Monitoring clonal evolution and resistance to EGFR blockade in the blood of metastatic colorectal cancer patients

Colorectal cancer tissue is used to define the molecular status of clinically relevant genes. For example, oncogenic mutations in KRAS and NRAS, which occur collectively in about 55% of the patients, are routinely assessed in tumor specimens as they predict lack of response to EGFR-_targeted monoclonal antibodies cetuximab and panitumumab². Mutations in BRAF are also associated with lack of response to EGFR blockade and frequently evaluated in metastatic CRC patients³. In colorectal tumors lacking RAS pathway mutations, HER2 or MET gene amplifications are being explored as biomarkers of response to drugs inhibiting the corresponding oncoproteins^4–6.

Tumor tissue genotyping has inherent limitations. It has been shown that the genomic profile of the primary tumor and the metastases are not always concordant due to the intrinsic molecular heterogeneity of the disease^1,7,8. Furthermore, commonly used chemotherapeutic agents as well as _targeted drugs can alter the tumor molecular landscape¹. To account for this, the genomic profiles of CRC patients should be evaluated repeatedly during the course of therapy. Several reasons prevent the reiterated use of tissue biopsies including the inherent risk of complications associated with these procedures 9. Moreover, biopsy samples are often limited and tissue processing might delay the initiation of treatment. To overcome these restrictions, we considered the analysis of circulating tumor DNA (ctDNA), a procedure known as liquid biopsy^10–15.

We assessed whether blood-based molecular profiles could be used to identify actionable _targets, to monitor drug resistance and to track tumor dynamics in CRC patients. We started by evaluating whether the genotype of RAS pathway genes of clinical relevance could be determined in circulating tumor DNA. To analyze ctDNA, we exploited Droplet Digital PCR (ddPCR)^16,17 and BEAMing¹⁸. Both technologies are based on micro compartmentalization of the PCR reaction and can detect mutant alleles with high sensitivity (0.01 to 0.001%)¹⁹.

We selected a total of 100 CRC cases (Supplementary Table 1 and Supplementary Figure 1) for which we collected matched pairs of tissue and blood samples. Droplet Digital PCR (ddPCR) was used to interrogate the mutational status of KRAS, BRAF and NRAS in plasma samples without prior knowledge of the tissue genotype. In 97/100 cases (97%) determinations of ‘RAS pathway mutations’ were concordant between tissue and blood (Supplementary Figure 1a and Supplementary Table 2). Two of the discordant cases (ONCG-CRC21 and ONCG-CRC55) were from patients with low tumor burden (lymph nodes-limited disease or lesions <1.5 cm in size) as assessed by imaging (data not shown).

In 8 cases, plasma analysis revealed mutations which were not detected in the matched tissue, thus denoting that the blood more comprehensively captures intra-patient disease heterogeneity (Supplementary Figure 1a and Supplementary Table 2). The mutational status did not correlate with the levels of circulating free DNA (genome equivalents, GE) indicating that the genotype does not overtly affect release of tumor DNA in the blood (Supplementary Figure 1b).

Presently, even when KRAS and NRAS mutant CRCs are excluded from treatment, only 20% of the patients benefit from monotherapy with EGFR _targeted agents^20,21. Using a cohort of 10 patients who received anti-EGFR antibodies regimens without achieving clinical benefit, we asked whether the molecular bases of primary resistance could be ascertained in blood. We started by determining the mutational status of the RAS pathway in ctDNA using the approach described above. As expected, KRAS mutations were absent in the entire cohort. Two patients (MOLI-CRC02 and MOLI-CRC06) displayed NRAS mutations (Figure 1a and Supplementary Table 4a). Scrutiny of the clinical history of these patients revealed that anti-EGFR treatment had been initiated before the introduction of NRAS testing in this clinical setting², thus explaining the findings.

We next sought to systematically identify genetic mechanisms of primary resistance to anti-EGFR treatment in the 8 patients in which the ‘RAS pathway’ was confirmed wild-type in plasma. To this aim, we analyzed the coding region of 226 genes, which we selected as being involved in colorectal cancer progression, oncogenic signaling and sensitivity or resistance to therapy (Supplementary Table 3). The approach relies on massive parallel next generation sequencing (NGS) of genomic DNA, which we adapted to capture and sequence circulating tumor DNA. To confirm that the NGS-blood analysis could identify tumor exclusive (somatic) genetic alterations, we determined the mutational status of APC and TP53, two genes frequently mutated in CRC. The analysis identified APC and/or TP53 mutations in circulating DNA, which were not present in the patients’ germ-line DNA, (Supplementary Table 5).

