Re-analysis of GEO Datasets GSE286094 & GSE286095

ATG7 Deficiency Reshapes Microglia Biology in Alzheimer’s Disease

Integrated single-cell and bulk RNA-seq analysis of Cai et al. 2025 (JEM)

Dataset: GSE286094 / GSE286095 · Mus musculus · Microglia · Atg7 fl/fl vs Atg7 dMG · 5xFAD AD model · 11 scRNA-seq + 6 bulk RNA-seq samples

Executive Summary

Key Findings

22 high confidence

Analysis Tracks

Quality → Trajectory → DE → Regulatory → Validation

Novel / Extending

17 orthogonal, 11 extending

Integrative Themes

Cross-cutting biological mechanisms

Central Mechanistic Model

ATG7 Deficiency Triggers Cascading Multi-System Failure in Microglia

Microglia-specific ATG7 deletion triggers a cascading failure across multiple biological systems. The primary mechanism is UPR impairment (especially the ATF6 arm), which creates a 'double vulnerability': cells simultaneously lose proteostasis protection AND gain ferroptotic susceptibility. In DAM cells, this manifests as a 'signaling desert' where 8 growth/survival pathways shut down, leaving only p53 and TNFa active — consistent with cells transitioning toward death rather than productive activation. This internal crisis is compounded by communication isolation (2:1 loss ratio of intercellular signals), loss of critical Adam10→Trem2 and Apoe→Lrp1 axes, and non-cell-autonomous immune remodeling (45% neutrophil depletion, 76% T cell enrichment). Importantly, ATG7 and 5xFAD effects are statistically ADDITIVE, not synergistic — the amplified disease response arises from combined additive effects rather than true biological interaction.

Novelty Classification

17 Orthogonal (novel)

11 Extends prior work

2 Confirms known

Of 30 key findings, 17 are entirely novel (orthogonal to prior work), 11 extend published results with new mechanistic detail, and 2 confirm established findings. Zero findings contradict the original study.

Integrative Themes

UPR-Ferroptosis Double Vulnerability

UPR impairment (especially ATF6) and ferroptosis susceptibility are anti-correlated at the single-cell level, creating coordinated vulnerability

5 supporting findings

DAM Signaling Desert & Communication Isolation

DAM cells experience coordinated shutdown of growth/survival pathways and lose twice as many intercellular interactions as they gain

5 supporting findings

Non-Functional Overexpression Pattern

Esr1 gene massively upregulated but TF activity paradoxically down; similar disconnect in JAK-STAT (gene up, signaling down)

2 supporting findings

Homeostatic Microglia Bifurcation

ATG7 loss splits homeostatic microglia into hyperactive (depleted) and senescent (accumulating) sub-populations with distinct signaling profiles

3 supporting findings

Non-Cell-Autonomous Immune Remodeling

Microglia-intrinsic autophagy loss remodels the brain immune landscape with selective cell type effects

3 supporting findings

Additive Biology with Pathway Saturation

Gene-level effects are additive but translation/ribosome pathways show sub-additivity, suggesting compensatory capacity saturation

2 supporting findings

Data Quality & Processing

90,475

Cells Retained

from 96,376 input

93.9%

Retention Rate

5,901 cells removed

2,884

Doublets Removed

3.0% of total

15,310

Genes Retained

3,000 HVGs selected

Bulk Samples

of 6 input (1 excluded)

wt3

Outlier Excluded

26× lower sequencing depth

scRNA-seq Cell Counts: Before vs After QC

QC filters applied: ≥500 genes, ≥1,000 UMIs, ≤10% mitochondrial reads, plus doublet removal via scrublet. AD_WT2 had the highest doublet rate (6.8%), while WT1 had the lowest (0.03%).

Bulk RNA-seq Library Sizes

Sample wt3 has only 3.8% of the median library size (1.26M vs 33.2M), making it an extreme outlier that was excluded from downstream analysis. Sample ko2 (3.8M) is low-depth but retained with monitoring.

Per-Sample QC Metrics

Sample	Group	Pre-filter	Post-filter	Removed	Retention %	Doublets	Med. Genes	Med. UMI	Med. Mito %
AD_KO1	Atg7_dMG-5xFAD	7,322	6,782	540	92.6%	179	1,475	3,437	1.29%
AD_KO2	Atg7_dMG-5xFAD	5,954	5,646	308	94.8%	58	1,479	3,686	1.54%
AD_KO3	Atg7_dMG-5xFAD	6,477	6,167	310	95.2%	146	1,396	2,942	1.34%
AD_WT1	Atg7fl/fl-5xFAD	10,513	9,927	586	94.4%	336	1,474	3,557	1.43%
AD_WT2	Atg7fl/fl-5xFAD	16,276	14,573	1,703	89.5%	1,099	1,274	2,893	1.48%
KO1	Atg7_dMG	13,745	12,465	1,280	90.7%	806	1,273	2,562	1.34%
KO2	Atg7_dMG	5,944	5,590	354	94%	111	1,248	2,367	1.23%
KO3	Atg7_dMG	5,855	5,554	301	94.9%	113	1,434	2,909	1.23%
WT1	Atg7fl/fl	12,005	11,805	200	98.3%	4	1,500	3,213	1.14%
WT2	Atg7fl/fl	8,494	8,280	214	97.5%	12	1,558	3,455	1.29%
WT3	Atg7fl/fl	3,791	3,686	105	97.2%	20	1,439	2,842	1.36%

Batch Integration Quality

3,000

HVGs Selected

scVI Latent Dims

Epochs Trained

816.6

Final Train ELBO

scVI integration (30 latent dims, 2 layers, negative binomial likelihood) successfully corrected batch effects across 11 samples while preserving biological variation. ELBO converged rapidly over 20 epochs, indicating stable model training.

High-Quality Input Data

The low overall cell loss (93.9% retention) and low median mitochondrial content (1.3%) indicate the scRNA-seq data was already high quality from the 10x pipeline. Doublet rates varied substantially, with AD_WT2 (6.8%) and KO1 (5.9%) highest, likely reflecting higher cell loading densities.

Bulk Outlier Exclusion

Sample wt3 was excluded due to a 26× lower sequencing depth that would bias DESeq2 size factor estimation and inflate false discovery rates. After exclusion, the 5-sample design (2 WT + 3 KO) retains adequate power, with PC1 (50.1%) capturing genotype differences.

Cell Atlas & Composition

Cell Types

at resolution 0.8

64.3%

Microglia

58,135 cells

8.7%

DAM Fraction

7,868 cells

5.7×

DAM Enrichment

5xFAD vs non-5xFAD

Leiden clustering at resolution 0.8 identified 27 clusters annotated into 12 distinct cell type categories using a hierarchical two-pass approach. Microglia comprise 64.3% of all cells, with 6 homeostatic sub-clusters (HM_1–HM_6) revealing greater heterogeneity than typically reported. DAM cells show 5.7× enrichment in 5xFAD groups, confirming disease-associated microglial activation. Non-microglia populations including neutrophils (9.8%), T cells (8.3%), and B cells (5.2%) were entirely unexplored by the original authors.

Cell Type Composition Across Genotype Groups

Percentage of each cell type across experimental groups. DAM is markedly expanded in 5xFAD groups (10.3–21.9%) while HM_1 dominates in Atg7^fl/fl controls (48.9%). Hover to see individual cell type percentages.

Cell Type Marker Scores by Cluster

Z-score of cell type marker gene module expression per cluster. Strong diagonal pattern confirms annotation quality. High scores (red) indicate marker enrichment; low scores (blue) indicate depletion. Prolif clusters show strong proliferation signatures alongside their lineage identity.

Blvrb as FTM DiscriminatorNovel

Ferritin-transporting microglia (FTM) cannot be reliably identified by Fth1/Ftl1 expression alone, as 99.6% of all cells express Fth1. Instead, Blvrb (biliverdin reductase B) serves as the key discriminator, expressed in approximately 32% of cells with strong enrichment in the FTM cluster. This practical annotation guidance has not been previously documented.

Homeostatic Microglia HeterogeneityNovel

Six distinct homeostatic sub-clusters (HM_1–HM_6) were resolved, compared to 1–3 typically reported in published scRNA-seq studies. HM_1 dominates in WT controls (48.9%) while HM_2 expands in ATG7-deficient conditions (32.0%), suggesting ATG7 loss shifts the homeostatic equilibrium towards an alternative transcriptional state.

Microglia Trajectory & Pseudotime

58,135

Microglia Analyzed

subsetted for trajectory

0.47

5xFAD Effect (rbc)

pseudotime rank-biserial

0.22

ATG7 Effect (rbc)

pseudotime rank-biserial

p = 0.17

DAM ATG7 Effect

not significant

PAGA graph abstraction and Palantir diffusion pseudotime computed on 58,135 microglia reveal a hub-like topology rather than a simple linear chain. Pseudotime ordering proceeds from HM (0.002) through DAM (0.012) and IFN (0.023) to TM (0.093), FTM (0.159), and Prolif (0.228). The 5xFAD amyloid pathology shifts microglia pseudotime upward (rbc = 0.47), indicating a larger proportion of cells in activated states. ATG7 deficiency has a smaller but significant effect (rbc = 0.22). Critically, within DAM cells, ATG7 status does not alter pseudotime (p = 0.17), suggesting ATG7 affects entry rates into DAM rather than the DAM transcriptional program itself.

PAGA Graph — Microglia Cluster Connectivity

PAGA abstracted graph of microglia subtypes. Node size reflects cell count; edge thickness reflects PAGA connectivity weight. The strongest connections are TM↔Prolif (0.41), IFN↔Prolif (0.37), and DAM↔IFN (0.35). FTM connects most strongly to TM (0.27), not to HM or DAM, suggesting FTM derives from the transitional state.

Median Pseudotime by Microglia Subtype

Palantir diffusion pseudotime ordering of microglia subtypes. HM cells occupy the earliest pseudotime, followed by a rapid transition to DAM and IFN. FTM and Prolif occupy the latest pseudotime positions, consistent with terminal differentiation states.

FTM Derives from Transitional MicrogliaNovel

Ferritin-transporting microglia (FTM) connect most strongly to transitional microglia (TM) via PAGA (w = 0.27), not to HM (0.03) or DAM (0.03). This suggests iron accumulation occurs specifically in microglia that have already entered a transitional/inflammatory state, rather than branching directly from homeostatic or disease-associated populations.

ATG7 as Gatekeeper, Not ReprogrammerNovel

ATG7 deficiency increases the proportion of activated microglia (rbc = 0.22) but does not alter pseudotime within DAM cells (p = 0.17). This dissociates the effect on cell state entry rates from within-state transcriptional changes, suggesting ATG7 acts as a gatekeeper controlling the probability of transitioning to DAM rather than modifying what DAM cells express.

Gene Expression Dynamics Along Pseudotime

889

Significant Genes

FDR < 0.05

6,546

Genes Tested

interaction model

Ribosomal Proteins

in top 30 interaction genes

ER Chaperones

in top 30 interaction genes

A genome-wide interaction model (expression ~ pseudotime × ATG7 status) identified 889 genes with significantly altered pseudotime dynamics in ATG7-deficient microglia (FDR < 0.05, from 6,546 tested). The top interaction genes are dominated by ribosomal proteins (9 of top 30) with negative interaction coefficients, indicating impaired translation machinery scaling during activation. ER chaperones (Hsp90b1, Sdf2l1, Manf) in the top 30 show positive interaction coefficients, reflecting compensatory upregulation in KO microglia along the activation trajectory. Two additional ER chaperones (Dnajb11, Ppib) rank just outside the top 30 with similar positive coefficients.

Top 30 ATG7 × Pseudotime Interaction Genes

Top 30 genes ranked by F-statistic from the ATG7 × pseudotime interaction model. Red bars indicate genes with increased expression along pseudotime in KO relative to WT (positive interaction coefficient); blue bars indicate decreased expression in KO. ER chaperones (Hsp90b1, Manf, Sdf2l1) show positive interaction, while ribosomal proteins and MHC-II genes show negative interaction.

