Astrocytic Nonsense-mediated mRNA decay regulates calcium signaling to support synapse function and restrain anxiety

Reviewer #1 (Public review):

Summary:

Lituma and colleagues investigate the role of NMD in astrocytes, an underexplored question given that prior work on NMD in the brain has focused exclusively on neurons. Using a tamoxifen-inducible, astrocyte-specific Upf2 conditional knockout (cKO) mouse, they report that loss of astrocytic NMD causes: (1) reductions in astrocyte cell volume and surface area across hippocampus, visual cortex, and prefrontal cortex; (2) decreased excitatory synapse density, reduced dendritic spine density, and impaired synaptic engulfment; (3) deficits in basal synaptic transmission and LTP, with selective impairment of mGluR-LTD; (4) elevated spontaneous calcium transients in astrocytes; and (5) anxiety-like behavior in the elevated plus maze (EPM) and contextual fear conditioning paradigms. Transcriptomic analysis of FACS-isolated astrocytes identifies 277 differentially expressed genes, ~40% of which carry canonical NMD-inducing features, implicating pathways linked to calcium signaling, phagosome formation, and glial development. A rescue experiment using the CalEx calcium extrusion pump demonstrates partial restoration of synaptic strength and anxiety behavior when astrocytic calcium is normalized.

The study addresses an important gap in our understanding of RNA regulation in glial cells, and the overall conceptual framework is well described. The experimental design is generally appropriate, and the multi-pronged approach lends the main claims a degree of validity.

Strengths:

(1) Novelty: This is the first study to systematically examine NMD function in astrocytes in vivo. The identification of astrocytic NMD targets via RNA-seq combined with an NMD-inducing feature classifier is a meaningful methodological contribution.

(2) Multi-method approach: The authors combine morphological analysis (Imaris 3D reconstruction), synaptic markers (PSD-95, LAMP2 engulfment assay), spine density measurements, acute slice electrophysiology, two-photon calcium imaging, behavioral testing, and transcriptomics. The convergence across these methods strengthens confidence in the claims.

Weaknesses:

(1) While the transcriptomic analysis is a valuable addition, the connection between specific NMD targets and the observed calcium phenotype remains largely correlational. The authors identify Gabbr2 and Adora1 as upregulated candidates with canonical NMD features and speculate that their elevated expression drives aberrant calcium signaling. However, no validation (e.g., qRT-PCR or protein-level confirmation) of these candidates is presented. The mechanistic pathway between NMD disruption and elevated calcium is thus inferred from pathway analysis rather than demonstrated. This is a significant gap between the transcriptomic and physiological arms of the study, and the authors should be more explicit about this limitation or, ideally, provide at least one validated target.

(2) The reduction in astrocyte surface area in cKO mice is interpreted as contributing to reduced synapse contact and engulfment capacity. This is a reasonable hypothesis, but the study does not directly demonstrate that reduced astrocyte territory correlates with reduced synaptic coverage at the level of individual cells or brain regions. The temporal sequence of these events is unknown. Do morphological deficits precede synaptic changes? Clarification and qualification of this causal chain in the Discussion would strengthen the manuscript.

(3) LFS-induced LTD is unaffected, while mGluR-LTD is reduced. This is intriguing and potentially informative about astrocyte contributions to distinct LTD mechanisms, but the difference receives limited discussion. Given the relevance of mGluR signaling to calcium dynamics and the identified pathway enrichments (GPCR signaling), this specificity deserves more attention.

(4) The CTRL + CalEx condition is included in the EPM experiment but not in the electrophysiology or calcium imaging experiments, making it difficult to fully assess whether CalEx itself has off-target effects on synaptic transmission or anxiety in wild-type animals. The CTRL + CalEx EPM data (Figure 7F) appears to show a modest reduction in open arm time relative to CTRL, which, if robust, would suggest that excessive calcium reduction in astrocytes is also anxiogenic. This finding would be physiologically relevant and deserves comment.

Reviewer #2 (Public review):

Astrocytes are highly responsive to their environment and play a range of critical roles in brain function. Lituma et al. theorize that one mediator of that responsiveness is the regulation of RNA stability. They therefore undertake an assessment of astrocytes missing Upf2, a protein required for mRNA degradation via nonsense-mediated decay. This is an interesting study, approaching astrocyte biology from a novel angle. The authors take on an ambitious set of experiments, spanning morphological assessment, synaptic engulfment, electrophysiology, behavior, and calcium imaging.

