DLG2 expression is low in high INSS stage NB and correlates to survival
We evaluated the association of DLG2 expression with INSS stage as well as patient outcomes, using the NB primary dataset 1 (GSE49710) obtained from the R2 Genomics Analysis and Visualization Platform (http://r2.amc.nl). When comparing DLG2 expression level between INSS stages, stage 4 tumors showed significantly lower DLG2 expression when compared to stage 1 tumors (log2 FC= 0.89, p< 0.001), stage 2 tumors (log2 FC= 0.83, p< 0.001) and stage 3 (log2 FC= 0.55, p< 0.005), with the only exception of stage 4s tumors (Fig. 1a). A similar trend was also observed in the Neuroblastoma primary datasets 2, 3 and 4 (Fig. S1a, b and d respectively), with the stage 4 tumors showing low DLG2 expression. There was also a significant difference in the expression of DLG2 in the high risk and low risk patients (log2 FC= 0.80, p< 0.001) (Fig. 1b). The same trend was also observed in dataset 4 (Fig. S1f). Overall survival outcomes (Fig. 1c) and event free survival (Fig. 1d) for the patients were determined using Kaplan-Meier analysis using NB primary dataset 1. High expression of DLG2 was associated with increased probability of both overall and event free survival (Fig. 1c-d, p<0.001). ). The same trend was also observed in NB primary dataset 2 (Fig. S2a and S2b). The gene expression data was subsequently clustered into 4 groups using the normalized total array expression and K means clustering method, and the survival of each of the 4 groups was then determined using Kaplan-Meier analysis. One of the groups showed a high survival, with 98% overall survival, while the other groups showed 82%, 47% respectively 37% survival (Fig. 1e). The groups were subsequently named after their overall survival and DLG2 expression was determined for each group. The group with 98% survival also had the highest DLG2 expression compared to 82% survival (log2 FC= 0.49, p <0.001), 47% survival (log2 FC= 0.71, p <0.001) and 37% survival (log2 FC= 0.90, p <0.001), there was also a significantly lower expression of DLG2 in the 37% survival group when compared to the 82% survival group (log2 FC= 0.41, p <0.05) (Fig. 1f). The similar K means clustering results was also observed in dataset 2 (Fig. S2c and S2d).
Gene set enrichment analysis shows DLG2 inversely correlates to cell cycle genes
Gene lists were created for the genes correlating to DLG2 expression in NB primary datasets 1 (GSE49710), 2 (GSE16476) and 3 (GSE3960), and an embryogenesis dataset 1 (GSE15744). The NB primary dataset 4 was not analysed due to the customized probe layout, and the cell line datasets were also not included. Gene set enrichment analysis of the genes correlating to DLG2 in the selected data sets showed that the pathways; cell cycle (p<0.0001), DNA replication (p<0.0001), Fanconi anemia (p<0.0001) and mismatch repair (p<0.0001) were enriched in the three analyzed NB primary datasets (Table 1-3). When these four identified pathways were compared to the enriched pathways in the embryogenesis dataset (Table 4), the Fanconi anemia pathway did not appear. The proportion of cell cycle genes in NB primary datasets 1, 2 and 3 were 71.8%; 31.4% and 47.7%, respectively; (p<0.0001), DNA replication genes 88.9%; 52.8% and 78.1%, respectively (p<0.0001), mismatch repair genes 78.3%; 40.9% and 70.0%, respectively (p<0.0001) and Fanconi anemia pathway genes 69.8%; 42.2% and 60.0%, respectively (p<0.0001) (Table 1-3). The embryogenesis dataset showed 41.7% cell cycle genes (p<0.001), 61.1% DNA replication genes (p<0.0001) and 63.6% mismatch repair genes (p<0.0001) (Table 4). The genes involved in the enriched pathways showed an overriding negative correlation to DLG2 expression. Concordance between the enriched pathways for the NB primary datasets 1, 2 and 3 was first determined using the intersection of these three datasets and subsequently compared to the embryogenesis dataset. The intersect of the cell cycle in the NB datasets included 25/71 genes (Fig. 2a), 13/35 genes for DNA replication (Fig. 2b), and 7/20 mismatch repair genes (Fig. 2c), with the concordant genes listed in the supplementary data (Table S1). When compared to the embryogenesis dataset the intersection included 14/61 genes for cell cycle (Fig. 2d), 8/27 genes for DNA replication (Fig. 2e) and 5/17 genes for mismatch repair (Fig. 2f); these concordant genes are found in Table 5.
