pmid stringlengths 8 8 | pmcid stringlengths 8 11 ⌀ | source stringclasses 2
values | rank int64 1 9.78k | sections unknown | tokens int64 3 46.7k |
|---|---|---|---|---|---|
40301674 | PMC12041382 | pmc | 118 | {
"abstract": "Coral-associated microorganisms provide crucial nutritional, protective, and developmental benefits, yet many functional traits remain unexplored. Phototrophic bacteria may enhance coral nutrition and reduce oxidative stress during bleaching via photosynthesis and antioxidant production. Despite this potential, their role in the holobiont’s energy budget and heat stress resilience is understudied. This review explores the functional traits and potential of phototrophic bacteria to enhance coral health and resilience under environmental stress.",
"conclusion": "Conclusion Phototrophic bacteria exhibit functional traits that could be beneficial and play a crucial role to enhance coral survival under stressful conditions. Their active phototrophic capacity suggests an underexplored potential contribution to coral nutrition or competition advantages against coral pathogens. Despite such potential, the extent of the contribution of phototrophic bacteria to the coral diet and overall health remains unclear, indicating a clear opportunity for studies incorporating experimental and isotopic analyses that can provide a more comprehensive understanding of their role. Furthermore, the significant production of anti-oxidant compounds (e.g., carotenoids) by phototrophic bacteria could also contribute to minimizing the concentration of excessive reactive oxygen species (ROS) within the coral holobiont during heat stress. These functional traits highlight the potential role of phototrophic bacteria as coral probiotics and encourage further studies targeting the specific contributions of phototrophic bacteria to the health of the coral holobiont and their underlying mechanisms."
} | 417 |
33996010 | PMC8098691 | pmc | 120 | {
"abstract": "There has been great interest in the fabrication of solid surfaces with desirable under-liquid wettability, and especially under-liquid dual-lyophobicity, because of their potential for widespread use. However, there remains the lack of a general principle to modulate the under-liquid wettability in terms of surface energy (SE). Herein, we found that the relative proportion between the polar and dispersive components in SE that reflects the competition between hydrophilicity and lipophilicity governs the under-liquid wettability of the solid surface. For the first time, we introduced hydrophilic–lipophilic balance (HLB) calculated solely based on the amount and type of hydrophilic and lipophilic fragments in surface molecules to rapidly predict the under-liquid wettability of a solid surface, thereby guiding the fabrication of solid surfaces with desirable under-liquid wettability. Accordingly, the under-liquid dual superlyophobic surfaces in a nonpolar oil–water-solid system were fabricated by grafting molecules with appropriate HLB values ( e.g. , 6.341–7.673 in a cyclohexane–water–solid system) onto porous nanofibrous membranes, which were able to achieve continuous separation of oil–water mixtures. This work provides reasonable guidance for the fabrication of solid surfaces with targeted under-liquid wettability, which may lead to advanced applications in oil–water–solid systems.",
"conclusion": "3. Conclusions In summary, 14 different surface molecules with different SE components were grafted onto silicon wafers via a covalent modification process, giving rise to modulated under-liquid wettabilities such as under-water lipophilicity/under-oil hydrophobicity, under-liquid dual-lyophobicity, and under-water lipophobicity/under-oil hydrophilicity. We found that the ratio of PSE to DSE that reflected the competitive relationship between hydrophilicity and lipophilicity of a solid surface was an appropriate parameter to describe under-liquid wettability. For instance, under-liquid dual-lyophobicity could be realized when PSE–PSE and DSE–DSE interfacial interactions maintained a relative balance. We further introduced an HLB-based criterion for rapidly predicting the under-liquid wettability of a solid surface exclusively based on the type and amount of hydrophilic fragments and hydrophobic fragments in the surface molecule. We found that surfaces with lower HLB values tended to be more lipophilic. In contrast, surfaces with higher HLB values led to more affinity with water. When the HLB value was located in a suitable range ( e.g. , 6.341–7.673 in cyclohexane–water-solid system), the under-liquid dual-lyophobicity of a solid surface was achieved. The under-liquid dual superlyophobic surfaces were successfully fabricated by affording proper HLB values onto the electrospun porous nanofibrous membranes, which were able to efficiently separate layered cyclohexane-water mixtures, as well as CTAB-stabilized cyclohexane-in-water and SDBS-stabilized water-in-cyclohexane emulsions. This work provides straightforward guidance for the fabrication of solid surfaces with desirable under-liquid wettability simply based on the components of surface molecules, which may provide new perspectives for applications in oil–water–solid systems, such as liquid separation, liquid–liquid interface assembly, heterogeneous catalysis, controlled bioadhesion, and anti-biofouling.",
"introduction": "1. Introduction Inspired by the unique wetting phenomena found in nature, 1–3 various lyophobic or lyophilic solid surfaces have been developed for the applications of liquid repellence, 4–7 transportation 8,9 or separation, 10–14 anti-fogging, 15 anti-biofouling, 16,17 catalysis, 18,19 and heat transfer. 20,21 In particular, there has been growing interest in the rational modulation of under-liquid wettability of solid surfaces, and especially the achievement of under-liquid dual-lyophobicity, to meet the requirements of environment- and energy-related applications. 22–26 Thus far, there have only been a few studies on wetting mechanisms to guide the modulation of under-liquid wettability. Tian et al. proposed that in an oil–water–solid system, the sum of the water contact angle in oil ( θ w/o ) and the oil contact angle in water ( θ o/w ) on a thermodynamically stable surface should be 180° in principle because the angles were supplementary to each other. Therefore, the under-liquid dual-lyophobic surface should be metastable in thermodynamics because both θ w/o and θ o/w on it were more than 90°. 27 The author also noted that the re-entrant geometric characteristic and the appropriate surface chemical composition were the key factors for realizing under-liquid dual superlyophobicity. Our previous work demonstrated the rational modulation of the under-liquid wettability of rough surfaces by changing the surface chemical composition. We further ascertained the thermodynamic metastability of the under-liquid dual-lyophobic surfaces by calculating the total interfacial energy at the solid–liquid interface. Consequently, in the nonpolar oil–water–solid system, the under-liquid wettability of the rough surfaces could be inferred according to their intrinsic water contact angle ( θ w ). 28 Chen's group demonstrated the restructuring behavior of specific surface molecules in different media ( e.g. , air, water, and oil), that is, the surface molecules would reorient and selectively expose their hydrophilic or lipophilic (oleophilic) parts to enhance the solid–liquid interaction and decrease the total interfacial energy, thereby leading to different wettabilities. 29–31 However, there remains a lack of a general criterion to address the surface composition influence, which allows rapid and qualitative prediction regarding the under-liquid wettabilities solely based on the components of surface molecules. In terms of the classical Young's equation, surface energies (SEs) governed by surface chemical compositions are commonly used to demonstrate the interfacial interaction between liquids and solid surfaces. 32 However, it was found that the further division of SE into the dispersive component (DSE) and polar component (PSE) based on the types of intermolecular forces was more suitable for evaluating interfacial adhesion and explaining some unique wetting phenomena. 33–37 For instance, our previous work demonstrated that the interaction between solid surfaces with high PSE and high-PSE liquids was more facile than that with high-SE liquids, forming a robust solid–liquid composite interface that could prevent the intrusion of immiscible low-PSE liquids. 10 Therefore, it is expected that the under-liquid wettabilities of solid surfaces could be rationally modulated by adjusting the relative proportion between PSE and DSE. In this work, we discovered that in the nonpolar oil–water–solid systems, the ratio of PSE to DSE, denoted as f , was an appropriate factor to describe the under-liquid wettability of solid surfaces: a lower f value tended denote under-water lipophilicity, whereas a higher f value led to under-oil hydrophilicity. Significantly, when the f value was located in a suitable range, the PSE–PSE and DSE–DSE interfacial interactions maintained their relative balance, resulting in under-liquid dual-lyophobicity. Here, for the first time, we introduce the concept of hydrophilic–lipophilic balance (HLB) to predict the under-liquid wettability of a given surface and guide the fabrication of a solid surface with desirable under-liquid wettability. The HLB value was calculated based on the type and the amount of hydrophilic and lipophilic fragments in the surface molecules ( Fig. 1a ). 38,39 Fig. 1 Schematic illustration of the molecular HLB criterion for rapidly predicting the under-liquid wettability of a solid surface based on the components of surface molecules as well as guiding the fabrication of the solid surface with desirable under-liquid wettability. (a) Solid surface molecule composed of hydrophilic and lipophilic fragments. (b–d) Under-liquid wettability of solid surfaces: (b) under-water lipophilicity/under-oil hydrophobicity, (c) under-liquid dual-lyophobicity, and (d) under-water lipophobicity/under-oil hydrophilicity. We found that in the nonpolar oil–water–solid system, when the HLB value increased, the solid surface was more hydrophilic, and the under-liquid wettability changed in turn, including under-water lipophilicity (oleophilicity)/under-oil hydrophobicity ( Fig. 1b ), under-liquid dual-lyophobicity ( Fig. 1c ), and under-water lipophobicity (oleophobicity)/under-oil hydrophilicity ( Fig. 1d ). Furthermore, under-liquid superlyophobic surfaces were successfully fabricated by grafting surface molecules with appropriate HLB values onto rough substrates via a simple chemical modification process. Among them, the representative cyanopropyl-terminated porous nanofibrous membrane efficiently separated layered oil–water mixtures and surfactant-stabilized emulsions.",
"discussion": "2. Results and discussion 2.1 Under-liquid wettabilities of solid surfaces with different components of SEs The surface wettability is determined by the surface topography and the chemical composition. To eliminate the effect of surface geometry on the under-liquid wettability of solid surfaces, smooth silica wafers were used as the substrates on which 14 different molecules were grafted via a covalent modification or plasma process (Table S1 † ). The total SEs of these surfaces, as well as the polar (PSE) and dispersive (DSE) components in the SEs, were estimated based on the contact angle (CA) data and fitted using the OWRK (Owen, Wendt, Rabel, and Kaelble) method (Table S2 and Note S1 † ). 36,40 Fig. S1 † shows four representative fitted curves of glycidyloxypropyl-, methacrylate-, and phenyl-terminated surfaces, and a polydopamine-coated surface. The well-matched fitting curves indicate that the OWRK method is reasonable and adequate to evaluate SE components, and the results are listed in Table S3. † The under-liquid wettabilities of the surfaces with different SEs were investigated in the nonpolar oil–water–solid systems. Taking the cyclohexane–water–solid system as an example, we measured the oil contact angle ( θ o/w ) and water contact angle ( θ w/o ) of these surfaces when they were immersed in water and oil, respectively (Table S4 † ). Fig. 2a shows that the under-water lipophilic surfaces (red dots) and under-liquid dual-lyophobic surfaces (yellow dots), especially for the glycidyloxypropyl- and iodopropyl-terminated surfaces, are overlapped with the increase in SE, suggesting that the under-liquid wettabilities of the solid surfaces cannot be accurately described using the SE as the parameter. Fig. 2 Under-liquid wettabilities of solid surfaces with different surface molecules. (a) Relationship between under-liquid wettabilities of solid surfaces and their total SEs. No intact region corresponding to the under-liquid dual-lyophobicity is observed. (b) Relationship between under-liquid wettabilities of solid surfaces and their f values. The under-liquid wettability of solid surfaces can be divided into three separate regions based on their f values. (c) Relationship between under-liquid wettabilities of solid surfaces and their HLB values. The under-liquid wettability of solid surfaces can also be classified into three independent regions according to their HLB values. The shadows in (b) and (c) are attributed to the lack of suitable modulations of surfaces with f values in the ranges of 0.254–0.390 and 1.037–1.410, and HLB values in the ranges of 5.346–6.341 and 7.673–7.770, respectively. Note: ■: θ o/w , ♦: θ w/o , red: under-water lipophilicity and under-oil hydrophobicity, yellow: under-liquid dual-lyophobicity, blue: under-water lipophobicity and under-oil hydrophilicity. It is known that the polar–polar and dispersive–dispersive interfacial attractions at the solid–liquid interface can be treated independently, which leads to the hydrophilicity and lipophilicity of solid surfaces, respectively. Herein, the ratio of PSE to DSE of the material surface, denoted as f , was employed as a new parameter to demonstrate the competitive affinity interaction of the solid surface between water and oil. In Fig. 2b , the under-liquid wettabilities of the surfaces can be successfully classified into three separate regions according to their f values: (1) f ≤ 0.254 (red dots), and the DSE-dominated SE affords the surfaces more affinity to oil, showing under-water lipophilicity/under-oil hydrophobicity; (2) 0.390 ≤ f ≤ 1.037 (yellow dots), and moderate affinity of the surfaces to both water and oil leads to under-liquid dual-lyophobicity; (3) f ≥ 1.410 (blue dots), and the higher PSE content in SE results in under-water lipophobicity/under-oil hydrophilicity. Therefore, the under-liquid wettability of solid surfaces can be described more accurately using the f value rather than by the total SE. There are two black shadows in Fig. 2b because of the lack of modified surfaces with f values in the ranges of 0.254–0.390 and 1.037–1.410. We then explored the effect of the f value on the under-liquid wettability in terms of the solid–liquid interface interaction. In the OWRK method, the interface interaction, called adhesion work ( W a ), can be expressed as the following equation: 36,40 1 where W d a and W p a denote the adhesion work generated by the DSE–DSE and PSE–PSE interfacial attraction, respectively. γ d S , γ p S , and γ d L , γ p L represent the DSE and PSE of a solid surface and the liquid, respectively. As shown in Note S2, † the competitive affinity to the solid surface between water and nonpolar oil can be expressed as the ratio of W a at the water–solid interface ( W aSW ) to that at the oil–solid interface ( W aSO ): 2 where f S denotes the f value of the solid surface. Therefore, the ratio of W aSW to W aSO is proportional to the square root of the f S value. Specifically, solid surfaces with higher f S values exhibit stronger interfacial affinity to water; conversely, lower f S values denote a stronger interfacial affinity to oil. This result further verifies that the ratio of PSE to DSE can accurately reflect the competitive relationship between hydrophilicity and lipophilicity of a solid surface, and determines the under-liquid wettability in a nonpolar oil–water–solid system. 2.2 The relationship between the hydrophile–lipophile balance (HLB) values and the under-liquid wettabilities To rapidly and conveniently predict the under-liquid wettabilities, we introduced the concept of hydrophilic–lipophilic balance (HLB) based on the components of surface molecules, which is widely used to evaluate the emulsifying and solubilizing properties of surfactants. 39 Herein, the modified molecules outside the silicon atoms that govern the surface chemical compositions of silica wafers were divided into hydrophilic fragments and the hydrophobic fragments, which were chosen for calculating the HLB values. The equation is expressed as: 3 where the HLB group number of the surface molecules is calculated by the following equation: 41,42 4 HLB group number = −0.337 × 10 5 × V x + 1.5 n where V x denotes the atomic volume data (Table S5 † ), and n relates to the number of water molecules participating in the solvation of different types of surface fragments. The calculated group numbers of lipophilic fragments ( e.g. , –CF 3 , \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 CH–, and –CH<) and hydrophilic fragments ( e.g. , –NHCONH 2 , –C \n \n\n<svg xmlns=\"http://www.w3.org/2000/svg\" version=\"1.0\" width=\"23.636364pt\" height=\"16.000000pt\" viewBox=\"0 0 23.636364 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(0.015909,-0.015909)\" fill=\"currentColor\" stroke=\"none\"><path d=\"M80 600 l0 -40 600 0 600 0 0 40 0 40 -600 0 -600 0 0 -40z M80 440 l0 -40 600 0 600 0 0 40 0 40 -600 0 -600 0 0 -40z M80 280 l0 -40 600 0 600 0 0 40 0 40 -600 0 -600 0 0 -40z\"/></g></svg>\n\n N, and –N C O) are listed in Tables S6 and S7, † which are generally less and greater than zero, respectively. 41,42 Accordingly, the calculated HLB values of these surfaces are given in Table S8. † When combining the HLB values of these surfaces with their under-liquid wettabilities (Table S4 † ), the θ o/w increases with the increase in the HLB value (square in Fig. 2c ) in the cyclohexane–water–solid system; in contrast, the θ w/o decreases (rhombus in Fig. 2c ). Therefore, we propose an HLB-based criterion for predicting the under-liquid wettabilities of solid surfaces: (1) HLB ≤ 5.346 (red dots), and the presence of abundant lipophilic fragments results in under-water lipophilicity/under-oil hydrophobicity; (2) 6.341 ≤ HLB ≤ 7.673 (yellow dots), and the similar affinity of the surface to oil and water leads to under-liquid dual-lyophobicity; (3) HLB ≥ 7.770 (blue dots), and the solid surfaces possess more hydrophilic fragments, thereby demonstrating under-water lipophobicity/under-oil hydrophilicity. The black shadows in Fig. 2c appear due to the lack of modified surface molecules with HLB values in the ranges of 5.346–6.341 and 7.673–7.770, respectively. The applicability of the HLB-based criterion to the prediction of the under-liquid wettability in other nonpolar oil–water–solid systems was also tested. Fig. S2 † provides the θ w/o and θ o/w of these modified surfaces in the hexadecane–water–solid system. The under-liquid wettability of surfaces can also be divided into three independent regions, and the boundary is the same as that in the cyclohexane–water–solid system, which is caused by the similar PSE and DSE components of hexadecane to that of cyclohexane. In addition, the HLB values of some reported smooth solid surfaces with known chemical compositions ( e.g. , hydroxyl-, cyanopropyl-, perfluorooctyl-, perfluorodecyl-, octadecyl-, and SU8-terminated surfaces, as well as polydopamine-coated surfaces) were calculated. 24,27 In Table S9, † all the calculated results are in good agreement with the experimental results, indicating that the HLB-based criterion can reasonably predict the under-liquid wettability of a given surface. Unfortunately, the HLB theory might be unsuitable for calculating the values of surface molecules containing ionic groups because the n values are defined as 9 and 6 for anionic and cationic groups, respectively, leading to extremely high HLB values. 42 This indicates that such surfaces possess an ultra-high affinity for water, thus demonstrating under-water lipophobicity/under-oil hydrophilicity. Note that the surface grafted by the uncompensated benzenesulfonate (BS − ) group exhibits under-liquid dual-lyophobicity, which goes against the HLB-based criterion. 31 Therefore, in this work, the under-liquid wettability of a solid surface containing only non-ionic surface molecules was discussed. 2.3 Fabrication of under-liquid dual superlyophobic surfaces According to the Wenzel model, the lyophilicity and lyophobicity of the surfaces can be greatly increased via roughening the lyophilic and lyophobic materials, respectively. 43 Herein, two types of rough surfaces with different geometries, (i) a vertical silicon nanowire array (SiNW, Fig. 3a and b ) and (ii) a SiO 2 –TiO 2 porous nanofibrous membrane (STPNM, Fig. 3c and d ), were selected as the substrates on which molecules with HLB values in the under-liquid dual lyophobic range (6.341–7.673, e.g. , cyanoethyl-, mercaptopropyl-, and aminopropyl-terminated molecules) were modified to achieve under-liquid dual-superlyophobicity. In Table S10, † both θ o/w and θ w/o on these rough surfaces are much larger than those on the smooth surfaces (Table S4 † ), and almost all of them exhibit under-liquid dual superlyophobicity. Exceptionally, the smaller θ w/o on the cyanoethyl-terminated SiNW (129.6 ± 5.8°) may be caused by the vertical microstructure of the SiNW that is not conducive to maintaining the solid–liquid interface that will repel the immiscible liquid. 27 By contrast, the θ w/o on cyanoethyl-STPNM reaches 151.2 ± 1.5°, suggesting that the network structure composed of the randomly stacked fibers is more beneficial for holding the infused liquid. This result is analogous to the effect of some unique surface topographies, such as re-entrant geometry, on the ability of the surfaces to repel both water and oil in the oil–water–solid or air–liquid–solid system. 6,27 Fig. 3 Morphologies of two rough substrates and the separation capacity of the CSTPNM. (a and b) SEM images of SiNWs showing top and side views, respectively, of the randomly arranged vertical silicon nanowires. (c and d) SEM and TEM images of STPNM, respectively, demonstrating the entangled fibers and hierarchical porous structure. (e and f) Profiles of the under-water oil droplet and under-oil water droplet on the CSTPNM, respectively, indicating the under-liquid dual superlyophobicity. (g) Demonstration of continuous oil–water separation. Two CSTPNMs prewetted with oil (cyclohexane, red) and water (colorless) were fixed onto two outlets of a T-shaped dual-channel apparatus. Cooperating with the automatic feeding by a peristaltic pump, cyclohexane/water mixtures could be continuously separated. (h) The separation efficiency of CSTPNM measured during the 4 h separation process. The error bars representing the s.d. were obtained from the test results of at least five replicates. 2.4 The separation capacity of the under-liquid dual superlyophobic membrane To evaluate the oil–water separation capacity of the under-liquid dual superlyophobic membrane, the cyanopropyl-STPNM (CSTPNM) with θ o/w of 159.4 ± 2.7° and θ w/o of 157.4 ± 2.6° ( Fig. 3e and f ) was used as an example to continuously separate a layered cyclohexane/water mixture. The CSTPNMs were fixed on two outlets of a T-shaped dual-channel apparatus, which were prewetted by a small quantity of water and cyclohexane, realizing the prewetting-triggered under-water superlipophobicity and under-oil superhydrophobicity, respectively ( Fig. 3g ). The CSTPNM allowed the passage of the infused liquid itself but repelled another liquid. Cooperating with the automatic feeding by the peristaltic pump, a mixture of water (colorless) and cyclohexane (red) could be continuously separated. The separation efficiency was evaluated via analysing the residual content of cyclohexane and water in the two collected liquids ( Fig. 3h ). The separation efficiencies for both oil and water were greater than 99.5%, and there was no apparent attenuation after a lengthy separation of 4 h. Furthermore, the surfactant-stabilized emulsions, including the CTAB-stabilized cyclohexane-in-water and SDBS-stabilized water-in-cyclohexane emulsions, were prepared to verify the emulsion separation capacity of CSTPNM. The as-prepared emulsions were milky white, and large amounts of micron-sized droplets were observed in the view (Fig. S3 † ). After the filtration process, the filtrates of both oil-in-water (separated by the water-prewetted CSTPNM) and water-in-oil (separated by the cyclohexane-prewetted CSTPNM) emulsions became transparent, and the densely-packed droplets in emulsions were entirely removed, indicating the high efficiency of the membranes for separating emulsions."
} | 5,988 |
39803200 | PMC11725385 | pmc | 121 | {
"abstract": "ABSTRACT Coral reefs worldwide are threatened by increasing ocean temperatures because of the sensitivity of the coral‐algal symbiosis to thermal stress. Reef‐building corals form symbiotic relationships with dinoflagellates (family Symbiodiniaceae), including those species which acquire their initial symbiont complement predominately from their parents. Changes in the composition of symbiont communities, through the mechanisms of symbiont shuffling or switching, can modulate the host's thermal limits. However, the role of shuffling in coral acclimatization to heat is understudied in coral offspring and to date has largely focused on the adults. To quantify potential fitness benefits and consequences of changes in symbiont communities under a simulated heatwave in coral early life‐history stages, we exposed larvae and juveniles of the widespread, vertically transmitting coral, \n Montipora digitata , to heat stress (32°C) and tracked changes in their growth, survival, photosynthetic efficiency, and symbiont community composition over time relative to controls. We found negative impacts from warming in all fitness‐related traits, which varied significantly among larval families and across life‐history stages. Larvae that survived heat exposure exhibited changes in symbiont communities that favored symbionts that are canonically more stress tolerant. Compared to larvae, juveniles showed more rapid mortality under heat stress and their symbiont communities were largely fixed regardless of temperature treatment, suggesting an inability to alter their symbiont community as an acclimatory response to heat stress. Taken together, these findings suggest that capacity for symbiont shuffling may be modified through ontogeny, and that the juvenile life stage may be less flexible and more at risk from climate warming in this species.",
"conclusion": "5 Conclusion Overall, we observed changes in symbiont communities in the early life stages of \n Montipora digitata \n in response to heat stress. We could not determine if these changes were due to an active or passive mechanism. However, our results suggest juvenile stages of \n M. digitata \n are more susceptible to heat stress compared to the larval stage. To determine if shuffling is indeed an active acclimatory mechanism, higher‐resolution time series sampling of early life stages should be conducted. As rare taxa increased in abundance in larvae that were heat stressed, future studies should also examine the degree to which background symbiont communities can be inherited, which will require larger cross designs. As \n M. digitata \n is a vertically transmitting species, the degree to which rare‐heat tolerant species are inherited in offspring may be an indicator of their future resistance to heat stress, which will play a critical role in their survival in a rapidly changing climate.",
"introduction": "1 Introduction Across the tree of life, symbiosis fuels biodiversity and many species engage in these life‐long partnerships to increase their fitness through mechanisms like nutrient exchange, shelter, or chemical defenses (Sachs et al. 2004 ). However, symbiotic relationships are not always equitable or stable over ecological timescales as observed in the nutritional exchange underpinning life on coral reefs, the coral‐algal symbiosis, which is particularly sensitive to environmental perturbation (Kiers et al. 2003 ; Kiers et al. 2010 ; Davy, Denis, and Weis Virginia 2012 ). Ocean warming currently represents the greatest threat to the persistence of coral reefs globally, and as the main driver of warming, human‐induced climate change is increasingly leading to more severe and frequent bleaching and mortality events on reefs (van Woesik et al. 2022 ). Coral bleaching is defined as the breakdown of the partnership between photosynthetic dinoflagellate symbionts (Symbiodiniaceae) and their coral hosts from environmental stress. Generally, this can be caused by high and persistent temperature and high light (van Woesik 2001 ). If stressful conditions persist, this can result in mortality of the host due to the loss of symbiont‐derived nutrition or destruction of host tissues (Glynn 1993 ). One key driver of determining bleaching thresholds in coral individuals and populations is the composition of Symbiodiniaceae hosted by the coral (Baker 2004 ; Berkelmans and Van Oppen 2006 ). Dinoflagellates in the family Symbiodiniaceae are classified into 15 genera with most capable of forming a symbiosis with coral (LaJeunesse et al. 2018 ; Yorifuji et al. 2021 ; Nitschke et al. 2022 ). Corals vary in the degree of specificity in their symbiotic partnerships (Sampayo et al. 2016 ; Elder et al. 2023 ) and the range of potential Symbiodiniaceae partners is diverse (LaJeunesse et al. 2018 ). Some hosts maintain simultaneous relationships with multiple symbionts, whereas others are more specific (Baker 2003 ; Little, van Oppen, and Willis 2004 ; Baird et al. 2007 ; Howells et al. 2020 ; Davies et al. 2023 ). Importantly, differences in these communities drive variation in host fitness, including heat and light tolerance, and growth rates (Putnam et al. 2012 ; Swain et al. 2017 ; Cunning, Silverstein, and Baker 2018 ; Matsuda et al. 2022 ; Davies et al. 2023 ). For example, an increase in the relative abundance of Durusdinum trenchii (formerly D1a, LaJeunesse et al. 2018 ) in three coral species resulted in increased heat tolerance (Cunning, Silverstein, and Baker 2018 ). In addition to their phylogenetic diversity, the coral‐Symbiodiniaceae symbiosis can be modified through the ecological mechanisms by which corals acquire and dynamically regulate their symbiont communities which also influences thermal resistance and resilience of the holobiont. There are two main mechanisms by which symbiont communities change in hospite , namely shuffling and switching. Symbiont shuffling refers to changes in the relative abundance of community members already in residence (Baker 2001 ; Baker et al. 2004 ). Generally, this involves a reduction in the abundance of a dominant symbiont due to an environmental change which provides an opportunity for a numerically rarer symbiont(s) to increase in relative abundance (Quigley et al. 2022 ). This process should, by definition, result in an increase in host fitness and may be adaptive (Baker et al. 2004 ). Switching refers to the ability of a host to replace an existing symbiosis by selecting for a novel partner from the environment (Sørensen et al. 2021 ). Increased abundance of symbionts in the opportunistic and generally stress‐tolerant genus Durusdinium is the canonical example of shuffling following heat stress (Berkelmans and Van Oppen 2006 ; Quigley et al. 2022 ), again emphasizing the role of symbionts in reef resilience (Berkelmans and Van Oppen 2006 ; Quigley et al. 2022 ). However, our knowledge of the functional relevance of shuffling and switching is generally limited to adult coral and has only been examined in early life‐history stages in a few studies (Quigley, Willis, and Kenkel 2019 ; Terrell et al. 2023 ). Symbiont communities in adult corals are also influenced by the mode of symbiont acquisition (Fabina et al. 2012 ). In corals, there are three known mechanisms for symbiosis initiation: vertical, horizontal, or mixed‐mode, with the majority of coral employing horizontal transmission (Quigley et al. 2018 ; Baird, Guest, and Willis 2009 ). Horizontally transmitting corals must acquire their algal symbionts from the environment each generation. Vertical transmitters, on the other hand, obtain their symbiont community from a maternal source, often through the infection of oocytes before fertilization or planula during gestation (Davy and Turner 2003 ; Hirose and Hidaka 2006 ; Padilla‐Gamiño et al. 2012 ). Mixed mode transmission refers to the ability for corals to inherit their symbiont community from a maternal source with the additional ability to acquire symbionts from the environment during development (Ebert 2013 ). Thus far, coral species have generally been categorically described as either vertical or horizontal transmitters (Baird, Guest, and Willis 2009 ), although mixed mode transmission was recently described in a canonical vertical transmitter (Quigley et al. 2018 ). A better understanding of transmission mode is critical because it affects the long‐term fidelity of the symbiotic association (Ebert 2013 ; Quigley et al. 2018 ; Dixon and Kenkel 2019 ). Vertically transmitted symbioses are generally thought to be co‐evolved associations, in which the diversity of symbionts in the host coral is lower (Fabina et al. 2012 ), and the ability of the symbiont to live outside the host is restricted (Krueger and Gates 2012 ). However, in the vertically transmitting coral Montipora digitata , symbiont communities in offspring are more diverse compared to adults (Quigley, Willis, and Bay 2017 ), and alterations in symbiont community composition in adults due to stress are reflected in oocytes, supporting the potential for transgenerational inheritance of shuffled algal communities over time (Quigley, Willis, and Kenkel 2019 ). This suggests that vertically transmitted symbiont communities are more flexible than originally thought, and dynamic shifts in the complement of algal symbionts passed on to offspring may confer fitness benefits in variable environments (Björk et al. 2019 ). Taken together, there is now evidence that flexible symbiotic partnerships may confer greater adaptive and acclimatory potential on the coral holobiont (Torda et al. 2017 ). However, the majority of our current understanding regarding the fitness impacts of flexible symbiont associations comes from studies on adult life stages (Baker 2001 ; Berkelmans and Van Oppen 2006 ; Mieog et al. 2007 ). Larval life stages of marine invertebrates have higher energetic demands (Pechenik 1999 ) which are further exacerbated by heat stress (Przeslawski, Byrne, and Mellin 2015 ). For example, coral larvae show increased respiration rates, decreased survival rates and decreased photosynthetic efficiency (Fv/fm) under heat stress (Putnam et al. 2013 ; Ross et al. 2013 ). Furthermore, heat‐induced differences in energetic demands of coral larvae can also vary across different family and population‐based crosses (Dixon et al. 2015 ; Zhang et al. 2019 , 2023 ), underscoring that larval energetic costs to heat stress have a heritable basis. At the molecular level, coral larvae respond to heat stress with decreased expression profiles of heat‐stress responsive genes, changes in oxidoreductase activity, and cell death (Rodriguez‐Lanetty, Harii, and Hoegh‐Guldberg 2009 ; Polato et al. 2010 ; Dixon et al. 2015 ). At the level of symbiosis, a switch from immune suppression to immune activation was observed upon initiation of symbiosis in heat‐stressed larvae of a horizontally transmitting, \n Acropora digitifera \n which coincided with reduced larval survival when initiating symbiosis with their dominant symbiont (Kitchen et al. 2022 ). However, \n Acropora tenuis \n larvae subjected to heat stress were observed to have greater survivorship rates when exposed to mixed communities of symbionts in equal portions ( Cladocopium sp., Durusdinium sp., Fugacium sp., and Gerakladium sp.) over a two‐week period (Matsuda et al. 2022 ). This makes our understanding of this nutritional endosymbiosis even more essential, particularly for vertically transmitting coral species as cooperation between host and symbiont is predicted to increase between both partners in this context (Douglas 1998 ; Sachs and Wilcox 2006 ; Nalepa 2020 ). In addition, the physiological consequences of the trans‐generationally inherited community shifts observed in oocytes on later coral life stages remain unknown (Quigley, Willis, and Kenkel 2019 ). We exposed multiple cohorts of coral larvae and juveniles to heat stress and monitored changes in their physiology, survival and Symbiodiniaceae communities over time to evaluate the relationship between physiological metrics of fitness and symbiont community composition over coral ontogeny. In addition to reductions in photosynthetic efficiency of symbionts, and size and survival of the host, we show that heat stress in larval samples increased Symbiodiniaceae community alpha diversity through increasing abundances of Symbiodinium , Durusdinium , and Fugacium spp. Alternatively, heat‐stressed juveniles showed a limited capacity to change their symbiont communities. Finally, we show that increased community diversity of Symbiodiniaceae in maternal corals is reflected in the family of offspring and may have fitness consequences.",
"discussion": "4 Discussion In this study, we demonstrate changes in symbiont communities in the early life‐history stages of a common, vertically transmitting coral, Montipora digitata , in response to heat stress. Specifically, we found symbiont communities differed between temperature treatments in larvae but detected no differences in juveniles derived from the same bulk culture fertilization. Although we cannot confirm whether these changes in larvae are due to active (i.e., shuffling by the host animal) or passive (i.e., increase in opportunistic symbionts or differential susceptibility among symbiont community members) mechanisms, we did observe significant differences in survival duration between larvae and juveniles, with larvae surviving more than twice as long as juveniles. Moreover, symbiont communities in heat‐stressed larvae became dominated by representatives of canonically stress‐tolerant genera. Finally, we also show that increased maternal symbiont diversity is reflected in offspring. Overall, our results indicate that larvae can survive twice as long when compared to juveniles under the same warming conditions, potentially driven by symbiont shuffling. This suggests that the juvenile life stage may be more at risk from climate warming due to limited flexibility. 4.1 Life Stage Specific Differences in Physiology and Symbiont Community Diversity in Response to Thermal Stress Although both larvae and juveniles were dominated by the same C15 DIV, larvae survived much longer on average and their symbiont community composition showed greater diversity in the heat treatment compared to juveniles. There is ample evidence to show that symbiont communities drive host physiology in coral adults and to a lesser extent offspring (Quigley et al. 2022 , 2023 ; Terrell et al. 2023 ), underpinned by differences in symbiont tolerance to stress (Swain et al. 2021 ). \n Symbiodinium microadriaticum \n and Durusdinium trenchii , for example, tend to produce less reactive oxygen species (ROS) in culture, a molecular response associated with coral bleaching, when exposed to heat stress, compared to Breviolum minitum and Cladocopium goreaui (Lesser 2019 ). Additionally, these dynamics within hosts can start as early as gametogenesis in vertically transmitting species, as changes in symbiont communities within oocytes of the same species were detected after a mass bleaching year (Quigley, Willis, and Kenkel 2019 ). Taken together, we postulate that the altered symbiont community in larvae of the vertically transmitting \n M. digitata \n may have afforded them a fitness advantage which then allowed them to persist longer under heat stress. Assuming the symbiont community shift led to direct gains in heat tolerance in larvae, this suggests that either the maternal colonies or the larvae have the capacity to actively rearrange their symbiont communities (as an acclimatization mechanism) to cope with heat stress. In horizontally transmitting coral species, the capacity to shuffle symbiont communities in response to heat stress events has been shown in juvenile (Terrell et al. 2023 ) and adult life stages (Berkelmans and Van Oppen 2006 ; Ross et al. 2013 ), indicating an acclimatory response to heat stress is available to some corals. Similar studies are limited for vertically transmitting coral species. Work on another congener, Montipora capitata , provides insight into this capacity. \n M. capitata \n is a vertically transmitting coral in which individuals generally host either Cladocopium (which are more susceptible to thermal stress), Durusdinium (which are less susceptible to thermal stress), or some combination of both (Cunning, Silverstein, and Baker 2018 ; Dilworth et al. 2021 ). However, no changes to symbiont communities were observed for either \n M. capitata \n colonies when exposed to short‐term stress (Dilworth et al. 2021 ). More recently, however, corals of this species previously and recently sampled along heat stress extremes sites along K āne‘ohe Bay showed mixed communities of Durusdinium and Cladocopium that changed in dominance and with stress levels (de Souza et al. 2022 ), perhaps indicating the capacity to shuffle in response to a sufficiently intense environmental effect. A short‐term heat stress did not result in symbiont shuffling in early life stages of M. captitata ; however, there is a capacity to inherit symbiont communities that resemble the dominant community present in the parental coral (Harris et al. 2022 ), a unique feature of vertically transmitting coral (Harris et al. 2022 ; Quigley, Willis, and Kenkel 2019 ). Taken together, these results suggest that extreme heat and vertical transmission (Bright and Bulgheresi 2010 ) can lead to heritable changes in symbiont community composition in vertical transmitting corals over generations. Indeed, this was the case in \n M. digitata \n , where symbiont communities in oocytes were observed to change in parallel to their maternal sources in response to a mass bleaching event (Quigley, Willis, and Kenkel 2019 ). Here we expand on these findings to show that \n M. digitata \n larvae are also able to change their symbiont communities in response to a thermal stress. Alternatively, juveniles lacked this capacity, which is more aligned with the fixed symbiont communities observed in later developmental stages of other vertically transmitting corals. More work is needed to disentangle symbiont establishment and winnowing compared to the mechanisms of shuffling and switching. Finally, although these changes in the symbiont communities appear to be acclimatory, we cannot conclusively determine if these changes preceded differential mortality and so cannot tease apart the influence of these two processes. Alternatively, larvae may be more robust compared to juveniles for reasons unrelated to symbionts. Larvae may be more resistant generally because of their positive buoyancy from high lipid content early in life, which exposes them to harsh environmental conditions such as high ultraviolet radiation and temperature at the sea surface (Glynn 1993 ; Wellington and Fitt 2003 ; Rodriguez‐Lanetty, Harii, and Hoegh‐Guldberg 2009 ; Aranda et al. 2011 ; Gleason and Hofmann 2011 ). Moreover, during the motile larval stage, they are actively exposed to both surface and benthic conditions from several days to multiple weeks (Ritson‐Williams et al. 2009 ), forcing them to withstand highly variable environmental conditions. Metamorphosis is also an energetically costly process that depletes larval energy reserves and may result in more susceptible juvenile stages (Edmunds, Gates, and Gleason 2001 ; Ritson‐Williams et al. 2009 ). The enhanced survival of the larvae compared to juveniles under heat stress may therefore result from either or both the change in symbiont community and an overall robustness of larvae. In summary, we hypothesize that lower survival of juveniles may be driven by lack of an ability to adjust symbiont communities combined with diminished energetic reserves post‐metamorphosis, suggesting the juvenile stage may be the most susceptible life‐history stage for corals. Interestingly, we found Symbiodiniaceae community composition in \n M. digitata \n juveniles to be highly stable regardless of heat exposure. This is in contrast to other studies in which shuffling in juvenile corals has been repeatedly confirmed during initial symbiont acquisition (Little, van Oppen, and Willis 2004 ; Yorifuji et al. 2017 ; Cumbo, Baird, and van Oppen 2013 ), and through development (Quigley et al. 2017 , 2020 ; Terrell et al. 2023 ), only stabilizing later in life (Abrego et al. 2008 ). These changes through juvenile ontogeny are generally referred to as winnowing (Abrego et al. 2008 ). Symbiont communities can change during this winnowing period and are characterized by increases in the abundance of a diversity of symbionts. As part of this, some opportunistic symbionts can be taken up; but communities generally stabilize through time to resemble adult communities, either due to competition or initiation of immune responses (McIlroy et al. 2019 ; Abrego et al. 2008 ). Although we saw community differences in \n M. digitata \n larvae when exposed to heat stress, the contrasting stability of symbiont communities in juveniles under the same conditions suggests that winnowing in this species occurs in larvae and is fixed after this life stage. This further reinforces the notion that the juvenile stage may be the most susceptible to stress. Further work is needed to characterize the dynamics of the symbiosis during this important ontogenetic transition. 4.2 Variation in Fitness Among Larval Families and the Role of Symbiont Community Diversity To better understand parental contributions to fitness differences, we undertook controlled genetic crosses. Previous work in this species showed a concordance between symbiont communities in parents and their eggs (Quigley, Willis, and Kenkel 2019 ), and we were expecting similar patterns in larvae—as we indeed observed here. Symbiont community is a heritable trait (Quigley, Willis, and Bay 2017 ; Quigley, Willis, and Kenkel 2019 ), which implies that at least some familial effects will be present. This has been demonstrated in a number of species in the Indo‐Pacific ( \n Acropora tenuis \n and \n Montipora digitata \n , Quigley, Willis, and Bay 2017 ; \n Seriatopora hystrix \n , Quigley et al. 2018 ). Importantly, we also showed that differences in symbiont communities among families are associated with differences in survival in larvae. In particular, larvae from cross WT3xC1 exhibited lower average survival and had the most disparate background symbiont community. This may have been due to the maternal influence of WT3, which had a distinct endosymbiont community dominated by C15‐C15dq‐C15dr when compared to the other three parental colonies. Therefore, poorer offspring survival could be due to increased abundance of opportunistic Cladocopium spp. variants (Howe‐Kerr et al. 2020 ). We lacked the ability to measure the degree to which symbiont community differences among families can change in response to heat stress due to low sample sizes. It may be that an increase in potentially opportunistic symbionts would increase or decrease the capacity for coral early life stages of \n M. digitata \n to alter their performance under heat stress."
} | 5,804 |
39803200 | PMC11725385 | pmc | 121 | {
"abstract": "ABSTRACT Coral reefs worldwide are threatened by increasing ocean temperatures because of the sensitivity of the coral‐algal symbiosis to thermal stress. Reef‐building corals form symbiotic relationships with dinoflagellates (family Symbiodiniaceae), including those species which acquire their initial symbiont complement predominately from their parents. Changes in the composition of symbiont communities, through the mechanisms of symbiont shuffling or switching, can modulate the host's thermal limits. However, the role of shuffling in coral acclimatization to heat is understudied in coral offspring and to date has largely focused on the adults. To quantify potential fitness benefits and consequences of changes in symbiont communities under a simulated heatwave in coral early life‐history stages, we exposed larvae and juveniles of the widespread, vertically transmitting coral, \n Montipora digitata , to heat stress (32°C) and tracked changes in their growth, survival, photosynthetic efficiency, and symbiont community composition over time relative to controls. We found negative impacts from warming in all fitness‐related traits, which varied significantly among larval families and across life‐history stages. Larvae that survived heat exposure exhibited changes in symbiont communities that favored symbionts that are canonically more stress tolerant. Compared to larvae, juveniles showed more rapid mortality under heat stress and their symbiont communities were largely fixed regardless of temperature treatment, suggesting an inability to alter their symbiont community as an acclimatory response to heat stress. Taken together, these findings suggest that capacity for symbiont shuffling may be modified through ontogeny, and that the juvenile life stage may be less flexible and more at risk from climate warming in this species.",
"conclusion": "5 Conclusion Overall, we observed changes in symbiont communities in the early life stages of \n Montipora digitata \n in response to heat stress. We could not determine if these changes were due to an active or passive mechanism. However, our results suggest juvenile stages of \n M. digitata \n are more susceptible to heat stress compared to the larval stage. To determine if shuffling is indeed an active acclimatory mechanism, higher‐resolution time series sampling of early life stages should be conducted. As rare taxa increased in abundance in larvae that were heat stressed, future studies should also examine the degree to which background symbiont communities can be inherited, which will require larger cross designs. As \n M. digitata \n is a vertically transmitting species, the degree to which rare‐heat tolerant species are inherited in offspring may be an indicator of their future resistance to heat stress, which will play a critical role in their survival in a rapidly changing climate.",
"introduction": "1 Introduction Across the tree of life, symbiosis fuels biodiversity and many species engage in these life‐long partnerships to increase their fitness through mechanisms like nutrient exchange, shelter, or chemical defenses (Sachs et al. 2004 ). However, symbiotic relationships are not always equitable or stable over ecological timescales as observed in the nutritional exchange underpinning life on coral reefs, the coral‐algal symbiosis, which is particularly sensitive to environmental perturbation (Kiers et al. 2003 ; Kiers et al. 2010 ; Davy, Denis, and Weis Virginia 2012 ). Ocean warming currently represents the greatest threat to the persistence of coral reefs globally, and as the main driver of warming, human‐induced climate change is increasingly leading to more severe and frequent bleaching and mortality events on reefs (van Woesik et al. 2022 ). Coral bleaching is defined as the breakdown of the partnership between photosynthetic dinoflagellate symbionts (Symbiodiniaceae) and their coral hosts from environmental stress. Generally, this can be caused by high and persistent temperature and high light (van Woesik 2001 ). If stressful conditions persist, this can result in mortality of the host due to the loss of symbiont‐derived nutrition or destruction of host tissues (Glynn 1993 ). One key driver of determining bleaching thresholds in coral individuals and populations is the composition of Symbiodiniaceae hosted by the coral (Baker 2004 ; Berkelmans and Van Oppen 2006 ). Dinoflagellates in the family Symbiodiniaceae are classified into 15 genera with most capable of forming a symbiosis with coral (LaJeunesse et al. 2018 ; Yorifuji et al. 2021 ; Nitschke et al. 2022 ). Corals vary in the degree of specificity in their symbiotic partnerships (Sampayo et al. 2016 ; Elder et al. 2023 ) and the range of potential Symbiodiniaceae partners is diverse (LaJeunesse et al. 2018 ). Some hosts maintain simultaneous relationships with multiple symbionts, whereas others are more specific (Baker 2003 ; Little, van Oppen, and Willis 2004 ; Baird et al. 2007 ; Howells et al. 2020 ; Davies et al. 2023 ). Importantly, differences in these communities drive variation in host fitness, including heat and light tolerance, and growth rates (Putnam et al. 2012 ; Swain et al. 2017 ; Cunning, Silverstein, and Baker 2018 ; Matsuda et al. 2022 ; Davies et al. 2023 ). For example, an increase in the relative abundance of Durusdinum trenchii (formerly D1a, LaJeunesse et al. 2018 ) in three coral species resulted in increased heat tolerance (Cunning, Silverstein, and Baker 2018 ). In addition to their phylogenetic diversity, the coral‐Symbiodiniaceae symbiosis can be modified through the ecological mechanisms by which corals acquire and dynamically regulate their symbiont communities which also influences thermal resistance and resilience of the holobiont. There are two main mechanisms by which symbiont communities change in hospite , namely shuffling and switching. Symbiont shuffling refers to changes in the relative abundance of community members already in residence (Baker 2001 ; Baker et al. 2004 ). Generally, this involves a reduction in the abundance of a dominant symbiont due to an environmental change which provides an opportunity for a numerically rarer symbiont(s) to increase in relative abundance (Quigley et al. 2022 ). This process should, by definition, result in an increase in host fitness and may be adaptive (Baker et al. 2004 ). Switching refers to the ability of a host to replace an existing symbiosis by selecting for a novel partner from the environment (Sørensen et al. 2021 ). Increased abundance of symbionts in the opportunistic and generally stress‐tolerant genus Durusdinium is the canonical example of shuffling following heat stress (Berkelmans and Van Oppen 2006 ; Quigley et al. 2022 ), again emphasizing the role of symbionts in reef resilience (Berkelmans and Van Oppen 2006 ; Quigley et al. 2022 ). However, our knowledge of the functional relevance of shuffling and switching is generally limited to adult coral and has only been examined in early life‐history stages in a few studies (Quigley, Willis, and Kenkel 2019 ; Terrell et al. 2023 ). Symbiont communities in adult corals are also influenced by the mode of symbiont acquisition (Fabina et al. 2012 ). In corals, there are three known mechanisms for symbiosis initiation: vertical, horizontal, or mixed‐mode, with the majority of coral employing horizontal transmission (Quigley et al. 2018 ; Baird, Guest, and Willis 2009 ). Horizontally transmitting corals must acquire their algal symbionts from the environment each generation. Vertical transmitters, on the other hand, obtain their symbiont community from a maternal source, often through the infection of oocytes before fertilization or planula during gestation (Davy and Turner 2003 ; Hirose and Hidaka 2006 ; Padilla‐Gamiño et al. 2012 ). Mixed mode transmission refers to the ability for corals to inherit their symbiont community from a maternal source with the additional ability to acquire symbionts from the environment during development (Ebert 2013 ). Thus far, coral species have generally been categorically described as either vertical or horizontal transmitters (Baird, Guest, and Willis 2009 ), although mixed mode transmission was recently described in a canonical vertical transmitter (Quigley et al. 2018 ). A better understanding of transmission mode is critical because it affects the long‐term fidelity of the symbiotic association (Ebert 2013 ; Quigley et al. 2018 ; Dixon and Kenkel 2019 ). Vertically transmitted symbioses are generally thought to be co‐evolved associations, in which the diversity of symbionts in the host coral is lower (Fabina et al. 2012 ), and the ability of the symbiont to live outside the host is restricted (Krueger and Gates 2012 ). However, in the vertically transmitting coral Montipora digitata , symbiont communities in offspring are more diverse compared to adults (Quigley, Willis, and Bay 2017 ), and alterations in symbiont community composition in adults due to stress are reflected in oocytes, supporting the potential for transgenerational inheritance of shuffled algal communities over time (Quigley, Willis, and Kenkel 2019 ). This suggests that vertically transmitted symbiont communities are more flexible than originally thought, and dynamic shifts in the complement of algal symbionts passed on to offspring may confer fitness benefits in variable environments (Björk et al. 2019 ). Taken together, there is now evidence that flexible symbiotic partnerships may confer greater adaptive and acclimatory potential on the coral holobiont (Torda et al. 2017 ). However, the majority of our current understanding regarding the fitness impacts of flexible symbiont associations comes from studies on adult life stages (Baker 2001 ; Berkelmans and Van Oppen 2006 ; Mieog et al. 2007 ). Larval life stages of marine invertebrates have higher energetic demands (Pechenik 1999 ) which are further exacerbated by heat stress (Przeslawski, Byrne, and Mellin 2015 ). For example, coral larvae show increased respiration rates, decreased survival rates and decreased photosynthetic efficiency (Fv/fm) under heat stress (Putnam et al. 2013 ; Ross et al. 2013 ). Furthermore, heat‐induced differences in energetic demands of coral larvae can also vary across different family and population‐based crosses (Dixon et al. 2015 ; Zhang et al. 2019 , 2023 ), underscoring that larval energetic costs to heat stress have a heritable basis. At the molecular level, coral larvae respond to heat stress with decreased expression profiles of heat‐stress responsive genes, changes in oxidoreductase activity, and cell death (Rodriguez‐Lanetty, Harii, and Hoegh‐Guldberg 2009 ; Polato et al. 2010 ; Dixon et al. 2015 ). At the level of symbiosis, a switch from immune suppression to immune activation was observed upon initiation of symbiosis in heat‐stressed larvae of a horizontally transmitting, \n Acropora digitifera \n which coincided with reduced larval survival when initiating symbiosis with their dominant symbiont (Kitchen et al. 2022 ). However, \n Acropora tenuis \n larvae subjected to heat stress were observed to have greater survivorship rates when exposed to mixed communities of symbionts in equal portions ( Cladocopium sp., Durusdinium sp., Fugacium sp., and Gerakladium sp.) over a two‐week period (Matsuda et al. 2022 ). This makes our understanding of this nutritional endosymbiosis even more essential, particularly for vertically transmitting coral species as cooperation between host and symbiont is predicted to increase between both partners in this context (Douglas 1998 ; Sachs and Wilcox 2006 ; Nalepa 2020 ). In addition, the physiological consequences of the trans‐generationally inherited community shifts observed in oocytes on later coral life stages remain unknown (Quigley, Willis, and Kenkel 2019 ). We exposed multiple cohorts of coral larvae and juveniles to heat stress and monitored changes in their physiology, survival and Symbiodiniaceae communities over time to evaluate the relationship between physiological metrics of fitness and symbiont community composition over coral ontogeny. In addition to reductions in photosynthetic efficiency of symbionts, and size and survival of the host, we show that heat stress in larval samples increased Symbiodiniaceae community alpha diversity through increasing abundances of Symbiodinium , Durusdinium , and Fugacium spp. Alternatively, heat‐stressed juveniles showed a limited capacity to change their symbiont communities. Finally, we show that increased community diversity of Symbiodiniaceae in maternal corals is reflected in the family of offspring and may have fitness consequences.",
"discussion": "4 Discussion In this study, we demonstrate changes in symbiont communities in the early life‐history stages of a common, vertically transmitting coral, Montipora digitata , in response to heat stress. Specifically, we found symbiont communities differed between temperature treatments in larvae but detected no differences in juveniles derived from the same bulk culture fertilization. Although we cannot confirm whether these changes in larvae are due to active (i.e., shuffling by the host animal) or passive (i.e., increase in opportunistic symbionts or differential susceptibility among symbiont community members) mechanisms, we did observe significant differences in survival duration between larvae and juveniles, with larvae surviving more than twice as long as juveniles. Moreover, symbiont communities in heat‐stressed larvae became dominated by representatives of canonically stress‐tolerant genera. Finally, we also show that increased maternal symbiont diversity is reflected in offspring. Overall, our results indicate that larvae can survive twice as long when compared to juveniles under the same warming conditions, potentially driven by symbiont shuffling. This suggests that the juvenile life stage may be more at risk from climate warming due to limited flexibility. 4.1 Life Stage Specific Differences in Physiology and Symbiont Community Diversity in Response to Thermal Stress Although both larvae and juveniles were dominated by the same C15 DIV, larvae survived much longer on average and their symbiont community composition showed greater diversity in the heat treatment compared to juveniles. There is ample evidence to show that symbiont communities drive host physiology in coral adults and to a lesser extent offspring (Quigley et al. 2022 , 2023 ; Terrell et al. 2023 ), underpinned by differences in symbiont tolerance to stress (Swain et al. 2021 ). \n Symbiodinium microadriaticum \n and Durusdinium trenchii , for example, tend to produce less reactive oxygen species (ROS) in culture, a molecular response associated with coral bleaching, when exposed to heat stress, compared to Breviolum minitum and Cladocopium goreaui (Lesser 2019 ). Additionally, these dynamics within hosts can start as early as gametogenesis in vertically transmitting species, as changes in symbiont communities within oocytes of the same species were detected after a mass bleaching year (Quigley, Willis, and Kenkel 2019 ). Taken together, we postulate that the altered symbiont community in larvae of the vertically transmitting \n M. digitata \n may have afforded them a fitness advantage which then allowed them to persist longer under heat stress. Assuming the symbiont community shift led to direct gains in heat tolerance in larvae, this suggests that either the maternal colonies or the larvae have the capacity to actively rearrange their symbiont communities (as an acclimatization mechanism) to cope with heat stress. In horizontally transmitting coral species, the capacity to shuffle symbiont communities in response to heat stress events has been shown in juvenile (Terrell et al. 2023 ) and adult life stages (Berkelmans and Van Oppen 2006 ; Ross et al. 2013 ), indicating an acclimatory response to heat stress is available to some corals. Similar studies are limited for vertically transmitting coral species. Work on another congener, Montipora capitata , provides insight into this capacity. \n M. capitata \n is a vertically transmitting coral in which individuals generally host either Cladocopium (which are more susceptible to thermal stress), Durusdinium (which are less susceptible to thermal stress), or some combination of both (Cunning, Silverstein, and Baker 2018 ; Dilworth et al. 2021 ). However, no changes to symbiont communities were observed for either \n M. capitata \n colonies when exposed to short‐term stress (Dilworth et al. 2021 ). More recently, however, corals of this species previously and recently sampled along heat stress extremes sites along K āne‘ohe Bay showed mixed communities of Durusdinium and Cladocopium that changed in dominance and with stress levels (de Souza et al. 2022 ), perhaps indicating the capacity to shuffle in response to a sufficiently intense environmental effect. A short‐term heat stress did not result in symbiont shuffling in early life stages of M. captitata ; however, there is a capacity to inherit symbiont communities that resemble the dominant community present in the parental coral (Harris et al. 2022 ), a unique feature of vertically transmitting coral (Harris et al. 2022 ; Quigley, Willis, and Kenkel 2019 ). Taken together, these results suggest that extreme heat and vertical transmission (Bright and Bulgheresi 2010 ) can lead to heritable changes in symbiont community composition in vertical transmitting corals over generations. Indeed, this was the case in \n M. digitata \n , where symbiont communities in oocytes were observed to change in parallel to their maternal sources in response to a mass bleaching event (Quigley, Willis, and Kenkel 2019 ). Here we expand on these findings to show that \n M. digitata \n larvae are also able to change their symbiont communities in response to a thermal stress. Alternatively, juveniles lacked this capacity, which is more aligned with the fixed symbiont communities observed in later developmental stages of other vertically transmitting corals. More work is needed to disentangle symbiont establishment and winnowing compared to the mechanisms of shuffling and switching. Finally, although these changes in the symbiont communities appear to be acclimatory, we cannot conclusively determine if these changes preceded differential mortality and so cannot tease apart the influence of these two processes. Alternatively, larvae may be more robust compared to juveniles for reasons unrelated to symbionts. Larvae may be more resistant generally because of their positive buoyancy from high lipid content early in life, which exposes them to harsh environmental conditions such as high ultraviolet radiation and temperature at the sea surface (Glynn 1993 ; Wellington and Fitt 2003 ; Rodriguez‐Lanetty, Harii, and Hoegh‐Guldberg 2009 ; Aranda et al. 2011 ; Gleason and Hofmann 2011 ). Moreover, during the motile larval stage, they are actively exposed to both surface and benthic conditions from several days to multiple weeks (Ritson‐Williams et al. 2009 ), forcing them to withstand highly variable environmental conditions. Metamorphosis is also an energetically costly process that depletes larval energy reserves and may result in more susceptible juvenile stages (Edmunds, Gates, and Gleason 2001 ; Ritson‐Williams et al. 2009 ). The enhanced survival of the larvae compared to juveniles under heat stress may therefore result from either or both the change in symbiont community and an overall robustness of larvae. In summary, we hypothesize that lower survival of juveniles may be driven by lack of an ability to adjust symbiont communities combined with diminished energetic reserves post‐metamorphosis, suggesting the juvenile stage may be the most susceptible life‐history stage for corals. Interestingly, we found Symbiodiniaceae community composition in \n M. digitata \n juveniles to be highly stable regardless of heat exposure. This is in contrast to other studies in which shuffling in juvenile corals has been repeatedly confirmed during initial symbiont acquisition (Little, van Oppen, and Willis 2004 ; Yorifuji et al. 2017 ; Cumbo, Baird, and van Oppen 2013 ), and through development (Quigley et al. 2017 , 2020 ; Terrell et al. 2023 ), only stabilizing later in life (Abrego et al. 2008 ). These changes through juvenile ontogeny are generally referred to as winnowing (Abrego et al. 2008 ). Symbiont communities can change during this winnowing period and are characterized by increases in the abundance of a diversity of symbionts. As part of this, some opportunistic symbionts can be taken up; but communities generally stabilize through time to resemble adult communities, either due to competition or initiation of immune responses (McIlroy et al. 2019 ; Abrego et al. 2008 ). Although we saw community differences in \n M. digitata \n larvae when exposed to heat stress, the contrasting stability of symbiont communities in juveniles under the same conditions suggests that winnowing in this species occurs in larvae and is fixed after this life stage. This further reinforces the notion that the juvenile stage may be the most susceptible to stress. Further work is needed to characterize the dynamics of the symbiosis during this important ontogenetic transition. 4.2 Variation in Fitness Among Larval Families and the Role of Symbiont Community Diversity To better understand parental contributions to fitness differences, we undertook controlled genetic crosses. Previous work in this species showed a concordance between symbiont communities in parents and their eggs (Quigley, Willis, and Kenkel 2019 ), and we were expecting similar patterns in larvae—as we indeed observed here. Symbiont community is a heritable trait (Quigley, Willis, and Bay 2017 ; Quigley, Willis, and Kenkel 2019 ), which implies that at least some familial effects will be present. This has been demonstrated in a number of species in the Indo‐Pacific ( \n Acropora tenuis \n and \n Montipora digitata \n , Quigley, Willis, and Bay 2017 ; \n Seriatopora hystrix \n , Quigley et al. 2018 ). Importantly, we also showed that differences in symbiont communities among families are associated with differences in survival in larvae. In particular, larvae from cross WT3xC1 exhibited lower average survival and had the most disparate background symbiont community. This may have been due to the maternal influence of WT3, which had a distinct endosymbiont community dominated by C15‐C15dq‐C15dr when compared to the other three parental colonies. Therefore, poorer offspring survival could be due to increased abundance of opportunistic Cladocopium spp. variants (Howe‐Kerr et al. 2020 ). We lacked the ability to measure the degree to which symbiont community differences among families can change in response to heat stress due to low sample sizes. It may be that an increase in potentially opportunistic symbionts would increase or decrease the capacity for coral early life stages of \n M. digitata \n to alter their performance under heat stress."
} | 5,804 |
39803200 | PMC11725385 | pmc | 122 | {
"abstract": "ABSTRACT Coral reefs worldwide are threatened by increasing ocean temperatures because of the sensitivity of the coral‐algal symbiosis to thermal stress. Reef‐building corals form symbiotic relationships with dinoflagellates (family Symbiodiniaceae), including those species which acquire their initial symbiont complement predominately from their parents. Changes in the composition of symbiont communities, through the mechanisms of symbiont shuffling or switching, can modulate the host's thermal limits. However, the role of shuffling in coral acclimatization to heat is understudied in coral offspring and to date has largely focused on the adults. To quantify potential fitness benefits and consequences of changes in symbiont communities under a simulated heatwave in coral early life‐history stages, we exposed larvae and juveniles of the widespread, vertically transmitting coral, \n Montipora digitata , to heat stress (32°C) and tracked changes in their growth, survival, photosynthetic efficiency, and symbiont community composition over time relative to controls. We found negative impacts from warming in all fitness‐related traits, which varied significantly among larval families and across life‐history stages. Larvae that survived heat exposure exhibited changes in symbiont communities that favored symbionts that are canonically more stress tolerant. Compared to larvae, juveniles showed more rapid mortality under heat stress and their symbiont communities were largely fixed regardless of temperature treatment, suggesting an inability to alter their symbiont community as an acclimatory response to heat stress. Taken together, these findings suggest that capacity for symbiont shuffling may be modified through ontogeny, and that the juvenile life stage may be less flexible and more at risk from climate warming in this species.",
"conclusion": "5 Conclusion Overall, we observed changes in symbiont communities in the early life stages of \n Montipora digitata \n in response to heat stress. We could not determine if these changes were due to an active or passive mechanism. However, our results suggest juvenile stages of \n M. digitata \n are more susceptible to heat stress compared to the larval stage. To determine if shuffling is indeed an active acclimatory mechanism, higher‐resolution time series sampling of early life stages should be conducted. As rare taxa increased in abundance in larvae that were heat stressed, future studies should also examine the degree to which background symbiont communities can be inherited, which will require larger cross designs. As \n M. digitata \n is a vertically transmitting species, the degree to which rare‐heat tolerant species are inherited in offspring may be an indicator of their future resistance to heat stress, which will play a critical role in their survival in a rapidly changing climate.",
"introduction": "1 Introduction Across the tree of life, symbiosis fuels biodiversity and many species engage in these life‐long partnerships to increase their fitness through mechanisms like nutrient exchange, shelter, or chemical defenses (Sachs et al. 2004 ). However, symbiotic relationships are not always equitable or stable over ecological timescales as observed in the nutritional exchange underpinning life on coral reefs, the coral‐algal symbiosis, which is particularly sensitive to environmental perturbation (Kiers et al. 2003 ; Kiers et al. 2010 ; Davy, Denis, and Weis Virginia 2012 ). Ocean warming currently represents the greatest threat to the persistence of coral reefs globally, and as the main driver of warming, human‐induced climate change is increasingly leading to more severe and frequent bleaching and mortality events on reefs (van Woesik et al. 2022 ). Coral bleaching is defined as the breakdown of the partnership between photosynthetic dinoflagellate symbionts (Symbiodiniaceae) and their coral hosts from environmental stress. Generally, this can be caused by high and persistent temperature and high light (van Woesik 2001 ). If stressful conditions persist, this can result in mortality of the host due to the loss of symbiont‐derived nutrition or destruction of host tissues (Glynn 1993 ). One key driver of determining bleaching thresholds in coral individuals and populations is the composition of Symbiodiniaceae hosted by the coral (Baker 2004 ; Berkelmans and Van Oppen 2006 ). Dinoflagellates in the family Symbiodiniaceae are classified into 15 genera with most capable of forming a symbiosis with coral (LaJeunesse et al. 2018 ; Yorifuji et al. 2021 ; Nitschke et al. 2022 ). Corals vary in the degree of specificity in their symbiotic partnerships (Sampayo et al. 2016 ; Elder et al. 2023 ) and the range of potential Symbiodiniaceae partners is diverse (LaJeunesse et al. 2018 ). Some hosts maintain simultaneous relationships with multiple symbionts, whereas others are more specific (Baker 2003 ; Little, van Oppen, and Willis 2004 ; Baird et al. 2007 ; Howells et al. 2020 ; Davies et al. 2023 ). Importantly, differences in these communities drive variation in host fitness, including heat and light tolerance, and growth rates (Putnam et al. 2012 ; Swain et al. 2017 ; Cunning, Silverstein, and Baker 2018 ; Matsuda et al. 2022 ; Davies et al. 2023 ). For example, an increase in the relative abundance of Durusdinum trenchii (formerly D1a, LaJeunesse et al. 2018 ) in three coral species resulted in increased heat tolerance (Cunning, Silverstein, and Baker 2018 ). In addition to their phylogenetic diversity, the coral‐Symbiodiniaceae symbiosis can be modified through the ecological mechanisms by which corals acquire and dynamically regulate their symbiont communities which also influences thermal resistance and resilience of the holobiont. There are two main mechanisms by which symbiont communities change in hospite , namely shuffling and switching. Symbiont shuffling refers to changes in the relative abundance of community members already in residence (Baker 2001 ; Baker et al. 2004 ). Generally, this involves a reduction in the abundance of a dominant symbiont due to an environmental change which provides an opportunity for a numerically rarer symbiont(s) to increase in relative abundance (Quigley et al. 2022 ). This process should, by definition, result in an increase in host fitness and may be adaptive (Baker et al. 2004 ). Switching refers to the ability of a host to replace an existing symbiosis by selecting for a novel partner from the environment (Sørensen et al. 2021 ). Increased abundance of symbionts in the opportunistic and generally stress‐tolerant genus Durusdinium is the canonical example of shuffling following heat stress (Berkelmans and Van Oppen 2006 ; Quigley et al. 2022 ), again emphasizing the role of symbionts in reef resilience (Berkelmans and Van Oppen 2006 ; Quigley et al. 2022 ). However, our knowledge of the functional relevance of shuffling and switching is generally limited to adult coral and has only been examined in early life‐history stages in a few studies (Quigley, Willis, and Kenkel 2019 ; Terrell et al. 2023 ). Symbiont communities in adult corals are also influenced by the mode of symbiont acquisition (Fabina et al. 2012 ). In corals, there are three known mechanisms for symbiosis initiation: vertical, horizontal, or mixed‐mode, with the majority of coral employing horizontal transmission (Quigley et al. 2018 ; Baird, Guest, and Willis 2009 ). Horizontally transmitting corals must acquire their algal symbionts from the environment each generation. Vertical transmitters, on the other hand, obtain their symbiont community from a maternal source, often through the infection of oocytes before fertilization or planula during gestation (Davy and Turner 2003 ; Hirose and Hidaka 2006 ; Padilla‐Gamiño et al. 2012 ). Mixed mode transmission refers to the ability for corals to inherit their symbiont community from a maternal source with the additional ability to acquire symbionts from the environment during development (Ebert 2013 ). Thus far, coral species have generally been categorically described as either vertical or horizontal transmitters (Baird, Guest, and Willis 2009 ), although mixed mode transmission was recently described in a canonical vertical transmitter (Quigley et al. 2018 ). A better understanding of transmission mode is critical because it affects the long‐term fidelity of the symbiotic association (Ebert 2013 ; Quigley et al. 2018 ; Dixon and Kenkel 2019 ). Vertically transmitted symbioses are generally thought to be co‐evolved associations, in which the diversity of symbionts in the host coral is lower (Fabina et al. 2012 ), and the ability of the symbiont to live outside the host is restricted (Krueger and Gates 2012 ). However, in the vertically transmitting coral Montipora digitata , symbiont communities in offspring are more diverse compared to adults (Quigley, Willis, and Bay 2017 ), and alterations in symbiont community composition in adults due to stress are reflected in oocytes, supporting the potential for transgenerational inheritance of shuffled algal communities over time (Quigley, Willis, and Kenkel 2019 ). This suggests that vertically transmitted symbiont communities are more flexible than originally thought, and dynamic shifts in the complement of algal symbionts passed on to offspring may confer fitness benefits in variable environments (Björk et al. 2019 ). Taken together, there is now evidence that flexible symbiotic partnerships may confer greater adaptive and acclimatory potential on the coral holobiont (Torda et al. 2017 ). However, the majority of our current understanding regarding the fitness impacts of flexible symbiont associations comes from studies on adult life stages (Baker 2001 ; Berkelmans and Van Oppen 2006 ; Mieog et al. 2007 ). Larval life stages of marine invertebrates have higher energetic demands (Pechenik 1999 ) which are further exacerbated by heat stress (Przeslawski, Byrne, and Mellin 2015 ). For example, coral larvae show increased respiration rates, decreased survival rates and decreased photosynthetic efficiency (Fv/fm) under heat stress (Putnam et al. 2013 ; Ross et al. 2013 ). Furthermore, heat‐induced differences in energetic demands of coral larvae can also vary across different family and population‐based crosses (Dixon et al. 2015 ; Zhang et al. 2019 , 2023 ), underscoring that larval energetic costs to heat stress have a heritable basis. At the molecular level, coral larvae respond to heat stress with decreased expression profiles of heat‐stress responsive genes, changes in oxidoreductase activity, and cell death (Rodriguez‐Lanetty, Harii, and Hoegh‐Guldberg 2009 ; Polato et al. 2010 ; Dixon et al. 2015 ). At the level of symbiosis, a switch from immune suppression to immune activation was observed upon initiation of symbiosis in heat‐stressed larvae of a horizontally transmitting, \n Acropora digitifera \n which coincided with reduced larval survival when initiating symbiosis with their dominant symbiont (Kitchen et al. 2022 ). However, \n Acropora tenuis \n larvae subjected to heat stress were observed to have greater survivorship rates when exposed to mixed communities of symbionts in equal portions ( Cladocopium sp., Durusdinium sp., Fugacium sp., and Gerakladium sp.) over a two‐week period (Matsuda et al. 2022 ). This makes our understanding of this nutritional endosymbiosis even more essential, particularly for vertically transmitting coral species as cooperation between host and symbiont is predicted to increase between both partners in this context (Douglas 1998 ; Sachs and Wilcox 2006 ; Nalepa 2020 ). In addition, the physiological consequences of the trans‐generationally inherited community shifts observed in oocytes on later coral life stages remain unknown (Quigley, Willis, and Kenkel 2019 ). We exposed multiple cohorts of coral larvae and juveniles to heat stress and monitored changes in their physiology, survival and Symbiodiniaceae communities over time to evaluate the relationship between physiological metrics of fitness and symbiont community composition over coral ontogeny. In addition to reductions in photosynthetic efficiency of symbionts, and size and survival of the host, we show that heat stress in larval samples increased Symbiodiniaceae community alpha diversity through increasing abundances of Symbiodinium , Durusdinium , and Fugacium spp. Alternatively, heat‐stressed juveniles showed a limited capacity to change their symbiont communities. Finally, we show that increased community diversity of Symbiodiniaceae in maternal corals is reflected in the family of offspring and may have fitness consequences.",
"discussion": "4 Discussion In this study, we demonstrate changes in symbiont communities in the early life‐history stages of a common, vertically transmitting coral, Montipora digitata , in response to heat stress. Specifically, we found symbiont communities differed between temperature treatments in larvae but detected no differences in juveniles derived from the same bulk culture fertilization. Although we cannot confirm whether these changes in larvae are due to active (i.e., shuffling by the host animal) or passive (i.e., increase in opportunistic symbionts or differential susceptibility among symbiont community members) mechanisms, we did observe significant differences in survival duration between larvae and juveniles, with larvae surviving more than twice as long as juveniles. Moreover, symbiont communities in heat‐stressed larvae became dominated by representatives of canonically stress‐tolerant genera. Finally, we also show that increased maternal symbiont diversity is reflected in offspring. Overall, our results indicate that larvae can survive twice as long when compared to juveniles under the same warming conditions, potentially driven by symbiont shuffling. This suggests that the juvenile life stage may be more at risk from climate warming due to limited flexibility. 4.1 Life Stage Specific Differences in Physiology and Symbiont Community Diversity in Response to Thermal Stress Although both larvae and juveniles were dominated by the same C15 DIV, larvae survived much longer on average and their symbiont community composition showed greater diversity in the heat treatment compared to juveniles. There is ample evidence to show that symbiont communities drive host physiology in coral adults and to a lesser extent offspring (Quigley et al. 2022 , 2023 ; Terrell et al. 2023 ), underpinned by differences in symbiont tolerance to stress (Swain et al. 2021 ). \n Symbiodinium microadriaticum \n and Durusdinium trenchii , for example, tend to produce less reactive oxygen species (ROS) in culture, a molecular response associated with coral bleaching, when exposed to heat stress, compared to Breviolum minitum and Cladocopium goreaui (Lesser 2019 ). Additionally, these dynamics within hosts can start as early as gametogenesis in vertically transmitting species, as changes in symbiont communities within oocytes of the same species were detected after a mass bleaching year (Quigley, Willis, and Kenkel 2019 ). Taken together, we postulate that the altered symbiont community in larvae of the vertically transmitting \n M. digitata \n may have afforded them a fitness advantage which then allowed them to persist longer under heat stress. Assuming the symbiont community shift led to direct gains in heat tolerance in larvae, this suggests that either the maternal colonies or the larvae have the capacity to actively rearrange their symbiont communities (as an acclimatization mechanism) to cope with heat stress. In horizontally transmitting coral species, the capacity to shuffle symbiont communities in response to heat stress events has been shown in juvenile (Terrell et al. 2023 ) and adult life stages (Berkelmans and Van Oppen 2006 ; Ross et al. 2013 ), indicating an acclimatory response to heat stress is available to some corals. Similar studies are limited for vertically transmitting coral species. Work on another congener, Montipora capitata , provides insight into this capacity. \n M. capitata \n is a vertically transmitting coral in which individuals generally host either Cladocopium (which are more susceptible to thermal stress), Durusdinium (which are less susceptible to thermal stress), or some combination of both (Cunning, Silverstein, and Baker 2018 ; Dilworth et al. 2021 ). However, no changes to symbiont communities were observed for either \n M. capitata \n colonies when exposed to short‐term stress (Dilworth et al. 2021 ). More recently, however, corals of this species previously and recently sampled along heat stress extremes sites along K āne‘ohe Bay showed mixed communities of Durusdinium and Cladocopium that changed in dominance and with stress levels (de Souza et al. 2022 ), perhaps indicating the capacity to shuffle in response to a sufficiently intense environmental effect. A short‐term heat stress did not result in symbiont shuffling in early life stages of M. captitata ; however, there is a capacity to inherit symbiont communities that resemble the dominant community present in the parental coral (Harris et al. 2022 ), a unique feature of vertically transmitting coral (Harris et al. 2022 ; Quigley, Willis, and Kenkel 2019 ). Taken together, these results suggest that extreme heat and vertical transmission (Bright and Bulgheresi 2010 ) can lead to heritable changes in symbiont community composition in vertical transmitting corals over generations. Indeed, this was the case in \n M. digitata \n , where symbiont communities in oocytes were observed to change in parallel to their maternal sources in response to a mass bleaching event (Quigley, Willis, and Kenkel 2019 ). Here we expand on these findings to show that \n M. digitata \n larvae are also able to change their symbiont communities in response to a thermal stress. Alternatively, juveniles lacked this capacity, which is more aligned with the fixed symbiont communities observed in later developmental stages of other vertically transmitting corals. More work is needed to disentangle symbiont establishment and winnowing compared to the mechanisms of shuffling and switching. Finally, although these changes in the symbiont communities appear to be acclimatory, we cannot conclusively determine if these changes preceded differential mortality and so cannot tease apart the influence of these two processes. Alternatively, larvae may be more robust compared to juveniles for reasons unrelated to symbionts. Larvae may be more resistant generally because of their positive buoyancy from high lipid content early in life, which exposes them to harsh environmental conditions such as high ultraviolet radiation and temperature at the sea surface (Glynn 1993 ; Wellington and Fitt 2003 ; Rodriguez‐Lanetty, Harii, and Hoegh‐Guldberg 2009 ; Aranda et al. 2011 ; Gleason and Hofmann 2011 ). Moreover, during the motile larval stage, they are actively exposed to both surface and benthic conditions from several days to multiple weeks (Ritson‐Williams et al. 2009 ), forcing them to withstand highly variable environmental conditions. Metamorphosis is also an energetically costly process that depletes larval energy reserves and may result in more susceptible juvenile stages (Edmunds, Gates, and Gleason 2001 ; Ritson‐Williams et al. 2009 ). The enhanced survival of the larvae compared to juveniles under heat stress may therefore result from either or both the change in symbiont community and an overall robustness of larvae. In summary, we hypothesize that lower survival of juveniles may be driven by lack of an ability to adjust symbiont communities combined with diminished energetic reserves post‐metamorphosis, suggesting the juvenile stage may be the most susceptible life‐history stage for corals. Interestingly, we found Symbiodiniaceae community composition in \n M. digitata \n juveniles to be highly stable regardless of heat exposure. This is in contrast to other studies in which shuffling in juvenile corals has been repeatedly confirmed during initial symbiont acquisition (Little, van Oppen, and Willis 2004 ; Yorifuji et al. 2017 ; Cumbo, Baird, and van Oppen 2013 ), and through development (Quigley et al. 2017 , 2020 ; Terrell et al. 2023 ), only stabilizing later in life (Abrego et al. 2008 ). These changes through juvenile ontogeny are generally referred to as winnowing (Abrego et al. 2008 ). Symbiont communities can change during this winnowing period and are characterized by increases in the abundance of a diversity of symbionts. As part of this, some opportunistic symbionts can be taken up; but communities generally stabilize through time to resemble adult communities, either due to competition or initiation of immune responses (McIlroy et al. 2019 ; Abrego et al. 2008 ). Although we saw community differences in \n M. digitata \n larvae when exposed to heat stress, the contrasting stability of symbiont communities in juveniles under the same conditions suggests that winnowing in this species occurs in larvae and is fixed after this life stage. This further reinforces the notion that the juvenile stage may be the most susceptible to stress. Further work is needed to characterize the dynamics of the symbiosis during this important ontogenetic transition. 4.2 Variation in Fitness Among Larval Families and the Role of Symbiont Community Diversity To better understand parental contributions to fitness differences, we undertook controlled genetic crosses. Previous work in this species showed a concordance between symbiont communities in parents and their eggs (Quigley, Willis, and Kenkel 2019 ), and we were expecting similar patterns in larvae—as we indeed observed here. Symbiont community is a heritable trait (Quigley, Willis, and Bay 2017 ; Quigley, Willis, and Kenkel 2019 ), which implies that at least some familial effects will be present. This has been demonstrated in a number of species in the Indo‐Pacific ( \n Acropora tenuis \n and \n Montipora digitata \n , Quigley, Willis, and Bay 2017 ; \n Seriatopora hystrix \n , Quigley et al. 2018 ). Importantly, we also showed that differences in symbiont communities among families are associated with differences in survival in larvae. In particular, larvae from cross WT3xC1 exhibited lower average survival and had the most disparate background symbiont community. This may have been due to the maternal influence of WT3, which had a distinct endosymbiont community dominated by C15‐C15dq‐C15dr when compared to the other three parental colonies. Therefore, poorer offspring survival could be due to increased abundance of opportunistic Cladocopium spp. variants (Howe‐Kerr et al. 2020 ). We lacked the ability to measure the degree to which symbiont community differences among families can change in response to heat stress due to low sample sizes. It may be that an increase in potentially opportunistic symbionts would increase or decrease the capacity for coral early life stages of \n M. digitata \n to alter their performance under heat stress."
} | 5,804 |
39803200 | PMC11725385 | pmc | 122 | {
"abstract": "ABSTRACT Coral reefs worldwide are threatened by increasing ocean temperatures because of the sensitivity of the coral‐algal symbiosis to thermal stress. Reef‐building corals form symbiotic relationships with dinoflagellates (family Symbiodiniaceae), including those species which acquire their initial symbiont complement predominately from their parents. Changes in the composition of symbiont communities, through the mechanisms of symbiont shuffling or switching, can modulate the host's thermal limits. However, the role of shuffling in coral acclimatization to heat is understudied in coral offspring and to date has largely focused on the adults. To quantify potential fitness benefits and consequences of changes in symbiont communities under a simulated heatwave in coral early life‐history stages, we exposed larvae and juveniles of the widespread, vertically transmitting coral, \n Montipora digitata , to heat stress (32°C) and tracked changes in their growth, survival, photosynthetic efficiency, and symbiont community composition over time relative to controls. We found negative impacts from warming in all fitness‐related traits, which varied significantly among larval families and across life‐history stages. Larvae that survived heat exposure exhibited changes in symbiont communities that favored symbionts that are canonically more stress tolerant. Compared to larvae, juveniles showed more rapid mortality under heat stress and their symbiont communities were largely fixed regardless of temperature treatment, suggesting an inability to alter their symbiont community as an acclimatory response to heat stress. Taken together, these findings suggest that capacity for symbiont shuffling may be modified through ontogeny, and that the juvenile life stage may be less flexible and more at risk from climate warming in this species.",
"conclusion": "5 Conclusion Overall, we observed changes in symbiont communities in the early life stages of \n Montipora digitata \n in response to heat stress. We could not determine if these changes were due to an active or passive mechanism. However, our results suggest juvenile stages of \n M. digitata \n are more susceptible to heat stress compared to the larval stage. To determine if shuffling is indeed an active acclimatory mechanism, higher‐resolution time series sampling of early life stages should be conducted. As rare taxa increased in abundance in larvae that were heat stressed, future studies should also examine the degree to which background symbiont communities can be inherited, which will require larger cross designs. As \n M. digitata \n is a vertically transmitting species, the degree to which rare‐heat tolerant species are inherited in offspring may be an indicator of their future resistance to heat stress, which will play a critical role in their survival in a rapidly changing climate.",
"introduction": "1 Introduction Across the tree of life, symbiosis fuels biodiversity and many species engage in these life‐long partnerships to increase their fitness through mechanisms like nutrient exchange, shelter, or chemical defenses (Sachs et al. 2004 ). However, symbiotic relationships are not always equitable or stable over ecological timescales as observed in the nutritional exchange underpinning life on coral reefs, the coral‐algal symbiosis, which is particularly sensitive to environmental perturbation (Kiers et al. 2003 ; Kiers et al. 2010 ; Davy, Denis, and Weis Virginia 2012 ). Ocean warming currently represents the greatest threat to the persistence of coral reefs globally, and as the main driver of warming, human‐induced climate change is increasingly leading to more severe and frequent bleaching and mortality events on reefs (van Woesik et al. 2022 ). Coral bleaching is defined as the breakdown of the partnership between photosynthetic dinoflagellate symbionts (Symbiodiniaceae) and their coral hosts from environmental stress. Generally, this can be caused by high and persistent temperature and high light (van Woesik 2001 ). If stressful conditions persist, this can result in mortality of the host due to the loss of symbiont‐derived nutrition or destruction of host tissues (Glynn 1993 ). One key driver of determining bleaching thresholds in coral individuals and populations is the composition of Symbiodiniaceae hosted by the coral (Baker 2004 ; Berkelmans and Van Oppen 2006 ). Dinoflagellates in the family Symbiodiniaceae are classified into 15 genera with most capable of forming a symbiosis with coral (LaJeunesse et al. 2018 ; Yorifuji et al. 2021 ; Nitschke et al. 2022 ). Corals vary in the degree of specificity in their symbiotic partnerships (Sampayo et al. 2016 ; Elder et al. 2023 ) and the range of potential Symbiodiniaceae partners is diverse (LaJeunesse et al. 2018 ). Some hosts maintain simultaneous relationships with multiple symbionts, whereas others are more specific (Baker 2003 ; Little, van Oppen, and Willis 2004 ; Baird et al. 2007 ; Howells et al. 2020 ; Davies et al. 2023 ). Importantly, differences in these communities drive variation in host fitness, including heat and light tolerance, and growth rates (Putnam et al. 2012 ; Swain et al. 2017 ; Cunning, Silverstein, and Baker 2018 ; Matsuda et al. 2022 ; Davies et al. 2023 ). For example, an increase in the relative abundance of Durusdinum trenchii (formerly D1a, LaJeunesse et al. 2018 ) in three coral species resulted in increased heat tolerance (Cunning, Silverstein, and Baker 2018 ). In addition to their phylogenetic diversity, the coral‐Symbiodiniaceae symbiosis can be modified through the ecological mechanisms by which corals acquire and dynamically regulate their symbiont communities which also influences thermal resistance and resilience of the holobiont. There are two main mechanisms by which symbiont communities change in hospite , namely shuffling and switching. Symbiont shuffling refers to changes in the relative abundance of community members already in residence (Baker 2001 ; Baker et al. 2004 ). Generally, this involves a reduction in the abundance of a dominant symbiont due to an environmental change which provides an opportunity for a numerically rarer symbiont(s) to increase in relative abundance (Quigley et al. 2022 ). This process should, by definition, result in an increase in host fitness and may be adaptive (Baker et al. 2004 ). Switching refers to the ability of a host to replace an existing symbiosis by selecting for a novel partner from the environment (Sørensen et al. 2021 ). Increased abundance of symbionts in the opportunistic and generally stress‐tolerant genus Durusdinium is the canonical example of shuffling following heat stress (Berkelmans and Van Oppen 2006 ; Quigley et al. 2022 ), again emphasizing the role of symbionts in reef resilience (Berkelmans and Van Oppen 2006 ; Quigley et al. 2022 ). However, our knowledge of the functional relevance of shuffling and switching is generally limited to adult coral and has only been examined in early life‐history stages in a few studies (Quigley, Willis, and Kenkel 2019 ; Terrell et al. 2023 ). Symbiont communities in adult corals are also influenced by the mode of symbiont acquisition (Fabina et al. 2012 ). In corals, there are three known mechanisms for symbiosis initiation: vertical, horizontal, or mixed‐mode, with the majority of coral employing horizontal transmission (Quigley et al. 2018 ; Baird, Guest, and Willis 2009 ). Horizontally transmitting corals must acquire their algal symbionts from the environment each generation. Vertical transmitters, on the other hand, obtain their symbiont community from a maternal source, often through the infection of oocytes before fertilization or planula during gestation (Davy and Turner 2003 ; Hirose and Hidaka 2006 ; Padilla‐Gamiño et al. 2012 ). Mixed mode transmission refers to the ability for corals to inherit their symbiont community from a maternal source with the additional ability to acquire symbionts from the environment during development (Ebert 2013 ). Thus far, coral species have generally been categorically described as either vertical or horizontal transmitters (Baird, Guest, and Willis 2009 ), although mixed mode transmission was recently described in a canonical vertical transmitter (Quigley et al. 2018 ). A better understanding of transmission mode is critical because it affects the long‐term fidelity of the symbiotic association (Ebert 2013 ; Quigley et al. 2018 ; Dixon and Kenkel 2019 ). Vertically transmitted symbioses are generally thought to be co‐evolved associations, in which the diversity of symbionts in the host coral is lower (Fabina et al. 2012 ), and the ability of the symbiont to live outside the host is restricted (Krueger and Gates 2012 ). However, in the vertically transmitting coral Montipora digitata , symbiont communities in offspring are more diverse compared to adults (Quigley, Willis, and Bay 2017 ), and alterations in symbiont community composition in adults due to stress are reflected in oocytes, supporting the potential for transgenerational inheritance of shuffled algal communities over time (Quigley, Willis, and Kenkel 2019 ). This suggests that vertically transmitted symbiont communities are more flexible than originally thought, and dynamic shifts in the complement of algal symbionts passed on to offspring may confer fitness benefits in variable environments (Björk et al. 2019 ). Taken together, there is now evidence that flexible symbiotic partnerships may confer greater adaptive and acclimatory potential on the coral holobiont (Torda et al. 2017 ). However, the majority of our current understanding regarding the fitness impacts of flexible symbiont associations comes from studies on adult life stages (Baker 2001 ; Berkelmans and Van Oppen 2006 ; Mieog et al. 2007 ). Larval life stages of marine invertebrates have higher energetic demands (Pechenik 1999 ) which are further exacerbated by heat stress (Przeslawski, Byrne, and Mellin 2015 ). For example, coral larvae show increased respiration rates, decreased survival rates and decreased photosynthetic efficiency (Fv/fm) under heat stress (Putnam et al. 2013 ; Ross et al. 2013 ). Furthermore, heat‐induced differences in energetic demands of coral larvae can also vary across different family and population‐based crosses (Dixon et al. 2015 ; Zhang et al. 2019 , 2023 ), underscoring that larval energetic costs to heat stress have a heritable basis. At the molecular level, coral larvae respond to heat stress with decreased expression profiles of heat‐stress responsive genes, changes in oxidoreductase activity, and cell death (Rodriguez‐Lanetty, Harii, and Hoegh‐Guldberg 2009 ; Polato et al. 2010 ; Dixon et al. 2015 ). At the level of symbiosis, a switch from immune suppression to immune activation was observed upon initiation of symbiosis in heat‐stressed larvae of a horizontally transmitting, \n Acropora digitifera \n which coincided with reduced larval survival when initiating symbiosis with their dominant symbiont (Kitchen et al. 2022 ). However, \n Acropora tenuis \n larvae subjected to heat stress were observed to have greater survivorship rates when exposed to mixed communities of symbionts in equal portions ( Cladocopium sp., Durusdinium sp., Fugacium sp., and Gerakladium sp.) over a two‐week period (Matsuda et al. 2022 ). This makes our understanding of this nutritional endosymbiosis even more essential, particularly for vertically transmitting coral species as cooperation between host and symbiont is predicted to increase between both partners in this context (Douglas 1998 ; Sachs and Wilcox 2006 ; Nalepa 2020 ). In addition, the physiological consequences of the trans‐generationally inherited community shifts observed in oocytes on later coral life stages remain unknown (Quigley, Willis, and Kenkel 2019 ). We exposed multiple cohorts of coral larvae and juveniles to heat stress and monitored changes in their physiology, survival and Symbiodiniaceae communities over time to evaluate the relationship between physiological metrics of fitness and symbiont community composition over coral ontogeny. In addition to reductions in photosynthetic efficiency of symbionts, and size and survival of the host, we show that heat stress in larval samples increased Symbiodiniaceae community alpha diversity through increasing abundances of Symbiodinium , Durusdinium , and Fugacium spp. Alternatively, heat‐stressed juveniles showed a limited capacity to change their symbiont communities. Finally, we show that increased community diversity of Symbiodiniaceae in maternal corals is reflected in the family of offspring and may have fitness consequences.",
"discussion": "4 Discussion In this study, we demonstrate changes in symbiont communities in the early life‐history stages of a common, vertically transmitting coral, Montipora digitata , in response to heat stress. Specifically, we found symbiont communities differed between temperature treatments in larvae but detected no differences in juveniles derived from the same bulk culture fertilization. Although we cannot confirm whether these changes in larvae are due to active (i.e., shuffling by the host animal) or passive (i.e., increase in opportunistic symbionts or differential susceptibility among symbiont community members) mechanisms, we did observe significant differences in survival duration between larvae and juveniles, with larvae surviving more than twice as long as juveniles. Moreover, symbiont communities in heat‐stressed larvae became dominated by representatives of canonically stress‐tolerant genera. Finally, we also show that increased maternal symbiont diversity is reflected in offspring. Overall, our results indicate that larvae can survive twice as long when compared to juveniles under the same warming conditions, potentially driven by symbiont shuffling. This suggests that the juvenile life stage may be more at risk from climate warming due to limited flexibility. 4.1 Life Stage Specific Differences in Physiology and Symbiont Community Diversity in Response to Thermal Stress Although both larvae and juveniles were dominated by the same C15 DIV, larvae survived much longer on average and their symbiont community composition showed greater diversity in the heat treatment compared to juveniles. There is ample evidence to show that symbiont communities drive host physiology in coral adults and to a lesser extent offspring (Quigley et al. 2022 , 2023 ; Terrell et al. 2023 ), underpinned by differences in symbiont tolerance to stress (Swain et al. 2021 ). \n Symbiodinium microadriaticum \n and Durusdinium trenchii , for example, tend to produce less reactive oxygen species (ROS) in culture, a molecular response associated with coral bleaching, when exposed to heat stress, compared to Breviolum minitum and Cladocopium goreaui (Lesser 2019 ). Additionally, these dynamics within hosts can start as early as gametogenesis in vertically transmitting species, as changes in symbiont communities within oocytes of the same species were detected after a mass bleaching year (Quigley, Willis, and Kenkel 2019 ). Taken together, we postulate that the altered symbiont community in larvae of the vertically transmitting \n M. digitata \n may have afforded them a fitness advantage which then allowed them to persist longer under heat stress. Assuming the symbiont community shift led to direct gains in heat tolerance in larvae, this suggests that either the maternal colonies or the larvae have the capacity to actively rearrange their symbiont communities (as an acclimatization mechanism) to cope with heat stress. In horizontally transmitting coral species, the capacity to shuffle symbiont communities in response to heat stress events has been shown in juvenile (Terrell et al. 2023 ) and adult life stages (Berkelmans and Van Oppen 2006 ; Ross et al. 2013 ), indicating an acclimatory response to heat stress is available to some corals. Similar studies are limited for vertically transmitting coral species. Work on another congener, Montipora capitata , provides insight into this capacity. \n M. capitata \n is a vertically transmitting coral in which individuals generally host either Cladocopium (which are more susceptible to thermal stress), Durusdinium (which are less susceptible to thermal stress), or some combination of both (Cunning, Silverstein, and Baker 2018 ; Dilworth et al. 2021 ). However, no changes to symbiont communities were observed for either \n M. capitata \n colonies when exposed to short‐term stress (Dilworth et al. 2021 ). More recently, however, corals of this species previously and recently sampled along heat stress extremes sites along K āne‘ohe Bay showed mixed communities of Durusdinium and Cladocopium that changed in dominance and with stress levels (de Souza et al. 2022 ), perhaps indicating the capacity to shuffle in response to a sufficiently intense environmental effect. A short‐term heat stress did not result in symbiont shuffling in early life stages of M. captitata ; however, there is a capacity to inherit symbiont communities that resemble the dominant community present in the parental coral (Harris et al. 2022 ), a unique feature of vertically transmitting coral (Harris et al. 2022 ; Quigley, Willis, and Kenkel 2019 ). Taken together, these results suggest that extreme heat and vertical transmission (Bright and Bulgheresi 2010 ) can lead to heritable changes in symbiont community composition in vertical transmitting corals over generations. Indeed, this was the case in \n M. digitata \n , where symbiont communities in oocytes were observed to change in parallel to their maternal sources in response to a mass bleaching event (Quigley, Willis, and Kenkel 2019 ). Here we expand on these findings to show that \n M. digitata \n larvae are also able to change their symbiont communities in response to a thermal stress. Alternatively, juveniles lacked this capacity, which is more aligned with the fixed symbiont communities observed in later developmental stages of other vertically transmitting corals. More work is needed to disentangle symbiont establishment and winnowing compared to the mechanisms of shuffling and switching. Finally, although these changes in the symbiont communities appear to be acclimatory, we cannot conclusively determine if these changes preceded differential mortality and so cannot tease apart the influence of these two processes. Alternatively, larvae may be more robust compared to juveniles for reasons unrelated to symbionts. Larvae may be more resistant generally because of their positive buoyancy from high lipid content early in life, which exposes them to harsh environmental conditions such as high ultraviolet radiation and temperature at the sea surface (Glynn 1993 ; Wellington and Fitt 2003 ; Rodriguez‐Lanetty, Harii, and Hoegh‐Guldberg 2009 ; Aranda et al. 2011 ; Gleason and Hofmann 2011 ). Moreover, during the motile larval stage, they are actively exposed to both surface and benthic conditions from several days to multiple weeks (Ritson‐Williams et al. 2009 ), forcing them to withstand highly variable environmental conditions. Metamorphosis is also an energetically costly process that depletes larval energy reserves and may result in more susceptible juvenile stages (Edmunds, Gates, and Gleason 2001 ; Ritson‐Williams et al. 2009 ). The enhanced survival of the larvae compared to juveniles under heat stress may therefore result from either or both the change in symbiont community and an overall robustness of larvae. In summary, we hypothesize that lower survival of juveniles may be driven by lack of an ability to adjust symbiont communities combined with diminished energetic reserves post‐metamorphosis, suggesting the juvenile stage may be the most susceptible life‐history stage for corals. Interestingly, we found Symbiodiniaceae community composition in \n M. digitata \n juveniles to be highly stable regardless of heat exposure. This is in contrast to other studies in which shuffling in juvenile corals has been repeatedly confirmed during initial symbiont acquisition (Little, van Oppen, and Willis 2004 ; Yorifuji et al. 2017 ; Cumbo, Baird, and van Oppen 2013 ), and through development (Quigley et al. 2017 , 2020 ; Terrell et al. 2023 ), only stabilizing later in life (Abrego et al. 2008 ). These changes through juvenile ontogeny are generally referred to as winnowing (Abrego et al. 2008 ). Symbiont communities can change during this winnowing period and are characterized by increases in the abundance of a diversity of symbionts. As part of this, some opportunistic symbionts can be taken up; but communities generally stabilize through time to resemble adult communities, either due to competition or initiation of immune responses (McIlroy et al. 2019 ; Abrego et al. 2008 ). Although we saw community differences in \n M. digitata \n larvae when exposed to heat stress, the contrasting stability of symbiont communities in juveniles under the same conditions suggests that winnowing in this species occurs in larvae and is fixed after this life stage. This further reinforces the notion that the juvenile stage may be the most susceptible to stress. Further work is needed to characterize the dynamics of the symbiosis during this important ontogenetic transition. 4.2 Variation in Fitness Among Larval Families and the Role of Symbiont Community Diversity To better understand parental contributions to fitness differences, we undertook controlled genetic crosses. Previous work in this species showed a concordance between symbiont communities in parents and their eggs (Quigley, Willis, and Kenkel 2019 ), and we were expecting similar patterns in larvae—as we indeed observed here. Symbiont community is a heritable trait (Quigley, Willis, and Bay 2017 ; Quigley, Willis, and Kenkel 2019 ), which implies that at least some familial effects will be present. This has been demonstrated in a number of species in the Indo‐Pacific ( \n Acropora tenuis \n and \n Montipora digitata \n , Quigley, Willis, and Bay 2017 ; \n Seriatopora hystrix \n , Quigley et al. 2018 ). Importantly, we also showed that differences in symbiont communities among families are associated with differences in survival in larvae. In particular, larvae from cross WT3xC1 exhibited lower average survival and had the most disparate background symbiont community. This may have been due to the maternal influence of WT3, which had a distinct endosymbiont community dominated by C15‐C15dq‐C15dr when compared to the other three parental colonies. Therefore, poorer offspring survival could be due to increased abundance of opportunistic Cladocopium spp. variants (Howe‐Kerr et al. 2020 ). We lacked the ability to measure the degree to which symbiont community differences among families can change in response to heat stress due to low sample sizes. It may be that an increase in potentially opportunistic symbionts would increase or decrease the capacity for coral early life stages of \n M. digitata \n to alter their performance under heat stress."
} | 5,804 |
40019370 | PMC12021054 | pmc | 123 | {
"abstract": "Abstract Asymmetric structures have exhibited significant advantages in regulating wetting behavior. Nevertheless, the influence of this unique structural feature on anti‐icing performance remains to be further explored. In this work, static/dynamic anti‐icing performance is investigated on the asymmetric superhydrophobic structures fabricated by micro‐milling combined with electrodeposition. Notably, although the reduction of the degree of asymmetry increases the droplet adhesion force by augmenting the solid‐liquid interface, asymmetric structures can still enable the droplet to bounce off the surface through the horizontal Laplace force generated by the contact angle difference between the two sides of the droplet. On this basis, a dynamic behavior criterion for the droplet to detach from the surface is established at low temperatures. Molecular dynamics simulation indicates that the asymmetric structure can reduce the icing probability on the precursor film by inhibiting the nucleation and growth process of water molecules, decreasing the liquid‐ice interface, and reducing the adhesion under low temperatures. Generally, specific asymmetric structures with nucleation inhibition characteristics can reduce droplet adhesion and increase the driving force during the droplet retraction stage by enhancing the horizontal Laplace force, effectively improving the dynamic non‐wetting performance of the surface at even −40 °C.",
"conclusion": "3 Conclusion In this work, asymmetric hierarchical structures with super‐hydrophobicity were fabricated by micro‐milling combined with the electrodeposition method. The static anti‐icing performance indicates that the greater the degree of asymmetry of the hierarchical structure, the more solid‐ice interfaces were generated below the droplet, and the extra thermal pathways were constructed for heat transfer under low temperatures, improving the ice crystal growth rate inside the droplet. Meanwhile, the asymmetric structure (A‐30) with higher icing delay performance could not only ascend the temperature gradient between the droplet and the substrate by increasing the interface thermal resistance but also provide more time for the air to fully dissolve into the droplet, improving the impact speed of the air bubble during the melting process, effectively promoting the surface wetting state to revert to the Cassie model. The subsequent dynamic non‐wettability evaluation revealed that the energy dissipation of droplets on various superhydrophobic surfaces mainly manifested as adhesive dissipation, and the bounce behavior of droplets was dominated by the competitive relationship between the contraction driving force induced by asymmetric structure and the freezing rate of precursor film induced by low temperature. Further force analysis clarified that the decrease in asymmetry degree led to a significant enlargement in the solid‐liquid interface, increasing the adhesion force of droplets on a superhydrophobic surface. However, the difference in contact angles between the two sides of a droplet induced by asymmetric structure generated a horizontal Laplace force, which determined whether the droplet could bounce off the surface. On this basis, a dynamic behavior criterion for the droplet to detach from the surface was established at low temperatures. Furthermore, combined with molecular dynamics analysis, it was verified that asymmetric structure could reduce the icing probability in the precursor film by inhibiting the nucleation and growth process of water molecules, decreasing the liquid‐ice interface, and reducing the adhesion force of droplets under low temperatures. Generally, specific asymmetric structures with nucleation inhibition characteristics could simultaneously reduce droplet adhesion and increase the driving force during the droplet retraction stage by enhancing the horizontal Laplace force, effectively improving the dynamic non‐wetting performance of the surface at even −40 °C.",
"introduction": "1 Introduction Ice formation has become an important issue plaguing aerospace, wind electricity, rail transportation, and other fields due to its potential safety hazards and economic losses. [ \n \n 1 \n , \n 2 \n , \n 3 \n , \n 4 \n , \n 5 \n , \n 6 \n \n ] Considering the energy limitation and environmental friendliness of existing anti/de‐icing technologies, superhydrophobic materials have evolved as one of the ideal anti/de‐icing materials without energy consumption owing to their unique characteristics of low droplet adhesion, low ice nucleation rate, and low ice adhesion strength. [ \n \n 7 \n , \n 8 \n , \n 9 \n , \n 10 \n \n ] It is well known that the realization of anti/de‐icing function mainly depends on the coordinated control of structure and energy on a superhydrophobic surface. [ \n \n 11 \n , \n 12 \n , \n 13 \n \n ] Notably, previous studies have shown that microstructure usually has a more significant impact on the hydrophobicity and anti‐icing performance of materials. [ \n \n 14 \n , \n 15 \n \n ] Therefore, numerous research works have been carried out to investigate the influence of microstructure on the anti/de‐icing property. However, in consideration of the limitations of structural design and fabrication technology, conventional research on superhydrophobic structures is still focused on isotropic configurations, such as symmetric or irregular structures. [ \n \n 16 \n , \n 17 \n , \n 18 \n \n ] Inspired by the structure of Trifolium, a periodic micro‐pit structure with higher critical Laplace pressure was constructed in order to achieve the reduction in ice accumulation by delaying the icing time. [ \n \n 19 \n \n ] On this basis, an open micro‐cone structure with nanoparticles was designed to introduce two Cassie–Baxter states (CB I and CB II) into the surface by controlling the pinning behavior of the liquid during the cooling process, significantly enhancing the energy barrier between CB Wenzel states. [ \n \n 20 \n \n ] Moreover, the influence of the wetting fraction of infiltrated liquid on the nucleation behavior was explored in the presence of potential nucleation sites on the regular micro‐column structure, promoting the establishment of design criteria for anti‐icing structure. [ \n \n 21 \n \n ] \n Although the anti‐icing behavior of symmetric structures has been discussed from multiple scales and perspectives, the recent design concept of symmetric anti‐icing structures still focuses on prolonging the icing time. [ \n \n 22 \n , \n 23 \n , \n 24 \n \n ] Interestingly, the asymmetric structure can significantly affect the mechanical behavior of droplets on superhydrophobic surfaces, which is expected to reduce ice accumulation on the surface by changing the form of solid‐liquid contact. [ \n \n 25 \n , \n 26 \n , \n 27 \n , \n 28 \n \n ] It was clarified that a superhydrophobic surface composed of asymmetric triangular microstructures with a lower droplet adhesion possessed the ability to make tiny droplets spontaneously bounce off the surface in a specific direction due to the anisotropic adhesion characteristics. This was the realization of directional long‐distance transport of droplets for the first time. [ \n \n 29 \n \n ] Afterward, an asymmetric conical micro‐structure raised a “tip‐induced flipping” effect through the control of structure curvature and height gradient, which could further improve the droplet transport efficiency even in the condition of adverse Laplace pressure. [ \n \n 30 \n \n ] Inspired by the structure of araucaria, an asymmetric 3D curved structure was designed to realize droplet movement in the direction of enlargement system energy without external energy input, breaking the inherent recognition that the direction of fluid movement was mainly determined by the solid surface structure. [ \n \n 31 \n \n ] Subsequently, a tilted stepped mushroom‐like micropillar surface was introduced to reduce the contact time (7 ms) of low surface tension droplets (γ = 32 mN m −1 ) while controlling the droplet bouncing direction. [ \n \n 32 \n \n ] On this basis, specific asymmetric inclined fiber structures could effectively accelerate the motion of the water droplet under the wind field, showing application potential in dynamic anti/de‐icing fields. [ \n \n 33 \n \n ] It is worth noting that the current research about the influence of asymmetric structure on surface properties is still restricted to the directional motion control of droplets. Although researchers have tried to design spring‐like pillars, which can achieve the self‐ejection of droplets after freezing by relying on the releasement of elastic energy. [ \n \n 34 \n \n ] Insufficient attention and exploration have been paid to its effect on anti‐deicing performance, especially nucleation behavior and static/dynamic non‐wettability at low temperatures. Herein, static/dynamic anti‐icing performance was investigated on asymmetric superhydrophobic structures fabricated by micro‐milling combined with the electrodeposition method. Although the reduction of the degree of asymmetry increased the droplet adhesion force droplets by augmenting the solid‐liquid interface, asymmetric structures could still enable the droplet to bounce off the surface through the horizontal Laplace force generated by the contact angle difference between the two sides of the droplet. Molecular dynamics simulation indicates that the asymmetric structure could reduce the icing probability on the precursor film by inhibiting the nucleation and growth process of water molecules, decreasing the liquid‐ice interface, and reducing the adhesion force of droplets at low temperatures. This work confirms that specific asymmetric structures could simultaneously reduce droplet adhesion and increase the driving force during the droplet retraction stage by enhancing the horizontal Laplace force, providing theoretical guidance for reducing the probability of icing by controlling the contact time of droplets.",
"discussion": "2 Results and Discussion 2.1 Fabrication and Characterization of Asymmetric Micro‐Nanostructures An asymmetric array microstructure distributed on aluminum alloy was manufactured using micro‐milling technology. Subsequently, superhydrophobic treatment was performed on the asymmetric array microstructures by a one‐step electrodeposition method with cerous nitrate and stearic acid, as shown in Figure \n \n 1 a . Considering the comparability and manufacturability of the microstructure, the height of the microstructure was fixed and the degree of asymmetry was controlled by changing the structural angle, therein, the asymmetry degree is defined as the ratio of two sides of a microstructure. Three typical wedge‐shaped microstructures with an identical height of 50 µm and different angles are selected to investigate the effect of geometric parameters of asymmetric structures on non‐wetting performance under low temperatures (A‐20 sample with an angle of 20° and an asymmetry degree of 2.92, A‐30 sample with an angle of 30°and an asymmetry degree of 2 and A‐40 sample with an angle of 40°and an asymmetry degree of 1.56), as illustrated in Figure 1b . The contact angles of A‐20, A‐30, and A‐40 samples with superhydrophobic treatment are 169.74°, 170.82°, and 167.31°, respectively, while the sliding angles remain ≈2°. Figure 1 Fabrication and characterization of asymmetric micro‐nanostructures. a) Schematic diagram of the preparation process for hierarchical structure. b) The morphology of asymmetric micro‐nanostructures observed by SEM and AFM and the water contact angles on different surfaces are inserted into the corresponding SEM images. c) Surface chemical composition analysis by FTIR. d) GIXRD analysis of asymmetric micro‐nanostructure surfaces. e) XPS analysis of asymmetric micro‐nanostructures. The details of peaks O1 and C1 are indicated respectively. f) EDS analysis of asymmetric micro‐nanostructures. Scanning Electron Microscope (SEM) images reveal that the superhydrophobic surfaces are composed of staggered nanorods with submicron length and a width of ≈50 nm. Meanwhile, Atomic Force Microscope (AFM) results indicate that the root‐mean‐square roughness (Rqs) of these three samples are 531, 527, and 480 nm, respectively. This demonstrates that the superhydrophobic nanostructures superimposed on the asymmetric microstructures have little effect on the characteristic morphology of the hierarchical structures. Additionally, Fourier Transform Infrared Spectrometer (FTIR), Grazing Incidence X‐Ray Diffraction (GIXRD), and X‐Ray Photoelectron Spectroscopy (XPS) methods were employed to further confirm the composition of the superhydrophobic surface, as depicted in Figure 1c–e . The absorption peaks occurring at 2914.88 and 2850.41 cm −1 are recognized as the ─CH 2 group, and the absorption peaks reflecting ≈1542.09 and 1453.17 cm −1 are identified as cerium stearate. [ \n \n 35 \n \n ] \n According to GIXRD analysis, few intrinsically hydrophobic cerium compounds (such as CeO 2 ) exist on superhydrophobic surfaces. Corresponding XPS results also confirm that the energy spectrum peak near 284.28 eV is supposed to be ─CH 2 , while the peak ≈287.93 eV is deemed to be an O═C─O─ bond. Moreover, the energy spectrum peaks appear at 530.91 eV and 529.16 eV are authenticated as C═O and C─O─ bonds, respectively. [ \n \n 35 \n \n ] It is worth noting that Ce3d peaks were detected in the full spectrum (≈885.12 and 903.04 eV), verifying the existence of cerium in the material as Ce 3+ . [ \n \n 36 \n \n ] Furthermore, the atomic occupancy ratios of C, O, and Ce elements on the superhydrophobic surface are 88.2, 10.2, and 1.6%, respectively, with a relative ratio of 55.13:6.38:1 (similar to the 54:6:1 of [CH 3 (CH 2 ) 16 COO] 3 Ce), indicating that the main component of the superhydrophobic surface is cerium stearate, as shown in Figure 1f . 2.2 Static Anti‐Icing Performance Evaluation The same electrodeposition treatment was applied to an aluminum alloy plate as a comparison sample to evaluate the effect of asymmetric micro‐nanostructure on the icing delay characteristic. The droplet shows a recalescence phenomenon on the superhydrophobic plate after 505 s and completely froze within 14 s, as demonstrated in Figure \n \n 2 a . This indicates that the [CH 3 (CH 2 ) 16 COO] 3 Ce deposited on the surface is sufficient to preserve a large amount of air at a low temperature of −20 °C, effectively prolonging the process of droplets icing. However, the recalescence time of the A‐20 sample is increased to 537 s while the complete freezing time is synchronously expanded to 32 s. It can be inferred that the asymmetric micro‐nanostructure effectively improves the air‐capturing capability of the surface, reducing the heat transfer efficiency between the droplet and the surface. Notably, the A‐30 sample exhibits a superior anti‐icing property under low temperatures with a recalescence time of 1316 s and a completely frozen time of 84 s. In contrast, the recalescence time of the A‐40 sample is 328 s, which is only 29.7% of that of the A‐30 sample, and even lower than that of the superhydrophobic plate. It is deduced that low temperature induces extra water molecule nucleation at the interface since the A‐40 sample with a lower asymmetry degree has a more solid‐liquid interface (higher quantity of microstructure per length), weakening the icing delay effect of the surface. Figure 2 Icing delay behavior and ice‐melting process on different superhydrophobic surfaces. a) Icing delay process on various superhydrophobic surfaces at −20 °C. b) Icing delay process on various superhydrophobic surfaces at −30 °C. c) Icing delay process on various superhydrophobic surfaces at −40 °C. d) Schematic diagram of solid‐liquid contact behavior on different asymmetric micro‐nanostructures. e) Ice‐melting behavior on different superhydrophobic surfaces. When the environment temperature is reduced to −30 °C, the recalescence time of the superhydrophobic plate sharply decreases to 4 s, while the anti‐icing ability of these three surfaces with asymmetric micro‐nanostructures also declines to varying degrees, as shown in Figure 2b . Unfortunately, the recalescence time of A‐30 is also shortened to 33 s, which is due to the fact that lower temperature significantly increases the icing nucleation rate at the solid‐liquid interface, allowing the droplets to further wet the structure, resulting in a sudden reduction in anti‐icing ability. Meanwhile, the recalescence time of A‐20 and A‐40 samples is only maintained for 2–3 s, which is only 6–9% of that on A‐30 samples. It can be inferred that the descending temperature greatly reduces the air pressure retained inside the micro‐nanostructures, resulting in an impairment of anti‐icing ability for A‐20 and A‐40 samples. Figure 2c reveals that the recalescence and completely frozen time of the superhydrophobic plate are curtailed to 1 s and 10 s at −40 °C, respectively. Meanwhile, A‐20 and A‐40 samples both show a recalescence phenomenon within 2 s while the A‐30 sample can only delay the icing time to 5 s, indicating that lower temperature completely destroys the support effect of the air bubble inside the asymmetric micro‐nanostructure on the upper droplets. Generally, considering the geometric characteristics of the asymmetric micro‐nanostructure on the superhydrophobic surface, the less the asymmetry degree, the smaller the space occupied by droplets within the microstructure, as shown in Figure 2d . The microstructure with a lower asymmetry degree can retain more micro‐air pockets to hinder temperature transfer, leading to an improvement of anti‐icing performance. This is why the anti‐icing performance of the A‐20 sample is lower than that of the A‐30 sample. However, for droplets with a fixed size, the larger the asymmetry degree, the lower the solid‐liquid contact interfaces within the same contact radius. The lesser asymmetry degree tends to easily cause a wide range of direct temperature transfer from the low‐temperature substrate, promoting the nucleation of droplets, and resulting in an abnormal decrease in the icing delay time on the A‐40 sample. Subsequently, an icing‐melting test is performed to further evaluate the anti‐icing performance of asymmetric superhydrophobic microstructures (Figure 2e ). The freezing process of the droplets during the test is shown in the Supporting Information. It is discernible that the contact angle and sliding angle of the A‐30 sample can still be maintained at 167.31° and 3.3° after melting, respectively. In contrast, the A‐20 sample barely maintains the superhydrophobicity after the icing‐melting process (contact angle≈150.31°, sliding angle≈5.4°), while the A‐40 sample completely lost the superhydrophobicity (contact angle≈142.05°, sliding angle≈14.7°). It is affirmed that a tension gradient at the surface always prompts tiny bubbles trapped in the ice to the melt zone in a test process due to the Marangoni effect. The larger the temperature gradient, the greater the Marangoni force, leading to an inverted conical interface between the melted and non‐melted regions. These tiny bubbles can impact downward rapidly to restore the surface from the Wenzel state to the Cassie–Baxter state when the Marangoni force is greater than the sum of buoyancy and water resistance acting on the bubble. The critical conditions for bubble motion can be expressed as: [ \n \n 22 \n \n ] \n \n (1) \n ∫ 0 2 π d γ d T Δ T ⋅ r b max d α − 4 π 3 ρ g r b max 3 − π 2 C ⋅ ρ v b 2 r b max 2 = 0 \n where dγ/dT = 0.1 mN mK −1 , ΔT is the temperature difference between the surface and the top of the droplet, ρ is the density of water, g is the gravitational constant, r bmax \n is the maximum bubble radius that can move downward, C is the resistance of water (5.9 × 10 3 ) and ν b \n impact velocity of the bubble. It can be seen from this equation that the bubble impact velocity increases exponentially with the increase of the temperature gradient between the droplet and the surface. Afterward, the evolution law of interfacial thermal resistance is investigated by establishing a quasi‐static heat flux model in order to further explore the relationship between the Marangoni effect and the wetting behavior.\n \n (2) \n R C a s s i e = 1 π r d 2 sin 2 θ ⋅ δ c k c f m f n + δ m k m f m + δ n k n f m f n \n where r d \n is the radius of a droplet in the Cassie–Baxter state, θ is the apparent contact angle. k m \n , k c \n , δ m, \n and δ c \n represent the thermal conductivity and thickness of the substrate and superhydrophobic surface, respectively. f n \n , k n, \n and δ n \n are the area fraction, thermal conductivity, and microstructure height, respectively. Hence, the superhydrophobicity can effectively ascend the interfacial thermal resistance, which not only effectively delays the icing process of the droplet, but also increases the temperature gradient between droplets and the surface, giving rise to an augment of bubble impact velocity. Synchronously, the A‐30 sample with higher icing‐delay performance also allows the ambient air to dissolve into the droplets sufficiently, providing enough tiny bubbles during the subsequent melting process. 2.3 Dynamic Non‐Wetting Performance at Low‐Temperature The bouncing processes of droplets on these three typical samples are monitored in order to investigate the dynamic non‐wetting behavior of asymmetric micro‐nanostructure, and the superhydrophobic plate is adopted as a comparison, as shown in Figure \n \n 3 a . Previous research reveals that the retraction process is the main factor determining whether droplets can detach from the superhydrophobic surface. [ \n \n 37 \n \n ] Droplets can successfully bounce off the superhydrophobic plate within 11.3 ms after reaching the maximum spreading diameter. For the A‐20 sample with an asymmetry degree of 2.92, the retraction stage only takes 9.5 ms, which is 16% less than the retraction time on the superhydrophobic plate. Therefore, it is considered that the asymmetric micro‐nanostructure effectively decreases surface adhesion dissipation, shortening the contact time of droplets on the surface. Notably, the retraction stage of a droplet on the A‐30 sample is only 9.3 ms, even lower than that of the A‐20 sample. This indicates that the influence of the structure on the droplet adhesion behavior is directional due to its asymmetric characteristics. Structures with specific asymmetry exhibit the characteristic of inducing droplet contraction. Additionally, although the retraction time on the A‐40 sample increases to 10.4 ms, the contractive effect induced by asymmetric micro‐nanostructure still ensures that the contact time of the droplet on the A‐40 surface is shorter than that on the superhydrophobic plate. Figure 3 Dynamic non‐wetting behavior of various superhydrophobic surfaces. a) Droplet impact process on different superhydrophobic surfaces at room temperature. b) Energy dissipation during droplet impact process on various superhydrophobic surfaces. c) Droplet impact process on asymmetric micro‐nanostructures at −20 °C. d) Droplet impact process on A‐30 sample at −30 °C. e) Droplet impact process on A‐30 sample at −40 °C. Therewith, a equation is used to describe the energy conversion process in order to further explore the dynamic behavior of droplet impact: [ \n \n 38 \n \n ] \n \n (3) \n E k + E d = E a d h e n s i o n + E V + E ′ k + E ′ d \n where E k \n and E k ’ are the kinetic energy before and after the droplet contacts the superhydrophobic surface respectively, the E d \n and E d ’ are the surface energy before and after the droplet contact the surface. The E adhension \n is adhesive dissipative energy, and the E V \n is viscous dissipative energy. Therein, the E k \n can be expressed as:\n \n (4) \n E k = 2 3 π ρ R 0 3 V 0 3 \n where\n \n (5) \n E v = μ V 0 δ 2 τ Ω \n \n \n (6) \n δ = 4 R 0 2 R e \n \n \n (7) \n R e = ρ V 0 R 0 μ \n \n \n (8) \n Ω = π δ R max 2 \n \n The V 0 \n is the instantaneous velocity of the droplet upon contact with the surface, taken as 1 m s −1 , R 0 \n is the initial radius of the droplet, set as 1.22 mm, ρ is the droplet density of 1.0 × 10 3 kg m −3 , µ is the viscosity coefficient of water at room temperature, 1.0087 × 10 −3 Nsm −2 , and R max \n is the maximum spreading radius of the droplet during impact process. On the premise of ignoring the air resistance and volume change before and after impact, the kinetic energy of the droplet after impact can be equivalent to the gravitational potential energy when it bounces to the highest point:\n \n (9) \n E ′ k = 4 3 ρ π R 0 3 g H max \n \n Considering the surface energy of the droplet is constant, the total energy consumption ( E total energy loss \n ) of the droplet during the impact process can be denoted as:\n \n (10) \n \n \n And the energy recovery coefficient ε , which can intuitively reflect the degree of surface energy dissipation can be calculated as:\n \n (11) \n ε = E ′ k E k = 4 3 ρ π R 0 3 g H max / 2 3 π ρ R 0 3 V 0 2 = 2 g H max V 0 2 \n \n The calculation results indicate that the energy dissipation of droplets on all these samples is still dominated by adhesive dissipation. Particularly, the adhesion dissipation energy of droplets on the A‐30 sample is only 2.41 µJ, verifying the decisive role of droplet contraction characteristics induced by asymmetry structure on the bounce behavior during droplet impact, as illustrated in Figure 3b . Meanwhile, the viscous dissipation energy is always between 18.2% and 21.2% of the total dissipation energy during various impact processes, indicating that the internal flow inside the droplet has a slight influence on the bouncing process. Moreover, the energy recovery coefficients of different samples also confirm that asymmetric micro‐nanostructures can effectively reduce the contact process between droplet and surface, descending the adhesive dissipation energy. When the surface temperature drops below zero (−20 °C), the droplet on the A‐20 sample is unable to detach from the surface with a retraction time of 13.4 ms, as delineated in Figure 3c . Relevant researches suggest that an extremely thin precursor film is presented at the leading edge of the solid‐liquid contact line at low temperatures, and the practical wetting process is the movement of the apparent contact line on the precursor film. [ \n \n 39 \n , \n 40 \n , \n 41 \n \n ] Hence, the reduction of driving force for droplets bouncing off the surface may be due to the freezing of precursor film induced by low temperature, which changes the liquid‐solid interface into a liquid‐ice interface, resulting in an increasing adhesion dissipation. Unlike A‐20, although the A‐30 sample has a longer retraction time of 17.4 ms, the droplet can still detach from the surface. It is inferred that the droplet contraction induced by certain asymmetric structures provides an additional driving force and compensates for the adhesive dissipation caused by low temperature. Unfortunately, the droplet on the A‐40 sample could not escape from the surface in spite of the retraction time is only 13.3 ms. It can be considered that the bouncing behavior of droplets under low temperatures is determined by the competitive relationship between the contraction force induced by asymmetrical structure and the freezing rate of precursor film induced by low temperature. That is, the droplet can successfully detach from the low‐temperature surface when the contraction effect raised by the asymmetrical structure is greater than the liquid‐ice adhesion effect caused by precursor film freezing. Subsequently, the A‐30 sample with excellent dynamic non‐wettability under low temperatures is further investigated at −30 °C. The results clarify that the decrease in temperature ascends the energy dissipation during the retraction stage, resulting in a reduction in bouncing height from 6.32 to 5.73 mm with a retraction time of 44.4 ms, as shown in Figure 3d . Surprisingly, the droplet can still detach from the A‐30 sample at −40 °C. Although the bouncing height has been reduced to 3.73 mm, its droplet retraction time has been actually shortened to 27.3 ms. This may be due to the enlargement of viscosity and tension of water molecules at the liquid‐ice interface triggered by low temperatures, promoting the contraction behavior of droplets. However, the augment of viscosity and tension is still hard to compensate for the adhesion dissipation caused by the precursor film icing occurring at larger areas, leading to a weakening of dynamic non‐wettability under low temperatures. This conclusion is also verified by the bouncing test of A‐20 and A‐40 samples in the Supporting Information (Figures S1 and S2 , Supporting Information). Further analysis of energy dissipation for the A‐30 sample at various low temperatures reveals that the continuous decrease of temperature elevates the adhesion dissipation energy from 2.41 to 2.97 µJ, confirming that adhesion dissipation is still the main type of surface energy dissipation at low temperatures, as shown in Table \n \n 1 \n . Furthermore, the viscous dissipation is basically maintained at ≈0.7 µJ below −10 °C (Figure S3 , Supporting Information). This restates that the influence of temperature on the droplet movement for the A‐30 sample is realized through regulating the contraction behavior induced by the asymmetric structure and the liquid‐ice adhesion induced by the precursor film icing at the interface instead of controlling the viscosity and tension induced by interfacial heat transfer variation. Table 1 Energy dissipation of A‐30 sample surface at different temperatures. Temperature \n E total \n [µJ] \n E k ’ [µJ] \n E adhesion \n [µJ] \n E V \n [µJ] \n ε [µJ] Room Temperature 3.06 0.93 2.41 0.65 0.23 −10 °C 3.48 0.51 2.75 0.73 0.13 −20 °C 3.50 0.49 2.80 0.70 0.12 −30 °C 3.54 0.45 2.84 0.70 0.11 −40 °C 3.70 0.29 2.97 0.73 0.07 John Wiley & Sons, Ltd. 2.4 Mechanical Behavior of Droplet on Asymmetric Structures Considering the remarkable effect of adhesive dissipation on droplet dynamic non‐wetting process, the contact behavior of droplets on the asymmetric structure is further investigated by monitoring the adhesive force of droplets moving along the asymmetric structure. Therefore, the hydrophobicity and adhesion properties are isotropic due to the uniform micro‐nanostructure on the superhydrophobic plate. Only two directions perpendicular to each other are selected for adhesive force evaluation. Moreover, the movement direction of the droplet facing the slop of the asymmetric microstructure is denoted as direction 1, while the opposite direction is defined as direction 2. Meanwhile, the directions along the ridge line of the asymmetric microstructure are denoted as directions 3 and 4, respectively, as shown in Figure \n \n 4 a . Figure 4 Analysis of droplet mechanical behavior on asymmetric structures. a) Diagram of droplet adhesion test. b) Adhesion strength of droplets on superhydrophobic plate along directions 1 and 3. c) Adhesion strength of droplets on A‐20 along directions 1 and 2. d) Adhesion strength of droplets on A‐20 along directions 3 and 4. e) Adhesion strength of droplets on A‐30 along directions 1 and 2. f) Adhesion strength of droplets on A‐30 along directions 3 and 4. g) Adhesion strength of droplets on A‐40 along directions 1 and 2. h) Adhesion strength of droplets on A‐40 along directions 3 and 4. j) Force analysis of droplet on asymmetric structures. i) Diagram of droplet adhesion behavior on different superhydrophobic surfaces. The adhesion forces of droplets moving in two vertical directions on the superhydrophobic plate are 0.0186 N and 0.0193 N. It substantiates that the superhydrophobic micro‐nanostructures obtained by electrodeposition have isotropic wettability, as shown in Figure 4b . However, the adhesion force of the droplet moving along direction 1 on the A‐20 sample is 0.0006 N, which is only 13% of that in the opposite direction, as shown in Figure 4c . It is worth noting that the droplet adhesion forces are 0.0005 N and 0.00056 N in directions 3 and 4, respectively when droplets move along the ridge line of microstructures, which is similar to the droplet adhesion force in direction 1, as displayed in Figure 4d . Additionally, the adhesion force of the droplet moving toward direction 1 on the A‐30 sample is 0.0012 N, which is 17.9% of that in the opposite direction (Figure 4e ). Interestingly, the adhesive force of droplets rolling along the ridge line is between 0.0011 N and 0.0018 N, which is slightly higher than that moving toward the slop side, as demonstrated in Figure 4f . Notably, the adhesion forces of droplets on the A‐30 sample are always higher than that on the A‐20 sample in various directions. This means that an A‐30 sample with a lower asymmetry degree brings a larger solid‐liquid contact area, increasing the adhesion effect between the droplet and the surface. Moreover, the adhesive force is measured to be 0.006 N when the droplet moves toward direction 1 on the A‐40 sample, while the adhesive force of the droplet rolling in the opposite direction is 0.0099 N, as illustrated in Figure 4g . Simultaneously, the adhesive force of the droplets moving along the ridge line of the microstructure is only ≈0.004 N (Figure 4h ). The higher adhesion force on A‐40 sample re‐confirms that the decrease in asymmetry degree leads to a significant enlargement in the solid‐liquid interface, which is also consistent with the conclusion obtained from Figure 2 . Generally, the adhesive force of a droplet moving toward direction 1 is always smaller than that of a droplet rolling in the reverse direction on various asymmetry microstructures, demonstrating a tendency of the droplet to slide toward the slop of asymmetry microstructure. Nevertheless, the adhesion force of droplets shifting along the ridge line of the microstructure is similar. Considering the adhesion behavior of droplets on different surfaces, the force model of droplets is simplified, as shown in Figure 4i . The average droplet adhesion of the superhydrophobic plate is up to 0.0379 N, nevertheless, the adhesion force gradually increases with the decrease of asymmetry degree on asymmetric micro‐nanostructure surfaces (from 0.0016 N of A‐20 to 0.0060 N of A‐40). Relevant research reveals that the droplet is also subjected to the horizontal Laplace force induced by the surface tension in addition to adhesion force during the impact process, which can be expressed as: [ \n \n 42 \n \n ] \n \n (12) \n P L x = 2 γ sin θ i R \n where γ is liquid surface tension, R is the radius of curvature for a droplet edge and θ i \n is the contact angle at the edge. The lower the contact angle of the droplet, the greater the horizontal Laplace force. The contact angle at both sides of the droplet increases with the reduction of the asymmetry degree. Particularly, the deviation between contact angles does not vary linearly, where the contact angle difference on the A‐30 surface is higher by 11.69°, followed by the A‐20 sample with a deviation value of 9.95°, and the A‐40 sample indicates a lower difference value of 8.91°, as delineated in Figure 4j . Taking into account the distinction of horizontal Laplace force induced by contact angle deviation, the total horizontal Laplace force during the droplet retraction stage can be defined as:\n \n (13) \n P L = P L 1 − P L 2 = 2 γ sin θ 1 R 1 − 2 γ sin θ 2 R 2 \n \n Therefore, the A‐30 surface with a larger contact angle difference generates a greater contraction force (which is the contraction driving force described in the previous section). Considering that the A‐30 sample, which can promote the droplet bounce off the surface at low temperatures, possesses a higher adhesion force than the A‐20 sample, it is confirmed that the horizontal Laplace force induced by asymmetric structure is the main factor determining whether the droplet can bounce off the surface. Based on the above analysis and Equation 3 , the energy variation process of a droplet from the maximum spreading stage to bouncing off the surface can be expressed as:\n \n (14) \n E k + E d = E ′ k + E ′ d + E a d h e n s i o n + E V − E p \n where E p \n is the energy dissipation driven by Laplace force. Therein\n \n (15) \n E d = 1 4 π D max 2 σ ( 1 − cos θ ) \n \n \n (16) \n E ′ d = 4 3 π R 0 3 σ \n \n \n (17) \n E a d h e s i o n = ∫ ∫ f ¯ R d α d R = ∫ 0 2 π d α ∫ 0 R f ¯ R d R \n \n \n (18) \n E p = ∫ ∫ 2 γ sin θ 1 R 1 − 2 γ sin θ 2 R 2 R d α d R = ∫ 0 2 π d α ∫ 0 R 2 γ sin θ 1 R 1 − 2 γ sin θ 2 R 2 R d R \n where σ is the surface tension, D max \n is the maximum spreading diameter and the f ¯ is the average droplet adhesion on the cryogenic substrate. Subsequently, the E k ’ needs to be larger than zero if the droplet can successfully bounce off the surface. Hence, a dynamic criterion for droplets to bounce off the surface at low temperatures is initially established:\n \n (19) \n ∫ 2 π 0 d α ∫ R 0 2 γ sin θ 1 R 1 − 2 γ sin θ 2 R 2 RdR − ∫ 2 π 0 d α ∫ R 0 f ¯ RdR ≥ μ V 0 δ 2 τ Ω + 4 3 π R 0 3 σ − 1 4 π D max 2 σ ( 1 − cos θ ) \n \n 2.5 Evolution of Ice Nucleation on Asymmetric Structures Considering that the icing rate of precursor film on asymmetric micro‐nanostructures also plays an important role in droplet adhesion behavior under low temperatures, the icing process is evaluated by the nucleation rate in order to further analyze the effect of temperature on the icing behavior of precursor film on A‐30 sample with excellent dynamic non‐wettability: [ \n \n 43 \n \n ] \n \n (20) \n J ϕ = Φ K ( T s ) e − Δ G h e t e k B T s \n \n Ф is the contact fraction of the solid‐liquid interface and is defined as:\n \n (21) \n Φ = cos θ + 1 cos θ r + 1 \n where θ is the apparent contact angle of sample A‐30 (taken as 170°), θ r \n is the intrinsic contact angle of [CH 3 (CH 2 ) 16 COO] 3 Ce surface (set as 130°), T s \n is the substrate temperature, ΔG hete \n is critical nuclear barrier of non‐uniform nucleation at the solid‐liquid interface, k B is a Boltzmann constant of 1.38 × 10 −23 J K −1 , and the K is the diffusion kinetic flux of water molecules at the ice‐water interface which can be expressed as:\n \n (22) \n K T s = k B T s n h ⋅ e − Δ F d i f f k B T s \n \n Therein, n is the number density of water molecules at the ice‐water interface (3.34 × 10 28 ), h is the Planck constant of 6.63 × 10 −34 J s −1 , ΔF diff \n is the diffusion activation energy of water molecules across the ice‐water interface: [ \n \n 44 \n \n ] \n \n (23) \n Δ F d i f f = k B E T 2 T − T R 2 \n where E is defined as 892 K and T R \n is 118 K. [ \n \n 45 \n \n ] \n The critical nuclear barrier ( ΔG ) of the ice core is a function of temperature and interface energy. According to classical nucleation theory, the relationship between the free energy barrier of non‐uniform nucleation ( ΔG hete \n ) and the homogeneous nuclear energy barrier ( ΔG homo \n ) is as follows:\n \n (24) \n Δ G h e t e = Δ G hom o ⋅ f \n \n \n (25) \n Δ G hom o = 16 π γ I W 3 3 Δ G ν 2 \n \n \n (26) \n γ I W = 28.0 + 0.25 T s − 273.15 \n \n \n (27) \n Δ G ν = T m − T s T m Δ H ν \n where γ IW \n is the interfacial tension of the ice‐water interface, ΔG V \n is the difference in volume free energy between the ice and water phases, T m \n is the melting temperature of ice at standard atmospheric pressure (273.15 K), ΔH V \n is the volumetric enthalpy during water melting process (2.87 × 10 8 J m −3 ), and f can be calculated by: [ \n \n 43 \n \n ] \n \n (28) \n f = 1 2 + 1 2 1 − m x w + x 3 2 2 − 3 x − m w + x − m w 3 + 3 m x 2 2 x − m w − 1 \n \n \n (29) \n m = cos θ I W \n \n \n (30) \n x = R a R c \n \n \n (31) \n w = 1 + x 2 − 2 x m \n \n \n (32) \n cos θ I w = γ 1 cos θ 1 − γ w cos θ w γ I W \n \n \n (33) \n r c = − 2 γ I W Δ G ν \n where θ IW \n is the contact angle of the surface with roughness R a \n in the supercooled water, which is basically equivalent to the intrinsic contact angle of the superhydrophobic surface and R c \n is the critical nucleation size. The calculation results indicate that the ice nucleation rate at the solid‐liquid interface increases exponentially with the decrease in temperature, as shown in Table \n \n 2 \n . The ice nucleation rate is lower than 1.94 × 10 −13 at temperatures above −20 °C, meaning that only several water molecules can nucleate at the interface. However, the ice nucleation rate rises sharply to 2.79 × 10 18 as the temperature further drops below −30 °C, demonstrating a wide range of ice behavior exists in the precursor film. The higher icing rate significantly increases the energy dissipation during the droplet impacting process and greatly weakens the non‐wetting characteristics of the superhydrophobic surface at low temperatures. Table 2 Ice nucleation rate of A‐30 samples at various temperatures. Temperature [k] γ \n IW \n \n Δ G v \n [× 10 7 ] \n R c \n [× 10 −9 ] \n f \n \n Δ G hom o [× 10 −19 ] Δ F diff \n [× 10 −20 ] \n J Φ \n 263.15 0.0255 1.05 4.86 0.8216 25.2 4.05 ≈0 253.15 0.0230 2.10 2.19 0.8215 4.62 4.32 1.94 × 10 −13 \n 243.15 0.0205 3.15 1.30 0.8215 1.45 4.65 2.79 × 10 18 \n 233.15 0.0180 4.20 0.86 0.8214 0.55 5.05 7.69 × 10 26 \n John Wiley & Sons, Ltd. Subsequently, the icing process of water molecules on the superhydrophobic asymmetric structure is simulated by the molecular dynamics method in order to further explore the icing mechanism of the precursor film at the interface. The specific criterion for the nucleation process is provided in the Supporting Information. The simulation results show that nucleation sites appear at multiple locations above the superhydrophobic plate, and the mixed area of cubic ice (marked by blue molecules) and hexagonal ice (marked by yellow molecules) is formed. Afterward, several water molecules near the initial nucleation position are rearranged as the nucleation progresses, resulting in an abundant water molecule regularly arrayed in the vertical direction, as shown in the enlarged box (marked by blue) in Figure \n 5 a . Additionally, there are a certain number of water molecules regularly arranged in both the vertical direction and the general direction (illustrated in the enlarged box marked by orange) on the A‐20 surface (Figure 5b ). Diversely, only a partial of arrayed water molecules can be discovered in the vertical direction on the A‐40 surface (Figure 5d ). In particular, the A‐30 surface exhibits a frozen state with completely disordered, and few arrayed water molecules can be observed from any direction (Figure 5c ). The corresponding nucleation temperature and nucleation time reveal that the nucleation time of the superhydrophobic plate is 89.5 ns, which is slightly lower than that of the A‐20 surface (89.6 ns), as depicted in Figure 5e and Figure S4 (Supporting Information). However, the nucleation time of 88.7 ns on the A‐40 surface is significantly shorter than that on a superhydrophobic plate, while the ice nucleation time on the A‐30 surface is up to 91 ns. Figure 5 Evaluation of ice nucleation process on superhydrophobic surfaces. a) Molecular dynamics calculation of ice nucleation process on a superhydrophobic plate. b) Molecular dynamics calculation of ice nucleation process on A‐20 surface. c) Molecular dynamics calculation of ice nucleation process on A‐30 surface. d) Molecular dynamics calculation of ice nucleation process on A‐40 surface. e) Nucleation times and nucleation temperature for various superhydrophobic surfaces. f) Nucleation rate of various superhydrophobic surfaces. g) Statistics on the quantity of cubic ice and hexagonal ice on different superhydrophobic surfaces. Considering that the icing process is essentially a regular arrangement of water molecules, ice tends to nucleate and grow rapidly along an easy growth direction, and the change of growth orientation is to adapt to the limitation of nucleation space. This change will disrupt the original growth inertia of ice, hindering the growth of the ice layer along the easy growth direction, and resulting in a delay in nucleation and growth. Compared with the previous simulation for symmetrical structure, [ \n \n 46 \n \n ] the specific asymmetric structure can effectively induce the disordered arrangement of vast water molecules, promoting the continuous adjustment of ice crystal orientation during nucleation, significantly delaying the nucleation process and demonstrating the ability to suppress ice formation (Figure S5 , Supporting Information). The nucleation molecules at different icing times are calculated, as demonstrated in Figure 5f . The superhydrophobic plate exhibits a greater ice growth rate, which is 1.36 times higher than that on the A‐20 surface. Notably, the growth rate on the A‐40 sample is only 34.4% of that on the flat plate, indicating a certain superiority in inhibiting the icing process. Surprisingly, the A‐30 surface presents a lower growth rate, only 24.3% of that on the superhydrophobic plate. On this basis, further analysis of ice crystal type reveals that the cubicity (the proportion of cubic ice in all molecules) of ice molecules on the superhydrophobic plate is 60.3%, which is slightly higher than that of 59.57% on the A‐20 surface, as shown in Figure 5g . Meanwhile, the A‐30 surface has a superior cubicity of 66.72%, which is 1.93% larger than that of the A‐40 surface. As is well‐known, ice formation is the process in which water molecules fill the space without creating vacancies as much as possible. The limited growth space is mismatched with the ice type due to the unique spatial structure and growth orientation of cubic ice and hexagonal ice. [ \n \n 47 \n \n ] Hexagonal ice exhibits a preferred growth tendency due to its anisotropic structure, and its nucleation energy barrier is much lower than that of cubic ice. Therefore, it is affirmed that the specific asymmetric structure (A‐30) can effectively delay the nucleation process while inhibiting the growth of ice crystals by simultaneously perturbing the directional arrangement of water molecules and increasing the content of cubic ice. This is why the nucleation probability of droplets on the A‐30 surface is lower at low temperatures during the impact process. Generally, asymmetric structures can reduce the probability of ice formation in surface precursor films by suppressing the nucleation process of water molecules. Whereafter, less area of the precursor film is frozen due to a lower icing probability, leading to fewer liquid ice interfaces and a lower adhesion. It is considered that the ice suppression characteristics of asymmetric structure not only reduce the adhesion force at the solid‐liquid interface but also provide more time for droplets to escape from the surface. Particularly, the specific asymmetric structures (A‐30) can simultaneously reduce droplet adhesion and increase the driving force during the droplet retraction stage by enhancing the horizontal Laplace force. This also confirms that the asymmetry of the structure can promote the self‐ejection effect of the droplet at the multi‐scale level, effectively improving the dynamic non‐wetting performance of the surface under low temperatures."
} | 11,773 |
23423583 | PMC3575075 | pmc | 124 | {
"abstract": "In this study, we present a highly configurable neuromorphic computing substrate and use it for emulating several types of neural networks. At the heart of this system lies a mixed-signal chip, with analog implementations of neurons and synapses and digital transmission of action potentials. Major advantages of this emulation device, which has been explicitly designed as a universal neural network emulator, are its inherent parallelism and high acceleration factor compared to conventional computers. Its configurability allows the realization of almost arbitrary network topologies and the use of widely varied neuronal and synaptic parameters. Fixed-pattern noise inherent to analog circuitry is reduced by calibration routines. An integrated development environment allows neuroscientists to operate the device without any prior knowledge of neuromorphic circuit design. As a showcase for the capabilities of the system, we describe the successful emulation of six different neural networks which cover a broad spectrum of both structure and functionality.",
"introduction": "1 Introduction By nature, computational neuroscience has a high demand for powerful and efficient devices for simulating neural network models. In contrast to conventional general purpose machines based on a von-Neumann architecture, neuromorphic systems are, in a rather broad definition, a class of devices which implement particular features of biological neural networks in their physical circuit layout (Mead, 1989 ; Indiveri et al., 2009 ; Renaud et al., 2010 ). In order to discern more easily between computational substrates, the term emulation is generally used when referring to neural networks running on a neuromorphic back-end. Several aspects motivate the neuromorphic approach. The arguably most characteristic feature of neuromorphic devices is inherent parallelism enabled by the fact that individual neural network components (essentially neurons and synapses) are physically implemented in silico . Due to this parallelism, scaling of emulated network models does not imply slowdown, as is usually the case for conventional machines. The hard upper bound in network size (given by the number of available components on the neuromorphic device) can be broken by scaling of the devices themselves, e.g., by wafer-scale integration (Schemmel et al., 2010 ) or massively interconnected chips (Merolla et al., 2011 ). Emulations can be further accelerated by scaling down time constants compared to biology, which is enabled by deep submicron technology (Schemmel et al., 2006 , 2010 ; Brüderle et al., 2011 ). Unlike high-throughput computing with accelerated systems, real-time systems are often specialized for low power operation (e.g., Farquhar and Hasler, 2005 ; Indiveri et al., 2006 ). However, in contrast to the unlimited model flexibility offered by conventional simulation, the network topology, and parameter space of neuromorphic systems are often dedicated for predefined applications and therefore rather restricted (e.g., Merolla and Boahen, 2006 ; Serrano-Gotarredona et al., 2006 ; Akay, 2007 ; Chicca et al., 2007 ). Enlarging the configuration space always comes at the cost of hardware resources by occupying additional chip area. Consequently, the maximum network size is reduced, or the configurability of one aspect is decreased by increasing the configurability of another. Still, configurability costs can be counterbalanced by decreasing precision. This could concern the size of integration time steps (Imam et al., 2012a ), the granularity of particular parameters (Pfeil et al., 2012 ), or fixed-pattern noise affecting various network components. At least the latter can be, to some extent, moderated through elaborate calibration methods (Neftci and Indiveri, 2010 ; Brüderle et al., 2011 ; Gao et al., 2012 ). In this study, we present a user-friendly integrated development environment that can serve as a universal neuromorphic substrate for emulating different types of neural networks. Apart from almost arbitrary network topologies, this system provides a vast configuration space for neuron and synapse parameters (Schemmel et al., 2006 ; Brüderle et al., 2011 ). Reconfiguration is achieved on-chip and does not require additional support hardware. While some models can easily be transferred from software simulations to the neuromorphic substrate, others need modifications. These modifications take into account the limited hardware resources and compensate for fixed-pattern noise (Brüderle et al., 2009 , 2010 , 2011 ; Kaplan et al., 2009 ; Bill et al., 2010 ). In the following, we show six more networks emulated on our hardware system, each requiring its own hardware configuration in terms of network topology and neuronal as well as synaptic parameters.",
"discussion": "4 Discussion We have successfully implemented a variety of neural microcircuits on a single universal neuromorphic substrate, which is described in detail by Schemmel et al. ( 2006 ). All networks show activity patterns qualitatively and to some extent also quantitatively similar to those obtained by software simulations. The corresponding reference models found in literature have not been modified significantly and network topologies have been identical for hardware emulation and software simulation, if not stated otherwise. In particular, the emulations benefit from the advantages of our neuromorphic implementation, namely inherent parallelism and accelerated operation compared to software simulations on conventional von-Neumann machines. Previous accounts of networks implemented on the Spikey system include computing with high-conductance states (Kaplan et al., 2009 ), self-stabilizing recurrent networks (Bill et al., 2010 ), and simple emulations of cortical layer 2/3 attractor networks (Brüderle et al., 2011 ). In this contribution, we have presented a number of new networks and extensions of previous implementations. Our synfire chain implementation achieves reliable signal propagation over years of biological time from one single stimulation, while synchronizing and filtering these signals (Section 3.1 ). Our extension of the network from Bill et al. ( 2010 ) to exhibit asynchronous irregular firing behavior is an important achievement in the context of reproducing stochastic activity patterns found in cortex (Section 3.2 ). We have realized soft winner-take-all networks on our hardware system (Section 3.3 ), which are essential building blocks for many cortical models involving some kind of attractor states [e.g., the decision-making model by Soltani and Wang ( 2010 )]. The emulated cortical attractor model provides an implementation of working memory for computation with cortical columns (Section 3.4 ). Additionally, we have used the Spikey system for preprocessing of multivariate data inspired by biological archetypes (Section 3.5 ) and machine learning (Section 3.6 ). Most of these networks allocate the full number of neurons receiving input from one synapse array on the Spikey chip, but with different sets of neuron and synapse parameters and especially vastly different connectivity patterns, thereby emphasizing the remarkable configurability of our neuromorphic substrate. However, the translation of such models requires modifications to allow execution on our hardware. The most prominent cause for such modifications is fixed-pattern noise across analog hardware neurons and synapses. In most cases, especially when population rate coding is involved, it is sufficient to compensate for this variability by averaging spiking activity over many neurons. For the data decorrelation and machine learning models, we have additionally trained the synaptic weights on the chip to achieve finer equilibration of the variability at critical network nodes. Especially when massive downscaling is required in order for models to fit onto the substrate, fixed-pattern noise presents an additional challenge because the same amount of information needs to be encoded by fewer units. For this reason, the implementation of the cortical attractor memory network required additional heuristic activity fitting procedures. The usability of the Spikey system, especially for neuroscientists with no neuromorphic engineering background, is provided by an integrated development environment. We envision that the configurability made accessible by such a software environment will encourage a broader neuroscience community to use our hardware system. Examples of use would be the acceleration of simulations as well as the investigation of the robustness of network models against parameter variability, both between computational units and between trials, as, e.g., published by Brüderle et al. ( 2010 ) and Schmuker et al. ( 2011a ). The hardware system can be efficiently used without knowledge about the hardware implementation on transistor level. Nevertheless, users have to consider basic hardware constraints, as, e.g., shared parameters. Networks can be developed using the PyNN metalanguage and optionally be prototyped on software simulators before running on the Spikey system (Brüderle et al., 2009 ; Davison et al., 2009 ). This rather easy configuration and operation of the Spikey chip allows the implementation of many other neural network models. There exist also boundaries to the universal applicability of our hardware system. One limitation inherent to this type of neuromorphic device is the choice of implemented models for neuron and synapse dynamics. Models requiring, e.g., neuronal adaptation or exotic synaptic plasticity rules are difficult, if not impossible to be emulated on this substrate. Also, the total number of neurons and synapses set a hard upper bound on the size of networks that can be emulated. However, the next generation of our highly accelerated hardware system will increase the number of available neurons and synapses by a factor of 10 3 , and provide extended configurability for each of these units (Schemmel et al., 2010 ). The main purpose of our hardware system is to provide a flexible platform for highly accelerated emulation of spiking neuronal networks. Other research groups pursue different design goals for their hardware systems. Some focus on dedicated hardware providing specific network topologies (e.g., Merolla and Boahen, 2006 ; Chicca et al., 2007 ), or comprising few neurons with more complex dynamics (e.g., Chen et al., 2010 ; Grassia et al., 2011 ; Brink et al., 2012 ). Others develop hardware systems of comparable configurability, but operate in biological real-time, mostly using off-chip communication (Vogelstein et al., 2007 ; Choudhary et al., 2012 ). Purely digital systems (Merolla et al., 2011 ; Furber et al., 2012 ; Imam et al., 2012a ) and field-programmable analog arrays (FPAA; Basu et al., 2010 ) provide even more flexibility in configuration than our system, but have much smaller acceleration factors. With the ultimate goal of brain size emulations, there exists a clear requirement for increasing the size and complexity of neuromorphic substrates. An accompanying upscaling of the fitting and calibration procedures presented here appears impractical for such orders of magnitude and can only be done for a small subset of components. Rather, it will be essential to step beyond simulation equivalence as a quality criterion for neuromorphic computing, and to develop a theoretical framework for circuits that are robust against, or even exploit the inherent imperfections of the substrate for achieving the required computational functions."
} | 2,906 |
33859194 | PMC8050072 | pmc | 125 | {
"abstract": "Expanding the portfolio of products that can be made from lignin will be critical to enabling a viable bio-based economy. Here, we engineer Pseudomonas putida for high-yield production of the tricarboxylic acid cycle-derived building block chemical, itaconic acid, from model aromatic compounds and aromatics derived from lignin. We develop a nitrogen starvation-detecting biosensor for dynamic two-stage bioproduction in which itaconic acid is produced during a non-growth associated production phase. Through the use of two distinct itaconic acid production pathways, the tuning of TCA cycle gene expression, deletion of competing pathways, and dynamic regulation, we achieve an overall maximum yield of 56% (mol/mol) and titer of 1.3 g/L from p -coumarate, and 1.4 g/L titer from monomeric aromatic compounds produced from alkali-treated lignin. This work illustrates a proof-of-principle that using dynamic metabolic control to reroute carbon after it enters central metabolism enables production of valuable chemicals from lignin at high yields by relieving the burden of constitutively expressing toxic heterologous pathways.",
"introduction": "Introduction Valorization of lignin, a complex aromatic heteropolymer and the second most abundant component of terrestrial biomass, will be critical for the economic viability of lignocellulosic biorefineries 1 . Biological upgrading of lignin has been demonstrated with production of aromatic catabolic intermediates 2 – 7 and their derivatives 8 , as well as carbon storage compounds such as polyhydroxyalkanoates (PHAs) 9 , 10 and lipids 11 (Fig. 1 , red boxes). However, the sizes of individual chemical markets are typically at least an order of magnitude smaller when compared to fuel markets. With an estimated billion tons of plant biomass able to be sustainably grown in the United States for lignocellulosic biofuel production 12 , hundreds of millions of tons of lignin-rich feedstock could be available for valorization in the United States alone. Therefore, lignin will need to be converted into a wide array of products to avoid oversaturating individual chemical markets and to replace petroleum-derived incumbent molecules. Fig. 1 Biological upgrading of lignin by funneling depolymerized lignin aromatics toward value added products up and downstream of central metabolism. Solid colored and black arrows indicate known metabolic pathway steps for conversion of aromatic intermediates into central metabolites, with dotted black arrows indicating predicted metabolic pathway steps. Dotted gray lines indicate heterologous pathways to convert aromatic intermediates to value-added aromatic derivatives. Red boxed compounds are those whose production from deconstructed lignin have been demonstrated, including itaconate from this study. Acronyms used above: 4-HB (4-hydroxybenzoate), 3-MMA (3-methylmuconate), 2,5-PDCA (2,5-pyridinedicarboxylate), 2,4-PDCA (2,4-pyridinedicarboxylate), β-KA (β-ketoadipate), 2-HMSA (2-hydroxymuconate semialdehyde), 4-OMA (4-oxalomesaconate). To increase the portfolio of products that can be made from lignin, additional parts of metabolism will need to be targeted. The tricarboxylic acid (TCA) cycle is a potential source of valuable chemicals including succinate and citrate, but it has not yet been harnessed for lignin valorization. Indeed, TCA cycle-derived chemicals are ideal products for lignin valorization because aromatic carbon is typically funneled directly into this part of metabolism. Itaconic acid is an unsaturated dicarboxylic acid derived from cis-aconitate in the TCA cycle, with industrial uses including as an acrylate alternative and for the production of polymers. 13 Itaconic acid has been produced industrially from simple sugars, primarily glucose, since the 1950s 14 , 15 , and its potential to functionally replace several petroleum-derived commodity chemicals was highlighted by its selection as one of the top bio-based platform chemicals in several reports, including a 2004 United States Department of Energy report. 16 However, the relatively high cost of sugars makes itaconic acid production expensive, limiting it to use as a specialty chemical. Using cheap and abundant feedstocks, such as lignin, has the potential to reduce production costs 17 and enable much broader industrial use of itaconic acid. The saprophytic bacterium Pseudomonas putida KT2440 is a microbe of industrial interest 18 , 19 due to its robust metabolism 20 and tolerance to xenobiotics. 21 – 24 \n P. putida also has the ability to tolerate and catabolize a wide-range of aromatic compounds 25 which led to its recent use in upgrading depolymerized lignin into PHAs 10 , 26 , cis,cis- muconic acid 3 , 5 , 27 , and other intermediates in aromatic catabolism 7 . In P. putida , p- hydroxyphenyl (H) and guiacyl (G) lignin-derived aromatics are funneled via the β-ketoadipate pathway to acetyl-CoA and succinate (Fig. 2a ). This direct route to key TCA cycle intermediates suggests that high yields of TCA cycle-derived products such as itaconic acid should be possible from lignin. For instance, because the lignin-derived aromatic compound p -coumaric acid is catabolized into one succinate and two acetyl-CoA molecules, the theoretical maximum yield of itaconic acid is 1.33 mol itaconic acid/mol p- coumaric acid. Fig. 2 Two-stage production of itaconic acid from p -coumaric acid. a Simplified p -coumaric acid assimilation, β-ketoadipate, and tricarboxylic acid (TCA) cycle pathways in Pseudomonas putida KT2440 with modified or heterologous steps indicated by colored arrows, and connecting metabolites outlined. For simplicity some steps are not included. The cis (red arrow) and trans (green arrow) pathways for itaconic acid are indicated with involved genes, cadA ( cis ) & tad1/adi1 ( trans ) adjacent to the reaction their gene products perform. Isocitrate dehydrogenase activity, provided by the icd & idh gene products, is indicated by a purple arrow. b Simplified polyhydroxyalkanoate (PHA) biosynthesis pathway in P. putida KT2440. The PHA pathway, via fatty acid biosynthesis, competes with the TCA cycle for acetyl-CoA during nitrogen-limited conditions. c Production of itaconic acid from p- coumaric acid in shake flasks by P. putida strains constitutively expressing cadA under nitrogen-limited conditions. Strain name and their unique modifications are indicated above the charts. Cell density (OD600, gray diamonds), residual p -coumaric acid (mM, blue circles), and produced itaconic acid (mM, yellow triangles) are indicated. d Growth rates of P. putida strains containing icd & idh start codon mutations with or without constitutive cadA expression using p -coumaric acid as sole carbon source. Rates were determined by 48-well microtiter plate cultivation. c, \n d Data are represented as mean values ± standard deviation in three replicates. Source data underlying Fig. 2c and d is provided as a Source Data file. Growth phase production of itaconic acid may be challenging because itaconic acid disrupts bacterial growth via inhibition of enzymes in the glyoxylate shunt 28 and citramalate cycle 29 . An alternate approach is to use a two-stage process to decouple growth of the microbial catalyst from conversion of feedstocks to chemicals, which provides solutions to many problems present in growth-associated processes (e.g., product toxicity, slow catalyst growth). 30 Such processes often take advantage of the natural responses to nutrient limitations (e.g., nitrogen, sulfur, phosphate) and environmental shifts (e.g., O 2 limitation, temperature shifts) that prevent microbial growth while maintaining the metabolic reactions of interest. Coupling two-stage processes with dynamic metabolic control has the potential to entirely remodel metabolism. In this study, we engineer P. putida to produce a commercially relevant chemical at high yields and gram-per-liter titers from model aromatic substrates and corn stover-derived, alkali-pretreated lignin. We further develop multiple production pathways and developed a signal-amplified biosensor for two-stage production via dynamic metabolic control to increase efficiency and mitigate toxicity.",
"discussion": "Discussion Until recently, robust itaconic acid production was limited to sugar-utilizing fungi such as A. terreus and U. maydis , but in recent years, bacterial strains have been engineered to produce itaconic acid from glucose 44 , 45 , acetate 46 , and glycerol. 47 However, all efforts to engineer heterologous itaconic acid production relied on expression of the cis -pathway ( cadA ) from A. terreus . 31 In our work, the trans -pathway ( adi1/tad1 ) from U. maydis 41 outperformed the cis -pathway, likely due to the production of a thermodynamically favorable intermediate. Accordingly, the trans -pathway may improve performance in other organisms. Among bacteria, the itaconic acid yield with P. putida , 0.44 g/g from p- coumaric acid, compares well to an engineered E. coli that produced 0.5 g itaconic acid/g glucose. 45 Furthermore, 1.43 g/L itaconic acid was produced from a complex, depolymerized lignin stream, yielding 0.79 g itaconic acid/g detectable aromatic monomers. Of note, the yield during the production phase is substantially higher than the overall yield in all cases, reaching nearly 1.2 mol itaconic acid/mol p- coumaric acid (the theoretical maximum yield is 1.33 mol/mol) using trans - pathway strains (Fig. 4c , Table 2 ). Therefore, advanced feeding strategies with extended production phases could improve yields even further. Finally, this work demonstrates itaconic acid production in an engineered bacterium in minimal salts medium without the use of replicating plasmids, antibiotic selections, or expensive inducer chemicals ( e.g . IPTG), each of which will be critical for a commercial process. Dynamic metabolic control combined with two-stage production is promising approach for biological chemical production 30 , and the biosensor developed here can serve as a master regulator for additional dynamic metabolic control tools such as CRISPRi 48 , targeted proteolysis, and nested dynamic regulatory systems. Our nitrogen biosensor will be a valuable tool for future P. putida metabolic engineering because it is able to tune the amplitude of the nitrogen-starvation induced transcriptional response over an 89-fold range, increasing it by up to 60-fold over the original response. Despite its utility, to the best of our knowledge, lysY has not previously been used for metabolic engineering. The dynamic response of this biosensor, already allowing 200-fold induction, has the potential to be further tuned by altering lysY expression. Although not explicitly demonstrated here, one can modulate levels of sequentially utilized nitrogen sources such as NH 4 and NO 3 for auto-induction of gene expression in single-stage processes following depletion of NH 4 . Finally, nitrogen-responsive σ 54 promoters from P. putida similar to P urtA have been ported to E. coli 36 , suggesting this biosensor can work in other organisms expressing the activator NtrC. In addition to lignin, recent work has demonstrated the potential for P. putida to valorize other biomass streams – including thermochemical wastewater 49 and lignocellulosic sugar hydrolysates 50 . Here, we demonstrated itaconic acid production from multiple substrates that could be found in potential commercial feedstocks, including biodiesel waste (glycerol), plant biomass (sugars, aromatics, acetic acid), and lipids (octanoic acid). Therefore, the technology developed here can serve as a universal platform for the two-phase production of a portfolio of chemicals from alternate carbon sources beyond lignin."
} | 2,948 |
39110686 | PMC11342387 | pmc | 126 | {
"abstract": "Neuromorphic computing\nseeks to replicate the capabilities of parallel\nprocessing, progressive learning, and inference while retaining low\npower consumption by drawing inspiration from the human brain. By\nfurther overcoming the constraints imposed by the traditional von\nNeumann architecture, this innovative approach has the potential to\nrevolutionize modern computing systems. Memristors have emerged as\na solution to implement neuromorphic computing in hardware, with research\nbased on developing functional materials for resistive switching performance\nenhancement. Recently, two-dimensional MXenes, a family of transition\nmetal carbides, nitrides, and carbonitrides, have begun to be integrated\ninto these devices to achieve synaptic emulation. MXene-based memristors\nhave already demonstrated diverse neuromorphic characteristics while\nenhancing the stability and reducing power consumption. The possibility\nof changing the physicochemical properties through modifications of\nthe surface terminations, bandgap, interlayer spacing, and oxidation\nfor each existing MXene makes them very promising. Here, recent advancements\nin MXene synthesis, device fabrication, and characterization of MXene-based\nneuromorphic artificial synapses are discussed. Then, we focus on\nunderstanding the resistive switching mechanisms and how they connect\nwith theoretical and experimental data, along with the innovations\nmade during the fabrication process. Additionally, we provide an in-depth\nreview of the neuromorphic performance, making a connection with the\nresistive switching mechanism, along with a compendium of each relevant\nperformance factor for nonvolatile and volatile applications. Finally,\nwe state the remaining challenges in MXene-based devices for artificial\nsynapses and the next steps that could be taken for future development.",
"introduction": "Introduction In\nthis era of “Big Data”, there is a growing interest\nin exploring solutions based on artificial intelligence (AI) such\nas artificial neural networks (ANNs). However, these AI-software approaches\nare deployed in computing units based on the von Neumann architecture,\nwhich is reaching its inherent performance bottleneck, 1 due to data storage and computing units being physically\nseparated. On the other hand, the human brain is capable of memory\nstorage and learning in the same substrate, using its vast network\nof neurons and synapses to avoid data migration. The schematics in Figure 1 (a) compares the\ntwo computing architectures. Furthermore, the human brain only requires\napproximately 20 W (0.3 kWh per day) to operate, against the ever-increasing\nenergy hungry AI applications such as ChatGPT (260 KWh per day). Therefore,\nusing neuromorphic computing architectures that seek to emulate the\nbrain at the hardware level by mimicking the structure and function\nof biological neural networks in artificial computing systems or analog\ncircuitry is a highly attractive solution. 2 , 3 Figure 1 (a) Comparison\nbetween traditional (digital) von Neumann and (analog)\nneuromorphic computing architectures. (b) Main achievements in MXene\nresearch since discovery, with all MXene-based memristors in red,\ndepicting the number of citations (number of works and respective\ncitations obtained from Scopus, 4 taken\nin June 2024). One of the key components in neuromorphic\ncomputing is the dynamic\nweights of synapses that connect neurons and allow data to be classified,\nemulating the strength of biological synapses. The large-scale assembly\nof such adaptive switches into electronic systems is rapidly evolving\ndue to the recent discovery of the memristor. 5 A memristor is defined as a two-terminal device where resistive\nswitching (RS) allows dynamic conductance states to occur. This effect\nwas already shown to mimic advanced biological learning rules such\nas short/long-term synaptic plasticity, Hebbian learning, and dendritic\nintegration, among others. 1 , 6 − 10 The first memristive devices were based on metal–insulator–metal\nstacks that still display low reproducibility between devices and\ntemporal variations due to the intrinsic stochastic switching mechanism\nof the insulating materials used. 1 , 3 , 11 The integration of low-dimensional materials in memristors\nwas shown to enhance memory and neuromorphic properties such as higher\nswitching control, higher spatial, and temporal reproducibility but\nalso lower power consumption and fabrication cost. 7 , 12 − 14 Besides, the integration of two-dimensional (2D)\nmaterials (e.g., MoS 2 , WSe 2 , WS 2 ,\ngraphene) in memristors 15 provides easy\ndevice scaling, due to their atomic-scale thickness and the ability\nto form van der Waals heterostructures. 7 , 16 , 17 MXenes are a family of 2D materials discovered\nin 2011, 18 composed by transition metal\ncarbonitrides,\ncarbides, and nitrides obtained through etching the A layer from the\nMAX phase. 19 The name MXenes emphasizes\nthe morphological similarity to graphene and that this type of material\nis prepared from a MAX phase precursor. They have found use in a vast\nrange of applications such as water purification, 20 electromagnetic interference shielding, 21 transparent and flexible electrodes, 22 high-performance supercapacitors, biosensors 23 , 24 (e.g., SARS-COV-2 detector, 25 sweat-based\nsensors 26 ), and storage devices. 27 − 29 A timeline of the MXene research is shown in Figure 1 (b). Since 2019, the integration of MXenes\nin these neuromorphic devices has gained significant interest, especially\ndue to the obtained performance enhancements, along with their excellent\ncharge trapping capability and electrical conductivity. 11 , 15 In comparison with 3D materials for the switching layer, MXenes\nbenefit from the reduced dimension, which offers better device integration\nand further decreases operation voltages and power consumption. Furthermore,\nas will be seen in detail below, the introduction of MXenes in memristors 30 − 32 also leads to improved device performance. When compared to other\n2D materials, such as transition metal dichalcogenides (TMDs), MXenes\npresent unique chemical and physical properties such as hydrophilic\nsurfaces that enable high chemical stability, 33 , 34 metallic conduction, better optical transport properties due to\nthe higher density of states at the Fermi level, 35 , 36 and tunable surface terminations, which enable one to tailor properties\nsuch as work function or surface electronegativity. 34 , 36 − 39 Moreover, during the etching and assembly stage, a large number\nof variables can influence the MXene properties, namely, surface terminations,\ninterlayer spacing, flake size, or defects, meaning that even the\nsame MXene material can have different chemical and physical properties, 40 − 43 while at the same time there is an extremely large material family\nto explore. All of these considerations, coupled with versatile and\ninexpensive fabrication methods (spin/spray/dip-coating, printing) 29 , 44 − 46 that prevent unwanted defects and damage typical\nof high-energy deposition methods, 47 place\nMXenes as extremely promising materials for neuromorphic computing\nin-hardware. This review begins with the fabrication methods\nof MXene-based\nmemristors, analyzing the different etching and deposition methods,\nas well as variations in these processes. Then, the characterization\ntechniques more appropriate for each stage of these processes are\ndetailed. The influence of the fabrication process on the switching\nmechanism is analyzed together with the impact of the different tuning\nprocesses during the fabrication stage on these mechanisms. Finally,\nthe most promising applications of the vast MXene family in neuromorphic\napplications are discussed, in particular their performance as artificial\nsynapses, either in volatile or nonvolatile regimes."
