Introduction

The PTEN tumor suppressor functions as a phosphatase, which metabolizes PIP3, the lipid product of PI3-Kinase that antagonizes the activation of the oncogenic PI3K/AKT/mTOR signaling axis1,2. In addition to its role as a primary regulator of cancer cell growth, recent findings also implicate PTEN in mediating aspects of the immune response against cancer3,4.

PTEN is frequently lost, either partially affecting one copy of the gene (hemizygous loss; HemDel) or entirely affecting both copies of the gene (homozygous loss; HomDel), in a range of sporadic tumor types. A variety of mechanisms can cause PTEN gene inactivation, including missense and truncation mutations and deletions, as well as reduced expression of active PTEN protein mediated by promoter methylation, the effects of miRNAs, and the suppression of PTEN enzyme activity5. Loss of function of the PTEN tumor suppressor has been observed as one of the most common events in many types of cancer6, having the greatest selection pressure for deletion based on an analysis of 746 human cancer genomes7.

The Pten conditional knockout mouse model closely mimics human prostate initiation and progression8. Studies showed that loss of one copy of the Pten allele dramatically increases the rate of tumors9. It was proposed that haploinsufficiency of the Pten gene resulted in reduced levels of the functional protein that were insufficient for preventing tumor formation. Similar studies using a brain-tumor-specific model also showed that Pten haploinsufficiency accelerated the formation of glioblastomas10. Analysis of hypomorphic Pten mice demonstrated that as little as a 20% reduction of PTEN protein contributed to the development of prostate cancer in mice11. These and other mouse studies12 show that minor reductions in PTEN protein levels can profoundly influence tumorigenesis13. The findings from Pten murine tumor models are consistent with the precise cellular controls needed to regulate PTEN stability and function14, implying that even subtle changes in PTEN expression may have profound effects on cancer development and progression5.

Álvarez-Garcia et al.15 recently surveyed the solid tumor literature on PTEN loss resulting from a mutation within the coding region of the gene; hemizygous or homozygous deletions of the gene as assessed by loss of heterozygosity (LOH), reduced copy number in array CGH analysis or loss of PTEN by FISH. Most tumor-associated missense mutations resulted in a loss or significantly reduced phosphatase activity or protein truncations. PTEN was found to demonstrate distinct patterns of somatic mutation leading to loss of function across different tumor types. Most glioblastomas display loss of function of PTEN in agreement with data from TCGA showing PTEN deletions in 143/170 (85%) of glioblastomas16. Other human tumor types found to have frequent PTEN deletions and/or loss of protein are cancers of prostate (25–50%), endometrium (50%), ovary (30–50%), lung (35–55%), breast (30–40%) and colon (10–40%)17. Moreover, it was found that single-copy inactivation of PTEN was far more common than mutation or deletion of both copies15. The PI3K/AKT/mTOR consortium study further suggested that partial copy number PTEN loss was associated with a less favorable outcome than complete loss across all available TCGA data18. These data, together with the murine Pten studies of tumorigenesis, formed the rationale for this PanCancer study that aimed to address the idea that partial loss of function of PTEN by hemizygous loss may confer a greater selective advantage for a subset of tumors than complete loss of function by homozygous loss (reviewed in19).

Our previous study examining the genomic impact of PTEN loss using TCGA data demonstrated that primary prostate tumors harboring PTEN deletions had enhanced levels of aneuploidy and non-synonymous mutations20. However, it was unclear whether the genomic instability generated by loss of one copy of PTEN was associated with tumor progression. Other studies examining the impact of PTEN loss using TCGA and other public domain datasets do not distinguish between one copy loss and complete loss of the gene21,22,23,24. Similarly, most PTEN cancer biomarker studies using immunohistochemistry and fluorescence in situ hybridization (FISH) have not been rigorously standardized15,17,25,26. For these reasons, the impact of partial loss of function due to one copy loss of PTEN is unknown for most cancers.

This study analyzed the TCGA copy-number and transcriptomic data, comparing oncogenic and immunophenotypic features that distinguish hemizygous PTEN loss from homozygous loss across the pan-cancer cohort. Our analysis also examines the relationships between PTEN intact, hemi-, and homozygous loss with genomic features of tumor progression and disease outcome.

Results

The PTEN study group comprised 30 tumor types, including 9,793 cases analyzed from the TCGA Pan-cancer cohort. PTEN somatic copy number alterations (SCNA) were present in 3,619 (36.96%) tumors. Copy number deletions were observed in 2,903 (29.65%) tumors, with 2,484 (25.37%) having hemizygous deletions resulting in one copy and 419 (4.28%) having homozygous loss of both copies. Copy number gains as a duplication of the PTEN gene were evident in 716 (7.31%) tumors (Table 1).

