The hypothesis underlying this work was that TREM2/TYROBP plays a causal role in driving molecular networks in AD [11]. To test this hypothesis, we used Drosophila to identify molecular interactions between neuronal expression of Aβ42 or tau and glial expression of TREM2/TYROBP.
Figure 1 shows an overview of the design and data analysis in the present paper. Briefly, we developed transgenic fly models expressing or co-expressing human TREM2WT/TYROBP or TREM2R47H/TYROBP in glial cells, Aβ42 in neurons, or tau in the retina. These transgenic fly models were characterized for several phenotypic changes including Aβ42 accumulation, the status of tau phosphorylation levels, behavioral deficits, and neurodegeneration. Then transcriptome-wide gene expression in control and transgenic fly heads were profiled by RNA-seq to identify differentially expressed gene (DEG) signatures between different genotype groups. The GO and pathway terms enriched in the DEG signatures were identified. Several DEGs were validated by qPCR. Lastly, human orthologs of the fly DEG signatures were projected onto gene networks from human AD datasets to explore the relevance of these gene signatures to AD pathogenesis from a network prospective. Gene regulatory relationship was characterized for a highlighted inflammatory response subnetwork using Bayesian network analysis.
Establishment of transgenic flies co-expressing TREM2 (TREM2WT) or TREM2 with pathogenic R47H variant (TREM2R47H) with TYROBP in glial cells
In order to co-express human TREM2 and TYROBP in fly glial cells, we generated transgenic flies carrying wild-type (WT) human TREM2 (TREM2WT), TREM2 with AD-related R47H variant (TREM2R47H), or TYROBP under the control of a tissue-specific LexA operator [57]. Expression of each transgene was driven by a pan-glial driver Repo-LexA, and their messenger RNA (mRNA) expression was confirmed by RT-PCR analysis (Fig. 2a).
We found that, while the expression of TYROBP proteins was readily detectable by western blotting, TREM2 protein levels were undetectable, raising the possibility that ectopically expressed human TREM2 proteins may be unstable in fly glial cells perhaps because a binding partner that is required to stabilize TREM2 protein was absent (Fig. 2b). Indeed, when TREM2 and TYROBP transgenes were combined and co-expressed in fly glial cells, TREM2 proteins became readily detectable (Fig. 2c). A prior report showed that R47H mutation reduced the stability of TREM2 proteins [58]. We compared protein levels of TREM2WT and TREM2R47H in fly brains and found no significant difference between them (Fig. 2c). When TYROBP was immunoprecipitated from the lysate of bigenic fly brains co-expressing TYROBP and TREM2, TREM2 proteins (both TREM2WT and TREM2R47H) were also precipitated, indicating that TREM2 and TYROBP proteins interact and stabilize each other in fly glial cells (Fig. 2d).
In mammalian cells, TREM2 is cleaved by α-secretase, which results in production of N- and C-terminal fragments of TREM2 [59]. The N-terminal fragments of TREM2 are secreted (sTREM2) and promote inflammatory responses [60], while C-terminal fragments of TREM2 are further processed by γ-secretase [59]. Western blotting using an anti-TREM2 antibody detected the C-terminal fragment of both TREM2WT and TREM2R47H (Fig. 2e), suggesting that TREM2 is processed and that sTREM2 is produced in fly glial cells in a manner similar to that observed in mammalian cells.
FEA revealed that significant overlap between molecular pathways affected by neuronal expression of Aβ42 and those affected by glial expression of TREM2WT/TYROBP in fly brains
To gain insights into the effects of glial expression of TREM2/TYROBP in the fly brains at the molecular level, we generated RNA-seq data from the brain samples in control flies (control) and flies with co-expression of TREM2WT and TYROBP (TREM2WT/TYROBP). Differentially expressed genes between control and TREM2/TYROBP flies were identified using two criteria including fold change > 1.2 and a FDR < 0.05 in an analysis using linear models implemented using the R package LIMMA [61]. Expression of TREM2WT/TYROBP resulted in upregulation of 239 genes and downregulation of 373 genes (Fig. 3a and Additional file 2: Table S1).
