Fulfilling the promise of a circular economy, where waste byproducts can be reutilized in closed-loop processes, relies on the development of sustainable alternatives to non-renewable resources. In recent years, it has become increasingly evident that addressing this challenge requires harboring the immense potential of microorganisms (Aswani et al., 2024). Microbes carry out a variety of functions that could be harnessed for biotechnological purposes. For example, some microbes are capable of degrading plastics (Taghavi et al., 2021), transforming complex compounds like lignocellulose (often present in municipal waste) into biofuels (Cragg et al., 2015; Ling et al., 2014; Prasad et al., 2019; Senne De Oliveira Lino et al., 2021), producing biomaterials (Laurent et al., 2024), or synthesizing high-value molecules including drugs (Zhang et al., 2022), plant natural products (Cravens et al., 2019; Walls et al., 2023), or vitamins (Fang et al., 2017). Microbes are now even emerging as a potential sustainable source of food (Graham and Ledesma-Amaro, 2023).

Unlocking the full biotechnological potential of microorganisms requires us to be able to optimize the relevant functions they provide. Humans have been harvesting microbial functions for millenia, with evidence that early agricultural societies already produced fermented beverages around 7,000 years ago (Wang et al., 2021). Yet, it was only in recent decades that the engineering of microbial functions began to be performed in a rational and methodical manner, driven by technological advances in microorganism identification, characterization, and manipulation. The development of genetic engineering tools (gene cloning, recombinant DNA technology, or more recently CRISPR-Cas9) fueled the establishment and expansion of the field of synthetic biology. Synthetic biology has aimed to systematize the engineering of microbial systems with a dual aim: understanding the principles by which living organisms process information, and constructing synthetic biosystems with enhanced (or even novel) traits for concrete biotechnological applications (Khalil and Collins, 2010; Leonard et al., 2008). While progress has been made over recent years, the latter objective has remained elusive (Hanson and Lorenzo, 2023; Zakeri and Carr, 2015).

Synthetic biologists have been mostly concerned with the engineering of individual strains or particular subcellular units, such as proteins or metabolic pathways. Some of the earliest examples include the genetic manipulation of yeast (Hansen and Kielland-Brandt, 1996; Romano et al., 1985) and bacteria (McKay and Baldwin, 1990) for food and beverage fermentation, the development of enzymes with high substrate specificity or catalytic activity through directed evolution (Schmidt-Dannert and Arnold, 1999; Yano et al., 1998; You and Arnold, 1996), or the optimization of intracellular biochemical reactions for increased enantioselectivity (Reetz et al., 1997). More recently, engineered microorganisms have been developed for modern applications such as carbon sequestration (Hu et al., 2019) or cancer immunotherapy (Chowdhury et al., 2019).

Engineered microorganisms, however, present various important limitations. The expression of synthetic genetic circuits is often inevitably noisy and subject to crosstalk with the native machinery of the host cells — a problem that is accentuated as these circuits become more complex (Kwok, 2010; Slusarczyk et al., 2012). In addition, heterologous expression often comes at a fitness cost, making engineered microbial functions sensitive to purging by evolution or to exclusion from better-adapted competitor species. Even when fitness costs are negligible, mere genetic drift may disrupt the expression of synthetic functions if they do not confer a benefit to the host. Synthetic biologists have proposed formal rules for the design of biosystems (Slusarczyk et al., 2012) which attempt to mitigate the effects of noise (Perrino et al., 2021), crosstalk (Müller et al., 2019), or evolution (Bull and Barrick, 2017); however challenges remain to this day.

