TM2D genes regulate Notch signaling and neuronal function in Drosophila

TM2 domain containing (TM2D) proteins are conserved in metazoans and encoded by three separate genes in each model organism species that has been sequenced. Rare variants in TM2D3 are associated with Alzheimer’s disease (AD) and its fly ortholog almondex is required for embryonic Notch signaling. However, the functions of this gene family remain elusive. We knocked-out all three TM2D genes (almondex, CG11103/amaretto, CG10795/biscotti) in Drosophila and found that they share the same maternal-effect neurogenic defect. Triple null animals are not phenotypically worse than single nulls, suggesting these genes function together. Overexpression of the most conserved region of the TM2D proteins acts as a potent inhibitor of Notch signaling at the γ-secretase cleavage step. Lastly, Almondex is detected in the brain and its loss causes shortened lifespan accompanied by progressive motor and electrophysiological defects. The functional links between all three TM2D genes are likely to be evolutionarily conserved, suggesting that this entire gene family may be involved in AD.


Introduction
Alzheimer's disease (AD) is the most common neurodegenerative disease affecting the aging population and accounts for the large majority of age-related cases of dementia [1,2]. AD is pathologically characterized by histological signs of neurodegeneration that are accompanied by formation of extracellular plaques and intra-neuronal tangles. Numerous studies have identified genetic factors that contribute to AD risk and pathogenesis [3]. Rare hereditary forms of AD are caused by dominant pathogenic variants in APP (Amyloid Precursor Protein), PSEN1 (Presenilin 1) or PSEN2 (Presenilin 2). These three genes have been extensively studied using variety of experimental systems, and the resultant knowledge has led to greater understanding of how they contribute to the formation of extracellular plaques found in both familial and sporadic AD brains [4]. PSEN1 and PSEN2 are paralogous genes that encode the catalytic subunit of the γ-secretase, a membrane-bound intramembrane protease complex [5]. γ-secretase substrates include many type-I transmembrane proteins including APP as well as Notch receptors that play various roles in development and physiology [6,7].
While studies of genes that cause familial AD have been critical in providing a framework to study pathogenic mechanisms of this disorder, pathogenic variants in APP and PSEN1/2 are responsible for only a small fraction of AD cases [8]. Familial AD can be distinguished from more common forms of AD because most patients with pathogenic APP or PSEN1/2 variants develop AD before the age of 65 [early-onset AD (EOAD)]. The majority (>95%) of AD cases are late-onset (LOAD, develops after 65 years of age), sporadic and idiopathic in nature [9]. In these patients, it is thought that multiple genetic and environmental factors collaborate to cause damage to the nervous system that converges on a pathway that is affected by APP and PSEN1/2. To reveal common genetic factors with relatively small effect sizes, multiple genome-wide association studies (GWAS) have been performed and identified over 40 loci throughout the genome that confer increase risk to developing AD [3]. The most notable riskfactors are variant alleles in APOE [10]. Although the precise molecular mechanism by which different alleles of APOE increase or decrease the risk of AD has been extensively debated, a number of studies have proposed that this gene is involved in the clearance of toxic Aβ peptides [11]. A recent meta-analysis has also identified ADAM10 (encoding a β-secretase enzyme that cleaves APP and Notch) as an AD associated locus [12], suggesting that genes involved in familial EOAD and sporadic LOAD may converge on the same molecular pathway. Functional studies of these and other newly identified risk factors for AD are critical to fully understand the etiology of this complicated disease that lack effective treatments or preventions.
We previously reported that a rare missense variant (rs139709573, NP_510883.2:p.P155L) in TM2D3 (TM2 domain containing 3) is significantly (OR = 7.45, pMETA = 6.6x10 -9 ) associated with increased risk of developing LOAD through an exome-wide association study in collaboration with the CHARGE (Cohorts for Heart and Aging Research in Genomic Epidemiology) consortium [13]. This variant was also associated with earlier age-at-onset that corresponds to up to 10 years of difference with a hazard ratio of 5.3 (95% confidence interval 2.7-10.5) after adjusting for the common ε4 allele of APOE. Although the function of this gene in vertebrates was unknown and this missense variant was not predicted to be pathogenic based on multiple variant pathogenicity prediction algorithms including SIFT [14], PolyPhen [15] and CADD [16], we experimentally demonstrated that p.P155L has deleterious consequences on TM2D3 function based on an assay we established using Drosophila embryos [13]. The Drosophila ortholog of TM2D3, almondex (amx), was initially identified based on an X-linked female sterile mutant allele (amx 1 ) generated through random mutagenesis [17]. Although homozygous or hemizygous (over a deficiency) amx 1 mutant females and hemizygous (over Y chromosome) males are viable with no morphological phenotypes, all embryos laid by amx 1 hemi/homozygous mothers exhibit severe developmental abnormalities including expansion of the nervous system at the expense of the epidermis [18,19]. This 'neurogenic' phenotype results when Notch signaling mediated lateral inhibition is disrupted during cell-fate decisions in the developing ectoderm [20,21]. By taking advantage of this scorable phenotype, we showed that the maternal-effect neurogenic phenotype of amx 1 hemizygous females can be significantly suppressed by introducing the reference human TM2D3 expressed under the regulatory elements of fly amx, but TM2D3 p.P155L expressed in the same manner fails to do so [13]. This showed that the function of TM2D3 is evolutionarily conserved between flies and humans, and the molecular function of TM2D3 that is relevant to LOAD may also be related to Notch signaling. More recently, another rare missense variant (p.P69L) in this gene has been reported in a proband that fit the diagnostic criteria of EOAD or frontotemporal dementia [22], indicating that other TM2D3 variants may be involved in dementia beyond LOAD.
TM2D3 is one of three highly conserved TM2 domain containing (TM2D) proteins encoded in the human genome. The two other TM2 domain-containing proteins, TM2D1 and TM2D2, share a similar protein domain structure with TM2D3, and each protein is encoded by a highly conserved orthologous gene in Drosophila that has not been functionally characterized (CG10795 and CG11103, respectively) (S1 Fig). All TM2D proteins have a predicted Nterminal signal sequence and two transmembrane domains that are connected through a short intracellular loop that are found close to the C-terminus. Within this loop, there is an evolutionarily conserved DRF (aspartate-arginine-phenylalanine) motif, a sequence found in some G-protein coupled receptors that mediates their conformational change upon ligand binding [23]. The extracellular region between the signal sequence and first transmembrane domain is divergent in different species as well as among the three TM2D containing proteins. In contrast, the sequences of the two transmembrane domains as well as the intracellular loop is highly conserved throughout evolution as well as between the three TM2 domain containing proteins [24] (S1 Fig). The three proteins also have short C-terminal extracellular tails that are evolutionarily conserved within orthologs but vary among the three proteins (e.g. TM2D1 has a slightly longer C'-tail than TM2D2 and TM2D3). The molecular functions of these conserved and non-conserved domains of TM2D proteins are unknown.
Amx has been proposed to function at the γ-secretase cleavage step of Notch activation based on a genetic epistasis experiment performed previously [25]. Notch signaling activation is initiated by the binding of the Notch receptor to its ligands (Delta or Serrate in Drosophila) [26]. This induces a conformational change of Notch to reveal a cleavage site that is recognized by ADAM10 [encoded by kuzbanian (kuz) in Drosophila] [27]. Notch receptor that has undergone ADAM10 cleavage (S2 cleavage) is referred to as N EXT (Notch extracellular truncation) and becomes a substrate for γ-secretase [28]. N EXT that is cleaved by γ-secretase (S3 cleavage) releases its intracellular domain (N ICD ), which then translocates to the nucleus and regulates transcription of downstream target genes [29]. To determine how amx regulates Notch signaling, Michellod and Randsholt attempted to suppress the embryonic neurogenic phenotype of embryos produced from amx 1 mutant mothers by zygotically overexpressing different forms of Notch using a heat-shock promoter [25]. While N ICD was able to weakly suppress the neurogenic defect, N EXT was not able to do so, suggesting that Amx somehow modulates the cleavage step that invovles the γ-secretase complex. However, because the phenotypic suppression observed by N ICD in this study was very mild and since the authors used Notch transgenes that were inserted into different regions of the genome (thus N ICD and N EXT are likely to be expressed at different levels and cannot be directly compared), additional data is required to fully support this conclusion.
In this study, we generated clean null alleles of all three Drosophila TM2D genes using CRISPR/Cas9-mediated homology directed repair (HDR) and assessed their functions in vivo. Surprisingly, we found that CG10795 (TM2D1) and CG11103 (TM2D2) knockout flies are phenotypically indistinguishable from amx (TM2D3) null animals, displaying severe maternaleffect neurogenic phenotypes. We also generated double-and triple-knockout animals to determine whether these three genes have redundant functions in other Notch signaling dependent contexts during development. The triple-knockout of all TM2D genes did not exhibit any obvious morphological phenotypes but shared the same maternal-effect neurogenic phenotype similar to the single null mutants, suggesting these three genes function together. We also provide further evidence that Amx functions on γ-secretase to modulate Notch signaling in vivo, and further uncover a previously unknown role of this gene in the maintenance of neural function in adults.

