is created for the hybrid CNN.",
"introduction": "1. Introduction Memristor-based neuromorphic computing systems (NCSs) provide a fast, high-computational, and energy-efficient approach to neural network (NN) training and solving cognitive problems (pattern and speech recognition, big data processing, prediction, and so on) [ 1 , 2 ]. Memristors could be organized in large crossbar arrays (with critical dimensions down to 6 nm [ 3 ]) to perform vector–matrix multiplication in a natural one-step method by weighted electrical current summation (according to the Ohm’s and Kirchhoff’s laws) [ 4 ]. In contrast, being the most massively parallel operation in NN learning and inference, vector–matrix multiplication is extremely time- and energy-expensive in traditional von Neumann architectures [ 2 ]. Owing to this difference, memristor-based NCSs are of high interest. Memristors have already been successfully implemented for diverse NCS realizations, and such schemes as perceptrons [ 5 , 6 ], spiking [ 7 , 8 ], or long short-term memory [ 9 ] networks and others (including NN circuits of Pavlov’s associative memory) [ 10 , 11 , 12 ] have been demonstrated. Most of these NCSs are usually trained by various types of gradient descent learning algorithm, the hardware realization of which is challenging because of unreliable cycle-to-cycle (c2c) and device-to-device (d2d) variations of memristive devices [ 2 ]. Several approaches have been proposed to partially mitigate these problems, including reservoir computing [ 13 ] and convolutional [ 14 , 15 ] NNs. The latter one is of particular interest as it allows to reduce the number of required weights (i.e., memristors) compared with fully connected NNs and, at the same time, demonstrates excellent performance in object recognition and image processing [ 15 , 16 ]. Convolutional NNs (CNNs) consist of two main parts: the feature extractor (convolutional layers) and classifier (fully connected layers). Convolutional layers extract feature maps from the input images by applying filters (kernels) of different dimensions, which allows decreasing the number of inputs. Most studies on memristor-based CNNs use either software models of memristors to emulate both parts of CNNs [ 16 , 17 , 18 , 19 , 20 , 21 ] or fully hardware parts of CNNs [ 14 , 22 , 23 ]. However, it significantly complicates the evaluation of the memristor-based convolutional layer efficiency, which should not be neglected. Generally, CNNs are prone to learning untrustworthy features and overfitting. An illustrative example of this case includes a CNN trained to classify images of huskies and wolves, which instead learned background features of the images, such as the presence of snow [ 24 ]. This highlights the importance of so-called explainability of the NNs. One possible way to control the feature extractor weights is to train the convolutional layers ex situ via a traditional computing system, and then transfer them to the memristors (hybrid training) [ 22 ]. Multiple software algorithms may help visualize and study the extracted features in this case [ 25 ]. However, this approach implies that a high-cost training process alongside with its verification should be performed in the software before the transfer of the convolutional kernel weights to the hardware. Moreover, the convolutional kernels may need to be retrained if the classification problem is changed. In this work, we propose a general approach to the implementation of a memristor-based CNN—a hybrid network, consisting of a hardware fixed pre-trained and explainable feature extractor and a trainable software classifier. Unlike the hybrid training process, where the weight update of a hardware memristor-based NN is made according to the software ex situ training results at each training cycle [ 22 ], in our case, no additional training of the CNN memristive part is needed. The hybrid CNN possesses the advantages of both hardware systems in terms of energy and computational efficiency and software systems in terms of architectural flexibility. The usage of convolutional layers with universally recognized fixed kernels (horizontal and vertical) gives hope that such a hybrid network would be eligible for any given image classification problem, and only the weights of its classifier part would require some fine-tuning. Here, we test this approach on the Fashion-MNIST (F-MNIST) image recognition problem [ 26 ]. The main goal of this work is to estimate the efficiency of the fixed memristor-based convolutional layers compared with the ideal software trainable ones.",
"discussion": "3. Results and Discussion In this work, the LiNbO 3 (LNO)-based memristors were used, as they are of emerging interest [ 28 , 29 , 30 ], especially those with embedded metal nanogranules [ 31 ]. In our studies, the capacitor metal/nanocomposite/metal (M/NC/M) structures based on (Co-Fe-B ) x ( LiNbO 3 ) 100 − x NC were fabricated by ion-beam sputtering with a metal content x ≈ 8–25 at.% [ 31 ]. The NC films, along with the metal nanogranules of 3–6 nm, contained a large number of dispersed Co (Fe) atoms (up to ~10 21 –10 22 cm −3 ). At some optimal x ≈ 8–15 at.%, the M/NC/M structures manifest stable resistive switching (RS) through a multifilamentary mechanism [ 32 ]; demonstrate high endurance, long retention, and multilevel RS [ 33 , 34 ]; and can be successfully used in NCSs [ 27 , 35 ]. In this work, we fabricated NC structures with a thin built-in LNO layer near the bottom electrode (i.e., structures like M/NC/LNO/M), which plays a critical role in the realization of stable RS [ 32 ]. The arrays of the NC memristors were fabricated in a 16 × 16 crossbar architecture ( Figure 1 a). Then, their characteristics were studied in order to verify the eligibility of the NC crossbar memristors for the hardware realization of different 3 × 3 convolutional kernels. The results of transition electron microscopy (TEM) provided a descriptive picture of the layer thickness ( Figure 1 b) and composition (elemental maps are presented in Figure S4 ) for a single memristive element from a 16 × 16 crossbar array. High-resolution images ( Figure S5 ) confirmed the presence of a ~10 nm thick pure LNO layer near the bottom electrode. Figure 1 c presents the current–voltage ( I – V ) characteristics of all 18 memristors (9 rows and 2 columns) used in this work, five cycles for each one. As can be seen in Figure 1 c, while c2c variations for these memristors are negligible, the d2d variations are more pronounced. Two groups of devices could be selected from the I – V curves (shaded in different colors in Figure 1 c). Each group represents devices from different columns of the crossbar array. The variations in the resistive switching voltage of these groups are associated with different resistances of crossbar busses, which act as load resistances, and some additional voltage drops on them. The bus resistances should be reduced to decrease the d2d variations in a crossbar array. However, as shown below, all of these devices can operate as equivalent parts of a convolutional kernel. Another important memristive characteristic for the convolutional layer implementation is retention time—after the weights of all memristors are adjusted to represent the chosen kernel, they should not vary. As can be seen from Figure 1 d, the resistance drift from both initial states (high resistance state, R off , and low resistance state, R on ) is negligible compared with their difference R off — R on . It should be noted that the resistance values of the obtained memristors are not high enough (≤1 kΩ), probably because of the small thickness of the LNO layer in the M/NC/LNO/M structures under study. Figure 2 a illustrates the proposed NN architecture. The original F-MNIST images were binarized in order to simplify the implementation of the hardware part. The features were then extracted with either a horizontal/vertical filter or both at once (the example of the extracted features is presented in Figure 2 a). Then, the obtained feature matrices were flattened, normalized, and fed to the fully connected classifying layers (676 input neurons in the case of the CNN with one filter and 1352 in the case of the CNN with two filters). Figure 2 b demonstrates the hardware feature extractor implementation. The image was divided into 3 × 3 patches; each patch was then flattened and fed to the crossbar array (i.e., the corresponding voltages were applied, 1 V for “1” pixels of the patch and 0 V for “0” pixels). The crossbar array acted either as a horizontal or a vertical kernel, i.e., the weights of nine memristive devices were adjusted to represent the chosen normalized and flattened 3 × 3 kernel. In order to obtain negative weights of the kernels, two columns of the crossbar array were used (i.e., nine rows-inputs and two columns-outputs were used in this work, as specified in the figure), and the resulting output current equaled the difference between the currents from both columns. A more detailed discussion of the feature extractor implementation can be found in Section 2 . The proposed NN architecture ( Figure 2 a) was additionally simulated in the software for subsequent comparison with the hybrid NN results. Figure 3 a presents the accuracy of the one-filter NN during training, estimated on a validation dataset (the discussion of the initial dataset portioning can be found in Section 2 ). Figure 3 b demonstrates the results for the two-filter NN. Three main conclusions can be drawn from these figures. Firstly, the binarization of the images does not lead to an accuracy decrease, so such a simplification can be done. Secondly, the CNNs with fixed filters do not concede to the trainable ones dramatically. Finally, during the first training epochs, the trainable filter usage leads to high accuracy variations; thus, such CNNs concede to the CNNs with fixed filters at the beginning of the training process. These results authorized the creation of a hybrid CNN. Figure 4 a,b compare the results obtained from the hybrid NNs and full software systems with one and two filters correspondingly. The best simulation results from Figure 3 a,b were chosen to make the comparison under stringent conditions. It can be seen that the hybrid NNs with one filter considerably concede to the simulated NN with a trainable filter. However, the results after 100 epochs for the two-filter hybrid NN are comparable to the simulation results. A more complete set of features, obtained using both horizontal and vertical filters, was generated for each image in this case, leading to a higher accuracy score. This result raises hope that some expanded set of filters may be created, which would extract all of the most important features from the images without additional training of the filters, thus leading to the creation of a generic hybrid NN with enhanced classification accuracy. Generally, the F-MNIST dataset was developed in order to replace the conventional MNIST digit dataset. Some modern software NNs can classify the MNIST digit dataset with an accuracy >99%, which makes it too simple for the NN performance evaluation [ 36 ]. In contrast, quite elaborate software NN architectures are required in order to surpass 90% accuracy for the F-MNIST dataset [ 26 , 37 ]. Meanwhile, the number of trainable parameters for such NNs equals 500–700 k (grayscale F-MNIST). The same goes for the memristor-based CNNs, e.g., sixteen 9 × 9 convolutional kernels were needed to reach ~87% accuracy (binarized F-MNIST) [ 16 ]. Another memristive CNN example demonstrated ~93% with ~3.5 M parameters (grayscale F-MNIST) [ 38 ]. In our study, the test classification accuracy of the hybrid CNN with two filters equaled ~ 84%, while the number of trainable parameters equaled ~44 k. This is a high enough accuracy value for a hybrid system with such a small number of parameters. Most mistakes were made for the classes, which are almost indistinguishable in the case of the binarized images ( Figure 4 c). The reduced number of trainable parameters leads to a less sophisticated and more robust training process, so introduction of the memristors does not dramatically decrease the classification accuracy. Moreover, the influence of the variability introduction to the hardware CNN part on its performance was studied. The weights of the hardware filters (i.e., resistive states of the memristive crossbar) in the hybrid CNN are set once and are not tuned in the future. Thus, memristors should remain in their initial resistive states to produce a reliable output from the feature extractor. It can be seen in Figure 1 d that our memristors have good retention, i.e., their states are not changed in time even if no external voltage is applied. However, it is known that memristive systems are in general prone to variations in characteristics [ 2 ]. It is probable that, after a longer time, some variations may appear as a result of internal microscopic degradation or external impacts. Therefore, we examine the influence of such possible variations on the classification accuracy. In order to simulate the variation introduction to the CNN, each output current from the memristive convolutional layer was chosen from a normal distribution, for which the mean value equaled the experimental results and the coefficient of variation was chosen from 0 to 100%. Only a high coefficient of variation (e.g., 100%) led to a considerable degradation of the training process and a decrease in the test dataset accuracy. All obtained data are summarized in Table S2 and demonstrate the robustness of the hybrid CNN to the variations in the memristor characteristics: dispersion of 20% leads to only a 3% decrease in accuracy."
} | 4,039 |
34071134 | PMC8197075 | pmc | 336 | {
"abstract": "The triboelectric nanogenerator (TENG) is a newly arisen technology for mechanical energy harvesting from the environment, such as raindrops, wind, tides, and so on. It has attracted widespread attention in flexible electronics to serve as self-powered sensors and energy-harvesting devices because of its flexibility, durability, adaptability, and multi-functionalities. In this work, we fabricated a tubular flexible triboelectric nanogenerator (TF-TENG) with energy harvesting and human motion monitoring capabilities by employing polydimethylsiloxane (PDMS) as construction material, and fluorinated ethylene propylene (FEP) films coated with Cu as the triboelectric layer and electrode, serving in a free-standing mode. The tube structure has excellent stretchability that can be stretched up to 400%. Modifying the FEP films to obtain a superhydrophobic surface, the output performance of TF-TENG was increased by at least 100% compared to an untreated one. Finally, as the output of TF-TENG is sensitive to swing angle and frequency, demonstration of real-time monitoring of human motion state was realized when a TF-TENG was worn on the wrist.",
"conclusion": "4. Conclusions In summary, we designed a tubular flexible TENG with mechanical energy collection and human motion monitoring functions. PDMS was used as the construction material to provide 400% stretchability, and FEP film and pure water constituted the triboelectric layers of the TF-TENG. Through superhydrophobic surface treatment, the output performance of the TF-TENG doubled due to the increased surface charge density, and also the device became easier to be triggered. The TF-TENG is lightweight and can be comfortably worn on the wrist, by which different motion states, such as walking, jogging, and running, can be monitored and distinguished in real time, demonstrating its application potential in the field of smart sports. Currently, the TF-TENGs are wired to the external electrometer for signal recording. A fully integrated wireless system with the capabilities of signal recording, processing, analysis, and wireless data transmission will further elevate the adaptability and conveniences of this technology to fit into diversified application scenarios.",
"introduction": "1. Introduction Recently, with the rapid development of a new generation of digital information technology, the implementation of various portable and wireless electronic devices has changed the way people live [ 1 , 2 , 3 , 4 ], and one consequence is that the energy supply of these electronic devices has become a critical challenge. As a new energy harvesting technology, triboelectric nanogenerators (TENGs) are based on the coupling effect of triboelectrification effect and electrostatic induction [ 5 ], which was first invented by Wang and coworkers in 2012 [ 6 ] and was widely used to convert randomly distributed, irregular, weak, and low-frequency mechanical energy into electric power [ 7 , 8 , 9 , 10 , 11 , 12 ]. The TENGs have four fundamental working modes, namely contact-separation mode, sliding mode, single-electrode mode, and free-standing mode [ 13 , 14 , 15 , 16 ]. With proper working modes, TENGs can fit into different application scenarios, to serve as, for example, micro/nano energy harvesters [ 17 , 18 , 19 ] and self-power sensor systems [ 20 , 21 , 22 , 23 ]. The generation of triboelectric charges during the triboelectrification process is highly dependent on the two contact materials’ relative ability to gain or lose electrons, or their relative positions in the triboelectric series [ 24 ]. Recently, many efforts have been made to develop or modify materials to promote the performance of TENGs [ 25 , 26 , 27 ]. Most TENGs are based on a solid–solid contact, causing physical wear of the device with surface damages, which would lead to output performance degradation with time. Instead, liquid–solid interface-based TENGs can provide higher durability, flexibility, and a larger effective contact area for a higher output performance [ 28 , 29 , 30 ]. Lately, many wearable TENGs have been reported [ 31 , 32 , 33 , 34 , 35 , 36 ] for energy harvesting from the body or self-powered sensing for body movements; however, there is still lots of room for improving their wearing comfortability, lowing manufacturing cost, and extending application scenarios. In this work, we fabricated tubular flexible TENGs (TF-TENGs) with excellent stretchability and super hydrophobicity that were introduced at liquid–solid interfaces for triboelectrification in a free-standing mode. The effects of swing angles and frequencies on the output performance of TF-TENGs were first checked by using Cu pellets as the other triboelectric layer rolling in the tube. Then, after the treatment of super hydrophobicity, pure water could roll easily in the TF-TENGs, demonstrating a 100% performance improvement compared to the untreated one. Finally, a TF-TENG was wrapped on the wrist for human motion state detection, revealing its application potentials in the field of smart sports.",
"discussion": "3. Results and Discussion 3.1. As-Fabricated TF-TENG and Its Strechability Figure 2 a sketches the structure diagram of a TF-TENG, and Figure 2 b is the photograph of an as-fabricated TF-TENG with a length of 24 cm and a diameter of 14 mm. It shows that the PDMS tube is semitransparent and two Cu-FEP films are attached separately on its inner surface. We first tested the stretchability of the structure, which was measured by using a homemade sample stage as illustrated in Figure 2 c,d. Due to the limitation of movable range of the sample stage, we cut the PDMS tube and clamped two ends of the tube into the sample stage with a suspended length of 2 cm. It can be stretched up to 8 cm, 400% of the original length, without mechanical fracture, which demonstrates its excellent stretchability. 3.2. Working Principle of TF-TENG The working principle of the TF-TENG is depicted in Figure 3 , which is based on the freestanding mode. First, as an example to simplify the case, we used Cu pellets to work as the other triboelectric layer rolling on the FEP films. After the first cycle of rolling, as shown in Figure 3 <I>, due to the triboelectrification effect, positive charges are generated on the surfaces of the Cu pellets and the quantity of these positive charges is equal to the sum of the negative charges generated on two FEP films. At the same time, charges that are induced by electrostatic induction in the left and right Cu electrodes are equal in number while opposite in sign. Then, when the positively charged Cu pellets roll to the right FEP film by external force, electrons flow from the left electrode to the right electrode, generating current from the right to the left through the external load ( Figure 3 <II>). Until all the Cu pellets move to the right side, the transferred charges between the two electrodes will reach the maximum, as shown in Figure 3 <III>. Then, a backward moving of these Cu pellets from right to left results in a reverse current in the external load ( Figure 3 <IV>). When the Cu pellets reach the original position on the left FEP film, all of the negative charges in the right electrode will be driven to the left electrode ( Figure 3 <I>). By cycling the moving process of the Cu pellets between the surfaces of two FEP films, an alternating current is generated. 3.3. Output Performance Characterization of the TF-TENG In the as-fabricated TF-TENG, two Cu-FEP films with a size of 10 cm × 1.2 cm were attached to the PDMS tube inner surface with a separation distance of 2 cm, and Cu pellets with a diameter of 2 mm were first enclosed into the PDMS tube as the freestanding triboelectric layer to characterize the output performance of the TF-TENG. Figure 4 shows the schematic illustration of the experimental setup for performance characterization. To introduce a swing motion into the TF-TENG, the TF-TENG was first placed on an acrylic plate. One end of the plate was fixed, and the other end could swing up and down, which was controlled by a motorized linear stage (MTN300CC, Newport) that was connected via a setup of a wire and a fixed pulley. The swing angle α of the device is defined as the angle between the orientation of the acrylic plate and the horizontal line, which is indicated in Figure 4 . The output characteristics of open-circuit (OC) voltage, short-circuit (SC) current, and transferred charges of the TF-TENG were measured at a swing angle of 5° and a swing frequency of 0.25 Hz. As shown in Figure 5 a, the OC voltage of the FT-TENG is about 6.5 V. The quantity of transferred charges and SC current illustrated in Figure 5 b are 1.8 nC and 5 nA, respectively. Figure 5 c shows the relationship between the number of Cu pellets and the electric output performance of the TF-TENG. When the number of Cu pellets increases from 1 to 16, the OC voltage increases from 1.5 V to 24.1 V, and the SC current also increases from 1.2 nA to 10.5 nA, which shows a certain degree of linear relationship. This is because when the number of Cu pellets increases, the effective contact areas between the freestanding layer and the surface of the FEP film increase, which increases the quantity of transferred charge correspondingly with an approximately linear relationship. In addition, the influence of different swing angles on the electric output of the TF-TENG was investigated. When the swing angle increases from 5° to 26°, the OC voltage of the device is almost the same, but the SC current increases significantly, as shown in Figure 6 a,b. In this case, since the quantity of transferred charges is the same when the same number of Cu pellets are used, the OC voltage does not change as it is determined by the surface charge density and the electrode separation distance. The rolling speed of the Cu pellet is positively correlated with the swing angle, which leads to more electric charges passing through the external load per second when the swing angle increases, resulting in a larger SC current. The SC current under a swing angle of 26° is enlarged in Figure 6 c to reveal its good reproducibility. The relationship between the working frequency and the output performance has also been studied, as exhibited in Figure 7 . Under the condition that the other influencing factors remain unchanged (swing angle, number of Cu pellets, electrode separation distance, etc.), when the swing frequency increases from 0.158 Hz to 0.5 Hz, it can be seen that the quantity of transferred charges is about 8 nC and stays the same, while the SC current increases from 10 nA to 20 nA. The quantity of charge transferred remains unchanged because the effective contact area between the triboelectric layer and the FEP film has not changed. Therefore, an increase in frequency will only lead to an increase in the speed of charge transfer and thus an increased SC current. In order to evaluate the stability of the device, we measured the SC current of the TF-TENG for 3500 continuous working cycles with a swing angle of 15° and a swing frequency of 2 Hz, as shown in Figure 8 . The output current almost maintains at a constant value, indicating that the TF-TENG has good stability and durability. 3.4. Superhydrophobic Surface Treatment of the TF-TENG The Cu pellets with a certain weight cannot roll easily in a curved tube when the TF-TENG is worn on the human body. Hence, pure water was injected into the PDMS tube to replace Cu pellets and serve as the triboelectric layer material. Pure water has good shape adaptability and can flow under a small deflection angle, which can perfectly solve the problem met by Cu pellets to realize lightweight and more sensitive TF-TENGs. As shown in Figure 9 a, when 2 mL pure water was injected into the PDMS tube, an OC voltage of 4 V was obtained. To further elevate the responsibility of the device under a small mechanical disturbance, the surface of the FEP films were treated with NeverWet TM multi-surface liquid repelling treatment (Rust-Oleum, USA) to realize a superhydrophobic surface. After the treatment, the OC voltage of the TF-TENG was increased to 8 V, as shown in Figure 9 b, which is double the output of the untreated one. To understand the improvement in performance, we checked the surface of the FEP film. First, we measured the contact angle of the FEP film before and after the superhydrophobic treatment, with the results shown in Figure 10 a,b. It is clear that the contact angle of the pure water on the FEP film surface is increased from 57.7° to 122.5°, demonstrating a good superhydrophobic effect. To further check the surface morphology with optical microscope, Figure 10 c,d show that for an untreated FEP film, the surface is almost clear, while for a treated one, there are randomly distributed microparticles on the surface that were introduced by the sprayed NeverWet TM solution, which introduces the superhydrophobic behavior [ 38 ]. These almost uniformly distributed silica microparticles also increase the surface roughness and dielectric constant, which simultaneously further improve the triboelectric charge density on the FEP film surface [ 39 , 40 , 41 , 42 ], and thus the output performance of the TF-TENG. 3.5. Application of the TF-TENG in Human Motion Monitoring Generally, when people move at different speeds, the state of motion, including the frequency and amplitude of arm swings, is different. Accordingly, to prove the capability of the TF-TENG as a sensor for human motion monitoring, we wrapped the device around the wrist, as shown in Figure 11 a, and recorded the electrical signal changes when the human body is in different motion states under a natural swing of the arm, as shown in Figure 11 b–d. First of all, under different motion states, the motion frequencies are distinguished from each other, and it is also observed that the output SC current increases as the motion frequency increases, which makes the difference more remarkable. When the wearer swung the arm at 0.8 Hz, the measured SC current of the TF-TENG was 4 nA, while when the wearer changed to swing the arm at 1.8 Hz and at 2.7 Hz, the SC current increased to 10 nA and 20 nA, respectively. This means that the TF-TENG is sensitive enough to capture different body motions and can be used for real-time analysis of the human motion state. Figure 12 shows the recorded signal of SC current when the human body is in different motion states, including walking, jogging, and running. The frequency and amplitude of the sigal both change with different motion states; thus, it is easy to distinguish between them. The demonstration reveals that the TF-TENG can be used as a self-powered sensor for human motion monitoring, and expands the TENG’s application in self-powered sensing systems for smart sports."
} | 3,703 |
35425330 | PMC8979104 | pmc | 337 | {
"abstract": "It has been accepted generally that it is necessary to obtain the so-called surface superhydrophobicity on intrinsically hydrophobic materials. However, recent experiments have indicated that it could be possible to prepare superhydrophobic surfaces on intrinsically hydrophilic materials by creating adequate roughness. In this work, such a strategy for surface superhydrophobicity on hydrophilic materials with an intrinsic contact angle less than 90° was demonstrated thermodynamically based on a proposed 2-D analytical model. In particular, different (trapezoidal, vertical and inverse-trapezoidal) microstructures were employed to analyze their wetting states such as composite and noncomposite and superhydrophobic behavior as well as the previous corresponding experimental observations. Based on the thermodynamic calculations, it was demonstrated that for an overhang microstructure, intrinsic contact angle, which was restricted by the sidewall angle of micropillars, was not an independent parameter to affect superhydrophobicity. Furthermore, an overhang microstructure was critical to realize the transition from hydrophilicity to superhydrophobicity, and for such a transition, the sidewall angle should be less than the intrinsic contact angle where a positive free energy barrier could support the liquid/vapor interfaces and separate the Wenzel and Cassie states on such hydrophilic surfaces. Most importantly, it was found that for such hydrophilic surfaces, generally, the free energy of the noncomposite or Wenzel states were lower than that of the composite or Cassie states for those trapezoidal, vertical and inverse-trapezoidal microstructures, implying that once a noncomposite state was formed, it can hardly become a composite state, or in other words, even if superhydrophobic behavior was possible, it could be temporary or unstable.",
"conclusion": "4. Conclusions The possibility for the surface superhydrophobicity on hydrophilic materials with an intrinsic contact angle less than 90° was investigated thermodynamically based on a 2-D model. In particular, different (trapezoidal, vertical and inverse-trapezoidal) microstructures were employed to analyze their wetting states such as composite and noncomposite and superhydrophobic behavior as well as the previous corresponding experimental observations. The results show that for a overhang microstructure, intrinsic contact angle was not an independent parameter to affect superhydrophobicity. Furthermore, a overhang microstructure was critical to realize the transition from hydrophilicity to superhydrophobicity, and for such a transition, the sidewall angle should be less than the intrinsic contact angle where a positive free energy barrier could support the liquid/vapor interfaces and separate the Wenzel and Cassie states on such hydrophilic surfaces. Most importantly, it was found that for such hydrophilic surfaces, generally, the free energy of the noncomposite or Wenzel's states were lower than that of the composite or Cassie's states for those trapezoidal, vertical and inverse-trapezoidal microstructures, implying that once a noncomposite state was formed, it can hardly be become a composite state, or in other words, even if superhydrophobic behavior was possible, it could be temporary or unstable.",
"introduction": "1. Introduction It is well known that the wettability of solid surfaces is not only dependent upon their chemical compositions but also closely related to the micro/nano-structures on their surfaces. 1–3 Superhydrophobic surfaces with large water contact angles (CAs) and small contact angle hysteresis (CAH) have attracted strong interest in various industries due to their unique liquid-shedding or droplet-sliding properties over the past two decades. 4–8 The excellent wettability of these surfaces has shown wide potential applications, such as self-cleaning glasses, biological scaffolds, microfluidics, lab-on-a-chip devices, coatings for automotive and aerospace vehicles, and textiles. 9–11 Although hydrophobicity can be enhanced by a chemical modification that lowers the surface energy, contact angles larger than 120° have never been achieved for water on ideal flat surfaces. Therefore it is generally thought that all superhydrophobic surfaces result from originally hydrophobic substrates with surface microstructures. This has been demonstrated well, e.g. , for a CA of the order of 100–120° on such flat surfaces, a microstructured or rough surface shows an amplified CA as high as 150–175°. 12–15 However, some studies indicted that roughness can also lead to the superhydrophobicity on a hydrophilic surface. For example, for example, Otten et al. have found that a water droplet can be held by hydrophilic hairs on the leaves of Lady's Mantle. 16 Apart from leaves of natural plant, some artificial superhydrophobic surfaces have been prepared on some hydrophilic materials without low surface energy modification. 17–20 Furthermore, two approaches to achieve hydrophobicity on inherently hydrophilic surfaces have been developed. One is to make hydrophilic substrates with cavities and the trapped air in cavities can inhibit the liquid from wetting the surface. Abdelsalam et al. have proved that the contact angle of gold surfaces can be larger than 130° if the substrates are decorated with 400–800 nm pores. 18 The other one is to prepare re-entrant or overhang microstructures, such as T-shape or inverse-trapezoidal structures, on substrates. 19–22 Cao et al. showed that the overhang microstructures can induce superhydrophobicity on Si substrates with an intrinsic contact angle 74°. 19 Furthermore, it has proved that the T-shape microstructures can induce superhydrophobic behaviors on hydrophilic surfaces without any organic modifications. 22 Although the above results have experimentally demonstrated that the superhydrophobicity can be achieved on hydrophilic surfaces with very rough microstructures. However, the theoretical explanations have not been completely understood. For example, Liu et al. 23 studied closed/airproofed microstructures and some special topologies of the pillars or hairs on solid hydrophilic substrates, considering the effect of Laplace pressure and a certain geometric condition on the formation of Cassie's state. Marmur 24 theoretically analyzed the possibility of high contact angle from low contact angle surfaces for concave and convex roughness topographies and found that concave parts of roughness topographies may not enable a Cassie's state, while convex roughness features may enable the formation of hydrophobic surfaces from hydrophilic materials. On the contrary, Patankar 25 considered the energy of drops on surfaces with cavities, and theoretically explained the possibility of hydrophobic surfaces with cavities from hydrophilic materials. In addition, McHale 26 compared the surface free energy (FE) changes based on Young's law, and explained hydrophobic tendencies induced the roughness hydrophilic solid surfaces. Here it is noted that the above studies are in agreement with specific experimental observations, and can advance our understanding of superhydrophobic behavior on hydrophilic surfaces. However, these studies have never addressed the stability for such superhydrophobic behavior. In particular, the explanations and discussions about local surface curvatures in these studies are qualitative. Therefore, it is necessary to conduct a quantitative thermodynamic analysis on the superhydrophobicity on hydrophilic surfaces, especially, with a overhang microstructure. In this work, based on a proposed 2-D model, we mainly investigate and analyze thermodynamic states and wetting behavior of the above overhang microstructures with an aim at the possibility of superhydrophobicity on hydrophilic surfaces. By such an analysis of free energy states, the effects of topographical features and in particular, intrinsic hydrophilic surfaces on contact angle and contact angle hysteresis as well as composite or Cassie's state can be obtained in order to reveal the thermodynamic mechanism for the superhydrophobicity on hydrophilic surfaces.",
"discussion": "3. Results and discussion 3.1. Changes in free energy and free energy barrier with contact angle \n Fig. 4 shows two free energy curves of composite and noncomposite states for a pillar microstructure ( a = b = 2 μm, h = 2 μm). One can see that there is only one global free energy minimum for each curve, which is associated with the equilibrium contact angle (ECA/ θ E ) and exactly corresponds to generalized Cassie's contact angle ( e.g. , θ C = 114.4°) and Wenzel's contact angle ( e.g. , θ W = 69.68°), respectively. However, it should be noted that if drop advances from a position A to B or recedes from A to C, the contact angle changes and the local curve can show a fluctuation in free energy, as illustrated in the inset of Fig. 4 . These fluctuations demonstrate that the free energy curve contains multi-valued local minimum free energy and maximum free energy, indicating that such extremes represent metastable and unstable equilibrium states, which are related to various apparent contact angles. Therefore, the free energy barrier refers to the difference between local minimum and maximum in the direction of three-phase line motion, i.e. , advancing and receding. Furthermore, there are always two free energy barriers, i.e. , advancing free energy barrier and receding free energy barrier, connected to each contact angle value. Fig. 4 Variation of normalized free energy with contact angle for noncomposite (non) and composite (com) states ( L = 10 −2 m, a = b = h = 2 μm; intrinsic contact angle, θ Y = 80°; θ W = 69.68°, θ C = 114.4°). The inset shows an enlarge view of a segment of free energy curve illustrating the free energy barrier; positions A, B and C correspond to those in Fig. 2 . ΔAB and ΔAC represent the free energy barrier for retreating and advancing contact line, respectively. \n Fig. 5 shows the advancing and receding free energy barrier for both composite and noncomposite states based on the same geometrical parameters as shown in Fig. 1 . Hence, the advancing contact angle ( θ a ) and receding contact angle ( θ r ) as well as contact angle hysteresis defined as ( θ a − θ r ) can be determined by the intersecting values of advancing and receding curves with x -axis, respectively. As a result, for the composite state, the advancing ( θ a = 180°) and receding ( θ r = θ Y = 80°) contact angles as well as the maximum theoretical contact angle hysteresis defined as (contact angle hysteresis = θ a − θ r = 100°) can be determined. Fig. 5 Illustration of determination of the receding and advancing contact angles as well as contact angle hysteresis from the typical curves of advancing and receding free energy barriers for composite and noncomposite states ( L = 10 −2 m, a = b = h = 2 μm; intrinsic contact angle, θ Y = 80°). The contact angle hysteresis for composite state shown is the maximum value associated with zero free energy barrier on the advancing and receding branches of the free energy curves, whereas for the noncomposite state, the free energy barrier is negative. 3.2. Effect of sidewall angle Through eqn (22) and (24) , the normalized UHFEB of different surface structures could be calculated numerically. Here it should be indicted that because the free energy (J m −1 ) had been normalized with respect to γ LV (J m −2 ) in eqn (22) and (24) , the unit of normalized UHFEB should be meter. Fig. 6(a) shows a typical change of UHFEB curves on hydrophilic surfaces with inverse-trapezoidal pillars, vertical pillars, and trapezoidal pillars where their sidewall angle α varies, respectively. It could be seen that the value of normalized UHFEB did not change greatly with different penetration depth, and the normalized UHFEB curve would shift down with the increasing sidewall angle α . If the sidewall angle was 70° and 80°, the normalized UHFEB would be larger than 13.6 m and 3.7 m respectively even if the intrinsic contact angle was 85° ( Fig. 6(a) ). Here it is noted that as indicated above, although the unit of the normalized UHFEB was meter in the present work, in fact, it represented the free energy change. If the liquid/vapor interfaces moved down per unit height, its value could reach meter order. When the normalized UHFEB was always positive, and the system would prefer Cassie state. It meant that a composite interface could appear on the inverse-trapezoidal microstructure surface. Therefore, it proved the experimental observation that superhydrophobicity could be induced on hydrophilic surfaces. 19 Fig. 6 Effects of sidewall angle on normalized UHFEB with different penetration depth h on hydrophilic materials. ( a = 20 μm, b = 20 μm, H = 20 μm, θ Y = 85°, S = 10 −6 m 2 ; S was the sectional area of water droplet). (a) Three typical microstructures; (b) the inverse-trapezoidal pillar microstructure with different sidewall angles. For the inverse-trapezoidal pillar microstructure, Fig. 6(b) shows the critical curve (green curve) about wetting transition. If the sidewall angle increased from 83° to 85°, the normalized UHFEB would become negative and wetting transition would start. If the angle was larger than the critical value 84.104°, the negative UHFEB curve indicated a fully wetted interface would occur since the Wenzel state was more stable than the Cassie state. From the viewpoint of thermodynamics, the above results showed that the stability of a composite interface could be amplified with a decreasing sidewall angle, since a higher UHFEB could inhibit wetting transition. Hence, these results also theoretically explained the experimental phenomena why inverse-trapezoidal micro-textures superhydrophobic surface could perform excellent robustness (not the mechanical robustness). 31 From Fig. 6(b) , it was also seen that whether the Cassie state could appear mainly depended on the value of normalized UHFEB when penetration depth h was zero ( E 0 u ). \n Fig. 7 shows the effects of sidewall angle on the wetting states on hydrophilic and hydrophobic surfaces. One can see that the increasing intrinsic contact angle could bring a higher E 0 u , regardless of the dimensions of the microstructures, implying that a large intrinsic contact angle was preferred for a stable superhydrophobicity. The critical value of intrinsic contact angle (the curve intersecting with the E 0 u = 0 line) was 70.86° if the sidewall angle was 70°. The positive E 0 u indicated that a composite interface could appear if θ Y was in the range of 70.86° to 120°. If the sidewall angle increased from 70° to 110°, the critical value could also increase from 70.86° to 110.97°, implying that a small sidewall angle was crucial for the Cassie state even if the surface was hydrophobic (the critical value of θ Y larger than 90° for green and pink curves in Fig. 7 due to a large α ). Generally, Wenzel's equation indicated that a rough surface would become more hydrophilic if substrates were hydrophilic. However, the results in Fig. 7 show that the intrinsic contact angle, which was restricted by the sidewall angle of rough micro-textures, was not an independent parameter to determine the superhydrophobicity. Therefore, it was important to note that besides pillar width and spacing, the sidewall angle also played a very important role in the superhydrophobic behavior, especially for the transition from noncomposite to composite wetting states on hydrophilic surfaces. Fig. 7 Variations of normalized E 0 u with respect to intrinsic contact angles for different sidewall angles ( a = 20 μm, b = 20 μm, H = 20 μm, S = 10 −6 m 2 ). 3.3. Effect of pillar width and pillar spacing Variations in normalized E 0 u with respect to sidewall angle α was shown in Fig. 8 for different pillar dimensions. Apparently, effects of pillar width and pillar spacing on normalized E 0 u were quite similar. From Fig. 8(a) , as pillar width increased from 10 μm to 50 μm, the maximum value of normalized E 0 u decreased from 10.4 m to 6.5 m if the sidewall angle was 70°. And the maximum value of normalized E 0 u also decreased from 15.6 m to 3.4 m with increasing pillar spacing in Fig. 8(c) . These results indicated that the smaller pillar dimensions, the more stable composite states could be. However, in terms of inducing the transition from hydrophilicity to superhydrophobicity, such effects of pillar width and spacing could hardly become apparent. From the previous thermodynamic analysis, a small sidewall angle was crucial to obtain free energy barriers to separate the Wenzel state and Cassie state irrespective of solid surface chemistry. However, the critical value of sidewall angle (the curve intersecting with the E 0 u = 0 line) changed slightly with different pillar width and spacing in Fig. 8(b) and (d) . With pillar width changing, the critical value of sidewall angle ranged from 79.06° to 79.21°, while they were in the 77.62–79.60° range for different pillar spacing. The change of the critical value were only 0.15° and 1.98°, respectively for pillar width and pillar spacing varying in the range of 10 μm to 50 μm. It could be argued that since for preparing microstructures, the control of sidewall angle with such an accuracy is impossible experimentally, the pillar width and spacing could hardly play a crucial role in achieving the superhydrophobicity on a hydrophilic surface, but both affected its stability for this superhydrophobic surface. Fig. 8 Comparison of variations of normalized E 0 u with respect to sidewall angle for different pillar width ((a) and (b), b = 20 μm, H = 10 μm, θ Y = 80°, S = 10 −6 m 2 ) and pillar spacing ((c) and (d), a = 20 μm, H = 10 μm, θ Y = 80°, S = 10 −6 m 2 ). 3.4. Transition of the Cassie state on hydrophilic surfaces For the wetting transition conditions, the normalized UHFEB should be positive from eqn (22) and (24) . Based on eqn (24) , the sidewall angle α of pillars should be less than intrinsic contact angle if a positive E 0 u was expected. Apparently, if α was less than intrinsic contact angle θ Y , the change of the free energy Δ E could be positive, implying that positive free energy barriers could appear even if a droplet was placed on hydrophilic surfaces with overhang microstructures. Although it was energetically favorable for the droplet to wet the hydrophilic surfaces, the Cassie state could be metastable because extra free energy was needed to overcome the energy barriers during the wetting process. 32 The criterion of inducing the transition from hydrophilicity to superhydrophobicity was therefore derived as α < θ Y . In terms of the previous experimental results, Cao et al. 19 prepared an overhanging microstructure with a sidewall angle of 35.3° and 54.7° to achieve the superhydrophobicity on Si substrates with intrinsic contact angle 74°, which were also compatible with the present theoretical analysis. 3.5. Comparisons of free energy and free energy barriers between noncomposite and composite states The above results indicated that for a hydrophilic surface, superhydrophobic behavior can be achieved. Nevertheless, our further theoretical investigations on the direct comparisons of free energy and free energy barriers between noncomposite and composite states indicated that such a superhydrophobicity could be unstable or temporary. Fig. 9 shows the free energy change of a wetting system on the hydrophilic surface ( θ Y = 85°) for a inverse-trapezoidal microstructure. As seen, for both composite and noncomposite wetting states, their free energy first decreased and then increased, i.e. , there was a valley with a minimum value, respectively, where the so-called equilibrium contact angle located. It is very interesting to note that the curve for the composite states was above the one for the noncomposite states, i.e. , the free energy for the composite states was always higher than the one for the noncomposite states, implying that the composite states were metastable. This indicated that for such a hydrophilic surface, from the viewpoint of thermodynamics, a droplet on this surface would tend to wet the rough microstructure. Further calculations confirmed that there was a large positive free energy barriers between the composite and noncomposite states. The above results therefore provided a solid support for the difficult and complexity to realize the superhydrophobicity on hydrophilic surfaces or materials although now it has been recognized that any efforts in both theoretical and practical aspects could be feasible. Fig. 9 Free energy variations with contact angle ( a = 20 μm, b = 20 μm, H = 20 μm, θ Y = 85°, and S = 10 −6 m 2 ). To further understand the superhydrophobic behavior on intrinsic hydrophilic materials, the calculations were also extend to investigate the effects of various geometrical parameters for different intrinsic contact angles. For the simplicity, we employ the pillar microstructure as a typical example. Fig. 10 shows the free energy variations for the noncomposite and composite wetting systems with the intrinsic contact angle of 80° for different pillar heights. One can see that the curves of both noncomposite and composite states intersect at 180° for the same geometry, indicating the same energy state for the noncomposite and composite states. The free energy of composite states for the present system is always higher than that of noncomposite states for different pillar heights, implying that the composite states for this system were more unstable than the noncomposite states. Fig. 10 Comparison of variations of normalized free energy with apparent contact angle between noncomposite (non) and composite (com) wetting systems with different pillar heights ( h ) ( L = 10 −2 m, a = b = 2 μm; intrinsic CA, θ Y = 80°). In contrast, Fig. 11 shows the free energy variations with respect contact angle for noncomposite and composite wetting systems with an intrinsic contact angle of 120° for different pillar heights. One can see that the composite state was more stable that the noncomposite state for different pillar heights. In particular, compared Fig. 10 to Fig. 11 , it was important to note that the free energy of hydrophilic materials was higher than that of hydrophobic materials; the former had a magnitude of 10 −2 m, whereas the later had a magnitude of 10 −3 m. The above results indicated that the thermodynamic state for hydrophilic materials was more unstable than that for the hydrophobic materials. Therefore, even though a composite state for hydrophilic materials can be formed, the corresponding superhydrophobicity may be temporary and tend to transfer to hydrophilicity with time, especially, in case of external stimulus such as vibrational energy. 30 Fig. 11 Comparison of variations of normalized free energy with apparent contact angle between noncomposite (non) and composite (com) wetting systems with different pillar heights ( h ) ( L = 10 −2 m, a = b = 2 μm; intrinsic CA, θ Y = 120°). In order to further reveal the intrinsic effect of different materials (hydrophilic or hydrophobic), Fig. 12 and 13 show the free energy barrier variations for the noncomposite and composite wetting systems with the intrinsic contact angle of 80° and 120°, respectively. From Fig. 12 , one can see that the free energy barrier and the resultant contact angle hysteresis of the composite state could hardly depend on pillar height. It was important note that the negative free energy barrier indicated that there was not enough energy to provide a transition between noncomposite and composite states. Comparatively, from Fig. 13 , one also can see that for an intrinsic contact angle of 120°, the free energy barrier and the resultant contact angle hysteresis of the composite state could not depend on pillar height. However, for the noncomposite state, the contact angle hysteresis increased and there was a positive free energy barrier, implying that there was enough energy to provide a transition between noncomposite and composite states. Here it is worth noting that some experimental studies suggested that such a transition was plausible. 33,34 For example, for hydrophilic materials with an intrinsic contact angle of 70°, a transition from noncomposite to composite states could occur if the depth of surface topography pores (a similar parameter to the pillar height) was very large. This happens perhaps because the difference in free energy and barrier between the noncomposite to composite states was so small due to adequate roughness that the transition between the two states could be easily realized under the experimental conditions where external resources may be available. Fig. 12 Variations of normalized free energy barrier with apparent contact angle for different pillar heights of the microstructured surfaces for noncomposite (non) and composite (com) wetting states ( L = 10 −2 m, a = b = 2 μm; intrinsic CA, θ Y = 80°). Fig. 13 Variations of normalized free energy barrier with apparent contact angle for different pillar heights of the microstructured surfaces for noncomposite (non) and composite (com) wetting states ( L = 10 −2 m, a = b = 2 μm; intrinsic CA, θ Y = 120°)."
