Introduction

Despite advances in targeted and combinatorial treatment approaches, colorectal cancer (CRC) remains a leading cause of cancer-related death [1,2]. Approximately 50% of patients develop metastatic colorectal cancer (mCRC) during the course of the disease, with a median overall survival (mOS) of approximately 30 months [3]. In mCRC patients, a combination of cytotoxic chemotherapy and targeted therapy remains the treatment backbone [4–6]. The mortality associated with CRC is related to the development of metastases⁷ and, later, duration of treatment response is affected by secondary treatment resistance by acquired drug resistance, particularly to targeted therapies such as anti-epidermal growth factor receptor (anti-EGFR) monoclonal antibodies [7–10].

Activation of EGFR primarily signals through the mitogen-activated protein kinase (MAPK) pathway, regulating cellular proliferation via the RAS–RAF–MEK–ERK cascade [11], but it also engages parallel signaling pathways such as the PI3K–AKT pathway, which contributes to cell survival and resistance mechanisms [12].

Most studies have indicated that resistance to anti-EGFR therapy is predominantly associated with mutations affecting components of the MAPK pathway, particularly in KRAS, NRAS, and BRAF [13]. However, it should be noted that many of these findings have been derived from analyses of initial responders who developed secondary disease progression under anti-EGFR monotherapy without concomitant chemotherapy [14,15]. In addition to direct alterations within the MAPK pathway, resistance can arise through activation of bypass signaling pathways that restore downstream MAPK signaling despite EGFR inhibition. These include upstream activation of alternative receptor tyrosine kinases such as ERBB2, MET, STAT3, YAP and FGFR, as well as ligand-mediated activation of parallel pathways [16]. Furthermore, alterations in the PI3K–AKT pathway and other compensatory signaling networks may contribute to resistance by promoting cell survival independently of EGFR blockade [17]. Together, these well-established resistance pathways highlight the complexity of resistance biology and underscore the need for comprehensive molecular profiling.

Currently, two anti-EGFR monoclonal antibodies are approved for clinical use in metastatic colorectal cancer (mCRC): the chimeric IgG1 monoclonal antibody cetuximab and the fully human IgG2 monoclonal antibody panitumumab. Both agents bind to the extracellular domain of the epidermal growth factor receptor (EGFR), thereby preventing ligand binding, receptor dimerization, and downstream signaling activation [18].

Anti-EGFR antibodies have demonstrated clinical efficacy across different treatment lines in mCRC, either as monotherapy or in combination with cytotoxic chemotherapy [18]. Analyses of major randomized clinical trials have consistently shown that their benefit is largely restricted to patients with left-sided primary tumors and tumors lacking activating alterations in the RAS signaling pathway [19]. In the presence of RAS mutations, constitutive downstream pathway activation renders EGFR inhibition ineffective, thereby explaining primary resistance in these patients.

Accordingly, current international guidelines [4,5]recommend anti-EGFR antibodies in combination with a doublet chemotherapy backbone as a preferred first-line treatment option for patients with left-sided, RAS wild-type mCRC, or in second line if they have been indicated in the first line but have not been used yet.

More recently, data from the PARADIGM study [20] have suggested that molecular “negative hyperselection” using circulating tumor DNA (ctDNA) to exclude anti-EGFR resistance-associated alterations may further refine patient selection [21]. This strategy may help identify a subset of patients with right-sided tumors who could potentially benefit from anti-EGFR therapy; however, this approach has not yet been established as a clinical standard.

Current commonly available mutational profiles from tumor tissues, obtained through either resection or biopsy, are usually biased towards untreated and mostly primary CRC. In pretreated patients, profiles are usually taken after progressive disease has become clinically apparent, with the emergence of one or more therapy-resistant clones. This is due to the need for sufficient, high-quality tumor material for next-generation sequencing (NGS)-based analysis, and to avoid unnecessary invasive biopsies without clinical implications [22,23]. Therefore, especially subclonal mutations fostering partial resistance to therapy in well responding patients and thereby limiting further tumor shrinkage in order to achieve resectability of colorectal cancer liver metastasis are underrepresented in the majority of previously published datasets. In line, sequential analysis of cell-free tumor DNA (cfDNA) has indicated subclonal mutations in the epidermal growth factor receptor (EGFR), which most likely occur as a response to selective anticancer treatment pressure from anti-EGFR directed medications [15]. On average, five resistance mutations bypassing the need of EGFR signaling were found in cfDNA analysis. Nevertheless, analysis of cfDNA is skewed towards dominant clonal populations and therefore underrepresents the low number of surviving remaining cancer cells after successful pretreatment of colorectal cancer liver metastasis before resection. Due to their low tumor burden and functionally inactive state, they often lack cfDNA shedding [24].