} | 1,951 |
33525834 | PMC7289024 | pmc | 128 | {
"abstract": "Methanogens are anaerobic archaea that grow by producing methane gas. These microbes and their exotic metabolism have inspired decades of microbial physiology research that continues to push the boundary of what we know about how microbes conserve energy to grow. The study of methanogens has helped to elucidate the thermodynamic and bioenergetics basis of life, contributed our understanding of evolution and biodiversity, and has garnered an appreciation for the societal utility of studying trophic interactions between environmental microbes, as methanogens are important in microbial conversion of biogenic carbon into methane, a high-energy fuel. This review discusses the theoretical basis for energy conservation by methanogens and identifies gaps in methanogen biology that may be filled by undiscovered or yet-to-be engineered organisms."
} | 212 |
37056967 | PMC10086674 | pmc | 129 | {
"abstract": "Currently, major energy sources such as fossil fuels and nuclear fuels face various issues such as resource depletion, environmental pollution, and climate change. Therefore, there is increasing interest in technology that converts mechanical, heat, vibration, and solar energy discarded in nature and daily life into electrical energy. As various wearable devices have been released in recent years, wearable energy-harvesting technologies capable of self-power generation have garnered attention as next-generation technologies. Among these, triboelectric nanogenerators (TENGs), which efficiently convert mechanical energy into electrical energy, are being actively studied. Textile-based TENG (T-TENGs) are one of the most promising energy harvesters for realizing wearable devices and self-powered smart clothing. This device exhibited excellent wearability, biocompatibility, flexibility, and breathability, making it ideal for powering wearable electronic devices. Most existing T-TENGs generate energy only in the intentional vertical contact mode and exhibit poor durability against twisting or bending deformation with metals. In this study, we propose a sandwich-structured T-TENG (STENG) with stretchability and flexibility for use in wearable energy harvesting. The STENG is manufactured with a structure that can maintain elasticity and generate a maximum voltage of 361.4 V and current of 58.2 μA based on the contact between the upper and lower triboelectric charges. In addition, it exhibited a fast response time and excellent durability over 5000 cycles of repetitive pushing motions. Consequently, the STENG could operate up to 135 light-emitting diodes (with output) without an external power source, and as an energy harvester, it could successfully harvest energy for various operations. These findings provide textile-based power sources with practical applications in e-textiles and self-powered electronics.",
"conclusion": "Conclusions We fabricated a stretchable and fully flexible STENG for use in wearable energy-harvesting devices. Based on the friction generated between the micropatterned EcoFlex and acetate, the STENG harvested mechanical energy in the vertical contact, stretching, and rubbing modes. The output voltages and currents of 361.4 V and 58.2 μA in the contact–separate mode, 166.2 V and 23 μA in the stretching mode, and 119.5 V and 17 μA in the rubbing mode were obtained. This resulted in a 250% improvement in the output performance compared with the flat EcoFlex-based STENG without micropatterns. In addition, excellent durability was demonstrated without a drop in the output power during repetitive pushing motions for 5000 cycles. The STENG operated up to 135 LEDs using its output power in the vertical contact, stretching, and rubbing modes. These findings can provide a textile-based power source for practical applications in e-textiles and self-powered electronics in the future.",
"introduction": "Introduction With the rapid development of the Internet of Things (IoT) technology, the demand for portable and wearable electronic products has increased. With the emergence of the wearable electronics market, studies on energy technologies that can supply power to portable/small electronic products have been conducted. 1–3 In particular, energy-harvesting technology that regenerates mechanical energy discarded in the surrounding environment ( e.g. , body energy, vibration, heat, and electrons) into electrical energy is in the spotlight. Among these, studies on triboelectric nanogenerators (TENGs), which efficiently convert mechanical energy into electrical energy, are being actively conducted. 4,5 TENG produce electrical outputs based on the combined effects of electrostatic induction and contact–frictional electrification. 6,7 In triboelectric electrification, materials are positively or negatively charged when they come into contact with or are separated by an external force. Based on the type of friction material and electrode position, TENGs are divided into four operating modes: vertical contact–separation, lateral sliding, single-electrode, and free-standing. In the vertical contact–separation mode, the TENG comprises two different friction materials and generates an electrical output based on the contact and separation of the friction materials. In this mode, the TENG is attached to the soles of most shoes to generate electricity based on the movement of people in daily life ( e.g. , walking or running) and can be easily manufactured at a low cost with a simple structure. The operation of the lateral sliding mode is similar to that of the vertical contact separation mode; however, this mode generates an electrical output based on the movement of parallel translation in a state where two different friction materials are not separated. This mode exhibits a better output performance because it has a larger contact area than the vertical contact mode. The single-electrode mode exhibited the simplest mechanism of action. It requires only one electrode and the friction material can move freely without being constrained. However, the single-electrode mode is suitable for portable and self-powered systems because of its lower output performance compared with other modes. Unlike other modes, the freestanding mode comprises two electrodes and a vertical or horizontal friction material. In this mode, a high electrical output can be obtained because there is no need to maintain contact. 8 In this mode, the friction material moves freely. TENG technology possesses distinct advantages, such as a variety of material choices, productivity, wearability, wide usability, and low manufacturing cost. Recently, several studies have been conducted on textile-based TENGs (T-TENGs) with high performance and wearability, which can efficiently harvest energy based on human body motions. Textiles possess various advantages such as flexibility, elasticity, durability, permeability, light weight, and biocompatibility; therefore, they are used as friction materials for TENGs. 9–11 Moreover, textiles can efficiently harvest energy through friction with materials possessing different electron affinities; the greater the relative difference in the electron affinity, the higher the power generated. T-TENGs can harvest large amounts of energy based on the movements of the human body in daily life, such as arm waving, walking, running, and arm and knee bending. 12–15 Previous studies have indicated that T-TENGs can be coated with polyvinylidene fluoride (PVDF), PTFE, and polydimethylsiloxane (PDMS) to increase the frictional surface area. 16–26 Moreover, a high electrical output has been achieved using metals with hard properties as friction materials, for example, Au, Ag, and Cu. However, previous studies did not consider the durability against complex manufacturing processes and twisting or bending deformation compared with the output power. In particular, the T-TENGs generated high energy only in the intentional vertical contact mode. The generated power was high when the area was large; however, the lifespan was short. 27 Therefore, it is necessary to develop a T-TENG that can efficiently harvest energy in various modes, including the vertical-contact mode, and that does not deteriorate in terms of durability under the effects of shape deformation and external forces. Somkuwar's work 28 proposed a breathable-fabric-based TENG with an open-porosity polydimethylsiloxane coating. To enhance the triboelectric performance and wearable comfort, sacrificial templates, including insoluble NaCl, DBP, and soluble silicone oil, were applied to synergistically construct open porous structures. The open porous structure not only benefits the air permeability but also enhances the triboelectric output owing to the increased contact area through an application experiment. Chung's study 29 proposed a stretchable FTENG using polydimethylsiloxane and 2D-polyester fibers to improve the energy-harvesting performance. An MN-FTENG with a microneedle structure was fabricated using polymethyl methacrylate (PMMA) to develop a wearable device with high elasticity and conductivity. Experiments demonstrated that FTENGs with microstructures had approximately 34–37% higher output voltage and current than FTENGs without microstructures, and that motion detection could be performed based on the movements of large joints such as elbows and knees. Song's study 30 proposed a flexible large-scale fiber-based TENG using a knitted Chinese fabric coated with silver and a PDMS film. To improve the electrical performance, the proposed TENG inserted a pattern on the surface of the PDMS film using rough sandpaper from the microstructured arrays. Through experiments, it was verified that as the degree of surface coating of the sandpaper increased, the effective contact surface expanded to improve the electrical performance, and excessive fine pores could degrade the electrical performance. In this paper, we propose a flexible sandwich-structured T-TENG (STENG) that can harvest energy based on the motion of the human body by effectively using textile strength. To solve the problems associated with previous T-TENGs, one side of the STENG was coated with a micropatterned EcoFlex. Moreover, to generate output power during various operations, the stretchability was improved by attaching the acetate cloth of the winding structure to the textile side of the STENG. Based on the friction generated between the micropatterned EcoFlex coating and acetate, the STENG could harvest mechanical energy in the vertical contact, stretching, and rubbing modes. The STENG generates a maximum voltage of 361.4 V and power of 58.2 μA in the vertical contact mode, resulting in a 250% increase in the output performance compared to the non-patterned planar-EcoFlex-based STENG. Moreover, 135 light-emitting diodes (LEDs) are successfully operated using the output power of the STENG without an external power source, demonstrating excellent durability and potential applicability. The proposed STENG is a self-powered device that can supply power to small portable electronic devices and is expected to be widely used in energy-harvesting systems in the future.",
"discussion": "Results and discussion To prove the optimal output performance of the STENG, various patterns were coated on the surface of EcoFlex, and the outputs were compared. Fig. 4(a–d) show the SEM images of the EcoFlex non-patterned coating, microstructure of the crepe paper, air form, and Teflon cloth pattern. The SEM analysis of the fabricated pattern-based EcoFlex was conducted at 100 μm and 500 μm scale bars, showing the array of patterns, surface texture, and microstructure of the surface. Fig. 4a shows the surface of a textile coated with EcoFlex without a specific pattern. Fig. 4b shows a surface image of the textile coated with EcoFlex in an air-form pattern, and it was confirmed that the surface was irregular and uneven. A STENG using an air-foam-based EcoFlex has fine holes formed by air foam; thus, the air permeability and flexibility are improved. Fig. 4c shows the surface image of a textile coated with EcoFlex and a Teflon cloth pattern with a high negative charge affinity, showing a regular and even array. Fig. 4d shows the surface of a textile coated with microstructured EcoFlex using the microstructure of the crepe paper. The output voltage and current of the STENG were measured in the contact–separation mode by deploying four coating patterns on the surface of the EcoFlex. As shown in Fig. 4(e and f) , a relatively low voltage of 155.9 V and current of 11 μA were evident when EcoFlex without a pattern was used as the negative triboelectric layer. The STENG using an air-foam-based EcoFlex improved the air permeability and flexibility owing to air-foam formation; however, noise was severe owing to excessive micropores and irregular patterns, and the output voltage and current were relatively low. This structure shows a maximum voltage of 172.3 V and an unstable current of 19 μA. In the case of Teflon cloth pattern-based EcoFlex, a stable power with a voltage of 215.7 V and current of 28 μA was determined; however, the output performance was lower than that of the crepe-paper-microstructure-based STENG. Through electrical characterization experiments, it was demonstrated that EcoFlex, based on the microstructure of crepe paper, has a sophisticated geometric structure and regular arrangement, which can improve the frictional area and electrical output and increase the efficiency of the STENG power generation process. Fig. 4 Microstructured EcoFlex-based STENG with a high electrical output. (a–d) SEM images of non-patterned, air foam, Teflon cloth, and crepe paper show a scale bar of 1 μm. (e and f) Comparison between the output voltage and current under different EcoFlex patterns. \n Fig. 5 shows the mechanical and electrical characteristics of the STENG. As shown in Fig. 5a , the tensile strength was measured using a tensile tester (MCT-2150) to evaluate the mechanical properties of the STENG. Tensile tests can be used to measure the maximum stress until the material is segregated by the tensile load generated when it is stretched from both sides. A constant-speed tensile tester was used to obtain the force–strain diagram. As shown in Fig. 5b , the STENG increased the strain at a constant rate and generated a maximum load of 5.8 N. It was not damaged even at a high strain rate because of its highly flexible materials, such as acetate cloth and micropatterned EcoFelx, for the winding structure. The power density was measured by connecting a load resistance of 10–10 15 Ω to the STENG electrode. As shown in Fig. 5c , when the load resistance increased, the output voltage increased and then saturated; however, the output current decreased according to Ohm's law. The output power density is calculated as P = I 2 R . The maximum output power density is obtained when the load resistance is equal to the internal impedance of the STENG. As shown in Fig. 4d , a maximum power density of 792 mW m −2 was obtained at a load resistance of 10 7 Ω. Fig. 5 (a) Experimental setup for the tensile strength test. (b) Load–deformation curve. (c) Output voltage and current based on the external load resistance. (d) Power density across various loading resistors. To demonstrate the excellent mechanical durability of the STENG, the output voltage and current based on repetitive contact–separation motions were measured using a pushing tester. Using a pushing tester (JIPT-100), a constant pressure of 0.1 kgf was applied to the STENG in the vertical-contact separation mode. The pushing tester repeatedly applied pressure to the 1 × 1 cm area of the experimental sample at regular intervals and tested the durability by calculating the average value of the generated voltage and current. As shown in Fig. 6a , when a force of ∼0.1 kgf for 5000 cycles was applied, the output voltage exhibited an error range of up to 0.4 V. As shown in Fig. 6b , in the same experimental environment, the output current exhibited a low error range of up to 0.6 μA. To prove the durability of the STENG in the stretching and rubbing modes, the voltage and current were measured through repetitive motion for 60 s. As shown in Fig. 6 , the output voltage and current of the STENG in the stretching mode showed an error range of up to 20 V and 2.4 μA. Moreover, in the rubbing mode, the STENG showed a maximum voltage of 12 V and an error range of 3.2 μA current. The error was calculated by comparing the initial and later outputs. Therefore, the STENG exhibited excellent mechanical durability and stability as it produced a constant signal without a significant decrease in the electrical output, and its electrical output performance did not deteriorate even after repeated deformation for 5000 cycles. Fig. 6 Mechanical durability test of the STENG. (a) Output voltages of the cycled compressive force on the STENG over 5000 cycles. The voltage output in the initial phase and after 5000 cycles. (b) Current of the cycled compressive force on the STENG over 5000 cycles. Current in the initial phase and after 5000 cycles. To demonstrate the potential of the STENG, the output voltage generated by the motion of a human was measured. As shown in Fig. 7(a–d) , the STENG can obtain power in daily life by attaching it to a commercial coat or shoe sole. 35 Fig. 7a shows a STENG attached to the shoe sole. When a human walks, a constant voltage of ∼50 V is generated by the contact between the STENG and the heel ( Fig. 7b ). As shown in Fig. 7c , the STENG attached to the coat harvests energy based on the movement of the cloth. Fig. 7d shows the output generated by the STENG when the coat was moved, which resulted in an average voltage of 112 V. These experimental results demonstrate that the STENG can harvest biomechanical energy from various types of body movements. As shown in Fig. 7e , using the STENG, it was possible to turn on LEDs connected in series based on the power generated from the motion. Thus, STENG with excellent flexibility, durability, and stability can be used in portable and wearable power systems. Fig. 7 STENG harvesting energy from human motion. (a) STENG attached to the shoe sole. (b) Output voltage generated by the STENG mounted on the shoe. (c) STENG integrated with a coat harvesting energy from the movement of clothes. (d) Output voltage generated by the STENG attached to the cloth. (e) STENG can be used to continuously illuminate several LEDs via tapping."
} | 4,399 |
34051845 | PMC8164749 | pmc | 132 | {
"abstract": "Background During the acetogenic step of anaerobic digestion, the products of acidogenesis are oxidized to substrates for methanogenesis: hydrogen, carbon dioxide and acetate. Acetogenesis and methanogenesis are highly interconnected processes due to the syntrophic associations between acetogenic bacteria and hydrogenotrophic methanogens, allowing the whole process to become thermodynamically favorable. The aim of this study is to determine the influence of the dominant acidic products on the metabolic pathways of methane formation and to find a core microbiome and substrate-specific species in a mixed biogas-producing system. Results Four methane-producing microbial communities were fed with artificial media having one dominant component, respectively, lactate, butyrate, propionate and acetate, for 896 days in 3.5-L Up-flow Anaerobic Sludge Blanket (UASB) bioreactors. All the microbial communities showed moderately different methane production and utilization of the substrates. Analyses of stable carbon isotope composition of the fermentation gas and the substrates showed differences in average values of δ 13 C(CH 4 ) and δ 13 C(CO 2 ) revealing that acetate and lactate strongly favored the acetotrophic pathway, while butyrate and propionate favored the hydrogenotrophic pathway of methane formation. Genome-centric metagenomic analysis recovered 234 Metagenome Assembled Genomes (MAGs), including 31 archaeal and 203 bacterial species, mostly unknown and uncultivable. MAGs accounted for 54%–67% of the entire microbial community (depending on the bioreactor) and evidenced that the microbiome is extremely complex in terms of the number of species. The core microbiome was composed of Methanothrix soehngenii (the most abundant), Methanoculleus sp., unknown Bacteroidales and Spirochaetaceae . Relative abundance analysis of all the samples revealed microbes having substrate preferences. Substrate-specific species were mostly unknown and not predominant in the microbial communities. Conclusions In this experimental system, the dominant fermentation products subjected to methanogenesis moderately modified the final effect of bioreactor performance. At the molecular level, a different contribution of acetotrophic and hydrogenotrophic pathways for methane production, a very high level of new species recovered, and a moderate variability in microbial composition depending on substrate availability were evidenced. Propionate was not a factor ceasing methane production. All these findings are relevant because lactate, acetate, propionate and butyrate are the universal products of acidogenesis, regardless of feedstock. Supplementary Information The online version contains supplementary material available at 10.1186/s13068-021-01968-0.",
"conclusion": "Conclusions In the present research, it was shown that the dominant components in the media (lactate, acetate, propionate or butyrate) subjected to methanogenesis moderately modified the final effect of bioreactor performance in terms of methane production and substrate utilization, whereas strongly affected the methanogenic pathways. Isotopic analysis evidenced different contributions of acetotrophic and hydrogenotrophic pathways for methane production, i.e., acetate and lactate favored the acetotrophic pathway, whereas propionate and butyrate favored the hydrogenotrophic pathway. Most of the 234 MAGs (31 archaeal and 203 bacterial species) were identified as new species. The core microbiome is represented by five MAGs present in high relative abundance (two methanogens: Methanothrix soehngenii and Methanoculleus sp., three bacterial MAGs classified only at high taxonomic level) and 108 other MAGs with a low relative abundance. Considering the relative abundance and/or their predicted functional role (determined according to KEGG pathways), three MAGs were found as propionate specific; four MAGs as lactate specific; four MAGs as butyrate-specific; and three MAGs as acetate-specific microbes. Analyzing the core microbiome and the substrate-specific species, we hypothesize the substrate may first of all change the metabolic activity of the bacteria/methanogens, rather than the composition of the microbial community. This requires confirmation using metatranscriptomic analysis. Interestingly, we did not observe a reduction in methane production in the bioreactor fed with the propionate-rich medium. It may indicate that propionate commonly connected with inhibition of methanogenesis is rather an indicator than the cause of disturbances in anaerobic digestion. All these findings are relevant due to the fact that lactate, acetate, propionate and butyrate are the universal products of acidogenesis, regardless of feedstock.",
"discussion": "Discussion Effects of dominant products of acidogenesis: acetate, butyrate, lactate and propionate on methane formation Using culture-independent techniques and a long-term system, we traced the processing of four dominant non-gaseous products of bacterial acidic fermentations to methane and carbon dioxide for better understanding of the metabolic pathways and syntrophic cooperation between microorganisms in the methane-yielding communities. Our system allows examining the acetogenic and methanogenic stages of anaerobic digestion and helps understanding microbial processes in multi-stage systems processing organic matter to biogas. However, anaerobic digestion of methanogenic sludge inside the bioreactor should also be considered. The media subjected to methanogenesis were dominated by, or contained exclusively, one of the acidic fermentation products: lactate (M1), butyrate (M2), propionate (M3) or acetate (M4). Utilization of the substrate measured as % reduction of substrates COD revealed that acetate and lactate were used by the microbial communities more efficiently than butyrate and propionate, especially when the media contained exclusively one compound. With regard to the efficiency of methane production, the interpretation of the results is rather ambiguous. The highest methane production in Experiment 1 was achieved for butyrate processing with 60% substrate utilization; whereas in Experiment 2, for lactate processing with 74% substrate utilization. The tested substrates had a lower impact on the final performance of the bioreactors than expected. In contrast, large differences were observed in the results of isotopic analyses clearly showing that domination of acetotrophic or hydrogenotrophic pathways of methane synthesis is substrate dependent. Acetate is a direct substrate for acetotrophic methanogens. It explains its efficient utilization and domination of the acetotrophic pathway of methanogenesis confirmed by isotopic analyses. Acetate detected in the effluents from bioreactor M4 in both experiments as well as from the other bioreactors could come from non-utilized substrate or from anaerobic digestion of methanogenic sludge inside the bioreactor. Acetate could also be oxidized to carbon dioxide and hydrogen. However, since acetate oxidation is an endoergic reaction (ΔG 0' = + 94.9 kJ/reaction), it requires syntrophic cooperation with hydrogenotrophic methanogens, which results in ΔG 0' = − 36.3 kJ/reaction. Oxidation of butyrate and propionate are also endoergic reactions, with ΔG 0' = + 48.3 kJ/reaction and ΔG 0' = + 76.0 kJ/reaction, respectively, that become thermodynamically favorable with hydrogenotrophic methanogens, with ΔG 0' = − 17.3 kJ/reaction and ΔG 0' = − 22.4 kJ/reaction, respectively [ 101 ]. This explains the domination of hydrogenotrophic methanogenesis in M2 and M3 microbial communities revealed by isotopic analyses. Previously, we have concluded that lactate is oxidized mainly to acetate during the acetogenic step of AD and this includes the acetotrophic pathway of methanogenesis [ 18 ]. The present study confirms our previous results. Compared to butyrate and propionate, lactate is the most efficiently used substrate. It can be explained by the thermodynamics of lactate oxidation reactions, as it was discussed previously [ 18 ]. Briefly, lactate can be oxidized directly to acetate by many bacteria without the contribution of methanogenic Archaea according to the mechanism described for Acetobacterium woodii (ΔG 0' = − 61 kJ/mol) [ 17 ]. The fermentation of lactate to propionate is also a thermodynamically favorable reaction (ΔG 0' = − 169.7 kJ/reaction for Desulfobulbus propionicus ) [ 102 ]. Propionate formation from lactate was also observed in this study. Formation of propionate probably induces the metabolic pathways of propionate oxidation. To summarize the results of the study, (i) butyrate and propionate are transformed mainly to acetate, while lactate is transformed mainly to acetate and propionate; (ii) oxidation of acetate and lactate determines the acetotrophic pathway of methanogenesis, whereas oxidation of butyrate and propionate determines the hydrogenotrophic pathway of methanogenesis. Propionate is one of the major intermediates in AD. It is estimated that 6–35% of methane can be produced from propionate. Propionate degradation to methane draws the attention of many researchers, because accumulation of propionate is often observed in bioreactors with poor methane production [ 22 ]. Interestingly, inhibition of methanogenesis in the bioreactor fed with a propionate-rich medium was not confirmed in this study. It could indicate that accumulation of propionate should be considered an indicator, rather than a cause of disturbances in anaerobic digestion and biogas production. Furthermore, accumulation of propionate is related to changes in the anaerobic digestion process, unstable pH and temperature, overload of feedstock, too high a concentration of short-chain fatty acids and hydrogen partial pressure, improper reactor configuration and hydraulic retention time [ 22 , 103 , 104 ]. Lactate-, butyrate-, propionate- and acetate-selected microbial communities Metagenomic analysis revealed that the dominant bacteria and archaea identified in this study are probably unknown, uncultivable species. A striking distinctive feature is the high number of Archaea found in the examined microbial communities, 31 archaeal species out of 234 microbial species identified (13.2%). A recent study performed metagenomic binning starting from a range of different biogas reactors but, out of 1635 microbial species identified, only 61 Archaea (3.7%) were recovered [ 37 ]. However, other studies revealed a remarkable number of archaeal species, which can also represent a large fraction of the entire microbiome [ 44 , 105 ]. To determine how many MAGs reported in the present study represented new species not previously reported in the AD biogas database, a comprehensive Average Nucleotide Identity calculation was performed. Interestingly, 75.6% (177) of the MAGs identified here were not present in the global AD biogas database reported by Campanaro and colleagues, evidencing a very high level of new species recovered in this study. It can be speculated that the use of four diverse SCFAs as feedstock promoted the growth of species associated with the terminal part of the anaerobic degradation food chain (methanogenesis) where Archaea play a crucial role in methane production. On the other hand, most of the previous studies performed on the AD microbiome (those investigated by Campanaro and colleagues) did not focus on the acetogenic and methanogenic steps of the biogas food chain, where there are still many unknown microbes that were identified in the present project. As a confirmation of this finding, 20 out of 31 archaeal species identified here were not present in the global AD biogas database. 113 MAGs exhibited a comparable relative abundance in all bioreactors. They can be considered a sort of core microbiome with some dominant microbes such as Methanothrix soehngenii , Methanoculleus sp., representatives of Bacteroidales , Spirochaetaceae and unclassified bacteria. Interestingly, substrate-specific bacteria are not predominant in the microbial communities. Furthermore, the recognized acetate, lactate, butyrate, and propionate oxidizers described in the introduction are not in their majority present in the studied microbial communities form the bioreactors. The exceptions are butyrate-specific Syntrophomonas wolfei AS29adLBPA_159 (MQ) or lactate-specific Desulfovibrio desulfuricans AS29adLBPA_31. This is a confirmation that the vast majority of microorganisms, especially those requiring syntrophic growth and/or those having a very low growth rate, are yet to be isolated and cultivated. However, they represent the orders, families, and genera to which belong some recognized syntrophic bacteria involved in oxidation of products derived from acidic fermentations. These latter species were identified at high relative abundance in all bioreactors (e.g., Spirochaetaceae sp. AS29adLBPA_218, Bacteroidales sp. AS29adLBPA_33 or Synergistales sp. AS29adLBPA_149). Some other microbes were classified as “lactate-specific”, such as for example Deltaproteobacteria sp. AS29adLBPA_110 (MQ), Firmicutes sp. AS29adLBPA_210, Clostridiales sp. AS29adLBPA_49 (MQ); butyrate-specific Syntrophomonas sp. AS29adLBPA_81 (MQ), Synergistales sp. AS29adLBPA_149, Syntrophaceae sp. AS29adLBPA_197; propionate-specific Peptococcaceae sp. AS29adLBPA_161, Desulfobacteraceae sp. AS29adLBPA_204; acetate-specific Clostridiales sp. AS29adLBPA_6, Sphaerochaeta sp. AS29adLBPA_43 (MQ), Geobacter sp. AS29adLBPA_1 (MQ), Anaerolineaceae sp. AS29adLBPA_207. Comments on substrate-specific and rare methanogenic species are reported in the Results section; additionally, functional analysis of medium-quality (MQ) MAGs has to be considered with caution, but their enrichment in specific samples can provide suggestions on their putative role. Preference for specific substrates can be determined based on many different indicators, one being for sure the possibility to degrade a specific compound, but also syntrophic behaviors determined by compounds exchange between microbes are involved in shaping the structure of the microbiomes [ 51 ]. For this reason, it is expected that only a part of the “specialized microbes” is able to directly utilize the substrate provided; other species probably rely on the chemical compounds released. It can simply indicate a wide capability of one bacterium to use several different substrates, which should also be illustrated by investigation of genes expression (metatranscriptomics). Differences in genes expression seem to be indirectly supported by the results of stable carbon isotope composition of methane and carbon dioxide in the fermentation gas. Metagenomic analysis of the microbial communities clearly shows that dominant hydrogenotrophic or acetotrophic pathways of methane formation elucidated by isotopic analysis are not associated with changes in the contribution of methanogens utilizing hydrogen and carbon dioxide or acetate. The most dominant methanogens in all the bioreactors are Methanothrix soehngenii AS29adLBPA_138 (10–19%) and Methanoculleus AS29adLBPA_62 (5.5–11.8%), the latter being a Medium-Quality (MQ) MAG. Methanoculleus, similarly to other hydrogenotrophic methanogens, produced methane from carbon dioxide and hydrogen generated during syntrophic oxidation of SCFAs, this assumption being based on taxonomic assignment since the recovered genome of AS29adLBPA_62 was MQ and did not undergo functional analysis. It may indicate induction of genes of either the hydrogenotrophic or acetotrophic pathways of methane formation depending on the supplied substrate. Our explanation is as follows. Acetate as well as lactate, which is easily metabolized to acetate via the mechanism described for A. woodii [ 17 ], determines the acetoclastic pathway of methane formation in Methanothrix soehngenii . Butyrate and propionate initiate metabolic pathways found for methane-yielding microbial communities fed with ethanol, where the Methanothrix species reduced carbon dioxide to methane with electrons accepted via DIET [ 40 ]. This hypothesis should be confirmed by metatranscriptomic analysis of the microbial communities. Since, in addition to DNA, RNAs have been isolated from the same samples, the work on gene expression in the examined microbial communities is ongoing and the results will be included in a future report. Recently, DIET is being increasingly mentioned in the context of anaerobic digestion [ 10 , 106 ]. The microbial community fed with ethanol and dominated by the Methanotrix species exhibited an elevated abundance of the bacterial pilA gene and the methanogenic sludge showed a higher conductivity. In another study Methanospirillum hungatei was shown to form electrically conductive filaments that are analogs of e-pilli in Geobacter species. In the case of M. hungatei, it is the archaellum, whose core consists of tightly packed phenylalanines [ 107 ]. In our study the Methanospirillum _AS29adLBPA_21 strain was classified as a propionate-specific methanogen in the microbial community fed with the propionate-rich substrate. In the study by Barua and co-workers (2018) [ 106 ], addition of conductive carbon fibers to the bioreactors fed with butyrate- and propionate-containing media resulted in (i) an increase of methane production, (ii) a higher efficiency of substrate utilization, (iii) an increased contribution of electroconductive bacteria such as Desulfuromonas, Pseudomonas, Azonexus or Azovibrio that accompanied butyrate and propionate oxidizers, and (iv) a domination of Methanoseta species among the methanogens. The results indicated that DIET is involved in processing of propionate and butyrate by the microbial community."
} | 4,452 |
35104027 | PMC9303619 | pmc | 133 | {
"abstract": "Abstract Aims Fourteen percent of all living coral, equivalent to more than all the coral on the Great Barrier Reef, has died in the past decade as a result of climate change‐driven bleaching. Inspired by the ‘oxidative stress theory of coral bleaching’, we investigated whether a bacterial consortium designed to scavenge free radicals could integrate into the host microbiome and improve thermal tolerance of the coral model, Exaiptasia diaphana . Methods and Results \n E. diaphana anemones were inoculated with a consortium of high free radical scavenging (FRS) bacteria, a consortium of congeneric low FRS bacteria, or sterile seawater as a control, then exposed to elevated temperature. Increases in the relative abundance of Labrenzia during the first 2 weeks following the last inoculation provided evidence for temporary inoculum integration into the E. diaphana microbiome. Initial uptake of other consortium members was inconsistent, and these bacteria did not persist either in E. diaphana ’s microbiome over time. Given their non‐integration into the host microbiome, the ability of the FRS consortium to mitigate thermal stress could not be assessed. Importantly, there were no physiological impacts (negative or positive) of the bacterial inoculations on the holobiont. Conclusions The introduced bacteria were not maintained in the anemone microbiome over time, thus, their protective effect is unknown. Achieving long‐term integration of bacteria into cnidarian microbiomes remains a research priority. Significance and Impact of the Study Microbiome engineering strategies to mitigate coral bleaching may assist coral reefs in their persistence until climate change has been curbed. This study provides insights that will inform microbiome manipulation approaches in coral bleaching mitigation research.",
"introduction": "INTRODUCTION The persistence of coral reefs is threatened by elevated sea surface temperature (SST) and associated summer heat waves that are the result of climate change (Hughes et al. 2017 ). As the dominant primary producers and reef builders, scleractinian corals are critical components of coral reefs. Corals live in symbiotic associations with other organisms, including bacteria, protists, fungi, archaea and viruses (Blackall et al. 2015 ; Ricci et al. 2019 ; Ainsworth et al. 2020 ). These microbial associates contribute to the health of this complex host–microbe association, or holobiont (Rohwer et al. 2002 ). Crucial members of the coral holobiont are the dinoflagellate photosymbionts of the family Symbiodiniaceae. In this mutually beneficial relationship, corals provide a safe haven and inorganic nutrients, while the Symbiodiniaceae meet most of the coral’s energy needs through the transfer of photosynthate in a nutrient‐poor seawater environment (Muscatine et al. 1981 ; Yellowlees et al. 2008 ; Tremblay et al. 2014 ). The relationship between the coral host and Symbiodiniaceae can break down during periods of stress, resulting in the loss of the symbionts from the coral tissues, a process called coral bleaching. There are several hypotheses for the mechanisms that drive bleaching (Weis 2008 ; Cunning and Baker 2013 ; Wiedenmann et al. 2013 ; Wooldridge 2013 ), with a common theme being the overproduction of reactive oxygen species (ROS) by the algal symbiont and their toxic accumulation. Elevated SST inhibits Symbiodiniaceae photosynthesis via damage to photosystem II (PSII) (Warner et al. 1999 ; Tchernov et al. 2004 ). This impairment leads to increased levels of ROS, which, once generated, can trigger the oxidation of essential photosynthetic molecules (Wang et al. 2011 ; Mathur et al. 2014 ), thylakoid membranes (Tchernov et al. 2004 ; Roberty et al. 2016 ), and enzymes of the Calvin‐Benson cycle (Asada and Takahashi 1987 ; Lesser and Farrell 2004 ), thereby interfering with the supply of fixed carbon to the holobiont (Lesser 2004 ). Increased abundance of ROS in the chloroplast also reduces the biosynthesis of chlorophyll (Takahashi et al. 2008 ; Roberty et al. 2016 ). In the ‘oxidative stress theory of coral bleaching’ (Downs et al. 2002 ), ROS produced by the Symbiodiniaceae are thought to diffuse into host cells and activate signalling cascades resulting in the loss of Symbiodiniaceae from the coral host (Smith et al. 2005 ; Weis 2008 ; Davy et al. 2012 ; Suggett and Smith 2020 ). If separation of coral and Symbiodiniaceae occurs, it may be fatal for the coral unless the symbiosis is re‐established. Thus, with average SST predicted to rise further and associated summer heat waves increasing in frequency, severity and duration (Aral and Guan 2016 ), finding effective coral bleaching mitigation methods has become crucial. Many innovative approaches have been proposed to mitigate bleaching and enhance coral survival during thermal stress, including selective breeding of thermally tolerant coral genotypes and the creation of interspecific hybrids (van Oppen et al. 2017 ; Chan et al. 2018 ; Quigley et al. 2020 ), directed evolution of Symbiodiniaceae (Chakravarti et al. 2017 ; Buerger et al. 2020 ), and inoculation with bacteria to engineer the coral microbiome (van Oppen et al. 2015 ; Damjanovic et al. 2017 ; Peixoto et al. 2017 ; Blackall et al. 2020 ; Peixoto et al. 2021 ). Directly targeting ROS, however, has rarely been investigated, despite evidence that exposure to exogenous antioxidants can reduce Symbiodiniaceae loss (Lesser 1997 ), increase survival (Nesa and Hidaka 2009 ), reduce respiration rates (Lesser 1997 ), and decrease DNA damage (Nesa and Hidaka 2008 ; Majerová and Drury 2021 ) in scleractinian corals exposed to thermal stress. The cumulative results from these studies suggest that increased biologically available antioxidants could potentially mitigate bleaching and prolong the life of corals during climate change induced summer heat waves. Nevertheless, applying antioxidants directly onto corals may not be practical due to their short lifespan when exposed to seawater (King et al. 2016 ) and ultraviolet radiation (Compton et al. 2019 ), or desirable due to their potential impact on non‐target organisms. Instead, increasing natural antioxidant generation within the coral holobiont may be more effective and feasible. Microbiome manipulation has been suggested as a method for increasing natural antioxidant generation in the cnidarian holobiont (van Oppen et al. 2015 ; Epstein et al. 2019 ). Exposure to beneficial bacteria has reduced the impact of pathogens (Alagely et al. 2011 ) or environmental stressors such as oil pollution (dos Santos et al. 2015 ) on cnidarians. Inoculation of coral with native bacteria (Doering et al. 2021 ), including catalase‐positive isolates, has also improved bleaching tolerance in corals exposed to elevated temperature (Rosado et al. 2018 ; Santoro et al. 2021 ). Although the results of the latter studies were promising, the reason for improved bleaching tolerance was unclear due to the selected bacteria’s broad range of traits. Furthermore, the stressed corals may have been supported indirectly through heterotrophic feeding on the introduced bacteria (Houlbrèque et al. 2004 ; Meunier et al. 2019 ) as the control corals in both cases were starved. Consequently, the influence of antioxidant‐producing bacteria on cnidarian bleaching requires further investigation. A major challenge with bacterial inoculation therapies is transient colonization (Hai 2015 ), with continuous addition of bacteria required for long‐term maintenance. Given the scale of coral reefs, repeated dosing with bacteria is not a feasible long‐term strategy to mitigate coral bleaching. However, if the bacteria can form a stable association with the host, the benefits of inoculation may persist overtime. To assess the ability of introduced bacteria to form a stable presence in the cnidarian microbiome and evaluate their influence on host health and bleaching, we inoculated the sea anemone Exaiptasia diaphana with a bacterial consortium consisting of host‐derived free radical scavenging (FRS) bacteria (Dungan et al. 2021a ) prior to thermal stress. Metabarcoding of bacterial 16S rRNA genes was used to track the incorporation of the bacteria into the E . diaphana microbiome and changes in bacterial community structure across time and treatments. The holobiont’s bleaching response to thermal stress was assessed by measuring Symbiodiniaceae photosynthetic performance and cell densities, while net ROS was quantified in host tissues with a fluorescent reagent. The influence of host genotype and Symbiodiniaceae strain was also explored by comparing the responses of three E. diaphana genotypes and tracking Symbiodiniaceae community composition.",
"discussion": "DISCUSSION Given the frequency of coral bleaching events, assisted evolution strategies, including the application of bacteria with putative beneficial properties, should be considered in addition to action plans to reduce carbon emissions, since without intervention coral reefs will not survive predicted climate change conditions. In our investigation of a microbial engineering strategy, we found that three applications of a high or low FRS bacterial consortium, tested in parallel with a no‐inoculum control, showed evidence of short‐term uptake into the E . diaphana microbiome. However, the incorporation of the consortium members was inconsistent, and none persisted in the anemone microbiome over time. Consequently, the failure of the high FRS bacteria to confer improved host thermal tolerance may have been due to their inability to integrate into the host microbiome for the full duration of the experiment. Importantly, there were no apparent physiological impacts (negative or positive) on the holobiont following inoculation, thus showing that the induced shifts in the abundance of native anemone microbiome members were not detrimental to holobiont health. Uptake of FRS bacteria by E. diaphana was uneven Our dosing of 10 6 bacterial cells ml −1 of each species in both the high and low FRS consortia is greater than reported bacterial carrying capacity of E . diaphana , which range from 10 3 –10 5 bacteria anemone −1 (Costa et al. 2021 ; Dungan et al. 2021a ). An increase in anemone‐associated Labrenzia provided the strongest evidence for the uptake of FRS bacterial by E . diaphana . Labrenzia are naturally abundant (~5%) in all the AIMS genotypes (Hartman et al. 2020 ; Dungan et al. 2021b ) and are core members of the Symbiodiniaceae microbiome (Lawson et al. 2018 ). These bacteria may therefore have faced low inhibition from antagonistic interactions with resident bacteria (Rypien et al. 2010 ). Poor uptake of Micrococcus could have been caused by below‐target densities of those bacterial cultures in the inocula (Table 1 ), emphasizing the need to dose at consistently high cell densities, and possibly higher densities for some bacteria. Alteromonas are metabolically‐versatile copiotrophs (Pedler et al. 2014 ), rapidly responding to increases in dissolved organic matter and often dominating mesocosm experiments (McCarren et al. 2010 ). A. macleodii has been found to be highly abundant in A. salina feedstock used in our E . diaphana culture system (Hartman et al. 2020 ) and increased rapidly throughout the experiment, becoming dominant in inoculated and uninoculated anemones. This suggests that A. macleodii likely originated from the A. salina feedstock and multiplied due to the culture conditions. Uptake of FRS bacteria by E. diaphana was short‐lived Post‐inoculation increases in the relative abundance of the FRS bacterial ASVs and significant differences between the bacterial communities of inoculated and uninoculated anemones on Days 1 and 3 suggested successful incorporation of the FRS bacteria by the anemones. However, this was short‐lived. The failure of bacterial uptake by the host may be due to failure of the selected bacteria to be recognized as symbionts by the host, control of bacterial adhesion as described by the bacteriophage adherence to mucus model (Barr et al. 2013 ), or founder effects whereby existing bacterial communities influence the identity of new associates (Apprill et al. 2012 ). Further, the ciliated surface of E . diaphana may play a role in preventing bacterial adhesion (Costa et al. 2021 ). The three‐dose strategy employed in the present study was similar to previous cnidarian experiments. For example, Rosado et al. ( 2018 ) inoculated coral samples with a bacterial cocktail twice, 5 days apart, Damjanovic et al. ( 2019b ) inoculated coral larvae seven times, at 3–4 day intervals, and Doering et al. ( 2021 ) completed microbiome transplants three times over 3 days for Porites coral fragments. However, these studies did not assess the uptake of inoculated bacteria between doses. Although previous studies have demonstrated that bacterial inoculation can protect E . diaphana from pathogenic infection (Alagely et al. 2011 ; Zaragoza et al. 2014 ), incorporation of the selected bacteria into the host microbiome was not assessed. Recently, Costa et al. ( 2021 ) showed that E . diaphana is resistant to microbiome changes as transplantation of either the Acropora humilis or Porites sp. associated microbiomes failed to shift the community composition compared to Exaiptasia microbiome controls. For introduced bacteria to be incorporated into the host microbiome, they must be attracted by the host’s chemical cues, be recognized by the host, and be resistant to any antimicrobial compounds present (Krediet et al. 2013 ). Since our FRS bacteria were host‐derived, it is possible that they met these criteria, but equally feasible that these individuals were transiently associated. Regardless, introducing the FRS bacteria and creating a stable shift in the host’s microbial community proved challenging. In the few studies that have investigated the onset of bacterial symbiosis with coral hosts, the focus has been on changes in bacterial community structure over time (Apprill et al. 2009 ; Damjanovic et al. 2019a ), and there is evidence that some bacterial symbionts are taken up more readily than others (Apprill et al. 2012 ). Future studies that examine the chemical cues produced by coral hosts or bacteria and the chemicals involved in bacterial recognition will improve our understanding of the biochemical requirements for bacterial uptake, and hence our ability to induce stable changes (Kvennefors et al. 2008 ). Thermal stress, ROS and bleaching Thermally induced bleaching occurred for all genotypes as there was a significant reduction of Symbiodiniaceae cell density and significant decline in photosynthetic efficiency on experimental Day 43 for anemones in the elevated temperature condition compared to ambient. Significant reductions in F \n \n v \n / F \n \n m \n and Symbiodiniaceae cell density for all anemones at elevated temperature occurred regardless of inoculated or uninoculated treatment. While the ROS data suggest some neutralization of ROS, the overall variation in the ROS values advises caution not to overinterpret these results. The bleaching response of E . diaphana is thought to be initiated by an increased production of ROS by their algal endosymbionts during periods of stress (Lesser 1997 ; Weis 2008 ). Contrary to the ‘oxidative stress theory of bleaching’ (Downs et al. 2002 ), elevated temperature did not lead to an increase in net ROS (Figure 4 g–i). Instead, bleaching of the AIMS3 anemones at elevated temperature in the high and low FRS treatments was accompanied by significant declines in net ROS. This observation corresponds with other studies that have shown bleaching can occur without photosynthetically produced ROS (Tolleter et al. 2013 ) and with discrepancies in enzymatic antioxidant activity between host and symbiont tissue portions (Krueger et al. 2015 ). Our data support the suggestion that bleaching does not require an influx of ROS from the algal symbiont to the host (Nielsen et al. 2018 ), and raises questions about the importance of symbiont‐derived ROS in initiating bleaching in E . diaphana . Given the novel finding that elevated ROS levels were not associated with bleaching in E . diaphana , this organism may not be an ideal candidate for testing an FRS bacterial consortium to mitigate climate change in cnidarians. Bleaching susceptibility differed by genotype As the introduced bacteria were not retained by the anemones, no correlation can be drawn between differences in bleaching tolerance and inoculation. However, the variation in bleaching intensity between genotypes is noteworthy as AIMS2 lost fewer Symbiodiniaceae under elevated temperature compared to AIMS3 and AIMS4. It is possible that differences in bleaching tolerance were driven by differences in the in hospite Symbiodiniaceae communities harboured by each anemone genotype (Howells et al. 2012 ; Hawkins et al. 2016 ). The AIMS2 anemones, which did not bleach, harboured a distinct Symbiodiniaceae ITS2 type profile, B1‐B1o‐B1p. Very few AIMS3 individuals had this ITS2 type profile and bleached heavily, whereas 50% of the AIMS4 anemones had this ITS2 type profile and displayed high variation in their bleaching response. These data suggest that Symbiodiniaceae with the B1‐B1o‐B1p ITS2 type profile could confer some thermal resilience to the holobiont. Alternatively, the anemone genotypes may have different bleaching susceptibilities independent of their Symbiodiniaceae type profile (Gabay et al. 2019 ). Previous work on corals (Quigley et al. 2018 ) and GBR‐sourced E . diaphana (Tortorelli et al. 2020 ) has shown that the mechanism of recognition and incorporation of Symbiodiniaceae into the holobiont is influenced by both algal symbiont and host. Observations from the present study indicate that genotype and algal symbiont type will be critical in future experiments seeking to induce or mitigate bleaching in E. diaphana . Limitations of the study, and recommendations for future microbial engineering work Although there was evidence for increased relative abundance of some consortium members in the host microbiome, the key limitation of the present study was the inability of the bacteria to persist after inoculation. Bacteria delivery via bioencapsulation in A. salina is used in aquaculture to ensure dosed bacteria are ingested (Hai et al. 2010 ), and a method for corals using rotifers has been reported (Assis et al. 2020 ). Researchers using probiotics to treat stony coral tissue loss disease (SCTLD) in the field have also tested strategies such as weighted enclosures, paste applied directly to SCTLD lesions, and slow release beads to improve bacteria‐coral contact and maintain probiotic concentrations in the open marine environment (Smithsonian Marine Station 2020 ). Microencapsulation of beneficial bacteria in alginate has been explored in the aquaculture industry, with high efficiency (80% of bacteria survived alginate encapsulation), retention (40% of encapsulated bacteria survived storage at 22°C for 30 days), and storage survival (over 90% survival of bacteria after 1 month storage at 4°C) (Rosas‐Ledesma et al. 2012 ). These approaches may address the issues of dilution and inadequate uptake of putative beneficial bacteria by coral and warrant further investigation. Furthermore, of the nearly 1000 bacteria isolated from E . diaphana (Dungan et al. 2021a ) and the thousands of bacterial cells they can carry (Costa et al. 2021 ), only 12 were used in this study. More work is needed to explore the potential beneficial roles of the other bacteria, which may be possible with advances in metagenomic sequencing and the assembly of bacterial metagenome‐assembled genomes. Because closely related bacterial strains can have an identical sequence in the short 16S rRNA gene region examined with metabarcoding, such as the high and low FRS conspecific pairs used in this study, future studies could be enhanced by quantifying consortium members using qPCR with strain specific primers to determine whether each member was retained in the host microbiome after inoculation. An element of the present study that we recommend as good practice is the use of a negative inoculum (here, low FRS). Inclusion of a negative inoculum allowed us to account for differences in anemone response to thermal stress due to the introduction of an additional source of nutrition as heterotrophy can reduce the impact of thermal stress on corals (Grottoli et al. 2006 ; Aichelman et al. 2016 ). To date, few studies have investigated the ability of bacteria to increase cnidarian thermal tolerance. For this type of assisted evolution strategy to be feasible, interventions must minimize risk (NASEM 2019 ). One of the unknown risks in microbial engineering is its impact on holobiont physiology as the introduction of high numbers of microbes to the holobiont could trigger unintended consequences. Critically, our results suggest that inoculation with a consortium of host‐derived bacteria does not negatively impact anemone or Symbiodiniaceae physiology. Some members of our inocula remained with the host after dosing, but only for a short period of time. Therefore, while inoculation with FRS bacteria did not mitigate bleaching in heat‐exposed anemones, this cannot be attributed to an inability of the FRS bacteria to confer improved thermal tolerance, which remains unproven. Future studies that maintain elevated levels of introduced bacteria with appropriate controls to consider heterotrophy will provide clearer insights into the potential of the coral bleaching mitigation strategy proposed here."
} | 5,449 |
36341296 | PMC9595181 | pmc | 134 | {
"abstract": "Mechanical energy harvesting using piezoelectric nanogenerators (PNGs) offers an attractive solution for driving low-power portable devices and self-powered electronic systems. Here, we designed an eco-friendly and flexible piezocomposite nanogenerator (c-PNG) based on H 2 (Zr 0.1 Ti 0.9 ) 3 O 7 nanowires (HZTO-nw) and Ba 0.85 Ca 0.15 Zr 0.10 Ti 0.90 O 3 multipods (BCZT-mp) as fillers and polylactic acid (PLA) as a biodegradable polymer matrix. The effects of the applied stress amplitude, frequency and pressing duration on the electric outputs in the piezocomposite nanogenerator (c-PNG) device were investigated by simultaneous recording of the mechanical input and the electrical outputs. The fabricated c-PNG shows a maximum output voltage, current and volumetric power density of 11.5 V, 0.6 μA and 9.2 mW cm −3 , respectively, under cyclic finger imparting. A high-pressure sensitivity of 0.86 V kPa −1 (equivalent to 3.6 V N −1 ) and fast response time of 45 ms were obtained in the dynamic pressure sensing. Besides this, the c-PNG demonstrates high-stability and durability of the electrical outputs for around three months, and can drive commercial electronics (charging capacitor, glowing light-emitting diodes and powering a calculator). Multi-physics simulations indicate that the presence of BCZT-mp is crucial in enhancing the piezoelectric response of the c-PNG. Accordingly, this work reveals that combining 1D and 3D fillers in a polymer composite-based PNG could be beneficial in improving the mechanical energy harvesting performances in flexible piezoelectric nanogenerators for application in electronic skin and wearable devices.",
"conclusion": "4. Conclusions In this study, we reported a new conceptual strategy to design ceramic/polymer nanocomposites for mechanical energy harvesting applications, involving the embedding of HZTO nanowires and BCZT multipods in a PLA biopolymer. Through the simultaneous recording of the mechanical input (stress under finger tapping) and the electric outputs (open-circuit voltage, short-circuit current and electrical charges), we can easily tune the energy harvesting performance. The c-PNG device demonstrated an enhanced output voltage, current and volumetric power density of 11.5 V, 0.6 μA and 9.2 mW cm −3 , respectively, under cyclic finger imparting, alongside a high-pressure sensitivity of 0.86 V kPa −1 and fast response time of 45 ms at 1 Hz. Besides this, the c-PNG can generate a highly stable and durable output voltage, even after three months, proving its applicability to power commercial electronics (charging capacitors, glowing LEDs and powering a calculator). Multi-physics simulations using COMSOL Multiphysics indicate that the presence of BCZT-mp was crucial in enhancing the piezoelectric response of the c-PNG. Accordingly, the embedding of HZTO-nw (1D) and BCZT-mp (3D) in a PLA composite-based PNG paves the way for a new and effective strategy to design eco-friendly and flexible devices for powering small portable electronics and could be beneficial in enhancing mechanical energy harvesting properties.",
"introduction": "1. Introduction Since 2006, the concept of piezoelectric energy harvesting, proposed by Wang et al. , 1 using a piezoelectric nanogenerator (PNGs) has become a growing and a promising technology for converting random mechanical energy into electric energy using nanoscale piezoelectric materials. 2–4 Another biomechanical energy harvesting concept is the use of triboelectric nanogenerators. 5–8 However, despite their high electrical outputs, the impermanent nature of triboelectric charges limits their long-term durability. 9 Besides this, contrary to chemical batteries, PNGs can provide sustainable electrical energy. 2 In particular, ceramic/polymer piezocomposites for energy harvesting applications are considered to be a significant research field, providing the convenience of mechanical flexibility, adaptability for large mechanical forces, suitable voltage with sufficient power output, lower internal leakage current, lower manufacturing costs, and rapid processing compared to ceramic-based PNGs. 2,10–12 In this case, both the ceramic and polymer must be carefully selected. 2 However, the most high-performance piezoceramics used in industry, such as PbZr x Ti 1− x O 3 (PZT), are lead-based, toxic and environmentally unfriendly, and thus their integration in future applications will be restricted. 13–17 Similarly, the majority of polymers used in polymer-based PNGs, such as polydimethylsiloxane (PDMS) and polyvinylidene fluoride (PVDF), are petrol-based, not eco-friendly and the majority of them are not biodegradable. 3,18–20 Besides this, almost all PNGs require a poling process to promote the piezoelectric activity of the utilised materials. 19,21–23 Consequently, the durability of the output performances of these PNGs have been questioned. 24–26 Recently, we designed a self-poled and bio-flexible piezoelectric nanogenerator (BF-PNG) based on lead-free biocompatible Ba 0.85 Ca 0.15 Zr 0.10 Ti 0.90 O 3 (BCZT) nanoparticles that are functionalised with polydopamine and embedded in a polylactic acid (PLA) biodegradable polymer. 27 The BF-PNG was able to generate excellent electrical outputs under gentle finger tapping, and demonstrated outstanding mechanical robustness, stable and durable output even after one year. It has been reported that 1D-shaped piezoelectric nanowires are better than 0D ones for energy harvesting applications. 28,29 To tailor piezocomposites with high electrical outputs, the design of novel piezo-fillers with complex morphologies unlike 0D-nanoparticles and 1D-nanomaterials (nanowires, nanotubes, nanofibers, etc. ) is a leading topic. Previously, we reported the piezoelectric energy harvesting of lead-free H 2 (Zr 0.1 Ti 0.9 ) 3 O 7 nanowires embedded in a PLA matrix and observed enhanced output performances (open-circuit voltage of 5.41 V and short-circuit current of 0.26 μA) and could drive small commercial electronics under various human motions). 30 To further improve the piezoelectric energy harvesting properties of nanogenerators, the design of other complex morphologies is attracting considerable critical attention. Mainly, the use of 3D fillers ( e.g. flowers, stars, etc. ) is gaining interest in the scientific community. 31–33 More specifically, flower-like piezo-fillers with a 3D morphology play a vital role in enhancing piezoelectric harvesting performances. 32 Accordingly, the combination of 1D and 3D piezoceramics could enhance further the piezoelectric response of piezoelectric nanogenerators. From this perspective, we report a novel strategy to design ceramic/polymer nanocomposites for energy harvesting applications. This involves embedding H 2 (Zr 0.1 Ti 0.9 ) 3 O 7 nanowires (HZTO-nw) and Ba 0.85 Ca 0.15 Zr 0.10 Ti 0.90 O 3 multipods (BCZT-mp) in a PLA biopolymer. The effects of the applied stress amplitude, frequency and pressing duration on the electric outputs in the piezocomposite nanogenerator device (c-PNG) were investigated using a simultaneous recording of the mechanical input and the electrical outputs. Through this approach, enhanced output voltage and a short-circuit current of 11.5 V and 0.6 μA, respectively, were obtained under finger tapping. Multi-physics simulations using COMSOL Multiphysics indicated that the presence of the BCZT-mp was crucial in enhancing the piezoelectric response of the c-PNG. In addition, the c-PNG can generate stable and durable voltage even after three months. The c-PNG proved its applicability to generate enhanced output voltage under bending motions, and to drive low-power commercial electronics (light-emitting diodes (LEDs) and a calculator).",
"discussion": "3. Results and discussion 3.1. Structural and morphological properties \n Fig. 2a shows the FESEM micrograph of the filler nanopowder composed mainly of 1D-HZTO nanowires (HZTO-nw) with a high aspect ratio, and 3D-BCZT multipods (BCZT-mp). Fig. 2b presents a STEM-HAADF image of a single BCZT multipod with a 500 nm pod-length six-pod architecture. The composition of the BCZT multipods was confirmed by energy dispersive X-ray (EDX) analyses, as shown in Fig. S1 and S2 (ESI). † The growth mechanism of the BCZT multipods has previously been thoroughly discussed. 36 The X-ray diffractometry (XRD) pattern of the nanopowder is illustrated in Fig. 2c , wherein the presence of BCZT-mp and HZTO-nw can be clearly observed, and HZTO-nw and BCZT-mp can be observed to have crystallised in monoclinic ( C 2/ m ) and tetragonal ( P 4 mm ) phases, respectively. As depicted in the inset of Fig. 2c , the selected area electron diffraction SAED pattern clearly shows that the BCZT multipod is single crystalline and has a good crystalline phase. As shown in the SAED pattern, the lattice spacings determined from the diffraction spots of the SAED of the BCZT multipod are 3.999 Å and 2.834 Å, which match the (100) and (011) planes, respectively. The corresponding zone axis (ZA) of the SAED pattern is indicated in the top of the figure. Further structural, morphology and composition analyses of the BCZT multipods can be found in the literature. 36 Besides this, Fig. 2d presents an FESEM image of the cross-section of the 20 vol% HZTO-nw + BCZT-mp/PLA nanocomposite film, from which a dense film with 8 μm thickness is observed (the cross-section of the nanocomposite with low magnification is provided in Fig. S3 (ESI)). † In the FESEM images, a BSE detector was used to visualise the composition fluctuation in the nanocomposite film, where the whitest spots represent the BCZT multipods. Also, it can be clearly seen that HZTO-nw and BCZT-mp fillers are covered with PLA polymer matrix, indicating the excellent compatibility between PLA and the fillers. 38,39 Fig. 2 (a) FESEM micrograph of the raw nanopowder (inset shows the magnified view of the BCZT multipods and HZTO nanowires). (b) STEM-HAADF image of a single BCZT multipod. (c) XRD pattern of the raw nanopowder (inset is the SAED pattern of a single BCZT multipod). (d) FESEM image of a cross-section of the 20 vol% HZTO-nw + BCZT-mp/PLA nanocomposite film. The mechanical and thermal properties of PLA and 20 vol% HZTO-nw + BCZT-mp/PLA films are depicted in Fig. S4 † (ESI). † Thermogravimetric analysis (TGA) was used to evaluate the effect of HZTO-nw + BCZT-mp fillers on the thermal stability of the PLA biopolymer. Fig. S4a † shows the TGA curves and their respective derivative thermogravimetry (DTG) curves (inset of Fig. S4a † ) of the neat PLA and 20 vol% HZTO-nw + BCZT-mp/PLA films. Both TGA curves display one decomposition step between 240 and 350 °C. It is observed that PLA and 20 vol% HZTO-nw + BCZT-mp/PLA start to degrade from 240 and 250 °C, respectively, as defined from the onset degradation temperatures in the DTG curves. This improvement in the thermal behaviour (shift in the TGA curve of the nanocomposite film toward higher temperatures) is related to the presence of HZTO-nw + BCZT-mp as a reinforcing charge. 40,41 The mass residues at 600 °C are 0.63 and 7.79% for the PLA and 20 vol% HZTO-nw + BCZT-mp/PLA films, respectively. Meanwhile, Fig. S4b † depicts the typical stress–strain curves of the neat PLA and 20 vol% HZTO-nw + BCZT-mp/PLA films. Obviously, both samples exhibit a clearly distinguished yield point, as shown in Fig. S4b. † After loading the PLA with HZTO-nw + BCZT-mp fillers, the Young's modulus ( Y ) doubled from 26.33 to 53.05 MPa. Accordingly, the 20 vol% HZTO-nw + BCZT-mp/PLA nanocomposite films can resist high mechanical stress and exhibit a relatively high Young's modulus. 42 Actually, good mechanical properties of the piezocomposite film are highly required to preserve the designed piezoelectric nanogenerator from damage while applying mechanical impartation during the energy harvesting process. 43,44 3.2. Mechanical energy harvesting of the c-PNG To assess the biomechanical energy harvesting ability of the modified piezocomposite films, piezoelectric nanogenerators with different filler concentrations (0, 5, 10, 20, 30 and 40 vol%) were designed. Fig. 3a shows a schematic illustration of the developed flexible piezocomposite nanogenerator device (c-PNG), consisting of a 20 vol% HZTO-nw + BCZT-mp/PLA nanocomposite film sandwiched between two copper electrodes and embedded with Kapton® tape, while Fig. 3b shows a photograph of the flexible piezocomposite nanogenerator device under bending motion, proving its flexibility. To gain insight on the effects of HZTO-nw + BCZT-mp filler concentration on the piezoelectric performance, specifically the open circuit voltage ( V oc ), the fabricated nanogenerators were subjected to finger impartations of 2 N and the resulting V oc values were recorded and are presented in Fig. S5 in the ESI. † It was observed that the output voltage was amplified with an increase in the HZTO-nw + BCZT-mp concentration. This can be explained by an enhancement in the dielectric constant of the composite with filler addition, and nanocomposite films with a higher dielectric constant show good storage charge capability, which can lead to better piezoelectric properties. 45 Nevertheless, above 20 vol%, a decline in the piezoelectric performance is observed due to filler agglomeration, i.e. deterioration of the mechanical properties which limits the movement of PLA dipoles and eventually weakens the polarizability of the nanocomposite. 46 Subsequently, the focus will only be on a piezocomposite with 20 vol% HZTO-nw + BCZT-mp. Fig. 3 (a) Illustration and (b) photograph of the fabricated piezocomposite nanogenerator. The generated open-circuit voltage when the external circuit is connected in the (c) forward and (d) reverse directions. Interestingly, compared to other previously reported PNGs, our piezoelectric nanocomposite energy harvester does not require any poling process to promote its piezoelectric activity. To verify the purity of piezoelectric effects in the PNG device, a switching polarity test was conducted. 47 As presented in Fig. 3c , upward and downward output signals were obtained upon pressing and releasing motions, respectively, when the c-PNG device was forward-connected to the measurement kit. In contrast, the signal direction was inverted in the reverse connection, as shown in Fig. 3d . The observation of such signal output switching by changing the polarity confirms that the energy harvesting signals are the product of piezoelectricity from the flexible c-PNG device. 29,48 A dual column mechanical testing system and an electrical measurement unit was employed to simultaneously record the input stress and output voltage. First, we examined the effect of applied stress on the electrical output performances, and the results are given in Fig. 5a and b . Real-time simultaneous recording of the open-circuit voltage and the stress amplitude under finger tapping is presented in Video S1 in the ESI. † It can be clearly seen that increasing the mechanical input (mechanical stress) enhances the electrical output (open-circuit voltage) of the c-PNG. For instance, by applying 2.97 kPa, an open-circuit voltage of 5.73 V can be obtained, however, by increasing the mechanical stress to 9.90 kPa, an open-circuit voltage of 11.04 V can be reached. The value discrepancy of each peak can be ascribed to the different strain rate of the device during the press-release process. 47 Besides this, the observed asymmetry in the positive and negative voltage peaks is attributed to the difference between the external force applied on the device and the restoring force. 49 These results are consistent with the dependence of the crystal structure deformation on the applied mechanical stress. A simulation of the piezoelectric potential of the c-PNG device by finite element (FEA) method using the COMSOL Multiphysics 5.6 programming software under a mechanical stress of 2 N is depicted in Fig. 4 . Simulation calculations of the piezopotential involved solid mechanics, the electrostatic effect, and the piezoelectric effect. By applying a mechanical stress of 2 N, positive and negative piezopotentials of 13 and −3 V were found. From Fig. 4a , it can be seen that the effective stress occupies a large area in the HZTO nanowires, however, highly effective stress is mainly applied to the corners of the BCZT multipods (brighter colour) compared to that applied to the centre. This can be explained by the extension of the applied stress to the corner of the BCZT-mp under external pressure to the composite. This stress distribution discontinuity is believed to be related to the specific space architecture of BCZT-mp, which was not observed in HZTO-nw. Similar observations were reported by Jian et al. 32 Therefore, the high local stress at BCZT-mp may be the main reason for the significantly enhanced voltage output in the c-PNG. A confirmation of these observations is presented in Fig. S6 (ESI), † where we simulated the mechanical stress and piezoelectric potential under a compression of 2 N of c-PNG using different fillers. It was found that the presence of BCZT-mp is vital in enhancing the piezoelectric response of the c-PNG. Fig. 4 The simulated distribution by FEA of (a) mechanical stress and (b) piezoelectric potential under a compression of 2 N. Fig. 5 Results of the simultaneous recording of the open-circuit voltage and the applied stress under finger tapping. Effect of the applied stress (a) amplitude, (c) frequency and (e) duration on the open-circuit voltage. (b, d and f) The corresponding applied mechanical stress under finger tapping. The mechanical energy harvesting at various frequencies was studied using a constant stress of about 10.50 kPa ( Fig. 5c and d ). As observed, the generated output voltage is not impacted by increasing the operating frequency from 1 to 5 Hz, indicating that the c-PNG device has the potential to scavenge mechanical energy with adjustable frequency and amplitude in a natural environment. Also, the effect of the pressing duration on the open-circuit voltage was investigated and the results are provided in Fig. 5e and f . A demonstration of the real-time simultaneous recording of the open-circuit voltage and the applied stress duration under finger tapping is depicted in Video S2 (ESI). † From first sight, it can be observed that using high mechanical stress does not result in a high output voltage. However, a thorough observation of the output voltage indicates that applying high mechanical stress results in symmetric voltage peaks. Besides this, contrary to the “press-release” sequence, the “press-hold-release” sequence produces output voltage peaks with greater negative voltage. This indicates that when applying low mechanical stress, the charges are hardly fully-pushed from one electrode to another. 50 It is worth mentioning that the stress drop in the hold sequence is related to the finger trembling while pressing ( Fig. 5f ). The peak-to-peak voltage (Δ V p–p ) was found to increase with increasing stress amplitude and hold time. For instance, Δ V p–p is enhanced from 13.4 to 17.8 V upon increasing the applied mechanical stress from 9.9 to 62.5 kPa, respectively. Notably, Δ V p–p can be further improved by adding a hold sequence (pressing duration) in the input stress. For example, by applying only 45.4 kPa with a hold time of 3 s, the Δ V p–p can be boosted to 23.2 V, compared to that obtained by applying 62.5 kPa. The sensitive properties of the c-PNG at low pressures (Δ σ ≤ 10 kPa) were evaluated through the dynamic pressure sensitivity ( S ) according to S = Δ V P − P /Δ σ , where Δ V p–p and Δ σ are the differences in the generated peak-to-peak voltage and applied stress, respectively. 51,52 From Fig. S7 (ESI), † the generated output voltage changes almost linearly with respect to the increased implied pressure. The calculated sensitivity of the c-PNG was found to be 0.86 V kPa −1 , equivalent to 3.61 V N −1 , which is higher than several previously reported piezoelectric pressure sensors. 52–61 Another important parameter in sensors is the sensing response time. By applying a pressure of 10.50 kPa, the c-PNG can delivers a fast response of 45 ms at 1 Hz. The response time of the c-PNG increases as the frequency of the applied pressure decreases, mainly due to a decrease of the applied pressure speed. 62 These response times are lower than those of other reported piezoelectric sensors. 52,63–66 These outcomes make the c-PNG a suitable candidate for use in electronic skin applications and wearable devices. By employing the same approach of the simultaneous recording of the mechanical and electrical outputs, the dependence of the electrical charges and current with the applied stress can be deduced, as presented in Fig. 6a–c . It is clearly shown that the values of the electrical charges and the current change are stress-dependent, and applying high mechanical stress induces high electrical charges and current values. For instance, by applying 9.8 kPa, electrical charge and current values of 11.55 nC and 0.42 μA, respectively, are obtained. The calculated charge density of about 48.4 μC cm −2 is higher than that obtained in our previous work in a piezoelectric nanogenerator based on BCZT spherical nanoparticles and a PLA matrix. 27 As observed in the effect of the pressing duration on the open-circuit voltage, the charge and current can be further improved by adding the hold sequence in the mechanical input. Accordingly, our pioneering concept based on simultaneously recording the mechanical input and electrical outputs can be helpful to easily tune energy harvesting performances. Fig. 6 Real-time recording of the (a) electrical charges and (b) current and (c) the corresponding mechanical stress under finger tapping using the flexible piezocomposite nanogenerator device. The test results of the reliability and durability of the fabricated c-PNG device. (d) Open-circuit voltage and (e) the corresponding applied stress under successive finger tapping. (f) Stability and durability of the open-circuit voltage after three months under sewing machine impartations, and (g) illustration of the open-circuit voltage stability testing of the c-PNG using sewing machine impartations. The voltage-stability testing of the fabricated piezocomposite nanogenerator was performed under gentle finger tapping (10 kPa) at a frequency of 1 Hz by measuring simultaneously both voltage and mechanical stress, as illustrated in Fig. 6d and e . It is observed that the output voltage is mechanical stress-dependent, and since it is difficult to maintain constant stress under finger tapping for a prolonged time, we used a sewing machine as a constant stress source (see Fig. 6f and g ), as reported in our previous work. 27 After 3300 cycles, at a frequency of 23 Hz, constant voltage values were observed without any performance degradation or mechanical damage. Besides this, the durability of the output performance, which is by far the biggest downside of piezoelectric nanogenerators, was evaluated after 3 months of aging under 1800 tapping cycles using a sewing machine at a frequency of 23 Hz ( Fig. 6f ). A slight drop in the output voltage (<0.04 V) was detected without any mechanical damage of the piezoelectric nanogenerator during the tests. These outcomes prove the excellent performance stability of our mechanical energy harvester even after three months of aging. 3.3. Self-poling mechanism in the c-PNG Self-polarisation has already been reported to be an amazing technique by which to eliminate the complexity of the traditional electrical poling process for piezoelectric and ferroelectric materials-based energy harvesting devices. 25,67 Similar to self-poled PVDF-based PNGs, the self-poling aspects in PLA results from some degree of molecular alignment of the PLA chains along the length of the HZTO-nw and BCZT-mp. 25 In other words, the PLA molecules are self-polarised in a favourable direction via the dual effect of stress and surface charge induced polarisation without the application of an external electric field. It has been reported that the electroactive β-phase of PLA has flexible molecular chains containing 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 dipoles, where the crystal structure is characterised by a helical structure and the shear piezoelectricity at the molecular level originates from the dipole that accompanies the asymmetric carbon. 68 The amount of the β-phase in PLA could be increased by embedding piezoelectric fillers into the PLA matrix, thus enhancing the piezoelectric properties. In our case, the presence of the oppositely charged polar surfaces on HZTO-nw and BCZT-mp aggressively interact with the different C O dipoles of PLA, resulting in the developing of negative and positive charge densities over the nanocomposite surface, promoting the formation of a piezoelectric polar β-phase through surface charge induced polarisation. Besides this, the applied mechanical strain induces a potential in the HZTO-nw and BCZT-mp which additionally aligns the C O dipoles in the direction of the applied force through stress induced polarisation. Accordingly, the use of the PLA biopolymer eliminates the complexity of the traditional electrical poling process in piezoelectric energy harvesters. 3.4. Some applications of the c-PNG In biomechanical energy harvesting, an effective piezoelectric nanogenerator is regarded to generate electrical energy from various motions (compressive stress, bending, stretching, folding, etc. ). 25,69,70 To shed more light on this concept, the ability of the c-PNG to generate electricity from bending motions was investigated. As presented in Fig. 7a , our energy harvesting system can operate effectively under bending/unbending motions under a reverse connection mode, and can generate a maximum voltage of around 2.5 V, with a peak-to-peak voltage of >6.5 V. The results of the voltage-stability testing under finger bending/unbending are shown in Fig. S8 (ESI) † . Fig. 7 The test results of the feasibility of the fabricated c-PNG device. (a) The generated open-circuit voltage of the c-PNG device when attached to the forefinger and stressed by successive biomechanical bending motions. (b) The measured voltage and calculated instantaneous output power under varied external resistance loads in the range from 1 kΩ to 50 MΩ (the inset shows the electrical circuit diagram used to measure the harvested output voltage). (c) The charging curve of a 1 μF-capacitor using the piezocomposite nanogenerator under sewing machine impartations (the inset shows the electrical circuit diagram used to harvest the electric energy from the c-PNG device (top) and an enlarged view of the charging curve (bottom)). (d) Schematic circuit diagram of the LEDs and calculator powering using the c-PNG. (e–g) Photographic images of the process of the red LED glowing under finger bending/unbending using the c-PNG. (h–j) Captured photographs showing various LEDs being lit up by the electric energy generated from the c-PNG device. (k) Captured photograph of a commercial calculator powered by the c-PNG under successive hand slapping impartations (the inset shows the calculator display turning on). \n Fig. 7b illustrates the load voltage and instantaneous power of the c-PNG device recorded as a function of various external resistive loads ( R L ) from 1 kΩ to 50 MΩ, as shown in the schematic circuit diagram in the inset of Fig. 7b . Increasing the load resistance gradually increases the output voltage to the open-circuit value. This behaviour is characteristic of piezoelectric nanogenerators. 71,72 The instantaneous power calculated by P max = V 2 / R L increases with increasing resistance up to a maximum value ( P max ) of 17.5 μW under an external load of 4 MΩ and then decreases upon further increasing the resistance. The obtained P max value is among the highest reported for piezoelectric nanogenerators at reasonably low load resistance. 3,19,27 The corresponding volumetric power density was calculated to be 9.2 mW cm −3 . This value is far higher than the values reported in the overwhelming majority of the previous reports on composite film based piezoelectric nanogenerators. 21,27,72–74 It was apprehended that the experimentally measured current ( I , 0.6 μA) is lower than that estimated theoretically using the equation (2.1 μA). This discrepancy is generally due to the power consumption of internal resistance present in the measurement system. 25,75 To check the feasibility of the c-PNG to be integrated as a powering device for small portable electronic applications and flexible high energy density capacitors, several tests were conducted. First, the c-PNG device was used to charge a 1 μF-capacitor under sewing machine impartations at a frequency of 23 Hz. The top inset of Fig. 7c shows the schematic circuit diagram used to track the accumulated voltage during the capacitor charging, which includes a full-wave bridge rectifier, 1 μF capacitor, and the piezoelectric nanogenerator. The bottom inset presents an enlarged view of the voltage–time curve recorded during the capacitor charging process. The c-PNG can charge the 1 μF capacitor up to 2.96 V in a very short time span of 10 s, which corresponds to U e = 1/2 CV 2 , to a stored energy of 4.38 μJ and an equivalent energy density of 2.3 mJ cm −3 . Here U e , C , and V refer to the stored electric density, capacitor capacity, and the generated voltage, respectively. Besides the energy storage capability, we also demonstrated the realistic usefulness of the c-PNG with energy generating capability to drive different types of commercial electronics, such as light-emitting diodes (LEDs) and a calculator. Using the schematic circuit diagram in Fig. 7d , the power generated from the c-PNG under cyclic finger tapping or hand slapping impartations was stored in a 10 μF-capacitor, and then could turn on: one red LED under finger bending ( Fig. 7e–g and Video S3 (ESI)), † one red LED under finger tapping ( Fig. 7h and Video S4 (ESI)), † one blue LED and two blue LEDs ( Fig. 7i and j and Video S5 (ESI)), † and a calculator ( Fig. 7k and Video S6 (ESI)) † after successive hand slapping impartations for five min. The results of these tests verify the c-PNG as a capable energy harvesting device for regular and irregular excitations present in our living or harsh environments and as a powering device for small portable electronics."
} | 7,760 |
38818385 | PMC11137249 | pmc | 135 | {
"abstract": "Deep neural feedforward networks are effective models for a wide array of problems, but training and deploying such networks presents a significant energy cost. Spiking neural networks (SNNs), which are modeled after biologically realistic neurons, offer a potential solution when deployed correctly on neuromorphic computing hardware. Still, many applications train SNNs offline , and running network training directly on neuromorphic hardware is an ongoing research problem. The primary hurdle is that back-propagation, which makes training such artificial deep networks possible, is biologically implausible. Neuroscientists are uncertain about how the brain would propagate a precise error signal backward through a network of neurons. Recent progress addresses part of this question, e.g., the weight transport problem, but a complete solution remains intangible. In contrast, novel learning rules based on the information bottleneck (IB) train each layer of a network independently, circumventing the need to propagate errors across layers. Instead, propagation is implicit due the layers' feedforward connectivity. These rules take the form of a three-factor Hebbian update a global error signal modulates local synaptic updates within each layer. Unfortunately, the global signal for a given layer requires processing multiple samples concurrently, and the brain only sees a single sample at a time. We propose a new three-factor update rule where the global signal correctly captures information across samples via an auxiliary memory network. The auxiliary network can be trained a priori independently of the dataset being used with the primary network. We demonstrate comparable performance to baselines on image classification tasks. Interestingly, unlike back-propagation-like schemes where there is no link between learning and memory, our rule presents a direct connection between working memory and synaptic updates. To the best of our knowledge, this is the first rule to make this link explicit. We explore these implications in initial experiments examining the effect of memory capacity on learning performance. Moving forward, this work suggests an alternate view of learning where each layer balances memory-informed compression against task performance. This view naturally encompasses several key aspects of neural computation, including memory, efficiency, and locality.",
"introduction": "1 Introduction The success of deep learning demonstrates the usefulness of large feedforward neural networks for solving a variety of tasks, but the energy cost associated with such networks presents an ongoing problem (Strubell et al., 2019 ). Neuromorphic computing platforms and spiking neural networks (SNNs), which model the power efficient properties of neural networks, offer a possible solution (Christensen et al., 2022 ). While recent advances allow SNNs to be trained offline (Neftci et al., 2019 ), these approaches only benefit from energy-efficient inference even though training continues to be the dominant energy bottleneck for deep learning. Though there are many strategies for training SNNs, it is widely believed that the most effective technique will be a biologically plausible learning rule (Zenke et al., 2021 ). While reproducing biology is not a strict requirement, the engineering constraints of neuromorphic hardware naturally align with biological constraints. Namely, we identify three defining properties of biologically plausible learning rules that directly impact energy efficiency: locality, asynchrony, and real-time processing . These three properties reduce the communication overhead and coordination required by a neuromorphic chip which are large sources of power consumption (Christensen et al., 2022 ). Unfortunately, training spiking neural networks directly on hardware is challenging, since the driving factor behind deep learning's success—back-propagation—is not considered to be biologically plausible (Lillicrap et al., 2020 ). Specifically, it is unclear how neurons might propagate a precise error signal within a forward/backward pass framework like back-propagation. A large body of work has been devoted to establishing plausible alternatives or approximations for this error propagation scheme (Balduzzi et al., 2015 ; Scellier and Bengio, 2017 ; Akrout et al., 2019 ; Lillicrap et al., 2020 ). While these approaches do address some of the issues with back-propagation, implausible elements, like separate inference and learning phases, still persist in many cases. Our work joins a body of recent literature that addresses biological plausibility by suggesting fundamentally different approaches to training networks from back-propagation (Payeur et al., 2021 ; Meulemans et al., 2022 ; Aceituno et al., 2023 ). These approaches modulate local Hebbian updates using top-down signals based on alternative objectives such as optimizing a control policy. Our work is similar in that we propose a dramatically different training objective. In contrast, we rely on recent advances in deep learning that train feedforward networks by balancing an information bottleneck objective (Ma et al., 2019 ). Unlike back-propagation, where an error signal computed at the end of the network is propagated to the front (see Figure 1A ), this method, called the Hilbert-Schmidt Independence Criterion (HSIC) bottleneck, applies the information bottleneck to each layer in the network independently. Layer-wise optimization is biologically plausible as shown in Figure 1B . Compared to related work, where the performance of the final layer affects training of prior layers through top-down signals, our objective is fully localized at each layer. Figure 1 (A) Sequential (explicit) error propagation requires precise information transfer backwards between layers. (B) Parallel (implicit) error propagation uses only local information in combination with a global modulating signal. Biological rules of this form are known as three-factor learning rules (Frémaux and Gerstner, 2016 ). Our contributions include: We show that optimizing the HSIC bottleneck via gradient descent emits a three-factor learning rule (Frémaux and Gerstner, 2016 ) composed of a local Hebbian component and a global layer-wise modulating signal. The HSIC bottleneck depends on a batch of samples, and this is reflected in our update rule. Unfortunately, the brain only sees a single sample at a time. We show that the local component only requires the current sample, and that the global component can be accurately computed by an auxiliary network. The auxiliary networks acts as a working memory with post-processing, and the effective “batch size” corresponds to its capacity. We demonstrate the empirical performance of our update rule by comparing it against baselines on synthetic datasets as well as MNIST (LeCun et al., 1998 ) and CIFAR-10 (Krizhevsky, 2009 ). To the best of our knowledge, our rule is the first to make a direct connection between working memory and synaptic updates. We explore this connection in some initial experiments on memory size and learning performance. 1.1 Preliminaries and related work Several works have presented approximations to back-propagation. Variants of feedback alignment (Lillicrap et al., 2014 ; Liao et al., 2016 ; Akrout et al., 2019 ) address the weight transport problem. Target propagation (Ahmad et al., 2020 ; Frenkel et al., 2021 ) and equilibrium propagation (Scellier and Bengio, 2017 ) propose alternative mechanisms for propagating error. Yet, all these methods require separate inference (forward) and learning (backward) phases. More recently, deep feedback control methods (Meulemans et al., 2022 ; Aceituno et al., 2023 ) use top-down signaling from a controller to optimize forward and backward weights concurrently. Unlike prior methods which address biological plausibility piecemeal, these techniques are plausible by design. We follow this approach to creating plausible learning rules, but we differ by focusing on layer-wise objectives instead of top-down control. Table 1 shows a comprehensive comparison between learning rule definitions. Only direct feedback control (DFC) and our work satisfy all objectives, but they represent two different solutions to the problem of biologically plausible learning rules. DFC is framed as a control problem with continuous dynamics, and the resulting weight update requires multi-compartment neurons. The authors note that this makes their work better suited for analog neuromorphic hardware. In contrast, our rule can be mapped to both digital or analog hardware, since time is denoted by sequences of samples and not physical time. Additionally, we do not put constraints on the neurons required to implement the rule. Table 1 A comparison of various learning algorithms categorized by four properties. \n Learning algorithm \n \n Weight transport-free? \n \n Local? \n \n Asynchronous? \n \n Real-time? \n Back-propagation (BP) ✗ ✗ ✗ ✗ Feedback alignment (FA) ✓ ✗ ✗ ✗ Direct FA (DFA) ✓ ✓ ✗ ✗ Single sparse DFA (SDFA) ✓ ✓ ✓ ✗ Equilibrium propagation (EP) ✓ ✓ ✗ ✗ Target propagation (TP) ✓ ✓ ✗ ✗ Direct random TP (DRTP) ✓ ✓ ✓ ✗ Plausible HSIC (pHSIC) ✓ ✓ ✓ ✗ Direct feedback control (DFC) ✓ ✓ ✓ ✓ Our work ✓ ✓ ✓ ✓ Weight transport-free rules do not require separate forward and backward networks with aligned weight parameters. Local rules utilize only locally available information. Asynchronous rules do not require a full forward pass before updating the weights of each layer. Real-time rules operate on samples arriving sequentially in time (not batches). Layer-wise objectives (Belilovsky et al., 2019 ; Nøkland and Eidnes, 2019 ), like the one used in this work, offer an alternative that avoids the weight transport problem entirely. Moreover, our objective emits a biologically plausible three-factor learning rule which can be applied concurrently with inference. Pogodin and Latham ( 2020 ) draw similar intuition in their work on the plausible HSIC (pHSIC) learning rule. But in order to make experiments with the pHSIC computationally feasible, the authors used an approximation where the network receives a batch of 256 samples at once. In contrast, their proposed biologically plausible rule only receives information from two samples—the current one and previous one—which reduces the accuracy of the HSIC estimate. This motivates our work, in which we derive an alternate rule where only the global component depends on past samples, while the local component only requires the current pre- and post-synaptic activity. Furthermore, we show that this global component can be computed using an auxiliary network. This allows us to achieve performance much closer to back-propagation without compromising the biological plausibility of the rule. 1.1.1 Other uses of information theoretic objectives for spiking neural networks Information bottleneck and other information theoretic quantities have been used in the context of training SNNs before. Yang and Chen ( 2023a , b ) utilize an information bottleneck objective in the final layer of an SNN to train networks that are robust to noisy input distributions. Yang and Chen ( 2023a ) improves on the standard information bottleneck by considering higher order terms. Similarly, Yang et al. ( 2022 ) trains networks with an additional minimum entropy criterion to promote robust learning in SNNs. Still, all works rely on back-propagation through time (BPTT) and surrogate gradient descent to train their SNNs. 1.1.2 Hardware substrates for implementing biological neural networks While our work does not directly deal with hardware implementations of biological networks, our contributions are motivated by the possible power efficiency benefits of biologically plausible learning rules. As such, we will briefly discuss various platforms for physical realization of neuromorphic computing. The landscape of neuromorphic hardware is vast and varied. At one extreme, platforms like Intel's Loihi (Davies et al., 2018 , 2021 ) use conventional CMOS technologies to create a digital array of biological neurons. While such systems are useful for exploring SNN applications, it is widely accepted that the primary power efficiency of neuromorphic hardware will come from novel device technologies. The most common devices are memristors (Yan et al., 2023 ) and resistive memory (Bianchi et al., 2023 ). Less common substrates based on metal-organic transistors (Wang et al., 2023 ) and thermal-guiding structures (Loke et al., 2016 ) exist as well. These devices are designed to mimic various functions of a biological synapse especially its plastic conductance. While the specifics differ, all devices have electrical properties that allow the conductance to be adjustable. Successful demonstrations of neuromorphic device arrays show their ability to simulate SNNs at a much lower power consumption than conventional computing. Yet, all rely on purely local update rules which fail to scale to very deep networks. Circumventing this limitation requires non-local circuitry, and the goal of any biologically plausible rule, including ours, is to limit the power overhead of these components. 1.1.3 Notation We will briefly introduce the notation used in the paper. Vectors are indicated in bold and lower-case (e.g., x ). Matrices are indicated in bold and upper-case (e.g., W ). Superscripts refer to different layers of a feedforward network (e.g., z ℓ is the ℓ-th layer). Subscripts refer to individual samples (e.g., x i is the i -th sample). Brackets refer to elements within a matrix or vector (e.g., [ x ] i is the i -th element of x ).",
"discussion": "4 Discussion In this work, we proposed a three-factor learning rule for training feedforward networks based on the information bottleneck principle. The rule is biologically plausible, and we are able to scale up to reasonable performance on MNIST. We do this by factoring our weight update into a local component and global component. The local component depends only on the current synaptic activity, so it can be implemented via Hebbian learning. In contrast to prior work, our global component uses information across many samples seen over time. We show that this content can be stored in an auxiliary reservoir network, and the readout of the reservoir can be used to modulate the local weight updates. To the best of our knowledge, this is the first biological learning rule to tightly couple the synaptic updates with a working memory capacity. We verified the efficacy of our rule on synthetic datasets, MNIST, and CIFAR-10, and we explored the effect of the size of the working memory capacity on the learning performance. Even though our rule does perform reasonably well, there is room for improvement. The rule performs best when it is able to distinguish between different high dimensional inputs. The resolution at which it separates inputs is controlled by the parameter, σ, in the kernel function ( Equation 4 ). The use of a fixed σ is partly responsible for the slow down in convergence in Figure 5 . In Ma et al. ( 2019 ), the authors propose using multiple networks trained with the different values of σ and averaging the output across networks. This allows the overall network to separate the data at different resolutions. Future work can consider a population of networks with varying σ to achieve the same effect. Addressing the resolution issue will be important for improving the speed and scalability of the learning method. Additionally, our rule is strongly supervised. While the mechanism for synaptic updates is biologically plausible, the overall learning paradigm is not. Note that the purpose of the label information in the global signal is to indicate whether the output for the current sample should be the same or different from previous samples. In other words, it might be possible to replace the term k ¯ ( y 0 , y p ) in Equation (7) with a binary teaching signal. This would allow the rule to operate under weak supervision. Alternatively, we could use contrastive learning, where output distribution, Y , is replaced by the output of a different network. Ideally, this other network should process a different, but related modality (e.g., a visual network and auditory network that are trained against each other using a contrastive approach). Most importantly, while our rule is certainly biologically plausible, it remains to be seen if it is an accurate model for circuitry in the brain. Since rules based on the information bottleneck are relatively new, the corresponding experimental evidence must still be obtained. Yet, we note that our auxiliary reservoir serves a similar role to the “blackboard” circuit proposed in Mumford ( 1991 ). This circuit, present in the thalamus, receives projected connections from the visual cortex, similar to how each layer projects its output onto the reservoir. Furthermore, Mumford suggests that this circuit acts as a temporal buffer and sends signals that capture information over longer timescales back to the cortex like our reservoir. So, while it is uncertain whether our exact rule and memory circuit are present in biology, we suggest that an in-depth exploration of memory-modulated learning rules is necessary. Even in the absence of a biological counter-part, our rule captures important properties necessary for neuromorphic hardware—locality, asynchrony, and real-time processing. We achieve this by suggesting a fundamentally different objective training deep neural networks, in line with recent work. We hope this work prompts further exploration of novel, non-back-propagation-based approaches for learning."
} | 4,427 |
31900444 | PMC7082336 | pmc | 136 | {
"abstract": "Reef corals are mixotrophic organisms relying on symbiont-derived photoautotrophy and water column heterotrophy. Coral endosymbionts (Family: Symbiodiniaceae), while typically considered mutualists, display a range of species-specific and environmentally mediated opportunism in their interactions with coral hosts, potentially requiring corals to rely more on heterotrophy to avoid declines in performance. To test the influence of symbiont communities on coral physiology (tissue biomass, symbiont density, photopigmentation) and nutrition (δ 13 C, δ 15 N), we sampled Montipora capitata colonies dominated by a specialist symbiont Cladocopium spp. or a putative opportunist Durusdinium glynnii (hereafter, C- or D-colonies) from Kāne‘ohe Bay, Hawai‘i, across gradients in photosynthetically active radiation (PAR) during summer and winter. We report for the first time that isotope values of reef corals are influenced by Symbiodiniaceae communities, indicative of different autotrophic capacities among symbiont species. D-colonies had on average 56% higher symbiont densities, but lower photopigments per symbiont cell and consistently lower δ 13 C values in host and symbiont tissues; this pattern in isotope values is consistent with lower symbiont carbon assimilation and translocation to the host. Neither C- nor D-colonies showed signs of greater heterotrophy or nutritional plasticity; instead changes in δ 13 C values were driven by PAR availability and photoacclimation attributes that differed between symbiont communities. Together, these results reveal Symbiodiniaceae functional diversity produces distinct holobionts with different capacities for autotrophic nutrition, and energy tradeoffs from associating with opportunist symbionts are not met with increased heterotrophy.",
"conclusion": "Conclusion Our analyses of M. capitata , therefore, provide three central findings. First, symbiont communities are capable of influencing the isotope values of the holobiont and this fact should be considered in the interpretation and sampling of corals across ecological gradients where symbiont species are partitioned. Second, proxies for assessing greater contributions of heterotrophy relative to autotrophy are more complex than previously recognized [ 48 ] and variance originating in host and symbiont δ 13 C values—including proxies for nutritional plasticity (i.e., δ 13 C H-S )—may be influenced by physiological factors (i.e., symbiont densities, photoacclimation) unrelated to prey capture. Finally, more nuanced approaches are needed to disentangle the influences of environmental and biological effects on coral carbon and nitrogen isotope values, including compound specific analyses of fatty acids and amino acids [ 5 , 47 , 48 , 93 ] and accounting for shifts in biomass composition [ 33 ]. Therefore, our findings reinforce the conclusion that facultative shifts in heterotrophic nutrition where they occur are species-specific [ 48 , 98 , 99 ] and context-dependent, perhaps limited to more extreme physiological conditions or to geographic locations favoring mixotrophy (i.e., high near-shore productivity [ 47 ]). Further tests of symbiont community effects on isotope values and the energetic consequences of flexible symbiont associations in reef corals should be prioritized in future studies [ 100 ].",
"introduction": "Introduction Nutrient exchanges between scleractinian corals and dinoflagellate symbionts (Symbiodiniaceae, formerly Symbiodinium spp.) [ 1 ] underpin the success of reef-building corals as habitat engineers in coral reef ecosystems [ 2 ]. Reef corals are mixotrophic, reliant on the translocation of symbiont-derived compounds [ 3 , 4 ] and heterotrophy in support of respiratory demands, calcification, and tissue growth [ 5 – 7 ]. In exchange, Symbiodiniaceae receive metabolic byproducts (i.e., CO 2 , NH 4 + ) required for growth and photosynthesis [ 8 ]. However, reef corals are increasingly threatened by climate change and local stressors, which disrupt the coral-algae symbiosis and reduce ecosystem services provided by coral reefs [ 9 , 10 ]. Associations with stress-tolerant symbionts can provide increased stress tolerance for corals albeit at the expense of coral performance [ 11 ], potentially requiring greater heterotrophic feeding to meet energy demands. The genetic and functional diversity of Symbiodiniaceae shapes the energy balance and stress tolerance of reef corals. Taxonomic resolution of Symbiodiniaceae has been achieved using genetic markers, primarily the internal transcribed spacer 2 (ITS2) region of nrDNA [ 12 , 13 ], and has revealed distinct symbiont genera and species (formerly clades and subclades) [ 1 ] with different capacities to support coral nutrition [ 14 , 15 ] and tolerate environmental stress [ 16 ]. For instance, symbionts in the genus Symbiodinium (formerly clade A) and Durusdinium (formerly clade D) are common on shallow reef zones of the Red Sea and Caribbean [ 17 , 18 ] and are also tolerant of high light and temperature stress. In particular, Durusdinium is observed in higher abundance on reefs impacted by local stressors or a history of bleaching [ 16 , 19 – 21 ]. However, some members of Symbiodinium and Durusdinium have been identified as opportunistic generalist symbionts, assimilating and transferring less carbon and nitrogen compounds to their coral hosts compared to specialist symbionts with high host specificity, such as members of Cladocopium and Breviolum genera (formerly clade C and B, respectively) [ 14 , 15 , 17 , 22 , 23 ]. As a consequence, corals in association with stress-tolerant symbiont species may be less sensitive to environmental change but incur energetic tradeoffs [ 24 ], including reduced autotrophic nutrition, lower tissue, and skeletal growth rates, and attenuated reproductive output compared to more mutualistic symbionts (i.e., Cladocopium and Breviolum ) [ 11 , 25 – 29 ]. To cope with reductions in symbiont-derived nutrition from less-mutualistic symbionts, it has been hypothesized that coral hosts may shift towards greater heterotrophy to meet metabolic demands [ 30 , 31 ], as has been observed in corals under experimental thermal stress ([ 32 ], but see in situ [ 33 ]) and on high turbidity reefs [ 34 ]. However, intraspecific changes in symbiont communities also reflect Symbiodiniaceae niche specialization, with some species being more efficient in uptake of carbon or nitrogen under low or high light and/or temperatures [ 17 , 23 ]. Therefore, it is important to unravel the ecological contexts and effects of symbiont communities on coral nutrition to determine the capacity for corals to be nutritionally plastic or to cope with opportunistic symbiont effects on performance [ 17 , 30 ]. Environmental factors, such as light availability [ 35 – 37 ], water quality [ 38 ], temperature [ 39 ], and bleaching history [ 40 , 41 ] are important in shaping the composition of symbiont communities. Changes in photosynthetically active radiation (PAR, hereafter ‘light’) influence the ecological niche of reef corals [ 42 ], and many coral species exhibit shallow-to-deep transitions in symbionts in favor of those adapted to lower light levels. For example, shallow colonies of Seriatopora hystrix in Western Australia [ 43 ] are dominated by Durusdinium , whereas deeper colonies are more often dominated by Cladocopium symbionts. Similarly, shallow-to-deep transitions from Symbiodinium (shallow) to Cladocopium (deep) have been observed in Stylophora pistillata [ 17 ] in the Red Sea and Orbicella faveolata in the Caribbean [ 18 ]. Where depth and turbidity attenuate light, corals can rely on the photoacclimation potential of their endosymbionts [ 43 ] and/or heterotrophy to prevent energy deficits [ 34 , 44 ]. Carbon and nitrogen stable isotope analyses of tissues have been applied to understand the balance of autotrophy and heterotrophy in reef corals and anemone model systems. These studies are widely interpreted as evidence for greater relative contributions of heterotrophy relative to autotrophy in corals during periods of resource limitation (i.e., reduced light availability) [ 45 ] or symbiosis disruption (i.e., coral bleaching) [ 46 ] and in regions characterized by high primary production [ 47 ]. However, the capacity for nutritional plasticity is not shared by all corals and depends on the biology of the coral host [ 48 ] and the optimization of skeletal morphology for light absorption or prey capture [ 49 ]. Recent research using corals and cnidarian model systems has also shown that the capacity for corals to increase heterotrophic feeding is also determined by Symbiodiniaceae communities [ 17 , 30 ]. For instance, Cladocopium -associated Stylophora pistillata corals exhibited greater feeding rates but lower inorganic nitrogen assimilation rates than shallow Symbiodinium -associated colonies [ 17 ]. Similarly, Durusdinium- associated anemones did not shift towards greater prey capture when carbon translocation was low and exhibited lower feeding and digestion rates compared to anemones with Symbiodinium minutum symbionts [ 30 ]. These studies have shown symbiont community composition can influence the trophic ecology of corals and anemones with downstream implications for host performance. Due to uncertain boundaries in symbiont community distributions, few inquiries have examined the role of symbiont type on corals across natural environmental gradients (but see refs. [ 35 , 43 ]) or symbiont-driven effects on coral feeding rates and the balance between autotrophic and heterotrophic nutrition. Yet, such inquires are necessary to understand the metabolic tradeoffs from distinct host-symbiont associations and symbiont effects on coral energy budgets. Here, we examine changes in the physiology and heterotrophic capacity (carbon and nitrogen stable isotope analyses) of a Hawaiian reef coral ( Montipora capitata ) dominated by algal symbionts Durusdinium glynnii (ITS2: D1-4-6 [ 50 ]) or Cladocopium spp. (ITS2: dominated by C31 [ 51 ], but see ref. [ 52 ]) from <1–10 m during summer and winter. M. capitata shows depth-dependent shifts in symbiont communities (greater Durusdinium sp. at shallow depths [ 36 ]) and can exhibit stress-induced changes in nutritional modes [ 32 ]. Therefore, we predicted greater heterotrophic feeding in M. capitata under significant light attenuation [ 45 , 47 ] and during association with opportunistic generalist symbionts (i.e., Durusdinium ). We observed greater symbiont densities and lower carbon isotopic values indicative of lower rates of carbon fixation and autotrophy in corals associating with Durusdinium , but neither Cladocopium- nor Durusdinium -dominated colonies exhibited greater heterotrophic feeding across light gradients or seasons. Isotope values in corals across depth gradients have been previously explored ([ 47 , 49 ]); however, our study offers the first evidence of symbiont community effects on host isotope values along a light gradient, where isotope values serve as proxies for coral productivity and nutritional modes. These findings reveal distinct traits of coral holobionts (symbiont and photopigment concentrations, carbon isotope values) that are determined by interactions between symbiont communities and environmental drivers, but do not lead to shifts toward greater heterotrophic nutrition.",
"discussion": "Discussion Symbiodiniaceae diversity, function, and community physiological traits have a clear influence on coral stable isotope values and these effects manifested across a narrow depth gradient in relation to rapid light attenuation. Species-specific symbiont attributes, including higher symbiont densities, lower chlorophylls per cell, and lower δ 13 C values indicate lower rates of carbon assimilation/translocation and potential host exploitation by Durusdinium glynnii . However, neither environmental nor symbiont community effects resulted in greater heterotrophic nutrition, suggesting photoacclimation may be central to coping with energetic tradeoffs incurred by corals in low-light environments as well as in association with opportunistic symbionts, such as D. glynnii . Finally, we identify symbiont community function as an important, yet often overlooked, component to isotopic investigations into coral physiological ecology. Light attenuation in Kāne‘ohe Bay was rapid across a narrow depth gradient with maximum PAR at 1 m-depth across all sites (ca. 1400–2300 μmol photons m −2 s −1 ) attenuated to 50–350 μmol photons m −2 s −1 at 8 m-depth—equivalent to PAR attenuation observed at 40–70 m in coral reefs of the Red Sea [ 73 ] and 20–40 m in the Caribbean [ 74 , 75 ]. Therefore, depth-dependent changes in light availability is a principal force acting to partition M. capitata symbiont communities with a greater frequency of Durusdinium observed at shallow depths across seasons and reef locations. This depth-dependent distribution in response to reduced light availability can in part be explained by close proximities to multiple watersheds and high concentrations of fine-grained particles that dominate inshore Kāne‘ohe Bay reefs [ 54 ], which settle slowly and are easily re-suspended, creating significant and persistent light attenuation [ 76 ]. Globally, the prevalence of Durusdinium is higher on human-impacted reefs, including those experiencing high temperatures and/or recent thermal stress and high rates of sedimentation [ 77 ]. Therefore, the shallow-water dominance of D. glynnii in Kāne‘ohe Bay M. capitata suggests a greater capacity for this species to tolerate a combination of more frequent temperature anomalies [ 20 , 51 ] and high irradiances [ 35 ] compared to Cladocopium spp. In Hawai‘i, reports of Durusdinium have been limited to M. capitata from Kāne‘ohe Bay, which has a history of persistent human impacts (dredging, sewage outflow) and high frequency of thermal stress anomalies [ 20 , 78 ]. This combination of human impacts and thermal stress anomalies may have allowed a niche for Durusdinium to exploit, while also driving local adaptation in the coral hosts for increased tolerance to thermal stress [ 79 ]. While Durusdinium prevalence in shallow depths (ca. < 3 m) indicates a tolerance to elevated light and temperature, the rarity of D-colonies at depth also indicate Symbiodiniaceae niche partitioning in response to light availability and perhaps a limited capacity for Durusdinium to adapt to low PAR or in response to changing light spectra [ 73 , 80 ]. Intraspecific shifts in symbiont communities generally occur at deeper depths, for instance Stylophora pistillata transitions from Symbiodinium microadriaticum (ITS2: A1) (<10 m) to Cladocopium spp. (>40 m) in the Red Sea [ 17 ], and Seriatopora hystrix transitions from Durusdinium (<23 m) to Cladocopium (>23 m) in western Australia [ 35 ]. However, high turbidity in Kāne‘ohe Bay has resulted in M. capitata symbiont community depth zonation to within a few meters (<2 m) of the surface ([ 36 ]; this study). D-colonies had lower δ 13 C values, indicating greater isotope fractionation and/or lower rates of carbon assimilation and translocation to the coral host compared to C-colonies [ 14 , 15 ]. While δ 13 C values are also influence by tissue properties, sources of inorganic carbon, and the contribution (and composition) of respiration CO 2 to the internal carbon pool [ 81 ], patterns observed here provide indirect evidence consistent with lower rates of photosynthesis and carbon fixation. M. capitata did not show signs of changes in heterotrophic feeding in response to changing light conditions, seasons, or due to hosting different symbiont species. However, substantial and persistent effects of symbiont community on isotope values were observed, which in conjunction with physiological attributes of the holobionts, reveal differing capacities of Cladocopium spp. and D. glynnii to fix and translocate carbon to their hosts. Photoacclimation had a clear influence on M. capitata host and symbiont δ 13 C values, indicated by a positive relationship with symbiont densities in winter and a negative relationship with photopigmentation (pg cell −1 ) (Fig. S8 ), and these effects were more pronounced in C-colonies. Changes in symbiont densities and photopigmentation are important photoacclimation responses to maximize light-use efficiency in reef corals [ 42 ] and influence holobiont energetics, calcification [ 29 ], and sensitivity to thermal stress [ 82 ]. Changes in chlorophylls, however, are but one mechanism corals employ to optimize light capture and attenuate damage from photoinhibition. Photoacclimation in shallow corals can also include increased concentrations in photoprotective pigments of the symbiont (xanthophyll, β-carotene) and host (fluorescent proteins), or greater dissipation of absorbed light-energy through non-photochemical quenching (i.e., NPQ) [ 83 ]. Similar mechanisms of modulating pigmentation and light-energy dissipation are also used by corals in deep-water habitats acclimated to low (and variable) light intensity of different spectral compositions. In the case of the latter, host-derived photoconvertible red fluorescent proteins (pcRFPs) appear to contribute to low-light adaptation in corals through the transformation of blue-to-orange light wavelengths that benefit symbiont light capture in deep-water corals [ 84 ]. Such changes in host pigmentation have not been tested in M. capitata but may be of importance considering this species exhibits depth-dependent distributions of two distinct color morphs (orange vs . brown) [ 36 ]. We observed D-colonies to have 54–58% greater symbiont densities and 0.8–1.6‰ lower δ 13 C values compared to C-colonies; however, C-colonies had 46% higher chlorophyll cell −1 and showed greater potential to regulate areal and symbiont cell-specific chlorophyll concentrations in response to changing environmental conditions. While Symbiodiniaceae cell size can influence symbiont densities, the coccoid cell sizes for other Durusdinium and Cladocopium species overlap [ 1 , 50 ]. High symbiont densities increase symbiont self-shading, which can reduce internal light values by 90% relative to those at the colony surface [ 85 ] and cause declines in net photosynthesis through light and/or carbon limitations. In both cases, reductions in photosynthesis rates (inferred here by lower δ 13 C values) reduce nutritional benefits to the host [ 82 ]. Changes in photosynthesis relative to respiration (P:R) can also be influenced by symbiont communities and environmental contexts, where metabolic savings from lower respiration rates with increasing depth (higher P:R) transition to metabolic costs (lower P:R) in symbionts unable to acclimate to light limitations [ 35 ]. Therefore, the clear differences in symbiont traits and isotope values between C- and D-colonies provide evidence for different physiological and biochemical processes in these two holobionts. In addition, the greater ability for C-colonies to photoacclimate in response to changing light-availability indicate light as a driver in the niche partitioning of these symbiont species. Differences in symbiont densities can also be influenced by Symbiodiniaceae growth rates and/or responses to nutrient availability [ 86 ]. For example, symbiont-derived photosynthates stimulate the recycling of ammonium waste by the host and nitrogen incorporation into amino acids, serving as a negative-feedback loop regulating symbiont densities by limiting nitrogen to the symbiont [ 87 ]. Paradoxically, reduced symbiont photosynthesis increases nitrogen availability to the symbiont, and as a result, symbiont biomass and densities may increase. Similarly, corals exposed to sub-bleaching thermal stress experience reduced carbon transferred from their symbionts, but symbiont carbon and nitrogen assimilation and growth increases [ 18 ]. Symbiont community effects on host, in particular through effects on ammonium metabolism, may also influence nitrogen availability and increase symbiont growth and densities [ 31 ]. For instance, Exaiptasia pallida anemones infected with Durusdinium trenchii increased the transport of 15 N-labeled products through the urea cycle and had greater glutamine pool enrichment (a primary enzyme in the assimilation of ammonium) compared to Breviolum minutum -hosting anemones, which did not show urea cycle feedback but had greater glutamine synthetase enzyme abundance [ 31 , 88 ]. Such effects on host metabolism dictated by Symbiodiniaceae community may serve to regulate nitrogen availability and symbiont densities and should be further tested in reef corals. Poor cellular communication and compatibility between symbiont partners in opportunistic or heterologous symbionts can elicit host immune responses [ 88 ] and adversely affect host performance [ 29 ]. Reduced communication and host-symbiont compatibility, therefore, may explain why shifts in symbiont communities are host specific and are generally limited to corals living near limits of their environmental tolerance [ 23 ]. Nevertheless, different capacities for Symbiodiniaceae to assimilate carbon and nitrogen in support of host growth [ 11 , 14 , 25 , 29 ] appear to also extend to influences on host metabolism and resource allocation [ 31 , 87 ], with one possible outcome being slower rates of production benefiting symbiont growth and favoring opportunism. Identifying the metabolic tradeoffs from hosting different Symbiodiniaceae communities requires further study, but may prove to be mechanisms by which symbionts benefit while imparting a metabolic cost to the coral host. Montipora capitata tissue (host, symbionts) and skeletal δ 13 C values declined as light decreased, in agreement with increased carbon isotope fractionation as carbon fixation rates decline in low-light environments [ 45 , 61 ]. During periods of high coral photosynthesis the internal carbon pool becomes enriched in 13 C (higher δ 13 C values) as dissolved inorganic carbon enriched in 12 C is preferentially fixed [ 81 ]. As a result, the discrimination (i.e., isotope fractionation) against heavy isotopes is reduced and δ 13 C values of photosynthetic products increase [ 89 ]. Conversely, reductions in carbon demand/photosynthesis allow for greater isotope discrimination and overall lower δ 13 C values in photosynthetic products and the internal carbon pool. In our study of M. capitata , we observed spatiotemporal changes in δ 13 C values to correspond between the host and symbiont, resulting in limited relative differences in carbon isotope values (i.e., δ 13 C H-S ), a commonly applied metric for greater heterotrophy (δ 13 C H-S values < 0) relative to autotrophy (δ 13 C H-S values > 0) [ 45 – 47 ]. In fact, δ 13 C H-S values became more positive as depth increased, which is opposite to the expected positive correlations between depth and coral heterotrophic nutrition [ 45 ]. Considering δ 13 C H-S would be expected to decline as symbiont densities are reduced in stressed corals, we further explored the relationship between δ 13 C H-S and symbiont densities. Again, we find unexpected results: δ 13 C H-S generally increases and becomes more positive in corals as symbiont densities decline (Fig. S9 ). This relationship may reflect optimal symbiont densities that maximize net photosynthesis and reduce cell shading in healthy corals. Nevertheless, these findings do not support greater heterotrophy in corals in response to light attenuation or reductions in symbiont densities and cast doubt on the effectiveness of using δ 13 C H-S as a proxy for greater heterotrophic nutrition in healthy, non-bleached corals as well as those undergoing symbiont loss due to environmental stress. Similar disagreements in using δ 13 C H-S values to infer depth-dependent changes in coral nutrition in other studies have been clarified using compound-specific isotope analyses. For instance, Stylophora pistillata and Favia favus showed no change in δ 13 C H-S across depths (<60 m) in the Red Sea. However, examining the δ 13 C values in lipids between host and symbionts suggested an increase in heterotrophic carbon usage by S. pistillata in lipid synthesis below 20 m but consistently low δ 13 C-lipid values in F. favus across depths, suggesting high and invariable heterotrophic contributions to lipid biosynthesis in F. favus [ 48 ]. In bleached and post-bleaching recovered M. capitata and Porites compressa δ 13 C values showed poor relationships with bleaching history and did not suggest a greater capacity for heterotrophic feeding; instead δ 13 C values were best explained by an isotope mass balance accounting for differences in proteins:lipids:carbohydrates in coral tissues [ 33 ]. It is therefore important for uncertainties in isotope values to be further explored in symbiotic mixotrophic organisms and for caution to be applied in the interpretation of these data to infer trophic plasticity. In the present study, changes in host and symbiont δ 13 C values appear to instead be an effect of the interaction between light-availability and symbiont community, in agreement with rapid internal cycling and a shared carbon source of the symbiont partners ([ 49 , 61 ]). While M. capitata biomass and C:N was not influenced by symbiont types or environmental conditions, it is important for future studies to also consider the role of Symbiodiniaceae on coral metabolite profiles [ 31 , 90 , 91 ] which may influence tissue biochemical composition and isotope values [ 33 , 35 , 92 , 93 ]. Skeletal carbonate δ 13 C values (i.e., δ 13 C Sk ) are expected to become more negative as P:R decline [ 94 ] or as an effect of greater metabolic fractionation and increased respiratory-derived carbon with low δ 13 C values used in biomineralization [ 95 ]. However, M. capitata showed relatively small differences in δ 13 C Sk and δ 13 C H-S values across light environments and seasons, suggesting a dominance of autotrophic sources and the maintenance of nutrient recycling between partners. Corals also did not show a decrease in δ 15 N values with depth, as would be predicted as photosynthesis becomes light-limited or if corals are feeding on a prey source with low δ 15 N values [ 96 ]. Considering the δ 15 N values of heterotrophic sources in seawater (Fig. S4 ), corals and their symbionts most resemble the δ 15 N values of dissolved inorganic nitrogen (3.8–4.9‰ [ 33 ]) and not plankton end-members. Therefore, M. capitata appears to have maintained high rates of photosynthesis (or P:R) as to not show systematic decreases in δ 15 N or 13 C Sk values and does not show evidence for increased heterotrophy with depth. Predator 15 N-enrichment factors typically observed in food webs (~3.5 ‰) appear absent or substantially attenuated in corals. The cause for this attenuated trophic enrichment may be nitrogen recycling, wherein symbionts uptake host excreta with low δ 15 N values (i.e., 14 NH 4 ) for amino acid synthesis and translocate a greater proportion of 14 N-products to the host, effectively lowering δ 15 N values for the holobiont [ 62 ]. However, a degree of predator enrichment would still be expected as coral predators and their symbiont retain “heavy” isotope products and exchange “light” isotope products, unless internally recycled nitrogen dominates the host’s nitrogen budget. Alternatively, the isotopic signal originating in prey capture may be overwhelmed by internal and external DIN uptake and nitrogen recycling by Symbiodiniaceae. Further experimental measurements of isotopic compositions of nitrogen sources and downstream products in host and symbiont tissues (i.e., amino acids) may provide a way to clarify heterotrophic influences on coral isotope values (for carbon, see ref. [ 97 ])."
} | 7,006 |
21031386 | null | s2 | 137 | {
"abstract": "A bio-inspired approach for superhydrophobic surface modification was investigated. Hydrophilic conversion of the superhydrophobic surface was easily achieved through this method, and the superhydrophobic-hydrophilic alternating surface was generated by the method combined with soft-lithography. The resulting patterned surface showed high water adhesion property in addition to superhydrophobic property."
} | 101 |
20520829 | PMC2876029 | pmc | 138 | {
"abstract": "Background Cyanobacteria account for 20–30% of Earth's primary photosynthetic productivity and convert solar energy into biomass-stored chemical energy at the rate of ∼450 TW [1] . These single-cell microorganisms are resilient predecessors of all higher oxygenic phototrophs and can be found in self-sustaining, nitrogen-fixing communities the world over, from Antarctic glaciers to the Sahara desert [2] . Methodology/Principal Findings Here we show that diverse genera of cyanobacteria including biofilm-forming and pelagic strains have a conserved light-dependent electrogenic activity, i.e. the ability to transfer electrons to their surroundings in response to illumination. Naturally-growing biofilm-forming photosynthetic consortia also displayed light-dependent electrogenic activity, demonstrating that this phenomenon is not limited to individual cultures. Treatment with site-specific inhibitors revealed the electrons originate at the photosynthetic electron transfer chain (P-ETC). Moreover, electrogenic activity was observed upon illumination only with blue or red but not green light confirming that P-ETC is the source of electrons. The yield of electrons harvested by extracellular electron acceptor to photons available for photosynthesis ranged from 0.05% to 0.3%, although the efficiency of electron harvesting likely varies depending on terminal electron acceptor. Conclusions/Significance The current study illustrates that cyanobacterial electrogenic activity is an important microbiological conduit of solar energy into the biosphere. The mechanism responsible for electrogenic activity in cyanobacteria appears to be fundamentally different from the one exploited in previously discovered electrogenic bacteria, such as Geobacter , where electrons are derived from oxidation of organic compounds and transported via a respiratory electron transfer chain (R-ETC) [3] , [4] . The electrogenic pathway of cyanobacteria might be exploited to develop light-sensitive devices or future technologies that convert solar energy into limited amounts of electricity in a self-sustainable, CO 2 -free manner.",
"introduction": "Introduction Cyanobacteria are of profound biological and biogeochemical importance. Oxygenic photosynthesis carried out by primitive cyanobacteria transformed early Earth's reducing atmosphere into an oxidizing one 2.4 billion years ago and provided for the evolution of complex aerobic life below a protective ozone layer [5] . From the earliest fossil record up to the present, cyanobacteria have continued to thrive, adapt and support higher life by converting solar energy into energy-dense organic compounds. Diverse genera of mat-building and planktonic cyanobacteria are found all over the world, from temperate ponds to some of the driest and most inhospitable environments imaginable, where they serve key ecological roles in energy transduction, nitrogen fixation and as pioneer species [2] . Indeed, these solar-powered prime movers of global nitrogen and carbon cycling probably represent the most important primary producers in the ocean and they colonize barren rock as new land is created through volcanic activity [6] . An astonishing 50% of the planet's biological nitrogen fixation is attributable to tropical and subtropical marine cyanobacteria [7] . In light of their tremendous ecological importance, deeper investigation into the mechanisms by which cyanobacteria convey solar energy to the environment is warranted. Solar energy reaches the Earth at the rate of the 178,000 TW [8] of which 0.2% to 0.3% is harnessed by cyanobacteria [1] . The amount of energy that passes through cyanobacteria exceeds by more than 25 times the energy demand of human society (∼15 TW); roughly 1,000 times the energy produced by all the nuclear plants on Earth. On a global scale, cyanobacteria fix an estimated 25 Giga tons of carbon from CO 2 per year into energy dense biomass [1] , [9] . The purpose of this study is to investigate whether cyanobacteria possess light-dependent electrogenic activity (i.e. are capable of depositing electrons to the extracellular environment in response to illumination), to test whether the electrogenic activity is generic and to probe the mechanism of the electrogenic response.",
"discussion": "Discussion The current studies show that cyanobacteria exhibit light-dependent electrogenic activity and that this activity is conserved among diverse genera. Previously, only chemotrophic bacteria were found to display intrinsic electrogenic activity [4] , [16] . In electrogenic chemotrophs such as Geobacter sulfurreductens , electrons are derived from oxidation of organic compounds and transported via a respiratory electron transfer chain (R-ETC) to extracellular terminal electron acceptors [3] . In Geobacter , the electron transfer to the environment was shown to be mediated by a diverse group of c-type cytochromes and possibly involved electrically-conductive microbial nanowires [17] – [19] . In previous studies, exogenous electron mediators such as 2-hydroxy-1, 4-naphtoquinone were employed for recording electrogenic activity from cyanobacterial cultures [20] , [21] . Because this synthetic quinone is capable of intercepting electrons at numerous sites along the R-ETC and the P-ETC, it was difficult to determine whether cyanobacteria possess natural mechanism(s) for electron discharge to the extracellular environment and, if they do, whether R-ETC or P-ETC is the source of electrons. The present work expands our basic understanding of bacterial electrogenesis and demonstrates that diverse cyanobacteria exhibit light-dependent electrogenic activity. In contrast to electrogenic chemotrophs, cyanobacterial electrogenic activity was observed in the absence of any exogenous organic fuel and was driven entirely by the energy of light. We show that the P-ETC is the source of electrons and that PQ plays a central role in electron flow from cyanobacteria to the extracellular environment; in this case the MFC anode. In cyanobacteria PQ is known to be present in substantial amounts in both cellular and thylakoid membranes [9] , [22] . Inhibition of PS-II by DCMU or CCCP rapidly reduced the light-dependent electrogenic activity in Nostoc and Lyngbya , providing support that electrons originate from PS-II-mediated photolysis of water ( Fig. 5 ). CCCP inhibits electron flow through PS-II but can also uncouple the electron-transport-dependent ATP synthesis by inactivating ATPase [23] . DCMU, however, is known to be a specific PS-II inhibitor that blocks binding of PQ to PS-II [24] , [25] . Blocking cytochrome b6f activity with DBMIB resulted in a significant increase in electrogenic activity. DBMIB is a PQ analogue that binds cytochrome b6f and prevents electron transfer from reduced PQ to cytochrome b6f [22] , [26] , [27] . The opposing effects of DCMU and DBMIB indicate electrons probably exit P-ETC downstream of PS-II and upstream of cytochrome b6f. PQ is the only P-ETC component located between these two P-ETC complexes. Furthermore, the rescue of DCMU-induced inhibition by duroquinone, which has a very high capacity as a PS-II electron acceptor and can shuttle water-derived electrons from PS-II [14] , strongly support the mechanism that electrons originate from PS-II-mediated photolysis of water. Inhibition of electrogenic activity by PMA was likely due to interference of PMA with the Q-cycle of electron transfer [12] , [13] . In the Q-cycle, electrons are transported from PS-I to ferredoxin, and then fed back to PS-I through PQ and cytochrome b6f complex ( Fig. 5C ). P-ETC inhibitors acted in a similar fashion in both Nostoc and Lyngbya implying that the mechanism responsible for electrogenic activity is conserved. Although the biological function of electrogenic activity in cyanobacteria is not yet clear, it is tempting to speculate that it could help regulate the redox state of PQ by shedding excess energy to the environment when PQ becomes over reduced in high light. Future studies will examine this possibility. An independent approach that employed lights of different color strongly supported a direct link between electrogenic activity and the cyanobacterial photosystem. The fact that electrogenic response was observed only under red or blue light is consistent with the notion that the electrons are donated to the extracellular environment via the P-ETC ( Fig. 6 ). Green light, which is not absorbed by light-harvesting pigments, was unable to induce a positive light response. The electrogenic response under red light was found to be almost as high as that under the reference white light. This result is consistent with the fact that red light absorbing phycobilisomes play a major role in light harvesting in cyanobacteria. The differences in the ratio of photosynthetic pigments in individual genera could explain the different amplitude of the electrogenic response between Nostoc and Lyngbya under the blue light ( Fig. 6 ). While it is tempting to speculate electrogenic activity may provide a means of shedding excess energy under intense light, other functions, such as a role in carbon fixation, might also be possible. For efficient carbon fixation, cyanobacteria actively import HCO 3 \n − using membrane-spanning transporters [28] . Interestingly, some cyanobacteria that lack systems for CO 2 uptake still possess HCO 3 \n − uptake systems, indicating HCO 3 \n − is the preferred inorganic source for carbon fixation [29] . One of the reason for preferential HCO 3 \n − uptake is that the HCO 3 \n − anion is much less membrane permeable than CO 2 , which readily diffuse out of cells. Once inside the cell, cytosol accumulated HCO 3 \n − is converted to CO 2 for carbon fixation by carboxysomal carbonic anhydrase (HCO 3 \n − +H + >H 2 CO 3 >CO 2 +H 2 O). In the present study, pH and dissolved oxygen rose to high levels after illumination of diverse genera ( Figure S3 ) and a voltage spike was observed whenever the light turned on. One possibility, to be followed up on by future studies, is that electrogenic activity could relate to the carbon concentrating mechanism and intracellular filling of the inorganic carbon reserves upon illumination. Indeed, previous researchers have suggested CO 2 entering cells can become trapped intracellularly by conversion to HCO 3 \n − through a P-ETC associated pathway [29] . When pure CO 2 was administered to MFCs electrogenic activity dropped sharply before rebounding within a few minutes indicating electron donation to the extracellular environment had been temporarily interrupted (data not shown). In the future, MFCs can serve as tools to study electrogenic activity as it may relate to the carbon concentrating mechanism. In anaerobic chemotrophs, electrogenic activity is a constituent part of bacterial respiration, whereas in cyanobacteria we speculate electrogenic activity might help cells adapt to unfavorable light conditions. To cope with the adverse effects of intense sunlight cyanobacteria have evolved several protective mechanisms [30] . To prevent ultraviolet damage mat-building cyanobacteria synthesize sunscreens such as scytonemin which accumulates in trichome sheaths [31] . Inducible non-photochemical quenching can limit the solar energy conveyed to PS-II by increasing the amount of light dissipated as heat [32] – [34] . Negative phototaxis triggered by increasing ROS concentration allows some motile cyanobacteria to shield themselves from intense sunlight and gain exposure to optimal intensity light by burrowing downward [35] , [36] . Despite this broad diversity, the currently known adaptive responses may not be effective against rapid fluctuations in light intensity, suggesting other protective mechanisms could also exist. Considering that light-dependent electrogenic activity is conserved among diverse genera of cyanobacteria, one possibly is that electrogenic activity might serve as a protective mechanism to adverse environmental conditions, such as rapid fluctuation or high intensity of sunlight. Our study found that a cyanobacteria-containing biofilm consortium exhibited electrogenic activity in a manner similar to that of individual cultures ( Fig. 4 ). DGGE analysis of 16S and 23S rRNA genes indicated the presence of several mat building cyanobacteria. Gene sequence analysis of the electrogenic pond biofilm indicated these cyanobacteria were phylogenetically most similar to Phormidium , Leptolyngba and Pseudanabaena . A green algae and chemotrophic bacteria were also indicated. Minor microbial constituents of the biofilm may not have been detected by DGGE analysis. Nevertheless, each of these three cyanobacterial genera showed light-dependent electrogenic activity when grown as individual cultures ( Fig. 1 ). Cyanobacteria are the most successful mat-building organisms. They form the topmost, aerobic layer of microbial mats where access to light, atmospheric CO 2 and N 2 is greatest [37] . A layer of oxidized iron may separate the cyanobacterial oxygenic layer from a lower anoxygenic layer composed of purple sulfur and green sulfur bacteria. In marine mats, anaerobic sulfate-reducing bacteria can be found throughout the mat below the top layer of cyanobacteria. Sulfate-reducing bacteria play a major role in decomposing organic materials produced by cyanobacteria. The joint metabolic activity of microorganisms in mats results in steep gradients of light, oxygen, carbon dioxide, pH, and redox potential [37] . One untested possibility is that rather than merely shedding excess solar energy to the abiotic environment, cyanobacteria might donate excess water-derived electrons to biofilm symbionts. The ability of cyanobacteria to donate electrons directly to the extracellular environment (i.e. anode) was illustrated by the decreasing anode potential observed when a Nostoc containing half MFC was exposed to light ( Fig. 3 ). An anode without cyanobacteria displayed no such response, indicating this phenomenon is mediated by cyanobacterial cells. The relatively low yield of electron harvesting by extracellular acceptors observed in our experiments was in part due to high amounts of dissolved oxygen present at concentrations substantially exceeding those found in natural cyanobacterial mats. However, the yield of electron harvesting could be improved by as much as 4.5-fold simply by changing the nanostructure of the anode surface. In natural cyanobacterial mats, the yield of electron discharge is likely to be variable, depending on the chemical environment, the physical properties of electron acceptors, the intensity of solar radiation, the concentration of dissolved oxygen and other factors. Even if a very modest yield is used for estimating the average rate of electron discharge, the transfer of solar energy to the environment via cyanobacterial electrogenic pathway could proceed at the rate of ∼9 TW on a global scale. Therefore, the electrogenic pathway appears to be an important microbiological conduit of solar energy into the biosphere and could have significant impact on a global scale. Anticipated applications of the electrogenic activity of cyanobacteria described here might be the biological conversion of solar energy to electrical energy or self-sustainable light sensors. Although at present the conversion yield is quite low, future studies on improvements in anode design, genetic manipulations of P-ETC or strain selection will answer the question of whether self-sustainable, CO 2 free technologies based on the light-dependent electrogenic activity of cyanobacteria are feasible."
} | 3,914 |
35541649 | PMC9080775 | pmc | 140 | {
"abstract": "To understand the effect of chemical composition on the anti-icing properties of a nanostructured superhydrophobic surface (SHP), four SHP surfaces were prepared on glass, which was initially roughed by a radio frequency (RF) magnetron sputtering method and then modified with HDTMS (a siloxane coupling agent), G502 (a partially fluorinated siloxane coupling agent), FAS-17 (a fully fluorinated siloxane coupling agent) and PDMS (a kind of polysilicone widely used in power transmission lines). Results show that the anti-icing properties of these four SHP surfaces in glaze ice varied wildly and the as-prepared SHP surface which was modified with FAS-17 (SHP-FAS) demonstrated a superior anti-icing/frosting performance. Approximately 56% of the entire SHP-FAS remained free of ice after spraying it for 60 min with glaze ice, and the average delay-frosting time (the time taken for the whole surface to become covered with frost) was more than 320 min at −5 °C. Equivalent model analysis indicates that Δ G , defined as the difference in free energy of the Cassie–Baxter and Wenzel states, of the SHP-FAS is much lower than the other three SHP surfaces, giving priority to Cassie state condensation and the self-transfer phenomenon helping to effectively inhibit the frosting process by delaying the ice-bridging process, which is beneficial for improving the anti-frosting property. This work sheds light on and improves understanding of the relationship between anti-icing and anti-frosting properties and is helpful in making the optimum selection of a surface modifier for improving the anti-frosting/icing performances of a SHP surface.",
"conclusion": "4. Conclusion We have studied the anti-icing and anti-frosting properties of four SHP surfaces with different chemical compositions but the same nano-scale structures. Results show that SHP-FAS demonstrates the optimal anti-icing/frosting property compared to the other three SHP surfaces. Only 44% of the SHP-FAS surface froze after spraying it for 60 min in glaze ice and the frosting time is effectively delayed by over 320 min at −5 °C. Investigating the dynamics of the condensed water droplets has shed light on the source of the varied anti-icing/frosting property, which is determined by the mode of condensation on the SHP surfaces which have been modified with various surface modifiers. As the Δ G values are both negative for SHP-FAS and SHP-HDTMS, water vapor is prone to condense in the Cassie state leading to a self-transfer movement. The self-transfer movement can help sweep the SHP surface free of condensed water droplets and also inhibit the ice-bridging rate, which contributes to an enhancement in the anti-icing/frosting properties of the SHP surfaces. We can conclude that the chemical composition has a large impact on the anti-frosting/icing performance of the SHP surfaces, which can be greatly improved by choosing an optimum surface modifier.",
"introduction": "1. Introduction Ice accumulation adversely affects numerous commercial sectors including aircrafts, telecommunications, automotives, wind turbines and power lines. 1 A glass insulator is one of the most important components in a transmission line and the icing of an insulator surface may lead to serious consequences, such as flashover incidents, collapsing of towers and power failures. 2,3 Traditional anti-icing methods including AC-hot de-icing methods and manual de-icing methods include problems with time consumption and low efficiency. Meanwhile, the irregular shape of insulators hinders the development of automatic de-icing devices. Therefore, measures to inhibit the accumulation of ice on surfaces which are exposed to cold regions are urgently needed. Inspired by the “lotus effect”, 4,5 superhydrophobic (SHP) surfaces are believed to be potential candidates for achieving anti-icing properties because of their excellent water repelling abilities. Considerable interest has been generated for utilizing SHP surfaces for anti-icing applications. 6 The SHP surface has been proven to be helpful in improving anti-icing properties by delaying ice formation, 7 enhancing the dynamic anti-icing behavior of water droplets impacting the SHP surface, 8 reducing the ice adhesion strength, 9 and so forth. Slippery liquid-infused porous surfaces (SLIPS) also demonstrate an excellent ice adhesion force. But the SLIPS may not be suitable for application on transmission lines because of the high voltage. 10–13 Huang et al. prepared a SHP surface by mixing silica sol and fluorinated acrylate copolymers and found that ice formation was delayed by 90 min at −5.6 °C. 14 Ruan et al. studied the anti-icing properties of SHP aluminum surfaces and found that the freezing temperature of static water droplets was lowered from −2.2 °C down to −6.1 °C. 15 Meanwhile some other researchers doubt the practical application of SHP surfaces in anti-icing applications. 16,17 They claim that frost formation could significantly compromise the icephobic properties of SHP surfaces, and frost nucleation could occur on all areas of the SHP surface leading to a loss of superhydrophobicity at low temperatures and high humidity. 18 The difference in the micro/nanostructures is believed to be the main reason for the above results. Recently, sporadic reports have shown that SHP nanostructures or hierarchical structures can effectively retard frosting. 19–21 Wen et al. designed a composite micro/nanostructured surface by using poly(vinylidene difluoride) (PVDF) polymer in combination with ZnO materials via heat-pattern-transfer and crystal-growth techniques, which displayed excellent anti-fogging and icing-delay properties. 22 Guo et al. investigated the dependence of the anti-frosting properties on the nano–micro/structure surface. A long frosting delay could be achieved for more than 185 min on the nanohair surface with a ratchet period of 290 μm under a temperature of −10 °C. 23 Hence, providing a SHP surface with optimum nano-scale textures is essential for improving the anti-icing properties. The general process for fabricating SHP surfaces involves two steps: construction of the micro/nanostructures and modification with low-surface-energy materials. 24,25 Apart from the textures, we believe that the surface modification, as one of the main fabrication steps, also has a large effect on the anti-icing properties of the SHP surfaces especially on the same desired nano-scale structures. 26–29 Commonly used surface modifiers mainly include long-chain alkanes, partially or fully fluorinated silanes and polydimethylsiloxane. Although considerable work has been conducted on modifying the structures in order to enhance the anti-frosting/icing properties of the SHP surfaces, a systematic investigation on the effect of chemical composition, endowed by various surface modifiers, on the frosting/freezing process has rarely been incorporated. Another key factor, apart from intrinsic parameters such as textures and chemical compositions, leading to a differentiation in the anti-icing properties is the variation in experimental methods. The experimental methods commonly used to characterize anti-icing properties of SHP surfaces include water droplet impact dynamics, delay-icing time of water droplets, the freezing temperature of water droplets or anti-fogging properties et al. However, these experimental conditions are far from real freezing weather conditions with regards to water droplet size, ambient temperatures, rainfall, relative humidity and so forth. The same roughness which prevents the accumulation of ice under certain conditions can be detrimental in other environments. 30 A single nanostructure SHP surface showed a robust anti-icing performance owing to supercooled micro-droplets partly rebounding off its surface in a wind tunnel. 31 Therefore, the accurate evaluation of anti-icing properties should be based on a reference environment. 31 Both field and laboratory investigations show that glaze ice is the most dangerous type of ice, associated with the highest probability of flashover in transmission lines. 32,33 However, studies on the anti-icing properties of SHP surfaces with various surface modifiers in glaze ice are still sorely lacking, and these are vital for the design of anti-icing coatings which can be applied to transmission lines. Besides, the SHP surfaces are exposed to an environment of low temperature and high humidity. Vapor condensation or the frosting process will inevitably occur during the freezing process of glaze ice. Investigations on the anti-frosting properties of a SHP surface is necessary in order to shed light on the anti-icing performance in glaze ice. As stated in our previous work, microstructure defects on SHP surfaces would cause an adverse effect on the anti-icing performance in glaze ice as the continually impacting supercooled micro-scale water droplets rapidly accumulate on these defects leading to a dramatic increase in the freezing area. 34,35 Therefore, to accurately evaluate the effect of surface modification on the anti-icing property, the as-prepared SHP surfaces should be carefully prepared to minimize defects. Compared to other methods, a radio frequency (RF) magnetron sputtering method represents one of the simplest and most effective methods for easily developing proper nano-scale structures. 36–38 A SHP ZnO surface modified with HDTMS was prepared by an RF magnetron sputtering method as reported in our previous work, and exhibits excellent anti-icing performance. 39 The focus of the present work is to further improve the anti-icing properties of a ZnO SHP surface by selecting an optimum surface modifier. The excellent process stability of RF magnetron sputtering is beneficial for obtaining uniform nano-scale textures and for decreasing the number of defects as much as possible, which is helpful to accurately evaluate the effect of surface modifiers on the anti-icing properties. In this study, an RF magnetron sputtering method was used to fabricate ZnO nanostructures, and four typical surface modifiers were used to modify the ZnO nanostructure to obtain superhydrophobicity. The surface morphology, chemical composition and wettability were investigated by corresponding methods. An artificial climate chamber was used to evaluate the effect of surface modifiers on the anti-icing properties of the nanostructured SHP surfaces in glaze ice. Furthermore, macroscopic and microscopic frosting processes of the four SHP surfaces were studied on a Peltier-based platform.",
"discussion": "3. Results and discussion 3.1 Surface morphology and wettability \n Fig. 1 shows the SEM images of the samples before and after surface modification. After an annealing treatment at 400 °C, the Zn layer was transformed into ZnO with a hexagonal crystal phase (shown in Fig. S2 † ). The as-prepared ZnO SHP surfaces exhibited evident surface roughness with ZnO nanostructures randomly distributed on the whole surface. As shown in Fig. 1b with a higher magnification, the diameters of these nanostructures ranged from 42.1 nm to 135.1 nm. As shown in the SEM images of SHP-HDTMS, SHP-FAS, SHP-PDMS and SHP-G502, there is no apparent change of surface morphology observed after surface modification. Although some organic thin film may appear on the surfaces, the surface morphology remains relatively consistent, which is further confirmed by the measured surface roughness results of shown in the ESI, Table S1. † The roughness average ( R a ) and the root-mean-square roughness ( R q ) of the four SHP surfaces are almost consistent and within a range of 470 ± 5% nm and 565.4 ± 8% nm. Therefore, we can reasonably consider that the four SHP surfaces possess the same surface morphology. Fig. 1 SEM images of the ZnO nanostructures before (a and b) and after modification with FAS-17 (c and d), HDTMS (e and f), G502 (g and h), and PDMS (i and j). As shown in Fig. S3, † the contact angles (CAs) of the glass surfaces modified with PDMS, G502, HDTMS, and FAS-17 are in the range of 93.7° to 108.9° demonstrating their hydrophobicity. 40 Water droplets easily adhered onto the glass surfaces modified with PDMS, G502, and HDTMS leading to sliding angles (SAs) larger than 66.8°. The SAs of glass surface modified with FAS-17 are lower than 20°. The as-prepared unmodified ZnO surface indicates a completely superhydrophilic property with water droplets almost completely drilling into the textures. Fig. 2 shows the CAs and SAs of the four as-prepared SHP surfaces. All of the surfaces exhibit superhydrophobicity with CAs larger than 160° and SAs smaller than 10° as shown in Table S2. † Water droplets can easily roll off the four SHP surfaces. Fig. 2 CAs and SAs of the four as-prepared SHP surfaces. 3.2 Chemical composition Both EDS and XPS were used to analyze the chemical composition of the as-prepared samples before and after surface modification. The geometries of the FAS-17, G502, HDTMS, and PDMS ( n = 10) molecules are illustrated in Fig. S4. † As shown in Fig. S5, † the EDS spectrum of the ZnO nanostructures prior to surface modification shows no other peaks apart from those of Zn and O, indicating the existence of ZnO. After modification with FAS-17 or G502, C, Si and F elements were detected in the EDS spectrum and a significant F(1s) peak appeared at 689.0 eV in the XPS spectrum, proving the existence of FAS-17 or G502 molecules. 41 Both EDS and XPS spectra reveal the presence of C after modification with HDTMS and PDMS molecules indicating that the HDTMS and PDMS molecules had successfully attached onto the ZnO nanostructures. These results confirm the successful attachment of low-surface-energy molecules to the ZnO surface after surface modification. 3.3 Anti-icing performance in glaze ice To further investigate the effect of the surface modifiers on the anti-icing performance of the four SHP samples, an artificial climate chamber was used to simulate freezing weather. “Glaze ice” represents one of the most serious freezing conditions and thus was simulated to test the anti-icing performance of the four SHP surfaces. 32 A schematic of the artificial climate chamber is shown in Fig. S6 † and the experimental conditions are set according to Table 1 . Experimental conditions for glaze ice Ambient temperature (°C) −8 ± 1 Water temperature (°C) 4 ± 1 Wind velocity (m s −1 ) 3 Water flow rate (L h −1 ) 90 Water conductivity (μs cm −1 ) 255 Though the SA of the glass surface modified with FAS-17 is lower than 20°, micro-scale water droplets can easily adhere onto the surface leading to poorer anti-icing properties compared to those for SHP-FAS, which indicates that nano-scale roughness is essential for enhancing the anti-icing performance in glaze ice, as shown in Fig. S7. † To shed light on the effect of surface modifiers on the anti-icing performance of nanostructured SHP surfaces, the four as-prepared SHP surfaces were placed in the artificial climate chamber. Fig. 3 and 4 show the freezing process of the four SHP surfaces in glaze ice over a 60 min time period. The freezing areas in glaze ice are the average values of two groups of identical samples. Though the four SHP surfaces demonstrate excellent water repellency at ambient temperatures, their anti-icing properties in glaze ice vary greatly. Supercooled water droplets were found to easily adhere onto the SHP-PDMS and SHP-G502 surfaces, which soon became covered with isolated ice particles. Almost 90% of the SHP-PDMS and SHP-G502 surfaces had frozen within 30 min. The ice particles slowly grew into long column-like icicles after a prolonged spraying time. SHP-HDTMS shows a better anti-icing performance with 52.1% of the surface remaining unfrozen after 30 min which increased up to 78.4% after 60 min. SHP-FAS exhibits the best anti-icing performance. Only 44.3% of the SHP-FAS surface had frozen after 60 min which is half of that for the SHP-PDMS and SHP-G502 surfaces. Due to the hydrophilicity of the edges, when sprayed, the supercooled water droplets would firstly adhere onto the upper edges of the as-prepared SHP surfaces leading to the formation of icicles. It was found that the long icicle, as marked by a red rectangle in Fig. 4 , which grew from the upper side was not in close contact with the SHP-FAS surface but was at a wedge angle indicating a real freezing area smaller than 44.3%. Fig. 3 Freezing morphology of the four as-prepared SHP surfaces. Fig. 4 Freezing areas of the four as-prepared SHP surfaces in glaze ice. 3.4 Anti-frosting performance The vapor condensation/frosting process inevitably occurs during glaze ice formation because of the high humidity and low temperatures in the artificial climate chamber. The anti-frosting performances of the as-prepared samples were investigated on a Peltier-based platform. The sample temperature was set at −5 °C. As shown in Fig. S8, † the glass surfaces modified with PDMS, G502, HDTMS, and FAS-17 as well as the as-prepared unmodified ZnO surface by RF magnetron sputtering showed poor anti-frosting properties with entire surfaces frozen in less than 10 min. Due to the hydrophilicity of the as-prepared unmodified ZnO surface, frost grew much more densely compared to that on the bare glass surfaces which had undergone modification. The frosting morphologies of the samples at 60 min varied obviously, which is probably a result of the different surface modifiers. However, the four as-prepared SHP surfaces demonstrated improved anti-frosting properties as shown in Fig. 5 . It was found that SHP-PDMS exhibited the poorest anti-frosting performance with a delayed-frosting time (time for the entire surface to be covered with frost) of less than 45 min. Obvious condensed water droplets were found on SHP-G502, SHP-FAS and SHP-HDTMS whose surface colors changed from dark grey to white in 45 min. A large area of the SHP-G502 and SHP-FAS surfaces remained free of frost except for a small amount of frost accumulating at the edges which is not apparent from a micro-point of view. An evident frost layer formed on the SHP-HDTMS surface within 45 min and slowly covered the entire surface within 240 min. Interestingly, although frost on SHP-HDTMS appeared earlier than on SHP-G502, the frost propagated faster on the SHP-G502 surface leading to a 91% freezing area in 90 min. Compared to the other three SHP surfaces, SHP-FAS exhibited the best anti-frosting behavior. About 82% of the entire surface remained free of frost for 90 min and the frosting time was effectively delayed for about 320 min. Fig. 5 The frosting process of ZnO surfaces modified by PDMS, G502, HDTMS and FAS-17. As the event of frosting may randomly occur on chilled solid surfaces, over 20 trials were conducted on the as-prepared SHP surfaces. Fig. 6 shows the frosting area of the four SHP surfaces as a function of freezing time. We can conclude that the SHP-PDMS surface displays the worst anti-frosting behavior with the whole surface freezing in less than 25 min. SHP-FAS demonstrates the best anti-frosting performance with an average frosting time delay of over 320 min, which is approximately 12.8 times that of SHP-PDMS. It is worth noting that the frost propagation on the four SHP surfaces slowly grew towards the unfrozen regions and no other clear defects were found during the whole experiment. Therefore, the difference in the experimental results can be ascribed to the effect of chemical composition ruling out the impact of structures. It can also be concluded that the anti-icing performances of the as-prepared SHP surfaces are closely related to their anti-frosting properties. The as-prepared SHP surfaces which demonstrate excellent anti-frosting properties are prone to having better anti-icing performances in glaze ice. Fig. 6 Frosting area of the as-prepared SHP ZnO surfaces modified with FAS-17, G502, HDTMS, and PDMS. To clarify frost propagation on the four SHP surfaces, the microprocess of frosting on the four SHP surfaces was observed. Pictures captured at 10, 20, 30 and 40 min are displayed in Fig. 7 . Obvious condensed microdroplets formed on the four SHP surfaces in 10 min. The condensed water droplets on the SHP-PDMS surface soon froze within 10 min, while the droplet size kept growing on the other three SHP surfaces. It is obvious that the growth pattern of the condensed water droplets on the other three SHP surfaces is quite different. The condensate droplets on the SHP-G502 surface grew larger mainly by merging with nearby condensed water and then remained almost motionless, resulting in comparatively homogeneous droplets. On the contrary, distribution of condensed water droplets on the SHP-HDTMS and SHP-FAS surfaces varied with time and the water volume was much smaller than that on the SHP-G502 surface, indicating a fairly dynamic property of condensed water droplets on SHP-HDTMS and SHP-FAS. Fig. 7 Microscopic growth process of condensed water droplets on ZnO surfaces modified by PDMS, G502, HDTMS, and FAS-17. The dynamic process of condensed water droplets is shown in Fig. 8 . With a prolonged freezing time, condensed microdroplets grew larger via vapor condensation and by the coalescence of nearby condensed microdroplets on the SHP-PDMS and SHP-G502 surfaces. There was no evident self-transfer phenomenon observed on these two SHP surfaces. Therefore, the condensed microdroplets stayed motionless. However, self-transfer phenomena were detected on the SHP-HDTMS and SHP-FAS surfaces. Nearby condensed microdroplets on the SHP-HDTMS surface merged together and subsequently departed from their original locations as identified by a color change of the coalesced microdroplets resulting from a distance variation between the jumping water droplets and the lense. Meanwhile, as the jumping water droplets rolled along the surface they would pick up tiny droplets, impacting on the surface and sweeping it clean and dry. A similar phenomenon was observed on the SHP-FAS surface. Coalesced microdroplets would jump off the surface or stay in another position on the SHP-FAS surface after several bounces. Fig. 8 Coalescence of condensed microdroplets on the PDMS, G502, HDTMS and FAS-17 modified SHP surfaces. To further investigate the self-transfer phenomenon on the SHP-FAS surface, high speed cameras with a frame rate of 1000 frames per second was used. As shown in Fig. 9 , a tiny droplet jumped off the surface and subsequently merged with a nearby droplet inducing a second bounce. As shown in the ESI video S1, † continually jumping water droplets jump off and subsequently fall back on the SHP-FAS surface. The coalescence of condensed microdroplets and the subsequent self-transfer movement on SHP-HDTMS and SHP-FAS markedly alter the diameter range and location of the condensed microdroplets. As a rational inference, the growth and distribution status of the condensed microdroplets on these two SHP surfaces exerts a large effect on the anti-frosting performance. Fig. 9 Side view of the self-transfer phenomenon on FAS-17-modified SHP surfaces. Due to the hydrophilic properties of the sample edges, frosting occurs first at these regions. Fig. S9 † displays frost propagation on the four SHP surfaces which is illustrated in Fig. 10 . The self-transfer movement on SHP-HDTMS and SHP-FAS demonstrates an obvious contrast to that on SHP-G502 and SHP-PDMS. The condensed water droplets freeze quickly on the SHP-PDMS surface indicating a faster heat transfer rate at low temperatures. The frozen water droplets behave like “defects” leading to a dramatically increased frost propagation rate. The “anchoring effect” on SHP-G502 and SHP-PDMS show that condensed water droplets grow in the “Wenzel state” on these two SHP surfaces leading to increased water volumes. According to Δ Q = Cm Δ T , it may take more time for the condensed water droplets (A and B) to freeze, which is why the frost on SHP-PDMS appears later than on SHP-HDTMS and SHP-FAS. However, the increased size of the condensed microdroplets minimizes the gap between the frozen droplets (C) and the nearby condensed microdroplets (A and B) making the frost crystals grow more easily across to the nearby condensed water droplets, which accelerates the frost propagation rate. Due to the self-transfer movement and the absorption of condensed water droplets by the frost front, a large “blank area” exists on SHP-HDTMS and SHP-FAS, which effectively inhibits the frosting propagation rate. This is the reason why the frost propagation rate on SHP-G502 is faster than that for SHP-HDTMS and SHP-FAS. According to the dimensionless form of Gibbs energy: 42 1 G * = F −2/3 ( θ )[2 − 2 cos θ − cos θ Y sin 2 θ ] where F ( θ ) = (2 − 3 cos θ + cos 3 θ ), θ is the intrinsic contact angle on a flat surface and θ Y represents the contact angle on a rough surface. cos θ Y for droplets in the Cassie state and the Wenzel state can be described as r cos θ and f s (1 + cos θ ) − 1, respectively. f s is the solid–liquid fraction. FAS-17 mainly contains –CF 3 groups with a surface free energy of 6.7 mJ mol −1 which is lower than that of –CH 3 in HDTMS or other surface modifiers, which can better improve the intrinsic contact angle. The θ values for the flat surface modified with G502, HDTMS, FAS-17, and PDMS were measured to be 94.2°, 101.9°, 109.7° and 93.3°, respectively. f s can be calculated to be 4.1% according to the CA measurement at ambient temperature where water droplets are supposed to be in the Cassie state. Given that the surfaces have comparable roughness, we converted the ZnO nanostructures into cylindrical structures referring to the average diameter of ZnO nanostructures (82.4 nm), equivalent film thickness (7.5 μm), and f s (4.1%) as shown in Fig. S10. † The surface roughness r can be roughly calculated as 16.39. By defining Δ G = G Cassie − G Wenzel , we can calculate the Δ G values for SHP-G502, SHP-HDTMS, SHP-FAS, and SHP-PDMS to be 0.3244, −0.9540, −1.8615, and 0.5057, respectively. If Δ G is negative, water vapor is more likely to condense in the Cassie state. The Δ G values for SHP-G502 and SHP-PDMS are both positive indicating that water vapor is prone to condense in the Wenzel state which is in accordance with the phenomenon observed in Fig. 7 . The Δ G values for SHP-HDTMS and SHP-FAS are negative indicating that the water vapor condenses in the Cassie state and thus demonstrates a greater mobility because of the self-transfer phenomenon. Fig. 10 Schematic of frost propagation on the SHP surfaces. The excellent anti-frosting performance of SHP-HDTMS and SHP-FAS-17 can be attributed to the self-transfer movement. On one hand, the jumping droplets minimize the number of water droplets distributed on the surface, which in turn serves to slow down frost propagation by requiring longer inter-droplet ice bridges and sometimes even resulting in failed bridges owing to sufficiently spaced droplets completely evaporating before the bridges can connect. On the other hand, condensed water droplets are easily absorbed by the frost front because of a low adhesion force between the condensed microdroplets and the chilled SHP surface (Fig. S9 † ), leaving a large gap between the frost front and the condensed microdroplets. Thus, frost propagation is effectively inhibited by delaying the ice-bridging process. 21,43 These results demonstrate that the anti-frosting performance of SHP surfaces with the same nano-scale textures is largely affected by the surface modifiers. The surface modifier affects the condensation mode and subsequently the frosting propagation rate. The self-transfer movement plays a vital role in enhancing the anti-frosting performance of nanostructured SHP surfaces. By improving the anti-frosting property with the selection of FAS-17, the SHP surface can maintain superhydrophobicity at low temperatures which enhances the anti-icing property in glaze ice."
} | 6,999 |
26836847 | null | s2 | 141 | {
"abstract": "With the increasing demand for rare earth elements (REEs) in many emerging clean energy technologies, there is an urgent need for the development of new approaches for efficient REE extraction and recovery. As a step toward this goal, we genetically engineered the aerobic bacterium Caulobacter crescentus for REE adsorption through high-density cell surface display of lanthanide binding tags (LBTs) on its S-layer. The LBT-displayed strains exhibited enhanced adsorption of REEs compared to cells lacking LBT, high specificity for REEs, and an adsorption preference for REEs with small atomic radii. Adsorbed Tb(3+) could be effectively recovered using citrate, consistent with thermodynamic speciation calculations that predicted strong complexation of Tb(3+) by citrate. No reduction in Tb(3+) adsorption capacity was observed following citrate elution, enabling consecutive adsorption/desorption cycles. The LBT-displayed strain was effective for extracting REEs from the acid leachate of core samples collected at a prospective rare earth mine. Our collective results demonstrate a rapid, efficient, and reversible process for REE adsorption with potential industrial application for REE enrichment and separation."
} | 305 |
24358265 | PMC3866191 | pmc | 143 | {
"abstract": "To date, few analyses of mutualistic networks have investigated successional or seasonal dynamics. Combining interaction data from multiple time points likely creates an inaccurate picture of the structure of networks (because these networks are aggregated across time), which may negatively influence their application in ecosystem assessments and conservation. Using a replicated bipartite mutualistic network of arbuscular mycorrhizal (AM) fungal-plant associations, detected using large sample numbers of plants and AM fungi identified through molecular techniques, we test whether the properties of the network are temporally dynamic either between different successional stages or within the growing season. These questions have never been directly tested in the AM fungal-plant mutualism or the vast majority of other mutualisms. We demonstrate the following results: First, our examination of two different successional stages (young and old forest) demonstrated that succession increases the proportion of specialists within the community and decreases the number of interactions. Second, AM fungal-plant mutualism structure changed throughout the growing season as the number of links between partners increased. Third, we observed shifts in associations between AM fungal and plant species throughout the growing season, potentially reflecting changes in biotic and abiotic conditions. Thus, this analysis opens up two entirely new areas of research: 1) identifying what influences changes in plant-AM fungal associations in these networks, and 2) what aspects of temporal variation and succession are of general importance in structuring bipartite networks and plant-AM fungal communities.",
"conclusion": "Conclusions This network analysis has demonstrated several novel results. First, network metrics were generally consistent between replicate plots. This means that across spatial scales the effects we report here are conserved within treatment (succession) and across time, and because of this conservation can probably be extended to other AM fungal-plant systems, although more experiments at larger spatial scales are necessary. Second, succession can strongly influence mutualist network structure, primarily through a decrease in the number of interactions and an increase of the proportion of specialists within the community with increasing successional stage. Third, the AM fungal-plant mutualism structure is dynamic throughout the growing season as the number of links between species increases. Further, our analysis revealed that AM fungal and plant species partnerships change throughout the growing season, potentially reflecting shifts in biotic and abiotic conditions.",
"introduction": "Introduction The analysis of bipartite mutualistic networks is a powerful tool for understanding the structure and dynamics of mutualistic interactions with multiple partners. Analysis of networks can determine the proportion of specialists and generalists within a focal community, the interactions between seemingly unrelated species, system complexity and functioning, as well as identify the organisms most strongly influencing these community properties [1] . As a result, bipartite network analysis of mutualisms is a growing area of interest in a wide variety of fields and study systems, though the focus has predominantly been on interactions between plants and animals [2] . Despite the power of network analyses to examine community structure, these analyses have rarely been applied to experimental or seasonal temporal data (but see [3] , [4] – [7] ). Until recently, network analyses often combined data on the same communities from multiple years [8] , [9] , and seasons [9] to build a single network that assumed no differences between years and seasons. Temporal dynamics such as succession and seasonal variation are fundamental ecological processes, and understanding the influence of these processes on networks will help reveal the basic properties that structure communities. As a result, the incorporation of successional and seasonal dynamics in network analyses is crucial. Few studies have focused on temporal dynamics in networks between months and seasons and those that have focused on non-bipartite aquatic [4] , [5] and bi-partite plant-pollinator systems [7] , [10] , [11] . These latter studies have revealed strong shifts in network structure, particularly connectance (the proportion of realized links between species in the network [12] ), throughout the year, and increases in the number of links. This further supports the assertion that lumping interaction data from multiple time points in a season likely biases the analysis of bipartite networks [11] concealing aspects of network structure which could have strong implications for the application of networks in ecosystem assessments and conservation. The small number of studies examining seasonal dynamics in terrestrial systems is likely due to the time required to gain sufficient observations to produce a highly resolved network model. However, the prevalence of new rapid screening molecular techniques (such as high sample throughput techniques (e.g. T-RFLP [13] ), and cloning and sequencing [14] , [15] , and next generation sequencing [4] , [15] , [16] ) is rapidly improving our ability to gather data in a timely and cost-effective fashion. These technologies allow researchers to sample at distinct time points with sufficient resolution for the creation of replicated temporal quantitative networks in systems (such as arbuscular mycorrhizal (AM) fungal-plant networks) that have rarely been studied in detail before. The number of studies examining how networks differ between successional stages is even smaller than the number examining seasonal dynamics. We know of only two studies examining successional dynamics: an ant-plant network that was re-examined at the same site ten years after the initial analysis [6] , and a study of plant-pollinator networks along a 130 year chronosequence of glacier development [3] . Both studies found a decrease in the proportion of specialists with time due to increases in the numbers of partners within the network [3] , [6] , but found opposite patterns for link density (mean number of interactions per species [1] ) and connectance [12] . In the ant-plant system both connectance and linkage density increased with time [6] , and the authors argued that these changes were due not to increases in network size but instead increases in the interactions between particular members of the original community that grew over time (due to their invasive nature). By contrast, in the plant-pollinator system linkage density remained constant with time but connectance tended to decrease [3] , and the authors argued that this was due to an overall increase in network size. Both of these studies examined differences between successional stages but combined one [3] or two years [6] of data to make comparisons thereby ignoring seasonal differences within their sites. Thus, no previous netowrk anlayses have incorporated both seasonal and successional dynamics, and the few existing analyses of successional networks may have been biased by the lumping of seasonal data [10] , [11] . Network analysis stands to provide a great wealth of information about belowground mutualistic organisms such as AM fungi where direct observations have been greatly limited [17] . Rapid screening molecular techniques are now being used to assess changes in temporal dynamics between successional stages in AM fungal-plant interactions [13] – [16] , [18] . The AM fungal-plant mutualism is arguably the most important free-living mutualism on the planet—these fungi appear to have facilitated plant colonization of land in the early Devonian [reviewed in 19] , associate with more than 80% of all plant species [17] , and contribute to plant diversity in natural systems (reviewed in [20] ). AM fungi act as a secondary root system that aids in nutrient (predominantly phosphorus, some nitrogen, and trace minerals) uptake and improved water availability for host plants, and in return the fungi gain carbon from their host plants [17] . The effect of biotic (e.g. soil pathogens [21] ) and abiotic (including nutrient availability, pH, and light availability [15] , [22] ) factors on plant growth is also mediated by interactions with AM fungi.Using molecular techniques, researchers have shown that plants associate with multiple AM fungal species simultaneously, although some combinations of plant and fungal partners occur more frequently than others [14] , [23] . Seasonal temporal variation occurs in the AM fungal-plant mutualism [15] , [24] , and anthropogenic disturbance can strongly influence the abundance, diversity, and community composition of AM fungi in a system [25] , [26] . AM fungal spore community composition has also been shown to shift between successional stages [27] , [28] which may contribute to variation in plant-AM fungal associations [13] , [14] , [16] . As a result, both seasonal and successional dynamics influence AM fungal and plant communities. Recent analyses of AM fungal-plant networks have revealed that, like most bipartite mutualistic networks, they are nested (defined by Bascompte & Jordano [2] as “a pattern of interaction in which specialists interact with species that form perfect subsets of the species with which generalists interact.”) [29] , [30] , and, potentially depending on the number of AM fungal genera present in a system, may be modular (consisting of sub-groups of organisms more likely to interact within the sub-group than with organisms outside the sub-group) [29] . However, as with most mutualistic bipartite networks, we know little about how seasonal and successional dynamics within the plant-AM fungal mutualism alter network structure, and, when considering all bipartite mutualistic networks, we do not know if seasonal and successional dynamics interact to influence network structure. As a result, we set out to answer the following questions: First, does successional stage alter network properties? We predict, as suggested above, that in the absence of invasive species and in the older successional stage there should be more species and therefore higher connectance and linkage density. We also predict, as seen in the ant-plant [6] and plant-pollinator system [3] that specialisation will decrease with time since disturbance because network size will increase with time since disturbance. Second, do temporal dynamics within a growing season alter network properties? We predict, based on the previous research described above, that links per species and connectance will increase throughout the growing season. This change will likely be due to an increase in the number of links through time because of an increase in the number of organisms in the system between spring and fall. These changes should then result in an increase in AM fungal generality [the effective number of plants per AM fungi (1)] and hence a reduction in plant vulnerability [the effective number of AM fungi per plant (1)] within the network. Finally, do successional stage and seasonal dynamics interact to influence network properties? This last question is important for determining whether successional analyses of bipartite network structure are likely to be biased if temporal dynamics are not included. Many of our predictions concerning the effects of successional and temporal dynamics on network properties are similar (e.g. connectance), so we predict these effects will likely be magnified when seasonal and successional changes coincide. We addressed these questions in a replicated well characterized AM fungal-plant system in the Estonian boreonemoral forest ecosystem [13] , [14] , [16] , [18] , [31] – [35] .",
"discussion": "Discussion Successional stage influenced network structure In contrast to our prediction specialization was greater in the older than the younger successional stage, and connectance was lower in the older successional stage while there was no effect on linkage density. Robustness was higher in the older successional stage. These results suggest that there were fewer interactions between plant and AM fungal species in the older successional stage, but those interactions were more likely to be specialist interactions. These changes have produced a community with a greater robustness to perturbation (i.e. species extinction) than the community in the younger successional stage. Very little previous research has examined the successional dynamics of the AM fungal-plant association, and most of that research has focused on the influence of successional dynamics on AM fungal root colonization (without identifying species) or AM fungal spore diversity. In most systems AM fungal spore diversity increases with time but then decreases as forest systems develop [27] , although the opposite pattern was observed in the Brazilian tropical dry forest [28] . However, spore diversity is likely not a good predictor of AM fungal diversity in root systems [46] . Prior to the research at the Koeru forest, Estonia, no one had ever examined the actual associations between AM fungi and plants (instead of AM fungal spore diversity or root colonization (without species identity)) throughout succession, so this (and previous research in the Koeru forest system) is the first to document the influence of succession on arbuscular mycorrhizae. Our results both agree and disagree with previous bipartite network analyses examining successional dynamics in phylogenetically distinct systems. None of our results corresponded with those found in a comparison of ant-plant networks built at the same site in Mexico in 1990 and 2000 [6] , but some of our results agreed with an analysis of a plant-pollinator network conducted along a 130 year transect [3] . Whereas no change in link density with successional stage occurred in our network and the plant-pollinator network, there was an increase in link density with time in the ant-plant network. Connectance decreased with time since disturbance in our system and tended to decrease with time in the plant-pollinator system; however, connectance increased with time in the ant-plant system. The authors of the ant-plant study argue that the changes in connectance that occurred in their system were due primarily to changes in the plant community [6] whereas there was little overall change in the plant species sampled between sites in our system which may explain the differences between these two studies. By contrast, where we observed an increase in the proportion of specialists, both the analyses of the ant-plant and plant-pollinator systems observed a relative decrease in the proportion of specialists with time since disturbance. As a result, more network analyses examining these and other properties (e.g. modularity) are required to aid a general understanding of the role of succession in structuring bipartite networks. It is difficult to determine what differences between the successional stages influenced the AM fungal-plant network structure in our system. Norway spruce, the dominant tree in our study system, is not an AM fungal host, so removal of spruce should not directly impact on AM fungi in this system. However, clear-cutting is a very strong disturbance which imposes major changes in understory environmental conditions and vegetation in this forest system [34] . Moreover, common silvicultural practice in northern Europe (plantation and maintenance of monocultures) produces even-aged coniferous stands [47] altering the structure [33] and reducing the diversity [35] of the herbaceous field layer of predominantly AM fungal plant hosts. These factors produced a significant difference in the structure of the understory plant community in young successional stands compared to unmanaged old growth forest [31] , which may in turn have limited the interactions between plant and AM fungal species. As a result, under these management conditions, the time required to restore linkages to their former state is clearly longer than 25 years. The considerably greater time since a disturbance event in the old growth plots also resulted in higher network specialisation. Network analyses of specialisation do not consider the biological capability of individual species within the network, instead they consider specialisation at the network level. That is, given an expected number of interactions between species within the network, how frequently does the actual number of interactions between species fall below the expected number? Those species with fewer interactions than the expected number increase the degree of specialisation within a network analysis context. Although the other successional bipartite analyses did not produce this result [3] , [6] , disturbance has been shown in other systems to have a greater negative effect on functional specialists than generalists [48] . Our analysis cannot determine whether the differences in the specialisation within our system are due to the following non-exclusive hypotheses: 1) lack of preferred partners in the early successional stage, 2) loss of rare AM fungal phylotypes with disturbance in the early successional stage, or 3) selection for specialization by plants or fungi in the later successional stage. First, in the young successional stage AM fungi and plants may not be able to associate with preferred partners, and may therefore be forced to associate with a wider group of partners to avoid extinction. Post-disturbance associations may not reflect host preference but rather the local availability of fungal taxa (as AM fungi are dispersal limited and may not easily re-colonize disturbed habitats). If this were to be true, AM fungi that appear as specialists in the older successional stage would appear as generalists in the younger successional stage. Although we cannot definitively test this, it is somewhat supported by two lines of evidence: First, we observed a trend of increased associations with host plants among some AM fungal virtual taxa in the young successional stage (exhibited by Glomus VT00113, VT00115, VT00143, VT00160, VT00166), and, in two cases (AM fungal taxa VT00143 and VT00160), specialist taxa switched to a generalist strategy. Second, the greater number of unique interactions found in the older successional stage also supports the notion that AM fungal specialists in the older successional stage are generalists in the younger stage. The second hypothesis suggests that rare AM fungal taxa may be lost following disturbance in the early successional stage. Rare AM fungal phylotypes have a greater likelihood of local extinction due to chance in a disturbance event, suggesting that specialists in the old growth forest may be absent in the younger successional stage. The loss of specialist AM fungal host plants could also lead to the loss of specialist AM fungi. Supporting this, an earlier study in the Koeru forest area revealed a distinct set of AM fungi associated only with forest habitat specialist plants—plants which are less likely to be present in the young successional stage [16] . In contrast, our system has previously been shown to be nested [29] , [30] , and therefore we would expect specialist plants to be associating predomoninantly with generalist AM fungi. Thus the loss of a specialist plant species would then not lead to the loss of a specialist AM fungal species. Again, we could not test this second hypothesis directly. Indirectly, a count of the number of extreme specialists (those fungi with fewer than 20 interactions across all sampling times) between the two successional stages did not differ but the metrics of robustness suggested there was a greater extinction probability in the young successional stage. Finally, the old growth forests at this site have been undisturbed for 130–140 years [18] , [31] which may be long enough to allow for selection of specialization among plants and fungi at these sites [49] , [50] . Recent research conducted at the older successional stage site demonstrated that very few specialist AM fungal species colonized a novel plant host's roots [51] . This research either supports the notion that specialist associations may take time to develop, or that early successional plants (such as invasives) may also favor generalist AM fungi (if doing so allows them to quickly obtain partners in a new environment) [52] . If selection for host specialization does not result in speciation and involves the same species present in the younger successional stage, it would be difficult to separate this hypothesis from the first hypothesis presented above. If the evolution of specialization involved speciation (perhaps a less likely scenario) then it would be difficult to distinguish between this hypothesis and the second hypothesis presented above. As result, there are several possible (non-mutually exclusive) explanations for the differences in the proportion of specialists between the two sites. Seasonal dynamics influenced network structure AM fungal-plant network structure was also significantly altered by seasonal dynamics. As we predicted, the number of links between partners increased between the June and October samples. A similar pattern was observed in an arctic plant-pollinator network [7] . An increase in the number of associations between AM fungi and plants throughout a growing season has been shown in controlled greenhouse conditions (reviewed in [17] ), however it has never before been demonstrated in field conditions. Previous research suggests that AM fungi differ in their colonization strategies [53] and may vary associations with host plants based on nutrient availability [54] which would contribute to variation in link density throughout the plant growing season if fewer links were maintained during the winter. This is also supported, as we predicted, by the increase in generality seen throughout the growing season where the relative number of plant species associated with a single AM fungal species increased from the first to the third sampling time. In addition, different AM fungal species are adapted to different temperature regimes [55] , so fewer AM fungal species may be able to colonize roots during colder months but these same cold weather colonizers may also be able to maintain colonization at higher temperatures [56] . As a result, season-scale temporal variation is likely an important determinant for the structure of mutualistic networks [57] . Here we provide strong evidence of switching between partners by AM fungi and plants in a natural system. As already shown within this system, the same AM fungal species were observed across the growing season [58] . However, specific partnerships between plants and AM fungi were not consistent throughout the growing season, as supported by the high rate of turnover for links (on par with turnover rates between years [44] ). In particular, the turnover rate for AM fungi was significantly greater than for plants—a somewhat surprising result given that AM fungi might be expected to be more stable partners due to their limited dispersal capabilities. Liu et al. \n [24] also showed variation in the presence and absence of different AM fungal phylotypes at different time points throughout the growing season. Switching among partners by AM fungi and plants suggests that AM fungal species and/or plant species have niches within the mutualism. That is, different partners may be better adapted or suited for different growth stages, soil temperatures [55] , day lengths, or abiotic [17] or biotic stresses [21] , [59] . When the abiotic or biotic environment changes the niche may disappear and plants and/or fungi may take on or adapt to a niche that requires different partners. Switching between partners has consequences for our understanding of AM fungal-plant dynamics. The majority of studies focused on AM fungal-plant dynamics have been conducted under controlled conditions [17] , assumed constant AM fungal communities in the roots throughout the experiment, and ignored variation in niches between the greenhouse and the field. As a result, we encourage new research identifying what aspects of temporal dynamics influence switching of partners between AM fungi and plants. In contrast to our prediction there was no interaction between the effects of seasonal and successional dynamics on network properties. In particular, connectance was not magnified by both seasonal and successional dynamics. Conclusions This network analysis has demonstrated several novel results. First, network metrics were generally consistent between replicate plots. This means that across spatial scales the effects we report here are conserved within treatment (succession) and across time, and because of this conservation can probably be extended to other AM fungal-plant systems, although more experiments at larger spatial scales are necessary. Second, succession can strongly influence mutualist network structure, primarily through a decrease in the number of interactions and an increase of the proportion of specialists within the community with increasing successional stage. Third, the AM fungal-plant mutualism structure is dynamic throughout the growing season as the number of links between species increases. Further, our analysis revealed that AM fungal and plant species partnerships change throughout the growing season, potentially reflecting shifts in biotic and abiotic conditions."
