Detection of clinically actionable gene fusions by next-generation sequencing-based RNA sequencing of non-small cell lung cancer cytology specimens: A single-center experience with comparison to fluorescence in situ hybridization.

BACKGROUND: Genomic profiling is needed to identify actionable alterations in non-small cell lung cancer (NSCLC). Panel-based testing such as next-generation sequencing (NGS) is often preferred to interrogate multiple alterations simultaneously. In this study, we evaluate the utility of an RNA-based NGS assay to detect genomic alterations in NSCLC cytology specimens and compare these results to fluorescence in situ hybridization (FISH) testing. METHODS: A retrospective review was performed of 264 NSCLC cytology specimens that were concurrently tested for gene fusions by RNA-based NGS and ALK, RET, and/or ROS1 by FISH. RESULTS: Genomic alterations were detected in 29 cases by NGS, including ALK, RET, ROS1, NTRK, NUTM1, and FGFR3 fusions and MET exon 14 skipping alterations. Of the 20 cases with ALK, RET, and ROS1 fusions detected by NGS, 16 (80%) were concordant with the corresponding FISH results. Three cases showed discordance, where EML4::ALK (n = 2) and SLC34A2::ROS1 (n = 1) fusions were not detected by the corresponding FISH assay; one case with EZR::ROS1 was inadequate for FISH. No gene fusions were detected in 181 cases by NGS and 54 cases failed testing. The concordance rates for detecting ALK, RET, and ROS1 fusions using NGS and FISH were 97%, 100%, and 99.5%, respectively. CONCLUSION: RNA-based NGS can be used to detect gene fusions in NSCLC cytology cases with high concordance with FISH results. However, RNA-based NGS may have high failure rates and therefore a low threshold for reflexing inadequate cases to an orthogonal testing method is essential for comprehensive genomic profiling.

STAT5B mutations in myeloid neoplasms differ by disease subtypes but characterize a subset of chronic myeloid neoplasms with eosinophilia and/or basophilia.

STAT5B has been reported as a recurrent mutation in myeloid neoplasms (MNs) with eosinophilia, but the overall frequency and importance across a spectrum of MNs are largely unknown. We conducted a multicenter study on a series of 82 MNs with STAT5B mutations detected by next-generation sequencing. The estimated frequency of STAT5B mutation in MNs was low.

PTEN in triple-negative breast carcinoma: protein expression and genomic alteration in pretreatment and posttreatment specimens.

Background: Recent advances have been made in targeting the phosphoinositide 3-kinase pathway in breast cancer. Phosphatase and tensin homolog (PTEN) is a key component of that pathway. Objective: To understand the changes in PTEN expression over the course of the disease in patients with triple-negative breast cancer (TNBC) and whether PTEN copy number variation (CNV) by next-generation sequencing (NGS) can serve as an alternative to immunohistochemistry (IHC) to identify PTEN loss. Methods: We compared PTEN expression by IHC between pretreatment tumors and residual tumors in the breast and lymph nodes after neoadjuvant chemotherapy in 96 patients enrolled in a TNBC clinical trial. A correlative analysis between PTEN protein expression and PTEN CNV by NGS was also performed. Results: With a stringent cutoff for PTEN IHC scoring, PTEN expression was discordant between pretreatment and posttreatment primary tumors in 5% of patients (n = 96) and between posttreatment primary tumors and lymph node metastases in 9% (n = 33). A less stringent cutoff yielded similar discordance rates. Intratumoral heterogeneity for PTEN loss was observed in 7% of the patients. Among pretreatment tumors, PTEN copy numbers by whole exome sequencing (n = 72) were significantly higher in the PTEN-positive tumors by IHC compared with the IHC PTEN-loss tumors (p < 0.0001). However, PTEN-positive and PTEN-loss tumors by IHC overlapped in copy numbers: 14 of 60 PTEN-positive samples showed decreased copy numbers in the range of those of the PTEN-loss tumors. Conclusion: Testing various specimens by IHC may generate different PTEN results in a small proportion of patients with TNBC; therefore, the decision of testing one versus multiple specimens in a clinical trial should be defined in the patient inclusion criteria. Although a distinct cutoff by which CNV differentiated PTEN-positive tumors from those with PTEN loss was not identified, higher copy number of PTEN may confer positive PTEN, whereas lower copy number of PTEN would necessitate additional testing by IHC to assess PTEN loss. Trial registration: NCT02276443.

