Search

Identification of circulating tumor DNA (ctDNA) following curative intent therapies is a surrogate for microscopic residual disease for patients with metastatic colorectal cancer (mCRC). Preclinically, in micrometastatic microsatellite stable (MSS) colorectal cancer, increased TGFβ signaling results in exclusion of antitumor cytotoxic T cells from the tumor microenvironment. Bintrafusp alfa (BA) is a bifunctional fusion protein composed of the extracellular domain of the TGFβRII receptor (“TGFβ trap”) and anti-PD-L1 antibody. Patients with liver-limited, MSS mCRC and with detected ctDNA after complete resection of all known tumors and standard-of-care therapy were treated with 1,200 mg of BA intravenously every 14 days for six doses. The primary endpoint was ctDNA clearance. Radiographic characteristics at recurrence were compared using independent t tests to historical data from a similar cohort of patients with liver-limited mCRC who underwent observation. Only 4 of 15 planned patients received BA before the study was stopped early for loss of equipoise. There was no grade ≥3 adverse event. None of the patients cleared ctDNA. All patients developed radiographic recurrence by the first planned restaging. Although not detectable at prior to treatment, TGFβ3 was found in circulation in all patients at cycle 2 day 1. Compared with a historical cohort, patients administered BA developed more metastases (15 vs. 2, P = 0.005) and greater tumor volumes (9 cm vs. 2 cm, P = 0.05). Treatment with BA in patients with ctDNA-detected, liver-limited mCRC did not clear ctDNA and was associated with large-volume recurrence, highlighting the potential context-specific complexity of dual TGFβ and PD-L1 inhibition. Significance: Use of ctDNA to identify patients with micrometastatic disease for therapeutic intervention is feasible. Treatment with BA in patients with liver-limited mCRC and with detectable ctDNA after resection generated rapid progression. Approaches targeting TGFβ signaling must consider its pathway complexity in future immunotherapy combination strategies.

KRAS and NRAS mutations occur in 45% of colorectal cancers, with combined MAPK pathway and CDK4/6 inhibition identified as a potential therapeutic strategy. In the current study, this combinatorial treatment approach was evaluated in a co-clinical trial in patient-derived xenografts (PDX), and safety was established in a clinical trial of binimetinib and palbociclib in patients with metastatic colorectal cancer with RAS mutations. Across 18 PDX models undergoing dual inhibition of MEK and CDK4/6, 60% of tumors regressed, meeting the co-clinical trial primary endpoint. Prolonged duration of response occurred predominantly in TP53 wild-type models. Clinical evaluation of binimetinib and palbociclib in a safety lead-in confirmed safety and provided preliminary evidence of activity. Prolonged treatment in PDX models resulted in feedback activation of receptor tyrosine kinases and acquired resistance, which was reversed with a SHP2 inhibitor. These results highlight the clinical potential of this combination in colorectal cancer, along with the utility of PDX-based co-clinical trial platforms for drug development. Significance: This co-clinical trial of combined MEK-CDK4/6 inhibition in RAS mutant colorectal cancer demonstrates therapeutic efficacy in patient-derived xenografts and safety in patients, identifies biomarkers of response, and uncovers targetable mechanisms of resistance.

