A yeast functional screen predicts new candidate ALS disease genes

Yeast Screen to Identify RNA-Binding Proteins with Properties Similar to TDP-43 and FUS.

TDP-43 and FUS are both RNA recognition motif (RRM)-containing proteins (RRM proteins) (3) and both form cytoplasmic inclusions and are toxic when expressed in yeast (Fig. 1A) (10–13). Including FUS and TDP-43, there are at least 213 RRM proteins (Protein Families database ID PF00076) present in the human proteome (Dataset S1). This result raised the question of whether other human RRM proteins would show properties like TDP-43 and FUS (e.g., aggregation prone and toxic in yeast). If so, these properties might make these genes top candidates for unique ALS disease genes. We therefore designed a simple yeast functional screen to identify such genes (Fig. 1B). Of the 213 human RRM proteins, we were able to obtain a gene clone for 133. We cloned these 133 different human RRM-containing ORFs into yeast expression vectors as YFP fusion proteins, under the control of a galactose-inducible promoter, and individually transformed them into yeast cells. We have previously found that the addition of the YFP tag to TDP-43 and FUS does not affect the aggregation or toxicity properties (10–12). Fluorescence microscopy was used to determine the localization of each protein (nuclear, cytoplasmic, diffuse, foci; Fig. 1C and Dataset S1) and spotting assays were used to assess toxicity (Fig. 1D and Dataset S1). Some proteins localized to the nucleus (38/133) whereas others were diffusely localized in the cytoplasm (26/133). Interestingly, several others formed multiple foci in the cytoplasm in a pattern strikingly similar to that of FUS and TDP-43 (54/133). Of the proteins that accumulated in the cytoplasm, 38 were also toxic, including FUS and TDP-43 (Table 1). Thus, 38 of 133 human RRM proteins behave like FUS and TDP-43 in yeast cells.

Prion-Like Domains in TDP-43, FUS, and Several Other RRM Proteins.

To focus this list further, we used a bioinformatics approach. In addition to the RRM domain, FUS and TDP-43 share a glycine-rich domain and a bioinformatics-predicted prion-like domain (4). Like prion domains found in fungal prion proteins (e.g., Sup35, Ure2, and Rnq1), these domains are enriched in uncharged polar amino acids (such as asparagine, glutamine, and tyrosine) and glycine (14). In TDP-43, the predicted prion-like domain overlaps the glycine-rich domain; in FUS, a QGSY-rich region defines the prion-like domain, although there is some overlap with the glycine-rich domain. The prion-like domain is a shared feature that may be important, given the “prionoid” aggregation propensity of many proteins associated with human neurodegenerative disease (15). Moreover, the prion-like domains of TDP-43 and FUS are critical for the aggregation of these proteins (10–13, 16). Remarkably, using an algorithm to score 21,873 human proteins for likelihood of harboring a prion-like domain, FUS and TDP-43 ranked 13th and 43rd, respectively. We therefore interrogated the list of human RRM proteins to identify whether others ranked highly, using the prion domain prediction algorithm (14). Interestingly, 31 of the 213 human RRM proteins ranked in the top 250 (Dataset S1). Among these, FUS and TDP-43 ranked first and 10th, respectively. Of the 38 proteins that were toxic and formed cytoplasmic inclusions in yeast, 13, including FUS and TDP-43, scored highly for a prion-like domain (Table 1). Thus, using the combined yeast screen and prion-like domain analysis, we narrowed the list of RRM proteins by 10-fold (133 human RRM proteins → 38 that aggregate and are toxic in yeast → 13 that also contain a prion domain). We therefore focused on these proteins because they shared similar functional and structural features with FUS and TDP-43: (i) they formed cytoplasmic accumulations, (ii) they were toxic in yeast, and (iii) they contained a predicted prion domain.

Identification of TAF15 Sequence Variants in Patients with ALS.

