A Molecularly Annotated Model of Patient-Derived Colon Cancer Stem–Like Cells to Assess Genetic and Nongenetic Mechanisms of Resistance to Anti-EGFR Therapy

This article is featured in Highlights of This Issue, p. 725

In metastatic colorectal cancer (mCRC), monoclonal antibodies against the EGFR, associated with standard chemotherapies, significantly extended the median overall survival (from 20 to 30 months) in a fraction of patients (1). In the vast majority of such responsive cases, EGFR inhibition is effective in the absence of any genetic alteration of the receptor (2, 3). This exception to the “oncogene addiction” rule can be explained by the critical role that normal EGFR, physiologically stimulated by its ligand(s), plays in supporting colorectal cancer cell proliferation. This role is likely rooted in the cancer stem–like subpopulation, which, like its normal stem/progenitor counterpart, exquisitely depends on EGF signaling (4, 5). In mCRC patients refractory to EGFR inhibition, the tumor often harbors genetic alterations such as KRAS or BRAF mutations, hijacking the control of the proliferative pathway downstream EGFR (6). Recently, new less frequent genetic mechanisms of resistance were identified, thanks to the analysis of a large collection of xenopatients. These mechanisms include ERBB2/MET amplifications (7, 8), ERBB2 mutations (9), IGF2 overexpression (10), and EGFR mutations (11, 12). Such alterations represent actionable therapeutic targets, and, indeed, patients with ERBB2 amplification underwent a successful clinical trial (13).

Interestingly, as shown by whole-exome genetic analysis of an ample panel of xenopatients, in a fraction of refractory cases, a genetic mechanism of resistance to EGFR inhibition has not been identified (11). In such cases, as suggested by previous studies (10, 14), complete or partial resistance to cetuximab can be explained by the presence of autocrine or paracrine circuits of growth factors. Ligands of the EGF family, which fall in two main groups based on their ability to directly bind EGFR or another member of the EGFR family (15, 16), can provide such a mechanism of resistance, as they could either compete or bypass cetuximab inhibition (17, 18). For studying the outcome of paracrine signals, xenopatients may suffer from limitations, mainly because relevant growth factors can be underrepresented in the tumor microenvironment, or the murine ligands can fail to cross-react with human receptors. In a proof-of-concept study, we recently showed that xenopatient-derived cultures of stem-like cells (xenospheres) are a valuable system to address these issues in vitro and in vivo (5).

Xenospheres provide a unique counterpart of the original tumor and xenopatient, as they are amenable to in vitro quantitative measurement of proliferative responses, and, in vivo, they can regenerate phenocopies of the original tumor to provide a meaningful preclinical model (5, 19). In this study, we derived an ample collection of xenospheres and showed that they retain a remarkable genetic, biological, and pharmacologic correspondence with original mCRC xenopatients. In xenospheres lacking oncogenic activation of the RAS pathway, we could identify NRG1 as the main EGF family ligand able to substitute EGF in sustaining proliferation, and to induce primary resistance to cetuximab by engaging ERBB3. We could then show that broad inhibition of the EGFR family is more effective than selective inhibition of EGFR alone.

All procedures involving animal experimentation were approved by the Italian Ministry of Health and by the internal Ethical Committee for Animal Experimentation. Xenospheres were derived as previously described (5). Genomic DNA was extracted with the Wizard SV genomic DNA purification system (Promega) and analyzed for KRAS, NRAS, BRAF, PIK3CA mutations as previously described (8). Whole-exome sequencing (WES) analysis was performed by Personal Genome Diagnostics as previously described for the corresponding xenopatients, using the same settings and specific controls (11). Data concerning gene mutations and copy-number variations (CNV) were extracted and analyzed. CNVs include complete gene deletions and copy gains >4.

Cells were plated at clonal density (10 cells/μL) in ultra-low attachment 96-well plates (Corning-Sigma) in basal stem-cell medium (i.e., with 0.4 ng/mL of EGF). Note that 20 ng/mL of growth factors (EGF, bFGF, HGF, TGFα, HBEGF, EREG, AREG, NRG1, Peprotech) and inhibitors (10 μg/mL cetuximab, 10 μg/mL panitumumab, 0.5 μmol/L Gefitinib, 0.5 μmol/L Lapatinib, 2 μmol/L PLX-4720) were added at day 0. ATP production was measured using Cell Titer Glo (Promega) and GloMax 96 MicroplateLuminometer (Promega).

