Network Systems Pharmacology-Based Mechanism Study on the Beneficial Effects of Vitamin D against Psychosis in Alzheimer’s Disease

Alzheimer’s disease (AD) is a chronic neurodegenerative disease with significant financial costs and negative impacts on quality of life. Psychotic symptoms, i.e., the presence of delusions and/or hallucinations, is a frequent complication of AD. About 50% of AD patients will develop psychotic symptoms (AD with Psychosis, or AD + P) and these patients will experience an even more rapid cognitive decline than AD patients without psychosis (AD-P). In a previous analysis on medication records of 776 AD patients, we had shown that use of Vitamin D was associated with delayed time to psychosis in AD patients and Vitamin D was used more by AD-P than AD + P patients. To explore the potential molecular mechanism behind our findings, we applied systems pharmacology approaches to investigate the crosstalk between AD and psychosis. Specifically, we built protein-protein interaction (PPI) networks with proteins encoded by AD- and psychosis-related genes and Vitamin D-perturbed genes. Using network analysis we identified several high-impact genes, including NOTCH4, COMT, CACNA1C and DRD3 which are related to calcium homeostasis. The new findings highlight the key role of calcium-related signaling pathways in AD + P development and may provide a new direction and facilitate hypothesis generation for future drug development.

In a previous study, we have compared the frequency of medication usage among AD + P and AD-P patients and conducted survival analysis on time to psychosis for AD patients to identify drugs with beneficial effects 18 . The results of our analysis revealed a significant association between Vitamin D use and delayed onset of psychotic symptoms. In addition, through the analysis of gene expression data, we found that AD-and/or psychosis-related genes were enriched in the list of genes most perturbed by Vitamin D. This observation provides us with a novel direction for the mechanism study of AD and psychosis, and may inspire the development of drugs to prevent or treat psychosis in AD.
The role of Vitamin D in neurodegenerative diseases has been reported by many researchers. Six of the nine case-control studies found significant between-group differences illustrated by lower serum concentrations of 25-hydroxyvitamin D, a metabolite of Vitamin D 3 , in AD cases compared to control groups [19][20][21][22][23][24][25] . Thus, Vitamin D Insufficiency is considered as a risk factor for AD. However, Vitamin D's beneficial effect against AD + P was freshly discovered and its mechanism may provide a unique viewpoint in preventing and treating AD + P.
Network approaches have been used in predicting and identifying the disease genes in multiple studies and some of the results have been verified 26,27 . It is suggested that, in the viewpoint of network biology, drug targets tend to locate at the transition area from the essential hubs, e.g. proteins interacting with more partner proteins, to redundant peripheral nodes 28 , e.g. proteins interacting with fewer partner proteins. The rationale behind this is a balance of toxicity and efficacy regarding the potential influence of the targets on cellular function.
The aim of this study is to explore potential molecular mechanisms that underlie the beneficial effects of Vitamin D on reducing psychosis symptoms in AD patients and to identify potential drug targets for AD + P prevention or treatment by applying systems pharmacology approaches on analyzing their protein-protein interaction networks.

