The detailed analysis of the two comprehensively curated HD datasets, PerturbDB (1,218 genes) and the HTT Interactome (1,619 genes), is presented in Supplementary Material (Section 1 in Supplementary Datasheet 4), and we only summarize the main findings here. The validity of the causal relationship to mHTT pathology of the genes in PerturbDB, with their disparate cellular context and mHTT-driven endpoints, has long been debated. However, the major PerturbDB experimental platforms, in particular those of Drosophila in vivo phenotypes, mammalian mHTT aggregation, and cell culture toxicity, highly overlap each other despite their very different biological context (Supplementary Section 1.2; Table S2A in Supplementary Datasheet 4). Thus, there are likely common mechanisms mediating mHTT-induced dysfunctions despite the disparate cellular context and endpoints. Moreover, PerturbDB highly overlaps the HTT Interactome, at ~30% between the two (−log FDR ~131), and again more so for the three major experimental platforms (Tables S2A,B; Figures S3A,B in Supplementary Datasheet 4). As the HTT Interactome by and large represents normal functions of HTT (only a relatively small number of HTT interactors are specific for either mHTT or HTT), the high number of HTT interactors that can perturb mHTT endpoints suggests that mHTT can both drive gain-of-function pathologies and disrupt normal biology of HTT. The comprehensive nature of the HTT Interactome and its origin from multiple sources also make it more likely to fill biological gaps not studied in the more hypotheses-driven PerturbDB datasets. Thus, we conclude that these two HD datasets are suitable as foundational data with which to apply our database-enabled methodology and workflow to formulate hypotheses on mHTT pathogenic mechanisms, as shown in Figure 1.

As described in Supplementary Material Section 2 in Supplementary Datasheet 4, we first curated a large set of genes possessing neuronal and synaptic functions (SynapseDB) and found highly significant enrichment of the HD datasets (Table S3, Figure S4 in Supplementary Datasheet 4), not surprising for a neurodegenerative disease. We then found that a small set of genes critical for synaptic vesicle functions are found in the HD datasets (Table S4 in Supplementary Datasheet 4). While examining these genes, we encountered an interesting set of studies on exosomes (or more broadly, extracellular vesicle or EV) in neuronal functions in which many of these synaptic genes are implicated (Ataman et al., 2008; Korkut et al., 2009, 2013; Fainzilber et al., 2011; Koles et al., 2012; Kerr et al., 2014). Exosomes are 50–100 nm extracellular vesicles released and taken up by all cells both for intercellular communication and for disposal of cellular material (reviewed in Pant et al., 2012; Kalani et al., 2014). Exosome biology is only beginning to be studied intensively in the nervous system, both as a novel means of intercellular communication and as a pathogenic mechanism in neurodegeneration (Chivet et al., 2013; Costanzo and Zurzolo, 2013; Schneider and Simons, 2013; Rajendran et al., 2014; Tsilioni et al., 2014; Coleman and Hill, 2015). The “transmissibility” of misfolded and aggregated forms of pathogenic proteins for multiple neurodegenerative disorders, as has been shown for prions, together with possible involvement of exosome-like secreted vesicles, have generated a great deal of interest. Thus, the detection of α-synuclein, LRRK2, Aβ and tau, prions, SOD1, and TDP-43 in exosomes suggests one means by which these pathogenic proteins could be propagated in Parkinson's (PD), Alzheimer's (AD), Creutzfeldt–Jakob disease, and ALS. On the other hand, exosomes may also ameliorate the disruption of synaptic plasticity by Aβ via sequestration on surface proteins (An et al., 2013). Thus far no direct link between exosome and mHTT has been reported, although transfer of polyglutamine aggregates between neurons via “tunneling nanotubes” (Costanzo et al., 2013), and transneuronal propagation of mHTT that depends on the vesicular fusion machinery (Pecho-Vrieseling et al., 2014), and transfer between neurons and glia (Pearce et al., 2015) have been reported. Moreover, forward genetics studies in Drosophila larval NMJ showed that wnt signaling, in a bidirectional manner, mediates synaptic structural changes and a particular form of plasticity in response to spaced intensive interval stimulation (Ataman et al., 2008; Korkut et al., 2009, 2013; Fainzilber et al., 2011; Koles et al., 2012; Kerr et al., 2014). This NMJ synaptic plasticity requires coordination between presynaptic motor neuron and postsynaptic muscle cell in both anterograde and retrograde directions, whereby alteration in presynaptic neurotransmitter release is tightly coupled to the number of postsynaptic neurotransmitter receptors. This coordination is in turn mediated by release of exosomes bearing wnt binding partners from the motor neuron terminal and their uptake by the postsynaptic muscle fiber. Thus, in addition to the well-characterized inter-neuronal and synaptic communication systems, exosomes provide yet another level of control on a form of Drosophila NMJ synaptic plasticity. The importance of exosomes in well-characterized forms of synaptic plasticity and whether they contribute to synaptic deficits in neurodegeneration, however, is unclear.