The results of the analysis of the 226 genes were noteworthy: four cases (ONCGH-CRC01, ONCGH-CRC11, ONCGH-CRC06 and MOLI-CRC15) harbored ERBB2 amplification (Figure 1a and Supplementary Table 4b). We previously found that ERBB2 amplification is present in approximately 3% of CRC and correlate with lack of response to EGFR blockade⁵. These findings therefore validate the NGS-based experimental approach and increased our confidence in profiling plasma DNA to nominate molecular mechanisms of drug resistance. One of the ERBB2 positive samples also harbored FLT3 amplification; the latter has been previously detected in 3% of CRC (Kopetz, J Clin Oncol 32:5s, 2014; suppl; abstr 3509; www.cbioportal.org, TCGA Nature 2012). In one patient (ONCG-CRC13) we detected a somatic variant of the MAP2K1 gene (p.K57N) encoding the MEK1 protein (Figure 1a and Supplementary Table 4a). MAP2K1 mutations are present in 1.5% of colorectal tumors (www.cbioportal.org, TCGA Nature 2012) but have not yet been associated with resistance to EGFR blockade.

We next asked whether analysis of plasma DNA could be used to unveil mechanisms of acquired (secondary) resistance to cetuximab or panitumumab. To answer this question, we studied the blood of 16 patients who responded and then progressed upon anti-EGFR therapy (Figure 1b, Supplementary Table 4a and 4b). We and others previously reported that acquired resistance to anti-EGFR antibodies is associated with the emergence of RAS pathway mutations and that these mutations can be detected in the blood before disease progression is clinically manifest^15,22–24. Accordingly, in most of the patients (11/16), KRAS mutated alleles were detected in blood samples obtained at progression (Figure 1b and Supplementary Table 4a). We also detected EGFR ectodomain mutations (in 2 cases) and KRAS or MET gene amplification in four instances (Figure 1b and Supplementary Table 4b). We previously reported that amplification of KRAS and MET genes, as well as mutations in the extracellular domain of EGFR, sustain resistance to anti-EGFR antibodies in CRC patients^4,24–26.

The clinical management of patients who develop RAS pathway mutations, as a consequence of EGFR blockade, remains challenging. Most often these individuals receive additional lines of therapies based on chemotherapeutic agents alone or combined with anti-angiogenic drugs, and in some cases also monotherapy with the multikinase inhibitor regorafenib. It is unknown whether and how subsequent therapies affect RAS (resistant) clones, which were selected during treatment with anti-EGFR antibodies.

To address this issue we monitored the behavior of KRAS mutations and MET amplification during therapeutic schemes that are commonly administered at relapse after EGFR treatment. Resistant clones were tracked in the blood of patients who initially achieved benefit and then experienced disease progression after treatment with an anti-EGFR antibody, either in first-line in combination with chemotherapy or as monotherapy in third-line. For 5 patients who developed KRAS mutations and for one patient who acquired MET amplification, samples were available at baseline, at the time of first disease progression and at different time-points across subsequent lines of treatment (Figure 2 and Supplementary Table 6a).

The analysis revealed that KRAS mutant alleles, which emerge at the time of disease progression, decline when treatment with anti-EGFR antibodies is suspended and remain below the limit of detection across subsequent lines of treatment (Figure 2). These findings led us postulate that clonal evolution of tumor cell populations that survive treatment with anti-EGFR antibodies continues beyond the point of clinically-established resistance.

To mechanistically evaluate this possibility, we studied populations of CRC cells (DiFi) in which acquisition of resistance to cetuximab is accompanied by amplification of the KRAS gene¹⁴. To parallel withdrawal of EGFR blockade, which occurs when patients develop cetuximab resistance, two populations of DiFi resistant cells were cultured for 160 days in the absence of the anti-EGFR antibody (Figure 3a). Analogously to what we had observed in the blood of patients, KRAS declined significantly in both cell models (Student’s t-Test p value <0.01) when the anti-EGFR antibody was suspended (Figure 2 and Figure 3a, top panel). The cell populations, which experienced antibody withdrawal, regained partial sensitivity to cetuximab (Student’s t-Test p value ≤0.001) as compared to the population in which the drug pressure was maintained (Figure 3a, bottom panel and Supplementary Figure 2).

The decline of KRAS alleles we observed in blood upon withdrawal of EGFR blockade and the cell based functional experiments suggest that re-challenge therapies may be worthy for CRC patients who, upon relapse, suspend EGFR-_targeted agents. The scrutiny of our clinical databases retrieved 2 patients who responded to anti-EGFR antibodies and upon relapse received additional lines of therapy before receiving again anti-EGFR treatment with cetuximab or panitumumab (Figure 3b and 3c). Notably, re-challenge with EGFR blockade was clinically effective and triggered partial responses or long lasting stabilization (Figure 3b and 3c). This is consistent with literature data showing that CRC patients can respond to anti-EGFR re-challenge²⁷.

We next asked whether and to what extent intermittent treatment with anti-EGFR antibodies affected the genomic status of CRCs. To address this question we studied three additional patients who received multiple rounds of anti-EGFR antibodies and for which longitudinal tissue or plasma samples where available (Figure 4).

In the first patient (HMAR-CRC08) tumor tissue was available before the first administration of anti-EGFR, as well as at progression after a partial response of nearly 8 months. KRAS p.G12V was detected at resistance and declined when cetuximab plus irinotecan treatment was suspended, as assessed by plasma analysis (Figure 4a). The patient then received standard chemotherapy and when a new progression was detected, was re-challenged with cetuximab alone, achieving again a partial response, which lasted 18 months.

