Oncogenic drivers and therapeutic vulnerabilities in KRAS wild-type pancreatic cancer

Abstract

Purpose:

Approximately 8-10% of pancreatic ductal adenocarcinomas (PDAC) do not harbor mutations in KRAS. Understanding the unique molecular and clinical features of this subset of pancreatic cancer is important to guide patient stratification for clinical trials of molecularly targeted agents.

Experimental Design:

We analyzed a single-institution cohort of 795 exocrine pancreatic cancer cases (including 785 PDAC cases) with a targeted multi-gene sequencing panel and identified 73 patients (9.2%) with KRAS wild-type (WT) pancreatic cancer.

Results:

Overall, 43.8% (32/73) of KRAS WT cases had evidence of an alternative driver of the mitogen activated kinase (MAPK) pathway, including BRAF mutations and in-frame deletions and receptor-tyrosine kinase (RTK) fusions. Conversely, 56.2% of cases did not harbor a clear MAPK driver alteration, but 29.3% of these MAPK-negative KRAS WT cases (12/41) demonstrated activating alterations in other oncogenic drivers such as GNAS, MYC, PIK3CA and CTNNB1. We demonstrate potent efficacy of pan-RAF and MEK inhibition in patient-derived organoid (PDO) models carrying BRAF in-frame deletions. Moreover, we demonstrate durable clinical benefit of targeted therapy in a patient harboring a KRAS WT tumor with a ROS1 fusion. Clinically, patients with KRAS WT tumors were significantly younger in age of onset (Median Age: 62.6 vs. 65.7 years , p = 0.037). SMAD4 mutations were associated with a particularly poor prognosis in KRAS WT cases.

Conclusions:

This study defines the genomic underpinnings of KRAS WT pancreatic cancer and highlights potential therapeutic avenues for future investigation in molecularly directed clinical trials.

INTRODUCTION

Pancreatic cancer is the third leading cause of cancer-related mortality in the United States and exocrine pancreatic cancers including pancreatic ductal adenocarcinomas (PDAC) and pancreatic acinar cell carcinomas (PACC) comprise the majority of cases¹. Most patients present with advanced disease and quickly succumb to their illness. Although survival of patients with pancreatic cancer has improved with the advent of multi-agent chemotherapy, long-term responses are rare and the 5-year survival rates remain around 10%^2,3.

A number of studies over the last decade have sought to elucidate the genomic characteristics of pancreatic cancer^4–9. Approximately 90% of tumors harbor mutations in the KRAS oncogene. KRAS is a membrane-bound GTPase that transduces pro-survival and growth signals from cell surface receptors to intracellular effector pathways. Point mutations at key residues (G12, G13, Q61) lock KRAS in its GTP-bound activated (“on”) state. Recently, direct inhibitors of the KRAS^G12C protein have been developed that have shown promising activity in several malignancies including in pancreatic cancers^10–13. However, KRAS^G12C is found in only 1-2% of pancreatic cancers and acquired resistance remains a challenge.

In addition to activating mutations in KRAS, pancreatic cancers frequently harbor inactivating alterations in tumor suppressor genes (TSGs) including TP53, CDKN2A, and SMAD4. Together, these four genes define the core genetic backbone of pancreatic cancer and have been associated with prognosis in resected disease^14,15. Recent work has elucidated genetically defined subsets of pancreatic cancer harboring therapeutically tractable targets, such as tumors with homologous recombination deficiency (HRD)¹⁶. In particular, actionable alterations are enriched in the 8-10% of PDAC that do not harbor canonical activating mutations in KRAS^9,17. KRAS wild-type (WT) exocrine pancreatic cancer, hereafter referred to as KRAS WT pancreatic cancer, has been shown to harbor alternative drivers in the receptor tyrosine kinase (RTK)-RAS-MAPK signaling pathway, including activating BRAF missense mutations, in-frame deletions, or fusions^6,8,18–20 and RTK gene fusions involving ALK^8,21, FGFR2^7–8, NRG1^22–24, MET⁸, NTRK1⁸, RAF^18,25 and RET^8,26. These alternative MAPK drivers can be susceptible to oncogene-specific small molecule inhibitors^18,27 or in the case of BRAF alterations, downstream MAPK (e.g., MEK/ERK) inhibition²⁰. PACC, a rare histologic variant of pancreatic cancer, are frequently KRAS WT and can harbor fusions involving the RAF and RET oncogenes^25,26. As such, the KRAS WT subpopulation of pancreatic cancer currently represents a prime target for precision oncology intervention in this highly lethal disease^17,24.

Much of the data about KRAS WT pancreatic cancer and associated targetable genomic lesions has been derived from individual case reports or small case series^9,17. One recent systematic effort to characterize the genetic features of KRAS WT pancreatic cancer demonstrated mutations or fusions in BRAF, FGFR2, ALK, RET and NRG1 in these tumors, but additional investigations across broader patient cohorts and functional study of the oncogene dependence for these alternative drivers are needed⁹. Here, we performed a comprehensive clinical, genomic, and functional characterization of a large institutional cohort of KRAS WT pancreatic cancer, defining the molecular underpinnings of these cancers and highlighting potential therapeutic avenues for future investigation.

METHODS

Identification of KRAS WT and KRAS mutant pancreatic cancer cohorts

We identified 812 unique pancreatic cancer patient samples, collected between 2013 and 2020, for which genetic profiling was performed using OncoPanel^28,29, a somatic-only targeted DNA sequencing assay that detects single nucleotide variants (SNVs), small insertions and deletions (INDELs), copy number variants (CNVs), and structural variants (SVs) in >400 cancer-associated genes²⁸. Three versions of OncoPanel with overlapping gene sets were used over the duration of the study (Suppl. Table S1). We included patients with all histologies of pancreatic exocrine tumors including PDAC, adeno-squamous (PASC), and acinar cell carcinomas (PACC) as these have similar treatment paradigms. Patients with pancreatic neuroendocrine tumors were not included in this analysis. Patients with either localized or metastatic disease were included in this cohort. All patients had consented to an Institutional Review Board (IRB) approved protocol at Dana-Farber Cancer Institute permitting access to their clinical and genomic data. The study was conducted in accordance ethical principles in the Belmont report.

Cases were divided into KRAS WT and KRAS mutant cohorts based on the absence or presence, respectively, of an activating oncogenic KRAS mutation. To ensure that KRAS mutations were not missed in low cellularity pancreatic cancers, we performed extensive quality control (QC) checks. Samples were excluded from the analysis if: 1) estimated tumor purity was ≤5% or 2) if no apparently somatic alterations were called on OncoPanel, indicating that the neoplastic cell content of the specimen was very low. In addition, a mean target coverage >50x was required for case inclusion. A total of 795 cases from our original cohort of 812 met these QC metrics (Suppl. Fig. S1A).

Classification of oncogenic alterations

Mutations called by the OncoPanel analytical pipeline were annotated using the MafAnnotator tool from OncoKB³⁰ to classify variants for predicted or known oncogenicity as well as predicted functional effects. Once annotated, alterations were filtered for Oncogenic/Likely Oncogenic variants according to classification of the affected gene as an oncogene or tumor suppressor gene (TSG) according to the OncoKB Cancer Gene List, i.e., only gain-of-function (GOF) variants were retained for oncogenes and only loss-of-function (LOF) variants were retained for TSGs (Suppl. Fig. S1B–C). If a gene was classified as both an oncogene and a TSG, GOF and LOF variants were both retained. Copy number variants (CNVs) were considered Oncogenic/Likely Oncogenic if OncoPanel identified either amplification (≥6 copies) of a gene classified as an oncogene or deep deletion (2-copy loss) of a gene classified as a TSG. In summary, GOF mutations or amplifications in oncogenes and LOF mutations or deep deletions in TSGs were considered oncogenic alterations.

Identification of oncogenic fusions

We manually searched the OncoPanel SV data for fusions involving known RTK/RAS driver genes including ALK, BRAF, FGFR1-4, MET, NRG1, NTRK1-3, RAF, RET, and ROS1. Sequencing data for fusions was reviewed by a molecular pathologist (J.A.N), assessing the number and quality of split reads and the orientation of involved genes in support of the formation of a functional in-frame oncogenic fusion. As in prior studies³¹, RTK fusions were considered oncogenic if they were orthogonally validated (e.g., via RNA-based fusion assay, immunohistochemistry (IHC), or fluorescence in situ hybridization (FISH)) or if the fusion partner and breakpoint locations had been previously reported.