UPR genes show negative correlation with pseudotime in WT (Calr r = −0.16, Hspa5 r = −0.18) but near-zero in KO (r ≈ −0.02 to −0.05), indicating the UPR gradient is abolished. Solid lines = non-5xFAD; dashed = 5xFAD.

DAM signature genes (Trem2, Apoe) and iron-related genes (Fth1, Blvrb) show distinct pseudotime dynamics across genotypes. Blvrb (interaction coef = −1.79) marks FTM identity with altered trajectory dynamics in KO, while Trem2 and Apoe define DAM activation along pseudotime.

UPR Gradient Abolished in ATG7-KONovel

WT microglia dynamically modulate UPR gene expression along the activation trajectory (Calr r = −0.16, Hspa5 r = −0.18), progressively reducing ER stress response as they transition toward activated states. In ATG7-KO microglia, this gradient is completely flattened (r ≈ 0), suggesting autophagy is required for dynamic UPR remodeling during microglial state transitions — not just for maintaining UPR levels.

Impaired Translation Scaling & Compensatory ER Chaperones

Ribosomal proteins (9 in top 30) show systematically negative interaction coefficients, indicating KO microglia fail to upregulate translation machinery during activation. Conversely, ER chaperones (Hsp90b1 F = 230.0, Sdf2l1 F = 127.7, Manf F = 120.9) are compensatorily upregulated, suggesting an alternative protein quality control pathway engages when canonical autophagy-mediated proteostasis fails.

Differential Abundance Analysis

6,505

Neighborhoods

tested (Milo)

28.4%

Significant

1,845 at FDR < 0.1

−2.00

HM_1 Mean logFC

806 depleted in KO

+1.21

HM_2 Mean logFC

541 enriched in KO

Milo differential abundance testing on 6,505 neighborhoods (90,475 cells) with a GLM design of ~ disease_model + atg7_status. Positive logFC indicates enrichment in ATG7-KO; negative indicates depletion. The most striking finding is a homeostatic microglia bifurcation: HM_1 is strongly depleted (logFC −2.00) while HM_2 is enriched (+1.21), revealing a qualitative shift in homeostatic sub-states rather than simple depletion. DAM expansion (logFC +1.30), neutrophil depletion (logFC −1.00), and T cell enrichment (logFC +0.44) further define the ATG7-dependent immune landscape. Concordance with cluster proportions: Spearman ρ = 0.912.

Milo Differential Abundance by Cell Type

Mean neighborhood logFC per cell type from Milo (design: ~ disease_model + atg7_status). Green bars = enriched in ATG7-KO, red bars = depleted. HM_1 and HM_2 show the strongest and most divergent effects, highlighting the homeostatic microglia bifurcation. Neutrophil populations (Neutrophil_1/2/3) are consistently depleted despite ATG7 being knocked out only in microglia, suggesting indirect effects on myeloid cell recruitment.

Homeostatic Microglia BifurcationNovel

ATG7 loss does not simply deplete homeostatic microglia — it specifically shifts the compartment from HM_1 (logFC −2.00, 806 depleted) to HM_2 (logFC +1.21, 541 enriched). This qualitative rather than quantitative change suggests autophagy maintains a specific homeostatic transcriptional program; its loss shifts microglia to an alternative sub-state that may represent a “primed” or pre-activated configuration.

Neutrophil DepletionNovel

Neutrophils are significantly depleted in ATG7-KO brains (60 neighborhoods, mean logFC −1.00) after controlling for 5xFAD status. Since ATG7 is knocked out only in microglia (CX3CR1-Cre), this indirect effect suggests autophagy-deficient microglia alter the chemokine milieu, reducing neutrophil recruitment or retention — a non-cell-autonomous consequence not examined by the authors.

Differential Gene Expression

Comparisons

scRNA-seq pseudobulk DE

HM DEGs

ATG7 effect, all conditions

2.4×

Disease Amplification

146 vs 60 DEGs in KO vs WT

167

Bulk DEGs

20mo microglia (109↑ 58↓)

+5.85

Esr1 log₂FC

bulk 20mo (padj < 10⁻²⁸⁷)

+3.84

Cdkn2a log₂FC

p16, 20mo only (padj = 0.046)

Pseudobulk differential expression with PyDESeq2 across 7 comparisons identifies 41 DEGs in homeostatic microglia (C1: ATG7 effect, all conditions), with UPR genes Calr (log₂FC = −1.27) and Hspa5 (−1.12) confirmed downregulated. The most significant gene is Esr1 (estrogen receptorα), upregulated ~32-fold in KO microglia (log₂FC = 5.0, padj = 2.9 × 10⁻¹⁰⁷). The 5xFAD disease response is amplified 2.4× in KO background (146 vs 60 DEGs) with only 15 shared genes, indicating qualitatively altered disease response. Bulk RNA-seq at 20 months confirms 167 DEGs (109 up, 58 down) with Esr1 even more dramatic (log₂FC = 5.85) and senescence marker Cdkn2a (p16) now significant alongside persistent Cdkn1a (p21).

Pseudobulk DEGs by Comparison

DEG counts per comparison (PyDESeq2, FDR < 0.05, |log₂FC| > 1). C5 (5xFAD in KO HM) yields 146 DEGs — 2.4× more than C4 (5xFAD in WT HM, 60 DEGs) — demonstrating amplified disease response in ATG7-deficient background. C7 (ATG7 in FTM) shows only 10 DEGs, confirming FTM is largely ATG7-independent.

Bulk RNA-seq Volcano (20 Months)

167 significant DEGs from bulk RNA-seq of 20-month FACS-sorted microglia (2 WT vs 3 KO, wt3 excluded). −log₁₀(padj) capped at 50 for readability; Esr1 true value = 287. Dashed lines mark significance thresholds.

Bulk DEGs — 167 Significant Genes

Gene	log₂FC	padj ▲	baseMean	Direction
Esr1	+5.85	8.9e-288	4930.4	↑ KO
Pla2g7	+4.29	1.5e-124	2226.2	↑ KO
Ddost	-2.42	7.0e-43	3454.4	↓ KO
Wfdc17	+4.47	2.2e-19	298.2	↑ KO
Saraf	-1.80	3.1e-17	6243.4	↓ KO
Lyl1	-1.93	3.6e-16	848.4	↓ KO
Fscn1	-1.31	2.8e-15	3387.3	↓ KO
Serp1	-1.52	4.1e-13	1533.5	↓ KO
Mfsd1	-1.36	8.6e-13	1872.3	↓ KO
Tm2d1	-2.47	7.4e-12	365.2	↓ KO
Mrc1	+2.64	6.5e-11	757.5	↑ KO
Cx3cr1	-1.09	6.5e-11	143248.6	↓ KO
Ncln	-1.45	1.8e-10	1780.5	↓ KO
Fkbp1a	-1.57	2.0e-10	842.7	↓ KO
Slc39a4	+1.50	2.7e-10	1284.2	↑ KO
Anxa2	+2.58	3.1e-10	292.9	↑ KO
Slc2a8	-2.21	3.1e-10	457.4	↓ KO
A530064D06Rik	+2.97	7.6e-10	277.3	↑ KO
Osgin1	+1.69	2.7e-9	680.3	↑ KO
Atg7	-1.36	2.8e-9	854.7	↓ KO
Hoxc8	+7.93	3.2e-9	79.9	↑ KO
Ccnd2	+3.40	4.8e-8	132.4	↑ KO
Tm9sf2	-1.03	5.7e-8	2758.9	↓ KO
Ifnar2	-1.14	5.7e-8	1790.5	↓ KO
Cox5a	-2.18	1.2e-7	223.0	↓ KO
Acaa1a	-2.07	1.2e-7	737.3	↓ KO
Sdf2l1	-1.48	1.4e-7	2506.6	↓ KO
Ms4a7	+2.59	4.0e-7	421.0	↑ KO
Acss1	-1.29	4.1e-7	991.9	↓ KO
Vcam1	+1.29	4.1e-7	2186.2	↑ KO
Blvrb	+1.16	5.2e-7	2290.4	↑ KO
Fez2	-1.19	5.2e-7	1676.6	↓ KO
C4b	+2.12	5.4e-7	780.1	↑ KO
Capn11	+5.60	7.2e-7	89.6	↑ KO
Cyc1	-1.88	9.3e-7	653.5	↓ KO
Gdf3	+4.73	1.9e-6	61.0	↑ KO
Lyz1	+2.34	2.5e-6	155.2	↑ KO
Tmem154	+1.35	3.4e-6	678.9	↑ KO
Gpr165	-1.31	4.0e-6	928.2	↓ KO
Clec12a	+4.37	6.0e-6	91.1	↑ KO
Trpv4	+1.23	1.3e-5	572.1	↑ KO
Ryr2	+4.96	1.4e-5	74.9	↑ KO
Pigx	-1.57	1.4e-5	518.1	↓ KO
Xdh	+2.28	1.9e-5	208.0	↑ KO
Dab2	+1.37	2.1e-5	1176.3	↑ KO
Sell	+6.09	4.5e-5	43.5	↑ KO
Gmpr	-1.10	5.0e-5	948.7	↓ KO
Tmem205	+1.28	5.1e-5	434.5	↑ KO
Uck1	-1.58	5.1e-5	328.1	↓ KO
Mgl2	+3.04	5.3e-5	172.0	↑ KO
Cog8	-1.85	7.9e-5	227.1	↓ KO
Ccl7	+3.12	0.0001	118.1	↑ KO
Cd209f	+5.38	0.0001	40.6	↑ KO
Mgst1	+2.99	0.0001	164.5	↑ KO
Trp53i13	-2.54	0.0001	135.8	↓ KO
Cfp	+2.45	0.0002	131.8	↑ KO
Aoah	+2.41	0.0002	212.8	↑ KO
Gimap1	-2.09	0.0002	220.7	↓ KO
Ddo	+6.61	0.0002	30.9	↑ KO
Fam53a	-1.44	0.0002	373.3	↓ KO
Cdkn1a	+1.42	0.0002	728.6	↑ KO
Crlf2	-2.04	0.0002	285.1	↓ KO
Gstm1	+1.26	0.0003	729.6	↑ KO
Dnase1l3	+1.38	0.0004	446.2	↑ KO
Cp	+1.24	0.0005	382.9	↑ KO
Spsb1	-1.07	0.0005	1131.6	↓ KO
Cd93	+3.10	0.0006	83.0	↑ KO
Srd5a3	-1.10	0.0006	611.4	↓ KO
Pf4	+2.46	0.0006	148.9	↑ KO
Psme3	-1.37	0.0007	604.2	↓ KO
Cd244a	+1.48	0.0008	247.3	↑ KO
Zfhx4	+5.52	0.0008	55.6	↑ KO
Clec10a	+3.80	0.0009	60.0	↑ KO
Pdgfrb	+5.23	0.0009	24.6	↑ KO
Nrbf2	-1.48	0.0012	243.4	↓ KO
Ltbp1	+12.28	0.0012	26077.0	↑ KO
Pirb	+1.01	0.0013	1325.4	↑ KO
Ifitm2	+2.24	0.0014	127.9	↑ KO
Kirrel2	+5.88	0.0014	50.5	↑ KO
Gnaz	+4.82	0.0022	35.3	↑ KO
Gramd1b	+1.09	0.0022	507.4	↑ KO
Ripor2	+1.27	0.0022	420.4	↑ KO
Psmd3	-1.31	0.0022	506.8	↓ KO
Pmaip1	+2.22	0.0022	160.4	↑ KO
Ccl8	+2.56	0.0025	138.8	↑ KO
Tusc3	-1.51	0.0027	230.4	↓ KO
Iigp1	+2.07	0.0029	140.0	↑ KO
Pop5	-1.85	0.0031	130.4	↓ KO
Zdbf2	+4.54	0.0031	35.6	↑ KO
Vwa3b	+10.43	0.0034	2696.3	↑ KO
H2-Eb1	+3.45	0.0035	2042.6	↑ KO
Creld2	-1.10	0.0042	566.2	↓ KO
Colec12	+7.13	0.0044	23.0	↑ KO
Meis3	-1.42	0.0050	253.2	↓ KO
Msr1	+1.57	0.0052	180.2	↑ KO
Stap1	+5.22	0.0052	24.6	↑ KO
Klhdc8b	-1.05	0.0055	434.9	↓ KO
Klra2	+3.40	0.0058	40.6	↑ KO
Shroom4	+6.19	0.0066	25.6	↑ KO
Cpq	+1.00	0.0068	800.1	↑ KO
Il18bp	+1.45	0.0069	280.4	↑ KO
Gbp2	+1.16	0.0069	316.7	↑ KO
Dda1	-1.24	0.0069	312.8	↓ KO
Pygl	+1.58	0.0074	328.5	↑ KO
Vmn1r43	+4.66	0.0081	40.2	↑ KO
Gpatch11	+1.10	0.0083	323.4	↑ KO
Cxcl2	+3.41	0.0093	32.4	↑ KO
Pdia5	-1.22	0.0095	333.8	↓ KO
Gas7	+1.95	0.0101	219.9	↑ KO
Gm29666	+2.58	0.0116	129.4	↑ KO
Mfsd14a	-1.03	0.0116	348.5	↓ KO
Nedd4	+4.45	0.0116	26.1	↑ KO
Tssk5	+5.73	0.0125	28.0	↑ KO
Pcdh15	+4.64	0.0125	43.9	↑ KO
H2ax	-2.42	0.0132	67.0	↓ KO
Flot2	-1.11	0.0140	356.9	↓ KO
Plppr4	-3.04	0.0141	38.5	↓ KO
Fbxl15	-3.14	0.0151	40.4	↓ KO
Itgb2l	+4.23	0.0151	44.5	↑ KO
Dpysl4	+9.55	0.0151	245.5	↑ KO
Tbx19	+7.11	0.0165	886.2	↑ KO
Slpi	+3.02	0.0175	41.8	↑ KO
Akr7a5	-1.97	0.0179	108.6	↓ KO
Wt1	+5.35	0.0180	28.9	↑ KO
Clec4n	+3.62	0.0180	50.0	↑ KO
Tsen54	-1.56	0.0183	174.6	↓ KO
Ezr	+1.57	0.0183	171.2	↑ KO
Pdia3	-1.51	0.0191	21390.8	↓ KO
Ehd1	+1.64	0.0194	119.3	↑ KO
Sh3bgrl	+1.38	0.0202	433.1	↑ KO
P2ry14	+2.72	0.0226	71.0	↑ KO
Ccdc151	+4.69	0.0227	36.6	↑ KO
Dram1	+3.40	0.0251	33.2	↑ KO
Acp5	+4.33	0.0251	46.0	↑ KO
Kti12	-1.54	0.0252	128.5	↓ KO
Rap1gap	+4.00	0.0269	34.9	↑ KO
Ebpl	-1.31	0.0276	247.1	↓ KO
Tmem70	-1.28	0.0282	215.0	↓ KO
Foxp2	+4.96	0.0286	27.5	↑ KO
Nrg4	+7.99	0.0289	580.5	↑ KO
Serpinb6a	+1.66	0.0290	203.8	↑ KO
Rxrg	+1.12	0.0290	538.3	↑ KO
Col26a1	+4.51	0.0291	32.1	↑ KO
Hacd2	-1.03	0.0291	579.4	↓ KO
Adamts5	+9.65	0.0299	5802.4	↑ KO
Zpr1	-1.24	0.0299	248.2	↓ KO
Tfap2a	+7.33	0.0330	132.7	↑ KO
Skap1	+5.89	0.0330	20.9	↑ KO
Unc45b	+6.80	0.0335	19.5	↑ KO
Tbc1d10a	-1.35	0.0347	282.8	↓ KO
Syne1	+2.45	0.0347	48.0	↑ KO
Pla2g2d	+3.11	0.0371	45.6	↑ KO
Spg21	-1.02	0.0380	343.1	↓ KO
Ddhd1	+1.60	0.0391	190.6	↑ KO
H2-Aa	+2.88	0.0391	2862.9	↑ KO
Plbd1	+2.50	0.0414	123.2	↑ KO
Cd44	+1.25	0.0421	268.2	↑ KO
Krt20	+8.44	0.0428	340.7	↑ KO
Fcna	+3.17	0.0429	30.8	↑ KO
Ifi207	+1.44	0.0451	184.9	↑ KO
Samd9l	+1.37	0.0457	150.1	↑ KO
Tnfaip2	+1.03	0.0458	436.1	↑ KO
Cdkn2a	+3.84	0.0458	20.1	↑ KO
Zbed3	-1.69	0.0468	107.1	↓ KO
Serpinb8	+8.60	0.0485	24.1	↑ KO
Mto1	-1.22	0.0490	195.6	↓ KO
Cdh3	+5.10	0.0494	24.1	↑ KO

