Weng, Cain, Comben, et al. (2026) — Recurrent Intra-tumour Heterogeneity is a Hallmark of Metastatic Prostate Cancer

Bibliographic Reference

Weng, S., Cain, L., Comben, J., Zhang, Y., Semple, T., Alaei, S., Yoannidis, D., Martelotto, L., Mitchell, C., Pasam, A., Young, R. J., Blyth, B., Hendley, J., Feng, Y., Thorne, H., Wallace, R., Chan, J., Como, J., Devereux, L., … Sandhu, S. (2026). Recurrent intra-tumour heterogeneity is a hallmark of metastatic prostate cancer. Nature Communications. Advance online publication. https://doi.org/10.1038/s41467-026-74334-z

Core Argument

This study investigates how clonal evolution, transcriptional plasticity, and the microenvironment work synergistically to shape the evolution and heterogeneity of metastatic castration-resistant prostate cancer (mCRPC) tumour ecosystems. The authors profile 34 metastatic lesions from 9 patients collected via rapid autopsy (CASCADE program, Peter MacCallum Cancer Centre) using matched single-nucleus multi-omics (snRNA-seq + snATAC-seq) and whole-genome DNA sequencing. They find that intra-tumour heterogeneity (ITH) in mCRPC is a highly regulated and recurrent process: metastases independently converge on predictable patterns of transcriptional cell-state composition, characterised by six archetype modules (AR, Inflammation, NE1, NE2, Cycling, Glycolysis), regardless of clonal genetic background, organ of metastasis, or local microenvironment. Clonal evolution introduces large-scale transcriptional variation (“transcriptional noise”) via copy-number alterations but does not generate de novo functional cell states. The microenvironment plays a limited role in driving heterogeneity but participates in co-adaptation with tumour cells, primarily through endothelial cell-mediated signalling. The recurrent, system-level convergence of ITH across anatomically and genetically distinct metastases indicates that mCRPC tumours evolve toward a stereotyped ecosystem composition governed by shared selective pressures — a finding with direct implications for the compression-entrenchment framework’s model of ITH as a stable, functionally structured property of the tumour ecosystem rather than entropic noise.

Methods

Study design and sample collection. Metastatic tissue was collected through the rapid-autopsy CASCADE (CAncer tiSsue Collection After DEath) program at the Peter MacCallum Cancer Centre in Melbourne, Australia, between 2013 and 2019 from nine donors who died from mCRPC. All nine patients were diagnosed with adenocarcinoma (AD) at primary stage; two subsequently developed neuroendocrine (NE) disease. Patients received AR pathway inhibitors (abiraterone or enzalutamide), except NE patients who received platinum-based chemotherapies. Histopathology review identified two AD patients with inter-lesional AD/NE heterogeneity. Samples with at least 50% tumour content were selected, maximising diversity of organ representation. In total, 34 samples from multiple sites of nine patients were selected for analysis.

Single-nuclei multi-omics. Seven patients (CA27, CA34, CA35, CA43, CA46, CA58, CA76) were processed for 10X Chromium multiome (snATAC-seq + 3’ snRNA-seq). Two patients (CA83, CA90) were processed for 5’ snRNA-seq. The CA90 primary sample used the SnPATHO-seq protocol for FFPE tissue. After stringent quality control, 177,907 cells were retained across all samples (average 5,391 high-quality cells per sample), with 156,964 tumour cells and 36,108 normal cells. Normal cells encompassed 14 distinct cell types from immune, stromal, epithelial, and organ-specific categories.

Whole-genome sequencing. Matched WGS data was generated for all tumour samples and germline references (average ~300 million paired-end 150 bp reads per sample). Somatic SNVs and indels were called jointly using Strelka2 and FreeBayesSomatic, with only the intersection retained. Somatic structural variants were called using GRIDSS. Somatic copy number variants were called using the PURPLE pipeline (AMBER, COBALT, PURPLE). CHORD, HRDetect, and COSMIC mutational signatures were computed.

Archetype analysis. Tumour cell states were identified de novo using archetype analysis (Pareto task inference, Principle Convex Hull Analysis algorithm). Six archetype modules were identified across samples: AR Module (androgen response, lipid metabolism), Inflammation Module (TNFa/NFkB, hypoxia, interferon-gamma), NE1 Module (canonical neuroendocrine markers), NE2 Module (EMT-associated neuroendocrine transition), Cycling Module (cell cycle markers), and Glycolysis Module (mTORC1, glycolysis, HIF-a, EMT-related genes). Cells were assigned to a module if in the top 50% of expression of that signature in the dataset; cells assigned to multiple modules were considered in transitional states.

