Passenger co-deletion confounds glutaminolysis signatures anchored on PTEN loss: a cautionary case for location-aware signature design

Across fourteen tumor types the glutaminolysis signature decreased as PTEN copy-number loss increased, significantly in twelve after Benjamini-Hochberg correction, with the strongest effect in prostate cancer (Spearman rho = -0.54, adjusted P = 5.3 x 10^-37), and no tumor type showed the positive trend predicted by the signaling model. Prior evidence linking PTEN loss to glutaminolysis came largely from individual cell-line models and from prostate cancer, and no study had asked across many tumor types, with attention to gene dosage, whether the glutaminolytic transcriptional program actually tracks PTEN copy-number loss. Fourteen tumor types from the TCGA PanCancer Atlas selected by a pre-specified inclusion rule of at least ten tumors per dosage tier, analyzed by two-tailed Spearman correlation with multiple-testing correction, with effect sizes reported as Cliff's delta and bootstrap confidence intervals from 2,000 resamples; effect sizes varied widely, explaining 29.1 percent of variance in prostate but 0.5 percent in breast and 0.4 percent in stomach.

The association is governed by chromosomal position rather than pathway membership: GLUD1 and GOT1 lie at 10q23-q24 and are co-deleted with PTEN, thirty-seven neighboring genes carrying no glutaminolysis annotation tracked PTEN copy number with a mean correlation of 0.843 compared with 0.842 for GLUD1 and GOT1, while the three signature genes on other chromosomes averaged 0.034, and co-deletion fidelity declined monotonically with distance from the PTEN transcription start site (rho = -0.993). No prior study had used the chromosomal location of signature genes as a direct discriminator between a regulatory association and structural passenger co-deletion. Each of thirty-seven unannotated genes across 10q23-q24 was correlated with PTEN copy number using Ensembl (GRCh38) positions, with no nominal P value reported because neighboring genes share copy-number profiles; the deletions involved are focal, with a median span of co-deleted sequence of 0.72 Mb and 90 percent of homozygously deleted tumors confined within 2 Mb of PTEN.

Two adjustments give different answers: adjusting for genome-wide aneuploidy left the chromosome-10 associations essentially unchanged (for example GLUD1 partial rho -0.59 in prostate against an unadjusted -0.59), whereas conditioning on each gene's own copy number abolished the association, reducing the mean chromosome-10 coefficient from -0.321 to -0.022, while the off-chromosome-10 genes were unaffected throughout. A genome-wide aneuploidy score cannot isolate loss of a single chromosome arm, so this widely applied control would here be read as evidence against a structural explanation, whereas gene-level copy-number adjustment is the direct test. Spearman partial correlations were computed controlling separately for the per-sample aneuploidy score, a 10q arm-level proxy defined as the median copy number of interval genes more than 2 Mb from PTEN, and each gene's own copy number, allowing the three adjustments to be compared side by side.

A mutation-based natural experiment separated loss of PTEN function from loss of the chromosomal segment: pooled across all cohorts, GLUD1 was lower in PTEN point-mutant copy-neutral tumors than in copy-neutral wild-type tumors (435 against 3,561 tumors, difference -0.24, P = 1.0 x 10^-8), but this was driven by endometrial carcinoma, and excluding that lineage left no detectable difference (P = 0.22, equivalence P = 0.015), whereas hemizygous deletion (mean -0.44) and homozygous deletion (mean -0.89) lowered GLUD1 dose-dependently in every analysis; in DepMap, GLS gene-effect scores did not differ across PTEN dosage tiers (Kruskal-Wallis P = 0.74). A passenger gene should respond to deletion but not to mutation alone, and this natural experiment had not previously been used to test the PTEN-glutaminolysis association. Two-tailed Mann-Whitney U tests per tumor type and pooled after within-cohort z-scoring, with equivalence assessed by two one-sided tests against a margin of 0.25 within-cohort standard deviations fixed in advance; the endometrial exception was larger rather than smaller in microsatellite-stable tumors (difference -0.53, P = 1.2 x 10^-6), excluding hypermutation as its cause.

The result applies to transcriptomic analyses of solid tumors in which PTEN dosage is defined by copy number, and its central conclusion rests on gene-level GISTIC copy number measured in tumor DNA, which is unaffected by tumor purity and stromal dilution; the four-step screening procedure is proposed by the authors as a screening workflow rather than a validated instrument, with the first three steps requiring only data already in hand and the fourth requiring independent functional data such as CRISPR dependency screens, bearing in mind that dependency is itself context-dependent. For studies specifically aiming to measure transcriptional regulation of glutaminolysis in the context of PTEN loss, the authors suggest a chromosome-aware signature that excludes the 10q23-q24 passengers GLUD1 and GOT1 and relies on the off-chromosome-10 members (GLS, SLC1A5, GPT2), optionally supplemented with additional non-chromosome-10 glutaminolytic genes such as GLS2 or SLC38A1.

Copy-number-defined dosage does not capture point mutations, promoter methylation, or structural variants, so some functionally PTEN-deficient tumors will be classified as intact, biasing toward the null; epigenetic silencing can itself have metabolic consequences distinct from copy-number loss. Thresholded GISTIC calls can assign homozygous status within a broad low-level loss, while low tumor purity works in the opposite direction by masking true homozygous deletion, and both are sources of misclassification in the dosage tiers. The glioblastoma cohort had only thirteen PTEN-intact tumors, an unstable baseline. GLS is a weak dependency in most cell lines irrespective of PTEN status, so the DepMap result bounds rather than proves the absence of a functional dependency. Bulk RNA-seq is subject to tumor-purity and stromal-dilution differences, and GLS and SLC1A5 are expressed in immune cells, so the expression-level interpretation of the off-chromosome-10 sub-signature warrants caution. The basis of the residual prostate association is unresolved, and that analysis was examined after the pan-cancer result was in hand and is reported as exploratory. The mechanism by which mutation alone lowers GLUD1 in endometrial carcinoma is not established. Whether the same measurement confound arises at other deletion-prone loci is a question the present data cannot settle.