In vivo Validation of Bimolecular Fluorescence Complementation (BiFC) to Investigate Aggregate Formation in Amyotrophic Lateral Sclerosis (ALS)

Amyotrophic lateral sclerosis (ALS) is a form of motor neuron disease (MND) that is characterized by the progressive loss of motor neurons within the spinal cord, brainstem, and motor cortex. Although ALS clinically manifests as a heterogeneous disease, with varying disease onset and survival, a unifying feature is the presence of ubiquitinated cytoplasmic protein inclusion aggregates containing TDP-43. However, the precise mechanisms linking protein inclusions and aggregation to neuronal loss are currently poorly understood. Bimolecular fluorescence complementation (BiFC) takes advantage of the association of fluorophore fragments (non-fluorescent on their own) that are attached to an aggregation-prone protein of interest. Interaction of the proteins of interest allows for the fluorescent reporter protein to fold into its native state and emit a fluorescent signal. Here, we combined the power of BiFC with the advantages of the zebrafish system to validate, optimize, and visualize the formation of ALS-linked aggregates in real time in a vertebrate model. We further provide in vivo validation of the selectivity of this technique and demonstrate reduced spontaneous self-assembly of the non-fluorescent fragments in vivo by introducing a fluorophore mutation. Additionally, we report preliminary findings on the dynamic aggregation of the ALS-linked hallmark proteins Fus and TDP-43 in their corresponding nuclear and cytoplasmic compartments using BiFC. Overall, our data demonstrates the suitability of this BiFC approach to study and characterize ALS-linked aggregate formation in vivo. Importantly, the same principle can be applied in the context of other neurodegenerative diseases and has therefore critical implications to advance our understanding of pathologies that underlie aberrant protein aggregation. Supplementary Information The online version contains supplementary material available at 10.1007/s12035-020-02238-0.


Aggregate Formation in Disease
Protein misfolding and aggregation are hallmarks of many different neurodegenerative diseases; however, the precise mechanisms linking protein aggregation and neurotoxicity are largely unknown (reviewed in [1][2][3]). Importantly, such an aggregation pathology is not exclusively linked to the central nervous system, as many other diseases such as type 2 diabetes, inherited cataracts, and myopathies display similar protein abnormalities, making protein aggregation a key concept in biological processes [1,4].
Protein aggregation is hypothesized to be caused when a particular protein folds into a stable alternative or intermediate conformation and starts to accumulate intra-or extracellularly. Protein misfolding and aggregation is hypothesized to be influenced by several scenarios, including post-translational modifications like protein hyperphosphorylation or SUMOylation (reviewed in [3,5]), disease-linked mutations leading to changes in the amino acid sequence (reviewed in [6]), prion-like behaviors (the ability to transmit their misfolded shape onto wild-type variants of the same protein) (reviewed in [4,7]), pathological increase in protein concentration (also termed "protein supersaturation") [8], and/or deficiencies in the proteasome or autophagy (protein clearance pathways) [6]. In addition, cellular challenges such as an increase in oxidative stress (reviewed in [9][10][11]), increased endoplasmic reticulum stress [12,13] mitochondrial dysfunction (reviewed in [14,15]), and alteration of cytoplasmic membrane permeability [16] are all associated with an increase in the severity of the protein aggregation.

Protein Aggregation and Neurodegeneration
Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disease, which is characterized by protein aggregates in the upper and lower motor neurons and the degeneration of these neurons (reviewed in [7,17]). Approximately 10% of ALS patients have a known family history of the disease. To date, gene mutations are the only proven cause of ALS (reviewed in [18]); however, the etiology of the disease is complex, with significant clinical heterogeneity and a variable penetrance of ALS-linked mutations. In contrast to the clinical heterogeneity of the disease, mutations in the gene TARDBP (encoding for TDP-43) are unequivocally linked to the development of ALS and TDP-43 inclusions are the key pathological hallmark for nearly all (~97%) ALS patients [19][20][21]. Likewise, mutations in the fused in sarcoma (FUS) gene account for approximately 4% of familial ALS patients and FUS protein inclusions are also found in a subset of ALS patients (1%) [22][23][24][25]. TDP-43 and FUS are predominately nuclear proteins, which have wellcharacterized nuclear functions, such as transcription, mRNA splicing and poly-adenylation, miRNA biogenesis (reviewed in [26][27][28]). Under physiological conditions, both proteins are found in low levels in the cytoplasm [29], where they have emerging roles in mRNA stability and transport, regulation of translation, miRNA processing, stress response, mitochondrial and autophagy regulation, and synaptic function (reviewed in [30]). In ALS on the other hand, both TDP-43 and FUS are depleted from the nucleus and mislocalize to the cytoplasm [19,22,25]. Clinical verification of TDP-43 and FUS pathology are currently limited to postmortem tissue examination. However, these histological techniques provide only a static snapshot of the aggregation pattern at predetermined stages of the disease. This significantly limits the ability to investigate the dynamic molecular mechanisms that are believed to trigger aggregate formation, maturation and mislocalization into the cytoplasm.