Immunohistochemical detection of ROS1 and TRK expression

IHC was performed on 4-μm thick formalin-fixed paraffin-embedded (FFPE) sections following pressure cooker antigen retrieval (0.001 M citrate buffer; pH 6.0), using a rabbit anti-ROS1 monoclonal antibody (1:100 dilution; 40 min incubation; clone D4D6; Cell Signaling Technology, Danvers, MA) and rabbit anti-pan-TRK monoclonal antibody (1:100 dilution; clone EPR17341; Abcam, Cambridge, MA). The EnVision+ Detection System (Dako, Carpinteria, CA) was used as a secondary antibody for ROS1 and the Novolink Polymer Detection System (Leica, Buffalo Grove, IL) was used for TRK staining.

ROS1 fluorescence in situ hybridization (FISH)

ROS1 FISH was carried out on interphase nuclei on 5-μm sections of FFPE tissue. ROS1 rearrangement was evaluated using the Kreatech Repeat-Free Poseidon ROS1 Break Probe Kit (Kreatech Inc., Durham, NC), which contains two differentially labeled probes which flank the ROS1 gene, located at 6q22. Probes and nuclei were co-denatured simultaneously, followed by hybridization and washing, according to the manufacturer’s instructions. An intact ROS1 locus is represented by two fused red/green signals. However, if a ROS1 rearrangement has occurred, one copy of the red probe will be separated from the adjacent green probe by ≥1 signal diameter or the 5′ (green) probe signal will be lost.

Validation of ROS1 fusion and detection of additional gene fusions with RNA-based fusion detection

Total nucleic acid was isolated from macro-dissected FFPE tumor tissue as per assay guidelines. A clinically validated Anchored Multiplex PCR assay was used for targeted fusion transcript detection using next generation sequencing and the ArcherDx FusionPlex Solid Tumor Kit primers. A laboratory-developed algorithm was used for fusion transcript detection and annotation³².

Genomic alteration visualizations

All OncoPrints were generated using ComplexHeatmap (v2.6.2, RRID:SCR_017270). Bar, violin, and volcano plots were generated using ggplot2 (v3.3.3, RRID:SCR_014601). All visualizations were generated using R (v4.0.3). Reported percentages for each gene reflect the percentage of samples which have an alteration in the given gene out of the total number of samples with an OncoPanel version which included the given gene.

TP53 alteration investigation

TP53 missense variants at the most frequently mutated codons (R175, Y220, G245, R249 and R273) were considered as hotspot variants³³. All TP53 missense variants were assigned their combined phenotype score as determined in Giacomelli et. Al.³⁴. The difference in combined phenotype score between the KRAS cohorts was tested as described in Statistical analysis.

BRAF alteration clonality assessment

To infer clonality of BRAF alterations in KRAS WT cases we compared variant allele fraction (VAF) of BRAF alterations with assessed tumor purity (TP) on histological examination. Based on methodology established within the DFCI Gastrointestinal Cancer Center molecular tumor board, cases with a VAF/TP ≥0.3 were considered to have clonal BRAF alterations³⁵.

TMB comparison between cohorts

Samples annotated by OncoPanel with high microsatellite instability were excluded from the comparison. The cohort difference was tested as described in Statistical analysis.

Differential gene/pathway alteration testing

Differential gene alteration testing between the KRAS WT and KRAS mutant groups was performed by counting the number of altered samples, which we defined as those samples containing ≥1 alteration (SNV, INDEL, CNV, or SV) in the gene being tested. Differential pathway testing between the KRAS cohorts was performed by counting the number of altered samples, which we defined as those samples with ≥1 alteration in ≥1 gene in the pathway being tested (Suppl. Table S2). All tested pathways were mutually exclusive and contained no overlapping genes. For pathway testing, only genes covered across all versions of OncoPanel (v1-v3.1) were included for consistency. Only pathways which consisted of 3 or more genes and had an alteration rate of ≥10% in either the KRAS WT or KRAS mutant cohort were included. The differences in total alteration counts between cohorts were then tested with Fisher’s Exact Test and p-values were corrected for multiple hypothesis testing using the Benjamini-Hochberg procedure.

Statistical analysis

Statistical tests were performed using R (v4.0.3) and Python (v3.8.3) (IPython, RRID:SCR_001658). Fisher’s Exact Tests were carried out in Python using the “fisher_exact” function from the stats module of Scipy (v1.4.1). Mann-Whitney U Tests were carried out in R using the “wilcox.test” function. Corrections for multiple hypothesis tests were performed using the Benjamini-Hochberg procedure with the “fdr_bh” method for the “multipletests” function from the stastmodels (v0.12.2) Python package.

Arm level copy number analysis

Arm-level copy number changes in targeted sequencing data were determined using the ASCETS algorithm³⁶. Enrichment analysis of arm-level CNV changes was performed using a custom permutation test as previously described³⁷. Briefly, the arm-level changes were randomly permuted 100,000 times, such that sample and arm-level CNV rates were preserved in each permutation, to form a background alteration rate. The one-tailed p-value for each arm-level alteration was calculated by the fraction of trials in which the difference in number of oncogenic alterations between the two groups met or exceeded the observed difference. Alterations with p-value <0.05 were considered significant, and those with q-value <0.25 were considered significant after multiple testing correction.

Detection of low-level copy number gains and copy-neutral loss of heterozygosity (CN-LOH)

Heterozygosity status was evaluated following methodology described previously³⁸. Briefly, we used common germline SNPs from the Affymetrix GenomeWideSNP_6 array to identify allelic imbalance in the genes of interest. Only SNPs with coverage of greater than 50 reads in total and greater than 5 reads for minor allele were used for the analysis. Allelic imbalances were registered if any heterozygotic germline SNP had a minor allele frequency below 25%. Copy-neutrality was identified using copy-ratio values generated by the RobustCNV algorithm³⁸. Five percent of segments with the highest and lowest copy ratios were excluded as copy ‘gains’ or ‘losses’ respectively, and the remaining events were identified as copy-neutral. The absolute value of log(copy-ratio) threshold was 0.42 for our data set. The KRAS gene did not have any well covered Affymetrix SNPs in our data set and, instead, we searched the gnomAD database (https://gnomad.broadinstitute.org/about) for heterozygous SNPs that are detectable by OncoPanel. We identified SNP rs1137282 at position 12:25362777 that met our criteria and was used in the same manner as described above.

Organoid culture

Organoids were generated under IRB-approved protocols at Dana-Farber Cancer Institute (DFCI #14-408, 17-000, 03-189). All patients provided written informed consent and the studies were conducted in accordance with recognized ethical guidelines. Organoids were cultured at 37°C in 5% CO₂. Cells were seeded in growth factor reduced Matrigel (Corning; Cat. #356231) domes and incubated with human complete feeding medium: Advanced DMEM/F12-based-conditioned medium, 1x B27 supplement, 10 mM HEPES, 2 mM GlutaMAX, 10 mM nicotinamide, 1.25 mM N-acetylcysteine, 50 ng/mL mEGF, 100 ng/mL hFGF10, 0.01 μM hGastrin I, 500 nM A83-01, Noggin 100ng/mL, 1x Wnt-3A conditioned 10% FBS DMEM (50% by volume) and 1X R-spondin Conditioned Basal Medium (10% by volume)^39,40.

Organoid drug treatment and viability assay

Organoids were dissociated and 1000 single cells per well were seeded in 20μl of culture media, consisting of 10% growth factor reduced Matrigel (Corning; Cat. #356231) and 90% human organoid medium, into a 384-well ULA plate (Corning; Cat. #4588) and a plate representative of the cell viability at the day of treatment was seeded in parallel for growth rate normalization.