The prospect of engineering microbial communities (as opposed to single organisms) has emerged as a promising alternative (Großkopf and Soyer, 2014; McCarty and Ledesma-Amaro, 2019; Shong et al., 2012). Communities present several advantages with respect to individual strains. They enable the compartmentalization of functional components (i.e., division of labor), alleviating fitness costs on individual strains (Beck et al., 2022; Wang M. et al., 2022; Roell et al., 2019). Functions that emerge at the community level may also be more robust (or at least change in predictable ways) in the face of evolution of constituent species (McEnany and Good, 2024; Venkataram and Kryazhimskiy, 2023). Communities may often be able to resist invasions from external species (Wagner, 2022; Mickalide and Kuehn, 2019) or groups of species (Diaz-Colunga et al., 2022; Lechón-Alonso et al., 2021). Furthermore, the composition and function of microbial communities may be modulated without the need to genetically engineer any member species, minimizing concerns on the environmental or health hazards that genetically modified organisms may pose (EFSA et al., 2020).

Here we review a variety of strategies which have been proposed for optimizing the functions of microbial communities, many of them inspired by synthetic biology at the level of organisms and subcellular units. These strategies range from bottom-up approaches, where defined sets of (typically few) species are combined into consortia with the aim of maximizing a function, to top-down approaches, where a community (which can be of high complexity and even of undefined composition) is manipulated through rational interventions (San León and Nogales, 2022). We also review methods based on mathematical modeling, including mechanistic models (most notably metabolic models) as well as more recent data-driven models. With this, we seek to provide an accessible overview of the rapidly-expanding field of synthetic ecology. We aim to prompt readers to carefully consider the extent to which the lessons learned from synthetic biology should guide our path towards community-level engineering.

Perhaps the most commonly used strategy for community-level engineering has relied on the rational, bottom-up assembly of synthetic consortia. In short, based on some known traits of a set of microbial species/strains, a consortium is constructed with the aim of maximizing a target function, as well as (potentially) its ecological and evolutionary stability (Krause et al., 2014; Lajoie and Kembel, 2019; Li et al., 2024).

This trait-based approach is reminiscent of how early efforts at protein design were based on searching for optimal amino acid sequences starting from the basic principles of amino acid biochemistry (Song et al., 2023; Bryson et al., 1995) — i.e., the biochemical “traits” of each amino acid. Using biochemical reasoning to predict protein function from sequence can be seen as solving a “puzzle” by carefully examining each of the pieces, with each piece being a specific amino acid (Figure 1A, top panel). In the case of synthetic organisms, the pieces could be genes within an artificial plasmid, while for communities they would represent the different member species (Figure 1A, bottom panel).

The strategy of “solving the puzzle” has sometimes yielded good results. In some cases, consortia have been assembled by leveraging the natural capabilities of wild-type microbial species for performing specific tasks. For example, Park et al. used a two-species bacterial co-culture for the production of bioethanol by leveraging the natural ability of C. phytofermentans to hydrolyze cellulose and the potential of E. coli to ferment cellobiose catabolism byproducts into ethanol, respectively (Park et al., 2020). Alternatively, researchers have relied on the genetic manipulation of different member species/strains within a consortium. Examples include the construction of two-species/strain systems for the generation of photovoltaic energy (Zhu et al., 2019) or the production of resveratrol, a plant natural product (Camacho-Zaragoza et al., 2016). In this last example, two E. coli strains were engineered to express a complementary part of the resveratrol biosynthesis pathway. Genetic engineering has sometimes also served to enhance the stability of the community and facilitate the coexistence of its members, for instance through the imposition of obligate mutualisms (Pignon et al., 2024; Sgobba et al., 2018).

Beyond these and other biotechnological applications, it is worth noting that synthetic communities have also been used to address more fundamental questions in microbial ecology, such as how microbial interactions determine the structure and dynamics of communities (Hu et al., 2022; Van Vliet et al., 2022; Cordero and Datta, 2016). Researchers have used synthetic consortia as laboratory model systems to mimic natural communities in the soil (Coker et al., 2022) or associated with animal and human hosts (Bonillo-Lopez et al., 2024; Clark et al., 2021), among other contexts. This is again reminiscent of how synthetic organisms have also served purposes beyond biotechnology, such as providing insights on cellular information processing and signal transduction (Gao et al., 2023; Hanson and Lorenzo, 2023).