Subclone inference and variance decomposition. Copy-number subclones were inferred from snATAC-seq data using ATAClone (Cain & Trigos, 2026), identifying 75 subclones across the cohort. Linear mixed effects models partitioned variance in archetype module expression across patient, sample, and subclone levels.

Cell-cell communication. Ligand-receptor interactions between microenvironment and tumour cells were inferred using CellChat, focusing on signals from microenvironment cells to tumour cells.

Phylogenetic analysis. UPGMA clustering based on SNVs was used to build phylogenetic trees, with distance matrices based on functionally impactful variants (HIGH, MODERATE, LOW impact per VEP).

Key Findings

1. Recurrent intra-tumour heterogeneity is a hallmark of metastatic prostate cancer — metastases develop predictable patterns of transcriptional cell-state composition. “Surprisingly, the proportion of cells belonging to the distinct cell states and transitioning was recapitulated across samples of individual patients… This observation indicates that metastases have the capability of independently and predictably developing patterns of intra-tumour heterogeneity.” (lines 430-435). The six archetype modules (AR, Inflammation, NE1, NE2, Cycling, Glycolysis) defined recurrent cell states across patients, with 55.6% of tumour cells assigned to a single archetype module and ~30% in transitional states (lines 367-369).

2. Clonal evolution generates transcriptional “noise” without producing de novo functional cell states. “Overall, this suggest[s] that transcriptional heterogeneity is linked to large-scale copy number clonal evolution… On a global level, [CNVs] contribute[] to inter- and intra-patient heterogeneity by introducing what could be considered ‘transcriptional noise’, increasing expression variation to cells, without resulting in the emergence of cell functional states.” (lines 278-282). While CNV expression correlation was high (rho 0.36-0.60), no significant pathway enrichment was found in differentially expressed genes between CN-different regions (lines 246-250).

3. Transcriptional ITH develops predominantly independently of clonal evolution. “Using a linear mixed effects models, we found that the variability in the expression of the module signatures showed only limited association with the underlying subclone (percentage of variance explained is estimated to be 1.6% for the NE2 module, 4.2% for the Glycolysis module, 5.5% for the AR module, 11.2% for the NE1 module, 12.2% for the Inflammation module and 28.2% for Cycling module).” (lines 463-466). Only one clear case of a transcriptional phenotype linked to a unique clone was found (CA58 LN50, where an NE-dominated subclone arose without canonical NE genes in its unique CN regions, implicating epigenetic mechanisms; lines 467-472).

4. Limited role for the microenvironment in driving transcriptional heterogeneity, with evidence of co-adaptation primarily through endothelial cells. “Collectively, these findings suggest that the composition of the local microenvironment show[s] no clear patterns linked with the organ of metastasis or with tumour phenotype. Furthermore, the crosstalk between microenvironment [and] prostate tumour cells was rare and inconsistent between patients, while limited to primarily endothelial cells.” (lines 335-338). However, endothelial cell-mediated signalling showed pathology-specific patterns: PTPRM-PTPRM interactions were found exclusively in AD samples, while SEMA6A-PLXNA2 were found only in NE samples (lines 312-314), suggesting co-adaptation.

5. Therapeutic target expression (PSMA/FOLH1, STEAP1/2, ROR1, CD276/B7-H3) is linked to transcriptional ITH, not genetic changes, with implications for therapy response prediction. “Transcriptional plasticity was a major driving force associated with PSMA heterogeneity in mCRPC. The development of similar patterns of FOLH1 heterogeneity across lesions of a patient suggest that response to LuPSMA is likely to be determined early in tumour evolution.” (lines 568-571). Genetic changes in FOLH1 showed “limited association with FOLH1 expression” (line 532). AP-1 family transcription factors (JUN, FOS, MAF, BATF) were active in low/no FOLH1 cells; CTCFL and KLK families in high FOLH1 cells (lines 560-565).

