Desmoid tumors, also called desmoid-type fibromatosis, account for approximately 3% of all soft-tissue tumors and present most commonly around the median age of 37 to 39 years. They are histologically bland, consisting of spindle cells resembling fibroblasts without the nuclear pleomorphism characteristic of malignant sarcomas. Despite not metastasizing, desmoid tumors carry a high rate of local recurrence after surgical resection, and their unpredictable behavior makes prognosis and treatment planning genuinely difficult. The majority of cases are sporadic, though a small subset arises in the context of familial adenomatous polyposis (FAP), a germline APC mutation syndrome that dramatically increases desmoid risk.
The CTNNB1 driver: Mutations in the CTNNB1 gene, encoding beta-catenin, are found in the vast majority of sporadic desmoid tumors. The most frequent mutations affect exon 3 at codons T41A and S45F/S45P, disrupting phosphorylation sites that normally tag beta-catenin for proteasomal degradation. This stabilizes beta-catenin in the nucleus, where it promotes transcription of Wnt target genes driving tumor cell proliferation. CTNNB1 mutations at T41A and S45F/S45P are largely mutually exclusive, and the S45F mutation has been associated with a higher recurrence rate in some series.
The clinical problem this study addresses: Two prior transcriptomic studies by Salas and colleagues (36-gene microarray signature) and Kohsaka and colleagues (3-gene RNA-seq signature) proposed gene expression-based tools for predicting desmoid recurrence after surgery. However, these signatures were derived from different methodologies, used fresh-frozen versus FFPE tissue, and yielded discordant results when applied to independent cohorts. The key question this study asks is whether the inconsistency across published signatures reflects true biological differences between cohorts, or whether it reflects an underappreciated property of desmoid tumors themselves, namely profound intratumor heterogeneity that makes single-biopsy molecular profiling inherently unreliable.
The study, conducted at Stanford Health Care and Vanderbilt University Medical Center and published in Clinical Cancer Research in 2024, addresses this question through a combination of broad transcriptomic profiling across 20 patients and deep multiomic analysis of tumor specimens taken from multiple anatomic regions of tumors from three patients, covering primary tumors and successive recurrences.
The study used archival formalin-fixed, paraffin-embedded (FFPE) specimens from two separate analyses. The first involved RNA sequencing (RNA-seq) of 31 specimens from 20 patients with primary desmoid tumors, 13 of whom developed local recurrence after surgical resection. The second, more intensive analysis applied a full multiomic battery to 24 specimens from primary and recurrent tumors in 3 patients, with 7 to 9 specimens profiled per patient. All diagnoses were confirmed by a surgical pathologist (M. van de Rijn), and specimens were required to contain at least 90% lesional spindle cells before proceeding to molecular analysis.
RNA sequencing: Whole-transcriptome sequencing libraries were prepared from 1 microgram of total RNA using TruSeq Stranded Total RNA Library Prep Kit and sequenced on an Illumina HiSeq 2000 in 2x101 bp paired-end mode. Reads were mapped to the GRCh37/hg19 reference genome using Rsubread, and differential expression analysis was performed with DESeq2. Multiple regions of the same primary tumor were profiled for six patients to directly measure intratumor transcriptomic variability.
DNA copy-number and methylation profiling: Copy-number and allelic ratio profiling was performed on 75 ng of DNA using Infinium OncoScan FFPE Assay (Affymetrix), analyzed with Nexus Express for OncoScan3 using the SNP-FASST2 algorithm. DNA methylation was profiled on 500 ng of DNA using Infinium MethylationEPIC microarrays (Illumina), processed using the minfi and wateRmelon Bioconductor packages. Quality thresholds required greater than 90% of probes measured with P less than 0.05 on at least three beads.
Whole-exome sequencing and phylogenetics: Whole-exome sequencing was performed on 22 tumor specimens from the three deeply profiled patients, with matched normal tissues serving as germline controls. Libraries were sequenced at 150x median coverage. Somatic mutations were called using MuTect2, annotated with the Ensembl Variant Effect Predictor, and used to reconstruct phylogenetic trees of tumor evolution using the LICHeE algorithm. Phylogenetic analysis required 0% variant allele fraction (VAF) in normal tissue and at least 10% VAF in tumor, yielding a median of 149 somatic mutations per specimen (range 1-594). CTNNB1 mutations were validated by Sanger sequencing.