Efficacy and clinicogenomic correlates of response to immune checkpoint inhibitors alone or with chemotherapy in non-small cell lung cancer.

The role of combination chemotherapy with immune checkpoint inhibitors (ICI) (ICI-chemo) over ICI monotherapy (ICI-mono) in non-small cell lung cancer (NSCLC) remains underexplored. In this retrospective study of 1133 NSCLC patients, treatment with ICI-mono vs ICI-chemo associate with higher rates of early progression, but similar long-term progression-free and overall survival. Sequential vs concurrent ICI and chemotherapy have similar long-term survival, suggesting no synergism from combination therapy. Integrative modeling identified PD-L1, disease burden (Stage IVb; liver metastases), and STK11 and JAK2 alterations as features associate with a higher likelihood of early progression on ICI-mono. CDKN2A alterations associate with worse long-term outcomes in ICI-chemo patients. These results are validated in independent external (n = 89) and internal (n = 393) cohorts. This real-world study suggests that ICI-chemo may protect against early progression but does not influence overall survival, and nominates features that identify those patients at risk for early progression who may maximally benefit from ICI-chemo.

Intragenic EGFR::EGFR.E1E8 Fusion (EGFRvIII) in 4331 Solid Tumors.

Epidermal growth factor receptor variant III (EGFRvIII, the deletion of exons 2-7) is a recurrent intragenic EGFR::EGFR.E1E8 fusion that occurs in high-grade gliomas. The presence of EGFRvIII in other solid tumors has not been well characterized. We retrospectively reviewed advanced malignant solid tumor cases tested by a custom hybrid capture 610-gene next-generation sequencing platform from 2021 to 2022. EGFRvIII was identified in 17 of 4331 (0.4%) cases, including 16 of 238 (7%) brain tumors and 1/301 (0.3%) breast tumors. EGFRvIII-positive brain tumors were all glioblastoma IDH-wildtype, most with concurrent TERT promoter mutation (14 of 16), EGFR amplification (13 of 16), and EGFR mutation (8 of 16). The only EGFRvIII-positive breast lesion was a sarcomatoid neoplasm in a young female patient. A separate breast case tested outside our institution with reported EGFRvIII was noted in a young female patient with a malignant phyllodes tumor with stromal overgrowth. Microscopically, both EGFRvIII-positive breast tumors showed high-grade sarcomatoid morphology with brisk mitotic activity. In summary, EGFRvIII is rare, occurring primarily in glioblastoma and rarely in breast sarcomatoid neoplasm, with no instances identified in other tumor types in our series. This select group of patients may benefit from chemotherapy and/or targeted anti-EGFR therapy.

Clinicopathologic spectrum of myeloid neoplasms with concurrent myeloproliferative neoplasm driver mutations and SRSF2 mutations.

Myeloproliferative neoplasms (MPNs) are frequently associated with classic driver mutations involving JAK2, MPL or CALR. SRSF2 is among the most frequently mutated splicing genes in myeloid neoplasms and SRSF2 mutations are known to confer a poor prognosis in patients with MPNs. In this study, we sought to evaluate the clinicopathologic spectrum of myeloid neoplasms harboring concurrent MPN-driver mutations and SRSF2 mutations. The study cohort included 27 patients, 22 (82%) men and five (19%) women, with a median age of 71 years (range, 51-84). These patients presented commonly with organomegaly (n = 15; 56%), monocytosis (n = 13; 48%), morphologic dysplasia (n = 11; 41%), megakaryocytic hyperplasia and/or clustering (n = 10; 37%) and bone marrow fibrosis >MF-1 (17/22; 77%). About one third of patients either initially presented with acute myeloid leukemia (AML) or eventually progressed to AML. Eighteen (68%) patients had a dominant clone with SRSF2 mutation and nine (33%) patients had a dominant clone with a classic MPN-associated driver mutation. Our data suggest that the presence of an SRSF2 mutation preceding the acquisition of a MPN driver mutations is not a disease-defining alteration nor is it restricted to any specific disease entity within the spectrum of myeloid neoplasms. In summary, patients with myeloid neoplasms associated with concurrent SRSF2 and classic MPN driver mutations have clinical and morphologic features close to that of classic MPNs often with frequent dysplasia and monocytosis.

Immunohistochemical loss of enhancer of Zeste Homolog 2 (EZH2) protein expression correlates with EZH2 alterations and portends a worse outcome in myelodysplastic syndromes.