Immunoblottings were analyzed with the following antibodies: rabbit anti-phospho EGFR (Tyr1068; Cell Signaling Technology), rabbit monoclonal anti-EGFR (Cell Signaling Technology), mouse monoclonal anti-ERBB2 (Santa Cruz Biotechnology), rabbit anti-phospho ERBB3 (Tyr1289; Cell Signaling Technology), rabbit monoclonal anti-ERBB3 (Cell signaling Technology), rabbit anti-phospho ERK 1/2 (Thr202/Tyr204; Cell Signaling Technology), rabbit anti-ERK 1/2, rabbit anti-phospho AKT (Ser473; Cell Signaling Technology), rabbit anti-AKT (Cell Signaling Technology), rabbit anti-phospho S6 (Ser235/236; Cell Signaling Technology), rabbit anti-S6 (Cell Signaling Technology). Mouse monoclonal anti-vinculin (Sigma Aldrich) antibody was used as control of equal protein loading.

Immunophenotype was performed by incubating 2 × 10⁵ cells with appropriate dilutions of the following antibodies: anti–EGFR-PE (BD Biosciences), anti-ERBB3-APC (BioLegend). DAPI was added for dead cell exclusion. Samples were analyzed in a CyAN ADP (DakoCytomation). Data were processed using Summit 4.3 software (DakoCytomation).

KRAS^wt-Ctx-S xenospheres (CRC0059 and CRC0078) were stably transduced with a human NRG1–expressing lentiviral vector (Origene). NRG1 RNA expression was evaluated in vitro by qPCR using TaqMan gene expression probe Hs00247620_m1 on ABI PRISM 7900HT sequence detection system according to the manufacturer's instructions. Ubiquitin (Hs00824723_m1) and beta-actin (Hs99999903_m1) were used as housekeeper genes. In vivo NRG1 RNA expression was evaluated by RNA-ISH using RNAscope technology (Advanced Cell Diagnostic, Inc.) according to the manufacturer's instructions.

Dissociated xenospheres (5 × 10⁴ cells) were resuspended in a 1:1 mixture of stem-cell medium and matrigel (BD Bioscience), and subcutaneously injected into NOD/SCID male mice (Charles River Laboratories). When tumors reached an average volume of 400 mm³, mice were randomized and treated with 20 mg/kg cetuximab (Merck) twice weekly by i.p. injection, and/or with 150 mg/kg of lapatinib (Hospital Pharmacy), or 40 mg/kg of neratinib (Selleckchem), or 20 mg/kg of afatinib (Sequoia Research Products) by daily oral gavage. Tumor size was measured once weekly by caliper, and volume was calculated using the formula (4/3πd/2)²D/2, where d is the minor and D is the major tumor axis.

Tumor sections were formalin-fixed, paraffin-embedded, processed according to standard procedures, and analyzed with the following antibodies: anti–phospho-S6 rabbit monoclonal (Ser235/236; Cell Signaling Technology) and anti–phospho-ERK1/2 rabbit monoclonal (Thr202/204; Cell Signaling Technology).

Results were expressed as mean ± SEM. Statistical significance was evaluated by one-way ANOVA followed by Bonferroni Multiple Comparison Test, using GraphPad Prism software. P < 0.05 was considered statistically significant.

For more detailed methods, see Supplementary Experimental Procedures.

From mCRC xenopatients (8), we previously derived and characterized a small panel of “xenospheres” (5). These were shown to fulfill the requirements of colon cancer stem–like cells as, (i) in vitro, they retained self-renewal and pseudo-differentiative ability, and (ii) upon injection into immunocompromised mice, they formed tumors (“spheropatients”) that reproduced both the same histologic features and therapeutic response to cetuximab as the original tumors (Fig. 1A; ref. 5).

We now expanded our collection to obtain xenospheres from 58 different xenopatients (8–11, 20, 21), which were molecularly annotated for the presence of somatic mutations in KRAS, NRAS, PIK3CA, and BRAF genes (Fig. 1B; Supplementary Table S1), known to predict primary resistance to anti-EGFR therapy in patients and xenopatients (8). The frequency of KRAS mutations in xenospheres (35%) was similar to that found in both patients and xenopatients (8, 22), suggesting lack of in vitro selection. Xenospheres without mutations in any of the above 4 genes (“quadruple wild-type,” n = 32/58) could be further grouped according to the presence or absence of biomarkers recently associated with resistance to cetuximab in xenopatients, such as IGF2 overexpression (10), EGFR mutations in the extracellular domain (11, 12), ERBB2 amplification or mutations (8, 9), or PTEN deletion (ref. 11; Fig. 1B). Notably, 18 of 32 of these xenospheres (WT) were devoid of any of the above alterations.

