Visualizing the inner life of microbes: practices of multi-color single-molecule localization microscopy in microbiology

The emerging field of Single-Molecule Localization Microscopy (SMLM) allows to resolve biological structures at the nanometer scale and to monitor molecular interactions in the millisecond range. To tackle the diverse biological and technical demands of specific research questions, a growing number of practical SMLM tools have been developed over the last years. This is nicely illustrated by recent general reviews on super-resolution microscopy developments [1–3] as well as by reviews focusing on photoswitchable fluorophores needed for SMLM techniques [4–6].

However, since each research field has its own particularities, only a subset of the overall SMLM toolbox matches the given, field-specific requirements. An example of such a specific research area is the field of microbiology. Technical demands are largely shared by this field of biology, which encompasses all known microorganisms. Robust SMLM tools for studying microbial cell biology all face the challenges that microorganisms, in general, are (1) small and densely packed single-cell organisms protected by robust cell walls, (2) show rather low protein copy numbers often combined with specific autofluorescence or background of colorful pigments when compared with, e.g. mammalian cells and (3) possess rather fast growth rates accompanied by rapid metabolism rates. However, many model organisms provide widely established genetic modification tool sets facilitating genetic target labeling.

While current (microbial) SMLM studies mostly examine the dynamic and structural properties of a single target [7–9], one could argue that biological processes, in general, rely on interactions of multiple components. Therefore, establishing reliable methods for multi-color SMLM is becoming increasingly more important.

Hence, in this review, we exclusively focus on the multi-color single-molecule imaging and tracking studies on microbial cell biology published to date and discuss their utilized tools’ advantages and disadvantages as well as possible pitfalls. Moreover, we highlight recent and potential future developments within the field.

Fluorescence microscopy is a powerful tool to investigate biological systems as it allows to monitor the spatiotemporal behavior of virtually any, fluorescently labeled biomolecule of interest at high specificity. Nevertheless, even for specifically chosen, labeled molecules, details below 200 nm remain unresolved due to the diffraction barrier of light microscopy (Figure 1a, left).

SMLM techniques achieve a higher resolution than conventional fluorescence microscopy methods by controlling the fluorescent emission of individual fluorophores. By a ‘blinking signal’ strategy, images of varying, small subsets of fluorophores can be acquired over time and the centroid of each, individual fluorescent signal can be localized at high precision to create detailed molecular maps and super-resolved SMLM images at the nanometer scale (Figure 1a, right).

The most commonly used fluorophores in SMLM imaging are fluorescent proteins (FPs) and organic dyes. FPs offer the advantage of genetic labeling and consist of a β-barrel protein structure protecting the chromophore in its middle (Figure 1b, left). Organic dyes, on the other hand, usually offer higher fluorescence quantum yields than FPs and can be customized during their chemical synthesis, e.g. fine-tuning their spectral properties by a distinct delocalized π-electron system design and increasing their solubility and photostability by additional groups flanking the chromophore. Furthermore, their application is highly flexible as variable labeling groups can be added (Figure 1b, right).

Most critical in SMLM imaging is the tight control of fluorophore blinking in order to resolve individual fluorescent signals. In general, all SMLM methods can be categorized into reversible and irreversible blinking strategies (Figure 1c). For reversible blinking, two main strategies exist: fluorophores can either be imaged while reversibly binding and unbinding their targets (such as in Points Accumulation for Imaging in Nanoscale Topography (PAINT) microscopy [10]), or they can be photophysically or photochemically switched between a fluorescence-emitting and a dark state by specific light illumination and/or imaging buffers (such as in direct STochastic Optical Reconstruction Microscopy (dSTORM) [11] or by several Dark State Pumping and Recovery methods [12–16] (to which we, for simplicity, refer to by using the acronym DaStPuRe) (Figure 1c, left).

The second type of fluorophores (such as FPs used in photoactivated localization microscopy (PALM) [17]) is irreversibly photoactivated or -converted from a dark or initial fluorescent state of different color into a fluorescent state of desired read-out color (Figure 1c, right). The majority of irreversible photoactivating/converting fluorophores require UV-light illumination for changing their state. Nevertheless, recently, a novel mechanism called primed photoconversion was discovered where a special class of FPs was found to be photoconvertible via an intermediate dark state upon irradiation with less phototoxic blue and infra-red (IR) light [18,19]. Detailed reviews focusing on the photophysical and/or photochemical specifics to blink fluorophores in SMLM imaging can be found elsewhere [2,4–6].

