Older age at diagnosis is consistently associated with worse clinical outcomes in prostate cancer. We sought to characterize gene expression profiles of prostate tumor tissue by age at diagnosis. We conducted a discovery analysis in The Cancer Genome Atlas prostate cancer dataset (n = 320; 29% of men >65 years at diagnosis), using linear regressions of age at diagnosis and mRNA expression and adjusting for TMPRSS2:ERG fusion status and race. This analysis identified 13 age-related candidate genes at FDR < 0.1, six of which were also found in an analysis additionally adjusted for Gleason score. We then validated the 13 age-related genes in a transcriptome study nested in the Health Professionals Follow-up Study and Physicians’ Health Study (n = 374; 53% of men >65 years). Gene expression differences by age in the 13 candidate genes were directionally consistent, and age at diagnosis was weakly associated with the 13-gene score. However, the age-related genes were not consistently associated with risk of metastases and prostate cancer–specific death. Collectively, these findings argue against tumor genomic differences as a main explanation for age-related differences in prostate cancer prognosis.

Prevention Relevance:

Older age at diagnosis is consistently associated with worse clinical outcomes in prostate cancer. This study with independent discovery and validation sets and long-term follow-up suggests that prevention of lethal prostate cancer should focus on implementing appropriate screening, staging, and treatment among older men without expecting fundamentally different tumor biology.

Prostate cancer predominantly affects older men, with over 60% of new diagnoses occurring in men ages 65 or above and almost a quarter occurring in those over 75 (1). With increasing life expectancy and the availability of PSA screening, the incidence and prevalence of prostate cancer among older men increases.

Older age at prostate cancer diagnosis is consistently associated with worse cancer-specific survival (2–4). It remains unclear whether this association is driven by differences in clinical management, including PSA screening practices, or underlying biological differences in cancers by age of diagnosis. Clinicians tend to underestimate life expectancy in patients presenting at an older age (5), with a higher prevalence of undertreatment in this patient group. Some cohorts report a complete attenuation of the association between age of diagnosis and cancer-specific mortality following adjustment for disease risk and treatment modality (2), supporting the notion that differences in prognosis by age are explained by differences in clinical management (6). Indeed, in a national Swedish cohort (4), the association between age at diagnosis and prostate cancer–specific mortality was primary explained by difference in clinical management rather than biology. However, it was not possible to completely exclude potential bias introduced by differences in treatment selection driven by age at diagnosis and/or residual or uncontrolled confounding. Alternatively, tumors in older men may be biologically similar but rather diagnosed at a more advanced stage. Gleason grade progression over time, potentially reflective of changes in biologic features, is uncommon (7).

Yet, older age at diagnosis has been demonstrated across multiple cohorts to be associated with higher risk tumor upstaging and upgrading following prostatectomy (8–10) suggesting a tendency toward behaving more aggressively. In the Göteborg-1 Prostate Cancer Screening Trial of repeated prostate screening, age was an independent risk factor for Gleason grade 7 or above, with a higher impact than the effect of screening round, year of screening, or number of screening rounds (11, 12). One hypothesis is that older patients harbor tumors that may have a distinct genetic profile compared with tumors diagnosed in younger men. Indeed, transcriptome signatures associated with age have been described both in murine and human physiologic prostate tissue (13, 14). In other cancers, tumor aggressiveness is known to vary with age types [e.g., poorer prognosis and possibly different biology in younger women with breast cancer (15); poorer prognosis in older patients in some hematologic malignancies (16)].

Here, we sought to characterize the transcriptomic profile of prostate tumor tissue by age at diagnosis. Differences in mRNA transcript expression could suggest distinct etiologic mechanisms and potential differential treatment sensitivity of tumors diagnosed in older men.

Study populations

The Cancer Genome Atlas (TCGA, the discovery set) is a primarily cross-sectional study that performed comprehensive molecular profiling of primary prostate tumor samples. mRNA sequencing used Illumina HiSeq and expressed measurements as reads per kilobase of transcript per million mapped reads (17).

The validation set consisted of prospective prostate cancer cohorts nested within the Physicians’ Health Study I (PHS-I, n = 22,071), which began in 1982 as a randomized primary prevention trial among male U.S. physicians, and the Health Professionals Follow-up Study (HPFS), an ongoing cohort study of 51,529 male U.S. health professionals that started in 1986. The research, conducted in accordance with the U.S. Common Rule, was approved by Institutional Review Boards at Harvard T.H. Chan School of Public Health and Brigham and Women's Hospital, and those of participating registries as required. Participants gave written informed consent.

The subset of men with mRNA expression data within the HPFS and PHS (HPFS/PHS) were sampled through a cumulative incidence-sampled case–control study, comparing lethal (metastases or prostate cancer–specific death) with nonlethal (no metastases after >8 years of follow-up) prostate cancers. Whole transcriptome gene expression profiling (Affymetrix Human Gene 1.0 ST microarray) with preprocessing as described previously (18) was available for 420 tumor tissue samples, of which 379 had known ERG protein expression, a proxy for the TMPRSS2:ERG fusion (19).

Discovery analysis (TGCA)

Genes with low expression in tumor tissue [i.e., <10 copies in >1/3 of samples (20)] were excluded from discovery analysis, leaving 14,276 of 20,502 genes for analysis. Heteroskedasticity in count data from RNA sequencing was addressed using precision weighting with voom (R package limma 3.44.2; ref. 21). Two main confounders were included in models for primary analysis: ERG fusion status, determined in TCGA by RNA sequencing, and self-reported race. Oncogenic activation of ERG through TMPRSS2:ETS fusions, a key driver in prostate cancer pathogenesis (22), has a distinct transcriptional profile (23). Furthermore, race and genetic ancestry are associated with differential tumor gene expression across a number of cancer types, including prostate cancer (24), and age at diagnosis differs by race. In a secondary analysis, models were additionally adjusted for Gleason score (6, 3+4, 4+3, 8, and 9–10, with categorical coding) to identify gene expression differences by age independent of grade.

Validation analysis (HPFS and PHS)

Validation was assessed as directionality in gene expression differences by age, with concordance in directionality assessed via Cohen's κ, and in magnitude of sign-based gene set expression differences. Gene scores and gene set scores were assessed through linear regression. In the validation set, no correction for multiple testing was applied because null-hypothesis testing is already questionable for result replication without such corrections (25).

Gene analysis

Candidate age-related genes were identified by linear regression analysis with age at diagnosis as the exposure (continuous), mRNA expression as the outcome (continuous, after log2 transformation), and the confounders noted above. Models used between-gene empirical Bayes adjustment as implemented in the limma package. Candidate genes were identified using a Benjamini–Hochberg FDR < 0.1. We performed two sets of sensitivity analyses. First, we repeated gene-level analyses with age coded as a binary variable with a cutoff at 65 years of age. Second, we repeated analyses stratified by Gleason grade group (1–2 = score 3+4 or below vs. 3–5 = score 3+4 or above).

Gene score

Following candidate gene identification, an age-related gene score was generated for each tumor sample using the más-o-menos method (26), generating a sum of all candidate gene levels per sample after standardization of expression levels to mean 0 and SD 1 for each gene and cohort (TCGA and HPFS/PHS). Signs for each candidate gene were determined on the basis of the directionality of their association with age at diagnosis in TCGA. The association between gene score and age at diagnosis was summarized using linear regression.