Since TREM2/TYROBP signaling is known to promote survival of microglial cells [28], we evaluated whether ectopic expression of TREM2 and TYROBP induced any structural changes in the fly brain and/or significantly altered the number of glial cells or neurons. No significant alteration in the size or gross morphology of brain structures was observed in TREM2WT/TYROBP bigenic flies (Additional file 1: Figure S1). In addition, immunostaining of fly brains against a glial marker protein, Repo, or a neuronal marker, Elav, revealed that the numbers of glial cells or neurons were not significantly different between control and TREM2WT/TYROBP bigenic fly brains (Additional file 1: Figure S1). These results suggest that gene expression changes induced by ectopic expression of TREM2WT/TYROBP are not due to either structural defects or altered number of neurons or glial cells in the fly brain.
To identify the biological pathways that are affected by glial expression of TREM2WT/TYROBP, we performed functional enrichment analysis (FEA) for the DEG signatures using GO annotation. The genes upregulated by the expression of TREM2WT/TYROBP were significantly enriched (multiple testing corrected FET p value < 0.05) for pathways designated as “oxidoreductase activity, acting on paired donors, with incorporation or reduction of molecular oxygen,” “intracellular membrane-bounded organelle,” “electron carrier activity,” “heme binding,” “iron ion binding,” “oxidation-reduction process,” and “cellular response to heat” (Fig. 3b and Additional file 3: Table S2). In contrast, the genes downregulated by the expression of TREM2WT/TYROBP were enriched (corrected FET p < 0.05) in the pathways including “integral component of plasma membrane,” “extracellular region,” “extracellular matrix,” “structural constituent of chitin-based cuticle,” “extracellular space,” “potassium ion transport,” and “myosin light chain kinase activity” (Fig. 3b and Additional file 3: Table S2).
We next compared molecular pathways affected by glial expression of TREM2WT/TYROBP and those affected by neuronal expression of Aβ42 in fly brains. In mammals, the majority of Aβ peptides are produced from amyloid precursor protein (APP) in the late secretory pathway [62]. In our Aβ42 fly model, a signal sequence was fused to the N-terminus of Aβ42 [43] to target the peptide to the secretory pathway of neurons. Western blot analysis detected monomeric forms of Aβ42 as 4 kDa signals (Fig. 4a and [43]) and immunoprecipitation followed by mass spectrometry analysis confirmed that the fused signal peptide was correctly cleaved and intact Aβ42 peptides were produced [43]. Although this Aβ42 fly model directly expressed Aβ42 peptides in the endoplasmic reticulum, immuno-electron microscopy (Immuno-EM) detected Aβ42 signals in the secretory pathway, including ER, Golgi, and lysosomes [44], with minimal signals in the mitochondria and cytoplasm of neurons in Aβ42 fly brains. Moreover, secretion of Aβ peptides occurred in Drosophila cultured cells [44] and, in Drosophila brains, immuno-EM analysis occasionally detected Aβ42 accumulation in glial cells, suggesting that Aβ42 peptides were secreted from neurons and then taken up by glial cells [44]. The expression of Aβ42 in this model caused learning deficits followed by locomotor dysfunction and neurodegeneration with accumulation of detergent-insoluble Aβ42, in the brains. These results suggest that our Aβ42 fly model may recapitulate some aspects of Aβ42-mediated toxicity. Similar approaches have been utilized to generate transgenic Aβ42 fly models by other groups with consistent neurodegenerative phenotypes [63,64,65].
RNA sequence analysis in our Aβ42 fly model identified that neuronal expression of Aβ42 upregulated 437 genes and downregulated 485 genes in heads as compared to control flies (Fig. 3a, Additional file 2: Table S1). The upregulated DEGs were enriched in pathways including “endomembrane system,” “endoplasmic reticulum,” and “oxidation-reduction process” (Fig. 3b and Additional file 3: Table S2). By contrast, the downregulated DEGs were enriched in “electron carrier activity,” “oxidoreductase activity, acting on paired donors, with incorporation or reduction of molecular oxygen,” “intracellular membrane-bounded organelle,” “heme binding,” “oxidation-reduction process,” “iron ion binding,” “extracellular space,” “transferase activity, transferring phosphorus-containing groups,” “extracellular region,” “metabolic process,” “carboxylic ester hydrolase activity,” “cuticle pigmentation,” and “melanin biosynthetic process” (Fig. 3b and Additional file 3: Table S2).