The Drosophila studies identified eight genes critical for the release and two genes for the uptake of exosomes at the NMJ. We mapped these 10 genes to 13 human orthologs, adding family members for ADP-ribosylation factor, vacuolar proton pump ATPases, and dynamin due to the presence of these related mammalian orthologs in HD datasets. These genes are shown in Table 1, together with 14-3-3 proteins and the ESCRT complexes involved in the life cycle of exosomes (Colombo et al., 2013; Fraser et al., 2013), and their relationships to the HD datasets. Strikingly, all the mammalian orthologs of the Drosophila genes are also found in PerturbDB and/or the HTT Interactome (Table 1, rows 1–13). These are synaptic vesicle fusion and endocytosis genes (syntaxin 1A, RAB11, dynamin, and clathrin); vacuolar proton pump ATPases for organelle acidification; ADP-ribosylation factor; Myosin V; 14-3-3 proteins required for the release of exosomes containing LRRK2, a pathogenic PD gene (Fraser et al., 2013); and the ESCRT complexes required for endosomal trafficking through the multi-vesicular bodies (MVBs). Indeed, ALL 27 genes for exosome function in the neuronal context thus far have also been independently implicated as modifiers of mHTT-driven endpoints and/or as part of the HTT Interactome, with 21 out of the 27 in the former, 19 in the latter, and 13 found in both (Table 1). This strong concordance indicates a close relationship between exosome and HTT biology and mHTT pathology. Interestingly, the overlap with PerturbDB is almost exclusively in Drosophila in vivo phenotypes and the protein aggregation platforms (Table 1), two of three that stand out as most highly intersecting with other PerturbDB platforms and with the HTT Interactome. The 11 Drosophila genes are in synaptic vesicle biology or are members of the 14-3-3 and ESCRT families, and their concordance in larval NMJ exosome function and pan-neuronal mHTT phenotypes in adult brain implicates exosome biology in the complex behavioral phenotypes of mHTT. The overlap with the aggregation platforms, on the other hand, implicates protein metabolism and processing in exosome biology in the context of mHTT, perhaps due to the dual roles of exosome in protein cargo delivery/disposal and in intercellular communication.

To perform robust enrichment analyses and to widen the gene sets to test for specific mechanisms involving exosomes, we curated and combined PPIs for each of the mammalian exosome functional genes to produce a protein network important for exosome release and uptake. This Exosome Synaptic PPI set of 892 genes would be analogous to the HTT Interactome and therefore suitable for enrichment analysis (see Supplementary Datasheet 2 for the gene sets). The 14-3-3 genes and ESCRT complex members were not included as they form very large protein interaction networks on their own and are probably best analyzed separately (PPIs of these two families are nevertheless included in Supplementary Datasheet 2). We summarize here the main findings for the two HD datasets, while details are found in Supplementary Material (Section 3 in Supplementary Datasheet S4).