Supplemental Figure S2

Supplemental Table S3

Identification of circulating tumor DNA (ctDNA) following curative intent therapies is a surrogate for microscopic residual disease for patients with metastatic colorectal cancer (mCRC). Preclinically, in micrometastatic microsatellite stable (MSS) colorectal cancer, increased TGFβ signaling results in exclusion of antitumor cytotoxic T cells from the tumor microenvironment. Bintrafusp alfa (BA) is a bifunctional fusion protein composed of the extracellular domain of the TGFβRII receptor (“TGFβ trap”) and anti-PD-L1 antibody. Patients with liver-limited, MSS mCRC and with detected ctDNA after complete resection of all known tumors and standard-of-care therapy were treated with 1,200 mg of BA intravenously every 14 days for six doses. The primary endpoint was ctDNA clearance. Radiographic characteristics at recurrence were compared using independent t tests to historical data from a similar cohort of patients with liver-limited mCRC who underwent observation. Only 4 of 15 planned patients received BA before the study was stopped early for loss of equipoise. There was no grade ≥3 adverse event. None of the patients cleared ctDNA. All patients developed radiographic recurrence by the first planned restaging. Although not detectable at prior to treatment, TGFβ3 was found in circulation in all patients at cycle 2 day 1. Compared with a historical cohort, patients administered BA developed more metastases (15 vs. 2, P = 0.005) and greater tumor volumes (9 cm vs. 2 cm, P = 0.05). Treatment with BA in patients with ctDNA-detected, liver-limited mCRC did not clear ctDNA and was associated with large-volume recurrence, highlighting the potential context-specific complexity of dual TGFβ and PD-L1 inhibition.Significance:Use of ctDNA to identify patients with micrometastatic disease for therapeutic intervention is feasible. Treatment with BA in patients with liver-limited mCRC and with detectable ctDNA after resection generated rapid progression. Approaches targeting TGFβ signaling must consider its pathway complexity in future immunotherapy combination strategies.

Supplemental Table S1

Supplemental Table S2

Supplemental Figure S1

Supplemental Table S4

68-gene panel for which mutation, amplification, deletion, and fusion profiling was available.

Trends in various mutation allele frequencies tracked serially over treatment course for two patients on study.

Demographic information for patients on study.

Trends in cfDNA fraction (black) and radiographic changes (red) for ten other patients on study.

Analysis of BRAFV600E allele fraction by cfDNA was validated using two separate methodologies.

Heatmap showing fRMA level gene data from Affymetrix subsets in training and validation datasets.

Supplementary methods and results

Validation of CMS assay at the research molecular diagnostic laboratory. Each sample classified in to CMS subtype 1-4 or mixed (M) subtype.

Summary of sample-wise FF/FFPE Spearman Correlations. Summary of sample-wise paired FF/FFPE correlations across 472 CMS genes for batch 1 containing CRCSC samples (top), batch 2 containing non-CRCSC samples (middle), and all samples (bottom).

Supplementary Figure from Targeting RAS Mutant Colorectal Cancer with Dual Inhibition of MEK and CDK4/6

3A. Classification accuracy of CMS1 samples on training data set V1 using 4-fold cross validation as function of number of genes.3B.able-3B: Classification accuracy of CMS1 samples on validation data set (V2) as function of number of genes.3C. Classification accuracy CMS1 samples with out of sample (V2o) prediction set as function of number of genes.3D. Classification accuracy CMS1 samples with TCGA dataset (RNA-seq subset) as function of number of genes. 3E. Classification accuracy CMS1 samples with Affymetrix set (V2a) as function of number of genes.

Kaplan-Meier estimation of (A) overall survival of BRAFV600E mCRC NGS cohorts and (B) progression free survival of BRAFV600E mCRC NGS cohorts from BC Cancer and MD Anderson shows no difference in overall or progression free survival.

Comparison of overall survival based on primary tumor side using a multivariate proportional hazards model controlling for co-variates that differed based on tumor location.

Sensitivity analysis comparing mutation prevalence estimates by site of biopsy of tissue used for next generation sequencing panel in metastatic colorectal cancer.

Classification accuracy on validation data set, the entire set (V2), out-of-sample prediction (V2o), Affymetrix subset (V2a), and TCGA RNAseq subset (TCGA) for various methods as function of number of genes.

Supplemental legend

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.

Nanostring CMS Classifier Performance: Performance of Nanostring 100 gene CMS classifier applied to FFPE samples (6A and 6C) and FF samples (6B and 63D). The top two figures plot 4-class accuracy vs. classification confidence, with the dots marking individual samples either correctly (1.0) or incorrectly (0.0), with color indicat-ing correct CMS. The line contains a generalized additive model (GAM) fit to these data with 95% pointwise confidence bands, and demonstrates that samples classified with greater confidence were more likely to be cor-rectly classified. Figure 6C and 6D plot 4-class accuracy vs. RNA quality, defined as 0nt (FFPE) or RIN (FF). Note that there is little if any association of CMS accuracy with RNA quality, suggesting that the perfor-mance of classifier is robust to RNA quality in this study.

Prevalence of BRAFV600E mutated colorectal cancer by IHC testing within VCH population-based cohort.