With this list of 13 “FUS- and TDP-43–like” proteins in hand, we sought to test the hypothesis that these additional RRM proteins might contribute to ALS. We gave top priority to TAF15 [RNA polymerase II, TATA box binding protein (TBP)-associated factor, 68 kDa] because it ranked second of 213 human RRM proteins (and 22nd of all 21,873 human proteins) on the basis of the prion domain prediction algorithm (Table 1). Moreover, TAF15 belongs to the same protein family as FUS and is remarkably similar, especially within the RRM, the glycine-rich domain, and the C-terminal RGG domain- and PY-motif–containing region (Fig. 2A). Spotting assays showed that TAF15 expression was also toxic, albeit not as toxic as TDP-43 and FUS (Fig. 1D). Given these commonalities, we proceeded to sequence exons 13–16 of TAF15 (NM_139215), a region analogous to where many FUS mutations are located (3). These exons comprise the RGG- and PY-motif–containing C-terminal domain, which is important for nuclear localization of FUS (17–19). We performed complete sequencing of these exons in 735 individuals diagnosed with ALS and in 1,328 geographically matched healthy population control individuals (see SI Materials and Methods for patient and control demographic information). We found three patient-specific nonsynonymous missense variants (Fig. 2 B–D and Dataset S2), c.1258G > A, p.G391E; c.1308C > T, p.R408C; and c.1504G > A, p.G473E. These variants were found in individuals with ages of onset of sporadic ALS of 67 y, 47 y, and 68 y, respectively, and were all located in highly conserved regions of TAF15 (Fig. 2F). In addition, we identified one nonsynonymous missense variant that was present in both patients and controls (TAF15 c.1249G > A, p.R388H). The presence of this variant in control individuals suggests that it likely represents a benign variant, although functional studies are required to distinguish potentially damaging variants from benign variants (see below). Thus, nonsynonymous missense variations in TAF15 were detected in 4 of 735 North American ALS cases and 1 of 1,328 North American controls. We also identified a fifth variant (M368T; Fig. 2E) in an independent cohort of 351 Swedish patients with ALS and a sixth variant (G452E; Fig. 2A) in a cohort of 176 Australian patients with ALS. Because the TAF15 variants were identified in sporadic ALS cases, familial evidence for segregation with disease was not possible. Notably, however, TARDBP and FUS mutations have also been confirmed in apparent sporadic ALS cases (20). In addition, the parents of the affected individuals were not available to determine whether the mutations occurred de novo or were inherited.

ALS-Specific TAF15 Variants Form Cytoplasmic Foci in Spinal Cord Neurons.

To provide evidence that the specific sequence variants that we found in ALS cases might be pathogenic, we performed an unbiased assessment of all nonsynonymous variants identified in cases and controls. We designed an assay that potentially discriminates functionally deleterious variants from benign variants. ALS-linked mutations in TDP-43 and FUS have been shown to disrupt protein localization, leading to enhanced accumulation of ALS-linked variants in cytoplasmic structures (8, 9, 18, 21–23). Given this common feature, we assessed the effects of the TAF15 variants (Fig. 2A) on subcellular localization. We transfected myc-tagged WT or the mutant forms of TAF15 into spinal cord neurons cultured from rat embryos (Fig. 3A). The transfection efficiency in these primary neuronal cultures was too low to quantify protein expression levels; however, immunoblotting of HEK293T cells transfected with these same constructs confirmed that the WT and variant constructs are expressed at similar levels (Fig. 3C). WT or mutant TAF15 expression under these conditions did not result in toxicity; however, we did observe significant effects of the variants on localization.

The transfected WT and mutant TAF15 proteins were restricted to the nucleus in ∼40% of cells and present in both nucleus and cytoplasm in ∼60% of cells. The TAF15 mutations did not affect nuclear localization per se, but rather promoted the formation of cytoplasmic foci. WT TAF15 and the R388H variant, which was present in both patients and controls, localized to the nucleus and cytoplasm, in a diffuse pattern (Fig. 3 A and B). In contrast, all three of the ALS-linked mutant forms of TAF15 (G391E, R408C, and G473E) found in the North American patients with ALS showed a striking accumulation of TAF15 in cytoplasmic foci in dendrites and axons (Fig. 3 A and B). Thus, all of the TAF15 variants that we found in patients with ALS were deleterious; that is, they caused mislocalization of the protein to cytoplasmic foci within neurons, whereas the one variant found in controls was benign and behaved like WT. This unbiased assessment of all variants provides strong support for the variants that we found in ALS cases as being risk factors for ALS (3 damaging TAF15 variants of 735 ALS cases vs. 0 damaging TAF15 variants of 1,328 controls, P = 0.0451, Fisher's exact test). The fourth ALS-specific variant (M368T; Fig. 2E), found in an independent cohort of 351 Swedish patients with ALS, also affected localization in this assay (Fig. 3 A and B), but we do not include this variant in the above statistical assessment because we did not sequence an equivalent number of matched controls. The fifth ALS-specific variant (G452E; Fig. 2A), in 1 Australian patient with ALS of a cohort of 176 patients with ALS, was not present in 72 sequenced Australian controls or in our collection of 1,328 sequenced North American controls. We have not yet tested the effects of this variant on TAF15 localization in our functional assay. Unlike several of the ALS-linked FUS mutations, which are located in the conserved PY motif at the extreme C terminus of the protein and have strong effects on nuclear localization (18), the TAF15 variants that we have identified are not located in TAF15's PY motif (Fig. 2A), possibly explaining why they do not affect nuclear localization per se but seem to promote the accumulation of TAF15 in cytoplasmic foci. As additional variants in TAF15 are identified, this functional assay will be a powerful tool for assessing their potential pathogenicity.