The authors show convincing data that knocking out Upf2 in astrocytes impairs synaptic plasticity, affects behavior, and changes the complement of astrocytic mRNA. These results, in and of themselves, are intriguing and suggest that NMD is an important biological process in astrocytes, warranting further study.

My primary concern is whether the authors may be largely studying dying cells. The idea that NMD disruption has a dramatic effect on astrocyte morphology is an intriguing idea, but it is not fully established here. The nuclei in the example cKO morphology images appear small and/or fragmented. This raises concerns that the authors did not ensure that they had the full 3D morphology of the astrocyte in the section, and the cell is in part cut off, which would compromise any data on the morphology. The authors state that the tissue was sectioned at 70 um. The diameter of an astrocyte in the adult mouse brain is typically between 50 and 70 um. Unless astrocytes are perfectly positioned in the center of the slice, at this thickness, the majority of astrocytes will almost certainly be partially cut off. More detail on how cells were chosen and what quality control metrics were implemented would alleviate concerns here. An alternative possible explanation for these small/fragmented nuclei is that cKO astrocytes may be unhealthy to the point that they are actively dying. Using the transgenic ZsGreen label, the authors state that they observe a size change (Figure S4); this is not readily apparent and is not quantified in any way. It does appear from these images that there may be a loss of some astrocytes; cell death, which would also be an interesting finding, is a fundamentally different process than morphologic restructuring in living cells. The authors do attempt to count astrocytes (Figure S6B), but do so with GFAP. This is a fundamentally flawed approach. Because GFAP is not readily detectable in most healthy astrocytes in most gray matter regions, GFAP should not be used to quantify astrocyte numbers; this experiment should be repeated with a better marker, such as Aldh1l1, Sox9, etc.

Synaptic engulfment: This is an extraordinarily high degree of engulfment in the control animals compared to many published studies, leading to concern as to the technical approach. Indeed, the overall low level of PSD-95 signal in control conditions in adult mice is concerning as to the technical accuracy of the approach. It is unclear exactly how the investigators labeled the astrocytes; presumably via the ZsGreen label, but it is never stated, and the only images shown are the highly processed Imaris renderings. The small astrocytic processes, or leaflets, that make up the vast majority of the astrocytic arbor are on the order of 100nm in diameter. The processes shown in Figure 2B are, according to the scale bar, at least 20x that size. It is difficult to have much faith in these results as currently presented.

The signal-to-noise ratio of the GCaMP experiments is worryingly low, likely responsible for the abnormally low dF/F in all conditions and the lack of significant change between control and CalEx, when control astrocytes should show a much higher GCaMP signal than any CalEx-expressing astrocyte. That said, the higher Ca++ in Upf2 KO astrocytes is intriguing. Given the roles of elevated calcium in cell death, this may reflect cells that are unhealthy to the point that they are starting to die.

The authors conduct a FACS-based analysis of astrocytic mRNA from control vs Upf2-KO, with intriguing results. An important caveat, though, is that a large amount of astrocytic mRNA is in the processes. If mRNA stability is being actively and rapidly regulated, it seems likely that the mRNA in the processes would be the most relevant population of regulated mRNA. FACS-based approaches to astrocyte purification will, as robustly shown elsewhere, strip off those processes. Particularly given that the authors have shown that the processes may be the most actively changing astrocytic compartment with Upf2 KO, this is a strange choice of technique vs. something like Ribotag that would preserve the mRNA in processes. At least, there should be some discussion regarding using FACS for this analysis and the consequences for profiling mRNA in astrocytic processes.

Minor points:

(1) The use of the Aldh1l1-CreER mouse is a strong choice and has been shown to be highly astrocyte-specific. Combining that transgenic mouse with viruses driven by different forms of the GFAP promoter is quite bizarre in several ways. First, GFAP-dependent AAVs have been shown repeatedly to have significant neuronal leak. Second, these mice are, in all cases, receiving two different viruses, driven by different forms of the GFAP promoter, and the non-Cre virus is not Cre-dependent (vs. a much more standard approach of using a Cre-dependent second virus to ensure that all analyzed cells received both viruses). The authors mention that "this experimental design ensures that phenotypes are not caused by an acute effect of tamoxifen." It is certainly true that tamoxifen is not a biologically neutral molecule. However, the mice still receive tamoxifen, both in these morphology virus experiments and in almost all other experiments. This experimental approach is not inherently bad, nor does it necessarily invalidate the data (although the near-certain neuronal contamination due to the GFAP promoter-driven viruses is a concern). It is, however, convoluted in ways that appear unnecessary. If there is a strong rationale for this approach beyond the tepid explanation already present, it should be explicitly mentioned.