One day after seeding, organoids were treated with inhibitors (Suppl. Table S3) or DMSO for normalization in a randomized fashion with a Tecan D300e Digital Dispenser (Tecan Trading AG, Männedorf, Switzerland). In parallel, cell viability was assessed using the Cell-TiterGlo 3D Cell Viability assay (Promega; Cat. #G9683) on the plate representative of cell viability on the day of treatment. Post-treatment viability was assessed 6 days after incubation with inhibitors on a FLUOstar Omega microplate reader following a Cell-TiterGlo 3D Cell Viability assay, executed according to the manufacturer’s instructions.

For inhibitor combination experiments, percentage viability was calculated by normalizing treated wells to DMSO treated control samples. For single inhibitor experiments, the cell viability post-treatment was normalized to the growth rate specific to each cell line using the cell viability on the day of treatment, as described in Hafner et al⁴¹. Curves were fit with nonlinear sigmoid functions, forcing the low-concentration asymptote to 100% viability to represent the DMSO control treatment, using a four-parameter curve in GraphPad Prism (GraphPad Prism, RRID:SCR_002798). Biological duplicate or triplicate measurements were recorded for each inhibitor concentration. Synergy scores were calculated using SynergyFindeer⁴² and the Loewe additivity model was used to calculate synergy scores⁴³.

Germline analysis

The electronic medical record (EMR) of each patient was searched for evidence of germline genetic evaluation. Assessment of germline genetic testing for patients was based on either 1) the original germline genetic testing results report, if available, 2) notes from Dana-Farber Genetics & Prevention consultation regarding germline genetic testing results, or 3) provider notes in reference to the results of germline genetic testing. The most recent interpretation of any variants by the company that performed testing was accepted as accurate.

Only the results of multi-gene panel testing were included in the analysis, i.e., dedicated gene/variant testing was excluded unless a pathogenic alteration was detected and the specific gene and variant were reported. Germline genetic testing panel sizes ranged from 13-156 genes with ~40-85 gene panels being most common. If mention of genetic testing or testing via patient report was noted in the absence of clear documentation (e.g., original germline genetic testing report, Dana-Farber Genetics & Prevention consultation, detailed provider notes regarding results of germline genetic testing), it was not considered as having been performed. Minimum sufficient evidence was considered note of a ‘Positive’ test result (i.e., report of ≥1 pathogenic variant) including report of a specific gene and variant. Note of a ‘Negative’ test (i.e., report of no variants of any kind) without supporting evidence was not considered as having been performed. A test with only benign variants or variants of uncertain significance (VUS) was considered ‘Negative’.

Clinical and survival data analysis

Available clinical data was compiled for all 795 patients in the cohort using the PRISSMM method of electronic medical record abstraction⁴⁴. Survival analysis was performed using the survival R package (v3.3-1) and visualized using the survminer R package (v.0.4.9). Covariates that were significant in univariable models were selected for inclusion in further multivariable analyses using Cox proportional hazards regression. The proportional hazards assumption was checked and satisfied for all covariates. The number of oncogenic alterations in each patient was defined as the total number of genes harboring an oncogenic alteration, i.e., the presence of ≥1 oncogenic SNV/INDEL, CNV, or SV in the given gene.

Data availability

Deidentified genomic data for the cohort will be available via project GENIE (Suppl. Table S4).

RESULTS

Pancreatic cancer targeted sequencing cohort

We identified a total of 795 exocrine pancreatic cancer cases (PDAC = 785, PASC = 5, PACC = 5) that had undergone OncoPanel sequencing and passed our QC metrics including at least 50x mean target coverage (Methods). Of these cases, 73 patients had KRAS WT (9.2%) and 722 had KRAS mutant tumors (Suppl. Figs. S1A–C). 67 patients with PDAC had KRAS WT tumors (8.5%).

An overview of the genomic landscape of the entire cohort (n = 795) revealed frequent alterations in KRAS (90.8%), TP53 (72.7%), CDKN2A (32.7%), and SMAD4 (21%) at rates consistent with those seen in previously published large patient datasets (Suppl. Fig. S2A–B)^4–6,14,20.

Genomic landscape of KRAS WT pancreatic cancer

KRAS WT exocrine pancreatic cancers demonstrated alterations in TP53, BRAF, CDKN2A, GNAS, ARID1A and SMAD4, as well as a longer tail of alterations that occurred in less than 10% of tumors (Fig. 1A). Comparison of the KRAS WT (n = 73) and KRAS mutant (n = 722) groups revealed no significant differences in tumor purity (p = 0.87, Mann-Whitney U Test), average read depth (WT Median: 291.7, IQR (240.5, 352.5), Mutant Median: 307.9 IQR (251.7, 376.2)), p = 0.43, Mann-Whitney U Test), or OncoPanel version (p = 0.33, Fisher’s Exact Test) (Suppl. Fig. S2C–E), suggesting comparable sample and sequencing quality. Both KRAS WT and mutant groups had a median of 3 oncogenic alterations per sample [WT: Mean 3.3, Range (0, 28) versus Mutant: Mean 3.6, Range (1, 28)]. KRAS mutant tumors had a slightly higher tumor mutational burden (TMB) compared to WT tumors [WT (Median (IQR)): 4.56 (3.04-6.08) Mut/Mb versus Mutant (Median (IQR)): 6.05 (4.56-7.26) Mut/Mb, p = 0.00046] (Suppl. Fig. S3A).

Figure 1. KRAS WT pancreatic cancers harbor frequent BRAF alterations.

(A) Mutational landscape of KRAS WT pancreatic cancer displays frequent oncogenic activating alterations in BRAF and GNAS. Inactivating alterations in TP53, CDKN2A, and SMAD4 appear disproportionately fewer compared to the entire pancreatic cancer cohort (TP53: 36% WT versus 73% Cohort; CDKN2A: 18% WT versus 33% Cohort; SMAD4: 11% WT versus 21% Cohort) (see Suppl. Fig. 2A). Signature ‘N/A’ indicates mutational signatures were not assessable on OncoPanel as some versions of the assay predate the use of the mutational signature algorithm. The alteration % reflects only samples with coverage of the relevant gene. Genes not covered in earlier OncoPanel versions on this plot include PBRM1 (v1) and MTAP (v1/v2). x-axis: Deidentified patient IDs. The two cases with red boxes represent BRAF in-frame deleted organoids used for drug testing in Figure 3. In-frame indels in BRAF refer to in-frame deletions (N486_P490del, n = 9) and duplications (T599dup, n = 1). PDAC = Pancreatic Adenocarcinoma, PACC = Pancreatic Acinar Cell Carcinoma, PASC = Pancreatic Adenosquamous Carcinoma, TMB = Tumor Mutational Burden

(B) Gene-level comparison between KRAS WT (n = 73) and KRAS mutant (n = 722) cases reveals significant enrichment in BRAF, GNAS, and ARID2 alterations in the WT cohort as well as enrichment in TP53 alterations in the mutant cohort. Statistical significance was assessed using Fisher’s Exact Test with the Benjamini-Hochberg Procedure for p-value multiple hypothesis corrections.

(C) Top, Most BRAF alterations found in KRAS WT pancreatic cancer are functionally Class II i.e., display constitutive dimer-dependent kinase activity. Bottom, Pathogenic BRAF alterations reside within the BRAF tyrosine kinase domain and in-frame deletions in the β3-αC loop are the most frequent (n = 9).

(D) Class II are the most common functional class of BRAF alterations in KRAS WT pancreatic cancer (72%) unlike colorectal cancer and melanoma in which Class I BRAF alterations are predominant (77% and 75%, respectively). Percentages reflect proportion of BRAF alterations of a given class out of the total number of BRAF mutations. Data for melanoma (n = 242) and colorectal cancer (n = 304) are from the PROFILE cohort in cBioPortal.

