FUS gene is dual-coding with both proteins united in molecular hallmarks of amyotrophic lateral sclerosis

Novel functional coding sequences (altORFs) are camouflaged within annotated ones (CDS) in a different reading frame. We discovered an altORF nested in the FUS CDS encoding a conserved protein, altFUS. We thus demonstrate the dual-coding nature of the Amyotrophic Lateral Sclerosis (ALS)-associated FUS gene. AltFUS is endogenously expressed in human tissues, notably in the motor cortex and motor neurons of healthy controls and ALS patients. AltFUS inhibits autophagy, a pathological hallmark presently and incorrectly attributed to the FUS protein. AltFUS is pivotal in two other pathological hallmarks: loss of mitochondrial membrane potential and accumulation of FUS/TDP-43 cytoplasmic aggregates. Suppression of altFUS expression in a FUS-ALS Drosophila model protects against neurodegeneration. Thus, wild-type altFUS is essential for ALS-like phenotypes arising from mutated FUS. Some mutations found in ALS patients are overlooked because of their synonymous effect on the FUS protein, yet we showed they exert a deleterious effect via their missense consequence on the overlapping altFUS protein. These findings suggest that both proteins, FUS and altFUS, are involved in the aetiology and pathological hallmarks of ALS.


ALTFUS IS A NOVEL 170 AMINO ACID PROTEIN, ENDOGENOUSLY EXPRESSED IN CELL LINES AND TISSUES 85
We began by querying OpenProt 27 predictions for FUS canonical mRNA (ENST00000254108 or 86 NM_004960), which led to 8 predicted altORFs, either overlapping the coding sequence (CDS) or 87 within the 3'UTR (Extended Data Table 1). Amongst these, IP_243680 or altFUS, a 170 codon 88 altORF, presents convincing experimental evidence of expression (OpenProt v1.3). AltFUS 89 overlaps the FUS CDS in an open reading frame shifted by one nucleotide (Fig. 1a, Extended Data 90 according to OpenProt (Fig. 1b). Ensembl 32 annotates two transcripts as coding (FUS-201 and FUS-95 202), either for the 526 amino acid FUS protein or its 525 amino acid isoform. From OpenProt 96 prediction, these two transcripts also encode altFUS (IP_243680), or its 169 amino acid isoform 97 (IP_243691) respectively. Moreover, the second most abundant transcript in brain tissues and 98 nerves , representing about 20 % of all transcripts, is non-coding according to Ensembl,99 but OpenProt predicts it contains the altFUS CDS (Fig. 1b, Extended Data Fig. 1c). Thus, of the five 100 most abundant transcripts in brain tissues and nerves, two code for both FUS and altFUS proteins, 101 one codes for altFUS alone, and the remaining two are non-coding. 102 Published RIBO-seq data in Human, retrieved from the Gwips portal 33 , revealed an accumulation 103 of initiating ribosomes around the altFUS initiating methionine, in association with an increase in 104 the density of elongating ribosomes over altFUS CDS (Fig. 1c). These results suggest that altFUS is 105 translated. Similar results were observed in Mouse (Extended Data Fig. 1d). Additionally, we 106 retrieved nucleotide conservation scores (PhyloP) for FUS transcripts over 100 vertebrates. PhyloP 107 scores range from -10 (highly variable) to 10 (highly conserved). PhyloP scores over the FUS CDS 6 are under a constraint at the altFUS CDS locus (average score of 2.6 instead of 4 elsewhere on the 109 FUS CDS), which is consistent with a selection pressure across 2 overlapping frames (Extended 110 Data Fig. 1e) 34 . We then retrieved altFUS protein sequences over 84 species and observed a strong 111 protein conservation across mammals, and primates notably (75 to 99.4 % of sequence identity -112 Extended Data Table 2, Extended Data Fig. 1f and Extended Data Alignment File). Thus, AltFUS 113 is well conserved, with domains showing little to no sequence variations (Fig. 1d). 