} | 6,423 |
33971881 | PMC8112011 | pmc | 145 | {
"abstract": "Background The demand for biobased polymers is increasing steadily worldwide. Microbial hosts for production of their monomeric precursors such as glutarate are developed. To meet the market demand, production hosts have to be improved constantly with respect to product titers and yields, but also shortening bioprocess duration is important. Results In this study, adaptive laboratory evolution was used to improve a C. glutamicum strain engineered for production of the C 5 -dicarboxylic acid glutarate by flux enforcement. Deletion of the l -glutamic acid dehydrogenase gene gdh coupled growth to glutarate production since two transaminases in the glutarate pathway are crucial for nitrogen assimilation. The hypothesis that strains selected for faster glutarate-coupled growth by adaptive laboratory evolution show improved glutarate production was tested. A serial dilution growth experiment allowed isolating faster growing mutants with growth rates increasing from 0.10 h −1 by the parental strain to 0.17 h −1 by the fastest mutant. Indeed, the fastest growing mutant produced glutarate with a twofold higher volumetric productivity of 0.18 g L −1 h −1 than the parental strain. Genome sequencing of the evolved strain revealed candidate mutations for improved production. Reverse genetic engineering revealed that an amino acid exchange in the large subunit of l -glutamic acid-2-oxoglutarate aminotransferase was causal for accelerated glutarate production and its beneficial effect was dependent on flux enforcement due to deletion of gdh . Performance of the evolved mutant was stable at the 2 L bioreactor-scale operated in batch and fed-batch mode in a mineral salts medium and reached a titer of 22.7 g L −1 , a yield of 0.23 g g −1 and a volumetric productivity of 0.35 g L −1 h −1 . Reactive extraction of glutarate directly from the fermentation broth was optimized leading to yields of 58% and 99% in the reactive extraction and reactive re-extraction step, respectively. The fermentation medium was adapted according to the downstream processing results. Conclusion Flux enforcement to couple growth to operation of a product biosynthesis pathway provides a basis to select strains growing and producing faster by adaptive laboratory evolution. After identifying candidate mutations by genome sequencing causal mutations can be identified by reverse genetics. As exemplified here for glutarate production by C. glutamicum , this approach allowed deducing rational metabolic engineering strategies. Supplementary Information The online version contains supplementary material available at 10.1186/s12934-021-01586-3.",
"introduction": "Introduction Plastics still are synthesized primarily from natural gas and petroleum and only a fraction of 1% is bio-based. The demand for environmentally friendly alternatives is steadily increasing and the annual market volume of bioplastics is predicted to increase to 18% until 2025 [ 1 ]. Biopolyamides are gaining more interest for use in the textile and construction industries. Polyamides can be obtained either by condensation of dicarboxylic acids with diamines or by anionic ring-opening polymerization of lactams, the cyclization products of ω-amino acids [ 2 ]. Bio-based production of monomeric building blocks for polyamides has been established in metabolically engineered C. glutamicum and E. coli [ 3 , 4 ]. Fermentative production of the C4-ω-amino acid γ-aminobutyrate (GABA) [ 5 , 6 ] and the C5-ω-amino acid 5-aminovalerate (5AVA) has been established [ 7 , 8 ] and, e.g., ring-opening polymerization of 5AVA can be used to produce the polyamide 5 (PA 5) [ 9 , 10 ]. Moreover, diamines like putrescine [ 11 , 12 ] and cadaverine [ 13 , 14 ] as well as the dicarboxylic acids succinate and glutarate [ 15 – 17 ] were successfully produced in high titers. Glutarate, e.g., is used as a building block for polyamides such as PA 4.5 [ 18 ], PA 6.5, PA 12.5 [ 19 ] or PA 5.5 the latter of which is synthesized by polycondensation of the C5-dicarboxylic acid glutarate with C5-diamine cadaverine [ 20 ]. Notably, the C5 polyamide building blocks cadaverine, 5AVA and glutarate can be synthesized from a common precursor, the amino acid l -lysine. Industrial l -lysine production by fermentation with Corynebacterium glutamicum is operated at large scale with an annual production volume of about 2.6 million metric tonnes in 2018 [ 4 ]. Glutarate can be derived from l -lysine by four different pathways. All four pathways converge to 5-aminovalerate (5AVA), which then is converted to glutarate in two enzymatic steps catalyzed by GABA/5AVA aminotransferase (GabT) and succinate/glutarate semialdehyde dehydrogenase (GabD). The first pathway from l -lysine to 5AVA employs l -lysine-α-oxidase (RaiP) from Scomber japonicus that catalyzes oxidative deamination of l -lysine using molecular oxygen followed by spontaneous decarboxylation [ 21 ]. The second pathway to 5AVA combines oxidative decarboxylation by l -lysine monooxygenase (DavA) using molecular oxygen followed by desamidation by γ-aminovaleramidase (DavB) from Pseudomonas putida [ 20 ]. The third pathway is based on l -lysine decarboxylase from E. coli , putrescine oxidase PuO from Rhodococcus qingshengii , which requires molecular oxygen, and γ-aminobutyraldehyde dehydrogenase PatD from E. coli [ 8 ]. The fourth pathway does not require molecular oxygen as it cascades l -lysine decarboxylase, 2-oxoglurate-dependent putrescine/cadaverine transaminase PatA, and NAD-dependent γ-aminobutyraldehyde dehydrogenase PatD from E. coli [ 7 ]. The pathway combinations LdcC-PuO-PatD-GabT-GabD and LdcC-PatA-PatD-GabT-GabD couple conversion of l -lysine to glutarate either to one (GabT) or two (PatA, GabT) transaminase reactions, respectively, which generate l -glutamic acid from 2-oxoglutarate. Deletion of gdh , the gene for the major ammonium assimilating enzyme l -glutamic acid dehydrogenase [ 8 , 16 ], enabled flux enforcement (Fig. 1 , left panel), i.e., the metabolic setup in which growth requires production of glutarate. In general, GDH is active under nitrogen surplus conditions and has a low affinity towards its substrates ammonia and 2-oxoglutarate [ 22 ]. By contrast, the enzyme pair glutamine synthetase (GS) and l -glutamic acid-2-oxoglutarate aminotransferase (GOGAT, also known as l -glutamic acid synthase) synthesizes l -glutamic acid in an ATP dependent manner during ammonium starvation at ammonium concentrations below 5 mM [ 23 ]. The GS/GOGAT system is encoded by glnA for GS and gltBD for the large and small subunits of GOGAT (Fig. 1 , right panel). The net reaction of the combined activities of GS and GOGAT results in ATP and NADPH dependent conversion of 2-oxoglutarate to l -glutamic acid, while GDH only requires NADPH for reductive amination of 2-oxoglutarate to l -glutamic acid. It is known that GS/GOGAT can compensate for the lack of GDH [ 24 ] also at higher nitrogen concentrations (up to 40 mM) [ 25 ]. Fig. 1 Schematic representation of the metabolic pathway for glutarate production, flux enforcement by deletion of gdh (left panel) and ammonium assimilation by the GS/GOGAT system (right panel) . Gene names are shown next to enzyme reactions (arrows), gene deletions are indicated by red crosses. Enzymes from P. stutzeri (dark grey), E. coli (light grey) and native enzymes (orange) are highlighted. gabT , GABA/5AVA amino transferase; gabD , succinate/glutarate-semialdehyde dehydrogenase; ldcC , l -lysine decarboxylase; patA , putrescine transaminase; patD , γ-aminobutyraldehyde dehydrogenase; glnA , glutamine synthetase (GS); gltBD , l -glutamic acid-2-oxoglutarate aminotransferase (GOGAT); gdh , l -glutamic acid dehydrogenase Systems metabolic engineering proved successful to achieve high titer glutarate production by metabolically engineered C. glutamicum [ 16 , 17 , 26 ]. In this study, we aimed to accelerate glutarate production by evolutionary engineering. Adaptive Laboratory Evolution (ALE) allows to leverage natural selection to optimize a target property of a production strain without the requirement of a priori knowledge of the genetic background [ 27 , 28 ]. This approach is straightforward if a growth advantage can be selected for. This was easily implemented, e.g., when higher tolerance against a compound is sought or to improve substrate utilization and to optimize growth rates [ 29 , 30 ]. Moreover, it can also be used to identify non‐intuitive targets for strain engineering, and ultimately to gain a comprehensive understanding of biological pathway regulation [ 31 ]. In C. glutamicum , ALE allowed to accelerate growth of the wild-type [ 32 , 33 ], to increase tolerance towards higher temperatures [ 34 ] and methanol [ 35 , 36 ], to improve consumption of xylose and cellobiose [ 37 , 38 ], and to increase production of putrescine and ornithine [ 39 , 40 ]. We have chosen to apply ALE in order to accelerate glutarate production via the LdcC-PatA-PatD-GabT-GabD pathway since two of the involved transaminase reactions (PatA, GabT) provide l -glutamic acid from 2-oxoglutarate and, thus, compensate for the lack of GDH due to the deletion of its gene [ 8 , 16 ]. The resulting flux enforcement provides a selectable trait by linking metabolic productivity to growth. In this metabolic setup, the rate of growth (requiring l -glutamic acid) is coupled to the rate of glutarate production (providing l -glutamic acid) and selection of faster growing mutants yielded strains with increased volumetric productivity. Mutations identified by genome sequencing could be rationalized by reverse genetics. Moreover, in order to complement strain development, purification of glutarate from the fermentation broth using a combination of reactive extraction and reactive re-extraction was considered. For this purpose, we adapted an approach, which was previously used for the purification of itaconic acid and is based on an aqueous organic extraction system and tertiary amines as (reactive) extractants [ 41 ], to serve for the recovery of glutarate by identification of optimal (reactive) extractants and organic phases in dedicated screening experiments. During the reactive extraction, the amine extractant interacts with glutarate building a hydrophobic complex, which is then extracted to the organic phase (separation from impurities). The results reveal that the concept allows for an efficient separation of glutarate from a crude fermentation broth showing high yields and selectivities, opening the window for industrial production.",
"discussion": "Discussion In this study, flux enforcement coupling growth to l -glutamic acid production by C. glutamicum provided the basis to select strains growing and producing faster by adaptive laboratory evolution. Among the candidate mutations determined by genome sequencing two causal mutations were identified by reverse genetics. This approach almost doubled volumetric productivity and in fed-batch bioreactor cultures a titer of 22.7 g L −1 , a yield of 0.23 g g −1 and a volumetric productivity of 0.35 g L −1 h −1 were achieved. Purification of glutarate directly from the fermentation broth leading to yields of 58% and 99% in the reactive extraction and reactive re-extraction step, respectively, was established. This report is not the first on metabolic engineering of C. glutamicum for glutarate production [ 8 , 16 , 17 , 26 ], but the first example of improving volumetric productivity by flux enforcement and ALE. Notably, engineering of the GS/GOGAT system, which was identified here as crucial to accelerate glutarate production, has not been reported previously as metabolic engineering target for glutarate production by C. glutamicum . Strain engineering, e.g., by overexpressing ynfM encoding the recently discovered glutarate exporter, media optimization, e.g., by using mixtures of glucose and sucrose, and process intensification, e.g., by using a pH–stat feeding strategy in fed-batch cultures, have been described to boost glutarate production to titers of more than 100 g L −1 [ 17 ]. Thus, the mutation pair identified here (GltB E686Q and deletion of gdh ) complements the previously described metabolic engineering strategies for glutarate production by C. glutamicum . It has to be noted that yggB RNA levels were reduced in both ALE strains and this gene codes for a transport system involved in export of l -glutamic acid out of the C. glutamicum cell [ 56 ], although l -glutamic acid export is not abolished in its absence [ 57 ]. Since the flux enforcement strategy used here relied on gdh deletion, thus, the major enzyme for synthesis of l -glutamic acid is absent, the reduced yggB RNA levels may help to avoid loss of l -glutamic acid from the C. glutamicum cell by export. Possibly, deletion of yggB may increase stringency of the flux enforcement by gdh deletion. Transport engineering has proven important for improving C. glutamicum processes [ 58 ], not only regarding substrate uptake [ 59 ] or product export (e.g., ynfM , [ 17 ]), but also to avoid loss of intermediates (used here to avoid export of the intermediate cadaverine). The strains analysed here lacked gdh . The growth rate of a gdh deletion mutant is reduced due to a partially triggered nitrogen starvation response as evidenced by, e.g., partial adenylylation of GlnK, such that ammonium is assimilated by GS/GOGAT. However, since the intracellular l -glutamic acid pool is not completely restored and the GS/GOGAT has an increased energy demand (1 ATP per NH 3 fixed), growth in the absence of gdh is slower than in its presence. The SNP in gltB resulted in amino acid exchange E686Q in the large subunit of GOGAT (GltB E686Q ). The large subunit serves two functions allocated to two domains: hydrolysis of l -glutamine to NH 3 and l -glutamic acid on the one hand and combining the produced NH 3 with 2-oxoglutarate to produce a second molecule of l -glutamic acid on the other hand. The small subunit transfers electrons from the co-substrate NADPH. The SNP may interfere with binding of glutamine according to our inspection of the structure by Phyre2, CUPSAT and COACH-D [ 42 , 53 , 60 ]. Currently, it is unknown how the amino acid exchange E686Q in the large subunit of GOGAT supports faster glutarate production by the strains analysed here. Since it is known that at high NH 3 concentrations (up to 40 mM) GS/GOGAT compensates for the lack of GDH [ 25 ], it is conceivable that the GOGAT mutant GltB E686Q is active at higher nitrogen concentrations. The media used here contains 468 mM NH 3 , which is particularly relevant in the early growth and production phase. RNAseq analysis revealed higher expression of genes belonging to the AmtR regulon of nitrogen starvation in GluA T5 and the parental strain GluA T0 than in GluA T7, which carries GltB E686Q (Additional file 2 : Tables S1 and S2). Binding to operator DNA by homodimeric AmtR is not released by a small-molecule effector, but by GlnK when adenylylated at tyrosine residue 51 [ 61 ], likely in a 6:6 stoichiometric (AmtR 2 ) 3 –(GlnK 3 ) 2 complex [ 62 ]. The adenylation status of PII-type signal transduction protein GlnK is controlled by adenyltransferase GlnD under nitrogen starvation, which is most likely perceived as ammonium limitation [ 63 ]. Thus, ammonium starvation is less pronounced or less perceived in the GltB E686Q carrying mutant GluA T7 than in the native GltB carrying strains GluA T0 and GluA T5. Besides possible effects due to transcriptional and post-translational regulation, the higher K m of GltB E686Q for glutamine as compared to native GltB may provide a clue why GltB E686Q performed better regarding glutarate production. GOGAT competes with the transaminases GabT and PatA for the substrate 2-oxoglutarate, and according to the BRENDA database the K m for 2-oxoglutarate is lower for GOGAT from C. glutamicum subsp. flavum (0.06 mM) than for GabT from P. aeruginosa (0.75 mM) and PatA from E. coli (19 mM). Thus, these kinetic parameters suggest that GOGAT may outcompete the transaminases used in the synthetic glutarate pathway. The variant GltB E686Q selected by ALE showed a two-fold higher K m for glutamine, the other substrate of GOGAT. Thus, this mutation likely favors a higher ratio of 2-oxoglutarate conversion via the transaminases, which is in line with the observed increase in glutarate productivity. However, it has to be noted that while nitrogen-replete E. coli cells show relatively low intracellular glutamine (0.2 to 0.5 mM) and 2-oxoglutarate (0.1 to 0.9 mM) concentrations [ 64 ], C. glutamicum has been described to accumulate about 10 mM glutamine under nitrogen-replete conditions [ 65 ]. Future research will have to unravel the mechanism and to distinguish whether nitrogen metabolism and/or regulation is altered due to GltB E686Q . Two changes affected expression and activities of the heterologous 5AVA amino transferase GabT and glutarate-semialdehyde dehydrogenase GabD: the amino acid exchange P134L in GabD led to higher combined GabD-GabT in vitro activity in crude extracts and a higher PCN of plasmid pEC-XT99A- tetA(Z) Δ21bp - gabTD P134L that differed from pEC-XT99A- gabTD P134L by a 21 bp off-frame deletion in the antibiotic resistance marker tetA(Z) . The first change, GabD P134L , accelerated growth and increased the glutarate titer by about two fold to 0.12 ± 0.00 h −1 and 45 ± 2 mM, respectively. The second change, pEC-XT99A- tetA(Z) Δ21bp - gabTD P134L , improved the growth rate to 0.14 ± 0.00 h −1 (compare isogenic strains GluA RG2 and GluA RG3 in Fig. 4 ). Although the glutarate titer was not increased, the volumetric productivity was increased (Fig. 4 ). 5AVA remained a by-product of accelerated, flux enforced glutarate production. In the fermentations performed in batch and fed-batch mode, it was demonstrated that as long as residual glucose is present 5AVA accumulated, but once it is negligible, only glutarate accumulated. C. glutamicum possesses the native operon gabTDP on its chromosome and its expression is reduced in the presence of glucose, gluconate, and myo -inositol, presumably via the cAMP-dependent global regulator GlxR, for which a binding site is present downstream of the gabT transcriptional start site [ 66 ]. Thus, while plasmid-borne expression of ldcC , patD and patA was sufficient for conversion of lysine to 5AVA, plasmid-borne expression of gabTD was limiting for conversion of 5AVA to glutarate and glutarate production benefitted from expression of the native gabTDP from the chromosome. Since flux enforcement by gdh deletion is more efficient when coupled to one transamination reaction rather than to two [ 8 ], either pathways with just one transamination reaction shall be used or flux enforcement has to be accentuate further, e.g., by combined deletion or attenuation of gdh and gltBD . With the aim to ensure appropriate industrially applicable purification, reactive extraction followed by a reactive re-extraction step proved to be a successful recovery strategy for glutarate produced by fermentation. Reactive extraction systems containing any amount of T-C6 or high amounts of T-C8 (molar ratio of T-C8/glutarate = 6/1 or higher) led to the formation of a third solid phase. Increasing the polarity of the organic phase by addition of 1-dodecanol as (polar) modifier successfully prevented the third-phase formation. An optimal reactive extraction system was identified containing T-C6 as amine extractant at a molar ratio of 9/1 including 10 wt% of 1-dodecanol as polar modifier in the organic phase. Within these experiments, a yield of 58.1% was achieved. Only small amounts of l -glutamic acid and 5-aminovalerate were co-extracted within this step. Reactive re-extraction using both WSA’s led to maximum re-extraction yield of 99%. One reason that a full recovery of glutarate from fermentation broth could not be achieved in the initial reactive extraction step is the presence of strong electrolytes like chloride ions in the aqueous phase, which significantly lowers the reactive extraction performance [ 41 , 67 ]. Other acids as e.g. hydrochloric acid can form complexes with the amine extractant as well, hence competing with the carboxylic acid for the amine extractant and lowering reactive extraction yield [ 67 , 68 ]. Previous studies showed that the effect of sulfate ions on reactive extraction yield of carboxylic acids is less pronounced [ 41 , 67 ]. Therefore, choosing sulfuric acid in this study to adjust the pH of the aqueous phase before reactive extraction was beneficial for the efficiency of the process. Furthermore, it was shown in experiments investigating reactive extraction of carboxylic acids from fermentation broth, that exchanging the ammonium source NH 4 Cl for ammonium sulfate ((NH 4 ) 2 SO 4 ) did increase the extraction yield significantly [ 41 ]. Therefore, this approach could lead to higher extraction yields for glutarate as well and should be considered in future studies as omitting chloride in CGXII did not affect glutarate production in the evolved strain (Additional file 1 : Figure S3). In conclusion, recovery of glutarate from fermentation broth applying the reactive extraction/reactive re-extraction concept for the purification of carboxylic acids was successful, hence adding to the feasibility of the industrial applicability of glutarate production by C. glutamicum ."
} | 5,401 |
40091861 | PMC11974494 | pmc | 146 | {
"abstract": "ABSTRACT Coral reefs and their photosynthetic algae form one of the most ecologically and economically impactful symbioses in the animal kingdom. The stability of this nutritional mutualism and this ecosystem is, however, at risk due to increasing sea surface temperatures that cause corals to expel their symbionts. Symbioses with these microeukaryotes have independently evolved multiple times, and non‐coral cnidarians (e.g., sea anemones) serve as a valuable and insightful comparative system due to their ease of husbandry in the laboratory and their ability to shuffle different strains of their photosymbionts to acclimate to thermal conditions. This breadth of symbiont shuffling is exemplified by the sea anemone \n Anthopleura elegantissima \n , which naturally occurs in symbiosis with the dinoflagellate Breviolum muscatinei (formerly Symbiodinium ) or the chlorophyte Elliptochloris marina as well as being aposymbiotic. Here, we assembled a draft genome and used multi‐omics to characterise multiple physiological levels of each phenotype. We find that \n A. elegantissima \n has symbiont‐specific transcriptional and metabolomic signatures, but a similar bacterial community dominated by a single Sphingomonas species that is commonly found in the cnidarian microbiome. Symbiosis with either eukaryotic symbiont resulted in differential gene expression and metabolic abundance for diverse processes spanning metabolism and immunity to reproduction and development, with some of these processes being unique to either symbiont. The ability to culture \n A. elegantissima \n with its phylogenetically divergent photosymbionts and perform experimental manipulations makes \n A. elegantissima \n another tractable sea anemone system to decode the symbiotic conversations of coral reef ecosystems and aid in wider conservation efforts.",
"introduction": "1 Introduction Animals across the tree of life have established microbial symbioses that are of deep evolutionary origin (Bordenstein and Theis 2015 ; McFall‐Ngai et al. 2013 ; Zilber‐Rosenberg and Rosenberg 2008 ). One of the primary functions of and incentives to cooperate in these partnerships is establishing and maintaining a metabolic mutualism (Kiers et al. 2011 ; Leigh 2010 ). Animals are often limited in their biosynthetic capabilities and, as a result, rely on symbionts to supplement their diet with various nutrients. Aphids, for example, feed on plant sap that is low in essential amino acids and require the endosymbiont Buchnera spp. to offset this deficiency (Douglas 1998 ), while diverse marine invertebrates associate with bacteria that can convert reduced substrates into host biomass (Dubilier et al. 2008 ). The nutritional symbiosis with arguably the most widespread ecological and economic impact occurs between corals and their photosynthetic algae (Hoegh‐Guldberg et al. 2007 ; Hughes et al. 2003 ). Coral reefs are biodiversity hotspots that are metabolically sustained by their symbiotic relationship with dinoflagellates of the family Symbiodiniaceae (Berkelmans and van Oppen 2006 ; LaJeunesse et al. 2018 ; Stat et al. 2008 ). These protists reside in vesicles within the coral's endodermal cells and typically supply the host with ≥ 90% of its total energy, while the host provides them with a sheltered environment and inorganic nutrients (Davy et al. 2012 ). Symbiodiniaceae are phylogenetically and functionally diverse, and some corals shuffle the composition of Symbiodiniaceae strains with which they associate in response to ambient and anthropogenic changes in the abiotic environment (Cunning et al. 2015 ). The ability to host and shuffle Symbiodiniaceae extends to other members of the class Anthozoa. The sea anemone Exaiptasia diaphana , for example, is a well‐established experimental system for understanding cnidarian symbioses. Many of the cellular and molecular insights of cnidarian symbioses would not have been possible without \n E. diaphana \n (Davy et al. 2012 ; Weis et al. 2008 ). The sea anemone \n Anthopleura elegantissima \n is another long‐standing system that exemplifies the wide range of symbioses formed by cnidarians, which was first used to determine that symbionts transfer carbon to their cnidarian host (Muscatine and Hand 1958 ). This temperate, intertidal sea anemone naturally occurs in symbiosis with the dinoflagellate Breviolum muscatinei (formerly referred to as Symbiodinium ), the chlorophyte Elliptochloris marina (which has only been reported in species of Anthopleura ), or without symbionts (Figure 1 ; Lajeunesse and Trench 2000 ; Leutsch 2011 ; Muller‐Parker and Davy 2001 ). The symbionts that \n A. elegantissima \n associates with are primarily determined by latitude, with \n E. marina \n found along the upper latitudes and B. muscatinei observed in the lower latitudes of the geographic range for \n A. elegantissima \n (Secord and Augustine 2005 ). This biogeographic gradient coincides with temperature and irradiance across the Pacific Coast of North America, with \n E. marina \n preferring cooler, shaded habitats (Secord and Muller‐Parker 2005 ). This continental biogeographic signature can also be found along the vertical gradients of the intertidal zone; \n A. elegantissima \n is dominated by B. muscatinei high in the intertidal, \n E. marina \n low in the intertidal, and aposymbiotic individuals occur in dark crevices (Bergschneider and Muller‐Parker 2008 ; Dimond et al. 2011 ; Verde and McCloskey 2001 , 2002 ). FIGURE 1 Symbiotic phenotypes of \n Anthopleura elegantissima \n . The sea anemone \n A. elegantissima \n occurs naturally in symbiosis with (A) the dinoflagellate Breviolum muscatinei (brown), (B) the chlorophyte Elliptochloris marina (green), or (C) without symbionts (white). These two symbionts exhibit strikingly different physiologies that directly contribute to the life history strategies and fitness of \n A. elegantissima \n . The chlorophyte \n E. marina \n lives at a 4 × higher density and grows 8 × faster than the dinoflagellate B. muscatinei , while each B. muscatinei cell is about 2.5 × more productive and can translocate 5 × more carbon to the host than an \n E. marina \n cell (Dimond et al. 2013 ; Verde and McCloskey 1996 ). Moreover, the metabolic products received by the sea anemone from these two symbionts are notably different: \n E. marina \n primarily translocates amino acids, while B. muscatinei translocates glycerol and sugars (Minnick 1984 ; Trench 1971a , 1971b ); thus, these two symbionts serve different nutritional roles. As a result, these symbionts of \n A. elegantissima \n influence the host's balance between growth, asexual cloning, and sexual reproduction (Bergschneider and Muller‐Parker 2008 ; Bingham et al. 2014 ). Sea anemones hosting \n E. marina \n tend to reproduce sexually, while symbiosis with the more productive B. muscatinei promotes cloning by fission, a strategy that makes \n A. elegantissima \n a highly successful and spatially dominant member of the intertidal (Bingham et al. 2014 ). Despite the long history of using \n A. elegantissima \n as a cnidarian system for symbiosis, we lack a modern molecular, metabolic, and microbial understanding of the partnership between \n A. elegantissima \n and its two phylogenetically distinct photosynthetic microeukaryotes. Therefore, the goal of this study was to characterise these symbiotic phenotypes under stable conditions in a laboratory‐based mesocosm. We did this by generating a reference genome and then comparing sea anemones of each symbiotic phenotype using transcriptomics, metabolomics, and amplicon‐based sequencing of the bacterial community to characterise multiple physiological levels. We find that the sea anemone \n A. elegantissima \n exhibits distinct transcriptional and metabolic profiles in each symbiotic phenotype, while the bacterial community remains unchanged. Furthermore, we observe that the bacterial community of \n A. elegantissima \n is dominated by a single Sphingomonas species, which is commonly found in the cnidarian microbiome.",
"discussion": "4 Discussion Animals often rely on microbial symbionts for diverse nutrients and metabolic products that supplement their limited biosynthetic capability (McFall‐Ngai et al. 2013 ; Zilber‐Rosenberg and Rosenberg 2008 ). Nutritional symbionts of animals, for example, provide essential amino acids or reduce sulphur compounds that serve as electron donors to fix carbon dioxide autotrophically (Douglas 1998 ; Dubilier et al. 2008 ). Corals engage in a similar metabolic associations with the photosynthetic unicellular eukaryotes that provide them with the majority of their energy (Davy et al. 2012 ; Rosset et al. 2021 ). Efforts to decode the molecular dialogue have expanded in recent years to other cnidarian species, with a particular emphasis on the sea anemone \n E. diaphana \n (Baumgarten et al. 2015 ; Jacobovitz et al. 2023 ; Roberty et al. 2024 ). We find that the sea anemone \n A. elegantissima \n in each symbiotic phenotype has distinct transcriptional and metabolic, but not bacterial, signatures under laboratory conditions. Moreover, we find that \n A. elegantissima \n is dominated by a single Sphingomonas species, suggesting that this sea anemone could potentially have a functionally relevant prokaryotic symbiosis across these phenotypes. Signatures of symbiosis for \n A. elegantissima \n tended to follow one of three patterns. The first, and most pronounced, related to being in a general state of symbiosis. This was primarily observed at the transcriptional level, where the majority of differentially expressed genes (~70.5%) across diverse biological processes were between aposymbiotic sea anemones and those that hosted either of the photosynthetic symbionts. A symbiosis‐related transcriptional pattern is widely observed throughout animals, including sponges (Marulanda‐Gomez et al. 2024 ), worms (Pees et al. 2024 ), and squid (Kremer et al. 2013 ). Moreover, this pattern is also observed in the sea anemones \n A. elegantissima \n —based on a cDNA microarray of the holobiont (Rodriguez‐Lanetty et al. 2006 )—and \n E. diaphana \n (Lehnert et al. 2014 ) as well as the temperate coral \n Astrangia poculata \n (Changsut et al. 2022 ). The processes related to being in symbiosis are also commonly observed, which include metabolism and respiration (Turnbaugh et al. 2008 ), circadian rhythm (Nyholm and McFall‐Ngai 2004 ), immunity (Thaiss et al. 2016 ), neurogenesis (Giez et al. 2023 ), and reproduction and development (Carrier and Bosch 2022 ). \n A. elegantissima \n , thus, exhibits a similar organismal‐level response to being in symbiosis that is expected for other cnidarians. The second symbiosis signature for \n A. elegantissima \n relates to the specificity of associating with either of the two photosymbionts. Sea anemones harbouring the dinoflagellate differentially expressed genes predominantly relating to nucleotide biosynthesis amongst other GO terms, while those associating with the chlorophyte had a more even expression of many functional categories (e.g., cell size regulation, reproduction, and vitamin and amino acid transport). Processes identified from these differentially expressed genes translated to the metabolite profiles, as the metabolite profiles for holobionts containing B. muscatinei were glycerol and sugar based, while \n A. elegantissima \n associating with \n E. marina \n was primarily based on amino acids. These metabolic differences between the symbionts directly contribute to physiological and life history differences between sea anemones in each symbiotic phenotype (Minnick 1984 ; Trench 1971a , 1971b ). Sea anemones hosting \n E. marina \n tend to reproduce sexually, while being in symbiosis with the more productive B. muscatinei promotes cloning by fission (Bergschneider and Muller‐Parker 2008 ; Bingham et al. 2014 ). Consistent with this is that \n A. elegantissima \n that were in symbiosis with \n E. marina \n differentially expressed genes relating to reproduction and development, while sea anemones with B. muscatinei did not. The third signature for \n A. elegantissima \n was the lack of a symbiotic influence, which specifically related to the bacterial community associated with this sea anemone. A stable bacterial community dominated by a single Sphingomonas taxon was unexpected for three reasons. First, a major change in host physiological state commonly results in a compositional shift in the associated microbial community (Carrier and Reitzel 2017 ; Kohl and Carey 2016 ). Second, microbiome diversity spans from a single obligate symbiont to a diverse, mostly facultative community (O'Brien et al. 2019 ). Sea anemones and other cnidarians are commonly observed to associate with a diverse microbiome composed of 100–1000 s of microbial species (Bourne et al. 2016 ; McCauley et al. 2023 ). Third, it has been previously shown that \n A. elegantissima \n and \n E. marina \n both associate with a symbiont‐specific microbiome (Morelan et al. 2019 ; Röthig et al. 2016 ). Transferring animals into stable laboratory conditions commonly leads to compositional shifts and a reduction in the diversity of the host‐associated bacterial community (Carrier and Reitzel 2017 ; Kohl et al. 2014 ). Our mesocosm conditions were provided directly and independently with ambient seawater and, thus, this should have maintained a symbiont‐specific microbiome signature similar to what was reported for \n A. elegantissima \n that was sampled directly from the field (Morelan et al. 2019 ). An alternate explanation of a potential laboratory influence on the bacterial community is that this \n A. elegantissima \n population predominantly associates with a single Sphingomonas taxon. An animal species that associates with a primary microbial taxon that is otherwise expected to interact with a diverse microbiome has been observed [e.g., the sponge \n Halichondria panicea \n (Knobloch et al. 2019 )]. Sphingomonas is a commonly observed genus in the bacterial community of cnidarians, suspected to potentially cause a band disease in the elliptical star coral \n Dichocoenia stokesi \n , and can be a DNA extraction kit contaminant (McCauley et al. 2023 ; Morelan et al. 2019 ; Richardson et al. 1998 ; Salter et al. 2014 ). The latter is a possibility because this study did not include this type of control. If Sphingomonas were removed, then the bacterial community of \n A. elegantissima \n from near‐ambient mesocosms would have been nearly depleted of detectable bacterial taxa; thus, contamination seems unlikely. The functional role of Sphingomonas or a more diverse microbial community remains uncertain across each symbiotic phenotype for the sea anemone \n A. elegantissima \n . Collectively, this study has provided a reference genome for \n A. elegantissima \n as well as assessed whether this sea anemone has a unique transcriptional, metabolomic, and bacterial signature when in symbiosis with the dinoflagellate B. muscatinei , the chlorophyte \n E. marina \n , or when aposymbiotic. Providing this series of resources has set the framework to compare the \n A. elegantissima \n genome to other sea anemones, corals, and cnidarians (Zimmermann et al. 2023 ) to determine the host‐microbe dialogue during symbiont shuffling (e.g., in response to light; Cunning et al. 2015 ), how these partnerships break down in response to future climate conditions (Brown 1997 ), and to perform experimental infections with thermally resilient photosymbionts (Herrera et al. 2021 ). \n A. elegantissima \n is, thus, another tractable experimental cnidarian system that continues to provide insights into a symbiosis that is critically important for both temperate and tropical ecosystems."
} | 3,967 |
24809396 | PMC4014880 | pmc | 147 | {
"abstract": "Nanoscale inorganic electronic synapses or synaptic devices, which are capable of emulating the functions of biological synapses of brain neuronal systems, are regarded as the basic building blocks for beyond-Von Neumann computing architecture, combining information storage and processing. Here, we demonstrate a Ag/AgInSbTe/Ag structure for chalcogenide memristor-based electronic synapses. The memristive characteristics with reproducible gradual resistance tuning are utilised to mimic the activity-dependent synaptic plasticity that serves as the basis of memory and learning. Bidirectional long-term Hebbian plasticity modulation is implemented by the coactivity of pre- and postsynaptic spikes, and the sign and degree are affected by assorted factors including the temporal difference, spike rate and voltage. Moreover, synaptic saturation is observed to be an adjustment of Hebbian rules to stabilise the growth of synaptic weights. Our results may contribute to the development of highly functional plastic electronic synapses and the further construction of next-generation parallel neuromorphic computing architecture.",
"discussion": "Discussion Although synaptic plasticity is governed by a multifactor and multiform rule according to the type or location or function of the synapses, and the interaction between timing- and rate- and voltage-dependent processes is still under intense debate, independent STDP or SRDP function is widely utilised in computational and experimental neural networks to implement more complex cognitive functions, such as associative learning and pattern classification 39 40 41 . We see the implementation of activity-dependent synaptic plasticity in an electronic synapse as a solid step toward constructing neuromorphic systems, but these results still call for additional research efforts on large-scale integration 42 . On the one hand, the performances of electronic synapses, such as power consumption and device scaling, need to be improved 43 . On the other hand, novel network architectures are urgently demanded 44 . The coordinated developments of above two aspects would breed the success of memristor-based neuromorphic computing. In summary, in one Ag/AgInSbTe/Ag structure chalcogenide memristor-based electronic synapse, we have experimentally demonstrated the activity-dependent synaptic plasticity that is the basic phenomenon for learning in various neuronal systems. The spike-timing dependence of the four forms of plasticity were emulated, and long-term synaptic modification depends on the exact timing of the pre-and postsynaptic spikes. Spike-rate dependent changes and voltage-based modifications in synaptic plasticity were also performed, showing that LTP and LTP also depend on stimulation frequency and synaptic voltage. Moreover, synaptic saturation was observed in our electronic synapse, which is a crucial adjustment of Hebbian rules to stabilise the growth of synaptic weights. We believe that the demonstrated synaptic operation in this study, together with the booming development of sub-ns memristive devices with high density and low power consumption, will contribute to the construction of next-generation neuromorphic computing architecture that requires plastic electronic synapses."
} | 812 |
36959175 | PMC10036481 | pmc | 148 | {
"abstract": "Cable bacteria are centimeter-long filamentous bacteria that conduct electrons via internal wires, thus coupling sulfide oxidation in deeper, anoxic sediment with oxygen reduction in surface sediment. This activity induces geochemical changes in the sediment, and other bacterial groups appear to benefit from the electrical connection to oxygen. Here, we report that diverse bacteria swim in a tight flock around the anoxic part of oxygen-respiring cable bacteria and disperse immediately when the connection to oxygen is disrupted (by cutting the cable bacteria with a laser). Raman microscopy shows that flocking bacteria are more oxidized when closer to the cable bacteria, but physical contact seems to be rare and brief, which suggests potential transfer of electrons via unidentified soluble intermediates. Metagenomic analysis indicates that most of the flocking bacteria appear to be aerobes, including organotrophs, sulfide oxidizers, and possibly iron oxidizers, which might transfer electrons to cable bacteria for respiration. The association and close interaction with such diverse partners might explain how oxygen via cable bacteria can affect microbial communities and processes far into anoxic environments.",
"conclusion": "Conclusion and perspectives Cable bacteria attracting and intimately engaging with a diversity of other bacteria in sediment is yet another surprise finding in relation to the electrical currents first derived from geochemical anomalies and later ascribed not to conductive minerals or nanowires but living wires in the form of the centimeter-long cable bacteria 1 , 29 . Now a line of new and interesting questions for future research arises: How widespread and important are these interactions in situ? What specific molecules mediate the interactions, and how and where are they processed in the cells on either side? How much of the current in a cable bacterium does arise from the bacterial flocks around it, and what share may the flock gain of the energy conserved from the primary oxidation processes to the final oxygen reduction? Is the interaction beneficial or detrimental for the cable bacterium, and is it controlled in any way? Are there harder-to-observe interactions with other, non-motile cells in the natural sediment environment, and what is the full palette of microbial processes stimulated by the electrical shortcut to oxygen offered by cable bacteria?",
"introduction": "Introduction Cable bacteria are long filamentous bacteria that can transmit electrons over centimeter distances and thereby couple the oxidation of sulfide to the remote reduction of oxygen or nitrate 1 , 2 . They occur globally in marine and freshwater sediments and aquifers 3 – 5 and interfere directly with sulfur, oxygen, carbon, and nitrogen cycling 6 . Via pH gradients and electric fields induced by their relocation of electrons, they also indirectly influence iron, calcium, cobalt, and arsenic cycling and all ion fluxes in their habitats 6 – 9 . In freshwater sediments, they can cause a 4.5-fold stimulation of sulfate reduction and drastically lower methane emission 10 , 11 . Cable bacteria activity has also been linked to enhanced carbon assimilation of autotrophic sulfide oxidizers 12 and correlated with the distribution of iron-cycling bacteria in marine sediments 13 . These apparent associations have led to speculations that bacteria in anoxic sediment somehow may use cable bacteria as an electron conduit to oxygen 14 , 15 . Here we use a combination of microscopic observations, metagenome sequencing, laser microdissection, and Raman microscopy to demonstrate the dynamics and intimacy of this association, tentatively identify the bacteria involved, and propose a likely mechanism for electron transfer between associated bacterial flocks and cable bacterial filaments.",
"discussion": "Results and discussion Bacteria flock around cable bacteria and appear metabolically stimulated Cable bacteria from an enrichment of the freshwater strain Ca . Electronema aureum GS 16 were observed under semi-natural conditions on a microscope slide (a so-called trench slide), where oxygen diffused in from the edge of the coverslip, while organic matter, sulfide, and other nutrients were provided from sediment in a central compartment (trench) 17 , 18 . This setup established an observation zone where cable bacteria stretched out from the sediment to the oxic–anoxic interface, which was clearly delineated by a microaerophilic veil of motile, aerobic bacteria (Fig. S 1 ). In the anoxic part of the zone, up to 4 mm away from the oxic-anoxic interface, bacterial cells were discovered to swim in a flock around cable bacteria segments, generally outnumbering adjacent cable bacteria cells by 2.2:1 (Fig. 1A , Movie S 1 , Table S1 ). Detailed cell tracking showed chemotactic behavior towards the cable bacteria: the flocking cells were concentrated close to the cable bacteria filaments, with the highest cell densities within a distance of 20 µm but still increased densities until at least 50 µm away (Fig. 1B , Fig. S 2 ). There was no overt pattern of the flocking bacteria touching cable bacteria, but with the resolution limitations of conventional light microscopes and the dynamic nature of the interaction deterring higher resolution methods, we currently cannot exclude that touching might occur; but if so, it was rare and brief. The swimming velocity of the flocking bacteria was significantly enhanced within a distance of 20 µm (Fig. 1C ), indicating an increased proton motive force 19 – 21 . The fraction of bacterial cells detectable by fluorescence in situ hybridization (FISH) within 10 µm distance from cable bacteria was significantly higher than >10 µm away (Table S2 ), suggesting that they had a higher ribosome count and thus higher metabolic activity compared to bulk sediment bacteria 22 . Taken together, this indicates high metabolic rates within the flock and metabolic stimulation of the flocking bacteria when getting really close to the cable bacteria. Fig. 1 Documentation and main properties of bacterial flocks around cable bacteria. A Flocking bacteria attracted to a cable bacterium filament (center) (Scale bar, 10 µm). B Counts of swimming bacteria at different distances to the cable bacterium filament (960,071 counts of 3211 flocking bacteria in 12 video frames, mean distance 17.42 µm. The means of the individual samples ranged from 4.96 to 24.58 µm. C Difference in mean swimming speed of bacterial cells relative to their distance to the cable bacterium filament. The shaded blue area corresponds to a distance within 20 µm of a cable bacterium, shaded green to more than 20 µm. Welch’s two-sample t -test (two-sided) shows that the swimming speed of cells is significantly different between these two distance groups (indicated by *); p -value = 2.2e −16 ( N samples = 11, N cells = 2712). D Density plot of bacterial cell sizes from all samples ( n = 12), showing that the majority of interacting cells is small. E Phase contrast images of the different cell morphologies found. Source data are provided as a Source Data file. Bacterial flocks disperse when cable bacteria are disconnected from oxygen Flocks of swimming bacteria were commonly observed in our setup, i.e., on 17 of 21 cable bacteria in contact with oxygen; in contrast, flocks were never observed on cable bacteria not in contact with oxygen, supporting a dependency of the chemotaxis on a high-potential electron sink in the cable bacteria 18 . This was strikingly confirmed by the swift response of the bacterial flocks upon cutting the cable filament in two with a laser (Fig. 2 ): flocking around the part now cut off from its electrical connection to oxygen ceased within 13 ± 2 s (Table S3 ). When a cable bacterium filament was cut in the middle of a bacterial flock, the response was even more evident. The flocking bacteria immediately dispersed from the part of the filament that was no longer connected to oxygen but was still observed around the part that still retained its connection to oxygen (Movie S2 ). This instant response suggests that the flocking bacteria are attracted by a condition created by cable bacteria exclusively while they conduct electrons to oxygen, and not, for example, extracellular polymers excreted as part of their motility 17 . Fig. 2 Principle and example of results of the laser cut experiment ( n = 9). A Schematic representation. A cable bacterium connected to oxygen at the right side is cut near the oxic-anoxic interface using a laser microdissection microscope. B Result. Bacterial tracks on the suboxic part of a cable bacterium generated from a video before and after a cut. Red lines are tracks of swimming bacteria, and black lines and dots denote the position of the cable bacterium for every 50 frames. The cable bacterium moves more after the cut as if the filament begins its oxygen chemotaxis (Scale bar, 10 µm). Source data are provided as a Source Data file. Cable bacteria associates are morphologically, phylogenetically, and metabolically diverse The bacteria interacting with the freshwater cable bacteria were morphologically highly diverse. The majority were rod-shaped, vibrioid, or ovoid cells of 1–2 µm size (long axis), but a few distinct larger cell types, including straight rods, curved rods, and spirochete-like cells, were also recorded (Fig. 1D, E , Fig. S 3 ). To get insights into the likely identity and metabolic diversity of the bacteria forming the flocks, we assembled 27 good-quality genomes (Fig. 3 , Supplementary Datasets 1 – 3 ) from a metagenome prepared by sampling the cable bacteria-observation zone of one of the microscopy slides (Fig. S 1 ). We identified 25 genera from 6 phyla, which, based on genome annotation and comparison to their closest relatives, had four major unifying traits: motility, chemotaxis, organotrophy, and respiration (with oxygen/nitrate/nitrite) (Fig. 3 ). In addition, sulfide oxidation and autotrophy were common, while sulfate or iron reduction and iron oxidation were rare; methane oxidation was never identified. These data are consistent with a scenario in which the flocking bacteria can oxidize organic compounds, sulfide, and possibly Fe 2+ by delivering electrons to cable bacteria, i.e., they breathe via cable bacteria. While some of the extracted genomes do show genes for extracellular electron transfer (EET), and some of the closest relatives of those bacteria also exhibit EET capabilities (Fig. 3 ), there is no clear and consistent pattern for a known EET mechanism. Fig. 3 Genome-based phylogeny and selected key features of putative cable bacteria-associated bacteria. Data were derived from the annotation of metagenome-assembled genomes and from literature searches. EET, extracellular electron transfer. Source data are provided in Supplementary Datasets 1–3. Interspecies electron transfer via soluble intermediates likely explains the flocking Electrons can be exchanged between prokaryotic cells by direct contact, by transfer through protein nanowires or conductive materials, or mediated through soluble electron shuttles 23 . The apparent chemotactic behavior with constant swimming and no stop-and-touch behavior, as observed in some electron transfers to minerals 24 , suggests that interspecies electron transfer (IET) from flocking bacteria to cable bacteria is mediated through gradients of soluble intermediates. This was further supported by observations of redox states of cellular cytochromes with Raman microscopy 18 ; cells of the flocking bacteria captured and moved with a laser tweezer were significantly more oxidized when placed closer than 5 µm to the cable bacterium and more reduced when more than 50 µm away (Fig. 4 , Fig. S 4 ). The same significant change was observed with cells introduced into the slides from a culture of Acidovorax facilis DSM649, a cultured representative of one of the most abundant members of the associated bacterial community (Fig. 3 ; 96.03% average nucleotide identity (ANI) to genome bin GS.4 recovered in this study); cytochrome redox state did not change in control cells randomly moved with the laser tweezer (Fig. S 5 ). Fig. 4 Cytochromes in flocking bacteria are more oxidized when close to a cable bacterium. A Schematic representation of the experiment. A flocking cell is captured and moved right next to a cable bacterium filament, measured by Raman microscopy, then moved ~50–100 µm away and measured again after ~3 s. B The change in normalized intensity of the 750 cm −1 band for each individual cell when moved next to and away from the cable bacterium. The 750 cm −1 band is indicative of cytochrome redox state, with high values for reduced and low values for oxidized cytochromes. Band intensities and, thus, cytochrome redox states are significantly different between the two positions ( N native cells = 5, p -value = 0.015, indicated by *; N A.facilis = 8, p -value = 0.0387 indicated by **, two-sided t -test for dependent samples,); a.u., arbitrary units. Source data are provided as a Source Data file. We estimate that shuttle concentrations in the nM range are sufficient to support the respiration of the bacterial flock (Supplementary Note 1 , Table S4 ). Since aquatic sediments can contain high-potential shuttle compounds exceeding this concentration (e.g. >1 mg ml −1 humic acids in lake sediment 25 , or flavins in the nM range in marine sediment 26 ), shuttles do not have to be produced by cable bacteria or associated bacteria. Also, the turnover rate of a soluble mediator in this concentration range would be below 2 s (Supplementary Note 1 , Table S4 ). This rapid turnover suggests an immediate depletion of the oxidized mediator once cable bacteria stop re-oxidizing its reduced form and thus explains the rapid dispersal of the bacterial flock upon cutting the connection of the filament to oxygen. That cable bacteria provide a potent oxidized electron shuttle seems the only plausible explanation for the proliferation of other bacteria associated with cable bacteria in the anoxic sediment compartment 12 , 13 , the amount and vigor of motile cells surrounding cable bacteria (Fig. 1A , Movie S1 ), and their rapid dispersal after cutting (Movie S2 ). On the cable bacteria side, no known EET capabilities have been identified in cable bacteria genomes 27 , yet intact cable bacteria have successfully been connected to electrodes 28 , making it likely that an undescribed outer membrane electron conduit exists, which is able to receive electrons from a shuttle 28 . Conclusion and perspectives Cable bacteria attracting and intimately engaging with a diversity of other bacteria in sediment is yet another surprise finding in relation to the electrical currents first derived from geochemical anomalies and later ascribed not to conductive minerals or nanowires but living wires in the form of the centimeter-long cable bacteria 1 , 29 . Now a line of new and interesting questions for future research arises: How widespread and important are these interactions in situ? What specific molecules mediate the interactions, and how and where are they processed in the cells on either side? How much of the current in a cable bacterium does arise from the bacterial flocks around it, and what share may the flock gain of the energy conserved from the primary oxidation processes to the final oxygen reduction? Is the interaction beneficial or detrimental for the cable bacterium, and is it controlled in any way? Are there harder-to-observe interactions with other, non-motile cells in the natural sediment environment, and what is the full palette of microbial processes stimulated by the electrical shortcut to oxygen offered by cable bacteria?"
} | 3,943 |
38075285 | PMC10704147 | pmc | 152 | {
"abstract": "Spiking neural networks coupled with neuromorphic hardware and event-based sensors are getting increased interest for low-latency and low-power inference at the edge. However, multiple spiking neuron models have been proposed in the literature with different levels of biological plausibility and different computational features and complexities. Consequently, there is a need to define the right level of abstraction from biology in order to get the best performance in accurate, efficient and fast inference in neuromorphic hardware. In this context, we explore the impact of synaptic and membrane leakages in spiking neurons. We confront three neural models with different computational complexities using feedforward and recurrent topologies for event-based visual and auditory pattern recognition. Our results showed that, in terms of accuracy, leakages are important when there are both temporal information in the data and explicit recurrence in the network. Additionally, leakages do not necessarily increase the sparsity of spikes flowing in the network. We also investigated the impact of heterogeneity in the time constant of leakages. The results showed a slight improvement in accuracy when using data with a rich temporal structure, thereby validating similar findings obtained in previous studies. These results advance our understanding of the computational role of the neural leakages and network recurrences, and provide valuable insights for the design of compact and energy-efficient neuromorphic hardware for embedded systems.",
"conclusion": "5 Conclusion In this work we explored the effect of spiking neurons synaptic and membrane leakages, network explicit recurrences and time constants heterogeneity on event-based spatio-temporal pattern recognition. The main findings of our work can be summarized as follows: Neural leakages are only necessary when there are both temporal information in the data and explicit recurrent connections in the network. Neural leakages do not necessarily lead to sparser spiking activity in the network. Time constants heterogeneity slightly improves performance and reduces time constants tuning efforts on data with a rich temporal structure and does not affect performance on data with a spatial structure. This work supports the identification of the right level of model abstraction of biological evidences needed to build efficient application-specific neuromorphic hardware. This is a crucial analysis for advancing the field beyond state-of-the-art, especially when constrains on resources are critical (e.g., edge computing). In fact, when using digital neuromorphic architectures, IF neurons have been shown to be 2 × smaller and more power-efficient than formal Perceptrons (Khacef et al., 2018 ). It is nevertheless not clear how this gain evolves when adding a multiplier to implement a LIF or CUBA-LIF neuron. Further works will focus on implementing these two architectures in FPGAs for fast prototyping. In addition, IF neurons give the possibility to implement a digital asynchronous processing purely driven by the input, since there is no inherent temporal dynamics in the spiking neurons. On the other hand, LIF and CUBA-LIF neurons require algorithmic time-steps where the leakage is updated regardless of the presence of input spikes. Further works will explore the impact of both paradigms in energy-efficiency on the Loihi neuromorphic chip (Davies et al., 2018 ). Furthermore, it is important to mention that our results only hold in benchmarking so far. In a real-world scenario such as continuous keyword spotting, there can be more noise in the data but also in void. Hence, when using the IF neurons that do not have any leakage, this noise can accumulate and create false positives and degrade the performance. Indeed, the low-pass filtering effect of the spiking neurons leakages has been shown to eliminate high frequency components from the input and enhance the noise robustness of SNNs, especially in real-world environments (Chowdhury et al., 2021 ). In addition, given that the LIF model achieved a superior performance when compared to the CUBA-LIF, it is important to investigate where the latter could perform better. More complex tasks could show such a gain for the CUBA-LIF neuron, because of its current compartment which is an extra state that gives more potential for spatio-temporal feature extraction. Finally, spiking neural networks in neuromorphic hardware can be used beyond fast and efficient inference, by adding adaptation through local synaptic plasticity (Qiao et al., 2015 ; Khacef et al., 2022 ; Quintana et al., 2022 ). In this context, the impact of the leakage can be different, as the inherent temporal dynamics is required in some plasticity mechanisms (Brader et al., 2007 ; Clopath et al., 2010 ) for online learning.",
"introduction": "1 Introduction Over the last decade, Artificial Neural Networks (ANNs) have been increasingly attracting interest in both academia and industry as a consequence of the explosion of open data and the high computing power of today's computers for training and inference. The state-of-the-art performance of deep neural networks on various pattern recognition tasks has given neural networks a leading role in Machine Learning (ML) algorithms and Artificial Intelligence (AI) research. However, the technological drive that has supported Moore's Law for 50 years and the increasing computing power of conventional processors is reaching a physical limit and is predicted to flatten by 2025 (Shalf, 2020 ). Hence, deep learning progress with current models and implementations will become technically, economically, and environmentally unsustainable (Thompson et al., 2020 , 2021 ). This limit is particularly prohibitive when targeting edge applications in embedded systems with severe constraints in latency and energy consumption (Rabaey et al., 2019 ). Neuromorphic computing is a promising solution that takes inspiration from the biological brain which can reliably learn and process complex cognitive tasks at a very low power consumption. On the one hand, neuromorphic sensors are event-based sensors and capture information with a high temporal resolution and high spatio-temporal sparsity at low-latency and low-power consumption (Liu et al., 2010 ; Gallego et al., 2022 ). On the other hand, neuromorphic processors are asynchronous and use parallel and distributed implementations of synapses and neurons where memory and computation are co-localized (Mead and Conway, 1980 ; Chicca et al., 2014 ), hence adapting the hardware to the computation model (Schuman et al., 2017 ; Bouvier et al., 2019 ). Spiking Neural Networks (SNNs) are the third generation of artificial neural models (Maass, 1997 ) that are investigated to exploit the advantages of event-based sensing and asynchronous processing at the algorithmic level. Inspired from the neuroscience literature, Spiking Neural Networks (SNNs) show promising performance in embedded spatio-temporal pattern recognition (Davies et al., 2021 ). For example, compared to a conventional approach using formal neural networks on an embedded Nvidia Jetson GPU, SNNs on the Intel Loihi neuromorphic chip (Davies et al., 2018 ) achieve a gain in energy-efficiency of 30× in multimodal (vision and EMG) hand gesture recognition (Ceolini et al., 2020 ) and 500× in tactile braille letters recognition (Muller-Cleve et al., 2022 ), at the cost of a loss in accuracy depending on the application. Multiple models of spiking neurons have been proposed in the literature (Hodgkin and Huxley, 1952 ; Kistler et al., 1997 ; Izhikevich, 2003 ) and implemented in hardware (Indiveri et al., 2011 ) with different levels of biological plausibility and computational complexity. However, there is a lack of understanding of how each of the factors determining the biological neuronal response can be effectively used in learning and inference. A key question for advancing the field is therefore to identify the right level of abstraction inspired from biology to achieve the best inference performance within strict constrains in speed/latency and power efficiency on neuromorphic hardware. This work attempts to partially answer this question by studying the effect of spiking neurons leakages in feedforward and recurrent neural networks for event-based visual and auditory pattern recognition tasks, in terms of accuracy and spiking activity. Today, digital neuromorphic chips from academia and industry use both non-leaky [e.g., SPLEAT (Abderrahmane et al., 2022 ) and DynapCNN (Liu et al., 2019 )] and leaky [e.g., MorphIC (Frenkel et al., 2019 ) and Loihi (Davies et al., 2018 )] spiking neurons. Understanding the computational role of the leakages provides insights for the hardware architecture of neuromorphic processors as they require extra circuitry overheads (Khacef et al., 2018 ). In Section 2, we introduce the spiking neuron models and present the training methodology. In Section 3, we present our grid search experiments and a detailed analysis of the resulting performance trends across different spiking neuron models, leakage parameters, network topologies, as well as time constant heterogeneity. Finally, in Sections 4 and 5 we discuss the results, highlighting the main insights, limits and outlook of our work.",
"discussion": "4 Discussion In the neuro-scientific literature, it has been reported that leakages in biological neurons exist in many contexts such as synaptic transmission in the visual cortex (Artun et al., 1998 ) and sodium ion channels (Snutch and Monteil, 2007 ; Ren, 2011 ). Many spiking neuron models imitate this leaky behavior through an exponential decay in the synaptic current and membrane potential. Other models prioritize computational efficiency by removing the leakage. To tackle the lack in understanding of the effect of these leakages from the modeling perspective, we confronted three spiking neuron models with variable degrees of leaky behavior, namely the CUBA-LIF, LIF, and IF, in classification tasks with a number of degrees of freedom. We first trained SNNs using the three neuron models with a feed-forward network to classify visual patterns of written digits from the N-MNIST dataset and auditory information of spoken digits from the SHD datasets. Surprisingly, the IF model, despite the absence of leaky behavior and the resulting lack of inherent temporal dynamics, slightly outperformed the other models on the SHD by reaching an accuracy of 78.36 ± 0.87%, and closely matched the best of LIF model accuracy on the N-MNIST by reaching 97.50 ± 0.06%. CUBA-LIF on the other hand, had the inferior performance among the three models on both datasets despite its intrinsic temporal dynamics caused by both synaptic and membrane leaks. Both LIF and CUBA-LIF saw a drastic decrease in accuracy when τ mem is less than 420 ms , which leads to a fast decay in membrane potential and loss of information. We also found that CUBA-LIF reached its highest accuracies when its dynamics are close to those of the LIF. We conclude that leakages do not necessarily lead to improved performances even on temporally complex tasks when using feed-forward networks. In terms of sparsity, it is IF to see sparser activity in IF neurons and CUBA-LIF neurons with smaller values of τ syn than their LIF counterpart. Upon inspection of the trained weights distributions, it seems that BPTT is tailoring LIF neurons to have bigger weights, and hence more spikes. Therefore, leakages do not always lead to sparser activity. Furthermore, we noticed that very low spiking activity resulted in the worst classification performance on the SHD. Very high spiking activity associated with bigger τ syn values also resulted in a worsened performance. These results suggest that there is a sweet-spot where a sufficient amount of spikes produce an optimal classification accuracy. Overall, IF neurons are sufficient when using data without temporal information or a network without recurrence in terms of classification accuracy and sparsity. It suggests that the fundamental ingredient of spiking neurons is their statefullness, i.e., having an internal state with an implicit recurrence, even without leakage. Furthermore, they offer a better alternative if we consider digital neuromorphic hardware design that is based on application-specific needs. IF neurons could be very cheap in terms of hardware resources, as they only perform additions for the input integration and a comparison for the output evaluation. In contrast, the LIF and CUBA-LIF neurons require multipliers to implement the leakage in their current and/or voltage compartments as shown in Table 6 , thus resulting in more expensive hardware. Table 6 Number of multiplication, addition, and comparison operations per spiking neuron at each time step, where N is the number of inputs (feedforward and/or recurrent) to the neuron and P is the percentage of those inputs that receive a spike. \n Neuron model \n \n IF \n \n LIF \n \n CUBA-LIF \n Multiplications 0 1 2 Additions N × P N × P N × P +1 Comparisons 1 1 1 Next, we added explicit recurrent connections to the neurons in the hidden layer. Expectedly, we saw a big improvement in accuracy for the SHD that has a rich temporal structure and no improvement at all for the N-MNIST that has mostly spatial structure. However, recurrences did not have any impact on the IF neuron on both datasets. Therefore, we conclude that the inherent temporal dynamics introduced by the leakages are only necessary when we use both data with a rich temporal structure and a neural network with a explicit recurrence. The best SHD accuracies we were able to obtained in a RSNN were very close to state-of-the-art results (Dampfhoffer et al., 2022 ) such that we reached 82.74 ± 0.17% with CUBA-LIF and 83.41 ± 0.37% with the LIF. In terms of sparsity, we saw a bigger increase in spiking activity with the SHD than the N-MNIST. In both datasets, the CUBA-LIF neurons with the best time constants combinations added the smallest number of spikes, which gives them an advantage in sparsity compared to LIF neurons. Finally, we introduced heterogeneity in the considered spiking neurons by incorporating learnable time constants in the training process following two approaches: homogeneous initialization and random initialization. Heterogeneous training with homogeneous initialization slightly improved performance on the SHD, which has a complex temporal structure. The best SHD accuracies we obtained with heterogeneous training in RSNN were also very close to state-of-the-art results (Dampfhoffer et al., 2022 ) with 82.84 ± 1.17% for CUBA-LIF and 83.47 ± 2.12% for LIF. However, results are very sensitive to initial values of time constants. On the other hand, random initialization did not improve performance but proved it can be promising given the 83.18 ± 0.19% achieved by the LIF with the SHD. For the CUBA-LIF, however, further investigations are required to assess the impact of its current compartment."
} | 3,762 |
40346103 | PMC12064777 | pmc | 154 | {
"abstract": "Neuromorphic computing based on two-dimensional materials represents a promising hardware approach for data-intensive applications. Central to this new paradigm are memristive devices, which serve as the essential components in synaptic kernels. However, large-scale implementation of synaptic matrix using two-dimensional materials is hindered by challenges related to random component variation and array-level integration. Here, we develop a 16 × 16 computing kernel based on two-transistor-two-resistor unit with three-dimensional heterogeneous integration compatibility to boost energy efficiency and computing performance. We demonstrate the 4-bit weight characteristics of artificial synapses with low stochasticity. The synaptic array demonstration validates the practicality of utilizing emerging two-dimensional materials for monolithic three-dimensional heterogeneous integration. Additionally, we introduce the Gaussian noise quantization weight-training scheme alongside the ConvMixer convolution architecture to achieve image dataset identification with high accuracy. Our findings indicate that the synaptic kernel can significantly improve detection accuracy and inference performance on the CIFAR-10 dataset.",
"introduction": "Introduction Computing for data-intensive artificial intelligence spans multiple domains, such as image processing, natural language processing, smart transportation, and medical diagnosis 1 – 5 . The increasing complexity and sheer scale of the application scenarios have come with a voracious appetite for computing power 6 , 7 . Conventional hardware is reaching its limits in the planar scale, exposing inherent bottlenecks that hinder performance 8 , 9 . In this context, neuromorphic computing, which mimics the operation of biological neural networks, presents a promising solution to deal with the exponential growth in information data 10 , 11 . The established artificial neural network architectures, such as feedforward neural networks (FNNs) 12 , convolutional neural networks (CNNs) 13 , recurrent neural networks (RNNs) 14 , and Spiking Neural Networks (SNNs) 15 , are inherently reliant on the programmable weight matrix computations executed by synaptic computing kernels (SCKs). Synaptic arrays, which emulate biological synapses, are functional units responsible for carrying out matrix-vector multiplications and weight-update operations critical to neural network algorithms. The dynamical reconfiguration of synaptic weights is pivotal for enabling learning, inferencing, and decision-making capabilities with the bio-inspired computing architectures 16 , 17 . Thus, developing high-performance, energy-efficient, and scalable synaptic arrays is of paramount importance to the practical implementation and widespread adoption of artificial neural network models. Moreover, advancing device technologies and circuit architectures that faithfully reproduce the programmable synaptic functions remain a central focus in neuromorphic computing. In recent years, memristors have emerged as a compelling choice for SCK hardware architectures, attributed to their suitability for matrix multiplication calculations. However, traditional memristive crossbar arrays have exposed defects in the neuromorphic computation task of dense SCKs, such as leakage currents 18 , 19 and coupling between reading and writing operations 20 . To address these challenges, novel synaptic matrix architectures like one-transistor-one-resistor (1T1R) 21 – 23 , two-transistor-one-capacitance (2T1C) 24 , one-transistor-four-resistor (1T4R) 25 , and two-transistor-two-resistor (2T2R) 26 – 28 have been proposed. The 1T1R configuration offers independent control of the access transistor and the non-volatile synapse, enabling selective addressing and updating of individual synaptic weights while not disturbing the state of neighboring synapses 21 – 23 . This capability enhances the flexibility and practicability of neural computation. Further, the monolithic integration of two-dimensional (2D) materials with memristors in the 1T1R configuration can significantly enhance performance without increasing overhead, which is critical to minimizing spatial and energy budgets in neuromorphic systems. The 2T2R architecture, which supports differential weight representation, expands the design space for synaptic computing by allowing additional degrees of freedom in weight programming and modulation. This configuration potentially increases dynamic range and precision, making it well-suited for artificial neural network models with enhanced computational capabilities 26 – 30 . However, its inherent complexity of functionality needs more stringent performances, posing major integration challenges, particularly in three-dimensional (3D) heterogeneously stacked architectures. With the growing demand for compact and stackable 2T2R SCKs, 2D materials have attracted attention due to their atomic thinness and potential in low-stress integration 25 , 31 – 34 . Amidst this context, the maturity of large-scale fabrication for MoS 2 has positioned the 2D material as an enabler for heterogeneous integration and the realization of 3D computing architectures 9 , 35 – 37 . MoS 2 exhibits good carrier transport properties, high drive current, and continuously improving wafer-scale uniformity via chemical vapor deposition (CVD) 38 , facilitating the integration of 2D materials with functional components to implement advanced multifunctional devices and systems. Lu et al. 36 reported ten MoS 2 logic circuit tiers for monolithic 3D integration system by repeating the van der Walls (vdW) lamination process in the vertical direction. Xie and co-workers 25 demonstrated scale-level integration of 1T4R unit, although it remains a separate configuration that does not form a complete system. Kang et al. 34 mimicked the vertical heterogeneous integration with a MoS 2 (1T)–WSe 2 /h-BN (1R) structure, validating the feasibility of the SCK monolithic 3D heterogeneous integration process. However, there is a notable lack of research on reliability and yield of these integrated systems. The endurance and variability of neuromorphic devices based on heterogeneous integration of traditional materials and emergent 2D materials face significant scaling challenges. These challenges stem from the need for high-quality large-area synthesis, effective material transfer, and improved fabrication techniques 39 , which have hindered the progression of neuromorphic applications. Therefore, achieving high-reliability 2D heterogeneous stacked 2T2R SCK is of paramount significance. In this work, we introduce a heterogeneously integrated SCK based on 2D materials for multi-bit storage and image detection. Our approach leverages two MoS 2 field-effect transistors (FETs) as selectors and two Al 2 O 3 analog memristors to implement a 2T2R structure comprising two differential 1T1R units sharing the common source electrode. The 1.8 eV bandgap of MoS 2 facilitates low-leakage switching, making it a good selector for memory devices. Additionally, the compatibility of 3D oxide-based memristors with semiconductor fabrication processes enables the vertical stacking of multiple layers of memory cell layers. We use the 2T2R units to experimentally demonstrate 4-bit signed weighting characteristics of artificial synapses and low device variations. Furthermore, we successfully realize a 16 × 16 array incorporating the 2T2R structure, achieving a yield of 91.2% with 15 discrete conductance states. Yield is defined that two sets of 1T1R in each 2T2R can be written and erased normally, and then 15 signed weights of 2T2R units are achieved through differential implementation. This configuration employs pulse width, number, and amplitude to effectively represent analog input signals, resulting in a functional analog-conductance matrix that supports a non-von-Neumann architecture for the storage and computation layer. The testing of the 16 × 16 2T2R array validates the system’s reliability. With high device uniformity, we construct a QuantConvMixer neural network (QCMNN) for the CIFAR-10 recognition task, achieving an accuracy of over 85%, comparable to the ideal accuracy of 89.3%. This work demonstrates the feasibility of monolithic 3D heterogeneous integration of 2D materials with 3D oxides at low temperatures (<200 °C) and provides insights for developing multifunctional customized mega data computing hardware.",
"discussion": "Discussion In summary, we experimentally demonstrated a heterogeneously integrated array using 2D materials and oxide-based memristors as the synaptic kernel. The defect-controlled material transfer and low-temperature fabrication result in high yield (>91.2%). The low-temperature heterogeneous integration approach represents a viable system-level solution for monolithic 3D heterogeneous integration, addressing the computational challenges posed by the explosive growth of data. The core 2T2R unit in the array, mapping weight to the conductance difference between two memristors, facilitates the implementation of a signed weight SCK that is well-suited for CNNs. We constructed a QCMNN for the CIFAR−10 recognition task, achieving an accuracy exceeding 85%. Benchmarking results indicate that the 2T2R-based synaptic kernel outperforms the conventional structures in terms of accuracy and training speed. Moreover, this synaptic kernel has the potential to be extended to other matrix multiplication-based neuromorphic hardware, providing an efficient route to enhance the overall system performance."
} | 2,391 |
31138834 | PMC6538640 | pmc | 155 | {
"abstract": "One of the mechanisms of rapid adaptation or acclimatization to environmental changes in corals is through the dynamics of the composition of their associated endosymbiotic Symbiodiniaceae community. The various species of these dinoflagellates are characterized by different biological properties, some of which can confer stress tolerance to the coral host. Compelling evidence indicates that the corals’ Symbiodiniaceae community can change via shuffling and/or switching but the ecological relevance and the governance of these processes remain elusive. Using a qPCR approach to follow the dynamics of Symbiodiniaceae genera in tagged colonies of three coral species over a 10–18 month period, we detected putative genus-level switching of algal symbionts, with coral species-specific rates of occurrence. However, the dynamics of the corals’ Symbiodiniaceae community composition was not driven by environmental parameters. On the contrary, putative shuffling event were observed in two coral species during anomalous seawater temperatures and nutrient concentrations. Most notably, our results reveal that a suit of permanent Symbiodiniaceae genera is maintained in each colony in a specific range of quantities, giving a unique ‘Symbiodiniaceae signature’ to the host. This individual signature, together with sporadic symbiont switching may account for the intra-specific differences in resistance and resilience observed during environmental anomalies.",
"introduction": "Introduction Dinoflagellate algae from the family Symbiodiniaceae are one of the keystone taxa for coral reef ecosystems. Their importance lies in that in tandem to living free in the environment 1 , 2 , they also form photo-symbiotic associations with corals and several other invertebrates (e.g. 3 , 4 ). Their extraordinary diversity encompasses at least nine major lineages 5 , labelled A to I in the literature some of which were recently erected to genus level 5 : Symbiodinium (clade A), Breviolum (clade B), Cladocopium (clade C) and Durusdinium (clade D). Moreover, each lineage encompasses multiple distinct genetic types 6 , 7 . The acquisition of Symbiodiniaceae by corals is initiated during early life stages via vertical/maternal transfer or through horizontal pathways (reviewed in 8 ). This fine-tuned partnership between the metazoan coral and the dinoflagellates enables the coral-guild (i.e. the holobiont) to thrive in oligotrophic waters (e.g. 9 – 14 ). The collapse of this symbiosis is becoming a common global phenomenon as environmental anomalies are becoming more and more frequent and severe, leading to mass coral bleaching and mortality events globally 15 – 18 . The coral holobiont’s resilience to environmental stressors is in large part based on various genotypes and biological traits 19 . This includes the complex interaction and differential physiological responses of symbiotic Symbiodiniaceae populations that may be composed of one single genus or multiple genera and species 20 , 21 and the host 22 – 25 . A plethora of in situ and ex situ studies have reported highly dynamic coral-Symbiodiniaceae associations in response to environmental changes, contrasts or extremes 26 – 31 . This is a particularly complex mechanism to study, as the free-living Symbiodiniaceae communities may be altered depending on habitat quality 32 , 33 . Shifts that occur over the coral’s ontogeny have been related to physiological states (e.g. diseased vs. healthy, temperature resilience) 34 – 37 . These ecological observations support the idea that flexibility in the coral-Symbiodiniaceae partnership enables the coral holobiont to adapt rapidly to environmental stressors 38 , a paradigm that is encapsulated in the Adaptive Bleaching Hypothesis 39 . In contrast, the analysis of ancient DNA from octocoral species revealed a stable coral host-Symbiodiniaceae association over the last century 40 . Moreover, comparative genomic and transcriptomic analyses showed evidence of a metabolic continuum between the genomes of corals and associated symbionts, supporting the hypothesis that co-evolutionary mechanisms in corals play important roles in the maintenance and adaptation of the symbiosis 41 – 43 . In line with these observations, coral-Symbiodiniaceae associations can exhibit a high degree of specificity 44 – 47 at the level of both genera and species, and there is a lack of evidence for adult coral colonies to form stable partnerships with newly acquired exogenous Symbiodiniaceae 48 , 49 . Therefore, corals have to balance two apparently incompatible traits: Symbiodiniaceae genus fidelity versus Symbiodiniaceae genus flexibility 50 . Deciphering the extent of symbiont change potentially occurring among different coral species through shuffling (intrinsic changes) and/or switching (extrinsic changes) of Symbiodiniaceae communities 39 , requires the consideration of cryptic taxonomic units present at trace levels 21 , 51 – 53 . While an increasing attention has been directed towards these symbiotic cryptic populations since the development of high-sensitivity molecular techniques such as real-time quantitative Polymerase Chain Reaction (qPCR; 21 , 51 ) and Next-Generation Sequencing 54 – 59 , their ecological role in the holobiont’s performance is still poorly understood. To date, studies investigating cryptic Symbiodiniaceae populations have mostly focused on the presumably stress tolerant genus Durusdinium , and the potential for change in the relative abundance of pre-existing Symbiodiniaceae genotypes in response to heat stress 23 , 28 , 31 , 53 , 56 , 60 , 61 . Recent studies also using random sampling surveys revealed that possible de novo acquisition of exogenous Symbiodiniaceae species can occur associated with environmental changes 47 , 56 . In contrast, a long-term survey of diseased A. cytherea corals found that the acquisition of unusual Symbiodiniaceae genera by corals facing environmental stressors is sporadic rather than specific 36 . In the present study we evaluate the potential of Symbiodiniaceae genus dynamics as a rapid adaptive mechanism for the coral holobiont. We study the composition of Symbiodiniaceae communities and their quantitative regulation in tagged colonies of three coral species over a period of 10–18 months, among environmentally contrasting locations of Moorea (French Polynesia).",
"discussion": "Discussion The choice of the qPCR assay employed in the present study was based on its ability to detect Symbiodiniaceae genera in hospite at the level of ≤200 28S copies, which approximates the number of 28S copies in a single algal cell 50 . The combination of this highly sensitive method with a sampling every two months of 56 tagged colonies of three coral host species enabled us to identify putative inferences of both newly acquired Symbiodiniaceae genera (switching) and shifts in the abundance of genera (shuffling) over a period that spans multiple seasons. The overall dataset provides a comprehensive picture of the diversity and quantitative variations of Symbiodiniaceae in various coral hosts over time and in different environments. Random acquisition and short-term maintenance of Symbiodiniaceae The detection of de novo Symbiodiniaceae genera in all three coral species suggests that symbiont switching may be a common natural phenomenon in healthy adult scleractinians. Interestingly, while clade F was presents in the surrounding environment 50 , it was never detected in coral samples, indicative of an active control of symbiont uptake by the host. Previous studies suggested that Symbiodiniaceae switching occurs readily in juvenile stages 23 , while in adults it only happens when the health of the colony is compromised during infected experimental conditions 49 , 65 or in situ 36 . While exogenous symbiont acquisition may be more common in healthy adult corals than previously thought, the rate at which it occurs varies substantially among species. Of the three coral species studied here, P. acuta showed a significantly higher affinity to integrate sporadic clades compared to P. rus and A. cytherea . This interspecific difference may reflect divergent strategies for coping with fluctuating environmental conditions. A higher flexibility in symbiont switching in P. acuta may be a biological mechanism adapted to compensate for the dominance of thermally tolerant but low efficiency genus Durusdinium 66 – 70 in this coral’s microbiome. Supportive of the symbiont switching hypothesis our study also found that six out of twenty (i.e. 30%) Cladocopium -dominated P. rus colonies acquired a background population of Symbiodinium over several consecutive months at some study reef sites but not for all surveyed colonies and without any temporal patterns. These results contradict previous reports 71 – 73 that describe an exclusive association of P. rus with a single ITS2 genotype (C15 of genus Cladocopium ) and could suggest a potential ecological role of the cryptic Symbiodinium in this coral species. Thus, while it is impossible to rule out that some of the de novo genera are merely consumed Symbiodiniaceae that are not in symbiosis with the coral (see 47 ), these can also be the first steps of ‘symbiont switching’ (Fig. 2 ). Although apparently not driven by external factors (within the study period), such random sporadic and/or temporary symbionts switching might strongly impact the fitness and behavior of coral colonies. This has been previously shown for A. cytherea in which the external acquisition of Durusdinium was correlated by the loss of a Vibrio \n spp. 36 . Furthermore, our results suggest that the mechanism of acquisition and the maintenance of non-permanent Symbiodiniaceae genera is largely labile (e.g. change in Pattern Recognition Receptors, or the lack of their maintenance 74 , 75 ), colony dependent and extends the mechanism of symbiont acquisition/selection already observed in symbiont-free offspring 76 . Dynamics of permanent Symbiodiniaceae genera Variations of the abundance of permanent Symbiodiniaceae genera in correlation with environmental changes suggests that the holobiont can rapidly modulate its autotrophic activity by increasing its Symbiodiniaceae densities to counterbalance conditions for limited heterotrophy. For example, one event of putative Symbiodiniaceae shuffling was observed between April and June 2012 in all tagged P. acuta ( Durusdinium ) colonies and at all four reef sites, as well as in P. rus ( Cladocopium ) colonies at the Linareva (undisturbed) and Vaiare (disturbed) sites. This period was characterized by a decrease in sea surface temperature and in nutrient concentrations (i.e. inorganic nitrogen, phosphate, silicates) and an increase in phytoplankton concentration (micro- and pico- fractions). Previous reports (reviewed in 77 ) stated that increased inorganic nitrogen levels boost Symbiodiniaceae densities under high irradiance conditions, as they are normally nitrogen-limited, while increased inorganic phosphate does not have a major effect. Therefore, the negative correlation detected between inorganic phosphate and Symbiodiniaceae densities is unexpected and could be explained by the fact that our study was conducted in situ , or that other parameters not measured here drive symbiont dynamics. Alternatively, this result suggests novel mechanisms currently not fully understood. Only one potential shuffling event was detected in this study that supports the ABH 39 , 61 . Indeed, there were no clear patterns in symbiont dynamics in response to environmental changes, i.e. neither for changes in one particular environmental parameter or the combination of multiple parameters. This is consistent with previous ecological observations showing that even if coral species are able to host a large diversity of symbionts, they will not necessarily activate rapid partnership changes as a response to acute environmental change 36 , 47 , 49 , 65 , 78 . Although changes in the abundance of non-permanent Symbiodiniacea genera were observed in all studied coral species, our results indicate that these acquisitions are random and their maintenance limited, supporting the idea that even under environmental changes, established symbioses are highly stable over time with only transient modifications 47 . This finding has also been highlighted in a recent reciprocal transplant experiment along a depth gradient 78 in which corals that vertically transmit their symbionts reversed after several months to the original symbiotic communities even if these were well-suited to the transplantation depth. Long term stability of permanent clades: coral colonies with unique Symbiodiniaceae signatures The stability and host specificity for particular Symbiodiniaceae genera has been well-established for most coral taxa in both tropical 36 , 44 – 46 , 78 , 79 and high-latitudinal 47 , 80 , 81 regions. In the present study, both coral species that vertically transmit their symbionts (i.e. from parent to offspring) displayed strict host specificity for a single permanent clade ( Durusdinium in P. acuta and Cladocopium in P. rus ), whereas the horizontally transmitting coral species A. cytherea exhibited multiple host clade-patterns ( Durusdinium / Symbiodinium , Symbiodinium , Durusdinium or Cladocopium ). These findings corroborate the hypothesis that the reproductive strategy of corals plays a key role in the establishment of host-symbiont associations, with more specific and stable patterns characteristic for vertically transmitting ‘symbiont specialist’ corals 45 , 82 – 85 , compared with the enhanced flexibility often encountered in horizontally transmitting or ‘symbiont generalist’ corals 80 , 83 , 85 . Most notably, this study also reveals that the coral-Symbiodiniaceae specificity is a trait that is shaped at the colony scale. Permanent Symbiodiniaceae genera are regulated by the host in a non-random manner, within a specific density range, which we coined ‘Symbiodiniaceae signature’. This signature reflects high intraspecific variability in symbiont communities in all three-coral species, which enables distinguishing individual coral colonies by their microbiome. In non-stressful conditions of symbiosis the Symbiodiniaceae densities are regulated in the host at an equilibrium state to sustain both partners in a position of mutual benefit 86 . Therefore, based on the diversity of Symbiodiniaceae signatures observed in the present study, we suggest that the symbiosis equilibrium state might be host dependent. This host regulation can be synergistically affected by environmental filtering. For example, P. acuta and A. cytherea from the undisturbed reef of Linareva had higher Durusdinium quantities in their Symbiodiniaceae signatures. It is potentially one way for the hosts to compensate for the poor contribution of Durusdinium symbionts to the holobiont’s metabolism, as previously described in non-stressful conditions 53 , 68 , 69 . The control of Symbiodiniaceae community composition can be reached through intrinsic and extrinsic factors (reviewed in 87 , 88 ), including direct expulsion 64 , 89 or in hospite degradation of Symbiodiniaceae cells (e.g. necrosis, apoptosis; 87 ). Regardless of the control mechanism involved, each individual coral colony harbors a specific suite and abundance of Symbiodiniaceae genera that likely reflect on their unique plasticity in endosymbiont regulation. Our findings indicate that each holobiont is characterized by a host with a specific Symbiodiniaceae signature of permanent genera and a constitutive flexibility to associate with additional temporary or sporadic genera. In the context of the current model of Symbiodiniaceae-coral winnowing 14 , 90 – 94 , this new theory of Symbiodiniaceae signatures suggests the existence of regulatory mechanisms within the host that enable them to maintain and regulate their permanent Symbiodiniaceae genera homeostasis within a particular range. When identical genera are present but in different quantity ranges, this signature can be considered as distinct amounts of Symbiodiniaceae ‘genomes’ modulating the hologenome ( sensu 95 ), which could account for differences in the host’s behavior 92 , 96 . Additionally, the appearance of sporadic and/or temporary non-permanent genera in the host may contribute to the diversity of the coral hologenome (see 95 ). This diversity may contribute to explaining intraspecific differences in the ecological success of distinct holobionts (Hologenome Dependent Susceptibility) and may constitute an important factor underpinning their resistance or resilience to environmental stressors. As ocean warming progresses, more research on the ecological relevance of particular coral hologenomes, including the Symbiodiniacea signature within and among coral species (e.g. physiological or gene expression responses to multiple stressors), will be critically important to better predict the fate of coral reefs and to assist in the elaboration of effective conservation plans."
} | 4,277 |
31061118 | PMC6535031 | pmc | 156 | {
"abstract": "Significance Coral reefs are threatened by global bleaching, spurring a need to improve upon reef restoration practices. Yet the strong capacity for corals and their symbionts to acclimatize to their local environment has brought into question whether or not corals that are temperature tolerant in one setting will lose that tolerance elsewhere. We show that variation in bleaching resilience among intraspecific colonies is maintained in novel environments for four species, and can be used to construct bleaching-resistant coral nurseries for restoration. By focusing on the host genotype and symbiont genus and its importance in stock selection, we demonstrate a path forward for reef restoration in the face of climate change.",
"discussion": "Discussion Our common garden experiment showed the ability of heat-tolerant corals to produce a heat-tolerant nursery in a transplant site that experienced similar mean temperatures but lower variability in daily fluctuations. These results document that the powerful mechanisms of coral acclimatization and symbiont switching did not fully erode heat tolerance after transplantation. Even after 8 mo of common garden growth, transplants from parents with higher heat tolerance bleached two- to threefold less during the 2015 and 2017 bleaching events than did transplants from parents originally with low heat tolerance. In addition, we monitored bleaching in the original parents on their native reefs in 2015 and show that parent and transplant bleaching was highly correlated across genotypes. Physiological acclimatization is a type of phenotypic plasticity that can powerfully adjust physiology in the face of fluctuating temperature ( 5 ). Especially in marine settings, acclimatization reduces the impact of temperature on physiological shifts such as metabolic rate changes ( 23 ). Experimentally derived acclimation is widely seen in corals: heat tolerance has been seen to shift rapidly after exposure to increased temperature ( 17 , 18 , 24 ). Kingsolver and Huey ( 25 ) suggested that untangling the roles of phenotypic plasticity and evolutionary adaptation will often require field experiments in different environments. Here, we used common garden experiments to show that acclimatization to lower heat tolerance occurred once we transplanted corals to a reef site that experienced fewer hot days, but the relative rank of heat tolerance across individual hosts remained largely the same. Role of Holobiont in Bleaching and Recovery. Physiological response of corals to environment is a complex feature of the coral holobiont, including species ( 26 ), host genetics ( 27 ), host acclimatization ( 10 , 20 , 23 ), symbiont genetics ( 8 , 19 ), symbiont acclimatization, and an extensive microbiome. In our experiments, we tested corals from a common garden setting that were different species, came from different locations, and had different symbionts. There is a good correlation of nubbin bleaching in the nursery with both symbiont type and origin of the parent colony when analyzed separately ( SI Appendix , Fig. S8 ). However, there is a tight association between symbiont type and pool of origin in this location. The HV pool is dominated by symbiont genus Durusdinium , whereas the MV pool has a more even mix of Durusdinium and Cladocopium symbionts ( SI Appendix , Fig. S8 ; see also ref. 28 ). Both GLM and multiple linear regression analyses ( SI Appendix , Tables S1 and S2 ) show that both origin and symbiont genus play a significant role in determining the holobiont’s response. Our goal in these experiments was to assess whether the differences we see in parental colonies are also seen in the nurseries after 8–17 mo of growth, and our results show this to strongly be the case (e.g., see correlations in Fig. 3 ). However, three other features stand out in these analyses. First, colonies from the MV pool or with genus Cladocopium symbionts have a wider range of bleaching than do colonies from the HV pool or Durusdinium symbionts ( SI Appendix , Fig. S8 , ANOVA F = 51.3, P = 10 −29 ), reflecting a larger range of average bleaching scores in MVP than HVP. Because these data were from nubbins exposed to common garden conditions, this variation cannot be ascribed to local variation in environment. Several alternative sources of variation could be: genetic differences among the hosts, hidden variation among Cladocopium symbionts, and long-term impact of environmental history on nursery traits. Genetic differences among host colonies are well known in A. hyacinthus from this population ( 11 , 17 ), and parental genotypes at 114 loci correlate well with nubbin bleaching ( SI Appendix , Fig. S5 ). The parental genetics of the other species are not as well known. Analysis of transcriptome SNPs from symbionts among our parent colonies does indeed show variation within both symbiont genera ( SI Appendix , Fig. S6 , PC axis 2), even within the genus Durusdinium , that might relate to bleaching. Thus, even in this well-known system, the genetic basis for host or symbiont effects on variation in bleaching is just starting to be discovered. Our bleaching data also show significant symbiont switching from genus Cladocopium to genus Durusdinium during the bleaching event, both in parental colonies and also in the common garden nursery clones. This is in contrast to strong persistence of symbiont types after transplantation in previous studies ( 20 ), but is in line with a host of studies in other systems on symbiont dynamics ( 16 ). Our previous transplants never experienced bleaching conditions, and so the differences in our two sets of results may hinge on the impact of bleaching on different symbionts. Third, host genotype also played an important role in recovery. Recovery from bleaching has been shown to be dependent on both biotic and abiotic conditions ( 29 , 30 ). In our system, biotic factors of genotype and species continued to play important roles in rapid recovery. Despite widespread bleaching in 2015, 65% of bleached replicates regained baseline levels of pigmentation within 5 mo. For A. hyacinthus , recovery depended solely on the severity of bleaching (GLM P < 0.01). However, for A. gemmifera and P. damicornis , recovery was also dependent on the genotype (GLM P < 0.01). For example, both genotype GE_M and GE_P had an average bleaching severity across replicates of >90%, but GE_M had 0% recovery while GE_P had 80% recovery. Conservation Engineering: A Path for Coral Nurseries to Survive a Changing Climate. Selecting for resilience remains an important question in the face of climate change. One current approach to select for resilience in corals has been to wait for bleaching events to unveil winners ( 13 ). However, bleaching is known to vary in severity from event to event ( 9 ), making it challenging to observe the total amount of bleaching diversity within a specific system. Instead, there have been alternative calls for predictive assays for coral thermal tolerance ( 17 , 18 , 20 ). To be successful, such predictive tools should work across species, be simple, inexpensive, rapid, and deployable in remote locations. In addition, the practice of selecting natural diversity should also carefully consider potential impacts and identify whether the system can be managed in other ways, for example within adaptation networks ( 31 ). In our study, coral nurseries were prepared for climate change by choosing parental stock using simple proxies of microclimate, origin, and experimental heat response. The proxies yielded results quickly and inexpensively, producing a nursery that resulted in two- to threefold less bleaching. Combined, these results provide managers a path for building resilience before bleaching occurs. This active approach to adding resilience and conservation of diversity for corals, much like ecosystem engineering, combines an understanding of the physiological drivers of environmental stress with the practical results of field trials. Conservation engineering and resilient restoration is part of a trend occurring across natural gradients in forests, grasslands, mangroves, and eelgrass ( 32 – 36 ). In all systems, creating conditions for high resilience will only buy time while global greenhouse gas emissions are reduced and atmospheric content declines. However, for coral, adding resilience by mapping tolerance of individuals already living on a reef may jumpstart use of resilient colonies in future restoration, even while longer-term laboratory-based selective breeding proceeds ( 37 ). More examples are needed, in the Indian, Pacific, and Caribbean basins, of both the degree to which host genotype plays a role in heat tolerance ( 38 ) and the degree to which acclimatization and symbiont switching alters colony phenotypes. If detecting and using heat-tolerance traits is repeatable in other locations, this provides a promising pathway for local restoration efforts to utilize thermally resilient coral. That these thermally resilient corals may be within their own management areas is a hopeful step forward for reefs in a changing climate."
} | 2,289 |
33568484 | PMC7875522 | pmc | 157 | {
"abstract": "Oxygen reduction in a cable bacterium is condensed in a few terminal cells, while the rest of the filament lives oxygen free.",
"introduction": "INTRODUCTION The rise of molecular oxygen O 2 in the atmosphere during the Great Oxidation Event (2.4 to 2.0 billion years ago) posed a grand challenge to life ( 1 ). Some microorganisms evolved strategies to cope with the reactive oxygen species (ROS) superoxide, hydrogen peroxide, and hydroxyl radical that form during oxygen reduction and react with various cellular components causing detrimental effects and undermining cell viability ( 2 – 4 ). Other microorganisms, including the emerging eukaryotes, evolved to take advantage of oxygen as a very potent electron acceptor. Today, anoxia still prevails in wet environments such as sediments, biofilm, guts, and feces because of the poor solubility and slow diffusion of oxygen in water ( 5 ). Over the course of evolution, microorganisms have developed different molecular and behavioral strategies to live at the oxic-anoxic interface, where they compete for electron donors released from anaerobic processes and jointly condense the activity to a very narrow zone with low oxygen concentrations ( 5 , 6 ). A peculiar strategy circumventing the crowd at the interface is put in place by the filamentous cable bacteria. They exploit multicellularity and electric wires made of a yet unknown conductive material to spatially bridge anoxic and oxic environments with long-distance electron transport using sulfide as an electron donor ( 7 , 8 ). By means of gliding motility with a speed of up to 2.2 μm s −1 , cable bacteria have been observed to maintain their distinctive interface-spanning position with some segments curled around the oxic-anoxic interfaces and other segments connecting straight back to the sulfide source ( 9 ). With this partial exposure to oxygen, they do not align with the conventional thinking of organisms, metabolizing either with or without the presence of oxygen at any given time. Specifically, the unique allocation of the coupled half reactions of the basic redox processes to different cells raised the question of how energy conservation and subsequent biomass accrual are distributed among the cells. Genomic inspection found that cable bacteria do not have the conventional membrane-bound terminal cytochrome C oxidase linking oxygen reduction to energy conservation ( 10 ). Hence, it was proposed that they had optimized oxygen reduction for high speed in a few cells exposed to oxygen, while energy conservation and assimilation could be confined to the sulfide-oxidizing cells in the anoxic zone, free from the oxidative threat. A study of the succession of cable bacteria in marine sediment found that only 10% of the total cable bacteria biomass resided in the oxic zone of marine sediment ( 11 ). A conservative estimate based on the current density and the assumption that all cells in the oxic zone were equally active led to a cell-specific oxygen consumption rate 4 to 14 times higher than in other sulfide oxidizers of comparable cell size ( 11 ). Recently, it was observed that substantial biomass accrual was only seen in filament fragments retrieved from the anoxic zone after 19 to 24 hours of incubation with an amino acid tracer ( 10 ) or 15 N and 13 C tracers ( 12 ). Furthermore, cyclic voltammetry has showed substantial capacity to catalyze electrochemical oxygen reduction ( 12 ). The aim of the present study was to use an individual-based approach to test whether cable bacteria indeed have evolved a strategy to “get the best and avoid the rest” of oxygen metabolism by means of electric currents, extremely compacted consumption, and controlled positioning across oxic-anoxic interfaces. We directly observed live, cable bacteria at oxic-anoxic interfaces in microchambers with a clonal enrichment culture and obtained individual and, in turn, cell-specific oxygen consumption rates from their visible distortions of oxygen gradients.",
"discussion": "RESULTS AND DISCUSSION Individual oxygen consumption rates obtained from visible distortions of the oxic-anoxic interface Custom-made glass chambers (“trench slides”; Fig. 1A ) were designed to let cable bacteria migrate from anoxic sediment out into an observation window to meet oxygen near the air-exposed edge. As pure cultures of cable bacteria are not available, the slides were inoculated with a sediment-based enrichment of Candidatus Electronema sp. GS, a freshwater species of cable bacteria with a medium thickness of 1 to 2 μm ( 13 , 14 , 10 ). Various kinds of swimming bacteria rapidly formed a distinct, visible veil 0.3 to 2.6 mm from the edge ( Fig. 1 ). Linear oxygen concentration gradients from air saturation at the edge to zero in the veil, as visualized in two dimensions by planar optode–based chemical imaging ( 15 , 16 ), documented that the veil bacteria were accurate markers of the oxic-anoxic interface and consumed all the intruding oxygen within a zone of 10 μm and at oxygen concentrations below 5 μM ( Fig. 1, D and E ). The linearity of the oxygen gradient also demonstrated that the transport coefficient was not enhanced near the veil as would have been the case if advection caused by the veil bacteria was important here ( 17 ). The oxygen consumption of veils is controlled by the steady, diffusive flux of dissolved, reduced compounds from decomposition processes in the anoxic zone ( 5 ). Oxygen consumption by single cable bacteria crossing the veil generated distinct distortions as the veil-forming, swimming bacteria followed the local decrease in oxygen ( Fig. 1, A to C ). The geometry of the microchamber, knowing the oxygen diffusion coefficient, allowed the calculation of oxygen fluxes and, hence, of the oxygen consumption of the veil, as well as the oxygen consumption rate of the cable bacteria making distortions ( Fig. 1 and fig. S3). Fig. 1 Visualization of oxic-anoxic interfaces and the distortions caused by oxygen consumption of individual cable bacteria. ( A ) Cartoon of a microscope wet mount with a sediment-filled, central trench and a dynamic veil of bacteria (blue line) marking the oxic-anoxic interface. ( B ) Dark-field image of the bacterial veil situated between the edge of the coverslip and the sediment. ( C ) Close-up of a veil distortion generated by the oxygen consumption of a single cable bacterium pointed out by blue arrows. The light, partially out-of-focus, spots in (B) and (C) are scattered particles. ( D ) Overlay of a dark-field image showing the veil in a trench slide and the oxygen gradient as measured by the planar optode [see details in Materials and Methods; see fig. S2 (B and C) for extended and nonoverlaid images]. ( E ) Oxygen profile as obtained from the planar optode, showing that oxygen is consumed by the veil, which is therefore considered the oxic-anoxic interface. See also figs. S2 and S3. Not every cable bacterium actively crossing the veil showed visible signs of oxygen consumption (movie S1), and the recorded rates varied greatly with 68 pmol O 2 day −1 as the maximum (table S1). The individual variation makes sense, as electron flow in a cable bacterium depends on many factors, including length, success in synchronously reaching electron sources and oxygen, and the extent of damage to the internal electric wires ( 18 , 19 ). To record the metabolic rate of individual bacteria in real time, and even in a natural gradient environment, is exceptional. To the best of our knowledge, only one previous study achieved this type of measurement by means of oxygen microsensors on a single, giant (diameter, 220 μm) Thiomargarita namibiensis cell ( 20 ). The hitherto highest biomass-specific oxygen consumption rate of a prokaryote reported in literature belongs to Desulfovibrio termitidis . When incubated with hydrogen and 20 μM oxygen concentration, D. termitidis showed an oxygen consumption rate of 1570 nmol O 2 mg protein −1 min −1 ( 2 , 21 ). For cable bacteria, an even higher rate of 2200 nmol O 2 mg protein −1 min −1 for the oxygen-exposed part was calculated using standard factors to convert from the individual rates and the biovolume measures (see Materials and Methods). D. termitidis has both high- and low-affinity cytochrome c oxidases (genome accession number: GCA_000504305.1), which are involved in the energy conservation using hydrogen as an electron donor. Cable bacteria, however, do not have any canonical terminal oxidases, including those present in Desulfobulbus propionicus from the same family ( 10 , 13 ). In their absence, it has been proposed that a unique cytochrome-truncated hemoglobin fusion protein might be the catalyst for a condensed periplasmic oxygen reduction ( 13 ). The presence of conductive fibers in the periplasm ( 8 , 22 , 23 ) indirectly backs this hypothesis, as electrons would be unloaded in the periplasm onto the fusion protein, which has no apparent anchorage to the cytoplasmic membrane and therefore no means of energy conservation; in addition, oxygen would thus be quickly reduced in direct proximity of the fiber. However, the complete oxygen reduction pathway in cable bacteria is not obvious. Whether and how the truncated hemoglobins, the cytochrome moiety, or other enzymes contribute to the oxygen reduction to water requires elucidation. Cell-specific oxygen consumption By measuring the oxic part of the filament and assuming 3 μm as a conservative estimate of the cell length ( 13 ), the individual rate could be readily converted into an average per-cell oxygen consumption rate estimate (fig. S5A and table S1). Despite expectations, no statistically significant correlations between the cell-specific oxygen consumption rate estimates and the lengths of parts on either side of the oxic-anoxic interface were found ( Fig. 2 ). However, the large difference of the mean and median rate, 500 and 200 fmol O 2 cell −1 day −1 , respectively, reflected a clustering with a large cluster below 250 fmol O 2 cell −1 day −1 and a smaller cluster between 600 and 2140 fmol O 2 cell −1 day −1 (fig. S5). It is tempting to suggest that the low-rate cluster represents cable bacteria in a situation where the electron flow was limited on the donor side, while the high rates indicate oxygen limitation of the current and hence represent maximum cell-specific capacity (fig. S5, A and B). In a recent cyclic voltammetry experiment, an oxygen consumption rate of 6134 fmol O 2 cell −1 day −1 was estimated ( 12 ). The marine cable bacteria used in the latter had more than three times larger cell volumes (3- to 5-μm diameter, as opposed to the Candidatus ( Ca .) Electronema cells used in this study with an average diameter of 1.42 μm). Taking into account the cell volume difference, the high rates between 600 and 2140 fmol O 2 cell −1 day −1 obtained in this study indicate that the rates obtained from cyclic voltammetry are indeed realized in live cable bacteria with natural, internal currents. Fig. 2 Scatterplot of the cell-specific oxygen consumption rate versus the ratio between the lengths of the anoxic and oxic segment of cable bacteria. OCR, oxygen consumption rate. Behavior of cable bacteria at the oxic-anoxic interface Cable bacteria that had positioned themselves across the oxic-anoxic interface in the trench slides only had 8.1 ± 6.4% of their bodies protruding into the oxic zone (fig. S5B), meaning that only few cells were responsible for the oxygen reduction of the whole cable bacterium. This is in line with previous measures of 10% of the cable bacteria biomass in the oxic zone of marine sediment ( 11 ). The ratio between the distances from the veil to the furthest oxic cell and to the edge, respectively, was used to calculate the maximum oxygen concentration that cable bacteria experienced, considering a linear oxygen gradient as confirmed by planar optode–based imaging. Longer cable bacteria, extending far away from the oxic zone toward the sediment and with a steady current, tended to keep cells below 14% air saturation (movie S1), indicating a phobic response to oxygen with a threshold a little higher than observed in another sulfide-oxidizing, gliding filamentous bacterium, Beggiatoa spp. [5% air saturation; ( 24 , 25 )]. One cable bacterium, observed for more than 12 hours, kept the same position across the oxic-anoxic interface, dynamically readjusting to small concentration changes (movie S2). Generally, cable bacteria inverted the direction of motion for short-term readjustments to the oxygen concentration every 60 s. Shorter cable bacteria that transiently dipped in the oxic zone without consuming oxygen showed similar short-term readjustments (movie S1), suggesting that the taxis is a response to oxygen rather than to changes in potential electron delivery as reflected in wire electric potentials. This aligns with the large cluster of observations of cells running far below the capacity as discussed above. Motility controlled so tightly by oxygen being experienced by only the very end of the filament raises questions on how distant segments are informed when to reverse gliding. In the filamentous cyanobacterium Phormidium uncinatum , lateral transmission of transmembrane potential Δμ H+ from a transiently illuminated segment of the trichome initiates motility at the other end of the trichome at least 2 mm away and in a time frame that rules out any role of signaling by diffusion of molecules ( 26 – 28 ). Similar rapid responses over millimeters distance with filamentous Beggiatoa spp. also calls for electrical rather than chemical signaling ( 24 ). Cable bacteria may instead use global changes in the potential of the internal wires as a second type of electrical signaling coordinating the behavior of a multicellular, filamentous bacterium ( 14 , 29 ). By keeping less than a tenth of the filament in the oxic zone (table S1) ( 11 ), the few oxic cells must flare off a considerable electron load to oxygen, which should result in the production of ROS. The high expression of proteins such as catalases, superoxide reductase, rubrerythrin, and GroEL/ES chaperonins in a previous study suggest that cable bacteria are indeed exposed to substantial oxidative stress ( 10 ). ROS have detrimental effects on viability, growth, and cell motility, as many enzymes responsible for the anabolism are inhibited ( 2 , 3 , 6 ). Despite expression of superoxide reductase and rubrerythrin that can dispose of ROS ( 10 ), apparently irreversible structural changes of oxygen-exposed cells were observed during long-term observations (movie S2). Even if these cells were dying, this loss might, from an evolutionary point of view, still be worth the price ( 10 , 12 ), as >90% of the filament lies in an oxygen-free environment. This strategy allows cable bacteria to harvest the power of a superior electron acceptor such as oxygen while, at the same time, circumventing the negative consequences that come with its reduction. This study confirms that cable bacteria’s ingenious adaptation to life at oxic-anoxic interfaces encompasses electricity, putatively in multiple ways, to communicate among distant cells and assign the troublesome reduction of oxygen to a few cells with possibly the most intense oxygen consumption in biology. More live observations of individual cable bacteria may also help to understand other essential features of this unique form of life, such as management of energy storages, allocation of growth, and competition among cable bacteria."
} | 3,891 |
35706373 | PMC9539982 | pmc | 158 | {
"abstract": "Summary \n How mycoheterotrophic plants that obtain carbon and soil nutrients from fungi are integrated in the usually mutualistic arbuscular mycorrhizal networks is unknown. Here, we compare autotrophic and mycoheterotrophic plant associations with arbuscular mycorrhizal fungi and use network analysis to investigate interaction preferences in the tripartite network. We sequenced root tips from autotrophic and mycoheterotrophic plants to assemble the combined tripartite network between autotrophic plants, mycorrhizal fungi and mycoheterotrophic plants. We compared plant–fungi interactions between mutualistic and antagonist networks, and searched for a diamond‐like module defined by a mycoheterotrophic and an autotrophic plant interacting with the same pair of fungi to investigate whether pairs of fungi simultaneously linked to plant species from each interaction type were overrepresented throughout the network. Mycoheterotrophic plants as a group interacted with a subset of the fungi detected in autotrophs but are indirectly linked to all autotrophic plants, and fungi with a high overlap in autotrophic partners tended to interact with a similar set of mycoheterotrophs. Moreover, pairs of fungi sharing the same mycoheterotrophic and autotrophic plant species are overrepresented in the network. We hypothesise that the maintenance of antagonistic interactions is maximised by targeting well linked mutualistic fungi, thereby minimising the risk of carbon supply shortages.",
"conclusion": "Conclusions and future perspectives To our knowledge, our study is the first to assess how mycoheterotrophic plant species are embedded in mutualistic mycorrhizal networks. We found that mycoheterotrophic plants as a group interacted with a subset of the available fungal partners, and generally targeted fungi that were well connected to autotrophic plants. Although mycoheterotrophic species show overlap in their fungal associations, we found that they were indirectly linked to different sets of autotrophic plants, suggesting a potential mechanism to avoid competition by preferentially relying on different carbon sources (Gomes et al ., 2017b ). The phylogenetic relationships between the fungi, probably a proxy for fungal traits, had a significant influence on these nonrandom tripartite interactions. Therefore, we concluded that the persistence of mycoheterotrophs in AM networks is dependent on particular well connected ‘keystone’ mycorrhizal fungi, which provide the mycoheterotrophs with carbon from a wide range of plants. Our observations that fungi connected mutualistic and antagonistic networks in a nonrandom fashion and that well connect fungal nodes in AM networks were more prone to be targeted by mycoheterotrophs, are similar to those of Sauve et al . ( 2016 ) for a plant–pollinator–herbivore network when considering binary interactions. Further research is needed to assess whether this is a general feature of interactions within species‐rich communities, also when taking interaction strength into account. Our study emphasises the raising of awareness of considering multiple interaction types simultaneously (e.g. antagonistic and mutualistic) to deepen our understanding of complex biodiversity patterns (Losapio et al ., 2021 ). In contrast with ectomycorrhizal symbiosis, for which it has been known for decades that several plant species are able to combine photosynthesis and carbon uptake from fungi, in a strategy termed ‘partial mycoheterotrophy’ (Selosse & Roy, 2009 ), only recently this mode of life has been suggested to be widespread within the AM symbiosis. Giesemann et al . ( 2021 ) have shown that many photosynthetic understory plants are potentially able to take up carbon from associated AM fungi. Future work will enlighten us on whether these partially mycoheterotrophic plants rely on similar sets of fungi and rely on similar interaction patterns within the mycorrhizal network as the fully mycoheterotrophic plants in in the present study.",
"introduction": "Introduction Since the early history of life, interspecific mutualisms have been paramount in the functioning of ecosystems (Thompson, 2005 ; Bascompte & Jordano, 2014 ). Mutualisms can form complex networks of interdependence between dozens or even hundreds of species. A prime example of this ‘web of life’ is the 450‐million‐yr‐old mutualism between the majority of land plants and arbuscular mycorrhizal (AM) fungi (Strullu‐Derrien et al ., 2018 ). In this interaction, plants supply the AM Glomeromycotina and Mucoromycotina fungi with carbon, essential for fungal survival and growth. In return, the fungi provide their host plants with mineral nutrients and water from the soil (Smith & Read, 2008 ; Bidartondo et al ., 2011 ). One of the key characteristics of the AM interaction is its low interaction specificity: a mycorrhizal plant typically associates simultaneously with multiple fungi and a mycorrhizal fungus often associates simultaneously with multiple plants (Lee et al ., 2013 ). This creates complex underground networks in which plants of different species are indirectly linked through shared AM fungi (Toju et al ., 2015 ; Chen et al ., 2017 ). Despite this low specificity, there is evidence that networks of plants and AM fungi do not assemble randomly; instead the interactions can be affected by plant functional group (Davison et al ., 2011 ; Sepp et al ., 2019 ), and plant or fungal evolutionary relationships (Montesinos‐Navarro et al ., 2015 ; Chen et al ., 2017 ). Considerable progress has been made in recent years in dissecting the exchange of resources between plants and AM fungi and its regulation. Experimental work on low‐diversity systems has demonstrated that control is bidirectional, and partners offering the best rate of exchange are rewarded, suggesting that AM networks consist of ‘fair trade’ interactions of carbon‐for‐nutrient exchange (Bever et al ., 2009 ; Kiers et al ., 2011 ). However, the importance of reciprocally regulated resource exchange is questioned, as mycorrhizas also affect plant health, interactions with other soil organisms, host‐defence reactions and suppression of nonmycorrhizal competitor plants (Walder & Van Der Heijden, 2015 ). Also, strictly reciprocal regulation of carbon‐for‐nutrients exchange does not seem to apply to all AM interactions. For example, some exceptional plants behave as ‘cheaters’ (Selosse & Rousset, 2011 ; Walder & Van Der Heijden, 2015 ): mycoheterotrophic plants obtain carbon from root‐associated fungi and some species have replaced photosynthesis with carbon uptake from AM fungi (Leake, 1994 ; Merckx, 2013 ). Although the mechanism underpinning carbon transfer from AM fungi to mycoheterotrophic plants remains unclear, mycoheterotrophic plants are often considered cheaters of the mycorrhizal symbiosis because they have evolved from mutualistic ancestors (Merckx et al ., 2013 ) and exploit the AM symbiosis for soil nutrients and carbon without reciprocating, to our current knowledge (Selosse & Rousset, 2011 ), or without apparently being sanctioned by the fungal partners (Walder & Van Der Heijden, 2015 ). Moreover, it has been suggested that mycoheterotrophic plants may display a truly biotrophic parasitic mode, digesting the fungus colonising their roots (Imhof et al ., 2013 ). Despite these observations, it is unknown whether mycoheterotrophic plants have a true negative effect on their associated fungi, and we cannot rule out if they provide cryptic benefits to their symbionts. Within obligate mutualisms, the critical barrier to mutualism breakdown and to the evolutionary stability of the resulting cheater species is thought to be a requirement for three‐species coexistence: a cheater plant relies on a mutualistic partner – a mycorrhizal fungus – which simultaneously interacts with an autotrophic plant (Pellmyr & Leebens‐Mack, 1999 ). In species‐rich mutualisms, such as the AM symbiosis, for which multispecies coexistence is the rule, a high potential for the occurrence of these tripartite linkages is expected (Merckx & Bidartondo, 2008 ). Indeed, while evolution of cheating in specialised obligate mutualisms is relatively rare (Sachs & Simms, 2006 ), cheating in the AM symbiosis has evolved in more than a dozen of plant clades, including over 250 species that together occur in nearly all tropical and subtropical forests (Merckx, 2013 ; Gomes et al ., 2019a ). Previous work has shown that mycoheterotrophic plants target a subset of the mycorrhizal fungi available in the local community (Bidartondo et al ., 2002 ; Gomes et al ., 2017a ; Sheldrake et al ., 2017 ). Their associated fungal communities can vary in specificity: while many families of mycoheterotrophic plant species associate with fungi that are clustered in the Glomeromycotina phylogeny, large differences in the number of associated fungi and phylogenetic specificity are observed between species of mycoheterotrophic plants (Merckx et al ., 2012 ). This specificity can be shaped by the competitive interactions of the plants. Gomes et al . ( 2017b ) showed that, among mycoheterotrophs, plant species usually associate with more distantly related fungi than expected by chance, and in communities of co‐occurring mycoheterotrophic species, the phylogenetic diversity of the associated fungi increases with the extent of fungal overlap between the mycoheterotrophic species. This pattern may respond to an ecological mechanism driven by maximising co‐occurrence and avoiding competitive exclusion among mycoheterotrophic plants. However, whether partner choice of mycoheterotrophs is affected by the mutualistic interactions of their associated fungi with autotrophic plants is currently unknown. Here, we hypothesise that mycoheterotrophic plants preferentially associate with ‘keystone’ (Mills & Doak, 1993 ) fungi that are well connected to many different autotrophic plants simultaneously, as these fungi are potentially more resilient to perturbations (Bascompte & Jordano, 2007 ) and may be the most reliable source of carbon (Waterman et al ., 2013 ). In addition, as fungal traits play an important role in arbuscular mycorrhizal interactions – phylogenetically related AM fungi (assumed to have similar functional traits), preferentially interact with similar plant species (Chagnon et al ., 2015 ) – we hypothesise that if partner selection in tripartite networks is trait driven we will be able to detect the influence of the phylogenetic relationships of the fungi. We tested these hypotheses on a combined tripartite mycorrhizal network of co‐occurring mycoheterotrophic and surrounding autotrophic plants linked by shared AM fungi compiled by high‐throughput DNA sequencing. To place our results in the context of recent work on comparing different types of ecological interaction networks (Melián et al ., 2009 ; Fontaine et al ., 2011 ; Sauve et al ., 2013 ), we consider autotrophic plants to establish mutualistic interactions with AM fungi and mycoheterotrophic plants to form antagonistic interactions with AM fungi, although it remains unclear whether mycoheterotrophs have a negative impact on their associated fungi.",
"discussion": "Discussion We found that mycoheterotrophic plants as a group target with a subset of the fungi that are potentially available, however this subset of fungi is associated with all autotrophic plants detected in this study. The results of the network analysis indicate that this pattern is produced by a preference of the mycoheterotrophs for well connected fungi. Therefore, despite associating only with a subset of the local pool of fungi, the mycoheterotrophs indirectly reach a wide range of autotrophic plants through their shared fungi, potentially obtaining carbon from any of the autotrophic plants at the study site. Fungi not detected in the roots of mycoheterotrophs were generally connected to few autotrophic plants. Within the fungi that are shared between the mutualistic and antagonistic networks, we detected a significant ecological symmetry between the mutualistic and antagonistic interactions of the fungi: pairs of fungi that interact with overlapping sets of autotrophic plants also interact with overlapping mycoheterotrophic plants. The network analysis indicated that this pattern occurs more often than expected by chance. Based on the highest fungal overlap between each mycoheterotrophic species with different subsets of highly connected autotrophic species, this pattern appears to be driven by a further preference of mycoheterotrophs for fungi that are well linked to specific mutualistic plants. These autotrophic plants are the ultimate sources of the carbon that mycoheterotrophic plants take up from the fungi shared between autotrophs and mycoheterotrophs. Therefore, we suggest that the observed pattern reflects a strategy in which the maintenance of antagonistic interactions is maximised by targeting well linked fungi, thereby minimising the risk of carbon supply shortages. Fungal preferences of mycoheterotrophic plants We found that plant species identity had a significant influence on the fungal community composition, regardless of the plant type, which indicates that these communities are nonrandom subsets of the local fungal taxon pool. This supports previous evidence that co‐occurring plant species showed differences in selectivity towards available AM fungi (Davison et al ., 2011 ). Mycoheterotrophic plant species are known to select particular groups of fungi, often a narrower range than the surrounding autotrophic plants (Bidartondo et al ., 2002 ; Gomes et al ., 2017a , b ). Here we observed that five co‐occurring mycoheterotrophic plant species collectively associate with approximately half of the available fungal taxa (Fig. 4 ). Antagonistic interactions can therefore be supported by a relatively wide array of AM taxa, as shown previously (Merckx et al ., 2012 ; Gomes et al ., 2017b ; Sheldrake et al ., 2017 ). Although fungi from three different fungal families (Glomeraceae, Acaulosporaceae and Gigasporaceae) were detected in the roots of mycoheterotrophic plants, a clear preference for Glomeraceae taxa, and Rhizophagus irregularis relatives in particular, was observed. The taxa of this clade were the most frequently encountered in the roots of the autotrophic plants as well (Fig. 4 ). Rhizophagus contains some of the most globally widespread and common AM fungi (Kivlin et al ., 2011 ; Moora et al ., 2011 ; Davison et al ., 2015 ; Gomes et al ., 2018 ) although this finding is mostly derived from studies in temperate areas. Our results indicate that the tropical rainforest offers no exception to this pattern. Glomeraceae are usually not only the most dominant clade in natural AM communities, often accounting for c . 70% of all species (Montesinos‐Navarro et al ., 2012 ), but they also have been found consistently to include the most generalist AM fungi in other network studies (Montesinos‐Navarro et al ., 2012 ; Chagnon et al ., 2015 ; Chen et al ., 2017 ). The ability to interact with many autotrophic plant species can be a potential reason for why mycoheterotrophic plants generally target Glomeraceae fungi (Merckx et al ., 2012 ; Renny et al ., 2017 ). Ecological theory predicts that generalist species tend to have large distribution ranges (Brown, 1984 ) and, consequently, are less vulnerable to (local) extinction than specialised species (Schleuning et al ., 2016 ). Therefore, associations with generalist fungi may be advantageous for the evolutionary persistence of mycoheterotrophs. Furthermore, associations with multiple autotrophic plant partners may increase fungal resilience to disturbance, while mediating temporal fluctuations in carbon flow and interaction dynamics (Bennett et al ., 2013 ). Therefore, this would guarantee a continuous carbon supply to the entire network without the pronounced negative effects, even in the presence of antagonists. In addition, in the context of mycorrhizal fungi, which can be linked to different plant species simultaneously (Montesinos‐Navarro et al ., 2012 ), generalist fungi are therefore likely to be more reliable carbon sources for mycoheterotrophs. An alternative and perhaps not mutually exclusive explanation for why mycoheterotrophic plants preferentially target well connected fungi may be that these fungi are less effective in detecting and excluding nonphotosynthetic plant partners (Bruns et al ., 2002 ; Egger & Hibbett, 2004 ; Bidartondo, 2005 ; Walder & Van Der Heijden, 2015 ). In contrast with our local‐scale study, Perez‐Lamarque et al . ( 2020 ), who performed a global‐scale study on AM mycoheterotrophic plants, reported that these plants tended to interact with (globally) specialised fungi. Although their results may have been influenced by the limited availability of global data – data of less than c . 0.2% of all autotrophic AM plants were available – it is possible that the most‐connected fungi in our study are less well connected in other habitats. In this case, the pattern of global‐scale reciprocal specialisation between mycoheterotrophs and AM fungi might be influenced by the specific local environmental conditions under which mycoheterotrophy occurs, such as low soil fertility (Gomes et al ., 2019b ). Fungal links between mutualistic and antagonistic networks We detected that pairs of fungi that interact with similar sets of autotrophic plants share links with overlapping mycoheterotrophic plants. Therefore, there is a high level of interaction symmetry between mutualistic and antagonistic mycorrhizal networks. Also, we measured a significant influence of the fungal phylogenetic relationships on both the mutualistic and antagonistic interactions, showing that closely related fungi interact with similar autotrophic and mycoheterotrophic plants respectively. Because biotic interactions are mediated by functional traits, and most functional traits are evolutionarily conserved, a shared evolutionary history of fungi can serve as a proxy for functional similarity (Chagnon et al ., 2015 ). We therefore hypothesise that both mutualistic and antagonistic interactions are shaped partially by evolutionary conserved functional traits of the fungi. In this case, the apparent preference for members of the Glomeraceae family may indicate a higher reliance on ruderal AM fungi (Chagnon et al ., 2013 ) for both autotrophic and mycoheterotrophic plants. Moreover, multiple clades within this family seem to be preferentially associated with mycoheterotrophic plants, which could reflect more fine discrimination of traits that we cannot discern with the current knowledge on AM fungal strategies. In addition, the network analysis indicated that pairs of fungi shared a mycoheterotrophic and an autotrophic plant more often than expected by chance. This analysis solely indicates that the diamond‐shaped module is overrepresented in the empirical network (Fig. 2 ), without reference to species degree or species identity. However, our results support the idea that the observed pattern is driven by the tendency of mycoheterotrophic plants to target fungi that are well linked to autotrophic plants (Fig. 1b ). The autotrophic plants with the highest fungal overlap in relation to the mycoheterotrophic plants are among those with the highest ranked degree and phylogenetic species variability from the pool of detected autotrophic plant species (Fig. 3 ). Moreover, fungi with the highest number of interactions in the mutualistic network are also among the best connected fungi in the antagonistic network (Fig. 4 ). Our findings therefore reveal that mycoheterotrophic plants preferentially associate with fungi that are simultaneously linked to a wide range of autotrophic plants. Targeting well connected fungi in the mutualistic network could be a strategy for mycoheterotrophic plants to increase their resistant and resilient facing perturbations. Although many mycoheterotrophic plants share a large number of fungi with Aspidosperma sp., which has the highest normalised degree among the autotrophic plants, mycoheterotrophic plant species also indirectly associate with nonoverlapping sets of autotrophic plants, as indicated by their divergent positions in the plant–plant interaction network (Fig. 2b ). Therefore, building on the hypothesis that mycoheterotrophic plants maximise their coexistence by increasing the phylogenetic diversity of the AM fungi with which they associate as the overlap among co‐occurring species increases (Gomes et al ., 2017b ), the present study suggests that the differential preference of mycoheterotrophic species for connections with nonoverlapping autotrophic species may contribute to competition avoidance among mycoheterotrophic plants. Potential sampling biases Considering that rainforests are species‐rich ecosystems (ter Steege et al ., 2013 ), it is likely that, despite our efforts, the sampling of the belowground diversity of AM fungi, and their plant partners remained incomplete, in part because both plant identity and fungal communities of only 35% of autotrophic plants samples could be obtained. Therefore, not all autotrophic plant species present at the sites were included in the network and the reported patterns must be interpreted with caution. However, the large fungal overlap between the mutualistic and antagonistic networks suggested that it is unlikely that exclusive fungal connections of mycoheterotrophs to nonrepresented autotrophic plants are prevalent. Furthermore, the representation of roots from mycoheterotrophic and autotrophic plants was necessarily imbalanced, as whole and partial root systems, respectively, were collected. This probably has an impact on the completeness of the fungal communities of the autotrophic plant species, and made the use of read abundances to estimate interaction strengths impossible. Moreover, the choice of primer set can introduce biases in the discovery of fungal diversity (Lekberg et al ., 2018 ). We used the fITS7/ITS4 primer pair to characterise the fungal communities associated with the plants in our study, and found that mycoheterotrophic plants were associated primarily with fungi in the Glomeraceae family, which agrees with previous studies that used the SSU region (Merckx et al ., 2012 ; Renny et al ., 2017 ). Glomeraceae fungi have also been revealed to predominate in roots of autotrophic plant species (Davison et al ., 2015 ). Importantly, previous studies have highlighted the importance of sampling intensity (i.e. the number of possible interactions per node, which directly impacts the normalised degree per species) on network metrics (Blüthgen et al ., 2007 ; Dormann et al ., 2009 ). As the plant species in our study are represented by different numbers of samples, we assessed carefully any potential impacts of sampling bias on the results of the network analysis using multiple strategies. First, we randomised the mycoheterotrophic and autotrophic matrices to build the null models separately as the ratio of autotrophic and mycoheterotrophic species was only 4 : 1. Second, we calculated the number of modules over 100 rarefied matrices (which greatly reduced the number of discovered fungi for the mycoheterotrophic plants) and null models showed the presence of a network motif in 100% of the cases. Third, as the number of samples per species was also unequal across plant species, we repeated the module search procedure with random resampling of three individual samples per species and while discarding species for which fewer than three samples were available (even though this led to an unrealistic proportion of mycoheterotrophic to autotrophic plant species). This approach showed that the overrepresentation of the module in the network with all fungi was not influenced by unequal sampling across plant species. For the network with overlapping fungi, the empirical overrepresentation of the diamond‐shaped module was potentially influenced by unequal sampling, therefore we also verified the consistency of this result across multiple rarefactions depths by the use of multiple rarefied matrices at each depth (Gotelli & Colwell, 2001 ). While we acknowledge that the completeness of sampling in our study is not ideal – a typical challenge for any study on mycorrhizal diversity – we also considered several factors that allowed us to separate the statistical patterns in our data from the influence of sampling effort, both in terms of the plant species detected in the sampled roots, and in their potentially incomplete number of fungal associations. Whether our results are potentially influenced by spatial patterns in the distribution of autotrophs, mycoheterotrophs and fungi remains to be determined. Conclusions and future perspectives To our knowledge, our study is the first to assess how mycoheterotrophic plant species are embedded in mutualistic mycorrhizal networks. We found that mycoheterotrophic plants as a group interacted with a subset of the available fungal partners, and generally targeted fungi that were well connected to autotrophic plants. Although mycoheterotrophic species show overlap in their fungal associations, we found that they were indirectly linked to different sets of autotrophic plants, suggesting a potential mechanism to avoid competition by preferentially relying on different carbon sources (Gomes et al ., 2017b ). The phylogenetic relationships between the fungi, probably a proxy for fungal traits, had a significant influence on these nonrandom tripartite interactions. Therefore, we concluded that the persistence of mycoheterotrophs in AM networks is dependent on particular well connected ‘keystone’ mycorrhizal fungi, which provide the mycoheterotrophs with carbon from a wide range of plants. Our observations that fungi connected mutualistic and antagonistic networks in a nonrandom fashion and that well connect fungal nodes in AM networks were more prone to be targeted by mycoheterotrophs, are similar to those of Sauve et al . ( 2016 ) for a plant–pollinator–herbivore network when considering binary interactions. Further research is needed to assess whether this is a general feature of interactions within species‐rich communities, also when taking interaction strength into account. Our study emphasises the raising of awareness of considering multiple interaction types simultaneously (e.g. antagonistic and mutualistic) to deepen our understanding of complex biodiversity patterns (Losapio et al ., 2021 ). In contrast with ectomycorrhizal symbiosis, for which it has been known for decades that several plant species are able to combine photosynthesis and carbon uptake from fungi, in a strategy termed ‘partial mycoheterotrophy’ (Selosse & Roy, 2009 ), only recently this mode of life has been suggested to be widespread within the AM symbiosis. Giesemann et al . ( 2021 ) have shown that many photosynthetic understory plants are potentially able to take up carbon from associated AM fungi. Future work will enlighten us on whether these partially mycoheterotrophic plants rely on similar sets of fungi and rely on similar interaction patterns within the mycorrhizal network as the fully mycoheterotrophic plants in in the present study."
} | 6,847 |
35118361 | PMC8800117 | pmc | 159 | {
"abstract": "Summary Self-powered wearable devices, with the energy harvester as a source of energy that can scavenge the energy from ambient sources present in our surroundings to cater to the energy needs of portable wearable electronics, are becoming more widespread because of their miniaturization and multifunctional characteristics. Triboelectric and piezoelectric nanogenerators are being explored to harvest electrical energy from the mechanical vibrations. Integration of these two effects to fabricate a hybrid nanogenerator can further enhance the output efficiency of the nanogenerator. Here, we have discussed the importance of 2D materials which plays an important role in the fabrication of nanogenerators because of their distinct characteristics, such as, flexibility, mechanical stability, nontoxicity, and biodegradability. This review mainly emphasizes the piezoelectric, triboelectric, and hybrid nanogenerator based on the 2D materials and their van der Waals heterostructure, as well as the effect of polymer-2D composite on the output performance of the nanogenerator.",
"conclusion": "Conclusions and perspectives In this review, we have summarized the role of the emerging 2D materials in the most promising energy harvesting technologies which open up a window for developing the self-powered system. The usage of the 2D materials in energy harvesting technologies not only renders the flexibility and ultrathin thickness which makes them suitable for designing a very thin device by forming stacking structures but also demonstrates that the extraordinary properties and distinctive characteristics of these materials have brought significant advancement in the mechanical energy harvesters by enhancing device performance. However, the performance of the device formed by a single energy harvesting mechanism is still low which hinder its practical applicability; therefore, usage of the hybrid energy systems in which the integration of two or more energy-harvesting mechanisms and coupling of multiple energy sources is favorable as they result in a new advanced technology capable of working continuously and reliably, even when some energy sources are unavailable for a short period. Great progress regarding the usage of the 2D materials for piezoelectric and triboelectric nanogenerator has been achieved in terms of theoretical research and multifaceted application demonstrations, but there are several challenges for the development and application of 2D materials based energy harvesters. 1. Scalable production of ultrathin 2D materials: Although a large number of 2D materials have been prepared and studied, the ongoing efforts are still limited by low yield, stringent restriction in growth conditions (e.g., high temperature, gas control in CVD, high vacuum), and instability of the synthesized 2D materials. Therefore, the future research direction will include the rational design, high quality and scalable synthesis of 2D materials with desired structural features including crystal phase, size, thickness, etc. 2. Exploration of 2D materials: The number of theoretical studies reporting the piezoelectricity and triboelectrification behavior in the 2D materials are much greater than experimental ones. So, more experimental studies are required to efficiently increase the applications of 2D materials for flexible and wearable electronics. 3. Improvement in the output performance and efficiency: An extensive thorough understanding of the basic mechanism involved in piezoelectric and triboelectric effect is required, which will facilitate the optimization of the output performance of devices to achieve a higher energy conversion efficiency. 4. Enhancing the working stability and environmental adaptability: As some of the 2D materials are sensitive to external environment stimuli such as temperature and humidity that degrade the performance of the device( Xiong et al., 2018 ), the incorporation of 2D materials with other robust materials and the optimization of device structure and packaging process are some good ways for stable output performance under extreme conditions.",
"introduction": "Introduction The rapid growth in population and industrialization has aggravated the significant ecological deterioration and energy crisis worldwide due to the consumption and depletion of fossil fuels. Fossil fuels provide the most significant contribution as a common energy source, but their limited reserve in nature makes it difficult to provide it for the future generation. Therefore, several efforts have been made by the scientific community over the past decades by exploring sustainable, renewable, and green energy sources to fulfill the future demand of energy in an effective and environment-friendly way. Besides this, with the forthcoming Internet of things(IoT) and artificial intelligence era, the need for smart electronics with multiple functionalities, portable, flexible, and miniaturization concepts are highly desirable ( Liu et al., 2019 , 2021a , 2021b ). Till now, most of the energy requirements in the electronic devices are fulfilled by the batteries even though it endows several drawbacks due to their short life span, durability, short charging/discharging, heavyweight, rigid, bulky, and overheating nature ( Armand and Tarascon, 2008 ; Van Noorden, 2014 ). This results in an exacerbated overall performance in the context of portability and wearability of such devices. Because some of these electronic devices work only a few hours, energy harvesting/storage devices with high energy density and capacities are required to meet the growing demand for energy to address the emerging energy needs. Therefore, a self-powered electronic device, with the energy harvester as the source of energy that can scavenge the energy from ambient sources present in our surrounding to power up its electronic components and sensors, as well as storing the excess energy for later use for continuous and stable operation, has become a viable alternative for the cumbersome batteries that require frequent recharge/replacement. Such technologies are especially vital for medical applications. So far, various energy harvesting technologies have been developed as a means to supply power to these electronic devices to form a self-powered system, such as, photovoltaic cells for harnessing solar energy( Behura et al., 2019 ); piezoelectric( Jung et al., 2011 ) and triboelectric nanogenerator (TENG) ( Niu and Wang, 2015 ) for harnessing mechanical energy; thermoelectric( Xie et al., 2017 ) and pyroelectric generator( Yang et al., 2012 ) for harnessing thermal energy. Among these energy harvesting technologies, mechanical energy harvesting is of the utmost importance because of its omnipresence which is not restricted by weather, space, and time, unlike solar energy harvesters where the performance of solar cells is strongly dependent on the light illumination and weather conditions; and thermal energy harvesters, which has a disadvantage due to small thermal gradients and low efficiency. The advancement in the field of nanoscience and nanotechnology allows us to manipulate the materials at the nanoscale, a process deemed quite byzantine before; consequently, numerous two-dimensional (2D) materials have sprung up. However, a huge breakthrough was made with the isolation of graphene in 2004( Novoselov et al., 2004 ), which rekindled researchers' interest to find more 2D materials. Thus, several 2D materials come forward which includes transition metal dichalcogenides (TMDs), black phosphorus, transition metal oxide (TMO), transition metal carbides/carbonitrides (MXenes), hexagonal boron nitride, etc., which cover up a wide range of properties of metals, semiconductors, and insulators ( Kim et al., 2015 ; Mas-Balleste et al., 2011 ; Novoselov et al., 2016 ; Rao, 1989 ). The atomic-scale thickness of these 2D materials offers tremendous physicochemical properties, such as, flexibility, high surface-to-volume ratio, stretchability, transparency, biocompatibility, which allows us to fabricate very thin electronic devices, even in a stacked structure. Also, the strong in-plane covalent bonding and weak interlayer van der Waals interaction in the 2D materials provides high in-plane stability allowing us to produce it at the thickness of individual unit cell and freestanding in nature. These 2D materials offer properties that are quite different from their bulk counterpart, for example, several 2D materials with single atomic layers, such as MoS 2 , h-BN, MoSe 2 , WS 2 , WTe 2 , and WSe 2 showing piezoelectric properties that are not present in their bulk form and therefore have demonstrated application as piezoelectric nanogenerator ( Ares et al., 2020 ; Duerloo et al., 2012 ). TMDs are shown to absorb more incidental sunlight as compared to the traditional GaAs and Si because of their tunable bandgap and strong light-matter interaction properties( Jariwala et al., 2014 ; Wang et al., 2015a ). Furthermore, the self-assembly and abundance of micro/nanostructure in the 2D materials generate additional geometric sites for storage of guest ions and facilitate intercalation/deintercalation of ions, as a result, they are utilized as electrodes in the energy storage device such as lithium/sodium-ion batteries( David et al., 2014 ; Lukatskaya et al., 2013 ; Shi and Zhao, 2017 ). Figure 1 shows the different characteristics and applications of 2D materials. These unique optical, electrical, mechanical properties and their van der Waals heterostructure of the 2D materials are promising for energy harvesting and storage application in the electronic device and therefore motivate the present piece of work. In this review the basics of energy harvesting based on piezoelectricity, triboelectricity and the current status of the work which have been carried out in the field of ultrathin 2D nanogenerators fabricated using different 2D materials for self-powered devices will be discussed. Figure1 A schematic showing 2D nanomaterials with their characteristic and application 2D materials for energy harvesting Among the various 2D materials, graphene is the most investigated material for energy harvesting and advanced industrial applications. Graphene with sp 2 hybridization among its carbon atom provides tremendous properties, such as high charge carrier mobility, ultrahigh specific surface area, high optical transmittance, and exceptional mechanical properties which enable its application in numerous devices including supercapacitors, transistors, solar cells, resonators, batteries, and energy harvesting devices( Allen et al., 2010 ; El-Kady et al., 2016 ; Yin et al., 2014 ). Although, graphene is intrinsically non-piezoelectric in nature, it can be engineered by chemical doping of the atoms on the basal plane of graphene, which breaks the inversion symmetry and induces piezoelectricity in graphene ( Ong and Reed, 2012 ). Graphene derivatives such as graphene oxide (GO) have a high dielectric constant and Young’s modulus, enabling it to be a good option for effectively harvesting mechanical energy ( Bhavanasi et al., 2016 ). Also, reduction of GO to reduce graphene oxide transform the electronic band structure from insulator to semiconductor depending on the degree of reduction. These distinctive features help in overcoming the problem of high leakage current and make a promising candidate as the interfacial material for triboelectric nanogenerators to elevate the output voltage ( Eda et al., 2009 ; Que et al., 2012 ). Hexagonal boron nitride (h-BN) is analogous to graphene where an equal number of boron and nitrogen atoms are alternatively present in the honeycomb configuration. The highly polarized covalent bond endows h-BN with an asymmetric structure and electrical insulating properties which can be tuned by strain engineering, defect engineering, and chemical functionalization ( Tran et al., 2016 ; Weng et al., 2017 ). It also possesses many outstanding properties, such as high thermal stability, high in-plane thermal conductivity, high mechanical strength, chemical inertness, transparency, and so forth ( Boldrin et al., 2011 ; Watanabeet al., 2004 ). Also, owing to the atomically flat and danglingbonds–free surface, h-BN gathered considerable attention as an excellent dielectric substrate suitable for 2D materials to form vertically stacked heterostructures that can enhance the electrical and optical properties of 2D materials ( Xue et al., 2011 ). TMDs are the layered compounds with a generalized formula of MX 2 , where M is the transition metal, mainly, and X represents chalcogens such as S, SE, and Te. Each monolayer of TMD consists of three atomic layers in which the transition metal is sandwiched between two chalcogens atoms. Because of the interaction between s-P z orbitals, during exfoliation of the multilayer, single-layer band gaps widen, making transition from an indirect bandgap to direct bandgap in the electronic structure. This wide bandgap provides excellent photoluminescence and avoids current leakage in piezoelectric materials ( Chhowalla et al., 2013 ). Two-dimensional TMDs have a wide range of tunable properties and may be functionalized with a variety of polymers, making them ideal candidates for flexible and transparent electronics with improved mechanical efficiency. The class of 2D layered transition metal carbides, nitrides, and carbonitrides are referred to as MXenes, having the generalized formula M n+1 AX n , where M represents early transition metal (e.g., Ti, V, Cr, etc.); A is the element from group IIIA and IVA (e.g., Al, Si, Sn, etc.), and X refers to carbon, nitrogen, and n is an integer that can take values from one to three, are emerging 2D materials ( Naguib et al., 2014 ). MXenes exhibit several advantages owing to their high electric conductivity, strain-tunability, diverse surface chemistry, tunable bandgap, good stability, and excellent mechanical properties which facilitates its application in gas sensors ( Wu et al., 2019b ), energy storage devices( Anasori et al., 2017 ), photocatalysis( Guo et al., 2016 ) and flexible electronics( Gao etal., 2020 ). Also, because of the presence of the abundant –F group and oxygen-containing terminating functional group, MXenes exhibit outstanding metallic conductivity and electronegativity, making them a good substitute for electronegative materials that can improve TENGs’ output performance ( Dong et al., 2018 ). Moreover, the hydrophilic behavior of the MXenes provides better interaction with the polymer matrix that can enhance impact strength and the Young’s modulus and encourage their usage in composite materials ( Ling et al., 2014 ; Sobolčiak et al., 2017 ). In addition to the above-mentioned 2D materials, there are many others, such as black phosphorus, transition metal oxides, layered metal-organic framework, graphitic carbon nitride, layered covalent organic frameworks, and so on, which are being investigated for energy harvesting applications. The 2D materials exhibiting the piezoelectric properties are of great interest because of their high in-plane flexibility and piezoelectric coefficient as compared to the conventional piezoelectric materials that allow its application in flexible mechanical energy harvesting devices. Mechanical energy harvesters Among the various sources of energy, mechanical energy is the most exploited source of energy because of its accessibility, ubiquity, and abundance in our surrounding environment. Every motion that is present in our environment is considered as the potential source of the kinetic energy, such as vibrations, wind flow, river flow, blood flow, and movements associated with the human body such as walking, breathing, talking, typing, and so on, can be converted into electricity with these energy harvesters. Here, we will mainly discuss the two types of energy harvesters based on 2D materials, for scavenging the mechanical energy; the first is the piezoelectric nanogenerator and the other is triboelectric nanogenerator, which is an important way to supply the electricity in a wearable device. Also, comparasion between the piezoelectric and triboelectric nanogenerator is sumarrized in Table 1 . Table 1 Comparison of the piezoelectric and triboelectric nanogenerator Piezoelectric nanogenerator Triboelectric nanogenerator Device structure Principle Piezoelectric effect Contact electrification and electrostatic induction Source of energy Bending/mechanical vibration/force Vibration/sliding/rotation Impedancetype Capacitive Capacitive Structure Structural deformation of mechanical strain Structural deformation of relative displacement Materials Piezoelectric materials Triboelectric materials Advantage ➢ Low internal resistance ➢ High current, power, and energy density ➢ High sensitivity to stimuli ➢ easy integration ➢ easy miniaturization (macro to nanoscale) ➢ Diverse choice of materials ➢ Structural flexibility ➢ High efficiency ➢ Multiple working modes ➢ High output voltage ➢ Nature friendly Disadvantage ➢ Low efficiency ➢ Pulsed output ➢ High matched impedance ➢ Pressing frequency may affect energy density ➢ Some materials are toxic ➢ Low durability ➢ High internal resistance ➢ Low current, power, and energy density ➢ Pulsed output ➢ High frictional damage Application ➢ Body-implantable and patchable device ➢ Wearable device ➢ Wireless sensor node ➢ Wearable and portable electronics ➢ Self-powered device ➢ Self-sensing and monitoring ➢ Remote charging Piezoelectric nanogenerator (PENG) The piezoelectric nanogenerator works on the principle of the piezoelectric effect, which states that application of mechanical stress on the non-centrosymmetric materials deforms the crystal structure, resulting in the generation of polarization charges, thus creating a potential difference across the material. The generation of electric potential by application of mechanical stress is known as the direct piezoelectric effect and is widely implemented in energy harvesting systems. While in the converse piezoelectric effect, mechanical strain is induced in the material when subjected to electrical signal/voltage, and is applicable to vibration damping, acoustic emitters, and actuators. In general, the piezoelectric effect is a reversible process and the materials which exhibit piezoelectric properties are known as piezoelectric materials and these demonstrate plenty of applications in energy transducers, actuators, sensors, and energy harvesters ( Liao et al., 2014 ; Masmanidis et al., 2007 ; Rupitsch, 2018 ). The crystal structure of these piezoelectric materials is non-symmetric, but has a balanced equilibrium between charges such that the effects of the positive and the negative charges exactly cancel out each other. When these materials are subjected to mechanical stress or force, dipoles are no longer aligned in the way to cancel out the effect of each other, which results in the generation of a net polarization charge on the surface of the materials. The number of dipoles present in the piezoelectric material plays a crucial role in their performance and behavior; therefore, to acquire more dipoles these materials are subjected to a strong electric field near the Curie temperature. This process is known as the ‘poling’ and it imparts a net permanent polarization in the materials which also change with the applied stress, as a result further enhancing the piezoelectricity. Most of the ferroelectric materials show piezoelectric behavior, but the converse is not true ( Datta and Mondal, 2019 ; Mishra et al., 2019 ; Ramadan et al., 2014 ). The piezoelectric effect is the coupling phenomenon between the electrical and mechanical properties of the materials. The electrical behavior of the material is given as (Equation 1) D i = ∑ j = 1 3 ε ij E j or D = ε E where ε ij is electric permittivity, D i is electric displacement, E j is the electric field. Mechanical behavior is given by the relation between stress and strain under small deformation (Equation 2) T ij = ∑ k , l = 1 3 c ijkl S kl where c ijkl is the elastic stiffness of the material. The above Equation 2 can be interpreted reversely as (Equation 3) S ij = ∑ k , l = 1 3 s ijkl T kl where s ijkl is elastic compliance of the material. The reduced tensor notation of the coupled constitution equation of piezoelectric materials are depicted ( Wang et al., 2015b ) as (Equation 4) D i = E j ε ij T + d iJ T J (Equation 5) S I = d Ij E j + s IJ E T J where the superscripts T and E mean the coefficients at constant stress and electric and, d iJ is the piezoelectric strain constants, respectively. In general, the piezoelectric material has two functioning modes to begin with. The device is said to work in d 33 mode when the direction of the applied stress is the same as the direction of polarization, and the generated piezoelectric charge is given by (Equation 6) Q PENG = Aσ d 33 where σ is the stress along the direction of applied force, A is the area of the piezoelectric material, d 33 is the piezoelectric charge coefficient. The corresponding piezoelectric potential for the open circuit condition is obtained by (Equation 7) V PENG = tσ g 33 where t is the thickness of the piezoelectric material and g 33 is the piezoelectric voltage coefficient respectively. Second, when the direction of the applied stress is perpendicular to the direction of the polarization, the device is said to work in d 31 mode. For this, the piezoelectric charge and voltage can be expressed as (Equation 8) Q PENG = Aσ d 31 (Equation 9) V PENG = tσ g 31 Because these 2D piezoelectric materials provide ultrathin geometry and better electromechanical response, they have become the intriguing topic of research owing to the anticipated demand in diverse functional and small scale devices ( Wang et al., 2012 ; Wu et al., 2014 ). Blonsky et al. (2015) determined the piezoelectric coefficient of 2D metal dichalcogenide, III-V semiconductor materials, and metal oxides and showed that the crystal structure of these 2D materials are exhibiting either planar hexagonal, buckled hexagonal, or 2H structure, which makes them piezoelectric in nature because of the lack of inversion symmetry in these structures. They were found to exhibit only one independent in-plane piezoelectric coefficient (d 11 ) in planar hexagonal and 2H structure whereas buckled hexagonal structure is reported to have out-of-plane piezoelectric coefficient (d 31 ) in addition to the d 11 coefficient. Although non-centrosymmetric structure and existence of bandgap is a prerequisite condition for the materials to be piezoelectric, the piezoelectric properties can be altered through surface engineering such as adsorption of foreign atoms or introducing the in-plane defects, thereby making these materials piezoelectric in nature. An example of this instance is that pristine graphene is not piezoelectric, however, incorporation of adatoms and defects break down the inversion symmetry leading piezoelectric effect in graphene( Chandratre and Sharma, 2012 ; Ong et al., 2013 ; Ong and Reed, 2012 ). Also, black phosphorene with centrosymmetric structure and non-polar space group can induce piezoelectricity by surface oxidation with piezoelectric coefficients d 11 and d 12 are 88.54pmV -1 and -1.94 pmV -1 respectively( Li et al., 2018 ). Muralidharan et al. showed that mechanoelectrochemical stress-voltage coupling in black phosphorus nanosheets is capable of harvesting low frequencies ∼0.01 Hz with the peak power delivery of ∼42 nW/cm 2 with bending and pressing impulse ( Muralidharan et al., 2017 ). Lee et al. reported that in comparison to mechanically exfoliated WSe 2 bilayers with Bernal stacking, the WSe 2 bilayers manufactured using turbostratic stacking had stable piezoelectric characteristics and generated an output voltage of 85 mV( Lee et al., 2017 ). Triboelectric nanogenerator (TENG) The triboelectric effect is a contact electrification process in which certain materials become electrically charged after they come into contact with another material through friction. A few instances of the triboelectric effect are rubbing glass through fur, plastic comb through the hair, and rubbing a balloon on hair, which is generally caused by static electricity. Whether the material acquires positive charge or negative charge and the strength of the charges they acquire depend on the relative polarities, surface roughness, strain, and other properties of materials. Therefore, the material that has an affinity to gain electrons will become the negative charge and the other material will become the positive charge. Depending on the affinity of the material to gain or lose electrons, materials are placed in a systematic order in a series known as the triboelectric series ( Pan and Zhang, 2019 ; Wang et al., 2016 ). Although, the triboelectric effect is considered a negative effect that has brought various disastrous risks in the industry, human life, nature, electronics, it shows great potential in small scale energy harvesting. The Professor Wang group in 2012, invented the first triboelectric nanogenerator (TENG) which uses these static charges to harness ubiquitous ambient small scale mechanical energy ( Fan et al., 2012 ). TENG works on the combined properties of the triboelectric effect and electrostatic induction where contact electrification provides the polarized charge and the role of electrostatic induction is to convert mechanical energy to electrical energy. Therefore, when two materials with different electron affinity are brought in contact with each other by an external force, triboelectric charges with opposite polarities are induced on the surface of the materials. After removing the external force, the charged surface gets separated which will induce the potential difference on the electrode. Thus, by connecting the external load between these two electrodes, charges will flow through the outer circuit to screen out the electric field. A later renewed contact between these surfaces results in the disappearance of these triboelectric charges, thus giving rise to the charge to flow in opposite directions. So, by contacting and separating the materials by an external force, the charges will flow back and forth via the external circuit ( Niu and Wang, 2015 ; Wu et al., 2019a ). The output voltage generated by TENG is given by (Equation 10) V = − Q C ( x ) + V OC ( x ) which is also known as V-Q-x relation and represents the inherent capacitive behavior of TENG where V oc is the open-circuit voltage, C(x) is the capacitance between two electrodes, Q is transferred charge between two electrodes. The output current generated across the load is given by (Equation 11) I = dQ dt = C dV dt + V dC dt where C is the capacitance of the system and V is the voltage across electrodes. The first term in the equation corresponds to the current produced because of the change in voltage across the non-contact surface of the electrode with time and the second term corresponds to the current introduced by variation in capacitance because of cyclic contact and separation. Under the short circuit condition (SC), the charges which transfer across the electrode Q SC completely cancel out the voltage generated by this triboelectric charge, and the above equation becomes (Equation 12) 0 = − Q SC C ( x ) + V OC ( x ) and the fundamental relation among Q SC , C, and V OC is given by (Equation 13) Q SC ( x ) = CV OC ( x ) Suitable material selection and structure design are the two important factors to optimize TENG performance. Therefore, depending on the electrode configuration and the various ways in which triboelectric layers can be stacked, four major modes of TENG are put forward which are shown in Figure 2 . Figure 2 A schematic illustrating the structural design and working mechanism of the different modes in the triboelectric nanogenerator (TENG). As the output power of triboelectrification is influenced by the charge affinity, work function, dielectric constant, impurity content, surface irregularities ( Seol etal., 2018 ). The 2D nanomaterials are becoming promising in TENG due to their ability to achieve favorable electrical and mechanical properties. For example, the introduction of 2D materials such as MoS 2 , reduced graphene oxide (rGO) in the triboelectric layer can aid in trapping electrons and boost the electron density ( Parvez et al., 2019 ; Wu et al., 2017a ). This is also demonstrated by Xia. et al. where insertion of aligned graphene nanosheets in PDMS enhances the output of TENG 3 times as compared to pristine PDMS film–based TENG( Xia et al., 2017 ). Furthermore, a GO dispersion based single liquid electrode-based TENG is reported by Wu et al. where ripples, a large number of folds, and high surface area provided by GO dispersion provide the large defects resulting in the effective charge transfer that provide upgraded deformability, improved mechanical flexibility, and better mechanical properties ( Wu et al., 2019c ). 2D materials based mechanical energy harvesters Several materials have come to the fore for energy harvesting application, among which 2D materials are of burgeoning interest as these can exist in the monolayer or the layered structure with a thickness on the atomic scale. The properties of the 2D materials can be altered with the number of layers. In most TMDs, the bandgap decreases as the number of layers increases, and the shift from indirect to direct bandgap occurs from bulk to monolayer. Piezoelectric nanogenerator based on 2D materials The 2D piezoelectric materials are of tremendous interest in the field of energy harvesting because of their high flexibility and piezoelectric constant. Because of their improved piezoelectric capabilities as compared to conventional piezoelectric materials, 2D materials are well suited for the future of wearable electronics. Recently, numerous theoretical studies to determine the piezoelectric characteristics in 2D materials have been published ( Blonsky et al., 2015 ) and experimental calculations on the piezoelectric characteristics of 2D semiconductors have been started. The piezoelectric characteristics of MoS 2 are reported by Professor Zhong Lin Wang’s group ( Wu et al., 2014 ) where physical exfoliation has been used to form a monolayer MoS 2 on a Si substrate. They carried out their research using atomic force microscopy (AFM) and Raman spectroscopy. A second-harmonic generation (SHG) technique is also used to examine the crystallographic orientations of MoS 2 flakes. After that, Cr/Pd/Au electrode is coated on a MoS 2 film and transferred to the polyethylene terephthalate (PET) substrate to construct a flexible device. The crystal orientation of MoS 2 can be classified into two types as it is synthesized: armchair and zigzag, therefore, by using these two crystal orientation directions a monolayer MoS 2 based device is fabricated whose voltage and current measurements are carried out under periodic strain as illustrated in Figure 3 (A). A positive value of output current and voltage is observed when deformation is applied along the armchair direction as a function of applied tensile strain. Whereas, negative output occurs, when the strain is reduced, changing mechanical energy into electricity. The current and voltage of this device reached 27 pA and 18 mV under 0.64% strain ( Figure 3 (B)), indicating that the output performance improves with applied strain. These results confirmed that under varied strain circumstances, the output voltage and current can be increased significantly with an increase in strain. Based on this, the load resistance measurements are performed at 0.53% strain and the stability of the MoS 2 based piezoelectric device is determined ( Figure 3 (C)). The piezoelectric output as a function of the number of MoS 2 layers and bulk MoS 2 are also measured as shown in Figure 3 (D). These results show that even layer numbers do not show piezoelectric properties, and output performance decreases as the layer number increases; yet, interestingly, piezoelectric characteristics exist exclusively in an odd number of layers. This is the first experimental research to show that 2D materials have piezoelectric characteristics. In 2017, Kim et al. evaluated the piezoelectric coefficient of single-layer MoS 2 generated by chemical vapor deposition (CVD) using piezoresponse force microscopy (PFM) ( Han et al., 2019 ; Kim et al., 2016 ). In this work, studies on the PFM were carried out along a lateral direction, as indicated in Figure 3 (E). The crystal structure of MoS 2 is hexagonal where Mo or S atoms are placed along each side of the MoS 2 flake. The piezoelectric coefficient is unlikely to be the same in both directions as the atomic orientation of MoS 2 dictates whether it has an armchair or a zigzag structure. Surprisingly, because the electrical and mechanical states have a linear relationship, the slope of the solid line representing the fitted linear equation may be utilized to compute the piezoelectric coefficients. Earlier studies show that the force-distance curve can be used to determine the lateral piezoelectric coefficient, d 11 . By comparing and computing the piezoelectric response of α-quartz, the piezoelectric constant of MoS 2 may be derived. In the armchair direction, the d 11 of the MoS 2 is 3.78 p.m. V −1 , while in the zigzag direction, its value is 1.38 p.m. V −1 . These experimental results are also compatible with prior simulations published. Figures 3 F–3G shows that a maximum output voltage of 20mV in armchair direction and 10mV in zigzag direction is obtained under 0.5 Hz frequency and 0.48% strain where the value of output current in the armchair and zigzag direction is found to be 30 and 20pA, respectively. As a result, the piezoelectric constant, which varies depending on the atomic orientation of the MoS 2 generated via CVD, is significant. The multilayer TMDs have considerably reduced or abolished piezoelectricity because continuous growth leads to a stable stacking structure with alternate polarization directions in surrounding layers. The piezoelectricity in WSe 2 mono- and bilayer is reported by Lee et al. with both simulations and experimental observations ( Lee et al., 2017 ). The layer of WSe 2 with a size of 30–50 μm is grown by the CVD process and then transferred to a flexible substrate using the usual wet transfer procedure. Initially, the lateral PFM method was used to determine the piezoelectric constant of the monolayer WSe 2 , which is 3.26 ± 0.3 p.m. V −1 . PENG output performance is also tested at 45 mV and 100pA with constant strain (0.39%) at a strain rate of 40 mm/s. The piezoelectric characteristics of bilayer WSe 2 are produced by two ways, one by directly growing on a sapphire substrate (db-WSe 2 ) and another by transferring single-layer of WSe 2 onto another single-layer WSe 2 film (tb-WSe 2 ) are then examined. In contrast to the single-layer WSe 2 , the lack of centrosymmetric structure in bilayer WSe 2 with Bernal stacking is responsible for loosening of piezoelectric properties, as the polarity is completely neutralized in the stacking mode ( Hsu et al., 2014 ; Puretzky et al., 2016 ). The increase in degrees of freedom in bilayer symmetry in tb-WSe 2 manufactured by the transfer approach, on the other hand, resulted in piezoelectricity. The geometric relationship between the two layers was reduced when randomly stacked by the transfer method, allowing for a variety of stacking configurations, which increased the asymmetry. The results of the lateral PFM confirm this. The piezoresponse of monolayer-WSe 2 , α-quartz, and tb-WSe 2 are illustrated in Figure 3 (H). Although the piezoelectric coefficient of tb-WSe 2 is lower than that of α-quartz, it still exhibits piezoelectric properties. A piezoelectric generator was built based on this, and its output performance was examined. Under a 0.64% strain condition, the output voltage of single-layer WSe 2 is 90 mV. However, the performance of the device deteriorates when the value of strain exceeds 0.64%. The tbWSe 2 based PENG, on the other hand, produced a maximum voltage of 85mVat 0.95% strain is depicted in Figure 3 (I). The high Young’s modulus and sliding effect in TMDs interlayer offers superior mechanical properties in bilayer TMDs materials as compared to monolayer TMDs ( Bertolazzi et al., 2011 ; Liu et al., 2014b ). Figure 3 Pristine 2D material based piezoelectric nanogenerator (A) The piezoelectric voltage and current response of PNG based on monolayer MoS 2 under constant strain in the armchair and zigzag direction. (B) The output voltage and current dependency on the magnitude of applied strain in monolayer MoS 2 devices. (C) Variation in the piezoelectric voltage as a function of applied load resistance with 0.53% strain. (D) The piezoelectric output of MoS 2 flakes by varying the number of layers, as well as a bulk MoS 2 flake ( Wu et al., 2014 ). (E) A schematic showing the lateral PFM measurement configuration with the help of a non-conductive AFM tip to determine piezoelectricity in monolayer MoS 2 . (F and G) the output voltage response of a MoS 2 monolayer in the armchair and zigzag direction as a function of applied strain ( Kim et al., 2016 ). (H) The lateral PFM results of m -WSe 2 , tb-WSe 2 , and α-quartz with their piezoelectric coefficient are shown in the inset. (I) m -WSe 2 and tb-WSe 2 piezoelectric output voltage at 0.57 and 0.95% strain ( Lee et al., 2017 ). (J andK) The output voltage and current response of the pristine MoS 2 monolayer and S-treated MoS 2 monolayer ( Han et al., 2018 ) One of the large-area synthesis technologies for the synthesis of 2D materials is CVD, where defects are unavoidably produced during the synthesis process. These defects induce free electrons to generate a potential screening effect which reduces the piezoelectric effect by removing some of the piezoelectric potential produced by mechanical deformation. Sulfur (S) pores, for example, behave as n-type carriers in MoS 2 which produce a screening effect. Therefore, the carrier density of n-type carriers should be reduced to improve the piezoelectric characteristics of MoS 2 . Han et al. reported that the piezoelectric characteristics of MoS 2 can be improved by efficiently regulating the sulfur vacancies in the in-situ synthesis of MoS 2 monolayer using the CVD method by heat treatment with additional H 2 S( Han et al., 2018 ). The properties of MoS 2 are also systematically investigated before and after sulfur treatment. Despite having a piezoelectric coefficient of 3.06 ± 0.6 pmV -1 , MoS 2 exhibits unstable characteristics. The piezoelectric coefficient of S-treated MoS 2 is 3.73 ± 0.2 p.m. V −1 , which shows good agreement with the reported theoretical value ( Duerloo et al., 2012 ). Based on these findings, it is projected that after the sulfur treatment, MoS 2 ’s piezoelectric capabilities will improve. After that, a PENG is fabricated by transferring it to a flexible PET substrate and the output characteristics of the piezoelectric generator are investigated. Figures 3 J and K show that the S-treatment of MoS 2 can enhance the current up to three-fold, from 30 pA to 100 pA, whereas the voltage increasing from 10 mV to 20 mV shows two--fold enhancement in the voltage as compared to pristine MoS 2 . Also, a 10-fold enhancement in the maximum power is observed for S-treated MoS 2 . Therefore, we can say that defect treatment can improve the degraded piezoelectric characteristics of large-area 2D materials. Recently, many theoretical studies demonstrated a new type of TMD structure possessing an unconventional asymmetry sandwich construction that differs from conventional TMDs. The Janus structures consist of two layers of different chalcogen atoms sandwiching transition atoms between them ( Cai et al., 2019 ). The reflection symmetry about the Mo atom is broken in the MoSSe structure owing to the MXY type structure, resulting in out-of-plane electric polarization in addition to the significant in-plane piezoelectric effects seen in the standard MX 2 structure. The reflection asymmetry is caused by a difference in electronegativity between the two chalcogen atoms. Many theoretical studies on Janus structures have revealed extremely high out-of-plane piezoelectricity, as measured by DFT. As a result, researchers now have a new technique to investigate and improve the piezo-output of 2D materials in the field of energy harvesting. Although few reports are available on the 2D Janus van der Waals heterojunctions, they showcase the immense potential in energy harvesting devices. For example, with higher elastic moduli, Janus van der Waals heterojunctions of MoSSe–BlueP are discovered to be dynamically stable( Li et al., 2019 ). The piezoelectric response (out-of-plane) of these Janus van der Waals heterojunctions measured by the piezoelectric coefficient e 311 (0.081C/m) is higher than that of the Janus MoSSe monolayer (0.058C/m), implying that it has potential for piezoelectric applications. The 2D TMDC atomic layer provides various advantages over the bulk material which includes strong in-plane covalent bonding, weak van der Waals connection between layers, reduced dielectric screening out-of-plane, significant exciton binding energy, and so on. As a result, the heterostructure piezoelectric effect is enabled by the combination of multiple atomic layers. A type-II staggered gap alignment is formed by the p- and n-type van der Waals heterostructure of TMDC atomic layers with different work functions. Between p- and n-type atomic layers, the large band offset of conduction band minimum (CBM) and valence band maximum (VBM) causes strong electric polarisation and piezoelectric conversion. The piezoelectric characteristics of the MoS 2 /WSe 2 heterostructure were explored by first-principle calculations using density functional theory (DFT), by Felix Jaetae Seo group in 2018 ( Yu et al., 2018 ) as shown in Figures 4 A–4D. Among the pairings of transition metal ions (Mo and W) with dichalcogenides (SE and S), the MoS 2 /WSe 2 heterostructure has the highest band offset between CBM and VBM. The considerable electronic polarization between n-type and p-type atomic layers is implied by the large band offset between CBM and VBM. In a first principle, DFT analysis for a partly vertical heterostructure of a MoS2/WSe2 with a size of 3.0 × 1.5 nm, the output voltages for 4 and 8% tensile strains are 0.137 and 0.183 V, respectively. The significant electric polarization is responsible for the large output voltage of heterostructure atomic layers at only a few nanometer scales. Figure 4 2D materials van der Waals heterostructure for Piezoelectric Nanogenerator (A) Schematic illustration of the piezoelectric device fabricated by using heterostructure of WS 2 and MoS 2 layers. (B) The distribution of electrostatic potential between electrodes for the entire heterostructure region. (C) The distribution of electrostatic potential near the electrode. (D) Variation in output voltage as a function of strain( Yu et al., 2018 ) To take a step further, many researchers are focusing on the fabrication of nanogenerators based on the composite of 2D materials with polymers ( Singh and Singh, 2020 ). The number of research publications is increasing at a great rate in this field. Polymer-based piezoelectric materials, such as PVDF( Singh et al., 2018 ) or its comparable copolymers, such as P(VDFTrFE), because of their advantageous qualities of flexibility, appropriate mechanical strength, ease of manufacturing, and excellent chemical resistance, are extensively employed for piezoelectric applications. Composite material-based nanogenerators, which are typically made of 2D piezoelectric nanomaterials dispersed in a polymer matrix, are a promising option for large-scale, flexible energy harvester applications. Composite-based nanogenerators have several advantages, including a simple fabrication process, low cost, and mechanical robustness. M.C. Bhatnagar’s group reported the synthesis of PVDF/RGO nanocomposite films for the fabrication of piezoelectric nanogenerators in 2020 ( Anand et al., 2020 ). They used a sonication-assisted approach to make reduced graphene oxide (RGO) nanosheets and a solution casting method to make a PVDF/RGO nanocomposite thin film. An X-ray diffractometer and a transmission electron microscope were used to investigate the structural analysis and surface morphology of produced reduced graphene oxide (RGO) nanosheets. MicroRaman Spectroscopy (MRS) was used to investigate the RGO’s distinctive modes. XRD was used to investigate the structural characteristics of RGO-loaded PVDF nanocomposite films. A scanning electron microscope was used to examine the surface morphology of the nanocomposite films. The impact of incorporating RGO nanosheets into the PVDF matrix was investigated. They also looked at how RGO nanosheets affected the nucleation of the electroactive polar-phase, as well as the optical, structural, and electrical aspects of nanocomposite films. Fourier Transform Infra-Red (FTIR) has been used to characterize the generation of β-phase in nanocomposite films. They also measured the piezo-response (output voltage signal) when ultraviolet-visible light was shone on nanocomposite films by applying pressure with a human finger. The output voltage of RGO loaded nanocomposite films increased significantly from 0.886 to 1.915 V, indicating a considerable increase in piezo-response as shown in Figures 5 A–5E. RGO loaded nanocomposite PVDF films demonstrated a 13.8% increase in piezo-voltage with the same pressure action when exposed to ultraviolet-visible light. The Yuan Lin group reported reduced graphene oxide and PVDF-TrFE sheets based on piezoelectric nanogenerators in 2019( Hu et al., 2019 ). The rGO/PVDF-TrFE films were prepared using scrap coating and in situ electric polarization in this study, and wearable nanogenerators with improved piezoelectric properties and energy-harvesting performance were fabricated. The addition of rGO to PVDF-TrFE improves the-phase crystallinity and increases hydrogen bond formation as well as dipole interaction between rGO and PVDF-TrFE, further improving the energy-harvesting performance of these devices. The rGO/PVDF-TrFE WNGs have a maximum open-circuit voltage of 8.3 V and a short-circuit current of 0.6 A, resulting in a power density of 28.7 w/m 3 . The rGO/PVDF-TrFE nanogenerators had 1.6 times the open-circuit voltage and twice the power density of pure PVDF-TrFE-based devices, respectively ( Figures 5 F–5G). A piezoelectric nanogenerator made of poly(vinylidene Fluoride) nanofiber webs and MoS 2 is reported in 2016 by Dr. Dipankar Mandal group ( Maity et al., 2016 ). They have devised a simple and low-cost technology that allows for the large-scale manufacture of a few layers of MoS 2 . Furthermore, they disclose, for the first time, an ultrasensitive and efficient PNG constructed of a few layer MoS 2 nanosheets with electrospun PVDF NFW. This PNG can detect lightweight objects (such as a leaf or a match stick) falling from different heights ( Figures 5 H–5I, as well as voltage produced by finger touch ( Figure 5 (J)), indicating that it has enormous potential in industry for use as an automated frequency counter mat/belt to count batch fabrication while also detecting quality using only the external mechanical impact. The specific speech recognition potentiality has also been noted, implying a broader application as a potential nanosensor in the biomedical sector to meet national security demands. PNG also has a significant piezoelectric output response and quick capacitor charging ability when scavenging mechanical energy from acoustic signals such as music vibration. They discovered that MoS 2 nanosheets help to improve overall crystallinity and piezoelectric-phase composition in PVDF NFW, which is suitable for PNG construction. Furthermore, because of its inherent charge buildup and transfer characteristics, 2D-MoS 2 modulates the overall piezoelectric performance of PNG. PNG demonstrated a tremendous improvement in acoustic sensitivity that paved the way for noise detection ( Figure 5 (K)), particularly in self-powered mode, because of these improved functions of the resultant 2D-MoS2 contained PVDF NFW. Furthermore, PNG’s ultrafast capacitor charging capabilities, i.e., 9 V in 44 s is shown in Figure 5 (L) which indicates a possible utility for quickly powering up portable electronic equipment. Herein, a summary of the piezoelectric nanogenerator based on 2D materials is presented in Table 2 . Figure 5 Polymer-2D composite material based Piezoelectric nanogenerator (A–E) Piezoelectric Output response for different PVDF/RGO composite based nanogenerator and touch and release response from nanocomposite films( Anand et al., 2020 ). (F and G) Output voltage and current measurements of rGO/PVDF-TrFE based PENGs with varied rGO concentrations ( Hu et al., 2019 ). (H–J) The piezoelectric output voltage responses of leaf, match stick, and finger touch for different applied stress when falling on the upper surface of PNG. (K) Non-rectified open circuit voltage produced by PNG to demonstrate its application as a noise detector. (L) The charging response of PNG at different capacitors (capacitances, 1, 2.2, and 4.7 μF) under music running conditions ( Maity et al., 2016 ) Table 2 Comparative summary of the piezoelectric nanogenerator based on 2D materials Materials Electrode Fabrication method Device output Application Ref PVDF/rGO Aluminum Drop casting, thermal evaporation V oc = 1.91V Energy harvesting under UV-visible Light ( Anand et al., 2020 ) rGO/PVDF-TrFE Silver(Ag) Magnetron sputtering, Scrap coating V oc = 8.3V I SC = 0.6μA Power density = 28.7W/m 3 Human motion energy harvesting for powering wearable or portable electronic devices ( Hu et al., 2019 ) PVDF/MoS 2 Aluminum Electrospinning V oc = 14V self-powered biomedical nanosensors ( Maity et al., 2017 ) MoS 2 Palladium CVD V oc = 15 mV I sc = 20 pA Power density = 2 mW/m 2 Nanodevices, bioprobes, stretchable/tunable electronics ( Wu et al., 2014 ) hBN-PVDF Copper foil electrospinning V oc = 68V I SC = 0.1μA Power density = 53.2 μW/cm 2 Biomechanical Energy Harvester (Pedometer) ( Yadav et al., 2020 ) MoSe 2 Chromium/gold APCVD, photolithography, metal deposition (10nm Cr/100nm Au), and lift-off process V oc = 35mV Power density = 42 mW/m 2 Energy Harvesting from Human Activities ( Wang et al., 2021 ) MoSe 2 Chromium/gold CVD, photolithography, metal deposition (10nm Cr/100nm Au), and lift-off process V oc = 60mV Drive PH sensor and photodetector ( Li et al., 2020a ) Sulfur treated MoS 2 Chromium, Palladium, gold CVD V oc = 22mV I SC = 100pA Power density = 73 μW/m 2 – ( Han et al., 2018 ) Triboelectric nanogenerator based on 2D materials To understand the triboelectrification behavior of the 2D materials, several attempts have been made. For instance, the first flexible TENG based on the graphene (GTENG)( Kim et al., 2014b ) is reported by Prof. Kim’s group in which large nanosheets of graphene are synthesized via the chemical vapor deposition technique. The GTENG reported was also designed by using randomly stacked monolayer, bilayer, tri-layer, quad-layer, and regularly stacked (Bernal stacking) few layers of graphene on a copper and nickel foil. The output performance of GTENG is analyzed in terms of their work function and the electrical power generated by this GTENG is capable of driving liquid crystal display (LCD), light-emitting diodes (LEDs), and an electroluminescence (EL) display without any external energy source, paving the way for powering low power portable device and self-powered electronic system. Furthermore, this group in 2018 investigated the triboelectrification behavior of the various 2D materials such as MoS 2 , WS 2 , MoSe 2 , WSe 2 , graphene, and graphene oxide (GO) by fabricating the TENG based on these materials and proposed the position of these materials in the triboelectric series ( Seol et al., 2018 ). To do so, nanosheets of these 2D materials were prepared by chemical exfoliation of bulk phase in a liquid phase, and then to investigate the relative charging polarities of these 2D materials against the materials present in the triboelectric series a simple push-type TENG is fabricated. For this, first, nylon which is considered as the most positive material was taken as the friction material with MoS 2 and Cu was used as an electrode to measure the triboelectric output. Following that, several TENG combinations were prepared by using PTFE, PDMS, polycarbonate (PC), PET, mica, and nylon with 200 nm thick MoS 2 film and measurement indicate that MoS 2 is intermediate in the triboelectric series between PTFE and PDMS. Furthermore, the same procedure was carried out on WS 2 , MoSe 2 , WSe 2 , graphene, and GOwhich showed that MoS 2 and MoSe 2 exhibit negative polarity whereas other 2D materials have positive polarities. The output voltage and the current measurements of these 2D materials with nylon under the force of 0.3 kgf and 1Hz frequency are shown in Figure 6 (A). The highest output voltage and current values, i.e., 7.48V and 0.82μA achieved by MoS 2 based TENG, indicate that it has the highest negative triboelectric charging characteristics among the 2D materials investigated in the triboelectric study. The results are further verified by fabricating TENGs, with a combination of 2D materials with the most negative material, i.e., PTFE in the triboelectric series, where output voltage shows complete opposite behavior when compared to the 2D material-nylon-based TENG. Moreover, a comparison between the effective work function of 2D materials measured via KPMF mode and first principal stimulation is also performed to clarify their ordering in the triboelectric series. Therefore, based on the results obtained from the TENGs with various combinations, a triboelectric series of the 2D materials is proposed as shown in Figure 6 (B). Also, it has been observed that triboelectric characteristics can be altered by chemical doping. The addition of p-type dopant Gold chloride (AuCl 3 ) in MoS 2 increases its work function while n-type dopant benzyl viologen (BV) decreases the work function as a result changes the position of MoS 2 in triboelectric series. A flexible and wearable single electrode TENG based on layer by layer assembly of graphene multilayers is reported by Jung and Kim et al. ( JunáChung and TaeáPark, 2018 ). The layer by layer technique allows the fabrication of customized graphene multilayers at the nanometer scale to be formed on a flat, undulated, and textile surfaces. The performance of the flat and undulated GTENG is examined by varying the number of graphene layers. Furthermore, the textile GTENG’s high mechanical durability is verified by rolling and bending it up to 20,000 times, which facilitates its application in textile-based power sources. Figure 6 TENG based on pristine 2D materials (A) Open-circuit voltage and short-circuit current measurement of a various 2D materials with respect to nylon. (B) Triboelectric series of 2D materials corresponding with their structure ( Seol et al., 2018 ). (C) A schematic showing the flat and crumpled MoS 2 layer synthesized by laser-directed method on SiO 2 /Si wafer and utilizing it to fabricate TENG device. (D) The output power of the F-TENG and MC-TENG with varying the load resistance. (E) Image of triboelectric haptic sensor array on hand where silver(Ag) electrode is printed on crumpled MoS 2 . (F) A schematic showing the ability of triboelectric sensor signal with multi-touch ( Park et al., 2020 ) Although, PTFE is one of the most electronegative material in the triboelectric series towing to its insulating nature, it is only restricted to the single electrode mode configuration. Thus, to overcome these limitations, Dong et al. fabricated TENG based on titanium carbide MXenes (Ti 3 C 2 T x where T x stands for the surface functional group) ( Dong et al., 2018 ). First, a single electrode TENG with MXenes/Glass:PET-ITO TENG and PTFE:PET-ITO TENG are fabricated to confirm the position of MXenes in the triboelectric series which shows that both single electrode TENG showing a similar response with output voltage oscillating between −180 and 500 V. Moreover, the absence of the output voltage in the MXenes/Glass:PTFE TENG shows that they should be placed closed to each other in the triboelectric series as the electronegativity in both MXenes and PTFE arises because of the presence of the same functional group F. Further, to make this TENG flexible, the glass substrate is replaced by the PET-ITO film which can generate the open-circuit voltage in range ∼500–650 V and can charge a 1μF capacitance in 8 min to a voltage of 40V under 15N force at 2Hz frequency. The flexibility test is performed by clamping MXene/PET-ITO: PET-ITO TENG at 30° relative to the horizontal plane with its free end subjected to 1N strain to harvest mechanical energy from human activities such as typing, texting, walking, thumb motion, and so on. This work indicates that MXenes TENG is a viable candidate for powering wearable electronics and can be integrated with a variety of accessories such as wristband, elbow patches, and knee patches. Recently, to analyze the triboelectric behavior in MoS 2 , plane and crumpled layers of MoS 2 was synthesized by laser-directed thermolysis process as shown in Figure 6 (C) ( Park etal., 2020 ), where the surface morphology and strain in the MoS 2 layer are controlled by laser irradiation level such as laser with power higher than 2.62 Jcm −2 form crumpled 2D MoS 2 layer and laser power between 2.52 and 2.61 Jcm −2 form flat 2D MoS 2 layer. The most crumpled MoS 2 based TENG(MC-TENG) can generate ∼40% more power as compared to the flat TENG (F-TENG) with PDMS as counterpart triboelectric material. The F-TENG and MC-TENG power density is also measured as a function of load resistance, with maximum power corresponding to the 100M Ω resistance ( Figure 6 (D)). Also, to demonstrate the multi-touch and position-mapping capabilities, a self-powered flexible haptic sensor ( Figure 6 (E)) is also fabricated with a crumpled MoS 2 array patterned by Ag electrode lines with the help of the inkjet printer. Figure 7 (F) (left) shows multi-touch contact motion with a stylus pen across different pixels of a flexible sensor array and the feasibility of multi-touch detection via triboelectric voltage since there is no serious interface between the potential difference of touched and untouched MoS 2 pixels ( Figure 6 (F)(right)). Figure 7 2D materials based van der Waals heterostructure for triboelectric nanogenerator (A) A Skin-based flexible single-electrode triboelectric nanogenerator. (B) A schematic showing graphene/Cu heterostructure where graphene layer acts as an energy barrier to prevent oxidation of copper. (C–E) The open-circuit voltage measurement of flexible TENG by attaching it to different parts of the body. (F) The bending performance of a graphene/Cu/PDMS-based flexible electrode. (G and H) The output voltage and transfer charge quantity measurement of TENG is based on graphene/Cu/PDMS electrodes ( Li et al., 2020b ). (I) A schematic depicting the charge generation mechanism in TENG during charging and discharging cycles. (J) The output characteristic of the graphite–TMDC TENG electrode. (K) A schematic demonstrating the applicability of TENG by continues illumination of LEDs by voltage accumulation in a capacitor connected to a rectifying circuit ( Nutting et al., 2020 ) As we know, TENG can operate in two modes: direct contact mode and non-contact mode between two frictional materials. The direct contact mode has the disadvantage of deformation and durability because of the continuous friction between the two materials, whereas the non-contact mode has an issue with the electric discharge problem caused by the rough surface and the output power generated. Therefore, to overcome these issues, Prof. Kim’s group fabricated an ultrathin non-contact TENG by using a high dielectric constant friction material, i.e., calcium copper titanate which is coated with the 1,1,2H,2H-Perfluorooctyltrichlorosilane (FOTS) to enhance the surface charge and ultrathin graphene as an electrically conducting material which serves as electrode to solve the electric discharge problem ( Han et al., 2021 ). The charge injection and their maintenance are required in the non-contact TENG, therefore, a thin layer of conducting material is required with an ultra-flat surface on which calcium copper titanate can be deposited. Graphene is a good choice for this because it can also act as an electrode and an abutting material, but due to the presence of dangling bonds on its surface it is impossible to deposit the insulating material on it. Thus, to overcome this problem, a layer of hexagonal boron nitride is first deposited over the graphene which acts as a buffer layer for calcium copper titanate which has little structural resistance to graphene. To know the direction of flow of electrons the surface potential measurement of graphene, calcium copper titanate, and FOTS treated calcium copper titanate before and after charge injection is measured by Kelvin PFM. Because of the presence of fluorine functional groups in FOTS treated calcium copper titanate it has a negative charge on the surface while graphene shows a positive charge on its surface. This non-contact TENG with an elastomer spacer between the two materials to control the gap distance between two materials can generate open-circuit voltage of 15.1V and short circuit current of 420nA under the 1kgf vertical force. Table 3 summarize the triboelectric nanogenerator based on 2D materials. Table 3 Comparative summary of the triboelectric nanogenerator based on 2D materials Materials Electrode Fabrication method Device output Application Ref Graphene/PDMS ITO coated PTE CVD, spin coating V oc = 650V I SC = 12μA Power flexible and portable electronics ( Shankaregowda et al., 2016 ) Graphene/PDMS Graphene CVD, spin coating V oc = 47.1V I SC = 7μA Power = 130μW Self-powered wearable electronics ( Chu et al., 2016 ) PVDF-TrFE/Mxenes/Nylon11 Conducting fabric Electrospinning V oc = 270V Current density = 140 mA/m 2 Power density = 4.02W/m 2 Smart home appliances ( Rana et al., 2021 ) PDMS/MoS 2 Aluminum and Si Wafer laser-directed, inkjet printing V oc = 25V I SC = 1.2μA Self-powered flexible sensors ( Park et al., 2020 ) PVA-Mxene/silk fibroin Aluminum Electrospinning V oc = 118.4V Power density = 1087.6 mW/m 2 Real-time monitoring human activity, self-powered electronics ( Jiang et al., 2019 ) rGO/PVDF Aluminum Drop casting technique V oc = 0.35V Energy harvesting ( Kaur et al., 2016 ) Cellulose nanofibrils-phospherene/PET Gold Sputtering V oc = 5.2V Current density = 1.8 μA/cm 2 Energy harvesting ( Cui et al., 2017 ) Nylon11-MoS 2 /PVDF-TrFE- MoS 2 ITO:PET Spin coating V oc = 145V Current density = 350 μA/cm 2 Power density = 50 mW/cm 2 Energy harvesting ( Kim et al., 2019 ) As we know 2D materials are characterized by strong in-plane covalent bonding and weak van der Waals forces among the interlayers which allow us to isolate these layers. Thus, isolating the layers of the different 2D materials and restacking them on one another enables us to develop a new hybrid material with unique customized properties which depend on the sequence of layers without taking into account the crystal lattice mismatching and atomic commensurability. This type of heterostructure is referred to as van der Waals heterostructure which has been widely used in many fields including sensors, photodetectors, energy harvesting devices, and so on( Qiao etal., 2015 ; Rawatet al., 2019 ; Wu etal., 2017b ). Recently, Li et al. reported graphene/copper heterostructure as an active flexible electrode for the TENG fabrication ( Figure 7 (A)) as an alternative of the indium tin oxide (ITO) or noble metals based electrodes ( Li et al., 2020b ). Although copper displays similar performance as that of the commonly used electrodes, its chemical stability is a major concern because it can be oxidized to form copper oxide. Therefore, to improve the chemical stability and chemical reduction in the electrodeposited copper layer, graphene dispersion prepared by physical exfoliation is spin-coated on the surface of copper where the graphene layer acts as an energy barrier, preventing the copper from oxidizing ( Figure 7 (B)). Thus, a skin-based TENG is fabricated using the graphene/copper heterostructure as a flexible induction electrode and PDMS as the active material that can power 22 LEDs by finger typing. To demonstrate its application in wearable devices, the output voltage is measured by attaching it to different parts of the body as shown in Figures 7 C–7E. Furthermore, the sheet resistance of this flexible electrode is shown by the bending measurements ( Figure 7 (F)), and to ensure the reliability of TENG bending and folding measurements are done for 10,000 cycles which shows no significant changes in voltage and transfer charge quantity ( Figures 7 G–7H). Also, Nutting et al. used the mechanical abrasion technique to produce the thin film of the multilayer heterostructure by using different 2D materials ( Nutting et al., 2020 ) and showed that when the graphite/MoS 2 heterostructure is used as a TENG electrode instead of abraded graphite, it can significantly enhance the current up to 50% and produce a peak power of 5.7μW because of the insertion of charge trapping layer of MoS 2 . Figure 7 (I) depicts the charge transfer process of TENG via continuous tapping and releasing cycles, whereas Figure 7 (J) depicts the resulting open circuit voltage and short circuit current generated by this device. The energy produced by this nanogenerator is stored in a capacitor with the help of a bridge rectifier whose circuit diagram is shown in Figure 7 (K) and used to power up the LEDs. Although, there are a few reports available where heterostructure formed by the 2D material is used as an electrode but no reports are available where the heterostructure acts as an active triboelectric layer to harvest the mechanical energy. Plenty of materials are used to fabricate the TENG, among which the most explored are the polymer materials. Polymers such as polytetrafluoroethylene (PTFE), polyvinylidene fluoride (PVDF), and fluorinated ethylene propylene (FEP), which contain fluorine, have always been used as negative triboelectric materials owing to the strong attractive force of fluorine elements and polymers with electron donor groups such as nylon, wool, and silk are used as positive triboelectric materials. These polymers provide several merits in terms of their flexibility, scalability, stretchability, lightweight, and processability, but have poor mechanical properties and a low melting temperature which limits their application in the high temperature range. On the other hand, the 2D materials are known for their superior mechanical properties. As a result, the addition of even a small amount of 2D material in the polymer matrix can significantly boost its mechanical properties ( Kim etal., 2014a ; Papageorgiouet al., 2017 ). MXenes is considered a suitable alternative for the electronegative friction material in TENG because of the presence of the –F group. By using the highly electric conducting and electronegative properties of MXenes, Jiang et al. fabricated a flexible TENG where triboelectric pairs of MXenes film and silk film are synthesized by the electrospinning technique ( Figure 8 (A))( Jiang et al., 2019 ). To further enhance the flexibility, MXenes nanosheets are mixed in the polyvinyl alcohol (PVA) aqueous solution and electrospinned at 18kV voltage with a delivering rate of 18μLmin -1 . Moreover, to optimize value-added fraction of MXenes on the performance of TENG, different concentrations of the MXenes nanosheet are introduced in PVA nanofiber which shows that at volume fraction 30% MXenes can generate a maximum output voltage of 117.7V, and afterward output voltage increases at a slow pace as shown in Figure 8 (B). This all-electrospun TENG is further investigated with different contact frequency, applied force, achieving a maximum power density of ∼1087.6 mW/m 2 at 5MΩ input resistance. Figures 8 C–8F() shows the practical applicability of the device by locating the PVA/MXenes based TENG on the finger, wrist, knee, and throat for real-time monitoring of human activities without the need for external power supply. Reduced graphene oxide (rGO) also exhibits interesting properties and can be used to enhance the negative charge on the surface of the friction material because of the presence of oxygen-containing functional groups which are highly electronegative in nature ( Huang et al., 2011 ). By using this, Navjot et al. fabricated an arc-shaped single electrode TENG by using rGO nanorods (rGONRs)/PVDF film and aluminum foil ( Kaur et al., 2016 ). Also, the contact resistance of pristine rGONRs and rGONRs/PVDF composite is measured by a two-probe resistivity measurement method which shows an increase in the contact resistance in the nanocomposite. Furthermore, the charge storage and transport capabilities are examined with the cyclic voltammetry analysis where rGONRs/PVDF composite shows better results compared to pristine rGO. Among 2D materials, phosphorene is known for its high carrier mobility∼ 286cm 2 V −1 s −1 , anisotropic photoelectronic properties and non-zero bandgap that can be tuned with the number of layers ( Bagheriet al., 2016 ; Liu et al., 2014a ; Xia et al., 2014 ). However, the stability of the phosphorene is the major concern as it oxidizes under ambient conditions ( Wood et al., 2014 ). To prevent the oxidation, Cui et al. mixed the exfoliated phosphorene with tempo-oxidized cellulose nanofibrils to form a hybrid paper ( Cui et al., 2017 ). By using the hybrid paper as an active triboelectric layer and a thin layer of gold as an electrode, a TENG is fabricated which can generate an open circuit voltage of 5.2 V with 1.8μAcm −2 current density which is higher than cellulose-based TENG. Also, as we know, the triboelectric surface charge induces the opposite charge on the interface of the counter electrode, and owing to the presence of an electric field, the charge on the surface are drifted and combined with the opposite charge present on the interface which drastically affects the accumulation of the charge on the surface and decreases the triboelectric potential. This combination of charges can be blocked by introducing the charge trapping layer and the 2D materials are good candidates for it ( Wu et al., 2019a ). Thus, incorporating 2D materials into the triboelectric layer aids in increasing the effective surface area that captures and stores charge in the layer formed by 2D materials, thereby suppressing charge combination between the electrode and the triboelectric layer, resulting in improved triboelectric charge density and output performance of the nanogenerator. For example, Wu et al. reported that the introduction of MoS 2 monolayer in the polymer matrix ( Figure 8 (G)) can provide the charge trapping sites in TENG because of its appropriate energy level, quantum confinement effect, and large specific surface area, as a result preventing the combination of the charge to tune the output performance of TENG ( Wu et al., 2017a ). The rectified voltage of the TENG with or without filling monolayer MoS 2 is also measured which shows that a maximum voltage of 400 V can be reached by optimizing the filler ratio of MoS 2 while without MoS 2 only 30V is achieved by TENG. The short circuit current density also shows similar trends where the current density produced by MoS 2 monolayer is high as compared to TENG fabricated without MoS 2 as shown in Figure 8 (H). The power density of the TENG is also enhanced up to 120 times as compared with the TENG without the MoS 2 layer. These findings link a significant improvement in TENG electrical output to an effective electron capture process in monolayer MoS 2 , which not only has large specific surface areas but also can prevent the loss of triboelectric electrons when they are created. Kim et al. also demonstrated the enhancement in the current density by incorporation of the MoS 2 flakes in the ferroelectric polymer where Nylon/MoS 2 composite serves as triboelectric positive layer and PVDF-TrFE/MoS 2 composite as a triboelectric negative layer( Kim et al., 2019 ). The surface charge density of these composite layers is further enhanced by the electrical polling process because of the ferroelectric nature of the polymer. The TENG fabricated by these composite films can generate a power density of 50 mW/cm 2 . Xiong et al. reported a skin-touch-actuated washable textile TENG using black phosphorus in which nanosheets of black phosphorus are first encapsulated in the hydrophobic cellulose oleoyl ester nanoparticles (HCOENPS) and then the layer is knitted on the PET fabric by dip coating to fabricate the HBP fabric as one of the friction layers in TENG. Moreover, to make the device deformable, waterproof and insulating for safe use fabric electrodes are made by coating the PET fabric with a paste made by mixing silver flakes in PDMS, and then this electrode is encapsulated by HCOENPs dipped PET fiber. This textile-based wearable TENG is shown to harvest energy from voluntary and involuntary body movements, output voltage achieved by voluntary touch is 880V at ∼ 4Hz frequency and ∼5N force, and involuntary friction with skin can generate an output voltage of 60 V( Xiong et al., 2018 ). Figure 8 Polymer-2D materials based Triboelectric nanogenerator (A) Fabrication of the MXene doped PVA nanofibers and silk nanofibers film by electrospinning technique. (B) The output voltage of TENG with different concentrations of MXenes nanosheets in PVA nanofiber. (C–F) Real-time monitoring of the human activity by attaching TENG at different regions of the body ( Jiang et al., 2019 ). (G) A schematic of TENG with a MoS 2 as an electron-accepting layer. (H) The effect of the introduction of MoS 2 on the short-circuit current density and charge generation ( Wu et al., 2017a ). (I) Image of the conformal TENG attached to hand, accompanied by its magnified and SEM image. (J) The output voltage is generated by conformal TENG as a function of a number of fingers tapping. (K) A schematic diagram of an assistive communication system derived by conformal TENG( Chu et al., 2016 ) By enhancing the triboelectric characteristics of graphene with a surface texturing and plasma treatment, Chu et al. designed an ultrathin TENG which offers good adherence to the human skin ( Figure 8 (I)) ( Chu et al., 2016 ). The output performance of the TENG is increased from 3.4 to 130 mW by using a nanostructured surface with increased contact area because of plasma treatment and fluorinated nanostructured surfaces with higher fluctuations in electron affinity. A self-powered assistive communication system ( Figure 8 (K)) is also developed by using this conformal TENG for speech impaired people to translate the finger touch into language . The number of fingers touching the hand is detected and converted to Morse code by employing conformal TENG’s triboelectric performance, which is influenced by the effective contact area and electron affinity. Figure 8 (J) shows the open-circuit voltage (Voc) of the conformal TENG as a function of the number of fingers tapping on it at a contact pressure of 30–40 kPa. As the effective contact area has changed, different values of voltage (4.1, 8.3, and 12.5 V) are produced. Subsequently, a series of electric impulses is generated with a variety of finger contacts, which are recorded in voltammeter, transformed into corresponding characters, and transmitted to a smartphone. By optimizing the critical voltage separating dots and dashes and developing an enhanced classification algorithm, four different contact sequences to express the word “LOVE”, corresponding to Morse code were successfully transmitted to the smartphone, demonstrating its potential application as an assistive communication device to assist persons with impairment to express their thoughts more readily. These results showcase the potential application of conformal 2D materials based TENG in human-machine interfaces and assistive devices. Hybrid energy harvesters based on 2D materials The piezoelectric nanogenerator requires deformation of the material by application of mechanical stress (or strain) to harvest mechanical energy while, on the other hand, the triboelectric nanogenerator based on contact electrification needs continuation touching and separation between the triboelectric layers. Thus, by integration of these two effects, the output of the nanogenerator can be enhanced effectively. By using this approach, Sahatiya et al. reported a hybrid nanogenerator based on a paper by coupling of piezoelectric and triboelectric effect ( Sahatiya et al., 2018 ). The hybrid nanogenerator constitutes two parts; the top part is fabricated by growing MoS 2 layers on cellulose paper via hydrothermal technique followed by PVDF deposition on both sides, as PENG which is connected parallel to polyimide as the bottom part, serve as TENG. To measure the combined output of the PENG and TENG, two bridge rectifiers are used that can generate the open-circuit voltage of 50 V, short circuit current of 30nA with 0.18 mW/cm 2 average power. The sensitivity and the response of this hybrid nanogenerator to shape are demonstrated by writing different letters on it with the help of a stylus pen that shows different voltages corresponding to different alphabets, which depict their application in digital signature verification, security, Internet of things (IoT), etc. Lee et al. reported a highly stretchable hybrid nanogenerator based on polymer and carbon-based electrodes which utilizes Piezoelectric–Pyroelectric for energy harvesting ( Lee et al., 2014 ). Figure 9 (A) shows a schematic image of a hybrid nanogenerator(HNG) which mainly comprises three parts: micropatterned P(VDF-TrFE) as piezoelectric and pyroelectric material, micropatterned PDMS-CNT composite as bottom electrode, and graphene as top electrode for the fast thermal gradient. The compatibility of the HNG is demonstrated by attaching it with different body parts ( Figure 9 (B)) which shows its applicability in wearable, biomedical, and robotic sectors. The performance of the device is evaluated by measuring piezoelectric, pyroelectric output voltage separately, as well as after coupling of piezoelectric-pyroelectric effect under stretching-releasing/compressing-releasing cycles and for heating-cooling of HNG as illustrated in Figure 9 (C). Furthermore, the working mechanism for the integration of piezoelectric and pyroelectric effect is explained based on the polarization and electric dipole oscillation in the P(VDF-TrFE). Figure 9 2D materials based Hybrid nanogenerator (A–C) Schematic depicting a highly stretchable piezoelectric-pyroelectric hybrid nanogenerator affixed to several parts of the body with their output voltage in the stretch and release conditions( Lee et al., 2014 ). (D) The open-circuit voltage of PDMS/MXene based TENG as a function of frequency and sound pressure level. (E) The open-circuit voltage of PDMS/MXene based TENG with respect to different wind speeds. (F) A schematic illustration of PDMS/MXene based TENG to harvest multiple energy for developing a self-powered environmental visualization system( Liu et al., 2021c ). (G) A photoelectrochemical hydrogen system powered by TENG. (H and I) A schematic illustration of RD-TENG with an SEM image of PTFE structure in the inset (I) Current measurements at various rotational speeds under darkness or illumination ( Wei et al., 2020 ) A single fiber-based power system was fabricated by Bae et al. by coupling the energy harvesters with the energy storage unit along single polymethylmethacrylate (PMMA) fiber by utilizing ZnO nanowires (NWs) and graphene as electrodes ( Bae et al., 2011 ). The system mainly comprises three parts: a nanogenerator, a dye-sensitized solar cell (DSSC) to harvest mechanical and solar energy simultaneously, and a supercapacitor as an energy storage unit. Here, the ZnO NWs grown by the chemical method are used as active material in nanogenerators, as well as a core unit for DSSC and supercapacitors. The maximum current and open-circuit voltage generated by the nanogenerator is 2nA and 7mV. A floor-tile-based hybrid triboelectric and electromagnetic nanogenerator is reported by Islam et al. which transforms the mechanical energy generated from human footsteps into electrical energy ( Islam et al., 2020 ). A thin electron acceptor layer of MoS 2 is inserted between the two kapton layers and conductive aluminum is used as the two oppositely charged layers in the TENG which hybridized with EMG consist of neodymium ferromagnets and copper coils. The performance of the TENG-EMG hybrid tile was analyzed with different stepping frequencies 60, 90, and 120 BPM (beat per minute) of a footstep. Furthermore, the sensitivity of the hybrid tile is tested with different loads which depict that as the loading weight on the tiles doubles, the output voltage doubles, and the current increases by 0.2–0.4mA respectively. The hybrid energy harvester device can produce power up to 5W with 1200V peak voltage and 5mA the short circuit current which is much higher than the commercially sold Pavegan tiles which utilize piezoelectric and electromagnetism to scavenge the energy. This prototype represents a cost-efficient, sustainable and effective approach used for harnessing mechanical energy and makes it a promising candidate for large-scale implementation of these tiles by flooring in public areas. Recently, Liu et al. proposed a single structural layer-based multifunctional triboelectric nanogenerator that can simultaneously harvest mechanical and light energy ( Liu et al., 2021c ). This highly stretchable hybrid TENG is mainly composed of the PDMS/MXenes as an active triboelectric material and silver nanowire (Ag NW) as an electrode. Owing to the highly electronegative surface and outstanding surface plasmon-assisted hot-electron property of MXenes, this hybrid TENG can generate an open-circuit voltage of 453V and short-circuit current of 131μA under the light illumination with the light power conversion efficiency of 19.6%. Moreover, to demonstrate the stretching ability of the device, the output performance of TENG is investigated at different stretching strain levels, and a maximum stretchable ability of 123% is achieved for PDMS/MXene based TENG. Also, the sensitivity of the device as a function of the effect of the acoustic and wind energy is systematically investigated by examining the performance of TENG as a function of frequency, sound pressure level, and wind speed as shown in Figures 9 D and 9E(). Furthermore, to showcase its potential in human-machine interface, an environmental interaction visualization system is devolved by using TENG and color-tunable LED matrix ( Figure 9 F). Wei et al. reported a self-powered photoelectrochemical hydrogen production system ( Figure 9 (G)), using WO 3 /BiVO 4 heterojunction as photoanode to obtain the hydrogen which is driven by mechanical and solar energy ( Wei et al., 2020 ). For harvesting the mechanical energy, a rotational disc-shaped TENG(RD-TENG) ( Figure 9 (H) was fabricated which plays a dual function by harvesting mechanical energy from water and acting as an external bias for driving the PEC cell after transforming and rectification the output generated by RD-TENG. The current trend of hydrogen production systems under dark or illumination conditions is demonstrated in Figure 9 (I) ranging from 60 to 140rpm. Furthermore, it has been observed that when the rotation rate is 60 rpm, the hydrogen can only be produced with sunlight irradiation but as the rpm increases PEC cells can produce hydrogen even in dark conditions. Also, when the rotation rate is set at 160 rpm the hydrogen production rate increases from 5.45μLmin -1 to 7.27μLmin -1 under dark and illuminated conditions with conversion efficiencies of 2.43 and 2.59%, respectively. These results show that using TENG a self-powered PEC system can be made for hydrogen production without the need for external bias."