The Exosome Synaptic PPI is highly enriched in the HTT Interactome and PerturbDB, overlapping ~18% of each of them, and even more so the HD Common set of genes at 33% with high –log FDR values (Table 2). Thus, genes found in both HD datasets shows a higher relative percentage of overlap compared to either HD dataset alone. That the protein network of genes critical for exosome synaptic functions highly intersects the HTT Interactome further supports a link between exosome and HTT biology. The overlap with PerturbDB (the Drosophila in vivo phenotypes and mammalian mHTT aggregation are again the two highest overlapping, Table S5 in Supplementary Datasheet 4) in turn suggests that certain mHTT dysfunctions may involve exosome biology. Thus, these results strengthen and extend the hypothesis that synaptic and neuronal functions of exosomes are linked to HTT and mHTT. This hypothesis can be experimentally verified in the Drosophila NMJ and then extended to mammalian experimental platforms, and followed by elucidating therapeutic opportunities in exosome function linked to mHTT pathophysiology.

The contents of exosomes or extracellular vesicles have been profiled in a variety of cells and tissues, and these data are aggregated in public databases (see Section Methods for sources). We curated a set of proteins (Exosome ProteinDB, with 4,019 genes) and miRNA (755 in total) detected in exosomes from a variety of tissues and cells for enrichment analysis. Almost all of the tissues and cells reported are non-neuronal, with the exception of a set of 739 proteins detected in human CSF exosomes (Chiasserini et al., 2014). A detailed analysis of the intersection of this dataset with the HD data subsets is presented in Supplementary Material Section 3 in Supplementary Datasheet S4.

Despite the relative paucity of data from brain and neuronal tissues, the Exosome ProteinDB still highly intersects the HTT Interactome (51% of the latter, −log FDR ~228; Table 2), PerturbDB (42%, −log FDR ~92) and in particular the HD Common set (62%, −log FDR ~79). Half of the Exosome Synaptic PPI set (442 genes) is found in the Exosome ProteinDB, and both highly overlap the HD datasets (see Venn diagram in Figure 2). Indeed, only 35% of the Exosome Synaptic PPI set is not intersected. Thus, many HTT associated proteins may be either exosomal components or signaling molecules, and contribute to mHTT pathophysiology by altering exosome functions. Also noteworthy is that HTT itself is detected as protein in exosomes or extracellular vesicles from human B cells (Meckes et al., 2013), thymic tissue (Skogberg et al., 2013) and urine (Wang et al., 2012; Fraser et al., 2013), and as mRNA in exosomes from cancer cells (Skog et al., 2008; Hong et al., 2009), raising the interesting question as to whether mHTT might also be detected in exosomes in HD patients. If true, then HD joins the other neurodegenerative diseases where transmission of misfolded pathogenic proteins between cells may play an important role in disease pathology, and that the exosome content in HD patients should be studied as potential biomarkers for disease state and treatment status.

miRNA and their target genes have been implicated in neuronal functions and in neurodegenerative diseases (Higa et al., 2014). We curated two studies of differentially expressed miRNAs in HD: in striatum of two mouse HD models (Lee et al., 2011); and in human HD patient brain frontal cortex and caudate (Martí et al., 2010; Supplementary Datasheet 3) and found that 50–60% of differentially expressed miRNAs in HD are also found in exosomes (Table 3). Other miRNAs linked to HD, such as miR-22, a “perturbation” molecule due to its neuroprotective properties in an HD cell culture platform (Jovicic et al., 2013), and miR-214, miR-150, miR-146a, and miR-125b that target the Htt gene (Sinha et al., 2011), are all found in exosomes. Moreover, 7 of 11 miRNAs dysregulated in the frontal cortex of HD monkeys are found in exosomes (Supplementary Datasheet 3), including the miRNA-128a that targets HTT and HIP1 and is also downregulated in the brains of pre- and post-symptomatic HD patients (Kocerha et al., 2014). Finally, miR-26 in astrocyte-derived exosomes is involved in CNS diseases and synaptic plasticity (Lafourcade et al., 2016), and is also differentially expressed in HD. Conversely, neuronally released exosomes can regulate expression of the glutamate transporter GLT1 in astrocytes (Morel et al., 2013), and GLT1 is downregulated in HD, leading to excess glutamate and deficient ascorbate in the synapses that may contribute to HD (Miller et al., 2008, 2012; Huang et al., 2010). The intriguing roles of exosomes in glial-neuronal interaction and their potential involvement in HD are further discussed in Supplementary Material Section 4 in Supplementary Datasheet 4. Thus, both the proteins and miRNAs in exosomes are linked to HD and further support the importance of exosome biology for HD.

