Expression of DLG members with L27 domains were inversely correlated to survival and risk
The main difference between different proteins encoded by DLG family members and their isoforms is the presence or absence of an N-terminal L27 domain (Figure 1a). Unique DLG2 exons is used to encode the different DLG2 isoforms, the exon structure and initiation sites of the ζ, β, α, ε, δ and γ protein isoforms is presented in Figure 1b (15, 16). We evaluated the association of DLG family expression with event free survival and risk, using online NB patient dataset (GSE49710) and patient dataset (TARGET) obtained from the R2 Genomics Analysis and Visualization Platform (http://r2.amc.nl). Risk stratification showed higher expression in low risk NB for DLG1 (log2 fc =0.40, p< 0.001), DLG2 (log2 fc =0.68, p< 0.001) and DLG4 (log2 fc =0.72, p< 0.001), whereas DLG3 (log2 fc =-0.47, p< 0.001) showed lower expression in low risk NB (Figure 1c). The level of expression of DLG1 (Figure 1d) or DLG4 (Figure 1f) showed no difference in event free survival whereas high DLG2 expression was associated with a longer event free survival (p< 0.001) (Figure 1e).
DLG2 isoform 7/8 were downregulated in high stage neuroblastoma
We evaluated the expression levels in NB of the L27-domain containing DLG family members, DLG1, DLG2 and DLG4, by comparing the total gene expression and transcripts encoding the alpha or beta proteins, using RNAseq-data from the NB patient dataset (TARGET) obtained from the R2 Genomics Analysis and Visualization Platform (http://r2.amc.nl). The data was divided into INSS stage for DLG1, DLG2 and DLG4. DLG1 showed a decreased DLG1 isoform 1 (DLG1-iso1) (ENST00000452595), (encoding SAP97α protein), expression in stage 4 NB compared to the favorable stage 4s (log2 FC= 0.44, p< 0.001) with no difference between stage 3 and 4 (Figure 2a). Decreased DLG1 isoform 2 (DLG1-iso2) (ENST00000357674), (encoding L27-containing SAP-97β protein), expression in stage 4 NB compared to stage 4s (log2 FC= 0.44, p< 0.001) and between stage 3 and stage 4s (log2 FC= 0.76, p< 0.05), was also seen (Figure 2a). At the total DLG1 gene expression level a similar decrease in expression as the DLG1-iso2 transcript was observed between stage 4 and stage 4s (log2 FC= 0.80, p< 0.001) and between stage 3 and stage 4s (log2 FC= 0.76, p< 0.05) (Figure 2a). We confirmed the DLG1-iso1 expression by using an independent NB patient dataset (GSE16476) based on microarray data, also showing the similar DLG1-iso1 expression in the stage 1+2 and 4s tumor groups, both considered low risk tumors (Figure 2b).
When analyzing the TARGET data, DLG2 showed no difference in DLG2 isoform 1 (DLG2-iso1) (ENST00000376104), (encoding the truncated L27-containing SAP-93β), expression or DLG2 isoform 2 (DLG2-iso2) (ENST00000398309), (encoding the non-L27-containing PSD-93α), expression when comparing the stages (Figure 2c). At the total gene expression level (including all DLG2 isoforms) a decrease in expression was observed between stage 4 and stage 4s (log2 FC= 0.72, p< 0.001) (Figure 2c), indicating that isoforms accounting for this difference were not included in this analysis.