} | 21,239 |
35864984 | PMC9294628 | pmc | 161 | {
"abstract": "Neuromorphic systems aim to provide accelerated low-power simulation of Spiking Neural Networks (SNNs), typically featuring simple and efficient neuron models such as the Leaky Integrate-and-Fire (LIF) model. Biologically plausible neuron models developed by neuroscientists are largely ignored in neuromorphic computing due to their increased computational costs. This work bridges this gap through implementation and evaluation of a single compartment Hodgkin-Huxley (HH) neuron and a multi-compartment neuron incorporating dendritic computation on the SpiNNaker, and SpiNNaker2 prototype neuromorphic systems. Numerical accuracy of the model implementations is benchmarked against reference models in the NEURON simulation environment, with excellent agreement achieved by both the fixed- and floating-point SpiNNaker implementations. The computational cost is evaluated in terms of timing measurements profiling neural state updates. While the additional model complexity understandably increases computation times relative to LIF models, it was found a wallclock time increase of only 8× was observed for the HH neuron (11× for the mutlicompartment model), demonstrating the potential of hardware accelerators in the next-generation neuromorphic system to optimize implementation of complex neuron models. The benefits of models directly corresponding to biophysiological data are demonstrated: HH neurons are able to express a range of output behaviors not captured by LIF neurons; and the dendritic compartment provides the first implementation of a spiking multi-compartment neuron model with XOR-solving capabilities on neuromorphic hardware. The work paves the way for inclusion of more biologically representative neuron models in neuromorphic systems, and showcases the benefits of hardware accelerators included in the next-generation SpiNNaker2 architecture.",
"introduction": "1. Introduction A vast array of brain modeling techniques exist to simulate brain activity with a view to gaining understanding of the human brain. These techniques range from mathematical representations of individual molecules within neurons to whole-brain simulations. One widely used method for simulation of brain activity is through the use of neural networks. Spiking Neural Networks (SNNs) use biologically-inspired models of neurons to carry out computation with the aim of simulating neural activity and have applications in a number of research areas including computational neuroscience, machine learning, and robotics. State-of-the-art large scale SNN simulations such as those described by the Blue Brain Project (Markram et al., 2015 ) aim to mimic brain activity through the use of complex neuron models to advance understanding of the human brain. Scientists were able to accurately reproduce anatomical and physiological features of real biological networks when simulating 0.29 mm 3 of the rat brain. Despite these recent achievements in the complexity and scale of SNNs simulated, simulation of these SNNs consumes considerable power (megawatts) for simulation of very small regions of the brain (Markram et al., 2015 ). The full simulation involved over 30,000 different neuron models incorporating 13 different ion-channel models, each neuron comprised on average 20,000 differential equations representing synaptic connections and ion-channels. The full simulation required solving of over two billion equations for every second of biological time (Kumbhar et al., 2019 ). This power consumption is not required in the human brain which demonstrates a remarkable ability for large amounts of fine-scale computation at a fraction of the power (up to 20 watts) and much faster than SNNs simulated on conventional computer hardware which do not run in real-time (Cox and Dean, 2014 ). Energy-efficient neuromorphic systems are designed to mimic the brain and provide low-power platforms for simulation of SNNs, providing a potential solution for the high energy requirements of large scale simulations. As neuromorphic computing platforms target real-time large-scale simulations of SNNs, the biological plausibility of neuron models has been largely ignored in favor of simple, efficient neuron models such as the Leaky-Integrate-and-Fire (LIF) neuron model. Such models are favored due to the ease of solving the equations involved: the differential equations can be solved exactly with a small number of addition and multiplication operations. These simple neurons have allowed large-scale SNNs in real-time such as the cortical microcircuit simulated on SpiNNaker (Rhodes et al., 2020 ). This SNN simulated ≈1mm 2 of mamillian neocortex, and while this demonstrated the potential of neuromorphic hardware as a neuroscience research tool, the model does not exhibit the fidelity typically explored by the neuroscience research community. The LIF model falls short of biological plausibility in two main areas: membrane conductance and structure. Membrane conductance is described with a single term in the model but is actually governed by a number of different ion-channels spanning the neural membrane. The flow of ions through these channels gives biological neurons a wide range of firing capabilities not captured with the LIF model, e.g., the ability to respond to identical inputs differently depending on the current state of the neuron and its ion-channels. Structure is simplified in the LIF model to a single point, however in biology, neurons are complex and elongated and incorporate vast branched extensions called dendrites. Dendrites are active structures capable of generating their own action potentials and are believed to contribute significant computational function to biological neurons (Dayan and Abbott, 2005 ; Poirazi and Papoutsi, 2020 ). Neuron models can increase in complexity to capture these simplified biological features and a wide range of spiking neuron models exist. Hodgkin and Huxley ( 1952 ) described a biologically inspired model incorporating equations for sodium and potassium ion channels which govern the progression of the action potential. Other models, such as the Izhikevich model (Izhikevich, 2004 ), aim to capture certain biological characteristics with more efficient non-biologically plausible equations. However, this lack of biological plausibility takes away the ability to explore the effects of incorporating different ion-channels and more complex morphologies than a single point neuron structure. Accurate and efficient ion-channel modeling on neuromorphic hardware would therefore allow exploration of a wide range of biologically inspired models including multi-compartment models describing complex neural morphologies with dendritic compartments (Markram et al., 2015 ; Gidon et al., 2020 ). Implementation of more complex neuron models onto neuromorphic systems could provide low-power solutions for large-scale SNN simulations. Neuron models with increased complexity have been tested in analog and digital neuromorphic systems, demonstrating the importance of this kind of modeling. For example, individual ion-channels have been modeled in an analog circuit (Abu-Hassan et al., 2019 ). Here, the aim was to design a biologically accurate neuromorphic circuit that responds identically to a biological neuron under any injected current. The authors were able to reproduce biological voltage recordings with 94–97% accuracy. These neurons were built to demonstrate the potential for making synthetic neurons with therapeutic potential for implementation into the central nervous system, therefore do not easily scale up to large SNNs and do not incorporate structural morphology. However, this work demonstrates accurate representation of ion-channel models on neuromorphic systems. Multi-compartment neuron models have also been tested on neuromorphic systems. BrainScalesS (Schemmel et al., 2010 ) is an analog neuromorphic system that features an Adaptive-Exponential Integrate-and-Fire (AdExp) neuron model. Schemmel et al. ( 2017 ) and more rec ently Müller et al. ( 2022 ) and Kaiser et al. ( 2022 ) expanded this neuron model to capture dendritic computation in multi-compartment approaches. Intel's Loihi (Davies et al., 2018 ) also offers support for dendritic computation by offering the opportunity to model neurons with multiple compartments. Here, the additional compartments are effectively identical, the only difference being that the “somatic” compartment generates spike output and the “dendritic” compartments do not. While this does enable a concept of dendritic computation through the ability to distribute synaptic input across individual units, there is a lack of biological plausibility as dendrites are actually much more computationally complex, exhibiting non-linear processing of synaptic inputs (Gidon et al., 2020 ; Poirazi and Papoutsi, 2020 ). 1.1. Neuromorphic Hardware While a range of neuromorphic computing systems are currently developed across industry and academia (Schemmel et al., 2010 ; Benjamin et al., 2014 ; Furber et al., 2014 ; Merolla et al., 2014 ; Davies et al., 2018 ; Pei et al., 2019 ), the application of this technology remains limited. While these systems boast impressive performance figures in terms of energy and processing speed, their bespoke architectures are often tailored to particular applications, making it hard to adapt these systems to emerging research problems. The SpiNNaker neuromorphic computing system is selected as the research platform for this work, as its flexibility enables exploration of the target neural models, while constraints such as co-location of memory and processors mean findings remain relevant for the wider neuromorphic research community. The SpiNNaker system is currently an active research platform, with a 1M core machine operating and maintained by the University of Manchester, UK. In parallel to exploring SNN applications on this system, research and development into next-generation hardware is also on-going in the form of the SpiNNaker2 system (Mayr et al., 2019 ). The two platforms are explored in this work, implementing models on both the SpiNNaker system and a SpiNNaker2 prototype chip (Jib2), to enable comparison and evaluation of performance. 1.1.1. SpiNNaker SpiNNaker is a massively-parallel many-core digital computing platform, designed for large-scale real-time simulation of SNNs. The system comprises chips assembled into a two-dimensional mesh network, enabling the system to scale to 1M cores. Each individual chip houses 18 cores, network on chip (NoC) and external RAM controller; while each core contains an ARM968, direct memory access controller, communications controller, two timers and other peripherals. Each core has 32kB instruction and 64kB data tightly coupled memory (ITCM and DTCM, respectively), with single cycle access. Each chip has an additional 128MB shared memory, typically accessed via DMA, and used to store larger SNN data-structures such as synaptic matrices. Cores operate at 200MHz, running an event-driven operating system enabling efficient neural processing (Rhodes et al., 2018 ). Individual cores simulate a collection of neurons using software to solve mathematical models representing neural dynamics. These models are solved in discrete time, with the goal of matching the simulation timestep to the time required to process the state update, in order to achieve real-time simulation. Models are programmed in C, and compiled into ARM code using the GCC toolchain. As the core has no floating-point unit, all models are coded using fixed-point arithmetic, with the ISO standard accum type favored for the majority of variables. This 32-bit type is a signed representation, with 16 integer and 15 fractional bits, and lower/upper limits of 0.000030517578125 and 65535.999969482421875, respectively (Rhodes et al., 2018 ). While transcendental functions are also not supported in hardware, division and exponential functions are available in software, requiring approximately 100 clock cycles each. This framework enables real-time implementation of multiple neuron models, including the current- and conductance-based LIF and Izhikevich neurons (Rhodes et al., 2018 ). 1.1.2. Jib2—SpiNNaker2 Prototype While the architectural principles are similar, the goals of SpiNNaker2 are to increase the number of cores by a factor of 10, and to increase the number of simulated neurons by a factor of 50, while staying within the same power budget. The system will use an ARM cortex M4 core, with adaptive body biasing to enable increased clock frequencies during periods of high load—switchable from 150 to 300MHz. Additional performance increases are expected from inclusion of hardware accelerators for specific operations common in neural processing, including random number generation, e x , and a single-precision floating point unit (Mikaitis, 2020 ). The experiments reported in this work are performed on a SpiNNaker2 prototype system known as Jib2, containing 8 processing elements (PE) arranged in two quad processing elements (QPEs) (Höppner et al., 2021 ). Each PE has an ARM cortex M4 in addition to the above mentioned accelerators, and runs compiled C code in a similar fashion to SpiNNaker (Section 1.1.1), again compiled with the GCC toolchain. PEs each have 128kB of fast access SRAM, for combined instruction and data storage. Jib2 has variable voltage-frequency levels enabling low-power operation and workload-dependent scaling of clock frequency. The experiments reported in this work are performed with the core running with voltage-frequency settings of 0.5V–150MHz and 0.8V–300MHz. 1.2. NEURON Simulation Environment New models implemented on neuromorphic hardware need to be benchmarked again standard methods used in the industry in order to ensure the models are accurate and valid. NEURON is a widely used platform for simulation of individual neuron models and networks of neurons and was designed specifically to simulate equations describing nerve cells. NEURON was chosen as the benchmark for models as it is a standard tool in the research field. It provides an environment for implementing biologically realistic models with a focus on incorporation of multiple ion-channel models and complex branched neuronal morphologies (Hines and Carnevale, 1997 ). The activity of neurons is modeled using the cable equation in which neurons are treated as trees consisting of a number of compartments. Each compartment is an unbranched cable which can be split into sections and each section can contain its own biophysical properties through different ion-channels. Each section is described by its membrane potential and a set of coupled differential equations are solved for each section within a neuron to compute the evolution of membrane potential inside the neuron over time. The general form of the cable equation for each section, j is: \n (1) \n c j d v j d t + I m j = ∑ k v k - v j r j k \n where c j is the membrane capacitance of the section, v j is the membrane voltage of the section, t is time, the ionic component I m j includes all currents through ion-channels. ∑ k v k - v j r j k represents the sum of axial currents entering from neighboring sections, v k is the membrane voltage of the neighboring section and r jk is the resistive coupling between compartments. This differential equation is coupled to an additional set of differential equations describing the active states of any ion-channels incorporated into the model. This leads to a set of coupled differential equations which need to be solved at each simulation time step. NEURON uses a backward Euler implicit integration method as standard (Hines and Carnevale, 1997 ). Each time step update is divided into a set of operations which are performed in order to progress from one time step to the next. These operations include a spike delivery step where synapses are activated by incoming spikes, a matrix assembly step where the ionic and synaptic currents are calculated, a matrix resolution step in which the membrane potential is calculated, a state variable update step in which the ion-channel states are updated, and a threshold detection step in which membrane voltages are checked against threshold values to determine whether a firing condition has been met (Kumbhar et al., 2019 ). The NEURON platform was designed specifically to model systems of neurons incorporating easy to configure biological data (branched morphologies and ion-channel models) and is therefore widely used by the computational neuroscience community.",
"discussion": "5. Discussion This work has provided the first fixed-point implementation of ion-channel, Hodgkin-Huxley, and multi-compartment models on SpiNNaker neuromorphic hardware and the first profiling for both speed and accuracy of such models on SpiNNaker2 prototype neuromorphic hardware, demonstrating the improved performance of the next-generation system through the use of hardware accelerators and floating point arithmetic. The first demonstration of a two-compartment neuron model running on neuromorphic hardware that can solve the XOR problem using a single neuron is also presented through this work. Neuromorphic systems are designed to provide low-energy platforms for simulation of Spiking Neural Networks (SNNs) but in doing so biologically plausible neuron models have largely been ignored in favor of simple and efficient neuron models such as the Leaky Integrate-and-Fire (LIF) model. In contrast, focus in the computational neuroscience community has been on building models with a high degree of biological accuracy which are in turn accompanied by large computational costs, making the models difficult to scale into SNNs. This work bridges this gap by presenting two biologically inspired neuron models ( Figure 1 ), implemented efficiently and accurately on SpiNNaker and Jib2 neuromorphic platforms ( Figure 5 ): a single compartment Hodgkin-Huxley (HH) neuron ( Figure 2 ) and a multi-compartment neuron incorporating dendritic computation ( Figure 4 ). Both SpiNNaker and Jib2 are able to accurately model both neurons over time with identical spike times recorded on Jib2 and a reference model in NEURON (Hines and Carnevale, 1997 ) and spike times within 0.1ms on SpiNNaker. Manipulation of equations, pre-calculation of constants and the use of lookup tables (using 12 and 13.6kB of memory for HH and two-compartment models, respectively) enabled a significant speed up of simulation time of the models (approx 11× for both the single and two-compartment models— Table 1 ). This speed-up is further increased by 3× with implementation on the next-generation Jib2 neuromorphic chip, demonstrating the effectiveness of hardware accelerators for expressions such as exponential operations ( Table 1 ) when simulating biologically representative neurons. Comparison with neuromorphic implementations of the conventional LIF neuron model revealed that both the HH and the multi-compartment neurons were slower to simulate on neuromorphic hardware, due to the increased complexity of the models ( Table 1 ). However, the computational capabilities gained justify the increased expense of running the model, and the model on Jib2 is within an order of magnitude of the LIF neuron in terms of computation time. The underlying ion channel models directly correspond to biophysiological data, bringing increased biological relevance to models simulated on neuromorphic hardware. Furthermore, the presented HH model exhibits a wide range of firing characteristics which cannot be captured with LIF neurons ( Figure 6 ), and the inclusion of a dendritic compartment enables the a single neuron model to function as a multi-layer network. The multi-compartment model provides the first implementation of a single neuron model capable of solving the XOR problem on neuromorphic hardware ( Figure 7 ). This work has explored simulation and profiling of individual neurons, and their realization on neuromorphic systems. The ultimate goal of implementing these models is harnessing their ability to capture biologically representative features in large-scale SNNs, and opening up new applications in bio-inspired AI. To understand how the presented models would scale when included in large networks, it is useful to contrast performance with LIF neurons and biologically representative neural circuits previously evaluated on SpiNNaker. In previous work modeling cortical microcircuits comprising LIF neurons, it was shown that neuron and synapse processing could be parallelized effectively on multicore architectures such as SpiNNaker (Rhodes et al., 2020 ). Through this parallelization real-time simulation of cortical circuits containing 80k neurons and 300M synapses was demonstrated, with an energy per synaptic event of ≈ 0.6 μJ. The models presented here would impact the neuron processing, resulting in a ≈ 40× reduction in neuron density relative to LIF neurons to accommodate the increased model complexity. This indicates that approximately 40× more SpiNNaker chips would be required to simulate the same size of model, leading to the same factor increase in total energy consumption. Projecting these numbers on to SpiNNaker2 requires consideration of the updated performance achieved with the new hardware. The HH and two-compartment neurons occupy 48 and 55kB, respectively (of the 128kB fast-access SRAM for combined instruction and data storage on Jib2) with the instructions to update the neuron and the storage of constants, variables and LUTs. Increasing the number of neurons does not significantly increase the storage requirements, as the instructions for updating the neurons are the same and all neurons share common LUTs. While the number of neurons per core determines the amount of state variables to be stored, these datastructures are relatively small compared to those described above (assuming split neuron and synapse processing/storage as described above, Peres and Rhodes, 2022 ). Therefore the determining factor in the number of neurons which can be simulated on each core is the processing time. As it takes 0.73 μs to update a HH neuron and 1.09 μs for the two-compartment neuron using a 0.1 ms simulation timestep, assuming the goal of real-time simulation, an upper limit of 136 HH neurons or 91 two-compartment neurons could be updated by a single core while maintaining real-time execution. In reality this number is likely to be reduced to enable cores to perform auxilliary operations such as monitoring and data recording, reducing overall neuron density. However, this is likely to remain above the 64 neurons per core utilized in previous cortical simulations on SpiNNaker (Rhodes et al., 2020 ), enabling real-time cortical simulations containing biologically representative ion-channel-based neuron models (on SpiNNaker2). Furthermore, embedding these models within the SpiNNaker routing and communications fabric should facilitate further expansion of model sizes while maintaining real-time execution. This indicates that the cost of changing from LIF to multicompartment models on SpiNNaker2 will incur a 10× increase in energy, with the overall system significantly more energy efficient—LIF neurons have been profiled at 20pJ per synaptic event (Höppner et al., 2021 ). The model provides a framework for capturing and testing more biologically plausible neural dynamics in an efficient way. For example, different ion-channels can easily be substituted or added to the model, and more complex morphologies can be captured through inclusion of more dendritic compartments. Recent work has demonstrated the potential of multi-compartment neuron models to learn via a synaptic learning rule (Bicknell and Häusser, 2021 ), opening the door to the possibility of training the neuron models presented in this work within large-scale SNNs on neuromorphic hardware, in particular those featuring hardware accelerators to maximize efficiency. Significant computational capabilities are gained with each individual neuron model and neuromorphic architectures can provide energy-efficient platforms for simulations. While this work has focused on demonstrating feasibility through development of software models suitable for execution on SpiNNaker, the developed models also provide the first step toward algorithm-hardware co-design. Hardware requirements such as arithmetic operations and memory use have been identified, providing insights into how future neuromorphic systems could be tailored to further optimize execution."
} | 6,152 |
26213412 | null | s2 | 162 | {
"abstract": "In collectively foraging groups, communication about food resources can play an important role in the organization of the group's activity. For example, the honeybee dance communication system allows colonies to selectively allocate foragers among different floral resources according to their quality. Because larger groups can potentially collect more information than smaller groups, they might benefit more from communication because it allows them to integrate and use that information to coordinate forager activity. Larger groups might also benefit more from communication because it allows them to dominate high-value resources by recruiting large numbers of foragers. By manipulating both colony size and the ability to communicate location information in the dance, we show that larger colonies of honeybees benefit more from communication than do smaller colonies. In fact, colony size and dance communication worked together to improve foraging performance; the estimated net gain per foraging trip was highest in larger colonies with unimpaired communication. These colonies also had the earliest peaks in foraging activity, but not the highest ones. This suggests they may find and recruit to resources more quickly, but not more heavily. The benefits of communication we observed in larger colonies are thus likely a result of more effective informationgathering due to massive parallel search rather than increased competitive ability due to heavy recruitment."
} | 369 |
35181602 | PMC8872774 | pmc | 163 | {
"abstract": "Significance Kelp forests are declining worldwide due to varied combinations of environmental change and the trophic downgrading of urchin-controlling predators. These processes have increased the frequency and extent of rapid, nonlinear shifts to so-called urchin barrens whose ecological functioning and services are reduced relative to those of kelp forests. Understanding the factors that regulate kelp-forest tipping points and switches between states is key to their management. Here we demonstrate that substrate complexity (surface rugosity) determines both the existence of and dynamic transition between community states around San Nicolas Island, CA. Kelp-forest conservation and restoration efforts are growing internationally and may benefit from the consideration of substrate complexity in their strategies.",
"conclusion": "Conclusions. The processes and feedbacks that associate with substrate complexity undoubtedly extend well beyond those that we have discussed. For example, high-complexity transects are more species rich and exhibit a greater coverage of foliose red algae and sessile invertebrates than do low-complexity transects ( 35 ). As such, our results add to a rich ecological literature detailing the many means by which physical and biological complexity can modify species coexistence and the dynamics and functioning of ecological communities ( 54 – 57 ). It nonetheless remains an open question how globally widespread the importance of substrate complexity is, as changes in kelp-forest state certainly do occur irrespective of substrate complexity, especially at higher latitudes ( 4 , 58 – 60 ). We speculate that many of these large-scale changes in kelp-forest state are driven by phase shifts rather than switches between alternative stable states. For example, in the Northeast Pacific, phase shifts between forested and urchin-barren states are driven by changing environmental conditions, specifically the presence or absence of sea otters ( 3 , 61 ). Urchin predator diversity and environmental conditions that influence urchin recruitment and population structure also vary with latitude and across the globe ( 16 , 50 , 62 ). The importance of substrate complexity may thus be overridden by additional factors in region-specific ways. That said, our findings bear two points of consideration for management and restoration efforts that seek to mitigate or reverse kelp-forest loss ( 63 – 65 ). First, our work implies that both natural and artificial high-complexity reefs offer a means to increase the strength of stabilizing kelp-forest feedbacks. Reefs could be selected for conservation efforts or constructed to maximize the entrapment of locally produced and delivered drift algae, provide structure for urchins to shelter, and support a diversity of urchin-controlling predators. In the context of artificial reefs, we acknowledge there is no quick fix for ecological restoration ( 66 ) and that multiple interests are often at play (e.g., the desire to minimize man-made structures in marine protected areas). However, given our evidence that kelp-forest stability can vary at the scale of a 10 × 2-m transect and strong evidence that metapopulation dynamics driving kelp spore dispersal operate at much larger scales ( 67 ), we submit that strategically placed patchworks of natural and artificial reefs could serve as hotspots of emergent kelp-forest resilience. Second, the large overlap between urchin densities in the mixed kelp–urchin and urchin-barren states ( Fig. 4 ) emphasizes that urchin density alone is an insufficient predictor of urchin behavior and state stability. In particular, the rapid timescales of kelp and drift algae loss, and the rapid manner with which urchin behavior responds ( 24 ), indicate that bimodality in system state mirrors a bimodality in urchin grazing activity. Hence, the common practice of removing or culling urchins to reduce their abundance will decrease grazing rates only in the short term and will not alone restore feedback processes that confer kelp-forest stability. More specifically, the processes of kelp growth, reproduction, dispersal, senescence, and drift production, which are critical for achieving and stabilizing the mixed kelp–urchin state, as well as the counteracting processes of urchin immigration, settlement, and recruitment, which stabilize the barren state, are not affected by such direct, short-term means of urchin control. Instead, urchin removal is likely to be most effective for jump starting kelp recovery when efforts are focused upon high-complexity substrate and paired alongside local kelp-focused restoration (e.g., outplanting) and short-term drift enhancement to strategically protect out-planted kelp until local kelp growth and drift production are reestablished.",
"discussion": "Discussion At San Nicolas Island (SNI), high-complexity sites and transects did not exhibit alternative stable states of community composition, instead exhibiting 38 y of stable kelp–urchin coexistence resilient to perturbation. Urchins were common in these transects ( Fig. 4 E and F) , but rather than forming fronts or grazing actively in the open, as urchins are known to do during urchin-barren formation ( 24 ), these urchins were consistently tucked away in crevices and self-created pits ( 44 ). In contrast, low-complexity transects exhibited both mixed kelp–urchin and urchin-barren states that persisted for up to 12 y, with transitions between them being higher-velocity events in both directions ( Fig. 3 and SI Appendix , Figs. S1–S3 ). Urchins in these transects were observed to exhibit sedentary behavior when in the mixed kelp–urchin state, with urchin densities seen during mixed kelp–urchin periods overlapping considerably with those seen during urchin-barren periods ( Fig. 4 A – D ). Because high- and low-complexity sites are interspersed around the island, with adjacent sites of differing substrate complexity experiencing equivalent oceanographic conditions, these patterns are unlikely to be caused by unassessed covariates (see SI Appendix for discussions of chlorophyll a [ SI Appendix , Fig. S4 ], sea temperature [ SI Appendix , Fig. S5 ], wave height [ SI Appendix , Fig. S6 ], and sea urchin predator abundance [ SI Appendix , Fig. S7 ]). Instead, we hypothesize that high-complexity substrate permits stable kelp–urchin coexistence because it modifies the relative strength of both top-down and bottom-up regulatory feedbacks through an interplay of behavioral, interspecific, and oceanographic processes. Fig. 4. ( A–F ) Total red and purple urchin abundances partitioned by system state (transects combined by site) with sites arranged by increasing mean substrate complexity from Left to Right . Within each panel the algal-only state is represented by the letter A, the mixed kelp–urchin state by M, and the urchin-barren state by B. Red line segments delineate median urchin abundances. A high degree of overlap between the mixed kelp–urchin and urchin-barren states indicates that urchin density is not the exclusive driver of kelp-forest states. Complexity Modifies the Strength of Urchin-Regulating Feedbacks. We hypothesize that substrate-induced covariation between top-down and bottom-up effects on urchin behavior, recruitment, and mortality determines the propensity of kelp-forest communities to exhibit a single, resilient state versus multiple, alternative stables states between which switches occur. Urchin predators, such as sea stars and California sheephead ( Semicossyphys pulcher ), positively associate with high complexity at SNI ( Fig. 1 C and SI Appendix , Fig. S7 ). Their presence exerts direct mortality on urchins and modifies urchin behavior through a “landscape of fear” ( 14 , 16 , 45 ). High-complexity substrate also entraps drift algae in cracks, below ledges, and at the base of rocky outcrops. High-complexity substrate thereby retains and stabilizes the supply of drift, which urchins prefer to consume over live kelp, particularly during large wave events that otherwise result in net drift export ( 22 , 46 ). Urchins persist during periods of relatively low drift availability following storms due to their high longevity even when starved ( 47 ). For low-complexity substrates, the net loss of drift during storms elicits the urchin behavioral shift to actively wander and graze upon live kelp ( 21 , 23 , 24 ). Because lower-complexity substrates also have lower abundances of slow-to-reproduce predators, active urchin grazing following drift loss proceeds largely unchecked, with increasingly strong feedback mechanisms—including lower local production of drift and a greater cover of encrusting algae that acts as a cue for urchin settlement ( 48 )—stabilizing the urchin-barren state. Once in the urchin-barren state, large, density-dependent but stochastic disease outbreaks at high urchin densities ( 17 ) permit opportunities for kelp recovery. Low-complexity substrates are thereby predisposed to alternative stable states because the combination of low drift retention (a bottom-up effect) and low predator abundance (a top-down effect) promotes persistent changes in urchin behavior and demography. Substrate complexity determines not only the number of kelp-forest alternative stable states but also how perturbations cause shifts between them. The high velocity required to move between alternative stable states ( Fig. 2 and SI Appendix , Fig. S1 ) indicates that low-complexity transects exhibit a time-invariant, bimodal stability landscape with alternative stable attractors separated by dynamically unstable space ( SI Appendix , Fig. S2 ). That is, transitions from one stable attractor to another require a pulse perturbation, such as rapid kelp and drift loss due to large wave events or urchin mass mortality due to disease ( 8 , 49 ). Shifts between states occurred in both directions and occurred both synchronously and asynchronously at low-complexity sites around the island, even as high-complexity transects exhibited stable persistence ( Fig. 3 and SI Appendix , Fig. S1 ). It is therefore unlikely that the existence of alternative stable states at low-complexity sites reflects forcing from changes in environmental drivers, including gradual press perturbation changes that alter the shape of the stability landscape itself. Instead, our results indicate that the localized effects of stochastic pulse perturbations are state dependent and are modified at small scales by the stabilizing feedbacks associated with substrate complexity. The Algae-Only State as a Multigenerational Long Transient. Potential analysis indicated the existence of a third alternative algal-only state ( Fig. 2 B and C ), but the velocity dynamics indicate this to be a multigenerational period of transient dynamics that inevitably and smoothly leads to the mixed kelp–urchin stable state upon the demographic recovery of urchins. For all but a single exceptional transect (discussed below), this algal-only state followed disease-related urchin mass mortality. Lacking nearly any observable urchins when in the algal-only state ( Fig. 4 B and C ), transects varied widely in kelp abundance, producing numerous instances of within-state high-velocity community movement that represent the vast majority of instances where transect position along axis 2 did not associate with substrate complexity ( SI Appendix , Figs. S1 and S3 ). Fluctuations in kelp abundance decreased as urchins began recovering ∼6 to 8 y following their crash ( Fig. 3 B and C ). Whereas kelp reproduce and grow annually, urchins require several years to reach adult size ( 50 ); thus, we hypothesize that these transient dynamics are driven by the temporal lag between urchin mass mortality and the time required for local urchin recovery. Such dynamics are expected for slow–fast systems with strongly differing consumer and resource generation times ( 39 ). The multiyear nature of this transience highlights the limitation of potential analysis when additional temporal insight is lacking ( 51 ) and emphasizes the need for long-term monitoring to contextualize shifts in state and guide kelp-forest management and conservation ( 13 , 52 , 53 ). Low-Complexity Dynamics Conditional upon Surrounding Heterogeneity. Performing analyses at the transect level provided insight into variation that potential analysis would not have revealed at the site level ( 37 ), but also raises a question regarding the behavior of an exceptional transect. The transect, East Dutch 45R, is the only low-complexity transect to exhibit a persistent algal-only state ( Fig. 2 E and Q ). It experienced repeated perturbations from which it returned to the algal-only state with high velocity ( Fig. 3 E and SI Appendix , Fig. S2 ). These dynamics suggest that this transect’s algal-only state reflects a third stable attractor, rather than a long transient. This is an exception to our inference that substrate complexity is the sole predictor of kelp-forest stability at SNI, as we would expect this low-complexity transect to exhibit multimodality. We contend, however, that this exceptional transect reflects a deeper nuance to kelp-forest dynamics related to spatial scale, as it is the only low-complexity transect that is surrounded by otherwise high-complexity substrate. We hypothesize that the stabilizing effects of adjacent complex substrate spill over to confer this transect’s resilience. Larger expanses of low-complexity substrate—as surrounds all other low-complexity transects and sites of our study—lack this stabilizing spatial spillover. Manipulative experiments, such as urchin additions or the continual removal of drift from similar low-complexity areas that are surrounded by high-complexity substrate, are needed to test this hypothesis and determine the spatial scales to which the mechanism may apply. Conclusions. The processes and feedbacks that associate with substrate complexity undoubtedly extend well beyond those that we have discussed. For example, high-complexity transects are more species rich and exhibit a greater coverage of foliose red algae and sessile invertebrates than do low-complexity transects ( 35 ). As such, our results add to a rich ecological literature detailing the many means by which physical and biological complexity can modify species coexistence and the dynamics and functioning of ecological communities ( 54 – 57 ). It nonetheless remains an open question how globally widespread the importance of substrate complexity is, as changes in kelp-forest state certainly do occur irrespective of substrate complexity, especially at higher latitudes ( 4 , 58 – 60 ). We speculate that many of these large-scale changes in kelp-forest state are driven by phase shifts rather than switches between alternative stable states. For example, in the Northeast Pacific, phase shifts between forested and urchin-barren states are driven by changing environmental conditions, specifically the presence or absence of sea otters ( 3 , 61 ). Urchin predator diversity and environmental conditions that influence urchin recruitment and population structure also vary with latitude and across the globe ( 16 , 50 , 62 ). The importance of substrate complexity may thus be overridden by additional factors in region-specific ways. That said, our findings bear two points of consideration for management and restoration efforts that seek to mitigate or reverse kelp-forest loss ( 63 – 65 ). First, our work implies that both natural and artificial high-complexity reefs offer a means to increase the strength of stabilizing kelp-forest feedbacks. Reefs could be selected for conservation efforts or constructed to maximize the entrapment of locally produced and delivered drift algae, provide structure for urchins to shelter, and support a diversity of urchin-controlling predators. In the context of artificial reefs, we acknowledge there is no quick fix for ecological restoration ( 66 ) and that multiple interests are often at play (e.g., the desire to minimize man-made structures in marine protected areas). However, given our evidence that kelp-forest stability can vary at the scale of a 10 × 2-m transect and strong evidence that metapopulation dynamics driving kelp spore dispersal operate at much larger scales ( 67 ), we submit that strategically placed patchworks of natural and artificial reefs could serve as hotspots of emergent kelp-forest resilience. Second, the large overlap between urchin densities in the mixed kelp–urchin and urchin-barren states ( Fig. 4 ) emphasizes that urchin density alone is an insufficient predictor of urchin behavior and state stability. In particular, the rapid timescales of kelp and drift algae loss, and the rapid manner with which urchin behavior responds ( 24 ), indicate that bimodality in system state mirrors a bimodality in urchin grazing activity. Hence, the common practice of removing or culling urchins to reduce their abundance will decrease grazing rates only in the short term and will not alone restore feedback processes that confer kelp-forest stability. More specifically, the processes of kelp growth, reproduction, dispersal, senescence, and drift production, which are critical for achieving and stabilizing the mixed kelp–urchin state, as well as the counteracting processes of urchin immigration, settlement, and recruitment, which stabilize the barren state, are not affected by such direct, short-term means of urchin control. Instead, urchin removal is likely to be most effective for jump starting kelp recovery when efforts are focused upon high-complexity substrate and paired alongside local kelp-focused restoration (e.g., outplanting) and short-term drift enhancement to strategically protect out-planted kelp until local kelp growth and drift production are reestablished."
} | 4,470 |
32719508 | null | s2 | 165 | {
"abstract": "Self-healing materials are indispensable for soft actuators and robots that operate in dynamic and real-world environments, as these machines are vulnerable to mechanical damage. However, current self-healing materials have shortcomings that limit their practical application, such as low healing strength (below a megapascal) and long healing times (hours). Here, we introduce high-strength synthetic proteins that self-heal micro- and macro-scale mechanical damage within a second by local heating. These materials are optimized systematically to improve their hydrogen-bonded nanostructure and network morphology, with programmable healing properties (2-23 MPa strength after 1 s of healing) that surpass by several orders of magnitude those of other natural and synthetic soft materials. Such healing performance creates new opportunities for bioinspired materials design, and addresses current limitations in self-healing materials for soft robotics and personal protective equipment."
} | 247 |
24278392 | PMC3837703 | pmc | 166 | {
"abstract": "Thermally induced bleaching has caused a global decline in corals and the frequency of such bleaching events will increase. Thermal bleaching severely disrupts the trophic behaviour of the coral holobiont, reducing the photosynthetically derived energy available to the coral host. In the short term this reduction in energy transfer from endosymbiotic algae results in an energy deficit for the coral host. If the bleaching event is short-lived then the coral may survive this energy deficit by depleting its lipid reserves, or by increasing heterotrophic energy acquisition. We show for the first time that the coral animal is capable of increasing the amount of heterotrophic carbon incorporated into its tissues for almost a year following bleaching. This prolonged heterotrophic compensation could be a sign of resilience or prolonged stress. If the heterotrophic compensation is in fact an acclimatization response, then this physiological response could act as a buffer from future bleaching by providing sufficient heterotrophic energy to compensate for photoautotrophic energy losses during bleaching, and potentially minimizing the effect of subsequent elevated temperature stresses. However, if the elevated incorporation of zooplankton is a sign that the effects of bleaching continue to be stressful on the holobiont, even after 11 months of recovery, then this physiological response would indicate that complete coral recovery requires more than 11 months to achieve. If coral bleaching becomes an annual global phenomenon by mid-century, then present temporal refugia will not be sufficient to allow coral colonies to recover between bleaching events and coral reefs will become increasingly less resilient to future climate change. If, however, increasing their sequestration of zooplankton-derived nutrition into their tissues over prolonged periods of time is a compensating mechanism, the impacts of annual bleaching may be reduced. Thus, some coral species may be better equipped to face repeated bleaching stress than previously thought.",
"introduction": "Introduction Coral reefs are of critical ecological, economic, and cultural importance, providing ecosystem services with an estimated value of hundreds of billions of dollars annually [ 1 ]. Reef building corals exhibit mixotrophy, relying on both the photoautotrophic products of their endosymbiotic algae and the nutrients acquired through heterotrophic predation [ 2 ]. This mixotrophy results in a complex cycling of inorganic and organic carbon between the coral host, the skeleton it secretes, and its endosymbiotic algae [ 3 , 4 ]. However, during thermal bleaching caused by elevated seawater temperatures the coral-algae relationship breaks down and there is a dramatic reduction in the concentration of endosymbiotic algae [ 5 - 7 ] and/or the endosymbiotic algal pigments [ 8 , 9 ]. This results in a substantial reduction in the assimilation of photoautotrophically derived organic carbon [ 4 ]. At an ecosystem level these thermally induced events can result in mass coral bleaching events where over 90% of the coral in any one area become bleached, often leading to significant coral mortality [ 10 ]. The occurrence of mass bleaching events is predicted to increase in frequency [ 11 ] and threatens to reduce reefs globally by 60% [ 12 ]. However not all bleaching events will result in the mortality of the coral colony; some corals will bleach and recover, whilst others might not visibly bleach at all [ 13 - 15 ]. For the surviving coral colonies the period between successive bleaching events allows the opportunity to recover from the physiological impacts of the bleaching event, acting as a temporal refugium analogous to a spatial refugia [ 16 ]. Predicting the response of coral reefs to repeated bleaching events is dependent on both defining the size of this temporal refuge, and on understanding any adaptive strategies that the coral holobiont may employ to recover within the limits of the temporal refuge or to increase the size of this temporal refugia. One such adaptive strategy is the ability of the coral animal to host multiple clades of endosymbiotic algae [ 17 ] and that a switch to more thermo-tolerant clades of endosymbiotic algae increases the resistance of recovering reefs to future bleaching [ 18 ]. This increases the size of the temporal refuge. If the recovery period is greater than the temporal refugia, then bleaching is likely to occur before the coral has fully recovered, thus lowering the resilience of the coral to that bleaching event. Prolonged elevated levels of heterotrophy may present another adaptive strategy for increasing the resistance of corals to bleaching and in hastening recovery from bleaching. During a bleaching event photosynthetic rates of the holobiont may be reduced by up to 90% [ 9 , 19 , 20 ] radically reducing the energy available to the coral holobiont. In some species, thermal bleaching triggers a switch to increased heterotrophic feeding [ 21 ], and this trophic switching is an important determinant of colony survival after bleaching [ 22 ]. It has long been known that some species of reef corals can survive for long periods without sunlight [ 23 ]. Heterotrophically acquired carbon is important in tissue building in corals and anemones [ 4 , 24 ] and can reduce the severity of bleaching [ 25 ]. However the proportionate contribution of heterotrophy and photoautotrophy to the coral diet during long-term recovery from thermal bleaching and the importance of either pathway during this process is poorly understood. This study aims to understand the role that these respective pathways play in the recovery of corals over the course of almost a year following thermal bleaching. Using 13 C enriched dissolved inorganic carbon (DIC) in seawater to label the photoautotrophic pathways and 13 C enriched rotifers to label the heterotrophic pathway, the proportionate contribution of both sources of carbon was assessed in two species of Hawaiian coral for 11 months following an experimental bleaching.",
"discussion": "Discussion Understanding how corals respond to, and recover from, bleaching events is crucial if we are to better predict the impacts of global warming on coral reef ecosystems. Our data show that the recovery pattern of the trophic behaviour of the coral holobiont is complex and non-uniform. Although the photoautotrophic mechanism had recovered after 4 months, with Chl a and photoautotrophic carbon assimilation levels the same between bleached and non-bleached corals of both species, the assimilation of heterotrophic carbon was highly elevated in the bleached corals compared to the non-bleached controls even after 11 months of recovery. The failure of heterotrophic carbon assimilation to return to non-bleached levels even after 11 months of recovery suggests that either 1) bleaching induces an acclimatization response that could buffer them from future bleaching, or 2) the temporal refugia for corals from bleaching events is greater than 11 months long. We explore these findings in more detail below. The photoautotrophic system of both species was still in recovery after 1 month as demonstrated by the lower levels of carbon assimilation and Chl a in bleached relative to the non-bleached corals. This is consistent with previous observations of coral bleaching reducing photosynthetic rates in these species by 67-90% [ 9 ] and also reduced CZAR (contribution of zooxanthellae-acquired carbon to daily animal respiration) by approximately 60% in these species [ 21 , 31 ]. Associated with the impact on the photoautotrophic system there was a lower assimilation of carbon into the skeletal component in the bleached corals relative to the non-bleached, which is also consistent with previous studies that have shown a reduction or cessation of skeletal growth as a result of bleaching in these species [ 26 , 32 , 33 ]. After 4 months of recovery, both Chl a and photoautotrophic carbon assimilation rates indicated that photoautotrophy had fully recovered in M. capitata . At the same time, P. compressa had recovered Chl a and was assimilating significantly more photoautotrophically derived carbon than the controls. This may be due to P. compressa not increasing its feeding rates when bleached [ 21 , 31 ], and thus relying predominantly on photosynthesis to promote recovery. By 11 months, there were no significant differences in the assimilation of photoautotrophic carbon between bleached and non-bleached control corals for either species. Thus for these two coral species, photoautotrophy had recovered within 4 months of bleaching. These findings show that bleached corals had visibly recovered and photosynthetic pigment concentrations and photoautotrophic carbon assimilation were at normal or higher levels after only 4 months of recovery. This corresponds to field estimates of the duration of coral recovery based on appearance, pigment concentration, and photosynthetic activity which range from between 25 days to over 11 months [ 9 , 34 - 36 ]. While photoautotrophic carbon is clearly important for recovering corals, heterotrophic carbon specifically appears to be critical for the survival of corals during long-term recovery, and consequently may be the variable that defines the extent of the temporal refugia. During the first month of recovery, heterotrophic carbon assimilation in M. capitata either did not significantly differ between bleached and non-bleached controls or was slightly lower in bleached than in non-bleached controls. Yet, previous work has clearly shown that feeding rates of M. capitata dramatically increase following bleaching [ 21 , 31 ]. Thus, the extra heterotrophic carbon acquired by M. capitata in the early stages of recovery is not being assimilated but is being rapidly catabolized to meet metabolic demand and/or is lost via mucus or particulate organic matter. This is consistent with findings from bleached Hawaiian Porites lobata corals that also catabolize their heterotrophically acquired carbon [ 37 ] and findings by Tanaka et al (2009) showing that bleached corals lost heterotrophically acquired carbon through mucous production or as particulate carbon. In addition, preliminary measurements of dissolved organic carbon (DOC) fluxes in M. capitata suggests that it also releases DOC when bleached (Hughes & Grottoli unpublished). However further experimental work is required to test this. For P. compressa , the lack of a significant difference in heterotrophic carbon assimilation after 1 month of recovery is consistent with a lack of any changes in feeding rates in this species with bleaching [ 21 , 31 ]. However after 4 months of recovery, heterotrophic carbon assimilation by both the coral host and endosymbiotic algae of both species was dramatically higher in bleached corals compared to the non-bleached control corals. The trigger for this increase in heterotrophic carbon assimilation is unknown, but bleaching depletes specific lipid classes [ 38 ] and the physiological change may elicit this response. This extra heterotrophic carbon assimilation was still evident for both species even after 11 months of recovery for which there are two possible interpretations. Firstly, it has been previously shown that for these species the tissue biomass, lipid, protein and carbohydrate recovers within 8 months post bleaching [ 9 ]. This, combined with the evidence presented here that the photoautotrophic system had recovered within 4 months, suggests that increased heterotrophic assimilation during long-term recovery is an adaptive response that enhances production through heterotrophy which, could increase coral resilience to future bleaching events. Previous experiments have also shown that heterotrophic carbon is the carbon source for tissue building in corals and anemones [ 4 , 24 ] and that heterotrophic feeding by corals can diminish the severity of bleaching [ 25 ]. This hypothesis is also consistent with model scenarios predicting that heterotrophy may be an important determinant of colony survival after bleaching [ 22 ]. Alternatively, this heterotrophic compensation is evidence that the corals are still in recovery after 11 months despite the recovery of other physiological parameters. Optimal foraging theory [ 39 ] suggests that if the capacity to increase heterotrophic carbon assimilation was beneficial to non-bleached corals, then there would be no difference between the bleached and non-bleached corals. As the heterotrophic compensation was only observed in the bleached corals, it supports the interpretation that it is a direct response to the bleaching stress and is part of the recovery process. This is the first definition of the temporal refugia for a coral species based on these measurements and is considerably longer than previous estimates based on growth rates or reproductive tissue [ 26 , 33 , 40 ], suggesting that full recovery can take significantly longer than previously thought and that the temporal refugia from climate change required is greater than originally assumed. The response to bleaching events varies between species and within individuals of the same species [ 41 , 42 ]. This in part due to past thermal history whereby those corals having previously experienced thermal stress are less susceptible to future bleaching [ 43 ] and also in part due to the ability of some corals to adapt or acclimatise to thermal stress. The mechanisms of this adaption are poorly defined. Adaptive change by the holobiont to coral bleaching has been previously observed through the ability of the coral holobiont to shuffle or switch the endosymbiotic algae it houses, switching from less thermally tolerant clades to more tolerant clades following bleaching [ 44 , 45 ]. Another possible mechanism is a high degree of physiological plasticity in the relationship between the host and the endosymbiotic algae such as the up regulation of heat shock proteins [ 46 ] allowing a more stable relationship between the coral animal and endosymbiotic algae during thermal stress. In addition, heterotrophic plasticity has been shown to maintain physiological status in corals immediately following bleaching [ 21 ]. At a community scale, natural selection on ecological timescales has also been posited as a mechanism of adaption. Here we show for the first time that heterotrophic compensation persists for almost a year following bleaching, highlighting the long-term importance of heterotrophic carbon in coral physiology for 11 months after bleaching. The increased heterotrophic carbon assimilation following bleaching may 1) act as an adaptive strategy against further bleaching events by increasing the nutrient acquisition through heterotrophy and possibly reducing the dependence of the holobiont on the photoautotrophic system (however further experimentation is required to test this hypothesis) or 2) be a symptom of a coral still suffering negative effects of bleaching and for whom the size of the temporal refugia required is greater than 11 months. In light of these findings long-term recovery from bleaching is critically dependent on healthy plankton populations throughout the year. Healthy coral reefs are known to have a concomitant and dramatic impact on plankton populations in overlying waters, depleting pelagic diatoms and zooplankton by as much as 90% and 60%, respectively [ 47 ]. Increasing sea surface temperatures in the tropics over the past few decades have resulted in a steady decline in zooplankton abundance [ 48 ] with marked decreases in plankton during bleaching events on reefs [ 49 ]. In the future, a potentially chronic need for extra heterotrophic carbon by corals due to multiple and possibly annual bleaching events, combined with decreases in zooplankton populations due to warming, would ultimately limit the quantity and quality of plankton available on reefs needed to support recovery from bleaching and to build future resilience to repeated bleaching events."
} | 4,010 |
37614339 | PMC10442545 | pmc | 168 | {
"abstract": "Introduction The spiking neural network (SNN) is a bionic model that is energy-efficient when implemented on neuromorphic hardwares. The non-differentiability of the spiking signals and the complicated neural dynamics make direct training of high-performance SNNs a great challenge. There are numerous crucial issues to explore for the deployment of direct training SNNs, such as gradient vanishing and explosion, spiking signal decoding, and applications in upstream tasks. Methods To address gradient vanishing, we introduce a binary selection gate into the basic residual block and propose spiking gate (SG) ResNet to implement residual learning in SNNs. We propose two appropriate representations of the gate signal and verify that SG ResNet can overcome gradient vanishing or explosion by analyzing the gradient backpropagation. For the spiking signal decoding, a better decoding scheme than rate coding is achieved by our attention spike decoder (ASD), which dynamically assigns weights to spiking signals along the temporal, channel, and spatial dimensions. Results and discussion The SG ResNet and ASD modules are evaluated on multiple object recognition datasets, including the static ImageNet, CIFAR-100, CIFAR-10, and neuromorphic DVS-CIFAR10 datasets. Superior accuracy is demonstrated with a tiny simulation time step of four, specifically 94.52% top-1 accuracy on CIFAR-10 and 75.64% top-1 accuracy on CIFAR-100. Spiking RetinaNet is proposed using SG ResNet as the backbone and ASD module for information decoding as the first direct-training hybrid SNN-ANN detector for RGB images. Spiking RetinaNet with a SG ResNet34 backbone achieves an mAP of 0.296 on the object detection dataset MSCOCO.",
"introduction": "1. Introduction In recent years, significant progress has been made in deep learning research, which has become a primary tool for various computer vision tasks, such as object recognition, object detection, and semantic segmentation. Key technologies such as ResNet (He et al., 2016 ) and batch normalization (Ioffe and Szegedy, 2015 ) have enabled the construction of deep neural networks with numerous parameters and deep model structures, achieving high accuracy in the aforementioned tasks. However, the growing network complexity and data quantity make it increasingly expensive to train and deploy deep neural networks. Therefore, it is necessary to explore network models and computational paradigms that are more efficient than current artificial neural networks (ANNs). One of the main research directions is the spiking neural network (SNN), a bionic neuron model inspired by biological neuron models based on spiking signals (Gerstner and Kistler, 2002 ; Cheng et al., 2023 ; Yi et al., 2023 ). Researchers have paid considerable attention to SNN because of its high-energy efficiency on neuromorphic hardwares (Merolla et al., 2014 ; Davies et al., 2018 ). Due to the non-differentiability of the spiking signals, training high-performance SNNs is challenging. First, researchers utilized the spike-timing-dependent plasticity (STDP) (Song et al., 2000 ) rule to conduct the unsupervised training of SNNs. STDP is a biology-inspired process that adjusts the synaptic weights based on the relative timing of the presynaptic and postsynaptic neurons' action potentials. However, STDP cannot accomplish supervised learning for large-scale networks, which limits its practical application. Currently, there are two mainstream approaches to obtain deep SNN models: ANN-to-SNN conversion and direct-training. The ANN-to-SNN conversion method consists of two steps. First, an ANN model corresponding to the target SNN model is trained. Then, the connection between the firing rates of the SNN and the activation values of the ANN are used to establish a conversion formula to help convert the weight of the ANN model to that of the SNN. The accuracy of this conversion is largely determined by the simulation time steps of SNNs. The simulation time step is usually in hundreds or thousands to obtain an SNN with competitive performance, which results in the unacceptably high latency. The second method, direct-training, approximates the non-differentiable heaviside step function with a surrogate gradient and trains the SNNs directly through backpropagation. Researchers usually adopt the backpropagation through time (BPTT) framework, which is derived from RNN. Unlike ANN-to-SNN conversion, the direct-training method requires only a tiny time step. The network thus obtained has very low latency, making it superior in real-time scenarios. However, because of the complicated neural dynamics of SNNs and the non-differentiability of spiking signals, direct-training of SNNs requires further exploration on several crucial issues to achieve acceptable results on large-scale datasets, such as ImageNet (Deng et al., 2009 ) and MSCOCO (Lin et al., 2014 ). The first issue is the gradient vanishing or explosion problem, which restricts SNNs to shallow architectures. To solve this problem, a natural idea is to introduce residual learning from ANNs into the SNNs. Spiking ResNet (Lee et al., 2020 ) replaces the ReLU activation function in the residual block with spiking neurons such as the integrate-and-fire (IF) and leaky-integrate-and-fire (LIF) (Gerstner and Kistler, 2002 ). However, it has been found that such a spiking residual block of spiking ResNet cannot achieve identity mapping because of the complex dynamics of spiking neurons. On this basis, SEW ResNet (Fang et al., 2021a ) has used an element-wise function to modify the residual block and has successfully achieved identity mapping. However, the ADD function used in SEW ResNet introduces non-spiking signals, which no longer conforms the properties of SNNs and preventes SNNs from being deployed on neuromorphic hardware. Therefore, effective residual learning in SNNs remains a problem worth exploring. We believe that the shortcut connection with addition in the residual block enables the analog tensor in different levels to achieve lossless information fusion, which is the reason for the high performance of the ADD-based SEW ResNet. However, the spiking signal is binary, and its addition operation cannot be deployed. This motivates us to explore a better residual block structure to accomplish information fusion with only full-spike operations. Another problem is the decoding of spike trains, which determines the high-dimensional image features in object recognition. There are two schemes: temporal and rate coding. The former directly adopts spike times as the information carrier, which is efficient in large time-step systems. However, direct-training SNNs have tiny time steps, leading to low accuracy of temporal coding. The latter method uses firing rates as the information carrier. Many direct training methods (Fang et al., 2021a ; Zheng et al., 2021 ) have adopted it due to its high performance. However, Wu et al. ( 2021 ) found that rate coding produced a less smooth learning curve, reducing the final accuracy. Meanwhile, from the perspective of neuroscience, rate coding is unreasonable because it treats activation at each time step as equally important. In fact, spikes at different times and in different spaces may have different effects on the results, depending on the salient region (Itti et al., 1998 ; Yao et al., 2021 ). The inability to handle complex computer vision tasks well is another problem. Most existing approaches are limited to classification. Object detection, a fundamental task in vision, has widespread applications in many real-time scenarios. However, there are only a few direct-training spiking object detectors. Cordone et al. ( 2022 ) trained an SSD detector using spiking VGG, MobileNet, and DenseNet as the backbones. Kugele et al. ( 2021 ) constructed a similar detector using a spiking DenseNet. They both performed detection on event data. Neither case performed well in the large-scale MSCOCO datasets, indicating that their methods were not applicable to most existing vision systems. In addition, the gradient degradation problem was not addressed such that the deeper DenseNet achieved a worse accuracy in Cordone et al. ( 2022 ). To address the gradient vanishing or explosion problem, we implemented the identity mapping of the residual block under the constraints of spiking signals by proposing spiking gate (SG) ResNet. The inspiration mainly comes from GRU (Cho et al., 2014 ) and Highway Network (Srivastava et al., 2015 ). These works have shown that the gate mechanism can dynamically control the flow of information in the network and can significantly solve the gradient vanishing problem. In each basic block of SG ResNet, a binary selection gate is introduced to guide the information fusion of the spiking signals. As for the decoding scheme, we propose the attention spike decoder (ASD) to decode the spike output from SG ResNet more effectively. The ASD block is highly generalizable and can be applied to object recognition and detection tasks. The effectiveness of the SG ResNet and ASD block are evaluated on object recognition datasets, including three static image datasets and a neuromorphic dataset. In addition, we propose spiking RetinaNet using SG ResNet as the backbone and the ASD block for information decoding. This is the first direct-training hybrid SNN-ANN detector that can achieve good performance on the MSCOCO dataset. Our contributions are as follows: A spiking gate ResNet with full-spike operations is developed to solve the gradient vanishing in SNNs to make deep SNNs trainable. Furthermore, two appropriate formulations of the binary gate in SG ResNet are provided. An attention spike decoder is proposed to apply temporal, channel, and spatial attention to accumulate the information of spiking signals. This is an effective and general decoder for both object recognition and detection. Numerous experiments are conducted on both static image and neuromorphic datasets in the object recognition task to verify the effectiveness of the SG ResNet and ASD block. Spiking RetinaNet, which is a hybrid neural network, is proposed to combine the SG ResNet backbone with a detection head. The ASD block plays a vital role in spike decoding. We demonstrate that with a proper backbone and decoding, a direct-training SNN can perform well in object detection.",
"discussion": "5.3.4. Comparison and discussion on SEW ResNet Previously, SEW ResNet (Fang et al., 2021a ) is also a variant of spiking ResNet that analyzed and solved the gradient vanishing problem from the perspective of residual learning. They analyzed the reason for the gradient vanishing theoretically and proposed the element-wise function to solve this problem. However, the most effective one of their proposed element-wise functions is ADD, which makes the network no longer spiking. In our SG ResNet, a gate mechanism is introduced to solve gradient vanishing while ensuring that the network is still spiking. In this section, we compare SG ResNet with SEW ResNet using IAND and ADD. Experimental results are shown in Table 8 . To avoid the impact of the ASD module, the decoding scheme used in all methods is rate coding. Table 8 Ablation study on the relationship between SG ResNet and SEW ResNet on CIFAR-100 dataset. \n Network \n \n SG ResNet \n \n SEW ResNet (IAND) \n \n SEW ResNet (ADD) \n ResNet10 72.68% 71.96% \n 73.02% \n ResNet18 74.62% 73.89% \n 74.90% \n ResNet34 75.01% 73.8% \n 75.93% \n Values in the table represents the top-1 accuracy. The bold values indicate the maximum accuracies of the comparison. IAND is a binary operator that returns the inverse and of two inputs. The output of IAND operation with two spiking inputs remains a spiking signal. Thus, the SEW ResNet with IAND is a deployable network. Compared with SEW ResNet with IAND, our SG ResNet performs better at every depth. On the CIFAR-100 dataset, SG ResNet34 has achieved 1.21% higher accuracy than SEW ResNet34 (IAND). ADD is a binary operator that returns the addition of two inputs. However, SEW ResNet with ADD is more like an ANN rather than an SNN. As is expected, SEW ResNet (ADD) has the highest accuracy among the three methods. Based on the above results, the SEW ResNet (ADD) structure, which is similar to the original ResNet, can achieve the best performance without considering signal type. However, this does not necessarily mean that it is the best. Our SG ResNet can be considered as an better compromise that improves accuracy while adhering to the spiking signal constraint.\n\n6. Discussion and conclusion This study focuses on the issues to be solved during direct training of high-performance SNNs in object recognition and detection tasks. We introduced a binary gate mechanism and presented the spiking gate ResNet to form deep architectures in SNNs. This is the first time that a widely used gate mechanism in RNNs is being combined with SNNs in the structural design. Through gradient analysis, we prove that SG ResNet can overcome gradient vanishing or explosion problems. An attention spike decoder is also proposed to address the spiking signal decoding problem. Using SG ResNet as the backbone and the ASD module for information decoding, we propose spiking RetinaNet, which is the first direct-training hybrid SNN-ANN detector for RGB images. The experimental results show that SG ResNet with an ASD decoder outperforms most direct-training SNNs with the surrogate gradient method on the object recognition task. Furthermore, spiking RetinaNet has achieved a satisfactory performance in object detection with an energy efficient spiking backbone. Regarding the future research topics, the binary gate mechanism is non-trivial and valuable to be further explored, including the efficiency-performance trade-off of parameterized gate mechanism and binarization of gate signals. In addition, it will be quite helpful and contributive to investigate how to use gate mechanism in the residual connection of spiking transformer. Finally, downstream vision applications of spiking neural networks are also what we consider to be a crucial direction, including image segmentation, object detection, video recognition, optic flow estimation, and so on."
} | 3,554 |
36439945 | PMC9682266 | pmc | 169 | {
"abstract": "Developing intelligent neuromorphic solutions remains a challenging endeavor. It requires a solid conceptual understanding of the hardware's fundamental building blocks. Beyond this, accessible and user-friendly prototyping is crucial to speed up the design pipeline. We developed an open source Loihi emulator based on the neural network simulator Brian that can easily be incorporated into existing simulation workflows. We demonstrate errorless Loihi emulation in software for a single neuron and for a recurrently connected spiking neural network. On-chip learning is also reviewed and implemented, with reasonable discrepancy due to stochastic rounding. This work provides a coherent presentation of Loihi's computational unit and introduces a new, easy-to-use Loihi prototyping package with the aim to help streamline conceptualization and deployment of new algorithms.",
"introduction": "1. Introduction Neuromorphic computing offers exciting new computational structures. Decentralized units inspired by neurons are implemented in hardware (reviewed by Schuman et al., 2017 ; Rajendran et al., 2019 ; Young et al., 2019 ). These can be connected up to one another, stimulated with inputs, and the resulting activity patterns can be read out from the chip as output. A variety of algorithms and applications have been developed in recent years, including robotic control (DeWolf et al., 2016 , 2020 ; Michaelis et al., 2020 ; Stagsted et al., 2020 ), spiking variants of deep learning algorithms, attractor networks, nearest-neighbor or graph search algorithms (reviewed by Davies et al., 2021 ). Moreover, neuromorphic hardware may provide a suitable substrate for performing large scale simulations of the brain (Furber, 2016 ; Thakur et al., 2018 ). Neuromorphic chips specialized for particular computational tasks can either be provided as a neuromorphic computing cluster or be integrated into existing systems, akin to graphics processing units (GPU) in modern computers (Furber et al., 2014 ; Davies et al., 2021 ). With the right ideas, networks of spiking units implemented in neuromorphic hardware can provide the basis for powerful and efficient computation. Nevertheless, the development of new algorithms for spiking neural networks, applicable to neuromorphic hardware, is a challenge (Grüning and Bohte, 2014 ; Pfeiffer and Pfeil, 2018 ; Bouvier et al., 2019 ). At this point, without much background knowledge of neuromorphic hardware, one can get started programming using the various software development kits available (e.g., Brüderle et al., 2011 ; Sawada et al., 2016 ; Lin et al., 2018 ; Rhodes et al., 2018 ; Michaelis, 2020 ; Müller et al., 2020a , b ; Spilger et al., 2020 ; Rueckauer et al., 2021 ). Emulators for neuromorphic hardware (Furber et al., 2014 ; Petrovici et al., 2014 ; Luo et al., 2018 ; Valancius et al., 2020 ) running on a standard computer or field programmable gate arrays (FPGA), make it possible to develop neuromorphic network architectures without even needing access to a neuromorphic chip (see e.g., NengoLoihi 1 and Dynap-SE 2 ). This can speed up prototyping as the initialization of networks, i.e., distributing neurons and synapses, as well as the readout of the system's state variables on neuromorphic chips takes some time. At the same time emulators transparently contain the main functionalities of the hardware in code and therefore provide insights into how it works. With this understanding, algorithms can be intelligently designed and complex network structures implemented. In the following, we introduce an emulator for the digital neuromorphic chip Loihi (Davies et al., 2018 ) based on the widely used spiking neural network simulator Brian (Stimberg et al., 2019 ). We first dissect an individual computational unit from Loihi . The basic building block is a spiking unit inspired by a current based leaky integrate and fire (LIF) neuron model (see Gerstner et al., 2014 ). Connections between these units can be plastic, enabling the implementation of diverse on-chip learning rules. Analyzing the computational unit allows us to create an exact emulation of the Loihi hardware on the computer. We extend this to a spiking neural network model and demonstrate that both Loihi and Brian implementations match perfectly. This exact match means one can do prototyping directly on the computer using Brian only, which adds another emulator in addition to the existing simulation backend in the Nengo Loihi library. This increases both availability and simplicity of algorithm design for Loihi , especially for those who are already used to working with Brian . In particular for the computational neuroscience community, this facilitates the translation of neuroscientific models to neuromorphic hardware. Finally, we review and implement synaptic plasticity and show that while individual weights show small deviations due to stochastic rounding, the statistics of a learning rule are preserved. Our aim is to facilitate the development of neuromorphic algorithms by delivering an open source emulator package that can easily be incorporated into existing workflows. In the process we provide a solid understanding of what the hardware computes, laying the appropriate foundation to design precise algorithms from the ground up.",
"discussion": "5. Discussion This study was motivated by two goals. We hope to simplify the transfer of models to Loihi and therefore developed a Loihi emulator for Brian , featuring many functionalities of the Loihi chip. In the process of developing the emulator, we aimed to provide a deeper understanding of the functionality of the neuromorphic research chip Loihi by analyzing its neuron and synapse model, as well as synaptic plasticity. We hope that the analysis of Loihi 's spiking units has provided some insight into how Loihi computes. With the numerical integration method, numerical precision and related rounding method, as well as the update schedule, we were able to walk from the LIF neuron model down to the computations performed. For neurons and networks without plasticity we are able to emulate Loihi without error. Analyzing and implementing synaptic plasticity showed that, due to stochastic rounding, it is not possible to exactly replicate trial by trial behavior when it comes to learning. However, on average the weight changes induced by a learning rule are preserved. The main benefit of the Brian2Loihi emulator lies in lowering the hurdle for the experimenter. Especially in neuroscience, many scientists are accustomed to neuron simulators and in particular Brian is widely used. It makes a deep dive into new software frameworks and hardware systems unnecessary. The emulator can be used for simple and fast prototyping, as it improves the initialization time in all cases drastically and the execution time, when a read out is used. In addition, hardware specific complications, like distributing neurons to cores, or constraints like potential limits on the number of available neurons or synapses, or on the speed or size of read-out, do not occur in the emulator. While this will surely improve with new generations of hardware and software in the upcoming years, they can already be ignored by using the emulator. At this point it is important to note that not all Loihi features are included in the emulator, yet. In particular, the homeostasis mechanism, rewards, and tags for the learning rule are not included. In Table A1 , we provide a comparison of all functionalities from Loihi with those available in the current state of the emulator. Development of this emulator is an open source project and we expect improvements and additions with time. Note that a follow up project, called Brian2Lava has already started. 12 An important vision for the future is to flexibly connect front-end development environments (e.g., Brian , NEST, Keras, TensorFlow) with various back-ends, like neuromorphic platforms (e.g., Loihi , SpiNNaker, BrainScaleS, Dynap-SE) or emulators for these platforms. PyNN (Davison et al., 2009 ) is such an approach to unify different front-ends and back-ends in a more general way. Nengo (Bekolay et al., 2014 ), as another approach, does not provide the use of other simulators, but allows several back-ends and focuses on higher level applications (DeWolf et al., 2020 ). NxTF (Rueckauer et al., 2021 ) is an API and compiler aimed at simplifying the efficient deployment of deep convolutional spiking neural networks on Loihi using an interface derived from Keras. We think that ideally, one could continue to work in their preferred front-end environment while a package maps their code to existing chips or computer-based emulators of these chips. We expect an interface along these lines will play an important role in the future of neuromorphic computing and want to contribute to this development with our Brian2Loihi emulator. At least for now, with an emulator at hand, it is easier to prototype network models and assess whether an implementation on Loihi is worth considering. When getting started with neuromorphic hardware, to e.g., scale up models or speed up simulations, researchers familiar with Brian can directly deploy models prepared with the emulator. We hope that with this, others may find a smooth entry into the quickly emerging field of neuromorphic computing."
} | 2,352 |
35524438 | PMC9378539 | pmc | 170 | {
"abstract": "Abstract Lignin is a largely untapped source for the bioproduction of value‐added chemicals. Pseudomonas putida KT2440 has emerged as a strong candidate for bioprocessing of lignin feedstocks due to its resistance to several industrial solvents, broad metabolic capabilities, and genetic amenability. Here we demonstrate the engineering of P. putida for the ability to metabolize syringic acid, one of the major products that comes from the breakdown of the syringyl component of lignin. The rational design was first applied for the construction of strain Sy‐1 by overexpressing a native vanillate demethylase. Subsequent adaptive laboratory evolution (ALE) led to the generation of mutations that achieved robust growth on syringic acid as a sole carbon source. The best mutant showed a 30% increase in the growth rate over the original engineered strain. Genomic sequencing revealed multiple mutations repeated in separate evolved replicates. Reverse engineering of mutations identified in agmR, gbdR, fleQ , and the intergenic region of gstB and yadG into the parental strain recaptured the improved growth of the evolved strains to varied extent. These findings thus reveal the ability of P. putida to utilize lignin more fully as a feedstock and make it a more economically viable chassis for chemical production.",
"conclusion": "5 CONCLUSIONS AND FUTURE WORK In this study, we successfully engineered P. putida KT2440 derivatives for robust growth using syringate as the sole carbon source through a combined approach of rational strain design and adaptive lab evolution. Genome sequencing and genetic analysis revealed a few mutations that enhanced the strain's fitness. To better understand how each mutation alters cellular states, we currently are conducting omics studies, in particular transcriptomics and targeted proteomics. Meanwhile, improvement in the growth characteristics of the best mutant, 1‐5b, may be achieved through another round of combined efforts of rational design and ALE experiments in light of results from the omics studies. Lignin is an attractive carbon source for bioproduction due to the projected large volume from biofuel production. The efficient use of lignin plays a critical role in the bioeconomy. Our work reported here provides a solid basis for further strain improvements of P. putida KT2440, a promising chassis of bio‐industrial applications for lignin valorization.",
"introduction": "1 INTRODUCTION Lignin is a complex organic polymer that makes up 20%–35% of plant cell walls. It is polymerized from three monolignols: p ‐coumaryl alcohol, coniferyl alcohol, and sinapyl alcohol, which are categorized as the corresponding hydroxy (H), guaiacyl (G), and syringyl (S) lignin components (Figure 1a ; Feofilova & Mysyakina, 2016 ). Depending on the type of plants, the ratio of H, G, and S subunits in lignin varies, where the grass lignin consists of all subunits, the softwood lignin often has low S component (Gellerstedt & Henriksson, 2008 ). The heterogeneity of lignin makes it a difficult target for traditional catalytic and enzymatic refinement approaches (C. Li et al., 2015 ; Rinaldi et al., 2016 ; Schutyser et al., 2018 ; Sun et al., 2018 ; Zakzeski et al., 2010 ). As a result, most lignin streams generated in existing processing facilities are burned as a low‐grade fuel (Ragauskas et al., 2014 ). Recent development in the biofuel technology, in particular the potential use of lignocellulosics as the starting material, ignites research interests into the valorization of lignin streams through microbial conversions of its degradation products (Becker & Wittmann, 2019 ; Eltis & Singh, 2018 ; Linger et al., 2014 ). Breakdowns of lignin can lead to an assortment of compounds that preserve the core chemical structures of the three monolignols. Maximizing the value of lignin therefore calls for a complete utilization of all components in the lignin depolymerization mixture. Figure 1 Lignin synthesis and degradation pathways. (a) Schematics for the polymerization of monolignols (left) into polymeric lignin (center) followed by its depolymerization into model monomeric aromatic products (right). (b) Catabolic pathways of model depolymerization products in P. putida . The known reaction catalyzed by VanAB is marked with solid arrows. Potential reactions for the conversion of syringic acid into gallic acid using VanAB are shown using dashed arrows. (c) The native P. putida catabolic pathway for gallic acid through the intermediates, including 4‐oxalomesaconic acid (OMA) keto and enol forms, and 4‐carboxy‐4‐hydroxy‐2‐oxoadipic acid (CHA). Enzymes, gallic acid dioxygenase (GllA), OMA keto–enol tautomerase (GllD), OMAenol hydratase (GllB), and CHA aldolase (GllC). Several microorganisms, such as Pseudomonas putida (Salvachúa et al., 2020 ; Willett, 2019 ), Sphingobium sp. SYK‐6 (Araki et al., 2020 ; Gall et al., 2014 ; Meux et al., 2012 ; Sato et al., 2009 ), and Novosphingobium aromaticivorans (Bell et al., 2012 ; Cecil et al., 2018 ; Kontur et al., 2018 ; Perez et al., 2019 , 2020 ), are being investigated for their abilities to utilize lignin‐derived compounds. Among them, P. putida KT2440 has emerged as a suitable candidate for industrial bioprocessing of lignin due to its genetic tractability, high tolerance to industrial solvents, and readily available engineering tools (Martínez‐García & de Lorenzo, 2019 ; Nikel & de Lorenzo, 2018 ). KT2440 can utilize a large number of carbon substrates, including compounds obtained from the breakdown of lignin (Dos Santos et al., 2004 ; Nikel & de Lorenzo, 2018 ; Nikel et al., 2014 ). It encodes pathways to metabolize vanillate, ferulate, p‐ coumarate, and 4‐hydroxybenzoate which are the major degradation products from the G and H components of lignin (Harwood & Parales, 1996 ; Jiménez et al., 2002 ; Nogales et al., 2017 ) (Figure 1a,b ). Notably, it lacks the ability to use syringate or sinapate as a sole carbon source, which are compounds derived from the lignin's syringyl components (Figure 1a,b ). Successful engineering of S lignin metabolism in the KT2440 strain therefore will greatly boost its application in lignin valorization. In this report, we demonstrate the successful engineering of syringate metabolism in KT2440 through a combined approach of chromosomal overexpression of the vanillate demethylase (VanAB) and subsequent adaptive lab evolution (ALE). Experimental analysis of mutations identified in the ALE mutants provided insights into potential causal genetic and metabolic changes that led to the improved growth on syringate.",
"discussion": "4 DISCUSSION \n P. putida is a promising bacterial chassis for the production of value‐added chemicals from lignin depolymerization compounds (Martínez‐García & de Lorenzo, 2019 ; Nikel & de Lorenzo, 2018 ; Nikel et al., 2016 ). Here, we demonstrated the successful engineering of P. putida KT2440 for growth on syringic acid by first overexpressing the native vanAB genes in strain Sy‐1 followed by its ALE experiment with syringic acid as the sole carbon source. ALE has been used to identify novel approaches to solve complex engineering problems in P. putida , including to enhance tolerance to toxic compounds, such as anthranilate (Kuepper et al., 2020 ), ionic liquids (Lim et al., 2020 ), solvents (Kusumawardhani et al., 2021 ), and lignin‐derived acids (Mohamed et al., 2020 ), and to improve the efficiency of catabolic pathways for ethylene glycol, 1,4‐butanediol (W.‐J. Li et al., 2020 ; W. J. Li et al., 2019 ), and xylose (Lim et al., 2021 ). Our ALE process spanned over 3 months to improve an average growth rate of the ALE culture from 0.4 to 2.8 day −1 , which is a combined result of shortened lag phase and increased growth rate. One of the best evolved mutant Sy‐1 1‐5b has a lag phase of 8 h and a growth rate of 0.115 h −1 , an 83.3% reduction, and 34.1% improvement, respectively. Genome sequencing data of ALE population and ALE isolates with improved growth revealed a plethora of genetic changes in the dynamic evolution process. We took a combined approach of literature searching and experimental verification to identify mutations that could contribute to the improved fitness of Sy‐1 mutants in media with syringate. Among unique mutations, certain ones emerged at the early stage of the ALE process but did not further prevail. For example, a total of eight unique SNPs were observed in gene PP_0168 (Supporting Information: Table S2 ), which encodes a possible surface adhesion protein. Since the formation of biofilm is a commonly observed microbial behavior in suspension cultures, such mutations were hypothesized as neutral drift, which are irrelevant to the evolutionary pressure. Mutations in fleQ (three unique mutations) and fliF (two unique mutations) (Supporting Information: Table S2 ) have been previously identified as common mutation targets in ALE experiments. It was hypothesized that the reduced or loss of ability to express flagellar saves energy from producing an unnecessary cell organelle under the shake flask environment. Strains of P. putida have been previously generated lacking the majority of flagellar genes which results in enhanced energy metabolism (Martínez‐García et al., 2014 ; Mohamed et al., 2020 ). We also observed a large number of mutations that occurred in intergenic regions and a small number of SNPs in uncharacterized genes. Further rationalization of these mutations is challenging due to the lack of gene(‐protein) knowledge. We then proceeded with an approach of directly analyzing mutations in an ALE isolate with known improved growth on syringate. The three genomically sequenced ALE isolates had 4 (1‐5b), 10 (2‐5a), and 8 (3‐5d) mutations, respectively (Supporting Information: Table S2 ). We retrofitted individual mutations from 1 to 5b into the Sy‐1 strain and observed between a 9% and 25% improvement in growth rate along with increased biomass accumulation (Table 2 and Figure 4 ). Of the four mutations tested, the agmR SNP had the greatest impact on growth, including the highest growth rate and one of the highest maximum OD. The mutation was also found in 85.1% of the sequenced DNA in the final ALE culture of Replicate 1, indicating that it was favorably propagated under the selective pressure. The agmR gene encodes a LuxR family transcriptional regulator. As a relatively understudied protein, limited literature showed that AgmR may play a role in regulating the expression of coenzyme PQQ synthesis protein A (PqqA) and ABC transport genes PP_2667 and PP_2669 (Vrionis et al., 2002 ). PQQ is a key component in bacterial redox metabolism, and often acts as a cofactor in alcohol dehydrogenases. Changes in the intracellular level of PQQ may lead to the change in cellular redox state, which was shown to be an important factor in the bacterial metabolism of aromatic compounds (Henson et al., 2018 ). The ABC transport system encoded by PP_2267‐PP_2269 was hypothesized to function in the uptake of 2‐phenylethanol based on their essentiality for growth (Wehrmann et al., 2019 ). Structural similarity between 2‐phenylethanol and syringate indicates possible crossactivity towards syringate as the substrate. Both the biosynthesis of PQQ and a potential ABC transporter are relevant to syringate utilization and changes in their expression level could lead to the phenotypic change of the Sy‐1 agmR‐SNP strain. Introduction of the other three mutations in Sy‐1 led to similar changes in growth characteristics on syringate. A possible role in improving cells' energy metabolism (Martínez‐García et al., 2014 ) by the fleQ SNP is discussed above. Two additional fleQ mutations were observed in the final ALE culture and the isolate of Replicate 2. The gbdR SNP identified in 1‐5b was found as a dominant mutation in the final ALE culture of both Replicate 1 (90.2%) and Replicate 3 (90.5%). In addition, two other mutations in the gbdR gene were observed in the final ALE culture and the isolate of replicate 2. The high mutation frequency in gbdR indicates its possible role in regulating cellular function(s) that is relevant to syringate metabolism. As a transcriptional activator, the GbdR in P. aeruginosa activates the transcription of the cbcXWV gene cluster, which encodes a primary ABC transporter of choline (Malek et al., 2011 ; Wargo, 2013 ). The SNP at the intergenic region of the gstB and yadG genes in strain 1‐5b was also observed in the final ALE culture and the isolate of Replicate 3. It is possible that the mutation, which is upstream of the yadG , affects the expression level of this transporter protein. Although implied with transport‐related function, a direct link between the gbdR SNP, the gstB‐yadG SNP, and syringate catabolism requires further investigations. All four Sy‐1 strains with a single mutation from 1 to 5b demonstrated improved growth on syringate, but none of them reproduced the characteristics of the 1‐5b strain. In particular, an increased expression level of the vanA gene was observed in 1‐5b, but not in the reverse‐engineered strains (Figure 2b ). The results indicate that every mutation confers benefits to a limited extent. The drastic improvement observed in 1‐5b is likely due to additive or synergistic effects of mutations in multiple genes. A larger number of mutations were identified in isolates 2‐5a and 3‐5d from ALE Replicates 2 and 3, respectively. Besides sharing a few gene targets of mutation, the majority of the mutation sites in these two strains are not presented in strain 1‐5b. Further deconvoluting the genetic cause(s) of their phenotypic change is more challenging. The observation showed us that this engineering problem potentially has multiple solutions, which cannot be easily envisioned through a rational approach."
} | 3,472 |
35524438 | PMC9378539 | pmc | 170 | {
"abstract": "Abstract Lignin is a largely untapped source for the bioproduction of value‐added chemicals. Pseudomonas putida KT2440 has emerged as a strong candidate for bioprocessing of lignin feedstocks due to its resistance to several industrial solvents, broad metabolic capabilities, and genetic amenability. Here we demonstrate the engineering of P. putida for the ability to metabolize syringic acid, one of the major products that comes from the breakdown of the syringyl component of lignin. The rational design was first applied for the construction of strain Sy‐1 by overexpressing a native vanillate demethylase. Subsequent adaptive laboratory evolution (ALE) led to the generation of mutations that achieved robust growth on syringic acid as a sole carbon source. The best mutant showed a 30% increase in the growth rate over the original engineered strain. Genomic sequencing revealed multiple mutations repeated in separate evolved replicates. Reverse engineering of mutations identified in agmR, gbdR, fleQ , and the intergenic region of gstB and yadG into the parental strain recaptured the improved growth of the evolved strains to varied extent. These findings thus reveal the ability of P. putida to utilize lignin more fully as a feedstock and make it a more economically viable chassis for chemical production.",
"conclusion": "5 CONCLUSIONS AND FUTURE WORK In this study, we successfully engineered P. putida KT2440 derivatives for robust growth using syringate as the sole carbon source through a combined approach of rational strain design and adaptive lab evolution. Genome sequencing and genetic analysis revealed a few mutations that enhanced the strain's fitness. To better understand how each mutation alters cellular states, we currently are conducting omics studies, in particular transcriptomics and targeted proteomics. Meanwhile, improvement in the growth characteristics of the best mutant, 1‐5b, may be achieved through another round of combined efforts of rational design and ALE experiments in light of results from the omics studies. Lignin is an attractive carbon source for bioproduction due to the projected large volume from biofuel production. The efficient use of lignin plays a critical role in the bioeconomy. Our work reported here provides a solid basis for further strain improvements of P. putida KT2440, a promising chassis of bio‐industrial applications for lignin valorization.",
"introduction": "1 INTRODUCTION Lignin is a complex organic polymer that makes up 20%–35% of plant cell walls. It is polymerized from three monolignols: p ‐coumaryl alcohol, coniferyl alcohol, and sinapyl alcohol, which are categorized as the corresponding hydroxy (H), guaiacyl (G), and syringyl (S) lignin components (Figure 1a ; Feofilova & Mysyakina, 2016 ). Depending on the type of plants, the ratio of H, G, and S subunits in lignin varies, where the grass lignin consists of all subunits, the softwood lignin often has low S component (Gellerstedt & Henriksson, 2008 ). The heterogeneity of lignin makes it a difficult target for traditional catalytic and enzymatic refinement approaches (C. Li et al., 2015 ; Rinaldi et al., 2016 ; Schutyser et al., 2018 ; Sun et al., 2018 ; Zakzeski et al., 2010 ). As a result, most lignin streams generated in existing processing facilities are burned as a low‐grade fuel (Ragauskas et al., 2014 ). Recent development in the biofuel technology, in particular the potential use of lignocellulosics as the starting material, ignites research interests into the valorization of lignin streams through microbial conversions of its degradation products (Becker & Wittmann, 2019 ; Eltis & Singh, 2018 ; Linger et al., 2014 ). Breakdowns of lignin can lead to an assortment of compounds that preserve the core chemical structures of the three monolignols. Maximizing the value of lignin therefore calls for a complete utilization of all components in the lignin depolymerization mixture. Figure 1 Lignin synthesis and degradation pathways. (a) Schematics for the polymerization of monolignols (left) into polymeric lignin (center) followed by its depolymerization into model monomeric aromatic products (right). (b) Catabolic pathways of model depolymerization products in P. putida . The known reaction catalyzed by VanAB is marked with solid arrows. Potential reactions for the conversion of syringic acid into gallic acid using VanAB are shown using dashed arrows. (c) The native P. putida catabolic pathway for gallic acid through the intermediates, including 4‐oxalomesaconic acid (OMA) keto and enol forms, and 4‐carboxy‐4‐hydroxy‐2‐oxoadipic acid (CHA). Enzymes, gallic acid dioxygenase (GllA), OMA keto–enol tautomerase (GllD), OMAenol hydratase (GllB), and CHA aldolase (GllC). Several microorganisms, such as Pseudomonas putida (Salvachúa et al., 2020 ; Willett, 2019 ), Sphingobium sp. SYK‐6 (Araki et al., 2020 ; Gall et al., 2014 ; Meux et al., 2012 ; Sato et al., 2009 ), and Novosphingobium aromaticivorans (Bell et al., 2012 ; Cecil et al., 2018 ; Kontur et al., 2018 ; Perez et al., 2019 , 2020 ), are being investigated for their abilities to utilize lignin‐derived compounds. Among them, P. putida KT2440 has emerged as a suitable candidate for industrial bioprocessing of lignin due to its genetic tractability, high tolerance to industrial solvents, and readily available engineering tools (Martínez‐García & de Lorenzo, 2019 ; Nikel & de Lorenzo, 2018 ). KT2440 can utilize a large number of carbon substrates, including compounds obtained from the breakdown of lignin (Dos Santos et al., 2004 ; Nikel & de Lorenzo, 2018 ; Nikel et al., 2014 ). It encodes pathways to metabolize vanillate, ferulate, p‐ coumarate, and 4‐hydroxybenzoate which are the major degradation products from the G and H components of lignin (Harwood & Parales, 1996 ; Jiménez et al., 2002 ; Nogales et al., 2017 ) (Figure 1a,b ). Notably, it lacks the ability to use syringate or sinapate as a sole carbon source, which are compounds derived from the lignin's syringyl components (Figure 1a,b ). Successful engineering of S lignin metabolism in the KT2440 strain therefore will greatly boost its application in lignin valorization. In this report, we demonstrate the successful engineering of syringate metabolism in KT2440 through a combined approach of chromosomal overexpression of the vanillate demethylase (VanAB) and subsequent adaptive lab evolution (ALE). Experimental analysis of mutations identified in the ALE mutants provided insights into potential causal genetic and metabolic changes that led to the improved growth on syringate.",
"discussion": "4 DISCUSSION \n P. putida is a promising bacterial chassis for the production of value‐added chemicals from lignin depolymerization compounds (Martínez‐García & de Lorenzo, 2019 ; Nikel & de Lorenzo, 2018 ; Nikel et al., 2016 ). Here, we demonstrated the successful engineering of P. putida KT2440 for growth on syringic acid by first overexpressing the native vanAB genes in strain Sy‐1 followed by its ALE experiment with syringic acid as the sole carbon source. ALE has been used to identify novel approaches to solve complex engineering problems in P. putida , including to enhance tolerance to toxic compounds, such as anthranilate (Kuepper et al., 2020 ), ionic liquids (Lim et al., 2020 ), solvents (Kusumawardhani et al., 2021 ), and lignin‐derived acids (Mohamed et al., 2020 ), and to improve the efficiency of catabolic pathways for ethylene glycol, 1,4‐butanediol (W.‐J. Li et al., 2020 ; W. J. Li et al., 2019 ), and xylose (Lim et al., 2021 ). Our ALE process spanned over 3 months to improve an average growth rate of the ALE culture from 0.4 to 2.8 day −1 , which is a combined result of shortened lag phase and increased growth rate. One of the best evolved mutant Sy‐1 1‐5b has a lag phase of 8 h and a growth rate of 0.115 h −1 , an 83.3% reduction, and 34.1% improvement, respectively. Genome sequencing data of ALE population and ALE isolates with improved growth revealed a plethora of genetic changes in the dynamic evolution process. We took a combined approach of literature searching and experimental verification to identify mutations that could contribute to the improved fitness of Sy‐1 mutants in media with syringate. Among unique mutations, certain ones emerged at the early stage of the ALE process but did not further prevail. For example, a total of eight unique SNPs were observed in gene PP_0168 (Supporting Information: Table S2 ), which encodes a possible surface adhesion protein. Since the formation of biofilm is a commonly observed microbial behavior in suspension cultures, such mutations were hypothesized as neutral drift, which are irrelevant to the evolutionary pressure. Mutations in fleQ (three unique mutations) and fliF (two unique mutations) (Supporting Information: Table S2 ) have been previously identified as common mutation targets in ALE experiments. It was hypothesized that the reduced or loss of ability to express flagellar saves energy from producing an unnecessary cell organelle under the shake flask environment. Strains of P. putida have been previously generated lacking the majority of flagellar genes which results in enhanced energy metabolism (Martínez‐García et al., 2014 ; Mohamed et al., 2020 ). We also observed a large number of mutations that occurred in intergenic regions and a small number of SNPs in uncharacterized genes. Further rationalization of these mutations is challenging due to the lack of gene(‐protein) knowledge. We then proceeded with an approach of directly analyzing mutations in an ALE isolate with known improved growth on syringate. The three genomically sequenced ALE isolates had 4 (1‐5b), 10 (2‐5a), and 8 (3‐5d) mutations, respectively (Supporting Information: Table S2 ). We retrofitted individual mutations from 1 to 5b into the Sy‐1 strain and observed between a 9% and 25% improvement in growth rate along with increased biomass accumulation (Table 2 and Figure 4 ). Of the four mutations tested, the agmR SNP had the greatest impact on growth, including the highest growth rate and one of the highest maximum OD. The mutation was also found in 85.1% of the sequenced DNA in the final ALE culture of Replicate 1, indicating that it was favorably propagated under the selective pressure. The agmR gene encodes a LuxR family transcriptional regulator. As a relatively understudied protein, limited literature showed that AgmR may play a role in regulating the expression of coenzyme PQQ synthesis protein A (PqqA) and ABC transport genes PP_2667 and PP_2669 (Vrionis et al., 2002 ). PQQ is a key component in bacterial redox metabolism, and often acts as a cofactor in alcohol dehydrogenases. Changes in the intracellular level of PQQ may lead to the change in cellular redox state, which was shown to be an important factor in the bacterial metabolism of aromatic compounds (Henson et al., 2018 ). The ABC transport system encoded by PP_2267‐PP_2269 was hypothesized to function in the uptake of 2‐phenylethanol based on their essentiality for growth (Wehrmann et al., 2019 ). Structural similarity between 2‐phenylethanol and syringate indicates possible crossactivity towards syringate as the substrate. Both the biosynthesis of PQQ and a potential ABC transporter are relevant to syringate utilization and changes in their expression level could lead to the phenotypic change of the Sy‐1 agmR‐SNP strain. Introduction of the other three mutations in Sy‐1 led to similar changes in growth characteristics on syringate. A possible role in improving cells' energy metabolism (Martínez‐García et al., 2014 ) by the fleQ SNP is discussed above. Two additional fleQ mutations were observed in the final ALE culture and the isolate of Replicate 2. The gbdR SNP identified in 1‐5b was found as a dominant mutation in the final ALE culture of both Replicate 1 (90.2%) and Replicate 3 (90.5%). In addition, two other mutations in the gbdR gene were observed in the final ALE culture and the isolate of replicate 2. The high mutation frequency in gbdR indicates its possible role in regulating cellular function(s) that is relevant to syringate metabolism. As a transcriptional activator, the GbdR in P. aeruginosa activates the transcription of the cbcXWV gene cluster, which encodes a primary ABC transporter of choline (Malek et al., 2011 ; Wargo, 2013 ). The SNP at the intergenic region of the gstB and yadG genes in strain 1‐5b was also observed in the final ALE culture and the isolate of Replicate 3. It is possible that the mutation, which is upstream of the yadG , affects the expression level of this transporter protein. Although implied with transport‐related function, a direct link between the gbdR SNP, the gstB‐yadG SNP, and syringate catabolism requires further investigations. All four Sy‐1 strains with a single mutation from 1 to 5b demonstrated improved growth on syringate, but none of them reproduced the characteristics of the 1‐5b strain. In particular, an increased expression level of the vanA gene was observed in 1‐5b, but not in the reverse‐engineered strains (Figure 2b ). The results indicate that every mutation confers benefits to a limited extent. The drastic improvement observed in 1‐5b is likely due to additive or synergistic effects of mutations in multiple genes. A larger number of mutations were identified in isolates 2‐5a and 3‐5d from ALE Replicates 2 and 3, respectively. Besides sharing a few gene targets of mutation, the majority of the mutation sites in these two strains are not presented in strain 1‐5b. Further deconvoluting the genetic cause(s) of their phenotypic change is more challenging. The observation showed us that this engineering problem potentially has multiple solutions, which cannot be easily envisioned through a rational approach."
} | 3,472 |
35524438 | PMC9378539 | pmc | 171 | {
"abstract": "Abstract Lignin is a largely untapped source for the bioproduction of value‐added chemicals. Pseudomonas putida KT2440 has emerged as a strong candidate for bioprocessing of lignin feedstocks due to its resistance to several industrial solvents, broad metabolic capabilities, and genetic amenability. Here we demonstrate the engineering of P. putida for the ability to metabolize syringic acid, one of the major products that comes from the breakdown of the syringyl component of lignin. The rational design was first applied for the construction of strain Sy‐1 by overexpressing a native vanillate demethylase. Subsequent adaptive laboratory evolution (ALE) led to the generation of mutations that achieved robust growth on syringic acid as a sole carbon source. The best mutant showed a 30% increase in the growth rate over the original engineered strain. Genomic sequencing revealed multiple mutations repeated in separate evolved replicates. Reverse engineering of mutations identified in agmR, gbdR, fleQ , and the intergenic region of gstB and yadG into the parental strain recaptured the improved growth of the evolved strains to varied extent. These findings thus reveal the ability of P. putida to utilize lignin more fully as a feedstock and make it a more economically viable chassis for chemical production.",
"conclusion": "5 CONCLUSIONS AND FUTURE WORK In this study, we successfully engineered P. putida KT2440 derivatives for robust growth using syringate as the sole carbon source through a combined approach of rational strain design and adaptive lab evolution. Genome sequencing and genetic analysis revealed a few mutations that enhanced the strain's fitness. To better understand how each mutation alters cellular states, we currently are conducting omics studies, in particular transcriptomics and targeted proteomics. Meanwhile, improvement in the growth characteristics of the best mutant, 1‐5b, may be achieved through another round of combined efforts of rational design and ALE experiments in light of results from the omics studies. Lignin is an attractive carbon source for bioproduction due to the projected large volume from biofuel production. The efficient use of lignin plays a critical role in the bioeconomy. Our work reported here provides a solid basis for further strain improvements of P. putida KT2440, a promising chassis of bio‐industrial applications for lignin valorization.",
"introduction": "1 INTRODUCTION Lignin is a complex organic polymer that makes up 20%–35% of plant cell walls. It is polymerized from three monolignols: p ‐coumaryl alcohol, coniferyl alcohol, and sinapyl alcohol, which are categorized as the corresponding hydroxy (H), guaiacyl (G), and syringyl (S) lignin components (Figure 1a ; Feofilova & Mysyakina, 2016 ). Depending on the type of plants, the ratio of H, G, and S subunits in lignin varies, where the grass lignin consists of all subunits, the softwood lignin often has low S component (Gellerstedt & Henriksson, 2008 ). The heterogeneity of lignin makes it a difficult target for traditional catalytic and enzymatic refinement approaches (C. Li et al., 2015 ; Rinaldi et al., 2016 ; Schutyser et al., 2018 ; Sun et al., 2018 ; Zakzeski et al., 2010 ). As a result, most lignin streams generated in existing processing facilities are burned as a low‐grade fuel (Ragauskas et al., 2014 ). Recent development in the biofuel technology, in particular the potential use of lignocellulosics as the starting material, ignites research interests into the valorization of lignin streams through microbial conversions of its degradation products (Becker & Wittmann, 2019 ; Eltis & Singh, 2018 ; Linger et al., 2014 ). Breakdowns of lignin can lead to an assortment of compounds that preserve the core chemical structures of the three monolignols. Maximizing the value of lignin therefore calls for a complete utilization of all components in the lignin depolymerization mixture. Figure 1 Lignin synthesis and degradation pathways. (a) Schematics for the polymerization of monolignols (left) into polymeric lignin (center) followed by its depolymerization into model monomeric aromatic products (right). (b) Catabolic pathways of model depolymerization products in P. putida . The known reaction catalyzed by VanAB is marked with solid arrows. Potential reactions for the conversion of syringic acid into gallic acid using VanAB are shown using dashed arrows. (c) The native P. putida catabolic pathway for gallic acid through the intermediates, including 4‐oxalomesaconic acid (OMA) keto and enol forms, and 4‐carboxy‐4‐hydroxy‐2‐oxoadipic acid (CHA). Enzymes, gallic acid dioxygenase (GllA), OMA keto–enol tautomerase (GllD), OMAenol hydratase (GllB), and CHA aldolase (GllC). Several microorganisms, such as Pseudomonas putida (Salvachúa et al., 2020 ; Willett, 2019 ), Sphingobium sp. SYK‐6 (Araki et al., 2020 ; Gall et al., 2014 ; Meux et al., 2012 ; Sato et al., 2009 ), and Novosphingobium aromaticivorans (Bell et al., 2012 ; Cecil et al., 2018 ; Kontur et al., 2018 ; Perez et al., 2019 , 2020 ), are being investigated for their abilities to utilize lignin‐derived compounds. Among them, P. putida KT2440 has emerged as a suitable candidate for industrial bioprocessing of lignin due to its genetic tractability, high tolerance to industrial solvents, and readily available engineering tools (Martínez‐García & de Lorenzo, 2019 ; Nikel & de Lorenzo, 2018 ). KT2440 can utilize a large number of carbon substrates, including compounds obtained from the breakdown of lignin (Dos Santos et al., 2004 ; Nikel & de Lorenzo, 2018 ; Nikel et al., 2014 ). It encodes pathways to metabolize vanillate, ferulate, p‐ coumarate, and 4‐hydroxybenzoate which are the major degradation products from the G and H components of lignin (Harwood & Parales, 1996 ; Jiménez et al., 2002 ; Nogales et al., 2017 ) (Figure 1a,b ). Notably, it lacks the ability to use syringate or sinapate as a sole carbon source, which are compounds derived from the lignin's syringyl components (Figure 1a,b ). Successful engineering of S lignin metabolism in the KT2440 strain therefore will greatly boost its application in lignin valorization. In this report, we demonstrate the successful engineering of syringate metabolism in KT2440 through a combined approach of chromosomal overexpression of the vanillate demethylase (VanAB) and subsequent adaptive lab evolution (ALE). Experimental analysis of mutations identified in the ALE mutants provided insights into potential causal genetic and metabolic changes that led to the improved growth on syringate.",
"discussion": "4 DISCUSSION \n P. putida is a promising bacterial chassis for the production of value‐added chemicals from lignin depolymerization compounds (Martínez‐García & de Lorenzo, 2019 ; Nikel & de Lorenzo, 2018 ; Nikel et al., 2016 ). Here, we demonstrated the successful engineering of P. putida KT2440 for growth on syringic acid by first overexpressing the native vanAB genes in strain Sy‐1 followed by its ALE experiment with syringic acid as the sole carbon source. ALE has been used to identify novel approaches to solve complex engineering problems in P. putida , including to enhance tolerance to toxic compounds, such as anthranilate (Kuepper et al., 2020 ), ionic liquids (Lim et al., 2020 ), solvents (Kusumawardhani et al., 2021 ), and lignin‐derived acids (Mohamed et al., 2020 ), and to improve the efficiency of catabolic pathways for ethylene glycol, 1,4‐butanediol (W.‐J. Li et al., 2020 ; W. J. Li et al., 2019 ), and xylose (Lim et al., 2021 ). Our ALE process spanned over 3 months to improve an average growth rate of the ALE culture from 0.4 to 2.8 day −1 , which is a combined result of shortened lag phase and increased growth rate. One of the best evolved mutant Sy‐1 1‐5b has a lag phase of 8 h and a growth rate of 0.115 h −1 , an 83.3% reduction, and 34.1% improvement, respectively. Genome sequencing data of ALE population and ALE isolates with improved growth revealed a plethora of genetic changes in the dynamic evolution process. We took a combined approach of literature searching and experimental verification to identify mutations that could contribute to the improved fitness of Sy‐1 mutants in media with syringate. Among unique mutations, certain ones emerged at the early stage of the ALE process but did not further prevail. For example, a total of eight unique SNPs were observed in gene PP_0168 (Supporting Information: Table S2 ), which encodes a possible surface adhesion protein. Since the formation of biofilm is a commonly observed microbial behavior in suspension cultures, such mutations were hypothesized as neutral drift, which are irrelevant to the evolutionary pressure. Mutations in fleQ (three unique mutations) and fliF (two unique mutations) (Supporting Information: Table S2 ) have been previously identified as common mutation targets in ALE experiments. It was hypothesized that the reduced or loss of ability to express flagellar saves energy from producing an unnecessary cell organelle under the shake flask environment. Strains of P. putida have been previously generated lacking the majority of flagellar genes which results in enhanced energy metabolism (Martínez‐García et al., 2014 ; Mohamed et al., 2020 ). We also observed a large number of mutations that occurred in intergenic regions and a small number of SNPs in uncharacterized genes. Further rationalization of these mutations is challenging due to the lack of gene(‐protein) knowledge. We then proceeded with an approach of directly analyzing mutations in an ALE isolate with known improved growth on syringate. The three genomically sequenced ALE isolates had 4 (1‐5b), 10 (2‐5a), and 8 (3‐5d) mutations, respectively (Supporting Information: Table S2 ). We retrofitted individual mutations from 1 to 5b into the Sy‐1 strain and observed between a 9% and 25% improvement in growth rate along with increased biomass accumulation (Table 2 and Figure 4 ). Of the four mutations tested, the agmR SNP had the greatest impact on growth, including the highest growth rate and one of the highest maximum OD. The mutation was also found in 85.1% of the sequenced DNA in the final ALE culture of Replicate 1, indicating that it was favorably propagated under the selective pressure. The agmR gene encodes a LuxR family transcriptional regulator. As a relatively understudied protein, limited literature showed that AgmR may play a role in regulating the expression of coenzyme PQQ synthesis protein A (PqqA) and ABC transport genes PP_2667 and PP_2669 (Vrionis et al., 2002 ). PQQ is a key component in bacterial redox metabolism, and often acts as a cofactor in alcohol dehydrogenases. Changes in the intracellular level of PQQ may lead to the change in cellular redox state, which was shown to be an important factor in the bacterial metabolism of aromatic compounds (Henson et al., 2018 ). The ABC transport system encoded by PP_2267‐PP_2269 was hypothesized to function in the uptake of 2‐phenylethanol based on their essentiality for growth (Wehrmann et al., 2019 ). Structural similarity between 2‐phenylethanol and syringate indicates possible crossactivity towards syringate as the substrate. Both the biosynthesis of PQQ and a potential ABC transporter are relevant to syringate utilization and changes in their expression level could lead to the phenotypic change of the Sy‐1 agmR‐SNP strain. Introduction of the other three mutations in Sy‐1 led to similar changes in growth characteristics on syringate. A possible role in improving cells' energy metabolism (Martínez‐García et al., 2014 ) by the fleQ SNP is discussed above. Two additional fleQ mutations were observed in the final ALE culture and the isolate of Replicate 2. The gbdR SNP identified in 1‐5b was found as a dominant mutation in the final ALE culture of both Replicate 1 (90.2%) and Replicate 3 (90.5%). In addition, two other mutations in the gbdR gene were observed in the final ALE culture and the isolate of replicate 2. The high mutation frequency in gbdR indicates its possible role in regulating cellular function(s) that is relevant to syringate metabolism. As a transcriptional activator, the GbdR in P. aeruginosa activates the transcription of the cbcXWV gene cluster, which encodes a primary ABC transporter of choline (Malek et al., 2011 ; Wargo, 2013 ). The SNP at the intergenic region of the gstB and yadG genes in strain 1‐5b was also observed in the final ALE culture and the isolate of Replicate 3. It is possible that the mutation, which is upstream of the yadG , affects the expression level of this transporter protein. Although implied with transport‐related function, a direct link between the gbdR SNP, the gstB‐yadG SNP, and syringate catabolism requires further investigations. All four Sy‐1 strains with a single mutation from 1 to 5b demonstrated improved growth on syringate, but none of them reproduced the characteristics of the 1‐5b strain. In particular, an increased expression level of the vanA gene was observed in 1‐5b, but not in the reverse‐engineered strains (Figure 2b ). The results indicate that every mutation confers benefits to a limited extent. The drastic improvement observed in 1‐5b is likely due to additive or synergistic effects of mutations in multiple genes. A larger number of mutations were identified in isolates 2‐5a and 3‐5d from ALE Replicates 2 and 3, respectively. Besides sharing a few gene targets of mutation, the majority of the mutation sites in these two strains are not presented in strain 1‐5b. Further deconvoluting the genetic cause(s) of their phenotypic change is more challenging. The observation showed us that this engineering problem potentially has multiple solutions, which cannot be easily envisioned through a rational approach."
} | 3,472 |
35524438 | PMC9378539 | pmc | 171 | {
"abstract": "Abstract Lignin is a largely untapped source for the bioproduction of value‐added chemicals. Pseudomonas putida KT2440 has emerged as a strong candidate for bioprocessing of lignin feedstocks due to its resistance to several industrial solvents, broad metabolic capabilities, and genetic amenability. Here we demonstrate the engineering of P. putida for the ability to metabolize syringic acid, one of the major products that comes from the breakdown of the syringyl component of lignin. The rational design was first applied for the construction of strain Sy‐1 by overexpressing a native vanillate demethylase. Subsequent adaptive laboratory evolution (ALE) led to the generation of mutations that achieved robust growth on syringic acid as a sole carbon source. The best mutant showed a 30% increase in the growth rate over the original engineered strain. Genomic sequencing revealed multiple mutations repeated in separate evolved replicates. Reverse engineering of mutations identified in agmR, gbdR, fleQ , and the intergenic region of gstB and yadG into the parental strain recaptured the improved growth of the evolved strains to varied extent. These findings thus reveal the ability of P. putida to utilize lignin more fully as a feedstock and make it a more economically viable chassis for chemical production.",
"conclusion": "5 CONCLUSIONS AND FUTURE WORK In this study, we successfully engineered P. putida KT2440 derivatives for robust growth using syringate as the sole carbon source through a combined approach of rational strain design and adaptive lab evolution. Genome sequencing and genetic analysis revealed a few mutations that enhanced the strain's fitness. To better understand how each mutation alters cellular states, we currently are conducting omics studies, in particular transcriptomics and targeted proteomics. Meanwhile, improvement in the growth characteristics of the best mutant, 1‐5b, may be achieved through another round of combined efforts of rational design and ALE experiments in light of results from the omics studies. Lignin is an attractive carbon source for bioproduction due to the projected large volume from biofuel production. The efficient use of lignin plays a critical role in the bioeconomy. Our work reported here provides a solid basis for further strain improvements of P. putida KT2440, a promising chassis of bio‐industrial applications for lignin valorization.",
"introduction": "1 INTRODUCTION Lignin is a complex organic polymer that makes up 20%–35% of plant cell walls. It is polymerized from three monolignols: p ‐coumaryl alcohol, coniferyl alcohol, and sinapyl alcohol, which are categorized as the corresponding hydroxy (H), guaiacyl (G), and syringyl (S) lignin components (Figure 1a ; Feofilova & Mysyakina, 2016 ). Depending on the type of plants, the ratio of H, G, and S subunits in lignin varies, where the grass lignin consists of all subunits, the softwood lignin often has low S component (Gellerstedt & Henriksson, 2008 ). The heterogeneity of lignin makes it a difficult target for traditional catalytic and enzymatic refinement approaches (C. Li et al., 2015 ; Rinaldi et al., 2016 ; Schutyser et al., 2018 ; Sun et al., 2018 ; Zakzeski et al., 2010 ). As a result, most lignin streams generated in existing processing facilities are burned as a low‐grade fuel (Ragauskas et al., 2014 ). Recent development in the biofuel technology, in particular the potential use of lignocellulosics as the starting material, ignites research interests into the valorization of lignin streams through microbial conversions of its degradation products (Becker & Wittmann, 2019 ; Eltis & Singh, 2018 ; Linger et al., 2014 ). Breakdowns of lignin can lead to an assortment of compounds that preserve the core chemical structures of the three monolignols. Maximizing the value of lignin therefore calls for a complete utilization of all components in the lignin depolymerization mixture. Figure 1 Lignin synthesis and degradation pathways. (a) Schematics for the polymerization of monolignols (left) into polymeric lignin (center) followed by its depolymerization into model monomeric aromatic products (right). (b) Catabolic pathways of model depolymerization products in P. putida . The known reaction catalyzed by VanAB is marked with solid arrows. Potential reactions for the conversion of syringic acid into gallic acid using VanAB are shown using dashed arrows. (c) The native P. putida catabolic pathway for gallic acid through the intermediates, including 4‐oxalomesaconic acid (OMA) keto and enol forms, and 4‐carboxy‐4‐hydroxy‐2‐oxoadipic acid (CHA). Enzymes, gallic acid dioxygenase (GllA), OMA keto–enol tautomerase (GllD), OMAenol hydratase (GllB), and CHA aldolase (GllC). Several microorganisms, such as Pseudomonas putida (Salvachúa et al., 2020 ; Willett, 2019 ), Sphingobium sp. SYK‐6 (Araki et al., 2020 ; Gall et al., 2014 ; Meux et al., 2012 ; Sato et al., 2009 ), and Novosphingobium aromaticivorans (Bell et al., 2012 ; Cecil et al., 2018 ; Kontur et al., 2018 ; Perez et al., 2019 , 2020 ), are being investigated for their abilities to utilize lignin‐derived compounds. Among them, P. putida KT2440 has emerged as a suitable candidate for industrial bioprocessing of lignin due to its genetic tractability, high tolerance to industrial solvents, and readily available engineering tools (Martínez‐García & de Lorenzo, 2019 ; Nikel & de Lorenzo, 2018 ). KT2440 can utilize a large number of carbon substrates, including compounds obtained from the breakdown of lignin (Dos Santos et al., 2004 ; Nikel & de Lorenzo, 2018 ; Nikel et al., 2014 ). It encodes pathways to metabolize vanillate, ferulate, p‐ coumarate, and 4‐hydroxybenzoate which are the major degradation products from the G and H components of lignin (Harwood & Parales, 1996 ; Jiménez et al., 2002 ; Nogales et al., 2017 ) (Figure 1a,b ). Notably, it lacks the ability to use syringate or sinapate as a sole carbon source, which are compounds derived from the lignin's syringyl components (Figure 1a,b ). Successful engineering of S lignin metabolism in the KT2440 strain therefore will greatly boost its application in lignin valorization. In this report, we demonstrate the successful engineering of syringate metabolism in KT2440 through a combined approach of chromosomal overexpression of the vanillate demethylase (VanAB) and subsequent adaptive lab evolution (ALE). Experimental analysis of mutations identified in the ALE mutants provided insights into potential causal genetic and metabolic changes that led to the improved growth on syringate.",
"discussion": "4 DISCUSSION \n P. putida is a promising bacterial chassis for the production of value‐added chemicals from lignin depolymerization compounds (Martínez‐García & de Lorenzo, 2019 ; Nikel & de Lorenzo, 2018 ; Nikel et al., 2016 ). Here, we demonstrated the successful engineering of P. putida KT2440 for growth on syringic acid by first overexpressing the native vanAB genes in strain Sy‐1 followed by its ALE experiment with syringic acid as the sole carbon source. ALE has been used to identify novel approaches to solve complex engineering problems in P. putida , including to enhance tolerance to toxic compounds, such as anthranilate (Kuepper et al., 2020 ), ionic liquids (Lim et al., 2020 ), solvents (Kusumawardhani et al., 2021 ), and lignin‐derived acids (Mohamed et al., 2020 ), and to improve the efficiency of catabolic pathways for ethylene glycol, 1,4‐butanediol (W.‐J. Li et al., 2020 ; W. J. Li et al., 2019 ), and xylose (Lim et al., 2021 ). Our ALE process spanned over 3 months to improve an average growth rate of the ALE culture from 0.4 to 2.8 day −1 , which is a combined result of shortened lag phase and increased growth rate. One of the best evolved mutant Sy‐1 1‐5b has a lag phase of 8 h and a growth rate of 0.115 h −1 , an 83.3% reduction, and 34.1% improvement, respectively. Genome sequencing data of ALE population and ALE isolates with improved growth revealed a plethora of genetic changes in the dynamic evolution process. We took a combined approach of literature searching and experimental verification to identify mutations that could contribute to the improved fitness of Sy‐1 mutants in media with syringate. Among unique mutations, certain ones emerged at the early stage of the ALE process but did not further prevail. For example, a total of eight unique SNPs were observed in gene PP_0168 (Supporting Information: Table S2 ), which encodes a possible surface adhesion protein. Since the formation of biofilm is a commonly observed microbial behavior in suspension cultures, such mutations were hypothesized as neutral drift, which are irrelevant to the evolutionary pressure. Mutations in fleQ (three unique mutations) and fliF (two unique mutations) (Supporting Information: Table S2 ) have been previously identified as common mutation targets in ALE experiments. It was hypothesized that the reduced or loss of ability to express flagellar saves energy from producing an unnecessary cell organelle under the shake flask environment. Strains of P. putida have been previously generated lacking the majority of flagellar genes which results in enhanced energy metabolism (Martínez‐García et al., 2014 ; Mohamed et al., 2020 ). We also observed a large number of mutations that occurred in intergenic regions and a small number of SNPs in uncharacterized genes. Further rationalization of these mutations is challenging due to the lack of gene(‐protein) knowledge. We then proceeded with an approach of directly analyzing mutations in an ALE isolate with known improved growth on syringate. The three genomically sequenced ALE isolates had 4 (1‐5b), 10 (2‐5a), and 8 (3‐5d) mutations, respectively (Supporting Information: Table S2 ). We retrofitted individual mutations from 1 to 5b into the Sy‐1 strain and observed between a 9% and 25% improvement in growth rate along with increased biomass accumulation (Table 2 and Figure 4 ). Of the four mutations tested, the agmR SNP had the greatest impact on growth, including the highest growth rate and one of the highest maximum OD. The mutation was also found in 85.1% of the sequenced DNA in the final ALE culture of Replicate 1, indicating that it was favorably propagated under the selective pressure. The agmR gene encodes a LuxR family transcriptional regulator. As a relatively understudied protein, limited literature showed that AgmR may play a role in regulating the expression of coenzyme PQQ synthesis protein A (PqqA) and ABC transport genes PP_2667 and PP_2669 (Vrionis et al., 2002 ). PQQ is a key component in bacterial redox metabolism, and often acts as a cofactor in alcohol dehydrogenases. Changes in the intracellular level of PQQ may lead to the change in cellular redox state, which was shown to be an important factor in the bacterial metabolism of aromatic compounds (Henson et al., 2018 ). The ABC transport system encoded by PP_2267‐PP_2269 was hypothesized to function in the uptake of 2‐phenylethanol based on their essentiality for growth (Wehrmann et al., 2019 ). Structural similarity between 2‐phenylethanol and syringate indicates possible crossactivity towards syringate as the substrate. Both the biosynthesis of PQQ and a potential ABC transporter are relevant to syringate utilization and changes in their expression level could lead to the phenotypic change of the Sy‐1 agmR‐SNP strain. Introduction of the other three mutations in Sy‐1 led to similar changes in growth characteristics on syringate. A possible role in improving cells' energy metabolism (Martínez‐García et al., 2014 ) by the fleQ SNP is discussed above. Two additional fleQ mutations were observed in the final ALE culture and the isolate of Replicate 2. The gbdR SNP identified in 1‐5b was found as a dominant mutation in the final ALE culture of both Replicate 1 (90.2%) and Replicate 3 (90.5%). In addition, two other mutations in the gbdR gene were observed in the final ALE culture and the isolate of replicate 2. The high mutation frequency in gbdR indicates its possible role in regulating cellular function(s) that is relevant to syringate metabolism. As a transcriptional activator, the GbdR in P. aeruginosa activates the transcription of the cbcXWV gene cluster, which encodes a primary ABC transporter of choline (Malek et al., 2011 ; Wargo, 2013 ). The SNP at the intergenic region of the gstB and yadG genes in strain 1‐5b was also observed in the final ALE culture and the isolate of Replicate 3. It is possible that the mutation, which is upstream of the yadG , affects the expression level of this transporter protein. Although implied with transport‐related function, a direct link between the gbdR SNP, the gstB‐yadG SNP, and syringate catabolism requires further investigations. All four Sy‐1 strains with a single mutation from 1 to 5b demonstrated improved growth on syringate, but none of them reproduced the characteristics of the 1‐5b strain. In particular, an increased expression level of the vanA gene was observed in 1‐5b, but not in the reverse‐engineered strains (Figure 2b ). The results indicate that every mutation confers benefits to a limited extent. The drastic improvement observed in 1‐5b is likely due to additive or synergistic effects of mutations in multiple genes. A larger number of mutations were identified in isolates 2‐5a and 3‐5d from ALE Replicates 2 and 3, respectively. Besides sharing a few gene targets of mutation, the majority of the mutation sites in these two strains are not presented in strain 1‐5b. Further deconvoluting the genetic cause(s) of their phenotypic change is more challenging. The observation showed us that this engineering problem potentially has multiple solutions, which cannot be easily envisioned through a rational approach."
} | 3,472 |
37765758 | PMC10536645 | pmc | 173 | {
"abstract": "Over the past decade, the artificial neural networks domain has seen a considerable embracement of deep neural networks among many applications. However, deep neural networks are typically computationally complex and consume high power, hindering their applicability for resource-constrained applications, such as self-driving vehicles, drones, and robotics. Spiking neural networks, often employed to bridge the gap between machine learning and neuroscience fields, are considered a promising solution for resource-constrained applications. Since deploying spiking neural networks on traditional von-Newman architectures requires significant processing time and high power, typically, neuromorphic hardware is created to execute spiking neural networks. The objective of neuromorphic devices is to mimic the distinctive functionalities of the human brain in terms of energy efficiency, computational power, and robust learning. Furthermore, natural language processing, a machine learning technique, has been widely utilized to aid machines in comprehending human language. However, natural language processing techniques cannot also be deployed efficiently on traditional computing platforms. In this research work, we strive to enhance the natural language processing traits/abilities by harnessing and integrating the SNNs traits, as well as deploying the integrated solution on neuromorphic hardware, efficiently and effectively. To facilitate this endeavor, we propose a novel, unique, and efficient sentiment analysis model created using a large-scale SNN model on SpiNNaker neuromorphic hardware that responds to user inputs. SpiNNaker neuromorphic hardware typically can simulate large spiking neural networks in real time and consumes low power. We initially create an artificial neural networks model, and then train the model using an Internet Movie Database (IMDB) dataset. Next, the pre-trained artificial neural networks model is converted into our proposed spiking neural networks model, called a spiking sentiment analysis (SSA) model. Our SSA model using SpiNNaker, called SSA-SpiNNaker, is created in such a way to respond to user inputs with a positive or negative response. Our proposed SSA-SpiNNaker model achieves 100% accuracy and only consumes 3970 Joules of energy, while processing around 10,000 words and predicting a positive/negative review. Our experimental results and analysis demonstrate that by leveraging the parallel and distributed capabilities of SpiNNaker, our proposed SSA-SpiNNaker model achieves better performance compared to artificial neural networks models. Our investigation into existing works revealed that no similar models exist in the published literature, demonstrating the uniqueness of our proposed model. Our proposed work would offer a synergy between SNNs and NLP within the neuromorphic computing domain, in order to address many challenges in this domain, including computational complexity and power consumption. Our proposed model would not only enhance the capabilities of sentiment analysis but also contribute to the advancement of brain-inspired computing. Our proposed model could be utilized in other resource-constrained and low-power applications, such as robotics, autonomous, and smart systems.",
"conclusion": "5. Conclusions and Future Work In this paper, we introduced a novel, unique, and efficient neuromorphic NLP sentiment analysis model based on the SNNs, deployed on the SpiNNaker neuromorphic platform. Our goal was to enhance the NLP traits/abilities, by harnessing and integrating the SNNs traits, as well as deploying the integrated solution on neuromorphic hardware. Our proposed SNN-based sentiment analysis model was created in such a way: to be energy efficient and also to be faster than the ANN-based NLP models, while predicting the sentiment. Our proposed SNN model, converted from our ANN model, is trained and deployed on the SpiNNaker hardware, which enables leveraging energy efficiency, and inherent parallelism of SNNs. Our proposed SSA-SpiNNaker model achieved 100% accuracy, and only consumed 3970 Joules of energy, while processing around 10,000 words and predicting a positive/negative review. From these results and analysis, it is evident that parallel processing capabilities and low energy consumption, associated with our proposed SNN model, are indeed a promising avenue for NLP tasks, compared to the traditional ANN-based NLP models. Our research also clarifies the importance of our proposed SNN model by considering the neural dynamics and demonstrates that the spike-based computations impact the effectiveness of the overall NLP tasks. Furthermore, from our investigation on existing works, we could not find any models in the published literature that provided an SSA-SpiNNaker model similar to ours, demonstrating the uniqueness of our proposed model. As future work, we will further enhance our proposed SNN model and explore the applicability of SNNs to various NLP tasks, such as neural embeddings, as well as to tasks that are beyond sentiment analysis. Also, as future work, we are planning to introduce field programmable gate arrays (FPGAs)-based hardware architectures for our proposed models. This is mainly because our previous work and analysis ([ 54 , 55 , 56 , 57 , 58 ]) demonstrate that FPGAs are one of the best avenues to deploy/execute compute and data-intensive applications, such as SNNs, on resources-constrained devices. We will also incorporate dynamic reconfiguration techniques [ 59 , 60 ] to create dynamic reconfigurable architectures (similar to [ 61 , 62 , 63 ]) to integrate adaptability traits to our proposed models.",
"introduction": "1. Introduction In recent years, the Artificial Neural Networks (ANNs) domain has witnessed a significant adaptation of Deep Neural Networks (DNNs) across several fields, such as machine learning, computer vision, artificial intelligence, and natural language processing (NLP). DNNs are capable of accurately performing a wide range of tasks by training on massive datasets [ 1 ]. However, the energy consumption and computational cost required for training large volumes of datasets and for deploying the resulting applications have been of less importance; thus, they have been overlooked [ 2 , 3 ]. The DNNs typically consume high power and require large data storage [ 4 , 5 ]. Although there have been significant advancements in ANNs, ANNs were unable to achieve the same level of energy efficiency and online learning ability as biological neural networks [ 6 ]. Drawing inspiration from brain-inspired computing, one potential solution to address the issue of high-power consumption is to use the neuromorphic hardware with Spiking Neural Networks (SNNs). SNNs, often considered the third generation of neural networks, are emerging to bridge the gap between fields such as machine learning and neuroscience [ 7 ]. Unlike traditional neural networks that rely on continuous-valued signals, the SNNs work in continuous time [ 8 ]. In SNNs, the neurons communicate with each other using discrete electrical signals called spikes. Spikes model the behavior of the neurons more accurately and more biologically plausible than ANNs, thus making SNNs more energy efficient and computationally powerful than ANNs [ 9 ]. The neuron models of ANNs and SNNs differ from each other. For instance, ANNs do not have any memory and use sigmoid, tanh, or rectified linear unit (ReLU) as computational units, whereas SNNs have memory and use non-differentiable neuron models. Typically, large-scale SNN models consume high power and require high execution time when utilized/executed on classical Von Neumann architectures [ 10 ]. Hence, there is a need for high-speed and low-power hardware for executing large-scale SNN models. In this regard, existing neuromorphic platforms, such as SpiNNaker [ 11 ], Loihi [ 12 ], NeuroGrid [ 13 ], and TrueNorth by IBM [ 14 ], are expected to advance the applicability of large-scale SNNs in several emerging fields by offering energy-efficient high-speed computational solutions. SNNs have the functional similarities to biological neural networks, allowing them to embrace the sparsity and temporal coding found in biological systems [ 15 ]. However, SNNs are difficult to train because of their non-differentiable neuron models. In terms of speed performance, SNNs are inferior to DNNs. Nevertheless, due to the low power traits, SNNs are considered more efficient than DNNs [ 6 ]. Considering the aforementioned, in this research work, we propose a novel and unique neuromorphic NLP sentiment analysis model based on the SNNs, deployed on the SpiNNaker neuromorphic platform. Our proposed sentiment analysis model is created in such a way to be energy efficient and also to be faster than the ANN-based NLP models, while predicting the sentiment. Our experimental results illustrate that our proposed model is energy efficient and consumes low power when executed on the Internet Movie Database (IMDB) dataset to predict the reviews based on user inputs. Our results and analysis also demonstrate that our proposed model is more efficient and effective compared to the existing ANN-based sentiment analysis models in the published literature. In this paper, we make the following contributions: We introduce a novel, unique, and efficient sentiment analysis model for neuromorphic hardware using SNNs. Our proposed model is highly accurate, while detecting text and predicting the sentiment. Our proposed model is adaptable, energy efficient, and computationally effective. These traits make our model well suited for areas where energy efficiency is crucial and resources are scarce. We perform experiments to evaluate the feasibility and efficiency of our proposed model. Our results and analysis demonstrate that our model is efficient and accurate compared to the ANN-based sentiment analysis model. Our paper is organized as follows. Section 2 provides an overview of SNNs and their applications in various fields. In Section 3 , we discuss and present our proposed SNN model in detail, highlighting its unique features. In Section 4 , we present our experimental results and analysis, including the comparison of speed performance and energy consumption of our proposed SNN model with traditional existing ANN models. In Section 5 , we summarize, conclude, and discuss the potential impacts of SNNs in various fields and our future directions."