Little is known about mutational patterns in patients with colorectal cancer liver metastasis after anti-EGFR plus chemotherapy pretreatment. Therefore, we performed a targeted panel-based NGS of several colorectal cancer liver metastasis samples, pretreated with chemotherapy plus anti-EGFR therapy.

Results

Targeted next generation sequencing (NGS) was successfully conducted in 29 out of 30 previously published [25] colorectal cancer liver metastases samples (Fig 1, The list of panel gene targets can be found in the methods section). 9/29 (31%) and 20/29 (69%) of samples were collected pre-treatment/untreated and post-treatment, respectively. Treated patients received chemotherapy plus anti-EGFR agents in a neoadjuvant treatment approach between January 2008 and February 2014 before RAS mutations other than KRAS were identified as a predictive factor for anti-EGFR treatment (see Table 1 for a summary of previously published patient demographics and treatments). Median time from the start of systemic treatment to resection, was 7.6 months (range 4.06–11.15 months). 19.9% and 76.7% had right and left sided primary tumors, respectively.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Basic Patient Characteristics.

https://doi.org/10.1371/journal.pone.0351243.t001

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1.

Consort Diagram and Patient Flow: Study design, patient selection, baseline characteristics, and reasons for exclusion are illustrated. Treatment response to systemic therapy consisting of anti-EGFR antibodies in combination with chemotherapy is shown and categorized as tumor shrinkage (partial response – high, partial response – low) or poor response (stable disease, progressive disease). Surgical resection status for colorectal cancer liver metastasis following systemic therapy are additionally depicted.

Abbreviations: KRAS, Kirsten rat sarcoma viral oncogene homolog; NGS, next-generation sequencing.

https://doi.org/10.1371/journal.pone.0351243.g001

A total of 145 mutations were identified (mean, 5.2 ± 1.3 mutations per sample; range, 0–28; median, 2; IQR, 3.75). For the purpose of this analysis, only mutations with a variant allele frequency (VAF) greater than 3% were considered, ensuring that subclonal variants included in the study were reliably detectable. Mutation count did not correlate with the achievement of resectability (data not shown). NGS confirmed prior identified KRAS mutations in 4/29 patients (14.8%) and revealed further somatic alterations in RAS in 5/19 post-treatment samples. Surprisingly, post treatment/acquired RAS alterations did not influence successful R0 resection (4/5, 80%, all RAS mutant patients 4/9, 44.4%) or tumor shrinkage (Fig 2) in this highly preselected cohort, although none of the pretreatment KRAS mutated patients achieved R0 resection.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Patient RAS Status.

RAS status was evaluated at three distinct time points.The first status was based on patient history, based on different modes of analysis. The second extended RAS analysis was performed within the previously published study [25] in a standardized approach. The last analysis was based on the results of targeted NGS presented in this study with an allel frequency >3%.

Abbreviations: Rat sarcoma = RAS, KRAS = Kirsten rat sarcoma, HRAS = Harvey rat sarcoma, NGS = Next generation sequencing.

