Genetic evolution to tyrosine kinase inhibitory therapy in patients with EGFR-mutated non-small-cell lung cancer

Abstract

Background

Tumour heterogeneity impacts the efficacy of metastatic cancer treatment even if actionable mutations are identified. Clinicians need to understand if assessing one lesion provides reliable information to drive a therapeutic decision in non-small-cell lung cancer (NSCLC) patients.

Methods

We analysed inter-tumour heterogeneity from five autopsied individuals with NSCLC-harbouring mutations in the epidermal growth factor receptor (EGFR), treated with EGFR tyrosine kinase inhibitors (TKIs). Through a comprehensive next-generation sequencing (NGS) oncopanel, and an EGFR panel for digital droplet PCR (ddPCR), we compared metastases within individuals, longitudinal biopsies from the same lesions and, whenever possible, the primary naive tumour.

Results

Analysis of 22 necropsies from five patients revealed homogeneity in pathogenic mutations and TKI-resistance mechanisms within each patient in four of them. In-depth analysis by whole-exome sequencing from patient 1 confirmed homogeneity in clonal mutations, but heterogeneity in passenger subclonal alterations. Different resistance mechanisms were detected depending on the patient and line of treatment. Three patients treated with a c-MET inhibitor in combination with TKI lost MET amplification upon progression.

Conclusion

At a given point and under selective TKI pressure, a single metastasis biopsy in disseminated tumours from EGFR-mutated NSCLC patients could provide a reasonable assessment of actionable alterations useful for therapeutic decisions.

Background

Improvement of diagnostic tools and a better understanding of the molecular pathways underlying tumour formation and progression have significantly contributed to the design and development of targeted cancer therapies. In lung cancer, patient habits, histology, receptor tyrosine kinases (RTKs), tumour microenvironment, expression of genes and specific mutations are used as prognostic and predictive factors of response to a repertoire of therapeutic options [1–4]. Complemented with research autopsies (use of necropsies for medical research purposes), they provide a better understanding of tumour evolution and resistance mechanisms, giving new insights for designing tailored therapeutic approaches [5].

In non-small cell lung cancer (NSCLC), performing biopsies of overt metastases is often limited to few locations and usually not well perceived by clinicians. Moreover, the limited amount of material obtained represents an important caveat for standard molecular testing, potentially affecting result consistency and reproducibility. DNA sequencing through whole-exome sequencing (WES) or targeted panels, and ddPCR using solid or liquid biopsies, are emerging as powerful diagnostic tools aiding treatment decisions. These techniques require low sample amounts and provide high-resolution results with a quick turnover, but their routine application to cancer management is not yet fully established [6–8].

Most importantly, the higher sensitivity and specificity achieved by these techniques opens the door to properly characterising tumour heterogeneity, identifying subclonal tumour populations and following their evolution, representing a clear benefit for personalising the treatment not only at the moment of diagnosis but while monitoring the course of disease [9, 10].

In this report, we perform post-mortem multi-lesion sampling of five patients with EGFR-mutated NSCLC to characterise the genomic complexity of the disease upon progression (PD) to EGFR-TKI, and provide insights into the role of driver mutations and acquired resistance to pharmacological treatments.

Patients and methods

Patients

The study includes a cohort of five autopsied patients (Supplemental Table 1) diagnosed with metastatic lung adenocarcinoma carrying EGFR mutations and treated at least in first line with first-generation EGFR-TKI inhibitors (gefitinib or erlotinib), along with other lines of therapy.  The study was approved by the HUVH Ethics Committee of Clinical Research according to local guidelines and regulations with the reference PR(AG)494/2019. Informed written consent was obtained from all the patients, or their relatives, included in this study.

Methods

A total of 34 formalin-fixed paraffin-embedded (FFPE) tumour samples derived from five patients and evaluated by a pathologist (Supplemental Table 1) were used. In all, 25/34 showing >10% of tumour area were genotyped by NGS using the VHIO-300 capture panel (432 genes, Supplemental Table 2). Reliable copy number alterations (CNAs) were reported only for samples with at least 30% tumour area. Due to its higher sensitivity, ddPCR was carried on for all available necropsies independently of the tumour area. Healthy kidney tissue was used as a control for mutational noise resulting from the formalin fixation procedure. Samples from patient 1 were further analysed by WES, using autologous healthy kidney tissue for germline variant clean-up. For a full description of the procedures, please refer to Supplemental Methods.