Interestingly, this analysis revealed that eight of the 15 pathways (13 pathways for DEGs downregulated by Aβ42, three pathways for DEGs upregulated by Aβ42, one pathway is overlapped) enriched for the Aβ42 DEGs were also enriched for the TREM2WT/TYROBP DEGs in the same or opposite direction (Fig. 3b and Additional file 3: Table S2). For example, the pathways “extracellular space” and “extracellular region” were enriched for the genes downregulated by Aβ42 and by TREM2WT/TYROBP, while “oxidoreductase activity, acting on paired donors, with incorporation or reduction of molecular oxygen,” “oxidation-reduction process,” “electron carrier activity,” “heme binding,” “iron ion binding,” and “intracellular membrane-bounded organelle” were enriched for the genes downregulated by Aβ42 and for the genes upregulated by TREM2WT/TYROBP.
These unbiased analyses revealed that gene expression changes induced by neuronal expression of Aβ42 and glial expression of TREM2WT/TYROBP merged onto the same molecular pathways. However, in fly genome, there is no clear ortholog of either TREM2 or TYROBP. One possibility could be that glial cells sense Aβ42 and/or associated neuronal damages and then induce gene expression changes through endogenous signaling pathways. Ectopically expressed human TREM2/TYROBP may sense these damage-associated signals and impact the overlapping molecular pathways.
TREM2R47H/TYROBP induces gene expression changes similar to those by TREM2WT/TYROBP in the fly brains
TREM2 variants were originally identified as causative mutations in patients with Nasu-Hakola disease [66]. However, recent genetic analysis revealed that R47H variant of TREM2 is associated with a three- to fourfold increased risk for AD [14, 15]. To examine the impact of glial expression of TREM2R47H/TYROBP in the fly brains at the molecular level, we generated RNA-seq data from the head samples in flies with the co-expression of TREM2R47H and TYROBP (TREM2R47H/TYROBP) as described above [61]. Expression of TREM2R47H/TYROBP resulted in 290 upregulated genes and 365 downregulated ones (Fig. 3a and Additional file 2: Table S1). No significant alteration in either the size, gross morphology of brain structures, the numbers of glial cells, or neurons was observed in TREM2R47H /TYROBP bigenic flies (Additional file 1: Figure S1), suggesting that gene expression changes induced by ectopic expression of TREM2R47H/TYROBP is not due to either structural changes or altered number of neurons or glial cells in the fly brain.
At the pathway level, FEA using GO annotation revealed that genes upregulated or downregulated by TREM2R47H/TYROBP were enriched in the same categories as those with TREM2WT/TYROBP (Fig. 3b and Additional file 3: Table S2), although some of the pathways that were enriched in the DEGs in TREM2WT/TYROBP, such as “iron ion binding,” “myosin light chain kinase activity,” and “cellular response to heat,” were not significantly enriched in the DEGs in TREM2R47H/TYROBP (Fig. 3b and Additional file 3: Table S2). The DEG signatures from TREM2WT/TYROBP and TREM2R47H/TYROBP flies shared about half of their members and that 98% of those overlapped genes changed in the same direction (corrected FET p = 2.0 × 10−246, 25.1-fold for downregulated genes; corrected FET p = 1.9 × 10−133, 27.4-fold for upregulated genes) (Fig. 3c).
To quantify differences in gene expression induced by TREM2WT/TYROBP and TREM2R47H/TYROBP, we directly compared mRNA expression levels between these two groups and identified 145 upregulated genes and 157 downregulated genes in TREM2R47H/TYROBP compared to TREM2WT/TYROBP (Fig. 3a). Interestingly, at an FDR of 5%, the upregulated DEGs are significantly enriched for “odorant binding,” “sensory perception of chemical stimulus,” “defense response,” and “response to pheromone,” suggesting that these functional pathways were activated by R47H mutation in TREM2 (Fig. 3b and Additional file 3: Table S2). Among these genes, Drosophila Toll-4 gene (the closest ortholog of human TLR7) detected in the “defense response” module is of particular interest, since TREM2 family proteins are known to modulate Toll-like receptor signaling in mammals [67, 68].
We also compared molecular pathways affected by neuronal expression of Aβ42 and those affected by glial expression of TREM2R47H/TYROBP (Fig. 3a, Additional file 2: Table S1). At the pathway levels, four of the above 15 pathways enriched in the Aβ42 DEGs were enriched in the TREM2R47H/TYROBP DEGs. At the gene level, a significant overlap between Aβ42 DEGs and TREM2 R47H/TYROBP DEGs was observed (Fig. 3c, corrected p value ≤ 1.0 × 10−32, ≥ 7-fold).