A Zebrafish Model to Study Neurodegeneration
Zebrafish are an excellent model to study some of the underlying mechanisms as they develop ex utero (allows for straight-forward genetic modification), are transparent during early age (aiding microscopic analysis), and can be assessed for movement impairment at an early age [31]. For example, zebrafish embryos have previously been used to study the effect of expression of disease-causing mutations in TDP-43, SOD1, FUS, C9orf72, and CCNF [32][33][34][35][36][37][38][39][40][41][42][43][44][45][46]. We have previously used the zebrafish system to visualize the intraneuronal localization and spread of human TDP-43 in degenerating motor neurons [47]. However, a limitation of this study was the inability to determine whether the spread of the "pathogenic" TDP-43 released from the dying motor neurons ultimately resulted in interaction with other TDP-43 molecules in neighboring cells, such as the glia or healthy motor neurons. The mechanism through which degeneration spreads, and whether this occurs through prion-like properties, represents a key question in the field.

Bimolecular Fluorescence Complementation-a Tool to Visualize Protein Interactions Dynamically
A variety of luminescence-based techniques to trace protein interactions in living cells have been established, including Förster resonance energy transfer (FRET) [48], bioluminescence resonance energy transfer (BRET) [49], and bimolecular fluorescence complementation (BiFC) [50]. BiFC has been previously used to validate protein interactions [51]. It is based on the reassembly of the unfolded, complementary, non-fluorescent N-and C-terminal fragments of a split fluorophore, which are fused to the protein of interest. Upon interaction of the protein of interest, the split fluorophore fragments are brought into spatial proximity (generally < 7 nm), enabling the structural complementation of the fluorophore.
BiFC offers unique advantages to study protein-protein interactions such as aggregation, maturation, and mislocalization in living cells (reviewed in [51]). The fluorescence complementation results in a specific signal due to the intrinsic fluorescence reconstitution of the non-fluorescent fragments [50]. BiFC does not rely on energy transfer between fluorophores, as seen in FRET and BRET assays, and can be detected by employing conventional fluorescence microscopy [52]. Importantly, the technique can also be adapted to confirm protein interactions within a complex [52]. Potential artifacts, caused by cell lysis or cell fixation, are eliminated, as the protein-protein interactions are studied in living cells [53]. BiFC has previously been used in vitro to visualize proteinprotein interactions in the early stages of aggregate formation in neurodegenerative disease [54,55]. BiFC has also been adapted for in vivo use in the zebrafish to study protein interactions in cellular pathways during development [56][57][58][59]. Recently, studies have utilized a split red fluorescent protein (RFP) assay in C. elegans to investigate neuronal synapses [60] and a split GFP complementation assay to visualize alphasynuclein in zebrafish [61]. In addition, split luciferase and mVenus BiFC assays have been used to visualize αsynuclein oligomerization in mouse brains [62,63].
In this study, we validated a fluorescence complementation approach to investigate the formation and localization of ALS aggregates in vivo by combining BiFC with the advantages of the zebrafish model system. We further provide in vivo validation of an optimized BiFC construct and show the selectivity of this approach using competitive injections as a control for nonspecific protein interactions, which have previously been reported in BiFC assays [64]. We also demonstrate preliminary results for the dynamic aggregation of the ALSlinked proteins Fus and TDP-43, both wild-type and ALSlinked mutant forms, via fluorescence complementation. Taken together, our results confirm that BiFC can be utilized to study ALS-linked aggregate formation and conceivably spread in vivo.