https://doi.org/10.1371/journal.pone.0351243.g002

We further analyzed the publicly available The Cancer Genome Atlas (TCGA) – Colon Adenocarcinoma (COAD) cohort of the TCGA Pan-Cancer Atlas, which mostly consists of primary CRC samples and the liver metastasis samples from the MSK- Metastastic colorectal cancer cohort and compared mutation frequencies to our study. Somatic mutations (for a complete list of gene abbreviations and their full names, please refer to the table in the Methods section) in APC and TP53 were identified in 68.9% (20/29) and 86.2% (25/29) of cases, respectively (Fig 3). Commonly mutated genes such as PIK3CA (24.1%, 7/29), SMAD4 (20.6%, 6/29), ATM (17.2%, 5/29) and FBXW7 (13.7%, 4/29) did not significantly differ in frequency between our cohort and publicly available data (Fig 3). Intriguingly, we found a higher frequency of clonal and subclonal mutations in HNF1A (34.4%, 10/29), FGFR2 (31%, 9/29), VHL (27.5%, 8/29), PDGFR (27.5%, 8/29), ERBB2 (24.1%, 7/29), ABL1 (24.1%, 7/29), SMO (24.1%, 7/29), GNA11 (20.6%, 6/29), RET (20.6%, 6/29), STK11 (20.6%, 6/29), HRAS (20.6%, 6/29) and NRAS (17.2%, 5/29), especially in comparison to the mutational profile of colorectal cancer liver metastases (Fig 3 and Fig 4).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Missense Mutations in Colorectal Cancer Liver Metastasis.

Panel-based NGS was performed. Samples with at least one missense mutation with an allele frequency (AF) >3% were counted and are displayed as a percentage of total samples (n = 29). Frequencies are compared to publicly available MSK-IMPACT Clinical Sequencing Cohort (MSK, n = 428, only colorectal cancer liver metastasis samples by filter), and The Cancer Genome Atlas- Colon Adenocarcinoma cohort (TCGA-COAD n = 461, mostly primary tumors (23% of samples without reported sampling side). For abbreviations of genes please see the table in the methods section.

https://doi.org/10.1371/journal.pone.0351243.g003

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Variant Allele Frequency of Frequently Altered Genes.

Bars are indicating the median of allele frequency for all samples with allele frequency above 3% (technical threshold), with error bars indicating interquartile range (IQR). For full gene names, please refer to the table in the methods section. Left of the vertical line are the well described clonal colorectal cancer driver mutations, without relevant differences to public databases. Right of the vertical line are genes which were detected in a higher frequency in SNVs in our dataset in contrast to public databases.

https://doi.org/10.1371/journal.pone.0351243.g004

Next, we investigated if some of these more frequently observed mutations showed differential expression depending on therapy response (patients with a “partial response high” (PR-high, more than 50% tumor shrinkage according to RECIST) or “partial response low” (PR-low, 30–50% tumor shrinkage) compared to non-responders (stable disease (SD) and progressive disease (PD)) in patients with available sequential CT scans for response evaluation (27/29). While HNF1A, FGFR2, PDGFR, ABL1, SMO, RET and STK11 did not show a difference in mutation frequency between responders and non-responders (Table 2), VHL (+33% in responders, 7/21 vs. 0/6), ERBB2 (+28.6% in responders, 6/21 vs. 0/6) and GNA11 (+19% in responders, 4/21 vs. 0/6) showed mutations exclusively in responding patients. Successful R0 resection was performed in 14/21 (66,6%) patients with overall therapy response versus 2/6 (33,3%) in non-responders. Further, we analysed if the well-responding patients shared unique mutations not present in the non-responders. Here, 26 mutations were exclusively observed in responders, and 17/26 were exclusively observed in PR-high patients (Fig 5 and Table 3). Nevertheless, we did not observe any significant difference between mutational status and successful R0 resection of colorectal cancer liver metastasis (Table 4).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Mutations in Treatment Responders and Non-Responders.

https://doi.org/10.1371/journal.pone.0351243.t002

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Overlapping and Unique Mutations in Different Treatment Response Groups.

https://doi.org/10.1371/journal.pone.0351243.t003

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Mutations and resection status of colorectal cancer liver metastasis.

https://doi.org/10.1371/journal.pone.0351243.t004

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Venn Diagram: Mutations in Different Treatment Response Groups.