Yet, for biotechnology, the trait-based construction of synthetic communities presents important challenges. First, the traits expressed by an organism often depend on its ecological context, that is, which other species (and potentially at which abundances) may be present (Diaz-Colunga et al., 2024a; Yang, 2021). In addition, even if the contribution of each individual microbial cell to a community function was constant, species’ population sizes often vary differently across ecological contexts (Baichman-Kass et al., 2023; Sanchez et al., 2023), so the total functional contribution of a species may change depending on the presence/absence or abundance of other community members. Interactions between species can often present higher-order components, e.g., a third species may affect how two members of the community interact (Morin et al., 2022; Sanchez-Gorostiaga et al., 2019; Mickalide and Kuehn, 2019; Guo and Boedicker, 2016). In the “puzzle” analogy, it would be as if the shape of each piece changed every time we included a new one. Perhaps for this reason, the construction of synthetic consortia for biotechnology through trait-based approaches has been mostly limited to low-complexity communities, typically of two or three species/strains (Park et al., 2024; Park et al., 2020; Zhu et al., 2019). This also owes to the fact that the rational assembly of microbial consortia in high-throughput remains an experimentally tedious process. The development of new experimental methodologies (Diaz-Colunga et al., 2024a), including based on microfluidic devices (Kehe et al., 2019; Kehe et al., 2021), promises to facilitate this process and potentially expand the bottom-up approach to more complex communities.

Much like in the case of synthetic organisms, communities engineered through this trait-based approach may be disrupted by evolution, environmental fluctuations, or the influx of invader species (Amor et al., 2020; Shibasaki and Mitri, 2020). Several strategies have been proposed for enhancing stability. For example, it has been shown that synthetic communities engaging in division of labor exhibit increased stability when not only the fitness costs, but also the benefits of expressing a function are allocated evenly across member species (Wang M. et al., 2022). Stability may also be achieved by rationally modulating intercellular interactions (Deter and Lu, 2022; Karkaria et al., 2021; Kong et al., 2018; Wu et al., 2024), e.g., interspecies metabolic cross-feeding (Li et al., 2022; Park et al., 2024; Peng et al., 2024; Ziesack et al., 2019), or by imposing a defined spatial structure to physically separate different subpopulations (Wang L. et al., 2022).

It is notable how these challenges (eco-evolutionary stability, scalability beyond low-complexity constructs, etc.) are similarly faced when engineering single organisms. This is a direct consequence of the trait-based design of microbial communities being very explicitly based on engineering principles from synthetic biology (Johns et al., 2016; San León and Nogales, 2022) — which becomes particularly evident in the case of consortia made up by genetically engineered strains (Park et al., 2024; Sgobba et al., 2018; Camacho-Zaragoza et al., 2016). In this case, the reasoning is straightforward: the target function to express is encoded in a synthetic genetic circuit, but the burden of its expression is distributed across different strains (i.e., multiple “chassis”) instead of assigned to a single one. While the basic idea is appealing, we must consider its practical limitations if the end goal is to develop microbial communities for biotechnological purposes which can be deployed on a large scale. Over the past decades, synthetic biologists have devoted much effort to addressing the challenges of engineering single organisms (e.g., Perrino et al., 2021; Müller et al., 2019; Slusarczyk et al., 2012), with only limited practical success (Hanson and Lorenzo, 2023; Zakeri and Carr, 2015). As the field of synthetic ecology develops, we must carefully assess the risk of falling into a similar stage of stagnation.

Artificial selection, most notably directed evolution, has been used for decades to improve the traits of individual organisms or the properties of subcellular units such as enzymes (e.g., Schmidt-Dannert and Arnold, 1999; You and Arnold, 1996) (Figure 1B, top panel). Directed evolution mitigates many of the limitations of rational design approaches, as it bypasses the need for a detailed understanding of the underlying mechanisms that may determine the function/trait of a molecule/organism. The basic strategy can be conceptualized as an iterative exploration of a genotype space in search of optimal phenotypes, e.g., an exploration of a sequence-function landscape in the context of protein engineering.