Generation of Constructs
The BiFC fragments chosen for all experiments described here are the amino-terminal (N-terminal) half of the yellow fluorescent protein (YFP) Venus (Venus-1-155, referred to as VN155 hereafter) and the carboxy-terminal (C-terminal) half of the cyan fluorescent protein (CFP) mCerulean (mCerulean-155-239, referred to as CC155 hereafter). These fragments were fused to a protein of interest via a glycine serine (GGGS)3 flexible linker to allow for the normal biological function of the fusion protein [66]. The constructs were cloned into the mammalian, avian, xenopus, and zebrafish expression pCS2+ vector. This vector contains an SP6 promoter allowing for the in vitro transcription of mRNA.
TDP-43 BiFC Constructs Synthesis and subcloning of the gene fragments into pCS2+ plasmids were performed by GenScript (NJ, USA). Wild-type human TDP-43 was N-terminally fused to either VN155 or CC155 via a (GGGS) 3 flexible linker. VN155 or CC155 with a (GGGS) 3 flexible linker was ordered as controls. Gene fragments were inserted into BamH1 and Xba1 restriction sites of the pCS2+ plasmid. Site-directed mutagenesis was performed by GenScript (NJ, USA) to obtain the optimised Venus I152L constructs.

Confocal Microscopy BiFC Signal Intensity Measurement
Tissue-specific imaging at higher resolutions was performed on a Leica SP5 or Leica SP8 confocal microscope following protocols detailed previously [68,70]. Acquisition was performed with identical gain and laser power settings for each of the treatment groups. The mean fluorescence intensity was measured using ImageJ (freehand selection tool) on maximum projection images acquired in the muscle from three larvae expressing either wild-type or mutant (M337V) TDP-43. Fluorescence intensities were obtained for the whole cell, as well as the cytoplasm and nucleus specifically. To account for any intrinsic differences in fluorescence intensity, the ratio of the cytoplasm and the nucleus for each individual cell was calculated and used for statistical analysis.

Statistical Analysis
GraphPad PRISM® (GraphPad Software, Inc.) was used to analyze data and create figures. Statistical analysis on competitive control injection experiments ( Fig. 1) was performed based on pooled individual data from four independent experiments (n = 36) with an unpaired, two-tailed t test at a 95% confidence interval. Data was normalized to controls. Statistical analysis on noise and signal experiments (Fig. 2) was performed based on pooled individual data from three independent experiments (n = 36) with an unpaired, two-tailed t test at a 95% confidence interval. Data was normalized to controls. Statistical analysis on the signal to noise ratio was performed on averaged data from three independent experiments with an unpaired, two-tailed t test, and data was normalized to controls. Statistical analysis on the cytoplasmic/nuclear fluorescence intensity signal was performed on pooled data from 3 independent fish with n = 37 (wild-type) and n = 32 (mutant) cells analyzed with an unpaired, two-tailed t test at a 95% confidence interval. Values in figures show mean (bar graphs) or individual data points, and error bars in all figures represent standard deviation (SD), ****p < 0.0001, ***p < 0.001, **p < 0.01, *p < 0.05.