A clean null allele of amx fully recapitulates previously reported maternaleffect phenotypes
In a previous study, Michellod et al. reported the generation of a null allele of amx (amx m ) by generating flies that are homozygous for a deletion that removes amx and several other genes [Df(1)FF8] and bringing back a genomic rescue construct for Dsor1, an essential gene that lies within this locus [30]. The authors reported that zygotic amx m [actual genotype: Df(1)FF8; P (Dsor1 +S )] mutant flies have reduced eye size and nicked wing margin as adults, which are phenotypes that had not been reported in the classic amx 1 allele. Because the amx 1 allele is caused by a 5bp deletion that may still produce a product that encodes a portion of the N'-extracellular domain (amx is a single exon gene in Drosophila, hence will not be subjected to nonsense mediated decay), the authors concluded that amx 1 is a hypomorphic allele.
Since Df(1)FF8 has not been molecularly characterized and it was uncertain whether all of the phenotypes attributed to amx m are due to loss of amx alone, we decided to generate a clean null allele of amx (hereafter referred to as amx Δ ) using CRISPR/Cas9 technology [31,32] (Fig  1). We knocked-in a dominant wing color marker (y wing2+ ) to replace the coding sequence of amx using HDR [33] (Fig 1A). Insertion of this cassette was screened by the presence of the visible marker (dark wings) in a yellow mutant background and the targeting event was molecularly confirmed via Sanger sequencing. We also confirmed the loss of amx transcript by RT-PCR (S2A Fig). Similar to amx 1 , homozygous amx Δ females exhibit sterility and all embryos produced by these animals exhibit a neurogenic phenotype (Fig 1C and 1D). Both female fertility and neurogenic phenotypes of their progeny can be suppressed by introducing a 3.3 kb genomic rescue construct containing the amx locus (Fig 1A, 1C and 1E) [13]. Human TM2D3 expressed using the same regulatory elements has~50% activity of fly Amx (Fig 1C  and 1G), consistent with what we previously observed using amx 1 hemizygous females [13]. We did not observe any morphological defects in the eye and wing of homozygous and hemizygous amx Δ flies at all temperatures tested (between 18-29˚C). We also looked at other tissues that are often affected when Notch signaling is defective including the nota and legs [34,35] but we did not observe any morphological phenotypes in these tissues either. In summary, amx Δ is the first clean loss-of-function (LoF) allele of amx generated by CRISPR/Cas9, and phenotypically resembles the classic amx 1 allele rather than the amx m allele.

Null alleles of other TM2D genes exhibit the maternal-effect neurogenic phenotype similar to amx
Using a similar strategy that we used to generate amx Δ [33], we generated null alleles for CG11103 (TM2D2 ortholog, located on the X-chromosome) and CG10795 (TM2D1 ortholog, located on the 2 nd chromosome), two uncharacterized genes without official names (Fig 2). This time, we inserted a dominant body color marker (y body+ ) into the endogenous locus of CG11103 or CG10795 to individually knockout these genes (Fig 2B and 2F), and we phenotypically and molecularly characterized these alleles in a similar manner. We confirmed the loss of each transcript by RT-PCR (S2A Fig), confirming that the new lines we generated are true null alleles.
Both males and females that are hemizygous or homozygous null for CG11103 are viable and these animals do not exhibit any morphological defects. Similar to amx, CG11103 mutant females are fully sterile, and all embryos laid by these mothers exhibit neurogenic phenotypes (Fig 2C). Importantly, these phenotypes are rescued by a 1.5 kb genomic construct containing the CG11103 locus (Fig 2B, 2D and 2E). Given the phenotypic and molecular similarities with amx, we gave CG11103 the name amaretto (amrt), after the sweet Italian liqueur traditionally flavored with almonds. The knockout allele for this gene is referred to as amrt Δ hereafter.
Like amx and amrt, mutants that are homozygous null for CG10795 appear morphologically normal, but all females are sterile and their embryos exhibit a neurogenic phenotype ( Fig  2G). We were able to suppress these phenotypes using a~40kb fosmid transgene in which the CG10795 protein is C-terminaly tagged with multiple epitopes including GFP (Fig 2F, 2H and 2I) [36]. Given the phenotypic similarity to amx and amrt mutants, we named this gene biscotti (bisc), after the Italian biscuits traditionally made with almonds. The knockout allele for this gene is referred to as bisc Δ hereafter and the fosmid-based GFP-tagged genomic rescue transgene is referred to as bisc::GFP.

Triple knockout of TM2D genes is phenotypically similar to single gene knockouts
The embryonic neurogenic defect is a rare phenotype that is almost exclusively associated with genes that affect the Notch signaling pathway [20]. Given the importance of Notch signaling in most stages of development preceding adulthood [7], it was peculiar that all three TM2D gene mutants do not exhibit any obvious developmental morphological defects related to Notch signaling while all three alleles exhibit a robust maternal-effect neurogenic phenotype. To test whether this may be due to redundancy between the three genes, we generated a fly strain that lacks all three TM2D genes (Fig 3). membrane localization. Transmembrane 2 (TM2) domain is boxed in red. Hexagon denotes where 3xHA epitope is located in 3xHA::Amx protein. (C) Egg hatching assay shows that genomic rescue constructs can suppress embryonic lethality (amx Δ n = 857. amx Δ + amx n = 139. amx Δ + 3xHA::amx n = 1673. amx Δ + TM2D3 n = 257). Error bars show SEM. One-way ANOVA followed by Dunnett test. ���� = p-value�0.0001. (D-G) Embryonic nervous tissue (neuronal nuclei, Elav, green) of developing embryos. Embryos from homozygous amx Δ females (D) exhibit a neurogenic phenotype. This phenotype can be suppressed by wild-type amx (E), 3xHA::amx (F), or human TM2D3 (G) genomic rescue constructs. Scale bars = 100 μm in (D-G).
https://doi.org/10.1371/journal.pgen.1009962.g001 Because amx and amrt are both located on the X-chromosome (located in cytological regions 8D2 and 12C4, respectively, which are 19cM away), we recombined the two null alleles by following the dominant wing color and body color markers knocked into each locus ( Fig  3A). The amx, amrt double null flies (amx Δ amrt Δ ) in a yellow mutant background exhibit wild-type color wings (y wing2+ marker of amx Δ ) and bodies (y body+ marker of amrt Δ ). Absence of both amx and amrt transcripts in this line was verified by RT-PCR (S2B Fig). These double null adult flies also appeared morphologically normal similar to the single null animals ( Fig  3F, 3I and 3L), suggesting that these two genes do not play redundant roles during the development of imaginal tissues. Consistent with single null animals, the amx, amrt double null females were fully sterile, exhibiting a maternal-effect neurogenic phenotype (Fig 3C).
We next combined the bisc Δ allele on the second chromosome with the amx Δ amrt Δ mutant X-chromosome to generate a triple null mutant line (amx Δ amrt Δ ; bisc Δ ) (Fig 3A). Absence of transcripts of all three TM2D genes in this line was verified by RT-PCR (S2B Fig). This genetic manipulation still did not produce any adult animals with obvious Notch signaling related external morphological defects (Fig 3G, 3J and 3M), suggesting that these three genes also do not play redundant roles in these contexts. Consistent with the single and double null lines, the triple null mutant females are also completely sterile and their progeny show maternal-effect neurogenic defects (Fig 3D). We did not observe any differences in the severity of the neurogenic phenotype in the embryos from the single gene knockouts and the triple knockout mothers (Figs 1D, 2C, 2G and 3D), likely because the neurogenic defect is already strong in single gene knockout animals, precluding any additive or synergistic effects.
According to large-scale transcriptomic studies, mRNA of amx, amrt and bisc are expressed relatively ubiquitously with some enrichment found in the female ovary [37,38]. Within the ovary, these three genes are likely to be expressed in similar cell types including germline cells and follicle cells according to a large single cell RNA-sequence database that have recently been released on the bioRxiv preprint server [39]. To further test whether amx, amrt and bisc do not have overlapping functions, we assessed whether genomic rescue transgenes for amrt or bisc can rescue amx mutants. If TM2D family proteins perform similar molecular tasks, increased expression of one gene is expected to suppress the LoF phenotype of another gene. Homozygous amx Δ female flies with amrt or bisc::GFP genomic rescue transgenes (used to rescue amrt Δ and bisc Δ in Fig 2, respetively) are fully sterile, demonstrating that amrt or bisc are not able to substitute for the loss of amx. Together with our earlier finding that single TM2D gene knockout flies are phenotypically indistinguishable (Figs 1 and 2), and that no further phenotype or exacerbation is seen in the triple null mutant line (Fig 3), we conclude that these three genes are uniquely maternally required during embryonic neurogenesis and likely function together.
https://doi.org/10.1371/journal.pgen.1009962.g002 developmental contexts that are known to depend on Notch. To further study the function of TM2D proteins in vivo, we tested whether overexpression of Amx is sufficient to modulate Notch signaling (Fig 4). In addition to generating transgenes that allow the expression of a fulllength Amx protein with or without an N'-3xHA tag (placed immediately after the predicted signal sequence, 3xHA::Amx FL ) (Fig 4A) using the GAL4/UAS binary expression system [40], we generated a version of the N'-3xHA tagged construct that lacks the majority of the non-conserved N'-terminal extracellular domain (3xHA::Amx ΔECD ) (Fig 4D) to specifically test the function of the highly conserved TM2 domain through overexpression experiments.  Schematic of proteins generated from UAS-3xHA::amx FL and UAS-3xHA::amx ΔECD transgenes. A 3xHA epitope (grey hexagon) was inserted after a predicted signal sequence (SS, blue box). The majority of the extracellular domain (ECD, light brown box) was removed to generate 3xHA::amx ΔECD , consisting mostly of the TM2 domain (red box) tagged with the N-terminal 3xHA epitope. (B) Overexpression of 3xHA::Amx FL with pannier(pnr)-GAL4 has no effect on notum morphology. The absence of bristles along the midline is due to a very mild dorsal closure defect caused by the pnr-GAL4 insertion. (C) Expression of 3xHA::Amx FL in the posterior wing using engrailed (en)-GAL4 has no effect of the morphology of wings of adults raised at 29˚C. We first overexpressed these transgenes in the developing dorsal thorax using pannier-GAL4 (pnr-GAL4). While we observed that Amx FL does not cause any morphological defects (Fig 4B), we found that Amx ΔECD causes an increase in the number of mechanosensory bristles (Fig 4E), suggesting an effect on Notch mediated lateral inhibition. We then expressed the two proteins in the developing posterior compartment within wing imaginal discs using engrailed-GAL4 (en-GAL4). We observed notching of the posterior wing margin when Amx ΔECD was expressed (Fig 4F), while no such defect was seen upon expression of Amx FL (Fig 4C). To determine whether the wing notching caused by Amx ΔECD overexpression was indeed due to loss of Notch signaling, we visualized Notch activation using NRE-GFP (Notch Responsive Element-Green Fluorescent Protein), a synthetic in vivo Notch signaling reporter [41], as well as immunostaining of Cut, encoded by an endogenous downstream target gene of Notch activation in this context [42]. While overexpression of Amx FL did not affect NRE-GFP and Cut expression (Fig 4G-4G"), Amx ΔECD caused a reduction in both NRE-GFP and Cut expression within the en-GAL4 expression domain (Fig 4H-4H"). In summary, overexpression of full-length Amx did not affect developmental events related to Notch signaling in the thorax and wing, while overexpression of a truncated form that only carries the conserved TM2 domain and its short C'-tail inhibited Notch signaling in several developmental contexts.