Gene set analysis

Gene set enrichment was conducted using Gene Ontology (GO) Biological Process gene sets from the Molecular Signatures Database, version 6.2 (27). Analysis used the R package Camera, a competitive gene set testing method accounting for intergene correlations (28). In benchmarking analyses, Camera performs well for identifying biologically relevant gene set rankings, allows for confounder adjustment, and can be integrated into reproducible analytic workflows (21, 29). Given that control of type 1 error rate is notoriously difficult in gene set analyses, including for Camera (29), a more stringent FDR of < 0.05 was used. Directionality reflects higher or lower pathway expression relative to all other competing gene sets; directionality does not directly equal functional activation or inhibition of a pathway.

Gene set score

Following candidate gene set analysis, an age-related gene set score was generated for each tumor. First, scores for each gene set were generated using más-o-menos, as described above for the gene score analysis. Second, individual gene set scores across all identified gene sets per tumor sample were summed into an overall age-related gene set score.

Replication of previously reported age-related gene signatures

We investigated two previously reported signatures of aging in prostate tissue. Crowell and colleagues have previously reported an age-related luminal progenitor signature of 16 differentially expressed genes in physiologic prostate tissue of 24-month-old mice compared with postpubertal, 3-month-old young adult mice (13).

Liu and colleagues have reported a human age-related inflammatory gene signature of 21 genes in a progenitor-like subset of CD38lo luminal prostate cells (14) which was associated with worse prognosis in patients (30, 31).

We calculated scores for these gene signatures in TCGA and HPFS/PHS cohorts using más-o-menos.

Impact of age at diagnosis as a confounder

In HPFS/PHS, we compared the association between gene expression and lethal disease between unadjusted logistic regression models and those adjusted for age at diagnosis as a continuous (linear) term. For comparison, we repeated this analysis with adjustment for a random number drawn from a normal distribution.

Similarly, we compared the top-ranking gene sets for lethal disease and for Gleason score (≥ 4+3 vs. ≤ 3+4) from gene set analysis, using Camera as above, between models with and without adjustment for age at diagnosis.

Data availability

Data are available through a project proposal for the HPFS (https://sites.sph.harvard.edu/hpfs/for-collaborators).

Study population

The discovery set, TCGA, included 320 men with prostate cancer after applying exclusions (mainly due to lower quality RNA data; Supplementary Fig. S1). Median age at diagnosis was 62 years (interquartile range: 56–66). Prostate cancers diagnosed at older ages were more often in men self-identifying as White and accompanied by higher PSA at diagnosis, higher Gleason scores, and no TMPRSS2:ERG fusions (Table 1).

Table 1.

Characteristics of men with primary prostate cancer and high-quality gene expression data in TCGA, left; n = 320) and the HPFS/PHS (right; n = 374)a.

TCGAHPFS/PHS
Age at diagnosis (yrs)≤5555–65>65≤5555–65>65
N 74 152 94 31 144 199 
Age at diagnosis (yrs) 52 (49, 54) 61 (58, 63) 68 (66, 70) 53 (51, 55) 62 (60, 64) 69 (67, 72) 
Self-reported raceb 
 Asian 4 (5.4%) 2 (1.3%) 2 (2.1%)    
 Black 18 (24%) 17 (11%) 8 (9%)    
 White 52 (70%) 133 (88%) 84 (89%) 30 (100%) 140 (100%) 192 (100%) 
 Unknown    
PSA at diagnosis (ng/mL) 6.8 (5.0, 14.0) 7.2 (5.1, 10.9) 8.2 (5.2, 12.8) 8.1 (4.9, 11.8) 7.6 (5.3, 11.8) 8.0 (5.6, 12.5) 
Unknown 26 75 36 13 36 
Gleason grade 
 <7 21 (28%) 27 (18%) 11 (12%) 7 (23%) 25 (17%) 26 (13%) 
 3+4 24 (32%) 43 (28%) 31 (33%) 13 (42%) 52 (36%) 63 (32%) 
 4+3 12 (16%) 39 (26%) 26 (28%) 9 (29%) 37 (26%) 51 (26%) 
 8 5 (7%) 25 (16%) 13 (14%) 0 (0%) 15 (10%) 24 (12%) 
 9–10 12 (16%) 18 (12%) 13 (14%) 2 (6%) 15 (10%) 35 (18%) 
Clinical stage 
 T1/T2 47 (72%) 91 (72%) 44 (63%) 28 (97%) 129 (92%) 164 (86%) 
 T3 6 (9%) 11 (9%) 10 (14%) 1 (3%) 6 (4%) 11 (6%) 
 T4/N1 12 (18%) 24 (19%) 16 (23%) 0 (0%) 3 (2%) 1 (1%) 
 M1 0 (0%) 1 (1%) 0 (0%) 0 (0%) 2 (1%) 14 (7%) 
 Unknown 25 24 
ERG statusc 
 ERG-positive 40 (54%) 68 (45%) 38 (40%) 17 (55%) 81 (56%) 88 (44%) 
 ERG-negative 34 (46%) 84 (55%) 56 (60%) 14 (45%) 63 (44%) 111 (56%) 
TCGAHPFS/PHS
Age at diagnosis (yrs)≤5555–65>65≤5555–65>65
N 74 152 94 31 144 199 
Age at diagnosis (yrs) 52 (49, 54) 61 (58, 63) 68 (66, 70) 53 (51, 55) 62 (60, 64) 69 (67, 72) 
Self-reported raceb 
 Asian 4 (5.4%) 2 (1.3%) 2 (2.1%)    
 Black 18 (24%) 17 (11%) 8 (9%)    
 White 52 (70%) 133 (88%) 84 (89%) 30 (100%) 140 (100%) 192 (100%) 
 Unknown    
PSA at diagnosis (ng/mL) 6.8 (5.0, 14.0) 7.2 (5.1, 10.9) 8.2 (5.2, 12.8) 8.1 (4.9, 11.8) 7.6 (5.3, 11.8) 8.0 (5.6, 12.5) 
Unknown 26 75 36 13 36 
Gleason grade 
 <7 21 (28%) 27 (18%) 11 (12%) 7 (23%) 25 (17%) 26 (13%) 
 3+4 24 (32%) 43 (28%) 31 (33%) 13 (42%) 52 (36%) 63 (32%) 
 4+3 12 (16%) 39 (26%) 26 (28%) 9 (29%) 37 (26%) 51 (26%) 
 8 5 (7%) 25 (16%) 13 (14%) 0 (0%) 15 (10%) 24 (12%) 
 9–10 12 (16%) 18 (12%) 13 (14%) 2 (6%) 15 (10%) 35 (18%) 
Clinical stage 
 T1/T2 47 (72%) 91 (72%) 44 (63%) 28 (97%) 129 (92%) 164 (86%) 
 T3 6 (9%) 11 (9%) 10 (14%) 1 (3%) 6 (4%) 11 (6%) 
 T4/N1 12 (18%) 24 (19%) 16 (23%) 0 (0%) 3 (2%) 1 (1%) 
 M1 0 (0%) 1 (1%) 0 (0%) 0 (0%) 2 (1%) 14 (7%) 
 Unknown 25 24 
ERG statusc 
 ERG-positive 40 (54%) 68 (45%) 38 (40%) 17 (55%) 81 (56%) 88 (44%) 
 ERG-negative 34 (46%) 84 (55%) 56 (60%) 14 (45%) 63 (44%) 111 (56%) 

aData are median (interquartile range) or n (%).

bHPFS/PHS was restricted by race for confounding control in this analysis.

cERG status is from TMPRSS2:ERG fusion calls based on RNA sequencing (TCGA) or from genomically validated IHC for ERG protein expression (HPFS/PHS).