EZH2 coding mutation (EZH2MUT), resulting in loss-of-function, is an independent predictor of overall survival in MDS. EZH2 function can be altered by other mechanisms including copy number changes, and mutations in other genes and non-coding regions of EZH2. Assessment of EZH2 protein can identify alterations of EZH2 function missed by mutation assessment alone. Precise evaluation of EZH2 function and gene-protein correlation in clinical MDS cohorts is important in the context of upcoming targeted therapies aimed to restore EZH2 function. In this study, we evaluated the clinicopathologic characteristics of newly diagnosed MDS patients with EZH2MUT and correlated the findings with protein expression using immunohistochemistry. There were 40 (~6%) EZH2MUT MDS [33 men, seven women; median age 74 years (range, 55-90)]. EZH2 mutations spanned the entire coding region. Majority had dominant EZH2 clone [median VAF, 30% (1-92)], frequently co-occurring with co-dominant TET2 (38%) and sub-clonal ASXL1 (55%) and RUNX1 (43%) mutations. EZH2MUT MDS showed frequent loss-of-expression compared to EZH2WT (69% vs. 27%, p = 0.001). Interestingly, NINE (23%) EZH2WT MDS also showed loss-of-expression. EZH2MUT and loss-of-expression significantly associated with male predominance and chr(7) loss. Further, only EZH2 loss-of-expression patients showed significantly lower platelet counts, a trend for higher BM blast% and R-IPSS scores. Over a 14-month median follow-up, both EZH2MUT (p = 0.027) and loss-of-expression (p = 0.0063) correlated with poor survival, independent of R-IPSS, age and gender. When analyzed together, loss-of-expression showed a stronger correlation than mutation (p = 0.061 vs. p = 0.43). In conclusion, immunohistochemical assessment of EZH2 protein, alongside mutation, is important for prognostic workup of MDS.

Distinct Gene Mutations Are Associated With Clinicopathologic Features in Urachal Carcinoma.

OBJECTIVES: To investigate the gene mutational profile of urachal carcinoma in correlation with its clinicopathologic features. METHODS: We analyzed genetic mutations in 30 cases of urachal carcinoma by next-generation sequencing (NGS) test. Histologic slides and clinical data were reviewed. RESULTS: The patients included 21 men and 9 women, with a mean age of 53 years (range, 24-75 years). The urachal carcinomas included mucinous (11), enteric (10), signet ring cell (8), and high-grade neuroendocrine (1) subtypes. Targeted NGS analysis demonstrated genetic mutations in all the urachal tumors (mean, 2; range, 1-4). TP53 was the most mutated gene (25), followed by KRAS (9) and GNAS (8) genes. TP53 mutations were more common in the signet ring cell subtype (7/8), and GNAS mutations were present only in the mucinous (5/11) and signet ring cell subtypes (3/8) but not in the enteric subtype (0/10). KRAS mutations were significantly associated with cancer stage IV (P = .02) and younger patient age (P = .046). Furthermore, the presence of KRAS mutations in urachal carcinoma portended a poorer overall survival (P = .006). CONCLUSIONS: Urachal carcinoma demonstrates frequent gene mutations that are associated with distinct clinicopathologic features. Gene mutation may underlie the development and progression of this aggressive disease.

Landscape of NOTCH1 mutations and co-occurring biomarker alterations in chronic lymphocytic leukemia.