114 Based on the OpenProt database, AltFUS was identified in multiple proteomics datasets, with up 115 to 7 confident unique peptides, representing a 41 % sequence coverage (Fig. 1e, Extended Data 116 Table 1). To validate altFUS protein expression, we developed a custom antibody targeting two 117 unique altFUS peptides (Extended Data Fig. 2a) and tested it using three constructs: FUS, altFUS 118 and FUS (Ø) . The latter is a monocistronic FUS version, where all altFUS methionines are mutated 119 for threonines in a manner synonymous for FUS (Extended Data Fig. 2b-d). Thus, the FUS protein 120 sequence is unchanged, but the altFUS sequence does not contain any methionines. Transfection 121 of HEK293 cells revealed expression of both proteins, FUS and altFUS, from the FUS nucleotide 122 sequence (Fig. 1f). As expected, altFUS expression was lost with the monocistronic FUS (Ø) 123 construct. HEK293 cells transfected with a siRNA targeting FUS mRNA showed a significant 124 knockdown of both proteins, FUS and altFUS; whereas altFUS endogenous expression was visible 125 in scrambled control siRNA and mock transfected cells (Fig. 1g). These results validate the 126 specificity of the custom antibody and demonstrate altFUS endogenous expression in HEK293 127 cultured cells. 128 AltFUS endogenous expression was visible in control human tissues, HEK293 and HeLa cell lines 129 (Fig. 1h). Since FUS gene is associated with ALS, which predominantly affects the motor cortex in 130 addition to the ventral spinal cord, we retrieved motor cortex lysates from 3 ALS patients with a 131 was detected in all cases (Fig. 1i). Furthermore, as ALS is a motor neuron disease, we derived 133 functional ventral spinal motor neurons from induced pluripotent stem cells (iPSCs) 35,36 from 134 healthy controls and ALS patients carrying valosin-containing protein mutations (3 lines per 135 group). AltFUS endogenous expression was detected in all samples (Fig. 1j). We noticed that brain 136 ( Fig. 1h) and motor cortex lysates (Fig. 1i), as well as iPSCs-derived motor neurons (Fig. 1j), from 137 healthy controls and ALS patients, presented a higher band detected with the custom altFUS 138 antibody. This band is not present in cultured cell lines or other tissues. It could come from a non-139 specific signal or a post-translational modification on altFUS that is specific to the motor cortex 140 and spinal cord motor neurons ( Fig. 1h-j). We could also observe a lower band in the line 2 of 141 controls motor neurons, which may correspond to a degradation product or an initiation at a 142 downstream methionine in altFUS sequence. This band has never been observed in other samples 143 so far. We demonstrate that the FUS gene encodes two proteins, FUS and altFUS, both 144 endogenously expressed in the motor cortex and spinal cord motor neurons, to the two 145 archetypal regions predominantly affected in ALS. 146