Most studies introducing new multi-color SMLM tools are conducted in mammalian cell systems, as they contain defined and accessible nanostructures (such as cytoskeletal networks [20,21], clathrin-coated pits [22], nuclear pores [23,24] or the human glycine receptor [25]) to benchmark the tools (e.g. SMLM-suitable fluorophores such as in [2], table 2). Since a direct technology transfer to microbial targets—also from our own experience—often proves to be challenging, we compiled a systematic review of all multi-color SMLM work on microbial targets which are published to date and summarized them in Table 1 for bacteria and Table 2 for fungi. Furthermore, we assembled two visual collections of ‘best-practice’ examples: one for structural studies (Figure 2) and one for dynamic single-particle tracking (SPT) studies (Figure 3).

We use this systematic summary as a basis to discuss and compare the strengths, similarities and differences between current approaches and with respect to the inherent requirements for specific microorganisms. As one of the most crucial decisions when planning a new SMLM study is the choice of label and labeling technique and the available options are one of the most limiting factors in current study designs, we additionally compiled a figure introducing and explaining them for the different classes of biomolecules—proteins, carbohydrates, lipids and nucleotides—and discuss them along with our literature review below (Figure 4). The references in the caption of Figure 4b–d link the interested reader to original literature which uses those methods for targets in microorganisms.

At a first glance, our compilation demonstrates that multi-color SMLM work is still exceptional in microbiology as we count only 40 studies (Tables 1 and 2). Most multi-color studies either investigate only one target at high resolution accompanied by a diffraction-limited structural reference such as the nucleoid, cell membrane or a single-spot-forming protein cluster (15 studies, examples are Figure 3a–c[43,47,49]) or two targets at high resolution (20 studies, examples are Figures 2b,c,f,h–k,m and 3d,f [12,13,26,29,33,35,40,53,54,56]) also often supported by a reference (Figures 2e,l and 3e [28,41,60]). Three targets in SMLM resolution are rare (five studies, examples are Figure 2a,d,g [14,26,28]), and studies aiming at four or more targets are non-existent (to our knowledge).

The vast majority was conducted in bacteria (30 studies with 18 studies at least dual color and 12 studies single-color SMLM imaging plus a diffraction-limited reference, Table 1), with the model organism Escherichia coli having the biggest share (13 studies). In contrast, only 12 studies focused on fungi biology (eight at least dual color, four are single-color plus reference, Table 2), among which the studies on Saccharomyces cerevisiae dominated by number (seven studies).

In general, for both fungi and bacteria, the great majority of studies is conducted in model organisms where laboratory cultivation techniques are well established and for which a large number of tools, e.g. genetic manipulation strategies have been developed. Multi-color SMLM studies under more complex culturing conditions (e.g. under medical relevant conditions or using non-model strains and exploring, e.g. biofilms or co-cultures like the human microbiome) are scarce (Figure 2l) [30,41].

Furthermore, the studies are largely focusing on structural SMLM imaging exploring spatial molecular organizations (126 targets in total), mostly conducted in chemically fixed cells (102 fixed versus 24 live targets, Figure 2 and Tables 1 and 2). Multi-color SPT studies are rare (eight targets in total, Figure 3 and Tables 1 and 2). These SPT studies investigating molecular interactions were mostly conducted using single-color sptPALM accompanied by a diffraction-limited reference [43,47] and in one case super-resolving the nucleosome via dSTORM [26]. Investigating the dynamics of two targets was conducted by either orthogonal photoactivation modes in a subsequent manner [28] or bimolecular fluorescence complementation-PALM (BiFC-PALM), where two biological targets were each labeled with one component of a split FP [36].

Reviewing the imaging routines, fluorophore combinations were either imaged in parallel by splitting the signals of appropriate fluorophore pairs onto two areas of the camera chip (14 studies), or more commonly in a sequential imaging mode using the same detection path (28 studies). For the latter, also fluorophores of similar emission spectra can be used when separating them either by different photoactivation/conversion modes [28] or by sequential addition or exchange of probes (e.g. sequentially added dyes for PAINT imaging [26–28] or 5-ethynyl-2′-deoxyuridine (EdU) staining of DNA [26,27]. For both methods, this brings the advantage of avoiding chromatic aberrations. When adding fluorophores sequentially, a position-stabilizing autofocus system or a robust position-refinding routine (by, e.g., re-recognizing unique landmarks) is required.