At a mutational level, we observed a significantly lower rate of LOF alterations in TP53 (35.6% versus 76.5%; p-adjusted = 2.4x10⁻¹⁰) in KRAS WT tumors compared to KRAS mutant tumors (Fig. 1B). No differences in rates of hotspot vs. non-hotspot TP53 variants were identified between KRAS WT and mutant tumors (4/16 [25%] vs. 99/354 [27.9%], Fisher’s exact test, p = 1.00; Suppl. Table S5). In an orthogonal approach we used combined phenotype score³⁴ to assess the pathogenic impact of TP53 variants and found no significant differences between KRAS WT and mutant tumors (Mean score: 1.011 vs. 1.019, p = 0.93; Suppl. Table S5). KRAS WT tumors also had a lower incidence of LOF alterations in CDKN2A (17.8% versus 34.2%; p-adjusted = 0.14) and SMAD4 (11.0% versus 22.0%; p-adjusted = 0.36), although these did not reach statistical significance. KRAS WT pancreatic cancer also demonstrated significant enrichment for oncogenic alterations in BRAF (24.7% versus 0.3%, p-adjusted = 4.96x10⁻¹⁶), GNAS (15.1% versus 2.6%, p-adjusted = 0.0014), and ARID2 (6.8% versus 0.7%, p-adjusted = 0.005) (Fig. 1B, Suppl. Table S6).

The KRAS WT group was significantly enriched for tumors with a mismatch repair deficiency (MMR-D) mutational signature (5.5%, 4/73 WT versus 0.5%, 4/722 Mutant, p = 0.0035) (Suppl. Figure S2A). Review of genomic data from these cases revealed the presence of pathogenic alterations in MMR genes in 5/8 cases (62.5%), whereas the mechanism of MMR-D could not be determined from OncoPanel genomic data in the remaining 3 cases. 3/5 cases with MMR gene alterations were confirmed to have Lynch syndrome on confirmatory germline testing, 1 case had a somatic MMR-gene mutation, whereas the cause of the MMR gene mutation could not be determined in the last case due to lack of germline testing (Suppl. Table S7). Two KRAS WT pancreatic cancer cases had an APOBEC signature and one case had a tobacco smoke signature. These other signatures, though rare in KRAS WT tumors, were not identified in the KRAS mutant group (APOBEC: 5.4%, 2/37 WT versus 0/409 Mutant, p = 0.0067; Tobacco: 2.7%, 1/37 WT versus 0/409, Mutant, p = 0.083).

The frequency of specific CNVs differed between the KRAS WT and mutant groups. Deletions in the long arms (q) of chromosomes 3, 10 and 11 were enriched in KRAS WT pancreatic cancer (p < 0.05 & q < 0.25, permutation test ASCETS³⁶) whereas the KRAS mutant cases had a significantly higher rate of loss of the short arm (p) of chromosome 17, which includes the TP53 locus (p < 0.05 permutation test ASCETS³⁶; Suppl. Fig. S3B). Analysis of pancreatic cancer data from TCGA⁶ confirmed the enrichment of chromosome 17p deletion in KRAS mutant cases (Suppl. Fig. S3C). Chromosomal alterations on chromosomes 3q, 10q and 11q specific to the KRAS WT cases in our cohort were not confirmed in the TCGA cohort, however the KRAS WT sample size in that cohort was relatively small (n = 10). Interestingly, the chromosomal regions preferentially deleted in KRAS WT pancreatic cancer harbor important TSGs including PTEN, and DNA-damage genes including ATM, ATR and CHEK1 (Suppl. Table S8). Given the focused genomic coverage in OncoPanel, ploidy estimation is not possible in all cases with OncoPanel data, and these estimates are not validated or provided as part of the clinically reported dataset. In addition, we found that the accuracy of ploidy estimation using OncoPanel is highly dependent upon tumor content, and accuracy may well be limited in the lower tumor content cases included in this study.

BRAF mutations in KRAS WT pancreatic cancer

As BRAF alterations are known MAPK drivers^6,8,20,27, we sought to further characterize the pathogenic BRAF variants in the KRAS WT cohort (Fig. 1C). Eighteen out of twenty (90%) BRAF alterations identified in our dataset were found in the KRAS WT group. Of those in the KRAS WT group, 9 out of 18 (50.0%) BRAF alterations were in-frame deletions, 7 out of 18 (38.9%) were missense alterations, and one in-frame insertion and one SND1::BRAF fusion were identified (Fig. 1C, Fig. 2A). There were 6 unique BRAF SNV/INDELs identified in our KRAS WT cohort: p.N486_P490del (n = 9), p.V600E (n = 3), p.G469S (n = 2), p.L485F (n = 1), p.D594G (n = 1), and p.T599dup (n = 1). Notably, the SND1::BRAF fusion was identified in a case of PACC. In KRAS WT pancreatic cancers, most BRAF mutations (72.2%) were classified as Class II variants (Fig. 1C). This class of variants has been reported to exhibit dimer-dependent constitutive activity^45,46. This is in sharp contrast to melanoma and colorectal cancer in which the majority of BRAF altered cases harbor the canonical p.V600E allele (Class I) which possesses constitutive monomeric activity (Fig. 1D). Since Class II BRAF variants can signal as homo- or hetero-dimers, we assessed the copy number status of the BRAF locus in KRAS WT tumors with BRAF alterations. Raw copy number profiles were available for 15/18 tumors and 12/15 also had adequate coverage of germline SNPs to allow CN-LOH assessment. The majority of the cases (13/15, 86.7%) were copy-neutral at the BRAF locus and one sample each showed evidence of copy number loss and CN-LOH (Suppl. Table S9). Comparison of BRAF alteration VAF with tumor purity (TP) suggested that the majority of these mutations were clonally present (12/16, 75%, Suppl. Table S10). In an orthogonal approach, we compared the VAFs of KRAS mutations in the KRAS mutant cohort with the VAFs of BRAF mutations in the KRAS WT cohort and found no statistically significant difference in their distributions (p = 0.34, Kolmogorov Smirnov test, D-statistic = 0.24; Suppl. Fig. S3D). Since KRAS mutations are known to be clonal in pancreatic cancer this would suggest that BRAF mutations in KRAS WT pancreatic cancer are also likely clonal. Overall, both approaches suggest that BRAF alterations are likely clonal events in KRAS WT pancreatic cancer.

Figure 2. Alternative MAPK alterations including RTK fusions are frequent drivers in KRAS WT pancreatic cancer.

(A) Exon/intron structure of predicted RTK fusion transcripts resulting from rearrangements in KRAS WT tumors detected via DNA-based targeted sequencing using OncoPanel (n = 5) or via RNA-based fusion assay ArcherDx (n = 3). Shown are relevant RTK functional domains as encoded across exons.

(B) Immunohistochemistry with a pan-TRK antibody reveals diffuse membranous TRK staining supporting the presence of the detected TJAP1::NTRK1 fusion.

(C) Left, ROS1 break-apart fluorescence in situ hybridization (FISH). In this system, separation of the normally fused red/green signal is indicative of a genomic alteration. The loss of red signal suggests deletion of the 5’ end of ROS1 (Exons 1–32) which was also captured by sequencing (see Fig. 4D). Right, ROS1 IHC displays ectopic ROS1 expression confirming the presence of the detected SLC4A4::ROS1 rearrangement.

(D) Alternative (i.e., non-KRAS) MAPK alterations are identified in 32 out of 73 (44%) KRAS WT tumors with BRAF alterations and RTK fusions being the most frequent. Amplifications in other MAPK drivers including ERBB2, EGFR, KRAS and MET were also identified. Alternative MAPK drivers tend to be mutually exclusive pointing to a likely driver role. The alteration % reflects only samples with coverage of the relevant gene. Genes not covered in earlier OncoPanel versions on this plot include NRG1, RAC1, and RASA1 (v1/v2). PDAC = Pancreatic Adenocarcinoma, PACC = Pancreatic Acinar Cell Carcinoma, PASC = Pancreatic Adenosquamous Carcinoma, TMB = Tumor Mutational Burden

(E) Pathway-level comparison of KRAS WT and mutant pancreatic cancer shows enrichment in alternative MAPK genes in the WT cohort. Statistical significance was assessed using Fisher’s Exact Test with the Benjamini-Hochberg Procedure for p-value multiple hypothesis corrections.

The two BRAF mutations co-occurring with oncogenic KRAS mutations in the KRAS mutant cohort were p.G464V (Class II, 26% VAF) paired with KRAS p.G12D (27% VAF) and p.G596D (Class III, 37% VAF) paired with KRAS p.G12V (44% VAF). While Class III BRAF mutations have impaired kinase activity, they remain dependent on upstream signaling and are often observed co-occurring with RAS mutations. Based on estimated tumor purity and associated allelic fractions of both KRAS and BRAF mutations in both specimens, it appears these alterations are likely co-occurring in the same cell (Suppl. Table S11). The co-occurrence of a RAS independent Class II BRAF mutation with a KRAS p.G12D mutation in a single tumor is unexpected.

