USP26 regulates TGF‐β signalling by deubiquitinating and stabilizing SMAD7; not applicable in glioblastoma

Abstract

Comment on “USP26 regulates TGF‐β signaling by deubiquitinating and stabilizing SMAD7” by Kit Leng Lui et al.

The ultimate significance of molecular signalling lies with its clinical applicability and druggability in disease treatment. Extensive research into TGF‐β signalling gives rise to many clinical trials, and among them over 20 are for cancer treatment 1. Yet, little success is derived, indicating an apparent gap between molecular understanding and clinical reality. A paper by Kit Leng Lui et al 2 reports the involvement of the deubiquitinating enzyme USP26 as a key modulator of TGF‐β activity through stabilisation of the negative feedback inhibitor Smad7. In an attempt to show clinical significance, Kit Leng Lui et al 2 further suggest that the loss of USP26 correlates with high TGF‐β activity and poorer patient survival/prognosis in glioblastoma. Given limited progress for glioblastoma treatment 3, the finding would have been an important contribution to the field. Their biochemical finding in vitro, however, is not applicable to glioblastoma as reported in the study.

The tumour‐promoting role of TGF‐β signalling has been well established in late‐stage cancer, promoting cell transdifferentiation and metastasis 4. Clinically, high TGF‐β expression and activation of the TGF‐β receptor downstream signalling pathways confer to poorer patient prognosis in aggressive cancers 5, 6. In glioblastoma, Bruna et al 7 reported that high pSmad2 levels conferred to poor prognosis in a subset of patients. Further studies have strengthened this association through the analysis of TGF‐β‐responsive gene levels as a measure of TGF‐β signalling activation 8, 9, 10. Here, the authors claim in glioblastomas: (i) there is a negative correlation between USP26 expression levels and Smad2 phosphorylation, (ii) the expression levels of USP26 and (iii) the expression levels of TGF‐β signalling components TβRI, TβRII and Smad7 correlate significantly with glioblastoma patients’ survival. We refute these conclusions based on finding inaccuracies and flaws in their data analysis and presentation.

After establishing the biochemical pathway where USP26 negatively regulates TGF‐β signalling by stabilising Smad7 in vitro, the study by Kit Leng Lui et al 2 looks into whether the finding is applicable to human glioblastoma. In Fig 5A of this study (represented as Fig 1A here), the authors present a negative correlation between pSMAD2 and USP26 levels in tissue microarrays from human glioblastoma patients. The authors report a ρ correlation of −0.15 with high statistical significance, P < 0.0001. The high degree of spread of data presented in the figure is unlikely to yield such a highly significant result, and to test the accuracy of this information, we extrapolated the values of this figure to rerun this statistical analysis (Fig 1B). Our Spearman's analysis showed no significant correlation (ρ = < 0.001; P > 0.99) between USP26 and pSmad2. While we do acknowledge there will be inaccuracies between our replicated data and the published results due to lack of access to the original raw data, the close match of each data point in our reproduction (Fig 1B) with those in the authors’ original figure (represented as Fig 1A) disputes the validity of the original finding. Our analysis gives rise to the conclusion that there is no evidence of a correlation between USP26 and pSmad2.

Figure 1. pSMAD2 levels do not associate with USP26 expression in glioblastoma.

(A) Scatterplot published by Kit Leng Lui et al 2 showing an association between USP26 expression and pSMAD2 in tissue microarrays from human glioblastoma patients (n = 36). Two‐tailed Spearman's statistical analysis was performed in this study 2. (B) A reproduction of the data presented in Kit Leng Lui et al 2 showing no association between USP26 expression and pSMAD2 levels. Two‐tailed Spearman's statistical analysis with 95% confidence intervals for ρ shown.