Results

Heterogeneity in mechanisms of resistance to TKI therapy among different individuals

Research autopsy has been previously used to analyse tumour evolution in the context of selective pressure imposed by pharmacological treatment [5]. In this study, we assessed whether resistance mechanisms were conserved between five different patients presenting similar carcinomas, all diagnosed as EGFR mutant by Cobas assay (Roche Molecular Systems), and negative for TKI-resistant EGFR variants (T790M or insertions in exon 20).

All patients were treated with a first-generation EGFR-TKI inhibitor after diagnosis (Fig. 1). Upon PD to first line, four patients presented a MET amplification, and three of them presented more than one mechanism of resistance (Fig. 2 and Supplemental Document 1). In detail, patient 1 presented a KRAS amplification detected by NGS and a MET amplification (only detected by IHC). Patient 2 presented a MET and an EGFR amplification, as well as an EGFR T790M mutation. In patient 3, MET amplification was the sole resistance mechanism detected. We found no resistance mechanism in patient 4, likely because EGFR L861Q mutation is sensitive to TKIs but to a minor degree compared to classical mutations [11, 12]. Instead, the tumour adopted a CTNNB1 G34R pathogenic variant in all metastatic sites analysed at the time of death. Patient 5 harboured both, MET and EGFR amplification.

Fig. 1. Tumour evolution and clinical history for each patient.

Genotyping of multiple metastatic lesions from five EGFR-mutated NSCLC patients along the evolution of the metastatic disease. Panels on the right display the CT-scan measurement of tumour lesions, their evolution in size, treatments and sample collection times.

Fig. 2. Homogeneous driver genomic alterations in necropsies of EGFR-mutated patients.

Genomic analysis of the primary, metastasis and necropsies available for NGS analysis from all patients. Heatmap of the non-silent variants and CNAs from each case; colour key for MAFs and CNA is shown at the bottom, shape of the colour illustrates the method of detection (VHIO-300 panel by NGS or ddPCR). Important to note, all variants under 3% of MAF and detected by NGS have been added manually after bam file manual review.

As a second-line therapy, patients 1, 3 and 5 were enrolled in the same clinical trial and received gefitinib combined with the c-MET inhibitor capmatinib (Fig. 1 and Supplemental Document 1). All of them progressed but MET amplification was no longer detected. Instead, patient 3 harboured the resistance mutation T790M in EGFR (Fig. 2a), patient 5 combined the de novo T790M variant coexisting with an EGFR copy number gain.

Patient 1 showed new pathogenic mutations like FBXW7 S560X and NFE2L2 E79K, among other variants of unknown significance (VUS). Patient 2 did not acquire a new variant related to resistance to the second-line therapy with osimertinib. Patient 4 did not receive a second-line treatment (Figs. 1 and 2a).

In summary, we detect heterogeneity in resistance mechanisms to TKI therapies in EGFR mutant NSCLC, being MET amplification and EGFR T790M the most prevailing ones. Nevertheless, such mechanisms occurred in combination with others in three out of five of the first-line therapies with erlotinib or gefitinib.

Temporal heterogeneity in an individual

We assessed temporal heterogeneity during the course of disease in patients 1 to 4. Phylogenetic trees were generated with the available biopsies from different time points as well as the necropsies. Patient 5 was excluded from the analysis due to the only availability of a single necropsy for NGS profiling and no available primary tissue.

For patient 1, genetic profile history spans 8 years from the primary tumour to the autopsy, during which the patient received treatments of surgery + adjuvant chemotherapy, radiosurgery in the brain, erlotinib, gefitinib + capmatinib, chemotherapy, a second regimen with erlotinib, chemotherapy, and finally nivolumab (see Fig. 1 and Supplemental Document 1). Temporal genetic evolution is depicted in Fig. 3. The normal tissue of this patient revealed a likely pathogenic germline mutation in FANCA (E383fs), confirmed by WES of the normal tissue. The lung primary tumour harboured a driver EGFR L858R. Three mutations in the first metastatic branch were identified. Two of them are considered likely pathogenic, FBXW7 S560X and NFE2L2 E79K. FBXW7 is a critical tumour suppressor and a very frequently deregulated ubiquitin-proteasome related protein in cancer. Normally, FBXW7 protein controls proteasome-mediated degradation of oncoproteins such as cyclin E, c-Myc, Mcl-1, mTOR, Jun, Notch and AURKA [13], but its deletion contributes to gefitinib resistance [14] and other chemosensitive alterations [15, 16]. NFE2L2 encodes for Nrf2, a transcription factor involved in oxidative stress and frequently mutated in lung cancer. Nrf2 E79 residue is a recurrent hotspot for gain-of-function mutations that disrupt its binding ability to KEAP1, leading to Nrf2 nuclear translocation and increasing the expression of free radical defence genes which could interfere with radiation-induced DNA damage [17] and to mediate radiation resistance in vitro [18, 19] and in vivo [20]. A third mutation in PLCG2 was classified as VUS. In addition, malignant CNAs were found, such as amplification of KRAS, MDM2, FRS2 and CDH4, and loss of CDKN2A/B-AS1. In summary, in the brain metastatic lesions of patient 1, detected 4 years after initial diagnosis, homogeneity of the driver mutations was maintained.