BiFC Facilitates Monitoring of TDP-43 Aggregation In vivo
In order to determine if BiFC could be used to investigate protein aggregation in zebrafish, the aggregation prone and neurodegenerative disease-linked protein TDP-43 [19] was utilized in our assay. We have previously visualized the full version of the fluorescently tagged human TDP-43 and have determined that wild-type human TDP-43 is predominately located in the nucleus of zebrafish spinal cord motor neurons with approximately 20% found in the cytoplasm [47]. Full-length human wild-type TDP-43 (mRNA) attached with either of the two split-fluorophore halves (CC155 and VN155; Fig. 1) was injected into the one-cell stage of the zebrafish eggs to induce expression throughout the embryos. We first performed a dose response curve to establish a dose that did not result in any phenotypical abnormalities such as developmental death or uncharacteristic morphological features in the injected fish. Twenty picograms was established as an optimal injection dose that did not result in abnormal phenotypes (three independent experiments, minimum n = 15 per experiment, Supp. Fig. 1). We next assessed if these injections resulted in fluorescence complementation. Clear fluorescence reconstitution was observed in injected embryos 24 hpf (Figs. 1d and 3a). Importantly, we could observe a seemingly nuclear localization of the TDP-43 fluorescence signal as we would expect from previous studies with fluorescent full-length TDP-43 [47] (Fig. 3c-h). To confirm that the fluorescence signal was specific to TDP-43 interaction, we performed a series of control experiments. Embryos injected with one TDP-43-fused fluorophore fragment and the complementary non-TDP-43-fused fluorophore fragment (VN155-TDP-43 and CC155) resulted in a very dim background expression pattern (TDP-43 independent noise, Fig. 1e). Likewise, embryos injected with fluorophore fragments not fused to TDP-43 (VN155 and CC155) displayed negligible fluorescence, which was distributed homogeneously and clearly distinguishable from the signal resulting from fluorophore fragments fused to TDP-43 (Fig. 1f).
These control experiments demonstrated that spontaneous, unspecific complementation of fluorophore fragments (noise) was negligible while TDP-43-fused BiFC fragments showed a clear propensity for fluorescence reconstitution, likely indicating enhanced aggregate propensities for TDP-43, and the specificity of the assay.
If the fluorescence reconstitution is a result of aggregating TDP-43 in our in vivo system, we would expect that additional supplementation with TDP-43 mRNA not coupled to a fluorophore half results in reduced BiFC intensity. To test this, nonfluorescent TDP-43 was injected with or without the same amount of the BiFC-halves (Fig.  1g). Here, we applied a microplate reader-based approach to measure signal intensities in an unbiased fashion. Injection of 10 pg TDP-43 control mRNA (non-fused) in addition to the standard BiFC halves (dose of 10 pg VN155-TDP-43 + 10 pg CC155-TDP-43 mRNA) resulted in a decreased BiFC intensity as compared to standard injections (10 pg Scale bars represent 20 μm. b Quantitative comparison of signal-to-noise ratio using standard VN155 versus VN155-I152L fragments. Data was obtained using an unbiased plate reader. Data pooled and averaged from three independent replicates. c, d The background fluorescence (noise) was significantly reduced with the I152L mutation (c), while the maximum fluorescence intensity did not change (d). Data in c, d was obtained using an unbiased plate reader. Data pooled from three independent replicates. *P < 0.05; ****P < 0.0001; ns P > 0.05 VN155-TDP-43 + 10 pg CC155-TDP-43; Fig. 1d). The BiFC intensity in these injections was significantly dimmer (0.58 units ± 0.30, three independent experiments, n = 36 per experiment) compared to the signal in standard-TDP-43-BiFC-injected embryos (1 unit ± 0.35, three independent experiments, n = 36 per experiment; TDP-43 fused to fluorophore-fragments only; absolute intensity divided by autofluorescence intensity in noninjected controls and normalized to standard) (Fig. 1g).

Fluorescence Optimization In vivo Using an Isoleucine Substitution
To further optimize the TDP-43 BiFC efficiency and reduce background (nonspecific) fluorescence, a reported optimization for mVenus-based BiFC systems was exploited [64]. This optimization involves an I152L substitution in the N-terminal mVenus fragment which reduces the self-assembly of the fluorophore and should therefore reduce the background nonspecific fluorescence of the assay [64]. To determine if the approach would work for our in vivo TDP-43 BiFC system, the mVenus I152L substitution was introduced into our constructs. To compare the levels of TDP-43-independent selfassembly of the fluorophore, mRNA injections of TDP-43 fused to either the standard (VN155-TDP-43) or the optimized (VN155-I152L-TDP-43) mVenus fragment were performed in combination with the mCerulean (CC155) control (not fused to TDP-43; Fig. 2a, right column). In parallel, the intensity of the TDP-43-dependent fluorescence signal was assessed by injecting TDP-43 fused either to standard (VN155-TDP-43) or optimized (VN155-I152L-TDP-43) mVenus fragments together with the complementary, TDP-43-fused mCerulean (CC155-TDP-43) fragment (Fig. 2a,  left column). An unbiased plate reader approach was used to determine fluorescence intensities. The optimized I152L substitution decreased nonspecific background fluorescence (TDP-43 independent noise) and resulted in a 2-fold increased signal-to-noise ratio as compared to the standard VN155-TDP-43 construct (average mean from three pooled independent experiments, normalized to standard: 2.08 ± 0.59 and 0.99 ± 0.29, respectively, three independent experiments, n = 36 per experiment) without diminishing the signal intensity ( Fig. 2b-d). In summary, the data demonstrate that fluorescence reconstitution in our mRNA-based TDP-43 BiFC assay is TDP-43-interaction-specific, and that the introduction of the I152L substitution further improves the signal to noise ratio 2fold by reducing the non-specific fluorescence reconstitution.

