Write a comment

König et al. aim to understand the role of TDP-43 inclusions in Alzheimer’s Disease, how those inclusions may drive toxicity, and present candidate drugs for reversing that toxicity. Another goal is to develop novel biomarkers for AD using this information.

König et al. present compelling data that link TDP-43 pathology, cytosolic double-stranded RNA, and interferon induction in human patient tissue, iPSC-derived neurons, and a neuron-like cancer cell line. They then train their drug repurposing pipeline using cryptic exon expression to identify candidate drugs for AD with TDP-43 pathology. This pipeline identified a number of tyrosine kinase inhibitors. They show that inhibition of tyrosine kinase signaling can rescue dsRNA-induced cell death by decreasing interferon signaling. Finally, they establish a link between an established TYK2 loss of function variant and decreased CXCL10 levels in human plasma.

There are two major weaknesses in the paper. First, due to reasonable technical limitations, the authors use indirect measures of cdsRNA accumulation and/or TDP-43 pathology and draw conclusions that may be consistent with their hypotheses, but do not rule out others in a satisfactory way. This leads to contradictions in their proposed model between the first and last parts of the text as written. Second, they draw a number of conclusions based on their drug treatments at concentrations that are much higher (in some places >100-fold) than the IC50s reported in the paper. Thus, the specificity of their proposed mechanisms are not fully supported by their data. These two weaknesses can be addressed as follows:

Major Points:

1. The authors stratify patients and controls by pTDP-43 pathology but do not ever tell the reader what the stratification strategy is. This is especially important considering the small number of patients in this study.

2. It is important to show that TDP43 aggregation and/or cytosolic TDP43 and/or phospho-TDP43 is co-incident with dsRNA accumulation in Figures 1A and 1B. Quantitation of this would provide very strong evidence for their proposed model. How many TDP-43 positive cells are dsRNA negative and vice-versa? Does TDP43 aggregation/mislocalization induce dsRNA accumulation in cells adjacent to the cells with aggregation/mislocalization?

3. The authors propose that cdsRNA mediates toxicity at least in part via induction of interferon signaling. However, interferon induction can be due to many factors (as mentioned in lines 142-144 and 395-404 in the text), especially in AD. The connection between TDP43 pathology/cdsRNA accumulation and interferon induction would be much better supported if the data in Extended Data Figure 2 contained cdsRNA staining and a quantification of co-localization and/or intensity of pPKR, cdsRNA, and TDP43 pathology. Similarly, a comparison of transcriptomics (Figure 1D and Extended Data 4) from AD patients with and without TDP-43 pathology would make a much stronger argument in support of their model, rather than comparing AD+TDP43 pathology versus healthy controls.

4. To prove the specificity of their hits, the authors should validate their initial CRISPR-screen phenotype with knockdown or knockout of at least IFNAR2, IRF9, and TYK2. Using these lines, they should then show that treatment with the identified tyrosine kinase inhibitors does not rescue toxicity when the above genes are knocked out or knocked down. Similar experiments should be done with knockdown of JAK2 and JAK1 to show specificity for TYK2. Relatedly, the authors begin treatment with deucravacitinib at 10nM (50-fold higher than the reported TYK2 IC50, but also 10-fold higher than the JAK1 IC50) and see almost complete rescue of the polyIC survival phenotype. Showing a concentration dependence for rescue would provide much stronger evidence for the authors’ model.

5. The model in Figure 4D does not take into account other possible sources of neuroinflammation or the role of TDP-43 pathology, as noted in the text (lines 395-404). The model should be updated so that it does not appear the authors are proposing that only cdsRNA can cause neuroinflammation to cause disease.

6. While the data is sound, it is unclear how relevant ReN VM conditioned media are to human plasma levels of CXCL10. Providing some references that show that conditioned media can be extrapolated to human plasma biomarker levels would make the argument for CXCL10 as an AD biomarker much stronger. To directly compare the mechanisms with Figure 5a, the authors could use gene editing to make the identified SNP and measure CXCL10 levels. Alternatively, directly examining CXCL10 plasma levels for AD versus healthy controls would be sufficient.

Minor Points:

1. For Figure 1C, it would be important to show that TDP43 is mislocalized with TDP43 IF.

2. The heatmap in Extended Data 4 is much more informative than in Figure 1D

3. A sentence or two introducing the RenVM cells would clarify those cells’ relevance to the reader.

4. It is unclear how hits are called in the CRISPR screen. Please add this to the methods or in the results section.

5. It would strengthen the authors’ argument that cdsRNA toxicity is mediated by interferon if they showed a gene enrichment analysis (e.g. using ENRICHR: https://maayanlab.cloud/Enrichr/) of their screen hits.

6. It would help with clarity to have the drug targets listed with the drugs in Figure 3 and Extended Data 10.

7. The authors mention that cGAS-STING activation can play a role in dsRNA-induced toxicity. If possible (not needed for publication), it would shed a huge amount of light on the relative contribution of the authors’ proposed mechanism versus cGAS-STING dsRNA sensing to perform those experiments in a cell line that does expression cGAS-STING.

8. It may be worth discussing a recent paper that identified TYK2 (PMID 39528671) as a regulator of tau levels and turnover in the discussion section.

9. Generally the paper would be improved if more of the extended data were in the main figures.

10. I cannot offer expert feedback on the technical merits of the DRIAD-SP training or framework.

Competing interests

The author declares that they have no competing interests.