(2) The characterization of the knockout is incomplete. While the authors should be applauded for their attempts to phenotype the cells in which they observe Cre-mediated recombination, there are issues with their technical approach. Most importantly, and an issue that affects other analyses in the paper as well: the vast majority of astrocytes in the healthy cortex do not express GFAP. Therefore, using GFAP to claim high astrocyte specificity and efficiency is a fundamentally flawed approach. Second, MBP is a myelin marker, not a cytoplasmic marker, and would not successfully colocalize with a cytoplasmic marker like ZsGreen even if recombination in oligodendrocytes did occur. Third, recombination at one set of LoxP sites is not a reliable indicator of recombination at other sites. Recombination efficiency is highly dependent on the spacing between the LoxP sites and cannot be reliably extrapolated to other floxed genes without validation. Finally, the most likely culprit for off-target recombination with Aldh1l1-CreERT2 (or other astrocyte-selective Cres, and certainly the GFAP-based viral promoters) is neurons, which the investigators did not test for. Neuronal Aldh1l1-CreERT2 leak is most likely to occur in the hippocampus. With the images shown in Fig S3, it is unclear whether it is possible to convincingly colocalize Upf2 staining with a cytosolic marker of all astrocytes, such as Aldh1l1 or S100b, but such data would be more appropriate. An alternative approach to validation would be in situ hybridization.

(3) Supplementary Table 2 should include gene IDs, not just Ensemble IDs.

(4) It is not fully clear what the investigators are denoting as a spine in Figure 2E; the two images do not appear to have the large degree of difference that the quantification suggests. The oversaturation of the signal complicates assessment.

(4) A more detailed discussion of the rationale behind the timeline would be helpful. What is the half-life of Upf2, and how rapidly do NMD genes build up upon Upf2 disruption? In particular, in the case of virus experiments, the timeline is quite fast: ~2.5 weeks from injection to analysis. ssAAV expression takes over a week to reach appreciable levels.

Reviewer #3 (Public review):

Summary:

The authors investigate mRNA targets of the nonsense-mediated decay (NMD) pathway in astrocytes and link the dysfunction of NMD in astrocytes to aberrant synaptic transmission that has downstream effects on behavior. Specifically, they find a link between the aberrant synaptic transmission with elevated spontaneous calcium signaling in astrocytes, and functionally they demonstrate that manipulating astrocyte calcium signaling with CalEx modulates astrocyte calcium signaling towards wildtype levels and improves anxiety behavior. They investigate the astrocyte calcium signaling changes in Upf2 conditional knockout mice in several brain regions that have been linked to anxiety behavior, including the hippocampus and prefrontal cortex. They also observe aberrant astrocyte calcium signaling in the visual cortex, demonstrating that dysfunction of the NMD pathway in astrocytes has widespread effects on synaptic transmission in various brain regions. This work identifies, through RNA-Sequencing, potential mRNA targets of NMD in astrocytes, and shows that pathway enrichment of these targets highlights calcium signaling. Altogether, this work highlights the importance of the basic cellular process of NMD in astrocytes, which are known to have extensive local translation of proteins in their perisynaptic processes. NMD may be particularly important in astrocytes due to their intimate association of processes with neuronal synapses, and the authors suggest that alterations to NMD function in astrocytes may be an important avenue for future investigation in neurodevelopmental disorders.

Strengths:

Altogether, this work is a critical foundation for future research into astrocyte contributions to neurodevelopmental disorders. The authors do a thorough characterization of astrocyte conditional Upf2 knockout mice in several brain regions. They present a complete story that connects molecular events (NMD pathway regulation of mRNA degradation) to astrocyte regulation of circuit activity to organismal behavior. The electrophysiological analysis is thorough, and the manipulation of calcium activity ties astrocyte calcium activity to anxiety behavior. The RNA-sequencing dataset is useful to the scientific community and provides a resource of candidate molecules that might be dysregulated in neurodevelopmental disorders.