} | 6,368 |
25679534 | PMC4542026 | pmc | 339 | {
"abstract": "Recently, a novel electrogenic type of sulphur oxidation was documented in marine sediments, whereby filamentous cable bacteria (Desulfobulbaceae) are mediating electron transport over cm-scale distances. These cable bacteria are capable of developing an extensive network within days, implying a highly efficient carbon acquisition strategy. Presently, the carbon metabolism of cable bacteria is unknown, and hence we adopted a multidisciplinary approach to study the carbon substrate utilization of both cable bacteria and associated microbial community in sediment incubations. Fluorescence in situ hybridization showed rapid downward growth of cable bacteria, concomitant with high rates of electrogenic sulphur oxidation, as quantified by microelectrode profiling. We studied heterotrophy and autotrophy by following 13 C-propionate and -bicarbonate incorporation into bacterial fatty acids. This biomarker analysis showed that propionate uptake was limited to fatty acid signatures typical for the genus Desulfobulbus . The nanoscale secondary ion mass spectrometry analysis confirmed heterotrophic rather than autotrophic growth of cable bacteria. Still, high bicarbonate uptake was observed in concert with the development of cable bacteria. Clone libraries of 16S complementary DNA showed numerous sequences associated to chemoautotrophic sulphur-oxidizing Epsilon- and Gammaproteobacteria, whereas 13 C-bicarbonate biomarker labelling suggested that these sulphur-oxidizing bacteria were active far below the oxygen penetration. A targeted manipulation experiment demonstrated that chemoautotrophic carbon fixation was tightly linked to the heterotrophic activity of the cable bacteria down to cm depth. Overall, the results suggest that electrogenic sulphur oxidation is performed by a microbial consortium, consisting of chemoorganotrophic cable bacteria and chemolithoautotrophic Epsilon- and Gammaproteobacteria. The metabolic linkage between these two groups is presently unknown and needs further study.",
"introduction": "Introduction The traditional view of diffusion-controlled redox zonation in marine sediments has recently been challenged by the observation that microorganisms are capable of transporting electrons over cm-scale distances ( Nielsen et al., 2010 ). This long-distance electron transport is mediated by filamentous cable bacteria belonging to the Desulfobulbaceae that are proposed to catalyse a new electrogenic form of sulphur oxidation ( Pfeffer et al., 2012 ). Cable bacteria have recently been found in a wide range of marine sediment environments ( Malkin et al., 2014 ), and seem to be competitively successful, because they can harvest electron donors (sulphide) at cm depth in the sediment while still utilizing thermodynamically favourable electron acceptors such as oxygen and nitrate ( Nielsen et al., 2010 ; Marzocchi et al., 2014 ) that are only available in the first mm of coastal sediments. The current conceptual model of electrogenic sulphur oxidation (e-SOx) envisions a new type of metabolic cooperation between cells, where different cells from the same multicellular filament perform distinct redox half reactions. Anodic cells located in suboxic and anoxic sediment zones obtain electrons from sulphide and liberate protons (anodic half-reaction: ½ H 2 S+2H 2 O→½ SO 4 2− +4e − +5H + ). These electrons are then transported along the longitudinal axis of the filament to cells located near the sediment–water interface ( Pfeffer et al., 2012 ). At the thin oxic layer near the sediment surface, cathodic cells reduce oxygen and consume protons (cathodic half-reaction: O 2 +4e − +4H + →2H 2 O). The two half-reactions leave a distinct geochemical fingerprint in the sediment consisting of a shallow oxygen penetration depth, a cm-wide suboxic zone separating the oxic and sulphidic sediment horizons, and a characteristic pH depth profile, defined by a sharp pH maximum within the oxic zone and a deep and broader pH minimum at the bottom of the suboxic zone ( Nielsen et al., 2010 ). Laboratory time-series experiments ( Malkin et al., 2014 ; Schauer et al., 2014 ) show that a network of cable bacteria can rapidly (<10 days) develop in sediments, reaching high filament densities (>2000 m of filaments per cm −2 after 21 days; Schauer et al., 2014 ) with fast generation times of ∼20 h. Furthermore, the progressive downward growth of the cable bacteria closely correlates with the widening of the suboxic zone and a strong increase in biogeochemical rates, such as sedimentary oxygen consumption ( Malkin et al., 2014 ; Schauer et al., 2014 ). One could hypothesize that cable bacteria may have a similar metabolism to their closest cultured relative Desulfobulbus propionicus ( Pfeffer et al., 2012 ) that can efficiently grow as a chemoorganotroph in propionate-rich media while obtaining metabolic energy from oxidation of sulphide to elemental sulphur followed by sulphur disproportionation ( Widdel and Pfennig, 1982 ; Dannenberg et al., 1992 ; Fuseler and Cypionka, 1995 ; Pagani et al., 2011 ). It is presently unclear whether they are organotrophs (heterotrophs) or lithoautotrophs (chemoautotrophs). Here, we adopted a multidisciplinary approach to characterize the carbon metabolism in coastal sediments with e-SOx activity, resolving the carbon substrate uptake of both cable bacteria and their associated microbial community. We conducted a series of laboratory incubations, starting in March 2012, to track the temporal development of the cable bacteria network by microsensor profiling and fluorescence in situ hybridization (FISH), and quantified inorganic carbon fixation at various time points through biomarker analysis of phospholipid-derived fatty acids combined with stable-isotope probing (PLFA-SIP). In August 2012, we studied both inorganic carbon and propionate uptake by PLFA-SIP. In both months, we examined the linkage between carbon metabolisms and e-SOx activity through targeted manipulation treatments. The active microbial community in the March and August experiments was characterized by 16S complementary DNA (cDNA) clone libraries. The final experiment in May 2013 quantified the carbon tracer uptake of both inorganic (bicarbonate) and organic (propionate) substrates by individual cable bacteria filaments using nanoscale secondary ion mass spectrometry (nanoSIMS).",
"discussion": "Discussion Electrogenic sulphur oxidation The microsensor and FISH data obtained in our sediment incubations confirmed results obtained in a previous field study ( Malkin et al., 2014 ) and laboratory experiments ( Nielsen et al., 2010 ; Risgaard-Petersen et al., 2012 ; Schauer et al., 2014 ). The temporal development of the characteristic geochemical fingerprint of the e-SOx process is accompanied by the downward growth of a dense network of long filamentous cable bacteria that span the suboxic zone ( Malkin et al., 2014 ; Schauer et al., 2014 ). Cable bacteria densities at peak development were half of those recorded in a similar sediment incubation experiment (2380 m cm −2 , Schauer et al., 2014 ) but three- to sixfold higher than densities in the suboxic zone (0.3–0.8 cm) observed under field conditions (120 m cm -3 , Malkin et al., 2014 ). In addition, we estimate that e-SOx was responsible for ∼70% of the sedimentary oxygen consumption given the steep increase in DOU from days 6 to 13 and the sharp decrease in DOU 1 h after cutting the sediment. This estimate closely aligns with the contribution of 81% found by Schauer et al. (2014) and highlights the important role that e-SOx can play in sedimentary geochemical cycling. The drastic geochemical effects observed in the manipulation experiments, with either induced anoxia or cutting of the sediment at 0.3 cm depth, were consistent with previous observations ( Nielsen et al., 2010 ; Pfeffer et al., 2012 ) and confirm the current conceptual model of e-SOx in which electrons are transported from anodic cells to cathodic cells along the longitudinal axis of cable bacteria filaments. Carbon metabolism and growth of cable bacteria We used two 13 C-stable isotope approaches (PLFA-SIP and nanoSIMS) to study the carbon metabolism of the fast-growing cable bacteria. PLFA-SIP identified high incorporation of propionate in 15:0, 17:1ω6c and 17:1ω8c fatty acids that we attributed to be biomarkers for cable bacteria given that those resemble the fatty acid composition of Desulfobulbus spp. ( Taylor and Parkes, 1983 ; Pagani et al., 2011 ) to which cable bacteria are most closely related ( Pfeffer et al., 2012 ; Malkin et al., 2014 ). Propionate is steadily produced through mineralization of organic matter in sediments and serves as a major energy and carbon source for several Deltaproteobacteria ( Sorensen et al., 1981 ; Parkes et al., 1989 ; Purdy et al., 1997 ). Rates of propionate assimilation measured with PLFA-SIP (17 mmol C m −2 day −1 ) scale well with biomass accruement of the cable bacteria based on FISH counts (16 mmol C m −2 day −1 ), suggesting that cable bacteria could successfully grow on propionate in the environment. Moreover, based on 13 C-dissolved inorganic carbon measurements obtained from pore water, we estimated that only 2% of the propionate assimilated was respired, indicating that propionate was mainly used as a carbon source by the cable bacteria rather than being respired by sulphate-reducing bacteria. However, we found that propionate incorporation (7 mmol C m −2 day −1 ) estimated by nanoSIMS explained 44% of the observed biomass increase (FISH). Desulfobulbus spp. can also grow on lactate, acetate, ethanol, propanol and pyruvate ( Laanbroek and Pfennig, 1981 ; Widdel and Pfennig, 1982 ; Parkes et al., 1993 ; Pagani et al., 2011 ), and therefore it seems plausible that cable bacteria use other organic substrates besides propionate. An autotrophic metabolism is unlikely as the biomass increase calculated from 13 C-inorganic carbon uptake (nanoSIMS) only explains ∼15% of the observed cable bacteria growth. The limited incorporation of 13 C-bicarbonate measured in cable bacteria (∼20% of propionate uptake) is well within the range observed in heterotrophic microorganisms ( Roslev et al., 2004 ; Hesselsoe et al., 2005 ; Wegener et al., 2012 ) and inorganic incorporation can therefore most likely be attributed to anapleurotic reactions. The fast expansion of the cable bacteria filament network, as observed in our incubations, has been previously explained by an exponential growth mechanism with high generation times, and continuous and uniform cell divisions throughout the filament ( Schauer et al., 2014 ). When generation time is calculated based on 13 C-propionate incorporation (nanoSIMS) and FISH counts (using the same assumptions as Schauer et al., 2014 for the filament density at day 0), this provides the short generation time (20 h) previously reported by Schauer et al. (2014) . However, our 13 C-propionate incorporation (nanoSIMS) results indicate that the turnover of cathodic, oxygen-respiring cells is twice as fast as that of anodic, sulphide-oxidizing cells. This suggests different energy yields from each half redox reactions that lead to rapid-growing cathodic cells and less-efficient anodic cells. Moreover, estimation of growth efficiencies calculated from biomass increase (16 mmol C m −2 day −1 ) and COC (54 mmol O 2 m −2 day −1 ) suggest an efficiency of 30%. Previously, a ratio of 1:1 of oxygen to carbon was reported for the cable bacteria ( Schauer et al., 2014 ), but our results reveal higher energy dissipation from the transport of electrons along the cm-long filament. The metabolic differences between spatially and temporally distinct cable bacterial populations highlight the need for pure cultures to further study the growth characteristics of this recently discovered bacterium. As observed previously ( Pfeffer et al., 2012 ), the cutting manipulation at 0.3 cm had an immediate and drastic response in the geochemistry of the sediment, suggesting an instant arrestment of e-SOx, presumably because of physical disruption of the electron transport by the cable bacteria filaments. Our manipulation experiments however indicate that under some conditions, this impediment of e-SOx may not be permanent. Unlike the cutting experiment of Pfeffer et al. (2012) , we observed a partial restoration of the e-SOx geochemical signature 24 h after the cut at 0.3 cm. Cable bacteria filament network below the cut (which were no longer in contact with oxygen) managed to regain or maintain part of their suboxic carbon uptake activity that was substantially higher than in the anoxic treatment. The treatment with the deeper cutting depth at 0.8 cm did not show any sign of recovery of e-SOx activity after 24 h. These distinct effects between the two cutting depths at 0.3 and 0.8 cm suggest that the recovery potential of the cable bacteria network depends on the location of disturbance with a critical threshold of a couple of mm within 24 h. Re-establishment of e-SOx activity after cutting could be explained by fast regrowth at the upper terminal end of the cable filament, or perhaps, by reorientation, implying some form of motility as proposed by Schauer et al. (2014) . The mechanism by which the transport of electrons is re-established requires further study into chemotactic and motility capacities of the cable bacteria. Associated chemoautotrophic community Intriguingly, high rates of inorganic carbon fixation were measured in this study in concert with the downward development of the heterotrophic cable bacteria. Total inorganic carbon assimilation was sixfold higher than those reported for subtidal environments ( Enoksson and Samuelsson, 1987 ; Thomsen and Kristensen, 1997 ) and sandy intertidal sediments ( Lenk et al., 2011 ) but are well within the range of rates obtained in sulphidic salt-marsh creek sediments ( Boschker et al., 2014 ). Although inorganic carbon fixation was high in horizons dominated by e-SOx, this inorganic carbon is unlikely assimilated by the cable bacteria themselves, but rather by other chemoautotrophic bacteria. Clone libraries revealed that sulphur-oxidizing Epsilon- and Gammaproteobacteria were not only present in the oxic zone, but persisted throughout the suboxic zone. PLFA analysis with 13 C-bicarbonate labelling also provided a biomarker fingerprint that was consistent with chemoautotrophic sulphur-oxidizing Epsilon- and Gammaproteobacteria ( Inagaki et al., 2003 ; Takai et al., 2006 ; Li et al., 2007 ; Glaubitz et al., 2009 ; Sorokin et al., 2010 ; Labrenz et al., 2013 ). Hence, both clone libraries and PLFA-SIP suggest that sulphur-oxidizing bacteria from the Epsilon- and Gammaproteobacteria may be the main chemoautotrophic organisms in sediments with e-SOx. The difference in depth distribution of inorganic carbon incorporation between the March and August experiments might be related to the initial conditions in the seasonal hypoxic Marine Lake Grevelingen, where bottom waters were oxic in March (80% O 2 saturation) as opposed to anoxic in August (0% O 2 saturation). To conclude, the presence of the cable bacteria performing e-SOx process favoured the codevelopment of a strongly active chemoautotrophic community that extends down to cm of depth in the sediment. Manipulation experiments provided further evidence of a tight coupling between subsurface chemoautotrophic organisms and the electron transport by the cable filament network. Induced anoxia demonstrated that chemoautotrophs in the suboxic zone directly depend on the oxygen in the overlying water given the almost complete inhibition of 13 C-bicarbonate uptake throughout the sediment. Similarly, inorganic carbon incorporation drastically decreased below the cutting depth (in the 0.8 cm cutting depth treatment) and was only maintained in sediment layers where cable bacteria were still connected to oxygen. Inorganic carbon fixation was actually stimulated in the surface sediment layer after cutting, and this could be because of an increased potential for reoxidation after doubling in oxygen penetration once the cable network was disrupted. Finally, when the sediment was cut at 0.3 cm depth, bicarbonate uptake only slightly decreased (still maintained 50 to 70% of initial activity) as did propionate uptake by the cable bacteria, suggesting that chemoautotrophs partially recovered in synchrony with the re-establishment of the e-SOx process after 24 h. Our study therefore suggests that the complete oxidation of sulphide in e-SOx sediments may be a two-step process, regulated by a consortium of bacteria composed of chemoorganotrophic cable bacteria and sulphur-oxidizing chemolithoautotrophs, rather than by the cable bacteria alone. Given that in the deeper layers, oxygen and nitrate are absent ( Risgaard-Petersen et al., 2012 ; Marzocchi et al., 2014 ) chemolithoautotrophs have to use other electron acceptors to oxidize reduced sulphur compounds. Although these chemolithoautotrophs may disproportionate sulphur ( Grote et al., 2012 ; Wright et al., 2013 ), their metabolic link to the cable bacteria indicates they use the cable bacteria possibly as an electron sink by tapping on to the electron transport network via nanowires, nanotubes or fimbrae ( Widdel and Pfennig, 1982 ; Reguera et al., 2005 ; Dubey and Ben-Yehuda, 2011 ). Clearly, further studies are needed to confirm key autotrophic players and to target the exact mechanisms by which the observed activity of the chemolithoautotrophic bacteria is coupled to the electron transport network of the cable bacteria. Our results suggest that other bacteria may benefit directly from the electron transport by the cable bacteria."
} | 4,467 |
33994935 | PMC8115403 | pmc | 340 | {
"abstract": "Achieving multi-level devices is crucial to efficiently emulate key bio-plausible functionalities such as synaptic plasticity and neuronal activity, and has become an important aspect of neuromorphic hardware development. In this review article, we focus on various ferromagnetic (FM) and ferroelectric (FE) devices capable of representing multiple states, and discuss the usage of such multi-level devices for implementing neuromorphic functionalities. We will elaborate that the analog-like resistive states in ferromagnetic or ferroelectric thin films are due to the non-coherent multi-domain switching dynamics, which is fundamentally different from most memristive materials involving electroforming processes or significant ion motion. Both device fundamentals related to the mechanism of introducing multilevel states and exemplary implementations of neural functionalities built on various device structures are highlighted. In light of the non-destructive nature and the relatively simple physical process of multi-domain switching, we envision that ferroic-based multi-state devices provide an alternative pathway toward energy efficient implementation of neuro-inspired computing hardware with potential advantages of high endurance and controllability.",
"introduction": "1. Introduction The recent advancements of data-driven learning paradigm such as artificial deep neural networks (DNN) have achieved superhuman performance in various applications including image/pattern recognition, natural language processing, and developing autonomous intelligence (LeCun et al., 2015 ). However, the energy consumption of artificial intelligence (AI) implemented in today's computers is significantly higher compared to that of a human brain (Cox and Dean, 2014 ). The energy inefficiency of such hardware is largely attributed to the von-Neumann memory bottleneck due to the separation of memory and compute units and the limited on-chip memory density in computing hardware. For instance, DNNs are usually implemented in graphic processing units (GPUs), which desire large area and power consumption in presence of growing DNN model sizes and large amount of data to process. Neuromorphic computing is an emerging computing paradigm that aims for building bio-plausible computing systems in pursuit of brain-level efficiency in cognitive processing (Mead, 1990 ; Roy et al., 2019 ). Recently, remarkable implementations of neuromorphic hardware such as TrueNorth (Merolla et al., 2014 ) and Loihi (Davies et al., 2018 ) have been demonstrated based on complementary metal-oxide-semiconductor (CMOS) technologies. But CMOS-based technologies require large number of transistors to implement neuronal and synaptic functions, leading to increased cost of energy and area. On the other hand, emerging non-volatile memories (NVM) based on novel physical mechanisms can lead to significant reduction of leakage power and achieve high on-chip density compared to CMOS, while mimicking key neuro-synaptic functionalities for neuromorphic computing (Yu and Chen, 2016 ; Li et al., 2019 ). Particularly, the emerging NVM technologies have great potential to provide scalable and energy efficient building blocks for crossbar based in-memory computing by performing computation within memory arrays (Ambrogio et al., 2018 ; Ielmini and Wong, 2018 ). With crossbar computing, the product of an input vector (voltage) and a weight matrix (conductance) can be obtained from the accumulated output currents, following Ohm's Law. Such configuration leads to efficient hardware realization of matrix-vector multiplication (MVM) operations, which are ubiquitous in both bio-plausible computing and standard DNN models. Therefore, enabling crossbar computing will not only facilitate the development of neuromorphic fabrics, but also improve hardware efficiency of executing generic AI/machine learning algorithms. As for neuromorphic processing, several mathematical models have been proposed to describe the neuronal models and synaptic learning rules of biological nervous system (Hodgkin and Huxley, 1952 ; Izhikevich, 2003 ), laying the foundation for neuro-mimetic implementation in hardware. Figure 1 shows a hardware-implemented neural network with exemplary leaky-integrate and fire (LIF) neurons and synapses based on spike timing-dependent plasticity (STDP) or pulse-driven synaptic learning rules (Yu et al., 2019 ). Emulation of both neurons and synapses have been experimentally demonstrated using emerging NVMs (Tuma et al., 2016 ; Islam et al., 2019 ). Figure 1 Concept of a biological neural network and its hardware implementation in crossbar arrays. Left panel illustrates that two biological neurons are interconnected by a synapse. The strength of the synaptic connections (synaptic weights) can be modified depending on the relationship between the two neurons. A crossbar array implementing artificial neural networks containing neurons connected by synaptic devices is shown in the right panel. The center panel describes synapse (top) and neuron (bottom) models. The activation of a neuron is controlled by its membrane potential, and its dynamics can be described of a leaky integrate-fire (LIF) neuron model. An accurate description of the membrane potential desires devices that can represent analog state values. Neurons are interconnected by synapses, which can be put into crossbar devices with variable conductance states. Spiking timing dependence plasticity (STDP) and pulse driven potentiation/depression of a synapse is shown to illustrate one bio-plausible learning mechanism based on synaptic plasticity. In order to provide high density on-chip memory as well as efficient emulation of synaptic plasticity and neuron activations, it is desirable to have programmable multi-level NVM devices. The capability of multiple states per device will not only enlarge the capacity and precision of synaptic weight storage in neuro-inspired computing, but also lead to benefits in generic memory applications due to increased density. Furthermore, multi-level NVM devices can realize both emulation of the aforementioned bio-plausible neurons (Sharad et al., 2013 ; Burr et al., 2017 ) and in-device implementation of various analog neuron models such as (shifted) sigmoid (Siddiqui et al., 2019 ) and rectified linear units (ReLu) (Lashkare et al., 2018 ), which are frequently used in DNNs. Among the emerging NVMs, resistive random-access memory (ReRAM) (Hu et al., 2014 ) and phase change memory (PCM) (Boybat et al., 2018 ) can provide high memory density as well as multi-level cells. However, the electroforming process involving ion motion of ReRAM and the melting-crystallization process of PCM induce endurance and reliability issues: large variations among devices and sizeable drifts over time (Eryilmaz et al., 2015 ). The low endurance significantly limits the numbers of writes, preventing the use of ReRAM and PCM for training large-scale AI models. The large device variation not only makes it difficult to program such NVM devices to a desirable conductance states but also places challenges to differentiate and sense the multiple levels. Moreover, due to the large conductance drift over time in PCM, erroneous results may occur even for inference-only tasks when running a pre-trained model mapped in PCM crossbar arrays. At present, although various types of NVM devices have been proposed and studied, it is still challenging to provide a reliable, scalable, and energy efficient hardware solution for multi-level neuro-mimetic devices (Burr et al., 2017 ; Schuman et al., 2017 ; Yan et al., 2018 ; Chakraborty et al., 2020b ; Kim et al., 2020 ). In contrast, devices using ferroic (magnetic and ferroelectric) materials such as magnetic RAM (MRAM) (Bhatti et al., 2017 ) and ferroelectric RAM (FeRAM) (Ishiwara, 2012 ) can provide better endurance and more energy-efficient writing, leveraging the unique properties of spin or charge polarization. In particular, the spintronic materials are promising for high endurance due to the absence of physical ionic motion in magnetization switching, while ferroelectric field-effect transistors (FeFET) could offer superior CMOS compatibility. While spintronic (magnetic) and ferroelectric materials have been traditionally investigated for binary memory, there has been growing interest in exploiting them for multi-level neuromorphic devices for functionalities such as synaptic plasticity and membrane potential modulation in neurons. More interestingly, the switching mechanisms and memory effects of ferroic materials share remarkable resemblance with biological neural systems, suggesting a possible path of developing bio-plausible hardware primitives. However, although the NVM devices based on ferroic materials might have larger cell areas than ReRAM and PCM, they can still achieve better density compared to CMOS. In this review article, we explore ferroic materials as possible material of choice for neuromorphic with multi-level resistive devices. It is shown that spintronic and ferroelectric materials leveraging multi-domain switching dynamics enable devices to obtain multi-level states with improved controllability and endurance compared to ReRAM and PCM. The rest of this article is organized as follows. Section 2 focuses on spintronic devices, where both domain wall (DW) motion in ferromagnets and multi-domains induced in exchange-coupled heterostructures can be leveraged to achieve multi-level nanometer-scale devices. Section 3 focuses on ferroelectric devices, including FeFET and ferroelectric tunnel junctions (FTJ) as the basic configurations for generating ferroelectric multi-level states. Both fundamental material physics of multi-level devices as well as representative demonstrations of neural functionalities are covered. Section 4 reiterates the opportunities with ferroic multi-level devices toward developing neuromorphic hardware in comparison with other technologies. We will highlight key advantages and challenges for each of these technologies, followed by discussions and proposals regarding the potential pathways for addressing the challenges. We conclude this review with the proposal that through the co-design of device, circuit, and algorithm, multi-level ferroic devices could provide exciting opportunities of constructing large-scale and energy efficient cognitive computing systems.\n\n2. Spintronic Devices Spintronic materials have shown clear advantages for developing next generation non-volatile memory with potential combination of high speed, low power, and unparalleled endurance. In particular, magnetic tunnel junctions (MTJ) have been extensively investigated and demonstrated reliable memory read and write schemes in device dimension down to tens of nanometers (Ikegawa et al., 2020 ). As is illustrated in Figure 2 , a MTJ in MRAM memory cell comprises of two FM layer separated by a thin tunnel barrier. Conductance of MTJ under applied voltages will be high (low) when the magnetizations in the two ferromagnetic layers are parallel (anti-parallel) due to spin-dependent tunneling across the barrier (Parkin et al., 2004 ). In practical MTJ device stack, one of the FM layer is pinned by additional structures forming a reference layer (RL), while the other FM layer defined as free layer (FL) can be switched between the two states under external stimulus including external magnetic field, or spin-polarized current induced spin transfer torque (STT) (Slonczewski, 1996 ; Diao et al., 2007 ) and spin-orbit torque (SOT) (Liu et al., 2012 ), as illustrated in Figure 2B . Figure 2 (A) Magnetic tunnel junction (MTJ) and tunneling magnetoresistance (TMR). Resistance of MTJ depends on the relative orientations of the two ferromagnetic layers next to the tunnel barrier. High/low resistance states correspond to anti-parallel (AP)/parallel (P) configuration of the magnetic ordering in the free layer and reference layer. (B) Spin transfer torque (STT) and spin-orbit torque (SOT). STT is originated from spin polarized current going through an MTJ. The STT effectively switches spins by countering against the Gilbert damping of the free layer magnetic moments. SOT is a result of spin Hall effect at interface of ferromagnetic/heavy metal layers, where a charge current flowing along the heavy metal layer can induce a transverse spin current flowing into the adjacent ferromagnetic layer. While MTJs can naturally store binary information with high accuracy and thermal stability based on the bi-directional magnetizations, having controllable multilevels in the FL is more of interest for device level emulations of neuromorphic functionalities. Note that ferromagnetic materials are known to maintain long-range magnetic ordering with long retention and high stability against perturbations due to the strong exchange interactions among localized magnetic moments therein. Conversely, introducing stable multi-domain configurations in magnetic-based materials inevitably becomes difficult thanks to the need of countering the forming of long-range magnetic ordering, imperatively urging new mechanisms from material and device structure level for inducing multi-domains functionalities. In the following subsections, various approaches introducing multi-level devices are highlighted and some prototypes of demonstrating brain-inspired computing are discussed. 2.1. Multi-Level Spintronic Devices Based on Domain Wall Motion A natural path to multilevels in spintronics is to split the single-domain magnetizations as formed in MRAM devices to multiple domains (Fong et al., 2016 ). It is known that switching of magnetic thin films can involve mechanisms of nucleation formation and domain propagation, suggesting a possibility to generate intermediate multi-domain of magnetic textures between the bi-stable states. In order to have stable configurations of multiple domains in continuous ferromagnetic thin films, additional mechanisms are required to maintain the pinning of domain walls between spin-up and spin-down regions. The DWs can be pinned or displaced, depending on the combined effects of material properties such as exchange coupling among magnetic moments, shape anisotropy determined by device geometry, as well as local defects (Beach et al., 2006 ; Thomas et al., 2006 ; Emori et al., 2013 ). Conceptually, such domain wall motion (DWM) in a ferromagnetic thin film can generate a near-continuous variation of magnetic states from one direction to the other, resulting in variable resistance states of a magneto-resistive device described by a model with parallel resistors. 2.1.1. Device Fundamentals of Domain Wall Motion The idea of using DWM driven by current-induced torque has sparkled a plethora of studies built on the mostly matured STT-switching technology (Wang et al., 2009 ; Lequeux et al., 2016 ). It has been proposed and demonstrated that with special engineering of device geometry, non-volatile multi-level resistance states can be realized in an MTJ with perpendicular magnetic anisotropy (PMA). As is shown in Figure 3A , a DW in the free layer of MTJ with the shape of an elongated stripe can be displaced by applied electric currents, leading to modifications in MTJ resistance following the relationship: (1) G ( x ) = G P * x / L + G A P * ( 1 - x / L ) + G D W where G P and G AP are the parallel and antiparallel conductance respectively, and x is the domain wall position in a stripe of length L. STT-DWM is controlled by the magnitude and polarity of the spin polarized currents across the FL. The current-driven DWM device can potentially work as a two-terminal compact device following the STT-MRAM configuration. Furthermore, it was recently found that in magnetic heterostructures such as oxide/ferromagnetic/heavy metal stacks, a chiral DW with Néel configuration can be formed and stabilized in perpendicularly magnetized thin films due to Dzyaloshinskii-Moriya exchange interaction (DMI) at the FM/HM interface and the broken inversion symmetry in the heterostructure stack (Emori et al., 2013 ). It is observed that the Néel Wall can be efficiently driven by the spin orbit torque (SOT) originated from spin Hall effect (SHE) of the heavy metal layer, as is shown in Figure 3B . Therefore, DWM driven by SOT in a MTJ/HM heterostructures can be leveraged for programming the multi-level device conductance (Sengupta et al., 2016b ). While in general a magnetic field is needed for deterministic SOT switching of PMA materials due to the in-plane spin polarization (Liu et al., 2012 ), it is found that the interfacial DMI could effectively provide the desired magnetic field toward field-free SOT driven DWM (Emori et al., 2013 ). Current-driven DW motion in heavy-metal/ferromagnet/oxide structures is naturally explained by the combination of the SHE and DMI. The SHE produces the sole current-induced torque, and DMI stabilizes chiral DWs while permitting uniform DW motion with very high efficiency. The writing speed of DWM-SOT devices is characterized by the domain wall velocities, which increases with the current density up to saturation in absence of pinning sites. Although domain wall velocity can be as high as 10 2 m/s (Beach et al., 2006 ; Agrawal and Roy, 2018 ), in practice DW motion can be hindered due to pinning in magnetic thin films, making it challenging to have precise control and efficient manipulation of DWs (Thomas et al., 2006 ). The readout mechanism in the proposed device is similar to that of a STT-MRAM memory cell, while the SOT is generated from a lateral charge currents which has a separated flow path from the reading operation across MTJ stack. The major advantage of the three-terminal device is the decoupled read and write paths which will eliminate the read disturbance issue of MTJ and thus lead to significant improvement in device endurance. Recently it is confirmed experimentally that SOT-DWM can be used for artificial synaptic devices in MgO/CoFeB/Ta heterostructures. In the following subsection, various prototypes implementing neuromorphic functionalities based on DWM spintronic devices are discussed. Although we focus on SOT-driven configurations given the advantage of separate read/write paths, similar mechanism will also work for STT-driven configurations. Figure 3 Domain wall motion (DWM) based multi-level devices. (A) STT-driven DWM in a MTJ-based device. Lequeux et al. Scientific Reports 6, 31510 (2016), Copyright 2016 Author(s), licensed under a Creative Commons Attribution (CC BY) license. (B) SOT-driven DWM device. Dzyaloshinskii-Moriya exchange interaction across the heavy metal and ferromagnetic layer provides effective magnetic fields that ensure deterministic switching. Reproduced with permission from Sengupta et al. IEEE Transactions on Biomedical Circuits and Systems, 10, 1152–1160 (2016), Copyright 2016 IEEE. 2.1.2. Domain Wall Motion Based Neuromorphic Devices A direct application of SOT-driven DWM device will be crossbar implementation for MVM computing engine. Leveraging a linear dependence of conductance on the DW position which is subsequently linearly dependent on the driving electric current (before saturation), SOT-DWM devices in a crossbar array can have their DW position (and thus conductance states) accurately programmed to map a synaptic weight matrix, as is illustrated in Figure 4 (Sengupta et al., 2016b ; Sengupta and Roy, 2017 ). Parallel dot product of vector (voltage) and matrix (conductance) can be directly executed following the Kirchoff Current Law. Given the non-volatility of conductance states in those multi-domain devices, we could just set the devices once with pre-trained weight matrix and reuse the stored weights during inference, eliminating additional memory access and data transfer. The advantage of separated read and write paths in SOT-DWM is evidently demonstrated, as the MVM operations at inference only involves reading path and thus read disturbance to the states is minimized. The DWM-based multi-level device can also implement STDP, another important synaptic characteristics. With STDP learning rule, presynaptic spike arrival before the occurrence of postsynaptic spike leads to long-term potentiation (LTP) of the connecting synapse, whereas spike arrival after postsynaptic spike leads to long-term depression (LTD) of the same synapse. The magnitude of the relative change in synaptic strength (ΔW) decreases exponentially with the timing difference between the preneuron and postneuron spikes. A key step to realize STDP with a DWM-based spin device is to link the timing of the pre-neuron and post-neuron to the conductance change in the interconnected SOT-driven DWM synapse. One approach as illustrated in Figure 4 is to provide exponential variation of HM currents modulated by the timing of PRE and POST neurons in circuit, assuming that the DW displacement and the device conductance change is linearly dependent on the magnitude of the HM current. By biasing the interfaced transistor M STDP in the sub-threshold regime, current flowing through the transistor will vary exponentially with the gate voltage. For instance, in the case of LTP, the turning ON of “POST signal” gate combined with a linear increase of the “PRE signal” gate voltage, will lead to an exponentially varying programming current connecting the PRE and POST gates, depending on the timing window between PRE and POST signals. In presence of increases in the spike timing difference, the M STDP driven from cut-off to the sub-threshold region will decrease the HM current and thus the resultant conductance change ΔG exponentially. In order to ensure the programmability of DWM and resolution of STDP, it is required that the rise time of the M STDP 's gate voltage is much longer than the post-spiking duration. In the proposed configurations, the M STDP transistor with μs rise time of gate voltage and ns spiking duration were used (Sengupta et al., 2016a ). The capability of synaptic weight storage as well as STDP with simple peripheral circuits makes spintronic based hardware promising for emulating complex brain-like algorithms. Figure 4 (A) All-spin crossbar array with both synapse and neurons based on SOT-DWM devices. (B) DWM-based spintronic synaptic devices. (C) DWM-based analog and IF spiking neurons. Figures reproduced with permission from Sengupta and Roy, Appl. Phys. Rev. 4, 041105 (2017). Copyright 2017 AIP Publishing. In addition to synaptic plasticity, neuronal behaviors can also be realized based on the domain dynamics involving SOT-driven DWM. While a mono-domain SOT-MTJ or STT-MTJ is fully capable of making a stochastic spiking neuron based on the sigmoidal switching probability function of excitation currents (Sengupta et al., 2018 ), being able to generate intermediate state in device could enhance the versatility of spintronic neurons and greatly extend the capability of emulating complex neuron functionalities. Following the multi-domain magnetic configuration in a SOT-DWM device, an analog nueron can be implemented following a similar device structure of the synaptic device discussed above. Such analog devices with almost continuous output values can directly mimic the behaviors of saturating rectified linear units (saturating ReLU), or sigmoidal neurons which are predominantly applied in state-of-art deep artificial neural networks (Sengupta and Roy, 2017 ). Moreover, SOT-driven DWM could also utilize variations of positions in device to mimic IF neuron, which is an essential building block for hardware implementation of spiking neural networks. As is shown in the lower panel of Figure 4C , a DW located in the FL away from the MTJ sensing region may be pushed toward the MTJ region by applied current, in analogy to an increase of membrane potential due to accumulated intake of excitation. As soon as the DW enters the sensing area under the tunnel barrier, resistance change will be sensed following the magnetization switching due to DW motion and subsequently the output terminal will generate a spike. Therefore, displacement of DW positions enables the representation of changing membrane potentials in biological neurons, providing an increased level of bio-fidelity compared to a binary stepped neuron using single domain MTJ. 2.1.3. Prospect of Domain Wall Based Multi-Level Devices Compared to other memristive technologies, spintronic DWM devices provide a feasible means of leveraging material physics for low-power and high-endurance implementations of neuromorphic functionalities. In particular, the SOT-driven DWM device configuration enables realization of an all-spin neuromorphic computing block including both synapse and neuron units, while the low operation voltages as well as separated read and write paths at device level may lend further advantages. At present, DWM mechanism is one of the most investigated feature for developing spin-based neuromorphic devices. While huge potential of DWM devices has been demonstrated in both simulations and experiments, there are still challenges to address toward large scale practical implementation of such technology. For example, most DWM devices relies on precise control and sensing of a single domain wall, which often needs a quasi-1D device shape for confinement of the domain dynamics. Such constraint on device geometry lead to large footprints along the DW propagation direction (ranging from 1 to 10 μm), and thus hindering the deployment of the DWM devices beyond prototype demos (Cai et al., 2017 ; Jin et al., 2019 ; Siddiqui et al., 2019 ). Novel ideas such as introducing skyirmions have been investigated (Chen et al., 2018 ), potentially extending the current single wall based devices to multiple DW configuration (Song et al., 2020 ), but more work is needed to illustrate a viable path of scalability and controllability using skyrmions. Moreover, the assumption that DW displacement is linearly dependent on applied currents may not always hold in practice, due to the presence of random local device/material defects that may trigger irregular pinning/depinning, leading to erroneous results in real devices. These challenges in DWM devices also motivate the community in search of material-level mechanism beyond fully relying on motion of a single domain wall. As is discussed in the following subsection, multi-state devices built on exchange-coupled heterostructures could address some of these concerns and may pave a promising pathway for building scalable neuromorphic primitives. 