KRAS WT pancreatic cancer is enriched for receptor tyrosine kinase (RTK) fusions

We identified 5 cases with oncogenic BRAF or RTK fusions in the KRAS WT pancreatic cancer cohort based on OncoPanel: TJAP1::NTRK1, SLC4A4::ROS1, APP::NRG1, and FGFR2::KIAA1598 in PDAC and a SND1::BRAF fusion in a case of PACC (Fig. 2A & Suppl. Fig. S4A). The presence of the TJAP1::NTRK1 fusion was orthogonally validated using TRK immunohistochemistry (IHC) (Fig. 2B). The partner gene involved in the ROS1 rearrangement was unclear from the OncoPanel data, however subsequent RNA-based fusion analysis identified SLC4A4 as the partner gene. IHC performed on this case demonstrated ROS1 expression throughout tumor cells and ROS1 break-apart fluorescence in situ hybridization (FISH) testing further confirmed the ROS1 rearrangement (Fig. 2C). There was insufficient tissue for validation of the remaining fusions detected by DNA-sequencing, however, they fulfilled our criteria for inclusion due to predicted structures containing in-frame fusion transcripts in which key functional domains were present in the expected configuration (Fig. 2A, Suppl. Fig. S4A). Within our targeted sequencing cohort, RTK fusion events were significantly enriched in the KRAS WT group (6.8%, 5/73 WT versus 0.1%, 1/722 Mutant, p-adjusted = 3.2x10⁻⁵). Notably, none of the KRAS WT cases with an RTK fusion harbored another activating BRAF or MAPK pathway alteration. In the KRAS mutant group, an FGFR2::NPM1 fusion was identified in one case with a co-occurring KRAS p.G12D mutation (8% VAF).

Given the observation that oncogenic RTK fusions were enriched in KRAS WT tumors and with knowledge of the limitations of targeted DNA sequencing for detecting certain rearrangement events, we performed RNA-based fusion analysis³² on KRAS WT tumors that lacked a clear alternative MAPK driver and for which adequate tissue was available (n = 19) (Suppl. Fig. S4B). We discovered an additional 3 oncogenic RTK fusions in these cases (3/19, 15.8% positivity rate): APP::NRG1, KANK1::NTRK3, and DCBLD1::ROS1 (Fig. 2A). Thus, we were able to detect actionable oncogenic fusions in a total of 8/73 (11.0%) KRAS wild-type pancreatic cancers, supporting the recommendation for RNA-based fusion detection in patients with KRAS WT pancreatic cancer lacking a clear MAPK driver.

Enrichment of alternative MAPK drivers in KRAS WT pancreatic cancer

We observed that 43.8% (32 of 73) patients with KRAS WT pancreatic cancer in our cohort had evidence of an alternative MAPK driver other than KRAS. Alternative oncogenic MAPK drivers included amplifications of EGFR (n = 1), ERBB2 (n = 1), KRAS (n = 1), and MET (n = 1) and LOF mutations in negative regulators of the MAPK pathway including NF1 (n = 2) and RASA1 (n = 1) (Fig. 2D, Suppl. Fig. S5A, Suppl. Fig. S6). The vast majority of alternative MAPK driver alterations (30/32) were mutually exclusive, pointing to a likely driver role. In one case, an FGFR1 amplification co-occurred with a Class III BRAF alteration (p.D594G), with the latter presumably amplifying downstream MAPK signaling^45,46. A second case harbored both KRAS and NTRK1 amplifications.

Using previously curated gene lists for cancer-relevant pathways, we compared KRAS WT and mutant groups at the pathway level to identify potential differences which may not have been apparent in the single-gene analysis (Fig. 2E & Suppl. Table S2)^20,47. As expected from our initial findings of the presence of activating BRAF alterations and RTK fusions, the KRAS WT group was significantly enriched in alternative (i.e., non-KRAS) MAPK drivers (39.7%, 29/73 WT versus 5.1%, 37/722 Mutant, p-adjusted = 5.6×10⁻¹⁵) (Fig. 2D–E, Suppl. Fig. S5A). This enrichment remained significant even when BRAF alterations were removed from the comparison (16.5%, 12/73 WT versus 4.9%, 36/722 Mutant, p-adjusted = 0.0048).

No alternative MAPK driver could be identified in 56.2% (n = 41) of KRAS WT cases in our cohort. We hypothesized that genetic alterations leading to downstream activation of other pro-survival pathways may be enriched in these cases. Within the 41 cases without a clear MAPK driver alteration, 29.3% (12/41) harbored activating alterations in alternative oncogenic drivers such as GNAS, MYC, PIK3CA and CTNNB1. At a pathway level, there was no statistically significant difference between the KRAS WT and KRAS mutant cohorts for alterations in the PI3K pathway (13.7%, 10/73 WT versus 5.8%, 42/722 Mutant, p-adjusted = 0.083) (Fig. 2E, Suppl. Fig. 5B), the SWI/SNF pathway (17.8%, 13/73 WT versus 9.3%, 67/722 Mutant, p-adjusted = 0.11) (Fig. 2E, Suppl. Fig. S5C), the DNA damage repair (DDR) pathway (15.1%, 11/73 WT versus 12.9%, 93/722 Mutant, p-adjusted = 1) (Fig. 2E, Suppl. Fig. S5D) or the homologous recombination (HR) pathway (9.6%, 7/73 WT versus 10.4%, 75/722 Mutant, p-adjusted = 1) (Suppl. Fig. S5E). Furthermore, we found no clear gene-level or pathway-level genomic differences between KRAS WT pancreatic cancers which either harbored or lacked activating MAPK alterations in our cohort (Suppl. Tables S12–13). Since determination of copy number in low purity pancreatic cancer samples can be challenging, we re-examined the KRAS locus and liberalized our analysis to allow low-level copy number gains in MAPK genes (including KRAS) as potential drivers in these KRAS WT, MAPK negative tumors. We examined 40/41 cases with available raw copy number data. The KRAS locus showed a low-level gain in 1/40 cases, low-level copy number loss in 2/40 cases, and was copy-neutral in the remaining 37 cases. There was no evidence of copy-neutral loss of heterozygosity (CN-LOH) at KRAS. A total of 14/40 (35%) samples had evidence of low-level copy number gains in any MAPK pathway genes. For the samples which had low-level copy number gains, the median number of low-level gains was 2 per sample (Range: 1–5). Twenty-one cases (52.5%) displayed either copy number loss (n = 20), or CN-LOH (n = 8) in known TSGs (Suppl. Fig 7). To identify other rare or atypical MAPK activating mutations which may have been inadvertently removed due to stringent filtering criteria contributing to our identification of the KRAS WT, MAPK negative cohort, we further expanded our analysis to include all VUS variants in the MAPK pathway and assessed for potential oncogenicity using a variant effect predictor tool⁴⁸. We noted evidence of 13 additional SNVs in MAPK genes in 10/40 KRAS WT, MAPK negative cases however none of these scored highly for oncogenicity using the Ensembl variant effect predictor tool (Suppl. Table S14). In summary, the majority of cases (26/40, 65%) with KRAS WT, MAPK negative pancreatic cancer do not show evidence of genetic activation of the MAPK pathway, even after relaxing our initially stringent selection criteria.

To determine if alternative oncogenic drivers in this cohort were missed due to the targeted nature of coverage on the OncoPanel platform, we performed an in-silico analysis of whole exome sequencing (WES) data from the PACA-AU and PACA-CA cohorts^5,49. We initially restricted our analyses to the 447 genes covered by OncoPanel. Using similar methodology as used on the OncoPanel cases, 70/764 (9.2%) tumors were KRAS WT and 45/70 were KRAS WT and MAPK negative. To identify if broadening sequencing coverage to all exons and genes identified additional potential oncogenic drivers, we performed the same analysis using the complete WES data for the same samples. This analysis did not yield a substantial increase in the number of MAPK oncogenic drivers identified in KRAS WT pancreatic cancer (3/45, 6.7%). These additional cases had alterations in SPRED2 and RASAL3 in one case, and INSR and KSR2, respectively, in the other two cases.