The validation set from HPFS/PHS, where only few men self-reported Asian or Black race (n = 5), was restricted to White participants for confounding control (n = 374). HPFS/PHS participants with a median age of 66 years (interquartile range: 62–69) tended to be older at diagnosis than TCGA participants. Similar to TCGA, prostate cancers diagnosed at older ages tended to have a higher Gleason score and be more frequently ERG negative, without noticeable differences in PSA at diagnosis (Table 1). Within the validation set, men who were older at diagnosis had a higher risk of lethal disease [Supplementary Table S1; odds ratio: 1.76 per decade of age at diagnosis, 95% confidence interval (CI): 1.20–2.61].

Gene-level analysis

In the primary discovery analysis on an individual-gene level (adjusted for ERG status and race), 13 candidate genes were identified as differentially expressed by age at diagnosis in TCGA at FDR < 0.1 (Table 2; Fig. 1). In the secondary discovery analysis with additional adjustment for Gleason grade group, six genes were identified at FDR < 0.1, all of which were also identified in the primary analysis. By design, an age-related gene score based on the 13 candidate genes in primary analysis was correlated with age at diagnosis (Pearson r: 0.39; 95% CI, 0.29–0.48; Fig. 1A), as was the score based on the six genes from models with additional grade adjustment (r: 0.35; 95% CI, 0.25–0.45; Fig. 1C).

Table 2.

Age at cancer diagnosis (continuous) and differentially expressed genes after adjusting for TMPRSS2:ERG status and self-reported race (left) and additionally adjusting for Gleason grade groups (right).

ERG/race-adjustedaERG/race/Gleason-adjustedb
TCGAHPFS/PHSTCGAHPFS/PHS
GeneRatiocFDRdDirectionePGeneRatiocFDRdDirectioneP
ADAP2f 1.22 0.003 Down 0.85 NOP2f 1.14 0.029 Up 0.17 
MNXf 1.79 0.003 Up 0.16 TMEM201f 1.09 0.029 Up 0.77 
HLX 1.24 0.011 Up 0.97 ADAP2f 1.20 0.029 Down 0.82 
SLC10A5f 1.33 0.011 Up 0.74 MNX1f 1.66 0.029 Up 0.18 
TMEM201f 1.09 0.036 Up 0.95 TCOF1f 1.08 0.08 Up 0.27 
NOP2f 1.12 0.040 Up 0.16 SLC10A5f 1.29 0.09 Up 0.78 
ORM2 0.48 0.040 Down 0.54      
CX3CR1 1.34 0.040 Up 0.76      
TCOF1f 1.08 0.040 Up 0.29      
GATAD2A 1.07 0.040 Up 0.75      
LIMK1 1.09 0.06 Up 0.19      
FCGR1A 1.28 0.07 Up 0.78      
MDFI 0.71 0.07 Down 0.17      
ERG/race-adjustedaERG/race/Gleason-adjustedb
TCGAHPFS/PHSTCGAHPFS/PHS
GeneRatiocFDRdDirectionePGeneRatiocFDRdDirectioneP
ADAP2f 1.22 0.003 Down 0.85 NOP2f 1.14 0.029 Up 0.17 
MNXf 1.79 0.003 Up 0.16 TMEM201f 1.09 0.029 Up 0.77 
HLX 1.24 0.011 Up 0.97 ADAP2f 1.20 0.029 Down 0.82 
SLC10A5f 1.33 0.011 Up 0.74 MNX1f 1.66 0.029 Up 0.18 
TMEM201f 1.09 0.036 Up 0.95 TCOF1f 1.08 0.08 Up 0.27 
NOP2f 1.12 0.040 Up 0.16 SLC10A5f 1.29 0.09 Up 0.78 
ORM2 0.48 0.040 Down 0.54      
CX3CR1 1.34 0.040 Up 0.76      
TCOF1f 1.08 0.040 Up 0.29      
GATAD2A 1.07 0.040 Up 0.75      
LIMK1 1.09 0.06 Up 0.19      
FCGR1A 1.28 0.07 Up 0.78      
MDFI 0.71 0.07 Down 0.17      

aAll 13 genes with FDR < 0.1.

bAll six genes with FDR < 0.1.

cPer decade of age at diagnosis.

dFDR among all 14,276 genes in the discovery set (TCGA).

eDirection of expression difference and unadjusted P value in the validation set (HPFS/PHS) with microarray profiling. Direction is shown instead of ratios because of different dynamic range in expression levels compared with RNA sequencing in TCGA.

fGene appears in both analyses.

Figure 1.

Age at diagnosis and age-related gene scores, adjusting for TMPRSS2:ERG status and race (A and B) and additionally adjusting for Gleason score (C and D), for the discovery set (TCGA, A and C) and the validation set (HPFS/PHS, B and D).

Figure 1.

Age at diagnosis and age-related gene scores, adjusting for TMPRSS2:ERG status and race (A and B) and additionally adjusting for Gleason score (C and D), for the discovery set (TCGA, A and C) and the validation set (HPFS/PHS, B and D).

Close modal

In the validation set, restricted to 374 tumors with known ERG status, the correlation of the gene score with age at diagnosis was considerably attenuated both for score based on the 13 candidate genes (r: 0.10; 95% CI, 0.00–0.20; Fig. 1B) and the score based on the six genes (r: 0.08, 95% CI, −0.02 to 0.18; Fig. 1D). Directionality of difference in expression by age was concordant for 12 of the 13 individual candidate genes (Cohen's κ: 0.75; 95% CI, 0.33–1.00); null-hypothesis tests per gene had 0.17 ≤ P ≤ 0.97 (Table 2).

Associations of the 13 candidate genes with lethal prostate cancer had mostly null point estimates, and CIs were consistently not compatible with strong associations for any gene (Fig. 2A).

Figure 2.

A, Age-related genes (in SDs) and odds of lethal disease (metastases/prostate cancer death vs. absence of metastases over >8 years of follow-up) in HPFS/PHS. Genes marked with an asterisk (*) appear in both the ERG/race-adjusted and the ERG/race/Gleason-adjusted analysis. B, Gene set enrichment for GO biological pathways from ERG/race-adjusted analyses. The x axis has FDRs from TCGA (discovery set), with FDR correction among all gene sets; the y axis has FDRs from HPFS/HPFS (validation set), with FDR correction among all gene sets with FDR < 0.05 in TCGA, as in Table 3. Labeled gene sets are those with FDR < 0.1 in both sets and consistent directionality.

Figure 2.

A, Age-related genes (in SDs) and odds of lethal disease (metastases/prostate cancer death vs. absence of metastases over >8 years of follow-up) in HPFS/PHS. Genes marked with an asterisk (*) appear in both the ERG/race-adjusted and the ERG/race/Gleason-adjusted analysis. B, Gene set enrichment for GO biological pathways from ERG/race-adjusted analyses. The x axis has FDRs from TCGA (discovery set), with FDR correction among all gene sets; the y axis has FDRs from HPFS/HPFS (validation set), with FDR correction among all gene sets with FDR < 0.05 in TCGA, as in Table 3. Labeled gene sets are those with FDR < 0.1 in both sets and consistent directionality.

Close modal

In sensitivity analyses, we performed separate discovery and validation analyses among men with ERG-positive tumors (with TMPRSS2:ERG fusion) and ERG-negative tumors (Supplementary Table S2). Interestingly, there was almost no overlap between the age-associated candidate genes identified in the main analysis adjusting for ERG status compared with those identified in ERG-stratified analyses, with the exception of MNX1 in ERG-negative tumors. Results from the validation set in this stratified analysis were inconclusive. A sensitivity analysis with age at diagnosis coded as a binary variable (≥65 years vs. <65 years; Supplementary Table S3) had similar results to the main analysis that considered age as a continuous variable. A sensitivity analysis with stratification by Gleason grade group (1–2 = score 3+4 or below vs. 3–5 = score 3+4 or above) did not identify age-related genes either in low-grade or in high-grade tumors (Supplementary Table S4).