In conclusion, the putative roles of HTT as a scaffolding protein, its high intersection with synaptic proteins, its role in regulating vesicular organelle movement, trafficking, and transport (Zala et al., 2008, 2013; Roux et al., 2012; Liot et al., 2013; Fu and Holzbaur, 2014), and it being part of the presynaptic cytomatrix (Yao et al., 2014), are all possibly linked to various aspects of exosome biology. The impact of mHTT could occur anywhere along the life cycle of an exosome, from its biogenesis during endosome formation and maturation, processing and transport through the MVB, release from nerve terminals and other regions of neurons and likely other cell types, and ultimately uptake by postsynaptic neurons or by glia cells. The impact of mHTT could also be cell and tissue specific, depending on the roles of exosomes in those contexts. Many questions of exosome synaptic biology are addressable in the Drosophila NMJ platform, and these will drive the critical experiments to translate and validate in mammalian neuronal culture and other platforms, where exosome biology is less well-studied. The abundance of potential drug targets involved in exosome biology and the ability of exosomes to carry disease biomarkers also provide a promising path forward for HD drug discovery. A key question, however, is which functions mediated by exosomes are impacted by mHTT and contributes to disease.

The critical role of exosome in a form of synaptic plasticity at the Drosophila NMJ suggests a broader link to neuronal plasticity, but this has been little studied. Whether and how the impact of mHTT on synaptic activity drive disease pathology is also unclear, as many studies HD models have reported variable alterations in neuronal and synaptic activity in different experimental platforms, although there is a focus on the corticostriatal connections that are most vulnerable in HD (reviewed in Raymond et al., 2011; Sepers and Raymond, 2014). The effects of mHTT, however, are likely countered by compensatory mechanisms until clinical manifestation due to the slow and progressive nature of HD. In this context, a key synaptic regulatory mechanism, homeostatic synaptic plasticity (HmSP), is particularly intriguing but has been little studied in HD. Homeostasis is a fundamental feature of biological systems, and HmSP is a form of “rheostat” control allowing the nervous system to regulate repeated perturbations such as episodes of synaptic plasticity within an optimal biological range (Lee et al., 2014). Distinguishing features of HmSP include the multiple levels it operates in, from synapses to cells to networks, the slow time course of its induction and maintenance, and the likelihood of a wide range of tolerated operational set points (some have argued that “allostasis” is a more appropriate term and concept Sterling, 2012). There are multiple forms of HmSP, including synaptic downscaling and upscaling, in which chronic activation or inhibition lead to compensatory decrease or increase in postsynaptic responsiveness, respectively, often mediated by availability of postsynaptic AMPA receptors (Turrigiano, 1999, 2007, 2012; Siddoway et al., 2014). Notably, mHTT impacts AMPA receptor trafficking and impairs glutamatergic neurotransmission (Mandal et al., 2011). Presynaptic HmSP is mediated by compensatory changes in transmitter release due to chronic postsynaptic inhibition or activation, and involves retrograde signaling and presynaptic transcriptional alterations (Davis, 2013). Glia also regulate synaptic upscaling via TNF-α signaling by altering the balance of synaptic AMPA and GABA receptors (Pribiag and Stellwagen, 2013, 2014). HmSP is drawing increasing interest as a fundamental mechanism critical for neuronal functions, especially other forms of synaptic plasticity. HmSP is particularly attractive as a neurodegeneration mechanism because of its highly adaptable and context-dependent nature operating over a long time frame. Thus, while HmSP may compensate for mHTT synaptic deficits (as suggested in Rocher et al., 2016), dysfunctions in HmSP itself over time may also play pathogenic roles in HD.