} | 2,608 |
31285755 | PMC6588928 | pmc | 174 | {
"abstract": "Microbial communities are ubiquitous in nature and exhibit several attractive features, such as sophisticated metabolic capabilities and strong environment robustness. Inspired by the advantages of natural microbial consortia, diverse artificial co-cultivation systems have been metabolically constructed for biofuels, chemicals and natural products production. In these co-cultivation systems, especially genetic engineering ones can reduce the metabolic burden caused by the complex of metabolic pathway through labor division, and improve the target product production significantly. This review summarized the most up-to-dated co-cultivation systems used for biofuels, chemicals and nature products production. In addition, major challenges associated with co-cultivation systems are also presented and discussed for meeting further industrial demands.",
"conclusion": "Conclusions In recent years, construction of co-cultivation systems for biofuels and chemicals production has attracted more and more attention. Not only limited to simply mix the wild strains, co-cultivation has also expanded into synthetic biology. The introduction of synthetic intercellular communication into the cell engineering toolbox will open new frontiers and greatly contribute to the future success of synthetic biology and its applications. Although the production could be improved when using co-cultivation systems, challenges still exist. Currently, studies associated with co-cultivation systems are mainly constricted at the levels of exchange of intermediate metabolites. Other elements of environmental variation, such as energy flux, signal exchange and nutrient cycling are still unknown. Only based on the comprehensive understanding of the interaction among microbes, the improvement of robustness, stability and reproducibility can be further achieved.",
"introduction": "Introduction Pure cultures dominate the current industrial bioprocesses; however, they are confronted with challenges due to the increased requirement for higher efficiency of production and fulfillment of more complicated tasks. In nature, 99% microorganisms exist in the form of microbial consortia [ 1 ]. Inspired by the omnipresent natural microbial consortia, more attention has been paid on the bioprocess development of artificial ones, which pools different engineered microorganisms in one pot [ 2 – 4 ]. However, different from natural microbial communities, which exist mainly for the survival and growth in the environment, the artificial microbial consortia are specifically constructed to broaden the scope of feedstocks, enhance the productivity of target bio-products, etc. [ 5 – 7 ]. Diverse microbial communities within the same or different species have been set up to realize more complicated tasks [ 8 – 10 ]. In addition to treatment of wastewater, biodegradation of textile azo dye and dispose of contaminated soil, recently, co-cultivation systems were also applied to produce biofuels (bioethanol, biobutanol, biodiesel, etc.), bulk chemicals (lactic acid, 2-keto- l -gulonic acid, etc.) and natural products (alkaloids, polyketides, terpenes, flavonoid, etc.) [ 11 – 21 ]. These artificial microbial consortia interact mutually through the interaction of synergism, commensalism, competition, mutualism, etc. (Fig. 1 ) [ 1 ]. Elaboration of the underlying mechanism in microbial communities, such as the exchange of intermediate metabolites, cell-to-cell electrical connections, communications, etc. would guide the design of artificial microbial consortia and further improve the robustness and stability of the co-cultivation systems [ 22 – 25 ]. Accordingly, this review summarizes the superiority of co-cultivation systems compared with pure cultures and the most updated advances in artificial microbial consortia for the production of biofuels and chemicals from renewable sources. Nevertheless, further application and development of microbial consortia are still confronted with challenges, such as the uncharacterized microbial interaction mechanisms, etc. Fig. 1 The schematic diagram for interaction modes of artificial microbial consortia. The interaction modes of artificial microbial consortia, including a commensalism, b mutualism, c competition and d parasitism Advantages of co-cultivation systems over pure cultures Compared with pure cultures, co-cultivation systems could broaden the substrate utilization spectra. Lignocellulose is the most abundant sustainable recourses; however, due to the complexity of cellulose-degrading systems, single strain generally can not directly utilize it to synthesize valuable products [ 26 , 27 ]. In general, two common strategies were developed: one is the incorporation of target product synthesis modules into cellulolytic microbes to achieve product generation from lignocellulose; the other is the introduction of cellulase systems into product-generating microbes (Fig. 2 a, b) [ 28 , 29 ]. However, the long and complex pathways including cellulase secretion and/or product synthesis would burden the metabolic stress and lead to low amounts of product generated [ 30 , 31 ]. On the contrary, microbial consortia offer a simpler and more efficient approach to achieve this goal through the so-called consolidated bioprocessing (CBP), in which enzymes production, substrate hydrolysis and microbial fermentation are completed in one single reactor. For example, setting up co-cultivation systems including cellulolytic Clostridium sp. and non-cellulolytic Thermoanaerobacter sp. can achieve ethanol production from cellulose through CBP. Argyros et al. [ 32 ] set up an artificial C. thermocellum – T. saccharolyticum co-cultivation system, in which organic acids formation pathways were both removed in these two constituent strains. 38 g/L of ethanol was finally produced from 92 g/L of Avicel, which was approximately 80% theoretical maximum, indicating that C. thermocellum could be a cornerstone of a robust cellulolytic platform. On the other hand, the lagged utilization of pentose in both hexose and pentose mixtures is commonly found in most microbes, known as carbon catabolic repression (CCR), when bacteria are exposed to two or more carbon sources [ 33 ]. The sequential utilization of component sugars of lignocellulose materials would reduce the whole processes efficiency. Microbial consortia enable to rationally utilize different substrates based on the specific metabolic pathway. A novel binary culture can solve the problem flexibly, in which one could only consume glucose and the other could only consume xylose, shifting the interaction modes from the competition to the commensalism [ 34 ]. Fig. 2 Comparison between pure cultures and microbial co-culturing systems for butanol production used lignocellulose. Two strategies for achievement of butanol production from lignocellulose via CBP. a the “native cellulolytic strategy”, in which butanol synthetic pathway was introduced into cellulolytic microorganism; b the “recombinant cellulolytic strategy”, in which cellulolytic enzymes were constructed into solventogenic ones. c The strategy for microbial co-culturing systems including lignocellulolytic microorganisms and solventogenic bacteria \n When the biosynthetic pathway of target product is long and complicated, a large number of genes would be heterologously expressed in single strains. Generally, the biochemical properties and expression levels of introduced enzymes vary to a large extent. A single host cell cannot provide the optimal environment to perform the function well for all enzymes, while microbial consortia can provide diversified cellular environments for different enzymes. Especially, when a biosynthetic pathway is composed of both prokaryotic and eukaryotic enzymes, a combination of bacterial and fungal hosts would be highly advantageous over using either host alone [ 35 ]. In addition, excessive cellular resources consumption and overwhelming metabolic burden often lead to the impaired growth and/or poor biosynthetic behavior of single host strain [ 36 ]. Microbial consortia can reduce this metabolic burden through the strategy of labor division, which not only benefits the growth of individual strains, but also improves the performance of overall bio-production (Fig. 3 ) [ 37 ]. Furthermore, insufficient supply of precursors or excessive accumulation of intermediate products could both influence the end-products generation. In pure culture, the relative expression level of different genes is adjusted through promoter strength, gene copy number, ribosomal binding site etc. [ 38 ]. Building microbial consortia is a straightforward way to flexibly balance the biosynthetic strength through changing strain–strain ratios [ 39 ]. Fig. 3 Illustration of the advantages and challenges of co-cultivation systems \n In pure cultures, most strains have individual suitable conditions for the growth. If cultural conditions changed, the growth and metabolism of strains would be affected. Microbial consortia could endure more changeable environments, providing an important new frontier for industrial production [ 1 ]. In microbial consortia, environmental disturbance can be dynamically balanced and regulated due to the coordination and cooperation of different strains. The undesired interference within different pathway modules in host strains would also be reduced [ 40 ]. Modular compartmentalization offers a new effective approach to limit negative interaction between pathway modules and improve the biosynthesis performance. Hence, microbial consortia commonly possess higher stability and robustness to environmental perturbations. Biofuels production by using co-cultivation systems Bioethanol As an environmentally friendly and sustainable source, biofuels production including bioethanol, biobutanol and biodiesel has gained considerable interests [ 41 – 43 ]. Bioethanol was regarded as one of the most promising biofuels, particularly as a carbon-neutral liquid transportation fuel [ 44 ]. Solventogenic yeasts, such as Saccharomyces cerevisiae and some bacteria, such as Thermoanaerobacter species are widely used to produce ethanol [ 45 – 47 ]. However, the feedstock spectrum is limited to some starchy-based materials [ 48 ]. Compared to grain-derived feedstocks, lignocellulose is a more economically feasible alternative because of its abundance and low cost [ 49 , 50 ]. An artificial Escherichia coli binary culture was constructed for direct conversion of hemicellulose into ethanol. The final ethanol concentration reached 2.84 g/L, which is 55% of the theoretical yield [ 51 ]. In this binary system, one E. coli strain was engineered to hydrolyze hemicellulose to xylooligosaccharides through co-expression of two hemicellulase genes. Xylooligosaccharide-utilizing enzymes were then over-expressed in the other E. coli strain to realize the conversion of xylooligosaccharides into ethanol. This co-cultivation system distributed the metabolic burden through extracellular and intracellular expression of different functional enzymes, resulting in the improved ethanol production over pure cultures. Furthermore, cellulase system can also be built in a microbial consortium. For example, dual-microbe Bacillus /yeast system was developed for cellulosic ethanol production. Recombinant B. subtilis carries eight cellulosomal genes originating from C. thermocellum : one scaffolding protein gene ( cipA ), one cell-surface anchor gene ( sdbA ), two exo-glucosidase genes ( celK and celS ), two endo-glucanase genes ( celA and celR ), and two xylanase genes ( xynC and xynZ ). The partner Kluyveromyces marxianus KY3-NpaBGS carries a glucosidase ( NpaBGS ) gene from rumen fungus. Ultimately, 9.5 g/L of ethanol was produced from 20 g/L of cellulose (Table 1 ) [ 52 ]. Table 1 Biofuels and chemicals production by co-cultivation systems Strains Subtracts Fermentation modes Products Titer Time References C. thermocellum – T. saccharolyticum 92 g/L avicel Batch Ethanol 38 g/L 146 h [ 32 ] E. coli E609Y/pCRAXEXYL– E. coli KO11/pBBKXYN 10 g/L xylan Batch Ethanol 2.8 g/L 60 h [ 51 ] C. thermocellum – K. marxianus 20 g/L glucan Batch Ethanol 9.5 g/L 5 days [ 52 ] C. thermocellum – Thermoanaerobacter strains 20 g/L cellulose Batch Ethanol 6.6 g/L ~ 6 days [ 56 ] C. phytofermentans – S. cerevisiae 100 g/L cellulose Batch Ethanol 22 g/L 400 h [ 57 ] C. thermocellum – C. beijerinckii 88.9 g/L alkali extracted corn cobs Batch Butanol 10.9 g/L 200 h [ 65 ] E. coli strain BuT-3E– E. coli strain BuT-8L-ato 20 g/L glucose Batch Butanol 5.5 g/L 24 h [ 67 ] Chlorella minutissima – A. awamori 10 g/L glycerol Batch Palmitic (C16:0) 35.02 mg/L – [ 73 ] Chlorella minutissima – A. awamori 10 g/L glycerol Batch Oleic (C18:1) 24.21 mg/L – [ 73 ] R. glutinis – Scenedesmus obliquus 50 g/L glucose Batch Total lipid ~6 g/L 4 days [ 74 ] T. reesei – L. pentosus 50 g/L avicel Batch Lactate 34.7 g/L 215 h [ 76 ] E. coli ALS1073– E. coli ALS1074 22 g/L glucose + 33 g/L xylose Batch Lactate 37 g/L 24 h [ 77 ] E. coli P5.2– E. coli BC 20 g/L glycerol Batch Muconic acid 2 g/L ~ 48 h [ 78 ] E. coli P6.6– E. coli BXC 13.2 g/L glucose + 6.6 g/L xylose Batch Muconic acid 4.7 g/L 72 h [ 34 ] G. oxydans – K. vulgare 80 g/L d -sorbitol Fed-batch 2-Keto- l -gulonic acid 76.6 g/L 36 h [ 82 ] E. coli – S. cerevisiae Xylose Fed-batch Oxygenated taxanes 33 mg/L 120 h [ 35 ] E. coli C5– E. coli p168 20 g/L glycerol Fed-batch Flavan-3-ols 40.7 mg/L 54 h [ 39 ] \n Considering the complex of lignocellulose degradation enzymes, co-culturing cellulolytic microorganism with ethanol-producing one is a convenient and flexible approach to produce ethanol from lignocellulose through CBP. Cellulolytic C. thermocellum is a model organism for CBP; however, its application was limited due to the low ethanol yield [ 53 – 55 ]. Considering its efficient capability of cellulose degradation, C. thermocellum can be co-cultured with non-cellulolytic Thermoanaerobacter strains (X514 and 39E), which showed high efficiency of ethanol production [ 56 ]. The final ethanol production achieved at 7.56 and 6.59 g/L, respectively, which were significantly improved by 194–440%. The labor division is straightforward in this system: C. thermocellum is mainly responsible for cellulolysis, while Thermoanaerobacter sp. takes charge of ethanol production owing to its high-efficient ethanol production capability. The interaction within these two strains was through the exchange of intermediate metabolites. Similarly, a co-cultivation system, in which cellulose hydrolysis and ethanol production were conducted by C. phytofermentans and S. cerevisiae, was set up [ 57 ]. Glucosidase gene was overexpressed in S. cerevisiae to hydrolyze cellodextrin intracellularly. The connection of separated pathway modules was facilitated by the expression of intermediate cellodextrin transporters in the downstream S. cerevisiae . Finally, 22 g/L of ethanol was obtained from 100 g/L of cellulose using this artificial co-cultivation system. Biobutanol Biobutanol, a four-carbon and straight-chained alcohol is considered as more advanced biofuel over ethanol owing to its higher heating value, better inter-solubility, lower heat of vaporization, higher viscosity and lower corrosivity [ 58 – 61 ]. Generally, butanol was synthesized through traditional acetone–butanol–ethanol (ABE) fermentation process by solventogenic Clostridium sp. [ 62 , 63 ]. However, most clostridia could not directly utilize polysaccharides, such as lignocellulose due to the inexpression of polysaccharide-degrading enzymes [ 64 ]. Hence, construction of microbial consortia may be an ideal strategy to achieve direct butanol production from renewable feedstocks (Fig. 2 c). For example, a co-cultivation system composed of different solventogenic consortia ( C. thermocellum ATCC 27405 and C. beijerinckii NCIMB 8052) was set up, which could directly produce butanol from lignocellulose [ 65 ]. The reducing sugars hydrolyzed by C. thermocellum ATCC 27405 were simultaneously metabolized by C. beijerinckii for butanol production. Meanwhile, the consumption of sugars could alleviate the feedback inhibition and further improve the degradation efficiency of alkali extracted corn cobs (AECC) by C. thermocellum . After optimization of cultivation temperature, 19.9 g/L of ABE (3.96 g/L of acetone, 10.9 g/L of butanol and 5.04 g/L of ethanol) were obtained from 88.9 g/L of AECC in 200 h, indicating the highest solvent production from lignocellulose through CBP (Table 1 ) [ 65 ]. Different from ethanol production, butanol synthetic pathway is more complex [ 66 ]. Introduction of butanol synthesis modules in model microorganisms, such as E. coli, would burden the metabolic stress. Whereas, dividing butanol biosynthetic pathway into butyrate-producing and butyrate-conversion modules in one co-culture system is more feasible. 5.5 g/L of butanol was finally produced in E. coli – E. coli system, which is twofold higher than that using pure culture [ 67 ]. Notably, volatile fatty acids travel freely across the cell membrane, which was recycled between the upstream and downstream E. coli strains to facilitate butyrate and butyryl-CoA inter-conversion. Biodiesel Biodiesel is another environmental-friendly biofuel, which can provide robust, massive, and enduring energy supply [ 68 , 69 ]. Naturally, oleaginous algae are the well-known biodiesel producers [ 70 ]. However, several constraints hindered its further application. One major issue is the slow-growing rate and mutually incongruous nature of biomass and lipid accumulation [ 71 , 72 ]. Co-cultivation of algae–fungus was proposed as an alternative approach for biodiesel production. An oleaginous fungus Aspergillus awamori was co-cultured with Chlorella minutissima MCC 27 and C. minutissima UTEX 2219, respectively. These two oleaginous algae–fungus consortia contain photoautotrophic green algae and obligate heterotrophic fungi. This system can utilize pure glycerol instead of glucose, which could reduce the production cost. A 2.6- and 3.9-fold increase in biomass and 3.4- and 5.1-fold increase in total lipid yields were observed in the co-cultures compared to the axenic cultures. Furthermore, C16:0 (31.26–35.02%) and C18:1 (21.14–24.21%) fatty acids were the major composites, suggesting that this co-culture system is a promising strategy for biodiesel production [ 73 ]. Microalgae are sunlight-driven cell factories that convert CO 2 into lipids and O 2 through the photosynthesis process. The production of O 2 could further facilitate the growth of aerobic yeast, while the yeast mutually provides CO 2 to the microalgae accompanied with the production of lipids. 40–50% of biomass and 60–70% of total lipids were increased compared to the single-culture batch [ 74 ]. The co-culture could provide the symbiotic environment for algae and yeast growth together, and the trace elements released through the natural lysis of the cells could be further utilized for the enhancement of cell growth. The co-culture of O 2 provider S. obliquus and CO 2 provider R. glutinis can offer gas transportation to both sides. Taken together, microbial consortia can be constructed not only within the same species, but also in different genus, such as fungus–bacterium. Each member in microbial consortia interacts mutually through the exchange of metabolites. These microbial co-cultures provide the opportunity to achieve direct conversion of renewable sources into biofuel, maximization of substrate utilization rate, enhancement of yield and production, and reduction of process costs. However, as an immature but promising technology, application of microbial consortia for biofuel production at industrial scale still poses several challenges, such as the stability of microbial members in co-cultivation systems. More research efforts are still needed to develop more robust and stable microbial consortia that could be used for biofuels production. Bulk chemicals production by using co-cultivation systems Lactic acid In addition to biofuels, a wide range of bulk chemicals have also been produced using co-cultivation systems. Taking lactic acid, a versatile platform as an example, it is mainly produced from starchy-based materials or mono-sugars, which limits its large-scale production [ 75 ]. Recently, an artificial consortium composed of aerobic cellulolytic fungus Trichoderma reesei and lactic acid-producing bacterium Lactobacilli pentosus was metabolically constructed [ 76 ]. T. reesei acts as cellulose degraders, and L. pentosus is a robust lactic acid producer. The stable coexistence of these two strains is mainly based on the interaction of competitive cheater and cooperator. 34.7 g/L of lactic acid was produced from 5% (w/w) microcrystalline cellulose (Table 1 ). As mentioned above, CCR commonly occurs in most microbes when using lignocellulosic hydrolysate as the substrate. To overcome this obstacle, novel microbial consortia were constructed, in which one could only consume glucose and the other could only consume xylose (Fig. 4 ). The xylose-selective (glucose deficient) strain E. coli ALS1073 was constructed through the deletion of pyruvate formate lyase ( pflB ), glucokinase ( glk ), phosphotransferase system ( ptsG ), and IID Man domain of the mannose PTS permease ( manZ ); while the glucose-selective (xylose deficient) strain E. coli ALS1074 has a pflB and xylose isomerase ( xylA ) deletion. The microbial consortium could simultaneously convert xylose and glucose into 37 g/L of lactate with a yield of 0.88 g/g [ 77 ]. In addition, the conversion rates of each sugar can be individually modulated to optimize the overall process. Fig. 4 The strategy for improvement of bulk chemicals production by co-cultivation systems disengaged from competition interaction Muconic acid Muconic acid (MA) is another important bulk chemical; however, its production meets a challenge caused by the insufficient functional expression of enzymes due to the complex of synthesis pathway. Accordingly, an E. coli – E. coli binary consortium was constructed to achieve direct MA production from glycerol [ 78 ]. Two modules were constructed in different strains: the upstream strain E. coli P5.2 contained only the shikimate pathway ending with the synthesis of 3-dehydroshikimic acid (DHS); whereas E. coli BC was equipped with enzymes to assimilate and convert DHS into MA. To strengthen the penetration of the DHS into E. coli BC, ShiA permease, an endogenous E. coli membrane-bound transporter was overexpressed in strain BC under the control of a constitutive pyruvate decarboxylase promoter isolated from Zymomonas mobilis . Compared with the pure cultivation, co-cultivation can improve the production efficiency significantly. Finally, 2 g/L of MA with a yield of 0.1 g/g was produced in a batch bioreactor. This combination of pathway modularization and microbial co-cultivation shows strong potential for future metabolic engineering studies [ 78 ]. The bacterial consortium realized complex biosynthetic pathway engineering; however, the interaction within E. coli – E. coli is competition. Balancing the intermediate secretion and mixed sugars utilization could successfully overcome this limitation [ 34 ]. In this binary system, two E. coli strains were constructed individually to accommodate different pathway modules to reduce the metabolic stress in each strain. Effective regulation of the endogenous upstream pathway and expression of the challenging downstream heterologous enzymes were divided into two distinct cellular metabolic backgrounds, respectively. This E. coli – E. coli system also achieved simultaneous utilization of glucose and xylose (Fig. 4 ). Furthermore, a membrane-bound transporter was engineered to enhance the mass transfer of the pathway intermediate between the upstream and downstream strains. The microorganism consortium produced 4.7 g/L of MA with a yield of 0.35 g/g from glucose/xylose mixture, which is significantly higher than previous reports [ 34 ]. 2-Keto- l -gulonic acid Currently, the most representative case for chemicals production using microbial consortia is 2-keto- l -gulonic acid (2-KGA), which is the precursor of vitamin C (L-ascorbic acid), an essential nutrient to maintain normal physiological activities in mammals. 110,000 tons of vitamin C is produced annually through bio-processes [ 79 ]. Currently, 2-KGA is mainly produced through two-step fermentation process, in which sorbitol is converted to sorbose by Gluconobacter suboxydans first, and then 2-KGA is synthesized from sorbose by co-cultivating with B. megaterium and Ketogulonicigenium vulgare [ 80 , 81 ]. Recently, one step of 2-KGA production from d -sorbitol was developed (Fig. 4 ). In details, two sequential pathway modules were incorporated into G. oxydans and K. vulgare to achieve the conversion of D-sorbitol-to-sorbose and sorbose-to-2KGA, respectively, leading to a simplified one-step bioproduction process. G. oxydans was also metabolically engineered to reduce its competition against K. vulgare for sorbose. More importantly, the performance of this one-step process was comparable to the traditional two-step one with production and yield of 76.6 g/L and 89.7% within 36 h, respectively [ 82 ]. Not only limited to above-mentioned chemicals, co-cultivation systems are also applied for other bulk chemicals synthesis, such as succinic acid, butyric acid, etc. In construction of microbial consortia, the design of metabolic pathway is quite necessary, especially for the complex biosynthesis pathway to achieve labor division and reduce the metabolic stress. Engineering a membrane-bound transporter is also a rational way to enhance the mass transfer of the pivotal pathway intermediates between the upstream and downstream strains. In addition, reducing the competition interaction was also used in many co-cultivation systems, such as co-cultures of E. coli strains using different carbon source. Higher value-added chemicals production using co-cultivation systems Natural products (NPs) are important sources for some novel bioactive compounds, such as drugs and other higher value-added compounds [ 13 ]. Typically, NPs can be extracted from plants and animals, but the low yield hinders their application. In addition, some bacteria and fungi are also important sources for NPs [ 83 ]. The most successful examples for NPs production using microbial consortia are taxol and flavonoids. Taxol is a well-known plant-derived terpenoids, because it is a chemotherapy medication used to treat various types of cancer [ 84 – 86 ]. The extracted yields of taxoids from the bark of the pacific yew tree ( Taxus brevifolia Nutt.) were extremely low and limited. The biosynthesis of taxol involves at least 19 enzymatic steps starting from the universal diterpenoid precursor, geranylgeranyl diphosphate. The long and complicated pathway using pure culture would burden the metabolic capability, resulting in the production levels only maintained at μg/L levels [ 87 , 88 ]. Division of synthetic pathway into different strains, such as bacterium–yeast strains would significantly improve the production level [ 37 ]. For example, E. coli can be engineered to use xylose as the substrate and overproduce taxadiene, which was the scaffold molecule of paclitaxel; S. cerevisiae was then engineered to express cytochrome P450s (CYPs) owing to its advanced protein expression machinery and abundant intracellular membranes, which functionalized taxadiene by catalyzing multiple oxygenation reactions. As known, S. cerevisiae is deficient in xylose utilization; hence, when xylose was used as the carbon source, E. coli would metabolize xylose to produce acetate and taxadiene first, and then acetate was used as the carbon source for S. cerevisiae growth. Accompanied with the consumption of acetate by S. cerevisiae , taxadiene could be further converted into taxol. The strategy of labor division in this system led to 33 mg/L of oxygenated taxanes including a monoacetylated dioxygenated taxane [ 35 ]. This success system shows an important advantage for designing the expression system and pathway in different strains, as they can be constructed and optimized in parallel to significantly improve the product titer. Furthermore, the system could combine dual properties of rapid production of taxadiene in E. coli with efficient oxygenation of taxadiene by S. cerevisiae . Another typical example for NP production using microbial consortia is flavonoids, which also shows promising potential for pharmaceutical application [ 89 ]. The biosynthetic pathway from phenylpropanoic acids to flavan-3-ols was divided into the malonyl-CoA-dependent upstream module (phenylpropanoic acids to flavanones) and the NADPH-dependent downstream module (flavanones to flavan-3-ols). However, when this complicated pathway was expressed in pure cultures, flavan-3-ols titers from phenylpropanoic acids were very low. Chemler et al. [ 90 ] engineered E. coli binary system, which not only reduced the overwhelming metabolic burden, but also enabled to individually optimize the intermediate supply and co-factor provision in separate strains. After systematical process optimization, including carbon source, temperature, induction point, and inoculation ratio, 40.7 mg/L of flavan-3-ols was achieved with 970-fold flavonoids production improvement over the pure culture approach [ 39 ]. Except increasing the yields of previously described metabolites, microbial consortia can also induce new biosynthetic routes to bioactive metabolites [ 8 , 91 ]. For example, new diorcinol J(1) was produced from a marine isolate of the fungi Aspergillus sulphureus KMM 4640 and Isaria felina KMM 4639 [ 92 ]. New lipoaminopeptides could be biosynthesized from two different fungi, Mycogone rosea and Acremonium sp.; however, the new derivatives were not detected in pure cultures of either fungus, suggesting that chimeric pathways resulting from co-culture can also lead to new natural products. Challenges and further perspectives for co-cultivation systems Although many advantages existed for co-cultivation systems, advances and development of this emerging approach are still needed to address two critical challenges. One is how to maintain the stable co-existence of the constituent strains in the co-culture systems; the other is how to parallelly maintain the fermentation conditions, such as pH, temperature and oxygen supply. Different from natural microbial consortia existing for survival, the artificial co-cultures are constructed to optimize the production of target products. As such, the growth of involved co-culture members may be not compatible, often resulting in the competition for growth resources. In addition, the growth rates of microbial strains, especially different species vary to a large extent. As a result, co-cultivation of these species under a uniform growth condition can easily lead to the outgrowth of one specie over the other. Under such condition, adoption of microbial strains derived from the same species may be a better option. However, the general applicability of the same species is limited, as many biosynthesis processes require mixed biosynthesis capabilities from two or more different microbial species. Another alternative strategy is to engineer the co-culture members to grow and utilize separated carbon sources, reducing the growth competition and improving the growth compatibility. On the other hand, cooperative behavior must be robust to variations of environment, offering important insight for modular co-culture engineering design [ 93 – 95 ]. The design principles for microbial consortia are based on the interaction among microbial members, including cell–cell interaction, exchange of metabolites, etc. So far, most studies about microbial consortia mainly focused on the exchange of intermediate metabolites. However, due to the unknown genetic background of many wild-type species and uncharacterized microbial interaction mechanisms, the energy conversion efficiency of these microbial consortia was difficult to optimize, which greatly restricted their practical applications (Fig. 3 ). Except energy conversion, cell–cell interaction should also be emphasized [ 96 ]. For a desirable co-culturing system, positive interactions between two microorganisms are expected. The interactions between microorganisms in mixed culture environments may not always lead to desirable consequences. Hence, understanding the interactions between associated strains in artificial microbial consortia becomes more important. Synthetic biology tools, such as quorum sensing are being developed to manipulate the cell–cell interaction through signaling mechanisms, which shows great potential for growth and metabolic pathway coordination between the co-culture members in the future [ 97 , 98 ]. In addition, building cross-feeding interactions within the microbial consortia is also an advantageous approach to connect cells and distribute metabolic functions [ 99 ]. Based on the understanding of the interaction among microbes, the robustness, stability and reproducibility could be further improved [ 100 ]. In addition, rationally designing parental strains through utilization of a combinatorial metabolic engineering approach for optimizing cellular phenotype would become future trends [ 101 , 102 ]. Compartmentalization can effectively reduce the burden of fermentative strains, and microbial consortia could support plug-and-play biosynthesis of various target products. The co-culturing members can be engineered to specifically satisfy the need of the accommodated pathway modules, rather than the entire pathway. Also, the co-cultures can be easily programmed for new target biosynthetic pathways by re-organization or addition of the involved pathway modules/strains that have been pre-optimized for a specific part of the biosynthesis. A variety of products can be produced from the same upstream module by simply swapping the downstream modules. This intrinsic advantage of implementing modular design is well in line with the concept of modularity in synthetic biology and holds the potential of extensive applications in metabolic engineering."
} | 8,627 |
21977417 | PMC3148050 | pmc | 175 | {
"abstract": "The emerging field of biomimetics allows one to mimic biology or nature to develop nanomaterials, nanodevices, and processes which provide desirable properties. Hierarchical structures with dimensions of features ranging from the macroscale to the nanoscale are extremely common in nature and possess properties of interest. There are a large number of objects including bacteria, plants, land and aquatic animals, and seashells with properties of commercial interest. Certain plant leaves, such as lotus ( Nelumbo nucifera ) leaves, are known to be superhydrophobic and self-cleaning due to the hierarchical surface roughness and presence of a wax layer. In addition to a self-cleaning effect, these surfaces with a high contact angle and low contact angle hysteresis also exhibit low adhesion and drag reduction for fluid flow. An aquatic animal, such as a shark, is another model from nature for the reduction of drag in fluid flow. The artificial surfaces inspired from the shark skin and lotus leaf have been created, and in this article the influence of structure on drag reduction efficiency is reviewed. Biomimetic-inspired oleophobic surfaces can be used to prevent contamination of the underwater parts of ships by biological and organic contaminants, including oil. The article also reviews the wetting behavior of oil droplets on various superoleophobic surfaces created in the lab.",
"conclusion": "Conclusion Biomimetics allows one to mimic biology or nature and for engineers to develop materials and devices of commercial interest. Properties of biological materials and surfaces result from a complex interplay between surface morphology and physical and chemical properties. Hierarchical structures with dimensions of features ranging from the macroscale to the nanoscale are extremely common in nature and possess properties of interest. There are a large number of objects including bacteria, plants, land and aquatic animals and seashells, with properties of commercial interest. One focus of this article is on biomimetics inspired structured surfaces for low fluid drag. One of the models from nature is the lotus leaf with a surface covered with wax and with hierarchical structure which provides superhydrophobicity, self cleaning, and low adhesion. An aquatic animal, such as a shark, is another model from nature. Shark skin is covered by very small individual tooth-like scales called dermal denticles (little skin teeth), ribbed with longitudinal grooves (aligned parallel to the local flow direction of the water). These grooved scales reduce vortices formation present on a smooth surface, resulting in water moving efficiently over their surface. The artificial surfaces inspired by the shark skin and the lotus leaf have been created and the influence of structure has been reviewed by measurement of pressure drop and fluid drag for drag reduction efficiency. Oleophobic surfaces have the potential for self-cleaning and anti-fouling from biological and organic contaminants both in air and underwater applications. A model for predicting the contact angle of water and oil droplets has been reviewed. The surface tension of oil and organic liquids is lower than that of water, so to make the surface oleophobic in a solid–air–oil interface, a material with surface energy lower than that of oil should be used. The wetting behavior of water and oil droplets for hydrophobic/philic and oleophobic/philic surfaces in three phase interfaces is reviewed. For underwater applications, we have reviewed oleophobicity/philicity of an oil droplet in water on surfaces with different surface energies of various interfaces and contact angles of water and oil droplets in air. This article provides a useful guide for the development of biomimetic artificial surfaces with either low drag or self-cleaning/anti-fouling properties.",
"introduction": "Introduction Biologically inspired design, adaptation, or derivation from nature is referred to as ‘biomimetics.’ It means mimicking biology or nature. Nature has gone through evolution over the 3.8 billion years since life is estimated to have appeared on the Earth [ 1 ]. Nature has evolved objects with high performance using commonly found materials. These function on the macroscale to the nanoscale. The understanding of the functions provided by objects and processes found in nature can guide us to imitate and produce nanomaterials, nanodevices, and processes [ 2 ]. There are a large number of objects (bacteria, plants, land and aquatic animals, seashells etc.) with properties of commercial interest. Natural superhydrophobic, self-cleaning, low adhesion, and drag reduction surfaces Drag reduction in fluid flow is of interest in various commercial applications. These include transportation vehicles and micro/nanofluidics based biosensor applications [ 3 ]. To reduce pressure drop and volume loss in micro/nanochannels used in micro/nanofluidics, it is desirable to minimize the drag force at the solid–liquid interface. A model surface for superhydrophobicity, self-cleaning and low adhesion is the leaves of water-repellent plants such as Nelumbo nucifera (lotus) [ 2 , 4 – 11 ]. The leaf surface is very rough due to so-called papillose epidermal cells, which form papillae or microasperities. In addition to the microscale roughness, the surface of the papillae is also rough, with nanoscale asperities composed of three-dimensional epicuticular waxes which are long chain hydrocarbons and hydrophobic. The waxes on lotus leaves exist as tubules [ 10 – 11 ]. Water droplets on these hierarchical structured surfaces readily sit on the apex of the nanostructures because air bubbles fill the valleys of the structure under the droplet ( Figure 1a ). Therefore, these leaves exhibit considerable superhydrophobicity. Static contact angle and contact angle hysteresis of a lotus leaf are about 164° and 3°, respectively [ 12 – 13 ]. The water droplets on the leaves remove any contaminant particles from their surfaces when they roll off, leading to self-cleaning [ 5 ] and show low adhesive force [ 14 – 16 ]. Figure 1 Two examples from nature: (a) Lotus effect [ 12 ], and (b) scale structure of shark reducing drag [ 21 ]. Natural superoleophobic, self-cleaning, and drag reduction surfaces A model surface for superoleophobicity and self-cleaning is provided by fish which are known to be well protected from contamination by oil pollution although they are wetted by water [ 15 , 17 ]. Fish scales have a hierarchical structure consisting of sector-like scales with diameters of 4–5 mm covered by papillae 100–300 μm in length and 30–40 µm in width [ 18 ]. Shark skin, which is a model from nature for a low drag surface, is covered by very small individual tooth-like scales called dermal denticles (little skin teeth), ribbed with longitudinal grooves (aligned parallel to the local flow direction of the water) ( Figure 1b ). These grooved scales reduce vortice formation present on a smooth surface, resulting in water moving efficiently over their surface [ 2 , 19 – 22 ]. The water surrounding these complex structures can lead to protection from marine fouling and play a role in defense against adhesion and growth of marine organisms, e.g., bacteria and algae [ 11 , 23 ]. If oil is present on the surfaces in air or water, surfaces are known to be oleophobic and may exhibit self-cleaning and anti-fouling properties. Many sea animals including fish and shark are known to be oleophobic under water. Superoleophobic surfaces can also reduce significant losses of residual fuel in fuel tanks and pipes [ 15 , 24 ]. Roughness-induced superhydrophobicity, self-cleaning, low adhesion, and drag reduction Jung and Bhushan [ 21 ] created artificial surfaces inspired by the lotus leaf and shark skin and studied the influence of structure on pressure drop and fluid drag. One of the basic properties of interest in fluid flow is slip. The relative velocity between a solid wall and liquid is believed to be zero at the solid–liquid interface, which is the so called no-slip boundary condition ( Figure 2 , left) [ 25 – 26 ]. However, for hydrophobic surfaces, fluid film exhibits a phenomenon known as slip, which means that the fluid velocity near the solid surface is not equal to the velocity of the solid surface ( Figure 2 , right) [ 27 – 33 ]. The degree of boundary slip at the solid–liquid interface is characterized by a slip length. The slip length b is defined as the length of the vertical intercept along the axis orthogonal to the interface when a tangent line is drawn along the velocity profile at the interface ( Figure 2 , right). Recent experiments with surface force apparatus (SFA) [ 34 – 36 ], atomic force microscopy (AFM) [ 32 – 33 37 ], and particle image velocimetry (PIV) [ 38 ] techniques have reported slip lengths on hydrophobic surfaces: No slip was observed on hydrophilic surfaces [ 34 , 36 – 40 ]. Theoretical studies [ 41 – 44 ] and experimental studies [ 33 , 45 – 47 ] suggest that the presence of nanobubbles at the solid-liquid interface is responsible for boundary slip on hydrophobic surfaces. Figure 2 Schematic of velocity profiles of fluid flow without and with boundary slip. The definition of slip length b characterizes the degree of boundary slip at the solid–liquid interface. The arrows represent directions of fluid flow. Roughness-induced superoleophobicity The surface tension of oil and organic liquids is lower than that of water, so to create a superoleophobic surface, the surface energy of the solid surface in air should be lower than that of oil. For underwater applications, if an oil droplet is placed on a solid surface in water, the solid–water–oil interface exists. The nature of oleophobicity/philicity of an oil droplet in water can be determined from the values of surface energies of various interfaces and contact angles of water and oil in air. Many superoleophobic surfaces have been developed by modifying the surface chemistry with a coating of extreme low surface energy materials [ 20 , 48 – 54 ]. Tuteja et al. [ 54 ] showed that surface curvature, in conjunction with chemical composition and roughened texture, can be used for liquids with low surface tension, including alkanes such as decane and octane. Liu et al. [ 18 ] performed experiments in a solid-water-oil interface. They found that hydrophilic and oleophilic surfaces (solid–air–water interface and solid–air–oil interface) can switch to an oleophobic surface in water (solid–water–oil interface). As a result, oil contaminants are washed away when immersed in water. This effect can be employed for underwater oleophobicity and self-cleaning that can be used against marine ship fouling [ 17 ]. Jung and Bhushan [ 20 ] proposed a model for predicting the oleophobic/philic nature of surfaces and showed how the water and oil droplets in three phase interfaces influence the wetting behavior on micropatterned surfaces with varying pitch values as well as the shark skin replica as an example from an aquatic animal. Article objective This article reviews drag data on artificial surfaces inspired from shark skin and lotus leaf. Oleophobic and self-cleaning surfaces inspired from aquatic animals are then discussed."
} | 2,805 |
32265641 | PMC7105894 | pmc | 176 | {
"abstract": "Among the recent innovative technologies, memristor (memory-resistor) has attracted researchers attention as a fundamental computation element. It has been experimentally shown that memristive elements can emulate synaptic dynamics and are even capable of supporting spike timing dependent plasticity (STDP), an important adaptation rule that is gaining particular interest because of its simplicity and biological plausibility. The overall goal of this work is to provide a novel (theoretical) analog computing platform based on memristor devices and recurrent neural networks that exploits the memristor device physics to implement two variations of the backpropagation algorithm: recurrent backpropagation and equilibrium propagation. In the first learning technique, the use of memristor–based synaptic weights permits to propagate the error signals in the network by means of the nonlinear dynamics via an analog side network. This makes the processing non-digital and different from the current procedures. However, the necessity of a side analog network for the propagation of error derivatives makes this technique still highly biologically implausible. In order to solve this limitation, it is therefore proposed an alternative solution to the use of a side network by introducing a learning technique used for energy-based models: equilibrium propagation. Experimental results show that both approaches significantly outperform conventional architectures used for pattern reconstruction. Furthermore, due to the high suitability for VLSI implementation of the equilibrium propagation learning rule, additional results on the classification of the MNIST dataset are here reported.",
"conclusion": "Conclusions In this paper, the dynamics of memristor–based recurrent neural networks has been analyzed. The network is trained by using two different generalizations of the backpropagation algorithm adapted to the continuous domain and energy-based models. Such in situ training learning rules permit to the memristor–based neural network to continuously adapt and adjust the synaptic weights without the direct computation of the loss function's gradient. Although, further work is still necessary to find physical memristor devices/emulators approximating the proposed memristive synapse dynamics, this manuscript provides two learning rules for the weights' update that can be implemented by a series of discrete programming pulses. Simulated results make clear that both methods significantly outperform conventional approach used for pattern reconstruction. In addition, promising results are also obtained by using equilibrium propagation in performing classification tasks.",
"introduction": "Introduction In the last few decades, the search of innovative computing platforms that could offer new, ultra-low power processing methods and architectures has intensified. Neuromorphic computing approaches aim to go beyond the state-of-the-art in conventional digital processing by exploiting complex dynamics and nonlinear phenomena emerging from the physics of nonvolatile memory devices (e.g., memristors) (Chua, 1971 ; Strukov et al., 2008 ). The hallmark of this kind of devices is the peculiar analog signal storing capability that allows them to mimic the behavior of neural synapses. The processing is not only analog and different from current digital processors, but also enhances computing speed and power efficiency for large sets of sensor data. This has been achieved by combining memristor technology with advanced deep learning algorithms used to train neural networks. In supervised learning, one of the most popular method used for training feedforward neural networks is the backpropagation algorithm. Although it is considered a powerful technique, it is computationally expensive and is commonly labeled as biologically implausible. The generalization of this rule to continuous-time recurrent networks was first introduced by Almeida ( 1987 ) and Pineda ( 1988 ) who independently obtained the same results. Recurrent backpropagation aims to iteratively adjust the weight matrix of the network in order to let the system converge, for fixed input and initial state, to a desired attractor. As for feedforward neural networks, this is achieved by minimizing a particular loss function associated to the system parameters with the difference that the error signal is now backpropagated by introducing an associated differential equation. This allowed to avoid the direct gradient's computations and reduced the large number of required multiplications. However, the necessity of a side network for the propagation of error derivatives makes this technique still highly different from emulating the brain complex computation. This hypothesis is further supported by the fact that there is no known mechanism that could explain how an error message is propagated backwards through the same pathway of the incoming signal. Recently, Scellier and Bengio ( 2017 ) proposed an alternative solution to the use of a side network by introducing Equilibrium Propagation, a learning technique used for energy-based models. The advantage of this approach is indeed the requirement of just one kind of neural computation for the training phase of the network. Firstly, inputs are clamped and the network relaxes to a fixed point which corresponds to a local minimum of the energy function. Secondly, after introducing a small external error signal, the network relaxes to a new but close-by fixed point which now corresponds to a rather lower cost value. Even though the two methods seem quite different, it is easy to observe that both share the same goal, finding low-energy configurations that have low cost values. The aim of this work is to propose a novel (theoretical) analog computing platform based on memristor devices and recurrent neural networks that exploits the memristor device physics to implement two variations of the backpropagation algorithm. In the first section, it is provided a brief introduction on memristors and their peculiar properties useful for the physical implementation. In the second section, a general introduction on biological algorithms is presented with particular attention on recurrent backpropagation and equilibrium propagation. In the last section, the two techniques are compared with the existing algorithms used in pattern reconstruction providing results of their compelling efficiency. Lastly, it is shown the application of a memristor-based recurrent neural network trained with equilibrium propagation used for the classification of a small subset of the MNIST dataset. The choice of using only this learning rule was mainly dictated by the fact that using a side network, as in the recurrent backpropagation approach, would at least double the required IC area."
} | 1,699 |
26840035 | PMC4804745 | pmc | 177 | {
"abstract": "Abstract Scleractinian corals are assumed to be stenohaline osmoconformers, although they are frequently subjected to variations in seawater salinity due to precipitation, freshwater run‐off and other processes. Observed responses to altered salinity levels include differences in photosynthetic performance, respiration and increased bleaching and mortality of the coral host and its algal symbiont, but a study looking at bacterial community changes is lacking. Here, we exposed the coral Fungia granulosa to strongly increased salinity levels in short‐ and long‐term experiments to disentangle temporal and compartment effects of the coral holobiont (i.e. coral host, symbiotic algae and associated bacteria). Our results show a significant reduction in calcification and photosynthesis, but a stable microbiome after short‐term exposure to high‐salinity levels. By comparison, long‐term exposure yielded unchanged photosynthesis levels and visually healthy coral colonies indicating long‐term acclimation to high‐salinity levels that were accompanied by a major coral microbiome restructuring. Importantly, a bacterium in the family Rhodobacteraceae was succeeded by Pseudomonas veronii as the numerically most abundant taxon. Further, taxonomy‐based functional profiling indicates a shift in the bacterial community towards increased osmolyte production, sulphur oxidation and nitrogen fixation. Our study highlights that bacterial community composition in corals can change within days to weeks under altered environmental conditions, where shifts in the microbiome may enable adjustment of the coral to a more advantageous holobiont composition.",
"introduction": "Introduction Coral reefs are among the most diverse and productive ecosystems on the planet (Reaka‐Kudla et al . 1996 ) and provide a wide range of goods and services to approximately 500 million people in more than 100 countries (Wilkinson 2008 ). Coral reef ecosystems rely on the three‐dimensional carbon skeleton framework built by scleractinian corals. Scleractinian corals are metaorganisms—so‐called coral holobionts—that are composed of the coral host, its dinoflagellate endosymbionts (genus Symbiodinium ) and a diverse microbial assemblage consisting of fungi, bacteria, archaea and viruses (microbiome) (Rohwer et al . 2002 ). The bacterial microbiome has been shown to play important roles in coral health (Rosenberg et al . 2007 ), immunity (Ritchie 2011 ), as well as carbon, sulphur and nitrogen cycling (Rohwer et al . 2002 ; Raina et al . 2009 ; Lema et al . 2012 ). Moreover, it has been suggested that microbial assemblages facilitate the acclimation of coral holobionts to new environmental conditions (the probiotic hypothesis, Reshef et al . 2006 ). However, knowledge about the specific role of the vast majority of microbes, and in particular bacteria, to holobiont function is still limited (Lesser et al . 2004 ; Barott et al . 2011 ; Morrow et al . 2012 ; Bourne & Webster 2013 ). Although frequently exposed to changes in seawater salinity, for example due to precipitation, freshwater run‐off, periods of prolonged drought or desalination processes (Chartrand et al . 2009 ; Lirman & Manzello 2009 ; Roberts et al . 2010 ; Edge et al . 2013 ; Hédouin et al . 2015 ), scleractinian corals are assumed to be stenohaline osmoconformers with a limited ability to adjust to salinity fluctuations (Hoegh‐Guldberg & Smith 1989 ; Ferrier‐Pages et al . 1999 ; Kerswell & Jones 2003 ; True 2012 ; Hédouin et al . 2015 ). Nevertheless, studies show that some coral species are able to tolerate greater salinity fluctuations than others. For instance, Stylophora pistillata showed decreased respiration and photosynthetic rates at minor salinity decreases; whereas Siderastrea radians displayed a high resilience towards salinity changes (Ferrier‐Pages et al . 1999 ; Lirman & Manzello 2009 ). Further, Coles ( 2003 ) reported high‐salinity tolerance of corals in the Arabian Gulf and Red Sea, with some species surviving salinities of 48–50 practical salinity unit (PSU). While higher animals possess excretory systems to adjust for changes in salinity, most marine invertebrates, including scleractinian corals, are considered osmoconformers (Yancey et al . 2002 ; Evans 2008 ). Yet, in sea anemones changes in free amino acid pools are involved in osmoregulation (Shick 1991 ). Similarly, marine microorganisms (i.e. algae and bacteria) accumulate diverse molecules that serve as osmolytes upon increased salinity exposure (Csonka & Hanson 1991 ; Mayfield & Gates 2007 ). However, osmoregulation in the coral‐algal endosymbiosis represents a challenging scenario. The coral animal host has to equilibrate the external osmotic pressure with its intracellular environment, which is determined by its own metabolism and that of its algal symbionts (Mayfield & Gates 2007 ). In this context, the role of the coral‐associated bacteria is virtually unknown. To elucidate the role of coral‐associated bacteria to salinity changes, we exposed the coral Fungia granulosa to strongly increased salinities resulting from seawater reverse osmosis (SWRO) desalination concentrate. By collecting data from all holobiont compartments (i.e. coral host, symbiont algae, bacterial microbiome) using a combination of short‐ (4 h) and long‐term (29 days) experimental treatments, we aimed to disentangle compartment‐specific responses and to assess potential adaptation/acclimation processes by characterizing the initial response and long‐term effects on the coral holobiont.",
"discussion": "Discussion In this study, we assessed the effect of short‐ and long‐term hypersalinity exposure on the Red Sea coral F. granulosa . Our results indicate distinct short‐ and long‐term reactions of the coral holobiont. The short‐term experiment, aimed to measure the initial response, was characterized by an absence of changes in the bacterial community structure, but a significant reduced calcification and photosynthesis under strongly increased salinity levels. In contrast, the long‐term transect experiment indicated a putative acclimation response, since corals exposed to high salinity for 29 days did not exhibit measureable (photo)physiological effects or signs of bleaching (van der Merwe et al . 2014b ), but rather, displayed a significant shift in the associated bacterial community. Short‐term and long‐term coral physiology \n Fungia granulosa is a single‐polyp scleractinian coral that has been demonstrated to possess a slow growth rate (Chadwick‐Furman et al . 2000 ). Accordingly, our values for the calcification rate ( G ) were lower than those reported for other corals, for example for Stylophora subseriata (1.05–1.73 μmol CaCO 3 /cm 2 /h) (Sawall et al . 2011 ). It is of note that calcification effectively stopped at high salinities in the short‐term treatment. Unfortunately, calcification rates are not available for the long‐term treatment. However, it would be interesting to see whether calcification rates are not influenced in long‐term hypersaline exposure—as observed for photosynthetic efficiency. In line with previous studies, our short‐term incubation showed an overall oxygen increase, which was significantly lower for high‐salinity conditions (Gattuso et al . 1999 ; Manzello & Lirman 2003 ; Lirman & Manzello 2009 ). Further, a reduction of photosynthetic rates has been documented for hyper‐ and hyposaline scenarios before (Muthiga & Szmant 1987 ; Moberg et al . 1997 ; Ferrier‐Pages et al . 1999 ; Alutoin et al . 2001 ; Chartrand et al . 2009 ). However, long‐term ϕPSII values from our coral samples were in the same range as photosynthetic yields originating from corals in their natural environment. We could not observe any differences between hypersaline and ambient conditions in regard to the corals' photophysiology or bleaching status (van der Merwe et al . 2014b ). In accordance with data collected from our short‐ and long‐term experiments, an initial sharp decline in photosynthetic performance with subsequent recovery has been suggested as an acclimation pattern (Manzello & Lirman 2003 ; Lirman & Manzello 2009 ). Mayfield & Gates ( 2007 ) interpreted these patterns as an indication for osmoregulatory processes, also considering corals to generally tolerate slow salinity changes better than more rapid ones (Muthiga & Szmant 1987 ). Our physiological data are supporting corals as being able to adjust to salinity. We report distinct physiological effects on coral host (calcification) and Symbiodinium (decreased photosynthesis) in our 4‐h incubation. In the long‐term in situ experiment, we could not find any influence on the (photo)physiology of F. granulosa indicating acclimation of the coral holobiont to prevailing salinity levels. Coral microbiome restructuring To our knowledge, this is the first study that assessed coral bacterial microbiome structure under short‐ and long‐term exposure of corals to salinity changes. Healthy corals maintain mostly specific, stable and uneven microbial assemblages indicating selected microbiomes (Bourne et al . 2008 ; Meron et al . 2011a , 2012 ; Bayer et al . 2013 ; Bourne & Webster 2013 ; Jessen et al . 2013 ; Kelly et al . 2014 ). The microbial communities are diverse and contribute to pathogen inhibition due to production of antimicrobial substances as well as competition for space and nutrients (Klaus et al . 2007 ; Rosenberg et al . 2007 ; Thurber et al . 2009 ). In our experiments, we found no distinct bacterial community changes after 4‐h salinity exposure, which contrasts the measured physiological reactions of coral host and algal symbiont. At the same time, Apprill et al . ( 2009 ) measured doubling times of 10+ hours for coral‐associated bacteria, which may have affected our ability to determine a bacterial microbiome response in the short‐term experiment. Conversely, we found no apparent physiological reaction, but pronounced microbial community changes after a 29‐day hypersalinity treatment. All corals, except those from the long‐term hypersalinity treatment, revealed highly uneven bacterial microbiomes that were numerically dominated by a single, distinct OTU (i.e. OTU0001) that could be identified to the level family, namely Rhodobacteraceae . Bacteria from this family were repeatedly observed in healthy corals (Sunagawa et al . 2009 ; Ceh et al . 2012 ; Morrow et al . 2012 ; Bayer et al . 2013 ; Kellogg et al . 2014 ; Li et al . 2014 ), even though they have also been found to be associated with stressed corals and stressed sea urchins (Buchan et al . 2005 ; Sunagawa et al . 2009 ; Meron et al . 2011a , b , 2012 ; Godwin et al . 2012 ; Pantos et al . 2015 ). Additionally, bacteria in the family Rhodobacteraceae have been found to be enriched in corals isolated from deeper habitats (27 m) compared to their shallow counterparts (6 m) (Pantos et al . 2015 ), which may explain their dominance in healthy F. granulosa collected in this study from a depth of 15–18 m. Taken together, the presence of Rhodobacteraceae in a range of hosts denotes environmental flexibility. For this reason, it is challenging to assign a specific role. However, the high abundance of a distinct OTU of this bacterial family in F. granulosa specimens from all treatments but the hypersalinity long‐term treatment indicates that this taxon probably provides an important function to the coral holobiont. In contrast, Pseudomonas veronii , the ‘core’ microbiome member and the most abundant taxon in corals from the hypersalinity long‐term treatment, was present at a much lower abundance in corals from all other treatments (i.e. freshly collected coral, short‐term ambient salinity, short‐term hypersalinity and long‐term ambient salinity). As P. veronii was present in all corals, albeit at lower abundance, we argue that its increase under high salinity might signify a change of selection regime for this taxon under the altered environmental conditions and not an opportunistic association. The uniformity of all water samples, that is no significant differences in water samples over different treatments or time points, further supports a selective process for the changes in the coral microbiomes. The specific function of P. veronii remains to be determined. However, it seems to be a versatile taxon that has been isolated from distinct environments, for example natural freshwater springs, soil samples and wastewater filters where it has been shown to degrade a variety of simple aromatic organic compounds making it a beneficial bacterium for bioremediation of contaminated environments (Elomari et al . 1996 ; Nam et al . 2003 ; Onaca et al . 2007 ). More generally, bacteria in the genus Pseudomonas have repeatedly been shown to be abundant in hypersaline environments and display broad metabolic capacity (Fendrich 1988 ; Brusa et al . 2001 ; Sass et al . 2001 ; Isnansetyo & Kamei 2009 ). Among other bacterial taxa that increased in abundance in the long‐term hypersalinity treatment, we identified the coral pathogen Vibrio shilonii (OTU0264) and also some unclassified Alteromonadaceae taxa (Table S3, Supporting information). These taxa are presumably associated with coral stress and disease, but are known to reside in healthy corals as well (Rosenberg & Falkovitz 2004 ; Sunagawa et al . 2009 ). Taken together, bacterial microbiome restructuring under highsalinity levels as signified by loss of the numerically dominant bacterial taxon (i.e. OTU0001), the increase in P. veronii (i.e. OTU0010), as well as an overall increase in richness, evenness and diversity possibly indicates stress (Bourne et al . 2008 ; Garren et al . 2009 ; Sunagawa et al . 2009 ; Meron et al . 2011a , b ; Zhang et al . 2015 ). At the same time, major microbiome restructuring induced by environmental stress (i.e. high salinity) in the absence of a measurable physiological reaction of the coral holobiont may give support to the probiotic hypothesis (Reshef et al . 2006 ), that is a change of the microbiome to facilitate coral holobiont acclimation. Functional changes of bacterial communities indicate metabolic adjustment Mapping of differences in bacterial community composition to putative functional differences revealed a prominent increase in PHB storage as well as changes in nitrogen and sulphur cycling in long‐term hypersalinity samples in comparison with all other coral samples. PHB can be synthesized by microorganisms as a carbon reservoir in cells (Roberts 2005 ) and may be produced in response to various stressors, such as nutrient limitation, for example under nitrogen‐limiting conditions (Ayub et al . 2004 ; Soto et al . 2012 ). Interestingly, PHB has also been identified as an osmolyte in microorganisms (Doronina et al . 2000 ; Martin et al . 2002 ; Arora et al . 2006 ; Soto et al . 2012 ). Additionally, PHB production in Rhizobia with a potential benefit for plant cultivation in saline soil has been suggested (Arora et al . 2006 ; Ali et al . 2014 ). It is striking that Pseudomonas strains closely related to P. veronii are shown to produce PHB (Ayub et al . 2004 ; Yan et al . 2008 ; Soto et al . 2012 ), but even more so, the genome of P. veronii harbours the enzyme 3‐hydroxyisobutyrate dehydrogenase (Ramírez‐Bahena et al . 2015 ), which is part of the PHB metabolism (Hügler & Sievert 2011 ). This provides a putative functional link to the numerical dominance of P. veronii in the long‐term hypersalinity samples and potentially indicates functional adaptation/acclimation of the coral holobiont by alteration of its microbiome. Such functional changes were shown in the aphid Acyrthosiphon pisum where replacing the native gut bacteria Buchnera line LSR1 with line 5AY from a more thermotolerant aphid matriline conferred a dramatic increase in thermal tolerance (Moran & Yun 2015 ). Changes in sulphur cycling as suggested by an upregulation of ‘sulphur oxidizer’ and a downregulation of ‘sulphate reducer’ presumably indicate the enrichment of oxidized products in the sulphur metabolism. The coral holobiont is a major contributor to the production of dimethylsulphide (DMS), a central compound of the global sulphur cycle (Raina et al . 2013 ), which can become oxidized to dimethylsulphoxide (DMSO) (Sunda et al . 2002 ). DMSO has a stronger reactivity towards reactive oxygen species (ROS) than DMS, is more hydrophilic (allowing higher cellular concentrations) and can be further oxidized to the water‐soluble antioxidant methane sulphinic acid (Sunda et al . 2002 ). Hence, an increased production (accompanied by a decreased reduction) of DMSO acting as an ROS scavenger may enable the coral to cope with increased oxidative stress in Symbiodinium . In agreement with these patterns, increased oxidative stress accompanied by antioxidant production as a response to high salinity has been shown for algae and other plants (Gossett et al . 1996 ; Fadzilla et al . 1997 ; Hernández et al . 2000 ; Jahnke & White 2003 ). Another distinct pattern emerged from the metabolic profile of nitrogen‐related functions. We found processes that increase nitrogen availability for the holobiont to be enhanced (i.e. ‘dinitrogen‐fixing’ and ‘nitrogen fixation’), whereas processes that require the availability of fixed nitrogen were reduced (i.e. ‘ammonia oxidizer’ and ‘nitrite reducer’). This suggests an enhanced nutrient limitation of the coral holobiont (Rädecker et al . 2015 ). Nutrient limitation may be a consequence of an increased metabolism with enhanced nutrient requirements. Long‐term coral holobiont response may indicate acclimation Changes of the coral microbiome under changed environmental conditions were previously described (e.g. Klaus et al . 2007 ; Meron et al . 2012 ; Jessen et al . 2013 ; Kelly et al . 2014 ; Pantos et al . 2015 ), and that these changes are relevant to holobiont function was demonstrated by Moran & Yun (Moran & Yun 2015 ). In line with these studies, we interpret the here‐observed prevalent change of the coral microbiome in combination with a lack of an apparent stress response by the coral or symbiont in the long‐term hypersalinity treatment as indication for an acclimation response. This is supported by the putative functional changes we detected in the microbial community, that is upregulation of PHB as an osmolyte, alterations to the nitrogen cycle to compensate for nutrient deficiency, and synthesis of DMSO as a ROS scavenger. It is important to consider that the adjustments of the Fungia granulosa coral holobiont to a high‐salinity environment presumably require considerable energy and these energy requirements need to be taken into account when assessing the response of corals to changes in salinity. Taking the large biomass of the solitary coral F. granulosa into account, energy reserves may be sufficient for supposedly initial stress periods (as a response to the changing environmental conditions) and simultaneous acclimation. By comparison, commonly employed setups using small coral fragments in short‐term experiments may considerably underestimate coral resilience towards (salt) stress and might miss acclimation due to insufficient energy reserves of the coral fragment to sustain and acclimate to the stressor (Ferrier‐Pages et al . 1999 ; Kerswell & Jones 2003 ; Manzello & Lirman 2003 ; Chartrand et al . 2009 ; Lirman & Manzello 2009 ; Seveso et al . 2013 ). Taken together, we argue that changes in salinities lead to changes in the holobionts internal environment, which in turn affect microbiome structure by selecting for a more advantageous bacterial community composition as posited by the coral probiotic hypothesis (Reshef et al . 2006 ). Future studies should target the temporal stability of restructured coral microbiomes accompanied by physiological measures under enduring ‘stress’ conditions to unequivocally confirm the importance of the microbiome to coral holobiont function."
} | 5,039 |
34248491 | PMC8267251 | pmc | 178 | {
"abstract": "Spiking Neuromorphic systems have been introduced as promising platforms for energy-efficient spiking neural network (SNNs) execution. SNNs incorporate neuronal and synaptic states in addition to the variant time scale into their computational model. Since each neuron in these networks is connected to many others, high bandwidth is required. Moreover, since the spike times are used to encode information in SNN, a precise communication latency is also needed, although SNN is tolerant to the spike delay variation in some limits when it is seen as a whole. The two-dimensional packet-switched network-on-chip was proposed as a solution to provide a scalable interconnect fabric in large-scale spike-based neural networks. The 3D-ICs have also attracted a lot of attention as a potential solution to resolve the interconnect bottleneck. Combining these two emerging technologies provides a new horizon for IC design to satisfy the high requirements of low power and small footprint in emerging AI applications. Moreover, although fault-tolerance is a natural feature of biological systems, integrating many computation and memory units into neuromorphic chips confronts the reliability issue, where a defective part can affect the overall system's performance. This paper presents the design and simulation of R-NASH-a reliable three-dimensional digital neuromorphic system geared explicitly toward the 3D-ICs biological brain's three-dimensional structure, where information in the network is represented by sparse patterns of spike timing and learning is based on the local spike-timing-dependent-plasticity rule. Our platform enables high integration density and small spike delay of spiking networks and features a scalable design. R-NASH is a design based on the Through-Silicon-Via technology, facilitating spiking neural network implementation on clustered neurons based on Network-on-Chip. We provide a memory interface with the host CPU, allowing for online training and inference of spiking neural networks. Moreover, R-NASH supports fault recovery with graceful performance degradation.",
"conclusion": "5. Conclusion In this work, we proposed and evaluated a reliable three-dimensional digital neuromorphic system geared explicitly toward the 3D-ICs biological brain's three-dimensional structure toward the design of a cross-paradigm system. Spike timing patterns represent the information in the network, and learning is based on the local spike-timing-dependent plasticity rule. The proposed platform enables high integration density and slight spike delay of spiking networks and features a scalable design. R-NASH is a design based on the Through-Silicon-Via technology, facilitating spiking neural network implementation on clustered neurons based on Network-on-Chip. We provide a memory interface with the host CPU, allowing for online training and inference of spiking neural networks. Moreover, R-NASH supports faults recovery using our fault-tolerant mapping method by optimizing communication and migration costs. We presented the functionality of the system by performing the MNIST dataset classification. Moreover, the R-NASH platform is also presented with the mapping method and fault-tolerance features. The mapping method shows that it can easily outperform manual mapping. On the other hand, we also proposed a genetic algorithm for fault recovery in SNN. Although the proposed work is expected to move neuromorphic computing toward a real-world scenario on large-scale systems, further optimization, such as bit-width reduction, low-power optimization, is needed. Future works for NASH would focus on real chip fabrication and deployment in a real-world application scenario, such as hand gesture recognition, prosthetic, and robotic arm control. Multiple-objective optimization for several costs such as communication, operating temperature is also considered in our future works. In STDP learning, keeping the weight dynamic is also an essential issue in our future work to balance the area cost and neuron/weight complexity.",
"introduction": "1. Introduction The brain-inspired computing paradigm takes inspiration from the biological brain to develop energy-efficient computing systems for future information processing capable of efficiently executing highly complicated tasks, such as decision-making and perception. Spiking neural networks (SNNs) attempt to mimic the information processing in the mammalian brain based on parallel arrays of neurons that communicate via spike events. Different from the typical multi-layer perceptron networks, where neurons fire at each propagation cycle, the neurons in SNN model fire only when a membrane potential reaches a specific value. In SNN, information is encoded using various encoding schemes, such as coincidence coding, rate coding, or temporal coding (Levin et al., 2014 ). SNN typically employs the integrate-and-fire neuron model in which a neuron generates voltage spikes (roughly 1 ms in duration per spike) that can travel down nerve fibers if they receive enough stimuli from other neurons with the presence of external stimuli. These pulses may vary in amplitude, shape, and duration, but they are generally treated as identical events. To better model the dynamics of the ion channel in a biological neuron, which is nonlinear and stochastic, the Hodgkin-Huxley (Goldwyn et al., 2011 ) conductance-based neuron is often used. However, the Hodgkin-Huxley model is too complicated to be used for a large-scale simulation or hardware implementation. Software simulation of SNN (Hazan et al., 2018 ; Stimberg et al., 2019 ) is a flexible method for investigating the behavior of neuronal systems. However, simulation of a large (deep) SNN system in software is slow and cannot fully exploit the overall system performance. An alternative approach is a hardware implementation, which provides the possibility to generate independent spikes accurately and simultaneously output spikes in real time. Hardware implementations of SNNs (neuromorphic) also have the advantage of computational speedup over software simulations and can take full advantage of their inherent parallelism. Specialized hardware architectures with multiple neuro-cores could exploit the parallelism inherent within neural networks to provide high processing speeds with low power, which make SNNs suitable for embedded neuromorphic devices and control applications (Vu et al., 2019 ). In general, the neuromorphic hardware systems consist of multiple nodes (or clusters of neurons) connected via an on-chip communication infrastructure (Akopyan et al., 2015 ; Ogbodo et al., 2020 ). Expansion using a multi-chip system and off-chip interconnects is also a viable solution for scaling up SNNs (Akopyan et al., 2015 ; Davies et al., 2018 ). In recent years, integrating many neurons on a single chip while providing efficient and accurate learning has been investigated (Schemmel et al., 2010 ; Benjamin et al., 2014 ; Furber et al., 2014 ; Akopyan et al., 2015 ; Davies et al., 2018 ). The challenges that need to be solved toward designing an efficient neuromorphic system include building a small-size, parallel, and reconfigurable architecture with low-power consumption, an efficient neuro-coding scheme, and an on-chip learning capability. Moreover, since the number of neurons to be connected is at least 10 3 times larger than the amount of PEs (Processing Elements) that need to be interconnected on modern multicore/multiprocessor SoC platforms (Furber, 2016 ), the on-chip communication and routing network is another major challenge. In a modern deep neural network (DNN) design, one neural network layer is often a 2D structure. However, the “mimicked” network is generally a 3D structure. Therefore, mapping a 3D structure onto 2D circuits may result in either multiple long wires between layers or congestion points (Vu et al., 2019 ; Dang et al., 2020b ; Ikechukwu et al., 2021 ). An event-driven neuromorphic system relies on the arrival of spikes (action potentials) to compute (Purves et al., 2008 ). Therefore, the arrival times of action potentials are critical to allow accurate and consistent outputs. Since the shared bus is no longer suitable for multicore systems and point-to-point interconnects cannot serve a high fanout wires (Lee et al., 2008 ), moving to a new on-chip communication paradigm with the ability to extend to multiple-chip interconnects is needed. One of the consensuses of state-of-the-art architecture is to utilize the parallelism and scalability of 2D Network-on-Chip (NoCs) (Akopyan et al., 2015 ; Davies et al., 2018 ) and further extend it to multichip systems. In this approach, the neurons of the silicon brain are clustered into nodes that are attached to micro-routers. From another hand, semiconductor development is confronting the end of Moore's Law, which no longer allows us to reduce the feature size as we reach the atomic scale. To get to the “More than Moore” goal (Waldrop, 2016 ), heterogeneous integration is a suitable approach to integrate more transistors in the same die. One of the popular approaches is to stack the conventional 2D wafers together to form a 3D-chip (Banerjee et al., 2001 ). Another method is monolithic 3D-ICs that support multiple silicon layers based on small vias (Panth et al., 2014 ). The Through-Silicon Vias (TSVs) or Monolithic Intertier Vias (MIVs) constitute one of the main interlayer communication mediums. The 3D-Network-on-Chip (3D-NoC) (Ben Ahmed and Ben Abdallah, 2013 ) is also a promising approach that can further enhance the parallelism and scalability of multicore and neuromorphic systems. Figure 1 illustrates a potential mapping of an emulated silicon brain into 3D-ICs. Here, the anatomical architecture of Spaun indicates large brain structures, and their connectivity is illustrated with thick dark-yellow lines for communication between elements of the cortex. In contrast, thin lines show connectivity between Basal Ganglia and the cortex (Eliasmith et al., 2012 ; Vu et al., 2019 ). However, despite bringing several benefits of lower power, smaller footprints, and low latency, the integration of a neuromorphic system into 3D-ICs was not well investigated. Figure 1 Neuro-inspired 3D silicon brain. (A) The anatomical architecture of Spaun indicates major brain structures and their connectivity. (B) The architecture of Spaun, where thick dark-yellow lines illustrate communication between elements of the cortex while thin lines show connectivity between Basal Ganglia and the cortex. (C) A possible mapping of a Spaun system into 3D-IC. This paper presents a reliable three-dimensional digital neuromorphic system, named R-NASH, geared explicitly toward the 3D-ICs biological brain's three-dimensional structure, where information in the network is represented by sparse patterns of spike timing and learning is based on the local spike timing-dependent plasticity rule. R-NASH is based on the Through-Silicon-Via technology, facilitating the spiking neural network implementation on clustered neurons based on Network-on-Chip. Morever, R-NASH features efficient spiking neuron mapping algorithms to map the neurons into suitable R-NASH clusters based on a genetic algorithm (GA). Furthermore, R-NASH supports faults recovery with graceful performance degradation. The rest of this paper is organized as follows: section 2 presents related works. section 3 describes the proposed R-NASH platform and presents the neuron mapping method based a graph-based algorithm. section 4 presents the evaluation results. Finally, section 5 concludes this paper.",
"discussion": "4.6. Discussion We have presented the design and the platform of a novel 3D neuromorphic system (R-NASH). Moreover, we demonstrated the scalability of the proposed approach and its ability to tolerate faults during run-time. In this section, we discuss the existing problems and potential solutions: First, the 3D-ICs are expected to have higher operating temperatures than the 2D-ICs due to the silicon layers' organization. In addition, 3D-ICs suffer from thermal dissipation. Large-scale neuromorphic systems also introduce high-density power consumption, which leads to high temperature. Therefore, optimizing the operating temperature is one of the critical issues. In Arka et al. ( 2021 ), the authors presented a MOO approach to optimize the operating temperature for 3D multi-core systems. As an optimization method, the Genetic Algorithm can surely solve multiple objective optimization (MOO) problems (i.e., using NSGA-II Deb et al., 2002 ), however, as it is a complex issue, further investigation is needed. Second, the scalability of the address range in R-NASH is currently limited to 3-bit per address. To have a better range, having an extra bit to represent the address is necessary. For instance, by adopting a 64-bit format for 3D NoC flit, the extra 32-bit can be distributed into address, neural mask, or AER fields. In TrueNorth (Akopyan et al., 2015 ), the flit consists of the offset between the source and the destination address; therefore, there is virtually no limitation on the address range. However, this way of addressing has two drawbacks: (1) the memory access is no longer globally accessible as one node can only reach a specific range, and (2) there are limited options for long-range synapses. By having a limited access range of memories, downloading weight for inference need a different mechanism. Moreover, the fault-tolerance and the initial mapping issues are also under constraints. The STDP training has not been efficient so far for multiple layer neural networks. In Lee et al. ( 2018 ), the authors have presented a method to train the kernel of convolutional neural network with STDP the update mechanism is still complicated for full-hardware implementation. In Shi et al. ( 2021 ), a hardware-friendly method has been introduced; however, the procedure is based on layer-based error backpropagation, which is not bio-plausible. Consequently, STDP-based multiple-layer network training is still an open problem for future researches. In our R-NASH design, a neuron can send its spikes to a high number of downstream neurons. The routing can be multicasting by replicating the flits during operation. In the NI, we also support a relaying protocol that allows the incoming spikes to be relayed to the nearby cores. However, there is a limitation on the number of upstream neurons that can send spikes to a neuron. Since the synaptic SRAM size is unchanged after manufacture, the number of memory cells to store the weight is limited. Consequently, it not possible to extend the size of up-stream neurons. Tree topology has been considered as more efficient than mesh topology (Merolla et al., 2013 ) as the packet relayed peak at O ( n 3/2 ) for unicast mesh and O ( n ) for the multicast tree. On average, mesh unicast is still O ( n 3/2 ) while tree multicast drop to O ( n ). Consequently, lesser packets are being relayed in tree topology thanks to its natural structure. However, while shifting to 3D-ICs, mesh topology can naturally extend into the third dimension by simply adding the vertical connection (up and down). Meanwhile, tree-topology in Merolla et al. ( 2013 ) support left, right, and top directions, which cannot naturally extend into the third dimension. Also, as mesh topology can provide escape channels if there is faulty ports or faulty neuron, tree-topology is extremely sensitive with fault as a single fault can cut a part of the tree out of communication. The problem of large fan-in and fan-out is an important one to be addressed. Bamford et al. ( 2010 ) presented a method tailored for the address-event receiver that allows a large axonal fan-out structure for neuromorphic systems. The technique can help solve the sizeable fan-in issue that our neuromorphic system is limited on the SRAM size. Meanwhile, Zamarreño-Ramos et al. ( 2012 ) shows a multicasting mesh for AER that allows communication in neuromorphic systems via 2D-Mesh topology. Compared to our approach, this work provides destination-driven routing, which is similar to our routing but in 2D Mesh. They also offer source-driven routing as a multicasting approach. This method requires a look-up to help the router understand the branch structure of the routing path. The source-driven routing can be efficient; however, it requires a large memory block in each router for routing, which is problematic for large-scale systems. Moreover, as we focus on the fault-tolerant routing in this work, table-based routing has two significant drawbacks. First, it introduces extra faulty elements. The faulty routing tables change the routing path, which may create a deadlock or livelock scenario. Second, any change in neuron location leads to an extensive amount of table update in the system."
} | 4,219 |
27065784 | PMC4809886 | pmc | 179 | {
"abstract": "Conventional vision, auditory, and olfactory sensors generate large volumes of redundant data and as a result tend to consume excessive power. To address these shortcomings, neuromorphic sensors have been developed. These sensors mimic the neuro-biological architecture of sensory organs using aVLSI (analog Very Large Scale Integration) and generate asynchronous spiking output that represents sensing information in ways that are similar to neural signals. This allows for much lower power consumption due to an ability to extract useful sensory information from sparse captured data. The foundation for research in neuromorphic sensors was laid more than two decades ago, but recent developments in understanding of biological sensing and advanced electronics, have stimulated research on sophisticated neuromorphic sensors that provide numerous advantages over conventional sensors. In this paper, we review the current state-of-the-art in neuromorphic implementation of vision, auditory, and olfactory sensors and identify key contributions across these fields. Bringing together these key contributions we suggest a future research direction for further development of the neuromorphic sensing field.",
"conclusion": "Conclusion In this paper, we have reviewed some of the most significant research contributions toward the improvement of neuromorphic vision, auditory, and olfactory sensors. The distinctive properties of neuromorphic sensors, such as sparse data output and low power consumption have led to extensive research and commercialization. The concept of developing neuromorphic sensors by emulating neuro-biological sensing in silicon has been progressing for many years. More recently, path-breaking research in biological sensing has provided an impetus to developments in neuromorphic sensing, especially in vision and auditory sensors. Pioneering contributions such as DVS and DAVIS, and AEREAR2 have provided considerable progress toward a sensor design that simulates neuro-biological vision and auditory sensing. Accordingly, these have led to the development of several applications for these sensors aiming at replacing conventional sensors in vision and audition. What is lacking is research that provides benchmarks for olfactory sensor implementation and its performance evaluation. Subsequently, future development in neuromorphic sensing should focus on the correlation of inputs from different sensors and efficient pre-processing. With this review we have identified challenges for future research on neuromorphic olfaction, building on the advancements made in vision and audition. In addition to neuromorphic olfaction, future research directions should target neuromorphic sensing of parameters such as pressure, vibration, thermal, and magnetic field as well as their intercorrelated sensor fusion functions which would be ideal for applications such as the Internet of Things (IoT).",
"introduction": "Introduction The field of neuromorphic engineering has been developing rapidly over the last decade. With the growing trend toward embedding intelligence in day-to-day devices, we are constantly making our surroundings smarter and more adaptive to our behavior. However, this technological progression requires an ever increasing number of sensors and associated data storage (Tenore and Etienne-Cummings, 2011 ). Along with the data processing challenges, factors such as power consumption and financial viability limit the development of smart devices. Realizing these limitations in the late 1980s, Carver Mead introduced the concept of neuromorphic engineering. This interdisciplinary field addresses the underlying concepts of neurobiological architecture and mimics its implementation using aVLSI. Neurobiological architecture is a low power consuming system which learns through exposure; these attributes, along with sparse output are crucial design criteria for neuromorphic systems (Mead, 1990 ). Neuromorphic approaches have been applied in implementing neural processors, developing neural networks, and particularly in electronic sensing where novel methodologies have been developed (Chicca et al., 2014 ). In the mid 1980s, Max Delbrück, John Hopfield, Carver Mead, and Richard Feynman collaborated to exploit the non-linear current characteristics of transistors (Indiveri and Horiuchi, 2011 ). Carver Mead, further highlighted the excessive dissipation of energy through conventional computing methods and the limitation of using transistors merely as digital switching components. He proposed that the analog physical properties of transistors could be exploited to design adaptive and low power consuming sensors like silicon retina and cochlea (Mead, 1990 ). With these models as inspiration, neuromorphic concepts have been applied to vision sensors, auditory sensors, and olfactory sensors. Current sensor advances have been supported by massive parallelism, asynchronous processing and self-organization (Douglas et al., 1995 ; Liu et al., 2015 ). As described in Hasler and Marr ( 2013 ), analog implementation of neural-like systems are capable of approaching equivalence to biological systems in terms of power consumption and efficiency. Most of the research in neuromorphic sensors has involved aVLSI; however, with rapidly changing technology, digital electronics has also been applied to implement neuromorphic concepts, particularly as it has proved to be robust to internal and external noise (Sarpeshkar, 1998 ). Systems implemented using digital electronics are easily programmed and upscaled (Sarpeshkar, 2006 ). Regardless of whether analog or digital implementations are used, the lack of standards and benchmarks for the output of neuromorphic sensors may limit their development and adoption. In the same way that the interfacing for neuromorphic sensors uses standard Address Event Representation (AER; Boahen, 2000 ), a standardized method to evaluate sensor outputs could help establish appropriate benchmarks for further improvement. In this paper we review significant recent contributions to neuromorphic vision, auditory, and olfactory sensing and compare them to identify potential benchmarks for neuromorphic sensors."
} | 1,543 |
26967105 | PMC4989317 | pmc | 181 | {
"abstract": "Much research has been invested into engineering microorganisms to perform desired biotransformations; nonetheless, these efforts frequently fall short of expected results due to the unforeseen effects of biofeedback regulation and functional incompatibility. In nature, metabolic function is compartmentalized into diverse organisms assembled into robust consortia, in which the division of labor is thought to lead to increased community efficiency and productivity. Here we consider whether and how consortia can be designed to perform bioprocesses of interest beyond the metabolic flexibility limitations of a single organism. Advances in post-genomic analysis of microbial consortia and application of high-resolution global measurements now offer the promise of systems-level understanding of how microbial consortia adapt to changes in environmental variables and inputs of carbon and energy. We argue that, when combined with appropriate modeling frameworks, systems-level knowledge can markedly improve our ability to predict the fate and functioning of consortia. Here we articulate our collective perspective on the current and future state of microbial community engineering and control while placing specific emphasis on ecological principles that promote control over community function and emergent properties."
} | 331 |
23110450 | null | s2 | 182 | {
"abstract": "Dragline silk from orb-weaving spiders is a copolymer of two large proteins, major ampullate spidroin 1 (MaSp1) and 2 (MaSp2). The ratio of these proteins is known to have a large variation across different species of orb-weaving spiders. NMR results from gland material of two different species of spiders, N. clavipes and A. aurantia , indicates that MaSp1 proteins are more easily formed into β-sheet nanostructures, while MaSp2 proteins form random coil and helical structures. To test if this behavior of natural silk proteins could be reproduced by recombinantly produced spider silk mimic protein, recombinant MaSp1/MaSp2 mixed fibers as well as chimeric silk fibers from MaSp1 and MaSp2 sequences in a single protein were produced based on the variable ratio and conserved motifs of MaSp1 and MaSp2 in native silk fiber. Mechanical properties, solid-state NMR, and XRD results of tested synthetic fibers indicate the differing roles of MaSp1 and MaSp2 in the fiber and verify the importance of postspin stretching treatment in helping the fiber to form the proper spatial structure."
} | 272 |
30214936 | PMC6135543 | pmc | 183 | {
"abstract": "Resistive switching devices were used as technological synapses to learn about the spatial- and temporal-correlated neuron spikes.",
"introduction": "INTRODUCTION The human brain outperforms most of the existing artificial neural networks (ANNs) in terms of energy-efficient and error-tolerant computing ( 1 , 2 ). One of the main differences between ANNs and the human brain is the representation of information ( 3 , 4 ): While most ANNs represent input/output data as real-valued vectors, the human brain encodes information via binary spikes. Biological neurons follow an all-or-none rule, where a neuron emits a unanimous action potential, or spike, when the stimulus is large enough; otherwise it keeps silent ( 5 , 6 ). Spiking neural networks (SNNs) ( 7 , 8 ) were introduced to emulate the style of information processing in the human brain, although the representation methodology, that is, how the sensory information is coded by neuron spikes, is still under debate ( 9 , 10 ). In rate coding (fig. S1A) ( 11 ), the intensity of an external stimulus is represented by the spiking rate. However, information is carried by a train of spikes, thus resulting in relatively low information density and low energy efficiency. Recent biological studies provide evidence for spatiotemporal coding ( 12 , 13 ), where information is represented by the spatial and temporal occurrences of the spikes. The neuron representing the strongest stimulus spikes first, followed by spikes of neurons representing lower-intensity stimuli (fig. S1B) ( 14 ). This spatiotemporal coding enables a high density of information with relatively few neurons and spikes, hence with high energy/area efficiency ( 9 , 15 ). To emulate the high synaptic density and the energy efficiency of biological neural networks, nanoscale resistive switching devices ( 16 , 17 ), such as resistive random access memory (RRAM) and the phase change memory, were introduced. These devices are two-terminal nanoscale devices that can change their resistive switching in response to external stimuli, similar to the plasticity mechanism in biological synapses. Recently, a close resemblance with biological synaptic and neuronal behaviors was reported by adopting volatile RRAM devices ( 18 – 20 ). Because of the good scaling ability and compatibility with the silicon-based technology of today’s microelectronic systems, resistive switching devices might enable the construction of large-scale neural networks with complexity comparable to the human brain ( 21 , 22 ). There has been a hardware demonstration of both conventional ANNs with real-valued neuronal signals ( 23 , 24 ), directly implementing matrix-vector multiplication by cross-point arrays ( 25 – 28 ), and SNNs where the information is carried by spiking signals ( 21 ). However, the spike representation relied on a simple spatial coding, that is, neurons spike synchronously to form a spatial-only pattern ( 21 , 29 – 32 ). SNNs capable of spatiotemporal computing with RRAM synapses would greatly improve the energy and information efficiency of neuromorphic hardware, thus accelerating the progress toward human-like cognitive computing.",
"discussion": "DISCUSSION The SNN is considered to be the third generation of neural networks ( 7 ), following the first generation based on McCulloch-Pitts neurons ( 36 ) and the second generation relying on neural activation functions and gradient descent (for example, error backpropagation) as the core of the learning algorithm ( 46 ). The computational capability of the SNN is improved thanks to the brain-inspired spatiotemporal representation of spiking neurons containing multiple information, such as the timing of a specific sensory input and its relationship with other events. Previous works on RRAM-based SNNs showed learning and recognition of spatial-only patterns with no specific temporal information, which strongly reduces the capability of the SNN ( 21 , 29 ). For instance, a spiking pattern of 16 PREs and only 4 PREs spiking can result in 1820 possible spatial-only patterns, while the number of possible spatiotemporal patterns with a constant spike timing interval is 43,680, that is, 24 times higher. For the case of sound origin detection, only two PREs are needed to represent all possible azimuth angles. On the other hand, rate-based coding, which is extensively embraced in RRAM SNNs ( 47 , 48 ) and complementary metal-oxide semiconductor (CMOS) neuromorphic computing ( 49 ), has been questioned by many experimental observations in neuroscience ( 14 ); for instance, in the human brain visual cortex, the ability of visual pattern classification is about 100 ms, which would be too short to carry sufficient rate-based information given the low spiking rate of neurons of about 10 Hz ( 50 ). Spatiotemporal coding strongly reduces the number of spikes (hence, amount of energy) needed to represent a given analog information, compared to rate-based coding. In summary, a RRAM-based synaptic neuromorphic network is proposed to learn and recognize spatiotemporal patterns, including spike sequences and spike groups (for example, pairs) where the spike timing carries information. The time difference of spikes among different neurons provides spatiotemporal coding with high sparsity and high information capacity. Our experiments demonstrate that the RRAM-based SNN, combined with suitable neuron circuit and operation scheme, is capable of learning spatiotemporal patterns via STDP, followed by recognition, thanks to suitably potentiated/depressed synapses. The results provide one step forward toward biorealistic computing machines capable of paralleling the energy efficiency and computing functionality of the brain."
} | 1,426 |
33143169 | PMC7709699 | pmc | 184 | {
"abstract": "The term ‘biomimetic’ might be applied to any material or process that in some way reproduces, mimics, or is otherwise inspired by nature. Also variously termed bionic, bioinspired, biological design, or even green design, the idea of adapting or taking inspiration from a natural solution to solve a modern engineering problem has been of scientific interest since it was first proposed in the 1960s. Since then, the concept that natural materials and nature can provide inspiration for incredible breakthroughs and developments in terms of new technologies and entirely new approaches to solving technological problems has become widely accepted. This is very much evident in the fields of materials science, surface science, and coatings. In this review, we survey recent developments (primarily those within the last decade) in biomimetic approaches to antifouling, self-cleaning, or anti-biofilm technologies. We find that this field continues to mature, and emerging novel, biomimetic technologies are present at multiple stages in the development pipeline, with some becoming commercially available. However, we also note that the rate of commercialization of these technologies appears slow compared to the significant research output within the field.",
"conclusion": "3. Conclusions and Outlook In conclusion, we offer the following observations and recommendations for future work in this area, on the basis of evident gaps in the current literature: i. An environmentally conscious, biocide-free solution to biofouling is an attractive target, as it aligns with societal and political objectives. This is in contrast to the majority of commercially available solutions. ii. Future research should take into account the plurality of natural processes achieving the same goal. For example, the “lotus effect” and “ Salvinia effect” are two distinct hydrophobic strategies, with the latter better suited to submersible applications. iii. Alongside antifouling effectiveness, other salient physicochemical properties of proposed biomimetic solutions should be ascertained, so as to better identify suitable applications. For instance, many proposed solutions lack the durability required for external applications. iv. Topographical solutions are often beyond the economical limits of technology, despite many efficient structures existing in the natural world. v. The successful scaling and integration of antifouling strategies into industry-standard processes is required to promote adoption of these solutions. However, this is, as of yet, an under-explored avenue. vi. A multi-faceted approach perhaps holds the greatest promise of a widely applicable solution to biofouling. This broad survey of recent literature, particularly that of the past decade, confirms again that biomimetic approaches to the development of antifouling technologies undoubtedly holds great promise, and research in this area is gathering pace, with many areas still to be explored. However, it would seem to be the case that commercial technologies resulting from these approaches often are yet to be realised from much of the reviewed research. Perhaps this is a result of a long development pipeline, or, perhaps more likely, there is still (as yet) an incomplete understanding of how natural surfaces self-clean or control fouling in many areas, and where the mechanisms are largely understood, there are often difficulties in faithfully replicating the complex natural structures or chemistries at sufficient scale for commercial application with the current technologies at hand. The question remains as to what can be done to further advance or accelerate integration of biomimetic materials and bioinspired design into future viable and commercial antifouling technologies, or those other technological challenges mentioned at the beginning of this review. Perhaps it is again important to reiterate again that rather than merely “mimicking” existing natural solutions (by directly replicating and applying them in an analogous fashion to how they operate in the natural world), we should continue to strive to be truly “inspired” by natural materials and processes into new solutions for technological development.",
"introduction": "1. Introduction Biomimetics, a term attributed to Otto Schmidt, can be thought of as the study of structure and function in natural systems as inspiration for (sustainable) technological design and engineering [ 1 ]. Vincent provides an excellent overview of the development of the field, including the semantics and history of the term [ 2 ]. Research into how nature produces materials has seen rapid growth, with the realisation that natural processes can be very efficient, for example biomineralisation [ 3 ]. Natural self-assembly processes can also show precise control of surface structure or processes (from the macroscale to the nanoscale) and may also be accompanied by a high fault tolerance. However, as pointed out succinctly by Vincent, it is not sufficient to try and transfer lessons from nature into existing technology, but rather biomimetics hold the promise of more—a new way of looking at the development of technologies, where challenges are approached with an understanding taken from the natural world [ 2 ]. The natural world has had, by most accounts, some 3.8 Gyr of diverse development and evolution in order to refine such processes, and the materials subsequently produced [ 4 , 5 ]. It would seem natural to take inspiration from these materials and processes when creating nanomaterials or devices in the lab, perhaps efficiently resulting in novel technologies and approaches; however, it is clear that this path is often not quite as straightforward [ 2 ]. Natural processes result in sophisticated structures (often from very simple starting conditions) that utilise complex interplay between surface topography and chemical properties, for example, in order to create multi-functionality. Nano- or micro-scale structures or surfaces that have different length scales superimposed on one another (hierarchical), for example, are common in nature (discussed later) and provide many interesting possibilities when developing novel synthetic materials. As pointed out by Bushan, multifunctionality in natural materials is quite common, and properties such as superhydrophobicity, self-cleaning, drag reduction, thermal insulation, high adhesion strengths or even reversible adhesion, aerodynamic or hydrodynamic lift, incredible mechanical or structural strength (and strength to weight ratios), self-assembly, anti-reflection, structural colour, or self-healing are only some examples of the properties of these materials that are of possible commercial interest [ 6 ]. There are many excellent published studies and reviews that address the adoption of bioinspired or biomimetic design for these purposes, for example in the field of adhesion studies [ 1 , 6 , 7 ]. However, existing reviews precede a substantive rise in the number of studies describing biomimetic antifouling strategies. In this review, we emphasise important recent developments in the area of biofouling control and antifouling approaches and focus primarily those that have arisen within the past 10 years. Antifouling or self-cleaning surfaces are also a sub-set of these studies ( Figure 1 ), and the relationship between searching for inspiration from natural adhesives on one hand, and naturally anti-adhesive surfaces or materials on the other hand, is of particular interest. Inspiration may be drawn from organisms, primarily aquatic, that colonise or are subjected to colonisation; the former will develop processes of adhesion, while the latter will develop processes to combat adhesion. 1.1. What is Biofouling and Why Control it? Biofouling can be defined as the undesired attachment and growth of life on artificial surfaces [ 8 ]. However, there can be great differences in what might be considered as a significant level of biofouling between industries, e.g., in the marine environment, biofouling is often visible to the human eye as mussels, barnacles, or seaweed attached to surfaces such as the hull of a ship or the piling of a pier [ 9 ]. Additionally, different applications will impose differing constraints on other physical parameters of the antifouling material. Biofouling and biofilms are present and just as problematic in other industries—for example, in water purification systems, space flight [ 10 ], and medicine [ 11 ]—but are perhaps less obviously visible and arguably more difficult to control. Flemming provides a sample list of industries known to be affected by biofouling [ 8 ]. The common overarching incentives to preventing biofouling in many industries generally relate to either economic or human health impacts, or both—such as cleaning and the associated downtime (e.g., ship fouling and dry docking times [ 12 ]), contaminated raw materials, poor performance of critical technologies or components (e.g., heat exchangers or membranes), or shortened component lifetime (of membranes, etc.). Biofouling negatively impacts commercial marine industries and activities, slowing ship speed and increasing the annual fuel consumption of the global shipping fleet [ 13 ], and influencing the rate and extent of corrosion on offshore platforms or commercial pipelines for example. Any particular technologies that reduce drag on the hull of a ship in service would have many benefits, increasing fuel efficiency (and aiding in decarbonising global transport) and vessel availability, for example, and reducing invasive species translocation and life-cycle costs [ 1 ]. In a medical setting, bacterial biofilms and microbial fouling can pose a serious risk, on medical implants for example, and biofilm development on surfaces can be associated with nosocomial (hospital-acquired) infections [ 14 ] and development of antibiotic resistance in clinical or medical settings [ 15 ]. The most effective general commercial approaches to preventing biofouling in many settings could have, until recently, been categorised as strategies to attempt to kill and/or clean biofilms. Antifouling or anti-biofilm approaches in many affected industries have largely involved coatings containing biocides as active ingredients to kill or somehow deter fouling [ 16 ]. Unfortunately, it is well recognised that these biocidal formulations can have negative effects on the environment [ 17 ], and widespread introduction of new biocides has been legislatively regulated in Europe and many areas worldwide as a result of these impacts [ 18 ]. This has prompted academic and commercial interest in seeking out new approaches to preventing, reducing, or mitigating the effects of biofouling or biofilm development, particularly in the marine environment, or in membrane technology for water separation, for example [ 19 ]. Kyei et al. have recently reviewed the currently available methods for preventing and mitigating growth of algal and other organisms on marine structures in an environmentally friendly and cost-effective (and legal) manner [ 20 ]. 1.2. Biomimetics and Biofouling Control Biomimetic research and biofouling/antifouling research interests intersect at this point ( Figure 1 ), as some natural biological surfaces are self-cleaning or naturally possess some antifouling capability (or, in the case of living surfaces that prevent colonisation, a capacity to prevent epibiosis). Particularly interesting examples are those chemical defences of marine organisms—many of which are well-known, such as the secondary metabolites of the red seaweed Delisea pulchra [ 21 ]—that control unwanted epibionts (colonising organisms, or an organism living on another organism). Many of these are still under study [ 22 ]. As a recent example, Karnjana et al. have reported that extracts from a red seaweed ( Gracilaria fisheri) decreased Vibrio harvey biofilm formation [ 23 ]. The group have subsequently isolated and identified the active compounds [ 24 ]. Other possible approaches are categorised in general terms in Figure 2 below. These include the well-studied hydrophobic self-cleaning properties of certain plant species, such as the “lotus effect” as observed in Nelumbo nucifera (whose mechanisms are elucidated and effects artificially reproduced in [ 25 , 26 , 27 ]). Some of these approaches have made it out of the laboratory and into broader technical applications, including some interesting recent patents [ 28 ] and applications [ 29 ]. Of course, these strategies have drawbacks, and there are undoubtedly technical challenges to be overcome in many applications—for example, Flemming points out that the lotus effect, despite its aquatic origins, has not found widespread commercial application on surfaces submerged in water, as the methods to date are based on hydrophobicity requiring the presence of both water and gas phases (a liquid–gas interface), which is difficult to maintain [ 8 ]. However, other biomimetic air-retaining strategies such as the “ Salvinia effect” hold further potential in creating persistent air layers on surfaces such as a ship’s hull [ 30 ]. The four criteria that are considered important here for long-term air retention under water are hair-like structures, a hydrophobic chemistry, inclusion of topographic features with undercuts, and the elastic nature of the structures [ 31 ]. Novel methods of producing such surfaces that create a “ Salvinia effect” are currently being examined, including the use of vertically aligned carbon nanotubes [ 32 ], and Zhou et al. have recently reported a facile, repeatable method of fabricating Salvinia -like surfaces [ 33 ]. The potential applications of such materials appear widespread, and Busch et al. recently calculated that surrounding a hull with an air layer could lead to estimated savings of 32.5 million tons of fuel (some 13% of fuel consumption of the worldwide shipping fleet), or USD 18.5 billion and 130 million tons of CO 2e per year [ 34 ]. Busch et al. also point out that successful application of such air-retaining surfaces could have a combined drag reduction and antifouling effect on a ship’s hull. The authors then provided updated figures incorporating antifouling and drag reduction of 25%, resulting in saving some 40.6 million tons of fuel, or some USD 23.2 billion in cost reduction, or approximately 162.5 million tons of CO 2e . These figures starkly demonstrate that the potential of such technologies, even if not completely realised in practise, surely justify further investment and exploration of any natural surfaces that achieve such effects."
} | 3,655 |
34006640 | PMC8166027 | pmc | 185 | {
"abstract": "Significance Biofilms are a ubiquitous form of bacterial community. Within biofilms, bacteria communicate via chemical signals in a process called quorum sensing (QS). However, if signal production is nutrient-limited, then the nutrient-deficient interior of a biofilm cannot contribute to QS, which limits the ability of bacteria to assess their own population and behave accordingly. Numerical simulations of competitions among biofilm bacteria led us to discover a biophysical limit on the efficacy of nutrient-limited QS. In view of this limit, we conclude that to be most effective QS signal production should be a prized function that is not metabolically slaved.",
"discussion": "Discussion We find that when production of a nondecaying AI is limited by a diffusible nutrient from a remote source there exists a biophysical limit on the DR of AI concentrations that cells can experience. Using agent-based simulations of biofilm growth, we demonstrate an illustrative case in which QS-based matrix-production strategies can provide a large competitive advantage—but not if AI production is limited by nutrient availability. Importantly, this biophysical limit is essentially independent of the diffusivity of the AI, the size or shape of the cells, or of the concentration of the growth-limiting nutrient at its source. While our illustrative case provides a concrete mechanistic example of the effects of this biophysical limit, the limit is much more general and applies to all scenarios involving QS in nutrient-limited environments. In principle, nutrient-limited AI production could still be exploited by bacteria in several ways. For example, in a biofilm where the density of cells is high, bacteria could employ QS to infer the concentration of the diffusible nutrient at its source. This is because, for a nondecaying AI, the local AI concentration mirrors the nutrient concentration, so that a locally depleted nutrient with a high AI concentration would imply a large nutrient source. Further, even at lower cell densities, nutrient-limited AI could act as a single consolidated chemotactic signal that would indicate, via its negative gradient, the direction of the source of the nutrient. Alternatively, the relevant information for bacteria might be the number of growing cells in the vicinity rather than the total number of cells. Nevertheless, the advantages of QS seem to be much greater if AI production is not nutrient-limited. Our results suggest a metabolic interpretation of autoinduction, i.e., positive feedback of AI production from AI sensing, which is a well-established feature of many QS systems ( 2 , 30 – 33 ). It is not fully understood why autoinduction per se is desirable for cells to sense their local density. It could be presumed that a higher density of cells would necessarily result in a higher AI concentration, obviating the need for positive feedback on AI production. However, this presumption would not be correct if AI production were nutrient-limited: Above a threshold cell density, AI concentration would hit its maximum as given by Eq. 2 and provide no further information. From this perspective, autoinduction may simply represent one way of breaking the dependence of AI production on nutrient availability in order to evade the biophysical limit ( Eqs. 2 – 4 and SI Appendix , section S1A ). Another way of breaking this dependence would be for AI production to depend nonlinearly on nutrient availability, such that AI production per unit of nutrient consumed is high at low nutrient availability. In order to highlight the importance of metabolic dependence in QS, we have chosen to contrast the two extreme cases of constitutive AI production (which would be an example of such a nonlinearity) and strictly linear dependence. Real systems might fall between the two, and this is an area that merits further investigation including experimental study. In any case, either autoinduction or a nonlinear dependence of AI production on nutrients would suggest that QS is a prized metabolic function that is prioritized by the cell in nutrient-limited conditions. Our predictions can be tested experimentally. For example, to test our prediction that AI production must be privileged, the rates of AI production in nutrient-limited and nutrient-replete conditions can be compared. If AI production is higher in nutrient-limited conditions than predicted by a strict proportionality to growth rate, it would suggest that QS is a prized function. Generally, investigation of how AI production scales with growth rate under nutrient limitation would reveal the physiological importance of QS in different growth conditions. Further, understanding the joint effect of nutrient limitation and autoinduction would reveal how the cell integrates two vital pieces of information: its local nutrient availability and the presence of other cells. While strongly nutrient-limited AI production could be used by bacteria to infer some types of information about their environment, our main conclusion is that to reliably infer local cell density via QS, AI production should not be entirely metabolically slaved. We propose that AI production and QS are privileged bacterial functions and that despite the strong links between metabolism and AI production ( 16 , 17 , 34 , 35 ), and the substantial cost of production of some AIs ( 36 , 37 ), cells are able to decouple the two processes and regulate AI production largely independently of cell metabolism. More broadly, we hope that the results of this study will spur theoretical and experimental interest in the roles of metabolic dependency of QS signal production."
} | 1,414 |
28584701 | PMC5455293 | pmc | 187 | {
"abstract": "Coral microbiomes are known to play important roles in organismal health, response to environmental stress, and resistance to disease. The coral microbiome contains diverse assemblages of resident bacteria, ranging from defensive and metabolic symbionts to opportunistic bacteria that may turn harmful in compromised hosts. However, little is known about how these bacterial interactions influence the mechanism and controls of overall structure, stability, and function of the microbiome. We sought to test how coral microbiome dynamics were affected by interactions between two bacteria: Vibrio coralliilyticus , a known temperature-dependent pathogen of some corals, and Halobacteriovorax , a unique bacterial predator of Vibrio and other gram-negative bacteria. We challenged reef-building coral with V. coralliilyticus in the presence or absence of Halobacteriovorax predators, and monitored microbial community dynamics with 16S rRNA gene profiling time-series. Vibrio coralliilyticus inoculation increased the mean relative abundance of Vibrios by greater than 35% from the 4 to 8 hour time point, but not in the 24 & 32 hour time points. However, strong secondary effects of the Vibrio challenge were also observed for the rest of the microbiome such as increased richness (observed species), and reduced stability (increased beta-diversity). Moreover, after the transient increase in Vibrios, two lineages of bacteria ( Rhodobacterales and Cytophagales ) increased in coral tissues, suggesting that V. coralliilyticus challenge opens niche space for these known opportunists. Rhodobacterales increased from 6.99% (±0.05 SEM) to a maximum mean relative abundance of 48.75% (±0.14 SEM) in the final time point and Cytophagales from <0.001% to 3.656%. Halobacteriovorax predators are commonly present at low-abundance on coral surfaces. Based on the keystone role of predators in many ecosystems, we hypothesized that Halobacteriovorax predators might help protect corals by consuming foreign or “alien” gram negative bacteria. Halobacteriovorax inoculation also altered the microbiome but to a lesser degree than V. coralliilyticus , and Halobacteriovorax were never detected after inoculation. Simultaneous challenge with both V. coralliilyticus and predatory Halobacteriovorax eliminated the increase in V. coralliilyticus , ameliorated changes to the rest of the coral microbiome, and prevented the secondary blooms of opportunistic Rhodobacterales and Cytophagales seen in the V. coralliilyticus challenge. These data suggest that, under certain circumstances, host-associated bacterial predators may mitigate the ability of other bacteria to destabilize the microbiome.",
"conclusion": "Conclusions Corals and other marine organisms are in constant contact with an array of distinct microbes. In the face of this constant microbial challenge healthy host microbiomes are robust to change unless challenged with foreign agents or poor environmental conditions. In this laboratory system, a challenge by V. coralliilyticus resulted in a destabilized host microbiome in which opportunists bloomed, potentially further exacerbating the negative effects of the initial inoculation. Recent evidence also supports the hypothesis that commensal or mutualistic host-associated microbes offer protection against invasive pathogens either by depriving these alien bacteria of essential nutrients or acting as a physical barrier to host attachment ( Weyrich et al., 2014 ). Here we show additional evidence by which host-associated predatory bacteria protect the host microbiome through direct consumption of prey. The ability to manipulate the microbiome and therefore test various hypotheses about the principles that govern microbial community assembly, dynamics, and functions, especially in terms of how these relate to host health, remain a challenge for our field ( Waldor et al., 2015 ). As our ability to culture and test the effects of more coral microbial taxa improves, so will our methods to manipulate and track the dynamics of the microbiome in real time and under more realistic environmental conditions. Such efforts will allow us to gain a better understanding of the relationships among members of the microbiota and between the microbiome with the host, and which ideally will result in better understanding and management of bacterial mediated diseases.",
"introduction": "Introduction Coral reefs have experienced sharp declines in coral cover from environmental factors ( De’ath et al., 2012 ), temperature induced bleaching ( Fitt & Warner, 1995 ), and disease ( Bourne et al., 2009 ; Burge et al., 2014 ), with some areas of the Caribbean experiencing as much as 80% coral loss over the past several decades ( Gardner et al., 2003 ). While many studies have identified microbial consortia that increase in diseased corals (e.g., Gignoux-Wolfsohn & Vollmer, 2015 ), etiological agents are generally unknown for the majority of coral diseases ( Mouchka, Hewson & Harvell, 2010 ) and others are defined in broad terms as polymicrobial disease ( Cooney et al., 2002 ). Vibrio coralliilyticus is a well described model bacterium for the study of interactions between corals, the environment, and pathogenic bacteria ( Ben-Haim et al., 2003 ). Several V. coralliilyticus virulence factors are temperature-dependent and upregulated above 27 °C ( Kimes et al., 2012 ), and it has been suggested that host tissue invasion can only occur above this threshold ( Vidal-Dupiol et al., 2011 ). Given the continuous rise in sea surface temperatures due to global climate change ( Hoegh-Guldberg et al., 2007 ), and the projected increased variability of temperature extremes, it is likely that the incidence of infections by V. coralliilyticus and other temperature-dependent pathogens will increase ( Maynard et al., 2015 ). Bacterial communities of diseased corals are also known to have large numbers of opportunistic pathogens and secondary colonizers ( Gignoux-Wolfsohn & Vollmer, 2015 ). It has been hypothesized that the majority of coral disease may be the result of normally-benign coral microbionts that become opportunistic pathogens during physiological stress to the host ( Lesser et al., 2007a ). Thus, the linkages between infection by a primary foreign agents and secondary opportunistic infections remain an area of active exploration. Corals also form mutualistic and commensal partnerships with diverse microorganisms, ranging from endosymbiotic photosynthetic dinoflagellates ( Symbiodinium spp.), to consortia of archaea, fungi, and bacteria. Although the role of Symbiodinium in the coral holobiont is well studied, the exact roles of each member of the bacterial portion of the holobiont remains far from clear. Experiments and metagenomic analyses have provided some insights into the roles of individual members of the coral microbiome (e.g., Wegley et al., 2007 ). It has been suggested that some of these bacteria provide direct benefits to the coral host, such as nitrogen fixation by symbiotic Cyanobacteria in Montastraea cavernosa ( Lesser et al., 2007b ), or ammonia oxidation by archaea ( Beman et al., 2007 ). Other bacteria, particularly those in the coral surface mucus layer, are thought to provide a first line of defense against potentially invading foreign bacteria. Mucosal bacteria are thought to protect the host by several mechanisms, including production of antibiotics ( Ritchie, 2006 ), secretion of chemical compounds that inhibit pathogen metabolism ( Rypien, Ward & Azam, 2010 ), or competition for necessary resources and niche space ( Ritchie & Smith, 1997 ). Increasingly, viruses and phages are recognized as also playing a regulatory role in the holobiont by controlling microbial populations ( Barr et al., 2013 ; Soffer, Zaneveld & Thurber, 2014 ; Nguyen-Kim et al., 2014 ). We have recently described how the predatory bacteria Halobacteriovorax , also likely influences the diversity and dynamics of the microbial community in the coral surface mucus layer through consumption of a broad range of bacterial prey ( Welsh et al., 2016 ). Halobacteriovorax spp. are small, highly motile predatory bacteria that exhibit a biphasic lifestyle and prey exclusively on gram negative bacteria, including known coral pathogens ( Williams, Falkler & Shay, 1980 ; Welsh et al., 2016 ). Halobacteriovorax are the marine component of a group of delta-proteobacteria known as Bdellovibrio and like organisms (BALOs). In free-living attack phase, BALOs actively seek out prey in order to attach, burrow inside, and restructure their host cell into a rounded bdelloplast. This kills their prey and provides BALOs with an osmotically stable structure free from competition to utilize prey resources for growth and replication. A new generation of attack-phase predators then bursts forth from the bdelloplast to seek new hosts. Bacterial predators in the coral microbiome could be a type of top-down control, that directly alters the structure and function of the coral microbiome as demonstrated in other aquatic systems by bacterivorous predators (see reviews by Jürgens & Matz, 2002 ; Pernthaler, 2005 ; Matz & Kjelleberg, 2005 ). For example, we highlighted potential interactions of Halobacteriovorax and other members of the coral holobiont using co-occurrence network analysis of an in-field experimental time series of three coral genera, across three years, several treatments, and range of temperature conditions. These networks showed that Halobacteriovorax are core members of the coral microbiome, present in >78% of samples from three coral genera Porites, Agarica, and Siderastrea ( Welsh et al., 2016 ; Zaneveld et al., 2016 ). We also showed that isolated strains of coral-associated Halobacteriovorax prey upon known coral pathogens in cultured settings. Such antagonisms between predators and prey in the holobiont may have variable effects on the microbiome, such that they could be occlusive to pathogens or disruptive to the coral microbiome itself. Here we examine how a specific bacterial predator ( Halobacteriovorax ), a foreign bacterium ( V. coralliilyticus ), and a coral host ( M. cavernosa ) interact to affect the complex system of the coral microbiome in a laboratory-based system. While V. coralliilyticus is not associated with causing disease in M. cavernosa , our study still provides baseline experiments for predator and prey interactions on a coral host.",
"discussion": "Discussion Tracking microbial community dynamics after different bacterial challenges High-throughput sequencing allows researchers to more easily document membership dynamics and community topology, yet we often lack the ability to confirm causal relationships among them. Manipulative studies are necessary to link cause and effect. While some host-microbe models can be more readily manipulated (e.g., mouse gut, squid light organ, and rhizosphere), there remain considerable methodological barriers for many systems, especially those for which gnotobiotic (germ-free) host animals are not available. Here we used individual and combinatorial bacterial challenges to a coral host in order to ask three specific questions: (1) How can inoculation of a foreign bacteria ( V. coralliilyticus ) alter the microbiome of a compromised host in a laboratory setting (2) Can predators of this taxa prevent or ameliorate the downstream effects of the alien challenges, and (3) Does the predator itself affect the coral microbiome? Vibrio coralliilyticus is a known disease-causing pathogen of corals worldwide ( Ben-haim, Zicherman-keren & Rosenberg, 2003 ; Wilson et al., 2013 ) and has been documented to induce bleaching and tissue loss in some species of corals ( Ben-Haim et al., 1999 ; Ben-haim, Zicherman-keren & Rosenberg, 2003 ). Furthermore, experimental evidence has demonstrated that under increased thermal stress V. coralliilyticus concentrations rise dramatically in corals ( Tout et al., 2015 ). However, the changes, if any, that V. coralliilyticus challenge causes to the microbial communities normally present in corals was previously unknown. V. coralliilyticus represents an alien or potential pathogen in this study, as it has not been previously shown to be a member of Montastraea cavernosa microbiome, and it remains unknown if it can infect certain Caribbean corals like Montastraea cavernosa. Never the less, determining how bacterial challenge can alter the normal flora of a host may provide insight into whether mutualists are lost and additional antagonisms arise during an infection cycle and thus contribute to secondary negative effects on animal hosts. Here we show that a V. coralliilyticus inoculation not only changes its own relative abundance in the system (as would be expected) but also alters the microbiome in various ways, including increases in alpha and beta diversity ( Fig. 3 ). However, when these corals were challenged with the V. coralliilyticus in the presence of the predator, these effects were diminished and resulted in almost no changes in the normal coral microbiome. Furthermore, addition of just the bacterial predator did not change the community in a similar fashion to the V. coralliilyticus challenge, suggesting that exposure to different foreign bacterial taxa (as opposed to any gram negative bacteria) will likely elicit variable downstream responses in the microbiome, unlike taxa that core members. Addition of V. coralliilyticus led to an increase in relative abundance in a known group of opportunists of corals, the Rhodobacterales ( Fig. 4 ). The increase in Rhodobacterales persisted at later time-points, even after the abundance of V. coralliilyticus had declined in the tissues. Rhodobacterales sequence abundances have been linked to disease outbreaks in white plague diseased Siderastrea siderea and Diploria strigosa corals ( Cárdenas et al., 2012 ). Also, sequences from the family Rhodobacteraceae have been shown to increase by 4-folds at the lesion front of corals with white syndrome ( Pollock et al., 2017 ). Rhodobacterales are fast growing taxa, capable of quickly responding to increasing availability of amino acids ( Mayali et al., 2014 ), and could be responding to resources made available from cells damaged by V. coralliilyticus. Such a mechanism would explain associations between Rhodobacterales and many stressed or diseased corals. While the present study cannot distinguish whether these secondary Vibrio -induced Rhodobacterales are harmful to corals, the experimental framework used here could test this question in the future. More broadly, V. coralliilyticus challenged samples showed a wider variety of bacteria sequences within the tissues ( Fig. 3A ). It is likely these opportunist species gained access and/or established within the tissue shortly after Vibrio coralliilyticus inoculation, as the increases in observed species persisted for the duration of the experiment ( Table S2 ). However, the disproportionate impact of the V. coralliilyticus was not observed in samples challenged with the predator Halobacteriovorax suggesting that this is not a generalizable response to the wounding and addition of any kind of bacteria. Halobacteriovorax as a possible top down control of opportunists We have previously cultivated Halobacteriovorax from multiple-species of corals, and used long-term microbial time series data to show that, despite its low abundance, it is a core member of the microbiome of several coral genera ( Welsh et al., 2016 ). Here we used bacterial challenge experiments to demonstrate that Halobacteriovorax can protect its coral host by consuming its prey V. coralliilyticus . We found that the application of Halobacteriovorax at the same time as V. coralliilyticus can prevent detectable changes in the relative abundance of V. coralliilyticus in M. cavernosa coral tissue after challenge. The Halobacteriovorax alone treatment showed higher variance in the mean relative abundance of Vibrionales than the controls or combined treatment ( Fig. 4 ), but the mean was not significantly different than the controls. Co-inoculations of this predator with V. coralliilyticus showed no significant differences in the abundance of Vibrionales in coral tissues versus control inoculations at any time in the course of the experiment ( Fig. 4 purple lines). Thus it is likely that these predators consumed the Vibrio immediately or at the point of inoculation, and therefore provided a biotic barrier to the host tissues. The ability of Halobacteriovorax to mitigate inoculation, if added hours or days after a V. coralliilyticus challenge, remains unknown. At the same time the generality of this effect of the Vibrio and the predatory Halobacteriovorax remains untested, but could be evaluated using similar methods to those we describe here. For example, additional types of pathogens that show clear infection signs upon addition or the use of more commensal strains of bacteria as controls would strengthen support for this hypothesis. Phage have already been shown to be effective against V. coralliilyticus ( Cohen et al., 2013 ), and likely play a role in controlling natural populations of V. coralliilyticus in the environment, which is similar to what has been suggested for phage and V. cholerae ( Faruque et al., 2005 ). Phages also provide an antimicrobial function in the mucus layer of corals ( Barr et al., 2013 ; Barr et al., 2015 ) and are often considered the main top-down control mechanism of bacteria in some systems. However in certain circumstances, Halobacteriovorax predation has been shown to be a more dominant factor in bacterial mortality than viral lysis ( Williams et al., 2015 ). In addition, predatory bacteria are thought to play a major role in controlling pathogenic Vibrio in seawater and shellfish ( Richards et al., 2012 ). In our study we show predatory Halobacteriovorax sp. PA1 is effective against V. coralliilyticus BAA 450 and other Vibrio strains, offering further support to the hypothesis that bacterial predators are likely to play a role in controlling populations in the environment. In a similar fashion to phages, Halobacteriovorax thus mediates top-down control of pathogens by preventing initial invasion of the host. Microbiome manipulation validates previous network analysis predictions A small but growing body of research suggests Halobacteriovorax naturally occur and regularly interact with members of the coral microbiome. For example, a previous metagenomic study of P. astreoides from Panama reported that sequences similar to predatory Halobacteriovorax were among the most commonly identified bacterial annotations in the coral microbiome ( Wegley et al., 2007 ). Furthermore, we found Halobacteriovorax was present in ∼80% of samples collected approximately monthly from three genera of Caribbean corals across a three-year time span. Network analysis of 198 of these samples detected intriguing co-occurrences between these predators and other taxa ( Welsh et al., 2016 ). Here in the bacterial challenge study, we validated several of the co-occurrence patterns detected in our network analysis. For example, in our networks from Agaricia corals, Bdellovibrionales (the order of Halobacteriovorax ) positively co-occurred with both Vibrionales and Cytophagales in the field ( Welsh et al., 2016 ). Here we experimentally demonstrated that Halobacteriovorax directly alters the abundance of both of these taxa. We show here that a V. coralliilyticus challenge is associated with significant increases in Cytophagales relative abundance in vivo as well, suggesting there is a more direct interaction between these two taxa ( Table 2 ). This work lends support to the use of networks to provide a predictive understanding of the microbiome’s function and dynamics in natural systems Caveats and considerations While this limited laboratory study suggests that Vibrio coralliilyticus alters the microbiome of coral tissue, whether these downstream responses of the microbiome are the result of wounding and challenge with any putative pathogen remains untested. At the same time, both the absolute numbers of V. coralliilyticus added in this experiment were not environmentally relevant and whether V. coralliilyticus is a pathogen of M. cavernosa is still unknown, shedding doubt on whether these same effects would be seen in true infection scenarios. Yet the patterns of change in the community after addition of V. coralliilyticus remarkably mirror the changes we found in corals exposed to stressors in the environment. For example, we found that overall beta-diversity and the relative abundance of opportunists such as Rhodobacterales increase in corals under threat ( Zaneveld et al., 2016 ). This striking similarity of these disparate works can be interpreted in two ways: (1) any stress to a system, whether it be a direct bacterial challenge or exposure to imperfect conditions on a reef, drive these changes in the microbiome or that (2) specific infection by groups of foreign bacteria are responsible for such affects. To distinguish these alternatives, more comprehensive experiments should be conducted such as those using a variety of host species and putative pathogens, additional commensals for controls, challenge trials that have different dilutions and more environmentally relevant concentrations of bacterial inoculate, and evaluations of the changes in the bacterial community in terms of finer taxonomic resolution as well as absolute abundances (as opposed to relative abundance) measures."
} | 5,436 |
37273623 | PMC10233687 | pmc | 190 | {
"abstract": "To combine the advantages\nof elastic and nonelastic triboelectric\nmaterials, this work proposes a new type of triboelectric nanogenerator\n(TENG) based on stacking —the stacked FKM/PU TENG. By stacking\nthe elastomer polyurethane (PU) and the nonelastomer fluororubber\n(FKM), the FKM/PU TENG combines the inherent triboelectric characteristics\nof both materials and the unique elasticity of PU to achieve an output\nperformance that is much higher than that of the FKM-TENG or the PU-TENG.\nThe maximum instantaneous open-circuit voltage and short-circuit current\nof the FKM/PU TENG reach 661 V and 71.2 μA, respectively. Under\nthe limiting conditions of 3 Hz and maximum compression, this device\ncan attain a maximum power density of 49.63 W/m 3 and light\nmore than 500 LEDs. Therefore, stacking materials with different properties\ngives the FKM/PU TENG high output performance and great application\npotential, which can contribute to future development of discrete\nmechanical energy harvesting.",
"introduction": "1 Introduction Fossil energy has always\nbeen the most commonly used energy in\nthe world, and it now faces a depletion crisis. However, abusing fossil\nfuels contaminates the environment and has an impact on human society\nthat cannot be ignored. 1 , 2 Therefore, it is important to\ndevelop green energy and build clean energy-based systems to secure\na better future. The triboelectric nanogenerator (TENG) was first\npresented by Zhonglin Wang at the Georgia Institute of Technology\nin 2012. Its principle is converting mechanical energy into electrical\nenergy through triboelectrification and electrostatic induction. 3 It has been shown to have the advantages of high\npower density, high efficiency, lightweight, low cost, and simple\nmanufacturing for collecting random and low-frequency mechanical energy. 4 The TENG uses the charge generated by the friction\nbetween two materials with different electronegativities to generate\na potential difference, thereby driving electrons to move in the external\ncircuit to form a current. TENGs can collect different kinds of mechanical\nenergy such as vibration, rotation, swing, human motion, 5 − 8 wave, 9 − 11 and wind energy. 12 − 14 There are mainly\nfour TENG working modes: vertical contact separation,\nsingle electrode, independent layer, and sliding plane. Different\nmodes are suitable for collecting mechanical energy in different forms\nand different motion environments, so TENGs are widely used in all\naspects of life and production. Vertical contact separation is a common\nTENG working mode when different materials are placed between the\ntwo plates. The TENG output performance increases when there is a\ngreater discrepancy in electron-capturing ability between the triboelectric\nlayers. 15 In 2020, Salauddin et al. proposed\nan MXene/Ecoflex nanocomposite with high negative electrical properties\nand mechanical stability. Made of nanocomposite and waterproof fabric,\nthe TENG can achieve a much better maximum power density than the\ncombination of PTFE and Nylon. 16 TENGs\nmade of rigid materials cannot harvest mechanical energy well often\nbecause of the unsatisfactory contact and long-term wear on their\nsurfaces. Therefore, researchers have tried to use elastic materials\nto fabricate TENGs. Liu et al. implanted conductive polyaniline nanowires\n(PANI NWs) on a polyurethane foam (PU) through simple dilute aniline\nchemical polymerization and constructed an ES-TENG. 17 Elastic deformation allows the TENG to respond effectively\nto random motions with different amplitudes and directions. However,\nelastic materials are triboelectrically weaker than rigid materials.\nFinding a method to integrate the characteristics of elastic and inelastic\ntriboelectric materials would advance the TENG to a new level. In this work, a stacked FKM/PU TENG is proposed and discussed,\nwhich is mainly composed of polytetrafluoroethylene (PTFE), fluororubber\n(FKM), PU, and conductive copper. The operation mode of contact separation\nis selected, and the stacked structure is divided into Plate A and\nPlate B. Plate A consists of PTFE and copper to form a common triboelectric\ngeneration unit. Plate B uses FKM as the direct triboelectric surface\nlayer, and the PU is the bottom. The copper electrode is set between\nthem. The PU is electropositive material and also involved in the\nelectron transfer between FKM and PTFE when Plate A contacts and squeezes\nPlate B. Under continuous extrusion, PTFE obtains more electrons from\nFKM and PU. Moreover, adding an elastomer to the bottom of Plate B\nenhances the extrusion and the duration. The increased PU deformation\nenhances the output of Plate B. Compared with the traditional design\nof triboelectric nanogenerator, the stacked FKM/PU TENG effectively\nimproves the adhesion and slows down the surface wear by using the\nstacking structure of different triboelectric materials. Besides,\nunder the coupling effect of PU and FKM, more induced charges are\ngenerated on the conductive layer, and the output performance of the\nTENG is significantly improved. The stacked FKM/PU TENG can adapt\nto different extrusion environments and bring different output effects.\nIn the maximum compression condition, the maximum open-circuit voltage,\nshort-circuit current, instantaneous output power, the amount of transferred\ncharge within 1 s and power density can reach 661 V, 71.2 μA,\n2.23 μC, 17.02 mW, and 49.62 W/m 3 , respectively.\nFurthermore, the TENG can easily light 540 commercial LEDs and support\ncalculators for simple operations. This improved method of TENG is\nmore economical, convenient and efficient without complicated material\nprocession or charge implantation. Therefore, the potency of this\nbrand new triboelectric nanogenerator in mechanical energy collection\ndeserves exploration.",
"discussion": "3 Discussion This work has proposed a new design\nof TENG based on stacking FKM\nand PU to form a stacked FKM/PU TENG. This TENG incorporates the merits\nof the PU and FKM into a single triboelectric material. Preparing\nthis TENG is simple and inexpensive. Powered by a reciprocating mechanism,\nthe stacked FKM/PU TENG generates a short-circuit current of up to\n71.2 μA and open-circuit voltage as high as 661 V. The degree\nof PU deformation changes as acrylic plates are added to change the\nextrusion environment, so the output voltage and total transferred\ncharge of the TENG gradually increase from 11.2 to 589.9 V and 0.53\nμC to 2.24 μC, respectively. Even after intense impact,\nthe friction layer consisting of PTFE is not significantly worn, and V OC remains above 500 V. Under a load of 40 MΩ,\nthe FKM/PU TENG achieves a maximum instantaneous output power of 17.02\nmW and power density of 49.62 W/m 3 . Within 60 s, the low-frequency\ndriven stacked FKM/PU TENG can achieve a voltage of 10.2 V across\na 4.7 μF capacitor. Such good output characteristics are attributed\nto not only the triboelectric characteristics of FKM and PTFE but\nalso the synergy of the elastomer and nonelastomer in the stacked\nstructure. This structure improves the contact, extrusion, and collision\neffects of the working TENG. Increasing the degree of extrusion increases\nthe amount of induced charge in the PU layer, which also improves\nthe performance of the stacked FKM/PU TENG. Thus, the stacked structure\nenables the TENG to have a better output than the FKM-TENG and PU-TENG\ncombined. This paper also carried out an actual loading test. The\nstacked FKM/PU TENG performs outstandingly in lighting 540 commercial\nLEDs and driving calculators for simple computations, demonstrating\na strong application potential. This TENG can help future development\nof energy collection and sensing technology."
} | 1,913 |
35548755 | PMC9085514 | pmc | 191 | {
"abstract": "Triboelectric nanogenerators (TENG) have been proven to be effective for the collection of low-frequency vibrational energy in the environment. However, most polymer materials as friction layers are highly susceptible to mechanical damage during operation, which reduces the performance and lifetime of TENG. Herein, we report a high-performance, flexible triboelectric nanogenerator with reproducible self-healing electronic characteristics. Based on its soft and flexible polymers, the self-healing triboelectric nanogenerator (SH-TENG) can achieve a peak power of 2.5 W m −2 and triboelectric charge density of about 100 μC m −2 . High-conductance Ag nanowires (AgNWs) are semi-embedded in the polymer to fabricate all-in-one friction layers and for an enhanced self-healing process. Both the output voltage and current of the healed device can reach up to about 99% of their original values even after five cutting/healing cycles. The fabricated SH-TENG has excellent stability and flexibility, which presents a significant step towards the fabrication of reliable triboelectric nanogenerators with recoverability and low maintenance costs.",
"conclusion": "Conclusions In summary, we fabricated a high-performance, flexible triboelectric nanogenerator with reproducible self-healing electronic characteristics by embedding electrically and thermally conductive AgNWs in TPU and PDMS films. Based on the soft and flexible polymers, the SH-TENG achieved a peak power of 2.5 W m −2 and triboelectric charge density of 100 μC m −2 . By integrating AgNWs in the polymer layers, these films could effectively and rapidly restore from the breakage caused by cuts and scratches under NIR light irradiation. Also, repetitive cutting/healing processes at the same region were demonstrated without significant decline in the electrical performance of the SH-TENG. Furthermore, both output voltage and current of the healed device reached up to about 99% of their original values even after five cutting/healing cycles. These films can act as promising self-healing materials and are believed to have potential applications in flexible electronic devices with prolonged durability and reliability.",
"introduction": "Introduction With the increasing energy demand worldwide, finding new techniques for energy harvesting is the key to meeting human requirements for the development of modern society. 1,2 The abundant mechanical energy in the environment such as wind energy, hydroelectric energy and ocean energy has aroused people's attention due to its wide distribution and low cost. 3,4 In recent years, various mechanical energy harvesting technologies have been developed, which convert energy in the environment into usable energy. 5,6 Currently, based on triboelectrification and electrostatic induction, the TENG reported by Wang et al. 7 have proven to be effective for collecting low-frequency vibrational energy, 8 which is considered as a potential sustainable power source for future developments. 9,10 The significant progress in TENG has opened a new chapter of energy harvesting technology in the ultimate challenge of miniaturization, and significant advances in the basic principles and applications of TENG have been achieved. 11,12 Due to many desirable features of TENG, such as simple manufacturing process, 13,14 flexible material choice, 15 high output voltage 16 and ability to operate under very low frequency and irregular mechanical motions, 17,18 it has been used as a power source for many self-powered devices, which have the advantages of high efficiency, convenience, environmental friendliness and low cost. 19,20 Recent research on TENG has mainly focused on conventional electrodes and friction materials. However, in the collection process of mechanical energy, TENG has long been subjected to mechanical impact. These materials are susceptible to fracture and damage, which lead to a sharp decrease in their sustainability, safety, and lifetime. 20,21 These effects inevitably lead to material failure, reducing the performance and decreasing the service life of the TENG. Thus, the application scope of TENG is greatly limited. Several solutions to this problem have been proposed to reduce the damage during the operation to improve the cycle life of the TENGs, which indeed decreased mechanical failure through a non-contact or slight-contact mode. 22,23 However, soft-contact is an effective way to enhance the potential performance of the TENGs. Furthermore, with rapid development of materials science, flexible electronics and nanotechnology, self-powered intelligent sensor devices have mushroomed, which require the development of self-healing polymeric materials for TENGs to improve their safety, lifetime, energy efficiency, and environmental impact. 24 Thus, the exploration of self-healing electrodes and polymeric friction materials is in high demand for the design of new types of TENG based on completely flexible and healable materials, which can recover after mechanical damage. 21,25 Herein, a high-performance flexible triboelectric nanogenerator with reproducible self-healing electronic characteristics is fabricated with high stability and recoverability. The top and bottom electrodes are prepared by semi-embedding the silver nanowires into a polydimethylsiloxane (PDMS) film and thermoplastic polyurethane (TPU) film respectively, which are used to fabricate a contact-separation mode TENG. 26 In the contact-separation mode, the triboelectric nanogenerator can achieve a peak power of 2.5 W m −2 and the triboelectric charge density of about 100 μC m −2 due to its soft and flexible polymers. 27 When this self-healing triboelectric nanogenerator (SH-TENG) is cut off or scratched, PDMS and TPU with AgNWs show the ability of self-healing simply by linking the broken ends under near infrared (NIR) irradiation and recover the electrical properties. The healing process of the TPU is attributed to hot melting and that of PDMS is the result of hydrogen bonds. 28 In addition, the electrode is prepared by embedding AgNWs in the polymer, which prevents the AgNWs from falling off due to touching or bending and maintain good conductivity and stability after several healing cycles. Compared with the performance of the original sample, the open circuit voltage, short circuit current and transferred charge of the healing sample are very close. Furthermore, the film exhibits stable electrical properties after multiple cycles of mending the cracks in the same region via simple thermal treatment. The fabricated SH-TENG is believed to have potential applications to power electronic devices, which require enhanced durability and reliability. The results indicate that our fabricated films can act as a promising self-healing material and have great applications in electronics and industries.",
"discussion": "Results and discussion Preparation of self-healing PDMS and TPU films An AgNW-embedded TPU (AgNWs/TPU) film and an AgNW-embedded PDMS (AgNWs/PDMS) film were fabricated, as schematically illustrated in Fig. 1 . We chose two different suitable substrates to form the uniform AgNWs semi-inlaid polymer films. 29 The AgNW solution was casted onto the substrate to obtain the AgNW electrode layer, as shown in Fig. 1a . Subsequently, the TPU solution and liquid PDMS were coated on the AgNW film, as shown in Fig. 1b and c , respectively. Then, the TPU and PDMS films were cured in an air oven, as shown in Fig. 1d and e . The details of the fabrication method are described in the Experimental section. Attributed to the permeation of the liquid precursor into the AgNW network, 30 the AgNWs were semi-embedded tightly in the cured TPU and PDMS films. 31,32 The AgNWs/TPU and AgNWs/PDMS films were obtained by peeling them off the substrates, as shown in Fig. 1f and g , respectively, which reflects their good flexibility. Also, the AgNW-embedded surface was conductive and the opposite surface was nonconductive due to the thickness of the polymer layer. The resistances of the conductive surface were 1.4 Ω sq −1 and 13.9 Ω sq −1 for the TPU film and the PDMS film, respectively. Furthermore, the conductivity of the films would not decrease due to touching and bending motions. 33 As shown in Fig. 1h and i , the AgNW-embedded TPU and PDMS films were characterized by scanning electron microscopy (SEM), respectively, which shows that the embedded structure was formed uniformly over the entire region and the AgNWs embedded into the TPU and PDMS formed a network structure. Fig. 1 Schematic illustration of the fabrication process of the AgNW-embedded TPU film and PDMS film. (a) Casting AgNW solution on a substrate. (b, d and f) Illustrations of the fabrication of the AgNW-embedded TPU layer by casting TPU solution on the AgNW film. (c, e and g) AgNW-embedded PDMS layer prepared by the bar-coating process. (h) SEM image of the AgNW-embedded surface of the TPU film. (i) SEM image of the AgNW-embedded surface of the PDMS film. The cutting/healing process of the self-healing PDMS and TPU films A schematic illustration of the self-healing process is shown in Fig. 2 . The primary samples were placed onto a glass substrate, and the AgNW-embedded surface directly absorbed NIR light as the top side. 34 Subsequently, the AgNWs/TPU and AgNWs/PDMS films were cut across their entire width with a scalpel ( Fig. 2a ). Right after contacting the ends of the fracture as close as possible, the samples were positioned under a conventional NIR lamp, of which the power density delivered to the samples was ∼0.63 W cm −2 , which allowed self-healing to occur ( Fig. 2b ). To improve the healing efficiency of PDMS, we added 3 wt% graphite powder (GP) during the moulding process of the PDMS film, which gave the optimal experimental result for healing through the absorption of NIR light (Fig. S2 † ). 34,35 GP-PDMS composites were fabricated through simple mechanical mixing. The GP-PDMS film was also cut into strips using a scalpel and positioned under a conventional NIR lamp. Fig. 2 Schematic illustration of the cutting/healing process. (a) Cutting the AgNWs/TPU or AgNWs/PDMS films. (b) Self-healing process induced by contacting the ends of the fracture and exposure to NIR light irradiation. (c) Schematic illustration of the healed AgNWs/TPU or AgNWs/PDMS film. (d) Resistance changes across the fracture area after five cycles of the cutting/healing process for TPU film and PDMS film. (e and f) SEM images of the AgNW layer of the AgNW/TPU film and the AgNW/PDMS film after healing, respectively. Characterization of self-healing PDMS and TPU films The self-healing result is due to the characteristics of effective absorption and conduction of NIR light into thermal energy by GP and AgNWs. 36 It should be noted that the AgNWs acted as an electrical conductor to initiate the healing of the PDMS and TPU films, which consequently contributed to the electrical properties of the AgNW layer. 37 The TPU melted at the broken interface and bonded again to heal the fracture above the melting point T m of the polymer under the NIR lamp. For PDMS, the self-healing process was the result of the interaction among the abundant hydrogen bonds from the Si–OH at the end of the chains, which were demonstrated by FTIR in Fig. S5. † The high density of hydrogen bonding sites gave the elastomer a partial self-healing capability at the broken interface. 28,38 Due to the energy from the NIR lamp, the PDMS reconnected through the hydrogen bonds at the fracture of the samples. 39,40 For the performance of the SH-TENG, the electric healing conductivity of its electrode is also an important reference index, in addition to ensuring the friction layer contact area. As shown in Fig. 2d , after five cycles of the cutting/healing process, the resistance ( R ) across the fracture area only increased from 1.4 to 11 Ω sq −1 and 13.9 to 30 Ω sq −1 for the TPU and PDMS films, respectively. The SEM images in Fig. 2e and f show that the separated AgNW layer was closely connected after the cutting/healing of the TPU and PDMS films (Fig. S6 and S7 † ). To test the self-healing properties of the conductive films, the NIR light-induced healing of the AgNWs/TPU and AgNWs/PDMS films was examined by integrating the films into a circuit with a light-emitting diode (LED) and a constant DC power supply of 2.0 V, 34,41 as shown in Fig. 3a and d , respectively. When the AgNWs/TPU and AgNWs/PDMS films were cut across their entire width with a scalpel, the LED immediately turned off ( Fig. 3b and e , respectively). The cut resulted in a break in the electrical circuit, which broke the connection between the AgNWs and the bottom TPU and PDMS layer. This indicates that the AgNWs/TPU film and the AgNWs/PDMS film incurred severe damage. Interestingly, the LED became bright again after the cut was exposed to NIR light ( Fig. 3c and f ), which induced the healing process in the film under NIR light irradiation. After repetitive folding of the healed AgNWs/TPU film and AgNWs/PDMS film, the cut was not distinguishable by the naked eye by a difference in brightness, which shows that the conductivity was restored by the NIR light irradiation. Fig. 3 Photographs of the AgNW/TPU film and the AgNW/PDMS film connected to a circuit with an LED bulb. (a–c) As-prepared AgNW/TPU film in the circuit before and after healing. (d–f) As-prepared AgNW/PDMS film in the circuit before and after healing. Scale bars, 10 mm. A time-controlled experiment illustrated the healing process in Fig. 4 , which shows the optical microscopy images of the healing samples under NIR irradiation. The scratched TPU film is shown in Fig. 4a . Then, the ends of the fracture were contacted as close as possible ( Fig. 4b ). Under NIR light irradiation, the length and width of the scratches gradually decreased, and the region was healed to about 70% of the original area in 15 min ( Fig. 4c ). The optical microscopy images show that some scratches in the samples had completely disappeared (according to visual observation) with 30 min NIR light irradiation ( Fig. 4d ). The healing of TPU is a result of the thermal effect of the NIR lamp, but excessive irradiation could cause the embedded AgNWs to submerge in the TPU, which increases the resistance of the conductive side of the TPU film. The other scratches on the PDMS film are displayed in Fig. 4e . After 30 min of exposure to NIR light, the scratches completely healed, which was observed under the microscope ( Fig. 4h ). The results show that the healing process is induced by NIR light. With the driving force of thermal energy on the molecular mobility, the TPU film heated above the T m melted to heal and PDMS reconnected the hydrogen bonds. 42,43 However, healing was only possible when the crosslinks were photo-induced instead of just being heated, especially for PDMS, which has a higher melting point. 37,44 The healing process of the films demonstrates the recombination of the chemical bonds in the cross-linking networks in response to temperature and light stimuli. 45,46 Fig. 4 The healing process of an AgNW/TPU film and an AgNW/PDMS film. (a) Scratching TPU film. (b) Contacting the ends of the fracture. (c) Under the NIR light irradiation, the scratches gradually decreased. (d) The scratches fully healed. (e) Scratching PDMS film. (f) Contacting the ends of the fracture. (g) The scratches gradually decreased under the NIR light irradiation. (h) The scratches fully healed. Scale bars, 1 mm. Performance characterization of the self-healing TENG The electrical performance of the fabricated SH-TENG was carefully tested, and the triboelectric series of different materials was studied (Fig. S9 and S10 † ). 47 As shown in Fig. 5a , the TENG was fabricated by integrating the embedded AgNWs in a dielectric layer of TPU and PDMS with the whole thickness of ∼220 μm. TPU was utilized as the triboelectrically positive material, and PDMS was utilized as the triboelectrically negative material. The images of the fabricated parts of the SH-TENG are displayed in Fig. 5b and c . A copper wire was connected to the AgNW electrode using transparent tape for electrical measurements. Finally, the TPU and PDMS contacted face to face to form the SH-TENG. The operation principle of the SH-TENG is based on the contact-separation mode, 48,49 which is schematically depicted in Fig. 5d . When a mechanical force is applied to SH-TENG, charges are produced and transferred from the surface of the TPU layer to the PDMS layer, generating equivalent negative and positive charges on both surfaces. When separated by releasing the external force, electrons are transferred from the PDMS layer to the TPU layer due to their different potentials. This electric current stops flowing when it reaches electrostatic equilibrium. Therefore, AC electricity can be continuously generated by the periodical contact-separation between the insulating layer and bottom electrode. 50 Fig. 5 Structure and performance of the TENG for harvesting mechanical energy. (a) Schematic diagram of the experiment setup. (b and c) Photographs of the TPU and PDMS films. (d) Working principle of the TENG. (e–h) Performance characterization of the TENG. (e) Open-circuit voltage, (f) short-circuit current, (g) transferred charge number and (h) dependence of the current and instantaneous power on the external resistance load. (The triboelectric surface between TPU and PDMS is 2 cm × 2 cm.) Detailed electrical characterizations were carried out to investigate the output performance of the SH-TENG. The open-circuit voltage and short-circuit current of the SH-TENG were measured using a Keithley 6514 electrometer. Unless otherwise specified, the sizes of the SH-TENG used in this work were all 2 cm × 2 cm. Also, a V oc of 120 V and I sc of 11.6 μA were obtained by applying a stable driving force ( Fig. 5e and f ), respectively. The compressive force was approximately 5 N. The peak value of I sc corresponded to the process of pressing, and the positive voltage was generated because of the immediate charge separation. The voltage was maintained at a plateau until the subsequent pressing. Here, the amount of transferred charges ( Q ) is about 40 nC ( Fig. 5g ) and the triboelectric charge density is 100 μC m −2 . In fact, the effective output power of the TENG depends on the external load. Fig. 5h shows the dependence of the current on the external load resistance, which varies from 1 kΩ to 10 GΩ. It is clearly seen that as the resistance increased, the current amplitude decreased. The instantaneous power relied on the load of the TENG, as shown in Fig. 5h . This was calculated using expression 1. 1 P = I 2 R The instantaneous power exhibits a maximum value of about 1 mW at a resistance load of 60 MΩ, and the corresponding power density of the TENG is about 2.5 W m −2 . During the process of harvesting mechanical energy, TENG have to be constantly operated for a long term under massive and complex mechanical impacts, which causes breakdown and performance degradation in the device. However, owing to the self-healing functions of the polymer materials and AgNW electrode, the as-prepared TENG is recoverable. Most importantly, the excellent self-healing capability of the as-prepared TENG is observed in its output performance. We evaluate three primary parameters of SH-TENG ( Fig. 6a, b and 6c ), including I sc , V oc and Q , on the self-healing performance of the device, respectively. For one typical sample tested under a certain mechanical force, the results show V oc of 120 V, I sc of 11.6 μA and Q of 40 nC can be generated by the original device. Although both the applied impulse and the device dimension remain unchanged when the SH-TENG was cut into two fragments, the V oc , I sc and Q decreased to about 70 V, 6 μA and 20 nC, respectively. Such an obvious change in output performance may be attributed to the variation in the effective contact area between TPU and PDMS before and after cutting. Once the broken ends connected for self-healing, the V oc , I sc and Q were restored and approached the initial performance, which achieved the high healing efficiency of 99%. To further evaluate the repeatability of the healing process, we measured the electric output of the as-fabricated TENG after 5 cutting/healing cycles. The results show that the healed device still effectively worked with a negligible decrease in its performance ( Fig. 6d ), indicating the great recoverability of SH-TENG for converting mechanical energy into electricity with a prolonged lifespan, durability and robustness. In this regard, the self-healing and flexible SH-TENG possesses distinct advantages for practical applications. The SH-TENG shows the potential to adapt to complex mechanical stimuli when serving as an energy harvester and self-powered sensor for various motions. Fig. 6 Self-healing process illustrated by the electric output performance of the SH-TENG (a–c). Open-circuit voltage (a) short-circuit current (b) and transferred charges (c) of the self-healing TENG in the original, broken, and healed states. (d) Electric stability of the SH-TENG when subjected to repetitive cutting/healing cycles."
} | 5,317 |
38572083 | PMC10987894 | pmc | 192 | {
"abstract": "Energy recovery from low-strength wastewater through anaerobic methanogenesis is constrained by limited substrate availability. The development of efficient methanogenic communities is critical but challenging. Here we develop a strategy to acclimate methanogenic communities using conductive carrier (CC), electrical stress (ES), and Acid Orange 7 (AO7) in a modified biofilter. The synergistic integration of CC, ES, and AO7 precipitated a remarkable 72-fold surge in methane production rate compared to the baseline. This increase was attributed to an altered methanogenic community function, independent of the continuous presence of AO7 and ES. AO7 acted as an external electron acceptor, accelerating acetogenesis from fermentation intermediates, restructuring the bacterial community, and enriching electroactive bacteria (EAB). Meanwhile, CC and ES orchestrated the assembly of the archaeal community and promoted electrotrophic methanogens, enhancing acetotrophic methanogenesis electron flow via a mechanism distinct from direct electrochemical interactions. The collective application of CC, ES, and AO7 effectively mitigated electron flow impediments in low-strength wastewater methanogenesis, achieving an additional 34% electron recovery from the substrate. This study proposes a new method of amending anaerobic digestion systems with conductive materials to advance wastewater treatment, sustainability, and energy self-sufficiency.",
"conclusion": "4 Conclusion Simultaneously applying CC, ES, and AO7 in a biofilm system generated a synergistic effect on functional community domestication and methanogenesis. AO7 enhanced the acetogenesis of fermentation intermediates by acting as an exogenous electron acceptor and reshaped the bacterial community structure by enriching electroactive bacteria. CC and ES, in contrast, regulated the archaeal community assembly and promoted electrotrophic methanogen proliferation. The combined action of these three factors synergistically cleared the blockage in electron flow, constructing a more efficient metabolic network in low-strength wastewater methanogenesis.",
"introduction": "1 Introduction The objectives of wastewater treatment have evolved beyond environmental pollution control to include resource and energy recovery, driven by the global energy crisis and climate change concerns [ 1 , 2 ]. While anaerobic methanogenesis is a preferred method for harnessing the energy potential of wastewater, its engineering in the treatment of low-strength wastewater, characterized by a chemical oxygen demand (COD) of less than 1000 mg L −1 , has always been challenged by insufficient productivity and high process sensitivity [ 3 ]. Anaerobic digestion (AD) typically encompasses four stages: hydrolysis, fermentation, acetogenesis, and methanogenesis. Achieving high biogas productivity necessitates the formation of efficient methanogenic communities with unobstructed electron transfer pathways. However, under low-strength conditions, electron flow during AD, especially in acetogenesis and methanogenesis, is often impeded by the kinetic limitation of low substrate levels [ 4 ]. The metabolism rate of acetoclastic methanogens (AM) is generally very low, approximately a quarter of that of fermentative bacteria (FB), and further decreases when acetate is insufficient (with a half-saturation rate constant of 150 mg L −1 for AM) [ 5 ]. Additionally, in the absence of an adequate population of methanogens as hydrogen scavengers, the anaerobic oxidation of fermentation intermediates (mainly volatile fatty acids) to acetate (i.e., acetogenesis) becomes thermodynamically unfavorable [ 6 , 7 ]. This results in the trapping of electrons in fermentation products and exacerbates the starvation of methanogens. Therefore, it is imperative to enhance energy recovery from low-strength wastewater by employing appropriate technical means to regulate community development and overcome blockages in the electron transfer pathway during acetogenesis and methanogenesis. Historically, methods for directly regulating the growth and metabolism of methanogens have been somewhat limited. However, a recent breakthrough has revealed that the introduction of conductive materials (CM) into AD can expedite methanogenesis by inducing a direct interspecies electron transfer (DIET) access between electroactive methanogens and electroactive bacteria (EAB) [ [8] , [9] , [10] ], which is faster than interspecies electron exchange via diffusive electron carriers (such as H 2 and formate) [ 11 , 12 ]. CM amendment enriches methanogens, accelerates system startup, and promotes systemic efficiency and stability [ [13] , [14] , [15] ]. In most cases, CMs were introduced into AD by dosing particle materials, but this approach faced practical challenges such as high dosage and loss, difficult separation, and potential environmental risks [ 11 ]. Given the substantial capacity and rapid processing characteristic of low-strength wastewater treatment [ 16 ], employing fixed conductive carriers (CC), whose enhancing effects have also been confirmed [ 17 , 18 ], is more economically and operationally feasible than dosing particle materials [ 11 ]. Furthermore, drawing from the experiences in bioelectrochemical systems, the application of electrical stress (ES) based on conductive carriers, which here also serve as electrodes, can be beneficial to methanogenesis by supplying additional electrons or habitats with suitable redox potential [ [19] , [20] , [21] ]. A widely overlooked issue is that a well-developed EAB population is necessary (but insufficient) to intensify DIET-type methanogenesis [ 22 ]. Currently, efforts in promoting methanogenesis through regulating EAB metabolism are limited. EAB are phylogenetically diverse and functionally versatile (such as electricity generation, metal ion or organic pollutant reduction, etc.), and they are characterized by the ability of extracellular electron transfer through cytochromes, conductive pili, or redox-active shuttles [ 19 , 23 ]. Interestingly, observations from studies on pollutant remediation indicated that EAB could be easily enriched in systems with exogenous electrophiles serving as electron acceptors, such as sulfate [ 7 ], nitrate [ 24 ], ferric iron [ 25 ], and even certain organic pollutants [ 26 , 27 ]. This presents an opportunity to enhance EAB development and metabolism by using exogenous electron acceptors as a special domestication method. Furthermore, the role of electron acceptors in promoting the degradation of organic acids has been confirmed by recent studies [ [28] , [29] , [30] , [31] ], which potentially alleviates the blockage in acetogenesis in low-strength environments. Acid orange 7 (AO7) is a typical azo dye with electrophilic azo bond [ 32 ] and low toxicity to methanogenesis [ 33 ], and it has been reported to effectively accelerate the decomposition of volatile organic acids (especially propionate) [ 29 ]. Therefore, it was employed in this study as a model electron acceptor, aiming to enrich EAB and accelerate acetogenesis. This study proposed a novel strategy combining CC, ES, and AO7 to enhance methanogenesis in low-strength wastewater. Six modified anaerobic biofilters (as simplified pattern biofilm systems) treating artificial wastewater (COD ∼350 mg L −1 ) were started up under different conditions (with nonconductive or conductive carriers, with or without ES, with or without AO7). The efficacy of this strategy was assessed by evaluating methanogenic performance in the presence of ES/AO7 and subsequent removal of these factors start-up. The mechanism influencing community assembly and electron transfer was disclosed by bacterial and archaeal sequencing, electron flux balance analysis, and thermodynamic analysis. The findings contribute to an expanded understanding of electrical communication within the AD community and hold valuable insights for improving low-strength wastewater treatment processes.",
"discussion": "3 Results and discussion 3.1 Methanogenic performance Reactors started up under different conditions exhibited notably distinct methanogenic efficiencies ( Fig. 1 ). With the organic loading rate (OLR) increasing from 1.2 to 1.6 kg COD m −3 d −1 (correspondingly HRT from 8 to 6 h), the methane yield of each group significantly improved, indicating that methanogenesis was constrained by the low substrate level. This suppression was most pronounced in traditional AD (R1), with a methane production rate of less than 0.002 m 3 m −3 d −1 and a methane yield of less than 3.1 mL per g COD. Fig. 1 Average methane production rate and yield under a hydraulic retention time (HRT) of 8 and 6 h. Fig. 1 All three environmental factors (CC, ES, and AO7) demonstrated a certain promoting effect on methanogenesis ( P < 0.05). Specifically, compared to R1, AO7 alone (R2) led to a 2–3-fold increase in methanogenic productivity, while CC (R3) resulted in a 14.7–27.1-fold promotion under an HRT of 6 h. For the systems with CC, applying ES (R5) further slightly accelerated methanogenesis with up to a 33.0-fold increase compared to R1, while AO7 (R4) produced contradictory effects under two hydraulic conditions. Remarkably, the simultaneous application of CC, AO7, and ES (R6) caused a substantial boost in methanogenesis, surpassing the cumulative effects of each factor. Under an HRT of 8 h, the methane production rate and methane yield of R6 were 0.065 m 3 m −3 d −1 and 64.3 mL per g COD, respectively, 63.5 and 72.0 times higher than those of R1. These indexes increased to 0.127 m 3 m −3 d −1 and 116.9 mL per g COD under an HRT of 6 h. This outstanding enhancement highlights the synergistic stimulating effect of three factors on methanogenesis. In this study, to elucidate the regulatory role of CC, ES, and AO7, only extremely low biomass (less than 1% of the final biomass) was initially inoculated into the system. Biofilms proliferated and colonized spontaneously under unfavorable conditions, including constant low strength, high-rate operation, and non-optimal temperature. In addition, dissolved methane was not counted, which is estimated to account for approximately 40% of the total methane, according to Henry's law [ 38 ]. Despite the suboptimal overall performance due to these constraints, the current results fully demonstrate the effectiveness of the reinforcement strategy. This insight is enlightening for upgrading various biofilm-based methanogenic engineering of low-strength wastewater. 3.2 Performance without ES & AO7 To discern the direct contribution of ES and AO7 to methanogenesis, control tests removing ES and AO7 after regular operation were performed ( Fig. 2 ). Without ES and AO7, the decline in methanogenic performance in each group was minimal or even insignificant ( Table S3 ). This suggests that, after sufficient domestication, the methanogenic performance became independent of the continuous supply of ES and AO7. This implies that the effect of ES and AO7 was attributed to the enhancement of biofilms’ metabolic function rather than direct contributions to methanogenesis. Fig. 2 Methanogenic efficiency in control tests removing electrical stress (ES) and acid orange 7 (AO7). R stands for regular operation, R′ stands for control tests without ES and AO7. Fig. 2 Despite the remarkable promoting effect, adding AO7 (typically considered an organic pollutant) presented a potential risk of secondary pollution. Encouragingly, the results suggest that ES and AO7 can only be applied temporarily during the startup or adjustment phases. This reduces the safety risk associated with AO7 use, making the strengthening strategy more practical and economically feasible. 3.3 Methanogenic activity and archaeal content of biomass The biomass activity in the AD system is generally evaluated by specific methanogenic activity tests. However, considering the mediation of CC, the isolated biofilms' methanogenic activity may not accurately represent their in situ state. Therefore, in this experiment, an apparent methanogenic activity (AMA) was calculated based on the total biomass in the bioreactor and the in situ methanogenic performance in control tests without ES and AO7 ( Table 1 ). The total biomass was promoted by CC ( P < 0.05), likely attributed to its surface properties. ES did not significantly affect total biomass ( P > 0.05). AO7, however, reduced the wet and dry weight of total biomass ( P < 0.05) but did not impact protein amount ( P > 0.05). This discrepancy may be due to AO7's influence on forming or degrading intracellular reserve polymers, such as glycogen and trehalose, commonly found in sugar-fed dynamic AD systems [ 39 ]. Table 1 Total biomass, apparent methanogenic activity (AMA), and archaeal content in the biofilm of each group. Table 1 Group Total biomass (g) Daily methane production a (L d −1 ) Apparent methanogenic activity c ( L CH 4 per d per g biomass) Archaeal content d (copies per g biomass) Wet weight Dry weight Protein R1 12.09 ± 0.18 2.07 ± 0.52 0.07 ± 0.01 0.002 0.03 1.6 × 10 5 R2 10.33 ± 0.18 2.50 ± 0.06 0.08 ± 0.01 0.007 b 0.09 2.1 × 10 5 R3 25.11 ± 0.15 7.14 ± 0.13 0.18 ± 0.02 0.054 0.30 3.6 × 10 5 R4 10.85 ± 0.41 1.19 ± 0.28 0.12 ± 0.01 0.038 b 0.32 3.7 × 10 5 R5 24.57 ± 1.30 6.34 ± 0.77 0.17 ± 0.02 0.059 b 0.35 4.7 × 10 5 R6 10.99 ± 0.39 1.41 ± 0.06 0.16 ± 0.02 0.117 b 0.73 7.1 × 10 5 a Average methane production (dissolved methane not included) under an HRT of 6 h. b Data in control tests without ES and AO7. c Total protein amount used as total biomass in the calculation. d Analyzed by real-time qPCR with general archaeal prime. All factors (CC, ES, and AO7) positively influenced AMA, with R6 showcasing superior performance compared to other controls. The archaeal content in unit biomass mirrored the same trend ( Table 1 ) and exhibited a positive linear correlation with AMA ( Fig. S3 ). According to the archaeal sequencing results, over 90% of archaea were identified as methanogens. Consequently, the archaeal content in biomass can be approximated as the content of methanogens. Both AMA and archaeal content directly demonstrate that, under the mediation of CC, the synergistic application of ES and AO7 significantly promoted the proliferation of methanogens and enhanced the methanogenic function of the microbial community. 3.4 Electron flux analysis Based on the COD balance of all components, electron fluxes in three stages (fermentation, acetogenesis, and methanogenesis) were estimated for both R1 (conventional AD) and R6 (with ES and AO7 removed) ( Fig. 3 a and b). Inevitably, a significant portion of electrons (40–60%) was consumed for microbial growth and stored as intracellular reserve polymers. The insufficient decomposition of reserve polymers under short HRTs led to some wastage of electrons, while ES and AO7 can alleviate this problem. Fig. 3 Estimated electron flux in anaerobic digestion in R1 ( a ) and R6 ( b ). Fig. 3 In a conventional AD system (R1), there was almost no electron flow between metabolites after the initial electron distribution during the fermentation stage. Electrons in most fermentation products (except hydrogen) were directly lost to the effluent. This indicates that only fermentative biochemical reactions were sufficient, but the electronic pathways of subsequent acetogenesis and acetotrophic methanogenesis were severely obstructed, and the metabolic functions of related microbial populations were not fully realized. In R6, electrons originating from fermentation intermediates to acetate and hydrogen increased from 1% to 19%, and a substantial portion of electrons in acetate (71%) flowed to methane. This indicates that the modulation of CC, ES, and AO7 stimulated the electronic communication between microbial populations, scavenged the blockage of electron flow in acetogenesis and acetotrophic methanogenesis, and reduced the number of wasted electrons in reserve polymers and effluent. 3.5 Bacterial community assembly In an AD system, the electron communication between metabolites relies on the metabolism of various microbial populations, so understanding the microbial community's structure is crucial. Given the small initial biomass (less than 1% of the final biomass) and the extended culture time (over 120 days) in this experiment, it is reasonable to assume the microbial community assembly, particularly the evolution of functional populations, was dominated by the deterministic effect of environmental factors rather than stochastic effect. The community's responses to each factor can be elucidated through statistical analysis. Bray-Curtis analysis ( Fig. 4 a) highlighted a significant difference between groups without AO7 (R1, R3, and R5) and those with AO7 (R2, R4, and R6) in bacterial community structures. This indicates that AO7, rather than CC or ES, played a decisive role in bacterial community assembly. AO7 notably increased bacterial community diversity ( Fig. 4 b) and influenced the distribution of metabolic groups ( Fig. 4 c). Despite variations in genera composition (dominant genera abundance summarized in Table S4 ), the abundance of each metabolic population was quite similar in AO7-free (or AO7-adding) groups. Fig. 4 a , Bray-Curtis analysis of bacterial community structure. b , Shannon rarefaction plot. c , Metabolic group abundance and genera distribution. NC stands for nonconductive carrier, and CC stands for conductive carrier. Fig. 4 FB are fast-growing flora responsible for fermenting glucose or glycolytic products to various acids, CO 2 , and H 2 , mainly including lactic acid bacterium Lactococcus [ 40 ], acetate and propionate producer Alkaliflexus [ 41 ], Propionivibrio [ 42 ], and so on. FB was the most abundant in all groups, accounting for 74.5 ± 0.6% in groups without AO7 and 41.5 ± 2.2% in groups with AO7. In typical AD systems, the conversion of fermentation intermediates to acetate and hydrogen mainly relies on syntrophic hydrogen-producing acetogens (HPA) [ 43 , 44 ]. As the symbiotic partner, HPA's evolution and metabolism are regulated by methanogenesis and are difficult to develop in low-strength wastewater [ 6 , 7 ]. In this experiment, only three acetogenic genera were detected with extremely low abundances (1.0 ± 0.4% without AO7 and 3.2 ± 0.3% with AO7), including two HPA ( Candidatus Cloacamonas and Syntrophomonas ) [ [45] , [46] , [47] ] and one homoacetogen ( Acetobacterium ), which produces acetate from CO 2 and H 2 [ 48 ]. The scarcity of acetogens explained the blocked electron flow during acetogenesis in R1. AO7 significantly facilitated the enrichment of various EAB (from 7.1 ± 1.0% to 23.7 ± 3.2%). This group included genera such as Aeromonas , Geobacter , Enterococcus , and Raoultella , as well as serval sulfate-reducing bacteria ( Desulfovibrio , Desulfuromonas , Desulfitibacter , and Desulfobulbus ). Most genera have azoreductases, allowing them to utilize AO7 as an electron acceptor in anaerobic respiration [ [49] , [50] , [51] ]. Additionally, with the mediation of conductive materials, methanogens can act as electron acceptors for some EAB in DIET-type methanogenesis [ 12 , 52 ]. EAB are metabolically versatile, utilizing a variety of intermediates as electron and energy donors, which are oxidized either completely to CO 2 or partially to smaller organic compounds [ [53] , [54] , [55] ]. Interestingly, it is observed that conductive materials and electrodes with low external voltage are not as effective as soluble electron acceptors (like AO7) in enriching EAB (2% vs. 15% increase). This disparity can be attributed to the challenges associated with electron transfer. Compared to the well-diffused and cell-penetrable soluble acceptors, the electron transfer between extracellular solid mediators and EAB is more difficult and generally limited by spatial distance [ 12 , 56 ]. Another noteworthy observation is that the enrichment of EAB does not necessarily lead to improved methanogenic efficiency, as seen in the case of R4. The ecological relationship between EAB and methanogens (whether cooperative or competitive) mainly depends on environmental factors [ 57 ]. Even under the mediation of conductive materials, DIET-type methanogenic syntrophs may not be fully established. In general, in the absence of AO7, the bacterial community was “monopolized” by FB, while other populations faced challenges in development. In particular, the scarcity of acetogens led to a blocked electron flux during acetogenesis. The introduction of AO7 was crucial in reshaping the bacterial community, promoting diversity, and enriching EAB. This enrichment of EAB potentially contributes more electrons for methanogens, either through DIET with the mediation of CC or by facilitating the partially oxidizing intermediates to acetate. 3.6 Archaeal composition Different from the bacterial community, according to archaeal sequencing results ( Fig. 5 a), the archaeal community structure was mainly determined by CC and ES rather than AO7. The genera distribution further reveals distinct patterns ( Fig. 5 b). In groups with NC, like R1 and R2, Methanobrevibacter , a hydrogenotrophic methanogen (HM) [ 58 ], dominated with a relative abundance ranging from 36.4% to 76.2%. On the other hand, groups with CC (R3 to R6) were characterized by the dominance of another HM, Methanobacterium , accounting for 46.1–66.1% of the community. Methanobacterium is known for its ability to accept electrons from electrodes or syntrophic partners [ 59 ]. Fig. 5 a , Bray-Curtis analysis of archaeal community structure. b , Archaeal distribution at the genus level. Fig. 5 In a well-functioning AD system, AM typically dominates the methanogenic community, constituting about 70% of the population. However, due to its faster growth rate, HM tends to prevail over AM in low-strength environments. In R1, the content of AM was notably low, accounting for only 1.6%. The application of ES induced a significant increase in AM content. Methanosarcina , a typical AM that could participate in DIET-type methanogenesis [ 60 ], was specifically enriched (4.4% in R5 and 27.2% in R6). This enrichment well explained the promoted electron flux in acetophilic methanogenesis in R6. The response of Methanosarcina to ES and AO7 and its important role in promoting AMA were also confirmed by canonical correlation analysis between characteristic genera, environmental factors, and AMA ( Fig. S4 ). 3.7 The synergistic mechanism of AO7 and ES As previously mentioned, AO7 and ES profoundly affected the microbial community's composition and function, mainly attributed to their intervention on microbial metabolism in the biofilm formation (start-up or adjustment) stage. In anaerobic conditions, the –N \n \n\n<svg xmlns=\"http://www.w3.org/2000/svg\" version=\"1.0\" width=\"20.666667pt\" height=\"16.000000pt\" viewBox=\"0 0 20.666667 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(0.019444,-0.019444)\" fill=\"currentColor\" stroke=\"none\"><path d=\"M0 440 l0 -40 480 0 480 0 0 40 0 40 -480 0 -480 0 0 -40z M0 280 l0 -40 480 0 480 0 0 40 0 40 -480 0 -480 0 0 -40z\"/></g></svg>\n\n N- bond in AO7 can be biologically (in R2, R4, and R6) or electrochemically (in R6, less than 60% of the total reduction) reduced, resulting in the production and two reduction products, SA and 1-amino-2-naphthol (AN) (equation (1)) [ 33 ]. Considerable AO7 decolorization efficiencies ranging from 85% to 98% and high product recovery efficiencies in 84%–92% were observed in all groups (details in Table S5 ), suggesting a sufficient reduction of AO7. The reduction products of AO7 can be fully mineralized under aerobic conditions [ 61 ]. Therefore, the environmental risks associated with introducing AO7 can likely be mitigated by following an appropriate oxidation operation. Image 1 The respiration of anaerobic flora exhibits great flexibility, and various redox pairs with potential differences can generate electron flow in their respiratory chain and provide energy for their growth. In typical AD without exogenous electron acceptors, the acetogenesis of fermentation intermediates, such as lactate, butyrate, propionate, and ethanol, is generally accompanied by H + reduction, which is thermodynamically feasible only at very low hydrogen partial pressures. In the start-up stage, hydrogen consumers are usually lacking due to the slow growth of methanogens, especially in low-strength wastewater, leading to hydrogen inhibition on acetogenesis. This issue can be addressed by the addition of AO7. Fig. 6 a shows the reduction potentials of redox couples of some typical fermentation intermediates at a hydrogen partial pressure of 5 hPa. In the absence of AO7, only ethanol and lactate can be oxidized by H + , while the acetogenesis of butyrate and propionate cannot be carried out. Adding AO7 increases the potential of electron acceptors in the AD system, making the oxidation of various intermediates thermodynamically feasible. The specific reactions that take place depend on the metabolic properties of the bacteria present. For instance, among the EAB detected in this study, Desulfovibrio mainly produces acetate, which accumulates as the end product of the oxidation of lactate, malate, pyruvate, and other organics [ 62 ]. Desulfuromonas generally conducts complete acetate oxidation, and some species can also use propionate and lactate [ 53 , 54 ]. Desulfobulbus can propionate oxidizing [ 55 ], and Geobacter prefers acetate and ethanol [ 12 ]. Importantly, using AO7 as an acceptor is advantageous in providing metabolic energy because it can increase the reaction electromotive force. Therefore, bacteria with AO7-reducing ability generally have a higher growth rate and are easier to enrich in the system. It is worth noting that the role of AO7 may not be unique, and other substances with high reduction potentials (or wastewater containing electrophilic pollutants) may have similar functions. Beware that AO7, with a higher reduction potential than the redox couples Acetate/CH 4 and HCO 3 − /CH 4 (about −0.24 V), may compete with methanogens for acetate and hydrogen, which could explain the contradictory effects of AO7 on methanogenesis in R4. Similar issues arise with other electron acceptors like sulfate and iron ions [ 7 ]. Theoretically, whether electron acceptors competitively inhibit methanogenesis depends on various factors, such as the type of electron donors, the concentration of electron acceptors, and their degradation dynamics [ 63 , 64 ]. Their concentration should be carefully considered when applying exogenous electron acceptors for methanogenesis promotion. Using them as a short-term regulation rather than a continuous condition may be more beneficial. In many cases where AD was coupled with bioelectrochemical systems, the improvement in methanogenesis is mainly attributed to the direct contribution of various electrochemical reactions, such as hydrogen evolution, methane electrosynthesis, and anode respiration [ 20 , 65 ]. However, in this study, the situation was quite different. Despite the current generation by ES in R5 and R6, the low electrochemical efficiency (seen in Table S6 ) indicates that the promoting effect obtained was not primarily due to the direct contribution of electrochemical processes. The average currents in R5 and R6 were less than 1.5 mA (corresponding to a current density of less than 1.5 A per m 3 of reactor volume), and anode coulombic efficiencies were less than 1%. Moreover, the theoretical maximum contributions of circuit current to methanogenesis (assuming all electrons in the circuit are involved in methanogenesis) were less than 3%. It suggests that the electrochemical process’ direct contribution to organic matter degradation and methanogenesis was negligible. This observation explains why methanogenic efficiency was not significantly reduced after removing ES. Referring to the analyses of the archaeal community and electron flux, it can be inferred that ES promoted the electron flow in acetotrophic methanogenesis in a manner that excluded the current's direct contribution. This effect was likely due to alternations in the electrochemical state of the carrier surface. Without ES, the carrier potential was about −0.25 V (vs. SHE), and with ES, the potential at the cathode was about −0.5 V (vs. SHE). The low-potential carrier surface may influence the electrochemical gradient of attached methanogens for basic cellular functions, including chemo-osmotic transport and ATP synthesis [ 66 , 67 ]. Alternatively, the presence of an electric field force could cause a directional guiding effect on the originally disordered electron transfer on the carrier and make it easier for methanogens to establish electrical connections with EAB. Overall, the combination of CC, ES, and AO7 synergistically affected the assembly and metabolism of the anaerobic community in low-strength wastewater ( Fig. 6 b). AO7, acting as an exogenous electron acceptor, enriched EAB and promoted the degradation of organic acids. With the mediation of CC and the stimulation of ES, the electrical communication between EAB and methanogens is enhanced. This facilitated the establishment of a metabolic network involving EAB respiration and DIET methanogenesis, overcoming the blockage of electron flow and resulting in more efficient recovery of electrons from the substrate. Fig. 6 a , The possible redox reactions and their estimated electromotive force in the system with and without acid orange 7 (AO7). b , Hypothetical influencing mechanism of the conductive carrier (CC), electrical stress (ES), and AO7 on anaerobic digestion metabolic network. A1: intermediates, acetate, and hydrogen oxidized by electroactive bacteria; A2: electroactive bacteria takes AO7 as an electron acceptor; A3: direct interspecies electron transfer (DIET) between electroactive bacteria and methanogens under the mediation of CC and the stimulation of ES. Fig. 6"
} | 7,545 |
26927849 | PMC4771773 | pmc | 195 | {
"abstract": "Quorum sensing is a process of chemical communication that bacteria use to monitor cell density and coordinate cooperative behaviors. Quorum sensing relies on extracellular signal molecules and cognate receptor pairs. While a single quorum-sensing system is sufficient to probe cell density, bacteria frequently use multiple quorum-sensing systems to regulate the same cooperative behaviors. The potential benefits of these redundant network structures are not clear. Here, we combine modeling and experimental analyses of the Bacillus subtilis and Vibrio harveyi quorum-sensing networks to show that accumulation of multiple quorum-sensing systems may be driven by a facultative cheating mechanism. We demonstrate that a strain that has acquired an additional quorum-sensing system can exploit its ancestor that possesses one fewer system, but nonetheless, resume full cooperation with its kin when it is fixed in the population. We identify the molecular network design criteria required for this advantage. Our results suggest that increased complexity in bacterial social signaling circuits can evolve without providing an adaptive advantage in a clonal population.",
"introduction": "Introduction Quorum sensing is a mechanism of bacterial cell—cell communication that relies on the production, release, and group-wide detection of extracellular signal molecules called autoinducers. Quorum sensing enables populations of bacteria to coordinate changes in gene expression [ 1 , 2 ]. Bacteria often use quorum sensing to orchestrate the release of public goods (e.g., enzymes or surfactants) whose functions benefit the entire community [ 2 ], and to direct other cooperative behaviors such as transitions to more efficient modes of growth [ 3 ]. The cooperative nature of quorum sensing is susceptible to exploitation by mutant genotypes that do not contribute to cooperation but benefit from it [ 2 , 4 – 6 ]. Despite their immediate advantage over the wild-type, exploiting “cheater” genotypes will be eliminated in structured populations due to their negative effect on the average fitness of the community [ 5 , 7 – 11 ]. In bacteria, population structure can naturally arise in biofilms, where bacteria can grow without significant mixing [ 10 ], or during the formation of growth bottlenecks upon invasion into a new environment [ 11 – 14 ]. Many bacterial species employ multiple quorum-sensing systems that impinge on the activity of a shared transcriptional regulator. Each of the quorum-sensing systems encodes a specific receptor and autoinducer production gene with no or limited crosstalk [ 15 ]. In several species such as B . subtilis [ 16 ], V . harveyi [ 17 ], and its pathogenic relative, V . cholerae [ 18 , 19 ], the quorum-sensing systems are arranged in a parallel, seemingly redundant, architecture. That is, all the quorum-sensing autoinducer receptors funnel information into the same signal transduction pathway. It is unclear what the adaptive benefit is of harboring multiple, rather than a single, quorum-sensing autoinducer—receptor pair when the pairs function in parallel. Here, we combine modeling and experiments in B . subtilis and V . harveyi to show that a strain that has accumulated an additional quorum-sensing system reduces its cooperative investment in the presence of its ancestor, but resumes full cooperation in a clonal population. We show that this facultative cheating strategy requires a specific system integration design criterion; the novel receptor must have a dominant repressive effect on the ancestral quorum-sensing response in the absence of the novel autoinducer. We show that, additionally, this particular network design often leads to synergistic activation of the quorum-sensing response by the different autoinducers.",
"discussion": "Discussion In this work, we propose that bacteria possessing multiple quorum-sensing networks that control the identical response, which are commonly found in nature, are selected through a facultative cheating process. Facultative cheating has been described in the past as a strategy by which microorganisms exploit nonkin but return to cooperation in the presence of kin [ 32 ]. Such behavior has been described in fruiting body-forming amoeba and bacteria [ 29 , 33 ], but the underlying molecular processes that lead to it are unknown [ 34 ]; however, links to cell—cell signaling and facultative cheating have been suggested [ 35 – 37 ]. We predict that accumulation of multiple quorum-sensing systems requires a specific set of network design criteria, the functioning of which we explored in two diverse but well studied organisms. Specifically, the introduced novel receptor must repress the quorum-sensing response in the absence of the novel autoinducer, as occurs in the B . subtilis Rap-Phr and V . harveyi Lux quorum-sensing systems. In contrast to these systems that depend on repression, other quorum-sensing systems exist that act positively, in that the receptor functions as an activator upon autoinducer binding. Our model and experimental results explain why accumulation of parallel positively acting systems is selected against. Indeed, we do not know of any bacterium that possesses multiple activation-based quorum-sensing systems that function in parallel. Rather, activation-based systems are commonly organized in a hierarchy, in which one quorum-sensing system regulates the expression of a second system. A hierarchical network design is not fully redundant because the two quorum-sensing systems can control different genes. Further work will be required to define the benefits and possible evolutionary routes giving rise to quorum-sensing systems that function positively and are arranged as hierarchies. Beyond facultative cheating, other possible adaptive functions for possessing multiple quorum-sensing systems have been suggested. These include gains in information acquired about cell density, information about the frequency of phenotypes in the vicinal population, and access to information about physical flow conditions [ 38 – 41 ]. Our social selection model does not contradict those alternatives and may promote them by driving the initial fixation of the redundant network design, which can, subsequently, be further modified for other adaptive advantages. Several processes may limit the accumulation of quorum-sensing systems. First, each system contributes a signaling cost [ 5 , 42 ]. Second, the facultative return to cooperation may not be complete, leading to reduced benefit during exploitation in structured populations. Third, social selection of facultative characters is weak and can lead to variable mutation selection balance [ 34 ]. Finally, rareness of available systems and the need to integrate them appropriately into the existing network may limit the rate of accumulation. Further work will be required to define the importance of each of these mechanisms. Exploitation can also occur between species; not only between variants within species. For example, cooperative secretion of antibiotic degrading enzymes has been shown to lead to coexistence of secreting and nonsecreting genotypes, at both the species and interspecific levels [ 43 , 44 ]. Accumulation of additional quorum-sensing systems could also be used to exploit species that produce fewer signals. This ecological factor may contribute to the continuous selection for maintenance of multiple systems. More generally, our results point to the roles facultative cheating and kin recognition may have in the ecology of complex microbial communities."
} | 1,892 |
33684253 | null | s2 | 196 | {
"abstract": "The potential to convert renewable plant biomasses into fuels and chemicals by microbial processes presents an attractive, less environmentally intense alternative to conventional routes based on fossil fuels. This would best be done with microbes that natively deconstruct lignocellulose and concomitantly form industrially relevant products, but these two physiological and metabolic features are rarely and simultaneously observed in nature. Genetic modification of both plant feedstocks and microbes can be used to increase lignocellulose deconstruction capability and generate industrially relevant products. Separate efforts on plants and microbes are ongoing, but these studies lack a focus on optimal, complementary combinations of these disparate biological systems to obtain a convergent technology. Improving genetic tools for plants have given rise to the generation of low-lignin lines that are more readily solubilized by microorganisms. Most focus on the microbiological front has involved thermophilic bacteria from the genera Caldicellulosiruptor and Clostridium, given their capacity to degrade lignocellulose and to form bio-products through metabolic engineering strategies enabled by ever-improving molecular genetics tools. Bioengineering plant properties to better fit the deconstruction capabilities of candidate consolidated bioprocessing microorganisms has potential to achieve the efficient lignocellulose deconstruction needed for industrial relevance."