Taken all together, biological pathways affected by glial expression of TREM2WT/TYROBP or TREM2R47H/TYROBP in fly heads are similar but about 300 genes show significant difference in mRNA expression. Moreover, glial expression of TREM2R47H/TYROBP impacts several common molecular pathways affected by neuronal expression of Aβ42, though TREM2WT/TYROBP appears to impact many more other common pathways affected by neuronal expression of Aβ42.
Expression of TREM2/TYROBP in glial cells modifies molecular signatures induced by Aβ42 expression in neurons in fly brains
To investigate the effects of TREM2WT/TYROBP on phenotypes as well as gene expression signatures induced by Aβ42, we achieved neuronal expression of Aβ42 and glial expression of TREM2WT/TYROBP in fly brains by using two tissue-specific transgenes expression systems in Drosophila (Additional file 1: Figure S2A). Phenotypic characterization revealed that glial overexpression of TREM2WT/TYROBP did not affect either Aβ42 accumulation levels (Fig. 4a), courtship learning and memory (Fig. 4b), or Aβ42-mediated neurodegeneration (Fig. 4c); however, some exacerbation of Aβ42-mediated behavioral deficits was observed (Fig. 4d and Additional file 1: Figure S2B).
We next analyzed the effects of glial expression of TREM2WT/TYROBP on gene expression signatures induced by neuronal expression of Aβ42 in fly brains. RNA sequence analyses revealed that expression of Aβ42/TREM2WT/TYROBP resulted in upregulation of 533 genes and downregulation of 727 genes compared to control flies (Additional file 1: Figure S2C). Interestingly, comparison of FEA results between Aβ42/TREM2WT/TYROBP and Aβ42 flies revealed that seven of the 15 pathways enriched for the Aβ42 DEGs disappeared when TREM2WT/TYROBP was expressed in glia (Additional file 1: Figure S2D and Additional file 4: Table S3). These pathways include “electron carrier activity,” “oxidoreductase activity, acting on paired donors, with incorporation or reduction of molecular oxygen,” and “heme binding,” which were enriched in the DEGs downregulated by Aβ42 and in the DEGs upregulated by TREM2WT/TYROBP (Fig. 3b and Additional file 3: Table S2).
Using the same strategy, we also analyzed the effects of glial expression of TREM2R47H/TYROBP on phenotypes as well as gene expression signatures induced by neuronal expression of Aβ42 in fly brains (Additional file 1: Figure S2A). Similar to TREM2WT/TYROBP, glial overexpression of TREM2R47H/TYROBP did not affect Aβ42 accumulation levels (Fig. 4a), courtship learning and memory (Fig. 4b), or Aβ42-mediated neurodegeneration (Fig. 4c); although there is a trend toward subtle exacerbation of Aβ42-mediated behavioral deficits (Fig. 4d and Additional file 1: Figure S2B).
RNA sequence analyses revealed that expression of Aβ42/TREM2R47H/TYROBP resulted in upregulation of 661 genes and downregulation of 846 genes (Additional file 1: Figure S2C and Additional file 2: Table S1). Comparison of FEA results between Aβ42/TREM2R47H/TYROBP and Aβ42 flies revealed that nine of the 15 pathways enriched in the Aβ42 DEGs disappeared following glial expression of TREM2R47H/TYROBP (Additional file 1: Figure S2D and Additional file 4: Table S3). Among these nine pathways, six pathways were also disappeared following glial expression of TREM2WT/TYROBP, suggesting that the effects of TREM2R47H/TYROBP on Aβ42 were similar to that of TREM2WT/TYROBP by this analysis.
Taken together, glial expression of TREM2/TYROBP modifies molecular signatures induced by neuronal expression of Aβ42. Since glial expression of TREM2/TYROBP did not reduce Aβ42 levels (Fig. 4a), the observed changes in FEA results are not simply due to reduced response to Aβ42 in fly brains. In addition, since TREM2/TYROBP proteins are expressed in glial cells and Aβ42 peptides are expressed in neurons (Additional file 1: Figure S2A), this modulatory action likely reflects non-cell autonomous effects by TREM2/TYROBP.