We evaluated the expression level of all main DLG2 isoforms in NB, using the transcript data from the TARGET dataset based off GRCh37. We determined that the DLG2 isoforms with the highest expression were DLG2-iso2 (ENST00000398309) and DLG2 L27 only (ENST00000472545), with no or very low expression of isoforms 1 (ENST00000376104), 3 (ENST00000418306) or 4 (ENST00000280241) detected (Figure 2d). In this chromosome build DLG2-iso7 or 8 were not included and therefore could not be included in the analysis. The presence of DLG2 L27 only (ENST00000472545) indicated that isoforms 7/8 were likely expressed, but not captured in this expression data using this chromosome build. Using 22 primary NB samples, we could confirm by qPCR the DLG2-iso2 expression observed in the TARGET dataset (Figure 2e). We could also confirm that there was no expression of isoforms 3 or 4 in our samples. Isoform 1 as a truncated variant of isoforms 7 and 8 (Figure 1b), could not be uniquely identified by qPCR when compared to isoforms 7/8, and since the isoform 1/7/8 qPCR result showed the same result as the specific isoform 7/8 qPCR, we concluded that isoform 1 was not expressed in our samples (data not shown). No variation in the expression of DLG2-iso2 (ENST00000398309) was observed between the stages (Figure 2e), consistent with Figure 2c. The DLG2-iso7/8 (ENST00000650630) transcript had decreased expression in the stage 4 tumors when compared to the stage 1 and 2 tumors (log2 FC= 3.1, p< 0.05), reflecting the difference in total DLG2 expression between differently staged NB (Figure 2e).
DLG4 showed no decrease in isoform 1 expression between the stages using the TARGET dataset. Furthermore, there was no change in total DLG4 expression level between stages (Figure 2f). We confirmed the isoform expression by using an independent Patient dataset (GSE16476), using microarray data (Figure 2g).
To evaluate the total DLG gene expression in NB we determined the relative expression of all DLG family members. DLG1, DLG2 and DLG3 all showed similar expression levels with DLG4 and DLG5 having significantly higher expression (p<0.001) (Figure 2h).
DLG2 expression correlated to LIN7 family gene expression and NB samples formed clusters.
The L27-domain enables binding to other L27-domain containing proteins. An important L27-containing scaffolding protein in signaling complex formation is the LIN7 protein family. The relationship between DLG2 and DLG1 gene expression and the various LIN7 binding partners was examined using primary tumor data taken from the Z score of 159 tumor data sets on the R2 Genomics Analysis and Visualization Platform (http://r2.amc.nl). A positive relationship (Y = 0.82x - 0.05, P<0.001) between DLG2 and LIN7A across tumor datasets could be confirmed (Figure 3a). Clusters were formed based on the spatial coordinates of DLG2 and LIN7A expression. Medulloblastoma (6/7), Ewings sarcoma (2/2), glioma (6/7), pheochromocytomas/paragangliomas (2/2) and NB (5/5) all showed high DLG2 expression as well as high LIN7A expression. The remaining tumors with similar expression included other tumors of the CNS such as glioblastoma, primitive neuroectodermal tumors (PNET) and other brain tumors. Squamous cell carcinoma (2/2) showed high DLG2 expression with low LIN7A expression. The remainder of the tumor dataset, consisting of lung-, colon-, ovarian-, and breast cancers and various lymphomas tended to show low expression of both DLG2 and LIN7A (Figure 3a). A weak linear relationship could be established between DLG1 and LIN7A (Figure 3b), however no distinct tumor clusters could be formed. A positive relationship (Y = 0.70x + 0.07, P<0.0001) could be established between DLG2 and LIN7B across tumor datasets (Figure 3c). Ewing’s sarcoma (2/2) and NB (5/5) clustered with high DLG2 expression as well as high LIN7B expression. No linear relationship (Y = 0.66x + 0.08, P=0.23) between DLG1 and LIN7B across tumor datasets could be confirmed (Figure 3d). A positive relationship (Y = 0.97x + 0.04, P<0.0001) between DLG2 and LIN7C between tumor datasets could be confirmed (Figure 3e). Ewings sarcoma (2/2) and NB (5/5) clustered with high DLG2 expression as well as high LIN7C expression. Squamous cell carcinoma (2/2) clustered with high DLG2 expression and low LIN7C expression (Figure 3e). A positive relationship (Y = 1.6x + 0.00, P<0.05) between DLG1 and LIN7C across tumor datasets could be confirmed (Figure 3d), however distinct tumor clusters were not formed.