TAF15 Is Inherently Aggregation Prone.

These genetic and functional studies highlight a potential role for TAF15 in ALS pathogenesis. We next sought further functional evidence that TAF15 has properties similar to TDP-43 and FUS. We asked, does TAF15 spontaneously aggregate in vitro like TDP-43 and FUS (11, 12) and does TAF15 confer neurodegeneration when expressed in the nervous system, like TDP-43 and FUS (24–31)?

We purified bacterially expressed GST-tagged TAF15 as a soluble protein under native conditions, as previously described for TDP-43 and FUS (11, 12). TAF15, TDP-43, and FUS were all competent in RNA binding, suggesting that the RRM domains were correctly folded (11, 12). Upon addition of tobacco etch virus (TEV) protease to specifically remove the N-terminal GST tag and SDS-PAGE to assess purity and expected molecular weight (Fig. 4A), we found that TAF15 rapidly aggregated at 25 °C with gentle agitation (Fig. 4 B and C). Turbidity and sedimentation analysis revealed that TAF15 aggregated with kinetics similar to FUS and slightly more rapidly than TDP-43, assessed by turbidity (Fig. 4B). If TEV protease was omitted, then little aggregation occurred (Fig. 4B). Electron microscopy studies established that TAF15 rapidly accessed oligomeric forms (Fig. 4D), which would frequently adopt a pore-like conformation [Fig. 4D (small arrows) and E], similar to those formed by TDP-43 and FUS (Fig. 4 D and E) (11, 12), which are reminiscent of pathological oligomers formed by α-synuclein and amyloid-β (32). Furthermore, TAF15 also assembled into linear polymers with a cross-sectional diameter of ∼15–20 nm (Fig. 4D, large arrows) that increased in length over time and would often become tangled into large masses by 60 min (Fig. 4D). In general, the morphology of TAF15 aggregates was more similar to FUS (12) than to TDP-43 (11), which over this time frame formed shorter polymers that would clump together to form large masses (Fig. 4D) (11). Importantly, a human RRM protein, DND1, which did not aggregate and was not toxic in yeast (Fig. 1 C and D), remained soluble and did not aggregate in this in vitro assay, providing evidence that in vitro aggregation is not a property shared by all RRM proteins. Thus, similar to TDP-43 and FUS, and concordant with the yeast data, TAF15 is an inherently aggregation-prone protein.

It is likely that sophisticated cellular proteostasis mechanisms, not recapitulated with this in vitro assay, mitigate rapid TAF15 aggregation in vivo. However, age-associated decline in these quality control measures, along with environmental triggers (for example, injury or exposure to toxins), might allow TAF15 to aggregate in disease. In vitro aggregation assays similar to these have been tremendously powerful tools in defining basic mechanisms underpinning TDP-43 and FUS aggregation (11, 12) as well as the aggregation events in Parkinson's disease and Alzheimer's disease (33–35).

ALS-Specific Variants Accelerate TAF15 Aggregation in Vitro.

Next, we tested two of the ALS-linked mutant forms of TAF15 (G391E and R408C) as well as the variant that we also found in controls (R388H) in this aggregation assay (Fig. 4F). We noticed that if we omitted agitation during the in vitro aggregation reaction, we were able to slow down the aggregation process, allowing us to detect potential differences in aggregation between WT and mutant TAF15 proteins. R388H aggregated with similar kinetics to WT TAF15 (Fig. 4G). By contrast, the two ALS-linked mutants, G391E and R408C, aggregated with more rapid kinetics (Fig. 4G). This increased aggregation propensity might help explain why ALS-linked mutant TAF15 is more prone to accumulate in cytoplasmic inclusions in spinal cord neurons (Fig. 3) and supports the hypothesis that these are likely pathogenic variants.

TAF15 Up-Regulation Causes Neurodegeneration in Drosophila and ALS-Linked Mutants Have More Severe Effect on Lifespan.