TDP-43 BiFC Accumulation Is Tissue and Compartment Specific
Since mRNA levels are highly regulated during zebrafish early development [71], the BiFC system employed here provides a transient BiFC signal, as the injected constructs are degraded progressively. To determine the duration of TDP-43 BiFC expression, embryos were microinjected at one-cell stage and raised in the dark incubator. The brightness of the BiFC signal was assessed at multiple time points during that period, and individual fish were imaged at consecutive days using identical acquisition parameters. As expected, the fluorescence intensity peaked at 24-30 hpf and faded gradually during the following days (Fig. 3b). At 72 hpf, the BiFC intensity declined to~30% of the 24 hpf signal, indicating the expected mRNA degradation and ensuing fluorescent reduction (n = 5, Fig. 3b, c-h). This signal loss is in line with the expected growth-dependent decline in the levels of cellular mRNA and proteins in live tissue, particularly in early developmental stages [71,72].
TDP-43 has been previously reported to localize primarily in the nucleus [47,73]. To determine the localization of our TDP-43 BiFC constructs more precisely, coinjections were performed with mRNA that labels the nucleus of the cells (human nuclear protein histone H2B fused to full-length mCerulean3; H2B-mCerulean3). At 24, 48, and 72 hpf, the BiFC signal could be clearly observed in the muscle tissue as well as spinal cord motor neurons (Fig. 3a, c-h). Higher magnification confocal microscopy revealed that exogenous wild-type TDP-43 was predominantly located in the nucleus, while lower levels of the protein were also present in the cytoplasm, resembling the expression pattern of endogenous TDP-43 ( [73], Fig. 3c-h). To further analyze the compartmentalization and (mis)localization of TDP-43, we next assessed the BiFC fluorescence expression of wild-type versus mutant (M337V) TDP-43 (Fig. 4c-f). Quantitative analysis of confocal z-projections revealed that the mutant form of TDP-43 had a higher propensity to accumulate in the cytoplasm compared to wild-type TDP-43 (Fig.  4b). This shift in the cytoplasmic to nuclear fluorescence intensity was independent of the overall fluorescence expression in our fish (Fig. 4b).

Characterization of the Interactions of Mutant and Wild-type Fus Using BiFC
To assess if the fluorescence complementation approach is relevant for other ALS proteins, we next tested the ALSassociated protein Fus (Fig. 5a). Wild-type Fus BiFC mRNA constructs were injected into the one-cell stage of zebrafish embryos. Injections of VN155-Fus and CC155-Fus resulted, similar to the TDP-43 assay, in successful fluorescence complementation, demonstrating that wild-type Fus proteins interact and accumulate as distinct fluorescent puncta throughout the muscle cells (Fig. 5b, c).
Disease-linked mutations in FUS have been shown to lead to mislocalization of the predominately nuclear FUS protein into the cytoplasm and are a potential cause of pathological mechanism in ALS [22]. Therefore, we next generated an endogenous zebrafish fus construct that contained an equivalent human pathological FUS mutation (human FUS R521G aligns with zebrafish Fus R536G ) and a nonfunctional nuclear localization signal (to achieve Fus protein (mis)localization into the cytoplasm). Indeed, injections of the mutant Fus constructs VN155-Fus and CC155-Fus resulted in fluorescence complementation in the cytoplasm as discrete puncta/aggregates (n = 3, Fig. 5d, e; note the lack of nuclear BiFC signal compared to wild-type Fus). When we next coexpressed wild-type Fus and mutant Fus together, we found that these two proteins interact albeit being determined for different cellular compartments (n = 3, wild-type nuclear and mutant cytoplasmic; Fig. 5f, g). Additionally, we observed that fluorescence complementation was predominantly restricted to the nucleus, indicating that wild-type Fus has not only the capacity to interact with mutant Fus but conceivably recruits mutant Fus into the nucleus (Fig. 5f, g).
To verify that these interactions are specific to Fus-Fus protein interactions and not due to nonspecific complementation of the fluorophore fragments, wild-type and mutant Fus Nterminal fluorophore fragments (VN155-Fus) were coinjected with the C-terminal mCerulean fluorophore fragment fused to an mKate2 fluorophore (CC155-mKate2; red fluorescence). This approach let us observe the red fluorescence independent of complementation but would result in yellow mVenus complementation if the interaction of the fluorophore halves was nonspecific. Indeed, red fluorescence was visualized throughout the embryo, indicating successful injection and maturation of the fluorophore (Supp. Fig. 2). Importantly, little to no background fluorescence complementation was observed and no yellow fluorescence reconstitution was detected (Supp. Fig. 2), highlighting that the Fus fluorescence complementation was due to the specific Fus-Fus protein interactions. Taken together, these results demonstrate that BiFC can be applied to different ALS-linked proteins and effectively used to investigate their aggregate characteristics in vivo.