A successful experimental multi-color design needs a thought-out choice of fluorophore combinations and labeling methods. Well-established genetic tools (Figure 4a(i)), which include powerful methods like exploiting the microbes’ own homologous repair mechanism for chromosomal recombinant replacements of native target genes, exist for most model microorganisms [69]. Also, while being densely packed into the small microbial volume, absolute copy numbers of microbial protein of interests (POIs) can be rather low [70] which demands for highly specific labeling approaches. Performing statistics on all published studies summarized in Tables 1 and 2, it is thus not surprising that chromosomal tags were used for 90 out of 154 targets, 72 of which as recombinant replacements at the original locus under the original promoter and 15 as ectopic additional copies, integrated into the chromosome at another locus. These are FP fusions to a large extent (77 POIs, Figure 4b(v)) with only a minor portion of protein tags (10 chromosomal replacements, Figure 4b(vi)) and fluorescent repressor–operator system (FROS) arrays (4 ectopic additions). Additional 22 genetic fusions were introduced into the cells by plasmids carrying recombinant genes. Other targeting methods are only seldom applied: We find 14 immunofluorescence stainings (10 antibody stainings targeting native epitopes and 4 anti-GFP/-RFP nanobody probes, Figure 4a(iii) and b(ii)), 17 uses of on- and off-binding PAINT probes (Figure 4c(ii) and d(ii)), and 12 incubations with target-specific and fluorescently labeled compounds (Figure 4a(ii), b(ii–iv), c(i–iii), d(i–iii)). For the latter, various labeling methods were used, such as (1) in vitro-labeled compounds (e.g. in vitro-labeled carbohydrates specific for the cell wall (Figure 4c(i)) [30], target-specific drugs like WGA [37,41] and concanavalin A (Figure 4c(iii)) [53,54], actin-binding phalloidin or lifeact (Figure 4b(iii)) [15] or smFISH oligonucleotides (Figure 4d(i)) [34]), (2) target-binding fluorophores (e.g. TOTO-3 targeting DNA (Figure 4d(ii)) [37]), and (3) analogs carrying, e.g. alkene or azide groups that can be stained by click chemistry approaches afterwards (Figure 4b(ii,v), c(iv), d(iv)). For the latter, the most often used one is the thymidine analog EdU incorporating into nascent DNA (Figure 4d(iv)) [26,27,29]).

Turning the perspective to the selected molecules of interest, FPs strongly dominate the chosen combinations when imaging protein targets. Here, PAmCherry fulfills a special role: it is the only commonly applied photoactivatable FP of the DsRed family [5]. These photoactivatable FPs photoactivate from an initial dark, premature chromophore state into their fluorescent state in the ‘red’ part of the visible spectrum (∼580–660 nm wavelength range). In a multi-color experiment, PAmCherry thus stands out from the RFPs from the Kaede family (e.g. mEos2 or 3, Dendra2 or mMaple(3)) as these green-to-red photoconverting FPs fluoresce in the ‘green’ part of the visible spectrum (∼490–560 nm wavelength range) in their initial GFP-like form [5]. This makes PAmCherry an almost obligatory choice for (1) green/red dual FP pairings (being either paired with eYFP (Figure 2a,c) [12,14,31] or Dronpa (Figure 2b) [33]) and for (2) dual red FP pairings which are separated by orthogonal illumination modes when photoactivating PAmCherry by UV-light and photoconverting a FP from the Kaede family by primed photoconversion (only possible for threonine 69 variants [18], e.g. using mEos3.2-A69T (Figure 2d,e) [28,60], or Dendra2 (Figure 3e) [28]).

Another commonly used combination is far-red (∼650–730 nm wavelength range) dyes together with red FPs (10 studies). Here both PAmCherry and the Kaede-like proteins are equally popular choices as the green spectral channel can be neglected (Figures 2f,g,h,i and 3d). The choice of the far-red dye is dominated by AF647—also incredibly popular in single-color dSTORM experiments (see [2], table 2)—and leads to a remarkable count of 9 out of 10 studies (examples in Figures 2f,g,h,i,k and 3d).