Esr1: A Novel ATG7-Dependent GeneNovel

Estrogen receptor alpha is the single most significant gene in the entire analysis. At 5 months, Esr1 is upregulated ~32-fold in KO microglia (log₂FC = 5.0, padj = 2.9 × 10⁻¹⁰⁷), consistently across all 11 samples (12–27% expressing in KO vs 0.5–1% in WT). At 20 months, the effect amplifies further (log₂FC = 5.85, padj = 8.9 × 10⁻²⁸⁸). This finding was not reported in the original paper and may represent a compensatory neuroprotective mechanism, as estrogen receptor signaling promotes anti-inflammatory microglia phenotypes.

p21 → p16: Progressive SenescenceNovel

At 5 months, Cdkn1a (p21) is significantly upregulated (log₂FC = 1.62, padj = 4.4 × 10⁻¹⁴), indicating early senescence entry. By 20 months, Cdkn2a (p16INK4a) becomes significant (log₂FC = 3.84, padj = 0.046) alongside persistent Cdkn1a (log₂FC = 1.42). The emergence of p16 at 20 months establishes a temporal trajectory: early p21 activation followed by irreversible p16 accumulation, consistent with progressive cellular senescence in ATG7-deficient microglia.

Pathway Enrichment Analysis

1,348

Significant Pathways

scRNA-seq FDR < 0.25

−2.64

UPR NES (KEGG)

strongest scRNA signal

+1.94

Ferroptosis NES

DAM cells (KEGG)

141

Bulk Sig Pathways

20-month FDR < 0.25

GSEA preranked analysis across 7 pseudobulk comparisons × 3 pathway databases (KEGG, Hallmark, Reactome) identified 1,348 significantly enriched gene sets (FDR < 0.25) from 7,909 tests. Mouse–human ortholog mapping achieved 92% coverage via Ensembl BioMart. The heatmap below shows normalized enrichment scores (NES) for curated top pathways across all comparisons. UPR/ER stress pathways show the strongest and most consistent downregulation in ATG7-KO (NES −2.56 to −2.64), while ribosome/translation pathways are strongly upregulated in disease contexts (NES +2.70 to +2.89).

GSEA NES Heatmap — All Databases

Curated top pathways × 7 pseudobulk comparisons. C1–C3/C6–C7 are ATG7 KO vs WT effects in different cell types; C4–C5 are disease effects. Pathways sorted by mean NES (most negative at bottom). The block of blue in ATG7-KO comparisons highlights coordinated UPR/glycan downregulation; red clusters in disease comparisons show ribosome/translation upregulation.