Table 1. GSEA of genes correlated to DLG2 expression in NB primary dataset 1 (GSE49710)
Enriched KEGG Pathway
|
# correlated genes
|
# pathway genes in total
|
Percentage
|
Corrected p-value
|
Ribosome
|
124
|
133
|
93.2%
|
5.2e-25
|
Oxidative phosphorylation
|
93
|
120
|
77.5%
|
2.0e-10
|
Alzheimer s disease
|
115
|
162
|
71.0%
|
9.7e-09
|
Cell cycle
|
89
|
124
|
71.8%
|
2.1e-07
|
Huntington’s disease
|
126
|
187
|
67.4%
|
2.3e-07
|
DNA replication
|
32
|
36
|
88.9%
|
1.2e-06
|
Sphingolipid signaling pathway
|
84
|
120
|
70.0%
|
2.4e-06
|
Spliceosome
|
89
|
131
|
67.9%
|
8.2e-06
|
Fanconi anemia pathway
|
37
|
53
|
69.8%
|
1.9e-03
|
Mismatch repair
|
18
|
23
|
78.3%
|
4.3e-03
|
Table 2. GSEA of genes correlated to DLG2 expression in NB primary dataset 2 (GSE16476)
Enriched KEGG Pathway
|
# correlated genes
|
# pathway genes in total
|
Percentage
|
Corrected p-value
|
DNA replication
|
19
|
36
|
52.8%
|
2.1e-11
|
Ribosome biogenesis in eukaryotes
|
27
|
64
|
42.2%
|
8.4e-11
|
RNA transport
|
44
|
139
|
31.7%
|
2.1e-09
|
Cell cycle
|
38
|
121
|
31.4%
|
3.6e-08
|
Fanconi anemia pathway
|
19
|
45
|
42.2%
|
5.0e-08
|
Spliceosome
|
37
|
119
|
31.1%
|
8.0e-08
|
Pyrimidine metabolism
|
26
|
94
|
27.7%
|
1.4e-04
|
Homologous recombination
|
11
|
29
|
37.9%
|
2.1e-04
|
Mismatch repair
|
9
|
22
|
40.9%
|
2.8e-04
|
Non homologous end joining
|
5
|
11
|
45.5%
|
2.7e-03
|
Table 3. GSEA of genes correlated to DLG2 expression in NB primary dataset 3 (GSE3960)
KEGG Pathway
|
# correlated genes
|
# pathway genes in total
|
Percentage
|
Corrected p-value
|
Ribosome
|
57
|
70
|
81.4%
|
4.50e-20
|
DNA replication
|
25
|
32
|
78.1%
|
6.70e-09
|
Oocyte meiosis
|
42
|
80
|
52.5%
|
2.60e-05
|
Pyrimidine metabolism
|
34
|
63
|
54.0%
|
6.80e-05
|
Cell cycle
|
53
|
111
|
47.7%
|
1.10e-04
|
Mismatch repair
|
14
|
20
|
70.0%
|
1.50e-04
|
Spliceosome
|
45
|
95
|
47.4%
|
4.70e-04
|
One carbon pool by folate
|
9
|
12
|
75.0%
|
9.10e-04
|
Fanconi anemia pathway
|
15
|
25
|
60.0%
|
1.60e-03
|
Folate biosynthesis
|
8
|
11
|
72.7%
|
2.60e-03
|
Table 4. GSEA of genes correlated to DLG2 expression in the human embryogenesis dataset (GSE15744).