Artificial selection at the microbiome level can be similarly framed as a guided exploration of a community structure space (defined as the set of all possible community compositions in terms of species presence/absence or abundance) in search of states that optimize a target community-level function. Such mappings between community compositions and functions have received the name of ecological landscapes or community-function landscapes (George and Korolev, 2023; Sanchez et al., 2023; Skwara et al., 2023), by analogy with the concept of genetic fitness landscapes.

In its most general form, an exploration of a community-function landscape through directed evolution would involve (i) ranking a set of microbial communities based on a target community-level function, (ii) selecting the best-performing ones, and (iii) propagating them into a new set of “offspring” communities, which can be ranked and propagated again in subsequent rounds of selection (Figure 1B, bottom panel). Despite its conceptual simplicity, this premise was not directly tested until the early 2000s, when Swenson et al. attempted to select for microbial ecosystems which efficiently performed functions such as the degradation of 3-chloroaniline (Swenson et al., 2000a) or the promotion of plant growth (Swenson et al., 2000b). More recent efforts have aimed to select microbiomes with the ability to induce early or late flowering in plant hosts (Panke-Buisse et al., 2015), or to degrade environmental pollutants (Arias-Sánchez et al., 2024). Community-level artificial selection has been typically implemented as a top-down strategy, where communities are selected and propagated without necessarily dissecting their species-level composition (Chang et al., 2021; Chang et al., 2020; Swenson et al., 2000a; Swenson et al., 2000b; Xie et al., 2019; Panke-Buisse et al., 2015). More recently, bottom-up variations have also been proposed, where the species-level composition of a community is well defined and experimentally manipulated in each round of selection (Arias-Sánchez et al., 2024; George and Korolev, 2023). While there exist promising studies of in silico community-level artificial selection (Chang et al., 2021; Lalejini et al., 2022; Xie et al., 2019), experiments have generally yielded only modest functional improvements, in some cases barely exceeding typical day-to-day fluctuations.

Selection (whether artificial or natural, at the level of organisms or groups) requires that there exists variation in the target trait/function, and that this variation can be passed from parents to offspring — i.e., that the trait is heritable to some extent (Lewontin, 1970). When organisms reproduce, they pass their genetic information to the next-generation, such that those phenotypes that are (at least partially) determined by the organism’s genotype can exhibit some degree of heritability. The very same process can also reintroduce trait variation through mutation or recombination, upon which selection can further act.

Directed evolution of communities, however, entails important nuances with respect to that of single organisms (Arias-Sánchez et al., 2019; Blouin et al., 2015; Sánchez et al., 2021; Xie et al., 2019). Communities, unlike organisms, cannot naturally self-replicate, and thus generating an “offspring” community from a “parental” one requires the intervention of the experimenter. The first attempts at microbiome breeding addressed the issue of community-level reproduction by taking inspiration from earlier experiments of group-level selection in small animal populations [e.g., of chickens (Muir, 1996) or beetles (Wade, 1976; Wade, 1977)]. In short, offspring populations were initialized from a random sample of individuals from the highest-functioning parental ones.

Yet, this approach was less successful when applied to microbial communities (Swenson et al., 2000a; Swenson et al., 2000b). This may be explained, at least partially, by an effect of population size (Sánchez et al., 2021): If the number of individuals sampled from a parent population is very large, all offspring populations may end up being compositionally very similar to one another, as the effect of stochastic sampling becomes negligible (Blouin et al., 2015). This issue is typically more prominent in microbial communities than in animal populations, as the former often have very large population sizes (e.g., a single colony can contain millions of microbial cells). The loss of structural (and therefore functional) variation across communities can naturally obstruct further selection (Blouin et al., 2015; Chang et al., 2020). Other mechanisms could also lead to the exhaustion of variation or even to functional collapse: for instance, evolution of member species within a community could have such effects under certain conditions (Shibasaki and Mitri, 2020; Venkataram and Kryazhimskiy, 2023; Xie et al., 2019).