2.2. Multi-Level Spin Devices Based on Exchange-Coupled Systems In this section, we will focus on two categories of approach based on magnetic exchange-coupled heterostructures. The first type relies on antiferromagnetic (AF) ordering, and particularly its interaction with ferromagnetic ordering to introduce non-coherent response to external excitation such as spin currents. The idea of using AF materials to modify FM in bilayer F/AF blocks has already been used such as exchange bias in MTJ stack used in magnetic sensor and MRAM (Parkin et al., 1999 ), while the adoption of AF for neuro-inspired device level granularity is emerging very recently (Fukami et al., 2016 ). The other type is to introduce micro-structure modifications into magnetic thin films in order to facilitate the divisions of magnetic domains with the help from material segregation. While the underlying technique of fabricating continuous/granular exchange coupled composites has been successfully implemented for boosting the storage density of binary magnetic storage over the past decade, its potential adoption for neuromorphic applications is only recently proposed (Wang et al., 2019 ). We will discuss the basic material selection and proposed device structure, followed by highlights of recent demonstrations of realizing neuro-inspired functionalities such as memristive synapses. 2.2.1. Material Physics of Exchange-Coupled Systems AF materials, which by definition have local spins ordered in compensated patterns (e.g., anti-parallel with neighboring spins), can facilitate multi-domains due to the absence of long range exchange interaction and dipole fields. In polycrystalline AF metallic thin films such as PtMn or IrMn, AF grains are formed with a dispersion of crytalline orientation and grain size, leading to variation in the AF ordering orientation and distribution of switching energy barrier among AF grains. Therefore, an inhomogeneous exchange bias is expected at a interface of FM/AF bilayer heterostructure. In presence of external fields or spin currents, the nucleation of the different regions in the FM layer may be impacted differently by the adjacent AF domains underneath, leading to multiple domains with non-coherent nucleation or gradual switching of the whole device area characterized by sloped hysteresis curves (Fukami et al., 2016 ). By replacing non-magnetic heavy metal with AF such as PtMn, a perpendicularly magnetized FM can be switched by SOT in a analog fashion, suggesting an exciting potential of integrating into practical SOT-MTJ devices, as is shown in Figure 5A . Meanwhile, it is reported that the multi-domain behavior will vanish when device dimension reduced to about 200 nm (Kurenkov et al., 2017 ), although the physical AF grains are typically as small as 15 nm. This observation suggests that magnetic cluster size in the continuous FM layer remains significantly larger than the scale of grain size, even under an inhomogeneous exchange bias from the AF grains. It remains challenging to find a feasible way so that adjacent AF can induce FM domains down to the size of AF grains. Interestingly, devices built with AF-only materials recently have also demonstrated multi-level resistance states, though more work is needed to search for better sensing mechanism of AF order without assistance of FM (Olejník et al., 2017 ). Figure 5 (A) Exchange-coupled F/AF bilayer structures with multi-level states. The top panel illustrates the bilayer structure and magnetic configurations, while the bottom panel shows multi-level Hall resistance and hysteresis loop. Figure adapted with permission from Borders et al. Appl. Phys. Express 10 013007 (2017). (B) Exchange-coupled continuous-granular structures with multi-level states. The top panel shows the material structure of the continuous/granular composite, while the bottom panel shows multi-state magnetization under an increasing writing field. Figure adapted with permissions from Choe et al. IEEE Trans. Mag., 41, 3172–3174 (2005) Copyright 2005 IEEE, Tham et al. IEEE Trans. Mags. 43, 671–675 (2007) Copyright 2007 IEEE, and Wang et al. U.S. Patent Application No. 16/255,698. In addition to bringing in F/AF exchange interaction from polycrystalline AF layer, interlayer exchange coupling between two ferromagnetic layers can also lead to effective splitting of the ferromagnetic order, if microstructure modifications can be introduced into one of the two ferromagnets. In pursuit of high density data storage, such technique has been matured and successfully implemented as exchange coupled composite medium in the magnetic recording industry. As is shown in Figure 5B , state of the art storage medium can have perpendicularly magnetized alloys (such as CoPt) grow in columnar structures with grain size averaged about 7–8 nm (Choe et al., 2005 ; Tham et al., 2007 ). More importantly, the intergranular coupling between the ferromagnetic columns can be greatly suppressed by the non-magnetic segregating oxide material (such as SiO 2 and TiO 2 ), leading to a magnetic cluster size similar to the physical grain size. The grains with a finite switching field distribution will respond to external excitation non-coherently, generating multi-domain states. It has been recently revealed that intermediate magnetic states can be retained under gradually increased external magnetic field, demonstrating a possibility of making memristors based on the multi-domain switching dynamics (Wang et al., 2019 ). Further studies are needed for validating reliable read and write schemes in such composite devices. In order to integrate such granular structures into integrated electronic devices, heterostructure of MTJ/granular layer may be used. FL of MTJ can be coupled to the granular layer via interfacial exchange coupling in order to read out the averaged magnetization in the multi-granular layer. As for writing mechanism, heavy metals such as β-W or Ta could be deposited as underlayer next to the granular layer, and thus SOT-driven switching can be exploited in addition to STT approach. 2.2.2. Neuromorphic Devices Based on Exchange-Coupled Heterostructures The possibility of inducing multi-states based on antiferromagnetic and exchange-coupled heterostructures ignited growing interest given its potential in developing neuro-inspired devices and hardware primitives. The field is still evolving rapidly today with ongoing efforts in various directions. As for F/AF heterostructures, It is recently shown that the non-volatile analog device build on [Co/Ni]/PtMn can provide synaptic weight matrix of a simple Hopfield Model which can be trained on device and realize associative memory operation as illustrated in Figure 6A (Borders et al., 2016 ). As is shown in Figures 6B,C , bio-plausible functionalities such as STDP and synaptic plasticity in response to input pulse trains are also demonstrated with the F/AF heterostructure, where STDP were achieved with pre-neuron and post-neuron spikes represented by opposite polarities (Kurenkov et al., 2019 ). Moreover, AF-only material also demonstrated synaptic behavior in response to accumulated pulses, opening up the possibility of AF-only neuromorphic devices ( Figure 6D ; Olejník et al., 2017 ). As for continuous/granular heterostructures, although multilevel magnetic states have demonstrated in CoPt-based composites, device integration into compact memristive prototype with MRAM type of read/write remains to be shown further down the road. Figure 6 Neuromorphic implementation based on exchange-coupled heterostructures. (A) Modifications of synaptic weights stored in F/AF devices after training for pattern recognition. Figure reproduced with permission from Borders et al. Appl. Phys. Express 10 013007 (2017). (B) STDP demonstration based on the Hall resistance of F/AF devices. (C) Effect of input pulses on the synaptic state of F/AF devices. Figures reproduced with permission from Kurenkov et al. Advanced Materials 31, 1900636 (2019). (D) Resistance modifications of an antiferromagnetic material CuMnAs under pulses of various lengths. Figure reproduced from Olejník, K. et al. Nat. Commun. 8, 15434 (2017). Copyright 2017 Author(s), licensed under a Creative Commons Attribution (CC BY) license. 2.2.3. Prospect of Multi-Level Devices Using Exchange Coupled Systems The leverage of exchange-coupled magnetic systems in achieving multi-level resistive states for neuro-inspired devices brings in new momentum into neuromorphic spintronics. With utilization of fundamental material properties, those multi-level devices based on the non-coherent switching of multi-domains becomes less dependent on special device geometry, which is usually needed for the quasi-1D confinement of a single domain wall. Although both F/AF and ferromagnetic continuous/granular exchange coupled heterostructures hold great promises, critical challenges such as scalability of domains confined by grains and integration with TMR-like readout mechanism will have to be addressed. At present, the resistance change of several Ohms (or less than few percents) in most of recent device-level demonstrations are not feasible for CMOS circuits to sense and process in a integrated chip. Combined efforts of microstructure segregation and antiferromagnetic order on MTJ-based platform could potentially pave the way for providing scalable multi-level spintronic integrated devices for neuro-inspired computing.\n\n2.1. Multi-Level Spintronic Devices Based on Domain Wall Motion A natural path to multilevels in spintronics is to split the single-domain magnetizations as formed in MRAM devices to multiple domains (Fong et al., 2016 ). It is known that switching of magnetic thin films can involve mechanisms of nucleation formation and domain propagation, suggesting a possibility to generate intermediate multi-domain of magnetic textures between the bi-stable states. In order to have stable configurations of multiple domains in continuous ferromagnetic thin films, additional mechanisms are required to maintain the pinning of domain walls between spin-up and spin-down regions. The DWs can be pinned or displaced, depending on the combined effects of material properties such as exchange coupling among magnetic moments, shape anisotropy determined by device geometry, as well as local defects (Beach et al., 2006 ; Thomas et al., 2006 ; Emori et al., 2013 ). Conceptually, such domain wall motion (DWM) in a ferromagnetic thin film can generate a near-continuous variation of magnetic states from one direction to the other, resulting in variable resistance states of a magneto-resistive device described by a model with parallel resistors. 2.1.1. Device Fundamentals of Domain Wall Motion The idea of using DWM driven by current-induced torque has sparkled a plethora of studies built on the mostly matured STT-switching technology (Wang et al., 2009 ; Lequeux et al., 2016 ). It has been proposed and demonstrated that with special engineering of device geometry, non-volatile multi-level resistance states can be realized in an MTJ with perpendicular magnetic anisotropy (PMA). As is shown in Figure 3A , a DW in the free layer of MTJ with the shape of an elongated stripe can be displaced by applied electric currents, leading to modifications in MTJ resistance following the relationship: (1) G ( x ) = G P * x / L + G A P * ( 1 - x / L ) + G D W where G P and G AP are the parallel and antiparallel conductance respectively, and x is the domain wall position in a stripe of length L. STT-DWM is controlled by the magnitude and polarity of the spin polarized currents across the FL. The current-driven DWM device can potentially work as a two-terminal compact device following the STT-MRAM configuration. Furthermore, it was recently found that in magnetic heterostructures such as oxide/ferromagnetic/heavy metal stacks, a chiral DW with Néel configuration can be formed and stabilized in perpendicularly magnetized thin films due to Dzyaloshinskii-Moriya exchange interaction (DMI) at the FM/HM interface and the broken inversion symmetry in the heterostructure stack (Emori et al., 2013 ). It is observed that the Néel Wall can be efficiently driven by the spin orbit torque (SOT) originated from spin Hall effect (SHE) of the heavy metal layer, as is shown in Figure 3B . Therefore, DWM driven by SOT in a MTJ/HM heterostructures can be leveraged for programming the multi-level device conductance (Sengupta et al., 2016b ). While in general a magnetic field is needed for deterministic SOT switching of PMA materials due to the in-plane spin polarization (Liu et al., 2012 ), it is found that the interfacial DMI could effectively provide the desired magnetic field toward field-free SOT driven DWM (Emori et al., 2013 ). Current-driven DW motion in heavy-metal/ferromagnet/oxide structures is naturally explained by the combination of the SHE and DMI. The SHE produces the sole current-induced torque, and DMI stabilizes chiral DWs while permitting uniform DW motion with very high efficiency. The writing speed of DWM-SOT devices is characterized by the domain wall velocities, which increases with the current density up to saturation in absence of pinning sites. Although domain wall velocity can be as high as 10 2 m/s (Beach et al., 2006 ; Agrawal and Roy, 2018 ), in practice DW motion can be hindered due to pinning in magnetic thin films, making it challenging to have precise control and efficient manipulation of DWs (Thomas et al., 2006 ). The readout mechanism in the proposed device is similar to that of a STT-MRAM memory cell, while the SOT is generated from a lateral charge currents which has a separated flow path from the reading operation across MTJ stack. The major advantage of the three-terminal device is the decoupled read and write paths which will eliminate the read disturbance issue of MTJ and thus lead to significant improvement in device endurance. Recently it is confirmed experimentally that SOT-DWM can be used for artificial synaptic devices in MgO/CoFeB/Ta heterostructures. In the following subsection, various prototypes implementing neuromorphic functionalities based on DWM spintronic devices are discussed. Although we focus on SOT-driven configurations given the advantage of separate read/write paths, similar mechanism will also work for STT-driven configurations. Figure 3 Domain wall motion (DWM) based multi-level devices. (A) STT-driven DWM in a MTJ-based device. Lequeux et al. Scientific Reports 6, 31510 (2016), Copyright 2016 Author(s), licensed under a Creative Commons Attribution (CC BY) license. (B) SOT-driven DWM device. Dzyaloshinskii-Moriya exchange interaction across the heavy metal and ferromagnetic layer provides effective magnetic fields that ensure deterministic switching. Reproduced with permission from Sengupta et al. IEEE Transactions on Biomedical Circuits and Systems, 10, 1152–1160 (2016), Copyright 2016 IEEE. 2.1.2. Domain Wall Motion Based Neuromorphic Devices A direct application of SOT-driven DWM device will be crossbar implementation for MVM computing engine. Leveraging a linear dependence of conductance on the DW position which is subsequently linearly dependent on the driving electric current (before saturation), SOT-DWM devices in a crossbar array can have their DW position (and thus conductance states) accurately programmed to map a synaptic weight matrix, as is illustrated in Figure 4 (Sengupta et al., 2016b ; Sengupta and Roy, 2017 ). Parallel dot product of vector (voltage) and matrix (conductance) can be directly executed following the Kirchoff Current Law. Given the non-volatility of conductance states in those multi-domain devices, we could just set the devices once with pre-trained weight matrix and reuse the stored weights during inference, eliminating additional memory access and data transfer. The advantage of separated read and write paths in SOT-DWM is evidently demonstrated, as the MVM operations at inference only involves reading path and thus read disturbance to the states is minimized. The DWM-based multi-level device can also implement STDP, another important synaptic characteristics. With STDP learning rule, presynaptic spike arrival before the occurrence of postsynaptic spike leads to long-term potentiation (LTP) of the connecting synapse, whereas spike arrival after postsynaptic spike leads to long-term depression (LTD) of the same synapse. The magnitude of the relative change in synaptic strength (ΔW) decreases exponentially with the timing difference between the preneuron and postneuron spikes. A key step to realize STDP with a DWM-based spin device is to link the timing of the pre-neuron and post-neuron to the conductance change in the interconnected SOT-driven DWM synapse. One approach as illustrated in Figure 4 is to provide exponential variation of HM currents modulated by the timing of PRE and POST neurons in circuit, assuming that the DW displacement and the device conductance change is linearly dependent on the magnitude of the HM current. By biasing the interfaced transistor M STDP in the sub-threshold regime, current flowing through the transistor will vary exponentially with the gate voltage. For instance, in the case of LTP, the turning ON of “POST signal” gate combined with a linear increase of the “PRE signal” gate voltage, will lead to an exponentially varying programming current connecting the PRE and POST gates, depending on the timing window between PRE and POST signals. In presence of increases in the spike timing difference, the M STDP driven from cut-off to the sub-threshold region will decrease the HM current and thus the resultant conductance change ΔG exponentially. In order to ensure the programmability of DWM and resolution of STDP, it is required that the rise time of the M STDP 's gate voltage is much longer than the post-spiking duration. In the proposed configurations, the M STDP transistor with μs rise time of gate voltage and ns spiking duration were used (Sengupta et al., 2016a ). The capability of synaptic weight storage as well as STDP with simple peripheral circuits makes spintronic based hardware promising for emulating complex brain-like algorithms. Figure 4 (A) All-spin crossbar array with both synapse and neurons based on SOT-DWM devices. (B) DWM-based spintronic synaptic devices. (C) DWM-based analog and IF spiking neurons. Figures reproduced with permission from Sengupta and Roy, Appl. Phys. Rev. 4, 041105 (2017). Copyright 2017 AIP Publishing. In addition to synaptic plasticity, neuronal behaviors can also be realized based on the domain dynamics involving SOT-driven DWM. While a mono-domain SOT-MTJ or STT-MTJ is fully capable of making a stochastic spiking neuron based on the sigmoidal switching probability function of excitation currents (Sengupta et al., 2018 ), being able to generate intermediate state in device could enhance the versatility of spintronic neurons and greatly extend the capability of emulating complex neuron functionalities. Following the multi-domain magnetic configuration in a SOT-DWM device, an analog nueron can be implemented following a similar device structure of the synaptic device discussed above. Such analog devices with almost continuous output values can directly mimic the behaviors of saturating rectified linear units (saturating ReLU), or sigmoidal neurons which are predominantly applied in state-of-art deep artificial neural networks (Sengupta and Roy, 2017 ). Moreover, SOT-driven DWM could also utilize variations of positions in device to mimic IF neuron, which is an essential building block for hardware implementation of spiking neural networks. As is shown in the lower panel of Figure 4C , a DW located in the FL away from the MTJ sensing region may be pushed toward the MTJ region by applied current, in analogy to an increase of membrane potential due to accumulated intake of excitation. As soon as the DW enters the sensing area under the tunnel barrier, resistance change will be sensed following the magnetization switching due to DW motion and subsequently the output terminal will generate a spike. Therefore, displacement of DW positions enables the representation of changing membrane potentials in biological neurons, providing an increased level of bio-fidelity compared to a binary stepped neuron using single domain MTJ. 2.1.3. Prospect of Domain Wall Based Multi-Level Devices Compared to other memristive technologies, spintronic DWM devices provide a feasible means of leveraging material physics for low-power and high-endurance implementations of neuromorphic functionalities. In particular, the SOT-driven DWM device configuration enables realization of an all-spin neuromorphic computing block including both synapse and neuron units, while the low operation voltages as well as separated read and write paths at device level may lend further advantages. At present, DWM mechanism is one of the most investigated feature for developing spin-based neuromorphic devices. While huge potential of DWM devices has been demonstrated in both simulations and experiments, there are still challenges to address toward large scale practical implementation of such technology. For example, most DWM devices relies on precise control and sensing of a single domain wall, which often needs a quasi-1D device shape for confinement of the domain dynamics. Such constraint on device geometry lead to large footprints along the DW propagation direction (ranging from 1 to 10 μm), and thus hindering the deployment of the DWM devices beyond prototype demos (Cai et al., 2017 ; Jin et al., 2019 ; Siddiqui et al., 2019 ). Novel ideas such as introducing skyirmions have been investigated (Chen et al., 2018 ), potentially extending the current single wall based devices to multiple DW configuration (Song et al., 2020 ), but more work is needed to illustrate a viable path of scalability and controllability using skyrmions. Moreover, the assumption that DW displacement is linearly dependent on applied currents may not always hold in practice, due to the presence of random local device/material defects that may trigger irregular pinning/depinning, leading to erroneous results in real devices. These challenges in DWM devices also motivate the community in search of material-level mechanism beyond fully relying on motion of a single domain wall. As is discussed in the following subsection, multi-state devices built on exchange-coupled heterostructures could address some of these concerns and may pave a promising pathway for building scalable neuromorphic primitives.",
"discussion": "4. Discussion The multi-level device characteristics utilizing collective multi-domain dynamics of magnetization or electric polarization switching have successfully demonstrated device-level emulation of neural functionalities and could be leveraged to build robust and energy efficient bio-plausible hardware primitives for AI applications. Spintronic and ferroelectric devices could potentially provide some advantages compared with ReRAM/PCM and CMOS technologies. The spin-based devices require a lower programming voltage compared to ReRAM (Park et al., 2013 ; Adam et al., 2017 ) and PCM (Papandreou et al., 2011 ; Tuma et al., 2016 ). They also demonstrate higher endurance compared to ReRAM and PCM (Prenat et al., 2016 ; Li et al., 2019 ). While the multi-bit capability of spintronic devices may be limited by the low ON/OFF ratio (G On /G Off ~ 2–3), it is worthwhile to note that high precision weight matrices can be mapped to multiple crossbars in large scale implementation of in-memory computing. Moreover, multi-level cells having large number of states would typically require a higher precision readout circuitry to interface between digital and analog domains, dominating the power and area costs (Shafiee et al., 2016 ; Ankit et al., 2019 ). On the other hand, FeFET and FTJ could provide large dynamic ranges (G On /G Off ≤ 10 2 ) with numerous intermediate states. Hence, multi-level spintronic devices can be well suited for implementing frequently updated components such as neurons, while ferroelectric devices are considered to be more suitable for implementing analog synapses, given the device characteristics of large memory windows between states, low read/write energy, fast switching, and superior CMOS compatibility (Khan et al., 2020 ). In the following, we will elaborate particular challenges of implementing crossbar in-memory computing based on ReRAM and PCM materials, and highlight the strength in spintronic and ferroelectric device characteristics that could potentially address some of the challenges. A near-term scenario of NVM based neuromorphic computing is to execute AI inference tasks with pre-trained models mapped into crossbar arrays, while a more challenging scenario is to enable on-chip learning. As for the inference-only scenario, the crossbars are in read-only mode. The MVM operations will be executed based on the product of the input voltage and the weight matrix, which are stored as device conductances. In general, it is desirable to have large storage density of devices with distinctive states and strong data retention against thermal agitation and other relaxation mechanisms over an extendable time. PCM based on alloys such as Ge 2 Sb 2 Te 5 , although capable of high density with large dynamic range (G on /G off ~ 10 3 ), suffers resistance drift due to relaxation of the amorphous state (high resistance states). Such drift in device conductance significantly degrades the desirable data retention and thus requires additional circuit-level compensation scheme in real applications, leading to additional energy consumption and delay (Yu and Chen, 2016 ). As for filament-based ReRAM with oxides such as HfO x /TaO x , although it has advantages of compact cell/array size, large device variability (especially at high resistance states) can be a major hindrance (Yu and Chen, 2016 ; Li et al., 2019 ). The large device variation not only places challenges on the sensing circuit but also leads to a reduced number of bits per cell, even when the device-level conductance ON/OFF ratio is high (Chakraborty et al., 2020b ). On the other hand, the memory effect of ferroic (ferromagnetic or ferroelectric) orderings are well-poised given their advantages in storing information with superior retention. In particular, spintronic devices can store information based on the magnetizations in materials with strong anisotropy. The thermal stability of the bits stored in spintronic devices is governed by ratio of the energy barrier of switching over thermal fluctuation Δ = E b / k B T , where E b is the effective energy barrier of magnetization switching, k B is the Boltzmann constant and T is the operation temperature. With Δ ~ 60–80 in current spintronic devices, superior data retention has been demonstrated on the order of years in memory and data storage applications. In addition, the fact that electro-forming is overall a one-dimensional process suggests that device and cycle variations could always be an issue with filament-based device such as ReRAM. Note, multi-domain devices are fundamentally not limited to 1D process and in principle, can be modeled as parallel conduction channels. Such conceptually parallel channels can be less prone to variations, thanks to the averaging effects of the collective channels. Therefore, both spintronic and ferroelectric devices suffer less from device variability compared to filament based memristive technologies. Another pressing challenge of running MVM on crossbar arrays is the non-ideality associated with crossbar circuits. While large crossbars are desirable to utilize the massive parallelism in crossbar based MVM, the non-ideality due to voltage drop along wire resistance and other circuitry components (“IR” drop) will become more severe at larger crossbars, leading to non-ideal output current values compared to the ideal crossbar output current I = G · V. Note that using devices with low ON-state resistance will be more susceptible to the non-ideal IR drop, and thus a high ON-state resistance is desirable for accurately performing crossbar-based inference task (Chakraborty et al., 2020a ). PCM and ReRAM typically have ON-resistance of 10–100 kΩ, while FeFET can operate at ON-resistance of higher than 500 kΩ (Jerry et al., 2017 ). Emerging spintronic devices such as spin orbit torque (SOT)-MTJ can potentially work under MΩ ON-resistance, due to separated read/write paths (Doevenspeck et al., 2020 ). The ability of providing large ON-state resistance can potentially enable deployment of large crossbars with minimal impact on computing accuracy. On-chip learning and training using crossbar arrays of memristive devices are even more challenging, since both read and write operations will be involved and thus the device conductance in crossbar arrays will be re-written frequently throughout training. One major issue of both ReRAM and PCM is the limited endurance (10 6 –10 8 write cycles). The chalcogenide alloy in PCM experiences frequent expansion and contraction under heating during writing process, and the high probability of physical detachment at the interface between the alloy and the heating elements can lead to permanent defects in devices, leading to hard errors such as stuck at fault when device states are no longer changeable. As for the ReRAM, both intrinsic device variation originated from fabrication process and the writing/resetting of ReRAM devices can contribute to severe device defects that can significantly degrade the accuracy of the crossbar in-memory computing. In contrast, the endurance of ferromagnetic devices is much better in comparison with ReRAM and PCM, since switching in spin-based devices do not need electro-forming of conduction filaments (or forming conductive phase through heat-induced crystallization). The endurance of production-ready STT-MRAM can be as high as 10 14 –10 15 cycles with write latency of 10 ns, making MRAM much more competitive for data memory requiring frequent updates. And newly emerging SOT-MTJ equipped with separated read and write current paths could provide even more flexibility in design fast and high-endurance devices. The leverage of MRAM technology for neuromorphic computing can provide a viable path toward making crossbar arrays with high endurance for training and learning. Moreover, writing in PCM and ReRAM requires high energy consumption and latency due to the device characteristics. The phase transition in PCM needs substantial write current (~100 μA for a 20 nm device as reported in Kang et al., 2011 ) to heat and melt the material, and the switching speed is limited by the relatively slow crystallization process (>50 ns; Yu and Chen, 2016 ). As for the filament-based ReRAM materials, large device-to-device and cycle-to-cycle variability resulting from the stochastic nature of ion migration and variation of the filament shape and structure have always been a critical issue for training in ReRAM devices. Although single write of a ReRAM device can be fast and efficient, multiple write-verify cycles are needed to set a device into a desirable conductance level, effectively raising the cost of writing. Therefore, training directly on arrays of PCM/ReRAM devices can lead to high energy consumption and significant delay as well as degradation in computing accuracy due to device failure. The high cost of writing with ReRAM and PCM crossbars can be potentially mitigated by using devices based on spintronic and ferroelectric materials. Changes of device states in ferromagnetic and ferroelectric materials rely on magnetization or polarization switching, which can be more energy efficient since no crystallization/melting process or significant ionic motion are involved. Note that utilizing the multi-domain switching dynamics provides a statistical averaging effect over the multiple domains involved, reducing the device and cycle variations. In addition, the implementations of neurons, compared to synapses, typically have more relaxed requirement of density, but may have more stringent requirement on endurance due to the possibly frequent occurrence of activation. At present, most of the available ReRAM and PCM materials do not have the desirable endurance to execute neuron activation. Therefore, we believe that emerging ferroic devices and systems with superior endurance can have exciting opportunities in providing non-volatile neuronal devices in addition to synapses. Such implementation could potentially lead to significant improvement in energy efficiency due to reduced data movement between memory and neuron units. In spite of the tremendous progress in implementing synaptic as well as neuronal functionalities based on a plethora of multi-domain materials, further explorations are still needed to tackle several key challenges. First of all, large device footprints are typically desirable for demonstrating multi-domain switching mechanism in either spintronic or ferroelectric devices, limiting the synaptic memory density that can be implemented. In magnetic devices, multi-domain switching are found to vanish in ferromagnets as the device size decreased to 200 nm, even in presence of interlayer coupling to adjacent multi-domain antiferromagnets (Kurenkov et al., 2019 ). Therefore, achieving energy-efficient generation and control of multi-domains in a scalable spin-based material/device platform still requires further explorations. In ferroelectric devices, most demonstrations of memristive behaviors to date are done on devices with lateral size of 200 nm to few microns, but it remains to be seen if the multi-level polarization switching can still persist as devices dimension shrinks to the size comparable to size of a nucleation site in ferroelectric thin films. Moreover, the spintronic and ferroelectric technologies also have their own challenges to address. As for spintronic devices, continuous efforts are needed for smooth integration of multi-domain functional blocks with manufacturable and programmable devices such as MTJs. In particular, how to utilize novel mechanisms such as voltage-controlled magnetic switching, spin-orbit torques in emerging materials such as antiferromagnets, multiferroics, or granular composite structures for generating multi-level in a scalable integrated devices will be of significant interest. Magnetic devices are further challenged by limited readout resolutions due to the relatively small magneto-resistance effect in practical MTJs. Fortunately, emerging SOT-MTJ with separate read and write paths could lead to device optimized in new design space with possibly higher ON/OFF ratios (up to 6x larger TMR ratio) that could be better suited for neuromorphic computing (Ikeda et al., 2008 ; Doevenspeck et al., 2020 ). And novel materials such as half metal have been predicted to potentially boost the magnetoresistive signals in future (Bhatti et al., 2017 ). In devices built on ferroelectric thin films (thickness <10 nm), growing depolarizing effects and charge redistribution/trap at interfaces across the heterostructures significantly limit the retention and endurance of the remnant polarization states. Therefore, further investigations of materials and device physics will be crucial in achieving reliable and scalable multi-domain neuro-inspired hardware primitives toward sub-10-nm domains. The multi-level spintronic and ferroelectric devices, which enable MVM acceleration and efficient emulation of neural functionalities, provide building blocks for implementing large-scale computing system. Note that complex AI tasks such as image classifications or natural language processing usually need to run large DNN models comprising multiple layers, where the size of involved matrices can be significantly larger than typical crossbar sizes. Furthermore, in order to yield satisfactory prediction accuracy, the model parameters in AI algorithms such as DNN will need higher resolution (for example, 16–32 bit synaptic weights) compared to the highest possible bits per cell in synaptic devices (about 1–5). Therefore, the mapping of input vectors and weight matrices into hardware may involve multiple arrays or multiple columns (Shafiee et al., 2016 ). The bits of input vector are converted to voltages, and they can be streamed in time depending on the bit-precision of the digital-analog converters (DAC) (Ankit et al., 2019 ). The accumulated current in crossbar column obtained from the dot product of the input voltage vector and conductance matrix will be processed in combination with results from multiple time steps (bit streaming) and/or multiple columns/arrays (bit slicing) to execute multi-bit MVM operations. In order to minimize noise accumulation through these aforementioned multi-bit operations, digital circuitry such as shift and adder are needed, requiring the conversion of analog signals to digital domains via analog-digital converters (ADC). Moreover, processing of partial sums from multiple crossbar arrays are needed in the case of large matrices for generating the final outputs, which also desire ADCs in order to leverage the noise resilience of digital circuitry. As is shown in Figure 11 , crossbar arrays of multi-level devices provide building blocks for partitioning and mapping of a neural network model, so that workflows involving large and high precision models can still leverage the advantage of the underlying device characteristics of non-volatile multi-level memory. Crossbar arrays of devices with large number of bits per device will increase the requirement of ADC bit precision, leading to significant increase of power and area costs dominated by the ADC (Shafiee et al., 2016 ). To this effect, the footprint and energy cost of crossbar array itself may not be the bottleneck. While more crossbars will be needed to map the algorithmic models using devices with fewer bits per cell, better energy efficiency could be reached with low-precision ADCs. Therefore, depending on which factor dominates the system-level performance, a trade-off analysis could lead to optimized designs with different crossbar and ADC configurations. The impacts of device/circuit level settings such as device conductance range, crossbar size, number of states per cell have been demonstrated on the performance of implementing high level AI algorithms (Chakraborty et al., 2020a ), while algorithmic characteristics such as sparsity in model parameters can also influence the choice of setting and operation of the crossbar arrays. Therefore, systematic trade-off analysis and application specific co-design across the hierarchy of material, device, circuit, and algorithm will be crucial toward the optimization of system-level performance, and we hope that this review will inspire more research efforts on utilizing the emerging ferroic device technologies toward developing efficient neuro-inspired computing system comprising both hardware and software. Figure 11 In-memory computing architecture involving multiple tiles of interconnected crossbar arrays to process high dimensional MVM operations with high bit precision. (A) Illustration of the multi-tile architecture for executing large input vector with high bit precision where multiple crossbars will be used for mapping the input and weight matrix. (B) Illustration of the circuit of a single crossbar array comprising the array of synaptic devices in series with access transistors and peripheral analog and digital circuits. A multiplexer (MUX) and transimpediance amplifier (TIA) are directly connected to the crossbar columns, followed by sample and hold circuit (S&H) as well as ADC and shift and add circuitry."