Mutations were grouped based on their presence in the different treatment response groups and overlap is displayed in form of a venn diagram. Gene names according to the numbers are provided in Table 4. Abbreviations: PR-High = Partial Response-High (>50% tumor shrinkage according to RECIST), PR-Low = Partial Response-Low (>30% < 50% tumor shrinkage according to RECIST), SD = Stable Disease.

https://doi.org/10.1371/journal.pone.0351243.g005

Finally, to see if some of the mutations were related to previous treatment with chemotherapy and anti-EGFR monoclonal antibodies, we investigated whether some mutations were more frequent in the post-treatment samples (Table 5). FGFR2 mutations were significantly more often observed in post-treatment samples compared to pre-treatment samples (p = 0.0168, 9/20 vs. 0/9). Numerically, mutations in VHL (+24%), ERBB2 (+19%), SMO (+19%), GNA11 (+19%), RET (+19%), STK11(+19%) and HRAS (+19%) were also more frequent in post-treatment cohort, although lacking statistical significance (Table 5).

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Mutations in Pre-treatment versus Post-Treatment samples.

https://doi.org/10.1371/journal.pone.0351243.t005

Discussion

Despite advances in multimodal therapy of microsatellite stable (MSS) stage IV colorectal cancer, the overall prognosis is still dismal [1,2]. The current standard of care is a combinatorial treatment approach consisting of chemotherapy (mostly FOLFOX or FOLFIRI doublet regimens) and monoclonal antibodies (targeting VEGF(R) or EGFR pathway) [4–6]. Anti-EGFR targeted therapy is active but acquired secondary therapy resistance usually occurs within 8–10 months [26–28]. Initially, resistance to anti-EGFR monotherapy has been explained by intrinsic mechanisms of resistance, mostly by mutations affecting the RAS/RAF/MEK and PI3K/AKT/mTOR pathway [29–31]. In recent years, additional genetic and non-genetic negative predictors of anti-EGFR therapy response have been discussed [32].

Therefore, in this single-center retrospective study, a total of 29 macrodissected tumor samples from patients with primarily unresectable synchronous or metachronous colorectal liver metastases who received neoadjuvant chemotherapy in combination with either cetuximab or panitumumab as a conversion strategy to surgical resection were analyzed with a targeted NGS panel. NGS confirmed prior identified KRAS mutations in 4/29 patient samples and revealed further somatic clonal and subclonal RAS alterations in five post-treatment samples. Noteworthy, the presence of acquired RAS alterations did not influence tumor response or surgical resection success. Further, we identified recurrent (clonal and subclonal) mutations in responding tumors with higher frequencies compared to published datasets of primary and metastatic CRC (Fig 3). Especially, mutations in VHL, ERBB2 and SMO were exclusively observed in well-responding patients, although the results lack statistical significance due to the small sample size and remain therefore descriptive. Further, mutations in FGFR2 were significantly associated with post-anti-EGFR treatment samples, indicating a higher frequency of this mutation as a response to treatment, although due to selection bias this remains speculative. Importantly, none of the mutations were predictive for successful R0 resection; therefore, no prediction for routine clinical decision making in predicting successful conversion therapy was possible based on genetic features within our cohort.

Although we cannot directly link single nucleotide variants to therapy resistance, the higher frequency of subclonal variants, which are also described to be involved in anti-EGFR antibody resistance in literature [33], in ERBB2 [34], ERBB4 [35], FGFR1–3 [36–38], EGFR [15,39] itself and PDGFR [40] with the potential downstream activation of ERK in responders and/or post treatment samples is indicating an association between therapy and alternative activation of MAPK signaling pathway in this tumor cell clones, especially as they are way less frequently reported in the large MSK and TCGA cohorts. In line, many of the observed alterations, based on the existing clinical and preclinical evidence of their role in anti-EGFR resistance, were included in the panels for negative hyperselection (i.e., PTEN [41], MET [34] and ALK [42]) of patients, which is suggested for identification of patients without mutations indicating anti-EGFR therapy resistance in circulating tumor DNA for treatment with anti-EGFR combination regimens [43].

To the best of our knowledge, single nucleotide variants of the von Hippel-Lindau (VHL) gene have not yet been implicated in resistance to anti-EGFR antibody treatment, although they have been described in subpopulations of primary CRC [44]. Consequently, they may represent a compelling target in this patient population, necessitating further research.