To extend these findings from in vitro to in vivo, and analyze the effects of TAF15 in the nervous system, we used Drosophila. We and others have previously shown that directing TDP-43 or FUS expression to the fly nervous system causes neurodegeneration (24–28, 30, 31). A series of transgenic lines were generated that expressed WT and mutant human TAF15. Directing expression of TAF15 to the eye of the fly caused degeneration of the structure (Fig. 5A) and led to progressive loss of motility when directed to motor neurons (Fig. 5B). As for the in vitro aggregation assay (Fig. 4), the effect was specific because two other unrelated Drosophila RRM proteins, the human counterparts of which did not aggregate and were not toxic in yeast, did not confer neurodegeneration when up-regulated in Drosophila (Fig. 5C). Thus, TAF15 possesses activity sufficient to confer neurodegeneration in the nervous system, in a manner similar to that of TDP-43 and FUS. TAF15 also induced a markedly shortened lifespan when expressed in the nervous system (Fig. 5D). Notably, expression of ALS-linked TAF15 variants G391E and R408C, inserted at the same genomic location to ensure similar protein expression (Fig. 5E), had a more severe effect, resulting in an even shorter lifespan than that of flies expressing WT TAF15 (Fig. 5D), supporting the notion that these missense variants in TAF15 are potentially deleterious.

TAF15 Is Mislocalized in Motor Neurons of Patients with ALS.

We next performed immunohistochemistry on human postmortem spinal cord tissue to determine whether TAF15 was present in motor neurons and whether its localization was affected in ALS, as for TDP-43 and FUS. TAF15 was robustly expressed in spinal cord motor neurons and localized to the nucleus, as for TDP-43 and FUS (Fig. 6A). We next examined TAF15 localization in autopsy tissue from three sporadic ALS cases (without mutations in TAF15), all with confirmed TDP-43 pathology. Immunostaining of spinal cord sections from these patients with ALS revealed significant cytoplasmic staining in a punctate granular pattern (Fig. 6 B–H), which did not colocalize with TDP-43 inclusions. In contrast to TDP-43 (7), and similar to FUS (8, 9), we did not observe significant nuclear clearing of TAF15. The strong punctate granular localization for TAF15 was not seen in any of the control cases that we examined (n = 3). Because these sporadic ALS cases that we examined also contained prominent TDP-43 cytoplasmic inclusions, which did not colocalize with TAF15, future studies will be required to extend these initial findings and to determine whether TAF15 mislocalization in disease is an early initiator or rather a later consequence of motor neuron degeneration. Thus, TAF15 is expressed in a disease-relevant cell type and can be mislocalized to the cytoplasm in ALS, further supporting the notion that TAF15 can contribute to pathogenesis. Notably, a recent study has also observed TAF15 mislocalization in neurons and glial cells of patients with frontotemporal lobar degeneration (FTLD)-FUS (36).

Plasmids, Yeast Strains, and Media.

Strain and plasmid construction is detailed in SI Materials and Methods.

Human RNA-Binding Protein Plasmid Library.

We constructed a library of yeast expression plasmids containing 133 unique human RRM-containing ORFs, as detailed in SI Materials and Methods.

Yeast Transformation and Spotting Assays.

Yeast transformation and spotting assays were performed as described in ref. 10.

Microscopy.

Fluorescence microscopy was performed as described in SI Materials and Methods.

Prion-Prediction Algorithm.

Putative prion domains were predicted using a hidden Markov model as described in SI Materials and Methods.

Sequencing TAF15 in Patients with ALS and Controls.

ALS patient and control population demographics and TAF15 mutational analysis procedures are described in SI Materials and Methods.

TAF15 Protein Purification and in Vitro Aggregation Assays.

TDP-43 and FUS were purified as described (11, 12). WT and mutant TAF15 proteins were expressed and purified from Escherichia coli as GST-tagged proteins and in vitro aggregation assays were performed as described in SI Materials and Methods.

Drosophila Experiments.

Transgenic flies expressing human TAF15 were generated by standard techniques using the pUAST vector and analyzed as described in SI Materials and Methods.

Rat Primary Neuron Transfection and Immunofluorescence.

Primary neurons were isolated from rat embryos, cultured, and transfected as described in SI Materials and Methods.

Statistical Analysis.

Two-tailed Fisher's exact tests were used to evaluate genetic association between TAF15 sequence variants and ALS. We tested all TAF15 nonsynonymous missense variants detected in cases and controls in an assay that potentially discriminates functionally deleterious variants from benign variants (e.g., the subcellular localization assay reported in Fig. 3), then classified the variants as deleterious or benign on the basis of their properties in this assay, and then performed a statistical comparison of the frequency of functionally deleterious variants in cases vs. controls.

Immunohistochemistry.

Immunohistochemistry to detect TAF15 localization in postmortem spinal cord tissue was performed as described in SI Materials and Methods.