The study identified 14 differentially expressed genes between the 13 patients who developed local recurrence and the 7 who did not from the primary RNA-seq cohort of 20 patients. This 14-gene signature was then evaluated alongside the previously published signatures in the same dataset. When the authors examined multiple regions from the same primary tumor in six patients, they found that the expression levels of signature genes differed substantially between different sampling sites within individual tumors, creating a scenario where the same tumor could be classified as high-risk in one region and low-risk in another.
Salas 36-gene signature: Of the 36 genes in the Salas and colleagues microarray-derived signature, only the FECH gene showed statistically significant differential expression in the RNA-seq dataset (unpaired two-sided t test, P = 0.01). Critically, the direction of the association was reversed: in the Salas study, higher FECH expression predicted recurrence, whereas in this dataset, higher FECH expression was seen in non-recurring tumors. The authors also observed substantial intratumor variability in FECH expression across multiple regions of the same tumor in individual patients.
Kohsaka 3-gene signature: The Kohsaka and colleagues signature, based on elevated IFI6 and LGMN expression as markers of high recurrence risk, showed neither gene to be differentially expressed in this dataset (P = 0.8 and P = 0.4, respectively). Intratumor heterogeneity of IFI6 expression was especially marked in patients 2 and 20, and LGMN showed similar variability in patient 1. These failures to replicate are not necessarily explained by study design differences, since the observed intratumor variability would itself prevent reliable replication regardless of which tissue region happened to be biopsied.
Protein-level confirmation: IHC validation on a tissue microarray containing duplicate 1 mm cores from 13 of the 20 patients confirmed the transcriptomic heterogeneity at the protein level. Four markers were examined: EFL1 from the newly developed signature, and HDAC6, TFG, and hnRNPC from the Salas signature. TFG and hnRNPC were strongly expressed across all cores, while HDAC6 and EFL1 showed heterogeneous staining within and between tumors, with adjacent cores sometimes exhibiting clearly different staining intensities. This protein-level confirmation removes the possibility that the observed transcriptomic variability is purely a technical artifact of RNA quality differences between FFPE cores.
Patient 1 presented with a 19-cm mesenteric desmoid tumor and developed a 4.5-cm local recurrence one year after surgical resection with positive margins. No adjuvant therapy was given between the primary resection and recurrence. Six regions of the primary tumor and three regions of the recurrent tumor were analyzed by the full multiomic panel. This case was particularly informative because the lack of intervening treatment removes treatment-induced clonal selection as a confounding variable for any molecular differences observed between primary and recurrent disease.
Dual CTNNB1 mutations: Whole-exome sequencing revealed two distinct somatic mutations in the CTNNB1 gene. All regions of both the primary and recurrent tumors carried the T41A mutation, establishing it as a clonal (truncal) driver event. However, two regions of the recurrent tumor also carried the S45P mutation at a lower VAF than T41A, indicating that the S45P mutation arose as a subclonal event, likely during the clonal evolution from primary to recurrent tumor. These findings were validated at the DNA level by Sanger sequencing and at the RNA level by identifying the mutations in RNA-seq reads, representing a rare documented case of "two-hit" CTNNB1 mutations within a single tumor.
Genomic and epigenomic profiles: All sampled regions of primary and recurrent tumors showed normal diploid karyotypes with only a few DNA copy-number alterations, consistent with the overall genomic stability of desmoid tumors. In contrast, DNA methylation profiles were clearly distinct between the primary and recurrent tumors, indicating that substantial epigenomic reprogramming occurred during recurrence even in the absence of heavy treatment exposure. One region of the primary tumor had a divergent transcriptomic profile from the other five regions, showing enrichment of 12 genes in the beta-oxidation pathway in mitochondria (MSigDB annotation), suggesting focal metabolic reprogramming within a spatially restricted subpopulation of tumor cells.
Phylogenetic reconstruction: The LICHeE-based phylogenetic tree confirmed that the T41A mutation is a clonal, ancestral event shared by all primary and recurrent tumor regions, with most additional somatic mutations being private to individual regions. This branched evolution pattern, rather than a linear progression, supports a model of parallel clonal diversification rather than sequential replacement of one clone by another. Two candidate genes showing differential methylation between primary and recurrent tissues in this patient were HTRA3 and TCTEX1D1, though their functional significance in desmoid biology remains to be established.
