Relative distribution of sense and antisense RNA foci
The frequency of sense and antisense RNA foci was determined in five neuronal populations: Purkinje and granule neurons in the cerebellum, motor neurons of the spinal cord ventral horn, and neurons of the hippocampal dentate gyrus and CA4 subfield. These neuronal populations were chosen as they all exhibit neurodegeneration in C9ORF72-ALS and are characteristic of both motor (motor neurons) and extra-motor (cerebellum and hippocampus) pathology . Sequential sections of tissue from C9ORF72-ALS cases, non-C9ORF72 ALS cases, and controls were examined for RNA foci in a blinded manner. No RNA foci were observed in tissue from controls and non-C9ORF72 ALS cases. Forty Purkinje neurons, forty motor neurons, >200 granule neurons, >150 dentate gyrus neurons, and >100 CA4 subfield neurons were evaluated from four C9ORF72+ ALS and/or FTD cases. The average frequency of sense and antisense foci per cell is shown in Table 2 (raw data are shown in Supplementary Table 2). Comparison between cases showed that the frequency of sense and antisense RNA foci was positively correlated in all neuronal populations i.e., cases with more sense foci per cell also had more antisense foci per cell. The exception to this was the dentate gyrus neurons where case-to-case variability was smallest (Pearson correlation coefficient: cerebellar Purkinje neurons 0.99, cerebellar granule neurons 0.6, motor neurons 0.2, CA4 subfield neurons 0.65) (Table 2). In the cerebellar populations and motor neurons but not hippocampal neurons, there was a difference between the frequency of antisense and sense foci which was consistent between cases (representative images are shown in Fig. 1a). To determine whether this difference was statistically significant, the foci count was modelled as a Poisson distribution and performing a likelihood-ratio test revealed that, within each individual case, the frequency of antisense compared to sense RNA foci was significantly higher in Purkinje neurons (likelihood-ratio test p < 0.05) and motor neurons (likelihood-ratio test p < 0.05), but significantly lower in cerebellar granule neurons (likelihood-ratio test p < 0.05). The fact that sense and antisense foci were relatively more abundant in different neuronal populations is against an artefact caused by differences in affinity of RNA FISH probes.
As reported for sense RNA foci , we observed cytoplasmic antisense RNA foci even in post-mitotic mature cells such as motor neurons (Fig. 1b).
Relative distribution of DPRs derived from sense and antisense RNA sequences
Staining of poly-GA, poly-GR, poly-PA, poly-PR, and poly-PG protein was studied in cerebellar granule neurons and motor neurons from three C9ORF72-ALS cases. More than 1000 granule neurons and approximately 50 motor neurons were examined in a blinded experiment. Neuronal inclusions containing sense RNA derived DPRs were only observed in granule neurons, whereas inclusions containing antisense RNA derived DPRs were only observed in motor neurons (Fig. 2a). Above background staining for poly-PG was not observed in any cells. In motor neurons, inclusions were predominantly nuclear, but in granule neurons inclusions were predominantly cytoplasmic (Fig. 2a). Background staining was examined in control and non-C9ORF72-ALS cases.
As a further validation, and to extend, the conclusions of this study to a larger number of cases, dual staining of poly-GA, and poly-PA protein were examined in a further blinded experiment. Approximately 1000 granule neurons and 50 motor neurons were studied from six C9ORF72+ ALS and/or FTD cases including three cases not utilized in the earlier analysis. In each case, the correct protein was determined based on the frequency of observed inclusions (Fig. 2b). Modelling the number of neuronal inclusions as a Poisson distribution and performing a likelihood-ratio test revealed that the frequency of poly-GA inclusions was significantly higher in granule neurons (likelihood-ratio test p < 0.01) and the frequency of poly-PA inclusions was significantly higher in motor neurons (likelihood-ratio test p < 0.01). The average frequencies of inclusions containing poly-GA and poly-PA protein are shown in Table 3 (raw data are shown in Supplementary Table 3).
In our previous study  we showed that, at a cellular level, there was no significant correlation between the presence of sense RNA foci and the presence of sense RNA-derived poly-GA inclusions. In this study, we examined the relationship between the presence of antisense RNA foci and poly-PA inclusions in fifteen motor neurons from four C9ORF72-ALS cases (Supplementary Table 4). As for the sense species, there was no significant correlation between the two observations (χ
2, p = 0.83).