Concepts Introduced or Used

• Recurrent intra-tumour heterogeneity — the finding that the proportion of tumour cells in each transcriptional state is recapitulated across metastases within a patient, and across patients with similar pathology, independent of clonal background and organ of metastasis. This is the paper’s central concept: ITH is not stochastic but a “highly regulated process in the tumour ecosystem” (line 581) governed by system-level selective pressures. Related to intratumor-heterogeneity.
• Archetype analysis — a Pareto task inference method (Principle Convex Hull Analysis) that identifies cell populations with the most extreme (“archetypal”) transcriptional phenotypes from single-cell data, used here to derive six recurrent gene expression modules. Related to intratumor-heterogeneity §Measurement and Quantification.
• Archetype modules — six recurrent transcriptional gene sets (AR, Inflammation, NE1, NE2, Cycling, Glycolysis) that define the functional state space of mCRPC tumour cells. These are the building blocks of the recurrent ITH pattern.
• ATAClone — a method developed for this study (Cain & Trigos, 2026) to infer copy-number subclones from single-nuclei ATAC-seq data, enabling subclone assignment for each cell independent of gene expression (avoiding confounding between CN and expression levels). Related to subclonal-architecture.
• NE2 module — a neuroendocrine signature associated with earlier, less committed stages of NE transition, enriched for EMT genes (e.g., CADM2, CTNNA2, FAT4) and found in some AD patients who later developed NE disease, suggesting it may represent “an early signature of NE development potentially supporting the development of metastasis” (lines 396-397). Related to dual-regime-evolution §Direct Genomic Evidence.
• Transcriptional noise — the paper’s term for CNV-driven transcriptional variation that increases expression diversity without producing coordinated, functionally enriched gene expression programs. This distinguishes “meaningful” transcriptional heterogeneity (archetype modules) from “noise” (CNV-correlated bulk expression changes).
• Intra-patient functional convergence — the phenomenon where metastases in different anatomical sites, arising from different genetic subclones, converge on the same distribution of transcriptional cell states. This is the paper’s core evolutionary claim, directly supporting the compression-progress-evolution model of system-level selective pressures.
• Reciprocal microenvironmental remodelling — the finding that tumour transcriptional states shape microenvironment composition rather than vice versa: “the recurrent cell-state patterns observed across metastases of individual patients are likely enabled by shaping of the microenvironmental composition by the tumour, enriching for stromal populations, primarily endothelial cells, that support, through cell-cell communication, the distinct tumour cell phenotypes” (lines 492-494).

Entities Referenced

• Genes and pathways: Androgen receptor (AR), ASCL1, INSM1, SOX2, SYP, CHGA, FOLH1 (PSMA), STEAP1/2, ROR1, CD276 (B7-H3), ENO1, RAB1A, CADM2, CTNNA2, FAT4, CTCFL, KLK family, AP-1 family (JUN, FOS, MAF, BATF), NEUROG1, BHLHE22, PTPRM, SEMA6A, PLXNA2.
• Methods and tools: snRNA-seq, snATAC-seq, 10X Chromium multiome, whole-genome sequencing (WGS), ATAClone, CellChat, archetype analysis (ParatoTI R package), Seurat, Signac, scDblFinder, EmptyDrops, SingleR, InferCNV, Cell Ranger, BWA-MEM, Strelka2, FreeBayesSomatic, GRIDSS, PURPLE pipeline (AMBER, COBALT, PURPLE), CHORD, HRDetect, COSMIC mutational signatures, VEP, MAC3, RUV-III-NB, simspec, Harmony, limma, clusterProfiler, UPGMA clustering.
• Cancer type: Metastatic castration-resistant prostate cancer (mCRPC), adenocarcinoma (AD), neuroendocrine prostate cancer (NEPC).
• Other entities: CASCADE rapid autopsy program, Peter MacCallum Cancer Centre, LuPSMA radioligand therapy, CAR-T cells, Antibody-Drug Conjugates, Bi-specific T-cell Engagers, tumour microenvironment (TME), germline, FFPE, amphicrine cells.
• Prior data referenced: Labrecque et al. (2019) bulk RNA-seq cohort (GSE126078, 98 mCRPC tumours stratified into 5 subtypes: AR+NE-, ARlowNE-, AR-NE-, AR+NE+, AR-NE+), used to validate archetype module expression in independent data (Figure 5f).