Gene set analysis

In a discovery analysis on a gene-set level in TCGA, among 4,436 GO Process gene sets, 76 gene sets were differentially expressed by age at diagnosis beyond other genes sets at FDR < 0.05 in ERG- and race-adjusted models (Table 3). With additional adjustment for grade, 36 gene sets were differentially expressed, with considerable overlap among the top 10 gene sets (Table 3). By design, age-related gene set scores based on the 76 or 36 gene sets were (weakly) correlated with age at diagnosis (Supplementary Fig. S2A and S2C).

Table 3.

Gene set analysis for GO biological process gene sets associated with age at diagnosis, from models adjusted for TMPRSS2:ERG status and race (top) and models additionally adjusted for Gleason score (bottom).

TCGAHPFS/PHS
#Gene SetGenesaDirectionFDRbDirectionFDRb
ERG/race-adjustedc 
GO Sterol Biosynthetic Process 38/42/43 Down <0.001 Up 0.93 
GO Actin Myosin Filament Sliding 16/38/38 Down <0.001 Down 0.87 
GO Muscle Contraction 156/227/233 Down 0.004 Down 0.56 
GO Protein Folding in Endoplasmic Reticulum 10/8/11 Down 0.004 Down 0.10 
GO Multi Organism Metabolic Process 136/135/138 Up <0.001 Down 0.009 
GO Nuclear Transcribed mRNA Catabolic Process Nonsense Mediated Decay 113/117/118 Up <0.001 Down 0.002 
GO Ribosome Biogenesis 290/294/308 Up <0.001 Down 0.24 
GO Translational initiation 141/143/146 Up 0.001 Down 0.002 
GO rRNA Metabolic Process 240/245/255 Up 0.001 Down 0.17 
10 GO Ribosomal Large Subunit Biogenesis 47/47/49 Up 0.002 Down 0.16 
ERG/race/Gleason-adjustedd 
GO Sterol Biosynthetic Process 38/42/43 Down <0.001 Down 0.85 
GO Actin Myosin Filament Sliding 16/38/38 Down <0.001 Up 0.82 
GO Alpha Linolenic Acid Metabolic Process 12/13/13 Down 0.002 Down 0.42 
GO Ribosome Biogenesis 290/294/308 Up <0.001 Down 0.11 
GO rRNA Metabolic Process 240/245/255 Up <0.001 Down 0.09 
GO Multi Organism Metabolic Process 136/135/138 Up <0.001 Down 0.001 
GO Nuclear Transcribed mRNA Catabolic Process Nonsense Mediated Decay 113/117/118 Up <0.001 Down <0.001 
GO Translational Initiation 141/143/146 Up <0.001 Down <0.001 
GO Ribosomal Large Subunit Biogenesis 47/47/49 Up <0.001 Down 0.08 
10 GO Ribonucleoprotein Complex Biogenesis 407/412/440 Up <0.001 Down 0.18 
TCGAHPFS/PHS
#Gene SetGenesaDirectionFDRbDirectionFDRb
ERG/race-adjustedc 
GO Sterol Biosynthetic Process 38/42/43 Down <0.001 Up 0.93 
GO Actin Myosin Filament Sliding 16/38/38 Down <0.001 Down 0.87 
GO Muscle Contraction 156/227/233 Down 0.004 Down 0.56 
GO Protein Folding in Endoplasmic Reticulum 10/8/11 Down 0.004 Down 0.10 
GO Multi Organism Metabolic Process 136/135/138 Up <0.001 Down 0.009 
GO Nuclear Transcribed mRNA Catabolic Process Nonsense Mediated Decay 113/117/118 Up <0.001 Down 0.002 
GO Ribosome Biogenesis 290/294/308 Up <0.001 Down 0.24 
GO Translational initiation 141/143/146 Up 0.001 Down 0.002 
GO rRNA Metabolic Process 240/245/255 Up 0.001 Down 0.17 
10 GO Ribosomal Large Subunit Biogenesis 47/47/49 Up 0.002 Down 0.16 
ERG/race/Gleason-adjustedd 
GO Sterol Biosynthetic Process 38/42/43 Down <0.001 Down 0.85 
GO Actin Myosin Filament Sliding 16/38/38 Down <0.001 Up 0.82 
GO Alpha Linolenic Acid Metabolic Process 12/13/13 Down 0.002 Down 0.42 
GO Ribosome Biogenesis 290/294/308 Up <0.001 Down 0.11 
GO rRNA Metabolic Process 240/245/255 Up <0.001 Down 0.09 
GO Multi Organism Metabolic Process 136/135/138 Up <0.001 Down 0.001 
GO Nuclear Transcribed mRNA Catabolic Process Nonsense Mediated Decay 113/117/118 Up <0.001 Down <0.001 
GO Translational Initiation 141/143/146 Up <0.001 Down <0.001 
GO Ribosomal Large Subunit Biogenesis 47/47/49 Up <0.001 Down 0.08 
10 GO Ribonucleoprotein Complex Biogenesis 407/412/440 Up <0.001 Down 0.18 

aNumber of genes expressed in TCGA samples measured with RNA sequencing/Number of genes measured on HPFS/PHS Affymetrix microarray/Total genes in this GO Biological Process gene set.

bFDR among all 4,436 gene sets.

cTop 10 gene sets out of 76 gene sets with FDR < 0.05.

dTop 10 gene sets out of 36 gene sets with FDR < 0.05.

In the validation set (HPFS/PHS), the gene set scores were not correlated with age at diagnosis (Supplementary Fig. S2B and S2D). Among the 76 candidate gene sets, only two had consistent directionality and FDR < 0.1 across both cohorts: protein folding in endoplasmic reticulum (downregulated) and cellular response to zinc ion (downregulated; Fig. 2B). In 38 of 76 gene sets (50%), directionality of age-related expression differences was opposite in the discovery and validation set.

Replication of previously reported age-related gene signatures

Neither of the gene signatures previously described by Crowell and colleagues (13) or Liu and colleagues (14) was positively correlated with age at diagnosis in TCGA or HPFS/PHS (Supplementary Fig. S3).

Impact of age at diagnosis adjustment on association of gene expression and lethal disease

For association of expression levels of each individual gene and lethal disease, odds ratios were >10% different between unadjusted models and models adjusted for age at diagnosis in 1,924 (12%) of the 16,692 expressed genes (Fig. 3A). However, when adjusted for a random number, a similar proportion of estimates were >10% different (2,169 genes; 13%).

Figure 3.

A, Impact of adjusting for the confounder age at diagnosis on OR estimates for the association of gene expression (comparing extreme quartiles) and lethal disease (metastases/prostate cancer death vs. absence of metastases after at least 8 years of follow-up) among all 16,692 expressed genes, of which 1,924 genes showed >10% change in estimates when adjusting for age at diagnosis (left) and 2,169 genes showed >10% change when adjusting for a random number (right). Example genes with the largest changes in estimates are highlighted. Concordance between the top 50 GO biological process gene sets associated with lethal disease (B and C) and with Gleason score (D and E) when comparing models that are unadjusted and those adjusted for age at diagnosis in HPFS/PHS. B and C, show rank per gene sets from unadjusted and age-adjusted models. D and E, are concordance-at-the-top plots, that is, the concordance of the highest-ranked gene sets between unadjusted and adjusted models; the dashed line indicates agreement by chance.

Figure 3.