DLG2-isoform 7 expression controlled LIN7A expression and the DLG2-isoform 7 encoded protein could bind to LIN7A
To further evaluate the relationship that was established in Figure 3a between DLG2 and LIN7A gene expression, we determined the expression of LIN7A and DLG2-iso7/8 in NB primary samples. A strong positive correlation (R2=0.89, Y = 1 .1x - 0.06, P<0.001) between the expression of DLG2-iso7/8 and LIN7A for 22 primary NB tumors of varying stages was detected (Figure 4a). To determine if the relationship was causal we over expressed DLG2-iso7 or knocked down DLG2 expression by siRNA treatment in SKNAS NB cells. When DLG2-iso7 was over expressed LIN7A expression increased, and LIN7A expression decreased following DLG2 silencing (Figure 4b). The result was confirmed on protein level by Western blot (Figure 4c). When LIN7A was over expressed or silenced by siRNA we saw no difference in total DLG2 expression (Figure 4d). To determine if DLG2-iso7 or DLG2-iso2 bound directly to LIN7A we performed co-immunoprecipitation using co-transfected HEK-293 cells, showing that DLG2-iso7 but not DLG2-iso2 could bind to LIN7A (Figure 4e). This was expected as DLG2-iso2 lack the L27-domain, and this is the only thing that differs between these two isoforms.
We determined that over expression of either DLG2-iso2 or DLG2-iso7 resulted in a decrease in the percentage of cells in G1 phase, as well as an increase in the number of cells in G2/M phase (Figure 4f). DLG2-iso7, but not DLG2-iso2, over expression resulted in an increase in the percentage (12.6%, p< 0.001) of cells in S phase when compared to the control (Figure 4f).
LIN7A expression was low in high staged tumors and over expression changed the growth behavior of NB cells.
To further investigate the importance of LIN7A we evaluated the association of LIN family expression with survival and INSS stage, using online microarray data in the NB patient dataset (GSE49710) obtained from the R2 Genomics Analysis and Visualization Platform (http://r2.amc.nl). The data was divided into survival outcome; alive or deceased. LIN7A (log2 fc =1.06, p< 0.001) showed a decrease in expression in the deceased patients compared to the patients that survived (Figure 5a). LIN7B (log2 fc =0.43, p= 0.09) and LIN7C (log2 fc =0.20, p=0.66) showed no difference in expression (Figure 5a). The expression of LIN7A was then stratified by INSS stage. Stage 4 tumors showed the lowest expression compared to stage 1 (log2 fc =0.44, p< 0.01), stage 2 (log2 fc =0.44, p< 0.001), stage 4s (log2 fc =0.25, p< 0.05) and stage 3 (log2 fc =0.50, p< 0.01) (Figure 5b). Over expression of LIN7A in NB cells (SKNAS) resulted in slower proliferation compared to the control (Figure 5c, p<0.001), and we observed a decrease in the number of viable cells (Figure 5d, p<0.001) and an increase in the non-viable cell fraction (Figure 5d, p<0.001) in cells with increased LIN7A expression. LIN7A silencing in SKNAS cells resulted in an increase in cell proliferation (Figure 5c, p<0.01), with an associated increase in viable cell number, no effect in the non-viable cell number was observed (Figure 5d). The LIN7A over expression after expression plasmid transfection, and LIN7A silencing by siRNA treatment of NB cells (SKNAS) was confirmed by qPCR (Figure 5e). We detected increased gene expression of BAX (Figure 5f), no alteration in BCL2 gene expression (Figure 5g) and an increase in the ratio of BAX/BCL2 (Figure 5h), indicating an increased level of apoptosis, when LIN7A was over expressed. The opposite was seen when LIN7A were silenced, then we detected a decrease in BAX gene expression (Figure 5f), increased BCL2 expression (Figure 5g) and a decrease in the BAX/BCL2 ratio (Figure 5h). This effect could also be confirmed on protein level by Western blot (Figure 5i).