Method and Material
Gene dataset collection and pathway mapping. A network that includes both AD-and psychosis-related proteins were constructed and analyzed in order to study the crosstalk between them. AD-and psychosis-related genes were collected from multiple literatures and databases, including MetaCore from Clarivate Analytics (https://portal.genego.com/), GWAS Catalog for Genome Wide Association Studies (GWAS) (https:// www.ebi.ac.uk/GWAS/home) 29 and BaseSpace Correlation Engine (https://www.illumina.com/index-d.html) 30 . These gene names were then converted to protein names by batch search function in the UniProt database. The criteria for including genes in our study are described in supplementary material. Vitamin D-perturbed genes and antipsychotics-perturbed genes were collected from BaseSpace Correlation Engine (https://www.illumina.com/ index-d.html) 30 . Both down-and up-regulated genes were included into our analysis.
network was shown in the molecular action view with the medium confidence level (>0.4) 36 . The network containing AD-related proteins (AD network) and the network containing psychosis-related proteins (Psychosis network) were joined to form a combined network (AD-psychosis combined network) for further study. PPI networks containing Vitamin D-perturbed proteins (Vitamin D network) and antipsychotics-perturbed proteins (Antipsychotics network) are also generated respectively. Centrality measures of the nodes were introduced in network analysis to describe how the information will spread through the network. Two different kinds of centralities were included: Degree Centrality and Betweenness Centrality. Degree Centrality, as the most simple and direct, describes the number of connections of a particular node regardless of the direction and weight of the edges. Betweenness Centrality, as the centrality of control, represents the frequency at which a point occurs on the geodesic (shortest paths) that connect pairs of nodes. In another word, it quantifies how many times a particular node acts as a bridge linking two ends of the network.
Networks were processed and plotted with python package networkx 35 and Gephi 37 . The centrality of nodes in the network was calculated based on the built-in algorithm of networkx 35 . In detail, the degree centrality values were normalized by dividing by the maximum possible degree in a simple graph n-1 where n is the number of nodes in a network. The Betweenness centrality algorithm is from Ulkrik Brandes [38][39][40][41] .
In order to minimize the bias caused by the number of studies associated with different proteins, we use Betweenness centrality as our primary indicator in this study to learn more on the nodes' position in the network's structure, rather than the degree centrality of the nodes in the network.
Network analysis methods with centrality measures will first be examined with psychosis-related genes and known antipsychotics-perturbed targets. In order to do that, a combined network of psychosis network and antipsychotic network is constructed and the centrality measures are calculated as mentioned above. The connectivity parameters of known antipsychotic targets are examined to determine if they possess a significantly higher value.
To find sub-networks (communities) having different biological functions, community detection was further conducted in the combined network. The algorithm used for community detection was based on the Greedy Modularity Maximization method 42,43 . It begins with each node in its own community and joins the pair of communities that most increases modularity until no such pair exists.  Table 3. Characteristics of AD-and Psychosis-related PPI networks. www.nature.com/scientificreports www.nature.com/scientificreports/ Triple-focusing network approaches: identification of potential novel targets. Network analysis was further used to study a joint AD-psychosis-Vitamin D network in order to find potential drug targets for AD + P. The rationale of this approach was that the ideal potential targets should be in the overlapping part of PPI networks of AD, psychosis and Vitamin D because the function of the potential targets can modulate the crosstalk between AD and psychosis and can also be regulated by Vitamin D through the Vitamin D receptor which is a transcriptional factor modulating gene expression. Thus, after constructing the AD-psychosis combined network and Vitamin D network, we overlapped them to explore the connectivity of these three parts and the roles of the triple-overlapped proteins.
Triple-focusing on AD-, psychosis-related and Vitamin D-perturbed proteins can help us reduce the artificial bias caused by the different amount of studies of those proteins, and also limit the potential side effects caused by targeting those very well-studied proteins which are usually located at the essential hubs. The identified small groups of proteins will have the potential to act as targets for Vitamin D to modulate AD-and psychosis-related networks.