Neuronal Aβ42 and glial TREM2R47H/TYROBP synergistically downregulated genes associated with synaptic and immune function modules of the co-expressed gene networks from human AD brains
Gene co-expression network analysis has uncovered a number of co-expressed gene modules pathologically related to human complex diseases including AD [11]. To investigate the relevance of the gene expression signatures in Aβ42, TREM2WT/TYROBP, TREM2R47H/TYROBP, Aβ42/TREM2WT/TYROBP, and Aβ42/TREM2R47H/TYROBP fly brains to AD, we investigated their association with the 111 co-expressed gene modules derived from co-regulation analyses of brain gene expression in the Harvard Brain Tissue Resource Center (HBTRC) AD and controls. The modules were annotated by the GO or pathways that the modules were enriched for. To do this, the fly DEGs were first converted to human orthologs by using DIOPT (DIOPT score > 1) [69]. The enrichment analysis shows that the genes upregulated by Aβ42 were enriched in the modules associated with “extracellular region” and “chaperone” (Fig. 5a and Additional file 5: Table S4a) while no module was enriched for the downregulated DEGs (Additional file 5: Table S4a). Neither the TREM2WT/TYROBP nor the TREM2R47H/TYROBP DEG signature showed significant enrichment in any of the HBTRC modules. By contrast, the DEGs upregulated by Aβ42/TREM2WT/TYROBP were enriched in the “chaperone” module (FET p = 0.022, 3.2-fold; Fig. 5a, Additional file 5: Table S4a) while Aβ42/TREM2R47H/TYROBP expression was associated with downregulation of genes enriched in synaptic transmission, neuronal activities, and transmission of nerve impulses, with FET p = 0.001 (2.1-fold), 0.024 (1.9-fold), and 0.024 (1.6-fold), respectively (Fig. 5a, Additional file 5: Table S4a). These results indicate that neuronal expression of Aβ42 and glial expression of TREM2R47H/TYROBP synergistically downregulated genes known to be associated with AD pathology [11], thus supporting the prediction that TREM2/TYROBP play roles in AD pathogenesis.
To validate these findings, we performed qPCR analyses in five DEGs from the “synaptic transmission” module with known functions related to neuronal activity; Sh, SK, Shab, para, and Nmdar2 (fly orthologs for potassium voltage-gated channel subfamily A, potassium calcium-activated channel subfamily N, potassium voltage-gated channel subfamily B, sodium voltage-gated channel alpha subunits, and glutamate ionotropic receptor NMDA type subunits, respectively). We found that expression levels of Sh were slightly downregulated by Aβ42 expression alone, while those of Sh and Nmdar2 were slightly downregulated by expression of TREM2R47H/TYROBP alone (Fig. 5b). By contrast, expression levels of all five genes were significantly downregulated in Aβ42/TREM2R47H/TYROBP flies.
We further examined whether downregulation of para, a fly ortholog for sodium voltage-gated channel alpha subunits, modifies neuronal dysfunction in Aβ42 flies. Neuronal knockdown of para by RNAi significantly worsened Aβ42-induced locomotor deficits (Fig. 5c). Moreover, neuronal knockdown of para by itself caused modest decline in locomotor functions in flies (Fig. 5d).
Taken altogether, these network analysis results suggest that genes associated with synaptic transmission were synergistically downregulated by co-expression of Aβ42 and TREM2R47H/TYROBP, which may lead to neuronal dysfunction.
To further explore the association of the DEG signatures with early phases of AD, we intersect them with the co-expression network modules from an independent cohort in the ROSMAP study. Again, the modules were annotated by the GO/pathways that the modules were most enriched for. The result is summarized in Fig. 5a and Additional file 5: Table S4b. The genes downregulated by Aβ42/TREM2WT/TYROBP were enriched for an “inflammatory response” module (salmon; corrected FET p = 0.005, 2.6-fold) and an “organic acid metabolism” module (cyan; corrected FET p = 0.014, 2.5-fold). Note that the “inflammatory response” module salmon was ranked number 7 in relation to AD pathology after ranking ROSMAP modules using multiple sorting features, including module-trait correlations and enrichment for genes correlated with or differentially expressed regarding neuropathological and clinical traits Braak staging, global cognition, CERAD neuropathological category (and, by extension, NIA-Reagan score). Importantly, this inflammatory module is not enriched for the DEGs by either Aβ42 alone or TREM2WT/TYROBP alone, suggesting interactions between Aβ42 and TREM2WT/TYROBP at the level of gene expression.