In the second patient (ONCG-CRC69) clonal tracking revealed that during the first round of EGFR blockade with cetuximab a KRAS p.G13D mutant clone raised in blood and declined when treatment was suspended as the patient refused further therapy (Figure 4b and Supplementary Table 6b). Subsequently, at disease progression, the patient received panitumumab-achieving SD with tumor shrinkage, which lasted for nearly six months (Figure 4b). Notably, during the second round of EGFR blockade, the KRAS p.G13D allele increased again in blood and was accompanied by the emergence of another KRAS mutant clone (p.G12V).

The third patient (AOUP-CRC05) developed a KRAS p.G12D clonal variant during the initial challenge with panitumumab which translated into a partial response. Upon progression the patient received several salvage therapies during which a single plasma samples was collected and again displayed a decline of the KRAS allele that emerged through the initial EGFR blockade. More than a year from the initial response to the anti-EGFR antibody the patient was treated again with panitumumab achieving again a partial response. During panitumumab re-challenge the levels of the KRAS p.G12D raised again in plasma (Figure 4c and Supplementary Table 6b).

Variations of KRAS clones detected in blood and cell lines suggest that colorectal cancers treated with anti-EGFR antibodies display remarkable plasticity. This likely reflects dynamic clonal competition and provides molecular evidence that re-challenge with EGFR blockade after a withdrawal period is effective as it exerts genomic selection in metastatic colorectal cancers.

Currently, CRC patients are treated according to mutational landscapes ascertained using a small fragment of the primary tumor or an individual metastatic lesion. Essentially, a single molecular snapshot is taken for each patient and based on this knowledge several rounds of therapy are administered, often over many years. Rather than being based on out-of-date ‘molecular’ pictures, each round of therapy should ideally be instituted on a recent genetic profile as comprehensive as possible of the overall disease.

In this work we evaluated whether ctDNA can be used throughout the clinical management of CRC patients to gather real time updates on the molecular landscape of the disease. We find that mutational profiles determined in blood and tissues are highly concordant. In a relevant fraction of the cases (8%), analysis of circulating DNA identified KRAS or NRAS mutated alleles, which were not detected in tissue samples, underscoring the ability of liquid biopsies to better capture the heterogeneity of the overall disease. Our findings suggest that blood-based tests may more comprehensively interrogate the ‘RAS pathway’ mutational status than tissue-derived genotypes.

Only a small fraction (20%) of patients who receive anti-EGFR therapies benefit from treatment. The molecular bases of the lack of response in the remaining patients are largely unknown. We find that ctDNA analyses can often identify genetic alterations likely responsible for resistance to EGFR blockade including amplification of the HER2 and MET genes or mutations in MAP2K1. We also observe that CRC patients refractory to anti-EGFR antibodies, who do not carry RAS pathway mutations, frequently display clinically-actionable oncogenic events. This approach may therefore be used to rapidly identify rare population of patients likely to respond to _targeted agents.

In accordance with our previous findings and other reports, KRAS or NRAS mutated clones often emerge in the blood of CRC patients who respond to cetuximab or panitumumab therapies^14,28–30. At present, these individuals discontinue anti-EGFR antibodies treatment and receive salvage therapies based mainly on chemotherapeutic agents. We discovered that KRAS mutant clones decline in blood when anti-EGFR antibodies are withdrawn, suggesting that resistant cell populations are highly dynamic. To functionally substantiate these discoveries, we examined preclinical models of acquired resistance to cetuximab. Analogously to what observed in patients, we noted that KRAS clones decay in resistant populations upon antibody suspension.

These results, together with the clinical cases in which we documented effectiveness of re-challenge with anti-EGFR antibodies paralleled by rise and fall of KRAS mutant clones, provide the rationale for adaptive therapy strategies. In this regard discontinuous dosing strategy with BRAF inhibitors has been successfully attempted in preclinical models of BRAF driven melanomas³¹. Furthermore there is evidence that re-challenge with _targeted therapies can be effective in patients with different tumors types^27,32–34.

Our data will likely trigger further study to assess whether CRC cells who developed resistance to EGFR may display fitness disadvantage in the absence of the drug and whether this can be exploited to forestall the onset of lethal drug-resistant disease. Drug schedules, aimed at maintaining a stable population of drug-sensitive cells to suppress the growth of resistant clones through intratumoral competition, could be explored in ad-hoc clinical trials. Accordingly, we plan to investigate the re-introduction of anti-EGFR antibodies in patients who achieve benefit from cetuximab or panitumumab and subsequently display a decline of KRAS mutant clones in circulating tumor DNA.

In conclusion, our data indicate that blood, rather than tissue, can be employed to monitor -virtually continuously- the molecular evolution of metastatic colorectal tumors. The findings that the genome of CRC adapts dynamically to pulsatile drug schedules provide rationale for additional lines of therapy for patients who benefit from an initial challenge with anti EGFR antibodies.