NOTCH1 is one of the most frequently mutated genes in chronic lymphocytic leukemia and has emerged as a marker of poor prognosis. In addition to coding NOTCH1 mutations involving exon 34, non-coding NOTCH1 mutations involving the 3' UTR have been described in a limited number of chronic lymphocytic leukemia (CLL) patients and were associated with adverse outcomes. In this study, 1574 CLL patients were assessed using targeted sequencing with a 29 gene panel and the results were correlated with prognostic characteristics. NOTCH1 mutations were detected in 252 (16%) patients, including both coding (220/252, 14%), non-coding (24/252, 1.5%) and a mixture of coding and non-coding (8/252, 0.5%) NOTCH1 mutations. NOTCH1 mutations were more commonly seen in patients with unmutated IGHV, ZAP70 positivity and CD38 positivity. Mixed NOTCH1 mutations were also more commonly seen in patients with unmutated IGHV and ZAP70. There was no association between mixed NOTCH1 mutations and CD38 expression in this cohort. The most common cytogenetic alteration detected in patients with coding and mixed NOTCH1 mutations was trisomy 12, whereas del13q was the most common cytogenetic alteration detected in patients with non-coding NOTCH1 mutation. The most common gene mutations co-occurring with coding NOTCH1 mutations were: TP53 (23.2%), SF3B1 (16.4%) and SPEN (10%). The most common gene mutations co-occurring with non-coding NOTCH1 mutations were: SF3B1 11(34.4%), ATM 4(12.5%) and TP53 4(12.5%). CLL patients with clonal coding and non-coding NOTCH1 mutations had a significantly shorter time-to-first treatment than patients with wild type NOTCH1 (4.3 vs 10.0 years and 0.9 vs 10.0 years respectively, p < 0.05). Similarly, CLL patients with subclonal coding NOTCH1 mutations had a significantly shorter time-to-first treatment than patients with wild type NOTCH1 (5.6 vs 10.0 years, p < 0.05). CLL patients with subclonal non-coding NOTCH1 mutations also had a shorter time-to-first treatment than patients with wild type NOTCH1 mutations, however, the difference was not significant (5.1 vs 10.0 years, p = 0.15). These data confirm that both coding and non-coding NOTCH1 mutations carry adverse prognostic impact and need to be included in sequencing assays performed for the prognostic workup of CLL patients.

Acquired WT1 mutations contribute to relapse of NPM1-mutated acute myeloid leukemia following allogeneic hematopoietic stem cell transplant.

The role of WT1 protein in hematopoiesis and leukemogenesisis incompletely elucidated. WT1 overexpression is common in acute myeloid leukemia (AML); however, WT1 mutations occur in only about 10% of cases, with increasing incidence in the setting of relapse. In this study, we investigated the clinical and molecular characteristics of WT1 mutations in NPM1-mutated AML, to enhance our understanding of the biology and potential therapeutic implications of WT1 mutations. Our study cohort included 67 patients with NPM1 mutated AML and a median follow-up of 13.7 months. WT1 mutations were identified in 7% (n = 5) of patients at the time of initial diagnosis. WT1 mutant clones were presumed to be present as co-dominant clones in 3/5 and in subclonal populations in 2/5 cases based on variant allelic frequency (VAF) when compared with NPM1 mutation VAF. All WT1 mutations became undetectable at time of MRD-negative (NPM1-wild type) remission. None of these patients experienced relapse at the time of last follow-up (median, 15 months; range, 4.5-20.2 months). A total of 15/67 (22%) patients relapsed; among these patient, four (27%) relapsed with WT1 mutant AML. Three of four patients had undergone allogeneic hematopoietic stem cell transplantation (HSCT). None of these patients had detectable WT1 mutations at the time of initial diagnosis. WT1 mutations were presumed clonal in two cases and subclonal in the other two cases, based on VAF. Our results indicate that WT1 mutations contribute to relapse in NPM1 mutated AML, especially in the setting of HSCT. These findings suggest that emerging WT1 mutations may serve as a conduit for relapse in NPM1-mutated AML, and that sequential molecular profiling to evaluate potential emergent WT1 mutations during surveillance and particularly at relapse likely has prognostic value in patients with NPM1 mutated AML.

Concurrent Mutations in SF3B1 and PHF6 in Myeloid Neoplasms.

It has been reported that gene mutations in SF3B1 and PHF6 are mutually exclusive. However, this observation has never been rigorously assessed. We report the clinicopathologic and molecular genetic features of 21 cases of myeloid neoplasms with double mutations in SF3B1 and PHF6, including 9 (43%) with myelodysplastic syndrome, 5 (24%) with acute myeloid leukemia, 4 (19%) with myeloproliferative neoplasms, and 3 (14%) with myelodysplastic/myeloproliferative neoplasms. Multilineage dysplasia with ring sideroblasts, increased blasts, and myelofibrosis are common morphologic findings. All cases but one had diploid or non-complex karyotypes. SF3B1 mutations were detected in the first analysis of all the patients. PHF6 mutations occurred either concurrently with SF3B1 mutations or in subsequent follow-up samples and are associated with disease progression and impending death in most cases. Most cases had co-mutations, the most common being ASXL1, RUNX1, TET2, and NRAS. With a median follow-up of 39 months (range, 3-155), 17 (81%) patients died, 3 were in complete remission, and 1 had persistent myelodysplastic syndrome. The median overall survival was 51 months. In summary, concurrent mutations in SF3B1 and PHF6 are rare, but they do exist in a variety of myeloid neoplasms, with roles as early initiating events and in disease progression, respectively.