Fig. 3. Temporal heterogeneity of the disease and selective pressure of the applied therapies.

Phylogenetic trees were constructed with the primary, the metastasis and necropsies analysed by VHIO-300 panel and using the maximum parsimony method. For Patient 1, the trunks of the tree were rooted by a germline DNA sequence that did have a germline mutation depicted with (*). Trunk, branch and sub-branch lengths are proportional to the number of mutations. Dashed lines represent the most plausible phylogenetic construction upon the absence of primary tissue.

A second metastatic branch was depicted in the tree, which probably developed after treatment with gefitinib + capmatinib and/or chemotherapy, showing ATRX R840fs, CDC73 Q490E and NOTCH2 R452C mutations. The selective constraints of successive targeted treatments may have triggered new pathogenic mutations at PD, such as ATRX R840fs, rendering tumour clones with a more aggressive phenotype. This would correlate with the acceleration of tumour growth in the last 2 years of PD.

In the absence of a primary tissue to be analysed by NGS, phylogenetic trees of patients 2 and 3 were built according to the most plausible clonal evolution, counting on the EGFR mutational analysis data from the pathologists and the allelic frequencies from the liver metastasis reported by the VHIO-300 panel. Patient 2 debuted with the EGFR indel in the exon 19 variant, and likely with a TP53 and RBM10 truncating mutations, since their mutant allelic fraction (MAF) is close to the EGFR sensitising MAF. After PD to gefitinib + olaparib, the first metastatic branch presented a T790M-resistance mutation, and SOX9 and CSFR1 mutations classified as VUS were harboured at a MAF similar to the T790M variant in the liver metastasis. After 29 months of osimertinib treatment, the patient developed a second branch acquiring GNAQ mutation, present in both necropsies. From there, private mutations appeared in each necropsy. Interestingly, this sole example of private mutations in different lesions at the time of autopsy correlates with the only patient under oligoprogression, with active disease restricted to the liver.

Patient 3 was diagnosed with advanced lung adenocarcinoma, the EGFR L858R mutation was confirmed by cobas, and we postulate that the TP53 pathogenic deletion and the ABL2 variant considered as VUS, were present at diagnosis due to the similarity in allelic frequencies with the driver EGFR variant. After PD to gefitinib, the patient revealed a T790M mutation in EGFR, only tested after death and for the purpose of this study. Subsequent lines of treatment were avelumab plus crizotinib for 4 months and chemotherapy for 5 months, but her condition deteriorated rapidly and died. The NGS analysis of the necropsies did not reveal additional mutations.

In contrast to the other cases, patient 4 had a shorter follow-up time, only 14 month-span between primary surgical resection and death after PD to first-line gefitinib treatment. Correlating with its early stage, the primary tumour only harboured EGFR L861Q variant. This sensitising mutation was previously reported to be associated with shorter survival [11]. After PD to gefitinib, metastatic lesions were analysed at the time of autopsy. All necropsies harboured mutations in the WNT pathway (CTNNB1 G34R). Loss of GATA4 and CDKN2A/B and a TERT amplification were detected in three of the four samples (Fig. 2a).

In summary, the four cases showed temporal heterogeneity due to selective pressure, potentially caused by the treatment itself.

Spatial homogeneity in metastatic lesions at the time of autopsy

Regardless of the course of treatment, the same somatic variants were detected in all necropsies from each patient, except in patient 2, (Figs. 2 and 3 and Supplemental Table 3), suggesting a predominant spatial homogeneity at the time of autopsy. Six additional necropsies with a tumour area below the threshold for NGS analysis (<10%) were alternatively tested by ddPCR with an EGFR panel (technique with a higher sensitivity [0.01–0.05% LOD]). The analysis revealed identical mutations as in the remaining profiled necropsies within each patient (a complete set of EGFR mutations in necropsy samples can be found in Supplemental Table 3).