Germline susceptibility in KRAS WT versus KRAS mutant pancreatic cancer

Prior studies have suggested an increased incidence of pathogenic germline alterations in KRAS WT PDAC⁶. Therefore, we sought to determine whether there was an association between KRAS WT pancreatic cancer and pathogenic variants in germline susceptibility genes. Electronic medical record (EMR) review showed that germline testing results were available for a similar proportion of patients in both cohorts (24/73, 32.9% WT versus 257/722, 35.6% Mutant) and that pathogenic germline alterations were identified in a similar proportion of both cohorts (5/24, 20.8% WT versus 48/257, 18.7% Mutant) (Suppl. Fig. S8A–B, Suppl. Table S15). More specifically, there was no significant difference between the incidence of germline pathogenic variants in homologous recombination (HR) ‘Core’ genes (8.3%, 2/24 WT versus 6.6%, 17/257 Mutant, p = 0.67) nor in combined HR ‘Core’/’Non-Core’ genes (16.7%, 4/24 WT versus 12.1%, 31/257 Mutant, p = 0.517) as defined by Park et. Al¹⁶ (Suppl. Fig. S8C–D).

Dual pan-RAF/MEK inhibition demonstrates efficacy in BRAF-mutated pancreatic cancer organoid models

We derived pancreatic cancer organoids from two patients with KRAS WT pancreatic cancer from our current cohort (Fig. 1A, red box; Suppl. Table S16) who harbored a BRAF p.N486_P490del in-frame deletion and performed preclinical drug testing of MAPK targeted therapies. The BRAF^V600E specific inhibitor dabrafenib has no growth inhibitory activity on these organoid models (Fig. 3A). This has been attributed to dabrafenib and other similar BRAF^V600E inhibitors stabilizing the αC helix of the BRAF protein towards the “out” conformation¹⁹, while BRAF in-frame deletions fix the αC helix towards the “in” conformation leading to constitutive activation and the inability to bind dabrafenib¹⁹. In contrast, these organoids were sensitive to the pan-RAF inhibitor LY3009120 which can bind the “in” conformation of the BRAF protein (Fig. 3A). These organoid lines were also highly sensitive to the MEK inhibitor trametinib and the ERK inhibitor BVD-523 (Fig. 3A).

Figure 3. Dual pan-RAF and MEK inhibition demonstrates synergistic inhibition of pancreatic cancer patient-derived organoids harboring BRAF in-frame deletions.

(A) Two independent pancreatic cancer patient-derived organoids (PDOs) harboring BRAF in-frame deletions (p.N486_P490del) are resistant to the growth inhibitory effects of the BRAF^V600 inhibitor (dabrafenib) but remain sensitive to pan-RAF (LY3009120), MEK (trametinib), and ERK (BVD-523) inhibitors. Shown are growth rate-corrected dose sensitivity curves.

(B) Combination of pan-RAF inhibitor (LY3009120) with MEK (trametinib) or ERK (BVD-523) inhibitors shows synergistic activity in the PANFR0172_T2 PDO line whereas no synergy is displayed by combining a pan-RAF (LY3009120) inhibitor with a BRAF^V600 inhibitor (dabrafenib). Growth inhibition with various combinations of drug concentrations is displayed as a sigmoidal curve (Top) or heatmap (Middle). (Bottom) Calculation of synergy between different compounds using the Lowe additivity model.

We hypothesized that vertical MAPK inhibition with a pan-RAF inhibitor and a downstream MAPK inhibitor (i.e., MEK or ERK inhibitor) would display synergy in inhibiting the growth of cell lines with BRAF in-frame deletions similar to the synergy seen with dual BRAF and MEK inhibition in BRAF V600E mutant melanoma⁵⁰. Indeed, dual targeting of the MAPK pathway with the pan-RAF inhibitor LY3009120 and the MEK1/2 inhibitor trametinib showed strong synergy according to the Loewe additivity model (Fig. 3B & Suppl. Fig. S9A). We also observed more modest synergy with dual pan-RAF and ERK inhibition (Fig. 3B & Suppl. Fig. S9A). As expected, the addition of dabrafenib to the pan-RAF inhibitor LY3009120 had no added impact on growth inhibition for organoids with BRAF in-frame deletions. We tested these drug combinations in pancreatic cancer cell line, BxPC3, which harbors the same BRAF in-frame deletion as in our organoid models used above. Similar to the organoids, BxPC3 showed synergistic growth inhibition with the combination of pan-RAF inhibitor LY3009120 with both MEK1/2 or ERK inhibition, but not with dabrafenib (Suppl. Fig. S9B).

Durable clinical benefit from targeting a SLC4A4::ROS1 fusion in a patient with KRAS WT PDAC

An 83-year-old man was diagnosed with metastatic pancreatic adenocarcinoma (PAN00202). Molecular profiling of the tumor using OncoPanel revealed a two-copy deletion of CDKN2A (p16) and no KRAS mutation. Instead, OncoPanel detected a genetic rearrangement involving ROS1. ROS1 FISH revealed deletion of the 5’ portion of ROS1 supporting a rearrangement with preservation of the kinase domain (Fig. 2C). ROS1 IHC revealed diffuse cytoplasmic ROS1 expression within the adenocarcinoma, further supporting a functional ROS1 rearrangement (Fig. 2C). Given the patient’s advanced age and on the basis of these molecular findings, the patient began first-line therapy with crizotinib, a targeted inhibitor of ROS1, instead of chemotherapy (Fig. 4A). Restaging scans obtained after 2 months on treatment revealed a partial response with a 31.2% reduction in tumor size by RECIST v1.1 (Fig. 4B–C, Suppl. Fig. S10A). After nearly 8 months on treatment with crizotinib, restaging scans (Day 234) revealed progression of disease. A repeat biopsy of one of the progressing liver lesions was obtained to ascertain a potential resistance mechanism. Histologic evaluation revealed no overt morphologic changes in comparison with the pre-treatment biopsy (Suppl. Fig. S10B). ROS1 protein expression was maintained and repeat OncoPanel confirmed prior evidence indicative of the presence of a ROS1 rearrangement (Fig. 4D). A targeted RNA-based translocation assay³² was performed on the resistant sample which confirmed an in-frame fusion involving exons 1-24 of SLC4A4 (electrogenic sodium bicarbonate cotransporter 1) to exons 33-43 of ROS1, inclusive of the entire ROS1 kinase domain (Fig. 2A). No new mutations potentially suggestive of a genetic resistance mechanism were identified by the repeat OncoPanel analysis (Suppl. Fig. S10C). Treatment with cabozantinib, an alternative multi-targeted tyrosine kinase inhibitor against ROS1 and other kinases, was initiated given prior reports of its ability to overcome crizotinib resistance in vitro⁵¹. The patient had stable disease on cabozantinib for nearly 6 months (Day 430) (Figs. 4A–C, Suppl. Fig. S10A), although the patient needed multiple treatment breaks due to treatment-related toxicity. Upon disease progression on cabozantinib, the patient was subsequently re-trialed on crizotinib but had continued disease progression and was switched to systemic chemotherapy with gemcitabine/nab-paclitaxel. The patient eventually passed away 22 months after diagnosis. Thus, this KRAS WT pancreatic cancer patient harboring a SLC4A4::ROS1 fusion achieved approximately 14 months of clinical benefit from ROS1-targeted therapy.

Figure 4. Durable response to sequential ROS1 targeted therapy in a KRAS WT PDAC harboring a SLC4A4-ROS1 fusion.

(A) Treatment history of index patient with a SLC4A4-ROS1 fusion who received prolonged benefit from targeted therapy with upfront crizotinib followed by cabozantinib.

(B) Representative images with target lesions in the liver and lung (yellow arrow) showing initial response to crizotinib (Day 61) followed by disease progression at Day 234. Target lesions display overall stability between Days 234 to 361 while the patient was receiving cabozantinib.

(C) Longitudinal monitoring of 5 target lesions used to determine disease response while on ROS1-directed therapy.