Table 1 PTEN somatic copy number status (SCNA). HemDel: PTEN hemizygous deletion. HomDel: PTEN homozygous deletion. Dupl Duplication or gain across the Pan-cancer cohort of 30 tumors.
Full size table

Hemizygous PTEN deletions (HemDel) are more common than homozygous deletions (HomDel) across the pan-cancer cohort

For most of the common solid tumors, hemizygous PTEN deletions were much more frequent than homozygous deletions (Fig. 1). For example, melanoma (SKCM) has 205 (55.86%) hemizygous deletion compared to 28 (7.63%) homozygous loss. Breast cancer (BRCA) has a hemizygous deletion frequency of 269 (25.28%), but homozygous deletions are less common at 54 (5.08%) of tumors. There are interesting differences in lung subtypes, with squamous cell carcinoma (LUSC) having 217 (44.83%) hemizygous and 48 (9.92%) homozygous losses. However, adenocarcinoma (LUAD) had fewer hemizygous deletions, 140 (27.50%) but only 5 (0.98%) homozygous losses. For prostate cancer (PRAD), the rate of hemizygous loss was 73 (14.96%), slightly less than the homozygous loss rate of 85 (17.42%). The four tumor types (KIRP, THYM, THCA, and PCPG) with the lowest frequency of PTEN SCNA are characterized by a relatively more favorable outcome in the literature. Pheochromocytoma and paraganglioma (PCPG) are tumors of the autonomic nervous system. This tumor type has 11 (6.83%) PTEN duplication gains suggesting instability leading to PTEN acquisition rather than loss may be a feature of this tumor. Other tumors with more marked levels of PTEN SCNA gain are ovarian (OV) 81 (19.66%), esophageal carcinoma (ESCA) 23 (12.71%), lung adenocarcinoma (LUAD) 73 (14.34%), stomach adenocarcinoma (STAD) 48 (11.59%) and hepatocellular carcinoma of the liver (LIHC) 35 (9.70%).

Figure 1
figure 1

Relative percentages of PTEN Somatic Copy Number Alterations (PTEN SCNA) from the 23 most common solid tumors taken from the 30 tumor types analyzed in the Pan-cancer cohort (shown in Table 1). Hemizygous deletions (HemDel) = blue; Homozygous deletions (HomDel) = red, Intact (no PTEN SCNA) = green, and gain/duplication = brown. The tumors are ranked by their combined frequencies of PTEN deletion (HemDel + HomDel) with melanomas (SKCM) having the highest overall frequency of PTEN deletion (top) and Pheochromocytoma and Paraganglioma (PCPG) having the lowest (bottom). Duplication or PTEN gain (Dupl) was not included in the rank order determination. Plotted using ggplot2 (https://ggplot2.tidyverse.org/).

Full size image

PTEN expression levels closely correlate to gene copy number

To determine how closely PTEN deletions and gains affect PTEN expression, we analyzed transcript counts across the pan-cancer cohort to estimate gene expression levels related to PTEN copy number (Fig. 2a). We observed that tumors with PTEN HomDel had the lowest expression levels, with mean transcript counts that were one order of magnitude lower than the mean count observed in intact tumors (P < 0.0001). The HemDel tumors showed transcript counts lower than intact (P < 0.0001), and PTEN gains were associated with slightly higher expression levels than PTEN intact tumors (P < 0.0001). We compared PTEN intact to HemDel and HomDel in the cohort’s 23 most common solid tumors (Fig. 2b). The same reduced expression pattern is apparent in all tumor types, with PTEN HomDel having the lowest expression levels for most tumors. Similarly, PTEN HemDel exhibited intermediate expression levels between HomDel and PTEN intact tumors. These data agree with the general correlations observed between gene copy number and RNA expression levels using TCGA transcriptomic data reported recently27.

Figure 2
figure 2

(a) Combined analysis of the association between PTEN SCNA status and differential gene expression of PTEN for all 30 tumors in the Pan-cancer cohort. Boxplots show the relationship between PTEN normalized transcript counts Log2 (Y-axis) versus PTEN SCNA status (HemDel, HomDel, Dupl and Intact). The mean transcript count for PTEN intact was 18.78, HomDel was the lowest expression level at 17.15, and HemDel showed reduced expression at 18.26, but Dupl was slightly higher than intact at 19.20. The mean transcript counts for all three classes of SCNA significantly deviated from intact (P < 0.0001, t-test). (b) Association between PTEN SCNA status and differential gene expression of PTEN for the 23 most common solid tumors in the Pan-cancer cohort. Boxplots show the relationship between PTEN transcript counts Log2 (Y-axis) versus PTEN SCNA status HomDel (red), HemDel (blue), and Intact (green) for each tumor. For most of tumors, the mean transcript counts for both HomDel and HemDel deviated significantly from intact (P < 0.05). Plotted using ggplot2 (https://ggplot2.tidyverse.org/).