Truncated Amx inhibits Notch signaling at the γ-secretase mediated receptor cleavage step
To determine how Amx ΔECD inhibits Notch signaling when ectopically overexpressed, we performed similar epistasis experiments as [25], but with improved genetic tools (Fig 5). First, we generated several new UAS constructs expressing different forms of Notch and inserted them into the identical genomic location on the 2 nd chromosome using site specific ϕC31-mediated transgenesis to avoid positional effects [43,44] (Fig 5A). In addition to transgenes that allow expression of N ICD , N EXT and full-length Notch (N FL ), we also generated a ligand-independent form of UAS-Notch that still depends on ADAM10 and γ-secretase by deleting several epidermal growth factor-like repeats (EGF) of the extracellular domain that contains the ligand binding domain and Lin-12/Notch Repeats (LNR) within the negative regulatory region (N ΔEGF1-

18.LNR
) based on an earlier construct designed by [45]. When we overexpressed N ICD , N EXT or N ΔEGF1-18.LNR in the developing wing pouch using nubbin-GAL4 (nub-GAL4), we observed increased Cut expression throughout the wing pouch, indicating ectopic Notch activation ( Fig  5B-5E and 5I). Over-expression of N FL only showed a mild increase in Cut expression in limited regions of the wing pouch (S3A Fig), likely due to its ligand-dependence. These observations are consistent with previous reports using UAS-Notch transgenic lines generated using random (P-element mediated) transgenesis technology [46].
Next, we recombined nub-GAL4 and UAS-3xHA::amx ΔECD onto the same chromosome. The resultant genetic recombinants constitutively overexpresses Amx ΔECD in the wing pouch, with heterozygous adults exhibiting wing notching and homozygous animals completely lacking wings in adulthood (S4A and S4B Fig). We then crossed these nub-GAL4, UAS-3xHA::amx ΔECD flies to different UAS-Notch lines to determine if Amx ΔECD can modulate the ectopic Cut expression phenotype caused by Notch overexpression. We found that co-overexpression of Amx ΔECD significantly suppresses the induction of Cut expression caused by N ΔEGF1-18.LNR or N EXT (Fig 5G-5I), but had no effect on the same phenotype caused by N ICD (Fig 5F and 5I). This indicates that amx ΔECD genetically acts at the γ-secretase-mediated cleavage step of Notch activation, consistent with previous epistasis experiments performed in embryos using amx 1 [25].
To further understand how Amx ΔECD inhibits Notch signaling upon overexpression, we assessed the distribution of the Notch receptor through immunostaining using a monoclonal antibody that recognizes both cleaved and uncleaved forms of Notch [47]. Upon overexpression with decapentaplegic-GAL4 (dpp-GAL4), which is expressed in a limited domain within the wing pouch, we observed that Amx ΔECD causes wing notching (S4D Fig) that is accompanied by a dramatic upregulation of Notch receptor (Figs 5K-5K' and S5B). Wing notching was not observed when Amx FL was overexpressed (S4C Fig), consistent with the results obtained using en-GAL4 and nub-GAL4. To test whether a similar phenotype is seen upon loss of γ-secretase function, we knocked down Presenilin (Psn, which is orthologous to human PSEN1 and PSEN2) in the wing disc and assessed its effect on Notch protein levels. When we performed RNAi against Psn using a UAS-RNAi line that had been validated in a previous study [48] with dpp-GAL4, we found that knockdown of Psn shows an accumulation of Notch similar to Amx ΔECD over-expression but to a lesser extent (Figs 5L-5L' and S5C). Next, to determine if the Notch receptor was accumulating at the cell surface, we performed a detergent-free immunostaining using an antibody that recognizes the extracellular domain of the Notch receptor [49]. We found that the level of Notch present at the cell surface was increased upon overexpression of Amx ΔECD or Psn knockdown, however to a lesser extent compared to the total Notch staining (S5D-S5F ' Fig). Therefore, Notch receptors seem to accmumulate both at the cell surface and in intracellular compartments in these contexts, suggesting a defect in protein trafficking or turnover. To also test whether this Notch accumulation phenotype occurs when Notch cleavage is altered by another mechanism, we tested whether loss of ADAM10 also causes this defect. When we generated mutant clones of a null allele of kuz (kuz e29-4 ) [50] using MARCM (Mosaic analysis with a repressible cell marker) [51], we did not observe any alterations in Notch levels (S6 Fig), indicating that the increase in Notch levels seen upon over-expression of Amx ΔECD and Psn knockdown is rather specific.
To obtain additional insights into why overexpression of Amx ΔECD inhibits Notch signaling whereas Amx FL does not, we examined the expression and subcellular localization of the two proteins. We found that Amx ΔECD is expressed at significantly lower levels (~10%) compared to Amx FL (S7A and S7B Fig), suggesting that the lack of phenotype induced by Amx FL overexpression was not due to lower expression of this protein compared to Amx ΔECD . Furthremore, we found that while Amx FL exhibited a memebranous localization pattern when expressed in the developing wing disc (S7C Fig), Amx ΔECD was found in a more punctate pattern (S7D Fig). These differences in protein levels and localization may contribute to their different ability to modulate Notch signaling.
In summary, while over-expression of the full-length Amx protein did not cause any obvious defects, we serendipitously found that a truncated form of this protein that contains the small region that is highly conserved among all TM2D proteins can act as a potent inhibitor of Notch signaling when overexpressed in a several tissues including the wing disc, which is a tissue that endogenously expresses all three TM2D genes (S7E Fig). We also found through epistasis experiments that Amx ΔECD acts at the S3 cleavage step of Notch activation, suggesting that it likely regulates γ-secretase or factors/processes that facilitate this cleavage. This is further supported by our findings that overexpression of Amx ΔECD and knock-down of Psn, but not loss of kuz, both lead to accumulation of Notch both at the cell surface as well as intracellularly.
Amx ΔECD causes an increase of Notch protein levels (K') compared to overexpression of a neutral protein, LacZ (βgalactosidase) (J'). Knockdown of psn mediated by shRNA also results in mild increase Notch levels (L'), mimicking the effect of 3xHA::Amx ΔECD . DAPI marks nuclei of wing disc cells whereas mCherry labels the dpp-GAL4 expression domain, the boundary of which is indicated by white dash lines, in J-L. Scale bars = 50 μm in (B-H). Scale bars = 5 μm in (J-L). https://doi.org/10.1371/journal.pgen.1009962.g005