} | 16,645 |
36505713 | PMC9682623 | pmc | 341 | {
"abstract": "Among different types of anti-icing coatings, superhydrophobic surfaces have attracted extensive attention due to their excellent water repellency and low thermal conductivity. We report facile spraying time tuning to optimize the superhydrophobic (SHP) surface coating fabrication by a one-step spraying method of mixing SiO 2 nanoparticles with epoxy resin (EP), polyamide resin (PAI), and HFTMS. The wettability performance was optimized by adjusting spraying time from 0 s to 25 s to control surface morphology by adjusting surface morphology and line roughness. With spraying time of 20 s, SiO 2 molecular clusters on the superhydrophobic surface showed a maximum water contact angle (WCA) of 160.4° ± 1.3° and a sliding angle (SA) of 4.1° ± 1.0°. What's more, the effect of the coatings' icing behavior were studied by icing heat conduction; SHP-20 delayed the icing time for 410 s at −15 °C, and the icing performance of SHP-20 also declined with the decrease of temperature to −9 °C, −12 °C, −15 °C, and −18 °C. The WCA of SHP-20 can remain above 140.9° ± 1.8° after 40 abrasive 1000# sandpaper wear cycles. The results also provide a basis for the preparation of SHP and anti-icing characteristics.",
"conclusion": "Conclusions We have fabricated a super hydrophobic aluminum surface in which SiO 2 nanoparticles and HFTMS combine to form low surface energy substance and mixed with EP and PAI by one-step spraying method by adjusting spraying time. Our results show the performance of the coating with different spraying time changes in surface morphology and surface roughness, SHP-20 shows the highest water contact angle of 160.4° ± 1.3° and a sliding angle of 4.1° ± 1.0°. Due to their different wettability, we designed experiments and proved that the coatings have anti-icing performance with different wettability at −15 °C by simplifying the theory heat conduction. At the same time, the lower surface temperature can greatly short the icing time. The SHP-20 surface coating also shows good mechanical properties and durability performance. This study provides a low cost produced method to study the influence of surface morphology on wettability and anti-icing properties for various applications.",
"introduction": "Introduction Inspired by the special properties of natural biological surfaces including lotus leaf surfaces and butterfly wings, superhydrophobic (SHP) surfaces have applications in many emerging industries, including anti-icing, 1 self-cleaning, 2 fluid rent reduction, 3 etc. In the application of metal materials, the hydrophilicity of their surfaces often leads to material failure. HFTMS as a known and lowest surface energy modifier yet recorded (6.7 mJ m −2 was attained for a surface with regularly aligned closest-hexagonal packed –CF 3 groups) has been proved to be able to react with hydrophilic –OH group to form hydrophobic groups with low surface energy. 4,5 At the same time, it can also enhance the bonding ability between metal matrix and organic coating. 6,7 SiO 2 nanoparticles are a hydrophilic and attractive material for superhydrophobic coatings which have –OH group on the surface due to cheaper, 8 chemically stable, 9 harmless, and low thermal conductivity. 10 Thus, we can modify SiO 2 molecules to obtain a stable superhydrophobic coating and study its reaction mechanism. The surface with water contact angle more than 150° can be realized by adjusting the surface structure and surface energy. 11 Currently, a large number of superhydrophobic materials have been prepared on different substrates by chemical electrospinning, 12 laser etching, 13–15 electrochemical treatment, 16,17 physical or chemical deposition, 18,19 and self-assembly 20,21 to get superhydrophobic wettability. T. P. Manoj et al. 22 fabricated a superhydrophobic surface by changing the etching time of titanium alloy in acid, the coating exhibits a maximum WCA of 162.3° ± 1° with a sliding angle of 1°. A. Gaddam et al. 23 studied one step femtosecond laser processing to get single-tier nano and two-tier multiscale structure to learn more about the surface morphology and surface hydrophobicity, and the experimental results show that the multi-scale surface morphology features have good hydrophobicity. However, most of these methods require fine technology and damage of surface metal characteristics that cannot achieve large-scale production. In addition, many researchers obtain superhydrophobic surfaces by changing the surface energy parameters of coatings. 24,25 The coating prepared by spraying can be applied to most alloys and has low thermal conductivity, and spraying low surface energy substances on the surface can effectively obtain appropriate surface roughness that the surface has superhydrophobic properties economically. W. Guo 26 prepared a superhydrophobic surface with fluorinated epoxy and unmodified-nano SiO 2 and micron SiO 2 , the contact angle for water was 158.6° ± 1°. However, the anti-icing performance of the coatings has not been satisfactorily achieved. As a result, most superhydrophobic surfaces can be ideal candidates for anti-icing surfaces, 27,28 the micro–nano structure design and water-repellent properties of coatings have a great influence on the icing performance of droplets on the coating surface. 29,30 Further, delaying the freezing process of droplets mechanism with an incomplete description and understanding and the function of superhydrophobic coatings icing behaviors under different temperature which also limits its practical application. In this paper, we report a simple, economical preparation method by coatings in which the EP, PAI, and HFTMS were self-assembled onto SiO 2 particles, and study the reaction process. Experiment was designed as a one-step spraying method to get superhydrophobic coating (SHP) that studied the surface morphology of spraying time with 0 s to 25 s. The icing behaviors of SHP surfaces were optimized at −15 °C, and coatings with the best hydrophobicity were studied at the different temperature of −9 °C, −12 °C, −15 °C, and −18 °C through the self-made anti-icing experimental process. And the durability and mechanical properties of the surface with the best hydrophobicity were tested.",
"discussion": "Results and discussion Analysis of modification mechanism of fluorinated SiO 2 particles \n Fig. 2(a) shows the formation mechanism of superhydrophobic coating on fluorinated SiO 2 surface. The reaction is composed of HFTMS and SiO 2 nanoparticles. HFTMS is composed of 5-CF 2 - (surface energy of 18 mJ m −2 ) and 1-CF 3 (the surface energy is 6.7 mJ m −2 ), 5 this is the lowest surface energy of water repellent. During the reaction, its fluorinated group Si–OC 2 H 5 will react to produce silanol Si–OH. C 13 H 13 F 17 O 3 Si molecules would shrink with each other to form a covalent bond connection, and a low surface free energy hydrophobic group film can be effectively formed on the microstructure surface. As a result, HFTMS (C 13 H 13 F 17 O 3 Si) is mixed with it to obtain nanomaterials with a hydrophobic surface. In the process of low surface energy treatment, fluorinated treatment will not affect the surface morphology of SiO 2 particles. Fig. 2(b) shows the FTIR spectrum of fluorinated modified SiO 2 . It can be seen in Fig. 2(b) that the absorption peaks at 1073 cm −1 and 472 cm −1 belong to the characteristic absorption band of Si–O–Si. After fluorination modification, the peak of –OH at 2851 cm −1 and 2935 cm −1 of SiO 2 nanoparticles disappeared, the characteristic absorption peak of the –CF 2 bond appears near 1144 cm −1 , the characteristic absorption peak of C–F exists near 1243 cm −1 , and the epoxy group absorption peaks of epoxy resin and polyamine resin are at 937 cm −1 and the peak of 1730 cm −1 were the stretching vibration peaks of C \n \n\n<svg xmlns=\"http://www.w3.org/2000/svg\" version=\"1.0\" width=\"13.200000pt\" height=\"16.000000pt\" viewBox=\"0 0 13.200000 16.000000\" preserveAspectRatio=\"xMidYMid meet\"><metadata>\nCreated by potrace 1.16, written by Peter Selinger 2001-2019\n</metadata><g transform=\"translate(1.000000,15.000000) scale(0.017500,-0.017500)\" fill=\"currentColor\" stroke=\"none\"><path d=\"M0 440 l0 -40 320 0 320 0 0 40 0 40 -320 0 -320 0 0 -40z M0 280 l0 -40 320 0 320 0 0 40 0 40 -320 0 -320 0 0 -40z\"/></g></svg>\n\n O. 32 Therefore, it can be concluded that the low surface free energy groups have been successfully self-assembled to the microstructure surface, reducing the microstructure surface free energy to obtain a superhydrophobic surface. Fig. 2 (a) Formation mechanism of modified SiO 2 nanoparticles on the surface; (b) FTIR diagram of unmodified SiO 2 nanoparticles and fluorinated SiO 2 nanoparticles. Analysis of surface morphology of spraying time. The key factor in preparing SHP surfaces is to establish a rough surface morphology surface. Fig. 3 shows the surface morphology of aluminum alloy after different spraying time modification by SEM scanning microscope. It can be seen that the surface of unmodified aluminum alloy, with a smooth surface structure and only some regular protrusions and grooves from Fig. 3(a) . With the change of different spraying time, Fig. 3(b) shows the surface morphology of 5 s is relatively flat which can't form a structure. It is observed from Fig. 3(c) and (d) that the particles on the surface gradually agglomerate and adhere to the aluminum surface to form a rough microstructure like peak structure which built up by the spraying pressure that begin to appear on 10 s and 15 s. Fig. 3(e) shows the increase of spraying time leads to the gradual formation of particle agglomeration on the surface from flat, which is mainly attributed to the existence of fluorinated SiO 2 particles on the surface, and EP and PAI are easy to agglomerate to form obvious micro–nano structure. When the coating spraying time at 25 s, the excessively thick micro nano structure will cover micro–nano structure and the increase of solid–liquid area. Fig. 3 SEM images of SHP surface morphology with different spraying time: (a)SHP-0; (b) SHP-5; (c) SHP-10; (d) SHP-15; (e) SHP-20; (f) SHP-25. Analysis of surface wettability of spraying time The preliminary wettability and adhesion behaviors of SHPs were tested by the water contact angle (WCA) and sliding angle (SA), respectively. SiO 2 nanoparticles are hydrophilic inherently, after the fluorination reaction, the coating surface has certain different hydrophobicity by adjusting the spraying time. With the increase of spraying time between 0 s and 25 s, SHP with different surface morphology enables the interaction between the surface and water droplets differently. A trend of WCA shows increases first and then decreases at a small range in Fig. 4 . The WCA of the surface is less than 150° while the spraying time is less than 10 s, which means that changing the spraying time can adjust the WCA of the surface. Similarly, when the spraying time is 20 s, the hydrophobicity of the surface is 160.4° ± 1.3° which repels the water excellently. When the spraying time continues to increase, the hydrophobicity of the surface begins to decline which can be attributed to the following two reasons. On the one hand, the reason is related to the unstable surface morphology and leads to a large error in the surface wettability test results when the surface coating is too thick. On the other hand, if the coating thickness continues to increase, the internal stress may be larger, that causing cracks on the coating surface. Sliding angle explains the behavior of surface adhesion. Water droplets adhere to the surface (from 23.1° ± 1.5° to 12.0° ± 3.1°) while the spraying time is less than 10 s. With the spraying time reaching 20 s, the adhesion of the surface reaches the minimum which the sliding angle is 4.1° ± 1.0°. The phenomena indicates that the wetting state between SHPs and water droplets varied from spraying time. Fig. 4 The WCA and SA different spraying time of SHPs coatings with different spraying time. Surface roughness and wetting model spraying time of the coating To better understand the effect of spraying time on the surface coating, understand the surface morphology and the change law of surface wettability, we characterized the surface morphology and roughness parameters of SHPs with Ultra-Depth 3D Microscope in Fig. 5(a)–(g) . It can be seen that with the change in spraying time, the 3D morphology formed on the surface and roughness has an obvious difference. When the spraying time is only 5 s, the structure height on the surface is 5 μm, as the spraying time increases to 10 s and 15 s, the coating height increases by 21 μm and 40 μm, and a stable cluster structure is formed on the surface and the height reaches 60 μm, and when the spraying time is 25 s, the coating height changes to 98 μm. The coating height changes greatly, which is due to the destruction of the structure formed during the spraying process and the aggregation of a large number of micro and nanoparticles, and the poor mechanical properties of the coating formed by the aggregation of these particles. The changes in surface morphology affect the wettability change behaviors of the surfaces as shown in Fig. 5(g) . When the spraying time exceeds 5 s, the rough structure formed on the surface can obtain a larger solid–liquid contact area, which makes the surface wetting state from Wenzel state to Cassis–Baxter state. 31 The Laplace pressure formed in the rough structure of the surface is not enough to support the droplets to completely suspend on the fluorinated SiO 2 molecular surface, resulting in the formation of Wenzel state on the surface as the spraying time is less than 5 s. The experimental results show that the rough structure formed on the surface can directly support the water droplets leading to the Cassis–Baxter state of spraying time was 20 s, and with the increase of time, the gap will be filled with excess particles. Fig. 5 3D surface images of SHP with different spraying time (a) SHP-0; (b) SHP-5; (c) SHP-10; (d) SHP-15; (e) SHP-20; (f) SHP-25; (g) the schematic of the wetting model of SHP with different spraying time. Simultaneously, the changing trend of roughness is consistent with that of line profile as shown in Fig. 6 , and with the change of spraying time, the roughness R a changes from 0.056 μm to 6.627 μm, the roughness R q changes from 0.073 μm to 7.292 μm. The results show that with the change in spraying time, the surface morphology forms peaks and troughs with different spacing, and the corresponding air supported droplets captured in the troughs can be suspended on the coating surface. Fig. 6 Surface roughness of SHP with different spraying time: (a) SHP-0; (b) SHP-5; (c) SHP-10; (d) SHP-15; (e) SHP-20; (f) SHP-25. Analysis of anti-icing performance of the coating For its surface structure of coating modification, superhydrophobic surfaces, with excellent anti-icing properties, have a bright future in anti-icing applications. The reason why superhydrophobic surfaces have good anti-icing performance is the existence of a micro–nano structure, that makes the surface in the Cassie–Baxter state and thermal conductivity, water droplets can be suspended over the superhydrophobic surfaces and easily roll off. 32 As discussed above, the static water droplet icing change of SHPs coating was tested to reflect their anti-icing performance. The process of ice formation is mainly divided into the pre-cooling stage and the nucleation and growth stage of droplets. Fig. 7 shows the real-time status of the water droplet of 5 μL on the prepared surface of SHPs with different spraying time. Compared with the pre-cooling time (before ice nucleation), the time of ice growth is significantly shorter, once nucleation occurs, the ice will move from the bottom to the upper liquid phase. Fig. 7(a1) shows bare aluminum surface and SHPs coating surface central area are transparent. With the progress of the experiment, the surface temperature decreased gradually, the central area of drops became non-transparent at first after 32 s, and the water droplets on the surface quickly freeze for about 14 s. It is worth noting that the drop on the SHP-20 conforms to Cassie–Baxter wetting model becomes non-transparent after 370 s, and the water droplets on the surface freeze for about 40 s, the precooling time of surface droplet icing accounts for about 90% of the total ice delay time. Moreover, Fig. 7(a1)–(a6) shows the state of frozen droplets on different wettability surfaces, and with the increasing of spraying time, the freezing time of SHPs increased first and then decreases which is consistent with the change trend of wettability. Fig. 7 (a) Freezing process (−15 °C) of droplets on SHPs with different spraying time: (a1) SHP-0; (a2) SHP-5; (a3) SHP-10; (a4) SHP-15; (a5) SHP-20; (a6) SHP-25; (b) icing mechanism of coating surface; (c) the heat transfer process of coating. To further explain the process of the change of the wettability of the droplet icing, Fig. 7(b) shows the mechanism of the change of the freezing process of the superhydrophobic surface. When the temperature of the experimental plate is maintained at −15 °C and the ambient humidity is 50 ± 5 wt%, the Cassie–Baxter state still exists on the SHP when the freezing behavior occurs. With the progress of time, the wettability of the surface begins to decline, due to the air trapped on the hydrophobic surface condensing, and the droplets are not suspended on the surface at all. At this time, the surface is in Wenzel state. Surprisingly, when the surface coating restores at the ambient temperature, SHP regains its superhydrophobic property. The reason is the internal water vapor in the air melted and the droplets were suspended by the surface structure. It can be seen from Fig. 8 that the wettability of the droplets decreases significantly during the precooling process and the freezing process. After drying at a high temperature of 50 °C, the water contact angle of the coating surface decreases, but the decrease is not obvious. It is further proved that the mechanism of the droplet icing process is that the water vapor in the internal air condenses to form Wenzel state, which can still be restored after melting. Fig. 8 Change of water contact angle during droplet freezing and melting at different spraying time. The coating can improve the thermal resistance of the heat transfer process and delay the icing process. It is assumed that the droplet is always spherical and the heat transfer between the droplet and air is uniform. From Fig. 7(c) , shows the heat transfer process of coating surface and smooth surface. According to the heat conduction formula: 1 Q is heat transfer during the whole icing process; Q g is heat transfer in heat conduction process; Q l is heat transfer in natural convection process. is heat transfer in thermal radiation process. Based on Fourier's law, Newton's cooling law and Stephen Boltzmann's law, the heat transfer process equation can be described as: 2 α is the silica thermal conductivity is 0.27 W cm −1 K −1 , 33 R is the droplet radius, θ is contact angle of coating surface T d is the ambient temperature, is the temperature of trapped air T s is the droplet temperature. There is no air is trapped on the smooth surface, and the solid–liquid contact area is much larger than the superhydrophobic surface. The heat transfer process on the surface can be described as follows: 3 α 1 the thermal conductivity of aluminum is 237 W cm −1 K −1 . 34 According to the literature, 35 it is proved that the heat transfer between liquid–solid interfaces is the main form of droplet heat loss. It can be inferred that the low thermal conductivity and high wettability of the droplets on the coating surface are the main factors for the delayed icing of the droplets on the surface. To discuss the influence of the droplet icing delay performance accurately, the droplet icing delay performance was also characterized at −9 °C, −12 °C, −15 °C, and −18 °C, as shown in Fig. 9(a) and (b) . With the decrease of substrate temperature, the anti-icing ability of static droplets on SHPs surface decreases rapidly, the precooling time and the ice growing time of water drops on different SHPs surfaces reducing greatly while the substrate temperature is −9 °C and −18 °C. And it can be found that at lower temperatures, the droplet icing delay time of the superhydrophobic surface of the micro–nano composite structure is still significantly improved compared with the SHP-0, and still shows a higher icing delay performance. Fig. 9 (a) The pre-cooling time of water droplets on −9 °C, −12 °C, −15 °C, and −18 °C with different spraying time; (b) the growing time on −9 °C, −12 °C, −15 °C, and −18 °C with different spraying time. Coating durability and wear resistance test Durability of the SHP-20 was evaluated by exposing it to the air for 180 days. 36 Fig. 10(b) shows the WCA and SA slightly changed by less than 7.0° ± 3.4° during air exposure for 6 months, which indicates that the SHP-20 coating has long-term stability in the air. On the other hand, to test the wear resistance of the superhydrophobic coating, 50 sandpaper friction cycle tests were carried out according to the scheme of ref. 37 and 38 . Fig. 10(a) shows the water contact angle of the coating was tested respectively, and Fig. 9(b) shows the specific steps. It can be seen from Fig. 10(c) that after 40 times sandpaper friction cycle test, the WCA can maintain above 140.9° ± 1.8° and the SA change to 21° ± 2.1°, indicating that the coating has possessed good abrasion resistance. Fig. 10 (a) Abrasion resistance test of SHP-20 specific wearing steps; (b) water contact angles and sliding angles of SHP-20 exposed to the air for 180 days; (c) results of SHP-20 after 50 times of sandpaper friction cycle test."
} | 5,487 |
40156067 | PMC11954184 | pmc | 342 | {
"abstract": "Cyanobacteria bear great biotechnological potential as photosynthetic cell factories. In particular, hydrogenases are promising with respect to light-driven H 2 production as well as H 2 -driven redox biocatalysis. Their utilization relies on effective strain design as well as a balanced synthesis and maturation of heterologous enzymes. In a previous study, the soluble O 2 -tolerant hydrogenase complex from Cupriavidus necator ( Cn SH) could be introduced into the model cyanobacterium Synechocystis sp. PCC 6803. Due to its O 2 -tolerance, it was indeed active under photoautotrophic growth conditions. However, the specific activity was rather low indicating that further engineering is required, for which we followed a two-step approach. First, we optimized the Cn SH multigene expression in Synechocystis by applying different regulatory elements. Although corresponding protein levels and specific Cn SH activity increased, the apparent rise in enzyme levels did not fully translate into activity increase. Second, the entire set of hyp genes, encoding Cn SH maturases, was co-expressed in Synechocystis to investigate, if Cn SH maturation was limiting. Indeed, the native Cn SH maturation apparatus promoted functional Cn SH synthesis, enabling a threefold higher H 2 oxidation activity compared to the parental strain. Our results suggest that a fine balance between heterologous hydrogenase and maturase expression is required to ensure high specific activity over an extended time period. Supplementary Information The online version contains supplementary material available at 10.1186/s13068-025-02634-5.",
"introduction": "Introduction H 2 is considered a key element of future cyclic economies and is of major interest within the field of renewable energy [ 55 ]. Despite its potential for a decarbonized economy, 96% of H 2 production still relies on fossil resource usage. In addition to technical issues regarding storage and H 2 conversion, the big challenge is to develop sustainable ways for H 2 production [ 23 ]. Nature comes in by showing diverse processes for microbial H 2 production [ 22 ]. Many microorganisms can produce H 2 via dark and photo-fermentation [ 59 ], and oxygenic photosynthesis, deriving electrons from water oxidation, in principle can be coupled with H 2 production [ 9 , 10 ]. Most prominent is the application of either eukaryotic microalgae or cyanobacteria for such light-driven H 2 production (photo-H 2 ). In green algae, H 2 formation relies on [FeFe] hydrogenases, which show high turnover rates of up to 104 s −1 . They are, however, produced and active only under micro- or anaerobic conditions and are rapidly disintegrated in the presence of molecular oxygen [ 51 ]. In contrast, cyanobacteria typically feature bidirectional [NiFe] hydrogenases, which are not disintegrated in the presence of O 2 , but reversibly inhibited. The unicellular model cyanobacterium Synechocystis sp. PCC 6803 (hereafter Synechocystis ) harbors a pentameric enzyme ( Syn SH) composed of a hydrogenase module (HoxYH) and a diaphorase module (HoxEFU) [ 63 ]. Syn SH is associated to the thylakoid membrane by means of the HoxE subunit [ 7 ], which facilitates electron transfer from the photosynthetic electron transport chain to the diaphorase module HoxEFU via reduced flavodoxins and ferredoxins [ 20 ]. Syn SH is considered to work as an electron valve to compensate for transiently missing electron sinks such as the Calvin-Bassham-Benson (CBB) cycle and O 2 upon sudden switches from dark to light [ 1 , 44 ]. Recently, an involvement of Syn SH in electron balancing under oxic conditions has been proposed, indicating a multi-functional role of this enzyme in cyanobacteria [ 6 ]. The main limiting factors for applying cyanobacterial hydrogenases for photo-H 2 production are its O 2 -sensitivity, H 2 re-oxidation when C- and N-assimilatory pathways as native electron acceptors become active, and the competition with these for photosynthetically derived electrons [ 1 ]. During the past two decades, progress has been made to overcome these challenges [ 34 ]. Recently, photosynthetic electron flow towards H 2 formation instead of nitrate, CO 2 , and/or O 2 reduction has been targeted via metabolic engineering [ 2 , 11 , 21 , 31 , 45 ], and a direct coupling of HoxYH of Synechocystis to photosystem I (PSI) has been established [ 1 , 32 ]. The latter approach resulted in reduced competition with the downstream metabolic pathways and avoided H 2 uptake activity. Photo-H 2 production has also been facilitated via enzymatic O 2 -removal, though O 2 -sensitivity and electron transfer efficiency remain challenges to be addressed [ 51 ]. One possible approach to circumvent the O 2 problem is the utilization of natural O 2 -tolerant hydrogenases. In this respect, heterologous expression of functional [NiFe] hydrogenases tolerating up to 1–3% of O 2 has been achieved in Synechococcus elongatus [ 67 ]. Moreover, the “Knallgas” bacterium Cupriavidus necator (hereafter C. necator ) features a soluble [NiFe] hydrogenase ( Cn SH) even retaining full activity at 20% O 2 [ 26 , 37 ]. Cn SH has successfully been introduced in heterotrophic hosts [ 18 , 36 , 52 , 57 ] and, recently, also in Synechocystis [ 42 ]. In Synechocystis , Cn SH was continuously active during oxygenic photosynthesis, oxidizing H 2 independently of the O 2 concentration. The results revealed a tight interconnection of Cn SH with cyanobacterial metabolism. The engineered strain Syn_Cn SH + was able to use H 2 -derived electrons to fix CO 2 and fuel growth even in the absence of water oxidation activity [ 42 ]. However, due to the strict dependency of Cn SH on NADH as electron donor, H 2 formation was achieved only in the presence of glucose effecting an elevated cytosolic NADH/NAD + ratio. For application of Cn SH for photo-H 2 production, it is, therefore, required to either provide electrons in the form of NADH, change its electron donor specificity, or couple the hydrogenase module directly to PSI. Moreover, the specific Cn SH activity exhibited by Syn_Cn SH + was up to two orders of magnitude lower than that in its native host [ 42 , 52 , 53 ]. This can be explained by lower enzyme abundance and/or inefficient hydrogenase maturation. Interestingly, a parallel introduction of the C. necator maturation apparatus was not in all cases required to establish a functional Cn SH in heterotrophic hosts. Key differences among these studies include the genetic background of the host strains, growth conditions applied, and expression systems used for the multi-gene operon. To achieve Cn SH activity in Synechocystis , the expression of auxiliary genes for Cn SH maturation was not required, except for hox W encoding a HoxH-specific endopeptidase [ 42 ]. Obviously, hydrogenase maturation factors of Synechocystis, encoded by the 6 accessory genes hyp A1, hyp B1, hyp C, hyp D, hyp E, and hyp F [ 25 ], to some extent also enable functional Cn SH assembly in Synechocystis under aerobic conditions. However, the specific Cn SH activity in cell-free extract of Syn_Cn SH + was roughly 200 times lower than those reported for C. necator or recombinant E. coli or P. putida with Cn SH genes and corresponding hyp genes co-expressed. In this study, we characterized limitations for Cn SH specific activities in Synechocystis [ 42 ] and increased the specific activity three-fold by fine-tuning Cn SH multigene expression and co-expression C. necator hyp genes.",
"discussion": "Discussion In our previous work, we replaced the native hydrogenase of Synechocystis with the soluble, O 2 -tolerant [NiFe] hydrogenase from C. necator ( Cn SH) and showed Cn SH activity in vivo and in vitro. Thereby, Cn SH was shown to be active in the presence of O 2 and during photosynthetic water oxidation. The specific enzyme activity determined in cell-free extracts of Syn_Cn SH + , however, was up to two orders of magnitude lower than rates achieved in the native and heterotrophic hosts (Table 1 ) [ 41 , 52 , 53 ].\n Table 1 Specific H 2 -oxidation activities of Cn SH in soluble fractions of different host strains Strain Condition U g Prot −1 References C. necator H16 CFE a 800–8000 [ 53 ] [ 52 ] E. coli CFE 1200 [ 52 ] P. putida PC b 150 [ 41 ] Syn_Cn SH + CFE 18 40 c [ 42 ] This work Syn_P nrsB Cn SHg CFE 80 This work + p P rhaBAD Cn Hyp CFE 125 This work a CFE: cell-free extract b Permeabilized cells c Optimized experimental procedure based on freshly grown cells instead of frozen cells pellets In this context, it is important to note that strong and stable recombinant gene expression in cyanobacteria remains a challenge [ 29 ]. Previous studies indicated that expression levels are typically limited by slow transcription and translation, which are fundamentally controlled by the promoter and RBS elements, respectively [ 14 ]. Transcriptional control of the initially introduced C. necator hox operon relied on the native light-regulated psb A2 promoter and resulted in low expression levels. Recombinant gene expression using native psb A2 has been reported to suffer from strong dependence on light availability and growth status [ 19 , 68 ]. We thus aimed for a well-controllable promoter with a wide dynamic range of induction. Among the native metal-inducible promoters, Ni 2+ -dependent P nrsB has recently been characterized in Synechocystis as titratable and tight [ 13 ]. Together with P nrsB , the strong synthetic RBS* and a native terminator (T psbC ) were used for hoxFUYHWI expression [ 24 , 39 , 64 ]. To improve cloning efficiencies and enable fast screening of genetic elements, we established a modular GoldenGate-type cloning system similar to the recently reported CyanoGate [ 62 ]. Both plasmid- and genome-based expression of the designed operon resulted in enhanced Cn SH synthesis and a doubling of the specific H 2 -oxidation activity (from 40 to 80 U g Prot −1 ). Gene expression with RSF1010-based plasmids [ 60 ] is known to be superior to genome-based expression, enabling higher gene copy numbers per cell (~ 30 plasmids vs 2–20 chromosome copies in Synechocystis ) [ 65 ] and increased transcription during stationary phases [ 28 ], but suffers from a lower stability [ 30 ]. Indeed, higher Cn SH levels were achieved by plasmid-based expression, but maximally reached Cn SH activities were similar and differed regarding the optimal Ni 2+ concentration necessary to achieve them. Further, only genome-based expression allowed stable hydrogenase production over 3 days post-induction, qualifying genome-based Cn SH expression in Synechocystis as superior. The non-correspondence of Cn SH expression levels and activities (Fig. 3 ) indicates the presence of non-functional hydrogenases, which may be due to the absence of C. necator Hyp proteins. Indeed, it is well known that recombinant production of [NiFe] hydrogenase in heterologous hosts is challenging due to the complex maturation process [ 56 ]. Heterologous hydrogenase expression studies supported the hypothesis that the probability of obtaining a functional enzyme correlates with the abundance of homologous and heterologous Hyp proteins sharing a high degree of similarity [ 16 ]. Although Synechocystis and C. necator maturases show only 50–67% amino acid sequence homology [ 42 ], Cn SH maturation obviously is realized by the maturases of Synechocystis , but may suffer from a low efficiency. In studies conducted with E. coli as host strain, with an amino acid sequence identity between C. necator and E. coli hyp maturases of 18–45%, the omission of C. necator maturases severely reduced the recombinant hydrogenase activity [ 15 , 52 ]. We firstly hypothesized that the absence of a hypX gene homolog in Synechocystis could have compromised aerobic Cn SH maturation. However, the introduction of hyp X only into Syn _ PnrsBCn SHg did not influence the achieved hydrogenase activity, ruling out CO biogenesis as main limiting factor (Fig. S8). Therefore, it can be assumed that CO allocation is sufficient in Synechocystis . The introduction of the complete maturation system of C. necator into Synechocystis improved Cn SH expression level and activity. It is, however, important to note that a fine balance between hydrogenase and maturase gene expression seems essential to maximize functional Cn SH production. Elevated Hyp protein production or maturase activities appeared to negatively affect cell growth and hydrogenase activity so that their expression had to be quantitatively controlled. Thus, new genetic tools are needed to enhance and control heterologous gene expression. With the fine balance of recombinant multigene expression, the presence of Cn SH-dedicated auxiliary proteins enhanced its maturation efficiency in Synechocystis and, consequently, Cn SH activity. Further investigations, e.g., the separate expression of functional Cn Hyp complexes (HypCD, HypEF, HypAB) in Synechocystis and knockouts of endogenous maturases could be useful to determine the most efficient combination of maturases and to optimize heterologous Cn SH production in Synechocystis . As plasmid-based expression appeared to enable higher Cn SH levels in Synechocystis , genomic integration of hyp genes or the use of compatible plasmids [ 48 ] for hox and hyp gene expression may also be promising. In the present study, we increased Cn SH activity in Synechocystis 3.1-fold (from 40 to 125 U g Prot −1 ) in a two-step approach—firstly, by improving protein synthesis via expression system engineering and secondly by the introduction of the C. necator hyp operon. The optimal expression conditions for the C. necator hox and hyp operons resulted in 60% higher H 2 oxidation activity and enhanced in vivo stability compared to the expression of hox genes alone. This finding supports the hypothesis that the co-expression of C. necator maturases plays a crucial role in the formation of a functional recombinant hydrogenase complex. The rates obtained (125 U g Prot −1 ) are comparable to those achieved in heterotrophic hosts such as P. putida (~ 160 U g Prot −1 ) [ 41 ] (Table 1 ), paving the way for diverse applications of the O 2 -tolerant hydrogenase of C. necator in Synechocystis , e.g., photo-H 2 production or H 2 utilization to boost growth and/or biotransformation reactions."
} | 3,631 |
25912797 | null | s2 | 345 | {
"abstract": "The increased development of green low-carbon energy technologies that require platinum group metals (PGMs) and rare earth elements (REEs), together with the geopolitical challenges to sourcing these metals, has spawned major governmental and industrial efforts to rectify current supply insecurities. As a result of the increasing critical importance of PGMs and REEs, environmentally sustainable approaches to recover these metals from primary ores and secondary streams are needed. In this review, we define the sources and waste streams from which PGMs and REEs can potentially be sustainably recovered using microorganisms, and discuss the metal-microbe interactions most likely to form the basis of different environmentally friendly recovery processes. Finally, we highlight the research needed to address challenges to applying the necessary microbiology for metal recovery given the physical and chemical complexities of specific streams."