} | 370 |
36033861 | PMC9399729 | pmc | 197 | {
"abstract": "Bacterial biofilms are ubiquitous in natural environments and play an essential role in bacteria’s environmental adaptability. Quorum sensing (QS), as the main signaling mechanism bacteria used for cell-to-cell communication, plays a key role in bacterial biofilm formation. However, little is known about the role of QS circuit in the N-transformation type strain, Paracoccus denitrificans , especially for the regulatory protein PdeR. In this study, we found the overexpression of pdeR promoted bacterial aggregation and biofilm formation. Through RNA-seq analysis, we demonstrated that PdeR is a global regulator which could regulate 656 genes expression, involved in multiple metabolic pathways. Combined with transcriptome as well as biochemical experiments, we found the overexpressed pdeR mainly promoted the intracellular degradation of amino acids and fatty acids, as well as siderophore biosynthesis and transportation, thus providing cells enough energy and iron for biofilm development. These results revealed the underlying mechanism for PdeR in biofilm formation of P. denitrificans , adding to our understanding of QS regulation in biofilm development.",
"introduction": "Introduction Biofilms are one of the most prevalent and important forms of life for bacteria, in which cells are encased in the extracellular matrix that can serve as a barrier to multiple adverse environmental factors ( Steenackers et al., 2016 ; Jo et al., 2022 ). Previous studies showed that the environmental stresses that microbes would face can be highly variable and complex, including mechanical damage, antibiotics pressures, oxidative stresses, etc. ( Starosta et al., 2014 ). Thus, over 80% of bacteria would form a biofilm to resist environmental stresses and to make sure the whole community survives better ( Mohammad Reza, 2018 ). For example, under extreme conditions such as the deep sea, microorganisms would form microbial mats and produce more extracellular polymeric substances (EPS) to resist mechanical stress ( Bolhuis et al., 2014 ). Moreover, EPS could work as a protective barrier, which contributes to the lower sensitivity and higher resistance of biofilms to antibiotics ( Hoiby et al., 2010 ; Yan and Bassler, 2019 ), and biofilm also promotes the efficiency of resistance genes horizontal transfer ( Orazi and O’Toole, 2019 ). Typically, biofilm formation and maturation are regulated by various factors, including cell-to-cell communication and environmental factors. Quorum sensing (QS) is a widely used bacterial communication mechanism, by which bacteria could secret and sense signaling molecules called autoinducers to coordinate gene expression according to the population density ( Papenfort and Bassler, 2016 ; Mukherjee and Bassler, 2019 ). Bacteria use QS to precisely coordinate various of group behaviors, including biofilm formation, carbon metabolism, EPS production, virulence factors production, luminescence, etc. ( Davies et al., 1998 ; Hense and Schuster, 2015 ). Previous studies have shown that the mutation of QS system severely damaged the ability of biofilm formation, such as in Pseudomonas aeruginosa , Burkholderia cepacia , Streptococcus mutans , etc. ( Lewenza et al., 1999 ; Parsek and Greenberg, 2005 ). QS coordinates the initiation and maturation of biofilm by regulating the expression of a series of functional genes, including EPS production related genes, cell motility genes, and so on. Gilbert et al. demonstrated that the QS regulation protein LasR in P. aeruginosa could directly regulate the expressional level of the extracellular polysaccharide (Psl) biosynthetic genes ( Gilbert et al., 2009 ). Moreover, when the second QS circuit rhl was mutant, the production of another kind of polysaccharide Pel significantly decreased in P. aeruginosa ( Sakuragi and Kolter, 2007 ). In addition, QS could coordinate the expression of cell motility genes, such as IV pilus gene clusters which are associated with cell attachment during the early stage of biofilm formation, as well as the genes for flagella synthesis which are essential for biofilm matures as a mushroom-like structure ( Yu and Ma, 2017 ; Yan and Bassler, 2019 ). However, research on the mechanism of QS regulating biofilm formation is mostly concentrated in pathogenic microorganisms, such as P. aeruginosa , Staphylococcus aureus , etc., while studies on environmental bacteria are relatively rare. Paracoccus denitrificans is widely distributed in soil and water, possessing the ability of heterotrophic nitrification-aerobic denitrification (HNAD), and thus is taken as the model strain for nitrogen transformation research ( Ji et al., 2015 ). We have demonstrated that P. denitrificans harbored a LuxI/R-type QS circuit, PdeI/R ( Zhang et al., 2018 ). The AHL synthetase PdeI could catalyze the biosynthesis of N -hexadecanoyl-L-Homoserine lactone (C16-HSL), and PdeR protein is the corresponding regulatory protein which could bind with C16-HSL and regulate genes expression. It has been shown that P. denitrificans forms a peculiarly thin biofilm at the gas–liquid interface, which consisted of almost a monolayer of cells ( Yoshida et al., 2017 ). Biofilms for Paracoccus species have many important applications, especially in wastewater nitrogen removal bioreactors, while the detailed mechanism of biofilm formation in this genus is largely unknown ( Morinaga et al., 2020 ). Previous studies showed that in the pdeI mutant strain, cell aggregation was more obvious and bacteria formed thicker biofilm, while exogenous C16-HSL addition inhibited cell aggregation ( Morinaga et al., 2018 ). Nevertheless, the underlying mechanism of QS regulating biofilm formation in P. denitrificans still remains unclear, especially for the role of PdeR protein. In this study, we constructed the pdeR overexpression strain to explore the role of PdeR on biofilm formation of P. denitrificans PD1222 at the initiation stage. The results of physiological tests indicated overexpressed pdeR promoted cell aggregation and EPS production, and thus the pdeR overexpression strain formed thicker biofilm. Furthermore, through transcriptomic analysis as well as biochemical experiments, we demonstrated that PdeR promotes adenosine triphosphate (ATP) production and iron absorption during the initial stage of biofilm formation to provide sufficient energy and iron for biofilm development. The results of this study deepened our understanding of the QS regulation mechanism for biofilm formation of P. denitrificans , providing some useful references for the optimization of P. denitrificans in applications.",
"discussion": "Discussion Paracoccus denitrificans is widely spread in either natural or artificial environments, functioning as a driver of the nitrogen cycle ( Wang et al., 2021 ). To survive better, environmental bacteria usually exist in the form of biofilms. And it is well known that biofilm requires QS system regulation for its initiation and maturation (P. Stoodley et al., 2002 ). Previous studies showed P. denitrificans harbors a LuxI/R-type QS circuit, PdeI/R ( Zhang et al., 2018 ). However, little is known about the role and regulation mechanism of QS on biofilm formation of P. denitrificans PD1222. In this study, we constructed the pdeR overexpression strain and performed phenotypic as well as global transcriptomic studies to provide information on the function of PdeR in P. denitrificans PD1222. Results showed the pdeR overexpression strain PD-pdeR produced more EPS and formed thicker biofilm compared with the control strain PD-pBBR ( Figures 2 , 3 ). These results are in keeping with previous observational studies, for examples, the lasI mutant cells of P. aeruginosa could only form flat and undifferentiated biofilms ( Davies et al., 1998 ); and in Vibrio parahaemolyticus , the lack of QS regulator protein AphA reduced the EPS production and thus hampered cells’ ability of biofilm formation ( Wang et al., 2013 ). To explore the molecular mechanism of PdeR promoting biofilm formation, we performed transcriptome sequencing analysis. As the results showed, PdeR is a global regulation factor, it could directly or indirectly control the expression of multiple genes which are involved in various metabolic pathways ( Figures 4 , 5 ). However, exactly which genes are the key contributing to biofilm initiation needs to be further explored. In P. denitrificans , several molecules have been demonstrated with the ability to affect biofilm formation: that is, the adhesion protein BapA that initiates biofilm formation, the intracellular second messenger cyclic diguanosine monophosphate (Cyclic-di-GMP) and nitric oxide ( Kumar and Spiro, 2017 ; Yoshida et al., 2017 ). However, these related genes’ expressional levels were not affected by the overexpressed PdeR. Moreover, although three genes participating in O-Antigen nucleotide sugar biosynthesis pathway were up-regulated in PD-pdeR, the expressional level of other genes related to polysaccharide synthesis and flagella or pili assembly was not affected by PdeR much. Remarkably, we found that PdeR promoted a series of metabolic pathways participating in ATP production, including amino acids degradation pathway, fatty acid degradation pathway and oxidative phosphorylation pathway ( Figure 6 ). Similarly to our findings, Pisithkul et al. had also reported during biofilm development in Bacillus subtilis , several metabolic alterations which were hitherto unrecognized as biofilm-associated had been detected, such as the tricarboxylic acid (TCA) cycle, fatty acid biosynthesis and degradation, etc. ( Pisithkul et al., 2019 ). These results suggested bacteria tend to promote the degradation of organic substances and accelerate the production of energy during biofilm formation. This is understandable given the fact that biofilm formation needs a large number of extracellular polysaccharides, proteins and a wide variety of secondary metabolites ( Payne and Boles, 2016 ). The biosynthesis and transportation of these molecules are both energy-consuming, thus cells need to generate enough energy to support the synthesis and secretion of these essential substances ( Sutherland, 2001 ). Consistently, we indeed detected higher ATP concentration in PD-pdeR strain cells, indicating PdeR promoted ATP production and thus benefit bacterial biofilm formation. In addition to energy, siderophore, as the main tool bacteria used to obtain iron is also essential for bacterial growth as well as biofilm formation. Previous studies showed the growth rate of B. subtilis mutant strains that lack the function of siderophore biosynthesis was significantly inhibited ( Rizzi et al., 2019 ). Qin et al. demonstrated that the siderophore biosynthetic genes mutant cells of B. subtilis showed reduced biofilm formation, while the addition of exogenous iron chelator DHBA could restore mutant strains’ biofilm formation ( Qin et al., 2019 ). And in B. cepacia , the production of siderophore ornibactin was affected by the luxR homolog cepR ( Lewenza et al., 1999 ). Our results in this work are in accord with these previous studies indicating that the increased siderophore production and iron transportation is another key point of PdeR promoting bacterial biofilm formation. In conclusion, PdeR plays a key role in promoting P. denitrificans biofilm formation, mainly through accelerating ATP production and increasing iron transportation at the initiation period. These data brought information about detailed mechanisms that may at least in part explain how QS regulates biofilm formation of P. denitrificans , adding to our understanding of QS regulation in biofilm development."
} | 2,934 |
35755176 | PMC9230140 | pmc | 198 | {
"abstract": "In recent years, machine-learning techniques, particularly deep learning, have outperformed traditional time-series forecasting approaches in many contexts, including univariate and multivariate predictions. This study aims to investigate the capability of (i) gated recurrent neural networks, including long short-term memory (LSTM) and gated recurrent unit (GRU) networks, (ii) reservoir computing (RC) techniques, such as echo state networks (ESNs) and hybrid physics-informed ESNs, and (iii) the nonlinear vector autoregression (NVAR) approach, which has recently been introduced as the next generation RC, for the prediction of chaotic time series and to compare their performance in terms of accuracy, efficiency, and robustness. We apply the methods to predict time series obtained from two widely used chaotic benchmarks, the Mackey–Glass and Lorenz-63 models, as well as two other chaotic datasets representing a bursting neuron and the dynamics of the El Niño Southern Oscillation, and to one experimental dataset representing a time series of cardiac voltage with complex dynamics. We find that even though gated RNN techniques have been successful in forecasting time series generally, they can fall short in predicting chaotic time series for the methods, datasets, and ranges of hyperparameter values considered here. In contrast, for the chaotic datasets studied, we found that reservoir computing and NVAR techniques are more computationally efficient and offer more promise in long-term prediction of chaotic time series.",
"conclusion": "6. Conclusion In this paper, six different ML time-series forecasting approaches, including two gated RNN techniques, three variants of ESNs, and the NVAR approach, were tested to predict five chaotic time series, including the Mackey–Glass, Lorenz-63, bursting Morris–Lecar, Vallis ENSO, and experimental cardiac action potential time series. Although we considered relatively large but still limited numbers of datasets and methods, we found that the LSTM and GRU approaches, despite their high computational costs and in contrast to the ESN and NVAR methods, were incapable of forecasting the Mackey–Glass, Lorenz, and bursting Morris–Lecar time series more than a few steps into the future, and that increasing the network size did not significantly improve their performance. For the ENSO model, the GRU method could predict somewhat longer, but it did not compare favorably with the ESN and NVAR approaches. Three variants of ESNs were employed including the baseline ESN, the clustered ESN (CESN), and the hybrid physics-informed ESN (HESN). For the five datasets we used in this work, only one (ENSO) showed improvement by using a more complicated ESN architecture such as the clustered reservoir. In all the other cases, the baseline ESN demonstrated similar or better performance compared to CESN. In contrast, whereas the HESN provided the same level of prediction accuracy for the four synthetic time series, within the tested network sizes, it was the most successful approach for forecasting the experimental dataset, where it delivered more accurate predictions as measured by RMSE. Thus, incorporating the domain knowledge of a dynamical system if available may improve the prediction ability of the ESN technique and may help with obtaining good predictions using smaller network sizes. For the tested network sizes and datasets, the best prediction performance in the case of the Mackey–Glass, Lorenz, and bursting Morris–Lecar datasets was obtained by the NVAR method, which was recently introduced as the next generation of RC techniques and has been demonstrated to be as successful as optimized ESNs. For the ENSO dataset, NVAR’s prediction accuracy was only slightly lower than that of the most accurate method, CESN. A noticeable advantage of the NVAR technique over conventional ESNs is avoiding the explicit construction of randomly connected neurons and circumventing the intrinsic randomness that increases the sensitivity of the network to the hyperparameter values and initial parameters that remain untrained. Moreover, the number of hyperparameters is much smaller than for ESNs, which makes NVAR easier to tune. Such advantages may initially suggest that the amount of data required to train the NVAR model could be less than that needed for the conventional ESNs. However, our experiments showed in the case of the experimental cardiac voltage dataset, better performance was only obtained by embedding more delays and at the cost of more computational time and effort. Nevertheless, in general, this approach shows promise for efficient prediction of chaotic time series. To the best of our knowledge, this work is one of the first applications of this newly introduced technique to real-world experimental time series. Further studies in this area may reveal more of the potential of this approach. For instance, in this work, we used a quadratic polynomial functional to construct the nonlinear portion of the state vectors at each time step; however, other nonlinear functions such as higher-order polynomials could also be employed and studied. It should also be noted that our conclusions are based on a limited number of datasets and employed methods. Moreover, in each case, the optimum hyperparameters were obtained in a finite grid search process. Accordingly, it is possible that the same approaches could provide different results when applied to other datasets or when hyperparameters are determined using more extensive grids or different optimization techniques.",
"introduction": "1. Introduction Time series are important in many real-world applications, such as biology ( Bar-Joseph et al., 2003 ), finance ( Dingli & Fournier, 2017 ; Plagianakos & Tzanaki, 2001 ; Takahashi et al., 2019 ; Tsay, 2005 ; Zhao, 2009 ), climate science ( Ghil & Vautard, 1991 ), anomaly detection in computer networks ( Limthong, 2013 ) and social networks ( Gong et al., 2018 ), and energy ( Billinton et al., 1996 ; Bunn, 2000 ; Deihimi & Showkati, 2012 ). Accordingly, the analysis and prediction of time series data are of great importance and have been the focus of much research in the past few decades. In general, a time series represents a record of observations of a dynamical system at specific time intervals. Therefore, time series prediction involves determining the future evolution of a dynamical system, which can be especially challenging for chaotic dynamical systems. The states of such systems can be represented by chaotic time series, which are recognized by the orbital instability characteristic, where infinitesimal differences in the initial values bring about large differences in the time series behavior. Consequently, prediction of a chaotic time series is only feasible for a relatively short time before the appearance of orbital instability. For this reason, forecasting chaotic time series has remained a difficult task for the last few decades. Data-driven approaches, and machine-learning (ML) techniques in particular, have recently become the main approaches used for time-series forecasting ( Ahmed et al., 2010 ; Ben Taieb et al., 2012 ; Chandra et al., 2021 ; Chattopadhyay et al., 2020 ; Cheng et al., 2015 ; De Gooijer & Hyndman, 2006 ; Dubois et al., 2020 ; Kutz, 2013 ; Li et al., 2005 ; Tealab, 2018 ). In particular, recurrent neural networks (RNNs) are the mainstream architecture for analyzing sequential data, owing to their ability in interpreting temporal dependencies in the input time series ( Chandra et al., 2021 ; Elman, 1990 ; Elman & Zipser, 1988 ; Schmidhuber, 2015 ). The recurrent connections in such networks serve as a notion of memory, allowing them to embed temporal information. Despite the success of RNNs in modeling short-term temporal data and non-chaotic dynamical systems, the high computational cost of back-propagation through time and their vulnerability to the vanishing or exploding gradient problems have limited their applications. Gated RNN architectures were introduced to address some of these problems. More precisely, the memory cell architecture and the gating mechanism enable these networks to be more selective over the information that needs be remembered or forgotten, thereby enabling them to learn long-term dependencies in temporal sequences. Long short-term memory (LSTM) networks ( Hochreiter & Schmidhuber, 1997 ) and gated recurrent units (GRUs) ( Chung et al., 2014 ) are among the most widely used gated RNNs. An alternative approach to deal with time-series forecasting and modeling dynamical systems is reservoir computing (RC), a learning paradigm mostly implemented as echo state networks (ESNs) ( Jaeger, 2002 ; Lukoševičius & Jaeger, 2009 ; Sun et al., 2020 ). The RC paradigm is fundamentally derived from RNN concepts offering a streamlined training process, which remains limited to obtaining the output layer weights, while the rest of the parameter values are set randomly and remain untrained. Notwithstanding such a major simplification, ESNs have successfully been employed for multi-step-ahead prediction of nonlinear time series and modeling chaotic dynamical systems at low computational cost ( Bianchi et al., 2017 ; Han et al., 2021 ), triggering the development of several network topologies in recent years. For instance, clustered ESNs (CESNs) ( Deng & Zhang, 2006 ; Junior et al., 2020 ), where multiple sub-graphs of sparsely connected hidden units form the reservoir, and deep ESNs, where the reservoir consists of multiple sub-reservoir layers stacked hierarchically ( Gallicchio & Micheli, 2017 ; Gallicchio et al., 2017 ), are two widely used architectures. Hybrid ESNs (HESNs) are another category of RC techniques introduced in a physics-informed ML framework ( Oh, 2020 ; Willard et al., 2020 ), where additional inputs from physics-based mathematical models integrate corresponding domain knowledge into data-driven models ( Doan et al., 2019 ; Pathak, Hunt, et al., 2018 ). The successful application of ESNs, despite their random construction, in forecasting complex dynamical systems using time-series data triggered a series of recent research providing an interpretation of how RC techniques function. Recently, Bollt demonstrated how the RC with linear activation functions and linear readout layer shares similarities with the well-studied vector autoregressive (VAR) concept, while using a quadratic readout can be interpreted as nonlinear VAR (NVAR) ( Bollt, 2021 ). Later, Gauthier et al. further studied this similarity and introduced the next generation RC, where instead of explicitly generating a reservoir of randomly connected neurons, an NVAR machine is formed in which the feature vector consists of time-delayed observations of the dynamical system and is augmented by nonlinear functions of these observations. Accordingly, with this approach there are fewer hyperparameters to tune and the intrinsic random nature of ESNs is effectively avoided. This approach was employed for one-step-ahead forecasting of benchmark chaotic time series for both reconstruction and cross-prediction tasks ( Gauthier et al., 2021 ). In this work, we assess the capability of the mainstream gated RNN techniques; ESN architectures, including the clustered architecture and the physics-informed hybrid approach; and the NVAR approach for multi-step-ahead prediction of nonlinear time series describing chaotic dynamical systems. In particular, we compare the performance of these models for forecasting two frequently used benchmark chaotic time series, derived from the Mackey–Glass and Lorenz dynamical systems, two additional chaotic times series derived from a bursting Morris–Lecar neuron model and the Vallis El Niño Southern Oscillation (ENSO) system, and one real-world dataset consisting of a time series of irregular cardiac voltage traces obtained in ex-vivo experiments in terms of the prediction error and computational efficiency. Moreover, this experimental dataset is further used to evaluate the performance of NVAR against traditional RC approaches in more detail. This paper is structured as follows. Section 2 presents a summary of the modeling approaches used for forecasting chaotic time series in this research and provides details about the implementation of each model and the evaluation metrics employed in this study. These methods are applied to datasets whose characteristics are described in Section 3 . The results are presented and discussed in Section 4 , and Section 5 presents concluding remarks."
} | 3,135 |
35755176 | PMC9230140 | pmc | 198 | {
"abstract": "In recent years, machine-learning techniques, particularly deep learning, have outperformed traditional time-series forecasting approaches in many contexts, including univariate and multivariate predictions. This study aims to investigate the capability of (i) gated recurrent neural networks, including long short-term memory (LSTM) and gated recurrent unit (GRU) networks, (ii) reservoir computing (RC) techniques, such as echo state networks (ESNs) and hybrid physics-informed ESNs, and (iii) the nonlinear vector autoregression (NVAR) approach, which has recently been introduced as the next generation RC, for the prediction of chaotic time series and to compare their performance in terms of accuracy, efficiency, and robustness. We apply the methods to predict time series obtained from two widely used chaotic benchmarks, the Mackey–Glass and Lorenz-63 models, as well as two other chaotic datasets representing a bursting neuron and the dynamics of the El Niño Southern Oscillation, and to one experimental dataset representing a time series of cardiac voltage with complex dynamics. We find that even though gated RNN techniques have been successful in forecasting time series generally, they can fall short in predicting chaotic time series for the methods, datasets, and ranges of hyperparameter values considered here. In contrast, for the chaotic datasets studied, we found that reservoir computing and NVAR techniques are more computationally efficient and offer more promise in long-term prediction of chaotic time series.",
"conclusion": "6. Conclusion In this paper, six different ML time-series forecasting approaches, including two gated RNN techniques, three variants of ESNs, and the NVAR approach, were tested to predict five chaotic time series, including the Mackey–Glass, Lorenz-63, bursting Morris–Lecar, Vallis ENSO, and experimental cardiac action potential time series. Although we considered relatively large but still limited numbers of datasets and methods, we found that the LSTM and GRU approaches, despite their high computational costs and in contrast to the ESN and NVAR methods, were incapable of forecasting the Mackey–Glass, Lorenz, and bursting Morris–Lecar time series more than a few steps into the future, and that increasing the network size did not significantly improve their performance. For the ENSO model, the GRU method could predict somewhat longer, but it did not compare favorably with the ESN and NVAR approaches. Three variants of ESNs were employed including the baseline ESN, the clustered ESN (CESN), and the hybrid physics-informed ESN (HESN). For the five datasets we used in this work, only one (ENSO) showed improvement by using a more complicated ESN architecture such as the clustered reservoir. In all the other cases, the baseline ESN demonstrated similar or better performance compared to CESN. In contrast, whereas the HESN provided the same level of prediction accuracy for the four synthetic time series, within the tested network sizes, it was the most successful approach for forecasting the experimental dataset, where it delivered more accurate predictions as measured by RMSE. Thus, incorporating the domain knowledge of a dynamical system if available may improve the prediction ability of the ESN technique and may help with obtaining good predictions using smaller network sizes. For the tested network sizes and datasets, the best prediction performance in the case of the Mackey–Glass, Lorenz, and bursting Morris–Lecar datasets was obtained by the NVAR method, which was recently introduced as the next generation of RC techniques and has been demonstrated to be as successful as optimized ESNs. For the ENSO dataset, NVAR’s prediction accuracy was only slightly lower than that of the most accurate method, CESN. A noticeable advantage of the NVAR technique over conventional ESNs is avoiding the explicit construction of randomly connected neurons and circumventing the intrinsic randomness that increases the sensitivity of the network to the hyperparameter values and initial parameters that remain untrained. Moreover, the number of hyperparameters is much smaller than for ESNs, which makes NVAR easier to tune. Such advantages may initially suggest that the amount of data required to train the NVAR model could be less than that needed for the conventional ESNs. However, our experiments showed in the case of the experimental cardiac voltage dataset, better performance was only obtained by embedding more delays and at the cost of more computational time and effort. Nevertheless, in general, this approach shows promise for efficient prediction of chaotic time series. To the best of our knowledge, this work is one of the first applications of this newly introduced technique to real-world experimental time series. Further studies in this area may reveal more of the potential of this approach. For instance, in this work, we used a quadratic polynomial functional to construct the nonlinear portion of the state vectors at each time step; however, other nonlinear functions such as higher-order polynomials could also be employed and studied. It should also be noted that our conclusions are based on a limited number of datasets and employed methods. Moreover, in each case, the optimum hyperparameters were obtained in a finite grid search process. Accordingly, it is possible that the same approaches could provide different results when applied to other datasets or when hyperparameters are determined using more extensive grids or different optimization techniques.",
"introduction": "1. Introduction Time series are important in many real-world applications, such as biology ( Bar-Joseph et al., 2003 ), finance ( Dingli & Fournier, 2017 ; Plagianakos & Tzanaki, 2001 ; Takahashi et al., 2019 ; Tsay, 2005 ; Zhao, 2009 ), climate science ( Ghil & Vautard, 1991 ), anomaly detection in computer networks ( Limthong, 2013 ) and social networks ( Gong et al., 2018 ), and energy ( Billinton et al., 1996 ; Bunn, 2000 ; Deihimi & Showkati, 2012 ). Accordingly, the analysis and prediction of time series data are of great importance and have been the focus of much research in the past few decades. In general, a time series represents a record of observations of a dynamical system at specific time intervals. Therefore, time series prediction involves determining the future evolution of a dynamical system, which can be especially challenging for chaotic dynamical systems. The states of such systems can be represented by chaotic time series, which are recognized by the orbital instability characteristic, where infinitesimal differences in the initial values bring about large differences in the time series behavior. Consequently, prediction of a chaotic time series is only feasible for a relatively short time before the appearance of orbital instability. For this reason, forecasting chaotic time series has remained a difficult task for the last few decades. Data-driven approaches, and machine-learning (ML) techniques in particular, have recently become the main approaches used for time-series forecasting ( Ahmed et al., 2010 ; Ben Taieb et al., 2012 ; Chandra et al., 2021 ; Chattopadhyay et al., 2020 ; Cheng et al., 2015 ; De Gooijer & Hyndman, 2006 ; Dubois et al., 2020 ; Kutz, 2013 ; Li et al., 2005 ; Tealab, 2018 ). In particular, recurrent neural networks (RNNs) are the mainstream architecture for analyzing sequential data, owing to their ability in interpreting temporal dependencies in the input time series ( Chandra et al., 2021 ; Elman, 1990 ; Elman & Zipser, 1988 ; Schmidhuber, 2015 ). The recurrent connections in such networks serve as a notion of memory, allowing them to embed temporal information. Despite the success of RNNs in modeling short-term temporal data and non-chaotic dynamical systems, the high computational cost of back-propagation through time and their vulnerability to the vanishing or exploding gradient problems have limited their applications. Gated RNN architectures were introduced to address some of these problems. More precisely, the memory cell architecture and the gating mechanism enable these networks to be more selective over the information that needs be remembered or forgotten, thereby enabling them to learn long-term dependencies in temporal sequences. Long short-term memory (LSTM) networks ( Hochreiter & Schmidhuber, 1997 ) and gated recurrent units (GRUs) ( Chung et al., 2014 ) are among the most widely used gated RNNs. An alternative approach to deal with time-series forecasting and modeling dynamical systems is reservoir computing (RC), a learning paradigm mostly implemented as echo state networks (ESNs) ( Jaeger, 2002 ; Lukoševičius & Jaeger, 2009 ; Sun et al., 2020 ). The RC paradigm is fundamentally derived from RNN concepts offering a streamlined training process, which remains limited to obtaining the output layer weights, while the rest of the parameter values are set randomly and remain untrained. Notwithstanding such a major simplification, ESNs have successfully been employed for multi-step-ahead prediction of nonlinear time series and modeling chaotic dynamical systems at low computational cost ( Bianchi et al., 2017 ; Han et al., 2021 ), triggering the development of several network topologies in recent years. For instance, clustered ESNs (CESNs) ( Deng & Zhang, 2006 ; Junior et al., 2020 ), where multiple sub-graphs of sparsely connected hidden units form the reservoir, and deep ESNs, where the reservoir consists of multiple sub-reservoir layers stacked hierarchically ( Gallicchio & Micheli, 2017 ; Gallicchio et al., 2017 ), are two widely used architectures. Hybrid ESNs (HESNs) are another category of RC techniques introduced in a physics-informed ML framework ( Oh, 2020 ; Willard et al., 2020 ), where additional inputs from physics-based mathematical models integrate corresponding domain knowledge into data-driven models ( Doan et al., 2019 ; Pathak, Hunt, et al., 2018 ). The successful application of ESNs, despite their random construction, in forecasting complex dynamical systems using time-series data triggered a series of recent research providing an interpretation of how RC techniques function. Recently, Bollt demonstrated how the RC with linear activation functions and linear readout layer shares similarities with the well-studied vector autoregressive (VAR) concept, while using a quadratic readout can be interpreted as nonlinear VAR (NVAR) ( Bollt, 2021 ). Later, Gauthier et al. further studied this similarity and introduced the next generation RC, where instead of explicitly generating a reservoir of randomly connected neurons, an NVAR machine is formed in which the feature vector consists of time-delayed observations of the dynamical system and is augmented by nonlinear functions of these observations. Accordingly, with this approach there are fewer hyperparameters to tune and the intrinsic random nature of ESNs is effectively avoided. This approach was employed for one-step-ahead forecasting of benchmark chaotic time series for both reconstruction and cross-prediction tasks ( Gauthier et al., 2021 ). In this work, we assess the capability of the mainstream gated RNN techniques; ESN architectures, including the clustered architecture and the physics-informed hybrid approach; and the NVAR approach for multi-step-ahead prediction of nonlinear time series describing chaotic dynamical systems. In particular, we compare the performance of these models for forecasting two frequently used benchmark chaotic time series, derived from the Mackey–Glass and Lorenz dynamical systems, two additional chaotic times series derived from a bursting Morris–Lecar neuron model and the Vallis El Niño Southern Oscillation (ENSO) system, and one real-world dataset consisting of a time series of irregular cardiac voltage traces obtained in ex-vivo experiments in terms of the prediction error and computational efficiency. Moreover, this experimental dataset is further used to evaluate the performance of NVAR against traditional RC approaches in more detail. This paper is structured as follows. Section 2 presents a summary of the modeling approaches used for forecasting chaotic time series in this research and provides details about the implementation of each model and the evaluation metrics employed in this study. These methods are applied to datasets whose characteristics are described in Section 3 . The results are presented and discussed in Section 4 , and Section 5 presents concluding remarks."
} | 3,135 |
35755176 | PMC9230140 | pmc | 198 | {
"abstract": "In recent years, machine-learning techniques, particularly deep learning, have outperformed traditional time-series forecasting approaches in many contexts, including univariate and multivariate predictions. This study aims to investigate the capability of (i) gated recurrent neural networks, including long short-term memory (LSTM) and gated recurrent unit (GRU) networks, (ii) reservoir computing (RC) techniques, such as echo state networks (ESNs) and hybrid physics-informed ESNs, and (iii) the nonlinear vector autoregression (NVAR) approach, which has recently been introduced as the next generation RC, for the prediction of chaotic time series and to compare their performance in terms of accuracy, efficiency, and robustness. We apply the methods to predict time series obtained from two widely used chaotic benchmarks, the Mackey–Glass and Lorenz-63 models, as well as two other chaotic datasets representing a bursting neuron and the dynamics of the El Niño Southern Oscillation, and to one experimental dataset representing a time series of cardiac voltage with complex dynamics. We find that even though gated RNN techniques have been successful in forecasting time series generally, they can fall short in predicting chaotic time series for the methods, datasets, and ranges of hyperparameter values considered here. In contrast, for the chaotic datasets studied, we found that reservoir computing and NVAR techniques are more computationally efficient and offer more promise in long-term prediction of chaotic time series.",
"conclusion": "6. Conclusion In this paper, six different ML time-series forecasting approaches, including two gated RNN techniques, three variants of ESNs, and the NVAR approach, were tested to predict five chaotic time series, including the Mackey–Glass, Lorenz-63, bursting Morris–Lecar, Vallis ENSO, and experimental cardiac action potential time series. Although we considered relatively large but still limited numbers of datasets and methods, we found that the LSTM and GRU approaches, despite their high computational costs and in contrast to the ESN and NVAR methods, were incapable of forecasting the Mackey–Glass, Lorenz, and bursting Morris–Lecar time series more than a few steps into the future, and that increasing the network size did not significantly improve their performance. For the ENSO model, the GRU method could predict somewhat longer, but it did not compare favorably with the ESN and NVAR approaches. Three variants of ESNs were employed including the baseline ESN, the clustered ESN (CESN), and the hybrid physics-informed ESN (HESN). For the five datasets we used in this work, only one (ENSO) showed improvement by using a more complicated ESN architecture such as the clustered reservoir. In all the other cases, the baseline ESN demonstrated similar or better performance compared to CESN. In contrast, whereas the HESN provided the same level of prediction accuracy for the four synthetic time series, within the tested network sizes, it was the most successful approach for forecasting the experimental dataset, where it delivered more accurate predictions as measured by RMSE. Thus, incorporating the domain knowledge of a dynamical system if available may improve the prediction ability of the ESN technique and may help with obtaining good predictions using smaller network sizes. For the tested network sizes and datasets, the best prediction performance in the case of the Mackey–Glass, Lorenz, and bursting Morris–Lecar datasets was obtained by the NVAR method, which was recently introduced as the next generation of RC techniques and has been demonstrated to be as successful as optimized ESNs. For the ENSO dataset, NVAR’s prediction accuracy was only slightly lower than that of the most accurate method, CESN. A noticeable advantage of the NVAR technique over conventional ESNs is avoiding the explicit construction of randomly connected neurons and circumventing the intrinsic randomness that increases the sensitivity of the network to the hyperparameter values and initial parameters that remain untrained. Moreover, the number of hyperparameters is much smaller than for ESNs, which makes NVAR easier to tune. Such advantages may initially suggest that the amount of data required to train the NVAR model could be less than that needed for the conventional ESNs. However, our experiments showed in the case of the experimental cardiac voltage dataset, better performance was only obtained by embedding more delays and at the cost of more computational time and effort. Nevertheless, in general, this approach shows promise for efficient prediction of chaotic time series. To the best of our knowledge, this work is one of the first applications of this newly introduced technique to real-world experimental time series. Further studies in this area may reveal more of the potential of this approach. For instance, in this work, we used a quadratic polynomial functional to construct the nonlinear portion of the state vectors at each time step; however, other nonlinear functions such as higher-order polynomials could also be employed and studied. It should also be noted that our conclusions are based on a limited number of datasets and employed methods. Moreover, in each case, the optimum hyperparameters were obtained in a finite grid search process. Accordingly, it is possible that the same approaches could provide different results when applied to other datasets or when hyperparameters are determined using more extensive grids or different optimization techniques.",
"introduction": "1. Introduction Time series are important in many real-world applications, such as biology ( Bar-Joseph et al., 2003 ), finance ( Dingli & Fournier, 2017 ; Plagianakos & Tzanaki, 2001 ; Takahashi et al., 2019 ; Tsay, 2005 ; Zhao, 2009 ), climate science ( Ghil & Vautard, 1991 ), anomaly detection in computer networks ( Limthong, 2013 ) and social networks ( Gong et al., 2018 ), and energy ( Billinton et al., 1996 ; Bunn, 2000 ; Deihimi & Showkati, 2012 ). Accordingly, the analysis and prediction of time series data are of great importance and have been the focus of much research in the past few decades. In general, a time series represents a record of observations of a dynamical system at specific time intervals. Therefore, time series prediction involves determining the future evolution of a dynamical system, which can be especially challenging for chaotic dynamical systems. The states of such systems can be represented by chaotic time series, which are recognized by the orbital instability characteristic, where infinitesimal differences in the initial values bring about large differences in the time series behavior. Consequently, prediction of a chaotic time series is only feasible for a relatively short time before the appearance of orbital instability. For this reason, forecasting chaotic time series has remained a difficult task for the last few decades. Data-driven approaches, and machine-learning (ML) techniques in particular, have recently become the main approaches used for time-series forecasting ( Ahmed et al., 2010 ; Ben Taieb et al., 2012 ; Chandra et al., 2021 ; Chattopadhyay et al., 2020 ; Cheng et al., 2015 ; De Gooijer & Hyndman, 2006 ; Dubois et al., 2020 ; Kutz, 2013 ; Li et al., 2005 ; Tealab, 2018 ). In particular, recurrent neural networks (RNNs) are the mainstream architecture for analyzing sequential data, owing to their ability in interpreting temporal dependencies in the input time series ( Chandra et al., 2021 ; Elman, 1990 ; Elman & Zipser, 1988 ; Schmidhuber, 2015 ). The recurrent connections in such networks serve as a notion of memory, allowing them to embed temporal information. Despite the success of RNNs in modeling short-term temporal data and non-chaotic dynamical systems, the high computational cost of back-propagation through time and their vulnerability to the vanishing or exploding gradient problems have limited their applications. Gated RNN architectures were introduced to address some of these problems. More precisely, the memory cell architecture and the gating mechanism enable these networks to be more selective over the information that needs be remembered or forgotten, thereby enabling them to learn long-term dependencies in temporal sequences. Long short-term memory (LSTM) networks ( Hochreiter & Schmidhuber, 1997 ) and gated recurrent units (GRUs) ( Chung et al., 2014 ) are among the most widely used gated RNNs. An alternative approach to deal with time-series forecasting and modeling dynamical systems is reservoir computing (RC), a learning paradigm mostly implemented as echo state networks (ESNs) ( Jaeger, 2002 ; Lukoševičius & Jaeger, 2009 ; Sun et al., 2020 ). The RC paradigm is fundamentally derived from RNN concepts offering a streamlined training process, which remains limited to obtaining the output layer weights, while the rest of the parameter values are set randomly and remain untrained. Notwithstanding such a major simplification, ESNs have successfully been employed for multi-step-ahead prediction of nonlinear time series and modeling chaotic dynamical systems at low computational cost ( Bianchi et al., 2017 ; Han et al., 2021 ), triggering the development of several network topologies in recent years. For instance, clustered ESNs (CESNs) ( Deng & Zhang, 2006 ; Junior et al., 2020 ), where multiple sub-graphs of sparsely connected hidden units form the reservoir, and deep ESNs, where the reservoir consists of multiple sub-reservoir layers stacked hierarchically ( Gallicchio & Micheli, 2017 ; Gallicchio et al., 2017 ), are two widely used architectures. Hybrid ESNs (HESNs) are another category of RC techniques introduced in a physics-informed ML framework ( Oh, 2020 ; Willard et al., 2020 ), where additional inputs from physics-based mathematical models integrate corresponding domain knowledge into data-driven models ( Doan et al., 2019 ; Pathak, Hunt, et al., 2018 ). The successful application of ESNs, despite their random construction, in forecasting complex dynamical systems using time-series data triggered a series of recent research providing an interpretation of how RC techniques function. Recently, Bollt demonstrated how the RC with linear activation functions and linear readout layer shares similarities with the well-studied vector autoregressive (VAR) concept, while using a quadratic readout can be interpreted as nonlinear VAR (NVAR) ( Bollt, 2021 ). Later, Gauthier et al. further studied this similarity and introduced the next generation RC, where instead of explicitly generating a reservoir of randomly connected neurons, an NVAR machine is formed in which the feature vector consists of time-delayed observations of the dynamical system and is augmented by nonlinear functions of these observations. Accordingly, with this approach there are fewer hyperparameters to tune and the intrinsic random nature of ESNs is effectively avoided. This approach was employed for one-step-ahead forecasting of benchmark chaotic time series for both reconstruction and cross-prediction tasks ( Gauthier et al., 2021 ). In this work, we assess the capability of the mainstream gated RNN techniques; ESN architectures, including the clustered architecture and the physics-informed hybrid approach; and the NVAR approach for multi-step-ahead prediction of nonlinear time series describing chaotic dynamical systems. In particular, we compare the performance of these models for forecasting two frequently used benchmark chaotic time series, derived from the Mackey–Glass and Lorenz dynamical systems, two additional chaotic times series derived from a bursting Morris–Lecar neuron model and the Vallis El Niño Southern Oscillation (ENSO) system, and one real-world dataset consisting of a time series of irregular cardiac voltage traces obtained in ex-vivo experiments in terms of the prediction error and computational efficiency. Moreover, this experimental dataset is further used to evaluate the performance of NVAR against traditional RC approaches in more detail. This paper is structured as follows. Section 2 presents a summary of the modeling approaches used for forecasting chaotic time series in this research and provides details about the implementation of each model and the evaluation metrics employed in this study. These methods are applied to datasets whose characteristics are described in Section 3 . The results are presented and discussed in Section 4 , and Section 5 presents concluding remarks."
} | 3,135 |
35755176 | PMC9230140 | pmc | 199 | {
"abstract": "In recent years, machine-learning techniques, particularly deep learning, have outperformed traditional time-series forecasting approaches in many contexts, including univariate and multivariate predictions. This study aims to investigate the capability of (i) gated recurrent neural networks, including long short-term memory (LSTM) and gated recurrent unit (GRU) networks, (ii) reservoir computing (RC) techniques, such as echo state networks (ESNs) and hybrid physics-informed ESNs, and (iii) the nonlinear vector autoregression (NVAR) approach, which has recently been introduced as the next generation RC, for the prediction of chaotic time series and to compare their performance in terms of accuracy, efficiency, and robustness. We apply the methods to predict time series obtained from two widely used chaotic benchmarks, the Mackey–Glass and Lorenz-63 models, as well as two other chaotic datasets representing a bursting neuron and the dynamics of the El Niño Southern Oscillation, and to one experimental dataset representing a time series of cardiac voltage with complex dynamics. We find that even though gated RNN techniques have been successful in forecasting time series generally, they can fall short in predicting chaotic time series for the methods, datasets, and ranges of hyperparameter values considered here. In contrast, for the chaotic datasets studied, we found that reservoir computing and NVAR techniques are more computationally efficient and offer more promise in long-term prediction of chaotic time series.",
"conclusion": "6. Conclusion In this paper, six different ML time-series forecasting approaches, including two gated RNN techniques, three variants of ESNs, and the NVAR approach, were tested to predict five chaotic time series, including the Mackey–Glass, Lorenz-63, bursting Morris–Lecar, Vallis ENSO, and experimental cardiac action potential time series. Although we considered relatively large but still limited numbers of datasets and methods, we found that the LSTM and GRU approaches, despite their high computational costs and in contrast to the ESN and NVAR methods, were incapable of forecasting the Mackey–Glass, Lorenz, and bursting Morris–Lecar time series more than a few steps into the future, and that increasing the network size did not significantly improve their performance. For the ENSO model, the GRU method could predict somewhat longer, but it did not compare favorably with the ESN and NVAR approaches. Three variants of ESNs were employed including the baseline ESN, the clustered ESN (CESN), and the hybrid physics-informed ESN (HESN). For the five datasets we used in this work, only one (ENSO) showed improvement by using a more complicated ESN architecture such as the clustered reservoir. In all the other cases, the baseline ESN demonstrated similar or better performance compared to CESN. In contrast, whereas the HESN provided the same level of prediction accuracy for the four synthetic time series, within the tested network sizes, it was the most successful approach for forecasting the experimental dataset, where it delivered more accurate predictions as measured by RMSE. Thus, incorporating the domain knowledge of a dynamical system if available may improve the prediction ability of the ESN technique and may help with obtaining good predictions using smaller network sizes. For the tested network sizes and datasets, the best prediction performance in the case of the Mackey–Glass, Lorenz, and bursting Morris–Lecar datasets was obtained by the NVAR method, which was recently introduced as the next generation of RC techniques and has been demonstrated to be as successful as optimized ESNs. For the ENSO dataset, NVAR’s prediction accuracy was only slightly lower than that of the most accurate method, CESN. A noticeable advantage of the NVAR technique over conventional ESNs is avoiding the explicit construction of randomly connected neurons and circumventing the intrinsic randomness that increases the sensitivity of the network to the hyperparameter values and initial parameters that remain untrained. Moreover, the number of hyperparameters is much smaller than for ESNs, which makes NVAR easier to tune. Such advantages may initially suggest that the amount of data required to train the NVAR model could be less than that needed for the conventional ESNs. However, our experiments showed in the case of the experimental cardiac voltage dataset, better performance was only obtained by embedding more delays and at the cost of more computational time and effort. Nevertheless, in general, this approach shows promise for efficient prediction of chaotic time series. To the best of our knowledge, this work is one of the first applications of this newly introduced technique to real-world experimental time series. Further studies in this area may reveal more of the potential of this approach. For instance, in this work, we used a quadratic polynomial functional to construct the nonlinear portion of the state vectors at each time step; however, other nonlinear functions such as higher-order polynomials could also be employed and studied. It should also be noted that our conclusions are based on a limited number of datasets and employed methods. Moreover, in each case, the optimum hyperparameters were obtained in a finite grid search process. Accordingly, it is possible that the same approaches could provide different results when applied to other datasets or when hyperparameters are determined using more extensive grids or different optimization techniques.",
"introduction": "1. Introduction Time series are important in many real-world applications, such as biology ( Bar-Joseph et al., 2003 ), finance ( Dingli & Fournier, 2017 ; Plagianakos & Tzanaki, 2001 ; Takahashi et al., 2019 ; Tsay, 2005 ; Zhao, 2009 ), climate science ( Ghil & Vautard, 1991 ), anomaly detection in computer networks ( Limthong, 2013 ) and social networks ( Gong et al., 2018 ), and energy ( Billinton et al., 1996 ; Bunn, 2000 ; Deihimi & Showkati, 2012 ). Accordingly, the analysis and prediction of time series data are of great importance and have been the focus of much research in the past few decades. In general, a time series represents a record of observations of a dynamical system at specific time intervals. Therefore, time series prediction involves determining the future evolution of a dynamical system, which can be especially challenging for chaotic dynamical systems. The states of such systems can be represented by chaotic time series, which are recognized by the orbital instability characteristic, where infinitesimal differences in the initial values bring about large differences in the time series behavior. Consequently, prediction of a chaotic time series is only feasible for a relatively short time before the appearance of orbital instability. For this reason, forecasting chaotic time series has remained a difficult task for the last few decades. Data-driven approaches, and machine-learning (ML) techniques in particular, have recently become the main approaches used for time-series forecasting ( Ahmed et al., 2010 ; Ben Taieb et al., 2012 ; Chandra et al., 2021 ; Chattopadhyay et al., 2020 ; Cheng et al., 2015 ; De Gooijer & Hyndman, 2006 ; Dubois et al., 2020 ; Kutz, 2013 ; Li et al., 2005 ; Tealab, 2018 ). In particular, recurrent neural networks (RNNs) are the mainstream architecture for analyzing sequential data, owing to their ability in interpreting temporal dependencies in the input time series ( Chandra et al., 2021 ; Elman, 1990 ; Elman & Zipser, 1988 ; Schmidhuber, 2015 ). The recurrent connections in such networks serve as a notion of memory, allowing them to embed temporal information. Despite the success of RNNs in modeling short-term temporal data and non-chaotic dynamical systems, the high computational cost of back-propagation through time and their vulnerability to the vanishing or exploding gradient problems have limited their applications. Gated RNN architectures were introduced to address some of these problems. More precisely, the memory cell architecture and the gating mechanism enable these networks to be more selective over the information that needs be remembered or forgotten, thereby enabling them to learn long-term dependencies in temporal sequences. Long short-term memory (LSTM) networks ( Hochreiter & Schmidhuber, 1997 ) and gated recurrent units (GRUs) ( Chung et al., 2014 ) are among the most widely used gated RNNs. An alternative approach to deal with time-series forecasting and modeling dynamical systems is reservoir computing (RC), a learning paradigm mostly implemented as echo state networks (ESNs) ( Jaeger, 2002 ; Lukoševičius & Jaeger, 2009 ; Sun et al., 2020 ). The RC paradigm is fundamentally derived from RNN concepts offering a streamlined training process, which remains limited to obtaining the output layer weights, while the rest of the parameter values are set randomly and remain untrained. Notwithstanding such a major simplification, ESNs have successfully been employed for multi-step-ahead prediction of nonlinear time series and modeling chaotic dynamical systems at low computational cost ( Bianchi et al., 2017 ; Han et al., 2021 ), triggering the development of several network topologies in recent years. For instance, clustered ESNs (CESNs) ( Deng & Zhang, 2006 ; Junior et al., 2020 ), where multiple sub-graphs of sparsely connected hidden units form the reservoir, and deep ESNs, where the reservoir consists of multiple sub-reservoir layers stacked hierarchically ( Gallicchio & Micheli, 2017 ; Gallicchio et al., 2017 ), are two widely used architectures. Hybrid ESNs (HESNs) are another category of RC techniques introduced in a physics-informed ML framework ( Oh, 2020 ; Willard et al., 2020 ), where additional inputs from physics-based mathematical models integrate corresponding domain knowledge into data-driven models ( Doan et al., 2019 ; Pathak, Hunt, et al., 2018 ). The successful application of ESNs, despite their random construction, in forecasting complex dynamical systems using time-series data triggered a series of recent research providing an interpretation of how RC techniques function. Recently, Bollt demonstrated how the RC with linear activation functions and linear readout layer shares similarities with the well-studied vector autoregressive (VAR) concept, while using a quadratic readout can be interpreted as nonlinear VAR (NVAR) ( Bollt, 2021 ). Later, Gauthier et al. further studied this similarity and introduced the next generation RC, where instead of explicitly generating a reservoir of randomly connected neurons, an NVAR machine is formed in which the feature vector consists of time-delayed observations of the dynamical system and is augmented by nonlinear functions of these observations. Accordingly, with this approach there are fewer hyperparameters to tune and the intrinsic random nature of ESNs is effectively avoided. This approach was employed for one-step-ahead forecasting of benchmark chaotic time series for both reconstruction and cross-prediction tasks ( Gauthier et al., 2021 ). In this work, we assess the capability of the mainstream gated RNN techniques; ESN architectures, including the clustered architecture and the physics-informed hybrid approach; and the NVAR approach for multi-step-ahead prediction of nonlinear time series describing chaotic dynamical systems. In particular, we compare the performance of these models for forecasting two frequently used benchmark chaotic time series, derived from the Mackey–Glass and Lorenz dynamical systems, two additional chaotic times series derived from a bursting Morris–Lecar neuron model and the Vallis El Niño Southern Oscillation (ENSO) system, and one real-world dataset consisting of a time series of irregular cardiac voltage traces obtained in ex-vivo experiments in terms of the prediction error and computational efficiency. Moreover, this experimental dataset is further used to evaluate the performance of NVAR against traditional RC approaches in more detail. This paper is structured as follows. Section 2 presents a summary of the modeling approaches used for forecasting chaotic time series in this research and provides details about the implementation of each model and the evaluation metrics employed in this study. These methods are applied to datasets whose characteristics are described in Section 3 . The results are presented and discussed in Section 4 , and Section 5 presents concluding remarks."
} | 3,135 |
35755176 | PMC9230140 | pmc | 199 | {
"abstract": "In recent years, machine-learning techniques, particularly deep learning, have outperformed traditional time-series forecasting approaches in many contexts, including univariate and multivariate predictions. This study aims to investigate the capability of (i) gated recurrent neural networks, including long short-term memory (LSTM) and gated recurrent unit (GRU) networks, (ii) reservoir computing (RC) techniques, such as echo state networks (ESNs) and hybrid physics-informed ESNs, and (iii) the nonlinear vector autoregression (NVAR) approach, which has recently been introduced as the next generation RC, for the prediction of chaotic time series and to compare their performance in terms of accuracy, efficiency, and robustness. We apply the methods to predict time series obtained from two widely used chaotic benchmarks, the Mackey–Glass and Lorenz-63 models, as well as two other chaotic datasets representing a bursting neuron and the dynamics of the El Niño Southern Oscillation, and to one experimental dataset representing a time series of cardiac voltage with complex dynamics. We find that even though gated RNN techniques have been successful in forecasting time series generally, they can fall short in predicting chaotic time series for the methods, datasets, and ranges of hyperparameter values considered here. In contrast, for the chaotic datasets studied, we found that reservoir computing and NVAR techniques are more computationally efficient and offer more promise in long-term prediction of chaotic time series.",
"conclusion": "6. Conclusion In this paper, six different ML time-series forecasting approaches, including two gated RNN techniques, three variants of ESNs, and the NVAR approach, were tested to predict five chaotic time series, including the Mackey–Glass, Lorenz-63, bursting Morris–Lecar, Vallis ENSO, and experimental cardiac action potential time series. Although we considered relatively large but still limited numbers of datasets and methods, we found that the LSTM and GRU approaches, despite their high computational costs and in contrast to the ESN and NVAR methods, were incapable of forecasting the Mackey–Glass, Lorenz, and bursting Morris–Lecar time series more than a few steps into the future, and that increasing the network size did not significantly improve their performance. For the ENSO model, the GRU method could predict somewhat longer, but it did not compare favorably with the ESN and NVAR approaches. Three variants of ESNs were employed including the baseline ESN, the clustered ESN (CESN), and the hybrid physics-informed ESN (HESN). For the five datasets we used in this work, only one (ENSO) showed improvement by using a more complicated ESN architecture such as the clustered reservoir. In all the other cases, the baseline ESN demonstrated similar or better performance compared to CESN. In contrast, whereas the HESN provided the same level of prediction accuracy for the four synthetic time series, within the tested network sizes, it was the most successful approach for forecasting the experimental dataset, where it delivered more accurate predictions as measured by RMSE. Thus, incorporating the domain knowledge of a dynamical system if available may improve the prediction ability of the ESN technique and may help with obtaining good predictions using smaller network sizes. For the tested network sizes and datasets, the best prediction performance in the case of the Mackey–Glass, Lorenz, and bursting Morris–Lecar datasets was obtained by the NVAR method, which was recently introduced as the next generation of RC techniques and has been demonstrated to be as successful as optimized ESNs. For the ENSO dataset, NVAR’s prediction accuracy was only slightly lower than that of the most accurate method, CESN. A noticeable advantage of the NVAR technique over conventional ESNs is avoiding the explicit construction of randomly connected neurons and circumventing the intrinsic randomness that increases the sensitivity of the network to the hyperparameter values and initial parameters that remain untrained. Moreover, the number of hyperparameters is much smaller than for ESNs, which makes NVAR easier to tune. Such advantages may initially suggest that the amount of data required to train the NVAR model could be less than that needed for the conventional ESNs. However, our experiments showed in the case of the experimental cardiac voltage dataset, better performance was only obtained by embedding more delays and at the cost of more computational time and effort. Nevertheless, in general, this approach shows promise for efficient prediction of chaotic time series. To the best of our knowledge, this work is one of the first applications of this newly introduced technique to real-world experimental time series. Further studies in this area may reveal more of the potential of this approach. For instance, in this work, we used a quadratic polynomial functional to construct the nonlinear portion of the state vectors at each time step; however, other nonlinear functions such as higher-order polynomials could also be employed and studied. It should also be noted that our conclusions are based on a limited number of datasets and employed methods. Moreover, in each case, the optimum hyperparameters were obtained in a finite grid search process. Accordingly, it is possible that the same approaches could provide different results when applied to other datasets or when hyperparameters are determined using more extensive grids or different optimization techniques.",
"introduction": "1. Introduction Time series are important in many real-world applications, such as biology ( Bar-Joseph et al., 2003 ), finance ( Dingli & Fournier, 2017 ; Plagianakos & Tzanaki, 2001 ; Takahashi et al., 2019 ; Tsay, 2005 ; Zhao, 2009 ), climate science ( Ghil & Vautard, 1991 ), anomaly detection in computer networks ( Limthong, 2013 ) and social networks ( Gong et al., 2018 ), and energy ( Billinton et al., 1996 ; Bunn, 2000 ; Deihimi & Showkati, 2012 ). Accordingly, the analysis and prediction of time series data are of great importance and have been the focus of much research in the past few decades. In general, a time series represents a record of observations of a dynamical system at specific time intervals. Therefore, time series prediction involves determining the future evolution of a dynamical system, which can be especially challenging for chaotic dynamical systems. The states of such systems can be represented by chaotic time series, which are recognized by the orbital instability characteristic, where infinitesimal differences in the initial values bring about large differences in the time series behavior. Consequently, prediction of a chaotic time series is only feasible for a relatively short time before the appearance of orbital instability. For this reason, forecasting chaotic time series has remained a difficult task for the last few decades. Data-driven approaches, and machine-learning (ML) techniques in particular, have recently become the main approaches used for time-series forecasting ( Ahmed et al., 2010 ; Ben Taieb et al., 2012 ; Chandra et al., 2021 ; Chattopadhyay et al., 2020 ; Cheng et al., 2015 ; De Gooijer & Hyndman, 2006 ; Dubois et al., 2020 ; Kutz, 2013 ; Li et al., 2005 ; Tealab, 2018 ). In particular, recurrent neural networks (RNNs) are the mainstream architecture for analyzing sequential data, owing to their ability in interpreting temporal dependencies in the input time series ( Chandra et al., 2021 ; Elman, 1990 ; Elman & Zipser, 1988 ; Schmidhuber, 2015 ). The recurrent connections in such networks serve as a notion of memory, allowing them to embed temporal information. Despite the success of RNNs in modeling short-term temporal data and non-chaotic dynamical systems, the high computational cost of back-propagation through time and their vulnerability to the vanishing or exploding gradient problems have limited their applications. Gated RNN architectures were introduced to address some of these problems. More precisely, the memory cell architecture and the gating mechanism enable these networks to be more selective over the information that needs be remembered or forgotten, thereby enabling them to learn long-term dependencies in temporal sequences. Long short-term memory (LSTM) networks ( Hochreiter & Schmidhuber, 1997 ) and gated recurrent units (GRUs) ( Chung et al., 2014 ) are among the most widely used gated RNNs. An alternative approach to deal with time-series forecasting and modeling dynamical systems is reservoir computing (RC), a learning paradigm mostly implemented as echo state networks (ESNs) ( Jaeger, 2002 ; Lukoševičius & Jaeger, 2009 ; Sun et al., 2020 ). The RC paradigm is fundamentally derived from RNN concepts offering a streamlined training process, which remains limited to obtaining the output layer weights, while the rest of the parameter values are set randomly and remain untrained. Notwithstanding such a major simplification, ESNs have successfully been employed for multi-step-ahead prediction of nonlinear time series and modeling chaotic dynamical systems at low computational cost ( Bianchi et al., 2017 ; Han et al., 2021 ), triggering the development of several network topologies in recent years. For instance, clustered ESNs (CESNs) ( Deng & Zhang, 2006 ; Junior et al., 2020 ), where multiple sub-graphs of sparsely connected hidden units form the reservoir, and deep ESNs, where the reservoir consists of multiple sub-reservoir layers stacked hierarchically ( Gallicchio & Micheli, 2017 ; Gallicchio et al., 2017 ), are two widely used architectures. Hybrid ESNs (HESNs) are another category of RC techniques introduced in a physics-informed ML framework ( Oh, 2020 ; Willard et al., 2020 ), where additional inputs from physics-based mathematical models integrate corresponding domain knowledge into data-driven models ( Doan et al., 2019 ; Pathak, Hunt, et al., 2018 ). The successful application of ESNs, despite their random construction, in forecasting complex dynamical systems using time-series data triggered a series of recent research providing an interpretation of how RC techniques function. Recently, Bollt demonstrated how the RC with linear activation functions and linear readout layer shares similarities with the well-studied vector autoregressive (VAR) concept, while using a quadratic readout can be interpreted as nonlinear VAR (NVAR) ( Bollt, 2021 ). Later, Gauthier et al. further studied this similarity and introduced the next generation RC, where instead of explicitly generating a reservoir of randomly connected neurons, an NVAR machine is formed in which the feature vector consists of time-delayed observations of the dynamical system and is augmented by nonlinear functions of these observations. Accordingly, with this approach there are fewer hyperparameters to tune and the intrinsic random nature of ESNs is effectively avoided. This approach was employed for one-step-ahead forecasting of benchmark chaotic time series for both reconstruction and cross-prediction tasks ( Gauthier et al., 2021 ). In this work, we assess the capability of the mainstream gated RNN techniques; ESN architectures, including the clustered architecture and the physics-informed hybrid approach; and the NVAR approach for multi-step-ahead prediction of nonlinear time series describing chaotic dynamical systems. In particular, we compare the performance of these models for forecasting two frequently used benchmark chaotic time series, derived from the Mackey–Glass and Lorenz dynamical systems, two additional chaotic times series derived from a bursting Morris–Lecar neuron model and the Vallis El Niño Southern Oscillation (ENSO) system, and one real-world dataset consisting of a time series of irregular cardiac voltage traces obtained in ex-vivo experiments in terms of the prediction error and computational efficiency. Moreover, this experimental dataset is further used to evaluate the performance of NVAR against traditional RC approaches in more detail. This paper is structured as follows. Section 2 presents a summary of the modeling approaches used for forecasting chaotic time series in this research and provides details about the implementation of each model and the evaluation metrics employed in this study. These methods are applied to datasets whose characteristics are described in Section 3 . The results are presented and discussed in Section 4 , and Section 5 presents concluding remarks."
} | 3,135 |
35755176 | PMC9230140 | pmc | 199 | {
"abstract": "In recent years, machine-learning techniques, particularly deep learning, have outperformed traditional time-series forecasting approaches in many contexts, including univariate and multivariate predictions. This study aims to investigate the capability of (i) gated recurrent neural networks, including long short-term memory (LSTM) and gated recurrent unit (GRU) networks, (ii) reservoir computing (RC) techniques, such as echo state networks (ESNs) and hybrid physics-informed ESNs, and (iii) the nonlinear vector autoregression (NVAR) approach, which has recently been introduced as the next generation RC, for the prediction of chaotic time series and to compare their performance in terms of accuracy, efficiency, and robustness. We apply the methods to predict time series obtained from two widely used chaotic benchmarks, the Mackey–Glass and Lorenz-63 models, as well as two other chaotic datasets representing a bursting neuron and the dynamics of the El Niño Southern Oscillation, and to one experimental dataset representing a time series of cardiac voltage with complex dynamics. We find that even though gated RNN techniques have been successful in forecasting time series generally, they can fall short in predicting chaotic time series for the methods, datasets, and ranges of hyperparameter values considered here. In contrast, for the chaotic datasets studied, we found that reservoir computing and NVAR techniques are more computationally efficient and offer more promise in long-term prediction of chaotic time series.",
"conclusion": "6. Conclusion In this paper, six different ML time-series forecasting approaches, including two gated RNN techniques, three variants of ESNs, and the NVAR approach, were tested to predict five chaotic time series, including the Mackey–Glass, Lorenz-63, bursting Morris–Lecar, Vallis ENSO, and experimental cardiac action potential time series. Although we considered relatively large but still limited numbers of datasets and methods, we found that the LSTM and GRU approaches, despite their high computational costs and in contrast to the ESN and NVAR methods, were incapable of forecasting the Mackey–Glass, Lorenz, and bursting Morris–Lecar time series more than a few steps into the future, and that increasing the network size did not significantly improve their performance. For the ENSO model, the GRU method could predict somewhat longer, but it did not compare favorably with the ESN and NVAR approaches. Three variants of ESNs were employed including the baseline ESN, the clustered ESN (CESN), and the hybrid physics-informed ESN (HESN). For the five datasets we used in this work, only one (ENSO) showed improvement by using a more complicated ESN architecture such as the clustered reservoir. In all the other cases, the baseline ESN demonstrated similar or better performance compared to CESN. In contrast, whereas the HESN provided the same level of prediction accuracy for the four synthetic time series, within the tested network sizes, it was the most successful approach for forecasting the experimental dataset, where it delivered more accurate predictions as measured by RMSE. Thus, incorporating the domain knowledge of a dynamical system if available may improve the prediction ability of the ESN technique and may help with obtaining good predictions using smaller network sizes. For the tested network sizes and datasets, the best prediction performance in the case of the Mackey–Glass, Lorenz, and bursting Morris–Lecar datasets was obtained by the NVAR method, which was recently introduced as the next generation of RC techniques and has been demonstrated to be as successful as optimized ESNs. For the ENSO dataset, NVAR’s prediction accuracy was only slightly lower than that of the most accurate method, CESN. A noticeable advantage of the NVAR technique over conventional ESNs is avoiding the explicit construction of randomly connected neurons and circumventing the intrinsic randomness that increases the sensitivity of the network to the hyperparameter values and initial parameters that remain untrained. Moreover, the number of hyperparameters is much smaller than for ESNs, which makes NVAR easier to tune. Such advantages may initially suggest that the amount of data required to train the NVAR model could be less than that needed for the conventional ESNs. However, our experiments showed in the case of the experimental cardiac voltage dataset, better performance was only obtained by embedding more delays and at the cost of more computational time and effort. Nevertheless, in general, this approach shows promise for efficient prediction of chaotic time series. To the best of our knowledge, this work is one of the first applications of this newly introduced technique to real-world experimental time series. Further studies in this area may reveal more of the potential of this approach. For instance, in this work, we used a quadratic polynomial functional to construct the nonlinear portion of the state vectors at each time step; however, other nonlinear functions such as higher-order polynomials could also be employed and studied. It should also be noted that our conclusions are based on a limited number of datasets and employed methods. Moreover, in each case, the optimum hyperparameters were obtained in a finite grid search process. Accordingly, it is possible that the same approaches could provide different results when applied to other datasets or when hyperparameters are determined using more extensive grids or different optimization techniques.",
"introduction": "1. Introduction Time series are important in many real-world applications, such as biology ( Bar-Joseph et al., 2003 ), finance ( Dingli & Fournier, 2017 ; Plagianakos & Tzanaki, 2001 ; Takahashi et al., 2019 ; Tsay, 2005 ; Zhao, 2009 ), climate science ( Ghil & Vautard, 1991 ), anomaly detection in computer networks ( Limthong, 2013 ) and social networks ( Gong et al., 2018 ), and energy ( Billinton et al., 1996 ; Bunn, 2000 ; Deihimi & Showkati, 2012 ). Accordingly, the analysis and prediction of time series data are of great importance and have been the focus of much research in the past few decades. In general, a time series represents a record of observations of a dynamical system at specific time intervals. Therefore, time series prediction involves determining the future evolution of a dynamical system, which can be especially challenging for chaotic dynamical systems. The states of such systems can be represented by chaotic time series, which are recognized by the orbital instability characteristic, where infinitesimal differences in the initial values bring about large differences in the time series behavior. Consequently, prediction of a chaotic time series is only feasible for a relatively short time before the appearance of orbital instability. For this reason, forecasting chaotic time series has remained a difficult task for the last few decades. Data-driven approaches, and machine-learning (ML) techniques in particular, have recently become the main approaches used for time-series forecasting ( Ahmed et al., 2010 ; Ben Taieb et al., 2012 ; Chandra et al., 2021 ; Chattopadhyay et al., 2020 ; Cheng et al., 2015 ; De Gooijer & Hyndman, 2006 ; Dubois et al., 2020 ; Kutz, 2013 ; Li et al., 2005 ; Tealab, 2018 ). In particular, recurrent neural networks (RNNs) are the mainstream architecture for analyzing sequential data, owing to their ability in interpreting temporal dependencies in the input time series ( Chandra et al., 2021 ; Elman, 1990 ; Elman & Zipser, 1988 ; Schmidhuber, 2015 ). The recurrent connections in such networks serve as a notion of memory, allowing them to embed temporal information. Despite the success of RNNs in modeling short-term temporal data and non-chaotic dynamical systems, the high computational cost of back-propagation through time and their vulnerability to the vanishing or exploding gradient problems have limited their applications. Gated RNN architectures were introduced to address some of these problems. More precisely, the memory cell architecture and the gating mechanism enable these networks to be more selective over the information that needs be remembered or forgotten, thereby enabling them to learn long-term dependencies in temporal sequences. Long short-term memory (LSTM) networks ( Hochreiter & Schmidhuber, 1997 ) and gated recurrent units (GRUs) ( Chung et al., 2014 ) are among the most widely used gated RNNs. An alternative approach to deal with time-series forecasting and modeling dynamical systems is reservoir computing (RC), a learning paradigm mostly implemented as echo state networks (ESNs) ( Jaeger, 2002 ; Lukoševičius & Jaeger, 2009 ; Sun et al., 2020 ). The RC paradigm is fundamentally derived from RNN concepts offering a streamlined training process, which remains limited to obtaining the output layer weights, while the rest of the parameter values are set randomly and remain untrained. Notwithstanding such a major simplification, ESNs have successfully been employed for multi-step-ahead prediction of nonlinear time series and modeling chaotic dynamical systems at low computational cost ( Bianchi et al., 2017 ; Han et al., 2021 ), triggering the development of several network topologies in recent years. For instance, clustered ESNs (CESNs) ( Deng & Zhang, 2006 ; Junior et al., 2020 ), where multiple sub-graphs of sparsely connected hidden units form the reservoir, and deep ESNs, where the reservoir consists of multiple sub-reservoir layers stacked hierarchically ( Gallicchio & Micheli, 2017 ; Gallicchio et al., 2017 ), are two widely used architectures. Hybrid ESNs (HESNs) are another category of RC techniques introduced in a physics-informed ML framework ( Oh, 2020 ; Willard et al., 2020 ), where additional inputs from physics-based mathematical models integrate corresponding domain knowledge into data-driven models ( Doan et al., 2019 ; Pathak, Hunt, et al., 2018 ). The successful application of ESNs, despite their random construction, in forecasting complex dynamical systems using time-series data triggered a series of recent research providing an interpretation of how RC techniques function. Recently, Bollt demonstrated how the RC with linear activation functions and linear readout layer shares similarities with the well-studied vector autoregressive (VAR) concept, while using a quadratic readout can be interpreted as nonlinear VAR (NVAR) ( Bollt, 2021 ). Later, Gauthier et al. further studied this similarity and introduced the next generation RC, where instead of explicitly generating a reservoir of randomly connected neurons, an NVAR machine is formed in which the feature vector consists of time-delayed observations of the dynamical system and is augmented by nonlinear functions of these observations. Accordingly, with this approach there are fewer hyperparameters to tune and the intrinsic random nature of ESNs is effectively avoided. This approach was employed for one-step-ahead forecasting of benchmark chaotic time series for both reconstruction and cross-prediction tasks ( Gauthier et al., 2021 ). In this work, we assess the capability of the mainstream gated RNN techniques; ESN architectures, including the clustered architecture and the physics-informed hybrid approach; and the NVAR approach for multi-step-ahead prediction of nonlinear time series describing chaotic dynamical systems. In particular, we compare the performance of these models for forecasting two frequently used benchmark chaotic time series, derived from the Mackey–Glass and Lorenz dynamical systems, two additional chaotic times series derived from a bursting Morris–Lecar neuron model and the Vallis El Niño Southern Oscillation (ENSO) system, and one real-world dataset consisting of a time series of irregular cardiac voltage traces obtained in ex-vivo experiments in terms of the prediction error and computational efficiency. Moreover, this experimental dataset is further used to evaluate the performance of NVAR against traditional RC approaches in more detail. This paper is structured as follows. Section 2 presents a summary of the modeling approaches used for forecasting chaotic time series in this research and provides details about the implementation of each model and the evaluation metrics employed in this study. These methods are applied to datasets whose characteristics are described in Section 3 . The results are presented and discussed in Section 4 , and Section 5 presents concluding remarks."
} | 3,135 |
35755176 | PMC9230140 | pmc | 200 | {
"abstract": "In recent years, machine-learning techniques, particularly deep learning, have outperformed traditional time-series forecasting approaches in many contexts, including univariate and multivariate predictions. This study aims to investigate the capability of (i) gated recurrent neural networks, including long short-term memory (LSTM) and gated recurrent unit (GRU) networks, (ii) reservoir computing (RC) techniques, such as echo state networks (ESNs) and hybrid physics-informed ESNs, and (iii) the nonlinear vector autoregression (NVAR) approach, which has recently been introduced as the next generation RC, for the prediction of chaotic time series and to compare their performance in terms of accuracy, efficiency, and robustness. We apply the methods to predict time series obtained from two widely used chaotic benchmarks, the Mackey–Glass and Lorenz-63 models, as well as two other chaotic datasets representing a bursting neuron and the dynamics of the El Niño Southern Oscillation, and to one experimental dataset representing a time series of cardiac voltage with complex dynamics. We find that even though gated RNN techniques have been successful in forecasting time series generally, they can fall short in predicting chaotic time series for the methods, datasets, and ranges of hyperparameter values considered here. In contrast, for the chaotic datasets studied, we found that reservoir computing and NVAR techniques are more computationally efficient and offer more promise in long-term prediction of chaotic time series.",
"conclusion": "6. Conclusion In this paper, six different ML time-series forecasting approaches, including two gated RNN techniques, three variants of ESNs, and the NVAR approach, were tested to predict five chaotic time series, including the Mackey–Glass, Lorenz-63, bursting Morris–Lecar, Vallis ENSO, and experimental cardiac action potential time series. Although we considered relatively large but still limited numbers of datasets and methods, we found that the LSTM and GRU approaches, despite their high computational costs and in contrast to the ESN and NVAR methods, were incapable of forecasting the Mackey–Glass, Lorenz, and bursting Morris–Lecar time series more than a few steps into the future, and that increasing the network size did not significantly improve their performance. For the ENSO model, the GRU method could predict somewhat longer, but it did not compare favorably with the ESN and NVAR approaches. Three variants of ESNs were employed including the baseline ESN, the clustered ESN (CESN), and the hybrid physics-informed ESN (HESN). For the five datasets we used in this work, only one (ENSO) showed improvement by using a more complicated ESN architecture such as the clustered reservoir. In all the other cases, the baseline ESN demonstrated similar or better performance compared to CESN. In contrast, whereas the HESN provided the same level of prediction accuracy for the four synthetic time series, within the tested network sizes, it was the most successful approach for forecasting the experimental dataset, where it delivered more accurate predictions as measured by RMSE. Thus, incorporating the domain knowledge of a dynamical system if available may improve the prediction ability of the ESN technique and may help with obtaining good predictions using smaller network sizes. For the tested network sizes and datasets, the best prediction performance in the case of the Mackey–Glass, Lorenz, and bursting Morris–Lecar datasets was obtained by the NVAR method, which was recently introduced as the next generation of RC techniques and has been demonstrated to be as successful as optimized ESNs. For the ENSO dataset, NVAR’s prediction accuracy was only slightly lower than that of the most accurate method, CESN. A noticeable advantage of the NVAR technique over conventional ESNs is avoiding the explicit construction of randomly connected neurons and circumventing the intrinsic randomness that increases the sensitivity of the network to the hyperparameter values and initial parameters that remain untrained. Moreover, the number of hyperparameters is much smaller than for ESNs, which makes NVAR easier to tune. Such advantages may initially suggest that the amount of data required to train the NVAR model could be less than that needed for the conventional ESNs. However, our experiments showed in the case of the experimental cardiac voltage dataset, better performance was only obtained by embedding more delays and at the cost of more computational time and effort. Nevertheless, in general, this approach shows promise for efficient prediction of chaotic time series. To the best of our knowledge, this work is one of the first applications of this newly introduced technique to real-world experimental time series. Further studies in this area may reveal more of the potential of this approach. For instance, in this work, we used a quadratic polynomial functional to construct the nonlinear portion of the state vectors at each time step; however, other nonlinear functions such as higher-order polynomials could also be employed and studied. It should also be noted that our conclusions are based on a limited number of datasets and employed methods. Moreover, in each case, the optimum hyperparameters were obtained in a finite grid search process. Accordingly, it is possible that the same approaches could provide different results when applied to other datasets or when hyperparameters are determined using more extensive grids or different optimization techniques.",
"introduction": "1. Introduction Time series are important in many real-world applications, such as biology ( Bar-Joseph et al., 2003 ), finance ( Dingli & Fournier, 2017 ; Plagianakos & Tzanaki, 2001 ; Takahashi et al., 2019 ; Tsay, 2005 ; Zhao, 2009 ), climate science ( Ghil & Vautard, 1991 ), anomaly detection in computer networks ( Limthong, 2013 ) and social networks ( Gong et al., 2018 ), and energy ( Billinton et al., 1996 ; Bunn, 2000 ; Deihimi & Showkati, 2012 ). Accordingly, the analysis and prediction of time series data are of great importance and have been the focus of much research in the past few decades. In general, a time series represents a record of observations of a dynamical system at specific time intervals. Therefore, time series prediction involves determining the future evolution of a dynamical system, which can be especially challenging for chaotic dynamical systems. The states of such systems can be represented by chaotic time series, which are recognized by the orbital instability characteristic, where infinitesimal differences in the initial values bring about large differences in the time series behavior. Consequently, prediction of a chaotic time series is only feasible for a relatively short time before the appearance of orbital instability. For this reason, forecasting chaotic time series has remained a difficult task for the last few decades. Data-driven approaches, and machine-learning (ML) techniques in particular, have recently become the main approaches used for time-series forecasting ( Ahmed et al., 2010 ; Ben Taieb et al., 2012 ; Chandra et al., 2021 ; Chattopadhyay et al., 2020 ; Cheng et al., 2015 ; De Gooijer & Hyndman, 2006 ; Dubois et al., 2020 ; Kutz, 2013 ; Li et al., 2005 ; Tealab, 2018 ). In particular, recurrent neural networks (RNNs) are the mainstream architecture for analyzing sequential data, owing to their ability in interpreting temporal dependencies in the input time series ( Chandra et al., 2021 ; Elman, 1990 ; Elman & Zipser, 1988 ; Schmidhuber, 2015 ). The recurrent connections in such networks serve as a notion of memory, allowing them to embed temporal information. Despite the success of RNNs in modeling short-term temporal data and non-chaotic dynamical systems, the high computational cost of back-propagation through time and their vulnerability to the vanishing or exploding gradient problems have limited their applications. Gated RNN architectures were introduced to address some of these problems. More precisely, the memory cell architecture and the gating mechanism enable these networks to be more selective over the information that needs be remembered or forgotten, thereby enabling them to learn long-term dependencies in temporal sequences. Long short-term memory (LSTM) networks ( Hochreiter & Schmidhuber, 1997 ) and gated recurrent units (GRUs) ( Chung et al., 2014 ) are among the most widely used gated RNNs. An alternative approach to deal with time-series forecasting and modeling dynamical systems is reservoir computing (RC), a learning paradigm mostly implemented as echo state networks (ESNs) ( Jaeger, 2002 ; Lukoševičius & Jaeger, 2009 ; Sun et al., 2020 ). The RC paradigm is fundamentally derived from RNN concepts offering a streamlined training process, which remains limited to obtaining the output layer weights, while the rest of the parameter values are set randomly and remain untrained. Notwithstanding such a major simplification, ESNs have successfully been employed for multi-step-ahead prediction of nonlinear time series and modeling chaotic dynamical systems at low computational cost ( Bianchi et al., 2017 ; Han et al., 2021 ), triggering the development of several network topologies in recent years. For instance, clustered ESNs (CESNs) ( Deng & Zhang, 2006 ; Junior et al., 2020 ), where multiple sub-graphs of sparsely connected hidden units form the reservoir, and deep ESNs, where the reservoir consists of multiple sub-reservoir layers stacked hierarchically ( Gallicchio & Micheli, 2017 ; Gallicchio et al., 2017 ), are two widely used architectures. Hybrid ESNs (HESNs) are another category of RC techniques introduced in a physics-informed ML framework ( Oh, 2020 ; Willard et al., 2020 ), where additional inputs from physics-based mathematical models integrate corresponding domain knowledge into data-driven models ( Doan et al., 2019 ; Pathak, Hunt, et al., 2018 ). The successful application of ESNs, despite their random construction, in forecasting complex dynamical systems using time-series data triggered a series of recent research providing an interpretation of how RC techniques function. Recently, Bollt demonstrated how the RC with linear activation functions and linear readout layer shares similarities with the well-studied vector autoregressive (VAR) concept, while using a quadratic readout can be interpreted as nonlinear VAR (NVAR) ( Bollt, 2021 ). Later, Gauthier et al. further studied this similarity and introduced the next generation RC, where instead of explicitly generating a reservoir of randomly connected neurons, an NVAR machine is formed in which the feature vector consists of time-delayed observations of the dynamical system and is augmented by nonlinear functions of these observations. Accordingly, with this approach there are fewer hyperparameters to tune and the intrinsic random nature of ESNs is effectively avoided. This approach was employed for one-step-ahead forecasting of benchmark chaotic time series for both reconstruction and cross-prediction tasks ( Gauthier et al., 2021 ). In this work, we assess the capability of the mainstream gated RNN techniques; ESN architectures, including the clustered architecture and the physics-informed hybrid approach; and the NVAR approach for multi-step-ahead prediction of nonlinear time series describing chaotic dynamical systems. In particular, we compare the performance of these models for forecasting two frequently used benchmark chaotic time series, derived from the Mackey–Glass and Lorenz dynamical systems, two additional chaotic times series derived from a bursting Morris–Lecar neuron model and the Vallis El Niño Southern Oscillation (ENSO) system, and one real-world dataset consisting of a time series of irregular cardiac voltage traces obtained in ex-vivo experiments in terms of the prediction error and computational efficiency. Moreover, this experimental dataset is further used to evaluate the performance of NVAR against traditional RC approaches in more detail. This paper is structured as follows. Section 2 presents a summary of the modeling approaches used for forecasting chaotic time series in this research and provides details about the implementation of each model and the evaluation metrics employed in this study. These methods are applied to datasets whose characteristics are described in Section 3 . The results are presented and discussed in Section 4 , and Section 5 presents concluding remarks."
} | 3,135 |
35755176 | PMC9230140 | pmc | 200 | {
"abstract": "In recent years, machine-learning techniques, particularly deep learning, have outperformed traditional time-series forecasting approaches in many contexts, including univariate and multivariate predictions. This study aims to investigate the capability of (i) gated recurrent neural networks, including long short-term memory (LSTM) and gated recurrent unit (GRU) networks, (ii) reservoir computing (RC) techniques, such as echo state networks (ESNs) and hybrid physics-informed ESNs, and (iii) the nonlinear vector autoregression (NVAR) approach, which has recently been introduced as the next generation RC, for the prediction of chaotic time series and to compare their performance in terms of accuracy, efficiency, and robustness. We apply the methods to predict time series obtained from two widely used chaotic benchmarks, the Mackey–Glass and Lorenz-63 models, as well as two other chaotic datasets representing a bursting neuron and the dynamics of the El Niño Southern Oscillation, and to one experimental dataset representing a time series of cardiac voltage with complex dynamics. We find that even though gated RNN techniques have been successful in forecasting time series generally, they can fall short in predicting chaotic time series for the methods, datasets, and ranges of hyperparameter values considered here. In contrast, for the chaotic datasets studied, we found that reservoir computing and NVAR techniques are more computationally efficient and offer more promise in long-term prediction of chaotic time series.",
"conclusion": "6. Conclusion In this paper, six different ML time-series forecasting approaches, including two gated RNN techniques, three variants of ESNs, and the NVAR approach, were tested to predict five chaotic time series, including the Mackey–Glass, Lorenz-63, bursting Morris–Lecar, Vallis ENSO, and experimental cardiac action potential time series. Although we considered relatively large but still limited numbers of datasets and methods, we found that the LSTM and GRU approaches, despite their high computational costs and in contrast to the ESN and NVAR methods, were incapable of forecasting the Mackey–Glass, Lorenz, and bursting Morris–Lecar time series more than a few steps into the future, and that increasing the network size did not significantly improve their performance. For the ENSO model, the GRU method could predict somewhat longer, but it did not compare favorably with the ESN and NVAR approaches. Three variants of ESNs were employed including the baseline ESN, the clustered ESN (CESN), and the hybrid physics-informed ESN (HESN). For the five datasets we used in this work, only one (ENSO) showed improvement by using a more complicated ESN architecture such as the clustered reservoir. In all the other cases, the baseline ESN demonstrated similar or better performance compared to CESN. In contrast, whereas the HESN provided the same level of prediction accuracy for the four synthetic time series, within the tested network sizes, it was the most successful approach for forecasting the experimental dataset, where it delivered more accurate predictions as measured by RMSE. Thus, incorporating the domain knowledge of a dynamical system if available may improve the prediction ability of the ESN technique and may help with obtaining good predictions using smaller network sizes. For the tested network sizes and datasets, the best prediction performance in the case of the Mackey–Glass, Lorenz, and bursting Morris–Lecar datasets was obtained by the NVAR method, which was recently introduced as the next generation of RC techniques and has been demonstrated to be as successful as optimized ESNs. For the ENSO dataset, NVAR’s prediction accuracy was only slightly lower than that of the most accurate method, CESN. A noticeable advantage of the NVAR technique over conventional ESNs is avoiding the explicit construction of randomly connected neurons and circumventing the intrinsic randomness that increases the sensitivity of the network to the hyperparameter values and initial parameters that remain untrained. Moreover, the number of hyperparameters is much smaller than for ESNs, which makes NVAR easier to tune. Such advantages may initially suggest that the amount of data required to train the NVAR model could be less than that needed for the conventional ESNs. However, our experiments showed in the case of the experimental cardiac voltage dataset, better performance was only obtained by embedding more delays and at the cost of more computational time and effort. Nevertheless, in general, this approach shows promise for efficient prediction of chaotic time series. To the best of our knowledge, this work is one of the first applications of this newly introduced technique to real-world experimental time series. Further studies in this area may reveal more of the potential of this approach. For instance, in this work, we used a quadratic polynomial functional to construct the nonlinear portion of the state vectors at each time step; however, other nonlinear functions such as higher-order polynomials could also be employed and studied. It should also be noted that our conclusions are based on a limited number of datasets and employed methods. Moreover, in each case, the optimum hyperparameters were obtained in a finite grid search process. Accordingly, it is possible that the same approaches could provide different results when applied to other datasets or when hyperparameters are determined using more extensive grids or different optimization techniques.",
"introduction": "1. Introduction Time series are important in many real-world applications, such as biology ( Bar-Joseph et al., 2003 ), finance ( Dingli & Fournier, 2017 ; Plagianakos & Tzanaki, 2001 ; Takahashi et al., 2019 ; Tsay, 2005 ; Zhao, 2009 ), climate science ( Ghil & Vautard, 1991 ), anomaly detection in computer networks ( Limthong, 2013 ) and social networks ( Gong et al., 2018 ), and energy ( Billinton et al., 1996 ; Bunn, 2000 ; Deihimi & Showkati, 2012 ). Accordingly, the analysis and prediction of time series data are of great importance and have been the focus of much research in the past few decades. In general, a time series represents a record of observations of a dynamical system at specific time intervals. Therefore, time series prediction involves determining the future evolution of a dynamical system, which can be especially challenging for chaotic dynamical systems. The states of such systems can be represented by chaotic time series, which are recognized by the orbital instability characteristic, where infinitesimal differences in the initial values bring about large differences in the time series behavior. Consequently, prediction of a chaotic time series is only feasible for a relatively short time before the appearance of orbital instability. For this reason, forecasting chaotic time series has remained a difficult task for the last few decades. Data-driven approaches, and machine-learning (ML) techniques in particular, have recently become the main approaches used for time-series forecasting ( Ahmed et al., 2010 ; Ben Taieb et al., 2012 ; Chandra et al., 2021 ; Chattopadhyay et al., 2020 ; Cheng et al., 2015 ; De Gooijer & Hyndman, 2006 ; Dubois et al., 2020 ; Kutz, 2013 ; Li et al., 2005 ; Tealab, 2018 ). In particular, recurrent neural networks (RNNs) are the mainstream architecture for analyzing sequential data, owing to their ability in interpreting temporal dependencies in the input time series ( Chandra et al., 2021 ; Elman, 1990 ; Elman & Zipser, 1988 ; Schmidhuber, 2015 ). The recurrent connections in such networks serve as a notion of memory, allowing them to embed temporal information. Despite the success of RNNs in modeling short-term temporal data and non-chaotic dynamical systems, the high computational cost of back-propagation through time and their vulnerability to the vanishing or exploding gradient problems have limited their applications. Gated RNN architectures were introduced to address some of these problems. More precisely, the memory cell architecture and the gating mechanism enable these networks to be more selective over the information that needs be remembered or forgotten, thereby enabling them to learn long-term dependencies in temporal sequences. Long short-term memory (LSTM) networks ( Hochreiter & Schmidhuber, 1997 ) and gated recurrent units (GRUs) ( Chung et al., 2014 ) are among the most widely used gated RNNs. An alternative approach to deal with time-series forecasting and modeling dynamical systems is reservoir computing (RC), a learning paradigm mostly implemented as echo state networks (ESNs) ( Jaeger, 2002 ; Lukoševičius & Jaeger, 2009 ; Sun et al., 2020 ). The RC paradigm is fundamentally derived from RNN concepts offering a streamlined training process, which remains limited to obtaining the output layer weights, while the rest of the parameter values are set randomly and remain untrained. Notwithstanding such a major simplification, ESNs have successfully been employed for multi-step-ahead prediction of nonlinear time series and modeling chaotic dynamical systems at low computational cost ( Bianchi et al., 2017 ; Han et al., 2021 ), triggering the development of several network topologies in recent years. For instance, clustered ESNs (CESNs) ( Deng & Zhang, 2006 ; Junior et al., 2020 ), where multiple sub-graphs of sparsely connected hidden units form the reservoir, and deep ESNs, where the reservoir consists of multiple sub-reservoir layers stacked hierarchically ( Gallicchio & Micheli, 2017 ; Gallicchio et al., 2017 ), are two widely used architectures. Hybrid ESNs (HESNs) are another category of RC techniques introduced in a physics-informed ML framework ( Oh, 2020 ; Willard et al., 2020 ), where additional inputs from physics-based mathematical models integrate corresponding domain knowledge into data-driven models ( Doan et al., 2019 ; Pathak, Hunt, et al., 2018 ). The successful application of ESNs, despite their random construction, in forecasting complex dynamical systems using time-series data triggered a series of recent research providing an interpretation of how RC techniques function. Recently, Bollt demonstrated how the RC with linear activation functions and linear readout layer shares similarities with the well-studied vector autoregressive (VAR) concept, while using a quadratic readout can be interpreted as nonlinear VAR (NVAR) ( Bollt, 2021 ). Later, Gauthier et al. further studied this similarity and introduced the next generation RC, where instead of explicitly generating a reservoir of randomly connected neurons, an NVAR machine is formed in which the feature vector consists of time-delayed observations of the dynamical system and is augmented by nonlinear functions of these observations. Accordingly, with this approach there are fewer hyperparameters to tune and the intrinsic random nature of ESNs is effectively avoided. This approach was employed for one-step-ahead forecasting of benchmark chaotic time series for both reconstruction and cross-prediction tasks ( Gauthier et al., 2021 ). In this work, we assess the capability of the mainstream gated RNN techniques; ESN architectures, including the clustered architecture and the physics-informed hybrid approach; and the NVAR approach for multi-step-ahead prediction of nonlinear time series describing chaotic dynamical systems. In particular, we compare the performance of these models for forecasting two frequently used benchmark chaotic time series, derived from the Mackey–Glass and Lorenz dynamical systems, two additional chaotic times series derived from a bursting Morris–Lecar neuron model and the Vallis El Niño Southern Oscillation (ENSO) system, and one real-world dataset consisting of a time series of irregular cardiac voltage traces obtained in ex-vivo experiments in terms of the prediction error and computational efficiency. Moreover, this experimental dataset is further used to evaluate the performance of NVAR against traditional RC approaches in more detail. This paper is structured as follows. Section 2 presents a summary of the modeling approaches used for forecasting chaotic time series in this research and provides details about the implementation of each model and the evaluation metrics employed in this study. These methods are applied to datasets whose characteristics are described in Section 3 . The results are presented and discussed in Section 4 , and Section 5 presents concluding remarks."
} | 3,135 |
35755176 | PMC9230140 | pmc | 200 | {
"abstract": "In recent years, machine-learning techniques, particularly deep learning, have outperformed traditional time-series forecasting approaches in many contexts, including univariate and multivariate predictions. This study aims to investigate the capability of (i) gated recurrent neural networks, including long short-term memory (LSTM) and gated recurrent unit (GRU) networks, (ii) reservoir computing (RC) techniques, such as echo state networks (ESNs) and hybrid physics-informed ESNs, and (iii) the nonlinear vector autoregression (NVAR) approach, which has recently been introduced as the next generation RC, for the prediction of chaotic time series and to compare their performance in terms of accuracy, efficiency, and robustness. We apply the methods to predict time series obtained from two widely used chaotic benchmarks, the Mackey–Glass and Lorenz-63 models, as well as two other chaotic datasets representing a bursting neuron and the dynamics of the El Niño Southern Oscillation, and to one experimental dataset representing a time series of cardiac voltage with complex dynamics. We find that even though gated RNN techniques have been successful in forecasting time series generally, they can fall short in predicting chaotic time series for the methods, datasets, and ranges of hyperparameter values considered here. In contrast, for the chaotic datasets studied, we found that reservoir computing and NVAR techniques are more computationally efficient and offer more promise in long-term prediction of chaotic time series.",
"conclusion": "6. Conclusion In this paper, six different ML time-series forecasting approaches, including two gated RNN techniques, three variants of ESNs, and the NVAR approach, were tested to predict five chaotic time series, including the Mackey–Glass, Lorenz-63, bursting Morris–Lecar, Vallis ENSO, and experimental cardiac action potential time series. Although we considered relatively large but still limited numbers of datasets and methods, we found that the LSTM and GRU approaches, despite their high computational costs and in contrast to the ESN and NVAR methods, were incapable of forecasting the Mackey–Glass, Lorenz, and bursting Morris–Lecar time series more than a few steps into the future, and that increasing the network size did not significantly improve their performance. For the ENSO model, the GRU method could predict somewhat longer, but it did not compare favorably with the ESN and NVAR approaches. Three variants of ESNs were employed including the baseline ESN, the clustered ESN (CESN), and the hybrid physics-informed ESN (HESN). For the five datasets we used in this work, only one (ENSO) showed improvement by using a more complicated ESN architecture such as the clustered reservoir. In all the other cases, the baseline ESN demonstrated similar or better performance compared to CESN. In contrast, whereas the HESN provided the same level of prediction accuracy for the four synthetic time series, within the tested network sizes, it was the most successful approach for forecasting the experimental dataset, where it delivered more accurate predictions as measured by RMSE. Thus, incorporating the domain knowledge of a dynamical system if available may improve the prediction ability of the ESN technique and may help with obtaining good predictions using smaller network sizes. For the tested network sizes and datasets, the best prediction performance in the case of the Mackey–Glass, Lorenz, and bursting Morris–Lecar datasets was obtained by the NVAR method, which was recently introduced as the next generation of RC techniques and has been demonstrated to be as successful as optimized ESNs. For the ENSO dataset, NVAR’s prediction accuracy was only slightly lower than that of the most accurate method, CESN. A noticeable advantage of the NVAR technique over conventional ESNs is avoiding the explicit construction of randomly connected neurons and circumventing the intrinsic randomness that increases the sensitivity of the network to the hyperparameter values and initial parameters that remain untrained. Moreover, the number of hyperparameters is much smaller than for ESNs, which makes NVAR easier to tune. Such advantages may initially suggest that the amount of data required to train the NVAR model could be less than that needed for the conventional ESNs. However, our experiments showed in the case of the experimental cardiac voltage dataset, better performance was only obtained by embedding more delays and at the cost of more computational time and effort. Nevertheless, in general, this approach shows promise for efficient prediction of chaotic time series. To the best of our knowledge, this work is one of the first applications of this newly introduced technique to real-world experimental time series. Further studies in this area may reveal more of the potential of this approach. For instance, in this work, we used a quadratic polynomial functional to construct the nonlinear portion of the state vectors at each time step; however, other nonlinear functions such as higher-order polynomials could also be employed and studied. It should also be noted that our conclusions are based on a limited number of datasets and employed methods. Moreover, in each case, the optimum hyperparameters were obtained in a finite grid search process. Accordingly, it is possible that the same approaches could provide different results when applied to other datasets or when hyperparameters are determined using more extensive grids or different optimization techniques.",
"introduction": "1. Introduction Time series are important in many real-world applications, such as biology ( Bar-Joseph et al., 2003 ), finance ( Dingli & Fournier, 2017 ; Plagianakos & Tzanaki, 2001 ; Takahashi et al., 2019 ; Tsay, 2005 ; Zhao, 2009 ), climate science ( Ghil & Vautard, 1991 ), anomaly detection in computer networks ( Limthong, 2013 ) and social networks ( Gong et al., 2018 ), and energy ( Billinton et al., 1996 ; Bunn, 2000 ; Deihimi & Showkati, 2012 ). Accordingly, the analysis and prediction of time series data are of great importance and have been the focus of much research in the past few decades. In general, a time series represents a record of observations of a dynamical system at specific time intervals. Therefore, time series prediction involves determining the future evolution of a dynamical system, which can be especially challenging for chaotic dynamical systems. The states of such systems can be represented by chaotic time series, which are recognized by the orbital instability characteristic, where infinitesimal differences in the initial values bring about large differences in the time series behavior. Consequently, prediction of a chaotic time series is only feasible for a relatively short time before the appearance of orbital instability. For this reason, forecasting chaotic time series has remained a difficult task for the last few decades. Data-driven approaches, and machine-learning (ML) techniques in particular, have recently become the main approaches used for time-series forecasting ( Ahmed et al., 2010 ; Ben Taieb et al., 2012 ; Chandra et al., 2021 ; Chattopadhyay et al., 2020 ; Cheng et al., 2015 ; De Gooijer & Hyndman, 2006 ; Dubois et al., 2020 ; Kutz, 2013 ; Li et al., 2005 ; Tealab, 2018 ). In particular, recurrent neural networks (RNNs) are the mainstream architecture for analyzing sequential data, owing to their ability in interpreting temporal dependencies in the input time series ( Chandra et al., 2021 ; Elman, 1990 ; Elman & Zipser, 1988 ; Schmidhuber, 2015 ). The recurrent connections in such networks serve as a notion of memory, allowing them to embed temporal information. Despite the success of RNNs in modeling short-term temporal data and non-chaotic dynamical systems, the high computational cost of back-propagation through time and their vulnerability to the vanishing or exploding gradient problems have limited their applications. Gated RNN architectures were introduced to address some of these problems. More precisely, the memory cell architecture and the gating mechanism enable these networks to be more selective over the information that needs be remembered or forgotten, thereby enabling them to learn long-term dependencies in temporal sequences. Long short-term memory (LSTM) networks ( Hochreiter & Schmidhuber, 1997 ) and gated recurrent units (GRUs) ( Chung et al., 2014 ) are among the most widely used gated RNNs. An alternative approach to deal with time-series forecasting and modeling dynamical systems is reservoir computing (RC), a learning paradigm mostly implemented as echo state networks (ESNs) ( Jaeger, 2002 ; Lukoševičius & Jaeger, 2009 ; Sun et al., 2020 ). The RC paradigm is fundamentally derived from RNN concepts offering a streamlined training process, which remains limited to obtaining the output layer weights, while the rest of the parameter values are set randomly and remain untrained. Notwithstanding such a major simplification, ESNs have successfully been employed for multi-step-ahead prediction of nonlinear time series and modeling chaotic dynamical systems at low computational cost ( Bianchi et al., 2017 ; Han et al., 2021 ), triggering the development of several network topologies in recent years. For instance, clustered ESNs (CESNs) ( Deng & Zhang, 2006 ; Junior et al., 2020 ), where multiple sub-graphs of sparsely connected hidden units form the reservoir, and deep ESNs, where the reservoir consists of multiple sub-reservoir layers stacked hierarchically ( Gallicchio & Micheli, 2017 ; Gallicchio et al., 2017 ), are two widely used architectures. Hybrid ESNs (HESNs) are another category of RC techniques introduced in a physics-informed ML framework ( Oh, 2020 ; Willard et al., 2020 ), where additional inputs from physics-based mathematical models integrate corresponding domain knowledge into data-driven models ( Doan et al., 2019 ; Pathak, Hunt, et al., 2018 ). The successful application of ESNs, despite their random construction, in forecasting complex dynamical systems using time-series data triggered a series of recent research providing an interpretation of how RC techniques function. Recently, Bollt demonstrated how the RC with linear activation functions and linear readout layer shares similarities with the well-studied vector autoregressive (VAR) concept, while using a quadratic readout can be interpreted as nonlinear VAR (NVAR) ( Bollt, 2021 ). Later, Gauthier et al. further studied this similarity and introduced the next generation RC, where instead of explicitly generating a reservoir of randomly connected neurons, an NVAR machine is formed in which the feature vector consists of time-delayed observations of the dynamical system and is augmented by nonlinear functions of these observations. Accordingly, with this approach there are fewer hyperparameters to tune and the intrinsic random nature of ESNs is effectively avoided. This approach was employed for one-step-ahead forecasting of benchmark chaotic time series for both reconstruction and cross-prediction tasks ( Gauthier et al., 2021 ). In this work, we assess the capability of the mainstream gated RNN techniques; ESN architectures, including the clustered architecture and the physics-informed hybrid approach; and the NVAR approach for multi-step-ahead prediction of nonlinear time series describing chaotic dynamical systems. In particular, we compare the performance of these models for forecasting two frequently used benchmark chaotic time series, derived from the Mackey–Glass and Lorenz dynamical systems, two additional chaotic times series derived from a bursting Morris–Lecar neuron model and the Vallis El Niño Southern Oscillation (ENSO) system, and one real-world dataset consisting of a time series of irregular cardiac voltage traces obtained in ex-vivo experiments in terms of the prediction error and computational efficiency. Moreover, this experimental dataset is further used to evaluate the performance of NVAR against traditional RC approaches in more detail. This paper is structured as follows. Section 2 presents a summary of the modeling approaches used for forecasting chaotic time series in this research and provides details about the implementation of each model and the evaluation metrics employed in this study. These methods are applied to datasets whose characteristics are described in Section 3 . The results are presented and discussed in Section 4 , and Section 5 presents concluding remarks."
} | 3,135 |
27935590 | PMC5364354 | pmc | 201 | {
"abstract": "Methanotrophs play a key role in balancing the atmospheric methane concentration. Recently, the microbial methanotrophic diversity was extended by the discovery of thermoacidophilic methanotrophs belonging to the Verrucomicrobia phylum in geothermal areas. Here we show that a representative of this new group, Methylacidiphilum fumariolicum SolV, is able to grow as a real ‘Knallgas' bacterium on hydrogen/carbon dioxide, without addition of methane. The full genome of strain SolV revealed the presence of two hydrogen uptake hydrogenases genes, encoding an oxygen-sensitive ( hup -type) and an oxygen-insensitive enzyme ( hhy -type). The hhy -type hydrogenase was constitutively expressed and active and supported growth on hydrogen alone up to a growth rate of 0.03 h −1 , at O 2 concentrations below 1.5%. The oxygen-sensitive hup -type hydrogenase was expressed when oxygen was reduced to below 0.2%. This resulted in an increase of the growth rate to a maximum of 0.047 h −1 , that is 60% of the rate on methane. The results indicate that under natural conditions where both hydrogen and methane might be limiting strain SolV may operate primarily as a methanotrophic ‘Knallgas' bacterium. These findings argue for a revision of the role of hydrogen in methanotrophic ecosystems, especially in soil and related to consumption of atmospheric methane.",
"introduction": "Introduction Methanotrophs form an important sink for the greenhouse gas methane and they play a key role in keeping the atmospheric methane concentration in balance ( Murrell and Jetten, 2009 ). They limit the amount of biologically or geochemically produced methane escaping into the atmosphere. The most studied group comprises the aerobic methanotrophs that phylogenetically belong to the Gammaproteobacteria and the Alphaproteobacteria ( Hanson and Hanson, 1996 ). However, the methanotrophic diversity was extended when three research groups independently described novel thermoacidophilic methanotrophs isolated from geothermal areas in Italy, New Zealand and Russia ( Dunfield et al , 2007 ; Pol et al , 2007 ; Islam et al , 2008 ). These novel isolates represented a distinct phylogenetic lineage within the phylum Verrucomicrobia. They belong to a single genus for which the name Methylacidiphilum was proposed ( Op den Camp et al , 2009 ). Recently, a second genus of mesophilic, acidophilic verrucomicrobial methanotrophs was described, Methylacidimicrobium ( van Teeseling et al , 2014 ). In addition, data from cultivation-independent environmental studies ( Sharp et al , 2012 ; Sharp et al , 2014 ) indicated that methanotrophic Verrucomicrobia may be present in many more moderate-temperature volcanic ecosystems than assumed before. The discovery of a verrucomicrobial methanotroph was exciting, since for the first time the widely distributed Verrucomicrobia phylum, from which most members remain uncultivated, was coupled to a geochemical cycle ( Pol et al , 2007 ). Initial analyses of our isolate Methylacidiphilum fumariolicum strain SolV showed major differences with the classical methanotrophs, for example extreme acid tolerance, absence of typical membrane structures, distinct enzymes and use of rare earth elements for methane oxidation ( Pol et al , 2007 ; Pol et al , 2013 ; Keltjens et al , 2014 ). It has long been assumed that methanotrophs are very strict in their diet, consuming only methane or methanol and occasionally other C1 compounds ( Hanson and Hanson, 1996 ), despite the fact that most of the time a range of multi-carbon substrates is available in their environments ( Tsien et al , 1989 ). Stimulation of growth by multi-carbon substrates like acetate, malate and succinate has been reported already by Whittenbury et al , 1970 , but claims of growth on such substrates or even sugars for several isolated strains in subsequent years have been doubted (for a review see Semrau, 2011 ). Only in recent years convincing evidence has shown that acetate, pyruvate, ethanol, malate and succinate can be used for growth, but this seems restricted to alpha-proteobacterial methanotrophs ( Kelly et al , 2000 ; Kolb, 2009 ; Kappler and Nouwens, 2013 ; Knief, 2015 ). The ecological significance especially of acetate consumption would be the increased survival of these methanotrophs under natural conditions, where methane concentrations are very low and variable ( Belova et al , 2011 ). Moreover, methane oxidation might be possible at much lower concentrations when additional reducing equivalents from multi-carbon compounds are available, as speculated by Theisen and Murrell, 2005 . In many ecosystems another source of reducing equivalents is available as a potential energy source for methanotrophs, that is molecular hydrogen. The presence of putative hydrogenases has been detected in many different (sub)phyla; Proteobacteria, Firmicutes, Cyanobacteria, Aquificae, Euryarchaeota, Crenarchaeota and Verrucomicrobia ( Greening et al , 2016 ). Hydrogenase-encoding genes were also identified in the genomes of multiple proteobacterial obligate methane oxidizers and uptake of hydrogen has been reported for Methylosinus sp. de Bont (1976) , Methylocystis sp. and Methylococcus capsulatus Bath ( Csáki et al , 2001 ; Kelly et al , 2005 ). In methanotrophs it is believed that hydrogenases also help to ‘save' reducing equivalents by recycling the hydrogen gas that is produced during nitrogen fixation ( Bothe et al , 2010 ). Csáki and co-workers ( Csáki et al , 2001 ) showed increased H 2 production during nitrogen fixing conditions using a hup LS mutant of M. capsulatus Bath. Furthermore, soluble and membrane-bound hydrogenases have been reported for M. capsulatus Bath and hydrogen was shown to be able to supply reducing equivalents for the methane monooxygenase ( Hanczár et al , 2002 ). The fact that several proteobacterial methanotrophs contain genes encoding the large and small subunit of both uptake hydrogenases and the ribulose-1,5-bisphosphate carboxylase (Rubisco) suggests the possibility of autotrophic growth. But attempts to grow M. capsulatus Bath autotrophically (on hydrogen and carbon dioxide) in liquid media were not successful ( Dalton and Whittenbury, 1976 ; Taylor et al , 1981 ; Stanley and Dalton, 1982 ; Baxter et al , 2002 ). Autotrophic growth was observed on solid agar media, but no physiological studies have been reported to support this ( Baxter et al , 2002 ). M. fumariolicum SolV was shown to use the Calvin-Benson-Bassham (CBB) cycle for carbon fixation pathways ( Khadem et al , 2011 ), and is capable of fixing nitrogen gas ( Khadem et al , 2010 ). Key metabolic genes of the ribulose monophosphate and serine pathways are absent ( Pol et al , 2007 ). The full genome of strain SolV ( Anvar et al , 2014 ) revealed the presence of two hydrogenase types ( hup - and hhy -type). Many Group 1 hydrogenases within the hup type are known to support growth. While in general hydrogenases are very sensitive towards oxygen and function only in anaerobic respiration, the Group 1d hydrogenases are known for their relative oxygen tolerance and may support aerobic growth in ‘Knallgas' bacteria. The hhy type genes encode the recently discovered Group 5 hydrogenases, which are widespread in actinobacteria from soil and were supposed to be responsible for ‘high affinity' atmospheric hydrogen uptake ( Schäfer et al , 2013 ; Greening et al , 2014 ). The ‘Knallgas' bacterium Ralstonia eutropha also contains this type of hydrogenase that upon isolation appeared to be oxygen insensitive ( Schäfer et al , 2013 ). A role for this enzyme is as yet not established. When hydrogen was added to methane-consuming cultures of M. fumariolicum SolV, the hydrogen was also oxidized ( Pol et al , 2007 ). RNA-seq analysis of SolV cells at maximal growth rate (μ max ) compared to nitrogen fixing and oxygen limiting conditions showed up-regulation of the hup -type hydrogenase at low oxygen concentrations, whereas the hhy -type enzyme was constitutively expressed ( Khadem et al , 2012a ). The presence of the CBB cycle, together with the uptake hydrogenase prompted us to investigate the possibility to grow strain SolV as a real ‘Knallgas' bacterium without addition of methane. We used physiological experiments in combination with phylogenetic and transcriptome analysis to show that strain SolV can grow autotrophically on hydrogen and carbon dioxide.",
"discussion": "Discussion In their natural ecosystems the methanotroph Methylacidiphilum fumariolicum SolV and its close relatives encounter both methane and hydrogen under acidic (pH 1–2) and hot (50–70 °C) conditions with low availability of oxygen ( Chiodini et al , 2001 ; Pol et al , 2007 ). Physiological experiments with both batch and continuous culture of strain SolV described in this study showed that this bacterium can grow autotrophically on hydrogen and carbon dioxide, using the ‘Knallgas' reaction to gain energy and the Rubisco pathway for carbon fixation. No supply of methane is needed. When growing on methane, hydrogen oxidation was possible under ambient oxygen concentration, but growth on hydrogen and carbon dioxide was possible only below 1.5% oxygen. Highest growth rates were obtained when oxygen was very low (dO 2 <0.1%). The maximum specific growth rate on hydrogen (0.047 h −1 ) is about 60% compared to the growth rate on methane ( Pol et al , 2007 ). Assuming the same CO 2 :H 2 consumption ratio as measured at D=0.0124 h −1 , such a growth rate needs a specific hydrogenase activity of about 175 nmol H 2 .min −1 .mg DW −1 (the carbon content of DW was 51%). This matches well with the in vitro measured maximum hydrogen uptake rates (220 nmol H 2 .min −1 .mg DW −1 ). The measured yield on hydrogen is about half that on methane (3.4 vs 6.4 g DW/mole CH 4 ) ( Pol et al , 2007 ) and slightly lower than those reported for ‘Knallgas' bacteria like Ralstonia eutropha (4.6 g/mole H 2 ) ( Morinaga et al , 1978 ) or Hydrogenomonas eutropha (5 g/mole H 2 ) ( Bongers, 1970 ). Continuous cultures fed with hydrogen under low oxygen conditions remained stable for more than a year. Although previous studies have shown that co-oxidation of hydrogen together with methane was possible ( de Bont, 1976 ; Hanczár et al , 2002 ), to our knowledge this is the first time that a methanotroph is shown to grow on hydrogen/carbon dioxide only in liquid mineral medium. Baxter et al (2002) reported growth with hydrogen and carbon dioxide on agar plates but not in liquid cultures for several methanotrophs but autotrophy was not substantiated by physiological and biochemical evidence. The presence of a fully operational Calvin cycle with Rubisco as the key enzyme in verrucomicrobial methanotrophs seems to be a pre-requisite for autotrophic growth on hydrogen. Recently, it has been shown that microorganisms from a wide diversity of ecosystems, ranging from the hypoxic hydrogen-rich habitats of animal guts and bog soils to aerated soils and waters containing small quantities of H 2 , possess genes encoding different types of hydrogenases ( Greening et al , 2016 ). We showed that Methylacidiphilum sp. and Methylacidimicrobium sp. representing the acidophilic verrucomicrobial methanotrophs ( Op den Camp et al , 2009 ; Sharp et al , 2014 ; van Teeseling et al , 2014 ) contain genes encoding hydrogen uptake hydrogenases. According to the phylogeny of both the large and small subunits, the hup -type hydrogenases in the verrucomicrobial methanotrophs belong to the membrane-bound H 2 -uptake [NiFe]-hydrogenases (Group 1d), and the hhy -type hydrogenases (present only in strains SolV and Kam1) belong to Group 1 h/5 hydrogenases as they are not membrane anchored ( Greening et al , 2016 ). Based on the phylogenetic analysis, we showed that several alpha- and gammaproteobacterial methanotrophic species contain hydrogenase genes, and growth with hydrogen and carbon dioxide was reported on agar plates but not in liquid cultures for several methanotrophs ( Baxter et al , 2002 ). In methanotrophic bacteria that cannot fix CO 2 , hydrogenases may help to save reducing equivalents by recycling the hydrogen gas that is produced during nitrogen fixation, or supply reducing equivalents for the methane monooxygenase ( Csáki et al , 2001 ; Hanczár et al , 2002 ; Bothe et al , 2010 ). The small subunits of the hup -type hydrogenases contain the typical twin-arginine signal peptide distinctive for periplasmic membrane-bound Group 1 hydrogenases ( Palmer and Berks, 2012 ). The hup -type hydrogenases of the thermophilic strains have features typical for Group 1d. Their large subunit contains the typical L1 (xxRICGVCTxxH) and L2 (SFDPCLACxxH) motifs and their small subunit contains the conserved motif with two extra cysteines ( Greening et al , 2016 ). In Ralstonia eutropha these extra cysteines have been shown to ligate to an ‘exclusive' proximal 4Fe3S cluster that is supposed to render hydrogenases extra oxygen tolerance ( Fritsch et al , 2011 ). In contrast, HupS subunits of the mesophilic strains do not contain these extra cysteines. Instead cysteines seem to be arranged to coordinate a 4Fe4S proximal cluster like most Group 1 members (not being the Group 1d type), although one of the four cysteines is replaced by an asparagine. A similar change is also observed in Group 1f hydrogenases of Roseoflexis , Geobacter and Frankia spp. where the cysteine is replaced by asparagine or aspartic acid. Asparagine coordination to 4Fe4S is also present in Group 2a hydrogenases. Conserved cysteines for the medial FeS cluster (3Fe4S coordinating with three cysteines, as in Group 1b–g) and distal FeS cluster (4Fe4S, coordinating with three cysteines and one histidine, as in Group 1a–h) are present in the HupS of both mesophilic and thermophilic strains. The Group 1d hydrogenase in strain SolV was shown to be responsible for increased growth rates when dO 2 level was below 0.2% (see above). Furthermore, we showed that genes encoding an additional hydrogenase are present in the thermophilic strains SolV and Kam1. Based on the phylogenetic analysis of both its small and large subunits, this hydrogenase belongs to Group 1h/5 hydrogenases. They contain active site L1 (TSRICGICGDNH) and L2 (SFDPCLPCGVH) motifs that are highly conserved in Group 1h/5 hydrogenases and distinct from other Group 1–4 hydrogenases ( Greening et al , 2016 ). These hydrogenases are located at the cytoplasmic side of the membrane, since they lack the twin-arginine signal peptide in the small subunit. The small subunit also lacks the C-terminal membrane span alpha helix extension to anchor in a membrane. The proximal FeS cluster is reminiscent to the mesophilic counterpart (discussed above) and others in Group 1b, f, g in that it has only three cysteines available for coordination. The fourth position for a 4Fe4S cluster would then be the conserved Asp ( de Bont, 1976 ) as shown for the Group 1h/5 hydrogenase of Ralstonia eutropha ( Schäfer, 2014 ) and could be a possible clue to the extreme oxygen tolerance of group 1h/5 hydrogenases ( Supplementary Figure S5 ). The transcriptome analysis clearly showed that the Group 1d hydrogenase gene cluster was upregulated under oxygen-limited conditions consistent with higher hydrogenase activity and growth rates found at dO 2 values below 0.2%. Previously, we showed that these two genes were also upregulated under the N 2 -fixing condition ( Khadem et al , 2012a ). In contrast, the other hydrogenase encoding gene cluster (Group 1h/5) and all accessory proteins were similarly expressed under oxygen-limited conditions (continuous culture) and under oxygen excess conditions (batch culture). This clearly correlates with the constant hydrogen uptake (activity) found in respiration experiments for cells grown under higher oxygen concentrations (above 0.2% O 2 ). Therefore we may conclude that the observed growth with hydrogen found at these higher dO 2 values is supported exclusively by the Group 1h/5 uptake hydrogenase. Growth on hydrogen only sustained by a Group 1h/5 hydrogenase has not been reported before. Other microorganisms expressing Group 1h/5 hydrogenases activity co-metabolize hydrogen but have not been shown to grow on or benefit from this hydrogen consumption. Group 1h/5 hydrogenase activity has been reported for Actinobacteria like Mycobacterium smegmatis ( Greening et al , 2014 ; Berney et al , 2014 ) (2.5 nmol.min −1 .g DW −1 ; 10 μmol.min −1 .g protein −1 ), and Streptomyces strains ( Constant et al , 2010 ) (0.2–2 nmol.min −1 .g DW −1 ), whereas Meredith et al (2014) reported extremely higher V max values of 14–180 μmol.min −1 .g protein −1 for Streptomyces species. ‘Knallgas' bacteria like Ralstonia sp. need the Group 1d hydrogenase for growth and express relatively low Group 1h/5 hydrogenase activity ( Schäfer et al , 2013 ). Although the Group 1h/5 hydrogenase of strain SolV was active up to at least ambient O 2 , growth on hydrogen was only observed below 1.5% O 2 . Clearly other factors also determine the oxygen tolerance during growth on hydrogen. The affinity for hydrogen of this enzyme (K s 0.6 μM) was lower than those reported for Group 5 hydrogenases of M. smegmatis (50 nM) ( Greening et al , 2014 ) and Streptomyces species (10 nM) ( Constant et al , 2008 ); (40–400 nM) ( Constant et al , 2010 )). These Actinobacteria are held responsible for the high-affinity hydrogen uptake from the atmosphere ( Constant et al , 2010 ; Meredith et al , 2014 ). The proteobacterial R. eutropha expresses only low Group 5 hydrogenase activity and showed a low affinity for hydrogen (K m 3.6 μM, after purification) ( Schäfer et al , 2013 ). The hydrogen kinetics of the Group 1d hydrogenase of M. fumariolicum SolV, which is active only at low oxygen concentrations, could not be determined due to the constitutive expression of Group 1 h/5 hydrogenase. However, cells with both hydrogenases active exhibited an affinity constant of about 1 μM and we estimated that the apparent affinity constant of the Group 1d hydrogenase was higher than 1 μM but less than 2 μM, similar to that of group 1 containing proteobacterial ‘Knallgas' bacteria ( Ludwig et al , 2009 ). It should be realized that when using whole cells (or soils) to measure hydrogenase kinetics, only a so-called apparent affinity constant is obtained, which depends on the measured V max . This V max is not necessarily the V max of the hydrogenase enzyme but that of the overall reaction that may be limited by the follow-up steps being respiration and this may explain the large variations found in values (nM to low μM range) of closely related strains that contain the same hydrogenase ( Greening et al , 2014 ). As a consequence the existence of a high affinity hydrogenase remains doubtful. The mesophilic Methylacidimicrobium strains lack a Group 1h/5 hydrogenase and an O 2 -tolerant Group 1d hydrogenase. They only possess a hydrogenase closely related to group 1b hydrogenases. These are found mainly in microorganisms with an anoxic H 2 respiration or photosynthesis, or supposed to be involved in protection against oxygen like in Geobacter sulfurreducens ( Coppi et al , 2004 ). Thus we assumed that these mesophilic strains only could grow on hydrogen when oxygen is limited. Preliminary tests to grow Methylacidimicrobium tartarophylax 4AC on hydrogen showed that the hydrogenase is inactivated when the sensor reading dO 2 value is above detection limit (0.01% oxygen) confirming the high sensitivity of this hydrogenase to oxygen. The presence of an oxygen-insensitive (group 1h/5) hydrogenase is of great importance for hydrogen consumption in environments like oxic upper soil layers that have been reported to take up hydrogen from the atmosphere. Although growth of methanotrophic Verrucomicrobia on ppm level ambient hydrogen may not be likely, the simultaneous uptake of hydrogen at such low concentrations may occur while these bacteria grow on methane and possibly hydrogen produced in anoxic layers beneath. Perhaps more interestingly vice versa: reducing power of hydrogen may lower the threshold for oxidizing methane as described for methanol ( Benstead et al , 1998 ; Jensen et al , 1998 ). This threshold has been suggested to be determined by the availability of reductant needed for the first step in methane oxidation. This step is performed by a methane monooxygenase and hydrogen has been shown to be a suitable donor of reducing equivalents ( Hanczár et al , 2002 ). In the volcanic environment from where M. fumariolicum SolV was isolated, hydrogen concentrations in emitted gasses are much higher than those of methane ( Chiodini et al , 2001 ) and this hydrogen might explain the uptake of methane even at atmospheric levels that was reported by Castaldi and Tedesco (2005) . Greening and co-workers ( Greening et al , 2016 ) showed that the occurrence of [NiFe] aerobic H 2 -uptake hydrogenases in the phylum Verrucomicrobia is more widespread, and suggests that Verrucomicrobia play a role in the hydrogen cycle not only in areas that emit geothermal gasses, but also in ecosystems such as rice paddy soil ( van Passel et al , 2011 ), pasture soil ( Kant et al , 2011 ) and the human gut ( Derrien et al , 2004 ). In conclusion, M. fumariolicum SolV is a real ‘Knallgas' bacterium but also able to consume methane and hydrogen simultaneously using an oxygen-sensitive and oxygen-tolerant hydrogenase, respectively. Detailed physiological studies and transcriptome analysis of the Methylacidiphilum strains that can grow on hydrogen is required to understand this interesting group of hydrogenases. Oxygen sensitivity of hydrogenases is a major drawback in application ( Jugder et al , 2016 ). In view of the high oxygen tolerance of the Group 1 h/5hydrogenase from strain SolV, biochemical and biophysical studies of the purified hydrogenase are essential since this enzyme has a high potential in biotechnological applications."
} | 5,538 |
38283867 | PMC10821171 | pmc | 202 | {
"abstract": "At the sediment-water interfaces, filamentous cable bacteria transport electrons from sulfide oxidation along their filaments towards oxygen or nitrate as electron acceptors. These multicellular bacteria belonging to the family Desulfobulbaceae thus form a biogeobattery that mediates redox processes between multiple elements. Cable bacteria were first reported in 2012. In the past years, cable bacteria have been found to be widely distributed across the globe. Their potential in shaping the surface water environments has been extensively studied but is not fully elucidated. In this review, the biogeochemical characteristics, conduction mechanisms, and geographical distribution of cable bacteria, as well as their ecological effects, are systematically reviewed and discussed. Novel insights for understanding and applying the role of cable bacteria in aquatic ecology are summarized.",
"introduction": "1 Introduction Microorganisms are key ecosystem agents for biogeochemical elemental cycling, greenhouse gas generation, environmental pollutants degradation, and human health protection [ 1 , 2 ]. These functions are ultimately linked to microbial redox processes [ [3] , [4] , [5] ]. In water-sediment systems, the diffusion of dissolved oxygen (DO) results in a redox potential gradient with an oxidizing environment at the surface of the sediment and a reducing anoxic environment below. A low electrical current can be detected by connecting the deep sediment and the overlaying water using metal wires [ 6 ]. Notably, recent advancements have confirmed the occurrence of long-distance electrical currents (LDET) involving the coupling of sulfide oxidation with oxygen reduction at the surface of marine sediments (i.e., electrogenic sulfide oxidation (e-SO x )) [ 7 , 8 ]. Subsequent research has established that these currents were microbially mediated by a new type of multicellular filamentous bacteria, cable bacteria. These unique organisms form networks of bacterial filaments, connecting thousands of single cells end to end and spanning centimeters from the oxic surface to the anoxic subsurface of the sediment ( Fig. 1 ) [ 8 , 9 ]. Fig. 1 The e-SO x process of cable bacteria. Cable bacteria oxidize sulfide to sulfate, channeling electrons upward to oxygen (or nitrate) via conductive fibers connecting each single cell. The illustrated depiction on the right side expounds upon the proposed mechanisms of sulfide oxidation and electron transfer within a singular cell. SRB, sulfate-reducing bacteria; SQR, sulfide-quinone oxidoreductase; PSR, polysulfide reductase; Dsr, dissimilatory bisulfite reductase; Apr, adenosine phosphosulfate reductase; APS, adenosine-5-phosphosulfate; Sat, sulfate adenylyltransferase. Fig. 1 The electron transfer capability of cable bacteria, extending over centimeter ranges, surpasses any observed in organisms to date [ 10 ]. This remarkable phenomenon involves the guided current along cells via fibers in a common periplasm of the entire cable bacterium [ 11 ]. Although the molecular composition remains unclear [ 8 , 10 , 12 , 13 ], such long-range electron transfer adds a new dimension to our understanding of biogeochemistry and microbial ecology in aquatic sediments [ 7 , 8 ]. A growing number of findings suggest that cable bacteria are widely distributed in sedimentary environments, such as offshore oceans, mangrove wetlands, freshwater bodies, and underground aquifers [ [14] , [15] , [16] , [17] ]. Notably, cable bacteria actively influence the redox conditions of the aqueous phase covering the sediment [ 18 , 19 ]. For example, cable bacteria regulate the cycling of sulfur, phosphorus [ 20 ], manganese [ 21 ], and iron in seasonal hypoxic environments [ 18 ]. Furthermore, cable bacteria can also suppress methane release and serve as an electrical connection to oxygen for other bacteria flocking around them [ [22] , [23] , [24] ]. Moreover, filamentous microorganisms, putatively cable bacteria, have been seen to cover the deap sea sulfide chimneys [ 25 ]. Therefore, speculation arises regarding the potential of cable bacteria and many other electroactive microorganisms to support biogeochemical cycles in aquatic environments by forming a widespread redox network, especially for cable bacteria through LDET processes from the sediment's depths to its surface [ 22 ]. This review summarizes the current knowledge about the niches and functions of cable bacteria in aquatic ecosystems. Additionally, we delve into the key metabolism and LDET mechanisms of cable bacteria. Conclusively, we outline the research gaps and questions that need to be addressed to better understand or apply cable bacteria in aquatic environments."
} | 1,175 |
36342556 | PMC9640510 | pmc | 203 | {
"abstract": "Optoelectronic memristor is a promising candidate for future light-controllable high-density storage and neuromorphic computing. In this work, light-tunable resistive switching (RS) characteristics are demonstrated in the CMOS process-compatible ITO/HfO 2 /TiO 2 /ITO optoelectronic memristor. The device shows an average of 79.24% transmittance under visible light. After electroforming, stable bipolar analog switching, data retention beyond 10 4 s, and endurance of 10 6 cycles are realized. An obvious current increase is observed under 405 nm wavelength light irradiation both in high and in low resistance states. The long-term potentiation of synaptic property can be achieved by both electrical and optical stimulation. Moreover, based on the optical potentiation and electrical depression of conductances, the simulated Hopfield neural network (HNN) is trained for learning the 10 × 10 pixels size image. The HNN can be successfully trained to recognize the input image with a training accuracy of 100% in 13 iterations. These results suggest that this optoelectronic memristor has a high potential for neuromorphic application.",
"conclusion": "Conclusions In summary, the fully CMOS process-compatible ITO/HfO 2 /TiO 2 /ITO optoelectronic synaptic memristor was fabricated. High transmittance under visible light was realized to ensure photosensitization. Stable bipolar analog switching, beyond 10 4 s data retention, and endurance of 10 6 cycles were achieved as basic storage function. Synaptic functions including LTP, LTD, and photonic potentiation were established. The light-tunable behavior originates from light irradiation-induced Vo 2+ . Furthermore, after 13 cycles of iteration, the simulated HNN can successfully recognize the 10 × 10 pixels size image. This memristor shows great potential in the next generation of intelligent optoelectronic neuromorphic computing systems.",
"introduction": "Introduction Vast amounts of data storage and rapid information processing are desired nowadays [ 1 , 2 ]. With the gradual failure of Moore’s law and the limitation of the von Neumann bottleneck, the revolutionary computing technique, neuromorphic computing is developed as the next-generation computing system due to its high-efficient information processing with low power consumption [ 3 – 5 ]. In a neuromorphic computing system, the synapses are crucial for connecting neurons and enabling the brain to function; an efficient artificial synapse is the core component [ 6 , 7 ]. The two-terminal memristor is a promising candidate as an artificial synapse due to its compact synapse-like structure, low power consumption, high durability, easy integration, and unique nonlinear characteristic [ 8 ]. In general, most artificial neuromorphic computing systems are based on electrically excited memristors, which are limited by package density, parallel operation, and bandwidth [ 9 , 10 ]. The operating speed of electronic memristors is limited by the trade-off between bandwidth and interconnection density. Compared with electrical tuning, optical control is a simple and low power consumption method to store and process data in an unprecedented bandwidth and high-speed optical way [ 11 – 14 ]. It can achieve programming by converting light information into an electric response [ 15 ]. However, there are remaining challenges, for example, process issues. Fully CMOS process-compatible optoelectronic memristors were rarely reported. In addition, most optoelectronic memristors show a nonvolatile light-induced current decrease phenomenon under visible light [ 16 – 19 ]. However, in this study, photonic current potentiation is realized under 405 nm light irradiation in the fully CMOS process-compatible ITO/HfO 2 /TiO 2 /ITO optoelectronic memristor. Neuromorphic computing is also investigated in this device by presenting an online learning pattern recognition.",
"discussion": "Results and Discussion The RS characteristics of the device are depicted in Fig. 2 . An electroforming operation is required for the fresh device to initialize the subsequent RS behavior. When a positive sweeping voltage (0 → 10 V) with 1 mA compliance current ( I CC ) is applied during the electroforming process, the current gradually increases at about 6 V and reaches I CC ; thus, the device turns to the low resistance state (LRS), as shown Fig. 2 a. The electroforming voltage is a little high and can be decreased or even forming-free by decreasing the deposition film thickness [ 20 ] of HfO 2 layer or using metal doping [ 21 , 22 ]. After electroforming, under a negative sweeping voltage (0 → − 1.7 V), namely RESET process, the current gradually decreases, demonstrating that the device turns from LRS to a high resistance state (HRS). Then, under the SET (0 → 1.7 V) and RESET (0 → − 1.7 V) processes, the device can switch repeatedly between LRS and HRS with 1 mA I CC , as shown in Fig. 2 b. Both SET and RESET processes are analog switching, which is beneficial for neuromorphic computing [ 23 ]. The electrical switching phenomenon can be attributed to the formation of oxygen vacancies (V O 2+ ) conductive filaments during electric stimulation [ 23 ]; such a switching mechanism is widely accepted for explaining the conduction phenomenon of the memristors. The big difference between forming voltage and set voltage can be explained as follows: During the forming process, a positive voltage is applied to the ITO top electrode (TE), the oxygen ions (O 2− ) move toward TE and store in the TE, and the Vo 2+ -based conductive filament would be formed at HfO 2 /TiO 2 resistive layer and grow up to connect TE and bottom electrode (BE). The device turns to a low resistance state (LRS). This process needs a high voltage to cause a soft breakdown and generate the point defect of Vo 2+ due to the high resistance of the pristine device. During the reset process, a negative voltage is applied to the TE, and the O 2− ions move from the TE to the BE. The O 2− would combine with the Vo 2+ in the resistive layer to disrupt the conductive filament. The device is changed to a high resistance state (HRS). It is worth noting that only a part of the conductive filament, which exists at the near HfO 2 /TiO 2 interface, would be broken to achieve HRS [ 23 ]. Therefore, during the next set process, it only needs a much lower voltage to fix this part of the filament to provide LRS. Fig. 2 a Electroforming process of the device. b I–V curve. c Retention of the device at room temperature. d Endurance plot. e Set and reset voltage distributions. f On and off current distributions The retention characteristics of the device on LRS and HRS states are investigated, as shown in Fig. 2 c. The resistance values of both states maintain stability and show no obvious shift beyond 10 4 s. The endurance performance was also studied, and the result is indicated in Fig. 2 d. The endurance test shows that the switching characteristic of the device does not have any degradation with 10 6 switching cycles. The cycle-to-cycle variability of V set , V reset , I off , and I on is shown in Fig. 2 e, f, respectively. V set and V reset are extracted from 100 switching cycles, while I off and I on are extracted from 1000 switching cycles. These results show the extremely narrow distribution of operating voltage and current, meaning that the device shows excellent cycle-to-cycle uniformity. With low variability, the conductance of the memristor will be programmed precisely in the neural network, and calculation and iteration will be more efficient, which can achieve high accuracy and need fewer train epochs to compute. Biological synapses are the information transmission centers between pre-neurons and post-neurons, and the transmission process is completed by the transmission of neurotransmitters, which are between the presynaptic membrane and the postsynaptic membrane [ 24 ]. The spike potential or action potential of presynaptic neurons can be transmitted through synapses to generate postsynaptic potentials. The amplitude of postsynaptic potentials depends on the weight of the synapse [ 25 ]. Adjustable resistance allows the memristor to mimic the typical synaptic response of the brain [ 26 ]. The schematic diagram of the synapse and the structure of the device are shown in Fig. 3 a. After the above electrical test, the device was used for mimicking long-term potentiation (LTP) and depression (LTD) synaptic behaviors. As shown in Fig. 3 b, set pulses (+ 0.95 V, 10 μs) are applied for potentiation and reset pulses (− 1.2 V, 10 μs) are employed for depression, with a reading pulse (0.1 V, 1 ms). After repeating the set pulse scheme 100 times, the conductance increases gradually, 1.8 coefficient of nonlinearity (NL) potentiation is realized. Then, following the 100 times reset pulse scheme, the conductance decreases gradually and 0.54 coefficient of NL depression is revealed, as shown in Fig. 3 c. In addition, the device can be trained more than 50 stable epochs without degrading the dynamic range (1.04–1.12 mS), as shown in Fig. 3 d. These results indicate the potential of this memristor for neural network applications [ 27 , 28 ]. Fig. 3 a Schematic diagram of the synapse and the structure of the ITO/HfO 2 /TiO 2 /ITO synaptic device. b Pulsing schemes for potentiation and depression, respectively. c Gradual conductance modulation of potentiation and depression under successive pulse stimulation. d Stable 50 epoch potentiation and depression trainings Optoelectronic memristor has opened up one way for light-tunable synaptic weight to further transmit and process stimulus information [ 29 ]. The light-tunable synaptic activities are investigated as follows. As shown in Fig. 4 a, the initial current is about 10.5 μA on HRS. When a 405 nm, 100 mW/cm 2 light pulse is applied from 16 to 46 s, light information could be perceived and the current gradually increases to 14 μA. On LRS, the current also increases from 97 to 102 μA with 180 s light irradiation, as shown in Fig. 4 b. The transition time and switching energy efficiency under illumination are not good enough, but they can be improved by doping modification [ 30 – 34 ]. The photoresponse current stems from the light irradiation-induced oxygen vacancies (Vo 2+ ), which will be discussed later. For both LRS and HRS, with stronger optical pulse intensity, more e-/Vo 2+ pairs will be generated, and the accumulated Vo 2+ can form conductive filaments to increase the conductance. The difference between LRS and HRS during illumination is that the current amplification on LRS is less than that on HRS since there were already existing many Vo 2+ on LRS before illumination. The amount of light-induced Vo 2+ is relatively fewer compared with already existing Vo 2+ ; thus, the increase in current on LRS is less than that on HRS. With the same initial current level and illumination time/rise time, stronger optical pulse intensity will induce a higher maximum/final potentiation current. In other words, with the same initial current level and stronger optical pulse intensity, it needs a shorter illumination time/rise time to achieve the same maximum/final potentiation current. After removing the light, the maximum current will decay to the final current during the falling time, which is related to the spontaneous physical diffusion of Vo 2+ conductive filaments component into the switching layer, driven by interfacial-energy-related Gibbs–Thomson effect [ 35 ] and Rayleigh instability of nanosize CFs [ 36 ]. To minimize the interfacial energy, the filaments component slowly diffuses to the minimum energy positions and merges into larger clusters. The driving force for this process is the chemical potential gradient induced by a perturbation in the radius. The instability can be modeled by introducing a sinusoidal perturbation with a form \\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}$$r = r_{0} + \\delta \\sin (2\\pi z/\\lambda )$$\\end{document} r = r 0 + δ sin ( 2 π z / λ ) on the surface of cylindrical CF, 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}$$r_{0}$$\\end{document} r 0 is the initial CF’s radius, \\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}$$\\delta$$\\end{document} δ 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}$$\\lambda$$\\end{document} λ are the amplitude and wavelength of the perturbation, respectively, and z is the coordinate along the CF’s axis. The cylindrical CF will become unstable 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}$$\\lambda > 2\\pi r_{0}$$\\end{document} λ > 2 π r 0 . At a certain wavelength \\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}$$\\lambda_{m} = 2\\sqrt 2 \\pi r_{0}$$\\end{document} λ m = 2 2 π r 0 , there is a minimum characteristic time of perturbation \\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}$$(\\tau_{m} )$$\\end{document} ( τ m ) , which corresponds to the CF’s relaxation time from the initial cylinder to its final shape [ 36 ]. The multilevel storage capacity of a memristor under light irradiation is vital for a light-in-memory computing system, and the related result of the investigation is shown in Fig. 4 c. The current increases under various times of duration (2, 3, 6, 14, 28, 77 s). After removing the light, the current will not return to the initial state immediately but remains at a higher level. During the decay process/fall time, the current obeys this formula: 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}$$I = Ae^{( - t/\\tau )} + I_{{{\\text{final}}}}$$\\end{document} I = A e ( - t / τ ) + I final Fig. 4 Current response under the application of 405 nm wavelength light irradiation on a HRS and b LRS. c Multilevel storage realized by light irradiation. d Optical potentiation and electrical depression characteristics simulated by optoelectronic artificial synapse where A is the part of the unstable state current, which will be dissipated after a period of time when the light is off. The value of A depends on the light intensity; A will increase with stronger light intensity. \\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}$$I_{{{\\text{final}}}}$$\\end{document} I final is the final state current after removing the light, which depends on the filament morphology. \\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}$$A + I_{{{\\text{final}}}}$$\\end{document} A + I final means the initial current on the moment when the light is off. \\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}$$\\tau$$\\end{document} τ is the average time it takes for unstable state current to be dissipated, which means the time at which 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}$$e^{ - t/\\tau }$$\\end{document} e - t / τ is reduced to 1/ e . \\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}$$\\tau$$\\end{document} τ is related to the diffusion kinetics. Take the decay process from 350 to 545 s ( t = 350–545) as an example; \\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}$$A$$\\end{document} A , \\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}$$\\tau$$\\end{document} τ , 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}$$I_{{{\\text{final}}}}$$\\end{document} I final are fitted as 9.63123E−4 A, 51.32848 s, and 1.65104E−5 A, respectively. As shown in Fig. 4 c, 7 stable states are realized after different illumination times, indicating that the memristor has the capability of in situ optical sensing and storage. By designing successive light and electrical pulses, the potentiation and depression behaviors of artificial synapses can be simulated, as shown in Fig. 4 d. The current of the memristor increases during light irradiation (405 nm, 100 s) and decreases during the electrical pulse (− 1.7 V, 10 μs, 1000 times), corresponding to the LTP/LTD characteristics of a dynamic range of 26–36 μA. These results show that this memristor can simulate basic synaptic functions under external light signals. Information processing, such as learning, is vital to biological systems [ 37 ]. Tunable memristor conductance can simulate continuous modulated synapse weight to achieve efficient neuromorphic calculation and recognition functions [ 38 ]. Based on the optical potentiation and electrical depression of conductances, we employed the Hopfield neural network (HNN) to investigate the pattern recognition capability of the device; the HNN is a form of recurrent ANN (Hopfield, 1982, and Little, 1974) [ 39 ]. The simulated Hopfield neural network (HNN) is trained to learn the 10 × 10 pixels size image, as shown in Fig. 5 a. We used relative normalized memristor conductance of optical potentiation and electrical depression to carry out the weight map simulation; each pixel represents the conductance of a single synapse. Initially, each synapse is randomized to store information in the range between 0 (yellow color) and 1 (blue color) (Fig. 5 b) to form the noisy image. Then, the value of each pixel will be updated during the learning process. The outcomes of the images after 5 and 13 cycles are depicted in Fig. 5 c, d, respectively. The HNN can be successfully trained to identify the input image in 13 cycles of iteration (Fig. 5 e) with 100% accuracy. In general, the results show that the ITO/HfO 2 /TiO 2 /ITO device can be possibly used for neuromorphic applications. Fig. 5 Pattern recognition simulation. a Input image of 10 × 10 pixels size for recognition. b Noisy image after weight updating. c Recalled image after 5 cycles of iteration. d Recalled image after 13 cycles of iteration. e The evolution of accuracy versus number of iterations A detailed comparison between previously reported photoelectronic artificial memristor synapses and the present device is provided in Table 1 . On the whole, the comparison indicates that the performance of the present device is better than those of the previously reported synaptic devices. The light-tunable mechanism could be explained by the light irradiation-induced oxygen vacancies (Vo 2+ ) in the TiO 2 layer [ 30 ], which form conductive filaments to increase the conductance. As shown in Fig. 6 a, the energy band gap of TiO 2 is 3 eV, and the energy of 405 nm light is 3.0612 eV, which is higher than the TiO 2 energy band gap. Under 405 nm light irradiation, the electrons of neutral lattice oxygen will gain energy hγ , activated to the conduction band. This action leaves movable O 2− (counterpart Vo 2+ ) in TiO 2 film as shown in Fig. 6 b. The oxygen ions are combined into oxygen gas, contributing to generating more Vo 2+ . With continuous irradiation, Vo 2+ would accumulate in the TiO 2 layer. With enough amount of Vo 2+ , they will aggregate and form conductive filaments, thus realizing an optical conductance increase. The HfO 2 layer has no current response under 405 nm light irradiation since the energy band gap of HfO 2 is 5.7 eV, much higher than photon energy. Table 1 Comparison between previous artificial photoelectronic memristive synapses and this work Memristor # of conductance Endurance Retention time Light source In-sensor computing COMS compatible Refs. Ag/ZnO/ITO 25 500 10 4 s Visible light Yes No [ 40 ] W/MoS 2 /p-Si 20 15 150 s Ultraviolet Yes No [ 41 ] ITO/Nb:SrTiO 3 /Ag 100 NA 3 × 10 3 s Visible light Yes No [ 42 ] Al/TiN x O 2−x /MoS 2 /ITO 400 450 30 s Visible light Yes No [ 43 ] ITO/ZnO 1−x /AlO y /Al 30 1000 500 s Ultraviolet Yes No [ 44 ] ITO/HfAlO/TiN-NP/HfAlO/ITO 100 175 NA Ultraviolet Yes No [ 45 ] Al/TiS 3 /ITO 50 100 10 4 s Visible light Yes No [ 46 ] ITO/HfO 2 /TiO 2 /ITO 100 1000 16,300 Visible light Yes Yes This work Fig. 6 Light-tunable mechanism of the device. a Energy band diagram of TiO 2 film. b Formation of V O 2+ in TiO 2 film under the illumination of light"
} | 5,814 |
32431592 | PMC7214871 | pmc | 204 | {
"abstract": "Spiking neural networks (SNNs) are potentially highly efficient models for inference on fully parallel neuromorphic hardware, but existing training methods that convert conventional artificial neural networks (ANNs) into SNNs are unable to exploit these advantages. Although ANN-to-SNN conversion has achieved state-of-the-art accuracy for static image classification tasks, the following subtle but important difference in the way SNNs and ANNs integrate information over time makes the direct application of conversion techniques for sequence processing tasks challenging. Whereas all connections in SNNs have a certain propagation delay larger than zero, ANNs assign different roles to feed-forward connections, which immediately update all neurons within the same time step, and recurrent connections, which have to be rolled out in time and are typically assigned a delay of one time step. Here, we present a novel method to obtain highly accurate SNNs for sequence processing by modifying the ANN training before conversion, such that delays induced by ANN rollouts match the propagation delays in the targeted SNN implementation. Our method builds on the recently introduced framework of streaming rollouts, which aims for fully parallel model execution of ANNs and inherently allows for temporal integration by merging paths of different delays between input and output of the network. The resulting networks achieve state-of-the-art accuracy for multiple event-based benchmark datasets, including N-MNIST , CIFAR10-DVS , N-CARS , and DvsGesture , and through the use of spatio-temporal shortcut connections yield low-latency approximate network responses that improve over time as more of the input sequence is processed. In addition, our converted SNNs are consistently more energy-efficient than their corresponding ANNs.",
"introduction": "1. Introduction Spiking neural networks (SNNs) were initially developed as biophysically realistic models of information processing in nervous systems (Rieke et al., 1999 ; Gerstner et al., 2014 ), but they are also ideally suited to process data from event-based sensors (Posch et al., 2010 ; Liu and Delbruck, 2010 ; Furber et al., 2013 ; O'Connor et al., 2013 ; Osswald et al., 2017 ), and are natively implemented on various neuromorphic computing platforms (Schemmel et al., 2010 ; Furber et al., 2013 ; Merolla et al., 2014 ; Qiao et al., 2015 ; Martí et al., 2015 ; Davies et al., 2018 ). Their sparse and event-driven mode of computation makes them more energy-efficient and faster compared to conventional artificial neural networks (ANNs), and additionally allows for the use of spatio-temporal spike codes to represent complex relationships between features in the network. These hypothetical advantages can, however, only be completely exploited on hardware that supports fully parallel model execution , which means that spiking neurons operate independently from each other and their update is solely based on incoming spikes. This is different from typical ANN execution schemes, which update all neurons in a fixed order determined by the network architecture and at fixed discrete time steps. The goal of this article is to develop a framework for obtaining SNNs that run fully in parallel and achieve high accuracy, low latency, and high energy-efficiency on sequence processing tasks, in particular classifying streams of events from neuromorphic sensors. Sequence processing seems to be a natural fit for the execution mode of SNNs where every neuron has its own dynamics, but in practice it has proven to be very challenging to exploit this property to train SNNs on temporally varying input data. Even more, current state-of-the-art methods for SNN training are unable to yield competitive accuracies compared to ANNs even in the simpler case of static inputs (Pfeiffer and Pfeil, 2018 ), albeit the gap has become narrower over the past years due to better training algorithms, such as e.g., variants of backpropagation for SNNs (Lee et al., 2016 ; Wu et al., 2018 ; Shrestha and Orchard, 2018 ; Neftci et al., 2019 ). However, Deng et al. ( 2020 ) argue that SNNs in general are put at a disadvantage in tasks designed for ANNs, such as image classification, because of the information loss incurred during conversion of images to spike trains of finite time window length. SNNs should not be expected to outperform ANNs in terms of accuracy on frame-based tasks, but they may be advantageous in terms of memory and compute costs. SNNs should ideally always be evaluated on event-based datasets, where they are able to outperform ANNs by exploiting the spatio-temporal information encoding of event-streams. Consequently, in this article we use only event-based datasets to evaluate our SNN performance and report memory and compute requirements for our networks, as suggested in Deng et al. ( 2020 ). The currently most successful method for obtaining accurate SNNs is to train an ANN with conventional deep learning methods, and convert the resulting ANN architecture and weights into an equivalent SNN, translating analog neuron activations into proportional firing rates of spiking neurons (Cao et al., 2015 ; Rueckauer et al., 2017 ). Conversion methods have achieved the best known SNN accuracies for image classification tasks, such as MNIST , but they rely on the assumption that input patterns do not change for some time. This is required because firing rates in each layer need time to converge to their targets derived from ANN activations. Spikes are allowed to propagate instantaneously between layers of the network, since this speeds up convergence of firing rates in deeper layers, and there is no additional temporal information beyond rates encoded in spike trains. These assumptions are no longer valid when sequence processing tasks are considered, which require networks capable of temporal integration. Temporal integration means that information from different times of the input has to be integrated at a single point in time at the output of the network. In a multi-layer network this means that the network architecture as well as the propagation delays between layers become crucial to control not just what features of the input are computed, but also when information computed in other layers can be used to update the feature computation. Temporal integration is achieved with recurrent or temporal skip connections , which not only skip layers in depth-direction of the network, but also bridge time like recurrent connections. Since temporal skip connections, in contrast to recurrent connections, serve as shortcuts in time, and hence, reduce the latency of early approximate network responses, we omit recurrent connections in the following. Our goal is to obtain SNNs for model-parallel execution on actual neuromorphic systems, which requires assigning non-zero delays to all connections in the network. However, current ANN-to-SNN conversion methods are unable to deal with the case of time-varying inputs or with temporal skip connections with different propagation delays. The main contribution of this paper is to close these gaps by unifying ANN-to-SNN conversion with the recently introduced concept of streaming rollouts (Fischer et al., 2018 ), thereby greatly extending the applicability of SNN training methods to novel and important classes of applications. Since the inference graph of an SNN determines the way temporal information is being processed, its temporal structure needs already to be taken into account during ANN training (see Section 2.2 for details). In other words, it has to be ensured that information from all required parts of the input sequence and the resulting activations of intermediate layers arrives at the right time at the output neurons both during ANN training and after conversion to SNNs. With this novel method for rolling out and training ANNs before conversion to SNNs we obtain SNNs that efficiently and accurately solve sequence processing tasks, and yield approximate responses as early as possible. In the following, we describe our methods in detail and show experimental results that emphasize the advantages of our approach for event-based sequence processing tasks.",
"discussion": "4. Discussion We have presented a novel way of training efficient SNNs for sequence processing via conversion from ANNs. The crucial observation is the connection between axonal delays in the SNN and the rollout strategy in the ANN. Streaming rollouts of ANNs are shown to be a particularly good fit, as they closely resemble the fully model-parallel execution in SNNs. To unify the two approaches, we introduced several additions to the existing conversion approach, such as a more general weight rescaling scheme, a new way to calculate predictions in the SNN, rescaling of average pooling layers and axonal delays. As a result, we make ANN-to-SNN conversion applicable in a principled manner to input signals changing over time, including general time series and the special case of event-based input data. Due to the fact that the streaming rollout imposes constraints on the ANN during training our approach can be interpreted as a “constrain-then-train” approach for SNNs (Esser et al., 2015 ; Pfeiffer and Pfeil, 2018 ), for which the superior training mechanisms available for ANNs are combined with the efficiency of SNN execution. We identify and highlight in our experiments particular advantages of applying conversion to rolled-out networks. Our proposed training and conversion scheme results in SNNs that efficiently integrate temporal information, provide early approximate network outputs, and achieve state-of-the art results on the N-MNIST , N-CARS , DvsGesture and CIFAR10-DVS datasets with smaller networks than other approaches, and with SNNs that are consistently more energy-efficient than their ANN counterparts. A uniform weighting of the network outputs in the loss function enables good early and late performance compared to other weighting patterns, such that even for the first network output, the prediction is significantly above chance level. Our framework is flexible enough to allow different trade-offs between early and late performance by choosing different weight factors a k . In this study, for the first time, streaming rollouts were applied to realistic and large-scale time series data, and were shown to be competitive with other approaches on multiple widely used event-based vision tasks (see Tables 1 – 4 ). Although we use only delays of one rollout frame in our experiments, in principle, arbitrary delays can be incorporated into the network rollouts. This principle is useful to convert advanced ANN architectures with temporal convolutions (van den Oord et al., 2016b ; Bai et al., 2018 ) that require multiple delays when rolled out. This is a big advantage over previous conversion approaches (e.g., Cao et al., 2015 ; Rueckauer et al., 2017 ), which do not take delays of connections into account. For purely feed-forward SNNs on suitable hardware (Farabet et al., 2012 ; Pérez-Carrasco et al., 2013 ) a pseudo-simultaneous spread of information, i.e., all delays in the network are zero, is advantageous, but causes de-synchronization if information needs to be integrated over time. Our approach generalizes the work of Diehl et al. ( 2016 ), who have shown a conversion approach for Elman-type recurrent networks using fixed delays in the recurrent layer and zero delays for feed-forward connections. Note that although our experiments only show DenseNet architectures and therefore lead to a linear growth of the size of the temporal receptive field with network depth, this is not a general restriction of our approach. More complex network graphs, for example containing temporal convolutions or recurrent connections, lead to a super-linear growth of the temporal receptive field. Rescaling weights and biases during conversion by using percentiles instead of maximum values as upper limits for ANN activations increases the accuracy. However, the percentile values of activations calculated over all rollout frames may overestimate the size of ANN activations in single rollout frames and would, hence, decrease the effective resolution of the firing rate approximation (for details, see Section 2.4). For example, in our network rollouts, the activity increases with each rollout frame (which does not hold in general). This results in strong overestimations of activations for early network outputs, which in turn increase approximation errors and, hence, decrease accuracy (see e.g., Figure 3C ). As the activity increases with more and more paths from input to output of the network contributing to later network outputs, this approximation error decreases until an optimal effective resolution is reached when spiking activity is present in all parts of the network. Adaptively rescaling the SNN weights or firing thresholds could be a solution to alleviate this effect. This can be seen as a kind of homeostasis mechanism that keeps the overall firing rates of SNNs at a constant level. Instead of simply averaging event rates to obtain input frames, our approach generalizes to using more advanced features for event-based vision, such as time surfaces (Sironi et al., 2018 ), event spike tensors (Gehrig et al., 2019 ) or motion-based features (Clady et al., 2017 ). As use-cases for event-based vision are becoming increasingly challenging (Gallego et al., 2019 ), and neuromorphic hardware platforms become more mature (DeBole et al., 2019 ), our approach fills an important gap to provide powerful SNNs ready for deployment on those platforms. A major goal of our approach is achieving energy-efficiency, which we measure by the number of operations necessary to reach the desired performance. High efficiency during early inference is enabled by temporal skip connections and carefully choosing the weight factors a k in the loss function to achieve a good early accuracy without deteriorating the later peak accuracy. After ANN-to-SNN conversion, the SNNs are consistently more energy-efficient than their corresponding ANNs, and the achieved relative gain in efficiency is higher than, e.g., reported by Rueckauer et al. ( 2017 ). This may be due to the different neural architectures and the increased sparsity of the input in our study. The sparsity of a single frame increases with a decreasing time interval T F over which events are accumulated. To further increase the efficiency we ran multiple experiments including quantization and observed interesting dependencies between quantization levels, network architectures, energy-efficiency, and final accuracy. A thorough investigation would exceed the scope of this study and is left for future studies. In summary, our approach sets a new standard for spiking neural networks for processing spatio-temporal event streams both in terms of accuracy and efficiency. However, in this study, information is encoded with firing rates, the underlying principle of network conversions, and we did not exploit the potential of encoding information with spike times that potentially allow for even more energy-efficient solutions (for an overview, see Pfeiffer and Pfeil, 2018 ). We are excited to see our results as a competitive baseline for further studies in the direction of spike codes."
} | 3,846 |
39900749 | PMC11790713 | pmc | 205 | {
"abstract": "Under drought conditions, arbuscular mycorrhizal (AM) fungi may improve plant performance by facilitating the movement of water through extensive hyphal networks. When these networks interconnect neighboring plants in common mycorrhizal networks (CMNs), CMNs are likely to partition water among many individuals. The consequences of CMN-mediated water movement for plant interactions, however, are largely unknown. We set out to examine CMN-mediated interactions among Andropogon gerardii seedlings in a target-plant pot experiment, with watering (watered or long-term drought) and CMN status (intact or severed) as treatments. Intact CMNs improved the survival of seedlings under drought stress and mediated positive, facilitative plant interactions in both watering treatments. Watering increased mycorrhizal colonization rates and improved P uptake, particularly for large individuals. Under drought conditions, improved access to water most likely benefited neighboring plants interacting across CMNs. CMNs appear to have provided the most limiting resource within each treatment, whether P, water, or both, thereby improving survival and growth. Neighbors near large, photosynthate-fixing target plants likely benefited from their establishment of extensive hyphal networks that could access water and dissolved P within soil micropores. In plant communities, CMNs may be vital during drought, which is expected to increase in frequency, intensity, and length with climate change. Supplementary Information The online version contains supplementary material available at 10.1007/s00572-025-01181-z.",
"introduction": "Introduction It is projected that climate change will prolong periods of drought and alter precipitation patterns across the globe (Caretta et al. 2022 ), and although plants can respond to drought stress through morphological, physiological, and biochemical mechanisms (Fang and Xiong 2015 ), 70% of species additionally associate with arbuscular mycorrhizal (AM) fungi (Brundrett and Tedersoo 2018 ) that improve water relations during drought (Augé 2001 ; Augé et al. 2015 ). In some instances, the presence of AM fungi is detrimental to plant growth (as reviewed by Augé 2001 ), but much of what we understand about AM functioning during drought has been revealed by studies in which hosts are grown in the presence or absence of AM fungi. Although such studies have been invaluable, we know very little about how interconnecting networks of extraradical AM fungus hyphae, called common mycorrhizal networks (CMNs), influence plant interactions for water (Püschel et al. 2020 ). It has been established, though, that CMNs have the power to influence plant nutrition and growth, interactions, populations, and communities (Leake et al. 2004 ; Selosse et al. 2006 ; Horton 2015 ; van der Heijden et al. 2015 ). The power of CMNs in plant communities lies in their ability to be conduits of belowground biological markets through the partitioning of mineral nutrients among interconnected hosts. According to the largely accepted “reciprocal rewards” hypothesis (Kiers et al. 2011 ), large C provisioners receive more mineral nutrients, like P or N, in return from AM CMNs compared to small plants ( i.e. Lekberg et al. 2010 ; Hammer et al. 2011 ; Merrild et al. 2013 ; Zheng et al. 2015 ; Weremijewicz et al. 2016 ). CMN-mediated reciprocal rewards amplify competitive interactions among interconnected individuals because large individuals come to dominate a particular resource provided by CMNs, and if the resource is growth limiting, then this exchange suppresses the growth of smaller individuals. Isotopic tracing studies with prairie grasses ( e.g. Weremijewicz et al. 2016 ), agricultural ( e.g . Merrild et al. 2013 ), and model species ( e.g. Fellbaum et al. 2014 ), have indeed demonstrated this intensified, asymmetric, competition by CMNs among interconnected individuals. When large C provisioners are prevented from fixing, and ultimately provisioning C, then little to no 15 N is obtained from CMNs by hosts (Fellbaum et al. 2014 ; Weremijewicz et al. 2016 ). Occasionally, however, the reciprocal rewards hypothesis is not supported, and other biological phenomena may be at play, such as source-sink dynamics, host-fungus complementarity, and functional differences among different species of AM fungi (Walder and van der Heijden 2015 ). For example, Walder et al. ( 2012 ) found that the largest C provisioner received the least P and N from a CMN comprising a single AM fungus species, and in a follow up study, found that this could be attributed to the regulation of inorganic P transporters with different P affinities when associating with individual species of AM fungi (Walder et al. 2016 ). CMNs have additionally been found to mediate facilitative interactions. In an arid environment, movement of N from nurse plants to adult plants was detected, although it only made up 2.6% of the total N requirement of individuals (Montesinos-Navarro et al. 2016 ). CMN-mediated facilitative interactions have also been found among deep and shallow rooted plants, in which deeply rooted individuals hydraulically lift water that is then redistributed by extraradical mycorrhizal fungus hyphae within the soil matrix (Querejeta et al. 2003 ) and to interconnected shallow-rooted individuals (Egerton-Warburton et al. 2007 ; Singh et al. 2019 ). Much more is known about the role of CMNs in nutrient partitioning than water partitioning. We do know, however, that the presence of AM fungi can improve water uptake for plants through many different indirect mechanisms (Augé 2001 ). AM fungi improve mineral nutrient uptake for host plants, which ultimately improves physiological aspects of their growth that increase drought resistance, such as an increased expression of aquaporins (Bárzana et al. 2014 ), stomal conductance (Augé et al. 2015 ; Symanczik et al. 2018 ), leaf hydration (Yang et al. 2014 ) and mineral nutrient uptake (Ouledali et al. 2018 ). The AM symbiosis also improves plant osmoregulation, which entails actively decreasing plant water potentials to create a gradient that promotes turgor, stomatal opening, and photosynthesis (Ruiz-Lozano 2003 ; Wu et al. 2017 ; El-Samad and El-Hakeem 2019 ). Additionally, associating with AM fungi protects against drought-induced oxidative damage (Ruiz-Lozano 2003 ; Duc et al. 2018 ; Zou et al. 2021 ; Tereucán et al. 2022 ). AM fungi also alter root morphology by increasing lateral root growth (Gutjahr and Paszkowski 2013 ) which may benefit plants during drought by increasing the volume of soil for water uptake. Finally, soils with mycorrhizal plants are able to retain water soil moisture better than those with nonmycorrhizal plants, most likely because of the effects on soil structure, such as increased water stable aggregates and extraradical hyphal densities (Augé et al. 2001 ). The simple physical presence of AM fungus hyphae in the soil increases the hydraulic conductivity of soil (Bitterlich et al. 2018 ), particularly of loams with low water contents (Pauwels et al. 2023 ). Soils with AM fungus hyphae have greater water retention capacities than those without AM fungi, most likely because of an increase in pore space heterogeneity (Pauwels et al. 2020 ). Hyphae also can decrease the resistance of water movement from high to low water potentials and close large gaps among soil pores by growing across them (Bitterlich et al. 2018 ). Water sticking to the surface of AM hyphae moves through capillary action to host plants (Augé 2004 ). Arbuscular mycorrhizal fungus hyphae can also move water through direct mechanisms related to cytoplasmic streaming within hyphae (Allen 2007 ; Egerton-Warburton et al. 2007 ; Ruth et al. 2011 ; Püschel et al. 2020 ; Pauwels et al. 2020 ; Kakouridis et al. 2022 ), but the magnitude of this effect on plants is debated. Püschel et al. ( 2020 ) found that although Rhizophagus irregularis hyphae doubled the water acquired by Medicago truncatula , this was most likely due to more extensive root systems in mycorrhizal plants compared to nonmycorrhizal plants. The direct hyphal acquisition of water was low compared to plant transpiration requirements. Other studies have indicated that the direct movement of water via extraradical hyphae can be a significant component of a plant’s water uptake, reporting values as high as 12.3% and 17% in alfalfa under high and low water conditions (Wu et al. 2024 ), 20% in barley (Ruth et al. 2011 ), or 34.6% in wild oat (Kakouridis et al. 2022 ). Kakouridis et al. ( 2022 ) used dyes and 18 O to visually demonstrate that water can move directly from Rhizophagus intraradices cells across cell membranes and walls to Avena barbata cells. The contrasting findings within these studies suggest that the overall contribution of water by AM fungi could be affected by plant-fungus species combinations, substrate type (Pauwels et al. 2023 ), substrate pore size, root morphology (Allen 2007 ), and even fungal species (Ruiz-Lozano and Azcón 1995 ; Augé et al. 2015 ). It is evident, however, that extraradical hyphae can access and move water within the rhizosphere. In nature, extraradical hyphae are likely to interconnect multiple host plants within CMNs. Despite our knowledge about the mechanisms of water movement by AM fungi, the partitioning of water among individuals that are interconnected by AM hyphae is not yet resolved (Püschel et al. 2020 ); even though it is established that CMNs influence whether plant interactions will be competitive or facilitative. We set out to investigate how CMNs comprising a suite of AM fungus species mediate plant interactions among conspecifics under drought and watered conditions in a target plant experiment. To avoid effects on plant physiology and soil moisture due to the presence or absence of AM fungi within treatments, we grew all plants with AM fungi, but manipulated CMN interconnections among plants by keeping CMNs intact or severing them once a week. We chose conspecifics because of their similar rooting depths and resource demands, which would allow us to avoid confounding effects caused by differences in root systems and investigate the role of CMNs beyond that of hydraulic lifting. We hypothesized that if reciprocal rewards influenced water partitioning by CMNs, then we would see competition among target plants and their neighbors, in which large plants with more available C would receive more water in return from CMNs than small individuals. We predicted that this would result in a negative relationship between target and neighbor plant size when they were interconnected by CMNs. We also predicted that drought would decrease plant size due to C limitation, but intact CMNs would increase plant size overall by assisting in water movement among interconnected plants within pots.",
"discussion": "Discussion In our study, all plants were colonized by AM fungi, but only plants with intact CMNs benefited in both survival and growth. Contrary to our hypothesis, however, the partitioning of water by CMNs did not intensify competitive interactions between targets and neighbors. Despite the pressure of being surrounded by six individuals within a pot, by the end of the experiment, large targets were surrounded by large, rather than small, neighbors, indicating facilitative interactions were mediated by CMNs. Plants with intact CMNs had improved colonization rates, suggesting that extraradical hyphae improved mycorrhiza formation in neighboring root systems. These additional hyphae likely provided pathways for enhanced water and nutrient movement within pots, which would have benefited the growth of neighboring individuals. Watering also improved AM colonization, which increased plant growth, but a positive relationship between colonization and shoot P concentration only existed when CMNs were intact. Under drought conditions, plants most likely derived some other benefit from the AM symbiosis. Because most nutrients other than N were examined, the Law of the Minimum (von Liebig 1840 ) would suggest that either water, N, or both were limiting growth and/or survival in this treatment. Under drought conditions, mycorrhizal fungi can acquire significant amounts of N for their hosts, particularly from inorganic sources dissolved within soil solution (Püschel et al. 2023 ). In our experiment, either of these possible limitations were severe enough to considerably affect survival, and extraradical hyphae of CMNs likely alleviated these limitations. In plants and in soil surrounding roots, water moves passively from high to low water potentials, driven by transpiration. For example, hydraulic lifting by roots moves water from deep, saturated soil profiles towards shallower, water-limited ones via root systems, where it is redistributed (Allen 2022 ), particularly by CMNs to shallow-rooted individuals (Egerton-Warburton et al. 2007 , 2008 ; Saharan et al. 2018 ; Singh et al. 2019 , 2020 ), resulting in facilitative interactions between differently rooted species. In our study, though, conspecific individuals did not have such large differences in root system size, but facilitative interactions still developed among interconnected plants within pots. Factors that improved the passive movement of water, like increased hydraulic conductivity and decreased resistance due to the increased presence of extraradical hyphae in the soil (Augé et al. 2001 , 2007 ; Bitterlich et al. 2018 ; Pauwels et al. 2023 ), likely caused plants in the intact CMNs treatment to benefit. Because we did not find evidence for competitive interactions, reciprocal rewards were likely not at play in our experiment. Rather than dominating limiting resources via CMNs through C provisioning, C transfer by large plants likely fostered AM fungus hyphae instead, benefiting neighboring individuals with improved colonization rates and ultimately plant size and survival. An increase in the rate of AM colonization with watering and intact CMNs likely improved the transport capacity of water from hyphae to plants, resulting in larger growth. Increased rates of root colonization are associated with improved stomatal conductance rates, by about 46% on average (Augé et al. 2015 ). Soil colonization, like that by extraradical hyphal networks, can affect soil water retention abilities (Augé et al. 2001 ) and positively affect stomatal conductance (Augé et al. 2007 ). When plants were interconnected in CMNs within pots, the extraradical hyphae could both directly and indirectly provide water to hosts, such as by moving water directly to plants within hyphae (Kakouridis et al. 2022 ), altering the geometry of water movement within the soil matrices (Püschel et al. 2021 ), and maintaining moisture for longer in drying soils (Augé et al. 2001 ). In contrast, plants with severed CMNs had hyphae constrained to the resources within their respective cone-tainers. To assess for the potential effect of extraradical hyphae on soil moisture, we examined the rate of drying (% d −1 ) within either cone-tainer soil or interstitial substrate over the course of the experiment, but no main effects of CMNs were detected (SI Table 6 ). This result is most likely because our watering regime regularly brought all of the treatments, regardless of CMNs, back up to a designated level of substrate moisture, negating the ability for us to detect any potential subtle effects of extraradical hyphae on water holding capacity of substrates. Root colonization was decreased when CMNs were severed, and no relationships were found between colonization and plant size. Although we did not measure stomatal conductance in this experiment because of the initial focus on ecological, rather than physiological, consequences of CMNs, the increased root-to-shoot ratios of plants under drought stress, especially with severed CMNs, indicated a morphological response, most likely in search of water (Taiz et al. 2015 ). Water is likely not exchanged nor controlled by C provisioning by hosts because of the passive nature of water uptake by plants at the cellular level. The presence of AM fungi is known to enhance aquaporin expression in host plants, improving the permeability of water into hosts (Uehlein et al. 2007 ; Quiroga et al. 2019 ). At the periarbuscular membrane, aquaporins move water from fungus to plant via facilitated diffusion (Uehlein et al. 2007 ; Quiroga et al. 2019 ). In contrast, for example, nutrient transporters like H + /Pi symporters involve the active transport of nutrients (Dreyer et al. 2019 ; Wipf et al. 2019 ), and fungi have the capability to control the amount of P delivered to the host by either reabsorbing inorganic P or leaving it in the periarbuscular space (Balestrini et al. 2007 ; Walder et al. 2016 ; Wipf et al. 2019 ). Such mechanisms have not been demonstrated for water movement across aquaporins. CMN-mediated water movement, therefore, may not be under metabolic control, which may mean that CMN-mediated plant interactions for water cannot be competitive. The outcome of CMN-mediated plant interactions is likely to be influenced by the watering regime of an experiment because it ultimately influences the most growth-limiting resource. In previous, similar experiments, A. gerardii individuals were well watered and CMNs consistently mediated competitive, not facilitative, interactions (Weremijewicz and Janos 2013 ; Weremijewicz et al. 2016 , 2018 ). Singh et al. ( 2020 ) demonstrated that although hyphal networks can mediate facilitative interactions through hydraulic lifting under drought conditions, competitive interactions for mineral nutrients emerged among interconnected plants in well-watered conditions. Under drought stress, plants close stomata to minimize water loss through transpiration, limiting access to CO 2 (Taiz et al. 2015 ), which may stunt growth due to C-limitation. In contrast, well-watered conditions increase plant access to C, causing growth to increase but also to become limited by mineral nutrients, which may induce competition among plants. CMNs may then mediate competition among individuals by partitioning limiting mineral nutrients to hosts with different photosynthate provisioning abilities. In previous experiments, A. gerardii plants were watered regularly to saturation to promote growth (Weremijewicz and Janos 2013 ; Weremijewicz et al. 2016 , 2018 ). In this experiment, however, we created long-term drought conditions, and it is likely that in reality, our watered treatment may have been a moderate drought treatment. In the watered treatment, plants were watered to field capacity which would have caused soils to desiccate between waterings more quickly than if we were to water to saturation. In between waterings, the macropores (> 80 μm) of the substrates would have desiccated first, concentrating water and dissolved mineral nutrients to micropores (< 30 μm) that were inaccessible to roots and root hairs, but were accessible to hyphal networks (Allen 2022 ). In the watered treatment, hyphal networks likely increased access to water and P simultaneously. It is well established that the simple presence of AM fungi can improve drought avoidance of host plants by relieving P stress (Augé 2001 ), but Püschel et al. ( 2021 ) demonstrated that external AM hyphae are significant for intermediate-term P uptake during moderate droughts. Water found within soil micropores is important for plants because dissolved anionic nutrients, like P, become concentrated (Allen 2007 ), and AM hyphae can move this P directly to their hosts when under moderate drought stress (Püschel et al. 2021 ). We found that as root colonization by AM fungi increased, so did shoot P concentrations, but only for plants with intact CMNs in the watered treatment. In this treatment, the additional water access may have likely caused stomates to open, which would begin photosynthesis and create a demand for nutrients like P, which hyphal networks were able to access. This improved access to resources from intact CMNs likely contributed to improved growth and survival. In the severe drought treatment, intact CMNs increased survival, plant growth, and root colonization, and mediated facilitative plant interactions. Unlike in the watered treatment, however, a lack of relationships among variables makes the mechanisms behind these results less clear. Mineral nutrient analyses revealed that plants under drought stress built-up high levels of micronutrients, and few CMN effects were found. In fact, when CMNs had a statistically significant effect on nutrient levels, plants with severed CMNs had the highest concentrations, suggesting the most C-stressed individuals accumulated mineral nutrients, most likely because they were unable to allocate them towards growth. The lack of a relationship between AM colonization rates and plant size for both CMN treatments under drought stress may indicate a drought effect on the AM fungi themselves. In general, AM fungi may decrease sporulation rates and increase root colonization when water stressed (Augé, 2001 ). In our study, though, overall colonization rates were decreased under drought conditions, consistent with the findings of Püschel et al. ( 2023 ). Leyva-Morales et al. ( 2019 ) found that Rhizophagus intraradices decreases the surface area exposed to drought by decreasing fine hyphae and increasing the production of large-diameter hyphae instead. Additionally, plants have been shown to shunt C away from some fungus species that were present in our inoculum. For example, under drought stress, C. claroideum decreases colonization rates (Forczek et al. 2022 ; Geneva et al. 2022 ). On the other hand, F. mosseae may be beneficial to plants under drought conditions (Forczek et al. 2022 ; Lidoy et al. 2023 ), and C. pellucida is present in semi-arid tropical dry forests in Brazil (Pagano et al. 2013 ). Another possible explanation for the decrease in colonization rates under drought conditions is an increase in root production by hosts. Colonization rates are similar to concentration measurements by being a measurement of quantity per area. An increase in root production (as indicated by root-to-shoot ratios) by plants under drought may have “diluted” colonization rates for plants in this treatment. It can also take weeks to reach peak colonization by AM fungi (Graham et al. 1991 ), which may mean that colonization had not yet caught up with root production. Regardless of the potential effects of soil drying on AM hyphae, the presence of intact CMNs improved colonization of roots overall, and this increased access to CMNs likely fostered positive, facilitative plant interactions among target plants and neighbors. Before plants can interact with one another, they must first survive. Because all treatments included AM colonization in our study, our findings suggest that it is not just the simple presence of AM fungi within plant roots that can positively influence aspects of plant fitness such as survival and growth, but rather the networks of extraradical AM hyphae. The fitness benefits of AM fungi are often described as nutritive, but our study supports Delavaux et al.’s ( 2017 ) findings that non-nutritive benefits, like water access, can also be significant. In our experiment, all plants were colonized by AM fungi, but under drought conditions, survivorship increased three-fold for plants with intact CMNs. In a similar system, Weremijewicz et al. ( 2018 ) also found that CMNs improved survival of A. gerardii and Elymus canadensis . Carbon isotope analyses, that were only possible on C3 E. canadensis , revealed that CMNs improved water uptake, which increased stomatal opening and access to atmospheric CO 2 , altering its carbon isotopic signature. Apart from this study, however, few have reported the effects of CMNs on survival. Generally, there is a demand for tissue samples for post-harvest nutrient analyses, thus, imposed periods of drought purposefully prevent mortality. In fact, Gehring et al. ( 2017 ) suggested that most nonagricultural studies investigating AM fungi are too short in duration to understand the impacts of drought. In our experiment, plants experienced drought for 8 weeks, while in other greenhouse studies, droughts range from four days (Püschel et al. 2020 ) to five weeks (Püschel et al. 2021 ), while in the field, drought can be as long as twelve weeks (Miller et al. 1995 ). In nature, although plants interact via both CMNs and roots, as well as experience diffusion of water and mineral nutrients with rainfall, our model system was necessary to elucidate the specific role of CMNs under drought conditions. Although it is possible that the mechanical disruption of soil from the rotation of cone-tainers to sever networks may have had an effect in our experiment, previous experiments with the same experiment design but with a third treatment using solid, undisturbed cone-tainers to test for this artifact, have found that this treatment is statistically identical to the severed CMNs treatments (Weremijewicz and Janos 2013 ; Weremijewicz et al. 2016 ). We additionally excluded a sterile control from our experiment design because 1) AM fungi have strong physiological effects on plants that affect drought tolerance, and 2) we sought to examine the role of interconnecting, extraradical hyphae on plant interactions, not just the presence/absence of AM fungi. Additionally, A. gerardii is relatively dependent upon and highly responsive to AM fungi (sensu Janos 2007 ). Based on the outcomes of previous work, it likely would have grown poorly without AM fungi (Hartnett et al. 1993 ; Weremijewicz and Seto 2016 ), which would preclude it from interacting with neighbors (Hartnett et al. 1993 ). Our system also included a suite of AM fungus species, which may be necessary to reveal the role of CMNs under drought because the interactions between plants and fungi within CMNs are dynamic, with differences in functional traits and benefits received from different fungus species (Kiers et al. 2011 ; Thonar et al. 2011 ; Walder and van der Heijden 2015 ) and shifts in carbon allocation by hosts to beneficial fungus species under drought, like F. mosseae (Forczek et al. 2022 ). Although it is also possible that plants may have interacted for additional N from both organic and inorganic sources within the substrates (Püschel et al. 2023 ), we were unable to measure this nutrient due to insufficient plant tissue quantities. Regardless, it has been suggested that the P uptake by plants could be more compromised than N uptake under drought conditions (He and Dijkstra 2014 ; Püschel et al. 2021 ). In conclusion, CMNs interconnecting large plants with their neighbors promote plant survival and growth during periods of drought. Although caution must be taken when extrapolating results from carefully controlled greenhouse experiments, the results of our experiment may relate to some patterns seen in the field. For example, increased external AM hyphal production can aid the recovery of prairie plant communities after drought (Miller et al. 1995 ). This phenomenon may also explain why facilitative interactions across CMNs have been found in arid environments (Montesinos-Navarro et al. 2016 ). Although our findings are in contrast to many others that have found CMNs increase competition, one result has remained consistent – connection to CMNs, in general, benefits plant growth overall. Whether plants experience competition or facilitation, however, is likely to depend upon the limiting resource under the environmental conditions at the time of observation. If the effects of water stress can influence future generations, as (Puy et al. 2022 ) suggests, then by improving plant fitness, CMNs have important implications for plant interactions in a changing world."