Pathway Results Table287 entries

Pathway	Database	Comparison	NES ↑	FDR
Protein processing in endoplasmic reticulum	KEGG	C2: HM 5xF	-2.638	<0.001
Protein processing in endoplasmic reticulum	KEGG	C6: DAM all	-2.591	<0.001
Protein processing in endoplasmic reticulum	KEGG	C1: HM all	-2.558	<0.001
Protein processing in endoplasmic reticulum	KEGG	C3: DAM 5xF	-2.360	<0.001
Antigen Presentation: Folding, assembly and pep…	Reactome	C2: HM 5xF	-2.306	<0.001
Various types of N-glycan biosynthesis	KEGG	C1: HM all	-2.216	<0.001
Post-translational protein phosphorylation	Reactome	C2: HM 5xF	-2.206	<0.001
Antigen processing and presentation	KEGG	C2: HM 5xF	-2.201	<0.001
Staphylococcus aureus infection	KEGG	C7: FTM all	-2.173	0.0010
Hematopoietic cell lineage	KEGG	C7: FTM all	-2.140	0.0010
N-Glycan biosynthesis	KEGG	C1: HM all	-2.113	<0.001
Various types of N-glycan biosynthesis	KEGG	C2: HM 5xF	-2.088	0.0019
Antigen Presentation: Folding, assembly and pep…	Reactome	C6: DAM all	-2.064	0.0045
Antigen Presentation: Folding, assembly and pep…	Reactome	C3: DAM 5xF	-2.057	0.0061
Regulation of Insulin-like Growth Factor (IGF) …	Reactome	C2: HM 5xF	-2.043	0.0079
N-Glycan biosynthesis	KEGG	C6: DAM all	-2.017	0.0070
Antigen processing and presentation	KEGG	C6: DAM all	-2.009	0.0047
Antigen processing and presentation	KEGG	C3: DAM 5xF	-1.983	0.0049
Staphylococcus aureus infection	KEGG	C2: HM 5xF	-1.949	0.0107
Various types of N-glycan biosynthesis	KEGG	C4: Dis. WT	-1.947	0.0218
Other glycan degradation	KEGG	C1: HM all	-1.945	0.0021
Various types of N-glycan biosynthesis	KEGG	C6: DAM all	-1.938	0.0066
Post-translational protein phosphorylation	Reactome	C7: FTM all	-1.937	0.0930
Coagulation	Hallmark	C7: FTM all	-1.936	0.0055
Post-translational protein phosphorylation	Reactome	C1: HM all	-1.919	0.0093
Regulation of Insulin-like Growth Factor (IGF) …	Reactome	C7: FTM all	-1.917	0.0790
N-Glycan biosynthesis	KEGG	C4: Dis. WT	-1.906	0.0201
Hematopoietic cell lineage	KEGG	C6: DAM all	-1.906	0.0070
N-Glycan biosynthesis	KEGG	C2: HM 5xF	-1.899	0.0129
Other glycan degradation	KEGG	C7: FTM all	-1.899	0.0216
Regulation of Insulin-like Growth Factor (IGF) …	Reactome	C6: DAM all	-1.869	0.0204
Post-translational protein phosphorylation	Reactome	C6: DAM all	-1.847	0.0231
mTORC1 Signaling	Hallmark	C2: HM 5xF	-1.843	0.0063
Antigen processing and presentation	KEGG	C1: HM all	-1.823	0.0114
Antigen processing and presentation	KEGG	C7: FTM all	-1.812	0.0371
Regulation of Insulin-like Growth Factor (IGF) …	Reactome	C1: HM all	-1.806	0.0435
mTORC1 Signaling	Hallmark	C6: DAM all	-1.805	0.0029
Post-translational protein phosphorylation	Reactome	C3: DAM 5xF	-1.784	0.0363
Staphylococcus aureus infection	KEGG	C3: DAM 5xF	-1.779	0.0339
Antigen Presentation: Folding, assembly and pep…	Reactome	C1: HM all	-1.777	0.0590
Other glycan degradation	KEGG	C2: HM 5xF	-1.758	0.0413
Thyroid hormone synthesis	KEGG	C7: FTM all	-1.717	0.0596
Regulation of Insulin-like Growth Factor (IGF) …	Reactome	C3: DAM 5xF	-1.707	0.0757
Thyroid hormone synthesis	KEGG	C6: DAM all	-1.705	0.0595
Staphylococcus aureus infection	KEGG	C6: DAM all	-1.705	0.0546
Antigen Presentation: Folding, assembly and pep…	Reactome	C7: FTM all	-1.686	0.1673
mTORC1 Signaling	Hallmark	C1: HM all	-1.675	0.0143
mTORC1 Signaling	Hallmark	C3: DAM 5xF	-1.672	0.0222
Thyroid hormone synthesis	KEGG	C2: HM 5xF	-1.669	0.0754
Hematopoietic cell lineage	KEGG	C3: DAM 5xF	-1.642	0.1002
Staphylococcus aureus infection	KEGG	C1: HM all	-1.641	0.0748
N-Glycan biosynthesis	KEGG	C3: DAM 5xF	-1.641	0.0927
Angiogenesis	Hallmark	C6: DAM all	-1.607	0.0184
Other glycan degradation	KEGG	C6: DAM all	-1.599	0.0855
Various types of N-glycan biosynthesis	KEGG	C7: FTM all	-1.579	0.1414
Various types of N-glycan biosynthesis	KEGG	C3: DAM 5xF	-1.578	0.1311
Epithelial Mesenchymal Transition	Hallmark	C7: FTM all	-1.513	0.2073
Hematopoietic cell lineage	KEGG	C1: HM all	-1.510	0.1565
Thyroid hormone synthesis	KEGG	C3: DAM 5xF	-1.507	0.1697
Coagulation	Hallmark	C1: HM all	-1.505	0.0400
N-Glycan biosynthesis	KEGG	C7: FTM all	-1.497	0.1973
Protein processing in endoplasmic reticulum	KEGG	C7: FTM all	-1.492	0.1970
Thyroid hormone synthesis	KEGG	C1: HM all	-1.461	0.2225
Other glycan degradation	KEGG	C3: DAM 5xF	-1.447	0.1973
Arachidonic acid metabolism	KEGG	C7: FTM all	-1.441	0.2376
E2F Targets	Hallmark	C6: DAM all	-1.426	0.0667
G2M Checkpoint	Hallmark	C3: DAM 5xF	-1.406	0.0885
Angiogenesis	Hallmark	C3: DAM 5xF	-1.381	0.0990
Epithelial Mesenchymal Transition	Hallmark	C6: DAM all	-1.378	0.0891
G2M Checkpoint	Hallmark	C6: DAM all	-1.373	0.0848
Epithelial Mesenchymal Transition	Hallmark	C3: DAM 5xF	-1.358	0.1116
Coagulation	Hallmark	C3: DAM 5xF	-1.339	0.1198
Angiogenesis	Hallmark	C7: FTM all	-1.282	0.5024
Hematopoietic cell lineage	KEGG	C2: HM 5xF	-1.246	0.3710
G2M Checkpoint	Hallmark	C5: Dis. KO	-1.226	0.1648
Interferon Alpha Response	Hallmark	C6: DAM all	-1.217	0.1728
Coagulation	Hallmark	C6: DAM all	-1.201	0.1821
Coagulation	Hallmark	C2: HM 5xF	-1.181	0.3155
p53 Pathway	Hallmark	C7: FTM all	-1.140	0.5863
N-Glycan biosynthesis	KEGG	C5: Dis. KO	-1.084	0.4562
Apoptosis	Hallmark	C7: FTM all	-1.077	0.6007
Arachidonic acid metabolism	Reactome	C7: FTM all	-1.064	0.8825
E2F Targets	Hallmark	C3: DAM 5xF	-0.994	0.5669
mTORC1 Signaling	Hallmark	C7: FTM all	-0.945	0.7172
Interferon Alpha Response	Hallmark	C3: DAM 5xF	-0.926	0.6834
Other glycan degradation	KEGG	C4: Dis. WT	-0.858	0.8669
Arachidonic acid metabolism	Reactome	C1: HM all	-0.855	1.0000
Various types of N-glycan biosynthesis	KEGG	C5: Dis. KO	-0.819	0.8535
Antigen Presentation: Folding, assembly and pep…	Reactome	C5: Dis. KO	-0.670	0.9660
Protein processing in endoplasmic reticulum	KEGG	C4: Dis. WT	+0.626	1.0000
G2M Checkpoint	Hallmark	C2: HM 5xF	+0.745	0.9562
Endocrine and other factor-regulated calcium re…	KEGG	C4: Dis. WT	+0.804	0.9985
Regulation of RUNX2 expression and activity	Reactome	C5: Dis. KO	+0.822	0.8967
Oxidative Phosphorylation	Hallmark	C7: FTM all	+0.828	0.8314
Thyroid hormone synthesis	KEGG	C4: Dis. WT	+0.829	0.9688
Protein processing in endoplasmic reticulum	KEGG	C5: Dis. KO	+0.859	0.8194
Arachidonic acid metabolism	KEGG	C4: Dis. WT	+0.867	0.9291
Regulation of expression of SLITs and ROBOs	Reactome	C2: HM 5xF	+0.958	0.9921
Ribosome	KEGG	C2: HM 5xF	+1.005	0.5675
Oxidative Phosphorylation	Hallmark	C1: HM all	+1.024	0.4289
Epithelial Mesenchymal Transition	Hallmark	C1: HM all	+1.025	0.4424
Interferon Alpha Response	Hallmark	C2: HM 5xF	+1.065	0.4468
Nonsense Mediated Decay (NMD) enhanced by the E…	Reactome	C2: HM 5xF	+1.065	0.8203
Arachidonic acid metabolism	Reactome	C4: Dis. WT	+1.095	0.6400
Staphylococcus aureus infection	KEGG	C5: Dis. KO	+1.120	0.4840
Selenocysteine synthesis	Reactome	C2: HM 5xF	+1.147	0.6890
Arachidonic acid metabolism	KEGG	C1: HM all	+1.165	0.5582
Signaling by ERBB4	Reactome	C5: Dis. KO	+1.170	0.5914
SARS-CoV-1 modulates host translation machinery	Reactome	C2: HM 5xF	+1.242	0.5851
Oxidative Phosphorylation	Hallmark	C4: Dis. WT	+1.246	0.2367
Selenoamino acid metabolism	Reactome	C2: HM 5xF	+1.252	0.5751
Arachidonic acid metabolism	Reactome	C2: HM 5xF	+1.268	0.5672
G2M Checkpoint	Hallmark	C4: Dis. WT	+1.269	0.2109
Antigen Presentation: Folding, assembly and pep…	Reactome	C4: Dis. WT	+1.274	0.4665
G2M Checkpoint	Hallmark	C1: HM all	+1.275	0.2306
Nonsense Mediated Decay (NMD) independent of th…	Reactome	C2: HM 5xF	+1.290	0.5440
E2F Targets	Hallmark	C4: Dis. WT	+1.317	0.1759
E2F Targets	Hallmark	C5: Dis. KO	+1.320	0.0863
p53 Pathway	Hallmark	C4: Dis. WT	+1.328	0.1696
Angiogenesis	Hallmark	C1: HM all	+1.334	0.2019
Arachidonic acid metabolism	Reactome	C5: Dis. KO	+1.336	0.3534
E2F Targets	Hallmark	C1: HM all	+1.338	0.2133
Coagulation	Hallmark	C4: Dis. WT	+1.338	0.1662
Epithelial Mesenchymal Transition	Hallmark	C4: Dis. WT	+1.364	0.1420
Thyroid hormone synthesis	KEGG	C5: Dis. KO	+1.367	0.2024
Activation of the mRNA upon binding of the cap-…	Reactome	C2: HM 5xF	+1.371	0.5036
Interferon Gamma Response	Hallmark	C3: DAM 5xF	+1.381	0.1855
Formation of a pool of free 40S subunits	Reactome	C2: HM 5xF	+1.382	0.4967
Interferon Gamma Response	Hallmark	C6: DAM all	+1.382	0.1704
E2F Targets	Hallmark	C2: HM 5xF	+1.383	0.1283
Post-translational protein phosphorylation	Reactome	C5: Dis. KO	+1.390	0.2801
GTP hydrolysis and joining of the 60S ribosomal…	Reactome	C2: HM 5xF	+1.402	0.5003
Endocrine and other factor-regulated calcium re…	KEGG	C5: Dis. KO	+1.402	0.1724
Xenobiotic Metabolism	Hallmark	C7: FTM all	+1.422	0.0835
L13a-mediated translational silencing of Cerulo…	Reactome	C2: HM 5xF	+1.430	0.4742
Other glycan degradation	KEGG	C5: Dis. KO	+1.440	0.1447
Interferon Alpha Response	Hallmark	C7: FTM all	+1.441	0.0902
Apoptosis	Hallmark	C1: HM all	+1.459	0.1639
Reactive Oxygen Species Pathway	Hallmark	C2: HM 5xF	+1.476	0.0992
Apoptosis	Hallmark	C3: DAM 5xF	+1.480	0.1346
mTORC1 Signaling	Hallmark	C4: Dis. WT	+1.488	0.0632
Interferon Alpha Response	Hallmark	C1: HM all	+1.489	0.2014
Antigen processing and presentation	KEGG	C5: Dis. KO	+1.489	0.1064
Apoptosis	Hallmark	C4: Dis. WT	+1.503	0.0584
Oxidative Phosphorylation	Hallmark	C3: DAM 5xF	+1.506	0.1819
E2F Targets	Hallmark	C7: FTM all	+1.511	0.0728
p53 Pathway	Hallmark	C1: HM all	+1.517	0.2435
p53 Pathway	Hallmark	C2: HM 5xF	+1.520	0.0950
Xenobiotic Metabolism	Hallmark	C4: Dis. WT	+1.520	0.0547
p53 Pathway	Hallmark	C5: Dis. KO	+1.520	0.0167
mTORC1 Signaling	Hallmark	C5: Dis. KO	+1.522	0.0171
Apoptosis	Hallmark	C2: HM 5xF	+1.528	0.1133
Arachidonic acid metabolism	KEGG	C5: Dis. KO	+1.535	0.0833
Oxidative Phosphorylation	Hallmark	C2: HM 5xF	+1.552	0.1315
p53 Pathway	Hallmark	C3: DAM 5xF	+1.552	0.1822
Apoptosis	Hallmark	C6: DAM all	+1.553	0.0836
p53 Pathway	Hallmark	C6: DAM all	+1.557	0.0930
Ribosome	KEGG	C7: FTM all	+1.557	0.2795
Interferon Gamma Response	Hallmark	C7: FTM all	+1.562	0.0700
Angiogenesis	Hallmark	C5: Dis. KO	+1.564	0.0117
Nonsense Mediated Decay (NMD) enhanced by the E…	Reactome	C7: FTM all	+1.566	0.2439
Pentose phosphate pathway	KEGG	C4: Dis. WT	+1.572	0.1157
Interferon Gamma Response	Hallmark	C2: HM 5xF	+1.583	0.1357
Activation of the mRNA upon binding of the cap-…	Reactome	C7: FTM all	+1.605	0.2423
Formation of a pool of free 40S subunits	Reactome	C7: FTM all	+1.614	0.2358
Post-translational protein phosphorylation	Reactome	C4: Dis. WT	+1.620	0.1602
Hematopoietic cell lineage	KEGG	C5: Dis. KO	+1.621	0.0480
Arachidonic acid metabolism	KEGG	C2: HM 5xF	+1.627	0.1561
Regulation of RUNX2 expression and activity	Reactome	C4: Dis. WT	+1.633	0.1523
G2M Checkpoint	Hallmark	C7: FTM all	+1.633	0.0685
Selenoamino acid metabolism	Reactome	C7: FTM all	+1.636	0.2463
Regulation of Insulin-like Growth Factor (IGF) …	Reactome	C5: Dis. KO	+1.640	0.0789
Selenocysteine synthesis	Reactome	C7: FTM all	+1.652	0.2463
Activation of the mRNA upon binding of the cap-…	Reactome	C3: DAM 5xF	+1.658	0.3164
Regulation of Insulin-like Growth Factor (IGF) …	Reactome	C4: Dis. WT	+1.664	0.1386
Signaling by ERBB4	Reactome	C4: Dis. WT	+1.672	0.1349
Regulation of expression of SLITs and ROBOs	Reactome	C7: FTM all	+1.673	0.2387
Apoptosis	Hallmark	C5: Dis. KO	+1.676	0.0040
GTP hydrolysis and joining of the 60S ribosomal…	Reactome	C7: FTM all	+1.686	0.2373
L13a-mediated translational silencing of Cerulo…	Reactome	C7: FTM all	+1.686	0.2289
Epithelial Mesenchymal Transition	Hallmark	C2: HM 5xF	+1.692	0.1600
Reactive Oxygen Species Pathway	Hallmark	C7: FTM all	+1.692	0.0818
Pentose phosphate pathway	KEGG	C2: HM 5xF	+1.694	0.1315
Nonsense Mediated Decay (NMD) enhanced by the E…	Reactome	C3: DAM 5xF	+1.699	0.3100
Pentose phosphate pathway	KEGG	C1: HM all	+1.702	0.3203
Coagulation	Hallmark	C5: Dis. KO	+1.714	0.0027
Nonsense Mediated Decay (NMD) independent of th…	Reactome	C7: FTM all	+1.726	0.2407
Reactive Oxygen Species Pathway	Hallmark	C5: Dis. KO	+1.728	0.0025
Signaling by ERBB4	Reactome	C6: DAM all	+1.734	0.1718
Signaling by ERBB4	Reactome	C7: FTM all	+1.734	0.2408
SARS-CoV-1 modulates host translation machinery	Reactome	C7: FTM all	+1.742	0.2447
Angiogenesis	Hallmark	C2: HM 5xF	+1.746	0.2012
Arachidonic acid metabolism	Reactome	C6: DAM all	+1.752	0.1608
Pentose phosphate pathway	KEGG	C5: Dis. KO	+1.756	0.0181
Angiogenesis	Hallmark	C4: Dis. WT	+1.757	0.0094
Pentose phosphate pathway	KEGG	C7: FTM all	+1.762	0.1987
Pentose phosphate pathway	KEGG	C6: DAM all	+1.772	0.1054
Regulation of expression of SLITs and ROBOs	Reactome	C3: DAM 5xF	+1.795	0.2748
Endocrine and other factor-regulated calcium re…	KEGG	C6: DAM all	+1.817	0.0841
GTP hydrolysis and joining of the 60S ribosomal…	Reactome	C3: DAM 5xF	+1.842	0.2550
Xenobiotic Metabolism	Hallmark	C5: Dis. KO	+1.850	<0.001
Hematopoietic cell lineage	KEGG	C4: Dis. WT	+1.853	0.0376
Activation of the mRNA upon binding of the cap-…	Reactome	C1: HM all	+1.864	0.1573
Endocrine and other factor-regulated calcium re…	KEGG	C7: FTM all	+1.867	0.2211
Reactive Oxygen Species Pathway	Hallmark	C4: Dis. WT	+1.871	0.0028
Arachidonic acid metabolism	KEGG	C6: DAM all	+1.872	0.0930
Pentose phosphate pathway	KEGG	C3: DAM 5xF	+1.898	0.1528
Arachidonic acid metabolism	Reactome	C3: DAM 5xF	+1.899	0.2272
SARS-CoV-1 modulates host translation machinery	Reactome	C1: HM all	+1.901	0.1494
Regulation of expression of SLITs and ROBOs	Reactome	C1: HM all	+1.908	0.1495
Staphylococcus aureus infection	KEGG	C4: Dis. WT	+1.919	0.0230
L13a-mediated translational silencing of Cerulo…	Reactome	C3: DAM 5xF	+1.927	0.2077
Signaling by ERBB4	Reactome	C3: DAM 5xF	+1.945	0.2052
Selenoamino acid metabolism	Reactome	C1: HM all	+1.949	0.1493
Regulation of RUNX2 expression and activity	Reactome	C7: FTM all	+1.972	0.2766
Nonsense Mediated Decay (NMD) independent of th…	Reactome	C3: DAM 5xF	+1.973	0.2229
Formation of a pool of free 40S subunits	Reactome	C3: DAM 5xF	+1.975	0.2332
Selenocysteine synthesis	Reactome	C3: DAM 5xF	+1.977	0.2429
Nonsense Mediated Decay (NMD) enhanced by the E…	Reactome	C1: HM all	+1.980	0.1411
SARS-CoV-1 modulates host translation machinery	Reactome	C3: DAM 5xF	+1.980	0.2532
Interferon Gamma Response	Hallmark	C5: Dis. KO	+1.981	<0.001
Ribosome	KEGG	C3: DAM 5xF	+2.016	0.1175
Xenobiotic Metabolism	Hallmark	C6: DAM all	+2.021	0.0044
Ribosome	KEGG	C1: HM all	+2.023	0.2271
Reactive Oxygen Species Pathway	Hallmark	C1: HM all	+2.026	0.1297
Interferon Gamma Response	Hallmark	C1: HM all	+2.045	0.1731
Reactive Oxygen Species Pathway	Hallmark	C3: DAM 5xF	+2.057	0.0172
Oxidative Phosphorylation	Hallmark	C6: DAM all	+2.059	0.0033
Regulation of RUNX2 expression and activity	Reactome	C3: DAM 5xF	+2.070	0.2439
Regulation of RUNX2 expression and activity	Reactome	C6: DAM all	+2.071	0.0135
Reactive Oxygen Species Pathway	Hallmark	C6: DAM all	+2.073	0.0066
Selenoamino acid metabolism	Reactome	C3: DAM 5xF	+2.075	0.2936
Epithelial Mesenchymal Transition	Hallmark	C5: Dis. KO	+2.093	<0.001
Selenocysteine synthesis	Reactome	C1: HM all	+2.120	0.1413
Xenobiotic Metabolism	Hallmark	C3: DAM 5xF	+2.148	0.0192
Arachidonic acid metabolism	KEGG	C3: DAM 5xF	+2.154	0.0725
GTP hydrolysis and joining of the 60S ribosomal…	Reactome	C1: HM all	+2.155	0.1446
Antigen processing and presentation	KEGG	C4: Dis. WT	+2.158	<0.001
L13a-mediated translational silencing of Cerulo…	Reactome	C1: HM all	+2.164	0.1499
Xenobiotic Metabolism	Hallmark	C1: HM all	+2.169	0.1565
Nonsense Mediated Decay (NMD) independent of th…	Reactome	C1: HM all	+2.173	0.1638
Endocrine and other factor-regulated calcium re…	KEGG	C3: DAM 5xF	+2.177	0.1680
Oxidative Phosphorylation	Hallmark	C5: Dis. KO	+2.199	<0.001
Signaling by ERBB4	Reactome	C1: HM all	+2.203	0.1844
Formation of a pool of free 40S subunits	Reactome	C1: HM all	+2.205	0.1973
Interferon Alpha Response	Hallmark	C5: Dis. KO	+2.205	<0.001
Activation of the mRNA upon binding of the cap-…	Reactome	C6: DAM all	+2.250	<0.001
SARS-CoV-1 modulates host translation machinery	Reactome	C4: Dis. WT	+2.292	<0.001
Interferon Gamma Response	Hallmark	C4: Dis. WT	+2.331	<0.001
Interferon Alpha Response	Hallmark	C4: Dis. WT	+2.338	<0.001
SARS-CoV-1 modulates host translation machinery	Reactome	C6: DAM all	+2.346	<0.001
Activation of the mRNA upon binding of the cap-…	Reactome	C4: Dis. WT	+2.372	<0.001
Xenobiotic Metabolism	Hallmark	C2: HM 5xF	+2.373	0.0653
Signaling by ERBB4	Reactome	C2: HM 5xF	+2.443	0.0807
Regulation of RUNX2 expression and activity	Reactome	C1: HM all	+2.487	0.0665
Activation of the mRNA upon binding of the cap-…	Reactome	C5: Dis. KO	+2.508	<0.001
Endocrine and other factor-regulated calcium re…	KEGG	C1: HM all	+2.531	0.0075
Endocrine and other factor-regulated calcium re…	KEGG	C2: HM 5xF	+2.543	0.0209
Nonsense Mediated Decay (NMD) enhanced by the E…	Reactome	C6: DAM all	+2.546	<0.001
Regulation of expression of SLITs and ROBOs	Reactome	C6: DAM all	+2.552	<0.001
SARS-CoV-1 modulates host translation machinery	Reactome	C5: Dis. KO	+2.558	<0.001
GTP hydrolysis and joining of the 60S ribosomal…	Reactome	C6: DAM all	+2.581	<0.001
L13a-mediated translational silencing of Cerulo…	Reactome	C6: DAM all	+2.598	<0.001
Nonsense Mediated Decay (NMD) independent of th…	Reactome	C6: DAM all	+2.639	<0.001
Ribosome	KEGG	C6: DAM all	+2.640	<0.001
Selenoamino acid metabolism	Reactome	C4: Dis. WT	+2.651	<0.001
Selenoamino acid metabolism	Reactome	C6: DAM all	+2.653	<0.001
Regulation of RUNX2 expression and activity	Reactome	C2: HM 5xF	+2.654	<0.001
Formation of a pool of free 40S subunits	Reactome	C4: Dis. WT	+2.666	<0.001
L13a-mediated translational silencing of Cerulo…	Reactome	C4: Dis. WT	+2.668	<0.001
Selenocysteine synthesis	Reactome	C4: Dis. WT	+2.673	<0.001
GTP hydrolysis and joining of the 60S ribosomal…	Reactome	C4: Dis. WT	+2.675	<0.001
Regulation of expression of SLITs and ROBOs	Reactome	C4: Dis. WT	+2.677	<0.001
Selenocysteine synthesis	Reactome	C6: DAM all	+2.682	<0.001
Formation of a pool of free 40S subunits	Reactome	C6: DAM all	+2.694	<0.001
Ribosome	KEGG	C4: Dis. WT	+2.699	<0.001
Nonsense Mediated Decay (NMD) independent of th…	Reactome	C4: Dis. WT	+2.709	<0.001
Nonsense Mediated Decay (NMD) enhanced by the E…	Reactome	C4: Dis. WT	+2.714	<0.001
Regulation of expression of SLITs and ROBOs	Reactome	C5: Dis. KO	+2.783	<0.001
Nonsense Mediated Decay (NMD) enhanced by the E…	Reactome	C5: Dis. KO	+2.842	<0.001
GTP hydrolysis and joining of the 60S ribosomal…	Reactome	C5: Dis. KO	+2.868	<0.001
Ribosome	KEGG	C5: Dis. KO	+2.889	<0.001
Selenoamino acid metabolism	Reactome	C5: Dis. KO	+2.889	<0.001
L13a-mediated translational silencing of Cerulo…	Reactome	C5: Dis. KO	+2.890	<0.001
Formation of a pool of free 40S subunits	Reactome	C5: Dis. KO	+2.911	<0.001
Selenocysteine synthesis	Reactome	C5: Dis. KO	+2.928	<0.001
Nonsense Mediated Decay (NMD) independent of th…	Reactome	C5: Dis. KO	+2.928	<0.001