KEGG Pathway
|
# correlated genes
|
# pathway genes in total
|
Percentage
|
Corrected p-value
|
One carbon pool by folate
|
13
|
17
|
76.5%
|
3.3e-05
|
Spliceosome
|
56
|
118
|
47.5%
|
4.5e-05
|
DNA replication
|
22
|
36
|
61.1%
|
5.4e-05
|
ECM receptor interaction
|
36
|
70
|
51.4%
|
1.1e-04
|
Mismatch repair
|
14
|
22
|
63.6%
|
6.4e-04
|
Protein digestion and absorption
|
33
|
67
|
49.3%
|
6.9e-04
|
Glycolysis Gluconeogenesis
|
27
|
57
|
47.4%
|
4.8e-03
|
Salivary secretion
|
25
|
52
|
48.1%
|
5.0e-03
|
Pyruvate metabolism
|
18
|
35
|
51.4%
|
6.3e-03
|
Cell cycle
|
50
|
120
|
41.7%
|
6.3e-03
|
Table 5. Cell cycle, DNA replication and mismatch repair genes negatively correlated to DLG2 expression, common to NB primary datasets 1-3 (GSE49710, GSE16476 and GSE89413) and the human embryogenesis dataset (GSE15744).
Gene symbol
|
Gene function
|
Cellular process
|
BUB1, BUB1B
|
Mitotic checkpoint serine/threonine kinase
|
Mitotic checkpoint
|
CCNA2
|
Cyclin A2, G1/S to G2/M transition
|
Cell cycle regulation
|
CCNB1
|
Cyclin B1, G2/M transition
|
Cell cycle regulation
|
CDC20
|
Cell division, anaphase regulation
|
Chromosome separation
|
CDK4
|
Cyclin dependent kinase, G1/S phase
|
Cell cycle regulation
|
DBF4
|
E2F mediated regulation of DNA replication
|
DNA replication
|
ESPL1
|
Sister chromatid cohesion and separation
|
Chromosome separation
|
EXO1
|
Exonuclease 1
|
DNA repair
|
LIG1
|
DNA ligase, DNA replication and repair
|
DNA replication
|
MCM2, MCM3, MCM5, MCM7
|
Initiation of genome replication
|
DNA replication
|
MSH6
|
Mismatch recognition
|
DNA repair
|
POLD2
|
DNA polymerase delta
|
DNA replication
|
PRKDC
|
DNA-dependent protein kinase
|
DNA repair
|
POLA1, POLA2
|
DNA polymerase alpha
|
DNA replication
|
RFC3, RFC4
|
Replication factor, DNA elongation
|
DNA replication
|
TP53
|
Tumor suppressor, DNA binding
|
Cell cycle regulation
|
MYCN amplified tumors have low DLG2 expression
Gene expression of DLG2 was found to be significantly lower in MYCN amplified samples in 4 independent datasets (1, 2, 3, 4), including, two patient cohorts and two cell line cohorts (Fig. 3a-d). The NB primary dataset 1 was used to identify the decrease in DLG2 expression in MYCN amplified samples (log2FC= 0.7131, p<0.0001) (Fig. 3a), with NB primary dataset 2 used to confirm the result (log2 FC= 1.212, p=0.0005) (Fig. 3b). The NB cell dataset 1 (GSE89413) with 39 distinct NB cell lines (27 containing MYCN amplification) was used to confirm the lower DLG2 expression in MYCN amplified cell lines (log2 FC= 0.65441, p=0.0022) (Fig. 3c). In the NB cell dataset 2 (GSE28019) we controlled for variability in the tissue of origin of the cell lines by using only the cell lines derived from metastatic bone marrow, here the MYCN amplified cells also showed lower expression (log2 FC= 1,496, p=0.049) (Fig. 3d). We further noted that induction of MYCN in the NB cell dataset 3, (GSE16478), resulted in a decrease in the expression of DLG2 (Fig. 3e). A similar trend was also observed in the NB primary datasets 3 and 4 (Fig. S1c and S1e) with the MYCN amplified tumors showing low DLG2 expression.