A, Impact of adjusting for the confounder age at diagnosis on OR estimates for the association of gene expression (comparing extreme quartiles) and lethal disease (metastases/prostate cancer death vs. absence of metastases after at least 8 years of follow-up) among all 16,692 expressed genes, of which 1,924 genes showed >10% change in estimates when adjusting for age at diagnosis (left) and 2,169 genes showed >10% change when adjusting for a random number (right). Example genes with the largest changes in estimates are highlighted. Concordance between the top 50 GO biological process gene sets associated with lethal disease (B and C) and with Gleason score (D and E) when comparing models that are unadjusted and those adjusted for age at diagnosis in HPFS/PHS. B and C, show rank per gene sets from unadjusted and age-adjusted models. D and E, are concordance-at-the-top plots, that is, the concordance of the highest-ranked gene sets between unadjusted and adjusted models; the dashed line indicates agreement by chance.

Close modal

Comparing gene set analyses for lethal disease (Fig. 3B and D; Supplementary Table S5) and for Gleason score (Fig. 3C and E; Supplementary Table S6) generally showed high concordance among top-ranking gene sets regardless of whether analysis was done with or without adjustment for age at diagnosis.

This study characterized transcriptome profiles of prostate cancer tumor tissue by age at diagnosis, using independent cohorts for discovery and validation. We identified 13 age-related candidate genes that were differentially expressed by age at diagnosis in our discovery set and in the validation set, albeit substantially attenuated in the latter. Whether six of these genes were also differentially expressed by age when comparing tumors of the same Gleason score is less clear, given the inconclusive results in the validation set. However, the 13 genes were not clearly associated with progression to lethal disease, and thus the slight transcriptomic differences cannot explain the considerably worse prognosis of older men with prostate cancer.

The age-related genes identified here encode proteins credentialed as involved in prostate carcinogenesis. This applies to 12 of 13 candidate genes: MNX1 (32–34), HLX (35, 36), TCOF1 (37), GATAD2A (38), MDFI (39, 40), all have functions in DNA binding and/or transcriptional regulation; ADAP2 (41, 42), TMEM201 (43), LIMK (44) are involved in cytoskeletal scaffolding and cell movement; ORM2 (45), CX3CR1 (46), and FCGR1A (47) are involved in immune regulation; NOP2 (48, 49) is involved in cell cycle regulation. Functions include: MNX1, previously suggested to have higher expression in tumors from African-American men than European-American men, is upregulated by androgen and AKT signaling (50, 51) and promotes proliferation, migration, and invasion in cell lines (52). TCOF1 encodes a potential substrate for PIM1, a kinase with higher expression in metastatic tissue relative to primaries that promotes and maintains prostate tumorigenesis (53). HLX expression has been implicated with development of androgen resistance in cell lines (54). GATAD2A harbors genetic susceptibility loci for prostate cancer and other hormone-related cancers (55). MDFI is downregulated by miR-145, a miRNA that inhibits proliferation, migration, and invasion in cell lines (56). LIMK1 expression, previously reported to the associated with prognosis (57), is involved in pathways involving altering cell morphology and migration (58, 59). CX3CR1 is upregulated in malignant transformation and in response to tissue hypoxia (60, 61) and has been implicated in adhesion of prostate cancer cells to the bone marrow endothelium (62). Finally, NOP2 has also been implicated in cell invasion, migration, and metastasis (63). Collectively, the identified and validated genes from our analysis have credence for being involved in prostate carcinogenesis. In addition, functionally related genes may not have been identified because of measurement error and limited power.

Our finding of age-related gene expression differences is consistent with reports from previous studies, including across cancer sites in TCGA, that demonstrated differences in prevalence of genomic instability, somatic copy-number alterations, and somatic mutations (64, 65), and as well as global transcriptomic changes (64). A study in patients with clinical DNA panel sequencing demonstrated more common TMPRSS2:ERG fusions in tumors from men with early-onset cancer and potential additional differences in AR copy numbers as well as SPOP and ASXL1 mutations (66). Another study demonstrated a positive association between age at diagnosis and a clinically used prognostic genomic classifier, the Decipher score (67), which indicates more aggressive disease. However, this association was attenuated after adjustment for clinical factors (67). It is possible that the unadjusted association resulted from differential use of the genomic test by age, for example, physicians requesting the test in older patients particularly if they had more aggressive disease. To gauge internal and external validity of such findings, replication is useful. Our study benefitted from discovery and validation sets with distinct patient populations, different techniques for tissue handling (e.g., fresh frozen tissue vs. formalin-fixed, paraffin-embedded) and different platforms quantification of gene expression [microarray vs. RNA sequencing, the latter having a higher sensitivity in identifying differentially expressed genes (68, 69)]. We conjecture that replication of gene scores is more likely to reflect a robust finding. The lack of replication of other, previously reported age-related gene signatures (13, 14) could be explained by limited power, splice variants, differences between murine models and human tissues, differences between aging in normal compared with tumor tissue, confounding due to patient characteristics or cell subpopulation, relevance only to a specific subset of patients (effect modification), or that original findings were due to chance.

That age-related genes were not associated with lethal prostate cancer is relevant for other studies of gene expression and lethal disease. While adjusting for age in such analyses frequently can lead to changes in estimates, such changes are not proof that age is a relevant confounder. Indeed, similar magnitudes of change occurred after adjustment for a random number, and can result from noncollapsibility of odds ratios and hazard ratios (70, 71). Consequently, at least in analyses of gene expression and lethal disease in this study population, little if any bias ensues if age is omitted from regression models.

Our study is subject to potential limitations. First, we relied on tumors for which gene expression data were successfully obtained. Selection mechanisms from the target population of men with prostate cancer are complex and may attenuate true associations between age at diagnosis and gene expression or induce spurious associations. This consideration applies to the hospital-based design of TCGA and to the cumulative incidence-sampled case–control study within HPFS/PHS. Second, in our main analyses, we made an implicit assumption of linearity, such that transcriptional differences between, for example, ages 55 and 60 years would be identical to those between 75 and 80 years. However, similar genes were identified when comparing transcriptomes between men 65 years or older with those who were younger at cancer diagnosis. Third, our main analyses did not account for molecular subtypes of prostate cancer as potential effect modifiers. Our data suggest that age-related genes could differ, for example, between tumors with and without the TMPRSS2:ERG fusion, but stratified analyses had small sample sizes. Fourth, the findings from gene set analyses should be interpreted with caution, given the frequent inconsistency in directionality upon validation. Transcriptional signatures of prostatic aging may not be well captured by GO biological processes (27). Fifth, age-related changes on stromal tissue around the epithelial tumor nodule are not assessed here. Sixth, our study consisted predominantly of White men. Finally, the validity and reproducibility of our findings is reliant on the quality of genomic data in our study population datasets. Most of these points could be addressed in future studies in other populations.

In summary, in this study of gene expression in prostate cancer tissue in two large independent cohorts, we identified and validated an age-related transcriptional signature that is weakly associated with older age at diagnosis in prostate cancer. However, the identified signature was not clearly associated with risk of lethal disease and is unlikely to explain age-related differences in survival after prostate cancer diagnosis.

A. Plym reports grants from the Prostate Cancer Foundation (PCF), the Swedish Cancer Society, and the Swedish Society for Medical Research during the conduct of the study. S. Tyekucheva reports grants from NCI during the conduct of the study. K.L. Penney reports grants from NIH during the conduct of the study. L.A. Mucci reports grants from NCI and Prostate Cancer Foundation during the conduct of the study; other support from AstraZeneca and Janssen; personal fees from Bayer and Convergent Therapeutics outside the submitted work. No disclosures were reported by the other authors.