ALTFUS IS A MITOCHONDRIAL PROTEIN, INVOLVED IN ALS-ASSOCIATED LOSS OF MITOCHONDRIAL MEMBRANE 147
POTENTIAL 148 FLAG-tagged altFUS (altFUS-FLAG) displayed a strong co-localization with a common mitochondrial 149 marker, TOMM20 (Fig. 2a). Additionally, mitochondrial extracts showed an enrichment in altFUS-150 FLAG (Fig. 2b). Cellular fractionation of cells over-expressing untagged altFUS further validated 151 altFUS mitochondrial localization (Fig. 2c). The endogenous altFUS protein was found in the 152 mitochondrial fraction, although it displayed a weak cytoplasmic signal as well (Fig. 2d), consistent 153 with the immunofluorescence data (Fig. 2a). Furthermore, cells over-expressing altFUS showed 154 an altered mitochondrial network, with a significant increase in fragmented mitochondria 155 (globular) compared to mock cells that displayed more tubular structures (Fig. 2e,  and is linked to severe fALS and sALS cases 9 . In this construct, altFUS is still present and not 162 affected by the mutation. Similarly to FUS (Ø) , we also generated the monocistronic construct 163 FUS (Ø) -R495x, which contains synonymous mutations for FUS-R495x (Extended Data Fig. 3b), but 164 prevents altFUS expression (Fig. 1f). V5-FUS (Ø-FLAG) and V5-FUS (Ø-FLAG) -R495x did not express altFUS, 165 but only the FUS protein, wild-type or ALS-linked mutant R495x respectively (Extended Data Fig.  166 3c). We first investigated the effect of altFUS on the mitochondrial membrane potential using the 167 potential sensitive dye TMRE (Fig. 2g, h, Extended Data Fig. 3d). As previously described 38 , over-168 expression of bicistronic FUS or FUS-R495x led to a decrease in mitochondrial membrane 169 potential. The mitochondrial membrane potential remained normal when over-expressing 170 monocistronic FUS (Ø) or FUS (Ø) -R495x, underlining the role of altFUS. However, over-expression of 171 altFUS alone did not alter the mitochondrial membrane potential, which suggests both proteins 172 cooperate for this ALS pathological hallmark. 173 To further evaluate the possible implication of altFUS in ALS, we investigated its protein 174 interactors. Using stimulated emission depletion microscopy (STED), we observed that altFUS 175 localized in puncta following a cristae-like pattern inside the mitochondria, delimited using an 176 outer-membrane mitochondrial marker, TOMM20 (Fig. 2i, j). We then used size-exclusion 177 chromatography on mitochondrial extracts to isolate altFUS-FLAG macromolecular complexes ( Fig. 9 interacting proteins (Fig. 2l, Extended Data Table 3). A gene enrichment analysis to the Human 180 mitochondrial proteome identified three significantly enriched biological processes: autophagy-181 related pathways, mitochondrial metabolism and cellular response to stress (Fig. 2m). Disruptions 182 within these are pathological hallmarks of ALS 3,21,39,40 . 183