} | 236 |
39890833 | PMC11785774 | pmc | 346 | {
"abstract": "Artificial intelligences are indispensable social infrastructures, neural networks are embodiment methodologies, and neuromorphic systems are promising solutions for compact size and low energy. Memristors were first prepared for the synapse devices but incur energy consumption, and memcapacitors were next prepared but have small dynamic ranges of capacitance. In this research, we have developed a neuromorphic system using capacitor synapses. Here, multiple capacitors have binary-weighted capacitances and are controlled to be connected to intermediate signals. They are discharged through transistors, and when they fall below the threshold voltage, the output signals are inverted. After all, electric charges in the multiple capacitances are summed and measured by the inverting intervals, which is the same as multiply–accumulate operation. A large-scale integration chip is actually fabricated. The working is confirmed by MNIST, and the circuit-aware rounding improves the accuracy to 96%, indicating a sufficient possibility for practical applications, and the energy efficiency is 163 GOPS/W even by the 180 nm technology, indicating a great potential for low energy consumption.",
"conclusion": "Conclusion We have developed a neuromorphic system using capacitor synapses. In this neuromorphic system, multiple capacitors have binary weighted capacitance values of C 0 , C 1 = 2 × C 0 , C 2 = 4 × C 0 , and C 3 = 8 × C 0 . They are controlled by weight signals, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${w}_{i,j}={\\left(-1\\right)}^{s}\\left(\\sum_{k}{2}^{k}{w}_{k}\\right)$$\\end{document} , and an input signal, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${x}_{i}$$\\end{document} , to be connected to intermediate signals, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${y}_{j}^{+}$$\\end{document} or \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${y}_{j}^{-}$$\\end{document} , for the positive and negative signals, respectively. The connected capacitors, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${y}_{j}^{+}$$\\end{document} , and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${y}_{j}^{-}$$\\end{document} are preliminarily charged to V dd , and each signal is discharged through discharging transistors, Td, to GND. When \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${y}_{j}^{+}$$\\end{document} and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${y}_{j}^{-}$$\\end{document} gradually fall below the threshold voltage, the logic buffers invert the output signals, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${z}_{j}^{+}$$\\end{document} and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${z}_{j}^{-}$$\\end{document} . After all, electric charges charged in the multiple capacitances in one synapse element and all synapse elements in one row are summed for each positive and negative sign, namely, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${q}_{j}^{+}=\\left(\\sum_{i}\\left|{w}_{i,j}\\right|{x}_{i}\\right){C}_{0}{V}_{dd} \\text{ for } s=+$$\\end{document} and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${q}_{j}^{-}=\\left(\\sum_{i}\\left|{w}_{i,j}\\right|{x}_{i}\\right){C}_{0}{V}_{dd} \\text{ for } s=-$$\\end{document} . They are measured by the inverting intervals, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${z}_{j}^{+}$$\\end{document} and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${z}_{j}^{-}$$\\end{document} , namely, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${z}_{j}^{+}=\\left(\\sum_{i}\\left|{w}_{i,j}\\right|{x}_{i}\\right) \\text{ for } s=+$$\\end{document} and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${z}_{j}^{-}=\\left(\\sum_{i}\\left|{w}_{i,j}\\right|{x}_{i}\\right) \\text{ for } s=-$$\\end{document} , and from \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${z}_{j}={z}_{j}^{+}-{z}_{j}^{-}$$\\end{document} , \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${z}_{j}=\\sum_{i}{w}_{i,j}{x}_{i}$$\\end{document} is obtained, which is exactly the same as MAC operation used in neural networks. An LSI chip of the neuromorphic system is physically designed and actually fabricated. The working is confirmed by MNIST recognition, and the circuit-aware rounding improves the accuracy to 96%, which indicates a sufficient possibility of this neuromorphic system for practical applications. Moreover, the energy efficiency is 163 GOPS/W even by the Si CMOS 180 nm technology, which indicates a great possibility for low energy consumption.",
"introduction": "Introduction Artificial intelligences are indispensable social infrastructures for smart information worlds 1 , 2 , and neural networks are the most general embodiment methodologies with quite skillful biomimicries 3 , 4 . However, conventional frameworks of neural networks are large and complicated software that runs on high-spec and energy-consuming hardware such as Neumann-type supercomputers, which are not customized for neural networks 5 , 6 . Neuromorphic systems are promising customized solutions for compact system size and low energy consumption by device and hardware-level biomimetics 7 , 8 , which consist of processing elements of neuron elements and synapse elements 9 , 10 . Memristors 11 – 13 , variable conductance devices, were first prepared for the synapse devices 14 – 23 , but they are a kind of resistors and hence in principle incur energy consumption as Joule heating. Therefore, memcapacitors 24 , 25 , variable capacitance devices, were next prepared 26 – 28 , because memcapacitors are a kind of capacitors and hence in principle incur no energy consumption, but the dynamic ranges of the variable capacitance are not so large, and the operation voltages cannot be so large in order not to overwrite the memorized capacitances. Incidentally, although memcapacitors can be emulated by transistors 29 , 30 and circuits 31 , 32 , some complicated structures and driving are needed. Even in recent years, many researchers are continuously publishing neuromorphic systems using memristors 33 – 36 and memcapacitors 29 , 30 , 37 – 39 . However, in most cases, only the synapse elements are actually fabricated and the neuromorphic systems are just virtually simulated, the synapse elements are not integrated in the neuromorphic systems, only brief reports are released and detailed information are unknown, and so on. In this research, we have developed a neuromorphic system using capacitor synapses. In this neuromorphic system, multiple capacitors have binary weighted capacitance values, and they are controlled to be connected to intermediate signals. The connected capacitors and intermediate signals are charged, and each signal is discharged through transistors. When they fall below the threshold voltage, the output signals are inverted. After all, electric charges charged in the multiple capacitances in one synapse element and all synapse elements in one row are summed, and they are measured by the inverting intervals, which is exactly the same as the multiply–accumulate operation used in neural networks. A large-scale integration chip of the neuromorphic system is physically designed and actually fabricated. The working is confirmed by MNIST recognition, and the circuit-aware rounding improves the accuracy to 96%, which indicates a sufficient possibility of this neuromorphic system for practical applications. Moreover, the energy efficiency is 106 GOPS/W even by the Si CMOS 180 nm technology, which indicates a great possibility for low energy consumption. Neuromorphic system using capacitor synapses The neuromorphic system using capacitor synapses is shown in Fig. 1 . Synapse elements are aligned in a matrix array. Here, a synapse element at column i and row j is shown in detail. Multiple capacitors have binary-weighted capacitance values. Here, four capacitors have capacitance values of C 0 , C 1 = 2 × C 0 , C 2 = 4 × C 0 , and C 3 = 8 × C 0 . They are controlled by weight signals, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$w_{i,j} = \\left( { - 1} \\right)^{s} \\left( {\\mathop \\sum \\limits_{k} 2^{k} w_{k} } \\right)$$\\end{document} , to be connected through the designated wiring to an intermediate signal, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$y_{j}^{ + }$$\\end{document} or \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$y_{j}^{ - }$$\\end{document} . Here, the weight signals are five bits of one sign bit for either positive sign or negative sign, s , and four bits for a binary number, w 0 , w 1 , w 2 , and w 3 . The \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$y_{j}^{ + }$$\\end{document} is for the positive signal to be connected, and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$y_{j}^{ - }$$\\end{document} is for the negative signal to be connected. The connected capacitors, including those in other synapse elements, and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$y_{j}^{ + }$$\\end{document} and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$y_{j}^{ - }$$\\end{document} , are preliminarily charged to V dd and kept connected by an input signal, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$x_{i}$$\\end{document} , and each signal is discharged through discharging transistors, Td, to GND. Here, Td is composed of four serially connected n-type transistors to regulate the discharge current. When \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$y_{j}^{ + }$$\\end{document} and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$y_{j}^{ - }$$\\end{document} gradually fall below the threshold voltage, the logic buffers invert the output signals, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{j}^{ + }$$\\end{document} and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{j}^{ - }$$\\end{document} . The detailed architecture of the neuromorphic system is explained in the “ Methods ” chapter. After all, electric charges charged in the multiple capacitances selected by the weight signals and input signals in one synapse element are summed for each positive and negative sign, namely, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$q_{i,j} = \\left( {\\mathop \\sum \\limits_{k} 2^{k} w_{k} } \\right)C_{0} V_{dd} x_{i} = \\left| {w_{i,j} } \\right|x_{i} C_{0} V_{dd}$$\\end{document} , and those in all synapse elements in one row are summed in \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$y_{j}^{ + }$$\\end{document} and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$y_{j}^{ - }$$\\end{document} for each positive and negative sign, namely, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$q_{j}^{ + } = \\left( {\\mathop \\sum \\limits_{i} \\left| {w_{i,j} } \\right|x_{i} } \\right)C_{0} V_{dd} {\\text{ for }} s = +$$\\end{document} and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$q_{j}^{ - } = \\left( {\\mathop \\sum \\limits_{i} \\left| {w_{i,j} } \\right|x_{i} } \\right)C_{0} V_{dd} {\\text{ for }} s = -$$\\end{document} . They are measured by the inverting intervals of \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{j}^{ + }$$\\end{document} and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{j}^{ - }$$\\end{document} , namely, by re-defining \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{j}^{ + }$$\\end{document} and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{j}^{ - }$$\\end{document} as their own inverting intervals, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{j}^{ + } = \\left( {\\mathop \\sum \\limits_{i} \\left| {w_{i,j} } \\right|x_{i} } \\right) {\\text{ for }} s = +$$\\end{document} and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{j}^{ - } = \\left( {\\mathop \\sum \\limits_{i} \\left| {w_{i,j} } \\right|x_{i} } \\right) {\\text{ for }} s = -$$\\end{document} , where the proportional coefficient is presumed to be 1, and from \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{j} = z_{j}^{ + } - z_{j}^{ - }$$\\end{document} , the following equation is obtained with correct handling of the positive and negative signs, which is exactly the same as multiply–accumulate (MAC) operation used as a common step in neural networks 40 and neuromorphic systems 41 . 1 \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{j} = \\mathop \\sum \\limits_{i} w_{i,j} x_{i}$$\\end{document} Fig. 1 Neuromorphic system using capacitor synapses. The operation procedure of the capacitor synapse is shown in Fig. 2 . First, during the weight memorizing period, four capacitors, C 0 , C 1 , C 2 , and C 3 , are controlled by weight signals, s , w 0 , w 1 , w 2 , and w 3 , to be connected through the designated wiring to an intermediate signal, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$y_{j}^{ + }$$\\end{document} or \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$y_{j}^{ - }$$\\end{document} . Here, as an example, C 0 and C 3 are connected by w 0 and w 3 to \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$y_{j}^{ + }$$\\end{document} by s . Subsequently, during the preliminary charging period, the connected capacitors, C 0 and C 3 , and the intermediate signal, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$y_{j}^{ + }$$\\end{document} , are charged to V dd by a charging signal, c , where the unconnected capacitors are not charged, which avoids unnecessary energy consumption. Next, during the input writing period, an input signal, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$x_{i}$$\\end{document} , is inputted, and only the selected capacitors, C 0 and C 3 , are kept connected. Finally, during the output reading period, the intermediate signal, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$y_{j}^{ + }$$\\end{document} , is discharged through the discharging transistors, Td, switched on by the discharging signal, d , to GND. When the intermediate signal, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$y_{j}^{ + }$$\\end{document} , falls below the threshold voltage, the logic buffer inverts the output signal, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{j}^{ + }$$\\end{document} . Both \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{j}^{ + }$$\\end{document} and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{j}^{ - }$$\\end{document} are outputted for positive signs and negative signs, respectively, and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{j} = z_{j}^{ + } - z_{j}^{ - }$$\\end{document} is also obtained. Fig. 2 Operation procedure of the capacitor synapse. The working confirmation of the output reading is shown in Fig. 3 . Here, circuit simulation is outcarried with a transistor model 42 and circuit simulator HSPICE 43 . Voltage waveforms of the intermediate signal, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$y_{j}^{ + }$$\\end{document} , and the output signal, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{j}^{ + }$$\\end{document} , with variation of the ratio of the selected capacitors are shown in Fig. 3 a, and the ratio of the selected capacitors vs the inverting interval, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{j}^{ + }$$\\end{document} , is shown in Fig. 3 b. It is found that the inversion interval, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{j}^{ + } ,$$\\end{document} linearly depends on the ratio of the selected capacitors, and the inversion intervals are less than 30 μs, It is confirmed from this result that the MAC operation can be correctly performed in practically possible time. Fig. 3 Working confirmation the output reading. LSI chip of the neuromorphic system A large-scale integration (LSI) chip of the neuromorphic system is shown in Fig. 4 . The LSI chip is physically designed and actually fabricated. Here, a silicon (Si) complementary metal–oxide–semiconductor (CMOS) 180 nm technology is used. The computer-aided design (CAD) layout of the synapse element is shown in Fig. 4 a. The four capacitances are formed by parallel plate capacitances of a silicon oxide (SiO 2 ) thin film between two metal electrodes, whose capacitance density is 1 fF/μm 2 , and therefore they have capacitance values and area values of C 0 = 20 fF = 20 μm 2 , C 1 = 2 × C 0 = 40 fF = 40 μm 2 , C 2 = 4 × C 0 = 80 fF = 80 μm 2 , and C 3 = 8 × C 0 = 160 fF = 160 μm 2 . Most of the transistors are CMOS field-effect transistors (FET) and have an experienced dimension of W = 1.25 μm and minimum dimension of L = 0.18 μm, except that the discharging transistors are n-type metal–oxide–semiconductor (NMOS) FETs and have deliberated dimensions of W = 0.22 μm and L = 10 μm, which prolongs the inverting intervals of the output signals. Vdd = 1.8 V, which is the standard voltage for the Si CMOS 180 nm technology. Fig. 4 LSI chip of the neuromorphic system. The microscope photograph is shown in Fig. 4 b. It is of course found that the microscope photograph of the synapse element actually fabricated is exactly the same as the CAD layout of it. The overview photograph is shown in Fig. 4 c. The LSI chip is packaged through wire bonding in a ceramic package. The peripheral control circuit boards are shown in Fig. 4 d. The LSI chips are mounted in sockets and evaluated. Working confirmation by MNIST recognition The working confirmation by MNIST recognition.is shown in Fig. 5 . The modified National Institute of Standards and Technology database (MNIST) is the most well-known database of handwritten digits 44 and is already outdated for practical purposes, but it is still very convenient to evaluate the potential feasibility of emerging systems. The network architecture is shown in Fig. 5 a. First, the pre-process is conducted by a convolutional neural network (CNN) 45 , which is not done by the neuromorphic system but by an external program 46 – 48 . From the MNIST, 28 × 28 images are processed by a 3 × 3 kernel for edge detection, and they are reshaped to be 26 × 26 images. Moreover, they are processed by nine 3 × 3 kernels for direction detection, namely, kernels for the direction of 0, (1/8)π, (1/4)π, (3/8)π, (1/2)π, (5/8)π, (3/4)π, (7/8)π, and π, and they are reshaped and added with the original images to be nine 24 × 24 images. Furthermore, they are processed by a 2 × 2 kernel for max pooling, and they are reshaped to be nine 24 × 24 images. Next, the main process is conducted by a fully-connected neural network (FC) 49 , which is realized by the aforementioned LSI chip of the neuromorphic system using the capacitor synapses. \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$x_{0}$$\\end{document} – \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$x_{i}$$\\end{document} – \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$x_{143}$$\\end{document} and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{0}$$\\end{document} – \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{j}$$\\end{document} – \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{9}$$\\end{document} of the detailed architecture in Fig. 1 correspond to \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$x_{0}$$\\end{document} – \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$x_{143}$$\\end{document} and \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{0}$$\\end{document} – \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{9}$$\\end{document} of the FC in Fig. 5 a. The 24 × 24 images are inputted as input signals, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$x_{0}$$\\end{document} – \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$x_{143}$$\\end{document} , and inverting intervals are outputted as output signals, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{0}$$\\end{document} – \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$z_{9}$$\\end{document} , which corresponds to the labels of the handwritten digits of 0–9. Fig. 5 Working confirmation by MNIST recognition. Training of the FC is executed to determine the weight signals, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$w_{i,j}$$\\end{document} , as follows. First, a back-propagation method is as usual employed to theoretically determine synaptic weights, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$v_{i,j}$$\\end{document} , by an external program of Python simulation 50 . This is a merely normal method like the one described in textbooks 51 . Next, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$v_{i,j}$$\\end{document} is converted to \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$w_{i,j}$$\\end{document} by the following three fashions. This is because \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$v_{i,j}$$\\end{document} is mathematically a real number, whereas \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$w_{i,j}$$\\end{document} is only four bits except for the sign bit and can express only 0–15, and therefore suitable conversion is required. The weight conversion is shown in Fig. 5 b, and the conversion functions are shown in the left graph of Fig. 5 b. “Simple rounding” is a simple conversion by the following equation. \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$w_{i,j}$$\\end{document} is proportional to \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$v_{i,j}$$\\end{document} by the proportionality factor \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\alpha$$\\end{document} , discretized to an integral number n in 0– \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\beta$$\\end{document} , and limited within the range ± \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\beta$$\\end{document} , where \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\beta = 15$$\\end{document} here. 2 \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$w_{i,j} = {\\text{sign}}\\left( {v_{i,j} } \\right) \\times \\left\\{ {\\begin{array}{*{20}l} n \\hfill & {{\\text{if}}} \\hfill & {n - 1 < \\left| {\\alpha v_{i,j} } \\right| \\le n} \\hfill \\\\ \\beta \\hfill & {{\\text{if}}} \\hfill & {\\beta < \\left| {\\alpha v_{i,j} } \\right|} \\hfill \\\\ \\end{array} } \\right.\\quad 0 \\le {\\text{integer}}\\;n \\le \\beta$$\\end{document} In addition, “Stochastic rounding” randomly rounds up or down while retaining the tendency of rounding off to the nearest integral number. “Circuit-aware rounding” considers parasitic capacitances of the switching transistors for w 0 , w 1 , w 2 , and w 3 by the following equation. It is assumed that \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$C_{p} = \\gamma \\delta_{n} C_{0}$$\\end{document} where \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\delta_{n}$$\\end{document} is the number of the switching transistors in the off state when \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$w_{i,j} = n$$\\end{document} . 3 \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$w_{i,j} = {\\text{sign}}\\left( {v_{i,j} } \\right) \\times \\left\\{ {\\begin{array}{*{20}l} n \\hfill & {{\\text{if}}} \\hfill & {\\left( {n - 1} \\right) + \\gamma \\delta_{n - 1} < \\left| {\\alpha v_{i,j} } \\right| \\le n + \\gamma \\delta_{n} } \\hfill \\\\ \\beta \\hfill & {{\\text{if}}} \\hfill & {\\beta + \\gamma \\delta_{\\beta } < \\left| {\\alpha v_{i,j} } \\right|} \\hfill \\\\ \\end{array} } \\right.\\quad 0 \\le {\\text{integer}}\\;n \\le \\beta$$\\end{document} The histograms of \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$w_{i,j}$$\\end{document} are shown in the right graph of Fig. 5 b. Incidentally, the average of \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$w_{i,j}$$\\end{document} , \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\overline{{w_{i,j} }}$$\\end{document} , is 3.41, namely, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$\\overline{{w_{i,j} }} /\\beta = 3.41/15 = 22.7\\%$$\\end{document} , for the circuit-aware rounding, which is used later as a pragmatical ratio for the analysis of the energy efficiency. Finally, the weight signals, \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$$w_{i,j}$$\\end{document} , namely, w 0 , w 1 , w 2 , and w 3 , are uploaded to the neuromorphic system. The achieved accuracies are shown in the left table of Fig. 5 c. The champion accuracies are shown for each weight conversion with parameter optimization. First, the Python simulation achieves an accuracy of 97%, which is the theoretical highest baseline, because it is assumed that all the processing elements of neuron elements and synapse elements work perfectly. Next, the neuromorphic chip using the simple rounding as the weight conversion achieves 94%, while the stochastic rounding improves it to 95%, and the circuit-aware rounding further improves it to 96%. It is surprising that the degradation of the accuracy from the Python simulation to the neuromorphic system is only 1%. It is indicated from this result that the neuromorphic system using the capacitor synapses has a sufficient possibility for practical applications. The energy efficiency is also shown in the left table of Fig. 5 c, and the ratio of the selected capacitors vs the energy consumption is shown in the right graph of Fig. 5 c. First, the energy consumption for one inference for the pragmatical ratio of \\documentclass[12pt]{minimal}\n\t\t\t\t\\usepackage{amsmath}\n\t\t\t\t\\usepackage{wasysym} \n\t\t\t\t\\usepackage{amsfonts} \n\t\t\t\t\\usepackage{amssymb} \n\t\t\t\t\\usepackage{amsbsy}\n\t\t\t\t\\usepackage{mathrsfs}\n\t\t\t\t\\usepackage{upgreek}\n\t\t\t\t\\setlength{\\oddsidemargin}{-69pt}\n\t\t\t\t\\begin{document}$${w}_{i,j}$$\\end{document} of 22.7% is 884pJ, which is calculated from the circuit simulation, the equivalent operation number for the MNIST recognition is 144 OP, and therefore the energy efficiency is 144 OP/884 pJ = 163 GOPS/W. It is surprising that this energy efficiency is achieved by the Si CMOS 180 nm technology. According to Dennard scaling, the energy efficiency will be 951 TOPS/W if it progresses to 10 nm technology. It is indicated from this result that the neuromorphic system using the capacitor synapses has a great possibility for low energy consumption. The performance comparison with other state-of-the-art reports on neuromorphic systems implementing memristors and memcapacitors is shown in Table 1 . First, it is found that the accuracy in this research is roughly the same as those in the other state-of-the-art reports. Next, it should be noted that this research is based only on the Si CMOS technology, and therefore miniaturization can be promoted according to Dennard scaling, which is different from the other reports. As a result, as aforementioned, the energy efficiency will be 951 TOPS/W if it progresses to 10 nm technology, which will be the highest among this table. Table 1 Performance comparison with other state-of-the-art reports on neuromorphic systems implementing memristors and memcapacitors. Memristor 52 Memristor 53 Memristor 54 Memcapacitor 28 [This research] Neuromorphic system using capacitor synapses MNIST Accuracy − 97.73% 98.64% 95% 96% Technology 28 nm 65 nm 65 nm 180 nm 180 nm ⇨ Dennard scaling⇨ 10 nm Energy efficiency 95.3 TOPS/W 3.08 TOPS/W − 578 TOPS/W 264 TOPS/W 163 GOPS/W 951 TOPS/W"
} | 11,836 |
34215883 | PMC8788776 | pmc | 347 | {
"abstract": "Abstract Carbon sources represent the most dominant cost factor in the industrial biomanufacturing of products. Thus, it has attracted much attention to seek cheap and renewable feedstocks, such as lignocellulose, crude glycerol, methanol, and carbon dioxide, for biosynthesis of value-added compounds. Co-utilization of these carbon sources by microorganisms not only can reduce the production cost but also serves as a promising approach to improve the carbon yield. However, co-utilization of mixed carbon sources usually suffers from a low utilization rate. In the past few years, the development of metabolic engineering strategies to enhance carbon source co-utilization efficiency by inactivation of carbon catabolite repression has made significant progress. In this article, we provide informative and comprehensive insights into the co-utilization of two or more carbon sources including glucose, xylose, arabinose, glycerol, and C1 compounds, and we put our focus on parallel utilization, synergetic utilization, and complementary utilization of different carbon sources. Our goal is not only to summarize strategies of co-utilization of carbon sources, but also to discuss how to improve the carbon yield and the titer of target products.",
"introduction": "Introduction Metabolic engineering of microorganisms for converting simple carbon sources into value-added products has made great progress over the past decade. To make these biological processes outcompete chemical synthesis, the research focus now is shifting into the utilization of renewable, abundant, and cheap feedstocks to produce target products at higher titer, yield, and productivity. Lignocellulose is the most abundant renewable feedstock on earth (Alio et al., 2019 ; Kumar et al., 2017 ; Rol et al., 2019 ). Employment of lignocellulosic biomass as a carbon source to produce value-added compounds does not compete with the global food supply (Heo et al., 2019 ; Nanda et al., 2015 ). Lignocellulose hydrolysates mainly consist of glucose, xylose, and arabinose (Yu et al., 2018 ). For cost-effective generation of target products using lignocellulose hydrolysates, construction of efficient microbial cell factories for utilization of one or more of these monosaccharides has gained much attention. However, glucose serves as the preferred carbon source for most microorganisms, the presence of glucose severely impairs metabolism of other carbon sources such as xylose, arabinose, and glycerol, due to carbon catabolite repression (CCR) (Fujiwara et al., 2020 ). In recent years, various strategies including adaptive evolution of host strains, deletion of phosphotransferase system (PTS), evolution of transporter proteins, overexpression of sugar transporters, and replacement of the promoters of genes encoding pentose catabolic pathways have been developed to achieve efficient carbon sources’ parallel utilization by removal of the CCR (Chiang et al., 2013 ; Gonzalez and Antoniewicz, 2017 ; Kim et al., 2015 ; Reider Apel et al., 2016 ; Wang et al., 2018 ; Young et al., 2012 ). Moreover, in some cases, the mixed sugar co-utilization efficiency can be further improved by construction of a synergetic utilization mechanism based on the unique metabolic characteristic for each carbon source. For instance, in order to enhance the carbon yield, the glycolysis pathway was blocked to achieve noncatabolic utilization of glucose as a skeleton molecule for production of polysaccharides and glycosylated compounds (Wu et al., 2017 ). However, efficient transportation of glucose into the cells requires phosphoenolpyruvate (PEP) as the cofactor (Choe et al., 2017 ; Long et al., 2017 ). Owing to the high efficiency for generating PEP, glycerol was selected as the second carbon source to support cell growth and drive glucose uptake (Tang et al., 2020 ; Wu et al., 2017 ; Wu et al., 2018 ). By designing and establishing such a SynCar, the titer and yield of glucose-derived products can be significantly increased. In addition to lignocellulose, several cheap and accessible C1 carbon sources including methanol and carbon dioxide (CO 2 ) also serve as promising substrates, moreover, methanol can offer extra reducing power in industrial biomanufacturing. However, C1 compounds are difficult to use as the sole carbon source for cell growth and product formation by most of microbes, due to their limited metabolic efficiencies in heterotrophic microorganisms (Dai et al., 2017 ; Lesmeier et al., 2015 ; Muller et al., 2015 ). As an alternative, the introduction of another sugar to support cell growth and drive methanol or CO 2 uptake serves as an attractive strategy. In recent years, the development of mixed feedstock co-utilization mechanisms for efficient production of desired chemicals has made great progress. Our intent with this review is to summarize recent advances in co-utilization of carbon sources for high-performance generation of value-added compounds, with emphasis on the applied metabolic engineering strategies for co-utilization of two or more carbon sources, including glucose, xylose, arabinose, glycerol, methanol, and CO 2 (Table 1 ). We focus on rational utilization of these carbon sources, especially parallel utilization, synergetic utilization, and complementary utilization (Fig. 1 ), for improving titer, yield, and productivity of target production. We also discuss the current challenges confronted in this research area, and intend to provide future perspectives for economical biosynthesis of target products using cheap and renewable feedstocks. Fig. 1. Schematic representation of carbon sources’ co-utilization strategies in microbes. (a) Parallel carbon utilization strategy. (b) Synergetic carbon utilization strategy. (c) Carbon-supplement co-utilization strategy. Table 1. The Strategies for Co-utilization of Different Carbon Sources in Microbes. Target compound Substrate Titer (g/L) Yield (g/g) Productivity (g/L/h) Strategies Host Reference Succinate Glucose and xylose 107.0 0.75 0.60 Deletion of gene ptsG \n Escherichia coli \n (Zhu et al., 2020 ) 4-Hydroxymandelic acid Glucose and xylose 15.8 0.19 0.26 Disruption of gene ptsG \n E. coli \n (Li et al., 2016) n-Butanol Glucose and xylose 5.2 0.42 0.17 Disruption of gene ptsG ; overexpression of gene Zmglf ; construction of a coculture system containing glucose-selective strain and xylose-selective strain \n E. coli \n (Saini et al., 2017 ) Shikimate Glucose, xylose, and arabinose 136.9 0.46 2.85 Overexpression of genes of xylAB and araBAD for pentoses transportation and metabolism \n Corynebacterium glutamicum \n (Kogure et al., 2016 ) Butyrate Glucose and xylose 42.6 0.35 0.56 Overexpression of genes of xylT, xylA , and xylB for pentoses transportation and metabolism \n Clostridium tyrobutyricum \n (Fu et al., 2017 ) Glucaric acid Glucose and arabinose 0.50 0.76 0.01 Deletion of genes pgi and zwf , separation of the production process and cell-growth process \n E. coli \n (Shiue et al., 2015 ) Glucaric acid Glucose and xylose 1.19 0.73 0.02 Deletion of genes pgi and zwf , separation of the production process and cell-growth process \n E. coli \n (Shiue et al., 2015 ) Trehalose Glucose and xylose 5.55 0.26 0.05 Deletion of genes pgi and zwf , separation of the production process and cell-growth process \n E. coli \n (Wu et al., 2018 ) \n cis, cis -Muconic acid Glucose and xylose 4.09 0.31 0.06 PTS was replaced with the galactose permease/glucokinase system, separation of the production process and cell-growth process \n E. coli \n (Fujiwara et al., 2020 ) L-Tyrosine Glucose and xylose 1.34 0.32 0.01 Replace PTS system with the galactose permease/glucokinase system, separation of the production process and cell-growth process \n E. coli \n (Fujiwara et al., 2020 ) Ethanol Glucose and xylose 48.72 0.49 4.06 Control expression of ptsG , separation of the production process and cell-growth process \n E. coli \n (Sun et al., 2018 ) 1,4-Butanediol Glucose and xylose 12.0 0.26 (on xylose) 0.40 Overexpression of genes of xylABCDX for xylose assimilation \n E. coli \n (Tai et al., 2016 ) 1,4-Butanediol Glucose and arabinose 15.6 0.22 (on arabinose) 0.22 Overexpression of genes of araCDABE for arabinose assimilation \n E. coli \n (Tai et al., 2016 ) 1,4-Butanediol Glucose and galacturonate 16.5 0.33 (on galacturonate) 0.18 Overexpression of genes of udh, garD, and ycbC for galacturonate assimilation \n E. coli \n (Tai et al., 2016 ) Mesaconate Glucose, xylose and arabinose 14.7 0.74 (on xylose and arabinose) 0.31 Overexpression of genes of araE for arabinose assimilation \n E. coli \n (Bai et al., 2016 ) Poly (lactate-co-glycolate) Glucose and xylose 12.6 wt% NA NA Deletion of gene ptsG \n E. coli \n (Choi et al., 2016 ) Docosahexaenoic acid (DHA) Glucose and glycerol 5.7 0.06 0.04 Glucose and glycerol as the initial and the feed carbon sources \n Schizochytrium limacinum \n (Patil and Gogate, 2015 ) DHA Glucose and glycerol 9.67 0.09 0.10 Glucose and glycerol as the initial and the feed carbon sources \n Thraustochytriidae \n (Ye et al., 2020 ) DHA Glucose and glycerol 32.36 NA 0.34 Glucose as the initial carbon source and glucose and glycerol as the feed carbon sources \n Aurantiochytrium limacinum \n (Li et al., 2015 ) DHA Glucose and fructose 20.1 0.11 0.18 Glucose and glycerol as the initial and the feed carbon sources \n Aurantiochytrium \n (Yu et al.) 1,3-Propanediol Glucose and glycerol 13.47 0.27 0.18 Deletion of gene ptsG , introduction of an ATP-dependent galactose permease/glucokinase system \n E. coli \n (Yang et al., 2018a ) 1,3-Propanediol Glucose and glycerol 78.1 NA 1.63 Deletion of gene crr \n Klebsiella pneumoniae \n (Lu et al., 2018 ) 1,3-Propanediol Glucose and glycerol 92 NA 1.70 Optimization feeding ratio of glucose and glycerol to shift redox household \n Lactobacillus diolivorans \n (Lindlbauer et al., 2017 ) Trehalose Glucose and glycerol 8.2 0.86 (on glucose) 0.33 Deletion of genes pgi and zwf to divert glucose into the biosynthetic pathway, adding glycerol to rescue cell growth \n E. coli \n (Wu et al., 2017 ) D- myo -inositol Glucose and glycerol 76 1.07 (on glucose) 0.79 Deletion of genes pgi and zwf to divert glucose into the biosynthetic pathway, adding glycerol to rescue cell growth \n E. coli \n (Tang et al., 2020 ) Ethanol Methanol and xylose 1.891 0.36 0.02 Introduction of the modified serine cycle, using medium supplemented with methanol and xylose \n E. coli \n (Yu and Liao, 2018 ) Ethanol Formate and xylose 1.922 0.31 0.42 Introduction of the modified serine cycle, using medium supplemented with formate and xylose \n E. coli \n (Yu and Liao, 2018 ) Glutamate Methanol and xylose 0.09 0.01 NA Introduction of Mdh and RuMP genes, using medium supplemented with methanol and xylose, adaptive laboratory evolution \n C. glutamicum \n (Tuyishime et al., 2018 ) Ethanol Methanol and xylose 4.6 NA 1.53 Introduction of Mdh and RuMP genes, using medium supplemented with methanol and xylose, adaptive laboratory evolution \n E. coli \n (Chen et al., 2018 ) 1-Butanol Methanol and xylose 2.0 0.47 0.33 Introduction of Mdh and RuMP genes, using medium supplemented with methanol and xylose, adaptive laboratory evolution \n E. coli \n (Chen et al., 2018 ) Acetone Methanol and glucose 0.755 0.01 NA Introduction of Mdh and RuMP genes, using medium supplemented with methanol and glucose, adaptive laboratory evolution \n E. coli \n (Bennett et al., 2020 ) NA = not applicable."
} | 2,873 |
20949064 | PMC2951366 | pmc | 348 | {
"abstract": "Background Coral reefs worldwide are in decline. Much of the mortality can be attributed to coral bleaching (loss of the coral's intracellular photosynthetic algal symbiont) associated with global warming. How corals will respond to increasing oceanic temperatures has been an area of extensive study and debate. Recovery after a bleaching event is dependent on regaining symbionts, but the source of repopulating symbionts is poorly understood. Possibilities include recovery from the proliferation of endogenous symbionts or recovery by uptake of exogenous stress-tolerant symbionts. Methodology/Principal Findings To test one of these possibilities, the ability of corals to acquire exogenous symbionts, bleached colonies of Porites divaricata were exposed to symbiont types not normally found within this coral and symbiont acquisition was monitored. After three weeks exposure to exogenous symbionts, these novel symbionts were detected in some of the recovering corals, providing the first experimental evidence that scleractinian corals are capable of temporarily acquiring symbionts from the water column after bleaching. However, the acquisition was transient, indicating that the new symbioses were unstable. Only those symbiont types present before bleaching were stable upon recovery, demonstrating that recovery was from the resident in situ symbiont populations. Conclusions/Significance These findings suggest that some corals do not have the ability to adjust to climate warming by acquiring and maintaining exogenous, more stress-tolerant symbionts. This has serious ramifications for the success of coral reefs and surrounding ecosystems and suggests that unless actions are taken to reverse it, climate change will lead to decreases in biodiversity and a loss of coral reefs.",
"introduction": "Introduction Modern coral reef ecosystems are based on and maintained by the symbiosis between corals (Cnidaria: Hexacorallia: Scleractinia) and photosynthetic dinoflagellate symbionts (Alveolata: Dinophycea: Symbiodinium ). Rising sea surface temperatures (SSTs) [1] threaten this ecologically important symbiosis [2] – [5] as SSTs only slightly above the annual mean can result in a loss of the algal symbionts from the host, a phenomenon termed ‘bleaching’ [6] – [7] . Loss of symbionts deprives the coral of a major source of nutrients and severe bleaching can lead to coral death with significant ramifications for the reef ecosystem. Scleractinian corals vary in their susceptibility to bleaching and this may be a reflection of the symbiont type within the coral, as symbiont taxa exhibit different tolerances to stress [6] – [11] . It has been proposed that reef corals might recover from and adapt to bleaching events by acquiring more stress-tolerant symbionts from the surrounding environment [12] . However, evidence for changes in a colony's endosymbionts is lacking. The majority of corals initially obtain symbionts from the surrounding environment at the larval or single polyp stage [13] . Although multiple symbiont types are initially acquired, selectivity exists, as not all available symbiont strains are taken up and only a subset of those strains are retained [14] , [15] . For an adult coral to survive and subsequently recover after a bleaching event, the coral must either retain symbionts that can meet its minimum physiological requirements or acquire the necessary symbionts from the environment after bleaching. Some corals may naturally contain stress-tolerant symbionts that dominate the symbiosis, in which case bleaching should be minimal, and recovery rapid [6] , [7] , [16] . In fact, some corals recover from bleaching by repopulation from background stress-tolerant in hospite symbionts remaining within the host after the bleaching-induced stress (i.e., those that are usually present at low to undetectable levels within the host prior to bleaching) [7] , [10] . On the other hand, if corals lack these stress-tolerant symbionts, then post-bleaching recovery depends on the acquisition of a more stress-tolerant symbiont from the surrounding environment. While anemones and octocorals are able to acquire Symbiodinium from exogenous sources (i.e. the environment) [17] , [18] , this ability has not been demonstrated for the cnidarians that provide the structural foundation of the coral reef ecosystem, the scleractinian corals. Using the Caribbean finger coral Porites divaricata , we show that although scleractinian corals may have the ability to acquire exogenous Symbiodinium , the new associations were unstable.",
"discussion": "Discussion This study provides the first experimental evidence that although some scleractinian corals are capable of acquiring symbionts from the water column after a bleaching episode, this acquisition may be temporary. These findings have important implications about the response of corals to climate change. Firstly, these results establish that although scleractinian corals such as Porites divaricata are able to acquire novel Symbiodinium from the surrounding environment, the acquisition is transient, with the normal Symbiodinium assemblage being reestablished over weeks and months. The subsequent loss of the novel symbiont types over time may be due to an inability of the novel symbionts to multiply in the host or to compete with resident symbionts. Other instances where symbiont shifts have occurred have shown similar transitions back to the original symbiont community [10] , [11] , [28] , [29] . This study, as others [11] , [30] , suggests that the acquisition of new symbionts does not provide a stable mechanism of acclimatizing to increasing SSTs. P. divaricata appears unlikely to rely on symbiont switching to ameliorate the effects of climate change on reefs. Corals may acquire symbionts from the environment, but these could be transient infections that are not maintained in a stable symbiosis and thus provide little hope of enhancing a coral's ability to acclimatize to predicted temperature increase associated with global warming. Secondly, Symbiodinium differ in their physiological response to stressors [31] , and members of Symbiodinium D, presumed to be a stress-tolerant clade, are reported to predominate in corals subjected to elevated temperatures and thermal bleaching [6] , [7] , [9] , [10] , [22] , [32] . This has led to the hypothesis that Symbiodinium D may lessen the effects of climate change on reefs [6] . Since P. divaricata initially acquired Symbiodinium D, our results could be interpreted as supporting the hypothesis that Symbiodinium D may aid the coral host under conditions of thermal stress. However, P. divaricata did not readily acquire Symbiodinium D and its acquisition was transient and at low levels, only detectable with D-specific qPCR. An alternative hypothesis that our results do not reject is that Clade D is an opportunistic species that takes advantage of the heat stressed symbiosis (29), but further experimentation will be needed to accept or reject this hypothesis. Finally, these results demonstrate that the corals did not acquire Symbiodinium indiscriminately. When colonies were exposed to a range of Symbiodinium , only those colonies in the B224 treatment acquired the novel symbionts as detected using cp-23S-rDNA screening, indicating that only B224 was taken up in large numbers. Symbiodinium B224 may have been acquired more readily because it is a Clade B symbiont, the clade that naturally predominated in this host coral, and B224 may be physiologically similar to the original symbionts (B170). Because we did not measure physiological parameters, we do not know if B224 provided the host with interim benefits. Yet, we observed selectivity even within Clade B as Symbiodinium B224 was not retained when corals were returned to the field, and B211was not acquired at all. The findings presented here support the hypothesis that changes in the most abundant symbionts observed in post-bleaching recovery of adult scleractinians probably result from the survival and population growth of in hospite symbionts rather than the acquisition of novel types from the environment [7] , [33] , [34] . Multiple studies have demonstrated selectivity in symbiont acquisition during early ontogeny [14] , [15] , [35] and recent evidence indicates a genetic basis for this selection as reflected in differential gene expression in the presence of non-compatible symbionts [36] , [37] . Variation in symbiont tolerances to heat stress among within-clade symbiont types (i.e. [11] ) suggests that the host may acclimatize to environmental changes by shuffling symbiont composition toward closely related symbionts (intra-cladal types) where evolutionary divergences are not as great as between those symbionts of different clades. However, the ability to associate with novel symbionts would most likely have to evolve over time scales longer than the ecological times scales over which global warming is acting. Although we have demonstrated that an adult scleractinian coral, P. divaricata , can acquire novel symbionts from the environment, it is noteworthy that these were not maintained through time. If similar interactions are encountered in other host species, it would suggest that corals will not be able to acclimatize or adapt to global warming by changing symbionts. If so, only coral species that already host heat-tolerant symbiont strains will acclimatize to the increasing temperature and other stresses predicted worldwide over the next 30–50 years, although even these may not be able to tolerate the temperature increases that are predicted [1] . Thus, these findings suggest that unless action is taken to curb global warming, the outcome of this will be a loss of coral reef biodiversity, leading to reefs that are very different from those that now exist."