Discussion
We have optimized and verified a novel BiFC assay for the visualization of ALS-linked protein aggregation in zebrafish.
Our BiFC approach provides confirmation that the two neurodegenerative disease-linked proteins, TDP-43 and Fus, undergo fluorescence complementation in vivo. Additionally, our data highlights that the BiFC signal could be significantly improved using fluorophore optimization and overall demonstrates the successful implementation of BiFC to observe mislocalization and aggregate formation.
Previous studies have utilized the BiFC system in cell culture for visualization of neurodegenerative disease-linked proteins and to find unprecedented insights into the aggregation propensities of disease aggregates. Tau aggregation and interactions have been studied quantitatively in cell culture using a split GFP/BiFC-based approach [54,55,74]. This approach revealed that tau aggregation is favored when tau is phosphorylated at S396/S404 in the presence of increased GSK3beta activity [54]. A similar approach has identified that specific amino acids in the head and tail portions of the N-terminal domain of TDP-43 (E17A; E21A; R52A; R55A) mediate the interactions necessary for its self-assembly [75] and that the self-assembly of FUS is mediated through its low-complexity domain (amino acids 1-214) [76,77]. BiFC approaches have also been adapted to in vivo systems to study morphogen gradients and protein interactions in developmental biology  [56][57][58][59]. More recently, BiFC has been implemented to observed neurological processes [60,61]. In C. elegans, a split RFP assay differentially labeled subsets of synapses between the same interneurons and different neuronal partners, and demonstrated that the clr-1 gene is required for the formation of both sets of synapses [60]. Vicario et al. also demonstrated that Parkinson's disease-related mutations in α-synuclein resulted in altered submitochondrial localization of the protein using a BiFC approach [61]. These studies highlight the suitability and value of split-fluorescence approaches to uncover some of the fundamental molecular mechanisms that drive protein accumulation.
To verify if BiFC can be used to study the process of aggregate formation in zebrafish, we expressed both TDP-43 and Fus BiFC fusion proteins transiently in zebrafish embryos. Our study revealed fluorescence complementation of wildtype TDP-43 and wild-type Fus and the development of intranuclear aggregates, while mutant Fus BiFC fusion proteins demonstrated fluorescence complementation in cytoplasmic aggregates as previously reported in cell culture BiFC experiments [75,77]. Interestingly, co-expression of wild-type and mutant Fus BiFC fusion proteins resulted in fluorescence complementation in intranuclear aggregates, not cytoplasmic aggregates, contrary to previously reported findings from cell culture studies and postmortem tissue [77,78]. However, the Matsumoto et al. cell culture study did not use a BiFC approach and one possibility is that the intranuclear aggregation could represent an early phenomenon (nuclear maturation of the Fus BiFC signal is observed as discrete puncta in the cytoplasm. g The combination of mutant and wild-type Fus constructs reconstitutes primarily nuclear. Scale bars, 10 μm protein followed by cytoplasmic shuttling) that is too late to observe in postmortem tissue or cell culture models. These preliminary differences highlight the importance of studying these dynamic processes of aggregate formation and maturation longitudinally in living cells/organism such as the zebrafish. Our in vivo analysis of TDP-43 compartmentalization also confirmed previous cell culture observations that described increased mislocalization of mutant forms of TDP-43 [79]. As expected, the M337V mutation induced cytoplasmic mislocalization of TDP-43 in our zebrafish (Fig. 4). This further demonstrates the suitability of our BiFC approach to study protein interactions and mislocalization in a living animal.