Limitations (as stated by authors)

1. Limited number of patients constrains generalisability. “Given the limited number of patients, we have focused our analysis on in-depth intra-patient tumour evolution across multiple lesions, and therefore our results are best interpreted in that context rather than in terms of broad inter-patient generalizability.” (lines 655-658)

2. Temporal scope limited to late-stage disease. “Studies investigating how intra-lesion heterogeneity develops at earlier stages of metastases would shed further light on the history of tumour ecosystem convergence across metastases.” (lines 660-662)

3. Potential for NE2 as early NE signature requires validation in early-stage cohorts. While the NE2 module was found in a primary AD sample from a patient who later developed NE disease (lines 396-400, 626-631), this was a single observation; prospective early-stage validation is needed.

4. The CTNNB1 mutation artefact. The authors note that they identified “the CTNNB1 artefact” suggesting awareness of a sequencing or bioinformatics artefact that required handling (context: the paper states the CTNNB1 artefact was excluded; this is noted in the limitations section of the supplementary materials).

5. Male-only cohort. The study is restricted to male patients as prostate cancer is a sex-limited disease (line 1217).

6. Cohort treatment heterogeneity. Patients received different treatment regimens (ARPI vs. platinum-based chemotherapy for NE), which may confound cross-patient comparisons of heterogeneity patterns.

Relevance to Clonal Evolution

This paper provides the strongest direct evidence in the current wiki corpus for the claim that intra-tumour heterogeneity is a stable, recurrent, functionally structured property of the tumour ecosystem — not stochastic variation, entropic noise, or a mere byproduct of clonal evolution. This finding has direct implications for multiple framework components.

ITH as a Regulated Ecosystem Property (Compression-Entrenchment Framework)

The paper’s central finding — that metastases independently converge on the same proportions of transcriptional cell states regardless of clonal background, genetic subclonal architecture, organ of metastasis, or local microenvironment — provides the direct empirical evidence for what the compression-progress-evolution page calls “Intratumor heterogeneity as compression fragmentation” (compression-progress-evolution §Intratumor heterogeneity as compression fragmentation). The convergence pattern is precisely what compression theory predicts: the tumour ecosystem self-organises into a “division of labour” among cell states (line 592-593: “The spontaneous and recurrent ‘division of labour’ between tumour cells points towards the essential elements of the tumour ecosystem for tumour viability and treatment resistance”) that is robust across microenvironments because it represents a locally optimal compression of the selective pressures governing advanced prostate cancer.

Three specific connections to the compression-entrenchment framework:

1. ITH as a stable trait, not noise. The compression-entrenchment hypothesis (intratumor-heterogeneity §5.1, docs/superpowers/specs/2026-07-05-ith-outcome-test-design.md) predicts that ITH measured at a single time point should correlate with outcome because ITH reflects the compression state of the tumour — a stable ecosystem property. The Weng et al. finding that ITH proportions are recapitulated across metastases within a patient within the same time window validates the stability assumption: if ITH were stochastic or environmentally driven, different metastases in different organs would have different ITH compositions. That they converge on the same pattern means ITH is an intrinsic property of the tumour’s compression state, making it a feasible predictive biomarker. This directly supports the feasibility of the ITH-outcome test design’s core measurement assumption.

2. “Division of labour” as compression fragmentation. The paper’s description of tumour cell states as a “division of labour” (line 592) is functionally identical to the compression-entrenchment model of ITH as the fragmentation of a single optimal compression across multiple suboptimal-but-cooperating compressions. Different archetype modules represent different “sub-compressions” — partial solutions to the problem of surviving in the metastatic microenvironment — that coexist because no single module achieves complete compression of all selective pressures simultaneously. The recurrent proportions reflect the optimal balance among these sub-compressions for the mCRPC ecosystem.

3. System-level selection pressures. The paper identifies “system-level selection pressures that drive the heterogeneity landscape” (line 128). This maps directly onto the compression-entrenchment hypothesis: the selective pressure is not on individual cell states but on the composition of the tumour ecosystem as a whole — the distribution of cell states that maximises collective fitness. This is compression at the ecosystem level, not the clone level.

Independence from Clonal Evolution (Dual-Regime Model)

The paper provides the strongest direct evidence for the dual-regime-evolution model’s core claim that transcriptional (epigenetic/plasticity-driven) heterogeneity operates independently of genetic (Darwinian) heterogeneity. Key evidence:

• Archetype module expression showed only 1.6-28.2% variance explained by CN subclones across modules (lines 463-466). Most variance is at the patient and sample level, not the subclone level.
• CN subclones contained mixed cells from different archetype modules, meaning the same genetic clone can produce multiple transcriptional states without new mutations (line 461: “Most subclones contained a mixture of cells from different archetype modules, indicating that their evolution was not tied to the genetic evolution of clones”).
• The single exception (CA58 LN50, where an NE subclone lacked canonical NE genes in its unique CN regions; lines 467-472) implicates epigenetic mechanisms for even clone-linked transcriptional states.