To further support the argument that USP26 is important for glioblastoma survival outcome, the authors have included Kaplan–Meier analysis using the REMBRANDT dataset (public access: http://www.betastasis.com) displayed in fig 5D (reproduced in Fig 2A) of this study. The overall survival presented in the Kaplan–Meier curves is grossly overestimated compared to the 16.7–17.6% 2‐year and 5.2–5.8% 5‐year survival rates commonly reported for glioblastoma 11. By reanalysing these figures using the same REMBRANDT dataset used by the authors, it becomes apparent that the reported survival curves represent a mixture of glioblastoma, oligodendroglioma and astrocytoma patients, not just glioblastomas. These tumours are clinically and molecularly distinct, and therefore, the data presented are a misrepresentation of glioblastoma. Additionally, the authors have selected a threshold to present the most significant association of the data. After adjusting the set threshold to the 25% (Fig 2B) and 75% quartile (Fig 2C), the association no longer remains significant. The selection of data by the authors based on the significance level has biased their conclusions and has not accurately represented the trends in USP26 expression and patient survival. This is particularly important since there is no change of USP26 expression levels across the whole spectra of brain tumours (including normal, Fig 3G).

Figure 2. USP26 gene expression weakly associates with glioblastoma patient survival.

Kaplan–Meier curves of brain tumour patient survival and USP26 expression using REMBRANDT database (http://www.betastasis.com). (A) Reproduction of Kaplan–Meier curve presented in Kit Leng Lui et al 2 fig 5D (all tumours; n = 329). (B, C) Kaplan–Meier curves for all tumours (n = 329) using 25% (B) and 75% (C) quartiles for USP26 expression threshold. (D) Kaplan–Meier curves for glioblastoma tumours (n = 178) using the same threshold set by Kit Leng Lui et al 2 (expression threshold ≈ 5.08). (E, F) Kaplan–Meier curves for glioblastoma tumours (n = 178) using 25% (E) and 75% (F) quartiles for USP26 expression threshold. (G, H) Kaplan–Meier curves for astrocytoma (n = 102) (G) and oligodendroglioma (n = 49) (H). P‐value was obtained by log‐rank test. Platform Affymetrix HG U133 v2.0 Plus.

Figure 3. Expression of TGF‐β signalling components TGFBR1, TGFBR2 and SMAD7 does not associate with clinical survival in glioblastoma but is altered in glioblastoma compared to lower grade brain tumours.

(A–C) Kaplan–Meier curves of glioblastoma patient survival (n = 178) using the REMBRANDT database (http://www.betastasis.com). A 50% gene expression threshold was used for survival comparisons. TGFBR1 (A), TGFBR2 (B) and SMAD7 (C). P‐values were obtained by a log‐rank test. (D–G) Boxplot graphs of mRNA expression by brain tumour type using the REMBRANDT database (http://www.betastasis.com). TGFBR1 (D), TGFBR2 (E), SMAD7 (F) and USP26 (G). Data were compared using Kruskal–Wallis statistical analysis with Dunn's multiple comparison testing. Adjusted P‐values presented; * < 0.05, ** <  0.01, *** < 0.001, **** < 0.0001. Platform Affymetrix HG U133 v2.0 Plus. GBM, glioblastoma.

To correct these biases and inaccuracies, we have reproduced the graph using only glioblastoma patients from the REMBRANDT dataset (Fig 2D–F). Our findings agree with the authors’ reported trend with low USP26 expression conferring to poorer prognosis, although the effect on survival is minor and the clinical relevance remains unclear. In addition, we analysed the association in astrocytoma (Fig 2G) and oligodendroglioma (Fig 2H) for which USP26 expression did not associate with patient survival. While there may be a relationship between USP26 and glioblastoma survival, the data presented do not provide strong evidence of an important function for tumour progression.

Further to the survival data presented for USP26 gene expression, we have also identified that the Kaplan–Meier curves presented in Fig EV5 (reproduced in Appendix Figs S1A, S2A and S3A) have also been analysed using the same mixture of brain tumours as glioblastoma. The authors have concluded using these data that TGFBR1, TGFBR2 or SMAD7 all significantly associate with clinical survival in glioblastoma. We have re‐analysed this data to test the accuracy of these statements.