Results
Method verification with psychosis-related PPI network and antipsychotics-perturbed genes. Psychosis-related and antipsychotics-perturbed PPI networks are used to validate the network analysis methods we proposed. Characteristics of these two PPI networks and the combined network are shown below (Table 1). Five genes, DRD2, DRD3, HTR2A, OPRD1 and HTR7, are found shared by psychosis network and antipsychotics network.
The centrality measures of nodes in the psychosis and antipsychotics are calculated and the top ten nodes sorted by Betweenness values were shown in Table 2. As we expected, DRD2 and HTR2A, two major targets for current antipsychotics, were ranked as the first two proteins in our combined network when measured by Betweenness Centrality. If ranked by Degree centrality, ALB and FOS, two well-studied proteins, will have higher priority than HTR2A. The result revealed the great potential for proteins with a high Betweenness centrality being drug targets and provided a solid support for the method we proposed. Thus, the network analysis methods were applied to AD-and psychosis-related PPI networks.
The AD-psychosis combined PPI network. In order to acquire a better understanding of the connection between AD and psychosis, and to further explore the potential drug targets suggested by the previous analysis, a combined PPI network of AD and psychosis was generated. One thousand and sixty-one AD-related genes and 15,691 PPIs of their protein products together with 483 psychosis-related genes and 1,361 psychosis-related PPIs were collected as the basis of our network. Among all the proteins collected, 90 proteins were shared by both AD and psychosis, including proteins encoded by SEMA3A , TUSC3, RPN2, AMBRA1, BECN1, CACNA1C, SGK1,  ADAM10, GRIN2A, FYN, ANK3, TBXAS1, EFNA5, POLN, CHRNA3, NOTCH4, GRIA1, NTRK3, IQGAP2,  RELN, NOS1, GPC6, TCF7L2, TCF4, MGLL, DRD3, CHRNA2, PAK2, CTNNA2, COL25A1, COL12A1, AGER,  KIF26B, PPP2R2B, TEK, KALRN, PRKG1, KSR2, COLGALT2, MEIS1, SHISA9, ZKSCAN4, PTPRG, NKAPL,  CTNNA3, PDE4B, HFE, MSR1, CSMD1, COMT, APBA1, IMMP2L, ELAVL4, LRRTM4, CDH13, ZNF804A,  PBRM1, LRRN2, TEP1, STXBP5L, FHIT, SYNGAP1, ZSCAN31, TENM4, ABCB1, PLCL1, RBFOX1, FSTL5,  SORCS3, NKAIN2, GLIS3, NXN, MAGI2, MEGF10, MPP6, TSPAN18, FRMD4B, MTHFD1L, TMTC1, LIN28B,  UXS1, BICC1, ATXN7L1, EYS, GRAMD1B, TSPAN2, ENOX1, TMEM132D, CR1 and PCNX. The AD-psychosis combined network has 1,454 nodes and 16,948 PPIs. Characteristics of the combined network were most similar to those of AD network due to the disparity of the node numbers in AD-and psychosis-related PPI networks (Table 3).  www.nature.com/scientificreports www.nature.com/scientificreports/ Top 10 net-influencers in the combined network are shown in Table 3 based on their Degree and Betweenness centralities respectively. It is not surprising that the 3 centralities overlapped with each other substantially, since they all measure the importance of the nodes in the whole network from different angles, and it is apparent that the top 10 nodes do have very higher values when compared with the average value, 10-fold ratio at least. A better view is provided in Fig. 1 showing that only a few nodes take position at the upper-right corner. This phenomenon suggests that though there are 1,454 of nodes in the network, a small group of nodes, such as the top 10 nodes shown in the table, are extremely connected and play a critical role in the signaling process and information flow within the network.
After identifying the critical proteins in the network, the function of these proteins is our interest. We conducted pathway enrichment analysis to identify the underlying pathways participated by those proteins and therefore to establish a connection between proteins and their biological functions. Firstly, ten communities were detected as relatively separated components of the network (Fig. 2). Among the 10 detected communities, 7 communities, excluding 7, 8 and 9, contain enough nodes to be biologically meaningful. When sorting the network based on the community and the nodes' Betweenness, every community has one or a few nodes that possess a much higher Betweenness value and those nodes serve as the portal connecting the community to the other parts of the network (Fig. 3). Among the top 10 proteins we mentioned above (Table 4), APP, FYN, and INS are distributed into different communities as the leading nodes, which further illustrates the importance of these nodes in the combined network. These detected communities represent different biological pathways participating in the development of psychosis in AD. Secondly, protein-pathway mapping was conducted by comparing the proteins in the same community against the proteins in the pathways from online databases like KEGG.
The protein-pathway mapping returned a list of pathways associated with these 7 communities evidenced with very low False Discovery Rate (FDR) adjusted p-values, meaning that the proteins in these communities are highly accordant with proteins in these pathways recorded in the database. Pathways closely related to AD and neurological disorders (Table 5) were enriched in the list, including the Huntington disease pathway, Alzheimer's disease-presenilin pathway, p53 pathway and Alzheimer's disease-amyloid secretase pathway. Figure 4 provided a more direct overview of the results of protein-pathway mapping. Community 1 and community 2 were mapped to multiple pathways with high credibility. It is fairly understandable because these two communities contain the largest amounts of targets and may result in mismatches.

Community
Pathways (Pathway ID) P-value  www.nature.com/scientificreports www.nature.com/scientificreports/ Overlapping proteins between AD network and Psychosis network. Since the objective of this study is to study the development of psychosis in AD, we focused on the overlapping proteins between AD and psychosis. The net-influence parameters of the 90 overlapped proteins are shown in Table 6. Most proteins in the overlapping part possess Betweenness values above average which further supports their bridging role in the networks. Figure 5 shows the distribution of connectivity parameters of overlapping proteins. Even in the overlapping part of the network, the average Betweenness centrality remains relatively low and only a few nodes, like FYN and GRIA1, possess a much higher connectivity than other nodes. The distribution of Betweenness follows the same pattern as the whole network suggesting that even though 90 targets are found overlapped between psychosis and AD, only a few of them are the "bridges" for the transferring of information.