Null mutants of TM2D family genes have shortened lifespan
AD is an adult-onset age-dependent disease that worsens over time. While TM2D3 p.P155L has been associated with LOAD [13] and TM2D3 p.P69L has been recently reported in a proband with EOAD or frontotemporal dementia [22], there is no functional data that directly links this gene to an age-dependent neurological phenotype in any species. In Drosophila, knockdown of genes encoding subunits of the γ-secretase complex, Psn and Nicastrin (Nct), specifically in adult neurons causes a shortening of lifespan phenotype which is accompanied by an age-dependent motor function decline [48]. Considering our results suggesting that Amx is involved in a process that relates to γ-secretase (Fig 5), we asked whether loss of amx causes an age-dependent phenotype in flies (Fig 6). To asssess the effect on lifespan, we compared the longevity of amx Δ hemizygous male flies with control flies that also carry a genomic rescue construct (amx Δ + amx) to neutralize the effect of genetic backgroud [13]. We selected males for our analysis because amx loss does not affect male fertility, allowing us to ignore any changes in lifespan that may be caused by alterations in fecundity [52]. In contrast to the rescued control animals that exhibit a median lifespan of 51 days, the median lifespan of amx null flies is significantly shorter at 27 days (Fig 6A, p = 1.0x10 -11 ). We next tested whether human TM2D3 can substitute for the loss of amx in this context by introducing the humanized genomic rescue construct [13] into the amx null mutant background (amx Δ + hTM2D3). These humanized TM2D3 animals had a median lifespan of 33 days (Fig 6A). Although this effect seems to be statically significant (p = 9.5x10 -10 ) and suggests that human TM2D3 may have some activity in this context, we cannot rule out the possibility that other genetic factors may be contributing to this effect considering the weakness of the phenotypic rescue.
Next, considering that loss of amx is phenotypically indistinguishable from the loss of amrt and bisc, we assessed whether loss of amrt or bisc causes a similar lifespan shortening phenotype in adult flies. By comparing the lifespan of amrt Δ or bisc Δ to their respective genomic rescue controls, we found that both mutants are significantly short lived (S8A- S8B Fig). Median lifespan of amrt Δ was 48 days compared to 63 days in rescued control flies (p = 6.0x10 -11 ), and median lifespan of bisc Δ was 32 days compared to 48 days in rescued animals (p = 3.0x10 -12 ). The difference in the median lifespan of rescued animals for the three mutants could be due to differences in their genetic backgrounds. In summary, flies that lack amx, amrt or bisc are short lived, suggesting that TM2D genes contribute to maintaining a normal lifespan.

Amx is expressed in the adult brain and required in neurons to maintain a normal lifespan
To further study the role of TM2D genes in aging animals, we decided to perform additional experiments on amx null mutant flies, considering the link between its human ortholog to AD. Shortened lifespan seen in amx Δ animals could be due to a number of reasons, including defects in the nervous system or other organs functionally affected by the loss of amx. Based on publicly available microarray [37], RNA sequencing (RNA-seq) [38] and single cell RNAsequencing (scRNA-seq) [53] datasets, amx mRNA has been detected in the adult nervous system at low levels in a subset of neurons and glia cells (S9 Fig). To determine whether Amx protein can be detected in the adult nervous system, we generated an N'-tagged amx genomic rescue construct in which a 3xHA epitope is inserted immediately after the signal sequence of the Amx protein (Fig 1A and 1B), similar to the UAS-3xHA::amx transgene discussed earlier (Fig 4A). This tagged genomic rescue construct is able to rescue the female sterility and the maternal-effect neurogenic phenotype of amx Δ (Fig 1C and 1F), indicating the epitope tag does not have a major effect on Amx function. Next, we verified expression of 3xHA::Amx in the female ovary through immunofluorescent staining and western blot (S10 Fig). Based on (A) Lifespan assay shows amx Δ animals (black, n = 247) have reduced lifespan compared to amx Δ + amx controls (orange, n = 224). amx Δ + human TM2D3 flies (green, n = 234) have significantly longer lifespan than amx Δ animals, but shorter than control. Animals were reared at 25˚C; Log-rank test (Mantel-Cox), ���� = p<0.0001. (B) Restoring Amx expression specifically in neuronal cells rescues the reduced lifespan in amx null animals. LacZ overexpression was used as a control. Animals were reared at 25˚C; Log-rank test (Mantel-Cox), ���� = p<0.0001. (C) Western blot of amx Δ ; 3xHA::amx brains shows that 3xHA::Amx (predicted 35 kDa size) is expressed in the adult nervous system (arrow). Protein isolate from five brains was loaded per lane and Actin was probed as a loading control. (D) Schematization of the giant fiber electrophysiological recordings. Stimulating electrodes are inserted into the brain and recording electrodes record responses from the TTM and DLM muscles. (E-F) TTM muscles of 5d old amx Δ mutants immunostaining, we detected the highest signal of 3xHA::Amx in nurse cells of the ovary and observed that it localizes to the cell membrane as well as in intracellular puncta (S10A and S10B Fig). Based on western blot, we identified 3xHA::Amx at the predicted molecular weight [34.82 kDa (31.35 kDa for Amx, 3.47 kDa for the 3xHA tag)] (S10C Fig). Finally, we assessed the expression of 3xHA::Amx in the adult brain via immunostaining and western blot. While we did not observe a strong signal beyond background fluorescence based on immunofluorescence staining using an anti-HA antibody (S11 Fig), we detected 3xHA::Amx expression via western blot using brain extracts (Fig 6C). Therefore, in addition to amx mRNA that has been detected through multiple high-throughput transriptomic studies, Amx protein is also expressed in the adult brain, albeit at relatively low levels.
To determine a if amx is required neuronally, we reintroduced wild-type Amx specifically in neurons using nSyb-GAL4 [54] in amx null animals. We found that the lifespan defect of amx Δ can be significantly suppressed by this manipulation (p<1.0x10 -16 , Fig 6B). In conclusion, Amx is expressed in the adult nervous system at relatively low levels, and required in neurons to maintain a normal lifespan.

amx null mutants show progressive motor and electrophysiological defects
Considering its neuronal requirement, we next tested whether loss of amx causes age-dependent behavioral defects. Since neuronal knockdown of γ-secretase genes in flies was reported to cause age-dependent climbing defects [48], we assessed whether amx null animals also show this phenotype. We found that amx Δ mutants show a climbing defect as early as 5 days posteclosion, which progresses in severity as time goes on (S8C Fig), suggesting that this gene is not only required to maintain normal lifespand but also for healthspan [55].
To directly assess whether amx is required for maintenance of neural function, we performed electrophysiology on amx Δ animals and controls. The giant fiber system is a neuronal circuitry that is required for rapid escape responses in insects and has been used as a model circuitry to assess neuronal function in a quantitative manner in adult flies [56]. This pathway can be activated through direct stimulation of the brain and outputs of the circuit can be monitored by recording the responses from tergotrochanteral (TTM) and dorsal longitudinal (DLM) muscles (Fig 6D). For a given time point, we applied multiple stimulations at different frequencies (20, 50, and 100 Hz; Figs 6E-6H and S12-S14), and measured how well these muscles respond to each stimulation. Healthy neurons can follow the stimulations showing a 'probability of response' that is close to 1.0, but neurons that are unhealthy show decreased probability of response, which indicates failure of the signal to travel from the brain to the muscles [57,58].
At 5 days post-eclosion, amx Δ flies have a minor but significant failure rate at later stimulations at 100 Hz in the DLM muscle compared to control animals (rescued with a wild-type amx [amx Δ + amx] or N'-3xHA tagged amx [amx Δ + 3xHA::amx]; Fig 6F). Additionally, amx Δ animals show only slight, early stimulation failure rates in the TTM at 50 and 100 Hz at 5 days (Figs 6E and S12D). At 15 days post-eclosion, amx Δ animals still only show mild failure rate in the TTM at 100 Hz (S13E Fig) but begin to show a significant increase in failure rate compared to control animal recordings from the DLM at 50 and 100 Hz (S13D and S13F Fig). At 25 days post-eclosion, the defects in the DLM increase in severity for 50 and 100 Hz compared to (black) have a response similar to amx Δ + amx controls (orange) while DLM muscles have small but significant decrease in response probability. 3xHA::amx (blue) flies also perform as well as controls. (G-H) TTM and DLM response in 25d old amx Δ mutants is significantly reduced. DLM response of 25d old amx Δ + human TM2D3 (green) flies is reduced compared to controls (I). Multiple unpaired t-tests with Holm-Šídák correction for multiple comparisons. � = p<0.05. �� = p�0.01. ��� = p<0.001, ���� = p<0.0001. Error bars show SEM. Additional data can be found in S12-S14 Figs.
https://doi.org/10.1371/journal.pgen.1009962.g006 rescued controls (Figs 6H and S14D). At this time point, the TTM also begins to show a significant failure rate in amx Δ animals at 100Hz compared to rescued animals but not at lower frequencies (Figs 6G, S14A and S14C). Interestingly, unlike in the lifespan assay in which the human TM2D3 did not dramatically suppress the amx Δ mutant phenotype (Fig 6A) and in the fertility/neurogenesis assay in which the human TM2D3 showed~50% activity of the fly Amx protein (Fig 1C), we found that human TM2D3 is able to significantly suppress the electrophysiological defects close to the level of fly Amx (Figs 6E-6H and S12-S14). In summary, loss of amx causes an age-dependent decline in neuronal function, and this defect can be fully rescued by both fly and human TM2D3, indicating an evolutionarily conserved role of this gene in healthy aging.