185
Following on these results, we hypothesized that the inhibition of autophagy observed with ALS-186 linked FUS mutants may instead be attributed to altFUS. We used the mCherry-GFP-LC3 reporter 187 to track the autophagic flux by confocal microscopy (Extended Data Fig. 4a). Under basal 188 conditions, cells displayed red and yellow foci as expected (Fig. 3a). An accumulation of yellow 189 foci was observed when cells were treated with bafilomycin, an inhibitor of autophagy. Similarly, 190 cells over-expressing altFUS displayed a significant accumulation of yellow foci (Fig. 3a). Our 191 results were consistent with previously published data 11 as cells transfected with FUS or  R495x displayed a decreased autophagic flux (Fig. 3a, Extended Data Fig. 4b). This accumulation 193 of yellow foci was absent in cells that express monocistronic FUS constructs, thus lacking altFUS 194 expression (FUS (Ø) or FUS (Ø) -R495x). We used bafilomycin followed by LC3 probing to further 195 validate the impact of altFUS on autophagy (Fig. 3b, Extended Data Fig. 4c). Similarly, an inhibition 196 of autophagy was observed only in cells over-expressing altFUS. Furthermore, in cells over-197 expressing monocistronic FUS (Ø) -R495x, the inhibition of autophagy could be restored by co-198 transfecting altFUS (Fig. 3b). These results establish altFUS, rather than FUS, as the protein 199 responsible of the inhibition of autophagy. 200 Furthermore, altFUS interactome analysis suggested a role in the cellular stress response, which 201 is known to be altered in ALS with a TDP-43 cytoplasmic accumulation in 98 % of patients 41,42 . In 202 FUS-linked ALS and some sALS cases, FUS cytoplasmic aggregates or mislocalization are also 203 observed [43][44][45] . We demonstrated that cells over-expressing FUS-R495x displayed cytoplasmic 204 aggregates that were positive for both FUS-R495x and TDP-43 (Fig. 3c). In cells over-expressing 205 the monocistronic FUS (Ø) -R495x construct, thus lacking altFUS expression, FUS-R495x displayed a 206 more diffuse cytoplasmic localization, and TDP-43 remained in the nucleus (Fig. 3c). FUS 207 cytoplasmic aggregates were significantly more numerous and larger when altFUS was co-208 expressed (Fig. 3d, e). Accumulation of FUS-R495x and TDP-43 in cytoplasmic aggregates could 209 be reconstituted by co-transfecting altFUS and the monocistronic FUS (Ø) -R495x construct (Fig. 3c). In order to investigate the role of altFUS in ALS, we generated Drosophila models expressing either 217 the bicistronic, FUS and FUS-R495x constructs, or the monocistronic, FUS (Ø) and FUS (Ø) -R495x, 218 constructs. We used the Elav-GeneSwitch-GAL4 Driver strain, as previously described 46 , as it 219 allows for an inducible over-expression in motor neurons and avoids lethality at the larval stage 220 from FUS over-expression in the central nervous system 46,47 . First, we generated flies containing 221 the sequences for UASt-altFUS, UASt-FUS, UASt-FUS (Ø) , UASt-FUS-R495x or UASt-FUS (Ø) -R495x. 222 These flies were then crossed with the Elav-GeneSwitch-GAL4 driver strain (Fig. 4a). UASt-223 mCherry flies were used as controls. Selected F1 individuals were then divided into 2 groups with 224 equal proportions of males/females. The first group received standard food, while the other 225 received RU-486 treated food. The treatment induces a conformational change in the Elav-GeneSwitch driver, which allows activation of the UAS promoter and thus expression of the target 227 protein. We retrieved flies at selected time points to validate protein expression in the RU-486 228 treated population through time, while the controls showed no expression (Fig. 4b). 229 The motor neuron degeneration linked to ALS provokes a progressive locomotion loss measurable 230 with a well-described climbing assay 48 . The control populations did not show any significant 231 locomotion loss at day 1, 10 nor 20 ( Fig. 4c-e); similarly to the RU-486 treated control group 232 (mCherry transgenic flies - Fig. 4c). AltFUS flies did not show any significant locomotion loss 233 through time (Fig. 4c). This result is consistent with the in cellulo data showing altFUS alone is not 234 sufficient to provoke pathological hallmarks. As previously shown with this model 19 , the 235 bicistronic FUS flies, which express both FUS and altFUS proteins, displayed a significant 236 locomotion loss (Fig. 4d). Bicistronic ALS-linked FUS-R495x flies showed an even greater motor 237 neuron degeneration through time compared to FUS (Fig. 4e). Monocistronic FUS (Ø) (Fig. 4d)

245
As of today, over 50 mutations in the FUS gene have been associated with ALS 9 . However, most 246 of these locate at the carboxyl end of the protein and as such have no effect on altFUS. We 247 wondered whether mutations altering altFUS may have been overlooked as non-consequential in 248 the FUS reading frame. We retrieved FUS synonymous mutations found in ALS patients, with an 249 allelic frequency below 0.01 %, from previous studies and the ALS Variant Server 250 (http://als.umassmed.edu/ -Extended Data Table 4). The retrieved mutations clustered on the 251 altFUS locus, with 60 % of FUS synonymous mutations found in sALS patients and 50 % of FUS 252 synonymous mutations found in fALS patients, which is significantly higher than expected by 253 chance (34 %) (Extended Data Table 4 and GFP-FUS (R64P-FLAG) -S77= (Fig. 5a). All altFUS mutants still localized to the mitochondria (Fig. 5b). 257 To investigate whether these mutations may provoke an ALS phenotype, we quantified the 258 number of cells presenting TDP-43 aggregates. All 4 altFUS mutants showed a 1.8 to 2.4 fold 259 increase compared to wild-type altFUS (Fig 5c, d) The physiological function of altFUS is still unclear, although our work provides evidence for its 298 role in mitochondrial dynamics and the cellular response to stress. One mechanism put forward 299 in ALS is that the disease originates from a sub-optimal resolution of cellular stresses, which can 300 come from environmental sources or mutated proteins 40,56 . We have shown that altFUS, not FUS, 301 inhibits autophagy, most likely via its interaction partners (Extended Data Table 3 Throughout the neural conversion and patterning phase (D0-18) the neuroepithelial layer was 419 enzymatically dissociated twice (at D4-5 and D10-12) using dispase (GIBCO, 1 mg ml-1). 420