Full size image

Associations between hemizygous PTEN deletions and poor outcome

Analysis of the overall survival of the entire cohort stratified by PTEN deletion status showed that patients with tumors harboring PTEN HemDel had a similar poor outcome to those with HomDel (Fig. 3a). We next performed outcome analysis comparing PTEN intact to HomDel and HemDel for four solid tumors (Fig. 3b–e) to illustrate the diversity of effects of PTEN loss on survival. For colorectal carcinomas and head and neck cancers, the proportion of tumors with HomDel PTEN loss was relatively low, so these patients were not considered for comparative studies. In brain tumors, the PTEN intact have a steadily decreased survival probability, but both PTEN HomDel and HemDel were associated with a rapid decline in survival probabilities (Fig. 3b). In cervical tumors, there is a lower survival probability for PTEN HemDel compared to HomDel and intact (Fig. 3c). In stomach tumors (Fig. 3d), HemDel was associated with intermediate survival risk. In head and neck squamous cell carcinoma (Fig. 3e), the HemDel tumors were associated with reduced survival probabilities. TP53 mutation was more common than expected for all HemDel tumors across the cohort (Supplementary Table 1). Further analysis considering the combined effects of TP53 mutation with PTEN deletion status was performed (Supplementary Table 2). Similar hazard ratios were observed for TP53 monoallelic loss and PTEN HemDel, and there was no obvious additive effect when both mutations occurred together (Supplementary Table 2a). HNSCC tumors also showed similar hazard ratios for PTEN HemDel and TP53 monoallelic mutation without apparent synergy when both mutations were present (Supplementary Table 2b). These data suggest that one copy loss of either PTEN or TP53 results in equivalent effects on cancer outcome.

Figure 3
figure 3

Kaplan–Meier curves illustrate overall survival for both PTEN intact and PTEN deletions groups (HemDel and HomDel). Boxplots (a) Pan-cancer cohort shows a lower survival probability for PTEN HemDel compared to HomDel and intact (P < 0.0001). (b) Brain tumors (GBM + LGG) show that both HomDel and HemDel are equally associated with earlier events (P < 0.0001). (c) Cervical tumors show a lower survival probability for PTEN HemDel compared to HomDel and intact. (d) Head and neck cancer. (e) Stomach tumors. (f) Cox Model analysis for PTEN intact and PTEN deletions groups. Plotted using Survminer (https://rpkgs.datanovia.com/survminer/index.html).

Full size image

Relationships between hemizygous PTEN deletions and genomic instability

To understand the impact of PTEN loss on genomic features of cancer progression, we analyzed the total number of mutations, percent genome altered by losses and gains, intratumor heterogeneity, aneuploidy, and homologous repair defects across the pan-cancer cohort (Fig. 4a–d). Our combined analysis revealed that cancers harboring PTEN HemDel had the highest levels of nonsilent mutations, intratumor heterogeneity, and percent genome altered. Moreover, PTEN HomDel and HemDel presented higher levels of homologous repair defect compared with PTEN intact, but no difference was found between the two deletion types. When stratified by tumor type, many of the tumors harboring PTEN deletions had increased levels of nonsilent mutations, intratumor heterogeneity, and homologous repair defects (Supplementary Fig. 2a–c).

Figure 4
figure 4

PTEN deletions are associated with (a) increased mutation levels, (b) intra-tumor heterogeneity (ITH), (c) homologous repair defects (HRD), and (d) percent genome alteration. Hemizygous deletions of PTEN are associated with increased levels of genomic alterations compared with homozygous deletions. Intact PTEN showed the lowest levels of genomic alterations. Plotted using ggplot2 (https://ggplot2.tidyverse.org/).

Full size image

To compare the percentage of genome altered, we found that 65% (13/20) of the investigated tumor types presented elevated genomic alterations when harboring PTEN HemDel compared to PTEN Intact tumors (Supplementary Fig. 2d). Pan-cancer analysis showed aneuploidy levels were significantly elevated in HemDel compared to HomDel and both deletion types showed significantly increased aneuploidy compared to all intact tumors (Supplementary Fig. 3a). Similar intratumor heterogeneity and aneuploidy patterns were found in Dupl cases (Supplementary Fig. 4a,b). Analysis of aneuploidy in a subset of twenty solid tumors showed a general increase in aneuploidy associated with PTEN deletion. Hemizygous losses in bladder, cervix, colon, head and neck, liver, ovarian, pancreatic, and stomach cancer promoted increased levels of aneuploidy (Supplementary Fig. 3b). Collectively, these findings led us to hypothesize that PTEN HemDel may be associated with an altered transcriptome, potentially driving more genomic alterations.