A growing list of genes has been shown to play critical roles in various forms of HmSP, and some are known to also play key pathogenic roles in a number of CNS disorders. For example, polyglutamine expansion in the Cav2.1 subunit CACNA1A (OMIM ^*601011) of voltage-gated calcium channels (VGCCs), the mammalian ortholog of the Drosophila gene cacophony that is required for presynaptic HmSP (reviewed in Frank, 2014), causes the autosomal dominant spinocerebellar ataxia type 6 (SCA6, OMIM 183086), one of nine polyQ diseases that include HD. Other mutations in this gene are associated with Episodic ataxia, type 2 (OMIM 108500). Retrograde signaling in HmSP engages transcription factors that include SMAD and PAX members, and SMN1 (OMIM 600354), mutation of which causes SMA, another monogenic neurodegenerative disease. These transcriptional regulators at least partly modulate expression of potassium channels of the Kv1 and Kv4 families that together with VGCCs regulate the readily releasable pool of transmitters (Lazarevic et al., 2013; Frank, 2014). Mutations in KCNA1 (Kv1.1) cause Episodic Ataxia, Type 1 (OMIM 160120), and those in KCND3 (Kv4.3) cause the very rare Spinocerebellar Ataxia 19 (SCA19; OMIM 607346). Presynaptic HmSP signaling also includes the BLOC-1 and SNARE complexes at the active zone. Dysbindin (DTNBP1, OMIM 607145), a member of the BLOC-1 complex, has been implicated in schizophrenia. Finally, two genes required for postsynaptic HmSP are implicated in CNS disorders, MECP2 (OMIM 300005) in Rett Syndrome and FMRI (OMIM 309550; Chen et al., 2014) in the Fragile X Syndrome and Fragile X-associated tremor/ataxia. MECP2 deficiency alters HTT/HAP1 dependent axonal transport of BDNF and of APP (Roux et al., 2012). Thus, mutations of genes all along the pathway for HmSP lead to multiple CNS and neurodegenerative disorders, supporting the importance of this mechanism for normal functioning of the nervous system, and points to its potential significance in HD.

We curated the HmSP literature to produce a comprehensive list of 143 genes that are critical for the proper functioning of different forms of HmSP to perform enrichment analysis (see Supplementary Datasheet 2). These 143 genes are highly interconnected between themselves with only a handful unconnected (queried in Ingenuity IPA, results not shown). This HmSP Functional gene set highly intersects the HTT Interactome and PerturbDB, with 71 genes or 50% found in one or both of the HD datasets, and the relative % overlap again higher in the HD Common set (Table 4). Many of the 49 PerturbDB genes overlapping the HmSP set are highly connected nodal signaling genes (e.g., BDNF, mTOR, CAMKII, AKT, CREB1) and are as well-known in the HD as they are in neuronal functions and cell survival, suggesting HmSP as a novel mechanism. Of the 143 HmSP genes, 72 or 50% are found in the Exosome DB (which combines the Exosome Synaptic PPI and Exosome ProteinDB), linking exosome biology to HmSP. Of these 72 genes, 45 are found in either or both HD datasets. Finally, 20 genes are common in HmSP, exosome, and HD datasets. Ninety-eight of 143 HmSP genes are also found in the HD and/or exosome datasets, with only 45 not present in these datasets (Figure 3). This further suggests that exosome biology and HmSP are each linked not only to HTT function and mHTT pathogenesis but also to each other, and where they intersect may be important for HD.