C.D. Zhou: Conceptualization, formal analysis, funding acquisition, writing–original draft. A. Pettersson: Conceptualization, investigation, writing–review and editing. A. Plym: Data curation, validation, investigation, writing–review and editing. S. Tyekucheva: Data curation, investigation, writing–review and editing. K.L. Penney: Investigation, writing–review and editing. H.D. Sesso: Investigation, writing–review and editing. P.W. Kantoff: Investigation, writing–review and editing. L.A. Mucci: Conceptualization, investigation, writing–review and editing. K.H. Stopsack: Conceptualization, formal analysis, supervision, funding acquisition, writing–review and editing.

We thank the participants and staff of the HPFS and the PHS for their valuable contributions. In particular, we would like to recognize the contributions of Elizaveta Gazeeva, Siobhan Saint-Surin, Ruifeng Li, Betsy Frost-Hawes, Ann Fisher, and Eleni Konstantis. We appreciate the support from the Channing Division of Network Medicine, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, MA. In addition, we would like to thank the following state cancer registries for their help: AL, AZ, AR, CA, CO, CT, DE, FL, GA, ID, IL, IN, IA, KY, LA, ME, MD, MA, MI, NE, NH, NJ, NY, NC, ND, OH, OK, OR, PA, RI, SC, TN, TX, VA, WA, and WY. The authors assume full responsibility for analyses and interpretation of these data.

HPFS is supported by the NCI (U01 CA167552). This research was supported in part by the NCI through the Specialized Programs of Research Excellence program in Prostate Cancer (P50 CA090381), Cancer Center Support Grants (P30 CA008748, P30 CA006516), and research grants (R01 CA136578, to L.A. Mucci; 5R37 CA227190-02, to S. Tyekucheva, K.L. Penney, and L.A. Mucci). C.D. Zhou was supported by funding from the Fulbright Elsevier Data Analytics Award, the Harvard Horace Cecil Fisher Scholarship and the BUNAC Educational Scholarship Trust (BEST) Scholarship. A. Plym, K.L. Penney, L.A. Mucci, and K.H. Stopsack are Prostate Cancer Foundation Young Investigators. A. Plym was further supported by the Swedish Society of Medical Research. The Department of Defense supported K.H. Stopsack (W81XWH-18–1-0330).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.

Note: Supplementary data for this article are available at Cancer Prevention Research Online (http://cancerprevres.aacrjournals.org/).