Discussion
In this study, we functionally characterized TM2D genes through gene knockout and overexpression strategies in Drosophila melanogaster to gain biological knowledge on this understudied but evolutionarily conserved gene family that has been implicated in AD. We first showed that the knockout allele of amx (Drosophila homolog of TM2D3) generated by CRISPR is phenotypically indistingusiable from the classic amx 1 allele and displays female sterility and a maternal-effect neurogenic defect. Recently, we reported that this allele also shows a maternal-effect inductive signaling defect to specify the mesoectoderm during embryogenesis which is another Notch-dependent event [59], demonstrating that amx is maternally required for multiple Notch signaling dependent processes during embryogenesis. In addition, we generated the first knockout alleles of amrt (Drosophila ortholog of TM2D2) and bisc (Drosophila ortholog of TM2D1) and documented that each null allele phenotypically mimics the loss of amx. Furthermore, we revealed that the triple knockout of all three TM2D genes in Drosophila show identical maternal-effect neurogenic phenotypes without exhibiting other obvious Notch signaling-related developmental defects. Moreover, although the over-expression of the fulllength Amx do not cause any scorable defects, we serendipitously found that expression of a truncated form of Amx that lacks the majority of the extracellular domain can strongly inhibit Notch signaling in the developing wing imaginal disc, a tissue in which all three fly TM2D genes are expressed endogenously. Through genetic epistatic experiments using newly generated UAS-Notch transgenic lines, we mapped this inhibitory effect to the γ-secretase cleavage step of Notch activation. Subsequently, we found that amx null animals have a shortened lifespan, a phenotype that can be rescued by reintroduction of Amx in neurons. This shortening of lifespan phenotype was also seen in amrt and bisc null flies, suggesting that these three genes may also function together in the aging brain. Finally, through assement of climibing behavior and electrophysiological recordings of the giant fiber system, we showed that amx null flies show age-dependent decline in neural function. In summary, we demonstrate that all three TM2D genes play critical roles in embryonic Notch signaling to inhibit the epithelial-to-neuron cell fate transformation as maternal-effect genes, and that amx is required for neuronal maintenance in the adult nervous system, a function that may be related to the role of human TM2D3 in AD.
Within commonly used genetic model organisms, TM2D genes are found in multicellular animals (both in invertebrates and in vertebrates) but are absent in yeasts (e.g. Saccharomyces cerevisiae, Schizosaccharomyces pombe) and plants (e.g. Arabidopsis thaliana), suggesting that this family of genes arose early in the metazoan lineage. In humans and flies, there are three TM2 domain-containing genes (TM2D1, TM2D2, TM2D3 in Homo sapiens; bisc, amrt, amx in Drosophila melanogaster, respectively), each corresponding to a single gene in the other species. Interestingly, this 1:1 ortholog relationship is also seen in mouse (Tm2d1, Tm2d2, Tm2d3), frog (Xenopus tropicalis: tm2d1, tmd2d, tm2d3) zebrafish (Danio rerio: tm2d1, tm2d2, tm2d3) and worm (Caenorhabditis elegans: Y66D12A.21, C02F5.13, C41D11.9) (S1 Fig). In general, most genes have more paralogous genes in humans compared to flies (for example, one Drosophila Notch gene corresponding to four NOTCH genes in human) as vertebrates underwent two rounds of whole-genome duplication (WGD) events during evolution [60]. Furthermore, teleosts including zebrafish underwent an extra round of WGD [61], leading to formation of extra duplicates in 25% of all genes (e.g. one NOTCH1 gene in human corresponds to notch1a and notch1b in zebrafish). Hence it is interesting that each of the three TM2D genes remained as single copy genes in various species despite whole genome level evolutional changes, suggesting that there may have been some selective pressure to keep the dosage of these genes consistent and balanced during evolution.
Although the in vivo functions of TM2D1 and TM2D2 have not been studied extensively in any organism, several lines of studies performed in cultured cells suggest that these genes may also play a role in AD pathogenesis. Through an yeast-two hybrid screen to identify proteins that bind to Aβ42, Kajkowski et al. identified TM2D1 and referred to this protein as BBP (beta-amyloid binding protein) in their study [24]. They further showed that TM2D1 can also interact with Aβ40, a non-amyloidogenic form of Aβ, and mentioned that they have preliminary data that it also binds to APP (Kajkowski et al., 2001). The interaction between Aβ peptides and TM2D1 was shown to require the extracellular domain as well as a portion of the first transmembrane domain of TM2D1. Because overexpression of TM2D1 in a human neuroblastoma cell line (SH-SY5Y) increased the sensitivity of these cells to cell death caused by incubation with aggregated Aβ and since the DRF motif was found to be required for this activity, the authors of this original study proposed that TM2D1 may function as a transmembrane receptor that mediates Aβ-toxicity [24]. However, a follow-up study from another group refuted this hypothesis by providing data that TM2D1 is not coupled to G proteins using a heterologous expression system in Xenopus oocytes [62]. Therefore, although the physical link between TM2D1 and Aβ42 is intriguing, the significance of this interaction and its role in AD pathogenesis has been obscure.
To our surprise, loss of bisc/TM2D1 and amrt/TM2D2 were phenotypically indistinguishable from the loss of amx/TM2D3 in Drosophila. The zygotic loss of each gene did not exhibit any strong developmental defects into adulthood, despite their relatively ubiquitous expression pattern according to large transcriptome datasets [37,38]. Furthermore, the loss of either amrt or bisc caused a reduction of lifespan, again mimicking the loss of amx. More interestingly, the triple null mutants are viable and do not exhibit any morphological defect, suggesting that these genes are not required zygotically during development. In contrast, maternal loss of any single TM2D gene causes a strong embryonic neurogenic defect, which is also seen in embryos laid by triple knockout animals. Neurogenic defect is a classical phenotype in Drosophila that was originally reported in the mid-1930s [63,64], and the study of mutants that show this phenotype led to the establishment of the core Notch signaling pathway in the late 1980's and early 1990's [7,65]. Although the study of neurogenic phenotypes and genes has a long history, this phenotype is a very rare defect that has so far been associated with only 19 genes according to FlyBase [66], prior to this work. Seven genes show this defect as zygotic mutants [aqz, bib, Dl, E(spl)m8-HLH, mam, N and neur], seven genes are zygotically-required essential genes with large maternal contributions (hence the need to generate maternal-zygotic mutants by generating germline clones to reveal the embryonic neurogenic defect) [Gmd, Gmer, gro, Nct, O-fut1, Psn and Su(H)], one gene has only been investigated by RNAi (Par-1) and four genes including amx are non-essential genes that show maternal-effect neurogenic defects (amx, brn, egh, pcx). Hence, our study has revealed two new genes that are evolutionarily closely linked to amx in this Notch signaling related process.
The similarity of sequences and phenotypes caused by loss of amx, amrt and bisc suggests that the proteins encoded by these genes may function together. The lack of additive or synergistic phenotypes in the double and triple null mutant flies also support this idea. Interestingly, high-throughput proteomics experiments based on co-immunoprecipitation mass spectrometry (co-IP/MS) from human cells have detected physical interactions between TM2D1-TM2D3 [67] and TM2D2-TM2D3 [68], suggesting these proteins may form a protein complex. Further biochemical studies will be required to clarify the functional relationship between the three TM2D proteins. Two additional mammalian datasets further support our hypothesis that these three proteins functions together. First, all three TM2D genes were identified through a large scale cell-based CRISPR-based screen to identify novel regulators of phagocytosis [69]. Individual knockout of TM2D genes in a myeloid cell line was sufficient to cause a similar phagocytic defect based on the parameters the authors screened for (e.g. substrate size, materials to be engulfed). Although the authors of this study did not generate double or triple knockout cell lines to determine whether there were any additive or synergistic effects when multiple TM2D genes were knocked out, this suggests that these three genes may function together in phagocytosis. The authors further note that because these genes are broadly expressed in diverse cell types beyond phagocytic cells in the nervous system [70], they may play other roles in the brain, consisnet with our finding that amx is required neuronally to maintain a normal lifespan. It would be interesting to explore whether age-dependent phenotypes seen in TM2D null mutants can also be suppressed through glial specific rescue experiments to understand the in vivo significance of this group's in vitro findings. Second, preliminary phenotypic data from the International Mouse Phenotyping Consortium [71] indicates that single knockout of mice of Tm2d1 (https://www.mousephenotype.org/data/genes/MGI:2137022), Tm2d2 (https://www.mousephenotype.org/data/genes/MGI:1916992) and Tm2d3 (https://www. mousephenotype.org/data/genes/MGI:1915884) are all recessive embryonic lethal prior to E18.5. Although detailed characterization of these mice will be required and further generation of a triple knockout line is desired, the shared embryonic lethality may indicate that these three genes potentially function together in an essential developmental paradigm during embryogenesis in mice.