Continuing with our strategy of enrichment analysis of protein networks built on key mechanistic genes, we curated PPI sets for each of 65 known for their roles in mediating eight specific forms of HmSP (Table S6 in Supplementary Datasheet 4), then combining the PPI sets for each form of HmSP: (1) presynaptic HmSP with Eph signaling to calcium channels (VGCCs) and regulation of the readily releasable pools of transmitters; (2) retrograde signaling to presynaptic neuron; (3) transcriptional regulation in presynaptic neurons; (4) postsynaptic synaptic up-scaling and (5) down-scaling; (6) glial TNFα-mediated up-scaling; (7) retinoic acid- and the Fragile X gene FMR1 mediated up-scaling; and finally (8) calcium signaling via CAMKII and debrin for HmSP. The eight PPI sets show high degree of intersection with each other, with 50% or more of genes in each set in common with at least one other set (Table S7 in Supplementary Datasheet 4), indicative of multiple overlapping protein networks that are tightly controlled to coordinate the critical neuronal modulatory functions of the specific forms of HmSP. We then combined these eight sets into an HmSP database (HmSP DB) of 3,782 unique genes, and the enrichment analysis for this set is summarized here. The detailed analysis is found in Supplementary Material Section 5 in Supplementary Datasheet 4.

As expected, this larger representation of the HmSP functional set shows highly significant overlap with the HD datasets, at 51% of the HTT Interactome and 48% of PerturbDB, and again a higher 71% of HD Common (Table 5, Figure 4A). A similar pattern of overlap with HD datasets holds true for the combined Exosome DB and the 1,600 genes found in both HmSP and the Exosome DBs. Again, the top two PerturbDB platforms that overlap HmSP are the Drosophila in vivo phenotypes and mHTT aggregation platforms (Table S8 in Supplementary Datasheet 4). Conversely, for each of the HD datasets there is also an increase in the relative % overlap of the HmSP-Exosome Common set vs. the HmSP or Exosome DB alone (Figure 4B). A Venn diagram of the HD datasets and HmSP and Exosome DBs (Figure S5) further supports their close tripartite relationships. A similar pattern is observed for the PPI of each of the specific form of HmSP and their common genes with the Exosome DB (Supplementary Material Section 5, Tables S9A,B in Supplementary Datasheet 4). Notably, there are a multitude of known drug targets in these gene sets, many of which have been studied in the context of HD but not with HmSP or exosome biology as endpoints, and others that have not yet been implicated in HD. Notable are CDK5 and calcium channels at the presynaptic level, PLK2 and others at the postsynaptic dendrite, and β-catenin in both glial and synaptic compartments, all highly connected in networks with multiple druggable entry points that provide therapeutic opportunities once the link to the disease is elucidated.

The hypothesis that exosomes and HmSP are functionally linked and collectively impacted by mHTT can be tested in the well-characterized in vivo Drosophila and mammalian cell culture platforms. Drosophila larval NMJ is best suited to test whether presynaptic HmSP and exosome synaptic functions are linked, then each mechanism and where they intersect can be tested for the impact of expressing mHTT in motor neurons, glia or perhaps even the postsynaptic muscle cells. When the mHTT effects are characterized, then known HD modifiers can be tested for their impact on these endpoints. A concordance of modifiers between the mHTT-driven exosome and HmSP endpoints in the NMJ vs. behavioral phenotypes in the CNS would further support the hypothesis. Postsynaptic HmSP such as upscaling and downscaling are best studied in neuronal cell culture, and experimental verification of the impact of mHTT on postsynaptic HmSP is straightforward. Moreover, the role of exosomes in postsynaptic HmSP will require further detailed study before the question of mHTT's impact can be addressed. Success will lead to the development of new HD assays to test the effects of genetic and pharmacological perturbations using these novel mHTT readouts.