1.
Heinzer
H
,
Steuber
T
.
Prostate cancer in the elderly
.
Urol Oncol
2009
;
27
:
668
72
.
2.
Bechis
SK
,
Carroll
PR
,
Cooperberg
MR
.
Impact of age at diagnosis on prostate cancer treatment and survival
.
J Clin Oncol
2011
;
29
:
235
41
.
3.
Abdollah
F
,
Sun
M
,
Thuret
R
,
Jeldres
C
,
Tian
Z
,
Briganti
A
, et al
.
A competing-risks analysis of survival after alternative treatment modalities for prostate cancer patients:1988–2006
.
Eur Urol
2011
;
59
:
88
95
.
4.
Pettersson
A
,
Robinson
D
,
Garmo
H
,
Holmberg
L
,
Stattin
P
.
Age at diagnosis and prostate cancer treatment and prognosis: a population-based cohort study
.
Ann Oncol
2017
;
29
:
377
85
.
5.
Bratt
O
,
Folkvaljon
Y
,
Hjalm Eriksson
M
,
Akre
O
,
Carlsson
S
,
Drevin
L
, et al
.
Undertreatment of men in their seventies with high-risk nonmetastatic prostate cancer
.
Eur Urol
2015
;
68
:
53
8
.
6.
Greco
KA
,
Meeks
JJ
,
Wu
S
,
Nadler
RB
.
Robot-assisted radical prostatectomy in men aged >or =70 years
.
BJU Int
2009
;
104
:
1492
5
.
7.
Penney
KL
,
Stampfer
MJ
,
Jahn
JL
,
Sinnott
JA
,
Flavin
R
,
Rider
JR
, et al
.
Gleason grade progression is uncommon
.
Cancer Res
2013
;
73
:
5163
8
.
8.
Busch
J
,
Magheli
A
,
Leva
N
,
Ferrari
M
,
Kramer
J
,
Klopf
C
, et al
.
Higher rates of upgrading and upstaging in older patients undergoing radical prostatectomy and qualifying for active surveillance
.
BJU Int
2014
;
114
:
517
21
.
9.
Brassell
SA
,
Rice
KR
,
Parker
PM
,
Chen
Y
,
Farrell
JS
,
Cullen
J
, et al
.
Prostate cancer in men 70 years old or older, indolent or aggressive: clinicopathological analysis and outcomes
.
J Urol
2011
;
185
:
132
7
.
10.
Richstone
L
,
Bianco
FJ
,
Shah
HH
,
Kattan
MW
,
Eastham
JA
,
Scardino
PT
, et al
.
Radical prostatectomy in men aged > or =70 years: effect of age on upgrading, upstaging, and the accuracy of a preoperative nomogram
.
BJU Int
2008
;
101
:
541
6
.
11.
Godtman
RA
,
Carlsson
S
,
Holmberg
E
,
Stranne
J
,
Hugosson
J
.
The effect of start and stop age at screening on the risk of being diagnosed with prostate cancer
.
J Urol
2016
;
195
:
1390
6
.
12.
Godtman
RA
,
Kollberg
KS
,
Pihl
C-G
,
Månsson
M
,
Hugosson
J
.
The association between age, prostate cancer risk, and higher gleason score in a long-term screening program: results from the göteborg-1 prostate cancer screening trial
.
Eur Urol
2022
;
82
:
311
7
.
13.
Crowell
PD
,
Fox
JJ
,
Hashimoto
T
,
Diaz
JA
,
Navarro
HI
,
Henry
GH
, et al
.
Expansion of luminal progenitor cells in the aging mouse and human prostate
.
Cell Rep
2019
;
28
:
1499
510
.
14.
Liu
X
,
Grogan
TR
,
Hieronymus
H
,
Hashimoto
T
,
Mottahedeh
J
,
Cheng
D
, et al
.
Low CD38 identifies progenitor-like inflammation-associated luminal cells that can initiate human prostate cancer and predict poor outcome
.
Cell Rep
2016
;
17
:
2596
606
.
15.
Azim
HA
,
Partridge
AH
.
Biology of breast cancer in young women
.
Breast Cancer Res
2014
;
16
:
427
.
16.
Krok-Schoen
JL
,
Fisher
JL
,
Stephens
JA
,
Mims
A
,
Ayyappan
S
,
Woyach
JA
, et al
.
Incidence and survival of hematological cancers among adults ages≥ 75 years
.
Cancer Med
2018
;
7
:
3425
33
.
17.
Cancer Genome Atlas Research Network
.
The molecular taxonomy of primary prostate cancer
.
Cell
2015
;
163
:
1011
25
.
18.
Penney
KL
,
Sinnott
JA
,
Tyekucheva
S
,
Gerke
T
,
Shui
IM
,
Kraft
P
, et al
.
Association of prostate cancer risk variants with gene expression in normal and tumor tissue
.
Cancer Epidemiol Biomark Prev
2015
;
24
:
255
60
.
19.
Pettersson
A
,
Graff
RE
,
Bauer
SR
,
Pitt
MJ
,
Lis
RT
,
Stack
EC
, et al
.
The TMPRSS2: ERG rearrangement, ERG expression, and prostate cancer outcomes: a cohort study and meta-analysis
.
Cancer Epidemiol Prev Biomark
2012
;
21
:
1497
509
.
20.
Chen
Y
,
Lun
ATL
,
Smyth
GK
.
From reads to genes to pathways: differential expression analysis of RNA-seq experiments using Rsubread and the edgeR quasi-likelihood pipeline
.
F1000Res
2016
;
5
:
1438
.
21.
Law
CW
,
Alhamdoosh
M
,
Su
S
,
Dong
X
,
Tian
L
,
Smyth
GK
, et al
.
RNA-seq analysis is easy as 1–2-3 with limma, glimma and edgeR
.
F1000Res
2016
;
5
:
ISCB
.
22.
Ayala
G
,
Frolov
A
,
Chatterjee
D
,
He
D
,
Hilsenbeck
S
,
Ittmann
M
.
Expression of ERG protein in prostate cancer: variability and biological correlates
.
Endocr Relat Cancer
2015
;
22
:
277
87
.
23.
Berglund
AE
,
Rounbehler
RJ
,
Gerke
T
,
Awasthi
S
,
Cheng
C-H
,
Takhar
M
, et al
.
Distinct transcriptional repertoire of the androgen receptor in ETS fusion-negative prostate cancer
.
Prostate Cancer Prostatic Dis
2019
;
22
:
292
302
.
24.
Yuan
J
,
Hu
Z
,
Mahal
BA
,
Zhao
SD
,
Kensler
KH
,
Pi
J
, et al
.
Integrated analysis of genetic ancestry and genomic alterations across cancers
.
Cancer Cell
2018
;
34
:
549
60
.
25.
Lash
TL
.
The harm done to reproducibility by the culture of null hypothesis significance testing
.
Am J Epidemiol
2017
;
186
:
627
35
.
26.
Zhao
SD
,
Parmigiani
G
,
Huttenhower
C
,
Waldron
L
.
Más-o-menos: a simple sign averaging method for discrimination in genomic data analysis
.
Bioinformatics
2014
;
30
:
3062
9
.
27.
Liberzon
A
,
Subramanian
A
,
Pinchback
R
,
Thorvaldsdottir
H
,
Tamayo
P
,
Mesirov
JP
.
Molecular signatures database (MSigDB) 3.0
.
Bioinformatics
2011
;
27
:
1739
40
.
28.
Wu
D
,
Smyth
GK
.
Camera: a competitive gene set test accounting for inter-gene correlation
.
Nucleic Acids Res
2012
;
40
:
e133
.
29.
Geistlinger
L
,
Csaba
G
,
Santarelli
M
,
Ramos
M
,
Schiffer
L
,
Turaga
N
, et al
.
Toward a gold standard for benchmarking gene set enrichment analysis
.
Brief Bioinform
2021
;
22
:
545
56
.
30.
Sboner
A
,
Demichelis
F
,
Calza
S
,
Pawitan
Y
,
Setlur
SR
,
Hoshida
Y
, et al
.
Molecular sampling of prostate cancer: a dilemma for predicting disease progression
.
BMC Med Genomics
2010
;
3
:
8
.
31.
Taylor
BS
,
Schultz
N
,
Hieronymus
H
,
Gopalan
A
,
Xiao
Y
,
Carver
BS
, et al
.
Integrative genomic profiling of human prostate cancer
.
Cancer Cell
2010
;
18
:
11
22
.
32.
Ferrier
DEK
,
Brooke
NM
,
Panopoulou
G
,
Holland
PWH
.
The Mnx homeobox gene class defined by HB9, MNR2 and amphioxus AmphiMnx
.
Dev Genes Evol
2001
;
211
:
103
7
.
33.
Arkhipova
V
,
Wendik
B
,
Devos
N
,
Ek
O
,
Peers
B
,
Meyer
D
.
Characterization and regulation of the hb9/mnx1 beta-cell progenitor specific enhancer in zebrafish
.
Dev Biol
2012
;
365
:
290
302
.
34.
Wendik
B
,
Maier
E
,
Meyer
D
.
Zebrafish mnx genes in endocrine and exocrine pancreas formation
.
Dev Biol
2004
;
268
:
372
83
.
35.
Mullen
AC
,
Hutchins
AS
,
High
FA
,
Lee
HW
,
Sykes
KJ
,
Chodosh
LA
, et al
.
Hlx is induced by and genetically interacts with T-bet to promote heritable TH 1 gene induction
.
Nat Immunol
2002
;
3
:
652
8
.
36.
Bates
MD
,
Schatzman
LC
,
Lints
T
,
Hamlin
PE
,
Harvey
RP
,
Potter
SS
.
Structural and functional characterization of the mouse Hlx homeobox gene
.
Mamm Genome
2000
;
11
:
836
42
.
37.
Valdez
BC
,
Henning
D
,
So
RB
,
Dixon
J
,
Dixon
MJ
.
The Treacher Collins syndrome (TCOF1) gene product is involved in ribosomal DNA gene transcription by interacting with upstream binding factor
.
Proc Natl Acad Sci U S A
2004
;
101
:
10709
14
.
38.
Spruijt