Our attempts to unravel the function of Amx through overexpression of the full-length protein was uninformative since this manipulation did not cause any scorable phenotype. However, we serendipitously found that a truncated form of Amx, which only contains the most conserved region of TM2D family proteins, has the capacity to strongly inhibit Notch signaling during wing and notum development. These results were surprising because we did not see any wing or bristle defects in the triple TM2D gene family knockout flies, even though all three genes are endogenously expressed in the wing imaginal disc. Through epistasis experiments using a set of new UAS-Notch transgenic lines, we found Amx ΔECD to inhibit Notch signaling at the S3 cleavage step which is mediated by the γ-secretase complex. This data is consistent with earlier epistasis experiments performed on amx 1 in the context of embryonic neurogenesis [25], further supporting the idea that amx may regulate γ-secretase function in vivo. We further determined that over-expression of Amx ΔECD in the wing imaginal disc causes an accumulation of Notch protein within the cell and at the cell membrane. This phenotype is similar to what is seen upon knockdown of Psn in the wing imaginal disc, which is consistent with earlier findings showing Notch accumulation at the cell membrane in neuroblasts of Psn mutants [72]. In summary, we showed that ectopic over-expression of a portion of Amx that is conserved among all TM2D proteins causes a strong Notch signaling defect by disrupting a process that realtes to γ-secretase, providing additional links between Amx/TM2D3 and AD.
By aging the amx null male flies that are visibly indistinguishable from the control flies (amx null flies with genomic rescue constructs), we found that loss of amx causes a significant decrease in lifespan, a phenotype that we observed in amrt and bisc null flies. By generating a functional genomic rescue transgene in which Amx is tagged with an epitope tag, we found that this protein is expressed in the adult brain. The short lifespan defect of amx null flies was effectively rescued by reintroduction of Amx in neurons, demostrating the importance of this gene in this cell type. In addition to the lifespan defect, amx null flies also exhibited an agedependent climing defect, suggesting that their neural function declines with age. By further performing electrophysiological recordings of the giant fiber system, which is a model circuit that is frequently used in neurological and neurodegenerative research in Drosophila [56,[73][74][75], we found that there is indeed an age-dependent decline in the integrity of this circuit. Through this assay, we observed that the DLM branch of the giant fiber system begins to show failures earlier than the TTM branch. The DLM is activated by giant fiber neurons that chemically synapse onto PSI (peripherally synapsing interneuron) neurons through cholinergic synapses, which in turn chemically synapse onto motor neurons (DLMn which are glutamatergic) through cholinergic connections. The TTM, in contrast, is activated by giant fiber neurons that electrically synapse onto motor neurons (TTMn which are glutamatergic) through gap junctions, causing a more rapid response. Considering the difference in the sensitivity of the two branches, cholinergic neurons/synapses may be more sensitive to the loss of amx, a neuronal/synaptic subtype that is severely affected in AD in an age-dependent manner [76].
How does amx maintain neuronal function in aged animals and is this molecular function related to AD? One potential molecular mechanism is through the regulation of γ-secretase in the adult brain. By knocking down subunits of the γ-secretase complex, Psn and Nct, specifically in adult neurons, Kang et al. showed that reduction of γ-secretase function decreases lifespan, which was associated with climbing defects as well as histological signs of neurodegeneration [48]. The requirement of γ-secretase components in neuronal integrity has also been reported in mice [77][78][79][80][81], suggesting this is an evolutionarily conserved phenomenon. Interestingly, the role of the γ-secretase complex in neuronal maintenance is unlikely to be due to defects in Notch signaling because neurodegeneration has not been observed upon conditional removal of Notch activity in post-developmental brains in flies and in mice [82]. While the precise function of γ-secretase in neuronal maintenance is still unknown, several possibilities including its role in regulating mitochondrial morphology [80] and calcium homeostasis [83,84] has been proposed based on studies in C. elegans and mice. Investigating whether Amx does indeed regulate γ-secretase in adult neurons and whether it impacts the aforementioned processes will likely facilitate our understanding on how this gene regulates neuronal health. Furthermore, considering that TM2D3 and other TM2D genes have been proposed to function in phagocytic cells, and because phagocytosis process plays many roles beyond engulfment of toxic Aβ molecules in the nervous system [85], Amx may also be playing a role in engulfing unwanted materials that are harmful for the adult brain. For example, loss of the phagocytic receptor Draper in glia cells causes age-dependent neurodegeneration that is accompanied by accumulation of non-engulfed apoptotic neurons throughout the fly brain [86]. Interestingly, a recent study has shown that over-expression of phagocytic receptors can also promote neurodegeneration [87], indicating the level of phagocytic activity needs to be tightly controlled in vivo. Further studies of amx Δ mutants (as well as amx Δ amrt Δ ; bisc Δ triple mutants) in the context of phagocytosis will likely reveal the precise molecular function of Amx and other TM2D proteins in this process.
Finally, could there be any molecular link between the role of TM2D genes in Notch signaling (proposed based on experiments in Drosophila) and phagocytosis (revealed based on mammalian cell culture based studies), or are they two independent molecular functions of the same proteins? All TM2D proteins have two transmembrane domains connected by a short intracellular loop, making them an integral membrane protein. By tagging the amx genomic rescue construct with a 3xHA tag that does not influence the function of Amx, we observed that 3xHA::Amx is localized to the plasma membrane as well as intracellular puncta, which likely reflects intracellular vesicles. Interestingly in embryos laid by amx Δ mutant females, we observed a mild and transient but significant alteration in Notch distribution during early embryogenesis [59]. Moreover, we observed a strong accumulation of Notch when we overexpressed Amx ΔECD in the developing wing primordium. These data indicate that Amx may affect protein trafficking, which in turn may impact the processing of Notch by the γ-secretase complex. Indeed, Notch signaling is highly regulated by vesicle trafficking and alterations in exocytosis, endocytosis, recycling and degradation all impact the signaling outcome [88,89]. In fact, multiple studies have proposed that γ-secretase cleavage occurs most effectively in acidified endocytic vesicles [90,91]. Hence, while amx may be specifically required for the proper assembly or function of the γ-secretase complex, it may alternatively be necessary to bring Notch and other substrates to the proper subcellular location for proteolytic cleavages to occur efficiently. The subcellular localization differences observed between the punctate Amx ΔECD , which causes a dramatic Notch signaling defect accompanied by mistrafficking of Notch, and the membranous Amx FL , which does not have this effect, may further support the idea that Amx function as a trafficking module. Similar to Notch signaling, phagocytosis also requires coordination of many cellular trafficking events to expand the plasma membrane to form a phagophore, internalize the particle of interest to generate a phagosome, and fuse the phagosome to lysosomes to degrade its content [92]. By studying the role of TM2D genes and proteins in embryonic Notch signaling, phagocytosis and age-dependent neuronal maintenance, we will likely understand the precise molecular and cellular function of this evolutionarily conserved understudied protein family, which will likely lead to further understanding of molecular pathogenesis of AD and other human diseases. Considering the phenotypic similarities of amrt and bisc to amx in Drosophila embryonic neurogenesis, the similarities between TM2D1-3 in human cells in the context of phagocytosis, and the similarities of Tm2d1-3 knockout mice in the context of embryogenesis, we propose that studies of rare genetic variants, epigenetic regulators or proteomic changes in other TM2D genes may reveal novel risk factors or biomarkers in epidemiologic study of AD and other forms of dementia.