} | 2,473 |
39971951 | PMC11840090 | pmc | 349 | {
"abstract": "Biogas generation from organic waste by anaerobic bioreactors as renewable energy largely depends on microbial community and species interplays involved. This microbial networking is complex and time-dependent, influencing community succession and reactor performance, but remains unexplored due to the challenges in quantifying dynamics. We employed empirical dynamic modeling to analyze daily networking from a newly established bioreactor converting sucrose to biogas. Over time, microbial interactions within the three trophic (fermentative, syntrophic, and methanogenic) groups varied substantially more than between groups. Notably, versatile syntrophic bacteria like Syntrophorhabdus exhibited stronger interaction strength (0.14 ± 0.22) to hydrogen-dependent methylotrophic Methanomassiliicoccus than strictly syntrophic bacteria associated with butyrate (0.01 ± 0.01 for Syntrophomonas ) and propionate (0.00 ± 0.01 for Syntrophobacter ). The time-varying interaction networks were closely linked to the system performance dynamics, particularly concerning hydrogen concentrations. As community succession progressed, the stability of interaction network increased through time, accompanied by increased complexity and higher interaction strength. Causal analyses revealed intricate feedback involving catabolic energetics, community structure, and microbial interactions. These feedback mechanisms played a crucial role in regulating anaerobic degradation processes, thereby offering strategies for manipulating microbial interactions to enhance bioreactor stability and efficiency.",
"introduction": "Introduction Anaerobic degradation represents a pivotal biotechnological advancement, offering a sustainable solution for converting organic waste into renewable energy, which is a critical step toward achieving an energy-neutral future 1 . In this process, a consortium of trophic microbes collaborates to degrade organic wastes and convert them into valuable biogas products (primarily methane). For example, the anaerobic degradation of sucrose, a carbohydrate, into methane involves coordination among three trophic groups. First, fermentative anaerobes degrade sucrose primarily into C 2 -C 4 volatile fatty acids (VFAs) and H 2 . Subsequently, syntrophic bacteria and methanogenic archaea catabolize these metabolites into biogas through a syntrophic interaction. Notably, the oxidation of VFAs must overcome inherent thermodynamic limitations (Fig. 1A ) 2 – 4 . These closely intertwined microbial interactions profoundly influence the energy released from organic degradation, which benefits the growth and methane production of the interactions 5 , 6 . Fig. 1 The illustration conceptualizes the dynamics of microbial interaction networks influencing anaerobic degradation processes. A The anaerobic degradation process of sucrose by three primary trophic groups involves a series of steps. First, a fermentation step is mediated by fermentative bacteria (FB, yellow circle). Subsequently, a syntrophic oxidation step for fermentation products such as butyrate (Bu) and propionate (Pr) is responsible for syntrophic bacteria (SB, green circle). Finally, a methane-producing step from the methanogenic precursors, hydrogen/carbon dioxide (H 2 /CO 2 ) and acetate (Ac), is facilitated by methanogenic archaea (MA, red circle). B As the microbial communities in the reactor undergo succession after start-up, dynamic changes in microbial interactions among various trophic groups affect system performance (substrate catabolism) and community assembly over time. Each of these components can be described conceptually using an equation expressing the microbial growth, d X/ d t , as a function of microbial growth yield ( Y ), substrate catabolism (d S/ d t ), time-varying microbial interactions of various trophic groups ( f interaction ), and the environmental factors ( f env ). IS, the interaction strength of positive (blue arrow) and negative (red arrow) edges; VFAs, volatile fatty acids; Δ G ′, Gibbs free energy change for the reaction. Assuming constant environmental factors, changes in microbial growth or abundance are determined by both substrate catabolism and microbial interactions with time. Acclimation of an inoculated consortium into a functional community is a major challenge during the startup of anaerobic reactors. Although studies have examined the relationships between reactor performance, community composition, and potential microbial interactions 7 – 9 , the altered microbial interactions and their impact on system performance during succession remain unclear. Some studies have explored interaction networks among microbial species in anaerobic reactor systems using co-occurrence or model-based methods 10 – 12 . However, many of these studies have been limited by infrequent sampling intervals and the fact that correlation does not always imply interaction 13 , 14 . Moreover, assumptions of linear and static interactions under steady-state conditions may not adequately capture the complexity of anaerobic reactor systems composed of numerous microbes engaging in dynamical, time-varying networks. Those microbes are particularly likely to adapt to changing conditions during reactor startup, resulting in a time-varying interplay between microbial networking, bioenergetics, and community assembly. The time-varying community assembly arises from differential growth rates among microbes, which are regulated by energy allocation and interactions between the microbes, and changes in microbial assembly further influence microbial interactions and metabolite composition (Fig. 1B ). Despite its profound implications, the dynamical nature of this interplay remains elusive, primarily because of challenges related to quantifying large microbial networks and their dynamics over time in highly fluctuating reactor environments 15 – 18 . According to our review of relevant literature, the selection force, such as variations in temperature 9 and ammonia 8 , drive the deterministic microbial assembly processes. However, when operation parameters are maintained steady, the governing mechanism of microbial community succession and their associated time-varying interaction network in anaerobic reactors have not been explored. Thus, an attempt to unravel the intricacies of microbial networking and bioenergetics within the anaerobic degradation process would potentially open new avenues for optimizing biotechnology for waste minimization and the renewable biofuel industry. To capture the non-static and dynamical nature of microbial interactions, we employed a novel analytical framework known as empirical dynamic modeling (EDM) 14 , specifically designed for analyzing nonlinear dynamical systems. EDM offers a unique solution to analyzing such systems in that it directly evaluates microbial time-series data rather than assuming static statistical properties. Within the EDM framework, we employed the multiview distance regularized (MDR) S-map approach to reconstruct high-dimensional, time-varying interaction networks 19 . This approach can provide insights into crucial network topological properties, such as centrality, strength distribution, and derived stability measures (see Methods). Consequently, it serves as an effective tool for unraveling the dynamical networks involved in microbial succession. Moreover, an in-depth understanding of networking dynamics offers significant potential for improving reactor startup strategies. By leveraging these insights, we can further develop technologies to optimize reactor performance through targeted manipulation of microbial interactions. The MDR S-map approach was applied to reconstruct high-resolution, time-varying networks using 110 daily time-series datasets obtained during methane biogas production from anaerobic sucrose degradation in an upflow anaerobic sludge bed bioreactor after startup. This bioreactor system has long served as a crucial platform for advancing anaerobic biotechnology and deepening the understanding of microbiology and ecology. To further elucidate the mechanisms underlying the reactor dynamics, we applied the convergent cross-mapping (CCM) method 13 , 20 to identify causal relationships among various time-series features. Through CCM, we elucidated the causal relationships among interaction network properties, bioenergetic parameters, and community structure variables. Our primary objective was to elucidate the dynamical interplay between sucrose-to-methane bioenergetics and network properties to ultimately reveal the mechanisms governing community succession from reactor startup to the achievement of stable, high-performance anaerobic degradation operation. Such insights hold the potential to enhance renewable biofuel production and contribute to its long-term sustainability.",
"discussion": "Discussion By analyzing each day of a 110-day dataset through EDM, we revealed a complex causal interplay among reactor performance, bioenergetics (or metabolite profile), community structure, and properties of reconstructed networks. This analysis revealed a more robust causality within the methanogenic reactor system ( ρ ( L max ) = 0.68 ± 0.18) compared to natural systems (e.g., the causal relationship ( ρ ( L max ) = 0.19 ± 0.16) from temperature to ecosystem function 20 ). Our study provides the first comprehensive description of the intricate feedback among these aspects within a methanogenic reactor. Whereas previous studies have focused on the influence of single deterministic factors, such as temperature 9 , ammonia 8 , sludge bulking 7 , and ion strength 48 , on community assembly or reactor performance, we focused on the inherent dynamics and complexity of anaerobic reactor systems. Interestingly, even without exogenic factors (i.e., reactor operation at fixed parameters), significant changes in community structures and reactor performance were observed, emphasizing that microbial populations autonomously drive temporal variations in community assembly. These variations are underpinned by the complex feedback mechanisms between chemical reactions (metabolite and free energy), community structure, and species interactions. We observed increased stability in the microbial community as complexity and mean IS increased during the transition (Phase II) from the initial inoculum (Phase I) to a functional community (Phase III), effectively converting sucrose to biogas. The observed increase in community stability with the mean IS is consistent with findings from studies on bacterial communities in natural beach environments, where strong positive interactions were noted to stabilize microbial communities 19 . However, this contrasts with the findings of another study that weaker interactions drive community stability in fish communities 49 . This discrepancy may be attributed to the simultaneous increase in IS and the diversity of interaction types (i.e., interaction complexity); furthermore, our research indicates that interaction complexity may play a more critical role in determining stability. This finding aligns with the theoretical hybrid community model, which posits that greater interaction complexity, characterized by higher species richness and a wider variety of interaction types with more balanced proportions, enhances community stability 50 . Also, the study involving paddy soil samples analyzed using a co-occurrence network method suggested that the complexity of the microbial interaction network has a great impact on soil function than species diversity 51 . An increase in positive interactions was observed during Phase I, which coincided with environmental filtering and a relatively unstable microbial community and reactor performance. Subsequently, a notable rise in the proportion of negative interactions was observed during the transition from Phase II to III, which may have contributed to stabilizing community assembly and reactor performance in Phase III. This time-dependent pattern can be partially explained by Wigner’s semicircle law, suggesting that negative interactions, such as competition, can enhance community stability by forming more stabilizing negative feedback loops 52 , 53 . Furthermore, as the reactor operation proceeded, positive and negative edges increased, with the average negative strength outweighing the positive (Fig. 4 B and C ). This may have contributed to a sharp increase in stability from Phase II onward until the reactor system stabilized in Phase III. Notably, the timing of the transition from positive to negative interactions coincided with the alleviation of H 2 accumulation starting from Phase II, indicating a crucial regulatory link to temporal variations in H 2 levels. Previous research on anaerobic digesters demonstrated that H 2 injection increases the number of cooperative interactions 54 . The strengthened cooperative interactions may mitigate H 2 -related stress, which represents the thermodynamic barrier of the anaerobic catabolism. The analysis of interaction identities revealed greater variability in interaction types within trophic groups (coefficient of variation: 61 ± 33 for commensalism and 76 ± 17 for amensalism) than between them (31 ± 6 for commensalism and 40 ± 8 for amensalism between trophic groups) (Fig. 4 E and F ). The interactions within each trophic group were predominantly asymmetric. As the system stabilized, the fermentation group exhibited an increase in the proportion of amensalism (−/0) and a stronger mean negative IS (Fig. 4F ; Supplementary Fig. 12 ). Because of the high functional redundancy in saccharide fermentation abilities 55 , this upward trend indicates that various fermentative populations may have experienced elevated competition for sugar 56 . Conversely, the populations within syntrophic bacteria and methanogenic archaea demonstrated elevated commensalism (+/0) levels and distinct fluctuations across different phases (Fig. 4E ). This disparity may have arisen because different syntrophic bacteria and methanogenic archaea might use different substrates, depending on their chosen catabolic pathways, whereas the fermentative bacteria solely competed for sucrose. Additionally, commensalistic (+/0) partnership may benefit syntrophic bacteria and methanogenic archaea, enabling them to thrive under changing substrate/metabolite conditions (e.g., VFA concentration and H 2 partial pressure) and the thermodynamic edge 5 , 57 . Our network analysis explored the unique species interactions and associated dynamics within the network and provided novel insights into prominent interactions within three trophic groups. The findings indicated that Clostridium sensu stricto 6 played a crucial role in the network and exhibited strong, negative strength-out activity across the three phases, forming positive edges only in Phase I. The hub centrality of Streptococcus was reduced in Phase III, whereas the authority magnitude of Propioniciclava and uncultured Anaerolineae SJA-15 was increased in Phase II (Supplementary Table 3 ). These dynamic shifts in network characteristics partly explain the transition of dominant fermentation populations from Clostridium sensu stricto 6 and Streptococcus in Phase I to Propioniciclava in Phase III, reflecting the progression of sucrose fermentation. Compared with the syntrophic interactions between versatile syntrophic bacteria and methanogenic archaea, those involving strict SBOB and SPOB were weak (Supplementary Fig. 8 ). Additionally, the strict SBOB exhibited higher interaction activity (i.e., strength in) than the strict SPOB did in Phases I and II (Supplementary Table 3 ), aligning with the observed slow development of propionate degradation capability. The rate-limiting nature of syntrophic propionate degradation in the sucrose-to-biogas conversion process may stem from its higher free energy barrier for oxidation than that for butyrate. Efficient syntrophic propionate oxidation requires the establishment of structured consortia to enhance interspecies electron transfer 58 , 59 . However, most of the biomass in the reactor remained flocculant throughout the study period. Our results suggest a strong feedback loop between reactor performance and the stability of the interaction network, with superior reactor performance being linked to a more stable microbial community (Fig. 5 ; Supplementary Fig. 10 ). This finding suggests enhancing reactor efficiency and stability may be possible by managing microbial interaction complexity 16 , 60 , 61 . In particular, selecting appropriate seeding microbial consortia – such as those with high species diversity or taxa that promote network stability – could improve overall stability when initiating a new reactor. Another approach involves bioaugmentation, where specific microbial taxa are introduced to biogas-producing systems to enhance performance. For example, targeting microbes with favorable network characteristics such as a modest contribution to network stability (trace), significant influence on other microbes (hub), and strong connectivity (degree) – could be effective. The syntrophic bacteria Syntrophus and Syntrophorhabdus , which rank in the top 25% for hub characteristics and moderate for network trace (26% ~ 50%), and the hydrogen-dependent methylotrophic Methanomassiliicoccus , with high degree values and low trace (top 25% in degree and bottom 25% in trace) (Supplementary Table 1 ), exhibit high potential for such applications. These populations, crucial to microbial networks, are rarely highlighted in bioreactors treating the non-aromatic substrates. Given their pivotal role as the rate-limited step in converting organics into methane, further exploration of these microbes in stabilizing community interactions is warranted. Additionally, several lesser-known taxa, often referred to as “microbial dark matter,” were found to play crucial roles within the interaction network (Supplementary Table 1 ). For instance, uncultured Desulfobacterota and SAR324_clade_marine_group_B rank among the last 25% in trace, and the top 25% for both hub and degree, indicating strong network influence. However, these taxa show potential benefits for reactor performance; their unculturability and lesser-known nature present challenges for practical application. Considering their robust interaction activities, further investigation of their physiological and ecological functions using culture-dependent and omics approaches is necessary. The MDR S-map operates within the EDM framework and can be applied to all non-linear systems, including UASB reactors, batch anaerobic digestion processes, and other biological reactors. However, its application comes with technical limitations. Adequate time series data with equal intervals (e.g., daily or weekly) are required, and the necessary data length depends on factors such as data quality and the degree of stochasticity. This study successfully employed a nonlinear empirical dynamic modeling approach with high-resolution time series data to systematically explore the features of time-varying interaction networks and the catabolic performance of microbial consortia within a biogas-producing system post-startup. The results suggest that the community succession of microbial consortia from reactor startup to stable performance is driven by complex feedback mechanisms involving microbial interactions, community structure, and catabolic energetics. We identified that fermentative bacteria, which exhibited significant variations in community composition and interaction properties, may contribute to an unstable network structure. Conversely, microbes, such as syntrophic bacteria ( Syntrophus and Syntrophorhabdus ) and hydrogen-dependent methylotrophic Methanomassiliicoccus , likely play key roles in stabilizing network dynamics, thereby improving system performance. These insights provide an in-depth understanding of microbial interaction dynamics, enhancing our comprehension of microbial ecology and unlocking new opportunities for managing the stability and efficiency of sustainable biogas-producing systems. However, these findings should be validated in the future since the metabarcoding analysis used to assess the microbial community may introduce biases during amplicon preparation and sequencing, potentially affecting the accuracy of microbiomic state-space reconstruction."
} | 5,116 |
28691240 | null | s2 | 350 | {
"abstract": "There is no coating technology currently available to prevent the notorious biofilm formation issue. Here, a potential solution to fully address this tough issue is reported by developing a super-antifouling coating. The use of zwitterionic hydrogel (a double-sided tape) and commercial superglue is combined and a durable and ultrarobust antifouling zwitterionic (DURA-Z) coating is created that can be easily and universally applied on common substrates. Commercial superglue mostly for binding hydrophobic materials is used to strongly immobilize the superhydrophilic DURA-Z coating through interpenetration. DURA-Z coating effectively solves several key challenges preventing the current antifouling coatings from practical use, including difficult fabrication, low efficacy, poor toughness, and durability. The fabricated DURA-Z coating retains antifouling property after 90 d of immersion in water, 50 d of buffer shearing, and 30 d of water flushing, and after repeated knife scratch and sandpaper abrasion under 570 kPa. The DURA-Z coating achieves a rarely reported long-term biofilm resistance to both Gram-positive and Gram-negative bacteria and fungi: it remains almost \"zero\" microbe adhesion after continuously challenged by more than 10"
} | 312 |
39809739 | PMC11733134 | pmc | 351 | {
"abstract": "Reservoir computing advances the intriguing idea that a nonlinear recurrent neural circuit—the reservoir—can encode spatio-temporal input signals to enable efficient ways to perform tasks like classification or regression. However, recently the idea of a monolithic reservoir network that simultaneously buffers input signals and expands them into nonlinear features has been challenged. A representation scheme in which memory buffer and expansion into higher-order polynomial features can be configured separately has been shown to significantly outperform traditional reservoir computing in prediction of multivariate time-series. Here we propose a configurable neuromorphic representation scheme that provides competitive performance on prediction, but with significantly better scaling properties than directly materializing higher-order features as in prior work. Our approach combines the use of randomized representations from traditional reservoir computing with mathematical principles for approximating polynomial kernels via such representations. While the memory buffer can be realized with standard reservoir networks, computing higher-order features requires networks of ‘Sigma-Pi’ neurons, i.e., neurons that enable both summation as well as multiplication of inputs. Finally, we provide an implementation of the memory buffer and Sigma-Pi networks on Loihi 2, an existing neuromorphic hardware platform.",
"introduction": "Introduction Reservoir computing is a paradigm for computing with recurrent neural circuits that are inspired by observations in neuroscience 1 , 2 and has yielded efficient realizations of recurrent neural networks, an architecture ubiquitous in technical applications for processing multivariate time-series. Reservoir computing uses a neural dynamical system, the so-called “reservoir,” to map a time-series into a pattern in a high-dimensional state space, which is then fed into a one-layer neural network 3 , 4 . The one-layer network can be trained in a supervised fashion to perform tasks, such as classification or regression of time-series. The reservoir is thought to serve two purposes (Fig. 1 a): First, it is a memory buffer for the input signals, often a fading memory buffer if emphasis on the recent input history is desired. For buffering it is crucial that the dynamics of the reservoir are fixed. For example, the standard strategy is to use a recurrent network with fixed random connections 4 , 5 . Second, nonlinearities in the reservoir dynamics can enable rich feature spaces 6 , including nonlinear functions of the input signals, potentially leading to separability and generalization unachievable on the original signals. Fig. 1 Overview of three different representation schemes. a Traditional reservoir networks simultaneously buffer input signals and extract nonlinear features from them. b The product representation 12 computes the memory buffer and the higher-order features explicitly using concatenation and tensor product operations, respectively. c The proposed approach can form randomized distributed representations of higher-order features using the representations of the reservoir computing network. The solid lines denote the compulsory connections while the dashed lines are the optional ones. The diagrams only show the parts of the models involved in computing the representations of the feature space. In practice, however, the ability of reservoir networks to form rich feature spaces could be limited. For example, reservoir networks with common saturating neural activation functions mainly cause memory fading, and the resulting feature space still closely resembles those of linear recurrent networks 7 – 9 . Reservoir networks containing more neurobiological details, such as spiking neurons 1 , or synaptic connections with short-term plasticity as additional dynamic variables 10 , can create richer representations. However, the richness is difficult to adjust for serving a particular computational task in the best possible way. Illustrating these limitations of traditional reservoir computing, it has been recently shown 11 , 12 that a representation scheme that computes tensor products between time-delayed states of the input signals (Fig. 1 b) can empirically outperform traditional reservoir networks on an important task of predicting dynamical systems. In light of these limitations, here we investigate principled, configurable, and efficient ways to implement reservoir computing with nonlinear features on neuromorphic hardware. We propose a bipartite approach combining two generic neural circuits (Fig. 1 c): traditional reservoir networks for forming a memory buffer, and novel Sigma-Pi networks 13 , 14 for computing nonlinear features. We theoretically characterize the two essential operations for jointly representing feature spaces – concatenation and tensor product – and show that each operation results in a different similarity structure between the constructed representations. We formulate a concrete scheme based on randomized distributed representations for multivariate time-series prediction and demonstrate these networks implemented on the neuromorphic chip Loihi 2 15 . The proposed approach, which builds on ideas from vector symbolic architectures 16 – 18 and randomized kernel approximation 19 , 20 , can form representations with approximately the same similarity structure as concatenation or tensor product, but the dimensionality of the representation remains fixed. We evaluate the novel randomized distributed representations on the prediction of chaotic dynamical systems and show that often the same quality of predictions can be achieved with representations that need fewer dimensions than in the original, explicit approach 12 .",
"discussion": "Discussion Reservoir computing is a powerful and general paradigm for computing with randomized distributed representations in a recurrent neural network. It draws on principles of neural computation, and it has proven useful for a wide range of tasks 40 . Yet despite general guarantees on function approximation 6 , traditional reservoir computing is often difficult to interpret and optimize in practice. This motivates exploring modifications of the original architecture to achieve the same performance with less resources. For example, ref. 41 used reservoirs combined with time delays and ref. 42 used structured matrices to speed up updates of the reservoir. Another promising approach is to model higher-order polynomial features in the data explicitly. This idea is explored in refs. 11 , 12 , 24 who show that extracting higher-order features from time-series can dramatically improve the performance of reservoir computing models. While powerful, the dimensionality of this explicit formation of higher-order features grows exponentially with the order of the polynomial, making scaling to high-dimensional inputs difficult. In addition, large explicit feature spaces conflict with the classical motivation of parsimony in reservoir computing, and are less amenable for deployment on neuromorphic hardware. Here, we propose an approach that computes higher-order features implicitly and preserves the performance benefits of the explicit construction with reduced resource requirements. Furthermore, we show that this approach provides a principled way of approximating polynomial kernels with compact neural circuits and provides a proof-of-principle demonstration by implementing it on the neuromorphic chip Loihi 2. Polynomial kernel machines and polynomial regression are widely known and useful tools in machine learning. The earlier results in refs. 11 , 12 and our results enrich the repertoire of reservoir computing networks, by explicitly linking reservoir networks to the polynomial kernels. The theoretical connection between Volterra series and polynomial kernel regression 43 further supports the idea of using these representations for learning dynamical systems. Our approach is principled by building on classic work from the machine learning literature on approximating kernel machines with randomized representations 19 , 20 . Standard kernel machines avoid the exponential cost of computing all higher-order features but still have costs quadratic in the number of data points 25 (“Implicit realization via polynomial kernel machine” section). By contrast, we use randomized distributed representations of polynomial kernels, which capture the same similarity structure as explicitly forming the feature map, but in a “compressed” representation that is much more parsimonious. A crucial advantage of this approach over explicitly forming the polynomial features is that in the former, the polynomial features are stored “in superposition” 18 using fewer dimensions than would be required to explicitly represent them as in the latter. The price of this compression is that the distributed representation is approximate (“Kernel approximation guarantees for randomized representations” section): the similarity kernel recovered by the distributed representations is only a noisy version of the true kernel. The magnitude of this noise as a function of dimensionality can be quantified precisely 19 , 20 using the theory of concentration of measure, and the dimensionality required to achieve a small error of approximation is modest compared to explicitly representing the features. This theoretical analysis underlies our empirical findings that the proposed approach performs more accurate prediction with smaller dimensions (Fig. 4 ). Prior theoretical work on randomized kernel approximation left open the question of how to best implement concatenation and tensor products on computing hardware. Our approach addresses this gap for neuromorphic hardware, leveraging vector symbolic architectures 16 – 18 , 30 , an algebraic, dimensionality-preserving framework for forming compositional distributed representations. The binding and superposition operations of vector symbolic architectures correspond to approximate representations of concatenation and tensor product, respectively 32 . Our approach points out that two motifs, memory buffers and higher-order polynomial features, can be composed to form feature spaces and consolidated into two neural networks (see ref. 44 for an alternative proposal within a single network). For the second motif (“Randomized distributed representations of trajectories” section), we demonstrate that the recursive vector bindings correspond to computations of higher-order features of the time points in the memory buffer of the reservoir. In addition, we propose that a recurrently connected network of Sigma-Pi neurons 13 , 45 can implement the recursive binding (“Networks of Sigma-Pi neurons for tensor product and binding” section). While Sigma-Pi neurons are an idealized model there is experimental evidence for multiplication-like nonlinearity by individual nerve cells 46 . For predicting dynamical systems (“Experiments on CPU” section), the performance of our approach is either better than or equal to that of the product representation, echo state network, and multilayer perceptron baselines. Further, it improves the product representation using fewer dimensions with matching performance (e.g., Fig. 4 b). Alternatively, higher performance can be attained with moderately increased dimensions to accommodate the features of increased order (Fig. 4 c). These results emphasize the role of dimensionality as a tuneable hyperparameter of the proposed approach. Note that this is the only additional hyperparameter introduced beyond the hyperparameters in the product representation scheme (i.e., the choice of delayed states, order of polynomial features, and the regularization parameter) 12 . As follows from the results in Fig. 4 , the dimensionality of randomized representations does not require extensive tuning. A simple heuristic is to initially use the number of features in the product representation, a conservative estimate that can often be reduced in practice. Thus, the proposed approach introduces minimal overhead to the hyperparameter search space compared to that of the product representation scheme. In practice, the expected performance increases with increased dimensionality of distributed representations. Dimensionality is a way of controlling the trade-off between the performance and resource-efficiency of the model. To generate an efficient neuromorphic realization of binding, we utilize how distributed representations can be computed in terms of networks of Sigma-Pi neurons. We further explore randomly connected Sigma-Pi networks, Supplementary Material S-III , and discuss the trade-offs for different VSA models in “Networks of Sigma-Pi neurons for tensor product and binding” section. Importantly, such compositional distributed representations can be computed by networks of recurrently connected neurons, which further benefit from neuromorphic hardware acceleration. We implement the sparse block code 32 , 36 on the Loihi 2 neuromorphic chip 15 , 39 (“Experiments on neuromorphic hardware” section), which is advantageous because of the limited number of synaptic connections required. Notably, the representations computed by the neuromorphic realization performed very close to their CPU counterparts (Fig. 5 ). It is anticipated that these findings will further impact advances in developing neuro-inspired algorithms, circuits, and applications within the neuromorphic computing community."
} | 3,378 |
33994904 | null | s2 | 352 | {
"abstract": "Living systems have not only the exemplary capability to fabricate materials ("
} | 19 |
28696037 | PMC5743829 | pmc | 353 | {
"abstract": "Summary Leveraging nature's biocomplexity for solving human problems requires better understanding of the syntrophic relationships in engineered microbiomes developed in bioreactor systems. Understanding the interactions between microbial players within the community will be key to enhancing conversion and production rates from biomass streams. Here we investigate a bioelectrochemical system employing an enriched microbial consortium for conversion of a switchgrass‐derived bio‐oil aqueous phase ( BOAP ) into hydrogen via microbial electrolysis ( MEC ). MEC s offer the potential to produce hydrogen in an integrated fashion in biorefinery platforms and as a means of energy storage through decentralized production to supply hydrogen to fuelling stations, as the world strives to move towards cleaner fuels and electricity‐mediated transportation. A unique approach combining differential substrate and redox conditions revealed efficient but rate‐limiting fermentation of the compounds within BOAP by the anode microbial community through a division of labour strategy combined with multiple levels of syntrophy. Despite the fermentation limitation, the adapted abilities of the microbial community resulted in a high hydrogen productivity of 9.35 L per L‐day. Using pure acetic acid as the substrate instead of the biomass‐derived stream resulted in a three‐fold improvement in productivity. This high rate of exoelectrogenesis signifies the potential commercial feasibility of MEC technology for integration in biorefineries.",
"conclusion": "Conclusions Renewable H 2 production from biomass‐derived streams is approaching targets for practical application utilizing bioelectrochemical systems that leverage the emergent capabilities of microbial communities. A maximum productivity of 9.35 ± 1.73 L per L‐day with BOAP was achieved with a switchgrass‐derived pyrolysate. The productivity was increased threefold to 27.6 ± 5.29 l per L‐day using pure acetic acid, demonstrating the potential capability of this system. The enriched microbial community demonstrated efficient and simultaneous conversion of a wide range of compounds through synergistic division of labour strategy and multisubstrate syntrophy demonstrated by open‐circuit stimulus‐response, effectively directing the biomass electrons to intermediates such as acetic acid at an efficiency of 68.3%. However, the rate of fermentation and production of intermediates which could serve as substrates for exoelectrogenesis limited the system productivity. This study serves to provide a foundation from which to build on for understanding biocomplexity in bioelectrochemical systems for conversion of biomass‐derived streams and towards the development of community management and engineering strategies for enabling renewable hydrogen production.",
"introduction": "Introduction Many conversion technologies that could comprise the future bio‐economy are still under development, and rapid progress is needed in order to meet the growing need for renewable and carbon‐neutral energy sources. Renewable hydrogen supply and water management are among the important issues facing sustainable development of biorefineries, due to the high hydrogen demand for deoxygenation (Jones et al ., 2013 ) and potential for water limitations in areas with intensive agriculture. Additionally, hydrogen in and of itself is being pursued as a renewable fuel source due to the significant reductions in tailpipe emissions that are possible via fuel cell technologies and can also serve as an energy storage mechanism for off‐peak power (FCTO, 2016 ). Hydrogen production from renewable sources such as biomass, however, has been lagging (Rahman et al ., 2016 ). Strategies such as dark fermentation have made progress but experience low yields and carbon losses to side products and can struggle with more complex streams, while photofermentation poses operational and design challenges (Singh et al ., 2015 ). Bioelectrochemical systems offer a novel way to solve these problems by recruiting biocatalysis and electrocatalysis for efficient conversion of complex biomass resources (Borole, 2011 , 2015 , 2016 ). Engineering model organisms to convert biomass into usable bioenergy products via synthetic biology can be challenging (Cardinale and Arkin, 2012 ; Zuroff and Curtis, 2012 ), due to the complex nature of lignocellulosic biomass and the large spectrum of compounds that result from hydrolytic or thermochemical depolymerization (Kumar et al ., 2009 ). The complexity of nature can be harvested to develop efficient conversion systems for energy production by repurposing the biology to solve specific human needs. In natural anoxic environments, microbial communities have evolved to degrade biomass and recycle the energy present in complex organic carbon through interactions of two main factions: fermentative and respiring bacteria. The fermentative organisms break down larger carbon compounds resulting in end‐products that are utilized by respiring bacteria to reduce nitrate, sulfate, iron or solid metals, storing the energy in cellular biomass or reduced inorganic end‐products (Lovley, 1993 ; Nealson and Saffarini, 1994 ). Bioelectrochemical systems provide a controlled environment where these processes continue to take place, but couple the electron transfer to a solid electrode, providing a means to harvest the energy as electrons and subsequently as hydrogen or other products. While recent studies have expanded the understanding of anode microbial communities using simple fermentable substrates or domestic wastewater (Lalaurette et al ., 2009 ; Parameswaran et al ., 2009 ; Miceli et al ., 2014 ; Mahmoud et al ., 2016 ), few studies have focused on investigating the biocomplexity of engineered BESs utilizing more complex, biomass‐derived streams. There have been several studies utilizing biomass‐derived streams such as fermentation effluent or other agro/industrial waste, but have separated the fermentation from the MEC and did not focus on developing a mixed microbial community combining the fermentation and exoelectrogenesis steps (Lalaurette et al ., 2009 ; Lu et al ., 2009 ; Wang et al ., 2011 ; Marone et al ., 2017 ). Mahmoud et al . ( 2016 ) demonstrated the limitation of fermentation in treating more recalcitrant streams like raw landfill leachate directly in the MEC, requiring Fenton oxidation to improve biodegradability to enhance performance. Additional agro‐wastes like molasses and hydrolysates such as those from straw and corn stover conversion have been investigated directly in MECs (Thygesen et al ., 2011 ; Wang et al ., 2014 ; Shen et al ., 2016 ). Of these, only Thygesen et al . tracked compound levels with time and were able to identify microbial roles for xylan degradation and propionate and acetate production, but observed low performance. Additionally, recent studies using intermediates and end‐products generated during fermentation such as carboxylic acids and alcohols have investigated their role as substrates in bioanode. Use of propionate as a substrate in MEC has revealed that it goes through a two‐step process to produce current. Hari et al . ( 2016 ) have delineated the pathways of propionate conversion in MEC and reported that it is first transformed into acetate and formate/hydrogen, followed by exoelectrogenesis to produce current. Similarly for butyrate, acetate has been reported to serve as a primary branching point for uptake by exoelectrogens (Miceli et al ., 2014 ). Lastly, Parameswaran et al . ( 2009 ) demonstrated that using ethanol as the substrate, three interacting groups including fermentative bacteria, H 2 ‐scavenging bacteria and exoelectrogenic bacteria were needed for successful conversion of the substrate into electrons. Conversion of an aqueous fraction of biomass‐derived pyrolysate to electrons was recently demonstrated in a bioanode with high efficiency and productivity for renewable hydrogen production via microbial electrolysis cell (MEC; Lewis et al ., 2015 ; Lewis and Borole, 2016 ). In order to reach levels required for commercial consideration, unravelling the biocomplexity of such a system will be key to unlocking the potential of MEC technology (Ghimire et al ., 2015 ). This can support conversion of the billion‐ton biomass to biofuels and hydrogen. Thus far, studies in the literature investigating complex streams have been lacking in biocatalyst development and community interrogation. The first step in this process is to understand the multistep conversion process and the interactions among various functional groups to enable complete degradation. In order to accomplish this, an integrated approach utilizing shifts in electrochemical and substrate conditions and time‐course metabolite tracking are needed to provide insights into the resulting interactions that develop for conversion of complex substrates. Delineating the bioelectrochemical interactions and influence of process conditions on community composition can help establish the relationship between biocomplexity and system performance including yield, efficiency and rate of production of the desired products. In this study, we report on the interaction between multiple microbial groups including fermentative and exoelectrogenic groups within a high‐performing anode community processing switchgrass‐derived bio‐oil aqueous phase (BOAP). Experiments were conducted to study the behaviour of the bioanode community under two different control regimes, one focused on changing the substrate from a complex feedstock to a substrate ideal for exoelectrogens and the other on changing the poised potential. The ideal substrate was acetic acid, which is an intermediate generated from the complex substrate BOAP, thus interrelating the two parameters. The following coupled investigations were conducted to parse the effects of the interacting parameters:\n Conversion of BOAP under poised conditions, Conversion of acetic acid under poised conditions Conversion of BOAP under open‐circuit conditions to assess fermentative conversion, while restricting exoelectrogenesis \n The underlying hypothesis we investigate is that the formation of acetic acid from the complex BOAP substrate is rate limiting. Sequential operation of MEC at poised and open‐circuit (unpoised) conditions provides insights into the rate at which the carbon from BOAP is directed to intermediates for exoelectrogenesis such as acetic acid and subsequently into current. Hydrogen productivity and current density as well as efficiencies of the anode, cathode and hydrogen recovery were determined. Lastly, microbial community characterization was conducted to gain insights into the relative changes in fermentative, methanogenic and exoelectrogenic populations during these experiments to understand biocomplexity.",