Cellular components other than proteins are usually reliant on non-genetic targeting tools and thus are mostly investigated by dSTORM and PAINT studies. In case of dual dSTORM experiments relying on organic dyes, it becomes apparent that the membrane-permeable, spectrally distinct red/far-red dye combination TMR(-Star)/ATTO655 (Figure 2m, four out of six red/far-red pairings [35,38,39]) were preferred over AF647 paired with spectrally close dyes in a spectral demixing imaging mode (AF700 (Figure 2k) [53] or AF750 [55]) or green/red [30], green/far-red [15,37] or STORM activator/acceptor (Figure 2l) [41] dye combinations.

SMLM visualization of the cell membrane was mainly conducted by using the dynamic on- and off-binding of NileRed for PAINT imaging (8 out of 17 studies of the cell membrane, Figures 2a,d,j,g and 4c(ii)[14,26–29,32]), while labeling of DNA was performed using either (a) transiently binding dyes (e.g. JF646-/JF549- or JF503-Hoechst (Figures 2j and 4d(ii)) [29], (b) EdU stainings with AF647 (Figures 2g, 3d and 4d(iv)) [26,27], (c) blinking intercalator TOTO-3 (Figure 4d(ii)) [37]) or (d) diffraction-limited fluorescent probes (DAPI (Figures 2l and 3e) [28,41], Sytox Green [42,48] or Orange (Figure 2d) [28], Syto-16 (Figure 3b) [47] or propidium iodide [37]). The most popular diffraction-limited reference is DNA (nine times), while the co-imaged SMLM probe in most cases was a red FP (seven times, out of which four times PAmCherry was chosen [28,47,48]). One study combined membrane and DNA probes for dual-PAINT, accompanied by either a FROS array spot or a PAmCherry-labeled POI (Figure 2j) [29].

For the few triple-color studies, until now dual-FP/PAINT (Figure 2a,d) [14,28], dual-PAINT/PALM [29] or PALM/PAINT/dSTORM (Figure 2g) [26,27] approaches were established.

Based on our observations while comparing the 40 publications applying multi-color SMLM studies in microbiology, we compiled a best practice guideline (Figure 5a and Table 3). Whereas the rationale behind this guideline can be generally used to design multi-color SMLM experiments for any organism, our examples are focused on studying microbes. Importantly, single-molecule imaging and tracking methods can yield a manifold of detailed answers about individual molecules and their interactions at a high spatiotemporal resolution in situ, but they are not every-samples techniques. Experimental factors such as single-molecule sensitivity beyond (low) background, tight photoswitching control of fluorophores or reliable corrections for drift and chromatic aberrations are strong determinants for image quality. Achieving good results for several channels in multi-color imaging is multiplying the overall efforts to be undertaken. To further illustrate our general scheme, we thus as well added a practical example of study design based on our own experience (Figure 5b).

First of all, it is worthwhile to invest a lot of resources into the study's design and to precisely formulate and specify the biological question one aims to answer. This entails a profound knowledge of the underlying biological system and often goes in hand with a strong hypothesis about observations to be expected (Figure 5a, upper box). It should be clear whether the observation of a structure and/or the dynamics of how many targeted molecules leads to a relevant investigation and which spatiotemporal resolution is required to acquire the data aimed for (Figure 5a, TASK 1 and 2).

Here, a higher temporal resolution or measuring the axial position often goes hand in hand with the trade-off of a lowered lateral spatial resolution due to lowered signal-to-noise (S/N) ratio of individual single-molecule fluorescence [1]. Often, one can already ‘guestimate’ from the target characteristics and its cellular environment where some technical hurdles might appear, e.g. thick cell walls might hinder staining, low pH, e.g. in the periplasm, lowers fluorescence read-out or colorful microbial pigments (e.g. carotenoids, melanins or flavins [73]) can superpose fluorescence in certain spectral ranges. In cases where both target protein termini are functional domains (as often encountered for membrane receptors), genetic tagging of either of them will most likely interfere with the proteins’ biology. In such cases, internal loop structures could be a better-suited spot for recombinant fusions, as has been done for, e.g., MreB in E. coli [74–76].