} | 7,008 |
24714190 | PMC6270765 | pmc | 206 | {
"abstract": "The lotus plant is recognized as a ‘ King plant ’ among all the natural water repellent plants due to its excellent non-wettability. The superhydrophobic surfaces exhibiting the famous ‘ Lotus Effect ’, along with extremely high water contact angle (>150°) and low sliding angle (<10°), have been broadly investigated and extensively applied on variety of substrates for potential self-cleaning and anti-corrosive applications. Since 1997, especially after the exploration of the surface micro/nanostructure and chemical composition of the lotus leaves by the two German botanists Barthlott and Neinhuis, many kinds of superhydrophobic surfaces mimicking the lotus leaf-like structure have been widely reported in the literature. This review article briefly describes the different wetting properties of the natural superhydrophobic lotus leaves and also provides a comprehensive state-of-the-art discussion on the extensive research carried out in the field of artificial superhydrophobic surfaces which are developed by mimicking the lotus leaf-like dual scale micro/nanostructure. This review article could be beneficial for both novice researchers in this area as well as the scientists who are currently working on non-wettable, superhydrophobic surfaces.",
"conclusion": "5. Conclusions and Future Perspectives Wonderful things can result from continuously learning from Nature. The discovery of the principle behind the extreme superhydrophobicity of lotus leaves by Barthlott and Neinhuis at the end of 20th century added fuel to the then slow-burning research on superhydrophobicity. The superhydrophobicity of the lotus leaf is mainly due to the low surface energy provided by epicuticular wax crystalloids and the air pockets trapped in micrometer-scale papillae structure which minimizes the solid-water contact area. Research on the development of superhydrophobic surfaces exhibiting similar wetting properties to lotus leaves is a critical and promising subject in materials science. In this review article, we tried to cover the different wetting properties of lotus leaves and the abundant research work carried out (since 1997) on the development of superhydrophobic surfaces by mimicking the lotus leaf-like surface morphology. So far, perfect mimicry of the lotus leaf surface morphology has not yet observed/reported. The mimicry of lotus leaf micro/nanostructure is just limited to the achievement of high water contact angles and low sliding angles on the surface, however, serious steps are to be taken to achieve a durability similar to that of a lotus leaf. Most of the superhydrophobic surfaces developed by mimicking lotus leaf-like micro-nanostructure have used various polymers during synthesis due to their natural hydrophobicity and toughness. It has been well demonstrated that plenty of naturally occurring surfaces are superhydrophobic and they all are developed with polymers [ 123 , 124 ]. The templation method is promising to achieve near-perfect mimicry of the lotus leaf-like morphology, whereas other physical and/or chemical methods could partially mimic the surface structure. The mimicry was either near-perfect or partial, however the wetting properties were identical to those of the lotus leaf. A main challenge in the development of the superhydrophobic surfaces is the design of their surface micro/nanostructures and low surface energy. Off course, the lotus leaf-like surface morphology is ideal to achieve strong superhydrophobicity on solid surfaces but further research on many other aspects should be carried out in order to compete in the commercial market. Enormous amounts of research on lotus leaf-like superhydrophobic coatings have been successfully carried out on the small scale in the laboratory, however they can be ineffective when applied on the large scale due to several reasons. The mechanical durability or scratch resistance of the coatings in combination with optical transparency is needed for applications like self-cleaning coatings on windows and door glasses of buildings. The adherent coating is paramount in the application of anticorrosive coating on the metals. We believe that, instead of inorganic superhydrophobic coatings, the future superhydrophobic research should be directed towards polymeric superhydrophobic coatings which shows good adhesion with underlying substrates of any shape and size and can last longer with high mechanical durability and optical transparency. The demand for superhydrophobic coatings is growing, and this will perhaps be one of the most challenging and interesting research fields in materials science for the next two decades.",
"introduction": "1. Introduction Nature is the world’s giant research laboratory, freely open to all scientists from interdisciplinary fields ranging from Biology, Physics, Chemistry, Mathematics, Materials Science, and Engineering. Nature is fantastic and always ready to teach without hesitation. In Nature, biological micro/nanostructures have developed as the result of millions of years of evolution and their designs sustain many unique and unusual properties [ 1 ]. For the past two centuries a majority of the deliberate research and development has been inspired and literally copied from Nature. Numerous next-generation artificial advanced functional materials like nanomaterials and nanodevices have been developed with ease by precisely mimicking the chemical components, surface micro/nanostructures, and exceptional properties present in natural systems [ 2 , 3 , 4 ]. As the ‘perfect mimicry’ is always impossible, perfect copy of the desired features have not yet been reported. In Mother Nature, numerous plant leaf surfaces exhibit excellent water repellent properties [ 5 , 6 , 7 ]. Among them, the Sacred Lotus ( Nelumbo nucifera ) is a semi-aquatic plant, which has been popularly known as a symbol of purity in Asian culture for over 2,000 years due to its capability to remain clean. Compared to other natural plant leaves, lotus leaf is the most superhydrophobic, with a water contact angle higher than 160° and sliding angle lower than 5°, hence it always remain clean in muddy and dirty ponds. In the rainy season, when the raindrops fall on the surface of lotus leaves, they immediately bead up like shiny spherical balls and quickly roll off the surface collecting dirt and debris along the way [ 8 ]. On the lotus leaf surface, the adhesion between the water droplet and dust particle is stronger than the adhesion between the dusts and the surface, hence the spherical water drops pick up the dust particles while rolling off the lotus leaf. This extreme water repellency and self-cleaning performance of lotus leaf is famously known in the literature as the ‘ Lotus Effect ’ [ 9 ]. Furthermore, lotus leaves can retain their superhydrophobicity for a lifetime due to their self-healing function [ 10 ]. Before 1996, limited attention was paid to superhydrophobic surface research which was solely based on the relation between static water contact angle and rough surface geometry [ 11 , 12 , 13 , 14 , 15 ]. In 1997, two German botanists, Barthlott and Neinhuis, with the aid of a scanning electron microscope (SEM), revealed for the first time the unique dual scale micro/nanostructure of the lotus leaves and also studied the chemical material present on it [ 16 ]. The hierarchical micro/nanostructure of a lotus leaf surface is constituted of secreted low surface energy epicuticular wax crystalloids that uniformly cover the cuticular surface in a regular microrelief of about 1–5 µm in height. It is concluded that the combination of dual scale roughness and low surface energy wax allows air to be trapped under the floating water drops that contribute to the superhydrophobic and self-cleaning behavior of the lotus leaf. This revolutionary research provided two important guidelines for researchers to reactivate the research on superhydrophobic surfaces, one is roughening the low surface energy materials [ 17 , 18 , 19 , 20 ], and other is the modification of rough structures with low surface energy materials [ 21 , 22 , 23 , 24 , 25 ]. Thus the unusual surface wettability existing in Nature can be directly mimicked by controlling the surface geometrical microstructure and low surface energy of the surface. Since this discovery by Barthlott and Neinhuis, plenty of research articles [ 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 ] and reviews [ 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 , 39 , 40 ] have appeared on the superhydrophobic surfaces describing their adoption in various potential applications ranging from self-cleaning coatings for windshields of automobiles [ 41 ], optical devices [ 42 ], window glasses and solar panels [ 43 , 44 ], anti-fogging and anti-corrosive coatings [ 27 ], paints [ 45 ], hydrodynamic drag reduction [ 46 , 47 ], anti-icing [ 48 ] and interior fabrics [ 49 ]. In many review articles as well as in the research articles, the famous ‘ Lotus Effect ’ has been discussed and the lotus leaf-like micro/nanoscale binary structure is recommended for the development of superhydrophobic surfaces [ 2 , 3 , 6 , 7 , 8 , 36 , 38 , 41 ]. Here we made an attempt to provide a review article which describes in particular the state-of-the-art research into the synthesis of artificial superhydrophobic surfaces by mimicking the lotus leaf-like micro/nanostructure. We thoroughly summarize the different design and fabrication strategies as well as their potential applications in non-wetting surfaces. We also intended to bring in light the different wetting properties of lotus leaves other than just high water contact angle and low sliding angle. This review article is organized into five main sections. The first section presents a brief introduction to a natural superhydrophobic surface, lotus leaves, including the dependence of their surface wetting properties on the chemical composition and rough hierarchical surface morphology. In second section, the different wetting properties of a solid surface by a liquid are briefly introduced. The third section provides a comprehensive overview on the different wetting properties of lotus leaves studied other than high water contact angle and low sliding angle. The fourth section is again organized in two subsections, in first subsection, the superhydrophobic surfaces prepared by using lotus leaves as a biological template are discussed, while in other subsection different physical and chemical methods used for the preparation of lotus leaf-like micro/nanostructures are presented. Finally, we will present our perspective on the future directions and challenges in the fascinating research area of superhydrophobic surfaces developed by mimicking the lotus leaf-like surface morphology."
} | 2,690 |
37351423 | PMC10282152 | pmc | 207 | {
"abstract": "Working memory refers to the brain's ability to store and manipulate information for a short period. It is disputably considered to rely on two mechanisms: sustained neuronal firing, and “activity-silent” working memory. To develop a highly biologically plausible neuromorphic computing system, it is anticipated to physically realize working memory that corresponds to both of these mechanisms. In this study, we propose a memristor-based neural network to realize the sustained neural firing and activity-silent working memory, which are reflected as dual functional states within memory. Memristor-based synapses and two types of artificial neurons are designed for the Winner-Takes-All learning rule. During the cognitive task, state transformation between the “focused” state and the “unfocused” state of working memory is demonstrated. This work paves the way for further emulating the complex working memory functions with distinct neural activities in our brains.",
"conclusion": "4. Conclusion In this paper, a memristor-based working memory that is capable of exhibiting dual functional states is presented. To achieve this, an artificial synapse with a simplified Hebbian learning rule was designed based on the LTP/LTD properties of the Au/LNO/Pt memristor, which uses a single-crystalline LiNbO3 (SC-LNO) thin film as its insulating layer. Two types of artificial LIF neurons were implemented in the network to encode feature information to working memory through the WTA rule and produce persistent neuronal firing patterns. The results show that the proposed system can realize various neuronal events in working memory, including encoding, attention, and retrieval. This study demonstrates that the memristor-based working memory can exist in the dual functional states: the sustained neuronal firing and activity-silent working memory. This study paves the way for the development of advanced bio-plausible neuromorphic computing systems based on memristive neural networks. This research represents a significant step toward the development of advanced bio-plausible neuromorphic computing systems based on memristive neural networks. It is hoped that this work will inspire further research in this exciting and rapidly evolving field.",
"introduction": "1. Introduction Working memory is an essential brain function that allows for the temporary storage and manipulation of information required for cognitive tasks (Baddeley and Hitch, 1974 ; Morris, 1986 ; Baddeley, 1992 , 2010 ). For a long time, it was thought to be presented in the form of persistent neuronal firing during the delay period (Funahashi, 2017 ). However, recent studies have suggested that synaptic weight can also store information during the delay period, even if persistent neuronal firing has ceased (Mongillo et al., 2008 ; Stokes, 2015 ; Silvanto, 2017 ). This phenomenon is referred to as “activity-silent” working memory. In most previous studies, the sustained neuronal firing and “activity-silent” working memory have been modeled independently, and these mechanisms appear to be fundamentally opposed in principle. On the other hand, in recent years, several studies have provided insights into the interaction between sustained neuronal firing and “activity-silent” working memory. Manohar et al. have proposed a memory model that unites both persistent activity attractors and silent synaptic memory, which is applicable to many empirical phenomena (Manohar et al., 2019 ). Barbosa et al. have investigated the interplay between persistent activity and activity-silent dynamics in the prefrontal cortex using monkey and human electrophysiology data (Barbosa et al., 2020 ). In the past decades, a great number of efforts had been made for the hardware implementation of a wide variety of artificial neural networks (Misra and Saha, 2010 ; Capra et al., 2020 ; Nguyen et al., 2021 ; Ghimire et al., 2022 ). Recent works on silent synapses and artificial synapses have highlighted their potential for advancing the understanding of the nervous system and developing neuromorphic computing technologies (Loke et al., 2016 ; Go et al., 2021 ; Hao et al., 2021 ). Following this research trend, there is a growing anticipation to realize a highly bio-plausible neuromorphic computing system. Although working memory plays a vital role in biological neurocomputing (Wang et al., 2020 ), there have been only a handful of studies on the hardware implementation of working memory, and the existing research has mainly focused on the independent neural mechanism of working memory (Brown and Aggleton, 2001 ; Ji et al., 2022 ). Proposing a hardware design for working memory that is compatible with both sustained neuronal firing and activity-silent working memory can improve its biological plausibility and expand the breadth of its application. To achieve the aforementioned dual functional states of working memory, this paper proposes a hardware design for working memory based on memristors. Memristor is a non-linear two-terminal electrical device that has been extensively studied in the past decade, which is a key element used in artificial neural networks for synapses and neurons due to similarities in electrical behavior (Chua, 1971 ; Strukov et al., 2008 ; Thomas, 2013 ; Li et al., 2018 ; Camuñas-Mesa et al., 2019 ; Xia and Yang, 2019 ). In this work, we propose a memristor-based neural network to realize the dual functional states of working memory. To achieve this, the electrical characteristics of an Au/LNO/Pt memristor based on Single-Crystalline LiNbO 3 (SC-LNO) thin films is utilized. The use of the high-quality SC-LNO thin film results in several advantageous properties, including high switching uniformity, long retention time, stable endurance performance, and reproducible multilevel resistance states (Wang et al., 2022 ). An artificial synapse circuit with simplified Hebbian learning rule is implemented with Au/LNO/Pt memristor. A spiking neural network capable of realizing the winner-takes-all (WTA) functionality is constructed, which is utilized to achieve working memory working memory. State transformation between the “focused” state (sustained neuronal firing) and the “unfocused” state (activity-silent working memory) of memristor-based working memory is demonstrated. This hardware solution for bio-plausible working memory with dual functional states, leveraging the intrinsic electrical properties of memristors, has promising implications for the development of advanced bio-plausible neuromorphic computing systems.",
"discussion": "3. Results and discussion The functionality of the memristor-based working memory was evaluated using SPICE simulation, leveraging the electrical characteristics of Au/LNO/Pt memristors derived from experimental data. The memristive conductance was normalized to serve as synaptic weights. The working memory employed a total of 9 feature-selective neurons, with each group of 3 neurons corresponding to a distinct feature dimension, including color, orientation, and location. Each neuron was responsible for encoding different feature information within its respective dimension. Additionally, 3 freely-conjunctive neurons were fully-connected to the feature-selective neurons, resulting in a total of 27 feature-to-conjunctive and 27 conjunctive-to-feature synapses. To enable lateral inhibition, 6 synapses were established between freely-conjunctive neurons, with self-connections excluded. Lastly, 3 self-exciting synapses were connected to the freely-conjunctive network in a self-connected manner, for the purpose of inducing self-excitation. Figure 7 shows a typical sensory input for working memory. Three input features were selected: Obj 1 (color: red, orientation: −45 degrees, location: bottom-left), Obj 2 (color: yellow, orientation: 0 degrees, location: top-left), and Obj 3 (color: blue, orientation: +45 degrees, location: top-right). The sensory current was applied to corresponding feature-selective neurons, and the magnitude of the sensory current was adjusted to the same value for each of the 3 feature dimensions. The firing rate of the feature-selective neuron was proportional to the magnitude of the input current. An initialization time of 50 ms was allocated for stabilization at the beginning of working memory. Each feature input lasted for 100 ms, and there was a 50 ms resting time between each two features. The time scalar T p for the simplified Hebbian learning rule is defined as 1 ms. Figure 7 Sensory input of feature-selective neurons. Figure 8 shows the activity of each freely-conjunctive neuron. When the first feature input Obj 1 is activated, feature-to-conjunctive synapses modify their connectivity based on a simplified Hebbian learning rule, and freely-conjunctive neurons compete with each other. This process is known as encoding in working memory. Once the competition is completed, the winning neuron remains active even without the sensory input. The feature information is encoded into working memory and is presented in the form of persistent neuronal firing, which can be regarded as the “focus” state of working memory. Additionally, feature information is silently encoded into the synaptic weights of the feature-to-conjunctive synapses. When the second feature Obj 2 is activated, the previous attention is disturbed by the new input, and a new round of encoding occurs. Then, the neurons that won the competition in this round persistently fire, forming a new attention . Similarly, when the third feature input Obj 3 is activated, another “focus” state of working memory is formed as the persistent firing freely-conjunctive neurons map to Obj 3 . However, the “unfocused” state of working memory exists in the form of synaptic mappings and persists as long as it is not overwritten by new feature information during the working memory task. After a resting time of 50 ms, a sensory input composed of partial feature information of Obj 1 (colors: red) is applied. Despite using only a small fraction of the feature information, persistent neuronal firing occurs again, which corresponds to the retrieval of Obj 1 . This result demonstrates that the proposed memristor-based working memory system can successfully achieve the dual functional states of working memory and accomplish the working memory task. Figure 8 Firing rate of freely-conjunctive neuron."
} | 2,599 |
23761313 | PMC3659410 | pmc | 208 | {
"abstract": "Changes in communities of syntrophic acetate-oxidizing bacteria (SAOB) and methanogens caused by elevated ammonia levels were quantified in laboratory-scale methanogenic biogas reactors operating at moderate temperature (37°C) using quantitative polymerase chain reaction (qPCR). The experimental reactor was subjected to gradually increasing ammonia levels (0.8–6.9 g NH 4 + -N l −1 ), whereas the level of ammonia in the control reactor was kept low (0.65–0.90 g NH 4 + -N l −1 ) during the entire period of operation (660 days). Acetate oxidation in the experimental reactor, indicated by increased production of 14 CO 2 from acetate labelled in the methyl carbon, occurred when ammonia levels reached 5.5 and 6.9 g NH 4 + -N l −1 . Syntrophic acetate oxidizers targeted by newly designed qPCR primers were Thermacetogenium phaeum , Clostridium ultunense , Syntrophaceticus schinkii and Tepidanaerobacter acetatoxydans . The results showed a significant increase in abundance of all these bacteria except T. phaeum in the ammonia-stressed reactor, coincident with the shift to syntrophic acetate oxidation. As the abundance of the bacteria increased, a simultaneous decrease was observed in the abundance of aceticlastic methanogens from the families Methanosaetaceae and Methanosarcinaceae . qPCR analyses of sludge from two additional high ammonia processes, in which methane production from acetate proceeded through syntrophic acetate oxidation (reactor SB) or through aceticlastic degradation (reactor DVX), demonstrated that SAOB were significantly more abundant in the SB reactor than in the DVX reactor.",
"introduction": "Introduction Methane formation from acetate can proceed through two different mechanisms. The most commonly described involves aceticlastic methanogens that perform acetate cleavage for methane production. The second mechanism proceeds through syntrophic acetate oxidation ( Zinder and Koch, 1984 ). This pathway entails fermentation of acetate to hydrogen and carbon dioxide by syntrophic acetate-oxidizing bacteria (SAOB). Hydrogen utilizing methanogens then reduce carbon dioxide to methane. High ammonia levels, formed during the anaerobic degradation of protein-rich material, have been shown to be one important factor regulating the shift from aceticlastic methanogenesis to syntrophic acetate oxidation in mesophilic biogas processes ( Schnürer et al ., 1999 ; Schnürer and Nordberg, 2008 ). The shift is probably a consequence of the inhibitive effect of ammonia on the activity of the aceticlastic methanogens ( Koster and Lettinga, 1984 ; Sprott and Patel, 1986 ). The concentration of acetate, dilution rate and presence of the aceticlastic Methanosaetaceae are other factors suggested to have an impact on the development of syntrophic acetate oxidation ( Petersen and Ahring, 1991 ; Ahring et al ., 1993 ; Shigematsu et al ., 2004 ; Karakashev et al ., 2006 ). So far a restricted number of SAOB have been isolated and characterized, namely the mesophilic bacteria Clostridium ultunense ( Schnürer et al ., 1996 ; 1997 ) and Syntrophaceticus schinkii ( Westerholm et al ., 2010 ), the thermotolerant Tepidanaerobacter acetatoxydans ( Westerholm et al ., 2011 ), and the thermophilic bacteria Thermacetogenium phaeum ( Hattori et al ., 2000 ; 2005 ) and Thermotoga lettingae ( Balk et al ., 2002 ). Initially, a thermophilic bacterium ( Lee and Zinder, 1988 ) named Reversibacter was described, but unfortunately this bacterium was lost before its phylogenetic position could be established. Information about syntrophic acetate oxidation, the organisms involved, and their role in the methanogenic environment is currently limited. However, greater understanding of microbial dynamics in response to inhibitory compounds, such as ammonia, should facilitate further development and also optimization of the anaerobic treatment process. In the present study, primers targeting 16S rRNA gene sequences of known SAOB were designed. Quantitative real-time polymerase chainreaction (qPCR) analyses were then performed in order to determine changes in SAOB and methanogenic communities caused by elevated ammonia concentrations. Two mesophilic biogas reactors were included in the analysis. In one (experimental) reactor a shift from aceticlastic acetate degradation to syntrophic acetate oxidation had been established previously, while in the second (control) reactor aceticlastic methanogenesis was the main pathway for methane formation ( Schnürer and Nordberg, 2008 ). Two high ammonia processes, in which methane production from acetate proceeded through syntrophic acetate oxidation (reactor SB) or through aceticlastic degradation (reactor DVX), were also included in the investigation.",
"discussion": "Results and discussion Pathway for acetate degradation in the biogas reactors In the control reactor, acetate degradation was primarily through aceticlastic methanogenesis throughout the operating period. In the experimental reactor, which was subjected to gradually increasing ammonia levels, a shift from aceticlastic acetate degradation to syntrophic acetate oxidation was established between 225 and 442 days of operation, when the ammonia level reached 5.5 and 6.9 g NH 4 + -N l −1 ( Fig. S1 ). Labelling analysis with [2- 14 C]-acetate demonstrated occurrence of syntrophic acetate oxidation in reactor SB ( Schnürer and Nordberg, 2008 ; Ek et al ., 2010 ), while in reactor DVX the analysis indicated that aceticlastic methanogenesis was the main pathway for acetate degradation ( Fig. S1 ). The dominance of aceticlastic methanogenesis in reactor DVX was somewhat unexpected, as the ammonia concentration in this reactor exceeded the levels previously shown to cause development of syntrophic acetate oxidation ( Schnürer et al ., 1999 ; Schnürer and Nordberg, 2008 ). Parameters other than ammonia [e.g. substrate change, increased loading rate and decreased hydraulic retention time (HRT)] apparently had an impact on the mechanism developed for methane formation in reactor DVX. Primer specificity and detection of SAOB in samples with conventional PCR Specific primer sets for detection of 16S ribosomal RNA (rRNA) genes of C. ultunense , S. schinkii , T. acetatoxydans and T. phaeum ( Table 1 ) generated single PCR products from genomic DNA of the corresponding species. Furthermore, in PCR amplification with primer sets targeting C. ultunense , S. schinkii and T. acetatoxydans , products of the predicted length were generated from template DNA extracted from the experimental reactor on day 442 and day 642 of operation. From all other samples (control and experimental reactor) the quantity of amplified DNA was below the detection limit for visualization using ethidium bromide staining. Furthermore, PCR analysis with the primer set targeting the 16S rRNA gene of the thermophile T. phaeum did not generate any visible product from any reactor sample. Analysis of DNA extracted from samples from reactors SB or DVX only gave a positive result with the primer set targeting the 16S rRNA gene of S. schinkii . The primers all showed high specificity, as PCR products generated from all reactor samples were sequenced and shown to be identical (100% identity over 171, 237 and 127 bp respectively) to the sequences retrieved from pure cultures of the corresponding bacteria. Furthermore, all primer sets showed high specificity to the corresponding SAOB in an evaluation against the GenBank database using BLAST. Table 1 Primer sets and PCR programs used in the investigation Primer a Target species or group Sequence (5′→3′) b Position in target species c T m (°C) Amplicon size (bp) Cultf e Clostridium ultunense CCT TCG GGT GGA ATG ATA AA 56–76 57 127 Cultr e TCA TGC GAT TGC TAA GTT TCA 162–183 THACf d Syntrophaceticus schinkii ATC AAC CCC ATC TGT GCC 802–820 61 171 THACr d CAG AAT TCG CAG GAT GTC 955–973 Tpf d Tepidanaerobacter acetatoxydans AGG TAG TAG AGA GCG GAA AC 963–983 63 237 Tpr d TGT CGC CCA GAC CAT AAA 1182–1200 Thf e Thermacetogenium phaeum GGG TGG TGT GAA GCC ATC 795–813 68 175 Thr e AGG TCC GCA GAG ATG TCA AG 970–990 Tbf f Total bacteria GTG ITG CAI GGI IGT CGT CA 1048–1068 61 323 Tbr f ACG TCI TCC ICI CCT TCC TC 1371–1391 Mscf g Methanosarcinaceae GAA ACC GYG ATA AGG GGA 380–397 60 408 Mscr g TAG CGA RCA TCG TTT ACG 811–828 MMBf g Methanomicrobiales ATC GRT ACG GGT TGT GGG 282–299 66 506 MMBr g CAC CTA ACG CRC ATH GTT TAC 812–832 Mstf g Methanosaetaceae TAA TCC TYG ARG GAC CAC CA 702–721 61 164 Mstr g CCT ACG GCA CCR ACM AC 812–832 pAf h Bacteria AGA GTT TGA TCC TGG CTC AG 8–28 55 1534 pHr h AAG GAG GTG ATC CAG CCG CA 1542–1522 Arch46f i Archaea YTA AGC CAT GCR AGT 46–61 40 971 Arch1017r j GGC CAT GCA CCW CCT CTC 1017–999 a f, forward; r, reverse primer. b I, inosine. c 16S rRNA gene sequence. d Designed by Dr Neil Gray, School of Civil Engineering and Geosciences; Newcastle University. e Designed by Stefan Roos and Maria Westerholm, Department of Microbiology, Swedish University of Agricultural Sciences, Uppsala, Sweden. f ( Maeda et al ., 2003 ). g ( Yu et al ., 2005 ). h ( Edwards et al ., 1989 ). i ( Øvreås et al ., 1997 ). j ( Barns et al ., 1994 ). Primer sets targeting the 16S rRNA genes of S. schinkii and T. acetatoxydans were designed with Primrose version 2.1.7 ( Ashelford et al ., 2002 ) and for amplification of the 16S rRNA genes of C. ultunense and T. phaeum , primers were designed with Primer3, version 0.4.0 ( Rozen and Skaletsky, 2000 ). The primer specificity was evaluated against the GenBank database using blast ( Altschul et al ., 1990 ). PCR amplifications were conducted using a 25 µl mixture including 5 µl of 10x NH 4 buffer (Bioline, London, UK), 1.5 µl of 50 mM MgCl 2 , 1 µl of forward and reverse primer (10 µM), 1 µl of dNTPs (10 mM each), 0.2 µl of Taq DNA polymerase (Bioline, London, UK), 14.3 µl of sterile water and 1 µl of template DNA in each reaction. Alternatively Ready-To-Go PCR beads (GE Healthcare Buckinghamshire, UK), containing 25 pmol of each primer per 25 µl PCR reaction, were used. The PCR program consisted of: 95°C for 3 min, 30 cycles of 95°C for 1 min, annealing for 1 min at temperature shown above, and 72°C for 1 min, followed by 10 min at 72°C. Real-time PCR quantification of SAOB and methanogens All standard curves for the quantitative PCR analyses, constructed as described in Table S1 , had a linear correlation coefficient ( r 2 ) ranging between 0.985 and 0.999, and the calculated qPCR efficiency of the reactions varied between 86.2% and 108%. The qPCR analyses showed a distinct increase in C. ultunense , S. schinkii and T. acetatoxydans in the experimental reactor when the ammonia level increased above 3.3 g NH 4 + -N l −1 ( Fig. 1A ). The increase was confirmed by an additional assay with triplicate DNA samples from the experimental reactor on day 225 and day 442, which demonstrated a significant increase (un-paired t -test, P < 0.05) in C. ultunense from 4.1 ± 1.2 × 10 5 to 2.3 ± 0.9 × 10 7 gene abundance ml −1 , in S. schinkii from 6.3 ± 1.4 × 10 6 to 6.8 ± 2.1 × 10 9 gene abundance ml −1 , and in T. acetatoxydans from 4.7 ± 2.4 × 10 5 to 5.7 ± 0.4 × 10 10 gene abundance ml −1 . In parallel, a decrease in the abundance of acetate utilizing methanogens from the family Methanosarcinaceae occurred from day 225 onwards, when the ammonia concentration exceeded 3.3 g NH 4 + -N l −1 . However, the abundance of the acetate utilizing Methanosaetaceae declined after only 70 days of operation ( Fig. 1B ). Hydrogenotrophic methanogens of the order Methanomicrobiales initially decreased in abundance between days 70 and 142, but subsequently increased to around their initial abundance by day 642. It is possible that certain members of the Methanomicrobiales declined initially due to ammonia inhibition or pH change and subsequently (> 142 days) ammonia-tolerant members of the Methanomicrobiales were favoured as the ammonia concentration increased. The observed decrease in aceticlastic methanogens and increase in hydrogenotrophic methanogens in response to increasing ammonia levels, most likely caused by a comparatively higher tolerance of Methanomicrobiales to ammonia ( Koster and Lettinga, 1984 ; Sprott and Patel, 1986 ), have been reported at population level previously ( Angenent et al ., 2002 ). However, the present study represents the first detailed analysis of changes in both the population of methanogens and SAOB in response to increasing ammonia concentration. Fig. 1 (A) Abundance of syntrophic acetate-oxidizing bacteria in the control reactor (- - -) and the experimental reactor (—), as determined by qPCR analysis of 16S rRNA genes. C. ultunense •; S. schinkii ▴; T. acetatoxydans . (B) Abundance of methanogens in the control reactor (- - -) and the experimental reactor (—) as determined by qPCR analysis of 16S rRNA genes. Methanomicrobiales •; Methanosarcinaceae ▴; Methanosaetaceae . Genomic DNA was extracted from three replicate samples (0.3 ml each) from each reactor and sampling point, using the FastDNA Spin kit for soil (Qbiogene, Illkrich, France). The triplicate DNA samples were pooled and the qPCR was performed with a BioRad iCycler (Hercules, CA). Each reaction contained 3 μl DNA template, 1 μl of each primer (10 pmol μl −1 ), 5 μl of sterile water, 10 μl iQ Supermix PCR reagent (BioRad, Hercules, CA), and SYBR-Green I as the fluorescent DNA intercalating agent (0.2 μl of 100x concentrate, Invitrogen, UK). In qPCR analysis of the methanogenic communities the temperature cycle consisted of: 95°C for 7 min; 55 cycles of 95°C for 40 s; annealing at specific temperatures ( Table S1 ) for 1 min; and 72°C for 40 s. qPCR analysis of SAOB was performed applying the following conditions: 7 min at 95°C; 40 cycles of 95°C for 30 s; annealing at specific temperatures ( Table S1 ) for 1 min; and 72°C for 30 s. At the end of each qPCR assay, a temperature melt curve was performed to verify reaction quality (55–95°C, ΔT = 0.1°C s −1 ). Logarithmic values of the concentration of the16S rRNA gene were plotted against the threshold cycle (C t ) number and used for estimation of gene abundance in the unknown samples. In the control reactor, methanogen and SAOB abundance remained stable throughout the 642 days of operation ( Fig. 1 ). Total bacterial abundance in the control reactor and experimental reactor was stable (4.9 ± 1.8 × 10 10 and 3.7 ± 2.2 × 10 10 gene abundance ml −1 respectively) throughout the operating period. These results supported the presumption that the changes in the microbial communities in the experimental reactor were a consequence of increased ammonia concentration. 16S rRNA genes from T. phaeum were not detected in any of the reactors. This was not surprising, as the temperature range of this thermophilic bacterium is 40–65°C, with an optimum around 58°C. The conditions in the reactors, operating at 37°C, were therefore unfavourable for T. phaeum . In previous studies, C. ultunense , S. schinkii and T. acetatoxydans proved capable of withstanding rather high levels of ammonium chloride (∼ 8 g NH 4 + -N l −1 ) at neutral pH ( Schnürer et al ., 1996 ; Westerholm et al ., 2010 ; 2011 ), an ammonium level that has strong inhibitory effects on aceticlastic methanogens from the families Methanosarcinaceae and Methanosaetaceae ( Sprott and Patel, 1986 ; Hajarnis and Ranade, 1993 ). The ammonia tolerance of these syntrophic acetate-oxidizers probably gives them a competitive advantage in ammonia-stressed systems. These bacteria, in association with ammonia-tolerant hydrogenotrophic methanogens, may consequently adopt the role of dominant acetate consumers in environments where ammonia restrains aceticlastic methanogenic activity. In reactor SB, C. ultunense , S. schinkii and T. acetatoxydans were present at significantly ( t -test, P < 0.05) greater abundance than in reactor DVX ( Fig. 2 ). The total bacterial gene abundance in reactor SB (1.0 ± 0.4 × 10 11 ml −1 ) was also slightly higher than in reactor DVX (3.2 ± 0.8 × 10 10 ml −1 ). The comparatively low abundance of acetate oxidizers in reactor DVX agreed with the labelling analysis, demonstrating dominance of aceticlastic methanogenesis in this reactor. In contrast, there was no significant difference ( t -test, P > 0.05) in mean gene abundance of Methanosarcinaceae or Methanomicrobiales between reactors DVX and SB, and the Methanosaetaceae abundance was even significantly lower in reactor DVX. The high abundance of Methanosaetaceae in reactor SB was unexpected and contradicted results reported by Karakashev and colleagues (2006) , showing that acetate oxidation is the dominant pathway only in the absence of Methanosaetaceae . The relatively high abundance of Methanosarcinaceae and Methanosaetaceae in reactor DVX is also noteworthy, indicating occurrence of ammonia-tolerant aceticlastic methanogens in this reactor operating at a high ammonia concentration. However, the accumulation of VFA and the decline in pH demonstrated the instability of aceticlastic methanogenesis in the conditions under which reactor DVX was operated, thereby reflecting the importance of SAOB for the maintenance of process stability in methanogenic systems with high ammonia concentrations. Fig. 2 Abundance of syntrophic acetate oxidizing bacteria and methanogens in reactors SB and DVX. Acetate degradation proceeded through syntrophic acetate oxidation in reactor SB and via aceticlastic methanogenesis in reactor DVX. SB was a large-scale reactor operating with an average HRT of 56 days and was fed with slaughterhouse waste as main substrate ( Ek et al. , 2010 ). At the time of sampling the concentrations of volatile fatty acids (VFA) and ammonia-nitrogen in the process were 2.3 g l −1 and 5.3 g NH 4 + -N l −1 , respectively. DVX was a laboratory-scale reactor that was inoculated with sludge from the SB reactor. The process was fed with distiller's waste and operated with an average HRT of ∼40 days at approximately pH 7.8, 1.6 g VFA l −1 and 7.8 g NH 4 + -N l −1 . The OLR of DVX was initially 4 g VS l −1 day −1 and was then gradually increased and had reached 6 g VS l −1 day −1 when sampled. After ∼330 days of operation, high concentrations of VFA (4-5 g l −1 ) had accumulated in the process and the pH had started to decrease. Triplicate samples from reactors SB and DVX, taken on a single sampling occasion, were analyzed separately and the qPCR analysis was conducted as described in Fig. 1 ."
} | 4,649 |
39731342 | PMC11967857 | pmc | 210 | {
"abstract": "Abstract The biobased production of chemicals is essential for advancing a sustainable chemical industry. 1,5‐Pentanediol (1,5‐PDO), a five‐carbon diol with considerable industrial relevance, has shown limited microbial production efficiency until now. This study presents the development and optimization of a microbial system to produce 1,5‐PDO from glucose in Corynebacterium glutamicum via the l ‐lysine‐derived pathway. Engineering began with creating a strain capable of producing 5‐hydroxyvaleric acid (5‐HV), a key precursor to 1,5‐PDO, by incorporating enzymes from Pseudomonas putida (DavB, DavA, and DavT) and Escherichia coli ( YahK). Two conversion pathways for further converting 5‐HV to 1,5‐PDO are evaluated, with the CoA‐independent pathway—utilizing Mycobacterium marinum carboxylic acid reductase (CAR) and E. coli YqhD—proving greater efficiency. Further optimization continues with chromosomal integration of the 5‐HV module, increasing 1,5‐PDO production to 5.48 g L −1 . An additional screening of 13 CARs identifies Mycobacterium avium K‐10 (MAP1040) as the most effective, and its engineered M296E mutant further increases production to 23.5 g L −1 . A deep‐learning analysis reveals that Gluconobacter oxydans GOX1801 resolves the limitations of NADPH, allowing the final strain to produce 43.4 g L −1 1,5‐PDO without 5‐HV accumulation in fed‐batch fermentation. This study demonstrates systematic approaches to optimizing microbial biosynthesis, positioning C. glutamicum as a promising platform for sustainable 1,5‐PDO production.",
"conclusion": "3 Conclusion In this study, we successfully constructed and optimized a 1,5‐PDO biosynthesis system in C. glutamicum . Our initial approach incorporated both CoA‐dependent and CoA‐independent pathways for converting 5‐HV to 1,5‐PDO. Notably, the CoA‐independent pathway, which utilizes the CAR enzyme, was more effective in C. glutamicum . Subsequent efforts focused on enhancing the production of 5‐HV, a key intermediate in 1,5‐PDO biosynthesis. Through targeted genetic modifications and strain engineering, we developed the 5HV‐int strain, which significantly improved 5‐HV production while minimizing the accumulation of unwanted byproducts. Through this process, the introduction of the optimized CoA‐independent conversion module into the strain resulted in the creation of the 15PDO‐3 strain, which exhibited a marked increase in 1,5‐PDO production. Despite these advancements, the incomplete conversion of 5‐HV to 1,5‐PDO, which results in 5‐HV accumulation, was observed. To address this issue, we comprehensively screened different CAR enzyme candidates, ultimately identifying MAP1040 as the most effective enzyme for converting 5‐HV to 1,5‐PDO. Further optimization of MAP1040 through strategies such as substrate‐binding cavity engineering, polarity modulation, and substrate access tunnel engineering resulted in the M296E mutant, which significantly enhanced 1,5‐PDO production. Nevertheless, the pursuit of efficient 1,5‐PDO production revealed ongoing challenges during fed‐batch fermentation, particularly in achieving substantial titer improvements. A deeper investigation into the metabolic constraints suggested a potential scarcity of NADPH, which is heavily utilized in both the l ‐lysine and 1,5‐PDO biosynthesis pathways. To overcome this, we employed deep learning‐based analysis to screen various aldehyde reductase candidates and assess their preference for NADH or NADPH as reducing agents. Among the 25 candidates screened, eight demonstrated a preference for NADH, with the GOX1801 gene emerging as the most promising. The incorporation of GOX1801 resulted in a final 1,5‐PDO titer of 43.4 g L −1 , underscoring its effectiveness in improving production. These iterative metabolic engineering efforts highlight the potential for developing an efficient and robust bioprocess for 1,5‐PDO production. Our findings provide valuable insights into the optimization of biosynthetic pathways and emphasize the importance of enzyme engineering, cofactor balancing, and strain development in advancing industrial biotechnology.",
"introduction": "1 Introduction The term “global boiling” is now used to describe current climate change, as recent climate data highlight a significant departure from the conventional understanding of global warming. [ \n \n 1 \n \n ] This shift underscores the imperative of sustainable development, emphasizing the need to meet present needs without compromising the ability of future generations to meet their needs. [ \n \n 2 \n , \n 3 \n \n ] In line with this ethos, many metabolic engineers have directed their efforts toward developing efficient microbial cell factories capable of producing various value‐added products from sustainable resources. Among these, C5 platform chemicals, including 5‐aminovaleric acid (5‐AVA), glutaric acid (GTA), 5‐hydroxyvaleric acid (5‐HV), and 1,5‐pentanediol (1,5‐PDO), have garnered significant attention due to their industrial applications. These chemicals are pivotal in synthesizing various high‐value products and materials, contributing to the versatility and applicability of biobased processes. [ \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 12 \n , \n 13 \n , \n 14 \n , \n 15 \n , \n 16 \n , \n 17 \n , \n 18 \n , \n 19 \n \n ] \n Accordingly, considerable attention has been directed toward the microbial production of 5‐HV and 1,5‐PDO due to their diverse applications (Figure S1 , Supporting Information). Various metabolic pathways for 1,5‐PDO biosynthesis from glucose have been developed, employing different l ‐lysine conversion modules such as the cadaverine‐derived pathway and the traditional DavBA‐mediated pathway. Additionally, distinct 5‐HV conversion modules have been explored, including the CoA‐dependent pathway and the CoA‐independent CAR‐based pathway. In an early study, an artificial 1,5‐PDO biosynthesis pathway was constructed using the DavBA‐mediated l ‐lysine conversion module and CAR‐based 5‐HV conversion module, progressing through intermediates such as 5‐AVA, glutarate semialdehyde, 5‐HV, and 5‐hydroxyvaleraldehyde in recombinant Escherichia coli , resulting in 0.97 g L −1 1,5‐PDO production. [ \n \n 11 \n \n ] Another study introduced a CoA‐dependent 5‐HV conversion module in E. coli , where 5‐HV was converted to 1,5‐PDO via 5‐HV‐CoA and 5‐hydroxyvaleraldehyde through sequential enzymatic reactions involving CoA‐transferase, CoA‐acylating aldehyde dehydrogenase, and aldehyde reductase. This pathway achieved 3.19 g L −1 5‐HV and 0.35 g L −1 1,5‐PDO from glucose and l ‐lysine. [ \n \n 12 \n \n ] More recently, a cadaverine‐derived l ‐lysine conversion module was employed to synthesize 1,5‐PDO. In this approach, l ‐lysine was transformed into 5‐HV via intermediates such as cadaverine, 5‐aminovaleraldehyde, 5‐AVA, and glutarate semialdehyde. The subsequent conversion of 5‐HV into 1,5‐PDO was facilitated through a CAR‐mediated pathway. Using this design, an engineered E. coli strain harboring the pathway achieved a production titer of 9.25 g L −1 1,5‐PDO. [ \n \n 16 \n \n ] \n The microbial production of 1,5‐PDO is intricately linked to the l ‐lysine‐derived pathway, highlighting the potential of Corynebacterium glutamicum as a robust host strain. Known for its high efficiency in synthesizing l ‐lysine and other C5 chemicals, including 5‐AVA, GTA, and 5‐HV, C. glutamicum could serve as an ideal platform for developing 1,5‐PDO biosynthetic processes. [ \n \n 5 \n , \n 6 \n , \n 7 \n , \n 8 \n , \n 9 \n , \n 10 \n , \n 13 \n , \n 14 \n , \n 15 \n , \n 17 \n \n ] However, the bio‐based production of 1,5‐PDO presents significant technical challenges, primarily due to enzyme inefficiencies and energy demands across its biosynthetic pathways. [ \n \n 12 \n , \n 16 \n \n ] For 5‐AVA synthesis, two distinct pathways have been explored: the cadaverine‐based pathway and the traditional DavBA‐mediated pathway. The cadaverine‐based pathway is more energy‐ and electron‐efficient, generating fewer by‐products, recycling glutamate, and eliminating the need for molecular oxygen. These features make it advantageous for large‐scale anaerobic fermentation processes. However, in C. glutamicum , cadaverine is excreted as an end‐product or byproduct, thereby reducing overall pathway efficiency. [ \n \n 20 \n , \n 21 \n , \n 22 \n \n ] In contrast, the DavBA‐mediated pathway, although less efficient in terms of energy and electron utilization, has been reported to achieve higher levels of 5‐AVA production in C. glutamicum while minimizing significant byproduct accumulation. [ \n \n 5 \n , \n 7 \n , \n 10 \n , \n 17 \n \n ] For the conversion of 5‐HV to 1,5‐PDO, two artificial modules have been developed for this conversion: the CoA‐dependent module and the CAR‐based direct reduction module. The CoA‐dependent pathway is theoretically more energy‐efficient, consuming less ATP per reaction. However, its practical utility is constrained by the low activity and substrate specificity of key enzymes such as CoA transferases and CoA‐acylating aldehyde dehydrogenases, resulting in low product titers. In contrast, the CAR‐based module benefits from a lower thermodynamic barrier and irreversible reductions, providing a robust driving force for 1,5‐PDO production. This makes the CAR‐based pathway a more industrially applicable route when enzyme inefficiencies hinder the CoA‐dependent route, countering its disadvantage of lower energy efficiency. A comparative analysis of these modules reveals trade‐offs between energy efficiency and enzyme activity. The cadaverine‐based 5‐AVA module paired with the CoA‐dependent 1,5‐PDO module theoretically achieves the highest yield but is hampered by enzyme limitations. [ \n \n 16 \n \n ] Alternatively, the DavBA‐mediated 5‐AVA pathway coupled with the CAR‐based 1,5‐PDO module provides a better balance of pathway efficiency and enzyme activity, making it more suitable for implementation in C. glutamicum . This approach was exemplified in an engineered C. glutamicum strain capable of producing 5‐HV, a key precursor for the subsequent conversion to 1,5‐PDO. In this 5‐HV biosynthetic pathway, the initial three steps of the l ‐lysine catabolic pathway, mediated via 5‐aminovaleramide and encoded by the Pseudomonas putida davTBA genes, were followed by an intracellular reduction step catalyzed by the E. coli yahK gene. By integrating an artificial 5‐HV biosynthesis pathway and eliminating byproduct pathways via gabD2 deletion, the engineered strain produced 52.1 g L −1 of 5‐HV during fed‐batch fermentation, achieving a yield of 0.33 g g −1 glucose. [ \n \n 13 \n \n ] This result underscores the potential of C. glutamicum as a platform for efficient 1,5‐PDO biosynthesis. Here, we address the metabolic engineering of C. glutamicum to produce 1,5‐PDO ( Figure \n \n 1 \n ). We first compared the CoA‐dependent and CoA‐independent 5‐HV conversion modules to establish an efficient biosynthetic pathway for 1,5‐PDO. Subsequent metabolic engineering of the base strain was conducted to optimize the production of the key precursor, 5‐HV, thereby enhancing the overall synthesis of 1,5‐PDO. Iterative improvements were then applied to increase the efficiency of each step in the 1,5‐PDO biosynthesis pathway. We systematically evaluated various carboxylic acid reductases (CARs) and their mutants using rational enzyme engineering techniques to identify the most suitable enzyme for converting 5‐HV into 5‐hydroxyvaleraldehyde. Next, we focused on identifying the most efficient aldehyde reductase for converting 5‐hydroxyvaleraldehyde into 1,5‐PDO. Ultimately, fed‐batch fermentation of the engineered strain produced 43.4 g L −1 1,5‐PDO. Figure 1 Schematic diagram of the metabolic engineering project for the development of 1,5‐PDO‐producing C. glutamicum . The abbreviations used are GLU, glucose; PYR, pyruvate; OAA, oxaloacetate; LYS, l ‐lysine; 5‐AVAM, 5‐aminovaleramide; 5‐AVA, 5‐aminovaleric acid; GSA, glutarate semialdehyde; 5‐HV, 5‐hydroxyvaleric acid; 5HVA, 5‐aminovaleramide; and 1,5‐PDO, 1,5‐pentanediol.",
"discussion": "2 Results and Discussion 2.1 Construction of the 1,5‐PDO Biosynthesis System In our previous study, we identified the DavB‐DavA‐DavT‐YahK‐mediated pathway as the most promising 5‐HV biosynthesis pathway in C. glutamicum ; the fed‐batch fermentation of C. glutamicum harboring this pathway, along with gabD2 deletion, produced 52.1 g L −1 5‐HV. [ \n \n 13 \n \n ] Consequently, we investigated the further conversion of 5‐HV into 1,5‐PDO on the basis of the established 5‐HV biosynthesis pathway. The previously reported CoA‐dependent ( Clostridium aminobutyricum AbfT + Clostridium saccharoperbytulacetonicum Bld L273T + E. coli YqhD) [ \n \n 12 \n \n ] and CoA‐independent ( Mycobacterium marinum MMAR2117 + Bacillus subtilis PPTase + E. coli YqhD) [ \n \n 11 \n , \n 16 \n \n ] pathways were examined via the engineered C. glutamicum ΔgabD (pCES208H30DavTYahKDavBhisA + pBL712H30AbfTBld L273T YqhD or pBL712H30MMAR2117PPTaseYqhD) strain through 120 h of flask cultivation. However, only l ‐lysine, GTA, and 5‐HV were produced after cultivation, with no conversion to 1,5‐PDO (Figure S2 , Supporting Information). Additionally, the transformation efficiency was too low, likely due to the large plasmid size. Therefore, based on the previous finding, the l ‐lysine conversion module (P H30 DavBhisA) was integrated into the chromosome of the C. glutamicum Δ gabD strain by disrupting lysE to prevent l ‐lysine into the culture medium. [ \n \n 13 \n , \n 23 \n , \n 24 \n , \n 25 \n , \n 26 \n , \n 27 \n \n ] As a result, we engineered a recombinant strain, C. glutamicum Δ gabD Δ lysE ::P H30 DavBhisA (referred to as C. glutamicum AVA‐int‐gd). To verify the functionality of this integrated module, flask cultivations of the C. glutamicum AVA‐int‐gd strain were performed to assess its ability to produce 5‐AVA (Figure S3 , Supporting Information). Next, for 5‐HV production, the plasmid‐based expression module of davT and yahK was introduced to 5‐AVA‐producing C. glutamicum strain AVA‐int‐gd. Further flask cultivation of C. glutamicum AVA‐int‐gd harboring pCES208H30DavTYahK was performed to assess its ability to produce 5‐HV (Figure S3 , Supporting Information). After confirming that each strain could produce 5‐AVA and 5‐HV, we introduced downstream pathways for CoA‐dependent conversion (AbfT‐Bld L273T ‐YqhD) and CoA‐independent conversion (MMAR2117‐PPTase‐YqhD), resulting in the development of the 15PDO‐1 and 15PDO‐2 strains, respectively ( Figure \n \n 2 A ). However, the flask cultivation of strains 15PD0‐1 and 15PDO‐2 did not yield 15PDO (Figure S4 , Supporting Information). To assess pathway functionality, we performed further batch fermentations of the 15PDO‐1 (Figure 2B ) and 15PDO‐2 (Figure 2C ) strains. For the 15PDO‐1 strain, minimal 1,5‐PDO production (0.04 g L −1 ) was achieved, with a substantial amount of 5‐HV (20.16 g L −1 ) remaining without undergoing further conversion. For the batch fermentation of 15PDO‐2 harboring the CoA‐independent pathway, 0.20 g L −1 1,5‐PDO and 18.79 g L −1 5‐HV were produced. Although both strains produced only marginal levels of 1,5‐PDO, the pathways were confirmed to be functional in C. glutamicum , as in the case of E. coli . Among the two pathways, the CoA‐independent (CAR‐based) pathway was more effective for 1,5‐PDO generation in C. glutamicum . Consequently, further engineering efforts were undertaken on the basis of the 15PDO‐2 strain. Figure 2 A) Metabolic pathway for the biosynthesis of 1,5‐PDO and plasmid construction for the expression of heterologous genes encoding the 1,5‐PDO biosynthesis pathway. The abbreviations used are GLU, glucose; PYR, pyruvate; OAA, oxaloacetate; LYS, l ‐lysine; 5‐AVAM, 5‐aminovaleramide; 5‐AVA, 5‐aminovaleric acid; GSA, glutarate semialdehyde; 5‐HV, 5‐hydroxyvaleric acid; 5HV‐CoA, 5‐hydroxyvaleryl‐CoA; 5HVA, 5‐aminovaleramide; and 1,5‐PDO, 1,5‐pentanediol. B) Batch fermentation of the C. glutamicum 15PDO‐1 strain for 1,5‐PDO production. C) Batch fermentation of the C. glutamicum 15PDO‐2 strain for 1,5‐PDO production. 2.2 Improvement of the 5‐HV Production System in C. glutamicum and its Further Application to 1,5‐PDO Biosynthesis The present study aims to achieve high 1,5‐PDO production. To this end, it is crucial to establish an efficient 5‐HV production system that supports substantial metabolic flux toward 1,5‐PDO. This significance arises from the critical role of 5‐HV as the key precursor in 1,5‐PDO synthesis. The initially constructed 1,5‐PDO biosynthesis system was inefficient and suffered from low transformation efficiency because the final plasmid size exceeded 10 kb. As a result, further engineering of the C. glutamicum AVA‐int‐gd strain was performed by integrating a 5‐AVA conversion module (P H30 ‐DavTYahK) at the gabD3 site. Consequently, the C. glutamicum Δ gabD2 Δ lysE ::H30 davB His A Δ gabD3 ::H30 davTyahK strain ( C. glutamicum 5HV‐int strain) was developed ( Figure \n \n 3 A ). To summarize, building on the previously developed 5HV‐7 strain, [ \n \n 13 \n \n ] which features the deletion of the gabD2 gene and the introduction of plasmids pCES208H30DavTBhisA and pBL712H30YahK, additional engineering strategies were applied to further enhance 5‐HV production in C. glutamicum . These modifications included the deletion of the lysE gene, achieved through the integration of P H30 ‐DavBhisA, to block L‐lysine excretion and enhance L‐lysine conversion, as well as the deletion of the gabD3 gene, facilitated by the integration of P H30 ‐DavTYahK, to prevent GTA formation and promote the conversion of 5‐AVA.\nSubsequent flask cultivation of the strain without plasmid‐based expression of the 5‐HV biosynthesis module produced 3.4 ± 0.97 g L −1 5‐HV, which was 5.4 fold greater than that achieved with the 5HV‐7 strain. [ \n \n 13 \n \n ] Additionally, there was a significant reduction in intermediate metabolites: 0.34 g L −1 ± 0.007 l ‐lysine, 0.40 ± 0.03 g L −1 5‐AVA, and 0.28 ± 0.001 g L −1 GTA (Figure 3B ). Further batch fermentation of the 5HV‐int strain produced 16.93 g L −1 5‐HV. Other metabolites were present at only minimal concentrations: 0.38 g L −1 \n l ‐lysine, 0.67 g L −1 5‐AVA, and 0.48 g L −1 GTA (Figure 3C ). As a result, it is revealed that integrating P H30 ‐DavBhisA and P H30 ‐DavTYahK into the genome enabled pathways for a more stable and balanced expression. These targeted modifications significantly improved the efficiency of 5‐HV production, underscoring the effectiveness of precise genetic engineering in optimizing metabolic pathways and minimizing byproduct accumulation. Figure 3 A) Metabolic engineering strategy for the development of the C. glutamicum 5HV‐int strain. The abbreviations used are GLU, glucose; PYR, pyruvate; OAA, oxaloacetate; LYS, l ‐lysine; 5‐AVAM, 5‐aminovaleramide; 5‐AVA, 5‐aminovaleric acid; GSA, glutarate semialdehyde; 5‐HV, 5‐hydroxyvaleric acid; 5HVA, 5‐aminovaleramide; and 1,5‐PDO, 1,5‐pentanediol. B) Flask cultivation of the C. glutamicum 5HV‐int strain for 5‐HV production. All flask cultures were performed in triplicate. The measurements are presented as the means ± standard deviations. C) Batch fermentation of the C. glutamicum 5HV‐int strain for 5‐HV production. D) Metabolic engineering strategy for the development of the C. glutamicum 15PDO‐3 strain. The abbreviations used are 5‐HV, 5‐hydroxyvaleric acid; 5HVA, 5‐aminovaleramide; and 1,5‐PDO, 1,5‐pentanediol. E) Flask cultivation of the C. glutamicum 15PDO‐3 strain for 1,5‐PDO production. All flask cultures were performed in triplicate. The measurements are presented as the means ± standard deviations. F) Batch fermentation of the C. glutamicum 15PDO‐3 strain for 1,5‐PDO production. Next, the CAR‐based 5‐HV conversion module (MMAR2117‐PPTase‐YqhD), which achieved higher 1,5‐PDO production than the CoA‐based 5‐HV conversion module, was introduced into the C. glutamicum 5HV‐int strain (Figure 3D ). The resulting strain, designated 15PDO‐3, was then examined through flask cultivation, yielding 3.2 ± 0.27 g L −1 5‐HV and 0.3 ± 0.09 g L −1 1,5‐PDO (Figure 3E ), achieving the undetectable production of other metabolites. Further batch fermentation of the 15PDO‐3 strain was conducted to monitor cell growth and metabolite production over time (Figure 3F ), producing 9.37 g L −1 5‐HV and 5.48 g L −1 1,5‐PDO. Negligible quantities of l ‐lysine and 5‐AVA were detected, with no GTA accumulation observed. However, despite the successful production of 1,5‐PDO, a substantial amount of 5‐HV remained in the culture medium without further conversion to 1,5‐PDO. As a result, modification of the 1,5‐PDO conversion system was deemed necessary. 2.3 Examination of Different CAR Candidates to Improve the 1,5‐PDO Biosynthesis System It was confirmed that 1,5‐PDO can be successfully produced in the established system. However, a low production level of 1,5‐PDO was observed, alongside inefficient conversion of 5‐HV into 1,5‐PDO. To address the issue of the inefficient conversion rate of 5‐HV into 1,5‐PDO, efforts were directed toward optimizing the reaction module mediated by CAR, which converts 5‐HV into 5‐hydroxyvaleraldehyde. Consequently, 13 additional CARs from various microorganisms were tested (Table S1 , Supporting Information). These 13 different CARs were introduced into the 5HV‐int strain, along with the expression of PPTase and YqhD, resulting in the development of 13 distinct engineered C. glutamicum strains: C. glutamicum 15PDO‐4 to 15PDO‐16 ( Figure \n \n 4 A ). Further cultivation of the strains in flasks revealed that the conversion of 5‐HV into 1,5‐PDO by six out of the 13 CARs was effective. Among the six CARs, MAP1040 emerged as the most effective candidate for 1,5‐PDO production. The C. glutamicum 15PDO‐13 strain expressing MAP1040, PPTase, and YqhD produced 1.3 ± 0.62 g L −1 1,5‐PDO and 0.72 ± 0.21 g L −1 5‐HV. With the established C. glutamicum 15PDO‐13 strain, which achieved the highest 1,5‐PDO production in flask cultivation, further batch fermentation was conducted to examine whether MAP1040 could increase 1,5‐PDO production. This fermentation process produced 4.53 g L −1 5‐HV and 14.30 g L −1 1,5‐PDO (Figure 4B ). Compared with that of the C. glutamicum 15PDO‐3 strain, 5‐HV accumulation decreased 0.27 fold, whereas 1,5‐PDO production increased 2.6 fold. This suggests that the newly introduced MAP1040 successfully converted 5‐HV to 1,5‐PDO via 5‐valeraldehyde. Moreover, other byproducts, such as l ‐lysine, 5‐AVA, and GTA, were not detected. Figure 4 A) Flask cultivation of C. glutamicum 15PDO‐3 to 16 strain strains for 1,5‐PDO production. All flask cultures were performed in triplicate. The measurements are presented as the means ± standard deviations. B) Batch fermentation of the C. glutamicum 15PDO‐13 strain for 1,5‐PDO production. C) Fed‐batch fermentation of the C. glutamicum 15PDO‐13 strain for 1,5‐PDO production. Subsequently, fed‐batch fermentation of the C. glutamicum 15PDO‐13 strain was performed to increase 1,5‐PDO production (Figure 4C ). However, the fed‐batch fermentation results exhibited distinct patterns of cell growth reduction compared with those of previous C. glutamicum fed‐batch fermentations. [ \n \n 9 \n , \n 13 \n , \n 14 \n , \n 20 \n , \n 28 \n \n ] During fermentation, a notable reduction in cell growth was observed after 36 h, with the OD 600 decreasing from a peak of 103.52–89.72. This decline continued throughout fermentation, resulting in a final OD 600 of 38.71 after 75 h. Despite the consistent glucose feeding and the maintenance of its concentration at 10–20 g L −1 , there was no substantial increase in 1,5‐PDO production. By the 48th h of fed‐batch fermentation, the production of 1,5‐PDO and 5‐HV reached 16.33 and 17.01 g L −1 , respectively, with a productivity rate of 0.34 g·L −1 ·h −1 . After 72 h of fermentation, the final concentration of 1,5‐PDO reached 17.83 g L −1 , while that of 5‐HV increased significantly, peaking at 28.93 g L −1 . The total carbon yield of the metabolites was calculated as 0.32 mol mol −1 . These issues, including a sharp decline in cell growth and inefficient conversion of 5‐HV to 1,5‐PDO in the later stages of fed‐batch fermentation, highlight the need for further modifications to the established system. Therefore, additional exploration and adjustments were deemed necessary to increase 1,5‐PDO production. 2.4 Engineering a Carboxyl Group Reduction Module to Increase 1,5‐PDO Production The accumulation of 5‐HV in the culture medium indicated that the conversion of 5‐HV into 1,5‐PDO via 5‐hydroxyvaleraldehyde was a limiting step. This phenomenon is presumably due to 5‐hydroxyvaleraldehyde, a highly reactive intermediate formed from 5‐HV, which poses challenges in the 1,5‐PDO biosynthesis pathway because of its potential to interact with and modify enzyme structures. Aldehydes are highly electrophilic compounds known to exhibit reactivity with various biomolecules, including amino acids and cellular membranes. For example, glycolaldehyde has been reported to interact with the amino and thiol groups of amino acids, inducing protein cross‐linking and compromising structural stability. [ \n \n 29 \n \n ] Similarly, long‐chain alkanals such as hexaldehyde and 4‐hydroxynonanal have been shown to react with cell membranes, causing damage, and to modify proteins through interactions with lysine, cysteine, and histidine residues. These reactions often involve the formation of Schiff bases, resulting in structural modifications and functional inactivation of proteins. In the case of 5‐hydroxyvaleraldehyde, similar toxic effects are hypothesized due to its reactivity with cell membranes and specific amino acids within enzymes. To investigate these potential interactions, structural modeling and computational analysis using MoleOnline were performed to identify the aldehyde product release pathway in the R‐domain of the MAP1040. The analysis revealed several residues along the putative product pathway that are prone to aldehyde reactivity, including three lysines, one cysteine, and four histidines (Figure S5 , Supporting Information). These residues are highly susceptible to Schiff base formation and other covalent modifications, which could lead to structural destabilization or inactivation of the enzyme. This finding highlights the inherent challenge of aldehyde accumulation during bioproduction processes. The reactivity of 5‐hydroxyvaleraldehyde could contribute to both enzyme inactivation and cell viability reduction. Therefore, optimizing the reaction rate of 5‐hydroxyvaleraldehyde and its subsequent conversion to 1,5‐PDO is critical for maximizing the 1,5‐PDO productivity. In this regard, the catalytic activity of CAR (MAP1040) was fine‐tuned by modulating its catalytic efficiency. The catalytic efficiency of MAP1040 was altered by engineering the amino acid residues near the active site and substrate‐binding cavity of the enzymes. For example, S299 was replaced with a bulkier amino acid residue (valine) to reduce the size of the substrate‐binding cavity. Additionally, mutations, such as M296E, M422E, R272A, N465T, and M269F, were introduced to alter the size and polarity of the active site and substrate‐binding cavity ( Figure \n \n 5 A,B,C ). After each mutation was introduced into the enzyme, the resulting plasmids, pCES208H30MAP1040mutPPTase and pBL712H30YqhD, were transformed into the 5HV‐int strain, generating the 15PDO‐13(mutant) strains. Then, flask cultivation was performed for each strain (Figure 5D ). The introduction of the M296E, R272A, and S299V mutants resulted in 1,5‐PDO production levels comparable to those of wild‐type MAP1040. Specifically, the 15PDO‐13(M296E) strain produced 1.7 ± 0.82 g L −1 5‐HV and 1.14 ± 0.87 g L −1 1,5‐PDO, the 15PDO‐13(R272A) strain generated 2.14 ± 0.98 g L −1 5‐HV and 0.89 ± 0.32 g L −1 1,5‐PDO, and the 15PDO‐13(S299V) strain yielded 1.49 ± 0.09 g L −1 5‐HV and 1.15 ± 0.98 g L −1 1,5‐PDO. However, the other recombinant strains presented lower 1,5‐PDO concentrations. Therefore, batch fermentations of the 15PDO‐13(M296E; Figure 5E ), 15PDO‐13(R272A; Figure 5f ), and 15PDO‐13(S299V; Figure 5G ) strains were conducted to examine the effects of CAR activity variations on 1,5‐PDO productivity in detail. Figure 5 A) Structure of carboxylic acid reductase (MAP1040) modeled by AlphaFold2, which consists of three multiple domains. B) Mutation sites within the adenylation domain of the CAR. C) Close‐up view of the active site and substrate‐binding cavity. The substrate 5‐HV is colored in green, and AMP is colored in yellow. D) Metabolic engineering strategies devised for the development of efficient CAR mutants: flask cultivation of C. glutamicum 15PDO strains for 1,5‐PDO production. All flask cultures were performed in triplicate. The measurements are presented as the means ± standard deviations. E) Batch fermentation of the C. glutamicum 15PDO‐13(M296E) strain for 1,5‐PDO production. F) Batch fermentation of the C. glutamicum 15PDO‐13(R272A) strain for 1,5‐PDO production. G) Batch fermentation of the C. glutamicum 15PDO‐13(S299V) strain for 1,5‐PDO production. Remarkably, the 15PDO‐13(M296E) strain demonstrated a significant increase in 1,5‐PDO production, reaching a peak concentration of 23.5 g L −1 without the formation of byproducts, such as l ‐lysine, 5‐AVA, GTA, and 5‐HV (Figure 5E ). This strain also achieved a peak OD 600 of 195.1, approximately double that of the 15PDO‐13 strain. In contrast, the other strains exhibited similar or inferior 1,5‐PDO production compared with the 15PDO‐13 strain. This result indicates that the intracellular catalytic activity of the MAP1040_M296E variant is nearly optimal for converting 5‐HV into 1,5‐PDO via 5‐hydroxyvaleraldehyde. Then, on the basis of the favorable M296E mutant, (M296E/S299V) and (M296E/R272A) double mutants were generated to explore potential synergistic effects. However, subsequent batch fermentation of these strains did not improve 1,5‐PDO production (Figure S6 , Supporting Information). To understand the high product yield achieved during batch fermentation of the 15PDO‐13(M296E) strain, we performed a detailed analysis of the substrate‐binding cavity in the MAP1040(M296E) model ( Figure \n \n 6 A ) and conducted kinetic studies of both MAP1040 and its variant (Figure 6B ). Kinetic analysis, based on NADPH oxidation, revealed that the catalytic efficiency of the MAP1040(M296E) enzyme was 11 fold lower than that of the wild‐type enzyme (Figure 6B ). This finding suggests that the M296E mutation reduced the rate of 5‐hydroxyvaleraldehyde formation, likely preventing aldehyde accumulation within recombinant cells. Such a reduction in aldehyde levels may alleviate stress on both enzymes and cells. Figure 6 A) Close‐up view of the substrate‐binding cavity of the M296E mutant protein. The substrate 5‐HV is colored in green, and AMP is colored in yellow. B) Kinetic parameters of MAP1040 and its mutant (M296E). C) Substrate tunnel analysis of MAP1040 and its mutant (M296E). D) Fed‐batch fermentation of the C. glutamicum 15PDO‐13(M296E) strain for 1,5‐PDO production. Comparative analysis of the substrate tunnels in the wild‐type MAP1040 and M296E mutant revealed significant changes in polarity. Specifically, the substitution of methionine with glutamate at position 296 introduced a pronounced negative electrostatic potential to the substrate tunnel. This alteration is hypothesized to cause electrostatic repulsion with the carboxylate (COO⁻) group of 5‐HV, likely hindering the efficient formation of the ES complex in the active site (Figure 6C ). This corresponds to the previous steady‐state kinetic results, where the K m value of M296E was significantly lower than that of the wild‐type. This effect may explain the higher cell density observed during fermentation, as reduced aldehyde accumulation would alleviate cellular toxicity. Overall, these results suggest that maintaining CAR activity at a level that prevents accumulation of the toxic reaction intermediate, 5‐hydroxyvaleraldehyde, is crucial. Based on these findings, the 15PDO‐13(M296E) strain, which demonstrated the highest 1,5‐PDO production, was selected for further optimization and investigation through fed‐batch fermentation. During fed‐batch fermentation of the 15PDO‐13(M296E) strain, the glucose concentration was carefully maintained between 10 and 20 g L −1 (Figure 6D ). At the end of fermentation, the final concentrations of 1,5‐PDO and 5‐HV were 24.99 and 22.01 g L −1 , respectively, with no other byproducts detected. However, similar to the 15PDO‐13 strain, there was no significant improvement in 1,5‐PDO production despite continuous glucose feeding and maintenance within the specified concentration range. Notably, a substantial decrease in cell growth was observed after 39 h, mirroring the pattern observed in the fed‐batch fermentation of the 15PDO‐13 strain. After 39 h of fed‐batch fermentation, 1,5‐PDO production reached 21.39 g L −1 , with a yield of 0.12 mol mol −1 and a productivity of 0.55 g·L −1 ·h −1 . Concurrently, 14.34 g L −1 5‐HV was produced, and the total carbon yield of the l ‐lysine‐derived metabolites was 0.19 mol mol −1 . Beyond this point, no further increase in 1,5‐PDO production was observed, with the final titer reaching 19.35 g L −1 1,5‐PDO. On the other hand, 5‐HV production increased, reaching its highest titer of 22.01 g L −1 at the end of fermentation. While the overarching trends in the results mirrored those obtained from the fed‐batch fermentation of the 15PDO‐13 strain, fed‐batch fermentation of the 15PDO‐13(M296E) strain revealed a distinct outcome, with rapid cell growth, glucose consumption, and concurrent 1,5‐PDO production. This accelerated fermentation demonstrated that the MAP1040(M296E) mutant played a pivotal role in enhancing 1,5‐PDO production. However, the limited increase in 1,5‐PDO production rate despite continuous glucose feeding and the excess 5‐HV accumulation strongly suggest potential bottlenecks or limitations within the engineered pathway. 2.5 Engineering an Aldehyde Group Reduction Module to Achieve Elevated Levels of 1,5‐PDO Production It was found that both 15PDO‐13 and 15PDO‐13(M296E) strains exhibited limited increases in 1,5‐PDO production and a sharp decline in cell growth after the peak OD 600 was reached during the fed‐batch fermentations. Therefore, we hypothesized that the toxic aldehyde intermediate 5‐hydroxyvaleraldehyde might be responsible for inhibiting cell growth. In previous fed‐batch fermentations, continuous accumulation of 5‐HV was observed. This observation suggests that 5‐hydroxyvaleraldehyde, which may cause cellular toxicity, is preferentially oxidized to 5‐HV rather than reduced to 1,5‐PDO, a process that relies on NAD(P)H as a cofactor. To evaluate the hypothesis, the 15PDO‐13(M296E)‐X strain was developed by introducing MAP1040(M296E) and PPTase into the 5HV‐int strain, with yqhD expression omitted. In this setup, the final step in 1,5‐PDO production—the reduction of 5‐hydroxyvaleraldehyde to 1,5‐PDO—relied on the action of yahK , which was integrated into the genome of the 5HV‐int strain. Batch fermentation of the 15PDO‐13(M296E)‐X strain was conducted to examine the patterns of cell growth and metabolite production. Compared with the 15PDO‐13(M296E) strain expressing MAP1040(M296E), PPTase, and YqhD, the 15PDO‐13(M296E)‐X strain expressing MAP1040(M296E) and PPTase exhibited a decrease in both the maximum OD 600 (114.29) and 1,5‐PDO production (8.68 g L −1 ) (Figure S7 , Supporting Information). This outcome supports the hypothesis that 5‐hydroxyvaleraldehyde is toxic to the cell, leading to reduced cell growth. To alleviate this toxicity, the oxidation of 5‐hydroxyvaleraldehyde to 5‐HV may be the preferred approach. In light of these findings, the yqhD gene was expressed on the basis of a high‐copy‐number plasmid, pHCP. [ \n \n 30 \n \n ] \n To examine whether an increase in the gene copy number would increase 1,5‐PDO production, the modified high‐copy plasmid pHCPH30‐MCS was used to express the yqhD gene. Next, two plasmids, pHCPH30YqhD and pBL712H30MAP1040(M296E)PPTaseYqhD, were cotransformed into the C. glutamicum 5HV‐int strain, generating the C. glutamicum 15PDO‐13(M296E)‐H strain. Batch fermentation of this strain was then conducted. However, unlike the expectation that high‐copy number plasmids would lead to elevated gene expression and increased product yield, the batch fermentation results did not show an increased production rate (Figure S8 , Supporting Information), achieving 1,5‐PDO production of 3.54 g L −1 reached its highest titer at the end of fermentation. In contrast, 5‐HV accumulated at a high concentration of 14.24 g L −1 . The discrepancy between the expected and actual outcomes may arise from the high expression level of yqhD , which catalyzes both the reduction of glutarate semialdehyde to 5‐HV and the reduction of 5‐hydroxyvaleraldehyde to 1,5‐PDO, with NADPH serving as a cofactor. In the 1,5‐PDO biosynthetic pathway we constructed, the production of 1 mole of 1,5‐PDO requires a total of 7 moles of NADPH and 1 mole of ATP. This includes 4 moles of NADPH for l ‐lysine biosynthesis and an additional 3 moles of NADPH along with 1 mole of ATP for the conversion of l ‐lysine to 1,5‐PDO. Given the heavy reliance of the established metabolic pathway on NADPH, increasing the action of yqhD would further strain the available NADPH supply. Therefore, rather than amplifying yqhD expression, which intensifies NADPH consumption, a more balanced reaction step needs to be devised. This may involve enhancing NADPH regeneration systems or identifying alternative enzymes that function more efficiently under existing metabolic conditions. To address the NADPH limitation in 1,5‐PDO biosynthesis, several strategies were employed to enhance NADPH availability. These included introducing transhydrogenases, PntAB and UdhA from E. coli , which transfer electrons from NADH to NADP⁺; NADH kinases, Pos5 from S. cerevisiae , which phosphorylate NADH to generate NADPH; and glucose dehydrogenases, GDH from B. subtilis , which regenerate NADPH from NADP⁺. [ \n \n 19 \n , \n 31 \n , \n 32 \n , \n 33 \n \n ] Accordingly, these genes were incorporated into the C. glutamicum 15PDO‐13(M296E) strain, resulting in derivative strains: 15PDO‐13(M296E)ecp, 15PDO‐13(M296E)ecu, 15PDO‐13(M296E)scp, and 15PDO‐13(M296E)bsg (Figure S9a , Supporting Information). However, flask cultivations of the engineered strains adversely showed a significant decrease in 1,5‐PDO production (Figure S9b , Supporting Information). These results indicate that enhancing NADPH regeneration alone is insufficient to improve 1,5‐PDO production and may even have detrimental effects. This limitation is likely due to a combination of factors, including metabolic imbalances, competition for cellular resources such as ATP, and unanticipated interactions within the metabolic network. Furthermore, these findings highlight the challenges of static NADPH regulation strategies, which cannot adapt to the dynamic NADPH demands that occur during different phases of cell growth and production. [ \n \n 34 \n \n ] In the case of 1,5‐PDO biosynthesis, which imposes a high NADPH demand, such rigid approaches often result in imbalances in the NADPH/NADP⁺ ratio, leading to disruptions in cell growth and production efficiency. This underscores the importance of developing dynamic, context‐specific strategies for NADPH regulation to effectively balance redox homeostasis and meet the metabolic demands of high‐performance bioproduction systems. [ \n \n 34 \n \n ] \n Subsequently, various aldehyde reductases from different microorganisms which utilized NADH as a cofactor were assessed to tackle the heavy reliance on NADPH of yqhD . [ \n \n 35 \n , \n 36 \n , \n 37 \n \n ] Accordingly, the cofactor preference prediction for aldehyde reductase candidates was performed using the deep learning‐based DISCODE model. This model was pre‐trained on a diverse dataset of 7,132 NAD(P) + ‐binding sequences from the Swiss‐Prot database, encompassing a wide range of structural domains and ensuring the universality of its predictions. [ \n \n 38 \n \n ] Using the deep learning‐based DISCODE model, the cofactor preference for NAD(P) + was evaluated for 25 aldehyde reductases from various microorganisms on the basis of previous reports and the KEGG database. Among them, eight aldehyde reductases showed a preference for NADH (Table S2 , Supporting Information; Figure \n \n 7 A ). These eight aldehyde reductases (YihU, PA1146, RHA07897, CpnD, Gbd, ButA, GOX1801, and GOX2181) were then tested for 1,5‐PDO production through flask cultivation, replacing the yqhD gene. Consequently, C. glutamicum strains 15PDO‐13(M296E)Y, 15PDO‐13(M296E)P, 15PDO‐13(M296E)R, 15PDO‐13(M296E)C, 15PDO‐13(M296E)Gb, 15PDO‐13(M296E)B, 15PDO‐13(M296E)G18, and 15PDO‐13(M296E)G21 were developed by expressing MAP1040(M296E), PPTase, and each aldehyde reductase (YihU, PA1146, RHA07897, CpnD, Gbd, ButA, GOX1801, and GOX2181, respectively). After flask cultivation of each strain, GOX1801 was identified as the most effective aldehyde reductase for the final reduction of 5‐hydroxyvaleraldehyde to 1,5‐PDO, supported by the highest 1,5‐PDO production, with a final titer of 1.72 ± 0.22 g L −1 (Figure 7B ). A structure‐based computational analysis was performed to examine the cofactor binding site of GOX1801 (Figure 7C ). In typical NADP + ‐dependent enzymes, such as γ‐hydroxybutyrate dehydrogenase from Geobacter sulfurreducens (PDB: 3PDU), residues like N31 and R32 are conserved to interact with the phosphate group of NADPH. [ \n \n 39 \n \n ] In contrast, GOX1801 was found to possess alanine and proline at these positions, which likely limits its ability to effectively interact with phosphate. Instead, S35 is proposed to form hydrogen bonds with the hydroxyl groups of NAD + , thereby facilitating its stable association within the binding site. Figure 7 A) Deep learning‐based analysis of various aldehyde reductases. B) Flask cultivation of C. glutamicum strains for 1,5‐PDO production. All flask cultures were performed in triplicate. The measurements are presented as the means ± standard deviations. C) Structural analysis and cofactor docking simulation of GOX1801 in comparison with γ‐hydroxybutyrate dehydrogenase from G. sulfurreducens (PDB: 3PDU). D) Batch fermentation of the C. glutamicum 15PDO‐13(M296E)G18 strain for 1,5‐PDO production. E) Fed‐batch fermentation of the C. glutamicum 15PDO‐13(M296E)G18 strain for 1,5‐PDO production. Next, batch fermentation of the strain was performed (Figure 7D ). As a result, the C. glutamicum 15PDO‐13(M296E)G18 strain could produce comparable amounts of 1,5‐PDO compared with the C. glutamicum 15PDO‐13(M296E) strain, with a final titer of 21.2 g L −1 and no formation of byproducts. Although the maximum OD 600 was lower than that of C. glutamicum 15PDO‐13(M296E), we proceeded with fed‐batch fermentation to verify whether the strain could sustain continuous 1,5‐PDO production while maintaining its biomass. During fed‐batch fermentation of the 15PDO‐13(M296E)G18 strain, the glucose concentration was maintained within a range of 10–20 g L −1 (Figure 7E ). Unlike the previous fed‐batch fermentation patterns, the strain maintained its biomass at an OD 600 of ≈100 for 20 h, after which it gradually decreased. Furthermore, 1,5‐PDO production continued to increase, reaching a peak of 43.4 g L −1 with a yield of 0.48 mol mol −1 without any other by‐products detected. This represents the highest reported titer to date. This accomplishment highlights the success of our metabolic engineering strategy and highlights the potential of C. glutamicum as a viable platform for the industrial bioproduction of valuable chemicals, such as 1,5‐PDO."
} | 11,044 |
33322237 | PMC7764841 | pmc | 211 | {
"abstract": "In numerous fields such as aerospace, the environment, and energy supply, ice generation and accretion represent a severe issue. For this reason, numerous methods have been developed for ice formation to be delayed and/or to inhibit ice adhesion to the substrates. Among them, laser micro/nanostructuring of surfaces aiming to obtain superhydrophobic behavior has been taken as a starting point for engineering substrates with anti-icing properties. In this review article, the key concept of surface wettability and its relationship with anti-icing is discussed. Furthermore, a comprehensive overview of the laser strategies to obtain superhydrophobic surfaces with anti-icing behavior is provided, from direct laser writing (DLW) to laser-induced periodic surface structuring (LIPSS), and direct laser interference patterning (DLIP). Micro-/nano-texturing of several materials is reviewed, from aluminum alloys to polymeric substrates.",
"conclusion": "5. Conclusions Undesired ice formation poses an increasingly serious threat in many industrial fields, particularly in aeronautics and telecommunications installations. Currently, the main strategies for reducing the problem are still based on active traditional methods such as mechanical, electro-thermal and liquid hybrid approaches. These approaches are definitively inefficient, energy-consuming, expensive, and can cause systems failure. Passive anti-icing coatings have also been proposed as an alternative to reduce or even replace the traditional de-icing methods. However, the existing available techniques present some disadvantages, such as their short-lived effectiveness, high manufacturing costs, and, in the case of chemical methods, their high environmental impact. In recent years, several studies have demonstrated a correlation between superhydrophobicity and anti-icing properties. Based on the surface wetting theory, substrates characterized by surface micro–nanostructures and low surface free energy can well conform to the wetting regime described by the Cassie–Baxter model. Such surfaces exhibit high contact angles (CA), low hysteresis (CHA) and sliding angles (SA) and, consequently, low droplet adhesion. In sub-zero temperature environments, high CAs usually result in an increase of the droplet freezing time, while low CHAs and SAs lead to a short contact time between dripping droplets and the surfaces, thus hindering ice formation. Indeed, low ice adhesion has been reported on SHSs. Among all the technologies for changing surface topography and obtaining SHSs, laser micromachining is considered to have many advantages. It offers extreme flexibility with respect to the morphology of the structures and surface micro-/nano-features that can be created, without posing any restrictions as to the type of material to be treated. The possibility of scaling up such technology to structure large areas with a reduction in process time and cost is also very attractive for industrial implementation. Several approaches (i.e., DLW, to LIPSS and DLIP) in different ranges, from nano- to microscale, have been reported in the literature to produce SHSs with anti-icing properties. In all cases, a strong influence of the machined structures on the anti-icing behavior was found. Regardless of the materials, a texture smaller than the droplet size may result in an anti-icing behavior with a delay in the droplet freezing time. This can be explained with both ice nucleation theory and thermodynamics. In fact, the nucleation rate decreases when CA increases, delating the ice formation. Moreover, the air trapped between the water droplet and the rough surface, acting as a thermal insulator, contributes to a reduction in the heat transfer between the solid and the liquid surfaces. Similarly, it was demonstrated that the water repellence can induce supercooled water droplets dripped onto the SH material to bounce away rapidly from the surface before completely freezing. Finally, even though an ice film is eventually formed on the functionalized surface, the superhydrophobic wetting regime may induce a smaller contact interface of solid/ice, thus allowing a lower ice adhesion and the pursuit of the aim of preventing ice accumulation.",
"introduction": "1. Introduction Ice in its several forms, i.e., frost, glaze, rime, snow, can cause severe problems for locks and dams [ 1 ], solar panels [ 2 ], wind turbines [ 3 ], aircraft [ 4 , 5 ], heat pumps [ 6 ], power lines or telecommunication equipment [ 7 ], civil engineering materials [ 8 ], and oil platforms [ 9 ], especially when it adheres and accumulates. These problems can increase energy consumption, reduce the energy conversion efficiency, origin mechanical and/or electrical malfunctions, and define safety hazards [ 10 , 11 ]. Among all the mentioned fields, the aerospace industry is particularly affected by icing phenomena, which can occur both while aircraft are on the ground [ 12 ] and in the air [ 13 ]. In the first case, before flying, ice needs to be removed, since it alters aerodynamic properties, and ice fragments can be sheared off by aerodynamic drag and fly into the engines. In the latter case, the impact of supercooled water droplets found in clouds, depending on atmospheric conditions, may stick to the surface, creating a dangerous ice layer. Therefore, the overall recommendation of the safety authorities is to strengthen the ice protection systems (IPS) to all commercial aircraft [ 14 ]. These protection systems can be classified into two groups: anti-icing systems, which aim to prevent ice formation, and de-icing systems, which aim at the removal of already-accumulated ice or frost from a surface. In recent years, great attention has been paid to the anti-icing surfaces, as attested by the increasing number of research papers and patents related to the topic “anti-icing” ( Figure 1 ). It can be difficult to find a common definition of anti-icing, but some ubiquitous features must be exhibited by such surfaces, i.e., (i) inhibition of water condensation on the surface; (ii) inhibition of the incoming water’s freezing; and (iii) weakening of the adhesion strength of ice, when formed, so that it can be easily removed [ 16 ]. Different strategies have been employed to prevent ice accretion and to easily remove it, and they can be divided into active and passive approaches. Keeping the surface temperature above the freezing point by means of electro-thermal heating is an efficient active method to reduce the likelihood of ice formation or to enhance its melting [ 17 ]. The exploitation of the Joule effect for heating line conductors is at present recognized as the most efficient engineering approach [ 18 ], especially in the case of transmission lines. However, this strategy is extremely time and energy consuming. In addition, electromagnetic disturbance can be generated by the passage of electric flow, which could interfere with the operation of the apparatus. Another method is to use freezing point depressants. It is well known that salt depresses the freezing point to the eutectic point, facilitating the ice in melting. Instead, in order to prevent icing and frosting of water on aircrafts surfaces, organic liquids are exploited, thanks to their lower crystallization temperatures compared to plain water [ 13 ]. In particular, for commercial aircraft in the northern hemisphere, various de-icing fluids are used [ 12 , 19 ]. They are essential not just on the exposed surfaces of aircraft wings, but also on other aerodynamic areas such as under the wings and on the rear spar stabilizer areas. Two standard test methods, namely Wet Spray Endurance Test [ 20 ] and Boundary Layer Displacement Thickness [ 21 ] are employed in climatically controlled wind tunnels to give a measure of the remaining film thickness on a surface during its continuous removal under an increasing flow of air over it. Both tests are critical for safe flight operation and are closely associated with density, surface tension and viscosity of the exploited fluids. This notwithstanding, the use of chemical agents to prevent ice formation has many drawbacks. First, their effectiveness is short-lived. Therefore, for large surfaces, periodic treatments are required. Furthermore, extensive use of these liquids, besides having high costs, can cause environmental problems [ 22 ]. Other active de-icing techniques consist of using mechanical forces to remove ice accretion from surfaces. With this aim, direct scraping or mechanical removal by shock waves, vibrations, or the twisting of conductors are applied. However, these methods often require personnel to reach the lines and towers; furthermore, helicopters or even shotguns must also be exploited when the ice is less accessible [ 18 ]. Moreover, applying mechanical force during de-icing causes additional stress to the networks, which in some cases can lead to failure. Consequently, such methods are certainly neither safe nor efficient. In addition, many other anti-icing and deicing methods have been studied and applied, e.g., electro-impulse systems, shape memory alloys, and ultrasound technology [ 23 ]. In addition, hybrid solutions consisting of material with a combination of passive anti-icing and active deicing functionalities have also been reported [ 24 ]. In [ 25 ], the prevention of freezing above −14 °C without any source of power input was claimed in the case of incoming water by using a spray-coating of perfluorododecylated graphene nanoribbons (FDO-GNRs), whereas resistive heating represents an alternative active deicing method in more extreme sub-zero environmental conditions. Thus, most of the active or hybrid strategies striving against ice formation are apparently inefficient, energy-consuming, expensive, or dangerous to the environment. It is thus crucial to develop innovative, green, cost-effective, and efficient technologies for anti-icing and deicing [ 11 ]. A new development perspective for pure passive anti-icing methods is surface micro- and nano-structuring. Recently, several efforts have been devoted towards the modification of surface topography and/or chemistry in order to obtain superhydrophobic properties [ 26 , 27 ]. A close relationship between the water repellency of superhydrophobic surfaces (SHSs) and icephobicity has been demonstrated [ 28 ]. Indeed, the air trapped between the water droplets and the underlying surface texture allows the surfaces to exhibit not only high water contact angle (WCA) and low contact angle (CA) hysteresis [ 29 ], but also prevent ice formation because of reduced water adhesion. Consequently, using SHSs for anti-icing application is undoubtedly an inherently attractive strategy. Among the different technologies available for surface micro- and nano-structuring, short and ultrashort laser pulses have several advantages: (i) they offer extreme flexibility in the morphology of the structures and surface micro-/nano-features that can be created [ 30 , 31 , 32 , 33 ]; (ii) in principle, they have no limitations in terms of manufacturable materials, as long as the targets absorb the laser wavelength [ 34 , 35 ]; and (iii) it is possible to scale up such technology [ 36 ] to structure large areas with industrially relevant process times and costs [ 37 ]. A comprehensive review of the most relevant works on the use of laser technology to functionalize surfaces with anti-icing properties is still lacking and would be beneficial to assess the latest advances and evaluate how mature is such a technology to be employed at an industrial scale. In the first part of this review, the key concepts regarding wettability, superhydrophobicity and anti-icing properties will be highlighted. Then, the main laser techniques used to fabricate anti-ice surfaces will be illustrated, ranging from Direct Laser Writing (DLW) to Laser-Induced Periodic Surface Structuring (LIPSS) and Direct Laser Interference Patterning (DLIP). The principal laser treatments carried out to obtain superhydrophobic and anti-icing substrates will be discussed in depth either for metallic or polymeric substrates."
} | 3,028 |
36582703 | PMC9792797 | pmc | 212 | {
"abstract": "Increasing untreated environmental outputs from industry and the rising human population have increased the burden of wastewater and other waste streams on the environment. The most prevalent wastewater treatment methods include the activated sludge process, which requires aeration and is, therefore, energy and cost-intensive. The current trend towards a circular economy facilitates the recovery of waste materials as a resource. Along with the amount, the complexity of wastewater is increasing day by day. Therefore, wastewater treatment processes must be transformed into cost-effective and sustainable methods. Microbial fuel cells (MFCs) use electroactive microbes to extract chemical energy from waste organic molecules to generate electricity via waste treatment. This review focuses use of MFCs as an energy converter using wastewater from various sources. The different substrate sources that are evaluated include industrial, agricultural, domestic, and pharmaceutical types. The article also highlights the effect of operational parameters such as organic load, pH, current, and concentration on the MFC output. The article also covers MFC functioning with respect to the substrate, and the associated performance parameters, such as power generation and wastewater treatment matrices, are given. The review also illustrates the success stories of various MFC configurations. We emphasize the significant measures required to fill in the gaps related to the effect of substrate type on different MFC configurations, identification of microbes for use as biocatalysts, and development of biocathodes for the further improvement of the system. Finally, we shortlisted the best performing substrates based on the maximum current and power, Coulombic efficiency, and chemical oxygen demand removal upon the treatment of substrates in MFCs. This information will guide industries that wish to use MFC technology to treat generated effluent from various processes.",
"conclusion": "6 Conclusion and future directives This review showcases the recent progress of substrate exploration and its performance in MFC systems and highlights the success stories of various configurations and the operating parameters that affect the efficacy of the MFCs. It also emphasizes the gaps that need to be addressed in the near future with the proposed cutting-edge research methodologies. There have been promising improvements in the MFC performance, and several groups have already reported field trials. A few tests are currently underway to understand the practical applications of this technology. It is well understood that the overall performance is associated with bacterial kinetics, and therefore, it has intrinsic limitations. However, to enhance MFCs and their performance for practical applications, more studies need to be conducted with the best-performing substrates for the purpose of deploying systems in real-scale wastewater treatment plants. In our opinion, apart from wastewater and power generation, MFCs can be used in several other applications, depending upon the system architecture and utilization of EABFs. A few areas need to be explored and are further described below: \n i) Reactor configuration and system architecture \n The MFC is suitable for treating diverse wastewater sources. However, due to their lower efficiency, they can be installed in a hybrid bioelectrochemical wastewater treatment plant. The hybrid process train can be composed of several stages. The stages include an anaerobic digester and an electrochemical oxidative reactor, followed by a MFC stack. For this configuration, highly efficient electrodes need to be developed and investigated for cost-effective scale up of MFCs. \n ii) Separator \n The proton exchange membrane has been generally explored to test the various configurations of the MFCs; however, polymer-based membranes cannot be the best option for building large-scale device fabrication. The major concern is that the polymer membrane cannot withstand high hydrostatic pressure, and therefore, it is essential to refocus efforts on ceramic-based membranes for MFC applications. \n iii) Cathode catalyst \n The catalyst plays a pivotal role in MFC systems, and several high-performing non-precious metal catalysts have already been examined. The best candidates should be tested in the ceramic-based electrode assembly for a performance and longevity study. Photosynthetic bacteria could be used to make self-sustained cathodic compartments by feeding CO2 generated by an anode chamber. The identification of novel bacteria that enhance the MFC performance will also benefit from the practical utility of the device. \n iv) Biocatalyst and sensors \n Apart from power generation and wastewater treatment, the MFC has a potential application in environmental sensors. There has been a great focus on making various amperometric MFC sensors for online BOD , toxicity, etc. However, these sensors have relied on bacterial kinetics, and bacterial composition can be easily altered upon slight variation of the substrate, pH, and operating parameters, resulting in inaccurate measurements. 3D biofilm printing of a pure EAB can be a promising approach to making highly reliable sensors that can address this problem. A 3D printed biofilm is prone to the issues mentioned above; however, 3D biofilm printing research is in the early stages and needs extensive work for 3D printed biofilms in MFC-based sensors for precise response. 3D biofilm printing technology can be a promising solution for desired bacterial catalysts, including the loading of bacteria and a highly defined biofilm layer that may enhance power generation and wastewater treatment.",
"introduction": "1 Introduction Industrialization and modern lifestyles have exploited the environment, creating sanitation issues for human beings and ecosystems [ 1 , 2 ]. Industrial expansion boosts the economy but brings with it waste disposal problems. Environmental release of untreated wastewater from both industry and household use leads to environmental contamination due to algal blooms and eutrophication [ 1 ]. Wastewater treatment is an energy-intensive and resource-intensive process [ 1 , 3 , 4 ]. This process results in the release of greenhouse gases and volatile substances into the atmosphere, and the sludge generated also creates disposal issues [ 5 ]. As untreated waste water has many sources and tracks with local economic factors [ 6 ], treatment of wastewater should be transformed from a cost to a benefit, as is being considered with other “waste-to-wealth” initiatives [ 7 ], in which biotechnology is factoring in heavily [ 8 ]. Indeed, wastewater has a high energy content and can be used as an energy source [ 1 , 5 ]. This review aims to bring into focus the rapidly developing technology microbial fuel cells (MFCs) as a biotechnology that can naturally convert wastewater from various sources into energy, while remediating the waste stream. Given the range of waste streams produced around the world, the specific focus is placed on recent progress in MFCs using substrates from these diverse sources. Microbial fuel cells (MFCs) use electroactive bacteria for simultaneous power generation and wastewater treatment. The use of MFCs in the treatment of wastewater was first proposed in the late 20 th century, but since then, continuous developments have enhanced the productivity of this technology [ 9 , 10 ]. MFCs solve waste treatment issues by utilizing the chemical energy of waste material to generate electricity [ 11 , 12 , 13 ]. Compared to other waste management technologies such as conventional biological sewage treatment plants, MFCs offer a range of benefits, including operation at room temperature, minimized sludge production, and environmental friendliness while simultaneously producing electricity [ 1 , 14 ]. As shown in Figure 1 , a typical MFC consists of an anode and is separated by an ion-exchange membrane cathode (a cation exchange membrane (CEM)), or anion exchange membrane (AEM), [ 9 , 15 ]. Exoelectrogens form electroactive biofilms (EAB) usually reside on the anode surface, which act as biocatalysts for substrate oxidation to produce electrons and protons by coupling certain redox processes to their metabolic cycles. The electrons are transferred into the anode from the attached EAB via direct and indirect electron transfer mechanisms [ 16 ]. Due to the difference in potential between anode and cathodes, elections travel from the anode to the cathode through an external circuit where they perform work. Thus, the current and cell potential are critical parameters in determining power outputs. The current is also related to the rate of oxidation. Protons generated from substrate oxidation usually travel through a proton exchange membrane into the cathode chamber, where they combine with electrons, protons and terminal electron acceptors, which is usually oxygen through the oxygen reduction reaction, forming water. It has been considered that economically and environmentally sustainable systems will use air cathodes for oxygen reduction [ 17 ]. Figure 1 Schematic and working principle of the typical microbial fuel cell. Figure 1 There are two types of electron transfer mechanisms in MFCs: mediator and mediator-less [ 18 , 19 ]. The mediator-type MFC requires soluble electroactive molecules to shuttle electrons to the anode surface. The use of mediators poses economic and safety constraints on the MFCs. Mediator-less MFCs exploit the ability of metal-reducing bacteria such as Geobacter metallireducens , Aeromonas hydrophila , Shewanella putrefaciens , Rhodoferax , Klebsiella pneumonia to generate electricity without mediators via direct electron transfer through physical contact [ 18 , 19 ]. On the cathode side, the reduction of an electron to water can be catalyzed by metals such as platinum or palladium in a single-chamber air cathode setup. When using well-defined substrates, (e.g., glucose or acetic acid), known anode potentials can be used in conjunction with respective cathode potentials to calculate basic thermodynamic features of the cell, such as open circuit potentials which typically range between 0.3 and 0.8 V [ 20 ]. Recently, useful methods have been demonstrated that allow researchers to measure individual anode and cathode potentials which will be helpful in benchmarking fundamental substrates and their mixtures, as discussed in this work [ 21 ]. Substrates provide food to the bacteria in MFCs and hence play a significant role. Thus, substrates play a vital role in the development of biofilms. The biofilm can be modulated by optimizing the substrate feed, and different substrates have been used in MFCs. There are several strategies to modulate microbiome of MFCs, but a more effective strategy than optimizing feed would be by adopting particular start-up and operating conditions [ 22 ]. Several factors affect the efficacy of the MFCs, such as the type of substrates used (simple or complex), the electrode material, and the microorganisms used to oxidize substrate molecules [ 23 , 24 ]. MFCs can use substrates from diverse wastewater sources ranging from domestic [ 25 , 26 , 27 ] to agricultural [ 1 , 28 , 29 ], industrial [ 14 , 30 ], pharmaceutical [ 31 ], and animal [ 32 , 33 ], among others. It is easier for the EAB to metabolize liquid waste; solid waste is difficult to metabolize due to low hydrolysis, and heterogeneity could add to slow mass transport [ 34 ]. This review focuses on the recent expansion of MFC technology for treatment and electricity generation from an ever-expanding range of wastewater sources. For example, complex substrates are a mixture of different chemical compounds and often contain microbial communities [ 35 ]. These complex substrates primarily include domestic wastewater from municipal [ 36 , 37 ], kitchen, and food waste [ 38 , 39 ] sources. Industrial wastewater typically originates from distilleries [ 25 , 40 ], palm oil factories [ 41 ], dark fermentation systems [ 42 ], landfill leachate [ 43 ], petroleum industry [ 44 ], textile factories [ 45 , 46 ] and the chocolate industry [ 47 ]. Agricultural sources include waste from cellulose [ 48 , 49 ], soybean [ 50 ], molasses [ 51 , 52 ], lignocellulosic biomass [ 53 , 54 ] and tofu [ 3 ]. Animal waste includes waste from fish markets [ 55 ], swine [ 33 , 56 ], cattle [ 57 ], seafood processing [ 58 ], slaughterhouses [ 59 , 60 ], biogas slurry [ 61 ] and poultry. Fruit waste includes the peels of fruits [ 28 ] and juices [ 29 ]. Chemical waste comes from sources such as azo dye [ 45 , 62 ], ethanolamine [ 63 , 64 ], sulfide [ 65 ], nitrate [ 66 ], isopropanol [ 67 ], and pharmaceutical components [ 31 ]. Dairy waste [ 68 ] consists of cheese whey [ 69 ] and yogurt waste [ 70 ]. Also, applications of MFCs for greywater treatment in view of non-potable reuse has been reported [ 71 ] as have been options for groundwater remediation. It should be noted that many of these sources are based on specific processes or input parameters that may vary. This is especially the case for complex substrates with often unpredictable compositions (example kitchen waste and wastewater). All of these materials have been examined as substrate sources for MFCs in recent years and are reviewed in detail in the next section. This includes operating pH, and chemical composition of the substrate molecules, which directly influence the MFC performance. The potential to commercialize MFC technology can be dictated by a series of performance indicators, each of which are significantly influenced by substrate type. Here we consider the most important metrics to be (i) current density (CD), which is a direct indicator of the reaction kinetics; (ii) power density (PD), which is the most important indicator of the potential of the MFC to be used as a power source; (iii) Coulombic efficiency (CE), which describes the efficiency in converting redox molecules to electrons; and (iv) carbon oxygen demand (COD) removal efficiency (ΔCOD), which describes the ability to remove organic materials from waste streams, or roughly speaking, to clean the waste stream. While power production is an enticing direction for MFC, its role in future energy markets is still undefined. On the other hand, the potential to remediate waste streams without external energy requirements (nor their associated CO 2 emissions) stands out as an important capability for near-term MFC implementation. Therefore, with the goal of strategically incorporating MFC technology, COD removal efficiency may be the most important figures of merit in the near-term. As CD is a marker of the rate of COD removal, this should be considered closely as well. The literature shows that COD removal can be efficient down to concentrations as low as 100 mg L −1 , and that current densities become reduced and stop outright at 50 mg L −1 [ 72 , 73 ], but recent studies have demonstrated methods to reduce this limit by 50 times [ 74 ]. We also note that MFCs can be further improved using highly efficient electrodes with nanoactive materials, and ions selective ceramic-based membranes are to be developed and investigated for cost-effective scale-up of MFCs. Further, to reduce the cathode cost and improve cathodic performance, photosynthetic bacteria could be used to make self-sustained cathode. Upon these modifications, MFCs can be used for potential applications in environmental sensors using 3D-printed biofilms for large-scale power generation and deployment in the wastewater treatment process."
} | 3,899 |
39119458 | PMC11307446 | pmc | 213 | {
"abstract": "Direct training of Spiking Neural Networks (SNNs) on neuromorphic hardware has the potential to significantly reduce the energy consumption of artificial neural network training. SNNs trained with Spike Timing-Dependent Plasticity (STDP) benefit from gradient-free and unsupervised local learning, which can be easily implemented on ultra-low-power neuromorphic hardware. However, classification tasks cannot be performed solely with unsupervised STDP. In this paper, we propose Stabilized Supervised STDP (S2-STDP), a supervised STDP learning rule to train the classification layer of an SNN equipped with unsupervised STDP for feature extraction. S2-STDP integrates error-modulated weight updates that align neuron spikes with desired timestamps derived from the average firing time within the layer. Then, we introduce a training architecture called Paired Competing Neurons (PCN) to further enhance the learning capabilities of our classification layer trained with S2-STDP. PCN associates each class with paired neurons and encourages neuron specialization toward target or non-target samples through intra-class competition. We evaluate our methods on image recognition datasets, including MNIST, Fashion-MNIST, and CIFAR-10. Results show that our methods outperform state-of-the-art supervised STDP learning rules, for comparable architectures and numbers of neurons. Further analysis demonstrates that the use of PCN enhances the performance of S2-STDP, regardless of the hyperparameter set and without introducing any additional hyperparameters.",
"conclusion": "6 Conclusion In this paper, we proposed Stabilized Supervised STDP (S2-STDP), a supervised STDP learning rule for training a spiking classification layer with one spike per neuron and temporal decision-making. This layer can be employed to classify features extracted by a convolutional SNN (CSNN) equipped with unsupervised STDP. Our learning rule integrates error-modulated weight updates that align neuron spikes with desired timestamps derived from the average firing time within the layer. Then, to further enhance the learning capabilities of the classification layer trained with S2-STDP, we introduced a training architecture called Paired Competing Neurons (PCN). PCN associates each class with paired neurons connected via lateral inhibition and encourages neuron specialization through intra-class competition. We evaluated S2-STDP and PCN on three image recognition datasets of growing complexity: MNIST, Fashion-MNIST, and CIFAR-10. Experiments showed that our methods outperform state-of-the-art supervised STDP rules when employed to train our spiking classification layer. S2-STDP successfully addresses the issues of SSTDP concerning the limited number of STDP updates per epoch and the saturation of firing timestamps toward the maximum firing time. The PCN architecture enhances the performance of S2-STDP, regardless of the hyperparameter set and without introducing any additional hyperparameters. Our methods also exhibited improved hyperparameter robustness as compared to SSTDP. In the future, we plan to expand S2-STDP to multi-layer architectures, while maintaining the local computation required for on-chip learning. This includes exploring both feedback connections (Zhao et al., 2020 ) and local losses (Mirsadeghi et al., 2021 ).",
"introduction": "1 Introduction Artificial Neural Networks (ANNs) have gathered exponential attention across diverse domains in recent years (Abiodun et al., 2018 ). However, ANN training suffers from high and inefficient energy consumption on modern computers based on the von Neumann architecture (Zou et al., 2021 ). Spiking Neural Networks (SNNs) (Ponulak and Kasinski, 2011 ) implemented on neuromorphic hardware (Schuman et al., 2017 ; Shrestha et al., 2022 ) have emerged as a promising solution to overcome the von Neumann bottleneck (Zou et al., 2021 ) and enable energy-efficient computing. In particular, memristive-based neuromorphic hardware (Jeong et al., 2016 ; Xu et al., 2021 ) is an excellent candidate for ultra-low-power applications, potentially reducing energy consumption by at least one order of magnitude compared to state-of-the-art CMOS-based neuromorphic hardware (Milo et al., 2020 ; Liu et al., 2021a ), and by several orders of magnitude compared to GPUs (Yao et al., 2020 ; Li et al., 2022 ). However, direct training of SNNs on neuromorphic hardware comes with a major constraint: implementing network-level communication is challenging and requires significant circuitry overhead (Zenke and Neftci, 2021 ). Therefore, the involved learning mechanisms should rely on local weight updates, i.e., updates based only on the two neurons that the synapse connects. Training SNNs to achieve state-of-the-art performance is typically accomplished with adaptations of Backpropagation (BP) (Eshraghian et al., 2021 ; Dampfhoffer et al., 2023 ). However, these methods are challenging to implement on neuromorphic hardware as they rely on non-local learning. In addition, they employ gradient approximation to circumvent the non-differentiable nature of the spike generation function, which is suboptimal. Other approaches attempted to make gradient computation local, notably by utilizing feedback connections (Neftci et al., 2017 ; Zenke and Ganguli, 2018 ), or by employing a layer-wise cost function (Ma et al., 2021 ; Mirsadeghi et al., 2021 ). Yet, they do not solve the gradient approximation problem. Furthermore, all of the aforementioned BP-based methods rely solely on supervised learning, which increases the dependence on labeled data. We believe that machine learning algorithms should minimize this dependence on supervision by employing unsupervised feature learning (Bengio et al., 2013 ). Hence, an optimal classification system should include both unsupervised and supervised components, for data representation and classification, respectively. Spike Timing-Dependent Plasticity (STDP) (Caporale and Dan, 2008 ) is a gradient-free, unsupervised and local alternative to BP, inspired by the principal form of plasticity observed in biological synapses (Hebb, 1949 ). STDP solves the previously mentioned limitations of BP and is inherently implemented in memristor circuits (Querlioz et al., 2011 ; Schuman et al., 2017 ), which makes it suitable for on-chip training on memristive-based neuromorphic hardware (Saïghi et al., 2015 ; Khacef et al., 2023 ). Unsupervised feature learning with STDP has been extensively studied in the literature, particularly for image recognition tasks. Convolutional SNNs (CSNNs) trained with STDP have demonstrated the ability to improve data representation by extracting relevant features from images (Tavanaei and Maida, 2017 ; Ferré et al., 2018 ; Kheradpisheh et al., 2018 ; Falez et al., 2019b ; Srinivasan and Roy, 2019 ). However, to perform classification based on the extracted features, these solutions employ external classifiers, such as ANNs or support vector machines, which are incompatible with neuromorphic hardware. To leverage the potential of these CSNNs in neuromorphic hardware and enable end-to-end SNN solutions, spiking classifiers trained with local supervised learning rules must be designed. Ensuring compatibility between classifiers and CSNNs, particularly regarding the type of local rule employed, could significantly mitigate hardware implementation overhead. Although STDP is traditionally formulated for unsupervised learning, it can be adapted for supervised learning by incorporating a third factor, taking the form of an error signal that is used to guide the STDP updates (Frémaux and Gerstner, 2015 ). As a result, STDP enables end-to-end SNNs to perform classification tasks by combining unsupervised STDP for feature extraction and supervised STDP for classification (Shrestha et al., 2017 ; Thiele et al., 2018 ; Lee et al., 2019 ; Mozafari et al., 2019 ). Several supervised adaptations of STDP are reported in the literature (Ponulak and Kasiński, 2010 ; Shrestha et al., 2017 , 2019 ; Lee et al., 2019 ; Tavanaei and Maida, 2019 ; Hao et al., 2020 ; Zhao et al., 2020 ; Zhang et al., 2021 ; Saranirad et al., 2022 ). Yet, all of the aforementioned rules are designed to train SNNs with multiple spikes per neuron, which is undesirable because state-of-the-art CSNNs trained with unsupervised STDP usually employ one spike per neuron. For compatibility with these CSNNs, a spiking classifier trained with supervised STDP should adhere to this single-spike approach. In addition, it has been shown that using one spike per neuron with temporal coding presents several advantages for visual tasks, including fast information transfer, low computational cost, and improved energy efficiency (Rullen and Thorpe, 2001 ; Park et al., 2020 ; Guo et al., 2021 ). The literature exploring supervised adaptations of STDP for training SNNs with only one spike per neuron is limited. Reward-modulated STDP (R-STDP) (Mozafari et al., 2019 ) is a learning rule based on Winner-Takes-All (WTA) competition (Ferré et al., 2018 ) that modulates the polarity of the STDP update to apply a reward or a punishment. R-STDP has gained popularity notably for its simplicity, but it results in inaccurate weight updates as only the polarity of STDP is adjusted. Recently, Supervised STDP (SSTDP) (Liu et al., 2021b ) proposes a method to modulate, in the output layer, both the polarity and intensity of STDP with temporal errors, resulting in more accurate weight updates. When combined with the non-local optimization process of BP, SSTDP enables state-of-the-art performance in deep SNNs on various image recognition datasets. However, it has not yet been investigated in settings based on local learning, combining unsupervised STDP for feature extraction and SSTDP for classification. In addition, we claim that SSTDP faces two issues that may limit its performance. First, SSTDP training results in a limited number of STDP updates per epoch, which can lead to premature training convergence. Second, SSTDP training causes the saturation of firing timestamps toward the maximum firing time, which can limit the ability of the SNN to separate classes. In this paper, we focus on the supervised STDP training of a spiking classification layer with one spike per neuron and temporal decision-making. This classification layer is the output layer of an SNN equipped with unsupervised STDP for feature extraction, as illustrated in Figure 1 . The main contributions of this paper include the following: In a preliminary study, we analyze the behavior of SSTDP when used to train the classification layer of an SNN equipped with unsupervised STDP. We demonstrate that the rule encounters two issues that may limit its performance: the limited number of STDP updates per epoch and the saturation of firing timestamps toward the maximum firing time. To address the issues of SSTDP, we propose Stabilized Supervised STDP (S2-STDP), a supervised STDP learning rule that teaches neurons to align their spikes with dynamically computed desired timestamps derived from the average firing time within the layer. To further enhance the learning capabilities of our classification layer trained with S2-STDP, we introduce a training architecture called Paired Competing Neurons (PCN). This method associates each class with paired neurons and encourages neuron specialization toward target or non-target samples through intra-class competition. We evaluate the performance of S2-STDP and PCN on three image recognition datasets of growing complexity: MNIST, Fashion-MNIST, and CIFAR-10. Figure 1 Architecture of the SNN employed in this paper for image recognition tasks. First, the image is preprocessed and each pixel is encoded into a single floating-point timestamp using the latency coding scheme. Then, a Convolutional SNN (CSNN) trained with unsupervised STDP is used to extract relevant features from the image. The resulting feature maps are compressed through a max-pooling layer to reduce their size and provide invariance to translation on the input. Lastly, they are flattened and fed to a fully-connected SNN trained with a supervised adaptation of STDP for classification. Each output neuron is associated with a class and the first one to fire predicts the label. Training is done in a layer-wise fashion. This classification pipeline organized into three blocks provides a flexible framework for SNNs combining feature extraction and classification. In this paper, we focus on the classification layer block (C), which may be integrated after other encoding or feature extraction blocks based on latency coding. The remainder of this paper is organized as follows. In Section 2, we provide the necessary background information about the SNN employed in this study. In Section 3, we demonstrate experimentally the aforementioned issues of SSTDP, which we address with our contributions. In Section 4, we describe our spiking classification layer and our proposed training methods. In Section 5, we cover our results on image recognition datasets and provide an in-depth investigation of the key characteristics of our methods. In Section 6, we conclude the paper. The source code is publicly available at: https://gitlab.univ-lille.fr/fox/snn-pcn ."
} | 3,324 |
29497169 | PMC5832778 | pmc | 215 | {
"abstract": "Superhydrophobic surfaces have great potential for application in self-cleaning and oil/water separation. However, the large-scale practical applications of superhydrophobic coating surfaces are impeded by many factors, such as complicated fabrication processes, the use of fluorinated reagents and noxious organic solvents and poor mechanical stability. Herein, we describe the successful preparation of a fluorine-free multifunctional coating without noxious organic solvents that was brushed, dipped or sprayed onto glass slides and stainless-steel meshes as substrates. The obtained multifunctional superhydrophobic and superoleophilic surfaces (MSHOs) demonstrated self-cleaning abilities even when contaminated with or immersed in oil. The superhydrophobic surfaces were robust and maintained their water repellency after being scratched with a knife or abraded with sandpaper for 50 cycles. In addition, stainless-steel meshes sprayed with the coating quickly separated various oil/water mixtures with a high separation efficiency (>93%). Furthermore, the coated mesh maintained a high separation efficiency above 95% over 20 cycles of separation. This simple and effective strategy will inspire the large-scale fabrication of multifunctional surfaces for practical applications in self-cleaning and oil/water separation.",
"conclusion": "Conclusions In summary, MSHOs were easily fabricated through one-step coating methods of brush-coating, dip-coating and spray-coating. The obtained surfaces maintained their unique wettability after being abraded with sandpaper or scratched with a knife, thus exhibiting excellent robustness due to the good adhesion and mechanical properties of the added ER. The multifunctional surfaces showed excellent self-cleaning performance in both air and oil and were successfully used to quickly separate oil/water mixtures with a high separation efficiency. After 20 separation cycles, the separation efficiency remained above 95%, indicating the excellent reusability of the coated mesh. The facile fabrication strategy of the MSHOs with excellent robustness is believed to promote the application of self-cleaning surfaces in harsh and oily environments and also be useful for continuous and efficient oil/water separations.",
"introduction": "Introduction Over the last few decades, superhydrophobic surfaces have attracted considerable attention due to their diverse practical applications, such as self-cleaning 1 , oil/water separation 2 , corrosion resistance 3 , anti-icing 4 , anti-fogging 5 , anti-fouling 6 , anti-bacterial 7 , anti-reflection 8 , and drag reduction 9 . Many functional biological surfaces in nature, including lotus leaves 10 , rose petals 11 , butterfly wings 12 , and water striders 13 , among others 14 , possess unique wettability properties. Studies of these biological surfaces have indicated that the surface roughness originating from unique micro/nanostructures and the surface chemistry are two major factors that affect the surface wettability 15 – 18 . To date, superhydrophobic surfaces have been developed by a variety of methods, such as sol-gel methods 19 , lithographic processes 20 , casting 21 , electrospinning 22 , chemical vapour deposition (CVD) 23 , chemical etching 24 , dip-coating 25 , and templating 26 . Although these approaches allow for the manufacture of superhydrophobic surfaces, many limitations stemming from the complicated and time-consuming fabrication processes, expensive equipment, and restrictions of substrate materials prevent the large-scale application of most of these methods 27 , 28 . Thus, low-cost, facile methods for fabricating superhydrophobic surfaces for widespread applications are urgently needed. Brush-coating, dip-coating and spray-coating are typical methods that can meet the abovementioned requirements and facilitate the practical application of superhydrophobic surfaces. However, the mechanical weakness of the micro/nanostructures on the coating surface limits the applications of superhydrophobic coatings 29 – 31 . To overcome this limitation, many studies have employed an adhesive layer to bond the coating to a substrate and enhance the robustness of the surface. Two strategies are available for producing an adhesive layer. One of the strategies is useful but not perfect because the adhesive and coating need to be separately applied to the substrate 32 , 33 . This two-step coating strategy is time-consuming, and the thickness of the resulting coating is restricted, which limits the large-scale application of the resulting superhydrophobic surface. Another important strategy is to thoroughly mix the adhesive with the coating and then deposit the mixture on the substrate through a one-step approach 34 . This method can be used to effectively fabricate robust superhydrophobic surfaces and has great potential for large-scale application. Fluorinated reagents with low surface free energies are often applied when constructing superhydrophobic surfaces to improve the hydrophobicity of the material surface 35 – 38 . However, fluoro-containing compounds are expensive and harmful to the environment and human health 39 – 41 . Therefore, fluorine-free compounds that are inexpensive and environmentally friendly, such as alkyl silanes and long carbon chain organics, have been developed for the fabrication of superhydrophobic surfaces 42 – 44 . Furthermore, noxious organic solvents are often used to prepare coating suspensions during the fabrication of superhydrophobic surfaces, which is environmentally hazardous. Thus, environmentally friendly solvents should be used for the large-scale fabrication of superhydrophobic surfaces. Superhydrophobic surfaces are easily contaminated with oil, causing the loss of their self-cleaning abilities 45 . Manufacturing superamphiphobic surfaces that can repel both water and oil is a promising method for overcoming this disadvantage 46 . However, the applications of superamphiphobic surfaces are limited to specific fields. For example, in the field of machinery, certain parts such as bearings, chains and gears need to be lubricated with oil to reduce friction 32 . In anti-corrosion applications, oil-lubricated metal surfaces can resist rust for long periods by providing a barrier against air and moisture. However, because superamphiphobic surfaces repel oil, they cannot be lubricated by oil. In addition, fluorinated reagents are needed for the fabrication of almost all superamphiphobic surfaces. In recent years, only a few reports have demonstrated the preparation of robust superhydrophobic and superoleophilic surfaces that possess self-cleaning capabilities in both air and oil 32 , 33 . Furthermore, most of these robust self-cleaning surfaces that function in both air and oil are fabricated through time-consuming two-step methods. In addition to self-cleaning, another important application of superhydrophobic and superoleophilic surfaces is the separation of oil and water 47 – 49 . The separation capability of the surface is typically evaluated in terms of the separation efficiency and reusability 50 . Although various strategies for oil/water separation have been developed, several shortcomings still need to be overcome, such as fouling of the surface by oil and poor recyclability 51 – 53 . These shortcomings lead to poor performance in continuous oil/water separation for practical applications. Herein, we report a fluorine-free coating for fabricating robust superhydrophobic and superoleophilic surfaces that possess self-cleaning abilities in both air and oil and can be used in highly effective oil/water separations. This strategy used epoxy resin (ER) as the adhesive, providing the resulting surfaces with microscale roughness and robustness. Silica nanoparticles and dodecyltrimethoxysilane (DTMS) were employed to enhance the nanoscale roughness and reduce the surface free energy, respectively. The as-obtained paint-like suspensions were painted on glass slides and stainless-steel meshes as substrates through brush-coating, dip-coating and spray-coating methods (as illustrated in Fig. 1 ). The morphologies, chemical compositions, and wettability of the coated surfaces were characterized. Self-cleaning tests in both air and oil demonstrated that the self-cleaning performance of the coated surfaces was maintained, even in oily environments. The coated surfaces were confirmed to be robust through abrasion and knife-scratch tests. Finally, the coated stainless-steel meshes were used to separate various oil/water mixtures, and the obtained separation efficiency was greater than 93%. The coated meshes maintained a high separation efficiency (>95%) with a chloroform/water mixture after 20 repeated separation cycles. Thus, this easily-manufactured, eco-friendly and multifunctional surface with superhydrophobic and superoleophilic properties has great potential for practical application in diverse fields. Figure 1 Schematic illustration for the fabrication of MSHOs by one-step deposition methods.",
"discussion": "Results and Discussion Surface morphologies The surface morphologies of the brush-coated glass slides, pristine stainless-steel meshes and spray-coated meshes were characterized by field emission scanning electron microscopy (FESEM). As shown in Fig. 2a , the coated glass slide exhibited microporous structures, which formed from the rapid evaporation of anhydrous alcohol. As shown in the enlarged SEM image, the coating possessed a rough surface with hierarchical micro/nanoscale structures (Fig. 2a ). Meanwhile, numerous nanoscale papillae structures were observed. These hierarchical micro/nanoscale rough structures (HMNRs) played an important role in establishing the superhydrophobicity. Figure 2b depicts a typical image of the untreated stainless-steel mesh, which exhibited a smooth surface. After coating the mesh as described, the steel wires were completely covered by the coating material (Fig. 2c ). It should be noted that the presence of micropores could ensure the free passage of oil through the coated mesh. The surface morphology of the coated mesh was different from that of the coated glass slide in low-magnification owing to the difference of coating methods and substrates. Nevertheless, the magnified FESEM image of the coated mesh also showed the presence of HMNRs, which provided an important foundation for establishing the superhydrophobic properties. In addition, the average surface roughness of coatings on glass slides and meshes was measured using a laser scanning confocal microscopy (LSCM). As shown in Supplementary Fig. S1 , the average mean square roughness of the coated mesh (Ra = 3.93 μm) was larger than that of the coated glass slide (Ra = 6.78 μm). This is due to the uneven surface of the original meshes and the difference of the coating methods. Figure 2 Structural characterization of the coating and stainless-steel mesh. Low-magnification and enlarged FESEM images of ( a ) the brush-coated glass slide, ( b ) the original mesh, and ( c ) the spray-coated mesh. Chemical composition The differences in the chemical compositions of the unmodified hydrophobic coatings and the superhydrophobic coatings modified by DTMS were analysed by energy dispersive spectrometer (EDS) and fourier transform infrared spectrum (FTIR). Figure 3a,b show the EDS spectra of the coatings before and after modification with DTMS, respectively. The atomic ratio of Si increased from 10.09% (Fig. 3a ) to 11.58% (Fig. 3b ) with the addition of DTMS. As displayed in Fig. 3c , the characteristic peaks of the coating before modification were observed at 916 cm −1 for the -CH(O)CH- groups of ER, 1095 cm −1 for the Si-O-Si groups of SiO 2 and 1646 cm −1 for the C=O groups of polyamide (PA) 54 . In addition, after modification, the Si-O-Si (1095 cm −1 ), -CH2- (2850 cm −1 ) and -CH3 (2920 cm −1 ) stretching vibrations became stronger than those in the unmodified coating, thereby demonstrating that the coating was successfully modified with DTMS 42 , 55 . Figure 3 Chemical compositions of the coatings. EDS spectra of the coatings before ( a ) and after ( b ) modification with DTMS, respectively. ( c ) FTIR spectra of the coatings before and after modification with DTMS. Surface wettability The wettability of the coated glass slides and coated meshes was characterized by the water contact angle (WCA), oil contact angle (OCA) and water sliding angle (WSA). The effects of the surface micro/nanoscale structures and the chemical compositions on the wettability were also investigated. As shown in Fig. 4a , the original glass slide and mesh exhibited hydrophilic and hydrophobic properties, respectively. After introduction of the coating without DTMS, the hydrophilic glass slide became hydrophobic, with a WCA of 133.5° ± 2°. Meanwhile, the coated mesh became even more hydrophobic (WCA = 135.5° ± 3.2°) than the coated glass slide. Furthermore, the coated glass slides (WCA = 154° ± 1.7°) and meshes (WCA = 153.3° ± 1.4°) modified with DTMS showed superhydrophobic properties and had WSAs below 5°. Thus, the HMNRs of the coating surfaces were not sufficient for imparting superhydrophobic properties on the coated surfaces. However, the low surface free energy of the coating that resulted from modification with DTMS established superhydrophobic properties. In other words, both appropriate HMNRs and chemical compositions were essential for the fabrication of superhydrophobic surfaces. Furthermore, the coated surfaces were not only superhydrophobic but also showed superoleophilic properties, with an OCA of 0° (Fig. 4b,c ). Figure 4 ( a ) The water wettability of A original glass slides and stainless-steel meshes, B coated glass slides and meshes using the coating without DTMS, and C coated glass slides and meshes modified by DTMS. ( b ) and ( c ) are WCA and OCA images of coated glass slides, respectively. ( d ) and ( e ) are WCA and OCA images of coated meshes, respectively. To further confirm the superhydrophobic and superoleophilic behaviours of the coated surfaces, the bouncing or spreading of water and oil droplets were observed with a high-speed camera. Using a microinjector, water droplets with a volume of approximately 6.7 μL were dropped from a height of 35 mm. The water droplets then impacted the coated glass slide or mesh at a speed of 0.83 m/s. When the water droplet was dropped onto a coated glass slide, it bounced and completely left the coated surface within 12.5 ms (Fig. 5a ), and the coated glass slide was not wetted by the water droplet dyed blue with methylene blue. Similarly, the water droplet dropped onto a coated stainless-steel mesh bounced and completely left the coated surface within 11.83 ms, as shown in Fig. 5b . The coated mesh was also not wetted by the coloured water droplet. The observed water-repellent behaviours of the coated glass slide and mesh surfaces further demonstrated the superhydrophobicity of the coated surfaces. In addition, oil dropping tests were carried out with the same parameters as those for the water dropping tests described above. The oil droplets had a volume of approximately 6.7 μL and impacted the coated glass slide or mesh at approximately 0.83 m/s. As shown in Fig. 5c , the oil droplet completely spread out on the coated glass slide within 2.53 s. After impacting the coated mesh, the oil droplet spread out and permeated into the mesh within 60 ms. The rapid spread of the oil droplets on the coated surfaces further illustrated the superoleophilic properties of the coatings. These superhydrophobic and superoleophilic behaviours of the coating surfaces suggested that the coating could be useful for self-cleaning and oil/water separation applications. Figure 5 Water bouncing and oil spreading processes on the coated glass slides and stainless-steel meshes. Bouncing dynamics of a water droplet impacting ( a ) a coated glass slide and ( b ) a coated mesh. ( c ) Spreading processes of an oil droplet impacting (a) a coated glass slide and ( d ) a coated mesh. Self-cleaning in air or oil A series of self-cleaning tests were carried out to demonstrate the self-cleaning properties of the coated surfaces. In these experiments, purple sand was utilized as dirt to aid visualization. For the self-cleaning test in air, the dirt was obviously not removed from the uncoated glass slide, which was wetted and contaminated by the coloured water (Fig. 6a 1 –a 3 ). This result indicates that the uncoated glass slide did not have self-cleaning abilities. In contrast, no dirt or dyed water remained on the coated glass slide after the removal of dirt by rolling water droplets, illustrating the excellent water-repelling and self-cleaning properties (Fig. 6b 1 –b 3 and Supplementary Video S1 ). For the coated glass slides, air was trapped in the HMNRs instead of water, forming an air layer. This trapped air substantially decreased the contact area between the water droplet and the solid surface. Therefore, water droplets in the Cassie state easily rolled off of the superhydrophobic coated surfaces due to the low water adhesion properties of the surfaces. Figure 6 Self-cleaning tests in air or oil. Self-cleaning properties of ( a 1 – a 3 ) the original glass slide and ( b 1 – b 3 ) the coated glass slide in air. Self-cleaning properties of ( c 1 – c 3 ) the original glass slide and ( d 1 – d 3 ) the coated glass slide in oil. When self-cleaning tests were carried out in oil, the glass slides were first completely contaminated with oil and then partially immersed in oil. The tilt angle was approximately 12°. Most self-cleaning tests of coatings with unique wettability properties were performed in air, not in oil 32 . The changes in the physical and chemical properties of the coated surface after contamination with oil resulted in the weakening or even loss of the self-cleaning abilities. Herein, the self-cleaning properties of the superhydrophobic and superoleophilic coating are demonstrated. Self-cleaning tests were carried out on uncoated and coated glass slides contaminated with or immersed in oil (hexadecane). The pristine glass slide did not show self-cleaning abilities under these conditions (Fig. 6c 1 –c 3 ). However, for the coated glass slide contaminated with oil, the oil was locked in the HMNRs of the coating, forming a thin oil film (Fig. 7c ). Because the surface tension of oil is lower than that of water, the relatively stable oil film could not be replaced by water. Therefore, a slippery liquid-infused porous surface (SLIPS) that repelled water droplets was formed, and the water droplet could easily slide across the coated surface as a result (Fig. 7a and Supplementary Video S2 ). Furthermore, the water droplets removed the dirt without wetting the coated glass slide (Fig. 6d 1 –d 3 and Supplementary Video S3 ). When the coated glass slides were immersed in oil, the oil penetrated into the HMNRs of the coating. Due to the HMNRs, the low surface free energy of the coated surface and the surrounding oil, the dyed water droplet was spherical and easily rolled off of the tilted surface (Fig. 7b and Supplementary Video S2 ). No water adhered to the oil-contaminated coated surface that was partially immersed in oil (Fig. 6d 1 –d 3 and Supplementary Video S3 ). These results show that the coated surfaces still retained their self-cleaning properties when contaminated with or even immersed in oil. The self-cleaning tests of the coated glass slides were repeated in hexane and silicon oil to further demonstrate the self-cleaning properties. As shown in Supplementary Fig. S2 , the water droplets removed the purple sand without staining the coated surfaces, illustrating the excellent self-cleaning properties of the coating. The experimental results show that the multifunctional superhydrophobic and superoleophilic surfaces (MSHOs) with excellent self-cleaning properties in both air and oil possess broad potential for application in complex practical environments, especially oily environments. Figure 7 Movement of water droplets on a coated glass slide contaminated with or immersed in oil. ( a ) The water droplet slid on the coated glass slide contaminated by oil and ( b ) rolled on the coated glass slide immersed in oil. ( c ) Schematic of a water droplet moving on an oil-contaminated coated glass slide that is partially immersed in oil. Robustness of coatings Micro/nanoscale structures are essential for establishing superhydrophobic properties in coatings. However, these structures are mechanically weak and easily worn, which limits the widespread application of superhydrophobic coatings. The use of an adhesive to bond the coating to the substrate can enhance the robustness of the coating. However, if the adhesive and coating must be individually painted on the substrate, the coating process becomes relatively complicated, which will affect the efficiency of the large-scale production. Here, we added ER and PA as adhesives to the paint-like suspensions to improve the poor robustness of the resulting superhydrophobic coatings and simplify the coating processes. Then, sandpaper abrasion and cross-cut scratching tests were carried out to examine the robustness of the obtained coatings. A schematic illustration of the sandpaper abrasion test is shown in Fig. 8a . The dip-coated surface maintained its water repellency after abrasion with sandpaper (Fig. 8b and Supplementary Video S4 ). As shown in Fig. 8d , the WCA varied between 150° and 156° in the 50 abrasion cycles, indicating that the superhydrophobicity of the coatings was not easily destroyed under mechanical abrasion. Figure 8f shows only a small part of the coating surface becomes smoother (the red dashed frames) and most of the coating surface possesses micro/nanoscale roughness structures after 50 abrasion cycles. This phenomenon could be explained that only the top layer of the micropores on the coating surface was removed due to the robustness of the coating and the micro/nanoscale roughness structures in the micropores were exposed (Fig. 8e and Fig. 8f ). In addition, some micro/nanoscale roughness structures could be produced by sandpaper wearing. The cross-cut scratching test showed that the water droplets rolled off from the coating surfaces which were scratched by a knife without wetting the surfaces (Fig. 8c and Supplementary Video S5 ). Although the scalpel knife truly damaged the roughness structures along the scratch due to its sharpness and hardness (Fig. 8g and Supplementary Video S5 ), it did not significantly influence the excellent water repellency of the coating in function, which was extremely important for practical applications. The as-prepared coating surfaces were concluded to be robust based on the results of the sandpaper abrasion and cross-cut scratching tests. The good adhesion and mechanical properties of the cured ER might be responsible for the good robustness. Since the coating could be painted on the substrate in a one-step coating process and the resulting superhydrophobic surface was robust, the coating might be widely used in the large-scale manufacture of superhydrophobic surfaces. Figure 8 ( a ) Schematic illustration of the sandpaper abrasion test. Water-repellent behaviour of the dip-coated surface after ( b ) abrasion with sandpaper and ( c ) scratching with a knife. ( d ) WCAs of the dip-coated surface after every fifth abrasion cycle. SEM images of the dip-coated surface before ( a ) and after ( b ) abrasion with sandpaper and ( c ) scratching with a knife. Separation of oil/water mixtures Superhydrophobic and superoleophilic porous materials have potential for use in the separation of oil/water mixtures. Here, oil/water separation tests were performed to investigate the potential of the coated stainless-steel mesh for application in oil/water separation. Figure 9a shows the separation procedure for a mixture containing 20 mL of heavy oil (chloroform) and 30 mL of water. The chloroform quickly permeated through the coated mesh under only gravity and was collected in the beaker below, whereas the water was retained in the glass tube above the coated mesh (Supplementary Video S6 ). These experimental results reveal the ability of the coated mesh to efficiently separate heavy oil/water mixtures. During the separation of light oil/water mixtures, if the mixture was quickly poured into the same separation device as that shown in Fig. 9a , some oil eventually floated on the collected water, resulting in a low separation efficiency. This phenomenon occurred because the permeate flux of the coated mesh was within a certain range. Hence, a mixture containing 20 mL of light oil (hexane) and 30 mL of water was separated in the experimental setup shown in Fig. 9b . The hexane rapidly passed through the coated mesh and was collected in the beaker underneath, while the water was completely retained above the coated mesh (Supplementary Video S7 ). Figure 9 Separation of a ( a ) a heavy oil/water mixture and ( b ) a light oil/water mixture by using the coated mesh. The water was dyed blue with methylene blue, and the oil was coloured red with sudan III. The separation efficiency is often used to evaluate the ability to perform oil/water separation and is calculated according to the equation η = (m 1 /m 0 ) × 100%, where m 0 and m 1 are the mass of oil before and after separation, respectively 50 . As shown in Fig. 10a , the separation efficiency of the coated mesh with a variety of oil/water mixtures was above 93%. The most efficient separation by the coated mesh was accomplished with the chloroform/water mixture, which gave a separation efficiency above 97%. A small gap remained between the practical separation efficiency and 100%, which was mainly attributed to the volatilization of a small amount of oil, while another portion of the oil was absorbed by the coated mesh or adhered to the separation device. Due to the combined effects of the volatility, density and viscosity of the oils, the separation efficiency of the coated meshes with different oils varied slightly. Moreover, the reusability of the coated mesh was evaluated because this property is important in practical applications. As shown in Fig. 10b , the coated mesh maintained a separation efficiency above 95% with a chloroform/water mixture after 20 separation cycles, which indicated that the mesh had excellent reusability. Therefore the as-prepared coated mesh was proven to be useful for continuous and efficient oil/water separation. Figure 10 ( a ) Separation efficiencies of the coated mesh with various oil/water mixtures. ( b ) Separation efficiency of the coated mesh with a chloroform/water mixture during repeated experiments."
} | 6,698 |
28482132 | null | s2 | 216 | {
"abstract": "The factors affecting plant uptake of heavy metals from metalliferous soils are deeply important to the remediation of polluted areas. Arbuscular mycorrhizal fungi (AMF), soil-dwelling fungi that engage in an intimate exchange of nutrients with plant roots, are thought to be involved in plant metal uptake as well. Here, we used a novel field-based approach to investigate the effects of AMF on plant metal uptake from soils in Palmerton, Pennsylvania, USA contaminated with heavy metals from a nearby zinc smelter. Previous studies often focus on one or two plant species or metals, tend to use highly artificial growing conditions and metal applications, and rarely consider metals' effects on plants and AMF together. In contrast, we examined both direct and AMF-mediated effects of soil concentrations on plant concentrations of 8-13 metals in five wild plant species sampled across a field site with continuous variation in Zn, Pb, Cd, and Cu contamination. Plant and soil metal concentration profiles were closely matched despite high variability in soil metal concentrations even at small spatial scales. However, we observed few effects of soil metals on AMF colonization, and no effects of AMF colonization on plant metal uptake. Manipulating soil chemistry or plant community composition directly may control landscape-level plant metal uptake more effectively than altering AMF communities. Plant species identities may serve as highly local indicators of soil chemical characteristics."
} | 374 |
34631117 | PMC8479340 | pmc | 217 | {
"abstract": "Flocking is a fascinating phenomenon observed across a wide range of living organisms. We investigate, based on a simple self-propelled particle model, how the emergence of ordered motion in a collectively moving group is influenced by the local rules of interactions among the individuals, namely, metric versus topological interactions as debated in the current literature. In the case of the metric ruling, the individuals interact with the neighbours within a certain metric distance; by contrast, in the topological ruling, interaction is confined within a number of fixed nearest neighbours. Here, we explore how the range of interaction versus the number of fixed interacting neighbours affects the dynamics of flocking in an unbounded space, as observed in natural scenarios. Our study reveals the existence of a certain threshold value of the interaction radius in the case of metric ruling and a threshold number of interacting neighbours for the topological ruling to reach an ordered state. Interestingly, our analysis shows that topological interaction is more effective in bringing the order in the group, as observed in field studies. We further compare how the nature of the interactions affects the dynamics for various sizes and speeds of the flock.",
"introduction": "1 . Introduction Cohesive group formation is one of the most eye-catching displays in nature. It is observed among various species, such as, flock of birds [ 1 , 2 ], school of fishes [ 3 ], swarm of prey [ 4 , 5 ], colony of bacteria [ 6 ], aggregation of cells [ 7 ] and pedestrian crowd [ 8 ]. The instances of collective behaviours have also been demonstrated in non-living systems including multi-agent robots [ 9 ], vibrated discs [ 10 ], artificial microswimmers [ 11 ] etc. Until now, a great effort has been devoted to understanding the underlying universal mechanisms of such spontaneous emergence of ordered motion irrespective of the nature of the constituent entities. As a result, several theoretical approaches have been developed to study the collective dynamics [ 1 , 2 , 12 – 19 ]. Among these, the self-propelled particle-based model, since the seminal work of Vicsek [ 20 ], remains one of the most favourite choices as it could replicate both system-specific and universal features of the collective behaviours observed across a wide range of species. Moreover, because of its simplicity, one could easily test the model against the experimental observations. Several studies have revealed that collective motion can emerge from simple local rules of interaction among the individuals without a leader or any kind of a central control [ 21 ]. The most commonly discussed interactions include short-range repulsions and long-range attractions among the individuals, or the alignment of velocities along with the nearest neighbours [ 1 , 22 ]. Many intriguing questions, thus arise, e.g. how the individuals keep track of its neighbour in a large extended group, how far its range of interaction extends, or how many interacting neighbours do they require to establish such a coordinated motion of the whole group to flock together. Following Vicsek’s work, most of the self-propelled particle models have incorporated metric-based interaction rules according to which each individual interacts with its surrounding neighbours up to a certain metric distance. Hemelrijk and Hildenbrandt have demonstrated that metric-based interactions incorporating cohesion, alignment and separation rules can explain the shapes and patterns of fish schooling [ 23 ]. Using a simple model with metric rules, Couzin et al. have shown how efficient information transfer and decision-making can occur in animal groups [ 24 ]. Another self-propelled particle model introduced by Bhattacharya and Vicsek indicates the mechanisms of the synchronized landing of a flock of birds performing foraging flights [ 25 ]. There are also other metric-based interaction models that have given many insights into the orientational order, cluster formations, synchronization and spatial sorting in collective groups [ 26 – 28 ]. However, in contrast to the metric-based interaction rules, it has recently been suggested in an experimental finding that the birds in a flock tend to interact with a fixed number of nearest neighbours [ 29 ]; this has been termed as topological interaction in the subsequent literature. Ballerini et al. have analysed the trajectories of flocks of a few thousand starlings and have shown that each bird interacts with a fixed number of neighbours (on an average six to seven neighbours) irrespective of their metric distance [ 29 ]. Camperi et al. based on a self-propelled particles model have shown that the topological models are more stable than the metric ones, and maximal stability is attained when topological neighbours are distributed evenly around each individual, i.e. the neighbours are chosen from a spatially balanced neighbourhood [ 30 ]. Further, using network and graph-theoretic approaches coupled with a dynamical model, Shang and Bouffanais have studied the consensus reaching process with topologically interacting self-propelled particles. They have shown regardless of the group size, a value of close to 10 neighbours speeds up the rate of convergence to the consensus to an optimal level where all particles interact with the entire group [ 31 ]. Several studies on metric interactions and quite a few on topological interactions have been carried out; however, what kind of inter-agents interaction governs the macroscopic collective order among diverse species is not yet well-understood [ 32 , 33 ]. On the one hand, animals can estimate absolute distance by various methods, including retinal image size and different motional cues [ 34 ]. Therefore, estimating the metric distance among the surrounding neighbours could be a natural interaction. On the other hand, the sensory and cognitive limitation of an individual indicates that instead of interacting with all members, topological interaction with a few neighbours could play an important role in establishing collective order in the group. In this paper, based on a simple self-propelled particle model, we investigate how the metric versus topological interactions affect the emergence of ordered motion in a flock. We study the dynamics of flocking by both varying the range of interaction radius and the number of interacting topological neighbours among the individuals. Our theory predicts the existence of a certain threshold interaction radius for the metric ruling and a threshold number of interacting neighbours for the topological ruling for the whole group to reach an ordered state. We further explore how the nature of interactions influences the overall dynamics for different speeds and group sizes of the flock. It shows that the lower speed is more beneficial for the group to flock together compared with the higher speed. Furthermore, in the case of metric interaction, the threshold value decreases with an increase in flock sizes irrespective of the flock speeds. On the contrary, in topological interaction, increasing the group size increases the threshold value for flocks moving with the higher speed; however, it remains unaffected for flocks with a lower speed. Overall, the topological interaction turns out to be beneficial and effective in bringing order in the flock.",
"discussion": "4 . Discussion We have investigated, based on a simple self-propelled particle model, how the local rules of metric and topological interactions affect the dynamics of flocking. We have studied the emergence of order by varying the interaction distance and also the number of interacting nearest neighbours for various flock sizes and speeds. It is observed that in the case of the metric ruling, a certain threshold value of interaction radius is required to reach the ordered state. Similarly, for the topological ruling, a threshold value of the number of interacting neighbours is needed for flocking. Our study shows that the threshold value for a given flock size gets lowered for the group moving with a lower speed for both the metric and the topological interactions. Thus, the lower speed turns out to be beneficial since the individuals are required to interact comparatively with a smaller number of neighbours and at a shorter distance for the group to flock together. Furthermore, in metric interaction, increasing the group size decreases the threshold interaction radius required for flocking irrespective of the flock speeds. On the contrary, in the topological ruling, the variation in the group size does not affect the flocking dynamics in the case of lower speed. Our study shows in the case of flocks moving at a lower speed, the optimal number of topologically interacting neighbours is approximately 7–10 to achieve the ordered motion of the whole group, which has also been observed in other studies [ 29 , 31 ]; moreover, the number does not depend on the flock size as found in [ 31 , 36 ]. However, our analysis indicates that for groups moving at high speed, the optimal number of interacting neighbours require to be much higher to reach an ordered state, and also, the threshold number of interacting neighbours increases with the increase in the flock size, as shown in table S1 in the electronic supplementary material. Overall, the topological interaction turns out to be more efficient as compared with the metric interaction for the whole group to reach the order state. Moreover, the difference between the metric and the topological ruling becomes more pronounced with a decrease in the flock size. Furthermore, it is worth noting that apart from these two broad groups of local interactions—metric and topological rules—there are also studies considering a hybrid model of metric–topological interactions to understand the specific behaviours associated with various swarms [ 37 , 38 ]. Moreover, there are recent approaches by explicitly incorporating the sensory information such as visual sensing of the individuals or inferring the fine-scale social interaction rules among the individuals that help to get an insight into the coordinated motion of collectively moving groups [ 39 – 41 ]. However, it is quite challenging to construct the visual field and analyse the trajectories of every individual of large swarms in natural fields. Besides, the ways of communication in the groups may vary with changing environmental conditions. In such cases, these two broad groups of local interactions, namely metric and topological rules that indirectly incorporate the sensory cues and cognitive capabilities of individuals in a group, provide a great deal of information about the collective dynamics, as shown by our study. Our minimal model set-up could further be extended by incorporating other interactions as observed in nature, the effect of strong noise, or the presence of restrictive boundary or a predator attack and explore how it could influence the outcomes."
} | 2,745 |
39746986 | PMC11696043 | pmc | 218 | {
"abstract": "Neuromorphic computing holds immense promise for developing highly efficient computational approaches. Memristor-based artificial neurons, known for due to their straightforward structure, high energy efficiency, and superior scalability, which enable them to successfully mimic biological neurons with electrical devices. However, the reliability of memristors has always been a major obstacle in neuromorphic computing. Here, we propose an ultra-robust and efficient neuron of negative differential resistance (NDR) memristor based on AlAs/In 0.8 Ga 0.2 As/AlAs quantum well (QW) structure, which has super stable performance such as low variation (0.264%), high temperature resistance (400 °C) and high endurance. The NDR devices can cycle more than 10 11 switching cycles at room temperature and more than 10 9 switching cycles even at a high temperature of 400 °C, which means that the device can operate for more than 310 years at 10 Hz update frequency. Furthermore, the NDR memristor implements the integration feature of the neuronal membrane and avoids using external capacitors, and successfully apply it to the self-designed super reduced neuron circuit. Moreover, we have successfully constructed Fitz Hugh Nagumo (FN) neuron circuit, reduced hardware costs of FN neuron circuit and enabling diverse neuron dynamics and nine neuron functions. Meanwhile, based on the high temperature stability of the device, a voltage-temperature fused multimodal impulse neural network was constructed to achieve 91.74% accuracy in classifying digital images with different temperature labels. This work offers a novel approach to build FN neuron circuits using NDR memristors, and provides a more competitive method to build a highly reliable neuromorphic hardware system.",
"introduction": "Introduction The exponential growth in data volume and computational demand, coupled with the performance limitations of transistor-based computing systems, has spurred interest in alternative computing paradigms 1 . Researchers have proposed neuromorphic computing hardware that simulates the behavior of brain and biological neural network, which can produce greater performance improvements than digital computing in the rapidly growing field of massive data sets, information recognition, and classification 2 , 3 . Currently, the Leaky Integrate and Fire (LIF) and Hodgkin-Huxley (HH) models are predominant in neuron simulation. The LIF model, being the simplest, is widely used for spike neuron simulations but lacks the ability to replicate many biologically relevant neuron features critical for computational neuroscience 4 . The CMOS-based HH model accurately replicates biological neuron behavior but necessitates a complex circuit design 5 . Neurons in the brain exhibit oscillatory behavior and multiple neurons can oscillate synchronously. The Fitz-Hugh Nagumo (FN) model, a simplification of the HH model, is optimal for neural networks exhibiting synchronous neuron oscillation 6 . Besides, FN neuron model can simulate the operation of biological neurons more accurately than LIF neurons. Furthermore, the neural network with FN neurons can be used to solve many complex computational problems and can also be used to study the work of the human brain. Therefore, the FN neuron model is considered one of the most successful models in computational neuroscience 7 . However, digital transistor-based chips attempt to model complex equations representing neuron-rich nonlinear dynamics, thereby complicating them, and these models are currently hampered by computational bottlenecks 8 , 9 . On the other hand, various new devices have been used to simulate biological neurons, benefiting from their greater biological similarity and scalability than digital transistors. Usually, FN neural circuits require N-type negative differential resistance (NDR) memristors to be implemented 10 , while traditional S-type NDR memristors including threshold switch memristors 11 , phase change memristors 12 , and Mott memristors 8 are difficult to support the implementation of FN neural circuits. On the other hand, due to the randomness of conductive filaments and nucleation sites 13 – 15 , these traditional NDR memristor devices are often unstable and typically require capacitors to integrate neurons, which limits their practical applications in large-scale neural morphology computing systems. Therefore, finding more stable N-type NDR devices to construct neural circuits is a very necessary task. Fortunately, the resonant tunneling diode (RTD) is a typical N-type NDR device that adopts band controlled tunneling mechanism and is less affected by temperature, which makes it have better device stability 16 . Therefore, combining the NDR effect of RTD and the hysteresis characteristics of memristors is a new idea to develop suitable devices for neuron circuits. Here, we have carefully designed a memristor with NDR effect to construct the electronic equivalent of biological neurons. The NDR memristor functional layer is assumed by AlAs/In 0.8 Ga 0.2 As/AlAs with quantum-well (QW) structure via metal-organic vapour phase epitaxy (MOVPE) ensures device stability and reproducibility, which displays a volatile resistance switch and is locally activated under the state of a hysteresis NDR with current-voltage characteristics. A new fabrication technique based on conventional i-line photolithography for micron-scale high current density NDR memristor devices is also developed with accurate control over the hence characteristics. The NDR memristor shows high reliability and temperature stability and can be cycled stably at room temperature and 10 9 cycles at 400 °C. The NDR memristor is used to construct a simple Fitz Hugh Nagumo (FN) neuron circuit, which effectively proves the feasibility and advantages of designing neuron circuit without capacitor. This design not only reduces hardware costs but also enables diverse neuron dynamics and functionality. The new FN neuron circuit designed by us realizes 9 kinds of neuron functions, including phasic spiking, anodal break excitation, spike accommodation, subthreshold oscillations, class 1 excitable, all or nothing firing, tonic bursting, refractory period, and accommodation. We constructed an edge detection device based on FN neurons, which showed better characteristics than other edge detection techniques by comparison. A voltage-temperature fused multimodal impulse neural network is also constructed based on the high-temperature stability of the device, and the results show that our system can distinguish images with different temperature labels. With excellent device performance and simple neuron circuit design, it opens a road to the neuromorphic computing with full memristor.",
"discussion": "Discussion In summary, we have developed an ultra-robust NDR memristor using the AlAs/In 0.8 Ga 0.2 As/AlAs quantum well (QW) structure, which possesses low variation, high temperature resistance, high temperature and high yield for a memristor-based neuromorphic system. This NDR memristor exhibits exceptional stability, including low variation, high temperature resistance, and endurance exceeding 10 11 cycles, making it ideal for neuromorphic computing. The innovative aspect of this memristor is its ability to integrate the neuronal membrane’s functionality, eliminating the need for external capacitors in the neuron circuit. The research successfully demonstrates the construction of a Fitz Hugh Nagumo (FN) neuron circuit using two NDR memristors and an inductor. This approach significantly reduces the hardware cost and complexity of FN neuron circuits, while offering versatility in neuron dynamics and functions. Finally, we applied this FN neuron to a voltage-temperature fusion multimodal image recognition system that can classify images with different temperature labels at high temperatures with a classification accuracy of 91.74%. This work presents a groundbreaking method to build FN neuron circuits with NDR memristors, contributing to the development of highly reliable neuromorphic hardware systems, specifically addressing the challenges of computational efficiency and reliability."
} | 2,039 |
39451698 | PMC11506689 | pmc | 219 | {
"abstract": "Microbial fuel cells (MFCs) represent a promising technology for sustainable energy generation, which leverages the metabolic activities of microorganisms to convert organic substrates into electrical energy. In oil spill scenarios, hydrocarbonoclastic biofilms naturally form at the water–oil interface, creating a distinct environment for microbial activity. In this work, we engineered a novel MFC that harnesses these biofilms by strategically positioning the positive electrode at this critical junction, integrating the biofilm’s natural properties into the MFC design. These biofilms, composed of specialized hydrocarbon-degrading bacteria, are vital in supporting electron transfer, significantly enhancing the system’s power generation. Next-generation sequencing and scanning electron microscopy were used to characterize the microbial community, revealing a significant enrichment of hydrocarbonoclastic Gammaproteobacteria within the biofilm. Notably, key genera such as Paenalcaligenes , Providencia , and Pseudomonas were identified as dominant members, each contributing to the degradation of complex hydrocarbons and supporting the electrogenic activity of the MFCs. An electrochemical analysis demonstrated that the MFC achieved a stable power output of 51.5 μW under static conditions, with an internal resistance of about 1.05 kΩ. The system showed remarkable long-term stability, which maintained consistent performance over a 5-day testing period, with an average daily energy storage of approximately 216 mJ. Additionally, the MFC effectively recovered after deep discharge cycles, sustaining power output for up to 7.5 h before requiring a recovery period. Overall, the study indicates that MFCs based on hydrocarbonoclastic biofilms provide a dual-functionality system, combining renewable energy generation with environmental remediation, particularly in wastewater treatment. Despite lower power output compared to other hydrocarbon-degrading MFCs, the results highlight the potential of this technology for autonomous sensor networks and other low-power applications, which required sustainable energy sources. Moreover, the hydrocarbonoclastic biofilm-based MFC presented here offer significant potential as a biosensor for real-time monitoring of hydrocarbons and other contaminants in water. The biofilm’s electrogenic properties enable the detection of organic compound degradation, positioning this system as ideal for environmental biosensing applications.",
"conclusion": "4. Conclusions In this study, we presented a novel microbial fuel cell design with a capacity of 100 L that leverages hydrocarbonoclastic biofilms formed at the water–oil interface to enhance both renewable energy generation and wastewater remediation. The microbial community analysis revealed a substantial enrichment of Gammaproteobacteria , which increased from 65.4% in the initial inoculum to over 90% in the biofilms formed on the anode. The MFC developed in this study achieved a peak power output of 270 μW, with a corresponding internal resistance of 136.6 Ω, and demonstrated stable operation over extended periods. Static power measurements revealed a maximum output of 51.5 μW with an internal resistance of 1.05 kΩ. Additionally, the system’s ability to recover and maintain performance after deep discharge was confirmed, with the MFC capable of delivering power for up to 7.5 h before requiring a recovery period. Over a five-day testing period, the MFC consistently stored an average of 216 mJ per day, sufficient to support low-power applications such as remote sensing and environmental monitoring. These results suggest that hydrocarbonoclastic biofilm-based MFCs offer a promising dual-functionality system that contributes to sustainable energy generation and addresses environmental challenges related to wastewater treatment and oil spill remediation. The ability to generate power while simultaneously degrading hydrocarbons highlights the potential of this technology for use in autonomous sensor networks, where reliable and sustainable power sources are essential. This MFC design is distinct in its ability to operate autonomously, utilizing a biofilm that forms naturally at the water–oil interface. This feature eliminates the need for active maintenance or intervention, making it suitable for applications requiring long-term stability and minimal external input. Although the power output is lower than other hydrocarbon-degrading MFCs, the system’s ability to support remote sensing applications demonstrates its potential for environmental monitoring and remediation. Furthermore, the electrogenic capabilities of hydrocarbonoclastic biofilms suggest that this MFC design can be effectively integrated as a biosensor for detecting environmental pollutants, such as hydrocarbons, in real time. This dual functionality underscores the system’s applicability in both energy production and environmental monitoring. Future research should focus on scaling up this technology and exploring its integration into various environmental applications. This study provides a foundation for developing advanced MFCs that can serve not only as efficient energy generators but also as effective tools for environmental remediation. By combining these dual capabilities, our discovery paves the way for groundbreaking solutions that address both renewable energy needs and environmental sustainability challenges. Furthermore, exploring potential symbiotic interactions during biofilm formation, such as by designing experiments with an artificial community of two different species or by loading Gammaproteobacteria alone, could yield further insights. This may help determine whether similar or higher levels of energy generation can be achieved with a simplified community structure. In summary, this innovative approach positions MFCs as a promising and versatile technology for future applications in energy production, waste management, environmental biosensing, and ecological preservation, offering a valuable contribution to a more sustainable and resilient world while leaving room for further exploration of alternative solutions.",
"introduction": "1. Introduction Harnessing the metabolic activities of microorganisms to convert organic substrates into electrical energy, microbial fuel cells (MFCs) represent a promising technology for sustainable energy generation [ 1 , 2 ]. These bio-electrochemical systems offer a clean and cost-effective power source, and their potential applications span across wastewater treatment [ 3 ], desalination [ 4 ], and energy production [ 5 ]. The principle underlying MFCs involves microorganisms oxidizing organic compounds at the anode, producing electricity, and simultaneously contributing to the breakdown of organic waste [ 6 , 7 , 8 , 9 , 10 , 11 ]. Recent advancements in MFC technology have improved their efficiency and expanded their potential applications, particularly in wastewater treatment and low-power micro-grids suitable for powering applications such as robotics, lighting, and wireless sensor networks (WSNs) [ 12 , 13 , 14 , 15 , 16 , 17 ]. Among the various challenges in wastewater treatment, scenarios involving oil spills can be taken into account. Oil spills create a physical barrier that limits atmospheric gas exchange and reduces light penetration, particularly impacting the euphotic zone where photosynthesis occurs [ 18 , 19 , 20 ]. The reduction in oxygen availability is further exacerbated by the microbial oxidation processes that consume the remaining oxygen, leading to the formation of anoxic zones. These anoxic conditions hinder the activity of oxygen-dependent (oxygenotrophic) microorganisms, slowing down the degradation of hydrocarbons. Despite these challenges, successional microbial communities consisting of facultative aerobic or anaerobic hydrocarbon-degrading species can adapt to these conditions [ 21 , 22 , 23 ]. These bacteria form biofilms at the water–oil interface, which are three-dimensional structures that provide a protective and adaptive environment for microbial communities [ 24 , 25 ]. To further understand the complexity of these biofilms and the interactions within microbial communities, advanced techniques like water-in-oil droplet-based microfluidics offer a deeper understanding of microbial communities, allowing the detection of less dominant species that might be overshadowed by more prevalent hydrocarbon-degrading bacteria [ 26 ]. This approach could also help identify symbiotic interactions and key strains for improved energy conversion, enhancing the potential of microbial fuel cells in environmental remediation. Biofilms play a crucial role in the degradation of hydrocarbons under anoxic conditions. The biofilm matrix facilitates nutrient availability and protects the bacteria from environmental stresses, enhancing their ability to degrade hydrocarbons effectively [ 27 , 28 , 29 ]. The existence of biofilms at the hydrocarbon–water interface has been documented, but the structural organization and the mechanisms enabling energy flow from the oil–water interface to the water column remain areas of active research. In this study, we present an innovative MFC design that utilizes the electrogenic properties of hydrocarbonoclastic biofilms formed at the water–oil interface. By integrating these biofilms into the MFC architecture, we aim to take advantage of their role in electron transfer, exploiting them to achieve a self-sustaining cell that autonomously generates electrical energy. Moreover, MFCs based on hydrocarbonoclastic biofilms present a dual-functionality system that improves both sustainable power generation and advanced biosensing capabilities. These systems leverage the electrogenic properties of specialized biofilms to detect and to respond to environmental changes, making them highly effective for real-time monitoring and environmental remediation. A unique feature of the MFC design we propose is that the biofilm naturally forms at the water–oil interface, which requires precise engineering to position the positive electrode there in order to optimize biofilm growth and electron transfer. Moreover, it functions autonomously without requiring external intervention or maintenance. In this paper, we begin by introducing the background and objectives of the study, followed by a detailed explanation of the materials and methods, including the design of the microbial fuel cell and the approaches used for microbial community and electrochemical analyses. The results highlight key findings on biofilm formation, microbial community structure, and the electrochemical performance of the MFCs, with a particular focus on its long-term stability. Finally, we conclude by summarizing the insights gained from this research and discussing the potential applications of hydrocarbonoclastic biofilm-based MFCs."