Preparation of tissue lysates of the motor cortex of ALS patients 421
Approximately 100mg of motor cortex from 4 sporadic ALS and 4 C9orf72-ALS cases was lysed in 422 10x RIPA (50mM Tris HCl pH7.8, 150mM NaCl, 0.5% sodium deoxycholate, 1% NP40; 423 supplemented with protease inhibitors and EDTA) volume using TissueLyzer equipment (Qiagen). 424 Lysates were incubated on ice 20 minutes followed by centrifugation at 20,000xg for 20 minutes 425 at 4°C. Supernatant was taken as 'RIPA fraction' and pellets resuspended in RIPA and SDS (final 426 concentration of 2%). 3 sporadic ALS and 3 C9orf72-ALS samples were subsequently used as they 427 were sufficiently concentrated to load 100 ug of proteins onto SDS-page gels. 428

Mitochondrial extracts and cellular fractionation 429
Mitochondrial extracts were prepared as previously described 52 . Briefly, HEK293 cells grown up 430 to 80 % confluence, were rinsed twice with PBS and gathered using a cell scraper.

Mass-spectrometry analysis 503
Peptides were separated in a PepMap C18 nano column (75 μm × 50 cm, Thermo Fisher Scientific). 504 The setup used a 0-35% gradient (0-215 min) of 90% acetonitrile, 0.1% formic acid at a flow rate 505 of 200 nL/min followed by acetonitrile wash and column re-equilibration for a total gradient 506 duration of 4 h with a RSLC Ultimate 3000 (Thermo Fisher Scientific, Dionex). Peptides were 507 sprayed using an EASYSpray source (Thermo Fisher Scientific) at 2 kV coupled to a quadrupole-508

Autophagic flux measurements 532
The mCherry-GFP-LC3 was used to evaluate the autophagic vesicles within HeLa cells by confocal 533 microscopy. Before fusion with the lysosome, the LC3 molecules on the autophagosome display 534 a yellow fluorescence (combined mCherry and GFP fluorescence). After fusion, the GFP 535 fluorescence is quenched by the lysosomal pH, and as such the LC3 molecules display a red signal 536 (mCherry alone). This allows a visual representation of the autophagic flux in a given cell. Cells treated with 50 nM Bafilomycin for 4 hours were used as a positive control to validate each 538 independent experiment. Observations were made across 2 technical duplicates for each 539 biological condition, across 3 independent experiments (n=3). Alternatively, the autophagic flux 540 was also evaluated by LC3 probing before and after bafilomycin treatment (50 nM for 4 hours). 541 The quantification corresponds to the treated / untreated ratio of LC3-II abundance. 542

Cytoplasmic aggregates measurements 543
Images of HeLa cells were taken by confocal microscopy and then processed using the Image J 3D 544 Objects Counter plugin. FUS cytoplasmic aggregates were then quantified in number and size 545

Statistical analyses and representation 565
Unless otherwise stated, the statistical analysis carried was a two-way ANOVA with Tukey's 566 multiple comparison correction. The box plots represent the mean with the 5 to 95 % percentile. 567 The bar graphs represent the mean, and error bars correspond to the standard deviation. When 568 using parametric tests, normality of data distribution was verified beforehand using the Shapiro-569 Wilk test. 570

DATA AVAILABILITY STATEMENT 572
The OpenProt database is available at www.openprot.org. The GTEx portal is available via 573 www.gtexportal.org. The Gwips portal is available at www.gwips.ucc.ie. The proteomics data are 574 available on the PRIDE repository with the accession PXD------ (Fig 2l-m  where the data are represented as the fold-change compared to the GFP-FUS (FLAG) expressing cells. 858 Statistical significance is indicated above the bars (n=3, **** = p value < 0.0001). 859 860