Classification accuracy of different modelling strategies on training data set (V1) using 4-fold cross validation.

Kaplan-Meier estimation of overall survival of mCRC patients with BRAFV600E mutations detected by (A) IHC (BRAFIHC) and (B) NGS (BRAFNGS) demonstrates no survival difference between those with dMMR versus pMMR for BRAFIHC and BRAFNGS within individual groups.

Comparison of Classifiers. Comparison of random forest (RF) and single sample predictors (SSP) in Guinney et al. (2015) paper with our wSVM classifier with 472 genes, either forcing a single CMS call for each sample (all calls) or only when classified with high confidence

KRAS and NRAS mutations occur in 45% of colorectal cancers, with combined MAPK pathway and CDK4/6 inhibition identified as a potential therapeutic strategy. In the current study, this combinatorial treatment approach was evaluated in a co-clinical trial in patient-derived xenografts (PDX), and safety was established in a clinical trial of binimetinib and palbociclib in patients with metastatic colorectal cancer with RAS mutations. Across 18 PDX models undergoing dual inhibition of MEK and CDK4/6, 60% of tumors regressed, meeting the co-clinical trial primary endpoint. Prolonged duration of response occurred predominantly in TP53 wild-type models. Clinical evaluation of binimetinib and palbociclib in a safety lead-in confirmed safety and provided preliminary evidence of activity. Prolonged treatment in PDX models resulted in feedback activation of receptor tyrosine kinases and acquired resistance, which was reversed with a SHP2 inhibitor. These results highlight the clinical potential of this combination in colorectal cancer, along with the utility of PDX-based co-clinical trial platforms for drug development.Significance:This co-clinical trial of combined MEK-CDK4/6 inhibition in RAS mutant colorectal cancer demonstrates therapeutic efficacy in patient-derived xenografts and safety in patients, identifies biomarkers of response, and uncovers targetable mechanisms of resistance.

Supplementary Table from Targeting RAS Mutant Colorectal Cancer with Dual Inhibition of MEK and CDK4/6

Clinical characteristics of patients with BRAFV600E mCRC receiving first-line palliative chemotherapy.

5A. Classification accuracy of CMS 3 samples on training data set V1 using 4-fold cross validation as function of number of genes. 5B.Classification accuracy of CMS 3 samples on validation data set (V2) as function of number of genes. 5C. Classification accuracy CMS3 samples with out of sample prediction set (V2o) as function of number of genes. 5D. Classification accuracy CMS 3 samples with TCGA dataset (RAN-seq subset) as function of number of genes. 5E. Classification accuracy CMS3 samples with Affymetrix dataset (V2a) as function of number of genes.

List of genes included in the colorectal cancer CMS classifier

Multivariate analysis of interaction between BRAF and MMR status in patients with metastatic colorectal cancer.

6A. Classification accuracy of CMS 4 samples on training data set V1 using 4-fold cross validation as function of number of genes. 6B. Classification accuracy of CMS 4 samples on validation data set (V2) as function of number of genes. 6C. Classification accuracy CMS 4 samples with out of sample prediction set (V2o) as function of number of genes. 6D. Classification accuracy CMS 3 samples with TCGA dataset (RAN-seq subset) as function of number of genes. 6E. Classification accuracy of CMS 3 samples with Affymetrix data set (V2a) as function of number of genes.

Details of the data sets, including tissue type, total number of samples, total number of consensus samples are shown. FF=Flash frozen, FFPE:Formalin fixed paraffin embedded.

Sample-wise and Gene-wise correlation of FF/FFPE paired Samples: Scatterplot of gene-specific Spearman correlation of FF/FFPE computed separately for CRCSC samples and non-CRCSC samples , with red and blue marking the overall top 100 and top 200 genes, respectively, in terms of FF/FFPE correlation computed across all samples. Note the high level of agreement across batches that genes with high FF/FFPE correlations for one batch tended to also have high FF/FFPE correlation in the other batch, suggesting high FF/FFPE correla-tion is a characteristic of the gene/probe set.