Bulk RNA-seq GSEA — 20-Month Microglia

GSEA of 20-month FACS-sorted microglia (ATG7-KO vs WT, 5 samples) identified 141 significantly enriched pathways (FDR < 0.25) across KEGG (30), Hallmark (17), and Reactome (94). The top 20 pathways by absolute NES reveal the same UPR/ER stress downregulation (blue) and ribosome/translation upregulation (red) observed in 5-month scRNA-seq, with the IRE1α–XBP1 arm now specifically implicated. Prefix: [K] KEGG, [H] Hallmark, [R] Reactome.

Top 20 pathways from bulk RNA-seq GSEA ranked by absolute NES. UPR-related pathways (IRE1α, XBP1, ATF6, N-glycan) are consistently downregulated while ribosome/translation pathways dominate the upregulated set — an age-dependent compensatory translation response absent from 5-month scRNA-seq.

UPR/ER Stress — Strongest Signal

The UPR is the most consistently downregulated pathway across both platforms and ages: scRNA NES −2.64 (KEGG, 5 mo) and bulk NES −2.44 (KEGG, 20 mo). In aged microglia, the IRE1α–XBP1 arm is specifically implicated (NES −2.54), with ATF6 branches also significantly depleted. Co-downregulation of N-glycan biosynthesis and antigen processing reflects broader ER functional impairment beyond the UPR itself.