Transcriptomic changes associated with PTEN loss activate immune pathways

We next investigated the transcriptome of seven representative solid tumors with increased frequencies of PTEN deletions and presented more diversity in genomic changes. Differential gene expression analyses showed prostate tumors with PTEN HemDel presented 2076 upregulated and 2966 downregulated genes compared to PTEN Intact tumors (Supplementary Table 3). Prostate tumors with HemDel of PTEN presented several enriched pathways compared with PTEN intact, including microtubule-based movement and chromosome segregation (Supplementary Fig. 5). PTEN HemDel was also associated with distinct enrichment of immune pathways across the cancer types. In prostate tumors with PTEN HomDel the high number of DEGs was linked to cell cycle, DNA repair, and vascular process pathways (Supplementary Fig. 6). Interestingly, for HomDel in prostate cancer, the genes closely flanking PTEN, such as ATAD1, RNLS and KLLN, are concurrently downregulated with the PTEN gene consistent with the large genomic deletions of chromosome 10 previously observed in prostate cancer by FISH28 (see the vertical bar in Supplementary Fig. 6).

Brain tumors presented the most distinct pathway enrichment results, including the upregulation of several immune pathways when PTEN was deleted. We found 3562 up- and 10902 downregulated genes in these tumors when compared with PTEN HemDel vs. PTEN intact (Supplementary Table 3). GO-enriched analysis showed that several immune-related pathways were upregulated for PTEN HemDel tumors, such as granulocyte migration and chemotaxis, neutrophil migration, acute inflammatory response, humoral immune homeostasis, and many others (Supplementary Fig. 7). Similarly, when we compared the transcriptome of tumors with PTEN HomDel vs. those with PTEN intact, we observed 4106 up- and 7332 downregulated genes (Supplementary Table 3, Supplementary Fig. 8). We also observed that several immune-related pathways were significantly upregulated in brain tumors harboring PTEN HomDel, namely myeloid leucocyte migration, humoral immune response, leucocyte chemotaxis, and acute inflammatory response (Supplementary Fig. 8). In summary, most investigated tumors had immune-related pathways up- or downregulated when PTEN was lost. The RNA-sequencing expression level of immunomodulatory genes like CD274 (PD-L1, Programmed Cell Death), CTLA4 (Cytotoxic T-Lymphocyte Associated Protein 4), IDO1 (Indoleamine 2,3-Dioxygenase 1), and LAG3 (Lymphocyte Activating 3) are altered in breast, glioblastoma, head and neck, liver, ovarian, prostate, sarcoma, stomach, and uterine tumors (Supplementary Fig. 9). These findings led us to hypothesize whether PTEN genomic deletions may be associated with an altered tumor microenvironment, as determined from higher or lower abundance of immune cells surrounding cancerous lesions.

Stomach tumors harboring PTEN HemDel show downregulated immune pathways, such as adaptive immune response, regulation of lymphocyte activation, T cell activation, positive regulation of immune response, leukocyte-mediated immunity, lymphocyte proliferation, and several other pathways (Supplementary Fig. 10). Stomach tumors with PTEN HemDel also had downregulation of Indoleamine 2,3-Dioxygenase 1 (IDO1) gene (Supplementary Fig. 10). In contrast, activation of immune pathways was not apparent in stomach tumors harboring PTEN HomDel, but the sample size (N = 20) for homozygous loss tumors was reduced compared to HemDel tumors (N = 90) (Supplementary Fig. 11, Supplementary Table 3). For head and neck tumors, we also observed that tumors with PTEN HemDel had downregulated immune-related pathways, including adaptive immune response, activation of T-cell receptor, regulation of lymphocyte-mediated immunity, regulation of leukocyte-mediated immunity, and regulation of immune effector process (Fig. 5a–c). However, these results were not observed for the comparison between PTEN HomDel and PTEN intact, in which the sample size from homozygous loss tumors was limited (N = 15). However, the linked passenger genes MINPP1 and KLLN from chromosome 10 showed concurrent reduced expression with PTEN (Supplementary Fig. 12). Lastly, for colon tumors, we found that tumors with PTEN HemDel had downregulated immune-related pathways, including response to interferon-gamma, lymphocyte-mediated immunity, adaptive immune response, and natural killer cell-mediated immunity (Supplementary Fig. 13). However, effects on immune pathways were not observed for the comparison between PTEN HomDel and PTEN intact. As with other homozygous losses, there was reduced expression of a group of PTEN-linked passenger genes mapping to chromosome 10, including ALDH18A1, DNMBP, KLLN, WAPL, MINPP1, and ATAD1 (see the vertical bar in Supplementary Fig. 14). Collectively, the pathway analysis results draw attention to the role hemizygous PTEN loss may play in the anticancer immune response.