} | 2,720 |
40108150 | PMC11923061 | pmc | 220 | {
"abstract": "Memristor crossbar arrays (CBAs) based on two-dimensional (2D) materials have emerged as a potential solution to overcome the limitations of energy consumption and latency associated with conventional von Neumann architectures. However, current 2D memristor CBAs encounter specific challenges such as limited array size, high sneak path current, and lack of integration with peripheral circuits for hardware compute-in-memory (CIM) systems. In this work, we demonstrate a hardware CIM system leveraging heterogeneous integration of scalable 2D hafnium diselenide (HfSe 2 ) memristors and silicon (Si) selectors, as well as their integration with peripheral control-sensing circuits. The 32 × 32 one-selector-one-memristor (1S1R) array mitigates sneak current, achieving 89% yield. The integrated CBA demonstrates an improvement of energy efficiency and response time comparable to state-of-the-art 2D materials-based memristors. To take advantage of low latency devices for achieving low energy systems, we use time-domain sensing circuits with the CBA, whose power consumption surpasses that of analog-to-digital converters (ADCs) by 2.5 folds. The implemented full-hardware binary convolutional neural network (CNN) achieves remarkable accuracy (97.5%) in a pattern recognition task. Additionally, in-built activation functions enhance the energy efficiency of the system. This silicon-compatible heterogeneous integration approach presents a promising hardware solution for artificial intelligence (AI) applications.",
"introduction": "Introduction The introduction of CIM architecture has provided a promising solution to address the inherent inefficiencies related to data movement in the traditional von Neumann architecture 1 – 4 . Simultaneously, it has enhanced parallel computing capabilities and energy efficiency within the domain of artificial intelligence (AI) applications 5 – 8 . A crucial component within this architecture is the multiply-and-accumulate (MAC) unit 9 – 11 . Memristors have garnered significant attention as viable candidates for MAC operations in CIM due to their compact size 12 , high array density 13 , 14 , low energy requirements 15 – 18 , and compatibility with back-end-of-line (BEOL) integration 19 – 21 . In particular, memristors based on 2D materials have generated substantial research interest due to their properties that are essential for AI applications, especially the ultrathin thickness and low switching voltage for energy-efficient computing (Supplementary Table 1 ). However, the advancement of 2D materials-based memristive crossbar arrays (CBAs) face several challenges. While the synthesis and integration of 2D materials still face challenges 22 – 25 , recent advances have demonstrated successful wafer-scale and monolithic 3D integration of 2D materials, achieving remarkable performance in logic and memory applications 26 – 32 . This method is impractical for scaling up to large-scale integration with silicon platforms. Scalable methods such as chemical vapor deposition (CVD), molecular beam epitaxy (MBE), and atomic layer deposition (ALD) present challenges in achieving growth processes compatible with back-end-of-line (BEOL) technology, where the growth temperature must remain below 450 °C 33 . Alternative approaches involve the growth of 2D materials over a substrate without strict temperature constraints, followed by the transfer of the 2D material onto the desired substrate. However, conventional transfer processes may introduce impurities and doping effects into the 2D material due to the utilization of polymers and liquid media 34 , 35 . Furthermore, the implementation of CIM hardware utilizing 2D materials-based CBAs also faces challenges, including the lack of process integration, limited array size, speed constraints, and the absence of integration with peripheral control and sensing circuits. To enhance device selectivity within the CBA and mitigate leakage current, the incorporation of access selectors or transistors becomes crucial in the process integration 36 , 37 . Nevertheless, existing 2D materials-based memristor arrays predominantly focus on memristor-only (1 R) array configurations, thus lacking scalability 22 , 23 , 32 , 34 , 35 , 38 , 39 . The issue of array size presents another obstacle as large-scale CBAs are indispensable for weighted layers in neural networks. Fully-connected (FC) layers, for instance, necessitate hundreds of neurons, while convolution layers require hundreds of channels, both of which rely on large array sizes for efficient parallel computing 40 , 41 . Although some heterogeneous integration of 2D materials-based memristors have been reported, such as the graphene transistor/hexagonal boron nitride (h-BN) memristor-based 0.5T0.5 R single cell 42 , the MoS 2 transistor/h-BN memristor-based 1T1R single cell 43 , the Si transistor/MoS 2 memristor-based 2T1R single cell 44 , and the Si transistor/h-BN memristor-based 1T1R 5 × 5 CBA 45 , these demonstrations suffer from limited array sizes and fail to accommodate the complete weight mapping of a neural network. Moreover, the slow programming speed remains a challenge as state-of-the-art heterogeneous Si/2D integrated 1T1R arrays report a slow program time of 232 us, which fall short of the requirement for tens of nanoseconds in compute-in-memory operation 45 , 46 . Engineering the integration of 2D materials-based memristive CBAs with peripheral circuits at the hardware-level for CIM system has not been adequately addressed. Previous demonstrations have primarily focused on device features extracted from small CBAs without integrated peripheral circuits and have evaluated array functionality using simulation approaches 45 , 47 – 49 . Variable resistive states have been observed in HfSe 2 -metal compounds, showing potential for memristive devices 50 , 51 . Our research indicates that Molecular Beam Epitaxy (MBE) can achieve wafer-scale growth of 2D HfSe 2 thin films, enabling large-scale (1 R) memristor arrays 52 . HfSe 2 offers advantages in scalability and integration with silicon-based devices for constructing 1S1R arrays. The restricted bandgap of these semiconductors allows for lower set and reset voltages, improving energy efficiency compared to hexagonal boron nitride (hBN) 53 . A thinner semiconductor layer (3 nm), as shown in device cross-section TEM (Supplementary Fig. 26 ), shows better-switching characteristics than thicker hBN layers (6 nm) 45 . In this work, we demonstrate a hardware CIM system that leverages on heterogeneous integration technology and incorporates the design of low-energy peripheral circuits. The integrated crossbar array utilizes a 1S1R configuration, where each cell integrates a Si-based selector and a HfSe 2 -based memristor. To enable this integration, a low-temperature three-dimensional (3D) stacking process is employed, involving wafer-scale 2D HfSe 2 synthesis and wafer-scale metal-assisted transfer techniques, positioning the 2D memristor above the Si-selector, thereby ensuring compatibility with complementary metal-oxide-semiconductor (CMOS) process. As a result, we successfully fabricated a 1-kilobit (32 × 32) one-selector-one-memristor (1S1R) crossbar array (CBA). This integrated array exhibits significantly reduced sneak current and enhanced endurance when compared to other 2D materials-based memristors that lack access devices. In comparison to the current state-of-the-art 2D materials-based memristors integrated with selectors, our presented integrated array demonstrates a marked improvement in both response time and energy efficiency 45 , 54 . Furthermore, these enhancements are shown to be comparable to those achieved by conventional oxide-based memristors 9 , 21 , 55 – 57 , with the added advantages of superior switching voltage, faster switching and reduced thickness as shown in Supplementary Table 11 . Additionally, time-domain sensing peripheral circuits are designed by utilizing a time-to-digital converter (TDC) for energy-efficient reading and computing, which takes advantage of the rapid device speed. The CIM hardware system achieves full integration of the CBA with peripheral circuits and demonstrates high-accuracy CNN inference. Leveraging the nonlinear behavior of the TDC circuits, analog computing capabilities and in-built activation functions are developed, enhancing energy efficiency during computation for the CNN. Finally, this work suggests that semiconductors with limited bandgaps provide improved voltage and energy efficiency, opening routes for future low-voltage, high-speed memristor applications in logic and storage.",
"discussion": "Discussion In conclusion, we have presented a comprehensive implementation of CNNs using our hardware CIM system. The large-scale CBA achieved remarkable response speed, energy efficiency and extensive statistical analysis, leveraging the advantages of heterogeneous integration with 2D-Si materials. By integrating specialized sensing circuits, low-power, time-domain parallel readout capabilities were demonstrated. The combination of the central CBA design and peripheral circuit design facilitated efficient MAC operations within the integrated system. Our full-hardware CNN implementation successfully classified binary input images, highlighting the potential of our CIM system for AI applications. Moreover, the incorporation of analog computing and in-built activation functions expanded the capabilities of our CIM system, allowing effective handling of analog inputs and reducing data transfer during forward propagation. The hardware-based recognition process maintained high accuracy, demonstrating the robustness and performance of our system. These results manifest the potential of our CIM system as real-time accelerators, combining high-accuracy CNN inference with minimized energy consumption. This work contributes to the advancement of hardware-based AI systems, showcasing the effectiveness of 2D materials-based memristive CBAs and their fully integrated CIM hardware in enabling efficient and accurate neural network computations."
} | 2,521 |
39203380 | PMC11356306 | pmc | 221 | {
"abstract": "Symbiotic microorganisms in reef-building corals, including algae, bacteria, archaea, fungi, and viruses, play critical roles in the adaptation of coral hosts to adverse environmental conditions. However, their adaptation and functional relationships in nutrient-rich environments have yet to be fully explored. This study investigated Duncanopsammia peltata and the surrounding seawater and sediments from protected and non-protected areas in the summer and winter in Dongshan Bay. High-throughput sequencing was used to characterize community changes, co-occurrence patterns, and factors influencing symbiotic coral microorganisms (zooxanthellae, bacteria, and archaea) in different environments. The results showed that nutrient enrichment in the protected and non-protected areas was the greatest in December, followed by the non-protected area in August. In contrast, the August protected area had the lowest nutrient enrichment. Significant differences were found in the composition of the bacterial and archaeal communities in seawater and sediments from different regions. Among the coral symbiotic microorganisms, the main dominant species of zooxanthellae is the C1 subspecies (42.22–56.35%). The dominant phyla of bacteria were Proteobacteria, Cyanobacteria, Firmicutes, and Bacteroidota. Only in the August protected area did a large number (41.98%) of SAR324_cladeMarine_group_B exist. The August protected and non-protected areas and December protected and non-protected areas contained beneficial bacteria as biomarkers. They were Nisaea , Spiroplasma , Endozoicomonas , and Bacillus. No pathogenic bacteria appeared in the protected area in August. The dominant phylum in Archaea was Crenarchaeota. These symbiotic coral microorganisms’ relative abundances and compositions vary with environmental changes. The enrichment of dissolved inorganic nitrogen in environmental media is a key factor affecting the composition of coral microbial communities. Co-occurrence analysis showed that nutrient enrichment under anthropogenic disturbances enhanced the interactions between coral symbiotic microorganisms. These findings improve our understanding of the adaptations of coral holobionts to various nutritional environments.",
"conclusion": "5. Conclusions This study analyzed the differences in the microbial community composition, influencing factors, and co-occurrence patterns in D. peltata , seawater, and sediments under different anthropogenic disturbances. Anthropogenic disturbances reduced the diversity of coral symbiotic algae, increased the species composition of coral symbiotic bacteria, and altered the community structure of coral archaea. DIN is a key factor affecting coral microbial communities. Co-occurrence network analysis revealed the interactions and functions of the coral microbial communities, reflecting the interactions and connections between microbial taxa in large seawater, coral, and sediment environments. These findings suggest that coral microbiomes vary across environments, which may impair host immunity, enabling diverse opportunistic bacteria in coral symbionts. Also, coral keystone taxa do not necessarily coincide with their dominant taxa. This study expands our understanding of the effect of nutrient enrichment on D. peltata symbiotic microorganisms (zooxanthellae, bacteria, and archaea). It improves our understanding of the adaptive mechanisms of coral holobionts in response to environmental changes. Further research is needed to fully understand the structure and function of microbial communities in coral holobionts and to determine how microorganisms in the natural environment establish connections and interact with corals.",
"introduction": "1. Introduction Healthy coral reefs are among Earth’s most productive and biodiverse ecosystems [ 1 ], providing various ecosystem services to humans and performing essential functions [ 2 ]. However, the environment in coastal areas has been damaged globally by increased human activity, resulting in the degradation of many coral reefs [ 3 , 4 ]. Although many anthropogenic environmental stressors severely affect coral reef communities, the effects of nutrients on coral reefs are diverse [ 5 ]. Numerous studies have shown that the extent to which corals respond to nutrients depends on the type of nutrient enrichment. Nitrogen enrichment may reduce coral calcification levels but can increase the growth and development of Symbiodiniaceae to a certain extent. In contrast, phosphorus enrichment is more likely to increase the calcification rate of corals [ 6 ]. Appropriate ammonia acclimation can improve coral immunity [ 7 ]. Nitrate works synergistically with temperature [ 8 ], and nutrient enrichment reduces the ability of corals to resist bleaching [ 9 ]. Eutrophication also affects the distribution of allele frequencies with coral bleaching resistance [ 10 ]. Therefore, studying the adaptability of corals to complex nutrient-enriched environments could provide a theoretical basis and guidance for domesticating corals to withstand environmental stress. Coral stress resistance and adaptability are closely related to coral microbiota, which includes symbiotic algae, bacteria, archaea, fungi, and viruses. This complex ecological assemblage is known as the coral holobiont [ 11 ]. Research suggests that microbial communities that have coevolved with their coral hosts may help corals adapt to their environments [ 12 ]. As the dominant microorganisms in coral symbionts, symbiotic zooxanthellae provide the most nutritional requirements for their coral hosts [ 13 ]. Research has shown that coral holobionts adapt to environmental changes primarily by reconfiguring zooxanthellae components [ 14 ]. Increasing ambient temperatures cause the coralline zooxanthella genus to shift from Cladocopium (formerly Clade C) to Durusdinium (formerly Clade D) [ 15 ]. Moderate nutrient levels can increase symbiotic algal density and improve coral photosynthesis [ 16 ]. Increasing evidence has revealed the critical role played by specific bacterial species in prokaryotic communities in maintaining the fitness of the entire holobiont [ 17 ]. This type of bacteria is called the “core bacteria” of corals and remains stable within a specific time and space range [ 18 ]. Different bacterial community compositions help corals cope with changes in sea-surface temperature and eutrophication [ 19 , 20 ]. Crenarchaeota are the most abundant archaea found in corals. These archaea are primarily involved in the recycling and transforming of nutrients in corals [ 21 ]. This transformation results in changes in the archaeal communities of corals as nutrients enter the environment. Nutrient changes significantly alter species richness and community variability among coral microbiomes [ 22 ]. The structure of coral microbial communities is critical to host health; as the commensal microbial community changes, the health status of corals may also change [ 23 ]. Environmental stress may allow other organisms to replace stable microbiota, leading to the development of coral diseases [ 24 ]. Furthermore, increased pathogenic microorganisms can destabilize coral self-regulatory systems [ 25 ]. Therefore, characterizing the structure of symbiotic coral microorganisms is critical for understanding coral reef resilience and predicting ecological change. The coral communities in Dongshan Bay form the northernmost edge of the reef-building coral communities in China. Located near the shore, they are easily affected by anthropogenic activities under the combined stress of global climate change and regional nutrient enrichment. The overall coral community in Dongshan Bay showed low species richness (less than 10 species), with coral species usually having high environmental tolerance primarily dominated by slow-growing block corals or plate corals with large polyps [ 26 ]. This situation provides a unique context for exploring how stress-tolerant coral species respond to long-term nutrient enrichment, primarily through anthropogenic activities [ 27 ]. Duncanopsammia peltata (Esper, 1790) is widely distributed in the Indo-Pacific region and is a dominant coral species with ecological significance in Dongshan Bay [ 26 ]. Therefore, it can serve as a model species for understanding how corals adapt to changing environmental conditions [ 28 ]. A detailed analysis of the relationship between the coral symbiotic microorganisms and the environment in Dongshan Bay would be more conducive to the protection of corals. Previous studies have mainly elaborated on the effect of temperature on symbiotic coral microbial communities [ 29 ]. However, only some studies have focused on the response of symbiotic coral microorganisms to changes in nutrients. The present study aimed to comprehensively investigate the adaptation of D. peltata symbiotic microorganisms to nutrient enrichment by collecting D. peltata from different areas (protected and non-protected areas) at other times (August and December) in Dongshan Bay waters and analyzing the diversity, species composition, and co-occurrence of coral symbiotic microorganisms. High-throughput sequencing of a specific barcode region (ITS or 16S) was applied to determine the community composition and changes in zooxanthellae, bacteria, and archaea in D. peltata under different nutritional environments. Although high-throughput metabarcode sequencing has some potential limitations and challenges, such as limited resolution, PCR bias, and quantitative accuracy, it has significant advantages, such as high-throughput capability, wide applicability, high sensitivity, relative cost-effectiveness, and data reproducibility, which make it a powerful tool for the analysis of coral symbiotic microbial communities. We reveal the relationship between these different symbiotic microbial communities in response to environmental variations and identify the respective keystone taxa groups involved. This study shows the characteristics and functional relationships of community changes in coral symbiotic microorganisms under nutrient enrichment and provides a theoretical reference for coral ecological protection.",
"discussion": "4. Discussion 4.1. Coral Symbiotic Microorganisms Symbiodiniaceae has been partitioned into multiple genera to better reflect their origins and divergence, as determined by molecular dating [ 37 ]. Many studies have shown that corals are associated primarily with endosymbionts in the genus Cladocopium (Clade C) [ 38 , 39 ]. This is mainly because of the robust adaptability of Cladocopium to its surroundings, particularly the thermally resilient lineage known as Cladocopium C15, which is commonly observed in Porites species [ 40 ]. The data collected in this study clearly show that in the Symbiodiniaceae family of D. peltata , Cladocopium dominated in different environments, with C1 being particularly prominent [ 29 ]. C1 is widely distributed in the waters of northern China, Okinawa, and the Jeju islands. These areas are typically characterized by low sea temperatures and high human activity levels [ 41 , 42 , 43 ]. Therefore, D. peltata may have established a stable symbiotic relationship with C1 to adapt to subtropical environments. Studies have also found that the symbiosis between host corals and Symbiodiniaceae changes over time and space on a small scale, with environmental differences [ 44 , 45 ]. Although seasonal changes have little effect on the composition of symbiotic-dominant algae [ 46 ], the diversity of non-dominant algae living in symbiosis with corals is affected by seasonal changes [ 47 ]. In the present study, C1, C1c, C18, C3, and C1054 were consistently present in different environments, indicating that they were the dominant algae in D. peltata . However, the diversity of Symbiodiniaceae in December was lower than in August, suggesting that corals suffered from stress in December and lost some Symbiodiniaceae algae. One notable difference was that C1a was found only in AugNPA, which may indicate that higher sea surface temperatures positively affect coral Symbiodiniaceae diversity [ 48 ]. However, the specific mechanism underlying the change in the coral Symbiodiniaceae requires further experimental data. Despite the high level of bacterial diversity in corals, the dominant bacteria (relative abundance > 1%) show some stability in space and time [ 18 ]. We found that Proteobacteria, Bacteroidetes, and Cyanobacteria in D. peltata were the dominant bacterial species, consistent with the overall composition of coral bacteria worldwide [ 49 , 50 ]. As the surrounding environment changes, the symbiotic bacteria composition within corals undergoes dynamic changes [ 25 ]. For example, SAR324_ clade Marine _group_B, widely present in deep seawater (75–125 m), has highly plastic metabolic characteristics, including fixing inorganic carbon and metabolizing oxidized sulfur [ 51 ]. In the present study, this bacterium only appeared in AugPA, indicating that D. peltata may have carried out more active carbon metabolism and sulfide oxidation in this area. Environmental factors can significantly influence microbial composition and function within a specific location [ 52 ]. The NMDS diversity analysis results revealed significant differences in the ecological structures of the PA and the NPA. The Mantel test analysis revealed that nitrogen nutrients are essential in distinguishing the PA from the NPA. The alpha diversity results showed that bacteria in the NPA had a higher Shannon index than those in the PA ( p < 0.01), which is consistent with the increase in the microbial richness of coral reefs when facing stress [ 53 ], possibly because opportunistic bacteria take advantage of compromised immune systems to colonize corals [ 54 ]. Studies have shown that some bacteria enhance stress resistance in coral hosts [ 55 ]. For example, Firmicutes are a typical class of beneficial bacteria essential in improving corals’ stress tolerance [ 56 ]. This study revealed that in August or December, the abundance of Firmicutes in D. peltata in the NPA was higher than in the PA. This may be a beneficial bacterial adjustment by the coral in the face of environmental stress. This study identified Nisaea in AugPA as a biomarker [ 57 ], and Thalassotalea , Vibrio , and Woeseia pathogenic bacteria were found in AugNPA, DecPA, and DecNPA [ 58 , 59 , 60 ]. However, the beneficial bacteria Spiroplasma , Endozoicomonas , and Bacillus are still used as biomarkers in these three areas [ 55 ]. These results show that the AugNPA, DecPA, and DecNPA environments have a specific stress effect on corals. However, corals actively adjust their bacterial community structure while adapting to the environment, with beneficial bacteria being the dominant species. This indicates the independent adaptability of corals to their environment. Coral-associated archaea are often overlooked in genomic studies of coral microorganisms, primarily because of their low read count in genome sequencing compared to bacteria [ 61 ]. Furthermore, primers commonly used to target bacterial 16S rRNA genes often fail to detect archaeal 16S rRNA sequences [ 62 ]. Nonetheless, studies have shown that Archaea play a vital role in the nitrogen cycle in coral symbiotic functional systems [ 63 ], mainly through nitrification and denitrification processes, which help the mucus layer capture excess ammonium and thus become a nutrient sink [ 21 ]. Therefore, studying the coral archaea is crucial. This study’s analysis of the coral archaea showed that the phylum Crenarchaeota dominated the coral archaea. According to previous studies, the average proportion of Crenarchaeota in the environment is 79% [ 21 ], which is consistent with the results of this study. NMDS and mantle test analyses of archaea revealed that nitrogen content is an essential indicator of the composition of archaeal communities. Temperature was not a critical factor affecting microbial composition, possibly due to the adaptation of D. peltata to a lower temperature range. Although our understanding of archaeal function remains limited, genomic analyses have demonstrated their potential for nitrogen cycling, particularly the ammonium oxidation capabilities of Crenarchaeota [ 40 ]. Halobacteria typically exist in high-salt environments and adapt well to high salt concentrations [ 64 ]. A certain number of halobacteria were detected in the sampling area in August, consistent with the environmental parameters. In the December sampling area, methanogens such as Methanobrevibacter , Methanobacterium , and Methanococcoides were dominant. This indicates that the sampling area in December may be rich in organic waste or decomposition products of organic matter and may be anaerobic [ 65 ]. Coral symbiotic microorganisms participate in the circulation of nutrients and the maintenance of productivity, which is beneficial to the stability of the entire coral reef ecosystem [ 56 ]. 4.2. Effects of Environmental Factors on Symbiotic Microorganisms The core coral microbiome is crucial in adapting coral holobionts to environmental stresses [ 66 ]. The abundance of symbiotic coral microorganisms often changes in response to environmental influences [ 55 ]. In this study, the core zooxanthellae C1 ranged from 56.25% in AugPA to 44.22% in AugNPA, indicating that the relative abundance of core Symbiodiniaceae changed with environmental changes ( Figure 2 B). In a study of coral symbiotic microorganisms at different latitudes in China, the core group of symbiotic algae in the same coral changed at different latitudes [ 29 ]. The core bacteria SAR324_cladeMarine_group_B of the corals in this study ranged from 41.98% of AugPA to 0.23% of AugNPA, indicating that the coral core bacterial community changed from dominant bacterial species to rare bacterial species ( Figure 4 A). In this study, the core archaea Crenarchaeota of corals ranged from 90.67% in AugPA to 95.58% in AugNPA, indicating that the relative abundance of core archaea also changed with environmental changes ( Figure 5 C). However, its change sensitivity was lower than that of bacteria. Higher DIN may have caused the shift in the centrality of coral symbiont microorganisms in this study in AugNPA. Furthermore, in the face of environmental changes, core symbiotic microorganisms may interact synergistically to provide nutritional support for coral holobionts to adapt to their environments [ 67 ]. Environmental factors such as high nutrient salinity may drive core coral microbial communities from dominant to rare [ 68 ]. Climate change could disrupt the symbiotic relationship between corals and their symbiotic microorganisms, causing major changes in the composition of coral microbial communities, characterized by overgrowth of pathogenic microorganisms, which will lead to coral bleaching and/or disease [ 69 ]. In summary, the impact of environmental factors on coral symbiotic microbial communities is a relatively complex process involving the relative abundance of core microbial community members, the importance of rare microbial species, and the role of environmental factors. A better understanding of these effects will benefit the conservation and management of coral ecosystems. 4.3. Co-Occurrence Network Features Co-occurrence network analysis is widely used to analyze the interactive relationships of microbial communities. It provides an overall insight into the composition of microbial communities and a better understanding of the complex interactions between microorganisms [ 70 , 71 ]. The seawater and sediment network diagrams and topological structures ( Figure 7 A and Table S6 ) showed that AugNPA had a higher network complexity than AugPA, consistent with the network results for coral symbiotic microorganisms. Meanwhile, the higher number of nodes and a lower average degree in NPA indicated that the node connections in the network were relatively dispersed rather than concentrated on a few nodes. This shows lower network stability and less efficient information dissemination [ 72 ], indicating the vulnerability of seawater and sediment ecosystems in NPA. DecPA also had a higher network density and lower average degree, suggesting that DecPA has a similar ecological vulnerability to AugNPA. The differences in the microbial network relationships between PA and NPA illustrate the importance of establishing PA to maintain the balance and strength of marine ecosystems. To utilize the interactions between microorganisms to improve the stability of marine ecosystems, it is necessary to explore the experimental conditions by creating a suitable environment. We constructed a symbiotic community network based on zooxanthellae, bacteria, and archaea among coral symbiotic microorganisms. The results showed that the co-occurrence relationships of bacteria in different environments were complex and diverse, occupying a significant position in the co-occurrence network. In contrast, coral symbiotic zooxanthellae and archaea co-occurrence patterns were relatively stable. The topology of a network can visually represent the interactions between microorganisms [ 73 ]. In this study, the NPA had more nodes and edges and a higher network density and degree than the PA. The microbial network was more connected than the microbial network in December than in August. This suggests that coral microbial communities with more significant anthropogenic impact have more complex co-occurrence networks. Modularity represents the degree of microbial aggregation in blocks and is an essential indicator of microbial interactions [ 74 ]. All classified network structures had high modularity values (0.448–0.913), all greater than 0.4, indicating an apparent modular structure [ 75 ]. This modular feature helps reduce the impact of environmental chaos, with each module containing one or more internally tightly connected microbial groups [ 76 ]. The whole and December groups exhibited high-density but low-modular network structures that may be unstable under external interference [ 77 ]. Corals from the PA were more modular than those from the NPA, indicating that the PA microorganisms may gather more closely and be able to adjust more efficiently to changes in the external environment [ 78 ]. These results suggest that coral microbial networks heavily disturbed by humans are more fragile and unstable than those less disturbed [ 79 ]. In this study, most microbial interactions were positive across all networks. Positive microbial interactions can form cooperative or symbiotic alliances, allowing symbiotic coral microorganisms to exchange nutrients rapidly. However, intense positive interactions are often accompanied by significant internal environmental pressures, resulting in fragile internal coral environments of corals [ 80 ]. Therefore, positive interactions within coral microbial communities may harm coral hosts in areas of severe anthropogenic disturbance, even though they increase network stability. Further studies should be conducted to elucidate the mechanisms by which nutrient enrichment enhances interactions between coral symbiotic microorganisms. In summary, PA corals have a relatively stable microbial community network that helps to maintain homeostasis in their internal environment. Under environmental pressures, such as high temperature, high acidity, and eutrophication, the immunity of corals decreases, and the composition of the symbiotic microbial community undergoes significant changes. Some pathogenic bacteria invade the internal symbiotic environment of corals and cause coral bleaching [ 81 ]. Coral microbial diversity increases under environmental stress, enhancing microbial interactions and establishing relationships between adjacent organisms, leading to increased correlation, reduced stability, and reduced host adaptability [ 29 ]. This study provides an in-depth exploration of the possible impact of human factors on the symbiotic microbial interactions of D. peltata by creating a co-occurrence network. These results indicate that interactions between coral microbial communities become more complex under severe anthropogenic disturbance. 4.4. Composition of Coral Keystone Taxa Microbial interactions are key factors that determine the community structure and function [ 82 ]. In microbial communities, some key taxa may have low abundance in space and time but have a significant impact on the structure and function of the microbiome, either individually or as a group. These taxa play unique and critical roles in microbial communities and, if removed, lead to significant changes in microbial community structure and function [ 36 ]. This study identified some possible key groups of D. peltata symbiotic microorganisms using co-occurrence network analysis ( Table S8 ). These results indicated that not all keystone zooxanthellae taxa were dominant. However, these keystone taxa mainly exist in the NPA, and it can be inferred that the NPA has more complex food webs, symbiotic relationships, and ecosystem services [ 36 ]. Keystone taxa, among bacteria and archaea, play essential roles in the decomposition and transformation of organic matter, emphasizing their positive impact on material circulation and energy flow throughout the system."
} | 6,345 |
37513093 | PMC10385009 | pmc | 222 | {
"abstract": "The electrical characteristics and resistive switching properties of memristive devices have been studied in a wide temperature range. The insulator and electrode materials of these devices (silicon oxide and titanium nitride, respectively) are fully compatible with conventional complementary metal-oxide-semiconductor (CMOS) fabrication processes. Silicon oxide is also obtained through the low-temperature chemical vapor deposition method. It is revealed that the as-fabricated devices do not require electroforming but their resistance state cannot be stored before thermal treatment. After the thermal treatment, the devices exhibit bipolar-type resistive switching with synaptic behavior. The conduction mechanisms in the device stack are associated with the effect of traps in the insulator, which form filaments in the places where the electric field is concentrated. The filaments shortcut the capacitance of the stack to different degrees in the high-resistance state (HRS) and in the low-resistance state (LRS). As a result, the electron transport possesses an activation nature with relatively low values of activation energy in an HRS. On the contrary, Ohm’s law and tunneling are observed in an LRS. CMOS-compatible materials and low-temperature fabrication techniques enable the easy integration of the studied resistive-switching devices with traditional analog–digital circuits to implement new-generation hardware neuromorphic systems.",
"conclusion": "4. Conclusions To summarize, the electrical characteristics and resistive switching properties of CMOS-compatible SiO x -based memristive devices are reported. In the initial state and before the thermal treatment, the device stacks do not require electroforming, but cannot store the recorded state. This is attributed to the impact of conductive defects formed during the fabrication of the stacks. After the thermal treatment, the devices exhibit bipolar resistive switching with a gradual transition between multiple resistances typical of synaptic behavior. This synaptic switching causes a change in the dielectric losses in the devices (and hence in the resistive switching characteristics) in the low frequency range (<10 5 Hz) only. When switching to the high-resistance state at elevated temperatures (482 K), the filaments do not break completely. At low temperatures (77 K), no dielectric losses associated with the Ohmic leakages (through the filaments) are found. The mechanisms of current transport in different resistive states of the memristive device are suggested: the Schottky or the Poole–Frenkel mechanism in the high resistance state and trap-assisted tunneling in the low resistance state. The values of the density and activation energies of traps in the SiO x film are determined and are similar to those of typical defects constituting conductive filaments in silicon oxide.",
"introduction": "1. Introduction At present, artificial intelligence (AI) systems play a significant role in our everyday life. At the same time, more and more complex AI applications require the development of efficient neuromorphic (brain-inspired) computing technologies [ 1 ]. New AI hardware is subject to a number of requirements for compactness, energy efficiency, and compatibility with the conventional complementary metal-oxide-semiconductor (CMOS) fabrication process. A memristor [ 2 ] or a memristive device is a simple two-terminal electronic device, which is able to change its resistive state adaptively depending on the voltage applied and maintain it for a long time. The change in resistance (resistive switching between at least two states: a low-resistance state (LRS) and a high-resistance state (HRS)) usually occurs due to rupturing and restoring the conductive filaments inside the insulator film sandwiched between metal electrodes. The diameter of filaments can range from 1 to 100 nm depending on the materials used and the mechanism behind their formation [ 3 ]. Resistive-switching devices are suitable for applications as compact (nanometer-sized) and energy-efficient (femtojoules per switching operation) Resistive Random-Access Memory devices, which can be easily integrated back-end-of-line (BEOL) with the front-end-of-line (FEOL) CMOS circuitry [ 4 ]. Such devices are capable of not only storing the Boolean values encoded by the resistance values (LRS or HRS), but also allow the alteration of these ones inside the memory circuits implementing the so-called “non-von Neumann” or “in-memory computing” paradigm. In addition, a simple structure of the memristive device stack allows for the development of super-dense and potentially three-dimensional cross-bar arrays, which implement the vector–matrix multiplication operations underlying inference in traditional deep learning artificial neural networks [ 5 , 6 ] and new learning algorithms for training the spiking neural networks [ 7 ]. In the present work, the electrical characteristics and resistive switching properties of metal–insulator–metal (MIM) device stacks based on a SiO x film obtained through a low-temperature plasma-enhanced chemical vapor deposition (PECVD) method were studied with the final goal of integrating these stacks into modern CMOS circuits.",
"discussion": "3. Results and Discussion The results of the XPS characterization of the initial SiO x film on the substrate are shown in Figure 1 . One can see the oxygen fraction x in the SiO x film material varying from 1.8–1.9 at the film surface up to 2.5–2.6 at the interface with the substrate corresponding to various silicon oxide types revealed in the chemical composition of the film. The partial oxidation of the titanium nitride bottom electrode should also be noted. The measurements of the frequency dependencies of the equivalent circuit parameters of the memristive stack in the initial state demonstrate high Ohmic losses (up to tg δ ~ 30 at f = 1 kHz), a high value of dielectric permittivity ε ≈ 9.8 at f = 100 kHz, and a low value of parallel resistance R p ~ 10 kΩ at f = 1 kHz. The above values can be explained by the presence of conductive defects nonuniformly distributed in the insulator in the initial state. The localization of such defects in the form of conductive channels similar to filaments during the deposition of the top electrodes can probably be facilitated by the contact potential difference between the top electrode and the bottom one. This is due to the different values of the electron work function for Zr (4.05 eV [ 9 ]) and for TiN (4.3–4.65 eV [ 10 ]), as well as the low thickness of the SiO x film. As a result, the electric field strength in the insulator at a zero bias is high enough ((1.25–3) × 10 5 V/cm) to promote the filament formation. Figure 2 a shows the I – V curves of the memristive stack in the initial state. Under an applied voltage of +3 V, the stack switches from the initial LRS to an HRS ( Figure 2 a, curve 1). Repeated measurements of the I–V curves in the same conditions demonstrate the LRS is restored after few minutes ( Figure 2 a, curve 2). Earlier, the thermal treatment of memristive stacks was shown to result in an irreversible change in their electrical characteristics and to positively affect the resistive switching performance of these stacks positively in some cases [ 8 ]. Because the memristive stacks under study did not demonstrate stable resistive switching in the initial state, they were subjected to thermal treatment in dried air ambient in a sealed thermostat at 250 °C for 10 min. This results in a change in the nature of resistive switching and in an irreversible change in electrical characteristics. After the thermal treatment, the devices demonstrate a decrease in the dielectric losses at low frequencies, in the leakage currents, and in the ε value (down to ≈ 5.4 at f = 100 kHz). These changes are associated with the oxidation of excess Si during the thermal treatment. After this treatment, the stacks also demonstrate bipolar resistive switching with synaptic behavior ( Figure 2 b). Namely, the current values in multiple HRSs decrease gradually with increasing voltage. Figure 3 shows the frequency dependencies of the equivalent circuit parameters of a memristive device in the HRS obtained after the successive application of a positive voltage ( Figure 2 b). An increase in the voltage applied leads to a change in the equivalent circuit parameters in the frequency range f < 10 5 Hz. The increase in dielectric losses with a decreasing frequency is described by losses of the Ohmic nature [ 11 ], which are minimal after switching with a voltage of +7 V. In the same frequency range, R p decreases with increasing voltage. However, the essential shortcutting effect of the Ohmic leakage is also evident after switching with the voltage of +7 V. Thus, it is not possible to eliminate the effect of filaments on the electrical characteristics in the stack under study completely. It should be noted that the synaptic behavior revealed in the gradual resistance change also takes place under the negative voltage, which transfers the device to an LRS. In this case, the changes in the value of R p and in the dielectric losses in the low-frequency range also occur. It is important to note that, in both cases (HRS and LRS), there are no changes in R s at high frequencies. At f = 1 MHz, R s is ~18 Ω, the value of R s at a high frequency is determined by the resistance of the top electrode [ 12 ]. The absence of changes in R s at high frequencies indicates the absence of electrochemical processes leading to a change in resistance in the top electrode. Figure 4 shows an HR X-TEM image of a SiO x -based memristive device after the thermal treatment. According to Figure 4 , the SiO x film has an amorphous structure with nanocrystals identified as Zr 3 O (area 1) and ZrO 2 (area 2) through a comparison of the interplanar spacings in the HR X-TEM image with the literature data performed earlier [ 8 ]. Such nanocrystals can form during thermal treatment due to a partial oxidation of the top electrode layer and can play a positive role in the localization of conductive filaments, e.g., as electric field concentrators [ 13 ]. Note that a link between the HR X-TEM results and the electrical characteristics can be obtained in more detail by using a combined approach, for example using the first-principle calculation and electron energy-loss spectroscopy. This was performed by the authors of [ 14 ], who showed that oxygen in Bi 2 S 3 nano-networks is a source of traps. Moreover, the authors note that continuously distributed traps are responsible for the gradual resistive switching (i.e., with a synaptic behavior), while the presence of discrete states leads to the abrupt resistive switching. In order to determine the mechanisms of electron transport in the memristive device, the electrical characteristics and resistive switching are considered in a wide temperature range. Figure 5 shows the I–V curves of the memristive device measured at 77 and 482 K. The I–V curves in the LRS obey Ohm’s law. In an HRS, these curves are nonlinear. The current in an LRS exceeds the current in an HRS by two orders of magnitude at 482 K and by four orders of magnitude at 77 K. Figure 6 and Figure 7 show the frequency dependencies of the equivalent circuit parameters of the memristive device after the thermal treatment obtained after measuring the I–V curves shown in Figure 5 . At 77 K, almost no effect of the filaments on the equivalent circuit parameters of the device in an HRS is found in the low-frequency range. In this range, there are almost no dielectric losses associated with the Ohmic leakage. It probably may originate from the mechanical strain arising during cooling down the stack. Similar phenomena and the thermal resistive switching have been previously found in HfO x -based memristive devices [ 15 ]. The value of R p shunting the stack in the HRS was ~3.5 × 10 6 Ω at f = 1 kHz. After switching to the LRS, the value of R p decreases down to ~200 Ω, whereas the Ohmic losses increase due to the restoration of filaments. The value of R s at a high frequency, which is determined by the resistance of the top electrode, almost does not change during switching. The equivalent circuit parameters of the memristive device in an LRS at 482 K almost do not differ from those measured at 77 K. However, the low-frequency dielectric losses remain due to the presence of incompletely broken filaments in an HRS. Figure 8 a shows the temperature dependencies of the current flowing through the device in an HRS measured in the temperature range from 300 to 560 K at different reading voltages. One can see that these dependencies have an activation character and can be fitted by straight lines in the lg I (1/ T ) scales in the temperature range from 300 to 530 K. Moreover, the values of activation energy E a determined from the slopes of the fitting lines decrease with increasing reading voltages from 0.14 eV to 0.11 eV. The linear extrapolation of the dependence E a ( V 1/2 ) to zero gives a value of 0.15 eV. The character of the above dependences in an HRS can indicate the Schottky electron transport mechanism or the Poole–Frenkel one [ 16 ]. However, low values of activation energy can be attributed to the presence of filaments shunting the functional insulator layer. The impact of filaments cannot be excluded completely in the temperature range investigated. In an LRS, the nonlinear character of conduction remains even at low resistances (12.5 Ω at the room temperature, curve 2 in Figure 8 b). The conductivity is characterized by a power-law temperature dependence of the current with a small exponent (<1) depending on the resistance of the sample and, therefore, on the mode of switching from an HRS to an LRS. Such a weak temperature dependence of the current points to the trap-assisted tunneling electron transport mechanism [ 17 , 18 ]. It should be stressed here that the electron transport in an LRS is confined predominantly within a thin filament with a diameter down to ~1 nm. Note also that the I–V curves of memristive devices can be fitted using the Fowler–Nordheim (FN) law in a certain voltage range marked with a red straight line in Figure 9 . The effective electron mass m* can be estimated from the slope of the respective part of the I–V curve in the Fowler–Nordheim coordinates as 1.6 m 0 ( m 0 is the free electron mass). This value is much higher than the effective electron mass in silicon oxide ( m* ≈ 0.4∙ m 0 ) [ 16 ]. On the other hand, the Fowler–Nordheim dependence I ( V ) also points to the trap-assisted tunneling as the predominant electron transport mechanism in the filament since the logarithm of electron tunneling probability between the traps should be proportional to 1/ V [ 19 ]. All the above points to trap-assisted tunneling as the predominant mechanism of electron transport through the filaments. One can note that the number of points to fit within the FN parts of the I–V curves is rather small. Nevertheless, we think that considering the FN fit of the respective parts can be justified due to the following reasons. First, the shapes of the I–V curves are quite typical for the cases when the field electron emission in the MIM stacks is observed (see e.g., [ 20 ]). Typically, the FN parts are observed at a higher bias, and the bias range, for which the FN-type I ( V ) dependence is observed, is narrow, inevitably because of a rapid (super-exponential) growth of I with increasing bias V in the FN regime. When studying the I–V curves of the MIM stacks, the upper limit for the further increase of the bias is set by the electrical breakdown of the insulator (unlike the situation, e.g., when studying the field electron emission in a vacuum when the upper limit for the bias is set by the self-ion emission phenomenon). In turn, it is necessary to note the importance of taking into account the change in electron transport during resistive switching in modeling the behavior of memristive devices, as was performed in Ref. [ 21 ], which presents a physical model describing resistive switching taking into account the trap-assisted transition between the Schottky emission and the Fowler–Nordheim tunneling, and successfully reproducing the memristive behaviors occurring on the interface between Bi 2 S 3 nano-networks and F-doped SnO 2 . The research carried out by the authors of [ 21 ] made it possible not only to reveal several features of the memristive interface including the distribution nature of the traps, the barrier height/thickness, and so on, but also to provide a foundation on which the real interfacial memristor can be quantitatively modeled. Quantitatively, the trap-assisted tunneling mechanism was confirmed by the C–V and G–V measurements. After switching, the extended nonlinear regions were observed on the C–V and G–V curves. From these data, the concentration of traps and the electron transport mechanism in the filaments can be evaluated. The C–V and G–V curves measured at 77 K for different frequencies of the small test signal f ss are shown in Figure 10 . The C–V and G–V curves measured at different values of f ss agree with each other qualitatively. Initially, C p decreases abruptly when applying a negative voltage while the G/ω values increase, and G/ω >> C p . This corresponds to the formation of filaments in the insulator and the switching to an LRS. At a further increase in the voltage up to V = 0, the conductivity of the filaments (owing to the electrons injected from the top electrode) decreases, and the capacitance increases. At V = 0 V, C p becomes equal to the capacitance of the insulator C D . At a further increase in the voltage ( V > 0), the conductivity increases again due to the injection of electrons from the TiN bottom electrode, while the capacitance decreases accordingly due to an increase in the Ohmic leakage, so G/ω >> C p again. The application of a voltage of +3–+4 V leads to the rupture of filaments and switching to an HRS. In an HRS, the value of C p does not depend on the voltage applied and is equal to C D whereas the value of G/ω is small, and G/ω << C p at a sufficiently high f ss ( Figure 10 b,c). During the backward voltage sweep from +5 V down to –5 V, there is no switching to an LRS. This can be explained by the charge accumulation inside the insulator, which screens the external electric field applied. Also, note that, at f ss = 1 MHz, the conductivity does not depend on voltage in the course of the reverse voltage sweep ( Figure 10 c). A decrease in f ss leads to a stronger dependence of G/ω on V ( Figure 10 a,b). This indicates the filaments not to be destroyed completely in an HRS. One can estimate the volume density of the charge carriers trapped during switching N from the area inside the hysteresis loop in the G/ω – V curve ( Figure 10 )\n (1) N = ∮ G ω d V q 1 S · d , \nwhere q is the electron charge, S is the area of the top electrode, and d is the thickness of the insulator. Its density depends on f ss and is found to be (7.6 ± 0.6) × 10 20 , (5.2 ± 0.5) × 10 19 , and (1.2 ± 0.2) × 10 19 cm −3 for f ss = 10, 100, and 1000 kHz, respectively. From the C–V curves, one can estimate the effective density of electron traps N t occupied by the charge carriers during switching to an LRS: N t = C D ∙Δ V / q ∙ S ∙ d , (2) \nwhere ∆ V is the parameter shown in Figure 10 , which equals 4–5 V regardless of the value of f ss . The obtained value of N t ≈ 8× 10 18 cm −3 is close to the concentration of traps in silicon oxide originating from the Si–Si bonds [ 22 ]. At the same time, these traps are nonuniformly distributed in the oxide film with a much higher local concentration in the filaments formed at the places where the electric fields are concentrated. It should be noted that the measurements of a static capacitance at temperatures up to 600 K do not reveal any increase in capacitance due to the ion migration polarization. So far, one can conclude that the electron transport mechanism along the filaments in the memristive devices studied in the present work is the electron tunneling through the defects built inside the filaments. This is confirmed by the dependences of conductivity on the reciprocal voltage in an LRS, which can be fitted by straight lines in the lg G/ ω–1/ V scales ( Figure 11 ). The total density of polarizable charges in the memristive device and their activation energies were determined from the measurements of the depolarization current I D in an HRS [ 23 ]. An example of a depolarization curve is shown in Figure 12 . The data presented are obtained after the polarization of the stack at a constant voltage of +1 V and the temperature of 550 K followed by heating from 170 up to 550 K and cooling down to 430 K at zero voltage. These data allow for determining the surface density of depolarized charges as an integral over the area under the depolarization current curve [ 23 ]. The value of the surface density of the depolarized charges of 5 × 10 14 cm −2 is obtained, which corresponds to ~4 × 10 20 cm −3 in the bulk of the insulator. Several parts of the depolarization current curves can be fitted by the straight lines in Arrhenius scales that allows for determining the activation energies of carrier traps. The values of E a of 0.33 ± 0.01, 0.38 ± 0.01, and 0.69 ± 0.08 eV are obtained. The value of E a = 0.69 ± 0.08 eV can be ascribed to the electron trap associated with the Si–Si bond [ 24 ]. The depolarization current curve of the memristive device in an LRS is shown in Figure 13 . Two values of the activation energy of the traps are found in an LRS. These values are 0.53 ± 0.09 and 1.61 ± 0.14 eV. The latter value is close to the energy of thermal ionization from the Si–Si amphoteric trap (1.5–1.7 eV [ 22 , 25 ]). The traps with smaller activation energies can be ascribed to the O 3 = Si–Si = O 3 species. Let us now discuss how the obtained data correlate with the mechanism of the formation and destruction of conductive filaments in similar devices with a relatively thicker SiO x layer obtained by magnetron sputtering [ 26 ]. That mechanism was based on the migration of oxygen ions (O 2− ) leaving the oxygen vacancies (Si-Si bonds) that form the filament in SiO x film, the accumulation of oxygen ions in the bottom TiN electrode, and their return to the filament region with partial oxidation of the latter. It is obvious that all these processes should also take place in the devices under study; however, the small thickness and defect composition of the film obtained by the low-temperature method determine the specifics of these processes even in the as-deposited state, which is initially conductive. The structure of filaments can also change dramatically upon an additional thermal treatment, which manifests itself in the observed regularities of electron transport. Although the general dependencies, consisting of the semiconducting-like temperature dependence of conductivity in a low-resistance state and the activation nature of current in a high-resistance state, are reproduced, these studies allow us to look deeper inside the nature of electronic processes in Si-based filaments, which are reflected in the electrical characteristics of new CMOS-compatible memristive devices."
} | 5,873 |
21283583 | PMC3026804 | pmc | 223 | {
"abstract": "The architecture and properties of many complex networks play a significant role in the functioning of the systems they describe. Recently, complex network theory has been applied to ecological entities, like food webs or mutualistic plant-animal interactions. Unfortunately, we still lack an accurate view of the relationship between the architecture and functioning of ecological networks. In this study we explore this link by building individual-based pollination networks from eight Erysimum mediohispanicum (Brassicaceae) populations. In these individual-based networks, each individual plant in a population was considered a node, and was connected by means of undirected links to conspecifics sharing pollinators. The architecture of these unipartite networks was described by means of nestedness, connectivity and transitivity. Network functioning was estimated by quantifying the performance of the population described by each network as the number of per-capita juvenile plants produced per population. We found a consistent relationship between the topology of the networks and their functioning, since variation across populations in the average per-capita production of juvenile plants was positively and significantly related with network nestedness, connectivity and clustering. Subtle changes in the composition of diverse pollinator assemblages can drive major consequences for plant population performance and local persistence through modifications in the structure of the inter-plant pollination networks.",
"conclusion": "Conclusion and Perspectives In summary, the diversity and specific composition of the local pollinator assemblages had a significant effect on E. mediohispanicum population performance through shaping geographic variation in the architecture and topology of the intraspecific pollination networks. Nevertheless, there are some important caveats for an accurate view of the functional values of individual networks. First, since nodes in these networks are individual plants, their spatial location has to be included in the analyses. Positional effects can be important in determining specific patterns of interaction with certain pollinator groups. Second, while presence-absence of interactions can provide a broad view of interaction patterns, more robust estimates can be obtained with quantified observations, despite the increased sampling effort needed. Third, the pattern of shared pollinators revealed by our analysis is just a proxy for inferring potential mating events among the linked individuals. More detailed analyses using genetic markers to infer actual mating events in the population could provide a most interesting supplementary view by linking pollinator sharing patterns with actual mating events. Despite these potential limitations, our study revealed consistent trends across distinct populations unequivocally linking network complexity and local population performance. Our study adds a new dimension to the definition of pollinator effectiveness, and suggests that pollinators may be effective not only by enhancing individual seed production but also by generating thorough pollen dissemination at the population level via their influences on mating events among individual plants. Plant populations mostly visited by effective pollinators would produce more offspring than populations visited by low effective pollinators, an effect ultimately related to how individual plants build up complex networks of interaction with their mutualists. The fact that the pollinator fauna varies geographically in this system gives rise to geographic mosaics of network patterns, with distinct functional signatures on plant performance through fitness effects, as revealed by our results. Our results highlight how subtle changes in the composition of diverse pollinator assemblages can drive major consequences for plant population performance and local persistence through modifications in the structure of the inter-plant pollination network.",
"introduction": "Introduction The architecture and properties of many complex networks, such as molecular and metabolic [1] , [2] , neuronal [3] , genetic [4] , [5] , social [6] , and transportation networks [7] , [8] , play a significant role in the functioning of the systems they describe. For instance, the fixation of single nucleotide mutations, and gene duplications and deletions, are influenced by the architecture of the metabolic network in Saccharomyces cerevisiae \n [1] , where genes encoding enzymes with high connectivity and high metabolic flux have higher chances to retain duplicates in yeast evolution. Similary, the coupling between tie strengths and network topology has important consequences for the global stability of networks generated by mobile phone calls [6] . In these networks, properties like functional robustness, optimal transport, or minimal energy cost directly emerge from the network topology. Recently, complex network theory has been applied to ecological entities, like food webs or mutualistic plant-animal interactions [9] . Ecological networks have a well-defined architecture, since they are more nested than expected by random models [10] , [11] , [12] , they have a higher density of links, a shorter distance between species, and species are more clustered [13] , [14] , and thus have strong small-world properties [14] . Previous studies have shown that ecological networks are robust to random losses of species, but probably very sensitive to the loss of key mutualists [15] , to extinctions of phylogenetically-related species [16] , or to invasions by successful exotic species [17] , [18] . Unfortunately, despite the fast-paced information gain on ecological network structure, we still lack an unambiguous link between structural properties of these complex interaction networks and their functional consequences for the dynamics of the system. In contrast to communities (i.e., assemblages of co-occurring species), populations are groups of individuals interconnected by functions like mating, reproduction, social interactions, sharing of mutualistics, common defense against predators, etc. More importantly, at the network level good estimates of the system functioning, like population performance, population dynamics, demography, recruitment, genetic diversity, etc, may be obtained. However, network theory has been seldomly applied to the study of individual interactions within populations [19] , [20] , [21] , and hence there is not yet a clear view of the relationship between the network architecture and the functioning of individuals within populations. In animal-pollinated plants, the pattern of shared pollinators among individual plants in a population to some extent may be translated into a pattern of mating [21] . If most individuals share the same fauna of pollinators, the resulting pattern of pollen transfer within the population would be less structured than if subsets of plants share distinct groups of pollinators. In this latter scenario there are ample possibilities for assortative mating even in systems with generalist pollination systems. The existence of different levels of structure in the mating and pollination networks may drive significant variation across populations in fitness effects, demography, local genetic structure and gene flow. In this study, we use complex network theory to study the interaction between a plant species and its remarkably diverse assemblage of pollinators to show that some properties of the individual-based mutualistic networks (the interactions among individual plants based on the pattern of pollinator sharing) can pervasively determine the performance of the whole system. We empirically derived the interaction networks of the herb Erysimum mediohispanicum (Brassicaceae) individual plants and their pollinators in eight different populations. Afterwards, we checked whether the architecture of the networks was related to main pollinator assemblage descriptors (abundance, diversity and identity). Finally, we tested whether the network architecture did affect the performance of the plant populations, quantified using a very inclusive estimate, number of juveniles recruited per population.",
"discussion": "Results and Discussion Architecture of individual-based pollination networks We built up intraspecific, individual-based plant networks in eight well-studied plant populations using data on pollinator visitation collected during 2005 ( Table S1 ). We applied network theory tools to the study of these networks, to describe their structural properties [22] . Each individual plant in a population was considered a node, and was connected by means of undirected links to conspecifics sharing pollinators ( Fig. 1 ); i.e., a population-level networks describing how individual plants share pollinators, estimated from the unipartite projections of the bipartite plant-pollinator, two-mode networks ( 1 ). In this way, the graph representations not only describe the potential mating events among individuals in each population, but also the specific pollinator species involved in potential pollen transfer. Network architecture was described by means of nestedness, connectivity and transitivity. Collectively, these three measures capture the topology of network architecture. Network nestedness is a measure of the order of the whole system, and quantifies whether the species composition of small assemblages is a proper subset of the species composition of large assemblages. It was calculated using by means of NODF, a nestedness measure based on overlap and decreasing fills [23] . The connectivity of the network measures how individuals are connected to one another through the network [24] and it was described using two parameters: Normalized degree and connectance [22] . Transitivity is a network property that determines the easiness of spread of any factor across the network [24] (e.g., mating events through pollen transfer). Transitivity was estimated by means of the CC1 clustering coefficient, the fraction of connected neighbours around a given individual, an index that measures the local group cohesiveness [25] . All these parameters therefore bear biological meaning in terms of mating and reproductive events in the plant population. 10.1371/journal.pone.0016143.g001 Figure 1 The topology of individual-based ecological networks. Unipartite networks depicting the pattern of shared pollinators by individual plants (nodes) in each population studied (Em01 to Em25). The links among nodes depict the pattern of shared pollinator species (described in Fig. S1 ); i.e., two nodes are linked whenever they share a pollinator species. The network representation (layout) was generated with the Kamada-Kawai energy-minimization algorithm [29] . Each node represents an individual plant. In green, the most connected plants ( = hubs) in each population. The size of the node refers to the overall flower number displayed by that individual. Some individual plants in each population attracted an outstandingly high number of pollinator species, whereas most plants attract a moderate-to-low diversity of pollinators. Consequently, the nestedness values of our intraspecific networks, calculated as NODF, were significant for all plant populations (P<0.0001 in all cases, Table 1 ). Relative nestedness ranged between 0.065 and 0.137 for Temperature and between 0.015 and 0.898 for NODF ( Table 1 ). Nestedness values were very similar to those found in most interspecific plant-pollinator networks studied to date [10] , [12] . Nested networks are organized around a cohesive core of nodes where generalist plant species interact with generalist animal species [10] . High nestedness indicates the occurrence of asymmetric specialization, where the most specialist species interact with the most generalist ones. This is a property that makes the whole system more resistant to the loss of some particular interactions, and favours the persistence of rare, specialist species [10] . The high nestedness of the observed assemblages would consequently explain the high diversity of the E. mediohispanicum pollinator fauna, and a highly structured interaction pattern centered on distinct subsets of individual plants in each population that attracted a diverse pollinator assemblage. This may contribute to reduce intraspecific competition and enhance the number of coexisting pollinator species visiting the population [26] . The biological meaning of nestedness in our individual-based networks refers to how pollinator sharing patterns influence mating events among plants. Nestedness implies that individual plants visited by a restricted subset of pollinator species actually share these species with individuals visited by a higher diversity of pollinators. This potentially allows for a more thorough pattern of mating events, with individuals visited by a restricted number of pollinator species not necessarily being reproductively disconnected from the rest of conspecifics. This could increase the probability of successful reproduction and increased population average fitness. 10.1371/journal.pone.0016143.t001 Table 1 Among-populations differences in network topology. Populations N plants N pollinators Nestedness Relative Nestedness Normalized Degree Connectance Clustering Em01 63 36 13.08 * \n 0.516 0.225±0.019 * \n 0.221 * \n 0.750 ns Em02 69 41 15.63 * \n 0.397 0.366±0.024 * \n 0.360 * \n 0.753 * \n Em08 70 32 16.05 * \n 0.624 0.276±0.020 * \n 0.272 * \n 0.760 * \n Em21 80 37 22.10 * \n 0.898 0.379±0.024 * \n 0.374 * \n 0.804 * \n Em22 58 32 11.53 * \n 0.307 0.207±0.0.24 * \n 0.204 * \n 0.703 * \n Em23 63 39 11.84 * \n 0.294 0.222±0.021 * \n 0.219 * \n 0.773 * \n Em24 47 30 8.90 * \n 0.115 0.159±0.018 * \n 0.156 * \n 0.650 * \n Em25 52 32 8.27 * \n 0.015 0.159±0.016 * \n 0.156 * \n 0.697 * \n Connectance is the number of lines in a simple network, expressed as a proportion of the maximum possible number of lines. Degree is the average number of lines incident with a given node. All network metrics were compared with random-generated networks (see Online Full Methods ). *p<0.0001. The plants belonging to the same populations were tightly connected among them by sharing many flower visitors. Consequently, our networks exhibited higher connectivity values than expected randomly (P<0.0001 in all cases, Table 1 ). Network degree, which indicates the average percentage of conspecific plants to which a given plant is connected to, ranged between 15.9% (Em24 and Em25) and 36.7% (Em21). Network connectance, which indicates the proportion of potential inter-plant links that actually occur, ranged between 0.156 (Em24 and Em25) and 0.374 (Em21). These values suggest that the proportion of mating links among plants that were effectively realized was high, up to 37% ( Fig. 1 ). Therefore, individual populations showed ample variation in the degree of pollinator sharing among individual plants, with plants in some populations exhibiting a high overlap of pollinator visitors. Since pollinators mediate the mating system of the individuals by pollen transfer, this means that there exists ample variation among populations in the potential for pollen flow through the population and, presumably, the sizes of individual genetic neighbourhoods. Despite the high pollinator connection amongst plants from the same population, we found that plants may group in distinct subsets of individuals sharing more similar pollinators. In fact, network clustering was higher than expected by random (P<0.0001, Table 1 ), ranging from 0.650 (Em24) to 0.804 (Em21) ( Table 1 ). This pattern suggests a high structuring of the individual interaction with pollinators in each population. Variation in this network property indicates ample variation among populations in the potential for assortative mating- i.e., clusters of individual plants that tend to mate among themselves more frequently. Effects of pollinator characteristics on network architecture \n E. mediohispanicum is a very generalist plant visited by over 180 insect species belonging to six orders and as much as nine functional groups (Gómez et al 2007). The structure of the E. mediohispanicum pollination networks depended on the abundance, diversity, identity and type of insects visiting the flowers in each locality. Thus, local pollinator richness was significantly and positively associated with network connectivity and clustering (P<0.05, spatial autoregressive models, Table 2 ), but not with nestedness (P>0.1, Table 2 ). E. mediohispanicum populations with highly diversified pollinator assemblages (Em02, Em21; Fig. 1 ) characteristically showed more structured patterns of interaction, with a well-defined core of individual plants interacting with a diverse pollinator assemblage that included some generalist insects visiting the subset of more specialized plants. Populations with depauperated pollinator faunas (Em22, Em24, Em25; Fig. 1 ) lacked a distinct core of generalist plants and showed a less structured network pattern. Local pollinator abundance was also related positively to connectivity ( Table 2 ). These results suggest that an increase in pollinator abundance and, especially, diversity entail a stronger linking among co-occurring plants and a more even mating pattern across the population. In this way we can examine to what extent variation across populations in mean performance relates to local changes in the composition of the pollinator assemblage that translate into variations in the structure of the interaction networks. 10.1371/journal.pone.0016143.t002 Table 2 Correlates of pollinator diversity on network topology across the eight E. mediohispanicum populations. Abundance S obs \n Hurlbert's PIE Bray-Curtis Morisita-Horn Nestedness 0.01±0.01 −0.00±0.03 −1.64±0.82 ms −0.011 −0.049 Degree 0.18±0.05 *** \n 0.09±0.03 * \n −11.67±18.43 0.400 * \n 0.342 * \n Connectance 0.032±0.01 *** \n 0.02±0.01 * \n −1.61±3.14 0.298 0.332 Clustering 0.11±0.05 ms 0.01±0.002 ** \n −0.83±1.86 0.179 0.235 Figures show coefficients ±1 standard error obtained from spatially-explicit models (for pollinator abundance, S obs and Hurlbert's PIE indices) and partial mantels (method = spearman), controlling for geographic distance (for Bray-Curtis and Morisita-Horn dissimilitude indices). P-values obtained with 1000 permutations: ms = marginally significant, *p<0.05, **p<0.01, ***p<0.001. The type of most abundant pollinator in a given plant population determined the connectivity of the local networks ( Table 2 ). Whereas populations with many beeflies had high connectivity (5.319±2.174, t = 2.45, P = 0.050, r 2 = 0.50), populations with many beetles and hoverflies had low connectivity (−12.016±4.066, t = 2.96, P = 0.050 for beetles; −3.506±1.272, t = 2.76, P = 0.004 for hoverflies). That is, plants in populations where beeflies were abundant were more intensely connected among them, whereas populations with many beetles and hoverflies showed a more defined subdivision of plant individuals in groups with distinct pollinator fauna ( Fig. 2a ). This is a consequence of the contrasting foraging behaviour displayed by each pollinator type. Beeflies move indiscriminately across the complete set of plants of a given population, whereas the other pollinator groups show a foraging behavior with a high proportion of local movements among close individual plants (i.e., hoverflies) or with very few movements among plants and most movements among flowers of the same plants (i.e., beetles). In fact, both the functional specialization, an estimate of the average topological distance between two given nodes produced by any agent in a network [27] , as well as the hub degree, a metric that quantifies the ability of specific agents to connect distant nodes across the network [28] , were significantly higher in beeflies than in the other flower visitors ( Fig. 2b ). These insects have a more central role in the network, potentially mediate pollen flow among a larger number of individual plants when compared to the other pollinator types, and cause more opportunity for mating diversity ( Fig. 2b ). All these findings indicate that the pattern of pollinator-mediated connections (potential mating events) among co-occurring plants of a given population, and its resulting network architecture, is strongly determined by the type of pollinators visiting the flowers in that population. 10.1371/journal.pone.0016143.g002 Figure 2 Pollinator effects on network topology. a) Expected changes in network connectivity due to different functional groups of pollinators. Network graphs are depicted for three example populations, illustrating the relationships among individual plants when all pollinator species are included (left) and when only specific subsets are considered (right). Note how the “hub” plants spread all over the partial networks. b) Differences among pollinator type (from left to right, beeflies, beetles, hoverflies, bees and butterflies) in hub degree (F 4,16 = 21.83, P = 0.0001) and functional specialization (F 4,16 = 4.13, P = 0.036). Relationship between plant population performance and network architecture The topology of our networks had dramatic consequences for the performance of the populations. Variation across populations in the average per-capita production of juvenile plants was positively and significantly related with network nestedness, connectivity and clustering ( Fig. 3 ). Since these analyses were performed after controlling for local pollinator abundance and diversity (see Methods ), we can conclude that network architecture itself had a direct effect on population performance independently on any pollinator-mediated effect. In addition, network architecture did not only affect overall plant population performance, but also had significant effects on most intermediate fitness components ( Fig. S2 – S3 ). This striking outcome strongly suggests that geographic variation in the local structure of individual plant-pollination networks has a pervasive influence in the outcome of the mutualistic interactions of E. mediohispanicum plants in terms of population-level reproductive success. As far as we know, this is one of the first evidences of functional signals of the structural patterns of ecological networks: population variation in plant performance was unequivocally associated with variation in the way that interactions with pollinators were organized. Specifically, our data suggest that high values of nestedness, connectivity and clustering of interactions with mutualistic pollinators are beneficial to individual plants coexisting in local populations. Since we worked with intraspecific networks, where individuals of one plant species interact with each other through many shared pollinator species, our results indicate that the number of successfully recruited new adults in these populations into the next generation will depend on the way individual plants share pollinators locally during the current generation. Therefore, asymmetrical specialization and the existence of a core of supergeneralist plant individuals visited by a very diverse pollinator fauna composed of both generalist and specialist pollinators seems to be beneficial in terms of population seed production, presumably through a more thorough pattern of mating events serviced by a highly structured pollinator assemblage. In addition, our results also indicate that individual plants locally embedded in more connected networks are those producing more seeds. Highly connected networks are those composed of plants more intensely linked through pollinator sharing. Consequently, enhanced pollinator-mediated connectivity in our intraspecific networks probably results in a high frequency of mating events amongst all members of the population, increased gene flow across the entire population, and a reduction of local-scale population genetic structure. These results suggest that nested and highly connected local pollinator assemblages might result in highly structured networks of mating among individual plants, with potential lasting consequences for patterns of gene flow and genetic structure. 10.1371/journal.pone.0016143.g003 Figure 3 Relationship between network architecture and function. The complex networks of pollinator-mediated interactions among individual plants (e.g., mating events) benefit E. mediohispanicum population performance. Populations organized around a core of highly interactive plants (high nestedness) with individuals tightly connected through shared pollinators (high connectivity) within distinct groups exhibiting similar pattern of interactions (high clustering) have high performance (number of juveniles produced per plant). We have previously shown that the diversity of pollinators may affect the performance of the plant populations [29] . However, local pollinator diversity is a variable describing the whole system without explicitly taking into account the interactions of the elements of the system (ie, the individual plants). In the present study, we have found that these interactions are also important to determine the function of the system. This is because a high level of local pollinator diversity may happen when different plants are visited by highly contrasting insect assemblages or, alternatively, when all plants are visited by the same very diverse pollinator assemblage. Using a network approach may help to differentiate between both possibilities because it provides critical additional information on the system. Conclusion and Perspectives In summary, the diversity and specific composition of the local pollinator assemblages had a significant effect on E. mediohispanicum population performance through shaping geographic variation in the architecture and topology of the intraspecific pollination networks. Nevertheless, there are some important caveats for an accurate view of the functional values of individual networks. First, since nodes in these networks are individual plants, their spatial location has to be included in the analyses. Positional effects can be important in determining specific patterns of interaction with certain pollinator groups. Second, while presence-absence of interactions can provide a broad view of interaction patterns, more robust estimates can be obtained with quantified observations, despite the increased sampling effort needed. Third, the pattern of shared pollinators revealed by our analysis is just a proxy for inferring potential mating events among the linked individuals. More detailed analyses using genetic markers to infer actual mating events in the population could provide a most interesting supplementary view by linking pollinator sharing patterns with actual mating events. Despite these potential limitations, our study revealed consistent trends across distinct populations unequivocally linking network complexity and local population performance. Our study adds a new dimension to the definition of pollinator effectiveness, and suggests that pollinators may be effective not only by enhancing individual seed production but also by generating thorough pollen dissemination at the population level via their influences on mating events among individual plants. Plant populations mostly visited by effective pollinators would produce more offspring than populations visited by low effective pollinators, an effect ultimately related to how individual plants build up complex networks of interaction with their mutualists. The fact that the pollinator fauna varies geographically in this system gives rise to geographic mosaics of network patterns, with distinct functional signatures on plant performance through fitness effects, as revealed by our results. Our results highlight how subtle changes in the composition of diverse pollinator assemblages can drive major consequences for plant population performance and local persistence through modifications in the structure of the inter-plant pollination network."
} | 7,022 |
37666802 | PMC10477309 | pmc | 224 | {
"abstract": "Anaerobic digestion of municipal mixed sludge produces methane that can be converted into renewable natural gas. To improve economics of this microbial mediated process, metabolic interactions catalyzing biomass conversion to energy need to be identified. Here, we present a two-year time series associating microbial metabolism and physicochemistry in a full-scale wastewater treatment plant. By creating a co-occurrence network with thousands of time-resolved microbial populations from over 100 samples spanning four operating configurations, known and novel microbial consortia with potential to drive methane production were identified. Interactions between these populations were further resolved in relation to specific process configurations by mapping metagenome assembled genomes and cognate gene expression data onto the network. Prominent interactions included transcriptionally active Methanolinea methanogens and syntrophic benzoate oxidizing Syntrophorhabdus , as well as a Methanoregulaceae population and putative syntrophic acetate oxidizing bacteria affiliated with Bateroidetes (Tenuifilaceae) expressing the glycine cleavage bypass of the Wood–Ljungdahl pathway.",
"introduction": "Introduction Renewable natural gas (RNG), primarily composed of biogenic methane (CH 4 ) and carbon dioxide is an important non-fossil energy resource useful in the transition to a low carbon future 1 . Despite widespread adoption of technologies such as anaerobic digestion (AD) to produce biogenic CH 4 , industrial-scale AD converting organic waste (e.g., municipal black and gray water) to RNG tends to experience operational challenges, including (i) variable RNG yields, (ii) lower production efficiencies than theoretical values, and (iii) substantial amounts of solid residues that can be costly to dispose 2 . These challenges can confound the economics of RNG production and arise in part from a prevailing “black box” paradigm that does not fully consider the microbial communities, also known as microbiomes, driving AD conversion processes 3 . Discovering design principles that shape the network properties of AD microbiomes offers a new paradigm for optimizing RNG production and waste resource recovery. Over the past decade, high-throughput sequencing approaches have been used to describe microbial community structure, function and dynamics associated with AD at different operating scales 4 – 9 , including recent efforts to characterize the global microbiome of wastewater ADs 10 . Importantly, these studies indicate that the AD milieu supports regionally distinct microbial communities containing heterogeneous constituents engaged in conserved metabolic interactions driving organic waste conversion to CH 4 11 . In addition to microbial dynamics, considerable evidence also indicates that industrial-scale AD performance is strongly influenced by a variety of process parameters including substrate chemistry, loading rates, retention time, temperature, and metal micronutrient (e.g., Fe, Ni, Mo, Mn) concentration 12 – 15 . Taken together, an emerging scientific consensus 16 – 18 suggests that a new paradigm for understanding and improving AD systems should involve constructing microbial interaction networks in relation to physical and chemical parameter information under time-resolved conditions at relevant operating scales. This is particularly needed for industrial-scale operations which are difficult to access and study over unit time 4 , 8 , 18 – 25 . Here, we present a two-year time series study of the Lulu Island municipal wastewater treatment plant (WWTP) operated by Metro Vancouver in Richmond, British Columbia, Canada. Our study is intended to facilitate further understanding of microbial community structure, function, and dynamics in relation to industrial-scale RNG production. The Lulu Island WWTP uses standard practices and produces >5000 m 3 of RNG per day from mixed sludge as a starting material. Replicated mixed sludge samples were collected biweekly from Lulu ADs and archived for DNA and RNA extraction. Small subunit ribosomal RNA (SSU or 16S rRNA) gene amplicons were generated from 116 replicated samples, including two periods of standard AD operation, a period of AD operation with an additional allochthonous waste stream, and a period that used serial (instead of parallel) AD operation. The resulting amplicon sequence variant (ASV) data was used to identify a core set of microorganisms across the time series, construct a co-occurrence network to generate statistically informed hypotheses related to potential syntrophic interactions, and, in combination with process parameter information, identify indicator microorganisms associated with different process configurations and relevant WWTP conditions such as final nitrate levels, volatile solids destroyed, total RNG produced, and RNG methane content. A subset of time series samples representing each process configuration was also selected for metagenomic whole genome shotgun and metatranscriptomic sequencing to produce metagenome-assembled genomes (MAGs) and estimate expression levels based on transcript read mapping. MAGs were associated with cognate ASV nodes in the co-occurrence network. Ultimately, these multi-omic datasets and statistical approaches offered potential mechanistic explanations (through encoded and expressed functions) for metabolic interactions between co-occurring microorganisms in the AD, providing insight into the metabolic network driving RNG production in the Lulu Island WWTP.",
"discussion": "Results and discussion Lulu Island waste resource recovery ecosystem The Lulu Island WWTP operated by Metro Vancouver in Richmond, British Columbia, Canada (Longitude: −123.14498° or 123° 8’ 42” W, Latitude: 49.11491° or 49° 6’ 54” N) provides primary and secondary treatment of >30 billion liters of mixed-sourced wastewater from ~200,000 residents each year. Primary treatment includes a series of tanks where wastewater undergoes screening, aeration, mechanical separation, settling, and clarification. This effluent then enters a secondary treatment stream, where it is pumped through a trickling filter, solids-contact tank, secondary clarifier, and disinfection tank before being released into receiving water. A portion of sludge from the primary and secondary treatment process streams is mixed and thickened, then split equally into two mesophilic (38 °C) anaerobic digesters (ADs) manifesting a 30-day retention time. The ADs are typically operated in parallel with identical mixed waste inputs. Methane and other gases from the ADs are scrubbed to renewable natural gas (RNG) which is either used to generate heat for Lulu Island WWTP operations or sold to FortisBC, the local distributor of natural gas. Metro Vancouver provided physicochemical parameters and triplicate waste secondary sludge (WSS) samples from solid contact tanks and overflow digestate samples from anaerobic digesters (AD1 and AD2) on a biweekly basis between October 13, 2016 and December 12, 2018. During this time interval, ADs experienced four different process configurations (Standard Operation I, standard operation with chemically enhanced primary treatment sludge (CEPT Operation), Standard Operation II, and Serial Operation; Fig. 1A ) which differed both in upstream primary and secondary wastewater treatment steps and in the flow of material into the ADs. Each configuration represented experimental perturbations of municipal operations to determine how different process parameters contribute to RNG production, volatile solids conversion, influent, and effluent sludge characteristics (e.g., nutrient and metal composition, chemical oxygen demand) in relation to microbial community structure, function, and dynamics. This study is focused on data collected and analyzed from one of the ADs (AD1), which was sampled continuously throughout the two-year time series. Fig. 1 Summaries of the operating conditions and physicochemical conditions of the digester across the time series. A Wire diagram of the four process configurations across the two-year time series. SOI Standard Operation I, CEPT Chemically Enhanced Primary Treatment Operation, SOII Standard Operation II, SER Serial Operation, MS mixed sludge waste from primary and secondary treatment), CH 4 (biogas), BS biosolids, Labels “1” and “2” indicate the two anaerobic digesters (all samples for this study were taken from AD1). B Temporal physicochemical sparklines of AD performance. Methane % volumetric percent of biogas as methane, HRT hydraulic retention time, OLR organic loading rate, Biogas (total volumetric biogas generated), VFAs volatile fatty acids. Box plots of configuration groups represent 75 th , 50 th (median), and 25 th percentiles, with whiskers representing 90 th percentiles and outlier position as filled points above whiskers ( n = 16 for SOI, 10 for CEPT, 11 for SO II, and 6 for SER). Physicochemical parameter information Physicochemical parameter information was collected throughout the time series to evaluate both mixed sludge characteristics entering the AD as well as organic waste conversion processes e.g., solids destruction, VFA concentrations, RNG production (Supplementary Data 1 ). In pairwise comparisons, physicochemical profiles from the two Standard Operations (I and II) were indistinguishable (Supplementary Data 4 ). However, samples from the intervening CEPT Operation differed significantly from the two Standard configuration modes based on a sharp increase in organic loading (kg/m 3 ), which ultimately led to a temporary increase in total biogas production approaching 5739 m 3 (Fig. 1B ). The final Serial Operation was the most distinct configuration of the four because of a fundamentally different flow regime, initiated by a rapid shift to in-line (serial) rather than equivalent flow of mixed sludge split between the two ADs (Fig. 1A ). This caused the inflow rate (ML/day) into AD1 to increase sharply at the beginning of Serial Operation, along with a concomitant decrease in hydraulic retention time (days). During this interval, digestate characteristics such as volatile fatty acid concentration (mg/L), final ammonia in the effluent (mg/L), and percent CH 4 in RNG all increased. While the CH 4 to carbon dioxide (CO 2 ) ratio of RNG increased during the Serial configuration (indicating higher quality), the total volume of RNG decreased, suggesting a possible tradeoff in RNG quantity vs. quality which was then explored further from a microbiological perspective. Microbial community structure Given the observed impact of process configuration on Lulu Island AD RNG production, we explored the sample archive to identify relationships between physicochemical parameters, such as VFA concentrations and RNG production, and microbial community structure. Genomic DNA extracted from 43 replicated AD1 samples spanning the time series was used to generate amplicons targeting the V4 region of the bacterial and archaeal (prokaryotic) 16S rRNA gene. Resulting data sets had an average of 23,833 quality-filtered 250-bp paired-end reads per sample resolving 928 unique ASVs (Supplementary Data 5 ). Consistent with physicochemical parameter information, microbial community structure also differed across samples in relation to process configuration (Supplementary Data 6 ). Interestingly, while the conditions in Standard Operation II returned to the same state as in Standard Operation I (after the intervening CEPT Operation), the microbial community did not return to its previous structure, indicating formation of a new stable state. The majority of unique ASVs (92%) had >80% sequence identity to cognate sequences in the global Microbes in Wastewater Treatment Systems and Anaerobic Digesters (MiDAS) 16S rRNA gene database 10 , indicating that a phylogenetically and globally conserved set of microbial lineages are adapted to driving hydrolysis, fermentation and methanogenesis in the AD milieu (Fig. 2 ). Fig. 2 A Sankey diagram of microbial community structure shown as relative abundances of family-level taxa grouped into their respective domains and classes. The width of ribbons represents the cumulative relative abundances of all ASVs within each taxonomic lineage. The final column assigns ASVs to categories of percent identity of their V4 16S sequences to the MiDAS database. The two-letter codes on the plot represent taxonomic names. Domain : Ba Bacteria, Ar Archaea. Class : Cl Cloacimonadia, Ba Bacteroidia, De Deltaproteobacteria, Ot Other, Cs Clostridia, An Anaerolineae, Sp Spirochaetia, Ve Verrucomicrobiae, Be Betaproteobacteria, Mm Methanomicrobia, Th Thermococci, Mb Methanobacteria, Family : Cl Cloacimonadaceae, Sy Syntrophaceae, Ot Other, An Anaerolineaceae, Sp Spirochaetaceae, Ri Rikenellaceae, Le Lentimicrobiaceae, Ba Bacteroidetes (vadinHA17), Ru Ruminococcaceae, Pr Prolixibacteraceae, Pe Pedosphaeraceae, Bu Burkholderiaceae, Mf Methanofastidiosaceae, Ma Methanosaetaceae, Mp Methanospirillaceae, Mr Methanoregulaceae, Mm Methanomicrobiaceae, Mb Methanobacteriaceae. Despite high recall of ASVs to the MiDAS database, the structure of the Lulu Island AD community differed in several important ways from previously described datasets from mesophilic ADs (Fig. S1 ) 23 , 26 – 28 . Although Bacteroidetes, Firmicutes, and Proteobacteria were common and conserved community members, the most abundant taxonomic group in Lulu Island samples was Cloacimonetes (candidate phylum WWE1), with a cumulative relative abundance exceeding 25% in most samples across the time-series (Figs. S1 , S2 ). This phylum is typically found in low abundance in mesophilic ADs treating wastewater, although it has been observed to dominate bioenergy facilities processing crop residues, and some mesophilic WWTPs (typically with long retention times or high organic loading rates), where it likely plays a role in amino acid fermentation and syntrophic propionate oxidation 8 , 29 . In addition to Cloacimonetes, ASVs associated with the candidate phylum Marinimicrobia were also relatively abundant, reaching up to 5.4% in some samples. Although prevalent and active in marine ecosystems under low oxygen conditions 30 – 32 , Marinimicrobia are often observed in ADs where they may play a role in hydrogen production and nitrogen mineralization 33 , 34 . Additional candidate groups, including Atribacteria, Kiritimatiellaeota, and Hydrogenedentes, were also identified in Lulu Island samples at relative abundances approaching 1%, suggesting that these candidate groups are important AD community members whose metabolic roles are conserved across time 8 , 35 . Amplicon sequence variants affiliated with Methanogenic archaea had low relative abundances across the time series and included the families Methanofastidiosaceae (0.9%), Methanosaetaceae (0.5%), Methanospirillaceae (0.3%), Methanoregulaceae (0.2%), Methanomicrobiaceae (0.1%), and Methanobacteriaceae (0.1%). Methanofastidiosaceae, the most abundant methanogenic lineage in Lulu Island AD samples, is a candidate group that lacks several canonical methanogenesis pathway components, likely using methylated thiol compounds for energy and CH 4 production 33 . Methanosaetaceae, primarily represented by Methanothrix populations in Lulu Island samples, convert acetate to CH 4 without the need for syntrophic interactions, while the four remaining lineages are obligate hydrogenotrophic methanogens 36 – 38 that depend on syntrophic interactions with bacteria for electron equivalents driving CO 2 reduction to CH 4 39 . Relationships between AD microbial communities and physicochemical parameters In ecosystems with well-controlled resource inputs such as ADs with constrained feedstock, organic loading rate, retention time, temperature, etc., environmental conditions and microbial community structure can fluctuate together in stable patterns 4 , 40 , 41 . In the Lulu Island AD milieu, intervals with similar process configurations typically selected for similar microbial communities based on hierarchical cluster analysis (Fig. 3A ), while sampling timepoints with distinct physicochemical parameters selected for distinct communities. Notably, at transition points between process configurations, when physicochemical parameters rapidly changed, microbial community composition rapidly responded in a time period shorter than the two-week resolution of our time series sampling, wherein microbial communities at the initial timepoint of a new operating condition still resembled that of the previous configuration before a new a new stable state was reached by the next sampling point (Fig. 3A ). Fig. 3 Relationships between microbial community structure and physicochemical conditions. A The structures of the AD microbial community (ASV data) and the AD physicochemical data are expressed as Bray-Curtis dissimilarity plotted in a dendrogram. Identical samples in each data-type dendrogram are connected with a line and colored corresponding to their process configuration. B The combined table of ASV and physicochemical data underwent principal coordinate analysis and samples were plotted in the first two coordinate dimensions. The variance explained by each dimension is shown as a percent in the axis text. Samples are colored by their process configuration and ellipses sized by the 95% confidence of the centroid of each configuration group are laid in the background. The two-sided PERMANOVA table above the plot shows results of a test between the four configuration groups. By applying a dimension-reduction approach (i.e., canonical correspondence analysis) to a sample-by-sample distance matrix of the combined physicochemical parameter and ASV abundance information, process configuration was identified as a statistically significant variable in shaping the AD microbes and physicochemistry based on permutational analysis of variance tests (Fig. 3B ). Using these combined datasets, it was observed that samples taken directly after transition to a new process configuration were more similar to the previous configuration. Given that no significant decline in RNG production was associated with these transitions (Supplementary Data 1 ), it appeared that detectable shifts in microbial community structure did not necessarily degrade AD performance with respect to overall RNG yield. From these combined data, it was further determined from model fitting that the set of physicochemical parameters that most influenced microbial community structure was organic loading rate (kg/m 3 ) and the concentrations of volatile acids, ammonia, and nitrate (mg/L) in the AD digestate, accounting for 27.6% of the temporal variation in microbial structure observed across the time series (Supplementary Data 7 ). These key parameters are typically identified as the main drivers of microbial structure and activity in mesophilic WWTPs 42 . To account for population-level relationships between microbes and process configurations, ASV temporal distribution patterns were correlated with 28 measured physicochemical parameters across the time series. These results identified microbial populations and physicochemical parameters that were positively correlated (Fig. S4 ). One set of positively correlated variables included total biogas production (m 3 ), organic loading rate (kg/m 3 ), and several populations of Clostridia, Sphingobacteriales, Chloroflexi, and Spirochaetes. A second set of positive correlations, loosely associated with the first set, included AD hydraulic retention time (days), percent biogas composed of CO 2 , concentration of effluent nitrate (mg/L), and several populations of Bacteroidales, Proteobacteria, and acetoclastic methanogens affiliated with Methanothrix. These groupings support previous observations that certain heterotrophic microorganisms, typically those engaged in hydrolysis coupled to fermentation, as well as acetoclastic methanogenic archaea, are most competitive when organic loading rate and retention time are high 43 , 44 . Although total gas production was positively correlated with this set of conditions and taxa, the percentage of CO 2 was also high relative to other timepoints, indicating that a larger share of organic carbon input was metabolized to CO 2 rather than to CH 4 . This correlation of higher CO 2 percent with dominance of acetoclastic methanogens is not unexpected given the stoichiometric imbalance in CO 2 :CH 4 between different methanogenic pathways 45 , 46 . A third set of positively correlated variables included percent CH 4 , temperature (°C), volatile fatty acids (mg/L), and several populations of Clostridiales, Bacteroidetes, and hydrogenotrophic methanogens affiliated with Methanoculleus (Fig. S4 ). A fourth set, loosely correlated with the third, included BOD and COD, ammonia (mg/L) and suspended solids (mg/L), as well as hydrogenotrophic methanogens affiliated with Methanomicrobiales. Previous studies indicate that increases in both ammonia and fatty acid concentrations tend to select for hydrogenotrophic methanogenesis under mesophilic conditions 12 , 47 – 51 . Shifts from acetoclastic to hydrogenotrophic methanogenesis have also previously been observed in WWTP ADs 39 , 52 , 53 , reducing the methane potential of effluent 54 , 55 , and resulting in higher methane content of RNG 56 . Taken together, these results indicate significant coupling between physicochemical parameters and both population- and community-level microbial structure across process configurations. Methanogenic archaea and some putative syntrophic bacteria that cooperate to reduce CO 2 to CH 4 were strongly selected during the Serial Operation, while direct acetate-reducing methanogens and other fermentative and non-syntrophic bacteria were selected in other process configurations. Physicochemical parameter measurements generally corroborated these patterns, including higher ammonia associated with increased hydrogenotroph/syntroph abundances, higher VFA concentrations associated with increased abundances of fermentative bacterial lineages, and higher RNG CO 2 content associated with increased acetoclastic methanogen abundances. Subsequent investigation of the time series focused on identifying microbial indicators engaged in metabolic interactions with potential to influence RNG purity and yield. Identification of microbial indicators During the time series, >25 complete volumetric turnover events occurred based on an average AD retention time of 30 days. Despite this recurring bottleneck, a robust core microbiome could be identified in the Lulu Island AD containing >30% of identified ASVs (339) in at least 80% of samples. This core collectively accounted for 64.4% of total 16S rRNA gene sequence reads. Given that input wastewater to the Lulu Island WWTP varied in origin and upstream processing across the time series, evidence of a robust core microbiome supports the hypothesis that selection factors such as environmental filtering and ecological interactions were likely more important than the initial community composition of waste inputs and the founder effect 11 , 57 , 58 . Based on previous work, selection pressures that help maintain a core AD microbiome include strong physical constraints such as retention time and organic loading 59 , chemical constraints such as low trace metal concentrations and highly anoxic, reducing conditions 60 , and biological constraints such as metabolic interactions between co-occurring microorganisms 61 . Research to identify and characterize these modes of selection in the AD milieu are becoming increasingly important to the biotechnology sector, typically with the assumption that operational controls can be identified and leveraged to select for communities that improve RNG production 6 , 20 , 28 , 62 , 63 . Although a robust core microbiome persisted throughout the time series, the relative abundance of many taxa was significantly impacted by shifts in physicochemical parameters and process configuration. Through indicator species analysis, with configuration as the conditional variable and all ASVs tested for significant overrepresentation based on temporal abundance patterns, 138 indicator ASVs were identified, including both common and conditionally rare taxa (Fig. S3 ). Indicator ASVs were usually distributed such that each configuration was represented by a unique set of ASVs, even within a single family-level taxon, suggesting there may be subtle underlying diversification patterns supporting functional redundancy in the AD milieu. For example, while there were many indicators affiliated with the Syntrophaceae family of Proteobacteria, there were unique sets of Syntrophaceae indicator ASVs for Standard I, CEPT, and Standard II, and Serial Operations. Other taxa with indicator ASVs were more specific in their representation of a given process configuration. For example, 10/13 ASVs from the Rikenellaceae family of Bacteroidetes were affiliated with the Serial Operation. Another notable pattern was the partitioning of archaeal indicator ASVs between configurations based on methanogenic phenotype. For example, the sole archaeal indicator for Standard Operation I was a population of Methanosaetaceae, which perform acetoclastic methanogenesis, while the two archaeal indicators for Serial Operation were affiliated with Methanoregulaceae and Methanobacteriaceae, which only perform hydrogenotrophic methanogenesis. Taken together, indicator analysis at the ASV level provided further evidence that the activity of co-occurring populations, both as core and configuration-dependent consortia, helps shape the active community structure driving RNG production. Time-resolved correlation network analysis Significant correlations were identified between AD process parameters, and both microbial community structure and the abundance of specific microbial populations relevant to RNG production over time. Based on the distributed nature of biomass conversion to methane between different microbial populations in WWTP ADs, we hypothesized that different process configurations would not only select for different populations but also that metabolic interactions among and between populations would vary under selection. To test this hypothesis, a co-occurrence network based on normalized ASV abundances across the two-year time series was constructed from a sparse inverse covariance matrix. Metadata information about nodes in the network, such as indicator ASV status, functional information from a paired MAG, and connectivity to other nodes, were then mapped onto the network to identify configuration-dependent subnetworks containing ensembles with potential to drive RNG production. The ASV co-occurrence network was composed of 390 ASVs which had significant covariance across the time series (Fig. 4 ). The average clustering coefficient of the network (i.e., a measure of grouping or density among nodes) was 0.16, placing it well-within the range of microbial food webs or functional networks, and higher in connectivity than randomly produced associations between populations (Supplementary Data 2 ) 64 . ASVs which were indicators for a given configuration tended to be connected in the network (a result of co-variance between the indicator analysis and the co-occurrence model) and formed strongly correlated subnetworks within the larger parent network. Similarly, closely related ASVs (e.g., those with >97% 16S V4 rRNA homology) tended to coalesce into subnetworks, suggesting that populations with low phylogenetic distance tended to share similar temporal distribution patterns and conserved functional roles due to similar selection pressures (see nodes and edges in Supplementary Data 8 and Supplementary Data 9 ). Fig. 4 A weighted co-occurrence network of the microbial community depicting multiple-test-corrected biweight midcorrelation values (edges) of ASVs (nodes) across the two-year time series. Panels ( A ) and ( C ) depict the network wherein node sizes represent ASV mean relative abundance across all samples, while Panels ( B ) and ( D ) depict node sizes which represent transcript expression levels (TPM) of the representative node MAG. Nodes are shaded with color if they are indicator ASVs for a given process configuration, outlined in black if they are methanogen ASVs, and shaped as triangles if they have a representative MAG. Subnetworks are numbered and highlighted in Panels ( C ) and ( D ). Genome-resolved correlation network mapping Although taxonomically labeled ASVs can provide a way to predict trait-based information that is useful for inferring metabolic interactions, many AD microorganisms have poor taxonomic classifications and additional genome-resolved analysis is needed to assign potential functional roles 8 , 9 , 20 , 27 , 43 . Instead of using reference databases to infer functions from ASV nodes in the network, MAGs were generated from a set of 17 representative samples used for metagenomic whole genome shotgun sequencing across the time series with the goal of linking MAGs to ASVs and then ascribing functions to network nodes. The resulting metagenome datasets had an average assembled length of 657.4 Mbp, from which 40 high-quality MAGs (HQ: >90% complete and <5% contamination, with at least one full ribosomal RNA operon) and 475 medium-quality MAGs (MQ: >50% complete and <10% contamination, with at least one gene copy of the three ribosomal RNA subunits) were binned. Metagenome read mapping indicated that HQ and MQ bins represented on average 12.03% of the total quality base pairs sequenced per sample (Supplementary Data 2 ). The resulting MAGs were mapped onto cognate ASVs based on 16S rRNA gene sequence homology, enabling genome-resolved analysis of nodes within the network. A total of 233 ASVs (25.1% of total ASVs) could be paired with a unique MAG at a 16S rRNA gene sequence homology cutoff of >99.5% over the full length of the V4 region (~1 allowed mismatch). Of the 233 ASV-MAG pairs identified by homology, a total of 186 shared the same lowest-common ancestor taxonomy between the ASV V4 region and the MAG genomic sequence and were used for genome-resolved correlation network mapping (see methods). Thus, the ASV time-series data, enabled by high-resolution sampling, were used to build a network of co-occurring ASVs that characterized temporally coherent taxa which responded similarly to changes in WWTP process configuration. Then, metagenomic sequence information was used to overlay a functional architecture onto the network that used representative MAGs to describe potential metabolic interactions between co-occurring populations at the level of genes, reactions and pathways. The 186 ASV-MAG pairs were distributed throughout the parent network, suggesting good coverage of taxonomic lineages and process configurations (Fig. 4 ). In addition to linking ASVs to MAGs via 16S rRNA gene identity, metatranscriptomes were generated from the 17 samples with metagenomes (Supplementary Data 3 ), allowing assignment of gene expression information (as transcripts per million, TPM) to MAGs in the network (Supplementary Data 10 ). A notable observation from this analysis was that methanogens displayed a strong decoupling between the metagenome- and metatranscriptomes-based measured relative abundances regardless of methanogenic lineage 65 – 67 , most of which had low TPM in metagenomes but very high TPM in metatranscriptomes (Supplementary Data 10 ). However, when only considering bacterial MAGs and not methanogen MAGs, the correlation between metagenome TPM abundance and metatranscriptome TPM abundance was strong ( ρ = 0.53; t = 13.81; df = 472; p -value < 0.001), indicating that bacterial relative abundance observed in metagenome libraries was predictive of enzymatic activity (metatranscriptomes) in the AD milieu. Subnetworks of methanogenic consortia Taken together, the genome-resolved correlation network contained information about (1) temporal patterns of population abundance from the ASV time series, (2) indicator taxa for the various AD process configuration, (3) a subset of network nodes (47.7%) associated with MAGs, and (4) gene-level expression of each MAG. Six subnetworks of co-abundant methane-producing populations were identified within the parent network (Subnetworks 1–6), which were constructed by identifying hydrogenotrophic methanogen ASVs and all nodes connected by primary or secondary co-occurrence edges (one or two degrees of separation). These subnetworks represented ensembles with potential to drive RNG production across the time series, with supporting evidence for biogas-producing syntrophic metabolisms gleaned from metabolic pathway reconstruction and gene expression data associated with MAGs mapped onto cognate ASVs within subnetworks. Subnetwork 1 included four populations: three Rikenellaceae ASVs annotated as MiDAS species 240 (M.S.240) and M.S.3232, two of which were Serial Operation indicators (one of the Rikenellaceae indicator ASV had a corresponding MAG that was annotated to Tenuifilaceae in GTDB), and one Methanoregulaceae ASV (M.S.4938), which was also a Serial Operation indicator with an associated MAG (Supplementary Data 8 ). This small Subnetwork was isolated from the rest of the parent network and, given that 75% of nodes were indicator taxa, the organisms in this subnetwork were significantly more abundant during the Serial Operation when RNG purity was highest. The two MAGs in Subnetwork 1 were both high-quality, with completeness and contamination metrics of 93.3% and 1.4% (Tenuifilaceae; 3300028576_27) and 99.0% and 0% (Methanoregulaceae; 3300036947_39). Due to the strong co-occurrences during Serial Operation and the quality of the MAGs in Subnetwork 1, the encoded and expressed functions of the two genomes were examined for evidence of potential metabolic interactions. As expected, the Methanoregulaceae MAG encoded and expressed all genes necessary for hydrogenotrophic methanogenesis (e.g., mcrABCDG , genes Ga0377204_000018.861 − 865 and mtrABCDEFH , genes Ga0377204_000018.866 − 873) and lacked acetate kinase (ack) and phosphoacetyl transferase ( pta) needed for acetoclastic methanogenesis (Figure S5 ; Supplementary Data 11 ). The Tenuifilaceae MAG that mapped to the subnetwork ASV encoded and expressed a partial Wood Ljungdahl (WL) pathway ( fdh, fhs, fol, met ) but lacked the CODH/ACS complex considered necessary for canonical acetate-oxidizing syntrophy by reverse WL 68 (Figure S5 ; Supplementary Data 11 ). The MAG did, however, both encode and express genes necessary for oxidizing acetate to 5,10-methylenetetrahydrofolate via the glycine cleavage system (Ga0255340_1000079.146, Ga0255340_1000129.100, and Ga0255340_1003402.5), indicating a potential route connecting acetate to its partial reverse WL methyl branch 33 , 56 . While this glycine cleavage system syntrophic mode has been proposed in several taxa there is currently no definitive evidence of its activity 33 , 56 , 62 , 69 – 71 , and many organisms use this system in amino acid biosynthesis reactions and other diverse functions independent from the reverse WL pathway 72 . The Tenuifilaceae MAG also encoded and expressed organoheterotrophic functions common to other Bacteroidetes, including a sus-like biopolymer degradation and transport system and several peptidases (Supplementary Data 11 ), indicating that these three ASVs could also be co-occurring with Methanoregulaceae through non-syntrophic modes of interaction (e.g., by catalyzing the rate-limiting depolymerization steps upstream of methanogenesis or through co-selection because of similar preferences for the conditions in the AD under Serial Operation). Subnetworks 2, 3, and 5, centered around Methanobacterium lacus , Methanospirillium M.S.2576, and Methanospirillium M.S.2576, respectively, none of which had associated high-quality MAGs. Subnetworks 2 and 5 each had an ASV from a syntrophic bacterial lineage ( Smithella and Syntrophaceae) directly connected to the methanogen node (Supplementary Data 9 ), while Subnetwork 3 was an isolated group containing three nodes (two methanogen and one Paludibacteraceae ASVs). Subnetwork 2 was the only subnetwork containing hydrogenotrophic methanogens that also included a Cloacimonadaceae ASV despite the latter lineage exhibiting the highest total relative abundance in the time series analysis and representing 10.5% of nodes in the parent network. This suggests that Cloacimonadaceae are likely not closely associated with syntroph-dependent methanogens in the Lulu Island AD milieu. Subnetwork 4 contained a Clostridia ASV (Christensenellaceae R-7 M.S.240) with an associated MAG (3300028677_44), a Bacteroidales UCG-001 ASV (M.S.1138), a Synergistaceaae ASV (M.S.2022), and a Methanobacterium lacus ASV with an associated MAG (3300028677_53) that was also an indicator for Serial Operation (Supplementary Data 8 ). The two MAGs in Subnetwork 2 had completeness and contamination of 78.3% and 0.4% (Christensenellaceae) and 94.4% and 0.8% (Methanobacteriaceae), respectively (Supplementary Data 10 ), and the Christensenellaceae was directly connected with the syntrophy-dependent Methanobacterium methanogen. This Firmicutes family includes populations of peptide/amino acid fermenters and potentially H 2 -producers 73 , 74 , and though it has no isolated representatives, the Christensenellaceae R-7 lineage has been observed to comprise up to 6% of the AD community in some mesophilic digesters 10 . According to functional data from the Christensenellaceae MAG in Subnetwork 4, this population encoded and expressed several peptide and amino acid metabolism functions (Supplementary Data 11 ), including three major operons (two livFGHKM and one azlCD ) for branched-chain amino acid transport, over 15 metabolic amino/exo/endo/oligopeptidase enzymes, and the hallmark fermentation enzyme pyruvate:ferredoxin oxidoreducase (Ga0255346_1000279.56). Between the Bacteroidales (as a depolymerizer 75 ), Christensenellaceae (as a fermenter), Synergistaceae (as an SAOB 56 , 76 ), and Methanobacterium (as a hydrogenotrophic methanogen) ASVs, Subnetwork 4 retained the metabolic capacity to carry out hydrolysis, fermentation, acetate oxidation, and methanogenesis of complex organic material to RNG. The observation that this Methanobacterium ASV was also a statistical indicator for the Serial Operation further suggests that this ensemble was most abundant in time series analysis during the period of high local RNG purity. Subnetwork 6 was the largest co-occurring set of ASVs and included 25 nodes with 7 total ASV-MAG pairs (Supplementary Data 8 ). This Subnetwork was formed by four smaller but contiguous subnetworks. With four hydrogenotrophic methanogens in the 25-node subnetwork, Subnetwork 6 accounted for 40% of hydrogenotrophic methanogen ASVs in the parent network. Among the 25 nodes were also two indicator ASVs for Standard Operation II. Subnetwork 6 included several syntrophic lineages that had primary or secondary edges connected to the methanogen nodes, including two Syntrophomonas (both M.S.3971), one Syntrophorhabdus (M.S.998), and one Synergistaceae ( Thermovirga M.S.988) 56 , 77 . Based on 10,000 permutations of 25 random nodes from the parent network, the probability of observing four syntroph nodes and four hydrogenotrophic methanogen nodes by chance was 2.8%, showing that Subnetwork 6 was rich in syntrophic interactions driving RNG production. Nodes with representative MAGs in Subnetwork 6 were two Clostridia ASVs (D8A-2 lineage and Ruminococcus ) and a single ASV from Desulfobacterota ( Syntrophorhabdus M.S.998), Deltaproteobacteria ( Phaselicystis M.S.2086a), Paludibacteraceae (M.S.2677), Patescribacteria (Candidatus Falkowbacteria M.S.5033), and Methanolinea (M.S.4938). The directly co-occurring Methanolinea and Syntrophorhabdus MAGs were 99.0% complete with 0% contamination (3300036947_39) and 72.9% complete with 8.6% contamination (3300028576_34), respectively, and each had a full rRNA operon. The 16S rRNA gene sequence of the Syntrophorhabdus MAG (across the whole 1482 bp gene) was 93.48% similar to the best cultured representative Syntrophorhabdus aromaticivorans str . UI. Similar to this type strain, the Syntrophorhabdus MAG encoded and expressed benzoate-CoA ligase for benzoate degradation to benzoyl-CoA (Ga0255340_1019139.2) and two benzoyl-CoA reductase operons for converting benzoyl-CoA to the dienoyl-CoA intermediate (Ga0255340_1011848.9 and Ga0255340_1011848.11, plus Ga0255340_1031544.4 and Ga0255340_1031544.6), including two operons of the heterodisulfide reductase involved in this endergonic step; the benzoyl-CoA reductase enzymes were highly expressed (TPM = 17.53; 30 th most abundant ORF in the genome), indicating that this pathway is a critical metabolic step for Syntrophorhabdus (Supplementary Data 11 ). To complete the oxidation of benzoate to H 2 , CO 2 , and fatty acids, the MAG encoded and expressed syntenic genes for dienoyl-CoA hydration (Ga0255340_1005473.7), breaking and hydrolyzing the intermediate ring (Ga0255340_1005473.3 and Ga0255340_1005473.5), and finally for beta-oxidation (Ga0255340_1013965.3-7). Taken together, these observations provide an overlaid picture of population-level co-occurrence, metabolic interactions facilitating that co-occurrence, and gene expression data to support those metabolic interactions. This study also provides a model for conducting time series amplicon analysis coupled with genome-resolved correlation mapping to identify known and novel metabolic interactions underlying organic waste conversion to RNG with potential to define new design principles that improve RNG quality and yield at scale. These same design principles can in turn be extended to the production of other value-added compounds, hastening the transition from waste treatment to sustainable waste resource recovery."
} | 10,525 |
37502701 | PMC10369077 | pmc | 225 | {
"abstract": "Arbuscular mycorrhizal fungi (AMF) are ubiquitous in soil and form nutritional symbioses with ~80% of vascular plant species, which significantly impact global carbon (C) and nitrogen (N) biogeochemical cycles. Roots of plant individuals are interconnected by AMF hyphae to form common AM networks (CAMNs), which provide pathways for the transfer of C and N from one plant to another, promoting plant coexistence and biodiversity. Despite that stable isotope methodologies ( 13 C, 14 C and 15 N tracer techniques) have demonstrated CAMNs are an important pathway for the translocation of both C and N, the functioning of CAMNs in ecosystem C and N dynamics remains equivocal. This review systematically synthesizes both laboratory and field evidence in interplant C and N transfer through CAMNs generated through stable isotope methodologies and highlights perspectives on the system functionality of CAMNs with implications for plant coexistence, species diversity and community stability. One-way transfers from donor to recipient plants of 0.02-41% C and 0.04-80% N of recipient C and N have been observed, with the reverse fluxes generally less than 15% of donor C and N. Interplant C and N transfers have practical implications for plant performance, coexistence and biodiversity in both resource-limited and resource-unlimited habitats. Resource competition among coexisting individuals of the same or different species is undoubtedly modified by such C and N transfers. Studying interplant variability in these transfers with 13 C and 15 N tracer application and natural abundance measurements could address the eco physiological significance of such CAMNs in sustainable agricultural and natural ecosystems.",
"conclusion": "4 Conclusions and future perspectives A range of 0.02 to 41% (C) and 0.04 to 80% (N) of one-way transfer have been observed from donor to recipient plants through the determination of 13 C and 15 N signatures ( \n Tables 1 \n , \n 2 \n ). Interplant C and N transfers can affect not only the growth and competition between donor and recipient plants but also ecosystem stability. For example, Weremijewicz et al. (2016) observed that Andropogon gerardii plants in intact CMNs under sunlight acquired 9% of their N, but shaded plants (~35% photosynthetically active radiation) acquired only 1% N, from their conspecific neighbors. They suggested that AM fungi in CAMNs preferentially provide N to conspecific hosts of with fixed C or presenting the strongest sinks, thus potentially expanding asymmetric underground competition. Castro-Delgado et al. (2020) showed that the mycelium could transfer diverse compounds and signals among plants that would modify plant behavior in favor of protection of the whole network. In general, stable isotope tracing has provided an effective way to study the exchange of mineral nutrients between plants through CAMNs. Although 13 C and 15 N labeling techniques have demonstrated that CAMNs are an important pathway for the translocation of both C and N, the functioning of CAMNs in ecosystem C and N dynamics remains equivocal. To make an explicit link between nutrient transfer in CAMNs and nutrient cycling in ecosystems new approaches are needed. For example, a combination of high-throughput genome sequence techniques with model-based assessments could further identify the extent of CAMNs in interplant C and N translocation in natural and managed ecosystems ( Orwin et al., 2011 ; Zhou et al., 2021 ). The following issues about the physiological and ecological functions of AMF or CAMNs should be addressed. 1. Can 13 C and 15 N natural abundance, like 13 C and 15 N external labeling, be employed to detect C and N transfer? Study have shown that plant δ 13 C signatures could reflect the δ 13 C of the C sources of associated fungi and δ 15 N signatures could reflect the δ 15 N of N sources to plants ( Querejeta et al., 2003 ). However, the reliability of using 15 N natural abundance to estimate AMF-mediated N transfer has been recently questioned ( Choi et al., 2020 ; Jach-Smith and Jackson, 2020 ). 2. In what form are C and N transferred through CAMNs? Amino acids, lipids, or carbohydrates for C, amino acids or ammonium for N? Does a pollen development encompass a mechanism that is shared with CAMNs symbiosis? What may the two phenomena have in common ( Nouri and Reinhardt, 2015 )? Can fluorescent nanoscale semiconductors or quantum dots ( Whiteside et al., 2012 ) be combined with 13 C and 15 N labelling to trace the transfer of organic nutrients through CAMNs ( Govindarajulu et al., 2005 ; Parniske, 2008 )? 3. Can network theory and computer modeling ( Southworth et al., 2005 ; Wipf et al., 2019 ) simulate the direction and distribution of interplant C and N transfer facilitated by CMNs and thus predict both positive and negative effects of CMNs in natural and managed systems ( Alaux et al., 2021 )? 4. 15 N labeling showed that AMF could not directly decompose organic matter, but the interaction between AMF and other decomposers enhanced organic matter decomposition and hence the absorption of N by AMF ( Hodge and Fitter, 2010 ). However, how can CAMNs regulate the process of C and N translocation and absorption between AMF mycelia and host plants? In addition, a coupled concurrent C and N movement through CAMNs has not been reported. 5. What determines the net effect of CAMN-mediated interplant nutrient transfer on plant C assimilation and N metabolism? Does the transferred C and N affect the performance or fitness of the donor, receiver, or both? What is the ecological significance of CAMN mediated nutrient transfers in natural and managed ecosystems? Whether AMF-mediated interplant C and N transferred is agronomically important to managed ecosystems, including agroforestry, forestry, croplands, and grasslands, is debated ( Rillig et al., 2019 ; Ryan et al., 2019 ). How do modern agricultural practices, such as long-term organic farming, no-till, or fertigation affect the establishment and performance of CMNs and subsequent effects on fertilizer use efficiency, crop agronomic characters and productivity? 6. How the abundance and function of soil bacterial and other fungal communities could be manipulated and promoted through a CAMN-mediated interplant C and N transfers ( Bonfante and Anca, 2009 ; Yuan MTM. et al., 2021 )? Is plant C investment in AM fungal growth related to soil N acquisition within a CAMN? How is the N for C trade between mycorrhizal symbionts regulated if plants are linked through a CAMN? What determines the magnitude and direction of such C and N transfer within the same or different plant species in mono-species or mixed-species systems, particularly along their complete plant growth and development cycle? How exogenous and endogenous factors can interplay with CAMNs, and how a nutrient can impinge on AM symbiotic signaling and also on a later cellular program in host plants ( Nouri et al., 2014 ). 7. Irrespective of photosynthetic capabilities or N 2 -fixation characteristics of plant species, what the phylogenetic and functional diversity of plant species can benefit from nutrient transfer through CAMNs? These species would be in a diverse range as C 3 , C 4 , C 3 -C 4 , CAM and parasitic plants. Are there interactions between AM and EM networks on C and N transfers since some plants do have dual AM/EM associations ( Wang and Qiu, 2006 ; Teste et al., 2015 )? How can technical problems be overcome in demonstrating unequivocally that a C or N transfer directly occurs through CMNs rather than indirectly through root exudates or soils ( Zhang et al., 2019 ; Fall et al., 2022 ; Reay et al., 2022 )? 8. How will drivers of global environment change including elevated CO 2 concentration, N deposition, drought and temperature affect interplant C and N transfer through CAMNs? Each can have substantial impacts on the direction and magnitude of such C and N transfers and ultimately on resource sharing or competition ( Fellbaum et al., 2014 ; Řezáčová et al., 2018 ; Mickan et al., 2021 ). To answer these issues, it is important to keep in mind that mycorrhizal symbiotic benefits are interactively formed between plants and fungi under specific habitats and soil properties.",
"introduction": "1 Introduction 1.1 Arbuscular mycorrhiza Arbuscular mycorrhizas (AM) are formed between arbuscular mycorrhizal fungi (AMF) and roots of ~70% of ~391,000 higher plant species ( Wang and Qiu, 2006 ; Smith and Read, 2008 ; Brundrett, 2009 ; Brundrett, 2017 ; Brundrett and Tedersoo, 2018 ). Currently 25 genera and ~338 fungal species belonging to the sub-Phylum Glomeromycota form AMF globally ( Schüßler and Walker, 2010 ). AMF acquire soil nutrients, such as nitrogen (N), phosphorus (P) and other mineral nutrients, and transport them to their host plant in exchange for up to 20% of photosynthetically fixed carbon (C) ( Smith and Read, 2008 ; Roth and Paszkowski, 2017 ). In an arbuscular mycorrhiza, the intraradical mycelium (IRM) often penetrates root cortical cells to form arbuscules, while the extraradical mycelium (ERM) extends into soil, far beyond the root zone. The ERM forages for N, P, potassium and other soil nutrients, and translocates them to the IRM, where they are exchanged for C from the host ( Smith et al., 2009 ). The ERM is extensive enabling plant access to nutrient resources well beyond the root depletion zone ( Li et al., 1991 ). In addition, several findings revealed that sources of carbon for mutualistic AMF include fatty acids exported from the host plants, as well as lipids and sugars ( Pfeffer et al., 1999 ; Keymer et al., 2017 ; Jiang et al., 2017 ). 1.2 Common arbuscular mycorrhizal networks AMF are ubiquitous components of most soil ecosystems, where they grow through soil, colonize plant roots, and can form links between plants ( Newman et al., 1992 ; Newman et al., 1994 ; He et al., 2003 ; He et al., 2009 ; Molina and Horton, 2015 ). The plants suppling AMF with labile carbon often grow close together, primarily in multiple species communities. Because AMF exhibit little host specificity ( Smith and Read, 2008 ), and plant roots can thus be linked by a common AM network (CAMN) ( Wipf et al., 2019 ). Such CAMNs, being formed among individual plants of the same species or genus, or from different genera or families ( Ronsheim and Anderson, 2001 ; Southworth et al., 2005 ), are usually woven into an even larger network of fungi and roots in natural communities ( Smith and Read, 2008 ; Wipf et al., 2019 ). In this way, plant species within CAMNs may be joined together as a functional guild and become pathways for movement or transfer of nutrients ( \n Figure 1 \n ), including C ( Francis and Read, 1984 ; Martins, 1992 ; Martins, 1993 ; Watkins et al., 1996 ; Fitter et al., 1998 ; Mikkelsen et al., 2008 ; Voets et al., 2008 ; Walder et al., 2012 ), N ( Hamel et al., 1991a ; Hamel et al., 1991b ; He et al., 2003 ; He et al., 2009 ; Frey and Schüepp, 1993 ; Rogers et al., 2001 ; Moyer-Henry et al., 2006 ), P ( Tuffen et al., 2002 ; Smith and Smith, 2011 ; Merrild et al., 2013 ), arsenic (P analog, Meding and Zasoski, 2008 ), cadmium ( Ding et al., 2022 ), K ( Gao et al., 2021 ), cesium ( Meding and Zasoski, 2008 ; Gyuricza et al., 2010 ), rubidium (K analogs) and strontium (Ca analog) ( Meding and Zasoski, 2008 , and zinc ( Cardini et al., 2021 ). Water ( Egerton-Warburton et al., 2007 ) and genetic material ( Giovannetti et al., 2004 ) can also move within these networks. Movement of these materials can thus promote coexistence and biodiversity among plants ( Read, 1991 ; Smith and Read, 2008 ). Figure 1 Milestones in nutrient transfers through common mycorrhizal networks (CMNs) (Note that the year of a reference pointing to the green arrow bar is not to scale). Despite the considerable evidence of the functional role of CAMNs, they have not been directly visualized in natural ecosystems due to their cryptic, fragile, and microscopic nature ( Newman et al., 1994 ; Ronsheim and Anderson, 2001 ; Southworth et al., 2005 ; Wipf et al., 2019 ). Plants invest photosynthetic products to feed their fungal partners, which, in return, provide mineral nutrients foraged in soil by their intricate hyphal networks ( Bever et al., 2010 ). The Driver (AMF partners drive plant communities) and Passenger (AMF community dynamics follows changes in the host plant community) hypotheses were suggested to explain the mutual relationships of plant and AMF communities ( Zobel and Öpik, 2014 ). Research into this complex system of plant-fungus interactions indicates that plants and fungi can choose their trading partners ( Kiers et al., 2011 ; Walder et al., 2012 ). An understanding of the stoichiometry of C, N, or other nutrients mediated by CAMNs could better elucidate the potential roles of CAMNs in C and N functioning in plant-soil systems ( \n Figure 2 \n ), although at present CAMNs have not been directly visualized in natural ecosystems due to their fragile and microscopic nature. Application of high-throughput genome sequences or all sorts of omics and BONCAT-FACS (bioorthogonal non-canonical amino acid tagging + fluorescence-activated cell sorting, Couradeau et al., 2019 ) could have an in situ observation of these underground cryptic microorganisms. Meanwhile, the employment of other emerging technologies, such as cryo-scanning electron microscope (Cryo-SEM), DNA stable isotope probing (DNA-SIP), quantitative multi-isotope imaging mass spectrometry (MIMS), nanoscale secondary ion mass spectrometry (NanoSIMS), single-molecule electronic device and synchrotron radiation facility, could enable the mapping the interplant flow of 13 C and 15 N through CAMNs. Given this demonstrated autonomy and the key role that CAMNs play in interplant nutrient transfers and biodiversity in ecosystems, it is crucial to understand how nutrient resources (e.g., C, N, P, other elements, see \n Figure 1 \n ) are shared among plants through CAMNs. And whether there may be a mechanism between CAMNs and ecosystems by which a greater biodiversity is associated with a greater productivity. Figure 2 A conceptual framework of roles played by common mycorrhizal networks (CMNs) in regulating carbon (C) and nitrogen (N) flow or transfer within and between plants. 1.3 Application of isotopes of 13 C and 15 N labeling The abundance level of stable isotopes is theoretically expressed as delta (δ) in parts per thousand or per mil (‰), which is calculated as δ 13 C or δ 15 N (‰) = [(R Sample /R Standard ) –1] × 1,000, where R is the 13 C/ 12 C or 15 N/ 14 N (atom%) ratio of the sample and standard, and “Vienna”-Pee Dee Belemnite (0.0112372) or atmospheric N 2 (0.0036765) is their respective standard material. The 13 C and 15 N isotopic composition (also expressed as δ 13 C and δ 15 N signatures) of plant materials can provide information on (i) inputs of photosynthetic C or uptake of fertilizer N, (ii) plant N derived from N 2 fixation by symbiotic microorganisms, (iii) C or N cycling and (iv) the sources of N available for host plant growth ( Dawson et al., 2002 ). For instance, the δ 13 C or δ 15 N signatures in vegetation could reflect the relative availability of C sources to fungi and N sources to plants differing in isotopic composition ( Querejeta et al., 2003 ). Here we examine the unique and common characteristics of CMN-mediated interplant C and N transfers that are demonstrated by 13 C and 15 N labeling (sometimes referred to as 13 C and 15 N enrichment) or variations in their isotopic composition for exploring the beneficial functionality of CMNs in sustaining managed and natural systems in a changing climate ( Dawson et al., 2002 ; He et al., 2003 ; Querejeta et al., 2003 ; Moyer-Henry et al., 2006 ; He et al., 2009 ; Jalonen et al., 2009 ; Kurppa et al., 2010 ; Walder et al., 2012 ; Ren et al., 2013 ; Meng et al., 2015 ; Wang et al., 2016 ; Řezáčová et al., 2018 ; Wipf et al., 2019 ; Muneer et al., 2020a ; Alaux et al., 2021 ; Avital et al., 2022 ; Reay et al., 2022 ). 1.4 Calculation of carbon and nitrogen transfer from a donor to a receiver plant Estimates of C or N transfer from a donor to a receiver plant are based on the assumption that an equal proportion of applied and unapplied C or N are transferred. The percentage of total C or N transferred to the receiver (% N transfer ) is then assessed from the ratio of applied C or N in the receiver and total applied C or N in the receiver and donor. Based mostly on the calculations from Giller et al. (1991) ; Ikram et al. (1994) ; Johansen and Jensen (1996) , the following equations are commonly employed by almost all relevant studies to calculate C or N transfers. \n (1) \n % C t r a n s f e r o r N t r a n s f e r = 13 C c o n t e n t r e c e i v e r o r 15 N c o n t e n t r e c e i v e r × 100 / ( 13 C c o n t e n t r e c e i v e r + 13 C c o n t e n t d o n o r o r 15 N c o n t e n t r e c e i v e r + 15 N c o n t e n t d o n o r ) \n where 13 Ccontent plant or 15 Ncontent plant = atom% 13 Cexcess plant or 15 Nexcess plant × \n (2) \n t o t a l C p l a n t o r N p l a n t / a t o m % 13 C e x c e s s l a b e l e d C o r a t o m % 15 N e x c e s s l a b e l e d N \n and atom% 13 Cexcess plant or atom% 15 Nexcess plant = atom% 13 C plant or \n (3) \n a t o m % 15 N p l a n t a f t e r l a b e l i n g − a t o m % 13 C p l a n t o r 15 N p l a n t b a c k g r o u n d \n The amount of C or N (mg plant −1 ) transferred from the donor (C transfer or N transfer ) is calculated as: \n C t r a n s f e r o r N t r a n s f e r = % C t r a n s f e r o r % N t r a n s f e r × \n \n (4) \n t o t a l C d o n o r o r t o t a l N d o n o r / ( 100 − % C t r a n s f e r ) o r ( 100 − % N t r a n s f e r ) \n The % of C or N in the receiver derived from transfer (% CDFT or % NDFT) is calculated as: \n (5) \n % C D F T o r % N D F T = C t r a n s f e r o r N t r a n s f e r × 100 / t o t a l C r e c e i v e r o r t o t a l N r e c e i v e r"
} | 4,533 |
22666024 | PMC3355405 | pmc | 226 | {
"abstract": "Autoinducer signals enable coordinated behaviour of bacterial populations, a phenomenon originally described as quorum sensing. Autoinducer systems are often controlled by environmental substances as nutrients or secondary metabolites (signals) from neighbouring organisms. In cell aggregates and biofilms gradients of signals and environmental substances emerge. Mathematical modelling is used to analyse the functioning of the system. We find that the autoinducer regulation network generates spatially heterogeneous behaviour, up to a kind of multicellularity-like division of work, especially under nutrient-controlled conditions. A hybrid push/pull concept is proposed to explain the ecological function. The analysis allows to explain hitherto seemingly contradicting experimental findings.",
"conclusion": "3. Conclusions A mathematical model was developed interlinking spatially an autoinducer system with an the environmental factor (nutrients). The modelling results indicate that autoinducers can promote a highly adaptable, spatial phenotypic heterogeneity in spatially structured populations, e.g., within colonies and biofilms, instead of inducing homogeneous behaviour of the population. The differentiation into subpopulations, which can be interpreted as division of work, is amplified by the interplay with other regulative factors as nutrient supply. In contrast to heterogeneity caused by stochastic noise [ 24 , 25 ], this can only emerge in spatially structured populations, not in plankton. Beside push information about the potential strength of a regulated activity, integration of nutrient starvation or other environmental and physiological factors introduces a pull (demand) dimension into the information carried by autoinducers. We hypothesise that the aim of this autoinducer regulation architecture is to ensure the most efficient responses of the whole cell aggregate to changing environmental conditions, aiming to optimise the fitness of the whole colony. Environmental autoinducer concentrations thus carry a highly integrative information, optimised to allow for meaningful decisions of cells in their specific environment. We think it will be worthwhile to re-consider some biological questions from the perspective of the push/pull concept: The existence of multiple autoinducer systems in some species is more comprehensible by assuming that the efficiencies of different target behaviours depend on different independent pull aspects, varying over time and space. Multiple autoinducer systems allow for independent transportation of such different pull information. Note that the push information might not differ in this case. Non-inducibility via externally added autoinducer in certain growth phases could be connected with downregulation of autoinducer systems (e.g., autoinducer receptors) via pull control [ 26 , 27 ]. The role of the universal autoinducer AI-2 has been questioned—among others—due to its origin as an unavoidable waste product reflecting the activity of the methyl cycle pathway, whereas “true” autoinducers should simply reflect the presence of a cell [ 28 ]. However, with respect to the discussion given in this paper, the differences diminish. Temporal pattern of autoinducer controlled gene expression in batch cultures [ 26 ], which is believed to be connected to changes in the concentration of, e.g., nutrient and waste product, is in situ probably at least partly reflected by spatial patterns. The relevance of spatial structuring affects the development of treatment strategies for pathogens or beneficial bacteria on autoinducer level: For instance, aiming at the spatial structure of bacteria populations may be an alternative to changing the cell density. Inversely, spatial organisation will make a bacterial behaviour more effective and protect it from treatment influence. The presented hypothesis about fitness optimisation on colony level needs further investigation. Our generic basic model does not include any feedback of the regulated behaviour on the autoinducer regulation, nor a fitness analysis. Experiments on the level of energy balance or reproduction success are needed to verify the hypothesis. The general lack of quantitative data and mechanistic insight for the (often nonlinear) relation between multiple environmental factors and autoinducer system activity impedes an detailed understanding of the ecologic purpose of signalling. It can be overcome by differentiated quantitative data from chemostat and retentostat experiments. Furthermore, experiments analysing the spatial pattern within colonies associated with various environmental conditions as well as the interaction of autoinducer systems between colonies and between species in the light of gradient induced heterogeneity are highly desirable.",
"introduction": "1. Introduction Extracellular signalling via small diffusible compounds (autoinducers) is used by many bacterial species, including pathogens, mutualistic as gut commensals, and plant growth promoting bacteria. In brief, bacteria release autoinducers and simultaneously regulate target gene expression dependent on the environmental autoinducer concentration [ 1 ]. Most autoinducer systems include a positive feedback loop, often by autoinducers' control of their own production. In combination with nonlinearity (e.g., using oligomers of autoinducer-receptor complexes as active units), this often results in switch like regulative behaviour. Functionality of various autoinducer systems has been investigated in a number of mathematical modelling approaches [ 2 ]. Autoinducer regulation was originally assumed to be a strategy enabling synchronous and uniform life style switches of the whole bacterial population in dependency on the cell density. This strategy has been termed “quorum sensing” (QS) [ 3 ]. Within cell aggregates like biofilms or colonies, spatial gradients of autoinducers may emerge and result in a spatial organisation of autoinducer induction. Spatially heterogeneous induction of autoinducer regulated genes has indeed been found in some biofilm experiments. Model results suggest that the highest autoinducer concentration is always present in the colony centre or near the attachment surface of biofilms. However, different experimental studies reported strongest upregulation of autoinducer controlled genes both, near the bottom or at the top of biofilms, partly even for the same species [ 4 – 6 ]. These contradictions have not been explained, yet. Based on more indirect arguments, the few biofilm models, which include a nutrient/autoinducer connection, assume that a decreasing nutrient supply downregulates autoinducer activity [ 7 – 9 ]. Seemingly, this assumption is appropriate for some species. However, autoinducer induction often regulates responses to stress as starvation, and the scare existent quantitative analyses suggest non-linear or even non-monotonous relations ([ 10 ], Mellbye and Schuster, pers. comm.). At least in a certain range, nutrient deficiency promotes autoinducer activity. Necessarily, at very severe starvation autoinducer production will finally abolish. Potential nutrient gradients in cell aggregates/biofilms can affect further the spatio-temporal heterogeneity of autoinducer regulation. We developed a 3D model of autoinducer regulation in attached microcolonies based on a typical lux-type autoinducer system. The relation between nutrient and autoinducer production is integrated as experimentally reported in [ 10 ], which presents to our knowledge the best quantitatively analysed system ( Figure 1 ). The main objective of our study was to develop a model for a mechanistic analysis of the developing spatio-temporal heterogeneity of autoinducer systems in growing microcolonies, with a special focus on the influence of nutrients as an example for interacting environmental factors in order to explain the seemingly contradicting experimental findings. We will further discuss the potential ecological implications: Which information is sensed via autoinducers and what is the benefit? Although mainly based on data of lux autoinducer system in Vibrio fischeri , this study is not meant to provide a quantitative analysis of autoinducer regulated activities in V. fischeri , but to contribute generally to the development of the theory of autoinducer regulation under complex environmental conditions.",
"discussion": "2. Results and Discussion We consider a generic model focusing on the interaction between the autoinducer system and nutrient availability, under consideration of emerging gradients of both in a bacterial colony. Our main hypothesis is that spatial heterogeneity is mainly caused by an interplay of these two systems. 2.1. Model The cells we model possess an autoinducer system of lux-type with an AHL (acylhomoserine lactone) acting as an autoinducer. It binds to a receptor molecule (LuxR), the AHL-receptor dimerises. The dimers bind to the lux operon, where the autoinducer synthase (LuxI) and luminescence genes are up-regulated, but also to other target genes of the regulon [ 3 ]. We assume that the bulk fluid is large, so that no accumulation or depletion of released or consumed substances occurs. Loss of AHL is mainly due to diffusion, whereas degradation by lactolysis can be neglected under slightly acid conditions [ 11 ]. We focus on equilibrium situations, which seems reasonable as diffusion and change of induction state occur at faster time scales than, e.g., microcolony growth. We give a brief description of the mathematical model. We start with the spatial geometry. Subsequently we address nutrient N ( x, t ) availability and AHL A ( x, t ) regulation (see Table 1 ). 2.1.1. Geometry Let us consider a colony, whose fraction of volume occupied by cells is given by ρ ( x ). Note that we do not measure the number of cells per volume (cells per volume), but the fraction of the volume occupied by cells. Of course, both data are equivalent. The colony is centred around the origin of the three-dimensional space. We do not consider population dynamics: As population dynamics is slower than the processes we shall consider here, we are allowed to assume the population size to be fixed; the results of a growing colony resemble the results we obtain here. Furthermore, a rotationally symmetric setting is assumed, i.e. , ρ ( x ) = ρ (| x |). The colony has radius R, ρ (| x |) = 0 for |x| > R . 2.1.2. Nutrient N ( x, t ) is a generic replacement of all nutrients we ever need. We do not relate N ( x, t ) to the concentration of a specific nutrient like glucose. This nutrient is available in 100% (corresponding to N ≡ 1) if we go arbitrary far away from the colony. Within the colony it is consumed in a Michaelis–Menten way, generating nutrients gradients in dependency of diffusion properties. N t = D N Δ N − ρ ( x ) K cat , N N N + K m , N N ( t , x ) → 1 for | x | → ∞ 2.1.3. AHL Diffusion of AHL is formulated in a standard way. Only the production term of AHL is interesting. Following the usual modelling approach [ 12 ], we use a constant production rate to indicate the basic expression, and a Hill function that corresponds to the increased production rate in the induced state. The Hill coefficient specifies the cooperative effects of the regulatory pathway controlling the autoinducer production. The complete production rate is multiplied by the cell concentration. Additionally, it is also multiplied with a factor f ( N ), which reflects the modulation of the signalling system by nutrient availability. A t = D A Δ A + ρ ( x ) f ( N ) [ α + β A n A n + A τ n ] A ( t , x ) → 0 for | x | → ∞ The rational of parameter selection and numerical methods for simulation can be found in Section 4. 2.2. Effect of Nutrient Availability on AHL Regulation In order to understand the effect of nutrient availability on AHL production, we simulate the system with and without influence of nutrient on the AHL production (for the latter we take the function f ( N ) to one). In both cases, we obtain a well-known bifurcation diagram for systems with hysteresis: if the population size is above a critical number (this is, if the colony radius is above a critical magnitude) the colony becomes activated ( Figure 2 ). This critical radius is decreased by the influence of nutrient depletion as it is to be expected. The model that does not incorporate effects of nutrient depletion shows that the activation degree stabilises for larger colony radii, while the model that takes into account nutrient depletion, the activation degree breaks down (due to starvation effects), starting with the central cells (black curve). Eventually also the boundary cells (and therewith the complete colony) starve so much that it is completely deactivated (green curve). This latter behaviour can change if nutrients are continuously delivered, e.g., by flow in the bulk fluid (not shown). Even more interesting is the profile of nutrient, AHL concentration and AHL production rate over the distance to the colony centre (see Figure 3 ). In both nutrient scenarios the AHL concentration reaches its maximum at the colony centre and declines towards the colony border. The variation of AHL concentration between different locations within the colony is limited (please note that the y -axis does not start at zero). In vivo , limited gradients may be covered by, e.g., additional cellular regulation pathways, spatially heterogeneous biofilm structures and noise. Variation of AHL production is less than 10% without nutrient influence. Again, the highest production rate is found in the colony centre. But the AHL production rate varies significantly if the nutrient dependency of this rate is taken into account. This applies both for the difference between maximum and minimum production rate and the location of the maximum. The reason is the nutrient depletion that naturally starts at the centre of the colony. Depending on colony size, the maximum activation may be at the colony centre, somewhere between colony centre and colony boundary, or at the colony boundary. If, e.g., AHL production and luminescence (or other target activities) are connected, i.e. both under similar control of autoinducer and nutrients, a stripe of active cells moves from the centre to the colony boundary in case of a growing colony. We expect the same behaviour for different geometrical setups as flat layers of bacteria with varying thickness in a plane biofilm. This observation may explain the observed conflicting experimental results about spatial autoinducer regulation. First/strongest induction in autoinducer systems was reported near colony centre resp. at the biofilm bottom but also at/near upper surface [ 13 – 15 ]. Simple models of autoinducer regulation without influence of additional factors as nutrients usually predict highest activity only in the centre. Interestingly, for autoinducer regulation in Pseudomonas aeruginosa [ 4 – 6 ], for which highest autoinducer regulation activity was reported to occur at the top or at the bottom of biofilms, indeed an nutrient influence was experimentally shown (Mellbye and Schuster, personal communication). The intention of this paper was to promote the general understanding of the analysed regulative subsystem, not a complete, quantitative analysis of a specific strain of V. fischeri or another species. Autoinducer systems of other species may have other parameter values, but the general trends will hold. Higher AHL production rates per cell, higher cell densities within the colony or lower threshold concentration for the induction will decline the colony size needed to induce the system. A lower Hill factor, as in a system without multimerisation of LuxI/AHL complexes, decreases or destroys the bistable range. The relation between nutrients (or other environmental factors) and the autoinducer system is critical for the potential of heterogeneity development. Unfortunately, direct quantitative data are largely missing. More complex biofilm morphologies, e.g., channels and density differences between the border and the centre of the colony, and intercellular variability will complicate the activity pattern, but are not assumed to change the conclusions of this paper. Cheaters, which do not produce AHL but respond on it, reduce the average AHL production rate per cell and thus the local AHL concentration, respectively increase the colony size required to reach a certain AHL concentration. 2.3. Ecological Interpretation The main idea about the purpose of the autoinducer system is efficiency (cost effectiveness): The cell regulates relatively expensive actions via the autoinducer system. Compared to the regulated activities, the autoinducer molecules are quite cheap to produce, as they are small and usually active in low concentrations. They are used as a proxy in order to decide if it is worth to trigger these expensive activities. It is straight to understand that the concentration of autoinducer molecules is able to predict the concentration of for example exoenzymes. Thus, it is sensible to use autoinducers to regulate the exoenzyme production. The initial interpretation has been quorum sensing (how large is the population?), later accompanied by diffusion sensing (how large is the diffusible space or, more generally, what are the mass transfer properties?) [ 3 , 16 ]. These two concepts have been unified by efficiency sensing (is it efficient to start any action?) [ 17 ]. How is it possible to integrate the observation that nutrient availability (and perhaps also other environmental or intracellular conditions) affects autoinducer production into these interpretations? The idea is that the cells do not perform pure quorum/diffusion/efficiency sensing (an impartial exploration of the space), but, up to a certain degree, they also integrate information about their individual demands into the communication signal and respond to it. Formally, this is reflected in the model by the dependency of the AHL production on the nutrient function f(N) (see Section 4.1.3). A complete analysis about the effect of nutrients with respect to the efficiency of the regulated activities goes beyond the scope of this paper. However, we can draw some conclusions. In case of nutrient depletion, the increase of autoinducer production can be interpreted as kind of an emergency call. Starving cells may have an increased demand for a coordinated behaviour improving the supply of a limiting nutrient. The demand is communicated by increased autoinducer release. This reduces the number of required cells sufficient to reach the induction threshold. On the other hand, in a spatially structured population the outer cells of the colony, which are yet under less starvation stress, will contribute to the cooperation and thus improve also the conditions for the more stressed inner cells. The system allows for an efficiency optimised regulation on colony level by integration of the specific demands of each cell at its specific side into a spatially structured communication. If the population is very small, an emergency call will not have any effect—though there is an increased autoinducer production, the AHL concentration does not become supercritical. However, with growing colony size the call will be taken up by the colony. Also outer, well-fed cells of this colony start to produce exoenzymes such that the inner, starving cells are supplied with nutrient. Once the starvation becomes too severe, eventually the demand stops. From the efficiency point, increase of autoinducer signal release under starvation, which implies increased costs, is reasonable in spatially structured populations where cells develop different demands. In mixed plankton, every cell experiences the same nutrient concentration, thus a direct co-control of target genes by autoinducers and nutrients seems more cost effective. Spatial organisation of activity via autoinducers adds a new layer of cooperativity to this communication systems. This observation raises interesting questions about its functionality under the aspect of kin respectively group selection. Referring to the terminology in industrial production processes and following its recent transfer to biological processes, we propose the term “hybrid push/pull control” for this kind of regulation design ([ 18 , 19 ]). Roughly, “pull” in industrial processes (the actual demand of the buyers for a product) reflects the demand of a cell for the target behaviour, transported by a changing autoinducer production. “Push” in industrial processes (the (potential) strength of a company to produce this product) reflects the possible reached strength of the target behaviour. As the pull and push factors both influence the cost-effectiveness of the regulated activity, we hypothesise that the core purpose of this regulation system is to promote a gene control based on pre-assessment of the efficiency of the target gene resp. behaviour ( Figure 4 ). Beside nutrients, other intra- and extracellular factors affect the efficiency. Examples are other environmental factors (like pH, temperature), presence of competing or beneficial microbes, indicators for health state of hosts (relevant for potential pathogens), and the developmental state of a cell (like sporulation, mobility). From the efficiency point of view all these factors affect the actual need of the cells for the regulated target behaviour, their potential to contribute to it, or the opportunity that the regulated behaviour is effective (because, e.g., the host is weak). In fact, some of these aspects are known to be connected with autoinducer regulation pathways [ 20 – 22 ]. One idea in that context is that sensing the combined pull/push information carried by the autoinducers allows for a contextual interpretation of the state of the neighbouring cells relative to its own and of the push factors for each cell [ 23 ]. This results in an adaptive behaviour of cells within the growing colony, highly dynamic in space and time. Cooperativity and division of work can emerge. Their spatio-temporal flexibility goes beyond those of real multicellular organisms, e.g. because the cell differentiation with respect to division of work can be highly reversible. This remarkable phenotypic plasticity probably enables an adaptive life style optimisation of the entire colony under the current conditions with respect to the fitness."
} | 5,613 |
24194675 | PMC3806361 | pmc | 227 | {
"abstract": "We optimized and tested a postbioprocessing step with a single-culture archaeon to upgrade biogas (i.e., increase methane content) from anaerobic digesters via conversion of CO 2 into CH 4 by feeding H 2 gas. We optimized a culture of the thermophilic methanogen Methanothermobacter thermautotrophicus using: (1) a synthetic H 2 /CO 2 mixture; (2) the same mixture with pressurization; (3) a synthetic biogas with different CH 4 contents and H 2 ; and (4) an industrial, untreated biogas and H 2 . A laboratory culture with a robust growth (dry weight of 6.4–7.4 g/L; OD 600 of 13.6–15.4), a volumetric methane production rate of 21 L/L culture-day, and a H 2 conversion efficiency of 89% was moved to an industrial anaerobic digester facility, where it was restarted and fed untreated biogas with a methane content of ~70% at a rate such that CO 2 was in excess of the stoichiometric requirements in relation to H 2 . Over an 8-day operating period, the dry weight of the culture initially decreased slightly before stabilizing at an elevated level of ~8 g/L to achieve a volumetric methane production rate of 21 L/L culture-day and a H 2 conversion efficiency of 62%. While some microbial contamination of the culture was observed via microscopy, it did not affect the methane production rate of the culture.",
"conclusion": "5. Conclusions A thermophilic bioprocess with a pure culture of the hydrogenotrophic methanogen M. thermautotrophicus and with a continuous gaseous feeding scheme showed that the hydrogen gas transfer flux was limiting the conversion rates of carbon dioxide into methane. During an experimental period in which the hydrogen influent rate was increased, we observed increasing VMPRs of up to 49.2 L/L culture-day (1.33 g/L-h). However, this resulted also in lower hydrogen conversion efficiencies due to hydrogen bypassing. Therefore, careful optimization of the bioprocess is needed to maximize the hydrogen conversion efficiencies while maintaining a high enough conversion rate of carbon dioxide into methane. Pressurization of the headspace, indeed, increased the VMPR further to 65.6 L/L culture-day (1.79 g/L-h). Introducing methane into the influent gas with different relative ratios showed that methane itself was not inhibiting the conversion rates, but that it lowered the hydrogen partial pressures, resulting in lower methane production rates and hydrogen conversion efficiencies. The single culture of M. thermautotrophicus was also able to convert carbon dioxide from industrial, untreated biogas into methane with an external source of hydrogen gas. Chemical contaminants, such as hydrogen sulfide, and biological contaminants in biogas did not result in a reduction in performance (VMPR) even though some microbial contamination of the culture was observed. The biomass concentrations increased during the 8-day operating period to a maximum level of 8.0 g/L (OD 600 of 15.2). The stable operating conditions, the robust activity, and the relatively high biomass concentrations under industrial conditions pave the way to develop this technology further to upgrade biogas into renewable natural gas when sustainable hydrogen is available.",
"introduction": "1. Introduction Organic waste streams contain energy that is stored in biomass, which had originally been harnessed from the sun by photosynthesis. To prevent environmental problems during the release of these waste streams, biological treatment is necessary. At the same time, there is a growing interest in recovering this stored energy in more useful forms by converting the complex biomass into bioenergy sources that are direct replacements of fossil fuels [ 1 ]. The traditional route for this conversion with relatively energy-dense wastes is via methane fermentation in anaerobic digesters [ 2 , 3 ]. Anaerobic digesters consist of an open culture of microbial consortia (referred to here as a reactor microbiome) with a dynamic food web that includes bacterial hydrolysis, acidogenesis, and acetogenesis, as well as archaeal methanogenesis [ 1 ]. Anaerobic digestion is an ideal process for two reasons: (1) the product methane bubbles freely out of solution without costly separation; and (2) the anaerobic reactor microbiome harvests the maximum amount of free energy without oxygen by maximizing the production of methane (resulting in high conversion efficiencies) [ 4 ]. However, during digestion, both methane and carbon dioxide must be produced to balance the high oxidation number (i.e., number of transferable electrons per carbon) for methane with the low oxidation number for carbon dioxide. The resulting stoichiometric reactions equalize the oxidation state of the products to the oxidation state of the substrate since no alternative electron acceptors or donors are added to an anaerobic system [ 5 ]. Therefore, the carbon dioxide content depends on the substrate composition of the organic waste stream, and typically remains within a range of 30–50% [ 6 ]. The carbon dioxide in digester biogas is inert as a fuel and dilutes the energy content of the biogas, preventing the introduction of biogas as a renewable natural gas into the current natural gas pipeline infrastructure. With a 30–50% carbon dioxide content, biogas has an energy density of ~18–23 MJ per cubic meter [ 7 ], while natural gas has an energy density of 37 MJ per cubic meter. The low energy density of biogas requires modification of energy conversion systems and renders biogas an inefficient energy carrier for long-distance transportation and energy storage [ 7 ]. Therefore, most of the biogas is used at or near the point of production to run boilers or combined heat and power systems, such as engines or turbines, or, in a worst-case scenario, it is flared off. Since the electric power efficiencies are low (~35%), a large fraction of the energy is often lost as waste heat when it cannot be directly used in a local setting. To overcome the low energy content of biogas, three strategies have been developed to upgrade biogas (i.e., increase methane content) into renewable natural gas by: (1) removing carbon dioxide from biogas via postprocessing technologies; (2) supplying a reduced substrate (e.g., hydrogen gas) to the organic waste stream of the anaerobic digester with the goal to convert carbon dioxide into methane in situ ; or (3) converting carbon dioxide from biogas into methane via postprocessing technologies. The minimum methane content requirement for the product gas to be injected into the natural gas network differs from country to country and depends on specific gas system regulations. For example, within Denmark, the carbon dioxide content of the natural gas cannot exceed 2.5 mol% [ 8 ]. For the first strategy, physical gas separation methods (e.g., gas-to-liquid exchange, amine extraction, semipermeable membrane technology, pressure-swing adsorption, or their hybrid variants) have been used at industrial scales to remove carbon dioxide from biogas and to discard it [ 9 – 11 ]. For the second and third strategies, converting carbon dioxide in situ or as a postprocessing step, respectively, offers the advantage of simultaneously increasing the net methane production. In a laboratory-scale study, Luo et al. [ 12 ] supplied hydrogen gas in addition to a complex organic substrate (manure) to mesophilic anaerobic digesters (second strategy). In essence, they increased the oxidation state of the total substrate (manure and hydrogen gas) to close the gap with the oxidation state of methane, resulting in lower carbon dioxide content in the biogas. The biological hydrogen conversion led to a reduction in carbon dioxide content from 38% to 15% and a methane production increase of 22%. An unanticipated disadvantage was the increase in pH to 8.3, which led to inhibition of methanogenesis [ 12 ]. The authors continuously monitored the accumulation of short-chain carboxylic acids, because, thermodynamically, under anaerobic conditions, the oxidation of propionic acid and n -butyric acid, which are intermediate chemical species in the anaerobic food web, is not favorable when hydrogen partial pressures reach levels above 10 −2 kPa [ 13 ]. Therefore, regulating the supplementation of hydrogen gas to a constantly varying, complex organic waste stream will be difficult; undersupplementation will lead to an excessive carbon dioxide content in the biogas, while oversupplementation will lead to accumulation of short-chain carboxylic acids and unstable digester conditions with a reduced overall methane yield. For the third strategy of upgrading biogas in a postprocessing step, an abiotic or biological system could be used. For another possibly carbon-dioxide-rich industrial gas (i.e., synthetic combustion gas), Hoekman et al. [ 14 ] had explored the use of metal catalysts to convert carbon dioxide into methane. In addition, several pure-culture methanogens, including Methanothermobacter thermautotrophicus , were tested to convert carbon dioxide from a synthetic fermentation off gas into methane [ 15 ]. For biogas upgrading, Luo and Angelidaki [ 16 ] operated a thermophilic, laboratory-scale anaerobic digester while continuously feeding a synthetic biogas stream that consisted of 60% H 2 , 25% CH 4 , and 15% CO 2 (4 : 1 ratio of H 2 : CO 2 ) and achieved a maximum methane content in the product gas of ~95%. Over the operating period of one month, the reactor microbiome (open culture) in this digester became enriched with hydrogenotrophic methanogens, but other trophic groups remained active, such as homoacetogenic bacteria. In addition, a microbiome characterization found a diverse group of thermophilic, anaerobic methanogens, including a sequence with a 93% ID to M. thermautotrophicus [ 16 ]. In another study with CO gas fed into an anaerobic digester, sequences with a <93% ID to M. thermautotrophicus were abundant [ 17 ]. In all these studies, a synthetic gas was used rather than an industrial biogas. \n M. thermautotrophicus is a lithoautotrophic, thermophilic (40–70°C) methanogenic archaeon, which was first isolated as strain deltaH from sewage sludge at a wastewater treatment facility in Urbana, IL [ 18 ], and has a sequenced genome [ 19 ]. M. thermautotrophicus deltaH was described as a strict, obligate anaerobe with an optimal growth temperature of 65–70°C and pH of 7.2–7.6 [ 18 ]. Other related strains have been isolated, including strain Hveragerdi, which was isolated from an Icelandic alkaline hot spring [ 20 ]. M. thermautotrophicus conserves energy by using hydrogen to reduce carbon dioxide to methane and also uses carbon dioxide as its carbon source. Some knowledge with laboratory-scale bioprocessing with a pure culture of M. thermautotrophicus strain Hveragerdi has been published in two studies by Schill et al. [ 21 , 22 ]. The authors found that growth of M. thermautotrophicus with a continuous H 2 : CO 2 (4 : 1) and medium flow was not just dependent on the dilution rate, which is commonly accepted for chemostats with a liquid substrate, but that both gas influent rates and dilution rates needed to be taken into account. In addition, they observed and modeled that the gaseous substrate consumption (removal) rates positively influenced the effective gas transfer flux into the system by maintaining a low effective gas concentration. A higher gas transfer flux resulted in a considerably higher production rate than when modeled with a standard gas transfer coefficient ( k \n L \n a ) and the bulk liquid concentration ( c \n D ) values for the gaseous substrate with the lowest solubility. The highest k \n L \n a reported for this study was 2,300 h −1 with a biomass concentration of 4.84 gL −1 at their highest gas flow rates. The authors also found a hydrogen conversion efficiency (into methane and biomass) of 88% at their lowest gas flow rates [ 21 ]. In addition, thermodynamic calculations explained why heat dissipation will be higher than for other bioprocesses: (1) cell synthesis from carbon dioxide will require a high energy expense; and (2) the entropy will drop considerably because the small molecules of hydrogen and carbon dioxide are converted to macromolecules [ 23 ]. The same authors further refined their mathematical model and experiments and observed that the growth rates for M. thermautotrophicus have to be very small with a large heat production rate, which they called entropy-retarded growth [ 22 ]. Biogas upgrading strategies that utilize living microbes to convert carbon dioxide into methane have the advantage of relying on a self-replicating catalyst. To prevent conversion of hydrogen gas into acetate by homoacetogens, and a resulting efficiency loss of biogas upgrading, the inoculation of the bioprocess with a pure-culture archaeon will be advantageous compared to a reactor microbiome. Next, to be of practical use in industrial settings, the ideal archaeon inoculum must also be able to withstand: (1) accidental exposure to oxygen; (2) exposure to hydrogen sulfide (which is often present in biogas); (3) contamination from other bacteria or phages, which are present in the continuously fed biogas; (4) possible inhibition by a high methane content in biogas; and (5) intermittent supply of renewable hydrogen due to the fluctuating nature of wind and photovoltaic energy sources. A robust archaeon must also avoid byproduct production and enable a highly energy efficient process. Here, we studied biogas upgrading with a pure culture of M. thermautotrophicus fed with hydrogen gas in combination with biogas. We optimized bioprocessing and tested biogas upgrading by performing four experiments: Experiment 1 with a synthetic H 2 /CO 2 gas without methane; Experiment 2 with a synthetic H 2 /CO 2 gas without methane and a pressurized headspace; Experiment 3 with a synthetic biogas; and Experiment 4 with an industrial, untreated biogas from the Anheuser-Busch InBev facility in St. Louis, MO. Our findings suggest that biogas can be upgraded with a pure culture of M. thermautotrophicus fed by an external source of H 2 gas and that the function of a M. thermautotrophicus culture was maintained during an 8-day operating period using an untreated industrial biogas even though some microbial contamination was observed.",
"discussion": "4. Discussion 4.1. Comparisons between Experiments and Other Studies Pure-culture growth in continuous bioreactors with gaseous substrates is a function of the liquid medium dilution rate as well as the gas influent rate [ 21 ]. With relatively low dilution rates, this resulted in a very high biomass concentration for M. thermautotrophicus of ~5 g/L in Schill et al. [ 21 ] and in our Experiment 4 (with a hydrogen influent rate of 0.2 L/min and with industrial, untreated biogas) for which we achieved a dry weight concentration and OD 600 of 8.0 g/L and 15.2, respectively. Both single-culture studies (Schill et al. [ 21 ] and our study) showed biomass concentrations that were considerably higher than what is typically observed in chemostats with a liquid influent. As a result, the high hydrogen uptake rates of the robust biocatalyst resulted in a relatively high hydrogen gas transfer coefficient with hydrogen as the limiting gas, resulting in the superior effective gas transfer flux that supported the observed maximum VMPR of 163 L/L culture-day (4.39 g/L-h) ( Table 3 ) [ 21 ]. Our maximum VMPR with a H 2 /CO 2 gas mixture was ~1/3 of this rate ( Table 3 ). According to Experiments 1 and 2, the hydrogen gas transfer flux was limiting the conversion rate of our bioprocess. Therefore, the differences in operating conditions between the studies, such as a larger active volume of 3 versus 1.5 L and a lower mixing speed of 700 versus 1,000 rpm, resulted in considerably lower power to volume rations, causing less mixing activity, which explain the lower hydrogen gas fluxes that we observed compared to Schill et al. [ 21 , 22 ]. Because of these large effects on the hydrogen transfer flux by changing the operating conditions, care must be taken to maintain high conversion rates during scale up while limiting the parasitic energy input of mixing. Within our study, we found little difference in VMPRs of the M. thermautotrophicus culture whether it was fed with industrial biogas or synthetic biogas ( Table 3 ). Thus, upgrading of biogas shows promise for application in industrial biogas production sites. We did, however, observe a considerable drop in VMPR from ~50 to ~12 L/L culture-day when the influent gas stream contained methane ( Table 3 ). The experimental results are consistent with a role for methane as an inert gas in the system by increasing the gas flow through the reactor, decreasing the gas residence time, and decreasing the hydrogen gas transfer flux into the medium. This influence of methane as a diluent of the reactive gases (H 2 and CO 2 ) must be considered in the engineering design of industrial postprocessing biogas upgrading bioreactors. The VMPR of 12 L/L-day (0.3 g/L-h) is similar to rates that have been achieved with high-rate anaerobic digesters for easily degradable organic wastewater ( Table 3 ) with very high volatile suspended solids concentrations of 50 g/L due to biofilm (granular biomass) formation, even though the methanogens existed in a very diverse reactor microbiome [ 24 ]. Because of the need for high mixing intensities to overcome hydrogen transfer flux limitations, such high biomass concentrations via biofilm formation are not anticipated for our bioreactor system with gaseous substrates. Feeding hydrogen gas into a low-rate anaerobic digester with a reactor microbiome to upgrade biogas did not result in compatible VMPRs when compared to the single-culture studies with M. thermautotrophicus ( Table 3 ). A lower, but compatible, VMPR of 5.3 L/L/-day (0.14 g/L-h) was achieved by Luo and Angelidaki [ 16 ] with a postprocessing bioreactor system to upgrade biogas with an acclimated reactor microbiome of enriched thermophilic hydrogenotrophic methanogens ( Table 3 ). It is not clear at this point whether a further acclimated microbiome will eventually be able to achieve maximum VMPRs as high as observed with our single-culture of M. thermautotrophicus fed with biogas. Although, we anticipate that the observed acclimated homoacetogens in the reactor microbiome by others [ 16 ] would reduce the efficiency considerably, failing to ever achieve the performance of the pure culture archaeon. It is noteworthy that the study by Luo and Angelidaki [ 16 ] used a smaller active volume of 600 mL and a higher mixing intensity of 800 rpm than our study. The power to volume ratios were, therefore, closer to the study by Schill et al. [ 21 ] than to our study, while Schill et al. [ 21 ] obtained a far superior VMPR compared to Luo and Angelidaki [ 16 ] ( Table 3 ). 4.2. Implementation Further development of the bioprocess technology is required before implementation. We observed low hydrogen conversion efficiencies during the optimization of VMPRs through increases in the gas influent rates. The efficiencies must, however, be improved so as not to lose valuable hydrogen gas into the effluent gas and ultimately into the natural gas infrastructure. Our study showed that improving these efficiencies by lowering the hydrogen influent rates resulted in considerably lower VMPRs. Therefore, the performance should be improved by other improvements. With the increased headspace pressure in Experiment 2, we already showed that increasing the hydrogen partial pressure improved these efficiencies. Experiment 2 also showed that with a pressurized atmospheric pressure (122 kPa), the VMPR increased at the highest hydrogen influent rate rather than decreased, which we observed for atmospheric pressures in Experiments 1 and 2 ( Figure 3(a) ). Pressurization, thus, reduced the hydrogen mass transfer limitations that became apparent because of the shorter hydrogen residence times in the bioreactors with an increased H 2 influent rate. Further optimization studies with higher pressures are required though. Other measures to lengthen the gas residence times in the reactor vessel may also yield higher hydrogen conversion efficiencies by, for example, using a novel bioreactor configuration and by recycling effluent gas. The latter needs to be tested first as methane as an inert gas would decrease the hydrogen gas transfer flux as we have shown in Experiment 3. The resulting improved hydrogen conversion efficiencies must also result in the production of an effluent gas with a sufficient quality to introduce it into the natural gas grid. Here, we did not achieve such a quality, but with bioreactor design and operating condition improvements should be attainable while maintaining high VMPRs. In addition, before full-scale systems to upgrade biogas can be implemented, other research questions need to be answered, including what is the effect of a higher or lower (i) H 2 S concentration compared to the 7,000 ppm that was present in the industrial biogas here to test for possible sulfide toxicity or a deficiency of sulfur for metabolic growth, respectively and (ii) H 2 : CO 2 ratio compared to the 4 : 1 ratio that we used here? To produce a high-methane-content renewable natural gas as our effluent gas, a source of sustainable hydrogen gas must be utilized. The first implementation of this technology will likely use off-peak electric power production to generate hydrogen gas via intermittent electrolysis of water. It is well known that methanogens can be intermittently fed and that dormant cultures can be started up rapidly in large-scale digester systems [ 2 ]. Current excess electric power exists in areas with a high density of wind or photovoltaic energy, and therefore, biogas upgrading can be incorporated into an energy storage system that transfers excess electric power from the existing electric grid into the other existing grid—the natural gas grid, which has a vast energy storage capacity. The overall electric-to-chemical energy conversion efficiencies are estimated to be ~60% when waste heat is not utilized and ~80% when waste is utilized. However, a detailed energy balance and life cycle assessment study is needed to ascertain the energy efficiencies and carbon dioxide recycling gains of storing energy rather than switching off the renewable generators such as windmills."
} | 5,606 |
22439005 | PMC3306304 | pmc | 228 | {
"abstract": "Foundation species, such as kelp, exert disproportionately strong community effects and persist, in part, by dominating taxa that inhibit their regeneration. Human activities which benefit their competitors, however, may reduce stability of communities, increasing the probability of phase-shifts. We tested whether a foundation species (kelp) would continue to inhibit a key competitor (turf-forming algae) under moderately increased local (nutrient) and near-future forecasted global pollution (CO 2 ). Our results reveal that in the absence of kelp, local and global pollutants combined to cause the greatest cover and mass of turfs, a synergistic response whereby turfs increased more than would be predicted by adding the independent effects of treatments (kelp absence, elevated nutrients, forecasted CO 2 ). The positive effects of nutrient and CO 2 enrichment on turfs were, however, inhibited by the presence of kelp, indicating the competitive effect of kelp was stronger than synergistic effects of moderate enrichment of local and global pollutants. Quantification of physicochemical parameters within experimental mesocosms suggests turf inhibition was likely due to an effect of kelp on physical (i.e. shading) rather than chemical conditions. Such results indicate that while forecasted climates may increase the probability of phase-shifts, maintenance of intact populations of foundation species could enable the continued strength of interactions and persistence of communities.",
"introduction": "Introduction A few strong interactions often contribute disproportionately to maintaining the composition and function of an ecosystem by modifying both the physical conditions and species interactions within [1] , [2] , [3] . Key species can maintain ecosystem composition not only by forming biological habitats whose physical environment facilitates their own recruitment, but also by dominating competitors that would otherwise inhibit this process. Such organisms, variously called ‘foundation species’ or ‘ecosystem engineers’, create stable conditions for other dependent species [3] , [4] . The inhibition of competitors associated with contrasting physical conditions and species interactions, therefore, enhances the stability of systems centered on these foundation species [5] . As human activities continue to modify abiotic conditions, there is increasing concern that such strong interactions will be altered (e.g. the sea Pisaster ochraceus may be less effective at consuming mussels [6] ). Reduction in the strength of interactions could disrupt the persistence of entire biological communities, ranging from kelp forests to seagrasses and coral reefs in the marine realm, and grasslands to forested ecosystems in the terrestrial realm. In the marine realm, the coastal zone is an area in which high productivity and species diversity coincide with human activity and this area is set to be further influenced by the effects of a changing climate [7] . Altered land use and ensuing discharges to the marine environment elevate nutrient concentrations at local scales, with the extent of change ranging from strong enrichment in urban areas to little or no change in agricultural and natural systems [8] , [9] , [10] . These waters will also absorb approximately 30 percent of the atmospheric CO 2 produced by human populations globally, leading to gradual ocean acidification [11] , [12] . While there is recent recognition that these alterations of the physical environment will affect species interactions [13] , [14] , [15] , [16] experiments to date have not progressed sufficiently to identify how they will affect biological communities dominated by foundation species such as kelp. Australian kelp are habitat-forming species whose persistence has been enabled by their self-facilitation of recruitment through the competitive exclusion of opportunistic turf-forming algae [17] . When kelp canopies are lost, turfs rapidly colonise space and their sediment-trapping morphology inhibits the recruitment of juvenile kelp and re-formation of kelp forests [18] , [19] . Under conditions of severely elevated nutrients, these naturally-ephemeral turfs persist in fragmented canopies [10] , [20] to cause intergenerational decline and collapse of the kelp community [8] . Turfs, therefore, mediate the effect of nutrient-driven loss of kelp forests and often constitute a vital component in the indirect effects of pollution on habitat loss. Under moderate scenarios of nutrient pollution, it is possible that kelp forests can persist by continuing to exclude turfs [10] . Similarly, the elevation of CO 2 over the near-future may not alter the strength by which kelp suppress turfs. While susceptible to many other human-altered conditions, kelp meiospores are anticipated to germinate successfully under near-term enrichment of CO 2 conditions [21] . Furthermore, productivity of ensuing recruits and subsequent individuals may be increased under elevated CO 2 \n [22] . Evidence to date, however, suggests moderate increases of CO 2 facilitate greater covers and biomass of turf, potentially turning them from ephemeral to persistent habitats [16] , [23] . It remains unknown whether the competitive dominance of kelp over turf, (i.e. an interaction of particular concern to the regeneration of kelp) is likely to be reduced or increased under the combined influence of moderate nutrient and CO 2 pollution. We consider the model that elevated CO 2 may assist kelp sustainability despite the greater potential for turfs to persist. We tested the hypothesis that a foundation species would continue to suppress its key competitor under conditions of moderate forecasted levels of pollution which have the potential to favour its competitor's expansion. That is, we assessed if the competitive dominance of kelp over turfs [17] would continue under moderate forecasted levels of local (i.e. nutrient) and near-term global pollution (i.e. CO 2 ) and their known synergy [16] . If the strength of interactions involving foundation species are maintained despite the increasingly novel conditions brought about by human activities, then phase-shifts may be avoided. Such phase-shifts are not uncommon, but anticipating them has been problematic because many involve indirect effects [24] for which the impact of one species (e.g. kelp) on another (e.g. turf) requires knowledge of a third element that is inadequately understood (e.g. synergies among pollutants). Our study addresses a reasonably widespread challenge of forecasting the ecology of phase-shifts under future climates.",
"discussion": "Discussion Over 30 years ago, Harrison [34] suggested that there was a need to understand not only the behavior of a community under ‘normal or good conditions’, but also its response to unusual or stressful conditions. Since then, research considering the effects of stressful conditions created by human activities has often focused on identifying the community response to highly-modified conditions (e.g. [35] , [36] ). A more pressing contemporary concern, however, is whether moderate near-term alterations will be of a sufficient magnitude to drive changes in community interactions. Potential exists that near-term future conditions may reduce the capacity of foundation species to suppress competitors whose expansion would otherwise cause communities to shift to, and be maintained in, a contrasting state (e.g. [19] ). Whilst severe pollution, such as nutrient conditions associated with urban coasts [10] , is known to reduce the capacity of kelp forests to recover from disturbance (i.e. resilience) [19] , intact kelp forests may be quite stable in the face of similar sets of stressors, of a lesser magnitude, such as coasts associated with agriculture [10] . Although near-term forecasted environmental conditions are anticipated to facilitate competitors and increase the probability of loss of foundation species (e.g. the strong positive synergistic effect of increasing nutrient and CO 2 concentrations on turf [16] ), the current study suggests that where kelp canopies are retained their mere presence may be sufficient to continue to suppress a key competitor (e.g. turfs), despite the synergistic effects of moderate elevation of local (i.e. elevated nutrients) and global pollutants (i.e. forecasted CO 2 ). As the conditions that promote community resistance may be different from those that favour resilience, recognizing the factors that affect persistence rather than recovery could assist in forecasting their effects on these normally robust and diverse natural systems [37] . The synergistic responses of kelp competitors to multiple pollutants (i.e. turf response to CO 2 ×nutrients ( [16] , this study), supports the model that multiple stressors can combine to produce conditions which increase the likelihood of phase-shifts [38] . Consequently, researchers have been increasing their focus to identify those sets of stressors which combine to produce effects that cannot be anticipated by adding their isolated effects [39] . The frequency and magnitude of non-additive responses are surprisingly common, to the extent that our concept of resource limitation has shifted from an earlier paradigm of single-resource limitation [40] towards that of co-limitation by multiple resources [41] , [42] . While ‘limitation’ can be experimentally recognised by changing the rate of processes through addition or reduction of the single relevant factor, ‘co-limitation’ is recognised as the greater response to simultaneous enrichment of multiple factors than would be expected from the sum of their individual responses [42] . The repeated observation of an interaction between CO 2 and nutrients ( [16] , this study) indicates nutrients are not available in great excess relative to CO 2 , as a modest addition of CO 2 quickly produces a limitation on nutrients. It also appears CO 2 is not in great excess relative to nutrients, as an addition of nutrients quickly provokes a limitation on CO 2 . When CO 2 and nutrients are added together, CO 2 and nutrient limitation may alternate in numerous small incremental steps, ultimately producing a synergistic effect. This model may account for the observed synergy between CO 2 and nutrients in a similar way Davidson and Howarth [43] account for the prevalence of nitrogen and phosphorous interactions [44] . Whilst this synergy would appear relevant for canopy-gaps or locations experiencing canopy loss, it is less likely to be relevant in disrupting the persistence of intact kelp forests The mechanisms that allow kelp to suppress their competitors under conditions that would otherwise facilitate their spread may be useful to understand. Quantification of physiochemical conditions within the experimental mesocosms indicates that the mechanism driving kelp inhibition is alteration of the physical (i.e. shading) rather than chemical (i.e. nutrient or carbonate) conditions experienced by understorey species. The presence of kelp did not appear to modify either the nutrient status (i.e. ammonia, phosphate, NO X ) or carbonate chemistry of water within the mesocosms (i.e. pH, TA, p CO 2 , HCO 3 \n − ; see also Figure S3 for diurnal pH variation). We suspect, however, that the accelerated growth of turf in the absence of kelp is likely to obscure this potential effect by utilising the relatively moderately elevated nutrients. On biomass basis, turfs are naturally more productive (i.e. 44–77%) than surrounding canopy-forming algae in this system [45] . We consider that shading by kelp canopies provides a more powerful explanation of the suppression of turfs. This explanation is derived from classical experiments showing the effects of canopy-shade on understorey communities [46] and covers of turfs [17] , [47] . Where perennial canopy species are removed, algae adapted to high light conditions, such as turfs, are then able to utilise the increased light to expand their covers [46] , [48] . In contemporary algal assemblages the presence of intact kelp canopies reduces light reaching the substratum to a similar extent as that which was observed in our experimental mesocosms (i.e. a ∼95% reduction) [17] , [47] . The retention of populations of foundation species seems critical in ensuring maintenance of the primary mechanism that enables the continued dominance of kelp over its competitors, in this case shading. We do, however, recognise that this conclusion is based on the assumption that communities will remain intact, maintaining the strength of interactions, a particularly important assumption for assemblages whose structure is determined by a small number of interactions centred on a single foundation species [49] . The biotic factors that influence shading tend to vary, especially when the impacts of human activities, such as canopy removal, are considered [50] . While the delivery of light flecks to the understorey during canopy movement appears important in maintaining understorey productivity, when large amounts of light become available, such as when entire plants are removed from the substratum and a gap in the canopy is produced, the influence of the canopy may be reduced and persistence of ecosystems disrupted [51] . For example, as kelp canopies are thinned, reduced in size or fragmented, the associated environmental conditions (including light) become more similar to those experienced outside the canopy [52] . Under these conditions, turfs can expand to dominate space in assemblages and inhibit the recruitment of kelp [19] , [23] , leading to phase-shifts over multiple generations [53] . Key species can maintain ecosystem composition through strong interactions that are often self-stabilising because they create conditions that facilitate the persistence of entire ecosystems [54] . Given that species interactions are often mediated by environmental conditions [55] , [56] , human activities which modify the abiotic environment have the potential to disrupt these interactions and alter the species composition of ecosystems [7] , [15] . Where strong interactions maintain community structure by retarding the effects of environmental forcing, management of key species may assist in the retention of communities, even under forecasted global conditions (i.e. large-scale pollution and climate change). In conclusion, our results show the interaction between kelp and turf may be maintained under near-term future conditions, indicating the retention of intact forests may reduce the effect of moderate pollutant enrichment in these communities. Many communities are governed by a few strong interactions (e.g. presence of kelp forests) which exert disproportionately strong community-wide effects [3] . The maintenance of intact populations of foundation species may enable these habitats to persist despite forecasted climates that would otherwise appear to increase the probability of their loss."
} | 3,759 |
30367461 | null | s2 | 229 | {
"abstract": "Experimental tests of community assembly mechanisms for host-associated microbiomes in nature are lacking. Asymptomatic foliar fungal endophytes are a major component of the plant microbiome and are increasingly recognized for their impacts on plant performance, including pathogen defense, hormonal manipulation, and drought tolerance. However, it remains unclear whether fungal endophytes preferentially colonize certain host ecotypes or genotypes, reflecting some degree of biotic adaptation in the symbioses, or whether colonization is simply a function of spore type and abundance within the local environment. Whether host ecotype, local environment, or some combination of both controls the pattern of microbiome formation across hosts represents a new dimension to the age-old debate of nature versus nurture. Here, we used a reciprocal transplant design to explore the extent of host specificity and biotic adaptation in the plant microbiome, as evidenced by differential colonization of host genetic types by endophytes. Specifically, replicate plants from three locally-adapted ecotypes of the native grass Panicum virgatum (switchgrass) were transplanted at three geographically distinct field sites (one home and two away) in the Midwestern US. At the end of the growing season, plant leaves were harvested and the fungal microbiome characterized using culture-dependent sequencing techniques. Our results demonstrated that fungal endophyte community structure was determined by local environment (i.e., site), but not by host ecotype. Fungal richness and diversity also strongly differed by site, with lower fungal diversity at a riparian field site, whereas host ecotype had no effect. By contrast, there were significant differences in plant phenotypes across all ecotypes and sites, indicating ecotypic differentiation of host phenotype. Overall, our results indicate that environmental factors are the primary drivers of community structure in the switchgrass fungal microbiome."
} | 499 |
20000730 | null | s2 | 230 | {
"abstract": "Synthetic spider silk holds great potential for use in various applications spanning medical uses to ultra lightweight armor; however, producing synthetic fibers with mechanical properties comparable to natural spider silk has eluded the scientific community. Natural dragline spider silks are commonly made from proteins that contain highly repetitive amino acid motifs, adopting an array of secondary structures. Before further advances can be made in the production of synthetic fibers based on spider silk proteins, it is imperative to know the percentage of each amino acid in the protein that forms a specific secondary structure. Linking these percentages to the primary amino acid sequence of the protein will establish a structural foundation for synthetic silk. In this study, nuclear magnetic resonance (NMR) techniques are used to quantify the percentage of Ala, Gly, and Ser that form both beta-sheet and helical secondary structures. The fraction of these three amino acids and their secondary structure are quantitatively correlated to the primary amino acid sequence for the proteins that comprise major and minor ampullate silk from the Nephila clavipes spider providing a blueprint for synthetic spider silks."
} | 306 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.