The data presented for TGFBR1 and TGFBR2 by the authors show a significant association between what the authors refer to as glioblastoma patient survival and gene expression level (Appendix Figs S1A and S2A). Using the same threshold set by the authors, we have re‐analysed the gene expression survival for oligodendrogliomas, astrocytoma and glioblastoma separately. In contrast to the reported findings, TGFRB1 and TGFBR2 do not associate with glioblastoma patient survival (Appendix Figs S1B and S2B). This lack of association was also present when adjusting the threshold to the 50% and 75% quartiles (Fig 3A and B, Appendix Figs S1C and S2C). Only in oligodendroglioma and astrocytoma patients did TGFRB1 and TGFBR2 gene expression significantly associate with survival when applying the threshold set by the authors (Appendix Figs S1D and E, and S2D and E). The overall conclusions drawn by the authors are therefore due to the strong association of the TGF‐β receptor expression in the low‐grade gliomas but do not hold true for the high‐grade glioblastoma.

We next sought to test the reported association with clinical survival of SMAD7 gene expression in glioblastoma. When re‐analysed with the glioblastoma cohort at the same threshold set by the authors, we did not identify any significant association with survival (Appendix Figs S3A and B). Setting the threshold to the 75% quartile also did not associate with survival (Appendix Fig S3C); however, at the 50% quartile (Fig 3C) there was a slight association with glioblastoma survival that was statistically significant. Astrocytoma patients on the other hand significantly associated with SMAD7 expression levels (Appendix Fig S3D) which was not observed in the oligodendroglioma cohort (Appendix Fig S3E). The association reported by the authors is primarily attributed to its expression in the astrocytoma cohort with its association in glioblastoma remaining unclear.

The trends of the data suggest that the expression of the TGF‐β signalling genes varies between brain tumours. Using the REMBRANDT dataset, TGF‐β receptor expression is associated with poorer survival in oligodendroglioma and astrocytoma patients, but this did not translate to glioblastoma. Although the gene expression of TGF‐β receptors did not correlate with survival in glioblastoma, both TGFBR1 and TGFBR2 expression levels were found to be highly upregulated in glioblastoma compared to oligodendroglioma and astrocytoma tumours (Fig 3D and E). Furthermore, expression of inhibitory SMAD7 was also found to be downregulated in glioblastoma compared to these lower grade gliomas, suggesting enhanced activation of the TGF‐β signalling pathway (Fig 3F). These findings may indicate that dysregulation of TGF‐β signalling is a common feature in glioblastoma, while lower grade brain tumours have a wider range of expression with high TGFBR1, TGFBR2 and low SMAD7 expression conferring to more aggressive tumours. USP26 expression, however, is unchanged across these brain tumour types suggesting its expression may not be correlated with these TGF‐β signalling genes (Fig 3G).

In addressing the inaccurate clinical information provided in this study, we conclude that there is no evidence of an interaction between USP26 and TGF‐β signalling in glioblastoma patients. It is our opinion that gene expression of TGFBR1, TGFBR2 or SMAD7 does not associate with survival in glioblastoma in the REMBRANDT dataset.

To further examine the relationship of TGFBR1, TGFBR2 and SMAD7 expression in glioblastoma, we utilised the TCGA database (accessed via http://www.betastasis.com) and stratified patients based on the established glioblastoma molecular subtypes: classical, mesenchymal, proneural and neural 12. In these subtypes, TGFBR1 (Appendix Fig S4) and TGFBR2 (Appendix Fig S5) gene expression did not associate with survival; however, high SMAD7 expression (Appendix Fig S6) correlated with better survival in the mesenchymal subtype only. The importance of the molecular landscape and TGF‐β signalling has yet to be fully explored and may provide further context on its involvement in glioblastoma.

The reanalysis of the data has rebutted the conclusions put forth by the Kit Leng Lui et al 2. We do not identify any association between pSmad2 and USP26 levels, and the clinical gene expression has been misrepresented making their overall conclusion about a clinical association between USP26 and TGF‐β signalling invalid for glioblastoma. As such, the loss of USP26 does not correlate with high TGF‐β activity, nor confer poor prognosis in glioblastoma clinically. Overall, there is insubstantial evidence provided to implicate USP26 as an important factor in glioblastoma pathogenesis. Our own analysis of the data suggests that TGF‐β signalling should be examined by stratification of tumour grades and subtypes to more clearly identify its role in tumour progression.

Supporting information

Appendix

EMBO Reports (2020) 21: e47030