Figure 5
figure 5

PTEN HemDel in head and neck squamous cell carcinoma (HNSC) shows distinct transcriptome activity. (a) Transcriptome heatmap exhibiting clustering of top 50 DEGs (Log2Fold change > 0.58). The color scale in the heatmap represents the Z-score of the normalized read counts for each gene, where the red scale color indicates upregulated, and blue low expressed genes. Enrichment analyses were performed using clusterProfiler and p-adjusted value = 0.05 as the cutoff. (b) Enriched GO (BP) pathways of the upregulated genes. (c) Enriched GO (BP) pathways of the downregulated genes. The tables shown in (b) and (c) indicate the top 10 most significant Gene Ontology (BP) terms. (d) CIBERSORT‐derived immune‐cell abundance of 22 distinct cell subsets based on PTEN status and head and neck tumors. Y‐axis shows CIBERSORT relative scores. PTEN HemDel in head and neck tumors was linked to higher M0 and M2 macrophages, mast cells, and lower CD4 T, CD8 T, T regs, and NK cells. *p-value < 0.05; **p-value < 0.01, by Mann–Whitney test. n = 492. DC dendritic cell; NK natural‐killer cell; PCa prostate cancer; Treg regulatory T cell; M0 M0 macrophage; M1 M1 macrophage; M2 M2 macrophage. Plotted using pheatmap (https://rdocumentation.org/packages/pheatmap/versions/1.0.12).

Full size image

PTEN deletions are associated with an altered tumor immune microenvironment

The results from the tumor microenvironment comparison of PTEN HomDel, HemDel, and intact tumors for all 22 immune cell types from brain, prostate, head and neck, and stomach tumors were investigated. PTEN HemDel in brain tumors was linked to higher CD4 T memory cells and M2 macrophage abundance and lower M1 macrophages, monocytes, and CD8 T cells (Supplementary Fig. 15a). Similar results were seen in head and neck tumors, where PTEN HemDel showed higher M0 and M2 macrophages and mast cells and lower CD4 T-cell, CD8+ T-cell, T-regulatory cells, and NK cells (Fig. 5d). While in prostate tumors with PTEN HemDel, we observed an increase in M1 macrophages and a decrease in M0 macrophages and monocytes abundance (Supplementary Fig. 15b). Stomach tumors with PTEN HemDel had an increased plasma and M0 macrophages density and decreased density of CD8 T-cells (Supplementary Fig. 15c). Cervical tumors with PTEN HemDel were linked to higher M0 macrophages and an abundance of activated mast cells (Supplementary Fig. 15d). Similar results were seen in colorectal tumors, where PTEN HemDel showed a higher M0 macrophage density and lower CD8 T-cell abundance (Supplementary Fig. 15e).

Discussion

The presence of the PTEN protein is essential for controlling the tumor promoting activities of the PI3K/AKT pathway2. Deregulation of this signaling axis confers a strong selective advantage to tumors, so that loss of PTEN control by inactivating mutation or deletion is one of the most common somatic events in human cancer6. PTEN mutation is significantly related to advanced disease, chemotherapy resistance, and poor survival in patients with head and neck cancers, breast, melanoma, colorectal, esophageal, and prostate19,29,30,31,32,33,34 so that its role as a predictive and prognostic cancer biomarker is increasingly gaining prominence35,36,37. Moreover, there is increasing awareness of the role PTEN deficiency plays in shaping the immune landscape in different cancers3,4.

Various murine models have shown that the tumor-promoting abilities of loss of the Pten gene are strongly dose-dependent38. In some models, the Pten gene loss has been found to exhibit haploinsufficiency with reduced levels of the normal Pten protein resulting in a diversity of tumors in mice2. Thus, in contrast to the classical definition of a tumor suppressor gene that requires complete loss of both copies of the gene39, it appears that reduced amounts of PTEN protein caused by either monoallelic loss or by one inactivating mutation is sufficient to cause cancer (reviewed in19).