(D) Genome-level copy number profiles from initial diagnostic (Top) and crizotinib-resistant (Day 234, Middle) samples both show chromosome 6q deletion (arrowhead) leading to loss of the 5’ end of ROS1, indicative of a likely genomic rearrangement involving ROS1. (Bottom) Magnified view of the ROS1 Exon 1–32 deletion.

Clinical and prognostic features in KRAS WT pancreatic cancer

Having established key genomic features and associated therapeutic tractability in KRAS WT pancreatic cancer, we assessed clinical differences between the two cohorts and the impact of genomic features on outcomes in the KRAS WT cohort. The KRAS WT cohort had significantly higher rates of variant histologies, in particular PACC, compared to KRAS mutant cases (PACC: 6.8%, 5/73 WT versus 0.0%, 0/722 Mutant, p = 5.75x10⁻⁶). In addition, KRAS WT patients were significantly younger compared to those with KRAS mutant tumors (p-value = 0.037) (Table 1). In particular, young-onset cases (<45y) were significantly enriched in KRAS WT tumors (6.8%, 5/73 WT vs. 2.2%, 16/722 Mutant, p = 0.036). Interestingly, in addition to lack of KRAS mutation, all 5 young-onset cases also lacked alterations in TP53, CDKN2A, and SMAD4 (Suppl. Table S17). Notably, 1/5 patients had a defined alternative MAPK driver with the presence of an FGFR2-KIAA1598 fusion, but 3/5 cases had negative ArcherDx testing, consistent with the lack of fusion event detected by OncoPanel. The 5^th patient did not have enough tissue to allow for ArcherDx testing. Other oncogenic alterations were also rare in these cases (Suppl. Table S17). Only 1/5 patients underwent germline genetic testing which was negative for pathogenic germline variants. When we restricted our analysis to patients with no detectable pathogenic germline alteration, the age of onset of KRAS WT pancreatic cancer cases remained significantly younger (KRAS WT, n = 20: Mean 60, Range [33, 77], KRAS Mutant, n = 209: Mean 65, Range [37, 83], p = 0.035). Lastly, we did not identify any clinical, or genomic differences between patients with KRAS WT pancreatic cancer with early vs. late onset of disease (Suppl. Table S18).

TABLE 1.

Clinical features of KRAS WT and Mutant pancreatic cancers

		KRAS Status
Characteristic	Category	WT (n=73)	Mutant (n=722)	P-Value
Age At Diagnosis	Years	62.6 (12.1)	65.7 (9.7)	0.037
Sex
	Female	29 (39.7%)	312 (43.2%)	0.653
	Male	44 (60.3%)	410 (56.8%)
Race
	Other	3 (4.1%)	41 (5.7%)	0.789
	White	68 (93.2%)	669 (92.7%)
	Unknown	2 (2.7%)	12 (1.7%)
Smoking Status
	Current	5 (6.8%)	72 (10.0%)	0.576
	Former	34 (46.6%)	298 (41.3%)
	Never	33 (45.2%)	348 (48.2%)
	Unknown	1 (1.4%)	4 (0.6%)
Diabetes Status
	Yes	25 (34.2%)	236 (32.7%)	0.794
	No	47 (64.4%)	476 (65.9%)
	Unknown	1 (1.4%)	10 (1.4%)
Stage At Diagnosis
	Local	27 (37.0%)	323 (44.7%)	0.314
	Metastatic	43 (58.9%)	388 (53.7%)
	Unknown	3 (4.1%)	11 (1.5%)
Histology
	Ductal adenocarcinoma	67 (92%)	718 (99.5%)	<0.0001
	Acinar cell carcinoma	5 (7%)	0 (0%)
	Adenosquamous	1 (1%)	4 (0.5%)
Mismatch Repair Deficiency #
	MMR-Deficient	4 (5.5%)	4 (0.5%)	0.0035
	MMR-Proficient	69 (94.5%)	718 (99.5%)

^*p-values from Fisher’s Exact Test and Unknowns excluded

^#MMR status determined by presence of MMR-D signature on OncoPanel sequencing

Next, we evaluated the impact of clinical and genomic features on patient outcomes. Since our cohort consisted of patients with both localized and metastatic disease, we used overall survival (OS) as our outcome measure. As has been shown in other reports focused on resected pancreatic cancers, we found that a greater number of alterations in core pancreatic cancer genes (KRAS, TP53, CDKN2A, SMAD4) adversely impacted OS in all patients with pancreatic cancer (Fig. 5A)^14,15. In addition, cases with KRAS WT pancreatic cancer had an improved OS compared to the KRAS mutant cohort (Fig. 5B). This remained statistically significant after accounting for the impact of disease stage and age using a Cox proportional hazards regression (Suppl. Fig 11) [KRAS Mutant: HR 1.42, 95% CI 1.05-1.92, p-adjusted = 0.023]. We then assessed the prognostic implications of various clinical and genomic factors in KRAS WT cases and compared it to the prognostic value of the same variables in the KRAS mutant cohort (Figs. 5C–D). In addition to advanced disease stage, the presence of inactivating SMAD4 alterations was associated with a worse OS in both KRAS WT and mutant cohorts [KRAS WT: HR 7.90, 95% CI 2.33-26.73, p-adjusted <0.001 and KRAS Mutant: HR 1.56, 95% CI 1.27-1.91, p-adjusted<0.001]. Interestingly, CDKN2A alterations were associated with a worse OS in KRAS mutant cases (HR 1.23, 95% CI 1.02-1.48, p-adjusted = 0.028, KRAS:CDKN2A interaction p-value = 0.022) and were unexpectedly associated with an improved OS in the KRAS WT cohort (HR 0.35, 95% CI 0.16-0.80, p-adjusted = 0.0127). Lastly, there was no prognostic significance for BRAF (HR 0.70, 95% CI 0.33-1.52, p = 0.37) or MAPK alterations (HR 1.31, 95% CI 0.73-2.37, p = 0.39) in KRAS WT patients.

Figure 5. Impact of genomic alterations on pancreatic cancer survival.

(A) Increased number of core pancreatic cancer mutations (KRAS, TP53, CDKN2A, SMAD4) is associated with significantly worse overall survival in the entire pancreatic cancer cohort by Kaplan-Meier analysis. P-value calculated by the log-rank test.

(B) KRAS WT pancreatic cancer is associated with a significantly improved overall survival compared to KRAS mutant pancreatic cancer by Kaplan-Meier analysis. P-value calculated by the log-rank test.

(C-D) Association of clinical and genomic features with overall survival in patients with KRAS WT (C) and KRAS mutant (D) pancreatic cancers using Cox proportional-hazards regression. P-values calculated by the Cox proportional-hazards model, with significant p-values are annotated with a star.

(E) Precision therapy landscape for KRAS WT pancreatic cancer. Summary of targetable genomic alterations detected in KRAS WT pancreatic cancer and potential therapeutic options for genomic subgroups. i = Inhibitor

DISCUSSION

Activating mutations in KRAS are a hallmark genomic feature of PDAC and are present in approximately 90% of tumors. However, this leaves a small yet important proportion of PDAC tumors which lack activating KRAS mutations^4–9. We analyzed the clinical and genomic features of KRAS WT pancreatic cancer from a large single institution patient cohort. Overall, 9.2% of our cohort had KRAS WT pancreatic cancer. These cases were enriched for alternative MAPK drivers including BRAF activating alterations and RTK fusions and, overall, 43.8% of patients with KRAS WT pancreatic cancer harbored an identifiable alternative MAPK driver in our cohort. Further analysis with transcriptional, epigenetic, and proteomic profiling may yield insights and new therapeutic targets into these KRAS WT pancreatic cancers without clear alternative MAPK driver alterations. Recent clinical data demonstrated a survival advantage with the addition of the EGFR blocking monoclonal antibody nimotuzumab to gemcitabine in patients with KRAS WT pancreatic cancer⁵². Akin to the efficacy of EGFR monoclonal antibodies in RAS and RAF WT colorectal cancers, it is tempting to speculate that this combination may have particular efficacy in the subset of KRAS WT cases without obvious MAPK drivers where wild-type EGFR may mediate survival signaling.

