Page 24 - Read Online

Page 58 Ding et al. J Transl Genet Genom 2021;5:50-61 I http://dx.doi.org/10.20517/jtgg.2020.01

The relative abundance of immune cells for primary tumor samples from 1,232 patients in the 5 GRID data
sets is shown in Figure 4B. In young patients, there was a significantly (P = 0.028) greater abundance of
immune cells in primary tumors from patients with metastasis compared to primary tumor samples from
patients without metastasis; there were no significant metastasis-associated differences in the old patient
group. In middle-aged patients, abundance of immune cells from patients with metastasis was significantly
greater than that in the patients without metastasis (P < 0.0001, Supplementary Table 3). Four immune
cell types demonstrated striking age-related differences in abundance of immune cell type [Supplementary
Figure 3]. For the 687 patients from the four independent validation data sets (MC II, CC, TJU, and
MSKCC), the predicted iPAM risk scores for metastasis were significantly associated with the immune
-11
scores (Spearman rho = 0.25, P-value = 2.68 × 10 ), indicating that gene expression for the 36 iPAM genes
may capture information on immune responses in the tumor microenvironment that lead to metastasis.

DISCUSSION
Prostate-specific antigen screening has reduced death from CaP due to early detection while also leading
to over-treatment of low-risk CaP [24,25] . Three commercially available genomic classifiers (Oncotype
[15]
[26]
[27]
DX , Prolaris , and Decipher ) are used to predict metastatic CaP and guide initial treatment and/or
postoperative intervention. However, those classifiers were not designed for predicting metastatic disease
in young patients and accuracy of prediction for young men with CaP was not examined. This is the first
study to investigate young men with CaP as a distinct and unique patient group in both discovery and
validation of prediction signatures for early metastatic disease.

The accuracy of early metastasis prediction, measured by AUC of five-year survival ROC for the Decipher
in the three validation data sets (MC II, CC, and TJU) with follow-up time from date of RP are shown in
Supplementary Table 4. The AUC of five-year ROC generated from the iPAM classifier was higher than
that from the Decipher classifier. If only the six clinical variables were used to predict metastasis, AUC was
0.69 for both data sets; therefore, both the Decipher and iPAM classifiers showed substantial improvement
on prediction of early metastasis compared to the clinical classifier alone. Inclusion of the clinical variables
into the genomic classifier did not improve prediction accuracy for both classifiers. This suggests that both
genomic classifiers captured the prediction information provided by the clinical variables.

To develop the iPAM classifier, we selected DEGs associated with sample-type (tumor or benign) factor,
Gleason score factor, and metastasis factor. It is known that: (1) genes differentially expressed between
tumor and matched benign tissues reflect the genetic basis of tumorigenesis [28,29] ; and (2) genes differentially
expressed between low and high Gleason scores correlate with tumor aggressiveness . Therefore, in
[30]
addition to being used as prognostic markers, the iPAM genes selected from those DEGs are likely to
be functionally relevant to cancer progression. From Ingenuity Pathway Analysis (IPA), those DEGs
were enriched in pathways of immune response, cell adhesion, and degradation of extracellular matrix
[Supplementary Table 5]. Enrichment of pathways in immune response was among the up-regulated
DEGs identified only in the young group (highlighted pathways in Supplementary Table 5). Estimation
of abundance of immune cells in tumor and benign tissues from COH patients corroborated a role
of more pronounced immune responses to tumor development in young patients than in old patients
[Supplementary Table 2 and Figure 4A]. Ten of 36 iPAM genes were linked to immune-related pathways
[Table 3] and showed larger metastasis-associated differences (FDR < 0.10, Supplementary Table 6 in
the young group than in the old group from Decipher GRID samples [Figure 2C and Figure 2D and
Supplementary Figure 4]. Furthermore, the estimated abundance of immune cells in primary tumor samples
from the five Decipher GRID data sets was positively associated with the development of metastasis in
young and middle-aged patients with no significant association in old patients [Supplementary Table 3 and
Figure 4B]. The positive correlation between the immune scores (the abundance of immune cell types) and