Partial loss of function caused by PTEN hypomorphic mutations has been observed amongst PTEN hamartoma tumor syndrome (PHTS) patients40. A lower cancer risk and possibly a higher likelihood of neurodevelopmental phenotypes, specifically autism spectrum disorder (ASD), was associated with PTEN hypomorphic mutations.

A recent review across all solid tumors showed that hemizygous PTEN loss due to monoallelic deletion or point mutation occurred at a much higher frequency than biallelic loss15. Our PTEN analysis of TCGA data also shows that most tumors have a much greater frequency of HemDel than HomDel and that hemizygous loss is less favorable than homozygous loss for some tumor types. These data agree with previous genetic and proteomic studies of PI3K/AKT/mTOR axis mutations using TCGA data that showed outcomes associated with partial loss of PTEN were less favorable than outcomes for complete loss of the gene18. Our analysis of PTEN RNA transcript levels across the TCGA cohort shows good agreement with gene copy number, as reported recently by others27, so the HemDel losses we analyzed in this study would be expected to reduce RNA expression levels. Moreover, since one somatic PTEN deletion or point mutations leads to a reduction in RNA levels, then partial PTEN protein expression is expected in HemDel tumors.

PTEN deletions and protein loss are known to affect DNA damage responses and disrupt mitotic spindle architecture resulting in accumulation of mutations in other genes and increased chromosome instability in both murine models and human cancers41. As expected, our TCGA analysis showed both hemi- and homozygous losses increased genomic instability features in many tumors. However, HemDel tumors had significantly higher rates of non-silent mutations, intratumor heterogeneity, percentage genome altered, and aneuploidy compared to HomDel tumors. This diversity of genetic alteration is likely to facilitate genomic evolution and increase cancer progression, treatment resistance, and poor outcome42. Collectively, these data suggest that the elevated levels of genomic instability and somatic mutations in the tumors with hemizygous loss could provide a selective advantage for the acquisition of other driver genes and the less favorable outcomes observed across the TCGA cohort.

PTEN loss may be sufficient to cause tumorigenesis in some tissues but not in others. A recent TCGA analysis of collaborating somatic mutations in PTEN-defective solid tumors identified cancer genes such as TP53, APC, TTN, MUC16, PIK3CA, BRAF, CDH1, and KMT2D, irrespective of the tumor type37. When surveying the outcome in different cancers across the TCGA cohort, we found the consequences of HemDel loss were more pronounced in some tumor types, suggesting that other cellular contexts or unknown tumor-specific molecular features may contribute to progression when PTEN is partially deficient.

The most obvious effects on outcome for HemDel PTEN losses were observed in head and neck squamous cell carcinomas (HNSCC) harboring a mono-allelic somatic loss-of-function mutation in TP53. For this type, the hemizygous loss was associated with a less favorable overall survival with a significant HR of 1.67 (P = 0.023) (Fig. 3, Supplementary Table 2). PTEN loss is associated with poor outcomes in HNSCC30. HNSCC also has a high frequency of TP53 somatic mutations43. We used cBioPortal to investigate the co-occurrence frequencies of TP53 mutations with PTEN intact, HemDel, and HomDel in HNSCC. Interestingly, PTEN HemDel was significantly (Supplementary Table 1, P < 0.001) associated with TP53 mutation, and PTEN HomDel was under-represented in HNSCC with mutated TP53 (data not shown). These data suggest that there may be preferential selection for one copy of a PTEN gene in HNSCC so that reduced levels of the functional PTEN protein allow tumor cells to bypass the TP53-induced cell senescence pathway44. These observations in HNSCC are in keeping with the PTEN literature that suggests that single-copy inactivation of PTEN may actually be selected for during the progression of some tumor types since bi-allelic inactivation of PTEN has been shown to lead to senescence or cell death when TP53 is mutated5. Recent functional studies of the relationship between TP53 mutation and PTEN loss in HNSCC indicate that loss of regulation of both pathways could influence radio- and chemosensitivity45. These findings also agree with recent studies showing that PTEN loss mediates resistance to cetuximab in HNSCC, indicating the need for more treatment response PTEN predictive biomarker studies that also examines the role of HemDel events46.

Mouse models of HNSCC draw attention to the role of Pten in the evasion of cellular senescence and activation of cancer-related inflammation47. Our gene set enrichment analysis for HNSCC indicates that various pathways involved with immune activation are downregulated in PTEN HemDel tumors. Pathways indicated in our analysis were antigen presentation, interferon, interleukins, toll-like, neutrophil, B cell, and inflammasome signaling. More marked alterations to the immune response in HemDel HNSCC were also evident in our CIBERSORT analysis. There were decreased CD8 T cells and alterations to tumor-associated macrophages (TAMs) consistent with the immune tumor microenvironment being less favorable in hemizygous HNSCC. These findings are consistent with the emerging recognition that PTEN loss could be a useful predictive immune biomarker in HNSCC48.