"discussion": "Results and discussion H 2 Production from BOAP versus Acetic Acid in MEC Current and hydrogen production from two different substrates, BOAP and acetic acid was investigated to understand the transformation of BOAP in a bioanode. Acetic acid was chosen as a second substrate for investigation because this is a known intermediate for exoelectrogens and a common end‐product of fermentation reactions, although not the only one. A comparison of the current production from the two substrates has potential to reveal the relative rates of fermentation versus exoelectrogenesis in the MEC. While BOAP experiments were extended for 72 h, the results from BOAP for the first 24 h are also compared as acetic acid experiments did not run beyond this time. Total hydrogen productivity from BOAP at a concentration of 0.5 (g Chemical Oxygen Demand (COD)) l −1 was 4.44 ± 0.68 L per L‐day over the first 24 h. Over the same period, the hydrogen productivity using 0.5 (g COD) l −1 acetate was 9.05 ± 0.71 L per L‐day. At this concentration, the maximum H 2 production rate and current density for BOAP were 9.35 ± 1.73 L per L‐day and 8.76 ± 1.54 A m −2 respectively (Fig. 1 ). In comparison, the maximum productivity and current density reached a higher peak for 0.1 (g COD) l −1 acetic acid. They were 1.4‐ and 1.3‐fold higher than that of BOAP, reaching 13.33 ± 0.96 L per L‐day and 11.48 ± 2.94 A m −2 respectively. The overall amount of H 2 produced over the entire run was three‐fold higher with 0.5 (g COD) l −1 BOAP compared with 0.1 (g COD) l −1 acetic acid (Additional File 1: Table S2). This is not unexpected, as the BOAP experiment was fed with five‐fold more total COD. A comparison of the experiments with 0.5 (g COD) l −1 acetic acid and 0.5 (g COD) l −1 BOAP (Fig. 1 ) shows that the maximum H 2 productivity and current density increased to 2.9‐fold and 2.8‐fold, reaching 27.6 ± 5.29 L per L‐day and 24.7 ± 3.64 A m −2 respectively. The cumulative H 2 production from acetic acid over the duration of the experiment was 1.9‐fold higher compared with BOAP. Figure 1 (A) Hydrogen productivity and current density for batch experiments with BOAP and acetic acid ( AA ) as the substrate during first 24 h. (B) Efficiency during batch experiments with BOAP and acetic acid as substrate during first 24 h. Looking at efficiencies during the first 24 h, BOAP produced lower hydrogen recovery (HRE), anode coulombic efficiency (CE) and cathode conversion efficiency (CCE). The anode CE, HRE and CCE for BOAP were 71.22 ± 15.08%, 66.90 ± 12.69% and 94.16 ± 2.12% respectively. For 0.1 and 0.5 (g COD) l −1 acetic acid, CE improved by 6.8 and 13.4% while HRE increased by 6.5 and 16.9% respectively. However, as mentioned above, BOAP conversion was slower than for acetic acid, so experimental run time was continued beyond 24–72 h. This increased overall CE by 17.5% to 88.77% ± 2.7% (Additional File 1: Fig. S2). However, HRE and CCE were reduced to 62.81 ± 9.84% and 70.96 ± 13.25% respectively. The improvement in CE for BOAP with extended run time is discussed in subsequent sections and could be result of intracellular uptake/storage during the first 24 h of BOAP conversion. The reduction in CCE and HRE for BOAP is likely the main reason for a lower total volume of H 2 compared with 0.5 (g COD) l −1 acetic acid, despite similar anode efficiencies. This outcome results from a lower and more prolonged current output from BOAP, resulting in a lower average cell voltage for the run with BOAP, which reduces the efficiency of H 2 production at the cathode. A similar observation has been reported in the literature (Gil‐Carrera et al ., 2013 ; Lewis and Borole, 2016 ). The differences in current production from the two substrates, BOAP and acetic acid at the same concentration (0.5 (g COD) l −1 ) indicate that the bioanode community was limited by fermentation of BOAP. Second, the observation that 0.5 (g COD) l −1 BOAP could produce three‐fold more H 2 compared with 0.1 (g COD) l −1 acetic acid at high anode efficiency indicates that the microbial community is capable of breaking down BOAP into intermediates, which serve as substrates for exoelectrogens. Another potential cause for lower current output with BOAP may be perceived to be inhibition by toxic furanic and phenolic compounds present in BOAP. However, our previous work in collaboration with Georgia Institute of Technology using the same microbial inoculum has shown that inhibition by these individual compounds and mixture of these compounds begins to occur only at a concentration two orders of magnitude higher than those used in this study (Zeng et al ., 2015 , 2016). Thus, it is unlikely that inhibition is playing a significant role in limiting the BOAP conversion. Further evidence is provided in subsequent sections. These results provide the first evidence that fermentative processes in conversion of biomass‐derived liquids may be the limiting step in this process. Comparing hydrogen productivity and coulombic efficiency with those in the literature, Wang et al . were able to achieve a hydrogen production rate of 2.27 L per L‐day and CE of 95% using molasses wastewater in a single chamber MEC. Lu et al . ( 2009 ) reached a hydrogen production rate of 1.41 L per L‐day with a CE of 80% with fermentation effluent, which was further improved to 87% using lower applied voltage. Additionally, Li et al ., 2014 achieved a production rate of 3.43 L per L‐day by coupling to a first step of dark fermentation to produce VFAs and reached 72% CE (Li et al ., 2014 ). So despite the fermentation limitation identified in this study, the maximum productivities and efficiencies reached of 9.3 L per L‐day and 88.7% using the more recalcitrant BOAP stream compared with fermentation effluents and molasses wastewater. This demonstrates that fermentation step need not be separated from the exoelectrogenic step and that higher performance and efficiency can be achieved in a single MEC using a specifically enriched biocatalyst. BOAP intermediates generated during open‐circuit stimulus In order to determine and quantify the intermediates generated during fermentation of BOAP and to further test the hypothesis of fermentative limitations, another experiment with 0.5 (g COD) l −1 BOAP was carried out utilizing an open‐circuit stimulus‐response. This condition allows the system to reach open‐circuit voltage, preventing the carbon felt from acting as an electron acceptor, which halts exoelectrogenesis while enabling fermentation to proceed. During the interruption from 0 to 4 h, acetic acid accumulation was observed at a steady rate of 8.63 ± 0.13 mg h −1 (Fig. 2 ). The rate of acetic acid production may be slightly underestimated as part of it may be simultaneously consumed by exoelectrogens which have the ability to store charge (Freguia et al ., 2007 ). To assess the efficiency of acetic acid production from the compounds identified by HPLC, an electron equivalence analysis was conducted as described in the Experimental section. Approximately 43.20% of the electron equivalents present in the substrate were converted to acetic acid during the first two hours of open‐circuit stimulus, which increased to 68.3% by the end of 4 h. These results demonstrate that acetic acid is the major collective fermentation end‐product from the community during the conversion of BOAP. The remaining electrons not recovered at the end of 4 h in the aqueous effluent were likely taken up by the cells to form cellular biomass or stored internally as polyhydroxyalkanoates or intracellular metabolites. The electrochemical data collected after 4 h were evidence for the latter as the coulombic efficiency obtained after poising the electrode was > 100%. Analysis of the aqueous effluent by HPLC showed that although additional fermentation by‐products were present, they were generated during closed‐circuit experiments as well. Only acetate showed the trend representative of an exoelectrogenesis substrate via heavy accumulation during open‐circuit condition, and fast removal once repoised, further suggesting that the other intermediates are not dominant fermentation end‐products in our system. Additionally, their concentration was an order of magnitude lower than acetic acid, indicating that acetic acid was the dominant branching point to exoelectrogenesis. While it is possible that some of these compounds could serve as substrates for unknown exoelectrogens, many fermentation intermediates such as propionate, butyrate, ethanol and butanol have been shown to be unsuitable for direct exoelectrogenesis (Kiely et al ., 2011 ; Miceli et al ., 2014 ; Hari et al ., 2016 ). Additionally, as described in the community analysis section, some β‐Proteobacteria were found to persist during pure acetic acid experiments and thus could be diverting a small portion of acetic acid during open‐circuit stimulus. Figure 2 Acetic acid ( AA ) removal rates and hydrogen productivity during anode potential interruption experiment. OC : open‐circuit voltage, CC : set anode potential of −0.2 V versus Ag/AgCl reference electrode. Considering the acetic acid production rate of 8.6 mg h −1 during open‐circuit conditions compared with the acetic acid removal rate of 58.7 mg h −1 achieved using 0.5 (g COD) l −1 of pure acetic acid (Additional File 1: Table S3), it is clear that the exoelectrogenic microbial subpopulation is capable of converting acetic acid at higher rates than it is being produced from BOAP. Furthermore, after the circuit was closed following open‐circuit stimulus, the current production reached a higher level than what was achieved without any interruption of circuit, further indicating the exoelectrogenic subpopulation is capable of higher current output and that compounds within BOAP were not inhibitory to either of the subpopulation. Some intermediates produced from BOAP such as formate and lactate are likely substrates for exoelectrogenesis; however, they were not dominant products during the open‐circuit condition. The rate of exoelectrogenesis would be higher if they also serve as substrates for exoelectrogenesis. The observed results during open‐circuit stimulus compared with the closed‐circuit experiment thus demonstrate that the rate of fermentation was the limiting step in conversion of BOAP to current. Biotransformation of individual BOAP substrates under poised conditions While previous work has demonstrated significant removal of the main compounds within BOAP by the end of the run and at high efficiency (Lewis et al ., 2015 ), the relative rates and order of conversion over time of the various components of the complex mixture BOAP have not been reported previously. The implications of this are significant, as pure microbial cultures can struggle with many of the lignin‐derived compounds present in BOAP (Jarboe et al ., 2011 ), while microbial communities can convert complex biomass streams containing these compounds via emerging synergistic capabilities within the consortium. The composition of BOAP is outlined in Table S4. The main compounds within BOAP were all transformed simultaneously within 48 h, although at different rates (Fig. 3 ). Overall COD removal reached 58.4% by 24 h, and further increased to 74.8% by 72 h. For the fermentable substrates, levoglucosan had the highest initial removal rate of 16.59 ± 0.59 mg h −1 over the first 2 h of the batch run, followed by furfural with a rate of 1.35 ± 0.23 mg h −1 (Table 1 ). However, relative to starting concentrations, furfural had the highest initial removal percentage of 87.74 ± 1.33%, with levoglucosan reaching 58.93 ± 2.63%. 5‐hydroxymethylfurfural is another major fermentable compound present in BOAP, which was utilized at a lower rate initially with 28.88 ± 13.47% removal after two hours, but increased to 54.49 ± 6.46% after 10 h. This may be due to lower microbial density or intrinsic reaction rates. For the fermentation by‐products acetic acid and propionic acid, their initial removal rates were 6.55 ± 4.36 and 2.83 ± 0.80 mg h −1 respectively. However, because acetic acid is being produced through fermentation of the other compounds within BOAP simultaneously, its true removal rate is underestimated. Nevertheless, the observation that concentration of acetic acid never increased with time demonstrates that its removal outpaced production. This can also be the case for additional intermediate compounds produced during the conversion of BOAP as phenol and catechol have been identified intermediates from larger phenolic compounds in the literature (Zeng et al ., 2017) and their concentrations were found to fluctuate during our experiments. However, their concentration at the end of the experiment was lower than the starting concentration, indicating that they were still utilized by the anode consortium. Figure 3 Per cent removal of individual model compounds within BOAP as measured by HPLC . The legend refers to the hours at which samples were collected. Table 1 Removal rates of individual compound in mg h −1 during batch experiment with BOAP as the substrate during 2 h blocks of time for the first 10 h, followed by the following 14 h block Time 2 4 6 8 10 24 Levoglucosan 16.59 ± 0.59 7.83 ± 0.59 2.36 ± 0.04 0.04 ± 0.18 −0.01 ± 0.24 0.08 ± 0.06 Acetic acid 6.55 ± 4.36 6.48 ± 2.11 9.62 ± 0.85 8.38 ± 1.06 3.66 ± 1.05 0.24 ± 0.03 Propionic acid 2.83 ± 0.8 1.07 ± 0.63 2.97 ± 0.14 1.30 ± 1.37 1.45 ± 0.88 0.47 ± 0.03 HMF 0.55 ± 0.44 −0.05 ± 0.39 0.17 ± 0.28 0.14 ± 0.03 0.17 ± 0.15 0.08 ± 0.03 2(5H)‐furanone 0.39 ± 0.1 0.02 ± 0.02 0.00 0.00 0.00 0.00 Catechol 0.07 ± 0.07 0.03 ± 0.01 −0.03 ± 0.02 0.02 ± 0.01 −0.01 ± 0.03 −0.02 ± 0.02 Furfural 1.35 ± 0.23 0.19 ± 0.01 0.00 0.00 0.00 0.00 Phenol −0.01 ± 0.13 0.04 ± 0.04 0.01 ± 0.04 0.01 ± 0.06 0.00 ± 0.04 0.01 COD 32.43 33.01 43.43 10.42 N/A 3.19 John Wiley & Sons, Ltd To further understand the productivity, efficiency and biotransformation trends observed during the conversion of BOAP, the theoretical contributions from each compound identified by HPLC towards H 2 production were calculated via an electron equivalence calculation similar to that described in the previous section (Fig. 4 ). This calculation relies on the assumption that removal of the parent compounds results in their complete conversion to CO 2 , electrons and protons (Experimental Section). The bars on the y ‐axis show equivalent rate of hydrogen production if all electrons were recovered as hydrogen at the cathode. While assuming 100% conversion is not possible, visualization in this manner allows us to estimate the extent to which the observed results deviate from this condition in discrete time frames. The results show a lag in hydrogen production compared with the rate of substrate removal. This is not unexpected as the substrate concentrations measured at the various time points are indicative of disappearance of substrate and not necessarily complete conversion. Furthermore, comparing Fig. 4 A and B, we can see an inverse trend in the first 6 h, with contributions attributed to individual compound removal starting high and dropping off, while overall COD‐based contribution starts lower and increases with time. This is indicative of production of biotransformation intermediates or cellular storage, contributing to increasing COD removal from 0 to 6 h, followed by a decreasing trend thereafter. Similar to the anode electron balance described at the end of the previous section, coulombic efficiency from 6 to 24 h exceeded 100% during normal poised conditions, indicating that intracellular storage was being tapped in addition to the substrate present. Figure 4 (A) Comparison of hydrogen productivity obtained experimentally with that estimated via electron equivalence calculation for conversion of individual compounds within BOAP . (B) COD contributions to hydrogen productivity based on electron equivalence compared with observed hydrogen productivity. Community analysis The bioanodes used in these experiments had been exposed to BOAP and adapted to this substrate for > 2 years and have evolved and enriched for BOAP conversion and acetate oxidation [6]. Focusing first on two important groups in the microbial community, exoelectrogens and methanogens, the results show different trends depending on the substrate used (Fig. 5 A). Exoelectrogens, represented by the family Geobacteraceae, increased from 1.9% to 33.0%, when BOAP was used as the substrate. A similar trend was seen when pure acetic acid was used as the substrate, increasing the population density of the exoelectrogens from 15.6% to 54.0%. On the contrary, the population of methanogens showed an opposite trend. With BOAP as the substrate, the methanogenic Euryarchaeota increased from 2.2% to 17.2%, while their population decreased from 13.1% to 1.8% with acetic acid as the substrate. Figure 5 16S rRNA ‐based taxonomical classification of the MEC community for batch BOAP versus acetic acid experiments. Numbers 1 and 2 indicate samples collected at the beginning (1) and the end (2) of each batch series. (A) Bar chart showing taxonomy of the MEC anode community at the phylum level with subclassification of the Proteobacteria at class level. (B) Trends in Archaea versus Geobacter subpopulations observed with the two substrates. Two inferences can be derived from these results. First, batch additions of acetic acid as well as BOAP provide a large amount of acetate, which is preferred by Geobacter for growth and exoelectrogenesis, thus explaining their growth with both substrates. Second, the fact that the population of methanogens was only observed to increase when using the complex fermentable substrate and decreased with pure acetic acid indicates the methanogens present in the anode are not acetoclastic and are likely hydrogenotrophic. Thus, the methanogen population is mainly feeding on intermediates produced during the fermentation process such as H 2 and CO 2 rather than the end‐product acetate, which is predominantly used by the exoelectrogenic fraction(Freguia et al ., 2008 ; Ishii et al ., 2008 ). The growth of methanogens is a well‐documented issue in bioelectrochemical systems even at low organic loading rates, and thus poses a significant challenge for controlling their growth at higher, industrially relevant levels (Cusick et al ., 2011 ; Escapa et al ., 2013 ). The difference in exoelectrogen population at the start and end of BOAP and acetic acid experiments differed greatly, which highlights the effect of substrate and fermentation limitation. Performance in terms of efficiency and current output did not improve as the Geobacter population increased with BOAP as the substrate during this study. This is because the amount of Geobacter is not the main determining factor for performance in the BOAP‐fed system, but rather a consequence of the use of BOAP in the anode. An experiment conducted after acetic acid run using BOAP as the substrate showed similar hydrogen productivity as that prior to the use of acetic acid as substrate in the MEC(Additional File, Fig. S2). As demonstrated in previous section, fermentation of BOAP to end‐products like acetate is a limiting factor in the bioanode conversion process. Once the substrate was switched to pure acetate, the exoelectrogens no longer relied on fermentation to produce the substrates they need and current production increased threefold and the Geobacter population increased further to 54%. Thus, limited availability of acetic acid limits Geobacter growth and affects its population in the anode. This was clearly illustrated by the experiment which was conducted after the acetic acid run (Additional File 1: Fig. S3). This indicates that the MEC performance is not necessarily determined by the Geobacter population, but by the substrate used in the MEC. In addition to the changes in Geobacter and methanogen population, additional taxonomic groups including Firmicutes , Bacteroidetes and multiple classes of Proteobacteria also demonstrated trends as a function of substrate. Firmicutes and Bacteroidetes persisted during BOAP experiments at > 10% of the population, but were reduced significantly when acetic acid was used as the substrate. γ‐Proteobacteria also showed this trend but to a lower extent. This indicates that these microbes are active in the fermentation of parent or intermediate compounds from BOAP and cannot be sustained on acetate alone. Firmicutes have been found frequently in bioelectrochemical anodes when fermentable substrates have been used (Jung and Regan, 2007 ; Rismani‐Yazdi et al ., 2007 ) and this phylum houses many biomass degraders and glucose fermenters in Clostridia . Additionally, certain microbes within γ‐Proteobacteria and Bacteroidetes have also been found to only persist when fed with fermentable sugars (Ishii et al ., 2014 ). In contrast, β‐Proteobacteria declined during BOAP experiments but persisted during pure acetic acid, although it should be noted that this class did still remain at high levels during BOAP feeding despite the overall reduction (Fig. 5 ). Looking closer at the family level of β‐Proteobacteria (Additional File 1: Fig. S4), Rhodocyclaceae and Comamonadaceae have been identified in our previous work and house a wide metabolic range of microbes (Rismani‐Yazdi et al ., 2007 ; Borole et al ., 2009 ; Hesselsoe et al ., 2009 ; Xing et al ., 2010 ; Oren, 2014 ; Lewis et al ., 2015 ). While these families have been implicated in degradation of complex carbon compounds, they have also been found to have abilities in acetate utilization (Ginige et al ., 2005 ) and some members of Comamonadaceae have also been found to be capable of electricity generation (Xing et al ., 2010 ), which would explain both of their abilities to persist during pure acetic acid feeding. Emergent functionality in engineered community The conversion of phenolic and furanic compounds provides a significant challenge to fermentation of BOAP, as these classes of compounds, are known to be inhibitory to many microbes (Jarboe et al ., 2011 ). Fermentation of the furanic compounds, furfural and HMF have been found to produce intermediates such furoic acid, furfuryl alcohol, 2,5‐bis(hydroxymethyl)furan, requiring further biotransformation to be converted to acetic acid (Wierckx et al ., 2011 ; Zeng et al ., 2015 ). Additionally, phenolic compounds such as phenol and catechol are even more recalcitrant, with fermentation proceeding most slowly for these types of compounds in the experiments presented in this study. Many of these intermediates including phenol, catechol and furoic acid were found in MEC effluent when BOAP served as the substrate. The comparative studies with BOAP and acetic acid show that the microbial groups were established and enriched in the anode to collectively participate in a fermentative chain that is capable of oxidizing the complex carbon compounds within BOAP progressively to acetic acid. The evidence presented in this study suggests that the microbial community utilize a synergistic strategy through mutually beneficial division of labour and syntrophic exchange as depicted in Fig. 6 that results in emergent functionality in converting the wide array of compounds within BOAP. Division of labour was evident through the simultaneous conversion of identifiable compounds, which allows for parallel processing and allocation of compounds to various microbes with different functionality and metabolic capabilities. This type of cooperative interaction within microbial communities is seen in natural environments and in engineered settings, manifesting into various ways of enhancing overall substrate utilization (Hays et al ., 2015 ) (Fröstl and Overmann, 1998 ; Crespi, 2001 ; Briones and Raskin, 2003 ; Eiteman et al ., 2008 ). This may also help prevent the toxic effects of many of these compounds on other community members. Downstream from this initial division of labour, the co‐conversions of the fermentable substrates converge to acetic acid as demonstrated by open‐circuit stimulus‐response. This leads to syntrophic cross‐feeding from fermentative groups to exoelectrogenic groups for the generation of current. This type of interaction has been demonstrated in bioelectrochemical systems before with simple substrates and is the foundation for electricity generation from fermentable substrates (Freguia et al ., 2008 ; Parameswaran et al ., 2009 ; Kiely et al ., 2011 ). Compounds such as levoglucosan were strongly preferred within the BOAP mixture by the microbial community, which is not unexpected as it is a sugar derivative, leading to the fastest removal rate. One of the pathways through which levoglucosan can be degraded is using levoglucosan kinase to convert it to glucose 6‐phosphate (Kitamura et al ., 1991 ; Zhuang and Zhang, 2002 ; Dai et al ., 2009 ; Layton et al ., 2011 ). This pathway generates acetic acid, and thus, the conversion of levoglucosan to acetic acid can occur in a single organism. Conversely, conversion of the phenolic and furanic compounds may require multiple steps. Catechol, phenol and furoic acid were found in the MEC effluent and were found to fluctuate with time during the experimental run, indicating their production from other compounds present in BOAP. These compounds have been identified as intermediates in the conversion of methoxy phenols and furanic compounds in a MEC which used the same source of enrichment used in these studies (Zeng et al ., 2017). Thus, exchange of carbon between community members may be occurring at several levels for multiple compounds during biotransformation of the lignin and hemicellulose‐derived intermediates. Syntrophies that direct electrons away from the electrode were also evident. An increase in the population of Euryarchaeota which functions to redirect fermentation intermediates through methanogenesis was found. Nonetheless, CE for the BOAP batch runs reached > 80% demonstrating an efficient and robust community that can convert cellulose, hemicellulose and lignin‐derived pyrolytic intermediates including inhibitory compounds at appreciable rates, providing a foundation for further improvements to reach commercial targets. Figure 6 Schematic of possible pathways active in anode microbiome for conversion of fermentable compounds within BOAP . ‘F’ corresponds to fermentative bacteria, and ‘E’ corresponds to exoelectrogenic bacteria. Intermediate level 1 includes compounds such as phenol, catechol, furoic acid, which were observed experimentally. VA , vanillic acid; SA , syringic acid; HBA , hydroxybenzoic acid; HMF , hydroxymethylfurfural. Bioelectrochemical systems for biorefining applications The need for integrated solutions to the current challenges facing the world continues to grow, and new innovative approaches to fully utilize lignocellulosic feedstocks will be essential. Bioelectrochemical systems have the potential for integration into variety of bioenergy platforms to help meet this goal (Borole, 2011 , 2016 ). MECs offer the potential to produce hydrogen in an integrated fashion in biorefinery platforms and as a means of energy storage through decentralized production to supply hydrogen to fuelling stations, as the world strives to move towards cleaner fuels and electricity‐mediated transportation. This study integrates a number of novel features into one to address practical applications such as use of MECs in biorefineries. These include:\n An approach of comparing complex substrate, BOAP versus acetate, using open‐ and closed‐circuit conditions to understand fermentation versus exoelectrogenesis relevant to biorefinery streams Community focus to understand the interactions within functional microbial groups. This can allow identifications of conditions to promote positive interactions and ways to minimize negative interactions Overall integrative/systems approach, looking at metabolites (individual compounds within mixture), genomics and electrochemical data Use of high‐performing MEC design capable of achieving high productivity and CE without using high concentration of substrates \n The demonstrated maximum rates with BOAP in this study are nearing the targets needed for practical application (Sleutels et al ., 2012 ), and almost 30‐fold higher than that identified by US DOE Fuel Cell Technology Office as state of the art (EERE, 2015 ). Alternate technologies such as in vitro synthetic enzyme systems have achieved comparatively high yields and productivities of 29 L per L‐day (54 mmol H 2 per L‐h) utilizing both xylose and glucose from corn stover hydrolysate 44 . The study presented here is a big step towards realizing the use of renewable waste biomass as a feedstock versus sugars or natural gas for hydrogen production. As shown in this study, developing an emergent microbial community capable of efficiently producing hydrogen from all constituents of biomass can be accomplished, but further work is necessary to increase the productivity to 20 L per L‐day or more to enable commercial consideration. This can be achieved via targeted increase in fermentative population. The generation of electrons from waste biomass has significant implications for the production of high‐value chemicals as well. This can be done via integration of electrosynthesis at the cathode (Rabaey et al ., 2011 ; Borole, 2012 ) with the bioanode developed in this study. Future work will focus on utilizing deep sequencing techniques to better understand the interactions among active groups within the microbial community and additional environmental factors impacting performance. Building on methods previously established such as those using operational ‘shocks’ coupled to metatranscriptomic analysis identification of functional roles of different community members is possible (Ishii et al ., 2013 ). Providing the right environment for improving fermentation while reducing competing pathways such as methanogenesis can enhance the production of acetic acid from BOAP. Continued efforts in these areas can lead to the development of an optimal microbial community management strategy for developing stable and high‐performing electroactive biofilms while contributing to overall strategies for engineering microbial communities for additional industrial applications."
} | 10,359 |
28345029 | PMC5342658 | pmc | 356 | {
"abstract": "Genes involved in an ancient cellular stress response may have a protective role against heat stress on the reef.",
"conclusion": "CONCLUSIONS Temporary heat pulses during strong midday low tides on the reef triggered transcriptional changes and the activation of the UPR in A. hyacinthus . Repeated exposure to similar short-term spikes in temperature may increase coral thermal tolerance and may be especially beneficial in the lead-up to a chronic stress event. Our field experiment highlights the role of the UPR during these short-term stress events and suggests that it is the first line of defense corals initiate when coping with environmental stress. Whether this transcriptional mechanism is common across other coral species, in particular, those living in less environmentally variable reefs, is a topic for future investigation. The rapid expression changes of UPR genes from day to day in A. hyacinthus reveal the high synchrony coral physiology has with the surrounding environment. However, the physiological capacity of corals to quickly bounce back from short-term stress events and build up acclimatization will be tested in future oceans.",
"introduction": "INTRODUCTION The environmental stress continuum represents the range of abiotic conditions that can trigger a stress response in an organism. For scleractinian corals, the conditions along this continuum range from temporary mild stress, such as a spike in temperature in the middle of the day, to chronic stress ( 1 – 8 ) manifested as coral bleaching [reviewed by Douglas ( 9 ) and Lesser ( 10 )]. Repeated exposure to temporary mild stress on the reef may increase coral thermal tolerance, via acclimatization, and provide a mechanism for corals to withstand chronic stress that typically results in bleaching ( 3 , 11 – 13 ). Physiological and transcriptomic data show that coral acclimatization can occur within 1 to 2 weeks ( 14 , 15 ), suggesting that even weeklong increases in mean water temperature might play a protective role for corals on future reefs. Recently, Ainsworth and colleagues ( 16 ) showed that many of the thermal stress events on the Great Barrier Reef occurred in what they termed a “protective” pattern, where an initial pulse of warm water—potentially a trigger for adaptive acclimatization—is followed later by a stronger warm water episode (that is, temperatures exceed the local bleaching threshold) ( 16 ). Although many studies previously described a strong transcriptional response in corals to bleaching ( 11 , 13 – 15 , 17 – 21 ), none investigated whether natural tidal heat pulses on the reef cause transcriptomic changes, what temperatures are required, or what physiological mechanisms are subject to change. This information will be highly useful in identifying future temperature trajectories that might continue to provide the kind of bleaching protection that Ainsworth et al . ( 16 ) describe. For many organisms, stress triggers a physiological response that begins with alterations in gene transcription ( 22 – 24 ). This pattern is seen in corals before ( 17 , 18 ) and during bleaching in the laboratory ( 11 , 13 – 15 , 17 , 19 – 21 ). A handful of field-based studies identified significant gene expression alterations in corals faced with chronic environmental stress or disease outbreaks ( 25 – 27 ). However, identifying the environmental conditions that trigger the initial disruption to organismal homeostasis requires monitoring the same coral individuals as they experience diverse degrees of environmental stress. Using high-resolution transcriptomic and environmental profiling, we monitored transcriptional regulation in coral colonies on a reef exposed to a natural short-term pulse of warm water, analogous to the prestress period followed by a recovery period in the protective trajectory reported by Ainsworth et al . ( 16 ) ( Fig. 1A ). We used the water temperature changes that occur across a typical 2-week tidal cycle to impose a variety of daily environmental extremes on corals, monitored the environment, and tested for transcriptional changes. We selected a highly variable back-reef pool of Ofu Island, American Samoa, that has been shown to reach 34° to 35°C during summer daytime low tides ( 3 , 13 ). Similar high variability reef environments exist throughout the Pacific, such as Palmyra Atoll ( 4 , 5 , 28 ); Davies Reef, Great Barrier Reef ( 29 ); Heron Island ( 8 ); and Kaneohe Bay, Hawaii ( 30 ). We focused on colonies of the tabletop coral Acropora hyacinthus living in the U.S. National Park of American Samoa, where previous studies found that coral colonies adjusted to thermal stress conditions through physiological acclimatization and genetic adaptation ( 12 , 31 , 32 ). Our field experiment identified a pivotal temperature, above which these corals mounted a strong but temporary transcriptional response, and a group of genes with coordinated expression that increased only on days with the highest temperatures. This set of genes is enriched for unfolded protein response (UPR) proteins. The UPR is an ancient eukaryotic cellular pathway involved in detecting and responding to the early stages of physiological stress. In corals, this mechanism may have been co-opted as a first line of defense to heat stress. Fig. 1 Coral reef temperature and gene module expression profile. ( A ) Protective temperature trajectory as in the study of Ainsworth et al . ( 16 ). PS, prestress period; RP, recovery period; SST, sea surface temperature; MMM, the maximum monthly mean. ( B ) Three colonies of A. hyacinthus that were sampled. Gold box (left), AH06; blue box (middle), AH75; red box (right), AH88. ( C ) Temperature during the 1 hour before transcriptome sampling for each day of the 17-day time series (1300 to 1410; n = 8 time points). Temperature data are the average for the three colonies, which had near-identical temperature profiles (see fig. S5). ( D ) Expression profile for the UPR and calcium homeostasis module (also known as module RJ9), which was significantly associated with changes in the environment [analysis of variance (ANOVA), false discovery rate (FDR)–corrected, all P < 0.0001]. Expression levels are represented by eigengenes for each day. Each color represents a different colony. Gold, AH06; blue, AH75; red, AH88.",
"discussion": "DISCUSSION Using environmental transcriptional profiling and gene coexpression network analysis, we show that tabletop corals in field settings rapidly make transcriptional adjustments when faced with stress far below the threshold that induces bleaching. Our findings document the specific transcriptional changes that occur in A. hyacinthus in a back-reef environment during 2 days of mild stress events. Although we did not test for acclimatization in this experiment, previous studies showed enhanced thermal tolerance in acroporids after 7 to 11 days of similar temperature cycles in laboratory trials ( 14 , 15 ). In our study, expression changes were clearly visible above a maximum daily temperature of 30.5°C, and expression quickly decreased when temperatures fell below 30°C, suggesting that corals have a fine-tuned response mechanism to maintain homeostasis during periods of environmental stress. Temperature spikes trigger transcriptomic response On days reaching above 30.5°C, the sampled corals all increased expression of the UPR and calcium homeostasis module. On Ofu Island back reefs, corals routinely experience high temperatures but only for brief periods of time. In the area we sampled, temperatures reach or exceed 30.5°C on approximately 85 days in a typical year, for a total of about 2.5% of the time. Although the corals activated the UPR on days above 30.5°C, the response was short lived: The expression returned to baseline levels a day later when temperatures fell below 30°C ( Fig. 1 , C and D). The fine-tuned expression changes of the UPR and calcium homeostasis module from day to day follow an impulse-like pattern that is typical of expression changes in response to environmental stimuli; there was a temporary spike in expression associated with an environmental stimulus followed by a return to baseline levels once the stimulus was removed ( 23 ). Repetitive stress events, such as those that occur during strong midday low tides, may serve to increase coral thermal tolerance ( 11 – 13 ), which is especially important in the lead-up to a bleaching-inducing stress event ( 16 ). For example, daily fluctuations from 29° to 31°C for 7 to 11 days induced thermal acclimation in Acropora nana just as strongly as did constant exposure to 31°C ( 15 ). Transcriptomic data from laboratory bleaching experiments with corals from the same back reef ( 17 ) show a different pattern at higher temperatures than what we measured on the reef. After a 3-hour exposure to 34°C in the laboratory, genes in the UPR and calcium homeostasis module also increase, but expression levels for unbleached colonies (bleaching level 1 in Fig. 4 ) are higher than field expression on days 7 and 8. Colonies with substantial bleaching after a 34°C treatment show even higher expression levels of the UPR and calcium homeostasis module ( Fig. 4 ), suggesting that a transient UPR response at low temperatures might convert to a stronger response under more intense stress. Environmental correlates of temperature stress In addition to temperature, other stressors occur on days with strong low tides around midday and midnight: On the nights of days 7 and 8, the corals were exposed to pH 7.78 and DO saturation around 50%. At night, DO levels at the coral boundary layer can be significantly lower than levels in the surrounding water column ( 43 , 44 ). Furthermore, days 7 and 8 had the highest day-night variability in temperature, pH, and DO saturation, exposing the colonies to a wide range of these variables over a relatively short period of time. The daily variability in pH and DO levels is driven by biogeochemical cycles in the reef ( 4 – 6 , 28 , 45 ). The role of the UPR The UPR and calcium homeostasis module is enriched for genes associated with the UPR during ER stress. The ER serves three main functions in eukaryotes: It is the site of protein folding for newly synthesized secretory and membrane proteins, with at least one-third of the cell’s proteins passing through the ER ( 46 ); it stores intracellular calcium ions; and the membrane is the site of lipid and sterol biosynthesis ( 42 ). The UPR is an evolutionarily conserved set of signaling pathways that are activated when unfolded proteins accumulate in the ER ( 47 ). ER stress can be triggered by environmental stress, point mutations that affect protein folding efficiency, and loss of calcium homeostasis. When the stress is mild, the UPR initiates an adaptive response that restores proteostasis and homeostasis. However, when the stress is more severe and damage is irreparable, the UPR switches to a terminal response that ultimately results in apoptosis ( 46 ). As part of the adaptive response, the UPR reduces translation of most proteins in a cell ( 47 ) but induces transcription of a specific set of genes, including those that encode ER-resident chaperones and genes for proteins involved in ER-associated protein degradation ( 48 ). To reestablish homeostasis in the ER, the adaptive UPR induces changes that increase the ER’s protein folding capacity, increase the capacity to clear out misfolded proteins from the ER through the up-regulation of ER-associated protein degradation, and globally reduce de novo protein synthesis to minimize protein congestion in the ER ( 46 ). There are multiple genes in the UPR and calcium homeostasis module that encode protein products localized to the ER and involved in the UPR that are up-regulated during the strong midday low tides. There are ER-resident chaperones, such as heat shock 70 kDa, and co-chaperones, such as DnaJ homolog and calreticulin ( 42 ). Furthermore, ER degradation–enhancing α-mannosidase–like 1, which is essential to the ER-associated protein degradation activity of the UPR and is up-regulated during ER stress ( 49 ), is present in the UPR and calcium homeostasis module. One of the transcription factors that induce expression of several chaperones as part of the UPR is cyclic AMP–dependent transcription factor ATF-2 ( 42 ). This gene is present in the UPR and calcium homeostasis module, suggesting an association between its expression and those of the chaperones. There is also a second ER-localized transcription factor in the UPR and calcium homeostasis module, cyclic AMP–responsive element–binding protein 3–like protein 3, that activates some UPR target genes during ER stress ( 50 ). In our field experiment on corals, increased expression of multiple genes, whose protein products are involved in the UPR, suggests that a cellular response was under way to restore homeostasis to the ER. The high temperatures during the strong midday low tides were likely a mild stress that triggered a temporary loss of homeostasis in the ER and caused the induction of the UPR in A. hyacinthus . Our field results also suggest increased calcium ion–binding activity during the strong midday low tides. Calcium-binding proteins are an essential part of the ER’s function because the cell’s major store of intracellular calcium ions and ER stress can disrupt intracellular calcium homeostasis ( 42 ). Several studies show that genes involved in calcium ion signaling and homeostasis are differentially expressed in heat-stressed corals ( 13 , 14 , 19 – 21 ). In A. hyacinthus , we identified at least seven different genes in the UPR and calcium homeostasis module that are involved in calcium ion binding and calcium homeostasis: calcium-binding protein CML19 , calreticulin , calsequestrin-2 , calumenin , reticulocalbin-3 , SERCA2 , synaptotagmin-4 , and synaptotagmin-7 . SERCA2 is involved in controlling the influx and efflux of calcium ions in the ER and is activated by a UPR-associated transcription factor ( 42 ). The abundance of calreticulin, a calcium ion–binding chaperone in the ER, has a positive relationship with the concentration of calcium ions in the ER ( 51 ). Calumenin is a calcium-binding protein localized to the ER and involved in protein folding ( 52 ). In Acropora millepora , calumenin expression decreased in colonies that bleached after an extreme heat stress but increased in thermally acclimated colonies that did not bleach ( 14 ), possibly indicating that the thermally acclimated colonies experienced sub-bleaching stress, as we see a similar pattern in the field. The presence of multiple calcium ion–binding proteins in the UPR and calcium homeostasis module, as well as their up-regulation during the strong midday low tides, further supports the finding that homeostasis in the ER was disrupted. Our interpretations are based on the expression pattern and functional enrichment of the UPR and calcium homeostasis module. Future investigation of essential UPR genes in corals, through in situ hybridization and proteomics, would be valuable. Dynamic switching of the UPR and bleaching A key feature of the UPR is that it switches to a terminal response when the stress is extreme and/or persistent (that is, it becomes chronic) and the damage is irremediable. The balance between maintaining an adaptive UPR and switching to the terminal UPR involves the signaling dynamics of the UPR sensors. When ER stress becomes chronic, the PERK and IRE1α signaling proteins induce apoptosis through multiple cell death networks, including making changes to several caspases and activation of the canonical apoptosis pathway in the mitochondria ( 46 ). Ainsworth and colleagues ( 16 ) found that expression levels of six apoptosis genes in corals of Acropora aspera exposed to their protective temperature profile were more similar to nonstressed corals than corals exposed to their more severe heat stress that induced bleaching. Similarly, in our data, there are no signs that apoptosis was initiated—genes involved in apoptosis did not increase expression during the strong midday low tides. However, multiple apoptosis-related genes are up-regulated when corals are experimentally bleached, suggesting that the apoptosis pathway is part of bleaching physiology ( 13 , 14 , 17 , 19 – 21 ). The absence of a significant transcriptional signal of apoptosis in our field experiment leads us to conclude that the corals were exposed to a mild stress at the lower end of the environmental stress continuum and that the UPR remained capable of restoring proteostasis and homeostasis (that is, it remained adaptive). When corals are exposed to environmental stress, the ER may be the first cellular component to lose homeostasis, and ER stress may represent the first type of cellular stress that occurs. Activation of the UPR to cope with ER stress then represents the initial physiological response corals have to restore homeostasis to the ER and the organism. Stress-induced genes, such as HSP70 , tend to ramp up expression as stress is amplified or persists ( 53 – 55 ). In corals, thousands of additional genes are up-regulated after experimental bleaching ( 17 ). The fact that expression of the UPR and calcium homeostasis module is higher in bleached corals than our field corals and that thousands of additional genes are activated in bleached corals suggests that our field observation of the UPR represents the first line of defense corals initiate when coping with environmental stress. However, if the environmental stress persists and homeostasis is not restored to the ER, perhaps the UPR switches to a terminal response, which is then associated with the initiation of the physiological adjustments that are made during bleaching. Our hypothesis is that the UPR, with its dynamic ability to switch from an adaptive to a terminal response, is a first line of defense when corals are initially coping with environmental stress. However, if homeostasis cannot be restored, then the UPR would play an important role in initiating bleaching. Whether the cellular mechanisms known to regulate the terminal branch of the UPR in model systems also contribute to the physiological events leading to bleaching is a topic for future investigation that may elucidate some of the less understood mechanisms of bleaching. Protective heat pulses and bleaching acclimatization Our data do not fully explore the induction of heat tolerance from protective warm water pulses. Our use of a tidal cycle as a proxy for a natural heating event did not include enough days with temperature extremes to expect substantial acclimatization [see the studies of Bellantuono et al . ( 14 ) and Bay and Palumbi ( 15 )], nor was it followed by a period with temperatures above the local bleaching threshold. We also followed only one tidal cycle with a limited range of temperature extremes. Nevertheless, our data suggest that transient temperatures above 30.5°C trigger the first physiological response to stress, compared to the average summer water temperature of about 29.3°C for this location ( 1 ). For this species, bleaching does not occur until about 33°C for temperature-sensitive colonies and 34°C for acclimatized individuals ( 12 ). These values are difficult to put into context with the predictions of current bleaching models that focus on degree heating weeks ( 56 ), because current models emphasize broader temperature profiles collected remotely over longer time intervals than our coral temperature loggers. However, as the ability to characterize local reefs and their temperature microhabitats improves, it may be possible to define the daily temperature rhythms that are associated with protective heat signatures that lead to short-term heat tolerance ( 16 ). Our data suggest that there may be an additional type of protective thermal signature. On Ofu Island, strong low tides tend to be in the middle of the day year-round, exposing corals in the back reef to large swings in temperature. In laboratory experiments, these swings are just as effective at inducing thermal acclimation as are chronic high temperatures ( 15 ). The synchrony of low tides and midday high temperature has been offered as one of the reasons why Ofu corals are so temperature-resilient ( 3 ). Because these facets of the tidal cycles are broadly predictable and stable, it may be possible to create a map of where tidal protection is likely for corals."