CG
,
Luijsterburg
MS
,
Menafra
R
,
Lindeboom
RGH
,
Jansen
PWTC
,
Edupuganti
RR
, et al
.
ZMYND8 co-localizes with NuRD on target genes and regulates poly (ADP-ribose)-dependent recruitment of GATAD2A/NuRD to sites of DNA damage
.
Cell Rep
2016
;
17
:
783
98
.
39.
Sui
Y
,
Li
X
,
Oh
S
,
Zhang
B
,
Freeman
WM
,
Shin
S
, et al
.
Opposite roles of the JMJD1A interaction partners MDFI and MDFIC in colorectal cancer
.
Sci Rep
2020
;
10
:
8710
.
40.
Huang
B
,
Jiao
Y
,
Zhu
Y
,
Ning
Z
,
Ye
Z
,
Li
QX
, et al
.
Mdfi promotes C2C12 cell differentiation and positively modulates fast-to-slow-twitch muscle fiber transformation
.
Front Cell Dev Biol
2021
;
9
:
605875
.
41.
Shu
Q
,
Lennemann
NJ
,
Sarkar
SN
,
Sadovsky
Y
,
Coyne
CB
.
ADAP2 is an interferon stimulated gene that restricts RNA virus entry
.
PLoS Pathog
2015
;
11
:
e1005150
.
42.
Bist
P
,
Kim
SS-Y
,
Pulloor
NK
,
McCaffrey
K
,
Nair
SK
,
Liu
Y
, et al
.
ArfGAP domain-containing protein 2 (ADAP2) integrates upstream and downstream modules of RIG-I signaling and facilitates type I interferon production
.
Mol Cell Biol
2017
;
37
:
e00537
-
16
.
43.
Borrego Pinto
J
.
Role of novel nuclear envelope proteins involved in nuclear positioning during cell migration
.
Lisboa
:
Universidade de Lisboa
;
2010
.
44.
Lee
M-H
,
Kundu
JK
,
Chae
J-I
,
Shim
J-H
.
Targeting ROCK/LIMK/cofilin signaling pathway in cancer
.
Arch Pharm Res
2019
;
42
:
481
91
.
45.
Fang
T
,
Cui
M
,
Sun
J
,
Ge
C
,
Zhao
F
,
Zhang
L
, et al
.
Orosomucoid 2 inhibits tumor metastasis and is upregulated by CCAAT/enhancer binding protein β in hepatocellular carcinomas
.
Oncotarget
2015
;
6
:
16106
19
.
46.
Landsman
L
,
Bar-On
L
,
Zernecke
A
,
Kim
K-W
,
Krauthgamer
R
,
Shagdarsuren
E
, et al
.
CX3CR1 is required for monocyte homeostasis and atherogenesis by promoting cell survival
.
Blood
2009
;
113
:
963
72
.
47.
Xu
J
,
Guo
Y
.
FCGR1A serves as a novel biomarker and correlates with immune infiltration in four cancer types
.
Front Mol Biosci
2020
;
7
:
581615
.
48.
Pagliuca
FW
,
Collins
MO
,
Lichawska
A
,
Zegerman
P
,
Choudhary
JS
,
Pines
J
.
Quantitative proteomics reveals the basis for the biochemical specificity of the cell-cycle machinery
.
Mol Cell
2011
;
43
:
406
17
.
49.
Kosi
N
,
Alić
I
,
Kolačević
M
,
Vrsaljko
N
,
Milošević
NJ
,
Sobol
M
, et al
.
Nop2 is expressed during proliferation of neural stem cells and in adult mouse and human brain
.
Brain Res
2015
;
1597
:
65
76
.
50.
Zhang
L
,
Wang
J
,
Wang
Y
,
Zhang
Y
,
Castro
P
,
Shao
L
, et al
.
MNX1 is oncogenically upregulated in African-American prostate cancer
.
Cancer Res
2016
;
76
:
6290
8
.
51.
Das
M
.
MNX1: a novel prostate cancer oncogene
.
Lancet Oncol
2016
;
17
:
e521
.
52.
Li
Z
,
Wang
F
,
Zhang
S
.
Knockdown of lnc RNA MNX 1-AS 1 suppresses cell proliferation, migration, and invasion in prostate cancer
.
FEBS Open Bio
2019
;
9
:
851
8
.
53.
Ledet
RJ
,
Ruff
SE
,
Wang
Y
,
Nayak
S
,
Schneider
JA
,
Ueberheide
B
, et al
.
Identification of PIM1 substrates reveals a role for NDRG1 phosphorylation in prostate cancer cellular migration and invasion
.
Commun Biol
2021
;
4
:
36
.
54.
Marques
RB
,
Dits
NF
,
Erkens-Schulze
S
,
Van Weerden
WM
,
Jenster
G
.
Bypass mechanisms of the androgen receptor pathway in therapy-resistant prostate cancer cell models
.
PLoS One
2010
;
5
:
e13500
.
55.
Kar
SP
,
Beesley
J
,
Al-Olama
AA
,
Michailidou
K
,
Tyrer
J
,
Kote-Jarai
Z
, et al
.
Genome-wide meta-analyses of breast, ovarian, and prostate cancer association studies identify multiple new susceptibility loci shared by at least two cancer types
.
Cancer Discov
2016
;
6
:
1052
67
.
56.
Fuse
M
,
Nohata
N
,
Kojima
S
,
Sakamoto
S
,
Chiyomaru
T
,
Kawakami
K
, et al
.
Restoration of miR-145 expression suppresses cell proliferation, migration and invasion in prostate cancer by targeting FSCN1
.
Int J Oncol
2011
;
38
:
1093
101
.
57.
Huang
J
,
Wu
Y
,
Lin
Y
,
Cai
H
,
Chen
S
,
Sun
X
, et al
.
Up-regulation of LIMK1 expression in prostate cancer is correlated with poor pathological features, lymph node metastases and biochemical recurrence
.
J Cell Mol Med
2020
;
24
:
4698
706
.
58.
Ahmed
T
,
Shea
K
,
Masters
JRW
,
Jones
GE
,
Wells
CM
.
A PAK4–LIMK1 pathway drives prostate cancer cell migration downstream of HGF
.
Cell Signal
2008
;
20
:
1320
8
.
59.
Cai
S
,
Chen
R
,
Li
X
,
Cai
Y
,
Ye
Z
,
Li
S
, et al
.
Downregulation of microRNA-23a suppresses prostate cancer metastasis by targeting the PAK6-LIMK1 signaling pathway
.
Oncotarget
2015
;
6
:
3904
17
.
60.
Jamieson
WL
,
Shimizu
S
,
D'Ambrosio
JA
,
Meucci
O
,
Fatatis
A
.
CX3CR1 is expressed by prostate epithelial cells and androgens regulate the levels of CX3CL1/fractalkine in the bone marrow: potential role in prostate cancer bone tropism
.
Cancer Res
2008
;
68
:
1715
22
.
61.
Xiao
L-J
,
Chen
Y-Y
,
Lin
P
,
Zou
H-F
,
Lin
F
,
Zhao
L-N
, et al
.
Hypoxia increases CX3CR1 expression via HIF-1 and NF-κB in androgen-independent prostate cancer cells
.
Int J Oncol
2012
;
41
:
1827
36
.
62.
Shulby
SA
,
Dolloff
NG
,
Stearns
ME
,
Meucci
O
,
Fatatis
A
.
CX3CR1-fractalkine expression regulates cellular mechanisms involved in adhesion, migration, and survival of human prostate cancer cells
.
Cancer Res
2004
;
64
:
4693
8
.
63.
Sun
F
,
Wu
K
,
Yao
Z
,
Mu
X
,
Zheng
Z
,
Sun
M
, et al
.
Long noncoding RNA PVT1 promotes prostate cancer metastasis by increasing NOP2 expression via targeting tumor suppressor MicroRNAs
.
OncoTargets Ther
2020
;
13
:
6755
65
.
64.
Chatsirisupachai
K
,
Lesluyes
T
,
Paraoan
L
,
Van Loo
P
,
de Magalhães
JP
.
An integrative analysis of the age-associated multi-omic landscape across cancers
.
Nat Commun
2021
;
12
:
2345
.
65.
Li
CH
,
Haider
S
,
Boutros
PC
.
Age influences on the molecular presentation of tumours
.
Nat Commun
2022
;
13
:
208
.
66.
Chalmers
ZR
,
Burns
MC
,
Ebot
EM
,
Frampton
GM
,
Ross
JS
,
Hussain
MHA
, et al
.
Early-onset metastatic and clinically advanced prostate cancer is a distinct clinical and molecular entity characterized by increased TMPRSS2-ERG fusions
.
Prostate Cancer Prostatic Dis
2021
;
24
:
558
66
.
67.
Den
RB
,
Santiago-Jimenez
M
,
Alter
J
,
Schliekelman
M
,
Wagner
JR
,
Renzulli Ii
JF
, et al
.
Decipher correlation patterns post prostatectomy: initial experience from 2 342 prospective patients
.
Prostate Cancer Prostatic Dis
2016
;
19
:
374
9
.
68.
Zhao
S
,
Fung-Leung
W-P
,
Bittner
A
,
Ngo
K
,
Liu
X
.
Comparison of RNA-Seq and microarray in transcriptome profiling of activated T cells
.
PLoS One
2014
;
9
:
e78644
.
69.
Rao
MS
,
Van Vleet
TR
,
Ciurlionis
R
,
Buck
WR
,
Mittelstadt
SW
,
Blomme
EAG
, et al
.
Comparison of RNA-Seq and microarray gene expression platforms for the toxicogenomic evaluation of liver from short-term rat toxicity studies
.
Front Genet
2019
;
9
:
636
.
70.
Greenland
S
.
Noncollapsibility, confounding, and sparse-data bias. Part 1: The oddities of odds
.
J Clin Epidemiol
2021
;
138
:
178
81
.
71.
Greenland
S
.
Noncollapsibility, confounding, and sparse-data bias. Part 2: What should researchers make of persistent controversies about the odds ratio?
J Clin Epidemiol
2021
;
139
:
264
8
.