When gene expression levels were compared between Aβ42/TREM2/TYROBP and Aβ42 alone, the downregulated genes in Aβ42/TREM2WT/TYROBP were enriched in the “inflammatory response” module (salmon; corrected FET p = 0.015, 3.5-fold), while downregulated genes in Aβ42/TREM2R47H/TYROBP were enriched in the same “inflammatory response” module (salmon; corrected FET p = 5 × 10−4, 3.4-fold) and a “locomotion” module (yellow; ranked number 9; corrected FET p = 0.034, 1.9-fold).
In summary, for both TREM2WT/TYROBP and TREM2R47H/TYROBP, interaction with Aβ42 affected the “inflammatory response” pathway in flies. This is an interesting observation since neuroinflammation is implicated as a significant contributor to AD pathogenesis and is also consistent with the proposed anti-inflammatory consequences of TREM2 signaling in human microglia.
To test if the enrichment of synaptic transmission and inflammatory response modules was biased by the conserveness of these two pathways between fly and human, we analyzed the enrichment of known GO categories (based on the MSigDB gene sets) in the human orthologs of fly genes. As shown in Additional file 6: Table S5, the mostly enriched gene sets are big pathways including cytoplasm, metabolic process, nucleus, organelle part, and macromolecular complex, which account for 14.6%, 12.2%, 10%, 8.8%, and 7.2% of the 10,938 high confidence human orthologous genes (DIOPT score > 1; http://www.flyrnai.org/cgi-bin/DRSC_orthologs.pl) accordingly, with FDR adjusted FET p value < 1.2E-58. In contrast, the immune system genes only account for < 1.1% of the orthologs and were not enriched (FET p value ≥ 0.54), while synaptic transmission accounted for 1.0% of the orthologs and was marginally enriched (FET p value = 0.004). Thus, it is unlikely that the significant correlation with the inflammatory and synaptic modules in fly signatures were caused by an artifact of overrepresentation of these pathways in the fly-human orthologous genes.
In summary, since Aβ42 accumulation, TREM2/TYROBP activation, altered inflammatory response, and synaptic dysfunctions are all implicated in early phases of AD pathogenesis, Aβ42/TREM2/TYROBP flies may recapitulate some molecular signatures relevant to early stages of AD.
Molecular pathways affected by neuronal expression of tau do not overlap with those affected by glial expression of TREM2/TYROBP in fly brains
In the pathogenesis of AD, abnormal accumulation and toxicity of tau is believed to play a critical role in neurodegeneration. Thus, identification of molecular signatures induced by simultaneous activation of TREM2/TYROBP axis and accumulation of tau may provide important information underlying neurodegenerative process in AD.
We first compared molecular pathways affected by neuronal expression of tau and those affected by glial expression of TREM2/TYROBP in fly heads. We performed RNA sequence analyses and characterized gene expression signatures using an established fly model of human tau toxicity [70] in which expression of human tau causes progressive degeneration of photoreceptor neurons in the retina [71]. Expression of tau in photoreceptor neurons using GMR-GAL4 driver upregulated 384 genes and downregulated 418 genes in the heads compared to control flies (Additional file 1: Figure S3A and Additional file 2: Table S1). The upregulated DEGs in tau fly heads were associated with “endosome transport via multivesicular body sorting pathway,” “ESCRT III complex,” and “vacuolar transport” (Additional file 1: Figure S3B and Additional file 7: Table S6). In contrast, downregulated DEGs in tau fly heads were significantly enriched in “rhabdomere” and “striated muscle thin filament” (Additional file 1: Figure S3B and Additional file 7: Table S6).
We also analyzed gene expression changes caused by pan-glial expression of TREM2WT/TYROBP or TREM2R47H/TYROBP in the same genetic background carrying the GMR-GAL4 driver. Glial expression of TREM2WT/TYROBP upregulated 448 genes and downregulated 306 genes while TREM2R47H/TYROBP upregulated 475 genes and downregulated 426 genes (Additional file 1: Figure S3A and Additional file 2: Table S1). There were 29–52 genes common between tau DEGs and TREM2/TYROBP DEGs (Additional file 1: Figure S3C, corrected p value ≤ 10−10, ≥ 3.9-fold). However, we observed no pathway that was enriched in both DEG signatures (Additional file 1: Figure S3B and Additional file 7: Table S6).