Several strategies have been proposed to tackle this issue. Structural variation can be externally re-introduced into the offspring communities, for instance through the co-inoculation of invader species or groups of species, through the application of harsh population bottlenecks (Chang et al., 2021; Sánchez et al., 2021), or through the propagation of not only the highest-function community but also of sub-optimal ones (Xie et al., 2019). It has also been suggested that selection schemes inspired by evolutionary computing could aid in maintaining functional variation across communities, leading to better selection outcomes (Arias-Sánchez et al., 2024; Lalejini et al., 2022).

Community-level artificial selection, in principle, could substantially boost our ability to engineer microbiomes, the same way it revolutionized our ability to engineer enzymes (Schmidt-Dannert and Arnold, 1999). There is extensive theoretical and empirical evidence showing that artificial selection can act above the organismal level (Doulcier et al., 2020; Goodnight, 1990a; Goodnight, 1990b; Lewontin, 1970; Wade, 1976; Wade, 1977), and thus it is not immediately obvious why experiments of microbiome-level breeding have found only modest success in general. It is possible that current strategies for community-level selection may be limited by methodological aspects, such as the choice of community-level reproduction method or the maintenance of functional variation across communities. It is also possible that there exist more fundamental factors which may intrinsically limit community-level artificial selection. These could include stochastic fluctuations in species abundances, or the rapid emergence of “cheater” strains, especially when the expression of the target function is costly (Blouin et al., 2015; Shibasaki and Mitri, 2020; Smith and Schuster, 2019).

It is important to note that targets of selection, artificial or natural, do often have fundamental limits at the organismal scale. These limits are evidenced, for instance, by the common observation of diminishing returns at the level of organismal fitness — i.e., beneficial mutations having smaller positive fitness effects in genetic backgrounds which are already well-adapted (Chou et al., 2011; Schoustra et al., 2016; Wünsche et al., 2017). Diminishing returns have also been recently observed at the level of community function (Diaz-Colunga et al., 2024b), suggesting that when a community function is high, most interventions will tend to disrupt it rather than improve it. This could pose an obstacle towards community-level artificial selection. Such limitations may become more pronounced when the individual members of a community benefit from the expression of the target ecological function — for instance, when the function is the clearance of a toxin (Arias-Sánchez et al., 2024). When such organismal-level pressures exist, the function may strongly dictate the composition of the community in a very deterministic manner, leading to reduced across-community functional variation. This would then constrain further optimization through selection.

As promising as community-level directed evolution may be, further work is necessary before it can become a viable option for biotechnological applications. Are there intrinsic limits to selection at the level of microbial communities? If not, why has directed evolution not yet been successfully applied to microbiomes? If the reasons are purely methodological, there are perhaps reasons for optimism, as in silico simulations have proven useful to inform the design of selection protocols (Chang et al., 2021; Lalejini et al., 2022; Xie and Shou, 2021). These methods, however, remain to be tested empirically.

Most efforts for the optimization of microbial community functions have relied on manipulating its species/strain-level composition and/or the genetic architecture of community members. However, there is extensive evidence that abiotic environmental variables can strongly modulate the traits of individual microbes (Hu et al., 2021; Wasner et al., 2024; Yang, 2021), the interactions between species (Crocker et al., 2024; Ratzke and Gore, 2018), and therefore the dynamics, composition, and function of complex communities (Dal Bello et al., 2021; Estrela et al., 2021; Goldford et al., 2018; Hu et al., 2022; Silverstein et al., 2024; Sun et al., 2024). Thus, an alternative optimization strategy is to rationally engineer the environment that microorganisms inhabit (Sánchez et al., 2024; Silverstein et al., 2023; Silverstein et al., 2024) (Figure 1C).