Discrepancies in VUS were detected only in patient 2 among the lung and the liver necropsies. This particular case was the only situation of oligoprogression before death. Although the pathogenic variants were present in the liver and the lung lesions including the drivers in EGFR, only the liver tumour was clearly growing by CT-Scan. Indeed, the steady-state lesion in the lung harboured a likely pathogenic truncating mutation in CDKN1B N146fs, but it was absent in the liver. In general, more variability was observed in terms of CNAs for all the cases, but mainly coinciding with samples of limited tumour area (15–10%) (Fig. 2a).

Patients 2 and 3 presented the resistant EGFR mutation T790M together with the sensitising mutations in exon 19 and L858R, respectively. The ratio of this doublet was previously described as a potential predictive biomarker of response to osimertinib [21] and rociletinib [22, 23]. In both cases, the spatial homogeneity included a similar value for T790M/sensitising ratio among all the necropsies analysed (Supplemental Fig. 1).

In order to better explore the spatial homogeneity concept, we further analysed the necropsies from patient 1 and performed WES (Fig. 4a). As expected, all samples showed private mutations, although among them only PABPC1 R565W in the oesophagus specimen was identified as a driver by the Cancer Genome Interpreter tool [24] (Supplemental Table 4). This mutation is not reported by cBioPortal [25, 26] and has scarce information in the literature supporting its pathogenicity. The median MAF of variants ubiquitous in all four necropsies is almost three-fold higher than the variants present only in one single lesion (Fig. 4b). This indicates the clonal and oncogenic behaviour of common mutations compared to likely passenger private mutations. Here, we found that driver mutations are present in all lesions in each patient, and passenger mutations are unlikely to influence the evolutionary fate of the tumour.

Fig. 4. Spatial homogeneity reviewed by whole-exome sequencing.

a Venn diagram of the mutations detected by WES in necropsies from Patient 1. The number of variants detected in each subgroup are indicated and listed in Supplemental Table 4. b Mutant allelic fraction corrected by the tumour area of each sample and segregated according to the number of lesions were the mutation is present. The Median MAF of each group is highlighted with a black line and the precise number.

As previously stated [27], the genetic landscape of the tumour subpopulations is highly dynamic and even though not fully understood, a number of factors may influence this, such as exposure to targeted therapies.

In summary, according to this study of limited size, 4/5 patients showed no variation of clinically relevant alterations (i.e. driver mutations, mechanisms of resistance and actionable pharmacological targets) at the time of death.

Discussion

As any other population under evolutionary constraints, tumours evolve depending on the quality and intensity of selective pressures. The heterogeneity of the primary tumour cannot be fully understood due to technological limitations and, as of today, it is very difficult to know if some of the resistance mechanisms are due to de novo mutations or correspond to the proliferation of pre-existing resistant subclones that are below the detection threshold, as already described for EGFR T790M mutation in treatment naïve tumours [28] or rare pre-existing MET amplification [29, 30].

Recent advances in molecular techniques are emerging as key diagnostic and monitoring tools. NGS applications may be considered the gold standard for molecular profiling of patient’s tumour status. ddPCR, with a 0.01–0.05% limit of detection (compared to an average 5% in NGS) is employed as a powerful additional resource to identify subclonal and expected resistance mutations, such as EGFR T790M or C797S.