Biomarker analysis of colorectal cancer showed weak expression of the PTEN protein in primary tumors with metastasis33. Our study found that PTEN HemDel colorectal tumors had downregulated immune-related pathways. Functional activities associated with immune response included regulation of granulocyte–macrophage colony-stimulating factor production, neutrophil migration, and regulation of leukocyte-mediated cytotoxicity. These findings are consistent with recent studies that showed that PTEN mutation was associated with microsatellite instability subtypes and tumor mutational burden in colon cancer49. Both in vitro studies50 and recent analyses of colon tumors51 suggest that PTEN loss in colorectal carcinoma may activate PI3K and upregulate PD-L1 expression. These data are in keeping with other studies indicating that the PTEN/ PI3K axis may promote immune escape through regulating PDL1/PD1 expression3,52.

Brain tumors also presented distinct pathway enrichment findings, including the upregulation of several immune pathways for PTEN HemDel and HomDel losses. These data are consistent with a recent study showing PTEN loss in glioblastoma was associated with immunosuppressive expression signatures and failed response to anti-PD-1 immunotherapy53.

Our TCGA analysis of prostate tumors with PTEN HemDel identified elevated levels of genomic instability and increased M1 macrophages but decreased abundances of M0 macrophages and monocytes. Pathway analysis showed enrichment for response to type I-interferon, consistent with other recent findings showing that PTEN loss leads to an immunosuppressive microenvironment in prostate cancer (PCa)54. These conclusions also agree with our previous study that showed increased IDO1 and Tregs in PTEN-deficient prostate cancer tumor microenvironment55. Interestingly, analysis of PCa showed that the genes flanking PTEN, such as ATAD1, RNLS, and KLLN are concurrently down-regulated when the PTEN gene is homozygously deleted. Similar observations were made in both colon and head and neck tumors with homozygous deletions. These data suggest that large somatic deletions of chromosome 10 may often encompass closely linked genes whose expression changes could contribute to clinical phenotypes36.

Unfortunately, most previous PTEN cancer biomarker studies have not systematically analyzed mutations or deletions using molecular genetic methods that can distinguish between HemDel and HomDel mutations. Similarly, loss of the protein using immunohistochemistry as an alternative way of determining PTEN status has also not been studied using uniform, reproducible methods that can detect subtle changes in protein levels15,17,25,26. The end result has been that majority of published PTEN cancer biomarker studies have depended on a variety of simple two-way classification assays (PTEN loss vs. intact) that ignore the distinction between hemi- and homozygous losses.

Our study also has some limitations that need to be addressed in future PTEN analyses of the highlighted tumors. Our work was based entirely on in silico analysis of copy number data derived TCGA, and all the pathway and immune parameters were inferred from the associated transcriptome data. The analysis presented here cannot fully capture the impact of PTEN mutations based on the limitations of the TCGA dataset, which lacks prognostic or treatment information. Also, our expression analyses associated with PTEN status were, like all in silico TCGA studies, obtained from the bulk tumors containing heterogeneous tumor cells intermingled with a small subset of stromal cells and varying quantities of immune cells. Lastly, our analysis regarding PTEN status did not include PTEN inversions, fusions, inactivating somatic mutations, promoter methylations, and large-scale structural variants, which may be a source of bias in our evaluation of the impact of HemDel and HomDel effects.

Collectively, our findings based on TCGA, together with the extensive murine Pten studies by others8,9, suggest that loss of one copy of PTEN by mutation or deletion may be sufficient to promote tumorigenesis and anti-cancer immune evasion. However, for most human cancers, there have been limited biomarker studies on how hemizygous PTEN deletion and partial protein loss may influences outcome and treatment response.

Methods

Data download and processing

We initially downloaded raw and level 3 PanCancer normalized RNAseq, array-CGH, and SNV from 10,713 samples derived from 31 tumor types from The Cancer Genome Atlas (TCGA) cohort (www.gdc.cancer.gov). These initial cases were used to determine outcome, genomic alterations, and relative immune abundances. For the differential gene expression analysis (DEG), TCGA biolinks had 9793 cases were available from 30 primary tumors as detailed in Table 1.