Exploration of Vitamin D's beneficial effect through a triple-focusing approach.
In our previously published paper, Vitamin D was identified as a promising medication with a significant association with decreased occurrences and delayed onset of AD + P 18 . Therefore, we examined the relationship between the Vitamin D network and the AD-psychosis combined network. In total, 89 targets and 344 PPIs were collected in the Vitamin D network (Table 7). Among the 89 proteins, twenty-one are shared with the AD-psychosis combined network. Net influence parameters are calculated for these 21 targets and sorted by their Betweenness centrality values (Table 8).
After sorting by the Betweenness centrality, CACNA1C, COMT, NOTCH4 and DRD3 are ranked as the top four proteins. Their positions in the overlapping part of the combined network allow them to function more as a bridge to link different components of the network, which also suggests a therapeutic potential for AD + P. Therefore, these four proteins gained our special interest. One interesting thing is, when we look back at Fig. 1, these four targets fell into the middle distribution of values for Degree centrality and Betweenness centrality, which matched the conclusion that drug targets tend to be positioned at the transition area in a biological network 28 . Figure 6 shows the distribution of connectivity parameters of overlapping proteins between AD-psychosis combined network and Vitamin D network. The 21 overlapped nodes followed the distribution of the whole combined network and revealed several proteins with outstanding Betweenness centrality values. These proteins will tend to act as the "bridges" in communicating AD-, psychosis-related network and Vitamin D perturbed network and thus the potential explanation of the beneficial effects of Vitamin D against AD + P.

Discussion
The network analysis based on the protein-protein interaction data have presented us four potential targets encoded by genes CACNA1C, NOTCH4, COMT and DRD3 that may account for the beneficial effects of Vitamin D against AD + P. These four potential targets all possess high enough connectivity to alter the crosstalk between AD and psychosis. In addition, variants in CACNA1C, NOTCH4 and COMT had been reported to be associated with schizophrenia in GWAS studies [44][45][46] . Among them, the function of CACNA1C, NOTCH4 and COMT were reported to be closely associated with calcium homeostasis 47-53 which can be further associated with Vitamin D's effect. Similarly, after the  www.nature.com/scientificreports www.nature.com/scientificreports/ activation of DRD3 by dopamine, the Gβγ complex is released and can interact directly with voltage-gated calcium channels 54,55 . Except for NOTCH4, all are targeted by marketed drugs for different indications. Interestingly, DRD3 is one of the primary targets for antipsychotics in treating psychotic symptoms in schizophrenia or other neurological disorders 56,57 . Alternative splicing of DRD3 in the transcription process may result in encoding different isoforms that are functionally impaired 58 . Although limited, there is some support for targeting DRD3 in the treatment of AD + P 59,60 . However, verification of DRD3 or the other of these four potential targets for AD + P will require additional studies.
The beneficial effect of Vitamin D against AD have been widely reported. The protective effect of Vitamin D can be executed by reducing the oxidative and nitrosative damage caused by elevated levels of nitric oxide (NO) and inducible nitric oxide synthase (iNOS) in nerve cells 61 . There is also evidence suggesting an overlap between the disruptions of vitamin D pathways with amyloid pathology which can partially explain the protective role of Vitamin D in AD 62 . However, this study is the first study to explore the mechanism of Vitamin D's beneficial effect against AD + P. In this study, the triple-focusing approach we use can help minimize the bias caused amount of studies and restrain our scope at Vitamin D related potential targets.
There are limitations in this study. The PPI networks were constructed based on the protein-protein interaction data extracted from databases, thus they are limited by the amount and availability of data in the databases. Also, there is no direction information attached with most PPIs which means our PPI networks are undirected. Therefore, centrality measures can be biased by the direction information in actual situations.

conclusion
In this study, various approaches of network analysis are incorporated with systems pharmacology to provide a systematic overview on the crosstalk among AD, psychosis and Vitamin D at the molecular level. The triple-focusing network method helps us explore the designated mechanisms for Vitamin D's effects on AD + P and a potential explanation is provided: Vitamin D regulates several genes encoding proteins that play critical  Table 6. Overview of net-influencers for overlapping proteins (named by their genes) between AD network and Psychosis network.  www.nature.com/scientificreports www.nature.com/scientificreports/ roles in the overlapping part of the AD-psychosis combined network, which allow them maximally influence the signaling and information transfer process. In other words, proteins with high net-influence that localize at the triple-overlapped part of the AD, psychosis and Vitamin D network, like CACNA1C, COMT, NOTCH4 and DRD3, possess the ability to play an important role in the crosstalk among AD and psychosis by delivering Vitamin D's effect to the transiting hub connecting the AD network and psychosis network. Thus, the four identified potential targets can be crucial in explaining Vitamin D's beneficial effect against AD + P. To conclude, the results from this study provided a possible explanation of the beneficial effect of Vitamin D against AD + P and presented a new direction for drug development with four potential novel targets.

Data Availability
Full gene lists used in this study can be found in the supplementary material. The detailed PPI interaction data and full net-influence parameters list are available on request.