Myelodysplastic/myeloproliferative neoplasms-unclassifiable with isolated isochromosome 17q represents a distinct clinico-biologic subset: a multi-institutional collaborative study from the Bone Marrow Pathology Group.

Classification of myeloid neoplasms with isolated isochromosome i(17q) [17p deletion with inherent monoallelic TP53 loss plus 17q duplication] is controversial. Most cases fall within the WHO unclassifiable myelodysplastic/myeloproliferative neoplasms (MDS/MPN-U) category. The uniformly dismal outcomes warrant better understanding of this entity. We undertook a multi-institutional retrospective study of 92 adult MDS/MPN-U cases from eight institutions. Twenty-nine (32%) patients had isolated i(17q) [MDS/MPN-i(17q)]. Compared to MDS/MPN without i(17q), MDS/MPN-i(17q) patients were significantly younger, had lower platelet and absolute neutrophil counts, and higher frequency of splenomegaly and circulating blasts. MDS/MPN-i(17q) cases showed frequent bilobed neutrophils (75% vs. 23%; P = 0.03), hypolobated megakaryocytes (62% vs. 20%; P = 0.06), and a higher frequency of SETBP1 (69% vs. 5%; P = 0.002) and SRSF2 (63% vs. 5%; P = 0.006) mutations that were frequently co-existent (44% vs. 0%; P = 0.01). TP53 mutations were rare. The mutation profile of MDS/MPN-U-i(17q) was similar to other myeloid neoplasms with i(17q) including atypical chronic myeloid leukemia, chronic myelomonocytic leukemia, myelodysplastic/myeloproliferative neoplasm with ring sideroblasts and thrombocytosis, myelodysplastic syndrome and acute myeloid leukemia, with frequent concomitant SETBP1/SRSF2 mutations observed across all the diagnostic entities. Over a median follow-up of 52 months, patients with MDS/MPN-i(17q) showed a shorter median overall survival (11 vs. 28 months; P < 0.001). The presence of i(17q) retained independent poor prognostic value in multivariable Cox-regression analysis [HR 3.686 (1.17-11.6); P = 0.026] along with splenomegaly. We suggest that MDS/MPN-i(17q) warrants recognition as a distinct subtype within the MDS/MPN-U category based on its unique clinico-biologic features and uniformly poor prognosis.

Factors Impacting Clinically Relevant RNA Fusion Assays Using Next-Generation Sequencing.

CONTEXT.­: RNA-based next-generation sequencing (NGS) assays are being used with increasing frequency for comprehensive molecular profiling of solid tumors. OBJECTIVE.­: To evaluate factors that might impact clinical assay performance. DESIGN.­: A 4-month retrospective review of cases analyzed by a targeted RNA-based NGS assay to detect fusions was performed. RNA extraction was performed from formalin-fixed, paraffin-embedded tissue sections and/or cytology smears of 767 cases, including 493 in-house and 274 outside referral cases. The types of samples included 422 core needle biopsy specimens (55%), 268 resection specimens (35%), and 77 cytology samples (10%). RESULTS.­: Successful NGS fusion testing was achieved in 697 specimens (90.9%) and correlated positively with RNA yield (P < .001) and negatively with specimen necrosis (P = .002), decalcification (P < .001), and paraffin block age of more than 2 years (P = .001). Of the 697 cases that were successfully sequenced, 50 (7.2%) had clinically relevant fusions. The testing success rates and fusion detection rates were similar between core needle biopsy and cytology samples. In contrast, RNA fusion testing was often less successful using resection specimens (P = .007). Testing success was independent of the tumor percentage in the specimen, given that at least 20% tumor cellularity was present. CONCLUSIONS.­: The success of RNA-based NGS testing is multifactorial and is influenced by RNA quality and quantity. Identification of preanalytical factors affecting RNA quality and yield can improve NGS testing success rates.

An Overview of Characteristics of Clinical Next-Generation Sequencing-Based Testing for Hematologic Malignancies.