Raw RNAseq experiments were performed through Illumina HiSeq 2000 RNA Sequencing platform. Copy number data were obtained from array-CGH experiments performed with the Affymetrix Genome-Wide Human SNP Array 6.0. DNA sequencing was performed in Illumina Genome Analyzer DNA sequencing. Clinical data for progression-free intervals and overall survival were obtained for all patients from the TCGA cohort. All analyses were conducted with data derived from primary tumors.

Defining PTEN deletions

PTEN deletions were characterized by using GISTIC-derived categorical SCNA. In brief, GISTIC identifies a cutoff for copy number loss or gain using the log2 continuous data from each individual segments, such as genes and other regions56. These analyses were performed by the TCGA group for the PanCancer analysis. We obtained, for our study, the Level 3 copy number calls for PTEN gene resulting from whole genome detection of copy number changes or point mutations inactivating either one or both copies of the gene20. The calls were defined as HomDel (two copies with loss/mutation), HemDel (one copy loss or mutation), intact PTEN (both copies present without mutation), and duplications of the PTEN gene57. PTEN duplications consisted of the presence of gain of an extra copy of the entire gene or regions at 10q23.31 that included PTEN. Our initial analysis of PTEN somatic mutations showed that in keeping with the literature15, this class of inactivation of the gene was infrequent in the TCGA cohort (Supplementary Fig. 1). We, therefore, choose to focus our analysis solely on SCNA affecting PTEN.

Gene expression and enrichment analyses

Analysis of gene expression was performed on the RNAseq level. Raw read counts for TCGA-tumors were downloaded using FANTOM-CAT/recount2 (https://jhubiostatistics.shinyapps.io/recount/)58,59. The subset of samples was compiled using each patient's clinical information, and only primary tumors were used in the analysis. Study groups were divided based on PTEN status, HomDel (two copies with loss/mutation), HemDel (one copy loss or mutation), and Intact PTEN (both copies present without mutation). Differential expression (DE) analysis was carried out using DESeq2 (v 1.32.0)59,60, and PTEN status was used as the design factor. For enrichment analysis of DE genes, we used the clusterProfile package (v 3.18.1)61,62. We used a P-adjusted value < 0.05 as the cutoff for the enrichment results. All analyses were conducted in RStudio software (R Foundation for Statistical Computing, R v4.1.2) and all results were plotted by pheatmap (v 1.0.12) and ggplot264. All results are displayed with PTEN Intact as a reference for statistical analysis.

Immunogenic and genomic effects of PTEN deficiency

To determine the association between PTEN inactivation and acquired DNA changes in tumors, we investigated the presence of genome doublings, homologous recombination defects (HRD), aneuploidy levels, tumor mutational burden, and intratumoral heterogeneity. These data were downloaded from the PanCanAtlas database for The Immune Landscape of Cancer from TCGA (https://gdc.cancer.gov/about-data/publications/panimmune)65. Genome doubling status was determined through ABSOLUTE algorithm66, which measures tumor ploidy and purity based on copy number and mutational signatures. Ploidy levels were determined by the quantity of DNA that each cancer cell presents after undergoing several numerical and structural chromosome aberrations. Clonality calls were employed to determine intratumor heterogeneity scores, which determine tumor copy number and point mutations as aggregates of clonal and subclonal components having varying ploidy levels. Tumor purity is a measurement that considers the fractions of normal cells within the bulk mass of tumor cells. Aneuploidy levels were calculated based on the total length of gained and deleted chromosome arms divided by the genome size67. Similarly, HRDs were calculated based on the copy number calls in with large (> 15 Mb) non-arm-level chromosome imbalances with loss of heterozygosity, breaks between genes larger than > 10 Mb, and subtelomeric regions harboring allelic imbalances. Groups with distinct PTEN inactivation statuses were compared by employing the Kruskal–Wallis test in R.

Prognostic effect of PTEN-deficiency

To determine if PTEN HemDel and HomDel status has a significant impact in predicting recurrence and death events, we performed survival analysis using Survival package in R. Log-rank tests and Kaplan Meier curves were generated through the Survival and Survminer packages in R. In addition, Cox Regression univariate models were obtained for all tumors, grouped and separately by tumor type.

In silico analysis of immune abundance in tumor types

To estimate the immune cell landscape in each tumor, we used the CIBERSORT deconvolution method (https://cibersort.stanford.edu/), which is an established tool to determine the abundance of immune cells using whole‐transcriptome data68. We separately imputed normalized gene expression data of brain, prostate, head and neck, stomach, cervical and colorectal tumors from the TCGA cohort. All results are displayed with PTEN Intact as a reference.

Ethics approval and consent to participate

Research Ethics Committee of the Clinical Hospital of the Faculty of Medicine of Ribeirão Preto, number 3.088.034.