Ferroptosis in DAM Novel

Ferroptosis is enriched in ATG7-KO DAM cells (scRNA NES +1.94, KEGG) with co-enrichment of glutathione metabolism and iron transport. Bulk 20-month data shows ferroptosis trending positive (NES +1.45) alongside strong mineral absorption (NES +2.04), suggesting age-dependent ferroptotic vulnerability. This connects autophagy deficiency → impaired iron recycling → ferroptotic cell death in disease-associated microglia.

Cholesterol Homeostasis Novel

Cholesterol homeostasis is specifically downregulated in ATG7-KO DAM (Hallmark NES −2.11, FDR ≈ 0) but not in homeostatic microglia. This likely reflects disruption of the Trem2–APOE lipid metabolism axis central to DAM function. Combined with the finding that Apoe→Lrp1 signaling is lost in KO, the lipid supply chain from homeostatic to disease-associated microglia appears severed by autophagy deficiency.

ATG7 × Disease Interaction

Significant Genes

FDR < 0.1 in any cell type

14,169

Genes Tested (HM)

11,748 DAM · 11,181 FTM

0.126

Min Adjusted p

Htra3 in HM (log2FC +1.63)

290

Pathway Interactions

FDR < 0.25 (of 4,127 tested)

Important Null Result

ATG7 and 5xFAD Effects Are Statistically Additive — Zero Synergistic Interactions Detected

Formal interaction analysis (~ atg7_status × disease_model) across 3 microglia subtypes found zero genes with significant ATG7×5xFAD interaction at FDR < 0.1. The 2.4× amplification of disease DEGs in KO background (146 vs 60 from DE analysis) arises from additive combination of independent effects, not from a qualitatively new “second hit” program. The closest candidate, Htra3 (padj = 0.126), falls short of significance with only 11 samples and 7 residual degrees of freedom.

Gene-Level Interaction Results by Cell Type

Cell Type	Genes Tested	Nominal p<0.01	Nominal p<0.05	Min padj	ATG7 Main (FDR<0.05)	Disease Main (FDR<0.05)
HM	14,169	105	608	0.126	122	179
DAM	11,748	55	390	>0.99	170	204
FTM	11,181	30	223	>0.99	5	2

Despite strong main effects for both ATG7 (122 HM genes, 170 DAM genes at FDR < 0.05) and disease (179 HM, 204 DAM), the interaction term yields zero significant genes. The HM nominal hits (105 at p < 0.01) roughly match the expected false-positive count (~142 out of 14,169), consistent with the null hypothesis.

Top Pathway-Level Interactions (GSEA on Interaction-Ranked Genes)

DAM

FTM

GSEA on interaction-ranked genes reveals 290 significant pathways (FDR < 0.25) across 4,127 tested. Translation and ribosome biogenesis pathways show the strongest sub-additive (negative NES) interactions in both HM and DAM, indicating that the combined ATG7-KO + 5xFAD condition saturates the translational compensation response.

Additive Biology ModelNovel

The original paper framed ATG7-KO + 5xFAD as a synergistic “second hit” model. Our formal interaction analysis reveals the enhanced disease phenotype arises from additive contributions of two independent effects, not qualitatively new transcriptional programs. The 2.4× DEG amplification in KO background (146 vs 60 genes) reflects combined magnitude, not synergy. With only 11 samples (7 residual df), a ≥20-sample study would be needed to detect subtle interactions (power ∼3× higher).

Translational SaturationNovel

While gene-level effects are additive, GSEA reveals a consistent sub-additive interaction for translation/ribosome pathways (NES = −2.69 to −3.13 in HM and DAM). Both ATG7-KO and 5xFAD independently upregulate translation, but the combined condition cannot double the response — cells hit a translational capacity ceiling. This saturated compensation may indicate worse outcomes: autophagy-deficient microglia in AD cannot mount the full compensatory translation response that either insult alone would trigger.

Gene Signature Scoring

−0.73

UPR Effect in DAM

Rank-biserial correlation (ATG7 KO vs WT)

0.586

Ferroptosis Score

KO-5xFAD DAM mean (vs 0.540 WT)

−0.130

Ferroptosis–UPR r

Spearman anti-correlation in DAM

52 / 72

Significant Tests

Mann-Whitney U (FDR < 0.05)

Single-cell gene signature scoring for Ferroptosis (18 genes), Senescence (16 genes), and UPR (18 genes) across 90,475 cells and 6 microglia subtypes. UPR impairment is the dominant ATG7-dependent signature across all microglia subtypes (rank-biserial = −0.58 to −0.73), providing single-cell validation of the pseudobulk DE and GSEA findings. Ferroptosis vulnerability is selectively elevated in KO DAM and critically is disease-conditional: 5xFAD increases DAM ferroptosis only in the ATG7-KO background (padj = 0.041) but not in WT (p = 0.19).

ATG7 Effect on Signature Scores (KO − WT)

Each cell shows the difference in mean score between KO and WT for a given signature, cell type, and disease context. UPR shows massive, uniform reduction across all cell types (deep blue), while Ferroptosis is modestly elevated (light red), particularly in DAM. Senescence shows minimal ATG7 effect.

Ferroptosis vs UPR Score Anti-Correlation

● HM

◆ DAM

▲ FTM

■ IFN

◉ Prolif

Each point represents one cell type × condition combination. KO conditions (amber, red) cluster in the lower-right: higher ferroptosis, lower UPR. WT conditions (green, blue) cluster upper-left. The anti-correlation (DAM r = −0.130, p < 10⁻³⁰) is driven by KO conditions, revealing a mechanistically linked “double vulnerability” at the single-cell level.

Double Vulnerability: Ferroptosis–UPR Anti-CorrelationNovel

The negative correlation between Ferroptosis and UPR scores (DAM r = −0.130, HM r = −0.073, p < 10⁻³⁰) reveals that these are not independent consequences of ATG7 loss but are mechanistically linked at the single-cell level. Cells with the lowest UPR activity tend to have the highest ferroptosis susceptibility. The proposed mechanism: impaired UPR → reduced ER proteostasis → disrupted iron handling → ferroptotic vulnerability. This “double vulnerability” integrates the two main ATG7-dependent pathological axes and suggests that pharmacological UPR induction (e.g., TUDCA) could rescue both defects simultaneously.

Disease-Conditional Ferroptosis in DAMNovel

5xFAD increases DAM ferroptosis only in the ATG7-KO background (padj = 0.041) and not in WT (p = 0.19). Autophagy normally protects activated microglia from ferroptosis during disease through NCOA4-mediated ferritinophagy and glutathione recycling. When this protection is removed by ATG7 loss, amyloid-associated stress pushes DAM cells toward ferroptotic vulnerability. This has therapeutic implications: ferroptosis inhibitors (ferrostatin-1, liproxstatin-1) may specifically benefit autophagy-impaired microglia in AD.

Ferroptosis and Senescence Are Independent Programs

The near-zero correlation between Ferroptosis and Senescence scores (HM r = −0.008, DAM r = 0.014, both non-significant) demonstrates these are distinct pathological programs affecting different cell populations. Ferroptosis concentrates in DAM (highest scores), while senescence shows low-level, more uniform activation driven primarily by 5xFAD (HM padj = 2.9 × 10⁻¹⁶ for disease effect) rather than ATG7 status. This argues against a simple “cellular stress cascade” model and instead supports parallel, independent pathological pathways with distinct therapeutic targets.

Regulatory Networks

2,844

Significant TF–Cell Pairs

FDR < 0.05 out of 7,744 tests (36.7%)

−1.27

ATF6 Δ Activity in DAM

Strongest UPR TF reduction (padj ≈ 0)

DAM Pathways Down

Growth/survival signaling desert in KO DAM

Significant Pathway Tests

FDR < 0.05 out of 154 tests (36.4%)

Transcription Factor Activity Inference

ULM-based TF activity inference using mouse CollecTRI regulons (43,226 interactions, 1,165 TFs, 6,582 targets) across 90,475 cells reveals ATG7 loss causes a coordinated collapse of the UPR transcriptional program, with ATF6/ATF6B as the most severely affected TFs (diff = −1.27 in DAM, padj ≈ 0) — more affected than XBP1 or ATF4, pinpointing the ATF6 arm as the primary UPR branch compromised. Spi1/PU.1 (myeloid identity) is downregulated in 8 of 11 microglia subtypes, while MHC-II regulators (RFXAP diff = −0.68, RFXANK −0.60, RFX5 −0.55 in DAM) are strongly reduced. NRF2 shows compensatory upregulation in homeostatic microglia (HM_2 diff = +0.33) but fails in activated states. Strikingly, Esr1 TF activity is paradoxically reduced despite massive gene upregulation (log2FC = 5.0–5.85), revealing non-functional overexpression.

TF Activity Differences (KO − WT)

Top 28 TFs ranked by effect size in DAM. UPR TFs (Atf6, Atf6b, Xbp1) show the strongest and most consistent downregulation across all microglia subtypes. White dots (•) indicate significance at FDR < 0.05. Cell types ordered from homeostatic (HM_1–HM_6) through transitional (TM) to activated (DAM, FTM, IFN) and proliferating (Prolif). HM_1 shows the broadest upregulation of compensatory TFs (Lmo4, Mkx, Tgfb1i1), consistent with its Milo depletion pattern.

ATF6 Arm: Primary UPR Branch CompromisedNovel

ATF6 and ATF6B are the most severely affected UPR transcription factors in ATG7-deficient microglia (ATF6 diff = −1.27 in DAM, significant in all 10 microglia types tested), exceeding XBP1 (diff = −0.67) and ATF4 effects. Most autophagy–UPR crosstalk studies focus on the IRE1/XBP1 or PERK/ATF4 arms; the ATF6 arm being the primary branch compromised by autophagy loss in microglia is previously unreported. ATF6 controls ER protein folding capacity and ERAD — its collapse directly explains the downstream Calr and Hspa5 downregulation observed in differential expression analysis.

Esr1 Paradox: Non-Functional OverexpressionNovel

Despite massive Esr1 gene upregulation (log2FC = 5.0–5.85 across ages and platforms, 12–27% expressing in KO vs 0.5–1% in WT), Esr1 transcription factor activity is paradoxically reduced (diff = −0.15 to −0.37, significant in 3 cell types). The cells produce excess Esr1 mRNA as a failed compensatory attempt — the protein may lack necessary cofactors, ligand availability, or proper post-translational processing for functional estrogen receptor signaling. This “non-functional overexpression” pattern emerges as a recurring theme in ATG7-deficient microglia (also observed for JAK-STAT pathway components).

PROGENy Pathway Activity Inference

PROGENy uses experimentally-derived perturbation footprints (14 pathways, 6,398 mouse gene interactions, top 500 genes/pathway) rather than gene set membership, providing a functional view of signaling activity across 90,475 cells. DAM cells in ATG7-KO show a coordinated shutdown of 9 growth/survival pathways (PI3K diff = −0.35, JAK-STAT −0.29, TGFb −0.20, Hypoxia −0.20, VEGF −0.15, EGFR −0.15, WNT −0.15, NFκB −0.11, Androgen −0.08) with only p53 (+0.30) and TNFa (+0.10) upregulated — a “signaling desert” consistent with cells transitioning toward ferroptotic death rather than productive activation. Critically, JAK-STAT activity is DOWN in DAM (−0.29) despite GSEA showing JAK-STAT gene enrichment, revealing a gene-expression vs functional-signaling disconnect that parallels the Esr1 paradox.

PROGENy Pathway Activity Differences (KO − WT)

14 PROGENy pathways ranked by effect size in DAM. The DAM column shows a striking signaling desert pattern with 9 pathways significantly reduced (blue) and only p53 upregulated (red). HM_1 and HM_2 show distinct profiles explaining the Milo bifurcation: HM_1 is MAPK/PI3K-hyperactive while HM_2 is p53/NFκB-driven. HM_6 shows massive JAK-STAT activation (+1.14). White dots (•) indicate FDR < 0.05.