Taken together, these results suggest that molecular signatures induced by expression of tau are dissimilar to those induced by TREM2/TYROBP in fly heads at the functional pathway level.
Glial expression of TREM2/TYROBP exacerbated tau-mediated neurodegeneration
Next, we examined the effects of glial expression of TREM2/TYROBP on gene expression signatures as well as neurodegenerative phenotypes induced by tau expression. In order to achieve expression of tau in photoreceptor neurons and expression of the TREM2/TYROBP complex in glial cells simultaneously, we utilized two tissue-specific transgenes expression systems in Drosophila (Additional file 1: Figure S4A).
Expression of human tau in photoreceptor neurons causes progressive neurodegeneration in the lamina [71], the first synaptic neuropil of the optic lobe containing photoreceptor axons and abundant glial cells [72]. We observed that pan-glial expression of both TREM2WT/TYROBP and TREM2R47H/TYROBP significantly exacerbated this neurodegeneration, while pan-glial expression of TREM2/TYROBP alone (i.e. in the absence of neuronal tau expression) did not show neurodegeneration (Fig. 6a). We also examined whether glial expression of TREM2/TYROBP increased the levels of tau and/or tau phosphorylated at AD-related sites. Western blot analyses with pan-tau or phospho-tau specific antibodies did not detect significant increase in either tau levels or phosphorylation status of tau by glial expression of TREM2/TYROBP (Fig. 6b). These results suggest that glial expression of TREM2/TYROBP exacerbates tau-mediated neurodegeneration without affecting tau accumulation or phosphorylation status, consistent with recent report using TREM2 deficiency mice [73].
Analysis of the gene regulatory network in AD brains revealed that tau and TREM2/TYROBP synergistically downregulated genes overrepresented in the modules related to immune systems associated with AD pathogenesis
We generated RNA-seq data from tau/TREM2WT/TYROBP and tau/TREM2R47H/TYROBP flies and identified gene expression signatures in comparison with control flies. Expression of tau/TREM2WT/TYROBP upregulated 377 genes and downregulated 476 genes, while expression of tau/TREM2R47H/TYROBP upregulated 596 genes and downregulated 601 genes (Additional file 1: Figure S4B and Additional file 2: Table S1).
Most of the pathways enriched in these DEGs (Additional file 1: Figure S4C and Additional file 8: Table S7) were the same as those detected in either TREM2/TYROBP alone or tau alone (Additional file 1: Figure S3B and Additional file 7: Table S6). However, we observed that “proteolysis” and “UDP-glycosyltransferase activity” were uniquely enriched in the DEGs downregulated by tau/TREM2WT/TYROBP and tau/TREM2R47H/TYROBP, respectively (Additional file 1: Figure S3B and Additional file 8: Table S7). The “proteolysis” pathway contains proteases including angiotensin-converting enzyme (ACE), which have been associated with AD [74], and UDP-glycosyltransferases, enzymes associated with oligodendrocyte myelination, disruption of which has been implicated in neurodegeneration in AD [12].
To further explore the relevance of the gene expression signatures in tau, TREM2WT/TYROBP, TREM2R47H/TYROBP, tau/TREM2WT/TYROBP, and tau/TREM2R47H/TYROBP flies to AD, we investigated their association with the 111 co-expressed gene modules derived from co-regulation analyses of brain gene expression in the HBTRC AD and controls [11]. The enrichment analysis shows that no module was enriched for tau DEG signature. Moreover, neither the TREM2WT/TYROBP nor the TREM2R47H/TYROBP DEG signature showed significant enrichment in any of the HBTRC modules.
By contrast, the downregulated DEG signatures in tau/TREM2WT/TYROBP and tau/TREM2R47H/TYROBP were enriched in the “cadherin” module (corrected FET p = 0.035, 1.6-fold; Fig. 7a, Additional file 9: Table S8a) and the “extracellular region” module (corrected FET p = 0.023, 2.0-fold; Fig. 7a, Additional file 9: Table S8a), respectively. Since these two modules are predicted to be highly associated with AD pathology [11], our data suggest pathological interactions between tau and TREM2/TYROBP at the level of gene expression in flies.