} | 5,120 |
39854124 | PMC11923897 | pmc | 358 | {
"abstract": "Abstract Flexible memristors are promising candidates for multifunctional neuromorphic computing applications, overcoming the limitations of conventional computing devices. However, unpredictable switching behavior and poor mechanical stability in conventional memristors present significant challenges to achieving device reliability. Here, a reliable and flexible memristor using zirconium‐oxo cluster (Zr 6 O 4 OH 4 (OMc) 12 ) as the resistive switching layer is demonstrated. The optimization of the structural rigidity of the hybrid oxo‐cluster network by thermal polymerization allows the precise formation of dispersed conductive cluster networks, enhancing the repeatability of the resistive switching with mechanical flexibility. The optimized memristor exhibits endurance of ∼10 4 cycles and stable memory retention performance up to 10 4 s, maintaining a high I \n ON / I \n OFF ratio of 10 4 under a bending radius of 2.5 mm. Moreover, the device achieves a pattern recognition accuracy of 97.44%, enabled by highly symmetric analog switching with multilevel conductance states. These results highlight that hybrid metal‐oxo clusters can provide novel material design principles for flexible and reliable neuromorphic applications, contributing to the development of artificial neural networks.",
"conclusion": "3 Conclusions In conclusion, we systematically investigated the potential of Zr 6 ‐oxo clusters for the development of flexible memristor devices. The structural rigidity of the Zr 6 ‐oxo cluster, optimized through a thermally activated polymerization process at 150 °C, allows for precise control over the growth dynamics of metal clusters. This control facilitates the formation of Ag metal clusters, enhancing resistive switching reliability by preventing excessive CF paths. The optimized device demonstrated exceptional repeatability in resistive switching with an endurance exceeding 10 4 cycles and stable memory retention performance up to 10 4 s, even under mechanical stress, while maintaining a high I \n ON/OFF ratio. Furthermore, the device exhibited high reliability under mechanical bending with radii of 2.5, 5, and 10 mm. With highly linear analog conductance modulation, the memristor achieved a remarkable accuracy of 97.44% in pattern recognition simulations, revealing its potential as an advanced flexible memory component for artificial neural networks and neuromorphic computing. These results validate the robustness of the zirconium‐oxo‐cluster‐based memristor and highlight their potential in flexible memory applications for advanced computing systems.",
"introduction": "1 Introduction With the recent advancement of next‐generation computing technologies, such as artificial intelligence and machine learning, a growing demand for more efficient processing of large amounts of data exists. [ \n \n 1 \n , \n 2 \n , \n 3 \n \n ] However, the limitations of conventional computing concepts based on the von Neumann architecture are becoming increasingly apparent. [ \n \n 4 \n , \n 5 \n \n ] In this architecture, the physical separation between the central processing unit and the memory unit leads to significant energy and time costs in transferring data, referred to as the von Neumann bottleneck. [ \n \n 6 \n , \n 7 \n , \n 8 \n \n ] To overcome these inherent limitations, emerging memory devices, such as resistive random‐access memory (ReRAM), ferroelectric random‐access memory, and phase‐change memory, have been explored to build neuromorphic computing architectures inspired by the synaptic structure of the human brain. [ \n \n 9 \n , \n 10 \n , \n 11 \n \n ] Among these devices, ReRAMs, also known as memristors, are regarded as promising for neuromorphic applications that emulate the synaptic functions of biological neurons, owing to their high scalability, low power consumption, and high operating speed. [ \n \n 12 \n , \n 13 \n \n ] \n Memristors have a simple two‐terminal configuration, with the switching active layer sandwiched between an electrochemically active electrode (e.g., Ag, Cu) and an inert electrode (e.g., Au, Pt). [ \n \n 14 \n , \n 15 \n \n ] The mechanism for resistive switching between two distinct resistance states, i.e., a low‐resistance state (LRS) and high‐resistance state (HRS), is typically based on the formation and rupture of conductive filaments (CFs) between the two electrodes. Under a positive electric field, the atoms of the active metal are oxidized into ions. These ions migrate through the switching active layer, resulting in the formation of CFs, which corresponds to the LRS. The LRS can be switched to the HRS by applying an electric field of opposite polarity, which disrupts the CFs. Unfortunately, achieving high‐performance and reliable operation of filamentary memristors has been challenging owing to the stochastic nature of filament formation and rupture. [ \n \n 16 \n , \n 17 \n , \n 18 \n , \n 19 \n \n ] The primary issue is the arbitrary growth of filaments in the switching active layer. [ \n \n 20 \n \n ] Randomly created filaments can either remain permanently owing to excessive charge injection of metal ions or become thin and fragile in certain regions, leading to unpredictable switching behavior. [ \n \n 16 \n \n ] Consequently, the dispersed conductive pathways created by these randomly formed filaments may interfere with reliable switching, particularly during repeated measurements. [ \n \n 21 \n \n ] To address these challenges, numerous approaches have been reported to develop highly reliable memristor devices utilizing a wide range of materials, including oxides, 2D materials, nanomaterials, metal halides, and organic–inorganic hybrid materials. [ \n \n 22 \n , \n 23 \n , \n 24 \n , \n 25 \n , \n 26 \n \n ] Despite extensive research efforts, the development of flexible memristors faces significant challenges in achieving reliable operation and high performance, such as low cyclic endurance, low on–off ratio, and asymmetry in conductance change, while maintaining mechanical flexibility. Consequently, new strategies aimed at optimizing the structural design of the material are essential for achieving ideal characteristics in flexible synaptic devices, including reliable and linear analog conductance modulation. As a typical organic–inorganic hybrid material, metal‐oxo clusters (MOCs) are promising candidates for next‐generation electronic devices, exhibiting versatile functionalities owing to the unique characteristics resulting from the combination of inorganic and organic materials. [ \n \n 27 \n \n ] Compared to other materials, MOCs offer a distinct advantage in terms of versatility in the fabrication process and straightforward synthesis, achieved through various combinations of inorganic MOC cores and organic ligands. [ \n \n 28 \n \n ] Particularly, the ligands of MOCs can be precisely engineered to modulate their chemical, mechanical, and electrical properties, making them promising for next‐generation electronics such as high‐resolution nanopatterning and flexible electronic applications. Herein, we report a highly reliable and flexible memristor device fabricated using zirconium‐oxo (Zr 6 ‐oxo) clusters (Zr 6 O 4 OH 4 (OMc) 12 ) as the switching active layer. The Zr 6 ‐oxo clusters can be precisely controlled through a thermally activated polymerization process, which imparts structural rigidity and resists the formation of CFs or metal clusters. The switching mechanism of the Zr 6 ‐oxo‐cluster‐based memristor is fundamentally driven by the formation of a 3D conductive cluster network rather than conventional filaments, enhancing the reliability and stability of the performance of the memory device. The optimized memristor exhibited highly reliable operation with an outstanding endurance of >10 4 cycles, a high I \n ON / I \n OFF ratio of 10 4 , and excellent retention characteristics up to 10 4 s. Based on the nonvolatile memory properties of Ag/Zr 6 ‐oxo/Au memristors, we successfully demonstrated a flexible memristor device on a flexible polyimide substrate with an endurance of >5 × 10 3 cycles, an I \n ON / I \n OFF ratio of 10 4 , and excellent retention characteristics up to 10 4 s under a bending radius of 2.5 mm. Furthermore, reliable synaptic characteristics with highly linear analog conductance modulation were achieved, and the device exhibited a remarkable accuracy of 97.44% in pattern recognition simulation, revealing its potential as an advanced flexible memory component in artificial neural networks and neuromorphic computing.",
"discussion": "2 Results and Discussions The molecular structure of a Zr 6 ‐oxo cluster consists of an octahedral zirconium core coordinated by eight oxygen atoms, with methacrylate‐based ligands (OMc) ( Figure \n 1 a ). The Zr 6 ‐oxo clusters can be cross–linked through thermal polymerization, facilitated by the presence of methyl methacrylate ligands containing C═ C bonds at the periphery of the clusters. The potential mechanism for the cross–linking reaction of the Zr 6 ‐oxo cluster is depicted in Figure 1a . Mild heating activates the C═C bonds, generating reactive radical species that can easily induce cross–linking between the C═C bonds on the methacrylate ligands, leading to the formation of a cross–linked cluster network. Figure 1b shows the Fourier‐transform infrared spectroscopy (FT‐IR) results of the Zr 6 ‐oxo cluster thin film under various thermal annealing conditions. The ligand shows three main infrared (IR) signals in the range of 650–1650 cm −1 . Notably, the characteristic IR peak at 1246 cm −1 , corresponding to the C═C bonding (C(CH 3 )═CH 2 ) in the methacrylate ligands of the Zr 6 ‐oxo cluster, decreased with increasing thermal annealing temperature due to the thermal cross–linking process. In addition, a noticeable shift in the diffraction peak position to a lower angle was observed in the grazing incidence X‐ray diffraction (GIXRD) patterns, indicating a transition from C═C bonds to C─C bonds resulting from the thermal cross–linking reaction (Figure S1 , Supporting Information). These findings confirm that the Zr 6 ‐oxo cluster undergoes cross–linking through thermal polymerization, consistent with a previous report. [ \n \n 29 \n \n ] \n Figure 1 a) Schematic of the molecular structure and cross–linking reaction of Zr 6 O 4 OH 4 (OMc) 12 clusters through the thermally activated polymerization process. b) FT‐IR results of the Zr 6 ‐oxo cluster thin film under various thermal annealing conditions. c) Elastic modulus of the Zr 6 ‐oxo cluster thin film at different annealing temperatures. Schematic of the diffusion and migration of silver cations in the Zr 6 ‐oxo cluster thin film with d) low, e) intermediate, and f) high rigidity. The material rigidity of the Zr 6 ‐oxo cluster, utilized as a switching matrix, plays a crucial role in regulating ion diffusion and migration, thereby affecting the formation of internal conduction paths. A less rigid Zr 6 ‐oxo cluster is expected to facilitate faster ionic diffusion, whereas the cross–linked Zr 6 ‐oxo cluster, with increased structural rigidity, resists the formation of external metal filaments or clusters due to slower ionic diffusion and mechanical suppression of large Ag crystal growth within the matrix. Figure 1c shows the elastic modulus of the Zr 6 ‐oxo thin film annealed at various temperatures. The higher modulus values suggest that Ag ion migration in the cluster matrix can be effectively controlled. The increase in the elastic modulus, as observed from the force–distance curves obtained via atomic force microscopy (AFM) (Figure S2 , Supporting Information), with increasing temperature, further confirms a substantial enhancement in the mechanical rigidity of the Zr 6 ‐oxo thin films. This increased rigidity significantly affects the formation of conduction paths in the Zr 6 ‐oxo cluster switching matrix, as the enhanced mechanical constraint may inhibit the formation of continuous filaments, instead promoting dispersed and isolated Ag clusters across the film, as illustrated in Figure 1d–f . To explore the potential application of the Zr 6 ‐oxo cluster as a switching active layer, we designed a device structure with an Ag/Zr 6 ‐oxo/Au vertical crossbar electrode configuration, illustrated in Figure \n 2 a . The cross‐sectional scanning electron microscopy (SEM) image in Figure 2b confirms the successful integration of each layer within the memristor device. The resistive switching mechanism of the Ag/Zr 6 ‐oxo/Au device follows the typical electrochemical metallization process, depicted in Figure 2c,d . During a positive voltage sweep, Ag atoms oxidize into Ag + ions, which migrate through the Zr 6 ‐oxo matrix, slightly reducing its resistivity. Upon the subsequent reduction of Ag, conduction paths form within the Zr 6 ‐oxo matrix, leading to an abrupt resistance switch from the HRS to LRS. These conduction paths, created by Ag clusters, are maintained after the SET operation even without external bias. When a negative voltage is applied, the device switches from the LRS to HRS due to the reversible redox reaction and dissolution of Ag filaments. [ \n \n 31 \n \n ] Figure 2e shows the current–voltage ( I–V ) characteristics under various annealing conditions, exhibiting typical nonvolatile resistive switching behaviors during a DC voltage sweep. A compliance current of 5 × 10 −4 A was set during the positive voltage sweep to prevent permanent breakdown of the devices. During the negative voltage sweep, the compliance current was released to dissolve the conductive paths, which required a higher current to break the Ag conductive pathways. Notably, resistive switching characteristics were observed under all the annealing conditions, indicating a correlation with the film's density (Figure S3 , Supporting Information). As the annealing temperature increased, a noticeable decrease in the HRS (off‐current state) was observed, primarily due to the film densification, which restricted ion mobility and suppressed Ag crystal growth. [ \n \n 32 \n \n ] \n Figure 2 a) Schematic of the device structure featuring the Ag/Zr 6 ‐oxo/Au vertical crossbar electrode configuration. b) Cross‐sectional SEM image of the Ag/Zr 6 ‐oxo/Au device. Schematic representation of the switching mechanism in Zr 6 ‐oxo‐cluster‐based memristors between the c) HRS and d) LRS. e) I–V characteristics of memristor devices under three different annealing conditions. Representative cyclic endurance tests for a f) device annealed at 100 °C (device 1), g) device annealed at 150 °C (device 2), and h) device annealed at 200 °C (device 3). i) I–V characteristics of device 2 over 500 DC sweep cycles. j) Cumulative probability distributions of SET and RESET voltages for cycle‐to‐cycle measurements. k) Resistance distribution in the HRS and LRS of device 2 over 10 4 cycles. l) Memory retention characteristics of device 2. To further confirm the repeatability of the resistive switching characteristics, we performed endurance measurements by applying a series of SET pulses (2 V with a duration of 5 ms) and RESET pulses (−4 V with a duration of 5 ms) to the device over 100 cycles. Specifically, devices annealed at 100 °C (referred to as device 1) exhibited significant vulnerability, undergoing permanent breakdown after only 50 cycles of voltage sweep, as shown in Figure 2f . This phenomenon suggests that lower annealing temperatures might compromise the ability of the device to maintain stable conduction pathways owing to the overgrowth of CFs. In contrast, device 3 (Figure 2h ), which was annealed at 200 °C, exhibited random and unpredictable resistive switching characteristics. These fluctuations in the resistance state led to difficulties in defining the HRS and LRS, which are crucial for the reliable operation of memory devices. The challenge in establishing these distinct states arises from the insufficient formation of CFs, attributed to the restricted Ag ion diffusion and metallic Ag crystal growth at these higher annealing temperatures with increased mechanical rigidity. The device annealed at the optimized condition of 150 °C (device 2) demonstrated superior performance, showing reliable and stable switching behavior without any permanent breakdown (Figure 2g ). This robust performance represents the optimal condition for ensuring the durability and reliability of the device by effectively modulating ion transportation, which directly contributes to the growth dynamics of the metal clusters. Overall, the different switching characteristics among devices 1–3 were mainly due to variations in Ag diffusivity, resulting from differences in the ionic conductivity of Ag cations within the Zr 6 ‐oxo thin film layers of different densities. Consequently, the reliability of the resistive switching characteristics is highly dependent on the kinetics of the ion transport process in the switching active layer. The results from the I–V characteristics of the three different devices in continuous DC sweep mode are consistent with the cyclic endurance data (Figure S4 , Supporting Information). Figure 2i shows the I–V characteristics of the optimized device 2 over 500 repeated resistive switching cycles. The statistical data for the SET and RESET voltages are plotted in Figure 2j , showing a uniform SET/RESET switching voltage dispersion over 100 cycles. The SET voltage ( V \n SET ) is defined as the point where the current abruptly increases, indicating the transition to the low‐resistance state (LRS). Conversely, RESET voltage ( V \n RESET ) is defined as the point in the negative bias region where the current gradually decreases and reaches the fully “off” current state, corresponding to the high‐resistance state (HRS). The V SET was observed to range from 0.8 to 1 V, whereas the V \n RESET varied from −2.75 to −4 V. These switching voltages are strongly dependent on the thickness of the zirconium‐oxo cluster layer, as they directly influence the electric field required for ion migration. By optimizing the thickness of the zirconium‐oxo cluster layer, the memory window and operating voltage can be precisely adjusted (Figure S5 , Supporting Information). The endurance test, conducted using a series of SET pulses (2 V with a duration of 5 ms) and RESET pulses (−4 V with a duration of 5 ms) with a read voltage of 0.1 V, showed stable resistive switching operation with a high I \n ON/OFF ratio of 10 4 over >10 4 cycles (Figure 2k ). Additionally, both the HRS and LRS were maintained without memory degradation for up to 10 4 s with a high I ON/OFF ratio of 10 5 (Figure 2l ). To further evaluate the impact of temperature variations on device 2, measurements were conducted after raising the temperature to 50 °C to simulate thermal effects caused by consecutive operational cycles. As shown in Figure S6 (Supporting Information), the device maintained consistent resistive switching behavior with no significant changes in V \n SET , V \n RESET, and I \n ON/OFF ratio. These results suggest that adjusting the cross–linking density of the resistive switching matrix based on the Zr 6 ‐oxo cluster layer through thermal cross–linking is an effective strategy for achieving highly reliable resistive switching with enhanced cyclic endurance. To directly investigate the growth structure of the Ag conductive path in the Zr 6 ‐oxo cluster matrix, we fabricated lateral devices (L‐device n, n = 1–3 corresponds to 100, 150, and 200 °C, respectively) specially designed with a near‐microscale gap between the two electrodes ( Figure \n 3 a ). The formation dynamics of the Ag conductive paths were observed using field‐emission SEM (FE‐SEM). As the positive voltage bias increased, the conduction path between the Ag and Au electrodes, originating from Ag nucleation, gradually became larger and thicker (Figure 3b,c ). Consequently, the subsequent growth of the Ag nucleation is physically connected to the inert electrode, leading to LRS conductance. More importantly, the growth of the conduction paths varied significantly among the devices subjected to different annealing temperatures. For L‐device 1, the conduction path constituted a relatively large portion of the overall filament growth (Figure 3d ). In contrast, in L‐device 2, the conduction path appeared narrower (Figure 3e ), due to its densely cross–linked network with enhanced rigidity, which resulted from the higher annealing temperature of the Zr 6 ‐oxo matrix. On the other hand, for L‐device 3, filament growth appeared more difficult, likely hindered by suppressed cation mobility, resulting in incomplete conduction paths (Figure 3f ). The Ag nucleation in the Zr 6 ‐oxo cluster layer, structurally tuned by thermally activated polymerization, might have promoted the formation of Ag metal clusters rather than excessively grown conductive paths. These Ag metal clusters within the cross–linked Zr 6 ‐oxo matrix are expected to exist in discrete, localized regions, suggesting the potential for multiple resistance states. To confirm the feasibility of multilevel programming in our device, the compliance current was incrementally adjusted in a positive voltage step, achieving multiple LRS conductance levels (Figure S7 , Supporting Information). Additionally, multiple HRS conductance levels were observed through controlled RESET stop voltage ( V \n stop ). As V \n stop increased, the corresponding current level decreased in the retention curve, confirming the capability of achieving multiple HRS levels through controlled reset voltages. This behavior is further supported by FE‐SEM images (Figure S8 , Supporting Information), which illustrate the spatial distribution of Ag clusters under different reset voltage bias conditions. The formation of Ag metal clusters, facilitated by the controlled structural rigidity of the Zr 6 ‐oxo cluster, is believed to enhance the repeatability and stability of the resistive switching behavior. Figure 3 FE‐SEM images of lateral device 1 (L‐device 1) designed with a near‐microscale gap between the two electrodes, subjected to a voltage bias of a) 0 V, b) 1 V, and c) 5 V. FE‐SEM images showing the LRS of devices under various annealing conditions: d) L‐device 1 annealed at 100 °C, e) L‐device 2 annealed at 150 °C, and f) L‐device 3 annealed at 200 °C. Cross‐sectional high‐resolution TEM image of vertical memristor device 2 in the g) HRS and h) LRS. The inset shows Ag clusters embedded in the Zr 6 ‐oxo cluster layer. i) FFT pattern of the specific area shown in the inset of (h). To verify Ag nucleation in the Zr 6 ‐oxo thin film layer, bright‐field transmission electron microscopy (TEM) imaging was performed on the optimized memristor device 2 in both the HRS and LRS. For the Zr 6 ‐oxo layer in the HRS, Figure 3g shows the vertical structure of the Ag/Zr 6 ‐oxo/Au device, without noticeable Ag nucleation inside the Zr 6 ‐oxo layer. However, in the lattice‐resolved TEM image and its fast Fourier transform (FFT) pattern in Figure 3h,i , obtained after applying a voltage bias of 2 V to the Ag electrode, lattice fringes with a spacing of 0.205 nm were observed in the Zr 6 ‐oxo layer, indicating Ag nucleation growth along the Ag (200) crystal plane. Furthermore, we observed a shortening of the effective distance between the two electrodes, resulting from the growth of Ag ions at the surface of the counter electrode. [ \n \n 32 \n , \n 33 \n , \n 34 \n \n ] \n Based on a comprehensive understanding of the resistive switching properties of the Zr 6 ‐ oxo clusters, a flexible Ag/Zr 6 ‐oxo/Au device was fabricated on a polyimide substrate, as shown in Figure \n 4 a . The electrical performance of the flexible device, based on device 2, which demonstrated highly reliable cyclic endurance, was evaluated in a bent state. Figure 4b shows the I–V characteristics of the device over 100 consecutive DC sweep cycles. The flexible device consistently exhibited nonvolatile resistive switching behaviors, consistent with the results obtained from the flat device. In addition, the flexible device showed robust resistive switching performance under bending‐induced strains of 1.0%, 0.5%, and 0.25%, corresponding to bending radii of 2.5, 5, and 10 mm, respectively (Figure S9 , Supporting Information). The stain stress was calculated using the following equation:\n \n (1) \n ε = t s 2 r \n where t \n s and r represent the thickness of the substrate (50 µm) and the radius of curvature of the device in its bent state, respectively. [ \n \n 49 \n \n ] Furthermore, the device maintained a high on–off ratio of up to 10 5 after 500 bending cycles at a strain of 1.0% (bending radius of 2.5 mm), demonstrating excellent flexibility (Figure S10 , Supporting Information). The device exhibited stable switching endurance, with no noticeable performance degradation, maintaining a high I \n ON/OFF ratio of 10 4 over 5 × 10 3 cycles under a bending radius of 2.5 mm (Figure 4c ). Retention tests further confirmed that both the HRS and LRS were sustained for up to 10 4 s, with only slight performance degradation, while still achieving a high I \n ON/OFF ratio of 10 5 (Figure. 4d ). Compared to other flexible memory devices reported in the literature, the high I \n ON/OFF ratio of 10 4 under a bending radius of 2.5 mm, along with stable endurance over 500 bending cycles, demonstrates exceptional reliability for flexible applications (Table S1 , Supporting Information). These results confirm that the device operates reliably under various bending conditions with robust switching endurance, highlighting its potential for flexible memory applications. Figure 4 a) Schematic of the flexible memristor with the Ag/Zr 6 ‐oxo/Au device on a polyimide substrate. b) I–V characteristics of F‐device 2 over 100 DC sweep cycles. c) Resistance distribution in the HRS and LRS of the flexible device over 5000 cycles. d) Memory retention characteristics of the flexible device under bending‐induced strain of 1.0% (bending radius of 2.5 mm). Biological synapses play a crucial role in the learning process by responding to various external stimuli within the nervous system. Synaptic weights represent the strength of the connection between two consecutive synaptic neurons and are updated in an analog manner based on learning rules. This occurs as stimuli from the pre‐neuron reach the synaptic cleft and are transmitted to the post‐neuron ( Figure \n 5 a ). Figure 5b illustrates a schematic of a neural network for MNIST pattern recognition, consisting of three neuron layers with 784 input neurons, 300 hidden neurons, and 10 output neurons. In neuromorphic computation, achieving high linearity and symmetry in the potentiation and depression of conductance modulation is essential for ensuring the desired inference accuracy. Figure 5 a) Schematic of a biological synapse illustrating the concept of synaptic modulation using a memristor‐based artificial synapse. b) Schematic of an MNIST‐data‐based neural network simulation system. c) Analog conductance modulations for three different device conditions under 0.8/−0.8 V, 200 ns pulse conditions. d) Ten repeatable sequential conductance modulations for device 2 performed under 0.8/−0.8 V, 200 ns pulse conditions. e) Analog conductance modulations for F‐device 2 under 0.6/−0.6 V, 200 ns pulse conditions. f) Accuracy of pattern recognition for device 2 after 30 000 iterations. g) Comparison of pattern recognition accuracy under different conditions after 30 000 iterations. h) Comparison of various memristor devices in terms of linearity, I \n ON/OFF ratio, and endurance. The red star represents the performance of the proposed device in this study, demonstrating superior linearity and endurance compared to the referenced devices. To investigate the synaptic characteristics of devices 1, 2, and 3, which are based on metal clustering ion‐migration dynamics, conductance modulation was performed using a train of 50 consecutive positive pulses with an amplitude of 0.8 V, followed by 50 consecutive negative pulses with an amplitude of −0.8 V, both with pulse widths of 200 ns (Figure 5c ). To evaluate linearity, we defined the linearity factor as follows:\n \n (2) \n Linearity factor = Δ G m a x + Δ G m i n 2 / G m a x − G m i n p u l s e n u m b e r \n where Δ G \n max and Δ G \n min are the maximum and minimum changes in conductance between adjacent levels, respectively. The calculated linearity values for potentiation and depression were 1.33 and 1.87, respectively, for device 1; 1.04 and 1.05, respectively, for device 2; and 1.20 and 1.63, respectively, for device 3. Device 2 exhibited the most linear and symmetric conductance updates under identical pulse conditions, confirming its exceptional linearity. In contrast, devices 1 and 3 showed deviations from the ideal value of 1 during both potentiation and depression, indicating nonlinear and asymmetrical conductance updates. Figure 5d shows the stable repetitive behavior of 10 sequential conductance modulations, demonstrating reliable and predictable conductance updates in response to a series of potentiation and depression pulses. Additionally, under the conditions of device 2 on a flexible substrate (F‐device 2), the calculated linearity values for potentiation and depression were 1.11 and 1.29, respectively, demonstrating high linearity comparable to those observed on a rigid substrate (Figure 5e ). To further evaluate the performance of the Ag/Zr 6 ‐oxo/Au device for neuromorphic computing applications, we conducted a neural network simulation for MNIST pattern recognition, as shown in Figure 5b . The conductance modulation performances were modeled using the following equation:\n \n (3) \n G n = A + B e − C n \n where the pulse index, n , is evaluated before the synaptic weight update for each epoch. The ideal parameters A, B, and C were extracted by curve‐fitting the experimentally obtained conductance modulation data shown in Figure 5c . The MNIST dataset, comprising images of handwritten digits with 28 × 28 grayscale pixels, included 60 000 training samples and 10 000 validation samples. Synaptic weights applied between each neuron layer were converted from the conductance values in Figure 5c and expressed as w = ( G − G \n mean ) / \n G \n diff , where G \n mean = [ G \n P (0) + G \n D (0)] / 2 and G \n diff = G \n D (0) + G \n P (0). A training session was conducted for 30 000 iterations. The maximum inference accuracies of devices 1 and 3 were 90.60% and 91.72%, respectively (Figure S11 , Supporting Information). In contrast, device 2 exhibited a superior accuracy of 97.44% (Figure 5f ). Moreover, when device 2 was tested on a flexible substrate, it achieved a remarkable maximum inference accuracy of 93.57%. The maximum inference accuracy of device 2, on both standard and flexible substrates, was clearly higher than that of devices 1 and 3 (Figure 5g ), indicating its excellent performance for high‐precision neuromorphic computing. The backpropagation algorithm assessed the weights by assuming a linear change in conductance, determining the number of programming pulses required to adjust the weights. This approach leverages high linearity and symmetry in the conductance modulation process to ensure high precision and minimal fluctuation. Compared to previously reported conductive bridge random‐access memory devices, [ \n \n 32 \n , \n 35 \n , \n 36 \n , \n 37 \n , \n 38 \n , \n 39 \n , \n 40 \n , \n 41 \n , \n 42 \n , \n 43 \n , \n 44 \n , \n 45 \n , \n 46 \n , \n 47 \n , \n 48 \n \n ] the Ag/Zr 6 ‐oxo/Au device shows significantly improved properties (Figure 5h ). These advancements highlight the effectiveness of the proposed metal clustering ion‐migration dynamics. Key parameters such as endurance, on/off ratio, and linearity—critical for evaluating the reliability of synaptic devices—have been extensively studied, underscoring the superior performance of our device in various aspects (Table S2 , Supporting Information)."
} | 7,996 |
30254839 | PMC6142729 | pmc | 359 | {
"abstract": "In many biological structures, optimized mechanical properties are obtained through complex structural organization involving multiple constituents, functional grading and hierarchical organization. In the case of biological surfaces, the possibility to modify the frictional and adhesive behaviour can also be achieved by exploiting a grading of the material properties. In this paper, we investigate this possibility by considering the frictional sliding of elastic surfaces in the presence of a spatial variation of the Young’s modulus and the local friction coefficients. Using finite-element simulations and a two-dimensional spring-block model, we investigate how graded material properties affect the macroscopic frictional behaviour, in particular, static friction values and the transition from static to dynamic friction. The results suggest that the graded material properties can be exploited to reduce static friction with respect to the corresponding non-graded material and to tune it to desired values, opening possibilities for the design of bio-inspired surfaces with tailor-made tribological properties.",
"conclusion": "Conclusion In this paper, we have considered the frictional sliding over a rigid substrate of an elastic material characterized by a grading of selected mechanical properties (Young’s modulus and local coefficients of friction), focusing on the effects on the global static friction and the detachment process at the onset of sliding. The system has been investigated by means of numerical simulations using 2D SBM and 3D FEM to verify the results and to exploit additional insights provided by the two different approaches, after having tuned the SBM parameters in order to have a precise match of the frictional force curve obtained by FEM, in terms of slope and static friction threshold. The results show that grading of the mechanical properties can reduce the global static friction with respect to a non-graded material, due to an anticipated detachment process. In the case of a graded distribution of local friction coefficients, detachment begins from the region where thresholds on static friction are smaller, while in the case of a graded Young’s modulus distribution, detachment begins from the regions where it is larger. In both cases, the 2D SBM predicts a linear decrease of the global coefficient of friction as a function of the relative grading variation and symmetry with respect to the sliding direction. In contrast, FEM simulations display an anisotropic behaviour due to the effect of the vertical stress distribution, which can either enhance or counterbalance the effect of the grading. Thus, a greater reduction of static friction can be expected when the grading on local friction decreases along the sliding direction, or when a grading of Young’s modulus increases along the sliding direction. The effect on the global dynamic coefficient of friction, instead, appears to be underestimated numerically, and should be the object of further investigations. These results are not valid when the relative grading variation is greater than 30% with respect to the average. In this case, the time evolution of the tangential force changes radically. The time duration of the detachment phase increases due to the large variation between edges so that the force peak in the transition from static to dynamic friction can disappear completely. This interplay of effects produces a nonlinear reduction of static friction with the grading. A quasi-linear decrease can be obtained in the case of a triangular grading, which is symmetric with respect to the two edges, so that the anisotropy of the vertical stress distribution is less influential. In particular, this outcome can be achieved by setting this type of grading along the orthogonal direction with respect to the sliding direction. We have thus found that the SBM can capture the main effects of gradings on the static friction coefficient and describe the detachment process at the interface, with a much smaller computational cost than that required by FEM simulations. Therefore, it can be adopted for a rapid, initial estimation of static friction values. These results suggest that it is possible to realize bio-inspired materials with a gradient in the mechanical properties, imitating the graded Young’s moduli found in nature, or in the local frictional properties, e.g., by controlling the roughness or the microstructure, for the design of advanced sliding interfaces. A reduction in the static friction up to almost 30%, with respect to the corresponding non-graded material, can thus be achieved.",
"introduction": "Introduction Materials with a gradient in their physical or elastic properties are widely found in nature. Several known biological systems have developed specialized functionalities due to stiffness, density or composition gradients. Beetles, for instance, display setae with a graded stiffness that optimises the adhesive performance on rough surfaces [ 1 ]. Hardness and stiffness gradients are of fundamental importance in the biomechanics of contacts, since they allow increased resistance against wear, impact, penetration and crack propagation [ 2 – 7 ]. Bio-inspired solutions have thus been proposed for the design of advanced materials that mimic the hierarchical and graded structures found in nature, for use in engineering applications [ 8 – 9 ]. Functionally graded materials (FGMs) display a gradient in their elastic properties along one or more directions and have recently acquired great interest in technology [ 10 ]. Several authors have studied standard solid mechanics problems considering FGMs, for example, in the case of various loading conditions [ 11 – 13 ], and in problems involving fracture [ 14 – 17 ] or fatigue [ 18 ]. Recently, FGMs have also been applied to tribological studies, where it is well known that the behaviour of a system is governed by multiphysics and multiscale interactions [ 19 ]. The first application of graded materials to contact mechanics was proposed by Giannakopoulos and Suresh, who presented an analytical study of the indentation of materials with an exponential or power law variation of the Young’s modulus through the depth [ 20 – 21 ]. Giannakopoulos and Pallot then extended the analysis to 2D [ 22 ]. Graded substrates have also been considered in elastohydrodynamic lubrication problems [ 23 ]. More recently, the method of dimensionality reduction [ 24 – 25 ] has been extended to the axisymmetric frictionless contact of elastically graded materials [ 26 ], and solutions are also provided in the presence of adhesion [ 27 ]. In all these cases, the elastic gradients are considered with respect to the depth, with an exponential or a power law variation of the Young’s modulus, i.e., E ( z ) = E 1 e α z or E ( z ) = E 2 z β , respectively, where z is the depth coordinate and E 1 , E 2 , α and β are constants. The first extension to a lateral elastic gradient, to the best of our knowledge, was by Dag et al . who studied the problem both analytically, by reducing the equation describing the contact of a rigid flat punch to a singular integral equation, and numerically, through the finite-element method [ 28 – 29 ]. In this paper, we extend the previous work on 1D composite surfaces [ 30 ] to 2D geometries to show how it is possible to tune the macroscopic tribological properties through local variations of material and surface properties, i.e., Young’s moduli and friction coefficients, reducing static friction compared to the non-graded case. The results also allow the predictions of a discrete approach like the spring-block model [ 31 – 32 ] to be compared to those derived by explicit finite-element simulations. This provides useful insights to understand the frictional properties of graded materials, with the aim of designing smart tribo-materials and innovative solutions for sliding interfaces.\n\nIntroduction In this work, we investigate the effect of surface or material property gradients on the global coefficient of friction. The system taken into consideration is composed of an elastic plate, with a square base of side L and height H << L , which is driven from the top surface at constant velocity over a rigid substrate and subjected to friction. We study this system by means of two numerical methods: a 2D spring-block model (SBM) and 3D finite-element method (FEM) simulations. The two methods are complementary in many aspects, so that by using both it is possible to cross-check the results and obtain interesting insights from different approaches. The SBM is a two-dimensional approximation of the real system, so that effects due to the thickness of the layer are neglected. Specifically, any effect due to the vertical stress distribution cannot be captured. While these can be minimized in the case H << L , it is still useful to compare the results with FEM simulations, which can model this thin layer while maintaining a 3D approach. As we will show later, the comparison between the two methods will allow some concurrent effects to be disentangled that govern the global frictional behaviour. On the other hand, different formulations of SBM have been used in many recent studies to describe aspects of the transition from static to dynamic friction, the nucleation of rupture wave fronts, and the effects of patterning [ 32 – 36 ]. The SBM method is usually computationally faster than FEM, thus it is more practical for a qualitative understanding of these phenomena, but also includes approximations that must be verified to check whether all effects are correctly described. Thus, in each section, we will consider the two models with the same setup, that is, by choosing the closest conditions and parameter sets for the two approaches, and we will describe the effects predicted by them in the presence of graded materials.",
"discussion": "Results and Discussion Gradient in the local coefficients of friction We first consider a gradient in the local friction coefficients. In real systems, this can be realised in two ways. First, the surface can be polished in a spatially variable manner or using different processes in order to have a varying roughness and thus varying local friction coefficients. Secondly, a surface with a gradient in the frictional properties can be obtained by appropriately fabricating and arranging microscopic structures of variable geometries or sizes, giving rise to variable local friction coefficients [ 40 – 41 ]. In order to compare the results, we report the variations of the global static coefficient of friction as function of a grading distribution Δ in the local coefficient of friction, with respect to the value of the non-graded surface, using both SBM and FEM. The variation is computed as , where µ s,0 corresponds to the case Δ = 0. The absolute values of µ s are reported in the Supporting Information File 2 . The results are shown in Figure 4 for both SBM and FEM simulations. In general, in the presence of a gradient, the global static coefficient of friction of the surfaces in contact decreases with respect to the non-graded surface, although the mean values of the local friction coefficients are the same. Figure 4 Effect of a gradient in the local coefficients of friction on the global static coefficient of friction, expressed as percentage variation as a function of Δ. The shaded blocks schematically represent the value of the local coefficients of friction, which are higher for a darker shading. The reason for this lies in the progressive detachment of the contact surfaces, always starting from the side where the critical value of the local shear stress is reached (i.e., the static friction threshold). The first detachment of the sliding surface produces a detachment avalanche propagating towards the region with higher static friction threshold, as shown in Figure 5 (see also Supporting Information File 3 ). Analogous effects on the propagation of detachment fronts have also been studied experimentally [ 42 ]. Consequently, an increasing absolute value of Δ reduces the global static coefficient of friction with respect to the non-graded surface, up to an asymptotic value corresponding to the dynamic friction coefficient value. Thus, the gradient can completely remove the force peak observed at the transition from static to dynamic friction (see Figure 2 ). This is schematically shown in Figure 6 , where the time evolution of the friction force is reported for two different values of Δ. An additional effect is the deviation from linearity when approaching the static friction threshold, observed for the highest value of the gradient (i.e., Δ = 0.4) and similarly highlighted by both the SBM and the FEM simulations. Figure 5 Propagation of the detachment front at the static friction threshold (left) and immediately after (right), in units of the dimensionless longitudinal stress σ x / p , for a surface with a gradient in the local coefficients of friction, computed using the SBM method for the case Δ = 0.1. The irregularities of the detachment front are due to the statistical dispersion of the local coefficients of friction introduced in the SBM formulation. Figure 6 Dimensionless friction force as function of time for two values of Δ, calculated using the SBM and FEM methods. In both cases a larger value of Δ leads to the reduction or elimination of the initial force peak. Unlike the SBM simulations, which give symmetric results with respect to the case Δ = 0 since they are insensitive to the vertical stress distribution, FEM simulations display an anisotropic behaviour when considering a positive or a negative gradient. This is equivalent to considering the same sign of the gradient but with an opposite sliding direction. This result can be attributed to the vertical stress distribution at the contact interface: when friction is present, the normal pressure is reduced at the leading edge of the sliding plate and increased at the trailing edge [ 28 , 43 – 44 ]. Since the static friction thresholds depend not only on the local µ s but also on the local value of the normal pressure, due to this effect, a gradient of detachment threshold already exists. This must be added to the gradient of the local coefficients of friction. As an example, for Δ > 0, the static friction thresholds are increased at the leading edge of the surface, so that the effect of the vertical stress acts as a counterbalance, and the effective gradient is smaller than Δ. Conversely, for Δ < 0, the vertical stress accumulates with the gradient. For this reason, with the same absolute value but different sign of Δ, we can expect a different behaviour; in particular, that for a positive Δ value, the global static friction is greater than for the case of a negative Δ value. From the results of the FEM simulations, this is reproduced correctly at least for Δ < 0.3, as can be seen in Figure 4 . For higher values of the gradient, the results are opposite due to the large difference of friction between the edges. For Δ < 0 at the leading edge of the surface, static friction is already weak so that the effect of normal pressure reduction is less influential, while at the trailing edge, static friction is large due to the combination of a large local friction coefficient and increased pressure. The result is that the detachment process is inhibited with respect to the same positive Δ value. The effect of larger values of Δ on the detachment process is also investigated through the SBM method. As previously explained, the detachment front nucleates from the edge where the weakest thresholds are, and the maximum of the friction force during the time evolution occurs shortly after the detachment begins. However, when the gradient increases, the time necessary for the detachment front to propagate across the surface increases (see Figure 6 and Supporting Information File 4 ). For higher values of Δ, the contribution to the total friction force from the region with higher thresholds is more influential, so that the maximum of the friction force occurs later during the detachment process and not shortly after its beginning. Thus, while SBM cannot capture anisotropic behaviour emerging from 3D deformation effects occurring in the materials in contact, it is still useful to disentangle the effects of the gradient and the vertical stress distribution. The global dynamic friction coefficient µ k does not display any appreciable variation when calculated with FEM or SBM if compared to the flat surface, i.e., its variation is limited to within 1% as shown in Figure 7 . Again, the FEM results are anisotropic with respect to Δ, for the reasons explained above. However, the effect of the gradient on the dynamic friction cannot be fully captured by a formulation only based on an Amontons–Coulomb friction law, as in the case of the SBM. Therefore, a good match between the two methods cannot be achieved here and further investigations are needed. Figure 7 Effect of a gradient in the local coefficients of friction on the global dynamic coefficient of friction, expressed as percentage variation as a function of Δ. The shaded blocks schematically represent the value of the local coefficients of friction, which are greater for darker shading. We have also investigated the effect of changing the sliding direction with respect to the direction of the gradient, as shown in Figure 8 . Both the SBM and the FEM predict a greater global static coefficient of friction when switching from the 0° to the 90° direction, and this is evident especially for large values of Δ. The dependence of µ s on the angle, instead, is more complex for Δ < 0.3, especially for the 3D FEM, where the interaction with the vertical stress distribution must also be taken into account, as discussed previously. Figure 8 Effect of a gradient in the local coefficients of friction on the global static coefficient of friction, expressed as percentage variation as a function of Δ and for three sliding directions. The shaded square schematically represents the value of the local coefficients of friction, which are greater for a darker shading, and the considered sliding directions. Gradient in the Young’s modulus of the material The effect of a gradient in the Young’s modulus is qualitatively similar to that of the graded coefficient of friction discussed above. As can be seen in Figure 9 , the global µ s for the graded material is smaller than that for the case Δ = 0. However, while in the previous case, the reason for the modification of the global friction coefficient can be found in a smaller static friction threshold, in this case, a given lateral strain produces a corresponding tangential force that is greater on the side of the material with greater local Young’s modulus. Therefore, in this case, the static friction threshold is reached first where the Young’s modulus is greater. The detachment of the contact surfaces starts from this side and proceeds towards the region with smaller E , with a propagation similar to that already shown in Figure 5 , but in the opposite direction. Thus, the material portion of the sliding plate is in tension for Δ > 0 and in compression for Δ < 0. Qualitatively speaking, a positive gradient in the Young’s modulus is equivalent to a negative gradient in the local coefficients of friction. Figure 9 Effect of a gradient in the Young’s modulus on the global static coefficient of friction, expressed as variation as function of Δ. The shaded blocks schematically represent the value of the local Young’s modulus, which is greater for darker shading. Again, FEM simulations produce an anisotropic result with respect to positive and negative gradients, for the reasons discussed above. The redistribution of normal stresses is again related to the static friction threshold: when Δ > 0, the tangential force is greater at the leading edge of the slider, where E is higher and the thresholds are reduced due to smaller p values, so that the effective gradient is larger than Δ. Conversely, for Δ < 0 the gradient of the Young’s modulus is counterbalanced by the effect due to the vertical stress. Thus, for small |Δ| values, a greater global µ s is observed for Δ < 0, while for larger |Δ| values, this trend is inverted due to the mechanism described previously. It is remarkable that for Δ < 0, the FEM and SBM simulations predict a very similar behaviour, suggesting that, in this case, two opposite effects are almost cancelled, so that the 2D SBM results provide a good approximation of real values. Although in the previous case, the interplay of effects produced a non-trivial behaviour by varying the gradient, in this case, for Δ < 0, the agreement between FEM and SBM simulations suggests that the reduction of the global static friction with the gradient is approximately linear. The interplay of effects between the grading and the vertical stress distribution, which are both asymmetric with respect to the sliding direction, causes a non-trivial behaviour of the static friction as a function Δ. The effect of the vertical stress distribution can be reduced by designing a triangular grading, according to Equation 5 , so that for Δ > 0 the detachment begins at the centre of the surface and propagates towards the edges, and vice versa for Δ < 0. As in the previous case, no appreciable variation in the dynamic coefficient of friction is predicted, as shown in Figure 10 . One difference is that both the SBM and the FEM simulations predict a higher global µ k with respect to the case of non-graded materials. Again, the FEM results are slightly anisotropic as a function of Δ and, as in Figure 7 , a smaller global dynamic coefficient of friction is obtained for Δ < 0. Figure 10 Effect of a gradient in the Young’s modulus on the global dynamic coefficient of friction, expressed as percentage variation as a function of Δ. The shaded blocks represent schematically the value of the Young’s modulus, which is greater for a darker shading. The results presented in Figure 11 show the effect of a triangular gradient in the Young’s modulus on the global static coefficient of friction calculated via SBM and FEM. The SBM results display a symmetric behaviour. The corresponding detachment process is shown in Supporting Information File 5 and Supporting Information File 6 for the case Δ > 0 and Δ < 0, respectively (see also Figure 12 for an example). The FEM simulations predict a smaller µ s in the case of Δ < 0 because the effect due to the grading is superimposed on the effect of the vertical stress, so that the static friction thresholds are exceeded earlier compared to the case Δ > 0. However, in both cases, the static friction decreases linearly with Δ. When considering the 90° sliding direction, i.e., orthogonal to the grading, the results of the SBM and the FEM simulations are in good agreement. In this case, the effects due to the vertical stress are ininfluential, since the detachment process is symmetric with respect to the sliding direction. This suggests that with a proper combination of grading and sliding direction, it is possible to obtain a linear reduction of the static friction with the grading level, which would allow the global static friction of a surface to be conveniently tuned to a chosen value, reduced with respect to the corresponding non-graded surface. Figure 11 Effect of a triangular gradient in the Young’s modulus on the global static coefficient of friction, expressed as variation as function of Δ and for two sliding directions (0°, i.e., parallel, and 90°, i.e., perpendicular to the gradient). The shaded blocks represent schematically the value of the local Young’s modulus, which is higher for darker shading. Figure 12 Propagation of the detachment front at the static friction threshold (left) and immediately after (right), in units of the dimensionless longitudinal stress σ x / p , for a material with a triangular gradient in the Young’s modulus, computed using the SBM method for the case Δ = 0.2. The irregularities of the detachment front are due to the statistical dispersion of the local coefficients of friction introduced in the SBM formulation."
} | 6,064 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.