To model the effect of MYCN amplification on DLG2 we investigated the effect of forced expression of the orthologous Drosophila melanogaster dMYC gene using a UAS-dMYC construct ubiquitously induced by da-Gal4 in a fruit fly model, and investigated changes in dmDLG gene expression and protein level in the larvae. Overexpression of dMYC resulted in lower gene expression of dmDLG (log2 FC= 0.8431, p=0.001) (Fig. 3f). We confirmed the decrease in dmDLG also on protein level (Fig. 3g), and furthermore evaluated the effect of dMYC over expression on Cyclin A (cycA) and Cyclin B (cycB). Consistent with the previous literature (30), we could determine an increase in the expression of both cycA and cycB (Fig. 3g).
11q deletion correlates to low DLG2 expression
Gene expression of DLG2 was found to be significantly lower in 11q-deleted NB tumor data when MYCN amplification was excluded, in two independent neuroblastoma primary datasets (3 and 4). In primary NB dataset 3 (GSE3960), the 11q-deleted samples showed a significant decrease (log2FC= 0.851, p=0.0004) when compared to the 11q normal tumors (Fig. 4a). NB primary dataset 4 (GSE73517) confirmed that 11q-deleted tumors had lower expression of DLG2 compared to the 11q normal tumors (log2FC = 0.661, p=0.034) (Fig. 4b). Methylation array data showed that there was very low methylation of the DLG2 promoter region, no observable difference in the promoter region methylation pattern was observed in 11q-deleted NB compared to 11q normal in NB methylation dataset 1 (GSE73515) (Fig. 4c) and also no DLG2 promoter methylation in general in NB methylation dataset 2 (GSE120650) (Fig. 4d).
DLG2 silencing or over expression changes the growth behavior of NB cells.
Over expression of DLG2 in 11q-deleted NB cells (SKNAS) resulted in slower proliferation compared to the control (Fig. 5a, p<0.001). We observed a decrease in the number of viable cells (Fig. 5b, p<0.001) and an increase in the non-viable cell fraction (Fig. 5b, p<0.05) in cells with increased DLG2 expression. DLG2 silencing in 11q-deleted SKNAS cells resulted in a slight decrease in cell proliferation (Fig. 5a, p<0.05), with no effect in viable/non-viable cell numbers (Fig. 5b). We determined that over expression of DLG2 in SKNAS resulted in a decrease of the percentage of cells in G1 phase and an increase in the number of cells in G2/M phase (Fig. 5c). There was no difference in the cell cycle state for DLG2 silenced cells compared to control (Fig. 5c). This was repeated in 11q normal NB cells (NB69), where over expression of DLG2 resulted in slower proliferation compared to the control (Fig. 5d, p<0.001). There was a significant decrease in total viable cell number (Fig. 5e, p<0.001) with no increase in the nonviable fraction. Knockdown of DLG2 resulted in an increase in cell proliferation compared to the mock control in 11q normal NB69 cells (Fig. 5d, p<0.001). An increase in the total number of viable cells was also determined (Fig. 5e, p<0.001), with no change in the number of non-viable cells. As there was no alteration in the non-viable fraction after changing the DLG2 expression level, the changes in proliferation was not likely due to apoptosis, rather a change in cell cycle progression. Over expression of DLG2 resulted in an increased fraction of cells in G2/M phase (Fig. 5f, p<0.001) and the associated decrease in the fraction of cells in G1 phase (Fig. 5f, p<0.001) compared to the mock control, no difference in the number of cells in S phase was detected. Conversely, knockdown of DLG2 resulted in an increase in the number of cells in G1 phase and a decrease in the number of cells in S phase (Fig. 5f, p<0.001). Over expression and knockdown of DLG2 was confirmed using qPCR in both SKNAS and NB69 cell lines (Fig. 5g and h). To model the effect of DLG2 loss, we tested the effect of knockdown of the orthologous Drosophila dmDLG gene in a fruit fly model. RNAi-dmDLG knockdown resulted in a small decrease in dmDLG RNA expression in the larvae (Fig. 5i), the result was confirmed on protein level (Fig. 5j). To confirm if the effect on cell cycle proteins of RNAi-dmDLG knockdown was similar to that of dMYC over expression (Fig. 3f) we determined the protein expression of Cyclin A and Cyclin B by Western blot. Both Cyclin A and B showed increased expression when compared to the control (Fig. 5j).