Based on the increasing knowledge of intratumor heterogeneity [5, 31, 32], a challenging open question is which samples and at which time-point in the disease course will yield relevant information for clinicians. Research autopsy is becoming a key tool to advance precision medicine strategies, helping to understand dynamics of cancer evolution and therapeutic resistance [5]. The amount of tissue available in necropsies is sufficient to try multiple techniques to generate useful insights not reachable with biopsies or resection sampling, such as the extent of multiregional sampling, but it is still underused in the clinical practice [5]. Here, we report five cases whose necropsies, in combination with longitudinal data from biopsies, allowed for an extensive analysis to better understand tumour clonal evolution and mechanisms of resistance among patients, and also reported co-occurrence of more than one mechanism after first line with first-generation EGFR inhibitors, erlotinib and gefitinib. The MET amplification at PD to first-line therapy has been observed in other studies in about 15–20% of first-generation EGFR-TKI-resistant cancers [33]. In this study, two out of five patients harboured EGFR and MET amplification along with EGFR T790M mutations in the same lesions. This is in agreement with a recent publication where two or more subclonal genomic alterations are acquired upon resistance to the second-generation TKI osimertinib [34]. We cannot discard that the MET and/or EGFR amplifications were primary resistance mechanisms, but the time to PD at the first line in these patients indicates the contrary. Early progressions have been reported for patients with pre-existing MET amplification [29, 30, 34], while the patients from our cohort with a MET amplification upon PD were responsive to treatment during an average of 28.5 months (from 40 to 16 months). Interestingly, combined treatment of TKI and c-MET inhibitors lead to loss of MET amplification upon progression. Little is known about the resistance mechanisms to double inhibition of EGFR and c-MET therapies. Although, our results are in agreement with a recent work reporting a decrease of MET copy number in resistant MET-amplified NSCLC cell lines upon capmatinib treatment, along with downregulation of c-MET phosphorylation, which indicates its potent efficacy against c-MET-dependent tumours [35]. The optimal treatment approach for patients with NSCLC and EGFR mutations upon resistance to TKI is unclear, and therapeutic options are limited and under exploration. Our results, along with the osimertinib resistance published data [34], point towards the use of a combination of drugs for targeting multiple simultaneous resistance mechanisms.

Another important aspect for therapeutic guidance is the degree of heterogeneity of the resistance mechanisms to therapy within the multiple lesions of a given individual [36]. Our analysis of 22 necropsies from five individuals, 16 by NGS and 6 additional samples by ddPCR, revealed a prevalence of homogeneity in the genomic variants detected. From a total of 16 pathogenic variants and 9 VUS detected in all necropsies by NGS, we report homogeneity in all of them among the different metastasis of each patient, except for patient 2. Case 2 presented two pathogenic variants unique in the lung lesion and three VUS only present in the liver lesion. Interestingly, it is also the only case of oligoprogression before death. However, results do not reconcile genomic data with CT-scan, showing that the only active disease was in the liver but not in the lung metastasis. Unfortunately, due to the limited number of cases we are not able to provide a rule to determine when a single biopsy is sufficient. But we hypothesise that homogeneous progression of metastasis might advocate for this decision, while oligoprogression might encourage the clinician for a more careful approach with multiple biopsies. Lastly, we cannot discard the possible bias of the autopsy scenario. Whether close to death, spatial homogeneity among lesions could be more prominent. Other studies performed on necropsies do not point toward such bias in other tumour types, finding diverse resistance mutations in different necropsies from clear-cell renal or oesophageal carcinomas [37, 38].

The exome mutational analysis from patient 1 corroborates that multiple lesions in a patient are genetically divergent, with multiple private subclonal mutations. Pharmacological intervention may lead to higher temporal heterogeneity [39], increasing the chances of therapeutic failure. In spite of this heterogeneity, the spatial homogeneity observed in all cases except Case 2, and at specific time points, supports the ubiquitous nature of the clinically relevant alterations (i.e., driver mutations, mechanisms of resistance and actionable pharmacological targets) in patients with EGFR mutant NSCLC and upon TKI therapy. Thus, we consider that large targeted oncopanels for NGS are an extremely helpful tool to study resistance mechanisms and to guide therapeutic decisions, avoiding the cost, time and scientific knowledge for interpretation of the variants that WES requires.

We recognise the limitations of the study in the number of patients, the availability of samples and diversity in the treatments, which may lead to specific tumour evolution paths. Therefore, conclusions drawn should be interpreted with caution. Moreover, we understand that intralesion genetic heterogeneity is well documented, and we cannot discard hidden diversity in subclonal mutations or CNA. Nevertheless, such source of heterogeneity should affect the results among lesions. On the contrary, we report a high degree of homogeneity within necropsies. While preparing this manuscript, two articles [40, 41] from the same lab reported that homogeneity of functional mutations in driver genes is a common scenario in metastatic lesions from several untreated tumour types.

Following the same trend, a previous study reported a single resistant mutation after treatment with crizotinib in all metastatic sites [36]. On the other hand, other studies report mixed or opposite results: in one study heterogeneity in the acquired resistance mechanisms in two patients with NSCLC and treated with EGFR-TKIs was observed although RNA-seq was used for variant calling, with limited resolution [39]. Works conducted with a high number of samples in different tumour types also advocate for a diverse landscape of potential drivers and resistance mutations. In another study, necropsy analysis in metastatic breast cancer showed that different lesions in a patient acquired a variety of molecular mechanisms of resistance to PI(3)Kα inhibitor that converged in a functional loss of PTEN [42]. A possible explanation of the differences with our study, besides the tumour type and targeted therapy, may be that we are identifying a mechanism of seeding all metastatic sites by genetically identical circulating clones instead of convergent evolution [43].