An important feature of HD that we have yet to address is its inverse Q length dependency of the age-of-onset of motoric symptoms. The vast majority of PerturbDB readouts are driven by overexpressed mHTT generally of over 100 Qs. Lower Q lengths of most clinical cases (>40 Q) do not induce robust readouts. For example, HD patient fibroblasts with 49 and 70 Qs show a delay in receptor recycling to plasma membrane via clathrin-coated vesicles because of aberrant Rab11 activity (Li et al., 2009) and other downstream effects (Li et al., 2010, 2012), revealing a subtle deficiency in post-endocytic membrane recycling in HD. Similarly, we predict that the impact of lower Q length mHTT on exosome biology and HmSP are subtle and difficult to detect unless additional stressors push the system outside of its compensatory boundaries. We propose that HmSP not only compensates for the effects of mHTT throughout life, but that mHTT actually alters HmSP together with exosome functions, so that eventually the dysfunction in homeostatic mechanisms, perhaps in concert with aging, as the phenotypic consequences of the HTT mutation become progressively more apparent.

In conclusion, we hypothesize that exosomes are involved in certain aspects of HmSP, similar to its critical role in fly NMJ synaptic plasticity, and that WT HTT is involved in both. This tripartite functional link is then altered by mHTT, which eventually result in HD. Experimental verification of the mHTT effects on exosome and HmSP functions in the NMJ and neuronal cell culture platforms will enable further research of both these mechanisms in HD and aid the hunt for drug targets. This hypothesis does predict that WT HTT is involved in exosome biology and HmSP, conceivably linked to the many roles ascribed to HTT, such as transport and trafficking (Caviston and Holzbaur, 2009), ER stress (Vidal et al., 2011), and transcription (Moumné et al., 2013). Finally, our hypothesis may also be relevant for the HTT lowering strategies currently in clinical development that do not discriminate the WT from the mutant form of HTT, as it is plausible that lowering both forms of HTT too drastically may negatively impact the normal functioning of exosome and HmSP and possibly confound any positive effects due to the lowering of mHTT itself.

We examined whether protein networks of pathogenic genes for neurodegenerative diseases that share with HD the slow onset, progressive deterioration, and regional selectivity of pathologies are also linked to HmSP and exosome biology, particularly as many HmSP genes are already implicated in multiple CNS disorders. Several pathogenic PD genes are also part of the HD PerturbDB: SNCA (α-synuclein; Furlong et al., 2000; Corrochano et al., 2011), PARK2 (parkin; Rubio et al., 2009), and PARK7 (DJ-1; Sajjad et al., 2014), and more recently, but not yet curated into PerturbDB, PINK1 (Khalil et al., 2015). Thus, perturbation of these genes in their WT form can alter mHTT-driven outcomes, indicative of a close biological relationship between their functions and mHTT pathophysiology. To extend to other neurodegenerative diseases, we first curated well-known pathogenic genes for PD, AD, eight PolyQ diseases but excluding HD, ALS, and SMA into the Neurodegenerative Disease (NeuroD) PPI gene sets, as described in Methods Section Gene Sets and shown in Table 6. These range from the smallest set of 51 genes in the ATXN2 set (for SCA2) to the largest of 2,133 genes for APP (AD), with most others in the low hundreds of genes. We also combined each PPI into its cognate NeuroD to form larger disease specific gene sets for enrichment analyses. All the gene lists are in Supplementary Datasheet 2.