Drosophila strains and fly husbandry
Drosophila melanogaster stocks used in this study are listed in S1 Table. Some strains were generated in house for this study (see below), and others were obtained from Bloomington Drosophila Stock Center (BDSC) and other sources. Flies were kept on standard media and maintained at room temperature (21-23˚C). Crosses were performed at 25˚C in an incubator unless otherwise stated. Genotypes of the flies shown in each figure panel are listed in S2 Table.

Generation of amx overexpression transgenes
To generate the untagged UAS-amx transgene, we PCR amplified the open reading frame (ORF) of amx derived from the cDNA clone GH02974 (Genbank ID: AF181623, Drosophila Genomics Resource Center, Indiana University) with NotI and XbaI restriction sites on the 5' and 3' end, respectively, and subcloned this fragment into pUASTattB [93]. The cDNA templates used to generate the UAS constructs to overexpress 3xHA epitope tagged full-length (UAS-3xHA:: amx FL ) and truncated (UAS-3xHA::amx ΔECD ) Amx were synthesized by Genewiz. The amx ORF used for these constructs is identical to the GH02974 clone, which is also identical to the coding sequence of the genomic rescue construct we reported in [13]. A 3xHA epitope sequence (N'-YPYDVPDYAGYPYDVPDYAGSYPYDVPDYA-C') was inserted after the signal sequence (SS) as predicted by SignalP 4.1 (http://www.cbs.dtu.dk/services/SignalP-4.1) and Phobius (http:// phobius.sbc.su.se). A 968 bp fragment corresponding to 3xHA tagged full-length Amx was synthesized with NotI and XbaI restriction sites on the 5' and 3' end respectively. Furthermore, a 464 bp fragment corresponding to 3xHA::Amx ΔECD was designed based on the full-length 3xHA::Amx construct, but with the majority of the endogenous Amx N'-end removed leaving the SS, epitope tag, and TM2 domains intact. We subcloned the synthesized fragments into pUASTattB [93] via restriction enzyme cloning utilizing the NotI and XbaI cut sites. We then injected the final plasmids into VK37 (PBac{y[+]-attP}VK00037) attP docking site [43] and injected animals were crossed to y w and screened for the presence of the white + marker encoded in the pUASTattB backbone in a whitebackground. Positive animals were then crossed to SM6a to create a balanced stock. One line for each transgenic construct was chosen for this study.

Generation of a N'-tagged amx genomic rescue transgene
The 3xHA::amx genomic rescue transgene was designed and constructed based on the nontagged genomic rescue construct we reported in [13]. A~3.3kb region including the full amx gene has the ability to fully rescue the maternal effect neurogenic phenotype of amx 1 [13] and amx Δ (this study). We inserted a 3xHA sequence into the pattB-amx genomic rescue construct described in [13] via NEBuilder HiFi DNA Assembly. 3xHA was inserted after the predicted signal sequence (N'-ATMRLQRQCIVVNMRSAIVLIMIFVLTGIRNSET-C') to tag Amx at its N'. pUASTattB-3xHA::amx FL described above was used as a template to amplify and add appropriate homology arms to the SS-3xHA::Amx DNA sequence with the primers 5'-CCCC GCTCTATCTGACCAAAGCCACCATGAGGCTCCAACGAC-3' and 5'-AAAACTAAACT AAGAACGGACTACTATATGTAAAGTGAGCCATCCGC-3' using Q5 High-Fidelity 2X Master Mix (M0492S, NEB). These and other Oligo DNA were synthesized by Sigma Aldrich. The section of pattB-amx plasmid containing the amx regulatory elements was linearized by PCR using Q5 polymerase and primers 5'-CGTTGGAGCCTCATGGTGGCTTTGGTCAG ATAGAGCG-3' and 5'-GCTCACTTTACATATAGTAGTCCGTTCTTAGTTTAGTTTTA CAGGGGT-3', which add appropriate homology arms to allow for assembly with the 3xHA:: Amx fragment. The construct was assembled following the protocol described by NEBuilder HiFi DNA Assembly Master Mix (NEB Catalog #E2621S) and confirmed by Sanger sequencing. The validated construct was injected into embryos expressing ϕC31 integrase with a 2 nd chromosome attP docking site (VK37) [43]. We also injected a empty pattB plasmid into the same docking site to generate a control like (pattB[w+]). Transgenic flies were isolated based on eye color (w + encoded by the mini-white gene in pattB vector) and balanced over SM6a.

Generation of an amrt genomic rescue construct
Genomic DNA was isolated from y, w, iso#6; +/+; attP2{nos-Cas9} animals using PureLink Genomic DNA Mini Kit (Cat no. K1820-01, Invitrogen). Genomic region fully containing the amrt (CG11103) locus and neighboring sequences was amplified by PCR using the primers 5'-TATATACTCGAGcgcgaaactttcgatttcc-3' and 5'-TATATAGAATTCatcgaatgtagagatgggc-3' (small letters indicate annealing region) which added EcoRI and XhoI restriction sites for follow-up cloning into the pattB vector [93]. The pattB-amrt plasmid was injected into VK37 [43] docking site on the second chromosome. Eclosed animals were crossed to y, w and screened for the w + marker in the subsequent generation. Positive animals were then crossed to SM6a to create a balanced stock.

Generation of biscotti::GFP fosmid transgenic line
A~40kb genomic fosmid construct in which bisc is C'-tagged with GFP and other epitopes [36] (bisc::GFP, FlyFos021003, Drosophila TransgeneOme Resource ID: CBGtg9060C1139D) was obtained from Source BioScience. Bacterial colonies were provided to GenetiVision Corp. for DNA preparation, injection into VK33 (PBac{y[+]-attP}VK00033) on the third chromosome [43], selection, and balancing of fly lines using the TM3, Sb balancer. Three independent lines were generated and all three behaved in a similar manner. One line was chosen for the experiments performed in this study.

Creation of Notch overexpression transgenic lines
All transgenic constructs were generated by Gateway (Thermo Fisher Scientific) cloning into the pUASg-HA.attB plasmid [93]. First, we generated Gateway compatible plasmid that contains the full-length Notch open reading frame. We subcloned the full-length Notch (N FL ) open reading frame into the pDONR223 plasmid based on a cDNA clone provided by Dr. Spyros Artavanis-Tsakonas [97], which was mediated by a Gateway reaction using BP clonase II (Thermo Fisher Scientific, #11789100) following a PCR reaction and addition of attB sites to the amplicon [98,99]. Truncated Notch constructs were generated by Q5 site-directed mutagenesis (NEB) with the following primers: N EXT : 5'-GCGGCCAAACATCAGCTG-3' and 5'-GTGCATTTTGTTAATCCAAAAACAA ATCC-3'.
All constructs were fully sequenced (Sanger), and cloned into pUASg-HA.attB via LR clonase II (Thermo Fisher Scientific, #11791020). All expression constructs were inserted into VK37 integration site [43]. Two transgenic lines were established for each and one line each was used for this study.

Embryo collection, staining and imaging
Embryo collection, staining and imaging was performed as previously described [13]. In brief, virgin females that are homozygous for each or all TM2D gene mutations, with or without genomic rescue constructs, were crossed to males flies of the same genotype or Canton-S males, and allowed to mate in a vial for 24 hrs. Flies were then transferred to a bottle with a grape juice plate supplemented with active yeast and allowed to lay eggs overnight. Embryos were then gently collected using a paint brush and their chorions were removed by 1.5 minute incubation in 66% bleach. Dechorionated embryos were then washed with water and fixed in 4% paraformaldehyde/PBS(phosphate buffered saline)/n-heptane solution for 30 minutes at room temperature. Fixed embryos were washed and stored in 100% methanol at -20˚C until use. Fixed embryos were rehydrated and rinsed with 0.03% Triton-X in PBS (PBST). A primary antibody to label neuronal nuclei [anti-Elav, rat monoclonal (7E8A10) [100], 1:200, Developmental Studies Hybridoma Bank (DSHB)] was applied in a solution of PBST/5% normal donkey serum (NDS)/0.1% NaN 3 overnight at 4˚C. Embryos were further washed with PBST upon removal of the primary antibody, and a secondary antibody (donkey anti-rat-Alexa488, 1:500; Jackson ImmunoResearch #712-545-153) was applied for 1 hour at room temperature. Stained embryos were washed in PBST and mounted onto glass slides using Vectashield with DAPI (4',6-diamidino-2-phenylindole, Vector labs). Embryos were imaged using Ti2E Spinning Disc confocal microscope (Nikon) and images analyzed using NIS software (Nikon).