Furthermore, reflecting target abundances, their replenishment and accessibility can be of large importance: Is a native expression from the native gene locus possible and favorable (e.g. for measuring stoichiometry and cellular organization) or is an ectopic expression better suited for the planned investigation (e.g. exploring the DNA binding affinities of proteins in large statistics facilitated by overexpression and irrespective of their native copy number)? Is the time of POI folding or POI lifetime known? Too fast protein turn-over can prevent the use of FPs due to their typically rather long maturation time needed to properly arrange their chromophore and thus their ability to fluoresce [77,78]. In case of low molecular abundances or co-localization studies of several partners, highly efficient and specific labeling becomes a key factor. For example, two interaction partners, both labeled with a realistic efficiency of 50% would yield at best—when forming a permanent, static complex—only maximal 25% of positive co-localization (here we think of co-localization in an SMLM-specific definition of molecules being co-localized within an ‘interaction radius’ that takes localization precision and post-processing errors, labeling linkage distances, protein sizes and position of label attachment into account [79,80]). This observed co-localization of molecular partners easily drops further, e.g. for dynamic on/off-binding interaction equilibria dependent on different molecular conformations or when being misled by unspecific staining artifacts or lowered by slow FP maturation times. Here, counteractive knowledge on how to accumulate or arrest target molecules in certain molecular states by environmental changes (e.g. pH, temperature and nutrition), specific drug treatments (e.g. antibiotics compromising transcription, replication and cell wall organization) or by protein mutation might help the experiment. Additionally, target density easily restricts label choices. A clustered target in dense substructures demands for a high spatial resolution and tighter control of fluorophore read-out. Clustered targets can also be more challenging to label due to steric hindrances or label-dependent artificial aggregation artifacts as compared with an evenly distributed target within the cytoplasm [21,78,81–83].

When imaging living microorganisms, the shortest possible and least disturbing read-out option is desirable to avoid an excess of phototoxicity effects or changes in observed biology, e.g., the sensitivity of the investigated organism for specific wavelengths should be tested. Fast dynamics (e.g. free cytosolic Brownian diffusion or fast active transport) need a high temporal sampling to be resolved and to avoid confinement effects of the small microbial volumes. Contrarily, POIs slowed down by interactions with other cellular components (e.g. nucleoid-associated proteins or larger protein complexes) or by the viscosity of a certain compartment (e.g. in the membrane) allow for higher spatial resolution by the improved S/N of fluorescence read-out of slower acquisitions. Thus, it remains challenging but important to identify suitable sampling rates and read-out densities to correctly separate overlapping trajectories, to avoid confinement effects and to ensure being faster in read-out than biological alterations.

Transferring all mentioned requirements that can interfere with each other as best as possible into a practical experimental plan (Figure 5a, TASK 3) is the crucial key for a successful single-molecule sensitive microscopic study displaying its full potential. Thus, the more a priori knowledge of the measures needed to answer a specific biological question and of the characteristics of the target and its environment we have, the more straightforward is the selection of an appropriate combination of single-molecule targeting and read-out methods as discussed in the next paragraph (and in Figure 5a, lower box).

When imaging multiple targets by multi-color SMLM one first has to choose groups of targets being imaged in parallel so that the signal of two sorts of fluorophores is detected at the same time, and/or sequential imaging, where the different targets are measured independently from each other and fluorescent signals are detected in a successive manner. For structural studies conducted in fixed samples which can be regarded as ‘frozen biology’ snapshots of immobile targets, this choice is only dependent on the selected fluorophore and labeling combination. Here, sequential imaging allows for subsequent addition of fluorophores which reduces channel cross-talk and prevents preterm photobleaching of only later read-out fluorophores. Furthermore, chromatic aberration can be circumvented by ‘reusing’ the same color channel [14,26–28]. Parallel imaging, on the other hand, reduces imaging times and allows for parallel drift correction. Chromatic aberrations can also be avoided by spectral demixing [53,55]. Next to the extremes of sequential and parallel imaging modes also an alternating read-out mode by orthogonal activation schemes can be applied, either using one [41] or several color channels [31,79].

When imaging living samples, parallel imaging is the primary choice for highly dynamic samples but technically highly limited. Here, dual-color sptPALM of interaction partners in microbes has been achieved by BiFC [36]. When using sequential imaging, at least one target should be considered temporally invariant [13,14,28,32,39,54], forced to be immobile by fast fixation [26,29,31], or both targets have to be regarded to be in dynamic equilibria states for the time of the experiment [28].