Conclusions

Heterogeneity in the primary tumour contributes to subclonal diversity, this may not only overcome the severe bottleneck imposed by the metastatic process, but it may be a source of potential mechanisms to overcome effective pharmacological treatment. Based on our results, in TKI-treated advanced NSCLC, more than one resistance mechanism might co-occur in first-line treatments with TKIs and the analysis of a single metastasis biopsy at the time of PD might provide valuable information for patient monitoring as well as therapeutic decision-making.

Author contributions

AMM: conceptualisation, visualisation, methodology, formal analysis, data curation, writing, review and editing. EF: conceptualisation, methodology, formal analysis, supervision, data curation, writing, review, editing and funding acquisition. FMM: bioinformatic analysis and review of the manuscript. GC and JM: performed all DNA extractions, sequencing and ddPCR experiments. PN and IS: performed histopathological analysis and review of the manuscript. NMD, SC, AC, PI, NP and AN: review and editing of the manuscript. JMM: project administration, visualisation and review of the manuscript. AV: conceptualisation, methodology, formal analysis, data curation, review, editing and funding acquisition. MS: conceptualisation, visualisation, methodology, formal analysis, supervision, data curation, writing, review and editing. All authors provided comments on the manuscript and had final approval of the submitted version.

Funding information

This work was supported by the Spanish Cancer Association (AECC) Scientific Foundation (GCB14-2170) and partially by the Spanish Ministries of Health and Fondo de Investigación Sanitaria-Fondo Europeo de Desarrollo Regional (PI14/01248). MS was supported by an AECC Investigator award from the Spanish Cancer Association (AECC) Scientific Foundation (INVES19056SANS) and is currently supported by a Radix fellowship funded by IdISBa and Janssen.

Data availability

These data will be held at VHIO on secure servers. VHIO is supportive of data sharing and will endeavour to assist in requests for data sharing. All requests for access to the data will be formally requested stating the purpose, analysis and publication plans together with the named collaborators. All requests will be dealt with on a case by case basis in strict observance of applicable laws, rules and regulations.

Ethics approval and consent to participate

Approval to use these tumour specimens was obtained from the institutional review board and performed in accordance with the Declaration of Helsinki. All samples were obtained with the patients’ informed consent.

Consent to publish

No individually identifiable data are presented.

Competing interests

AMM provided consultation, attended advisory boards and/or speaker’s bureau for the following organisations: BMS, Roche, MSD, Pfizer, Boehringer-Ingelheim, AstraZeneca. EF provided consultation, advisory role and/or speaker’s bureau: AbbVie, AstraZeneca, Blueprint medicines, Boehringer-Ingelheim, BMS, Celgene, Lilly, Guardant Health, Janssen, Medscape, Merck KGaA, MSD, Novartis, Pfizer, Roche, Takeda, Touchtime. PN provided consultation, advisory role and/or speaker’s bureau: Bayer, MSD, Novartis, and Targos. IS provided consultation, attended advisory boards and/or speaker’s bureau for the following organisations: Roche, Abbvie, MSD, Pfizer, Takeda, BMS. SC provided consultation, attended advisory boards and/or speaker’s bureau for the following organisations: BMS, Roche, Pfizer, Boehringer-Ingelheim, MSD Oncology, Amphera. AC provided consultation, attended advisory boards and/or speaker’s bureau for the following organisations: BMS, Roche, Pfizer, Boehringer-Ingelheim, MSD Oncology, Kyowa Kirin, Celgene. PI provided consultation, attended advisory boards and/or speaker’s bureau for the following organisations: BMS, Roche, MSD, Boehringer-Ingelheim, MSD Oncology, Rovi, Kyowa Kirin, Grunenthal Pharma S.A. NP provided consultation, attended advisory boards and/or speaker’s bureau for the following organisations: BMS, Roche, Pfizer, Boehringer-Ingelheim. AN provided consultation attended advisory boards and/or speaker’s bureau for the following organisations: BMS, Roche, Pfizer, Boehringer-Ingelheim, Oryzon Genomics. AV has participated in advisory boards for BMS, Guardant Health and Bayer, and has provided consultation for Sysmex. The remaining authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41416-021-01558-9.