DAM Signaling DesertNovel

ATG7-deficient DAM cells show coordinated collapse of 9 growth/survival signaling pathways simultaneously, creating a catastrophic failure of the DAM signaling network rather than selective pathway impairment. PI3K (downstream of TREM2, diff = −0.35, padj ≈ 0) is the most severely affected, consistent with disrupted TREM2/APOE signaling and cholesterol homeostasis loss. Only p53 (+0.30) compensates, driving senescence rather than productive activation. This profile — simultaneous survival pathway loss with p53 gain — is consistent with cells transitioning toward ferroptotic death (confirmed in Gene Signatures analysis).

HM Bifurcation: Hyperactive vs SenescentNovel

The HM_1/HM_2 bifurcation identified by Milo now has distinct signaling profiles: HM_1 is hyperactive (MAPK +0.41, PI3K +0.17, JAK-STAT +0.16) while HM_2 is senescent (p53 +0.41, NFκB +0.21, JAK-STAT −0.29). HM_1 may be depleted through activation-induced burnout — excessive MAPK/PI3K signaling driving cells out of homeostasis — while HM_2 accumulates through p53-driven growth arrest. This provides a mechanistic explanation for why ATG7 loss produces opposite effects on two homeostatic microglia subtypes.

Cell-Cell Communication

2,641

Significant Interactions

In at least one condition (rank < 0.1)

243

Gained in KO

New or strengthened interactions

278

Lost in KO

Weakened or abolished interactions

2:1

DAM Loss Ratio

42 lost vs 21 gained — communication collapse

LIANA cell-cell communication analysis reveals that ATG7 loss fundamentally rewires intercellular communication in the brain immune microenvironment. Among 2,641 significant interactions, 278 are lost and 243 gained in KO. DAM cells show the most dramatic communication collapse, losing twice as many interactions as they gain (42 vs 21) — consistent with the "signaling desert" identified by PROGENy pathway analysis. Key lost signals include Adam10→Trem2 (critical for TREM2 cleavage) and Apoe→Lrp1 (lipid metabolism), while compensatory BAFF signaling (Tnfsf13b→Tnfrsf17) emerges as a novel microglia-to-microglia survival axis.

Net Interaction Changes by Cell Type Pair

Net interaction changes (gained − lost) per cell type pair in ATG7-KO vs WT. Green indicates net gain; red indicates net loss. TM as target shows the most consistent loss pattern, while IFN-originating signals tend to be gained.

Top Differential Ligand-Receptor Interactions

Top 30 gained (green, negative ΔRank) and top 30 lost (red, positive ΔRank) interactions in ATG7-KO vs WT. TM-originating signals dominate both gained interactions (via Adam10, Pdgfb, Vim) and lost interactions (via App→Cd74, S100a9→Itgb2).

Differential Interactions Table

Ligand	Receptor	Source	Target	ΔRank	\|ΔRank\| ▼	Direction
Vim	Cd44	TM	B_cell	-0.947	0.947	Gained
Gstp1	Traf2	Prolif	TM	-0.941	0.941	Gained
Pdgfb	S1pr1	TM	DAM	-0.923	0.923	Gained
Tnfsf13b	Tnfrsf13b	HM_1	TM	-0.920	0.920	Gained
Fadd	Traf2	Monocyte	TM	-0.917	0.917	Gained
Adam10	Notch2	TM	Prolif	-0.906	0.906	Gained
Mfng	Notch2	TM	IFN	-0.887	0.887	Gained
Cd200	Cd200r1	B_cell	TM	-0.835	0.835	Gained
Cd200r1	Cd200	TM	B_cell	-0.835	0.835	Gained
Hspa4	Tlr4	TM	BAM	-0.818	0.818	Gained
Pdgfb	Itgav	TM	IFN	-0.778	0.778	Gained
Pdgfb	Itgav	TM	HM_other	-0.749	0.749	Gained
Sema4d	Cd72	TM	T_cell	-0.703	0.703	Gained
Adam10	Notch2	TM	IFN	-0.647	0.647	Gained
Pdgfb	Itgav	TM	Monocyte	-0.645	0.645	Gained
Adam10	Tspan5	TM	Monocyte	-0.631	0.631	Gained
Adam17	Il6ra	TM	Neutrophil	-0.628	0.628	Gained
Adam17	Rhbdf2	TM	Monocyte	-0.621	0.621	Gained
Tnfsf13b	Tnfrsf17	HM_2	Prolif	-0.601	0.601	Gained
App	Cd74	TM	TM	+0.580	0.580	Lost
Gstp1	Traf2	DAM	TM	-0.569	0.569	Gained
S100a9	Cd68	B_cell	B_cell	+0.565	0.565	Lost
Ptdss1	Jmjd6	IFN	TM	-0.557	0.557	Gained
Sema4d	Cd72	TM	TM	+0.552	0.552	Lost
Tnfsf13b	Tnfrsf17	HM_2	HM_1	-0.549	0.549	Gained
App	Cd74	Neutrophil	TM	+0.548	0.548	Lost
Adam10	Cd44	TM	B_cell	-0.548	0.548	Gained
App	Cd74	Prolif	TM	+0.537	0.537	Lost
Copa	Cd74	Monocyte	TM	+0.524	0.524	Lost
Adam10	Notch2	TM	B_cell	-0.519	0.519	Gained
Adam10	Il6ra	TM	Neutrophil	-0.507	0.507	Gained
S100a8	Cd68	B_cell	T_cell	+0.507	0.507	Lost
App	Cd74	FTM	TM	+0.498	0.498	Lost
App	Cd74	BAM	TM	+0.492	0.492	Lost
Adam10	Tspan17	TM	HM_other	-0.488	0.488	Gained
Tgfb1	Lpp	Neutrophil	HM_other	+0.484	0.484	Lost
Ptdss1	Jmjd6	BAM	TM	-0.448	0.448	Gained
Tgfb1	Acvrl1	Neutrophil	IFN	+0.446	0.446	Lost
S100a8	Cd68	B_cell	B_cell	+0.440	0.440	Lost
S100a9	Itgb2	TM	TM	+0.439	0.439	Lost
App	Cd74	HM_other	TM	+0.424	0.424	Lost
Selplg	Itgb2	BAM	TM	+0.423	0.423	Lost
S100a8	Itgb2	B_cell	B_cell	+0.409	0.409	Lost
App	Cd74	Monocyte	TM	+0.400	0.400	Lost
Tnfsf13b	Tnfrsf17	HM_2	DAM	-0.383	0.383	Gained
S100a9	Itgb2	DAM	TM	+0.379	0.379	Lost
Ccl5	Ccrl2	Monocyte	IFN	-0.376	0.376	Gained
S100a9	Itgb2	BAM	TM	+0.368	0.368	Lost
Tnfsf13b	Cd40	HM_2	FTM	-0.367	0.367	Gained
S100a9	Itgb2	HM_2	TM	+0.366	0.366	Lost
Selplg	Itgam	DAM	TM	+0.346	0.346	Lost
S100a9	Itgb2	FTM	TM	+0.338	0.338	Lost
S100a9	Itgb2	IFN	TM	+0.328	0.328	Lost
S100a9	Itgb2	T_cell	TM	+0.325	0.325	Lost
App	Cd74	TM	Prolif	+0.311	0.311	Lost
S100a8	Itgb2	DAM	TM	+0.303	0.303	Lost
S100a8	Cd36	B_cell	BAM	+0.299	0.299	Lost
S100a8	Itgb2	BAM	TM	+0.298	0.298	Lost
S100a8	Itgb2	HM_2	TM	+0.295	0.295	Lost
S100a8	Itgb2	FTM	TM	+0.292	0.292	Lost

60 interactions: top 30 gained and top 30 lost by absolute ΔRank. Click column headers to sort.

Adam10→Trem2 Signaling Lost Novel

ADAM10-mediated TREM2 ectodomain shedding from TM to DAM is lost in ATG7-KO (ΔRank = +0.287). TREM2 cleavage is critical for DAM program activation and lipid sensing. This connects autophagy deficiency directly to the impaired TREM2 signaling axis in Alzheimer's disease.

Compensatory BAFF Signaling Novel

BAFF (Tnfsf13b→Tnfrsf17) emerges as a novel compensatory axis from HM_2 to DAM in ATG7-KO (ΔRank = −0.383). BAFF is typically a B cell survival factor; this microglia-to-microglia BAFF signaling may serve as a survival signal for signaling-impaired DAM cells, representing a potential therapeutic target.

Apoe→Lrp1 Lipid Supply Disrupted

Apoe→Lrp1 signaling from HM_2 to DAM is lost in KO (ΔRank = +0.115), providing cell-cell communication evidence for the cholesterol homeostasis pathway downregulation identified in GSEA (HALLMARK NES = −2.11). The lipid supply chain from homeostatic microglia to DAM is disrupted.

Cross-Dataset Validation

Shared DEGs

significant in both platforms

100%

Concordance

19/19 same direction, 0 discordant

0.22

Genome-wide r

Pearson, 11,842 genes

−0.921

Iron-MG rbc

FTM enrichment (padj ≈ 0)

Cross-validation of ATG7 effects between bulk RNA-seq (20-month FACS-sorted microglia) and scRNA-seq pseudobulk (5-month) across 11,842 shared genes reveals moderate but highly significant correlation (Pearson r = 0.22, p = 2.8 × 10⁻¹³⁰; Spearman ρ = 0.29). Among significant genes, correlation strengthens to r = 0.33. Of 19 DEGs shared between platforms, all 19 show perfect directional concordance (zero discordant), establishing a core age-independent ATG7 signature including Esr1, Cdkn1a, Blvrb, Sdf2l1, and Pdia5. Effects are moderately amplified at 20 months (median LFC ratio = 1.17). FTM signature validation confirms these cells score strongly on iron-MG consensus (rbc = −0.921) and MIMS-iron signatures, but not on ferroptosis signatures, establishing FTM as iron-homeostatic rather than ferroptotic.

Bulk vs scRNA-seq Log₂FC Correlation

Log₂FC comparison across platforms. Gold points: 19 genes significant in both (r = 0.87). Gray: 148 bulk-only DEGs. Blue: 22 scRNA-seq-only DEGs. Dashed line: regression through shared genes. All 19 shared DEGs fall in concordant quadrants (upper-right or lower-left).

FTM Signature Enrichment

FTM cells score strongly on iron-homeostatic signatures (green bars, *** = padj ≈ 0) but not on ferroptosis signatures (gray bars, NS). Effect size shown as −rank biserial correlation; positive values indicate FTM enrichment.

19 Cross-Validated DEGs

Gene	Bulk log₂FC	SC log₂FC	Bulk padj ▲	SC padj
Esr1	+5.85	+5.00	8.9e-288	2.9e-107
Pla2g7	+4.29	+2.16	1.5e-124	7.1e-16
Slc39a4	+1.51	+1.24	2.7e-10	3.6e-8
A530064D06Rik	+2.97	+2.20	7.6e-10	1.5e-26
Osgin1	+1.69	+1.45	2.7e-9	2.6e-8
Sdf2l1	-1.48	-1.00	1.4e-7	0.0004
Blvrb	+1.16	+1.25	5.2e-7	4.7e-19
Gdf3	+4.73	+2.02	1.9e-6	6.3e-6
Lyz1	+2.34	+2.86	2.5e-6	0.0472
Gpr165	-1.31	-1.03	4.0e-6	0.0001
Trpv4	+1.23	+1.05	1.3e-5	1.7e-6
Tmem205	+1.28	+1.24	5.1e-5	4.4e-18
Cdkn1a	+1.42	+1.62	0.0002	4.4e-14
Gstm1	+1.26	+1.36	0.0003	1.4e-14
Dnase1l3	+1.38	+1.17	0.0004	0.0027
Il18bp	+1.45	+1.48	0.0069	4.8e-21
Pdia5	-1.22	-1.02	0.0095	2.5e-5
Clec4n	+3.62	+1.16	0.0180	0.0099
Tnfaip2	+1.03	+1.99	0.0458	4.7e-20

Perfect Cross-Platform Concordance

All 19 shared DEGs between 5-month scRNA-seq and 20-month bulk RNA-seq show identical direction of change, with zero discordant genes across all 5 cross-dataset comparisons. This establishes a robust core ATG7 signature — including UPR impairment (Sdf2l1, Pdia5), iron metabolism (Blvrb), senescence (Cdkn1a), and estrogen receptor (Esr1) — conserved across ages and technologies.