CONTEXT.­: With the increasing integration of molecular alterations into the evaluation of hematologic malignancies (HM), somatic mutation profiling by next-generation sequencing (NGS) has become a common clinical testing strategy. Limited data are available about the characteristics of these assays. OBJECTIVE.­: To describe assay characteristics, specimen requirements, and reporting practices for NGS-based HM testing using College of American Pathologists proficiency testing survey data. DESIGN.­: The College of American Pathologists NGS Hematologic Malignancies Survey (NGSHM) results from 78 laboratories were used to determine laboratory practices in NGS-based HM testing. RESULTS.­: The majority of laboratories performed tumor-only (88.5% [69 of 78]), targeted sequencing of cancer genes or mutation hotspots (98.7% [77 of 78]); greater than 90% performed testing on fresh bone marrow and peripheral blood. The majority of laboratories reported a 5% lower limit of detection for single-nucleotide variants (73.1% [57 of 78]) and small insertions and deletions (50.6% [39 of 77]). A majority of laboratories used benchtop sequencers and custom enrichment approaches. CONCLUSIONS.­: This manuscript summarizes the characteristics of clinical NGS-based testing for the detection of somatic variants in HM. These data may be broadly useful to inform laboratory practice and quality management systems, regulation, and oversight of NGS testing, and precision medicine efforts using a data-driven approach.

Utilization of cytology smears improves success rates of RNA-based next-generation sequencing gene fusion assays for clinically relevant predictive biomarkers.

BACKGROUND: The use of RNA-based next-generation sequencing (NGS) assays to detect gene fusions for targeted therapy has rapidly become an essential component of comprehensive molecular profiling. For cytology specimens, the cell block (CB) is most commonly used for fusion testing; however, insufficient cellularity and/or suboptimal RNA quality are often limiting factors. In the current study, the authors evaluated the factors affecting RNA fusion testing in cytology and the added value of smears in cases with a suboptimal or inadequate CB. METHODS: A 12-month retrospective review was performed to identify cytology cases that were evaluated by a targeted RNA-based NGS assay. Samples were sequenced by targeted amplicon-based NGS for 51 clinically relevant genes on a proprietary platform. Preanalytic factors and NGS quality parameters were correlated with the results of RNA fusion testing. RESULTS: The overall success rate of RNA fusion testing was 92%. Of the 146 cases successfully sequenced, 14% had a clinically relevant fusion detected. NGS testing success positively correlated with RNA yield (P = .03) but was independent of the tumor fraction, the tumor size, or the number of slides used for extraction. CB preparations were adequate for testing in 45% cases, but the inclusion of direct smears increased the adequacy rate to 92%. There was no significant difference in testing success rates between smears and CB preparations. CONCLUSIONS: The success of RNA-based NGS fusion testing depends on the quality and quantity of RNA extracted. The use of direct smears significantly improves the adequacy of cytologic samples for RNA fusion testing for predictive biomarkers.

Development and Validation of a Gene Signature Classifier for Consensus Molecular Subtyping of Colorectal Carcinoma in a CLIA-Certified Setting.

PURPOSE: Consensus molecular subtyping (CMS) of colorectal cancer has potential to reshape the colorectal cancer landscape. We developed and validated an assay that is applicable on formalin-fixed, paraffin-embedded (FFPE) samples of colorectal cancer and implemented the assay in a Clinical Laboratory Improvement Amendments (CLIA)-certified laboratory. EXPERIMENTAL DESIGN: We performed an in silico experiment to build an optimal CMS classifier using a training set of 1,329 samples from 12 studies and validation set of 1,329 samples from 14 studies. We constructed an assay on the basis of NanoString CodeSets for the top 472 genes, and performed analyses on paired flash-frozen (FF)/FFPE samples from 175 colorectal cancers to adapt the classifier to FFPE samples using a subset of genes found to be concordant between FF and FFPE, tested the classifier's reproducibility and repeatability, and validated in a CLIA-certified laboratory. We assessed prognostic significance of CMS in 345 patients pooled across three clinical trials. RESULTS: The best classifier was weighted support vector machine with high accuracy across platforms and gene lists (>0.95), and the 472-gene model outperforming existing classifiers. We constructed subsets of 99 and 200 genes with high FF/FFPE concordance, and adapted FFPE-based classifier that had strong classification accuracy (>80%) relative to "gold standard" CMS. The classifier was reproducible to sample type and RNA quality, and demonstrated poor prognosis for CMS1-3 and good prognosis for CMS2 in metastatic colorectal cancer (P < 0.001). CONCLUSIONS: We developed and validated a colorectal cancer CMS assay that is ready for use in clinical trials, to assess prognosis in standard-of-care settings and explore as predictor of therapy response.