Patients 2 and 3 had more complex clinical histories with multiple recurrences and different treatment regimens, providing a framework for understanding how therapy-induced selection pressure shapes desmoid tumor evolution. Patient 2 had a 6.5-cm buttock desmoid tumor resected with positive margins, received radiotherapy, and then developed recurrences at 4, 10, and 11 years after initial resection, with imatinib treatment before the first profiled recurrence and vinblastine plus methotrexate between the second and third recurrences. Patient 3 had a 3.5-cm arm desmoid tumor followed by radiotherapy and multiple recurrences at 1, 4, and 5 years, with toremifene and meloxicam between recurrences 2 and 3.
Patient 2 - new mutations in recurrent disease: All seven specimens from primary and recurrent tumors of patient 2 carried the same CTNNB1 T41A mutation, confirming it as the ancestral clonal driver. The third recurrent tumor, which appeared 11 years after initial resection, showed three new mutations in DNA not present in the primary tumor or second recurrence: mutations in MKI67 (a marker of cell proliferation), MYOF (myoferlin, involved in membrane repair and vesicle trafficking), and HAUS1 (a component of the augmin complex involved in microtubule nucleation). RNA-seq data confirmed the MYOF mutation in one region of the third recurrence and showed that the subclonal HAUS1 mutation was present in only a fraction of tumor cells, revealing intratumor mutational heterogeneity even within a single recurrent lesion.
Patient 3 - focal genomic instability: All but one exome-sequenced specimen from patient 3 carried the CTNNB1 S45F mutation as the driver event. One region of the first recurrence and the entire second recurrence showed a high number of focal DNA copy-number alterations not present in other specimens. The second recurrence had a highly distinct epigenomic profile from all other primary and recurrent specimens, as shown by unsupervised clustering of methylation data. The primary tumor and first recurrence showed high intratumoral transcriptomic heterogeneity, while the second and third recurrences showed more homogeneous but distinct expression profiles compared to earlier tumors.
These data from patients 2 and 3 contrast directly with patient 1, where no adjuvant therapy was administered and the molecular profiles of the recurrent tumor were more similar to the primary tumor. This comparison supports the interpretation that treatment exposure accelerates clonal divergence and epigenomic reprogramming in desmoid tumors, potentially contributing to therapeutic resistance over successive recurrences.
A key comparative finding of this study is that different molecular layers showed distinct patterns of heterogeneity within and between tumors. Transcriptomic profiles showed the highest degree of variability both within individual tumors (intratumor) and between primary and recurrent tumors (intertumor). The expression of hundreds of genes differed substantially between multiple regions of the same tumor, making gene expression signatures particularly vulnerable to sampling bias at any given biopsy site.
DNA methylation patterns: In contrast to transcriptomics, DNA methylation profiles were generally similar between different areas within individual primary or recurrent tumors, indicating that epigenomic identity is relatively spatially homogeneous within a single tumor mass at a given timepoint. However, methylation showed substantial differences between the primary tumor and subsequent recurrences, particularly in patients 2 and 3. This temporal divergence in methylation suggests that the epigenome undergoes major reprogramming during recurrence, potentially reflecting adaptation to new microenvironments, treatment exposures, or clonal selection of epigenetically distinct subpopulations.
Mutational and copy-number profiles: Somatic mutation profiles were intermediate in their heterogeneity. CTNNB1 mutations were broadly clonal and present across all regions, establishing them as early driver events. Other somatic mutations were largely private to individual tumor regions or specific recurrences, consistent with branching clonal evolution. Copy-number alterations were generally sparse across all specimens, consistent with the known genomic stability of desmoid tumors, with the notable exception of focal copy-number gains and losses in specific recurrent specimens from patient 3.
This layered portrait of heterogeneity has direct implications for biomarker development and clinical trial design. Biomarkers derived from gene expression data are most susceptible to sampling bias and may require multi-region profiling or spatial transcriptomics to generate stable estimates. Methylation-based biomarkers may be more spatially robust for characterizing a given tumor at one timepoint, though they are still subject to temporal evolution across recurrences. Mutation-based approaches targeting CTNNB1 exon 3 represent the most reliable single-biopsy target, as this mutation is present in essentially all cells across all regions and timepoints.