Some limitations of the zebrafish-based BiFC system described in this study are the transient nature and ubiquitous expression that result from mRNA injections. During zebrafish early development, mRNA levels are highly regulated and the injected constructs, which encode the BiFC proteins, are degraded quickly [71]. Therefore, the applied mRNA system, which has the advantage of rapid detection, is transient and only allows for the detection of fluorescence in the very early stages of development (Fig. 3). DNA constructs designed to induce stable transgenesis are needed to allow for the longer-term visualization of disease aggregate maturation and its potential mislocalization. Additionally, in order to study the longer-term effects of ALS aggregates and their precise contribution to neurodegeneration the BiFC constructs could be expressed in neurons and glia, respectively. While we tested the toxicity of the BiFC constructs in our assay (Supp. Fig. 1), the general toxicity level is likely related to the protein of interest (TDP-43 in our assay). Nevertheless, we employed a series of assays to test for the specificity of these interactions and demonstrate that TDP-43 and Fus complementation occurs only when the proteins are fused to the non-fluorescent halves, and that this signal could be increased 2-fold by introducing a previously reported mVenus mutation (I152L) [64] (Fig. 2). These results demonstrate that the fluorescence complementation is both a specific and reliable tool to detect ALS-linked protein aggregation in vivo in zebrafish. Notably, this approach is not limited to ALS aggregates and may present a unique opportunity to investigate propagation of aggregates in other neurodegenerative diseases (such as αsynuclein in Parkinson's disease [80]).
The specificity and nonreversible nature of the fluorescence complementation offer unique opportunities to study aggregate maturation and the potential spread of TDP-43 in vivo. Current imaging techniques are limited by the quenching of an existing fluorescence signal after neurodegeneration (i.e., due to uptake by microglia [47]). The BiFC approach may open novel avenues to test the hypothesis that TDP-43 can spread horizontally throughout the spinal cord. Our zebrafish BiFC assay could also be adapted to study stress-granule formation in real time in vivo. Stressors involved in disease onset and progression (such as oxidative stress) could be applied to the embryos to determine if and how physiologically relevant cellular stress can exacerbate protein aggregation. Furthermore, a current hypothesis in the field is that TDP-43 aggregates form frequently through a process called liquid-liquid-phase-separation (LLPS [81]). This process is reversible and allows proteins to form membrane-less sub-compartments where specific short-term actions such as RNA processing can be performed rapidly. However, it is now believed that under certain conditions these liquid droplets can form into tightly clustered aggregates that can become pathological [82]. To the best of our knowledge, this BiFC approach may be one of the very few ways to assess how recruitment of mutant or pathological forms of TDP-43 can directly influence the droplet behavior of wildtype TDP-43 (or vice versa). Particularly transparent zebrafish offer some unique advantages to examine these pathways in detail as zebrafish embryos can be generated in large numbers, allow high-resolution fluorescence microscopy, and ultimately can provide a useful medium-throughput preclinical platform to screen for molecules that can modulate the aggregation of disease relevant proteins in the future.

Conclusions
Protein aggregation is hypothesized to underlie a variety of degenerative human diseases. However, the mechanistic connection between the processes of protein aggregation, tissue degeneration, and the progression and spread of degeneration is not well understood. Pathological aggregates likely vary across a structural spectrum ranging from small unstructured oligomers to well-defined cross-beta-sheet amyloid fibrils. Which forms of aggregate are most toxic and the mechanisms underlying this proteotoxicity are of immense interest to the field. Identification of the mechanistic and molecular underpinnings, and identifying small molecules with the capacity to modify these processes, may be an important step towards a potential therapeutic intervention applicable to ALS and a range of other human proteinopathies.