In the described KRAS WT cohort, BRAF^V600E, the most frequent BRAF mutation across all cancers, occurred less frequently than BRAF in-frame deletions and other BRAF Class II mutations. The underlying biology for this observation remains unclear but may reflect differences in mutational processes or downstream signaling pathways that may provide a selective advantage for Class II BRAF mutations. These findings have considerable therapeutic implications for targeting BRAF-mutated pancreatic cancer. To understand targeted therapy options for this unique disease subtype, we generated two independent patient derived organoids (PDOs) from tumors harboring BRAF in-frame deletions. These organoid lines were resistant to dabrafenib but highly sensitive to the pan-RAF inhibitor LY3009120, as previously described in two-dimensional cancer cell lines^18,19. Importantly, a combination of pan-RAF and MEK inhibition displayed synergy and was highly potent against PDOs with BRAF in-frame deletions. Historically, pan-RAF inhibitors have been associated with significant toxicity and been difficult to dose escalate in patients; however, novel pan-RAF inhibitors with improved safety profiles are now in clinical trials (e.g., NCT03284502)^53,54. Based on these data, further investigation of novel pan-RAF and MEK inhibitor combinations is warranted in this unique patient population harboring BRAF in-frame deletions, which constitute approximately 1% of all patients with pancreatic cancer.

RTK fusions were the second most frequent targetable MAPK lesion in KRAS WT pancreatic cancer and were mutually exclusive with BRAF mutations. Although recurrent RTK fusions have been described in various case series of KRAS WT pancreatic cancer, their overall prevalence in this group of patients is not clear^{22,25,55–57}. Our data suggests that approximately 11% of patients with KRAS WT tumors harbor RTK fusion oncoproteins. This finding and prior reports of therapeutic efficacy make a strong case for investigating the therapeutic targeting of RTK fusions in these patients. Furthermore, we describe a case of KRAS WT PDAC harboring a ROS1 fusion where the patient was treated with upfront targeted therapy with crizotinib based on the results of genomic testing. Notably, identification of fusions using DNA capture-based assays can be challenging³¹. In particular, NRG1 fusions, which have been described in several KRAS WT pancreatic cancer cases and are known to occur via complex genomic rearrangements^22–24, can be difficult to identify by capture-based DNA sequencing assays. Indeed, 3 out of 19 KRAS WT tumors negative for alternative MAPK drivers by OncoPanel harbored RTK fusions, which were identified using an RNA-based assay³². Hence, a comprehensive multimodal approach combining DNA- and RNA-based assays along with prudent use of IHC and/or FISH is necessary for the identification of RTK fusions and should be considered in KRAS/BRAF WT pancreatic cancer. Interestingly, in a study of a large cohort of KRAS WT pancreatic cancer which utilized combined DNA and RNA sequencing approaches in all cases, Philip et al. identified fusions in 21% of the entire KRAS WT cohort⁹. Hence, the true prevalence of RTK fusions in KRAS WT pancreatic cancer may be even higher than that found in our study.

In addition to the presence of alternative MAPK drivers, we identified additional features that distinguish KRAS WT pancreatic cancer from KRAS mutant cancers. KRAS WT pancreatic cancers have an overall lower tumor mutational burden than KRAS mutant tumors. Rates of TP53 alterations were statistically significantly lower in KRAS WT pancreatic cancers, and loss of function mutations in CDKN2A and SMAD4 were also observed to be less frequent in KRAS WT disease, although not reaching statistical significance in this cohort. A lower number of canonical TSG alterations has been observed to be associated with improved overall survival in pancreatic cancer¹⁴ and this may contribute to the improved survival outcomes seen in patients with KRAS WT tumors. Furthermore, mutations in GNAS were almost exclusively found in KRAS WT tumors. Since GNAS mutations are frequent in cancers arising from intraductal papillary mucinous neoplasms (IPMNs), we hypothesize that these tumors likely arose from IPMN precursor lesions⁵⁸.

Clinical-genomic correlation showed that patients with age of pancreatic cancer onset earlier than 45 years were significantly enriched in our KRAS WT cohort, a finding also supported by a recent study of early-onset pancreatic cancer⁵⁹. Varghese et. al., with a larger sample size of early-onset pancreatic cancer, showed that 15.9% cases were KRAS WT, and that 31.9% cases had evidence of pathogenic germline alterations; however, they did not statistically compare the distribution of KRAS WT tumors or pathogenic germline alterations in early- vs. late-onset pancreatic cancers. We noted that the presence of SMAD4 alterations had a significant adverse impact on OS regardless of presence or absence of KRAS mutations in patients. These results are interesting since SMAD4 loss in both in vitro and in vivo models of pancreatic cancer leads to increased tumor progression and metastatic potential^60,61, while prior studies focusing on resected patients alone have not found SMAD4 loss to be an independent predictor of patient outcomes^14,15. These incongruent results could be related to our larger cohort size or a differential impact of SMAD4 loss in localized vs. metastatic disease. Interestingly, alterations in BRAF or the MAPK pathway were not associated with OS in our KRAS WT cohort, whereas unexpectedly the presence of CDKN2A alterations was associated with an improvement in outcomes. The underlying reason for the discordant clinical impact of CDKN2A alterations between the KRAS WT and MUT cohorts is unclear and needs further validation in independent cohorts and additional mechanistic evaluation. In cases with CDKN2A deletion, the concomitant co-deletion of adjacent genes including MTAP, or a type 1 interferon gene cluster are known to impact tumor biology^62,63. Rates of genomic loss of these adjoining loci may be different between KRAS WT and MUT cohorts and also need further investigation. It is possible that the biologic impact of CDKN2A loss is modulated by concurrent activating mutations in KRAS.

Our study has several strengths, including careful clinical and genomic annotation from a large single institution database with relatively uniform treatment approaches and the ability to evaluate new therapeutic approaches in organoid models from patients with KRAS WT disease. However, this work also has several limitations. Given the high desmoplastic content and low tumor cellularity of some pancreatic ductal adenocarcinoma specimens, it is possible that we infrequently missed subclonal KRAS mutations, hence incorrectly classifying these as KRAS WT despite stringent QC metrics. Given deep sequencing to a depth of 312X mean target coverage depth for samples in this study, we are able to reproducibly detect KRAS mutations when present at 2% or greater allele fraction, thus, we feel confident that we were able to detect KRAS mutations when present in the vast majority of cases. Second, given the lack of concomitant transcriptional profiling in our cohort, we were unable to comment on variations in tumor cell states or potential microenvironmental differences between KRAS WT and mutant cases. A recent analysis of a cohort of KRAS WT pancreatic cancers profiled using a commercial platform showed similar genomic findings to our study but leveraged whole transcriptome profiling to suggest immune microenvironmental differences between the WT and mutant cohorts⁹. Another recent analysis of transcriptional profiling of a small number of KRAS WT pancreatic cancers surprisingly revealed transcriptional similarities to fusion-bearing KRAS WT cholangiocarcinomas⁶⁴. Hence, future efforts would benefit from combining cohorts with genomic, transcriptional, and possibly epigenetic profiling to achieve optimal understanding of the molecular basis and heterogeneity of KRAS WT pancreatic cancer.

In summary, we have characterized the clinical and genomic landscape of KRAS WT pancreatic cancer, defining two major categories of KRAS WT tumors, including those harboring an alternative MAPK driver and those without a MAPK driver (Figure 5E). These data present a framework for understanding precision oncology in KRAS WT pancreatic cancer and highlight potential genomically defined therapeutic avenues in these patients.

Statement of Translational Relevance.

Pancreatic cancers without KRAS mutations represent an important subset of this disease and often harbor alternative MAPK drivers which might be amenable to precision oncology approaches. However, the global clinical and genomic landscape of these tumors has not been deeply characterized. Using a large single institution case series, we describe the clinical and molecular features of 73 KRAS WT pancreatic cancers. Type 2 BRAF alterations and fusions in receptor tyrosine kinase associated genes occur frequently and 43.8% of all KRAS WT cases show evidence of an alternative MAPK driver. We provide preclinical evidence for dual RAF and MEK inhibition for tumors harboring BRAF in-frame deletions and demonstrate clinical therapeutic benefit by targeting ROS1 fusion in an index case. Our work provides foundational evidence and proof of concept pre-clinical studies for future clinical trials in this important subset of pancreatic cancer.