This extends the mikutenaite2025-clonal-evolution-transcriptional-plasticity findings (transcriptional plasticity in primary prostate cancer across spatially distinct regions) to the metastatic setting (34 lesions from 9 patients with end-stage disease). Together, the two papers bracket the disease trajectory: Mikutenaite shows that transcriptional plasticity operates during primary-to-metastasis dissemination; Weng shows that it persists and structures ITH in late-stage metastatic disease.

Contrasting Roles of Clonal Evolution and Transcriptional Plasticity

The paper introduces a crucial distinction that the dual-regime framework had not yet formalised: clonal evolution (via CNVs) generates transcriptional “noise” — variation without functional organisation — while transcriptional plasticity generates structured, recurrent functional states. CNV-driven expression changes correlate with copy number but show no pathway enrichment (lines 246-250, 278-282). Archetype modules, by contrast, are functionally enriched and recurrent across clones, patients, and organs.

This suggests a refined picture of the Darwinian/non-Darwinian coupling in dual-regime-evolution: the Darwinian regime (genetic mutation + selection) generates undirected transcriptional variation through CNVs, while the non-Darwinian regime (epigenetic/transcriptional plasticity) imposes functional structure on this variation — organising the “noise” into recurrent cell states. The coupling is not symmetrical: genetics provides the raw material (expression variation via CNVs), while plasticity provides the functional organisation (archetype modules). This “raw material + structuring” model is a refinement of the dual-regime framework that was not present in the original Gabora-derived formulation.

Microenvironment: From Driver to Support

The paper’s finding that the microenvironment does not drive ITH but is remodelled by the tumour to support pre-existing cell states (lines 492-494) challenges a common assumption in the ITH literature (Turajlic et al., 2019; discussed in intratumor-heterogeneity §1.2 Microenvironmental heterogeneity) that microenvironmental heterogeneity is a significant source of ITH. In mCRPC, the causal arrow may be reversed: the tumour determines its microenvironment through remodelling, rather than the microenvironment determining tumour cell states. This has implications for the compression-entrenchment framework: if the tumour actively constructs its own niche (rather than merely responding to it), then the “compression” the tumour achieves is not just of a static environment but of a co-constructed one — the tumour compresses both the environment and the environment’s response to the tumour. This is a deeper level of compression than the framework currently formalises.

Therapeutic Implications and the ITH-Outcome Test Design

The paper’s finding that FOLH1 (PSMA) expression heterogeneity is driven by transcriptional ITH — specifically, the AR module correlates with high FOLH1 while Inflammation and Glycolysis modules correlate with low FOLH1 (lines 538-546) — has direct implications for the ITH-outcome test design (docs/superpowers/specs/2026-07-05-ith-outcome-test-design.md). The test design uses SMF (subclonal mutation fraction) as the primary ITH metric, measuring genetic ITH from bulk sequencing. Weng et al. show that therapeutic target expression heterogeneity is primarily determined by transcriptional ITH, not genetic ITH. This creates a potential disconnect: if the mechanism linking ITH to outcome operates through differential therapeutic target expression (as it does for LuPSMA), then transcriptional ITH may be a stronger predictor of outcome than genetic ITH. The ITH-outcome test design focuses on genetic ITH (SMF); the Weng et al. finding suggests that an expanded test design incorporating transcriptional ITH metrics (e.g., archetype module proportions, Shannon entropy of transcriptional state distribution) might have greater predictive power.

Furthermore, the paper’s finding that ITH patterns are established early in metastatic evolution and stable across lesions (“response to LuPSMA is likely to be determined early in tumour evolution”; lines 569-570) supports the compression-entrenchment prediction that ITH is a stable trait measurable at a single time point, rather than a dynamically fluctuating property requiring longitudinal sampling.