FTM: Iron-Homeostatic, Not FerroptoticNovel

FTM cells score strongly on MIMS-iron and Iron-MG consensus signatures (rbc = −0.921) but not on ferroptosis signatures (FAS: NS, FerrDb: NS). Despite being labeled "ferritin microglia," their iron storage program is protective (anti-ferroptotic) rather than pathological. Fth1/Ftl1 are expressed in 99.6% of all cells and cannot discriminate FTM; Blvrb is the true marker.

ATG7: The Iron/Ferroptosis SwitchNovel

ATG7 loss increases ferroptosis scores in FTM (FAS padj = 3.9× 10⁻⁶) without altering iron-homeostatic identity (Iron-MG consensus: NS). This dissociation reveals autophagy as the switch between protective iron storage and ferroptotic vulnerability — impaired ferritinophagy converts sequestered iron into a source of toxic free iron. The effect is even stronger in DAM (FAS padj = 6.3 × 10⁻⁵¹).

Non-Microglia Immune Remodeling

32,340

Non-Microglia Cells

35.7% of total

0.55

Neutrophil OR

45% depleted in KO (padj ≈ 0)

1.76

T Cell OR

76% enriched in KO (padj ≈ 0)

5/6

Types with DEGs

216–927 DEGs per type

Among 32,340 non-microglia cells (35.7% of total), ATG7 deficiency in microglia reshapes the entire brain immune milieu through non-cell-autonomous effects. The reciprocal neutrophil depletion (OR=0.55) and T cell enrichment (OR=1.76) reveal a fundamental shift in peripheral immune cell recruitment. Monocytes, B cells, and BAMs are ATG7-neutral, demonstrating pathway-specific rather than general inflammatory remodeling. The disease (5xFAD) effect is independently powerful: neutrophil enrichment (OR=1.71) and BAM depletion (OR=0.64).

Non-Microglia Composition by Genotype

Percentage composition of 6 non-microglia cell types across genotype groups. Neutrophils dominate in Atg7fl/fl-5xFAD (37.8%) but are reduced in Atg7_dMG groups. T cells expand from ~19% in WT to ~28% in KO backgrounds.

Differentially Expressed Genes per Cell Type

Number of significant DEGs (KO vs WT, Wilcoxon test, top 30 per type shown) split by direction. Monocytes show the most DEGs (927), followed by neutrophils (492). Red = upregulated in KO, blue = downregulated.

Composition Tests: ATG7 and Disease Effects

Cell Type	Cells	OR (ATG7) ↓	padj (ATG7)	OR (5xFAD)	padj (5xFAD)
T cells	7,525	1.756	4.1e-100	0.702	1.1e-40
Other myeloid	3,906	1.144	1.2e-4	0.601	5.2e-49
B cells	4,750	1.023	0.562	1.270	7.4e-14
Monocytes	5,175	0.996	0.928	1.063	0.060
BAMs	2,097	0.995	0.928	0.640	2.4e-22
Neutrophils	8,887	0.549	2.1e-122	1.712	2.9e-100

Fisher's exact test for composition differences. OR > 1 = enriched in KO (ATG7 column) or 5xFAD (Disease column). Significant results (padj < 0.05) highlighted. T cells and neutrophils show the strongest and most significant ATG7 effects.

Non-Cell-Autonomous Immune ControlNovel

Microglia-specific ATG7 deletion indirectly depletes brain neutrophils by 45% while enriching T cells by 76%, revealing that microglial autophagy status controls the broader brain immune landscape. The selectivity is remarkable: monocytes, B cells, and BAMs are ATG7-neutral, indicating specific chemokine/cytokine pathways mediate this remodeling rather than general inflammatory changes.

Neutrophil Phenotype SwitchNovel

Remaining brain neutrophils in ATG7-KO upregulate MHC-I (B2m +0.62, H2-K1 +1.03, H2-D1 +0.42) while downregulating antimicrobial effectors (Tspo −0.96, Ifitm6 −1.08, Prok2 −2.29), indicating a shift from effector to antigen-presenting phenotype — consistent with the paradoxical gain of outgoing signaling despite numerical depletion (see Cell Communication).

Mechanistic Synthesis

Total Findings

Across 5 analysis tracks

Novel (Orthogonal)

Not reported in original paper

Extends Literature

Builds on known findings

High Confidence

8 medium confidence

Central Mechanistic Model

Novelty Classification

Of 30 findings, 17 are entirely novel (orthogonal to existing literature), 11 extend known biology, and 2 confirm previous reports.

Confidence Distribution

22 findings are supported by high-confidence evidence (multiple methods, strong statistics), while 8 are medium confidence (single method or smaller effect sizes).

Integrative Themes

UPR-Ferroptosis Double Vulnerability(5 findings)

UPR impairment (especially ATF6) and ferroptosis susceptibility are anti-correlated at the single-cell level, creating coordinated vulnerability

DAM Signaling Desert & Communication Isolation(5 findings)

DAM cells experience coordinated shutdown of growth/survival pathways and lose twice as many intercellular interactions as they gain

Non-Functional Overexpression Pattern(2 findings)

Esr1 gene massively upregulated but TF activity paradoxically down; similar disconnect in JAK-STAT (gene up, signaling down)

Homeostatic Microglia Bifurcation(3 findings)

ATG7 loss splits homeostatic microglia into hyperactive (depleted) and senescent (accumulating) sub-populations with distinct signaling profiles

Non-Cell-Autonomous Immune Remodeling(3 findings)

Microglia-intrinsic autophagy loss remodels the brain immune landscape with selective cell type effects

Additive Biology with Pathway Saturation(2 findings)

Gene-level effects are additive but translation/ribosome pathways show sub-additivity, suggesting compensatory capacity saturation

All Findings (30 of 30)

ID ↑	Track	Category	Finding	Novelty	Confidence
F01	T1	Cell Type Identification	27 clusters at resolution 0.8 resolve 6 HM sub-clusters, DAM, FTM, IFN, TM, Prolif, plus non-microglia (T cells, B cells, neutrophils, monocytes, BAMs)	Extends	high
F02	T1	Cell Type Identification	FTM identification requires Blvrb as discriminator — Fth1/Ftl1 expressed in 99.6% of all cells	Extends	high
F03	T1	Data Quality	Bulk RNA-seq sample wt3 is a clear outlier (1.26M reads, 27x lower depth) requiring exclusion	Novel	high
F04	T2	Trajectory	FTM connects most strongly to TM (PAGA w=0.27), not to HM or DAM, suggesting FTM derives from transitional microglia	Novel	medium
F05	T2	Trajectory	ATG7 affects DAM proportion but not DAM pseudotime position — cells entering DAM follow normal trajectory	Extends	medium
F06	T2	Gene Dynamics	UPR pseudotime gradient abolished in KO: WT microglia modulate UPR along activation but KO do not	Extends	high
F07	T2	Differential Abundance	Homeostatic microglia bifurcation: HM_1 depleted (logFC=-2.00) while HM_2 enriched (+1.21) in ATG7-KO	Novel	high
F08	T2	Differential Abundance	Neutrophils depleted in ATG7-KO brains (logFC=-1.01), indicating non-cell-autonomous effects of microglia autophagy loss	Novel	high
F09	T3	Differential Expression	Esr1 is the most dramatically upregulated gene in ATG7-KO microglia (log2FC=5.0 at 5mo, 5.85 at 20mo)	Novel	high
F10	T3	Differential Expression	UPR genes (Calr, Hspa5) confirmed as top downregulated genes across all comparisons	Confirms	high
F11	T3	Differential Expression	Disease response amplified 2.4x in KO background (146 vs 60 DEGs) with only 15 shared genes	Extends	high
F12	T3	Pathway Enrichment	Ferroptosis pathway significantly enriched in ATG7-KO DAM with co-enriched glutathione metabolism	Confirms	high
F13	T3	Pathway Enrichment	Cholesterol homeostasis specifically downregulated in ATG7-KO DAM (NES=-2.11) but not HM	Extends	high
F14	T3	Temporal Dynamics	Progressive senescence: p21/Cdkn1a at both ages, p16/Cdkn2a emerging only at 20 months	Extends	high
F15	T3	Interaction Analysis	ATG7 × 5xFAD interaction is statistically ADDITIVE at gene level (zero interaction genes at FDR<0.1)	Novel	high
F16	T3	Signature Scoring	Ferroptosis and UPR are anti-correlated at single-cell level, creating 'double vulnerability' in KO	Novel	medium
F17	T4	TF Activity	ATF6 arm is the most severely affected UPR branch (not IRE1/XBP1 or PERK/ATF4 as typically focused)	Extends	high
F18	T4	TF Activity	Esr1 TF ACTIVITY paradoxically DOWN despite massive GENE upregulation — non-functional overexpression	Novel	medium
F19	T4	TF Activity	Spi1/PU.1 myeloid master regulator down in 8/11 microglia subtypes — impaired myeloid identity	Extends	high
F20	T4	TF Activity	MHC-II TF regulators (RFXAP/RFXANK/RFX5) strongly reduced in DAM — upstream regulatory evidence for impaired antigen presentation	Novel	high
F21	T4	Pathway Activity	DAM 'signaling desert': 8 growth/survival pathways down, only p53 and TNFa up	Novel	high
F22	T4	Pathway Activity	HM bifurcation explained mechanistically: HM_1 = hyperactive signaling (MAPK/PI3K), HM_2 = senescent (p53/NFkB)	Novel	high
F23	T4	Cell Communication	DAM cells lose 2x more interactions than gain (42 lost vs 21 gained) — communication isolation	Novel	high
F24	T4	Cell Communication	Adam10→Trem2 (TM→DAM) lost — impaired TREM2 ectodomain shedding connects autophagy to TREM2 axis	Novel	medium
F25	T4	Cell Communication	Compensatory BAFF signaling (Tnfsf13b→Tnfrsf17) emerges HM_2→DAM in KO	Novel	medium
F26	T5	Cross-Validation	19 DEGs shared between 5mo scRNA-seq and 20mo bulk with perfect directional concordance (0 discordant)	Extends	high
F27	T5	Cell Type Characterization	FTM is iron-homeostatic, NOT ferroptotic — autophagy is the switch to ferroptotic vulnerability	Extends	high
F28	T5	Cell Type Characterization	Two distinct ferroptotic mechanisms: DAM (failed iron homeostasis) vs FTM (iron overload from impaired ferritinophagy)	Novel	medium
F29	T5	Non-Microglia Effects	Microglia-specific ATG7 loss depletes neutrophils 45% and enriches T cells 76% — non-cell-autonomous immune remodeling	Novel	high
F30	T5	Non-Microglia Effects	Neutrophils shift from effector to MHC-I-upregulated phenotype in KO brains	Novel	medium

Complete catalogue of 30 findings across 5 analysis tracks. Click column headers to sort. Use filter buttons to focus on specific novelty categories.

Methods Note

This synthesis integrates results from 5 analysis tracks: T1 (Quality & Cell Atlas), T2 (Trajectory & Dynamics), T3 (Differential Expression & Pathways), T4 (Regulatory Networks & Communication), and T5 (Cross-Validation & Integration). Novelty classification compares each finding against the original publication (Cai et al. 2025, JEM) and existing literature. Confidence is based on effect size, statistical significance, cross-method validation, and biological plausibility. Findings are classified as orthogonal (completely novel), extends (builds on known findings with new detail), or confirms (independently validates published results).

This report was generated with the assistance of AI. While every effort has been made to ensure accuracy, AI can make mistakes — please verify key findings against primary data before drawing conclusions.