Sample size and patient selection: The most significant limitation is the small number of patients in the multiomic analysis. Three patients provide the depth of multiregion, multi-recurrence profiling needed to characterize tumor evolution, but the specific patterns observed, including which molecular layers diverge and how quickly, may not be representative of the full spectrum of desmoid behavior. Patient 3, who showed the most extreme focal genomic instability, cannot be taken as typical. The primary cohort of 20 patients for the transcriptomic analysis provides a somewhat larger foundation for assessing signature heterogeneity, but is still modest for identifying a robust 14-gene recurrence signature.
FFPE tissue and methodology differences: The study analyzed FFPE specimens, whereas both the Salas and Kohsaka signatures were developed using fresh-frozen tissue or applied RNA-seq to different cohorts. FFPE RNA extraction is inherently noisier than fresh-frozen RNA due to RNA degradation and crosslinking, and some of the gene-level discrepancies observed between this study and prior work may partly reflect technical factors rather than purely biological heterogeneity. The authors acknowledge this as a source of uncertainty in directly comparing findings across studies.
Immune landscape and mismatch repair: The study examined mismatch repair protein expression by IHC (MLH1, MSH2, MSH6, PMS2) in tissue microarrays from patients 2 and 3, finding intact expression in all specimens, arguing against microsatellite instability as a driver of the observed mutational heterogeneity. CIBERSORT analysis of the RNA-seq data did not identify significant immune cell infiltrates in desmoid tumors, which the authors note may limit the relevance of immunotherapy-based approaches and of immune heterogeneity as a contributor to clinical behavior. However, the absence of immune signatures could also reflect a limitation of CIBERSORT's sensitivity for detecting stromal immune populations within FFPE-derived expression data.
Clinical outcomes data: The recurrence information used to build the transcriptomic signature was based on post-surgical follow-up from two academic centers. Variability in surgical technique, margin assessment, and follow-up interval could introduce outcome misclassification, potentially weakening the signal for any prognostic signature in this dataset.
The core clinical message of this study is that single-biopsy molecular profiling of desmoid tumors is insufficient to characterize their full molecular complexity. The prognostic signatures from Salas and colleagues and Kohsaka and colleagues, while developed with rigor from larger cohorts, may have failed to validate in this and other independent datasets partly because the tumors themselves contain spatially mixed populations with opposing prognostic profiles. This observation has immediate practical consequences for clinical trial design: if a patient is enrolled based on a biopsy classified as high-risk by one signature, a different biopsy from the same tumor might classify it as low-risk, leading to systematic heterogeneity in the treatment arm that would obscure any true therapeutic signal.
Multi-region biopsy and liquid biopsy: The study's findings motivate multi-region biopsy protocols for molecular characterization of desmoid tumors, particularly for research purposes. In practice, however, multiple core biopsies from a single desmoid tumor may not always be feasible given tumor location and size. Liquid biopsy approaches measuring circulating tumor DNA (ctDNA) derived from shed cells across the entire tumor mass are mentioned in prior literature as potentially providing a more integrated molecular snapshot. A companion study referenced in this paper demonstrated the utility of a combination approach for detecting ctDNA in leiomyosarcoma; similar methodology applied to desmoid tumors could potentially overcome sampling bias by integrating signals from heterogeneous tumor subpopulations.
Therapeutic resistance and treatment sequencing: The observation that recurrent tumors from patients 2 and 3 (who received adjuvant therapy) showed markedly divergent molecular profiles compared to their primary tumors, while patient 1's recurrence (without adjuvant therapy) was more molecularly similar to the primary tumor, suggests that treatment exposure drives clonal selection and molecular reprogramming. This has important implications for understanding resistance to currently available therapies, including the gamma-secretase inhibitor nirogacestat, which received FDA approval in 2023 for adult patients with progressing desmoid tumors. If treatment selects for epigenomically or transcriptomically distinct subclones, earlier molecular profiling of recurrent disease (rather than re-profiling from the archive of the primary tumor) may be necessary to guide treatment choices.
Spatial and single-cell approaches: Future studies deploying spatial transcriptomics or single-cell RNA-seq on desmoid tumor sections would map the heterogeneous cell populations identified here to their histological context, potentially identifying whether divergent transcriptomic profiles correspond to specific morphological zones or microenvironmental niches within the tumor. The integration of such spatially resolved data with the multiomic framework established in this study represents the natural next step toward understanding how intratumor heterogeneity in desmoid tumors is organized and how it evolves in response to therapy.