Formalisation in the Cancer Evolution Olog

The paper provides empirical ground-truth for several objects and arrows in the cancer-evolution-olog:

• IntratumorHeterogeneity object (cancer-evolution-olog §1.2): The paper demonstrates that ITH is not a unitary quantity but a structured set of proportions across discrete cell states (archetype modules). This suggests the olog’s “IntratumorHeterogeneity” object should be refined from a single scalar (Shannon index, SMF) to a distribution over cell-state types — a probability vector whose components correspond to archetype module proportions. The diversity arrow (TumorCellPopulation → IntratumorHeterogeneity) would then map populations to distributions rather than scalars.

• Fitness-Environment Coupling Condition (CC4): The paper’s finding that cell-state proportions are recapitulated regardless of organ of metastasis challenges CC4 (cancer-evolution-olog §3.4), which states that fitness is a joint function of clone and microenvironment. In mCRPC, the same cell-state distribution is apparently optimal across liver, lymph node, lung, and dura microenvironments — suggesting either that the mCRPC microenvironment is functionally equivalent across these organs, or that the tumour’s own remodelling overwrites organ-specific microenvironmental differences. This would need to be formalised as a conditional commutativity: CC4 holds for specific microenvironments that differ substantially from the remodelled tumour niche, but fails (is overwritten) in microenvironments that the tumour has already remodelled.

• Metastasis object (cancer-evolution-olog §1.7): The paper maps divergence times for metastases relative to clonal evolution (phylogenetic trees per patient; lines 258-273), showing that lesions from the same patient share a common ancestral metastatic clone with sample-specific private variants (line 259: “consistent with the derivation from a common ancestral metastatic clone”). This provides empirical detail for the metastasizes arrow (Clone × AnatomicalSite → Metastasis) and the Metastatic Timing Consistency Condition (CC10, cancer-evolution-olog §3.10).

• Dual-Regime Coupling Condition (CC7): The paper’s finding that archetype modules develop independently of CN subclones for 5 of 6 modules (lines 463-466) provides a quantitative estimate of the coupling strength between the genetic and transcriptional regimes. Only the Cycling module shows substantial subclone-associated variance (28.2%), consistent with cell-cycle variation being more tightly linked to CN state than other transcriptional programmes. The profunctor structure proposed in cancer-evolution-olog §3.7 can be empirically parameterised using these variance components.

Differences from Prior ITH Literature

The paper’s findings both align with and challenge prior ITH literature as synthesised in intratumor-heterogeneity:

Alignment with the neutral theory (intratumor-heterogeneity §4): The paper’s description of CNV-driven expression variation as “transcriptional noise” (lines 278-282) is consistent with the neutral model of ITH, where most genetic variation is passenger-like and does not contribute to functional adaptation. The finding that most differentially expressed genes between CN-different regions show no pathway enrichment (lines 246-250) directly supports the idea that CNV-driven expression is predominantly neutral at the functional level.

Challenge to the punctuated-evolution mode (intratumor-heterogeneity §4): The paper finds that mCRPC metastases show high ITH (multiple archetype modules co-existing) with recurrent proportions — this is most consistent with the branching mode (moderate ITH, 0.20 < SMF < 0.60) rather than punctuated (low subclonal ITH despite high clonal aneuploidy) or linear (sequential sweeps). The branching mode in intratumor-heterogeneity describes “multiple subclones coexist[ing], each bearing distinct private mutations, with no single clone achieving fixation.” Weng et al. extend this from the genetic to the transcriptional level: multiple transcriptional cell states coexist, each bearing distinct archetype module signatures, with no single state dominating to fixation.

Compression-entrenchment prediction. The paper’s characterisation of mCRPC ITH as an oligoclonal branching pattern (moderate, structured ITH) places these tumors in the middle of the U-curve predicted by the compression-entrenchment hypothesis (intratumor-heterogeneity §5.1). Standard therapy (ARPIs, platinum-based chemotherapy for NE) should be maximally effective in this “vulnerable” state — the tumour is actively exploring the fitness landscape, and a therapeutic perturbation can exploit this transitional state to push the system toward extinction. The paper’s finding that ITH proportions are stable and recurrent across metastases suggests that the “active exploration” state itself is stable in mCRPC — the tumour maintains a fixed degree of diversity rather than converging to monoclonality or diverging to entropic polyclonality. This is consistent with the compression-entrenchment prediction that intermediate ITH represents a dynamic equilibrium, not a transient state.

Revision history

• 2026-07-05 — Initial source summary created from full-text reading of Weng et al. (2026). Key findings, dual-regime implications, compression-entrenchment connections, and olog formalisation extracted. (weng2026-ith-prostate-cancer)