Historically, secondary mutations in the mitogen-activated protein (MAPK) pathway have been the primary factor in anti-EGFR therapy resistance in colorectal cancer or lung cancer [14,15]. Further, the development of treatment resistance has been mostly based on the model of therapy-resistant subclonal populations that ultimately outgrow and outnumber treatment-sensitive cells [14]. However, a recent study showed that this clonal progression of MAPK pathway mutations was present in 46% of patients with acquired resistance to anti-EGFR monotherapy (combination of third line clinical trials) compared to as few as 9.1% of patients treated with chemotherapy plus anti-EGFR antibodies in comparison to 5.7% in patient treated with chemotherapy alone [45], thus challenging the relevance of genetic resistance for patients treated in the first therapy line with combinatorial treatment approaches. Accordingly, we did not observe any effect of the observed mutations in the RAS pathway or with the potential to bypass EGFR signaling on treatment response or successful R0 resection. Further, it remains questionable if the subclonal mutations observed in this study are relevant for later therapy resistance or if they explain the survival of clones harboring those alterations. In line, a previous study reported that subclonal mutations rarely become clonal upon progression [45]. Several groups recently reported that transcriptomic mechanisms of resistance mediate cross therapy resistance, i.e., by epithelial-to-mesenchymal transition (EMT) to chemotherapy and anti-EGFR monoclonal antibodies, thereby being the major driver of resistance in this patient subgroup [45–47]. Therefore, the role of non-genetic resistance mechanisms is highlighted by the absence of relevant therapy resistance/response prediction in the patients with acquired RAS pathway alterations in our cohort.

Our study has several limitations. First, and most importantly, the small sample size and retrospective design limit the statistical power of the analysis. Consequently, the observed mutational patterns should be interpreted as descriptive and hypothesis-generating rather than confirmatory. Second, with the advances in whole-exome and whole-genome sequencing, the used NGS panel appears rather small, potentially missing additional interesting genomic alterations. As pre-treatment and post-treatment samples were derived from different patients, direct longitudinal comparisons were not possible. Therefore, the observed differences in mutational frequencies cannot be conclusively attributed to treatment exposure and may partly reflect inter-cohort biological variability. These findings should thus be interpreted as descriptive and hypothesis-generating. Last, this study is subject to selection bias due to the highly specific composition of the cohort. Only patients who were eligible for curative-intent liver surgery were included, either because of primary oligometastatic disease or due to a particularly favorable response to anti-EGFR-based conversion chemotherapy. Consequently, this population represents a biologically and clinically distinct subgroup of metastatic colorectal cancer patients with relatively favorable disease characteristics. Our findings may therefore not be generalizable to the broader metastatic colorectal cancer population, particularly to patients with unresectable or more aggressive disease. Nevertheless, this study contributes valuable insides into CRC behavior upon treatment with chemotherapy and anti-EGFR antibodies. Future studies in larger and more heterogeneous cohorts are required to validate these observations.

In conclusion, the presented study adds important data on clonal and subclonal mutations implicated in resistance to anti-EGFR monoclonal antibodies in patients with colorectal cancer liver metastasis successfully treated in the first line with chemotherapy and anti-EGFR antibodies. Additionally, the data highlight that no relevant conclusions can be drawn from genomic data for predicting clinical responses in this patient subgroup, and that the currently established molecular markers are sufficient for making decisions about first-line therapy. Therefore, further research focusing on non-genetic mechanisms of resistance, ideally using matched pre- and post-treatment samples, is urgently needed to enable more mCRC patients to undergo surgical resection with curative intention after neoadjuvant treatment and to identify prognostic markers for neoadjuvant treatment response.

Materials and methods

Targeted next generation sequencing

Sequencing libraries were prepared from tumor and non-tumor tissue with TruSeq® Amplicon – Cancer Panel (Illumina, San Diego, CA) according to the manufacturer’s instructions and paired-end sequencing was performed on a MiSeq (Illumina, San Diego, CA) with 2 × 150 base pairs (bp) read length.

Download:

• PNG
  larger image
• TIFF
  original image

https://doi.org/10.1371/journal.pone.0351243.t006

Supporting information

Acknowledgments

We thank all patients and their families, as well as the UCT Biobank for participating in this study. The results published in Fig 3 are in part based upon data generated by the TCGA Research Network: https://www.cancer.gov/tcga. Furthermore, the authors would like to thank Katrin Bankov for her methodological support.