Next to the biological constraints, technical factors restrict the repertoire of available fluorophore combinations that can be reliably used in multi-color SMLM (Figure 5a, lower box). Transferring the ideal list of desired fluorophore and labeling properties to the, in reality, still rather limited number of existing multi-color combinations of well-performing strategies usually only leaves a few choices—if any at all—and often requires simplification of the original experimental plans (returning to the first phase of study design as depicted in the upper box of Figure 5a).

Generally, as already introduced in Figure 4a, a fluorescent marker can be brought into the biological system by genetic fusions or by staining using immunolabeling, specific drugs, analogons or oligonucleotides dependent on the type of molecule of interest (protein, lipid, nucleic acid, etc.). Genetic fusions provide specific labeling in a one-to-one ratio which allows to visualize low abundant POIs or to quantify protein numbers (correcting for under- and overcounting effects [84]), whereas staining bears the risk of insufficient or unspecific labeling (e.g. due to dye charges, probe sizes or hydrophobicity). Only a few fluorescent dyes suited for SMLM studies are cell membrane-permeable, such as TMR, ATTO655 or the JF dye family [29,30,35,38,39]. Consequently, introducing dyes at high staining efficiencies into the crowded microbial organisms is challenging and dye delivery, as well as residual dye removal, needs to be assisted by membrane permeabilization or electroporation and cell wall digestion. Genetic fusions should be checked for growth and functionality deficiencies as some FPs tend to oligomerize [21,78,81–83] or might sterically hinder the protein function(s), e.g. shown for MreB in E. coli [74–76]. Finally, fluorescent dyes are commonly brighter and more photostable than FPs, improving read-out, but require specific switching buffers for photoblinking, which can be tricky to apply to live samples or can be toxic (e.g. depleting oxygen or adding strong reductants [2,6]). Here, different dyes might require different switching buffers which prevent their combined use. Also, FPs are influenced by switching buffers, e.g. oxygen removal prevents folding and reductants induce increased blinking [85,86]. Clever pairings of fluorophores and molecules of interest can compromise some drawbacks, e.g. when labeling and imaging a low abundance POI tagged with an FP first and only then staining the structural reference with a dye or when imaging the most dynamic POI by the brightest fluorophore of a chosen combination.

Nevertheless, taken all these limitations and requirements, it is not surprising that only a few fluorophore combinations perform well and lead to a full biological study after successfully passing TASK 4 and 5 in Figure 5a. These working combinations appear repeatedly in our literature review and often are specifically tailored. For example, dye staining is preferred for outer cell staining such as the cell wall, or brightness is often traded for probe specificity when using an FP tag for otherwise difficult-to-label POIs. In this respect, Tables 1 and 2 give a good overview of current working strategies and at the same time highlight the need for further probe developments.

AF

Alexa Fluor

BiFC

bimolecular fluorescence complementation

DaStPuRe

dark state pumping and recovery, illumination with high laser power forces the fluorophores into a dark state

dSTORM

direct stochastic optical reconstruction microscopy

EdU

5-ethynyl-2′-deoxyuridine

FP

fluorescent protein

FRET

fluorescence resonance energy transfer

FROS

fluorescent repressor–operator system

HILO

highly inclined and laminated optical sheet

JF

Janelia Fluor

PAINT

point accumulation for imaging in nanoscale topography

PALM

photoactivated localization microscopy

PC-PALM

PALM using primed photoconversion

POIs

protein of interests

ROI

region of interest

S/N

signal-to-noise ratio

smFISH

single-molecule fluorescent in situ hybridization

SMLM

single-molecule localization microscopy

SPT

single-particle tracking

sptPALM

single-particle tracking PALM

TIRF

total internal reflection fluorescence

UV-PALM

PALM using UV-light photoactivation/conversion

I.V., J.W. and U.E. performed the literature review and created the tabular overviews, I.V. and J.W. conceived the figures, I.V. and U.E. wrote the manuscript with the help of J.W.

The authors gratefully acknowledge the Max Planck Society, SYNMIKRO and the Fonds der Chemischen Industrie for financial support.

We sincerely hope that we did not overlook any multi-color SMLM study conducted in microorganisms and would like to apologize to the authors if we did not include them in this summary. We furthermore thank David Virant, Bartosz Turkowyd and Alexander Balinovic for discussions and critical reading of the manuscript. Finally, we would like to thank all authors that provided us with figure data from their original publications to compile the overview figures 2 and 3.

The authors declare that there are no competing interests associated with the manuscript.