One of the most paradigmatic examples of environmental engineering for microbial biotechnology is found in open fermentation systems (Li et al., 2014). In these, the premise is to manipulate a set of environmental conditions (e.g., pH, substrate availability, etc.) that will naturally select for microbial taxa which are able to carry out a desired function. The system is left unsterilized, open to the influx of environmental microbes, rather than inoculated with a specific set of strains previously chosen by the experimenter. This approach has been traditionally used in food and beverage fermentation, and has more recently been employed for applications such as the production of butanol from butyrate (Pinto et al., 2022) or the synthesis of enzymes (Qureshi et al., 2017) or lactic acid (Wang et al., 2016) from food waste.

The challenge of manipulating microbial environments lies in managing their highly multidimensional nature. Microbial growth and functional profiles may depend on a plethora of abiotic factors, including nutrient availability (Okano et al., 2019; Skonieczny and Yargeau, 2009; Zhu and Dai, 2024), temperature (Fu et al., 2022; Sun et al., 2024), pH (Pinto et al., 2022; Ratzke and Gore, 2018), the presence of antimicrobial compounds (Athamneh et al., 2014), or the spatial structure (or lack thereof) in their habitat (Pignon et al., 2024; Van Vliet et al., 2022). As an additional complication, the effect of these factors can often be highly non-additive. A paradigmatic example of this non-linearity is the observation that different antibiotics can act synergistically in combination (Cacace et al., 2023; Lázár et al., 2022; Yeh et al., 2006), or modify microbial responses to other environmental variables such as temperature (Cruz-Loya et al., 2021).

Ultimately, these challenges are similar to those faced by synthetic biologists when engineering single organisms (or communities) from the bottom-up. Interactions between environmental components are, at least conceptually, reminiscent of the interactions that exist between genes within an organism (or between member species within a community). These interaction networks can be very complex and of high dimensionality in all of these cases. Owing to this analogy, strategies for environmental design have drawn inspiration from the bioengineering of organisms or subcellular components (Sánchez et al., 2024). For example, genetic algorithms have been used to select optimal environments. In these cases, a set of environmental variables (pH, salinity, concentration of vitamins and minerals, etc.) were manipulated in each round of selection, and the best environments were propagated into subsequent generations. Vandecasteele et al. followed this principle to identify environments where a microbial community maximized the degradation of a synthetic dye (Vandecasteele et al., 2008). In another example, Kucharzyk et al. followed a similar approach to optimize the degradation of perchlorate, both when the function was performed by a single strain of Dechlorosoma sp. or by a complex microbial community (Kucharzyk et al., 2012).

Modeling tools can also be used to inform environmental design strategies. For example, Pacheco and Segrè developed a computational method combining metabolic modeling and genetic algorithms to find optimal environmental compositions for target community functions (Pacheco and Segrè, 2021). Mathematical models of community dynamics that are extensively used by microbial ecologists may explicitly incorporate environmental variables, e.g., the secretion of metabolic byproducts to the environment is specifically included in microbial consumer-resource models (Marsland et al., 2020b; Marsland et al., 2020a) or in some dynamic community-level metabolic models (Dukovski et al., 2021). These could be useful to inform the construction of environments which optimize microbial community functions such as the efficiency of substrate utilization or the production of specific secondary metabolites.

Data-driven models have also been developed which aim to infer optimal environmental compositions from partial observations (Chen et al., 2009; Jiménez et al., 2014; Kikot et al., 2010; Skonieczny and Yargeau, 2009; Zhou et al., 2023). These methods have the advantage of not requiring information on the specific biological mechanisms which may drive microbial responses to their environment. Yet, the typically vast number of potential environmental factors makes it so these models often must be trained with very limited empirical data. They have thus generally taken very simple forms, e.g., linear regressions with the variables being the effects of each environmental factor, at most incorporating pairwise interactions between environmental components (Jiménez et al., 2014; Kikot et al., 2010). In practice, this has limited their application to relatively simple settings including few environmental components. As is the case with the bottom-up assembly of communities, recent methodological advances could facilitate the construction of environments in high throughput (Diaz-Colunga et al., 2024a; Sánchez et al., 2024) and the expansion of the scope of these models.