Summary of the relative prognostic and mutational impact of tumor location in metastatic colorectal cancer. Sub-caption: Relative prevalence of mutations at each site was compared to the median prevalence across all sites and used to calculate relative font size for each gene name. Hazard ratios displayed in the figure utilize rectal location as the reference category.

Sensitivity analysis demonstrating optimal left vs right cut-point to maximize prognostic differences based on side in metastatic colorectal cancer. Sub-caption: Numbers in figure represent the hazard ratio followed by the 95% confidence interval within square brackets. The hazard ratio estimate occurs at the location used to divide left from right in each sensitivity analysis. For example, if the division between right and left occurs between the hepatic flexure and transverse colon, the hazard ratio estimate will be placed on the graph half way between the hepatic flexure and transverse colon (ie. 1.66).

Supplementary Data from Targeting RAS Mutant Colorectal Cancer with Dual Inhibition of MEK and CDK4/6

Purpose: Colorectal cancers are classified as right/left-sided based on whether they occur before/after the splenic flexure, with established differences in molecular subtypes and outcomes. However, it is unclear if this division is optimal and whether precise tumor location provides further information.Experimental Design: In 1,876 patients with colorectal cancer, we compared mutation prevalence and overall survival (OS) according to side and location. Consensus molecular subtype (CMS) was compared in a separate cohort of 608 patients.Results: Mutation prevalence differed by side and location for TP53, KRAS, BRAFV600, PIK3CA, SMAD4, CTNNB1, GNAS, and PTEN. Within left- and right-sided tumors, there remained substantial variations in mutation rates. For example, within right-sided tumors, RAS mutations decreased from 70% for cecal, to 43% for hepatic flexure location (P = 0.0001), while BRAFV600 mutations increased from 10% to 22% between the same locations (P < 0.0001). Within left-sided tumors, the sigmoid and rectal region had more TP53 mutations (P = 0.027), less PIK3CA (P = 0.0009), BRAF (P = 0.0033), or CTNNB1 mutations (P < 0.0001), and less MSI (P < 0.0001) than other left-sided locations. Despite this, a left/right division preceding the transverse colon maximized prognostic differences by side and transverse colon tumors had K-modes mutation clustering that appeared more left than right sided. CMS profiles showed a decline in CMS1 and CMS3 and rise in CMS2 prevalence moving distally.Conclusions: Current right/left classifications may not fully recapitulate regional variations in tumor biology. Specifically, the sigmoid-rectal region appears unique and the transverse colon is distinct from other right-sided locations. Clin Cancer Res; 24(5); 1062–72. ©2017 AACR.See related commentary by Dienstmann, p. 989

4A. Classification accuracy of CMS2 samples on training data set V1 using 4-fold cross validation as function of number of genes.4B. Classification accuracy of CMS2 samples on validation data set (V2) as function of number of genes.4C. Classification accuracy CMS2 samples with out of sample(V2o) prediction set as function of number of genes. 4D. Classification accuracy CMS2 samples with TCGA dataset (RNA-seq subset) as function of number of genes.4E. Classification accuracy CMS2 samples with Affymentrix dataset (V2a) as function of number of genes.

Algorithm showing strategy to de-velop the Nanostring based FFPE CMS classifier using Affymetrix gene expression data. FF: flash frozen, FFPE: Formalin fixed par-affin embedded, NFFPE: Nanos-tring FFPE, AFF: Flash frozen Affymetrix, GES: Gene expres-sion score, CRCSC: Colorectal cancer subtyping consortium

Kaplan-Meier estimation of overall survival of mCRC patients with BRAFV600E mutations detected by NGS (BRAFNGS) and two IHC (BRAFIHC) subsets: patients that received any NGS testing and patients that were never NGS tested.

Classification Confidence of Affymetrix Validation Samples vs. Classification Accuracy of 472-gene wSVM clas-sifier: Distribution of classification confidence (�i) across 713 validation samples (top), and classification accura-cy (1=correct, 0=incorrect) as a function of classification confidence (�i), with fitted loess curve and 95% confidence interval.

Multivariate analysis of Progression Free Survival of patients with BRAFV600E mCRC with therapy forced into model.

Classification accuracy for training data V1 (based on 4-fold cross val-idation, 2A) and the full validation data V2 (2B) for various methods as function of number