We first examined the overlap of each NeuroD PPI with the HTT Interactome, as they are all protein networks centered on pathogenic disease genes. Interestingly, there is significant overlap, particularly for the PD PPIs, with 59% of SNCA (α-synuclein) at the highest (−log FDR ~165) and 44% of PARK2 at the lowest (−log FDR ~40). While these genes only account for portion of PD cases, their role in protein folding, misfolded protein processing and mitochondrial functions suggests these mechanisms are important for both diseases. The microtubule-associated tau (MAPT) PPI also highly overlaps the HTT Interactome, at 57% with −log FDR ~44. The role of tau, in its multiple isoforms and phosphorylation states for cytoskeletal functions, has just recently been implicated in HD (Fernández-Nogales et al., 2014; Blum et al., 2015; Gratuze et al., 2015; Vuono et al., 2015). Some of the PolyQ disease genes, especially the VGCC subunit in SCA6 at 40% (−log FDR ~40), points to the importance of calcium dysregulation in HD (Giacomello et al., 2013). We then combined the PPI sets into their respective five disease sets (Table 7, Figure 5), finding again the highest overlap of the HTT Interactome with the PD set (48%, −log FDR ~224). Moreover, the NeuroD PPIs overlap PerturbDB with the HD Common set again showing higher % overlap than each HD dataset (Table 7, Figure 5; SMA at 4% each for the HD single sets and 6% for HD Common). As is the case for other datasets, within PerturbDB the Drosophila in vivo and mammalian aggregation platforms again most highly overlap the NeuroD PPI sets (Tables S10A,B in Supplementary Datasheet 4). Thus, the overlapping NeuroD protein networks contain many modifiers of mHTT-driven outcome measures, and perhaps many are also modifiers for other neurodegenerative diseases. Finally, the NeuroD PPI sets also significantly overlap each other, with about 72% of the PD and ALS sets intersecting at least one other NeuroD set (Table S11 in Supplementary Datasheet 4; Figure S6). Indeed, the subsets of each NeuroD PPI found in common with the HTT Interactome are even more highly overlap with each other (Figure S7), suggesting an even tighter interlocking protein network for neurodegenerative pathogenic genes centered on HTT that when perturbed at different points can contribute to pathogenesis in distinct diseases. More detailed analysis of the NeuroD PPIs is presented in Supplementary Material Section 6.1 in Supplementary Datasheet 4.

Not surprisingly, the NeuroD PPIs also highly overlap the Exosome and HmSP DBs (Table 8), with % overlap ranging from mid 30s for AD to 60s for PD and ALS, and −log FDR values generally in the hundreds. The intersection with the HmSP and Exosome DBs is higher for NeuroD PPIs that are in common with the HTT Interactome (+HTT) than those that do not (−HTT), as exemplified for PD (75% and 52% respectively, Figure 6). The same pattern holds true for the Exosome DB and the HmSP-Exosome Common set (Figure 6), most with −log FDR values over 100 (Table S12 in Supplementary Datasheet 4). The same pattern of increased relative % overlap is true and perhaps even more striking for the remaining NeuroD sets, as shown in Figure S8 (the SMA set increases from 48 to 73%; data not shown). These results suggest that there are more biological links between HmSP and exosomes and the NeuroD protein networks that overlap with the HTT Interactome compared to the parts of the network that do not.

Finally, the overlap of the NeuroD sets is higher for genes in the HmSP-Exosome Common set vs. those in each DB alone, with the exception of the PolyQ (Figure S9). This again supports a linkage between HmSP and exosome biology, and furthermore that where the two mechanisms intersect is particularly important for the pathophysiology of these neurodegenerative diseases. Again, when the NeuroD PPIs are segregated into the subsets that overlap the HTT Interactome, the relative % overlap is even higher (Figures S9A,B in Supplementary Datasheet 4). This is particularly prominent for the PolyQ PPIs, where the increased relative % overlap is obvious only for the +HTT subset. The same trend is found for the constituent subsets of HmSP (e.g., presynaptic, up-scaling and down-scaling etc.) as presented in Tables S13A–F in Supplementary Datasheet 4. Thus, as in the case for HD, there appears to be strong linkages between exosome biology and HmSP, and particularly where the two mechanisms intersect, with the other neurodegenerative diseases. As discussed for HD, experimental verification of the hypothesis is needed, using experimental platforms driven by the appropriate mutant NeuroD pathogenic genes, to determine if exosome and HmSP functions are indeed altered.