Weaknesses:

The study suffers from some overstated claims and a lack of statistical rigor in some experiments, as detailed below.

(1) The title states that "Astrocytic Nonsense-mediated mRNA decay regulates calcium signaling to support synapse function and restrain anxiety". The term "restrain anxiety" implies that the NMD pathway has a direct effect on a molecular switch to control anxiety. Anxiety behavior is a complicated process, controlled by many biological phenomena and synaptic transmission in the circuit as a whole, and is not directly linked to a specific NMD mRNA target. This title is overstating the findings of the study.

(2) In general, the first figures (1-2) suffer from low power (N = 3) and statistical rigor. The statistics are inflated by analyzing individual fields of view and per-cell data rather than performing the statistics on the average of biological replicates. It is preferable to show the biological replicate data so that readers can observe the natural biological variability between replicates.

(3) The claim that astrocytes have decreased engulfment of synapses in the Upf2 conditional knockout mice is not strongly substantiated by the data. The resolution of confocal microscopy and the static nature of histological images make it difficult to measure synaptic engulfment as an active process. Additionally, the metric of quantifying the % occupancy of PSD95 puncta within the total astrocyte volume may be skewed due to overall differences in cell size (shown in Figure 1). There is not much discussion of how a decrease in astrocyte engulfment of synapses may lead to decreased synapse number. To the contrary, one might expect decreased engulfment to result in increased synapse density.

(4) The authors use Gfap as a marker to count astrocyte cell number and assess if there are changes in cell number between genotypes (Figure S6). However, Gfap does not label all astrocytes in the cortex and, in fact, is rather an aberrantly expressed marker in conditions of inflammation, as opposed to the hippocampus, where Gfap is basally expressed in all astrocytes. In the cortex, there seems to be a trend for reduced Gfap in the conditional knockout mice, which may suggest differences in astrocyte molecular signatures rather than cell numbers. Another astrocyte marker, like Aldh1L1, will be more accurate to assess this question histologically.

(5) The authors state that "Preventing abnormally high basal calcium activity in NMD-deficient astrocytes restores normal excitatory synapse function...". However, this claim is not substantiated by the data. CalEx manipulation certainly shifts the input-output curve but does not restore to wildtype baseline levels (Figure 6E). Additionally, synapse number does not appear to be restored to wildtype levels (Figure 6D - although the p-value for this comparison is now shown). The investigators do observe improvements in anxiety phenotypes, suggesting there is some modulation of circuit activity, but the claim that CalEx manipulation restores baseline synaptic transmission is not supported.

Author response:

We thank the reviewers for their careful reading of our manuscript and for providing positive, constructive feedback. In particular, we thank the reviewers highlighting the several strengths of our study.

To address the reviewers’ major concerns, we will revise the presentation of our main findings (specifically data/animal vs data/ROI), provide more clarity in the Results, Methods, and Discussion sections, and modify the title to better reflect these nuances.

Additionally, we will perform the following new experiments:

(1) Astrocytic Marker Validation: To further confirm comparable astrocyte cell counts between the CTRL and Upf2-cKO conditions, we will perform immunostainings using Aldh1L1, Sox9, or S100b instead of GFAP.

(2) NMD Candidate Validation: To validate top candidate NMD target transcripts, we will perform immunostainings or qRT-PCR for Gabbr2, Adora1, S100b, or Cldn9.

(3) Sample Size Expansion: To strengthen the morphological and PSD-95 quantifications, we will increase the sample size (N) by incorporating additional animals.

(4) Mechanistic Timeline & Phenotype Linkage: We value the reviewer’s comment regarding the timeline of morphological and Ca²⁺ phenotypes. To gai insight into whether these phenotypes are independent or linked, we will perform 3D reconstructions in CalEx conditions to assess whether Ca²⁺ restoration rescues astrocyte morphology in Upf2-cKO mice. This will allow us to determine if increased Ca²⁺ activity is upstream of the morphological alterations. Taken together, we believe that incorporating these manuscript revisions will strengthen the clarity and conclusions of our work. We thank the reviewers for their time and careful evaluation of our study.