Mathematical and computational models have been widely used by microbial ecologists to address questions such as inferring species interactions from co-ocurrence networks (Faust and Raes, 2012), assessing microbial coexistence and community stability (Akjouj et al., 2024), explaining the emergence of community-level properties from complex interactions between species (Van Den Berg et al., 2022), or reproducing the relationships that exist between biodiversity and function within microbial communities (Marsland et al., 2020b). In the context of biotechnology, modeling has served to guide the construction of microbial consortia that efficiently deliver target functions (Figure 1D), such as the production of relevant metabolites (e.g., Jones et al., 2016; Clark et al., 2021).

Models can be cataloged according to different criteria (Van Den Berg et al., 2022). Here, for the sake of simplicity, we broadly classify them into two categories: (1) models based on species’ traits and interaction mechanisms, and (2) models which leverage statistical features of microbial communities to reproduce and predict community-level properties while remaining mechanism-agnostic (Figure 2).

Throughout this review, we have discussed the parallels that exist between the engineering of biological systems at the ecological scale and below. It is clear that synthetic biology tools have had an enormous influence on the development of strategies for the design and optimization of community-level microbial functions. In particular, a very common strategy has been to assemble synthetic consortia from the bottom-up, building on the available phenotypic information of the constituent parts (i.e., the species or strains). This approach has sometimes been informed by mechanistic models of ecological interactions, and/or relied on the genetic engineering of community members. While this is naturally a reasonable starting point when attempting to engineer microbiomes, it is also important to acknowledge its limitations.

Communities, like all biological entities, are complex systems, and thus their properties and functions are often difficult to explain simply from those of their parts. Applying engineering principles to their design is therefore nuanced, as in doing so there is an underlying assumption of modularity and scalability that may often not hold. This, of course, also applies at the organismal scale and below: a single cell is a very complex system in itself, and thus combining different organism types into a community means building complexity on top of complexity. Synthetic biology has arguably not yet fulfilled its promise of developing practical solutions for biotechnology on a large scale. It is thus imperative that we ask whether applying similar principles at the level of communities will yield better results, or if, on the other hand, there are fundamental limitations to this approach when it comes to biological systems.

Top-down strategies, as well as data-driven modeling, could in principle be more suitable for the task of prediction and optimization. Their main advantage lies in the fact that these methods do not rely on mechanistic information. For example, community-level directed evolution could be implemented without even characterizing the composition of the community nor the interactions between its member species. In practice, however, this strategy has only yielded modest results, and further work is necessary to identify the factors which may have limited its success. Data-driven models have been used extensively for predicting biological function below the ecological scale, but more rarely for optimizing microbial community functions. Their main limitation is the difficulty in extracting relevant biological insights from them, as these models often tend to operate as “black boxes”. In any case, this may be a lesser consideration if our primary focus is to optimize biotechnological processes.

The variety of available strategies for community-level engineering underscore the importance of choosing the appropriate approach if we wish to develop practical, viable, and sustainable solutions for open challenges in biotechnology. Taking inspiration in other areas of biology beyond microbial ecology can be fruitful, but we must carefully consider the limitations we may face. Synthetic ecology emerged as an extension to synthetic biology that promised to alleviate the limitations of the latter, in particular with respect to scalability and robustness. Yet, the rational design of microbial consortia has faced similar obstacles, perhaps because it has been approached using similar bottom-up thinking. As the field of synthetic ecology develops, it will be important to devise new optimization strategies that embrace and deal with the underlying complexity of microbial communities — and of biological systems at all scales.