In the co-expression network from ROSMAP, the DEGs downregulated by tau/TREM2WT/TYROBP and tau/TREM2R47H/TYROBP significantly overlapped with three modules (Fig. 7a, Additional file 9: Table S8b): the “inflammatory response” module (salmon; ranked number 7) (corrected FET p = 9.0 × 10−4, 3.3-fold and corrected FET p = 3.7 × 10−4, 3.0-fold, respectively); “locomotion” (yellow; ranked number 9) (corrected FET p = 4.4 × 10−3, 2.1-fold and corrected FET p = 0.04, 1.7-fold, respectively); and “organic acid metabolism” (cyan; ranked number 24) (corrected FET p = 5.7 × 10−3, 3.1-fold and corrected FET p = 0.02, 2.5-fold, respectively). These three modules also significantly overlapped with the Aβ42-related DEG signatures, as described above (Fig. 5a). The “locomotion” and “organic acid metabolism” modules were also enriched for DEGs downregulated by expression of tau alone (corrected FET p = 0.01, 2.0-fold and corrected FET p = 0.02, 2.8-fold, respectively), while the “inflammatory response” module salmon was enriched in the DEGs downregulated by expression of TREM2WT/TYROBP alone (corrected FET p = 4.1 × 10−5, 4.9-fold). The “inflammatory response” module (or the salmon module, ranked number 7) was also enriched in the DEGs from downregulated by tau/TREM2WT/TYROBP in comparison with tau (corrected FET p = 0.01, 3.6-fold), or by tau/TREM2R47H/TYROBP in comparison with tau (corrected FET p = 0.006, 3.4-fold). In the ROSMAP data (Additional file 10: Table S9), this salmon module had three members downregulated in AD brains, including GADD45A, FABP5, and BAALC-AS1, the first two of which were also downregulated in the present tau/TREM2WT/TYROBP and tau/TREM2R47H/ TYROBP flies (FET p = 9.2 × 10−5, 14.7-fold), consistent with the existence of substantial network consistency when human data and fly data are compared.
Of particular interest, the DEGs upregulated by tau/TREM2WT/TYROBP were enriched for another “inflammatory response” module (lightcyan; ranked number 8; corrected FET p = 0.04, 2.9-fold) in the ROSMAP network. This lightcyan module contained five AD GWAS loci, including CD33, INPP5D, MS4A4A/MS4A6A, RIN3, and TREM2 (Additional file 10: Table S9). Moreover, TYROBP was a member of this ROSMAP lightcyan module. This module was not enriched with DEGs upregulated by either tau alone or TREM2WT/TYROBP alone, suggesting that genetic interactions between tau and TREM2WT/TYROBP may induce this gene expression signature. Moreover, significant enrichment was not observed with the DEG signatures in tau/TREM2R47H/TYROBP flies, suggesting that the TREM2R47H variant may have weaker impact on this module than does TREM2WT.
Taken together, these results revealed that different components of the immune response system were either activated or inhibited by the tau/TREM2/TYROBP pathway. Since both the salmon and lightcyan modules were highly ranked for their predicted relationship to AD pathology, the present results highlighted the importance of inflammatory response subnetworks as potential targets for disease intervention.
As shown in Additional file 1: Figure S5, the salmon and lightcyan modules in the ROSMAP network were adjacent to each other in the cluster dendrogram, indicating that the two inflammatory response modules were highly related in the human data, even though they were regulated differently in tau/TREM2/TYROBP flies. Therefore, our fly models provide valuable biological insights into the human data that were not otherwise evident. To investigate the causal regulatory relationships among the inflammatory response module genes, we combined the genes from the two inflammatory response modules and overlaid the combined gene set onto a Bayesian causal network constructed from the ROSMAP data by using an approach described in our previous study [11]. Figure 7b shows the network structure of the 604 inflammatory response genes of which 270 genes have fly orthologs. Over one-third (96) of the 270 fly orthologs were differentially expressed in at least one of the fly transgenic models analyzed here, resulting in a 1.3-fold enrichment (p value = 2.5 × 10−4). TYROBP was highlighted as a key regulator for controlling a large number of downstream genes in this inflammatory response network: 17, 36, and 79 genes were in the immediate first, second, and third layer downstream of TYROBP, respectively. Therefore, the causal network analysis further validated the causal role of TYROBP and informed other novel key regulators, which modulate the inflammatory response pathways, such as, LAPTM5, MYO1F, CLIC1, and CSF1R, which were highlighted by a large node size in Fig. 7b.