Egg hatching assay
Egg hatching assay was performed as previously described [13]. Embryos were collected and dechorionated as described above. Dechorionated embryos were suspended in PBS and placed in 12-well cell culture dishes. The dishes were then placed in a 25˚C incubator for 24 hours. The ratio of hatched to unhatched embryos after the 24 hour period was recorded for each genotype and used to calculate hatching rate (%). This was repeated at least 3 times and statistical analysis and graph generation was performed using GraphPad Prism 9.0 software. We performed one-way ANOVA followed by Dunnett test or t-test. ���� = p-value �0.0001.

Imaging of adult flies
Heads and wings from adult flies were removed from the body using fine dissection scissors and imaged directly using a MZ16 microscope (Leica) with attached Microfire camera (Optronics) using ImagePro Plus 5.0 acquisition software (Media Cybernetics). Extended focus function was used to obtain deep focus images out of Z-stack images. Imaging of the dorsal thoraxes (nota) were described in [101] with slight modifications. In brief, legs, head, and abdomen were removed from thoraxes with fine dissection scissors. Then, the dissected thoraxes were then placed in 10% KOH at 95˚C. for 10 min to dissolve soft tissue. Thoraxes were then further trimmed with scissors prior to imaging and mounted on a glass slide with spacers using 75% glycerol/25% ethanol solution. Photos were taken using microscope system and imaging software as above.

Western blot of HA tagged Amx
To determine whether 3xHA::Amx is expressed in adult brains or ovaries, we performed western blot. Adult brains from y, w; VK37{pattB-3xHA::amx} flies were dissected out as described in [103]. For ovaries, we mated the y, w; VK37{pattB-3xHA::amx} females flies to male flies of the same genotype while supplying plenty of active yeast to stimulate oogenesis for two days prior to protein isolation. Ovaries were then dissected out from the abdomen in cold (4˚C) PBS. Dissected brains and ovaries were rinsed with cold PBS and placed immediately in cold 8M urea lysis buffer (8M urea, 10% glycerol, 0.5% SDS, 5% β-mercaptoethanol) with Halt Protease Inhibitor Cocktail 100X (Thermo Scientific, #78430) added before lysis. We homogenized the brains or ovaries were via pestle in 15 uL of 8M urea lysis buffer. Homogenate was incubated for 30 minutes on ice, and 2x Laemmli Sample Buffer (Bio-Rad, #1610737) was added prior to gel loading. Best results were obtained when avoiding heating/boiling protein sample. Homogenate was loaded directly into Mini-PROTEAN TGX 4-20% Precast Gels (Bio-Rad, #4561094). SDS-PAGE (Sodium Dodecyl Sulphate-PolyAcrylamide Gel Electrophoresis) was run for 30 minutes at 120V in Tris/Glycine/SDS buffer (BioRad, #1610732). Protein was then transferred onto PVDF membrane using Bio-Rad TransBlot Turbo system using the low molecular weight protocol. The membrane was blocked with 5% skim milk in 0.5% Tween-20/Tris-Buffered Saline (TBST) for 1 hour at room temperature and then washed 3 times for 5 minutes with TBST. A primary antibody (rat anti-HA, 1:1,000, Sigma, 3F10) was diluted in 5% fetal bovine serum/TBST and the membrane was incubated overnight at 4˚C. The membrane was again washed with TBST and a secondary antibody (donkey anti-rat HRP, 1:5,000, Jackson ImmunoResearch, #712-035-150) was applied in 5% milk/TBST for 2 hours at room temperature. The membrane was washed again and imaged using SuperSignal West Femto Maximum Sensitivity Substrate (Thermo Scientific, 34096) and ChemiDoc imaging system (Bio-Rad) using default settings. The same protocol was used to compare the levels of overexpressed 3xHA::Amx FL or 3xHA::Amx ΔECD from adult brains dissected from elav-GAL4/ +; UAS-3xHA::Amx FL or ΔECD animals.

Longevity assay
To determine whether loss of amx, amrt or bisc causes lifespan defects, we compared the longevity of male flies that lack each gene (amx Δ , amrt Δ or bisc Δ ) to respective control flies in which the null allele has been rescued with genomic constructs (VK37{amx}, VK37{amrt} or VK33{bisc::GFP}). For amx, we also tested whether N'-3xHA tagged Amx (y, w, amx Δ ; VK37 {3xHA::amx}) or human TM2D3 expressed under the control of fly amx regulatory elements (y, w, amx Δ ; VK37 {TM2D3}) can also rescue the mutant phenotype. Flies were reared and collected as described in [104]. Ten animals were housed together in a single vial and flies were flipped to a new vial with fresh food every 2-3 days. Vials were kept in a 25˚C incubator with a 12 hr light/dark cycle. Dead flies were recorded after every vial flip. Generation of the Kaplan-Meier curve and statistical analysis was performed with GraphPad Prism 9.0. We applied logrank test (Mantel-Cox), ���� = p�0.0001.

Climbing assay
8-10 flies that were age matched were placed in an empty vial and flies were tapped down to the bottom. The number of flies that successfully climbed past a 7 cm line from the bottom of the vial within 15 seconds were recorded. This was repeated 3 times for each group of flies. The graph was generated using GraphPad Prism 9.0.

Electrophysiological recordings of the giant fiber system
Electrophysiological recordings of the giant fiber system were performed with a protocol modified from [105]. Flies were first anesthetized on ice then transferred to a petri dish filled with soft dental wax; wings and legs were mounted in wax, ventral side down, using forceps. Five electrolytically sharpened tungsten electrodes were used: two for stimulating the giant fiber, one as a reference electrode, and two for recording from the TTM and DLM, respectively. To activate the giant fiber, two sharp tungsten electrodes were inserted into each eye and voltage stimulation was applied at different frequencies ranging from 0.5 Hz to 100 Hz. DLM and TTM responses were measured through the two electrodes implanted in the DLM and TTM.
For each adult fly, prior to applying high frequency stimulation on the giant fiber, low frequency stimulations at 0.5 Hz were applied after placing the two recording electrodes in TTM and DLM to ensure that the electrodes are recording from the proper muscles (the latency of responses for TTM: 0.8 ms and for DLM: 1.2 ms [105]. For the actual experiments, high frequency train stimulations of 20 pulses were delivered to the giant fiber at 20, 50 and 100 Hz in random order. Ten times repetitive stimulations were applied for each particular frequency train, interspersed with few minutes rests between two trains of stimuli (for 20 Hz, 1 minute resting; 50 Hz 2 minutes and 100 Hz 3 minutes). 0.5 Hz stimulations were used again after high frequency stimulation to confirm that electrodes were still in the proper muscle. The aforementioned process was considered as one biological sample. Stimuli of the crossing electrodes were fixed at a duration of 10 microseconds at 10-13 V of amplitudes through a stimulus isolation unit (Digitimer Ltd, model DS2A) and the frequency of train stimuli was controlled by LabChart Pro-8 acquisition software (ADInstruments). A microelectrode amplifier (A-M system, Model 1800) was used for all recordings. PowerLab 4/35 (ADInstruments) was used for data acquisition. The probability of responses for one biological sample, under particular frequency of giant fiber stimulation, due to a particular stimulus, was calculated from the proportion of successful responses (out of 10) for both TTM and DLM pathways. The difference of 'probability of responses' between control and experimental samples (pvalue) for each stimuli were calculated by multiple unpaired t-tests with Holm-Šídák correction for multiple comparisons using Graph Pad Prism 9.0. n.s. = not significant. � = p�0.05. �� = p�0.01. ��� = p�0.001. ���� = p�0.0001. and Amx ΔECD expression driven by pan-neuronal elav-GAL4 driver. Expected band size for Amx FL is 35 kDA. Expected band size for Amx ΔECD is 10 kDa. Since two bands are present around the expected size for Amx ΔECD , both signals were quantified in (B). (B) Amx ΔECD expression is~10x lower than Amx FL . t-test. ���� = p<0.0001. Error bars show SEM. (C-D) Immunostaining against HA tag (green) shows 3xHA::Amx FL has a membranous expression pattern (C) compared to a more punctate pattern seen with 3xHA::Amx ΔECD (D). (E) RT-PCR on mRNA extracted from wing discs from control animals (genotype: y w) shows that all three TM2D genes are endogenously expressed in this tissue. Scale bars = 5 μm in (C-D). (TIFF) S8 Fig. amrt and bisc mutants show a shortened lifespan phenotype while amx mutants exhibit age-dependent climbing defects. (A-B) Lifespans of amrt and bisc mutants compared to their respective controls. Animals were reared at 25˚C; Log-rank test (Mantel-Cox), ���� = p<0.0001. (A) amrt Δ animals (dark yellow, n = 147) have reduced lifespan compared to amrt Δ + amrt controls (pale dash yellow, n = 59). (B) bisc Δ animals (dark pink, n = 94) have reduced lifespan compared to bisc Δ + bisc::GFP controls (pale dash pink, n = 84). (C) amx Δ mutants (on a y w mutant background) exhibit climbing defects that worsen over time compared to controls (y w). The absence or presence of the white gene does not affect this phenotype.