Cellular distribution of RNA foci and RRM-containing proteins
We used confocal microscopy to validate in vivo some of the RRM-containing proteins that were found to interact with (CCCCGG)4 repeat RNA . We and others have previously demonstrated colocalisation of SRSF2, hnRNP A1, hnRNP H/F, and ALYREF with sense RNA foci [3, 17]. We set out to investigate the cellular distribution of the same proteins with respect to antisense RNA foci, and we also examined nucleolin and hnRNP K which are proposed to be specific binding partners of sense and antisense foci, respectively .
Approximately 50 cerebellar Purkinje neurons were examined in a blinded experiment, from a minimum of three C9ORF72-ALS cases. Simultaneous co-staining was carried out in parallel in non-C9ORF72 ALS cases and neurologically normal controls. For ALYREF, hnRNP A1, SRSF2, hnRNP H/F, and hnRNP K, the overall cellular distribution was not specifically altered in C9ORF72+ cases except for nuclear areas where colocalisation was demonstrated (Fig. 3a–e). Haeusler et al.  observed disruption of nucleolin expression from the nucleolus in cell models expressing expanded C9ORF72, but reported a variable distribution or nucleolin in C9ORF72+ CNS tissue. In agreement with this, we identified C9ORF72+ neurons which did and did not demonstrate an altered distribution pattern of nucleolin (Fig. 3f).
By IHC, we demonstrated colocalisation of SRSF2, hnRNP A1, hnRNP H/F, ALYREF, and hnRNP K in cerebellar Purkinje neurons with 34, 21, 3.4, 7.8, and 8.1 % of antisense RNA foci, respectively (Fig. 3a–e). In contrast, nucleolin was not observed to colocalise with antisense RNA foci (Fig. 3f). To validate the IHC findings, we performed in vitro Ultra-Violet (UV) crosslinking assays using radiolabelled synthetic (GGGGCC)5 or (CCCCGG)5 RNA oligonucleotides, and purified recombinant proteins synthesized in E.coli. Unlike IHC, this allows determination of direct and specific RNA:protein interactions via the formation of covalent bonds under UV light exposure. Both sense and antisense repeat RNA were observed to directly interact with hnRNP F, hnRNP A1, ALYREF, and SRSF2 proteins although the RNA-binding activity was not equal in all cases and for hnRNP A1 was relatively low (Fig. 4). In contrast, we failed to detect any direct interactions between sense or antisense repeat RNA and hnRNP K, suggesting that the previously observed colocalisation of hnRNP K with antisense RNA foci is not due to direct binding between hnRNP K and CCCCGG-repeats. The smeared appearance of certain of the proteins on the phospho image (Fig. 4) is likely to be due to the formation of covalently bonded oligomeric protein:RNA complexes. Multiple molecules of RRM-containing proteins bind to RNA oligonucleotides via inter and intra molecular interactions. We have demonstrated a similar effect previously .
Cellular distribution of RNA foci and TDP-43
We also examined the association of RNA foci with depletion of TDP-43 from the nuclei of motor neurons of seven patients with C9ORF72-ALS. Nuclear depletion and cytoplasmic mislocalisation of TDP-43 form the pathological hallmark of most subtypes of ALS, including C9ORF72-mediated disease . We have previously shown that the proportion of sense RNA foci+ motor neurons with and without nuclear TDP-43 is approximately equivalent (χ
2, p = 0.75) . As a direct comparison with this study, approximately fifty motor neurons were examined in FFPE sections from seven C9ORF72-ALS cases (Supplementary Table 5). Unlike sense RNA foci, the presence of antisense foci was significantly associated with nuclear loss of TDP-43. Seventy-seven percent of antisense foci+ motor neurons displayed loss of nuclear TDP-43 compared to 13 % of motor neurons without observable antisense foci (χ
2, p < 0.00001) (e.g. Fig. 5). A similar experiment in hippocampal CA4 subfield neurons did not reveal a significant correlation between the presence of antisense foci and nuclear loss of TDP-43, indeed no CA4 subfield neurons exhibited complete nuclear clearance of TDP-43 (data not shown).