ADVERTISEMENT
Importance of tumor sampling in transcriptomics-based risk stratification
Background
The variability of gene expression signatures across tumor sites has been previously documented [Stewart et al. 2017]. However, how this information can be exploited in practice is less obvious. To better understand the links between gene-based risk scores and tumor sampling, we performed morphology-based RNA extraction and we studied the risk scores' variability and their relative prognostic value in comparison with whole-tumor sampling. We deliberately avoided comparing the risk scores to each other, since our data did not allow significant comparisons.
Methods
From 99 colon tumors (hospital cohort), consecutive sections were used for whole-tumor and morphological-region RNA extraction. The following tumor morphological regions have been manually annotated in the virtual slides and macro-dissected: complex tubular, desmoplastic, mucinous, papillary, serrated and solid trabecular. Additionally, a number of tumor adjacent normal and stroma regions have been marked. To increase the statistical power, regions from the same tissue section were also considered grouped together according to their stromal and tumoral content. ESTIMATE [Yoshihara et al. 2013] was used to score the tumoral and stromal content from gene expression and Cox regression analysis was performed to assess the prognostic significance of the risk scores. 10 different risk scores were computed on the regions and whole tumor, respectively.
Results
In total, 99 whole tumor and 173 regional transcriptomics profiles were obtained, respectively. The patient ranking varied greatly across risk scores from whole-tumor (Spearman correlation -0.12 to 0.64), indicating a wide range of predictions. The morphological regions were grouped into stroma-rich (S) (e.g. desmoplastic) or tumor cell-rich (T) (e.g. serrated, solid trabecular) regions, according to their respective ESTIMATE scores. In general (7 out of 10), the risk scores were performing better in either S or T regions in comparison to whole-tumor performance. In multivariable analysis, including whole-tumor risk score and either S or T region scores, the regional scores were significantly better than whole-tumor scores (respective p < 0.05) for 5 out of 10 scores, in the other cases none of them achieving significance. Interestingly, selecting the worst prognosis among regions did not lead to an overall stronger predictor than either the S- or T-region one. The current limited cohort did not allow for the derivation and validation of a novel, region-based score.
Conclusions
The prognostic value of the risk scores varies across morphotypes and, in general, can be improved by a more targeted tumor sampling. Each score had a preference for one or the other type of regions (tumor- or stroma-rich), a consequence of their resptective derivation strategies. Consequently, a morphology-guided risk score construction may lead to stronger prognostic performance and a multi-region strategy may prove the most robust.
Legal entity responsible for the study
The authors.
Funding
This work was supported by Czech Science Foundation (GACR) through grant no. 19-08646S (VP).
Disclosures
All authors have declared no conflicts of interest.