The gene expression profiles of 28 randomly chosen matched xenospheres and xenopatients were compared to investigate the biological correspondence between the two models, showing a high level of overlapping between the transcriptome of matched samples (Fig. 1C and Supplementary Table S2). Overall, these findings show that in vitro–cultured colon cancer stem–like cells maintain both genetic and biological features of the tumor of origin.

In order to investigate whether xenospheres faithfully retain also the whole genetic make-up of xenopatients, 10 xenospheres derived from xenopatients that had undergone WES were analyzed using the same technological platform (Fig. 2A and Supplementary Tables S3 and S4; ref. 11). As the xenopatient WES was performed to identify new genetic biomarkers of anti-EGFR therapy resistance, all the cases were KRAS^wt (11). In xenospheres, the total number of CNVs ranged from 0 to 58 (with an average of 16, against 22 in the corresponding xenopatients), whereas somatic mutations ranged from 93 to 244 (with an average of 148, against 129 in xenopatients; Supplementary Table S5). The majority of CNV and somatic mutations were shared between xenospheres and matched xenopatients (Fig. 2B and C), with only few exceptions: CNV in CRC0574 (Fig. 2B), and somatic mutations in CRC0078 and CRC0254 (Fig. 2C). Shared lesions included also recently identified biomarkers of resistance to cetuximab such as ERBB2 amplification (CRC0080), ERBB2 mutation (CRC0151), EGFR^ecd mutation (CRC0104), and PTEN^del (CRC0394; Supplementary Table S1; refs. 8–12).

Interestingly, the somatic mutations shared between xenospheres and the matched xenopatients displayed an average allelic frequency of approximately 50%, whereas lesions private to either model displayed a frequency of approximately 30% (Fig. 2D). This is in accordance with the notion that genetic alterations required for the oncogenic phenotype (“driver” lesions) should be both shared within the same cell population (high allelic frequency) and conserved during the passage from tumor tissue to xenosphere culture. These features are consistent with the neutral tumor evolution model, recently proposed also for colorectal cancer (23, 24).

Moreover, somatic mutations common to xenospheres and xenopatients displayed highly similar allelic frequencies in the two models, further attesting that xenospheres faithfully retain the genetic make-up of xenopatients (Fig. 2E). Exceptions were CRC0078 xenospheres, which displayed a lower grade of correspondence with the original xenopatient, both in the number of shared somatic mutations and their allelic frequency correlation (Fig. 2C and E). This could be explained by two mechanisms: (i) in vitro selection of a pre-existing xenopatient subclone (due to sample bias or to selection during cell culture) and/or (ii) xenosphere genetic drift occurred during multiple in vitro passages. To identify the underlying mechanism, WES analysis of CRC0078 xenospheres was longitudinally compared with that of two further in vitro derivatives of the same lineage: (i) a secondary CRC0078 xenosphere (CRC0078^2nd), derived from the spheropatient, and previously shown to retain the same biological and tumorigenic properties of the primary xenosphere (5), and (ii) a primary cell culture selected in adherent standard cell culture conditions from the secondary xenosphere (CRC0078^ad; Fig. 3A). Although the total number of CNV was conserved in the three in vitro models (Fig. 3B), a progressive increase of somatic mutations was observed in CRC0078^2nd and CRC0078^ad (Fig. 3C). As the number of mutations shared between each model with the original xenopatient remained constant, this increase indicated accumulation of new private lesions. Notably, the average allelic frequency of such lesions was lower, as compared with the frequency of lesions shared with the original xenopatient (Fig. 3D), as observed in all other xenospheres (Fig. 2D). Moreover, the allelic frequencies of single genes shared between each in vitro model and the xenopatient (Fig. 3E), or between the different in vitro models (Fig. 3F), were highly conserved. These data suggest that CRC0078 xenospheres likely derive from a pre-existing xenopatient subclone and can accumulate new mutations that, however, for their low allelic frequency, are supposed to be mainly passengers.

Collectively, WES analyses indicate that xenospheres qualitatively and quantitatively reproduce, and stably retain over multiple in vitro passages, the landscape of relevant genetic alterations of the original tumors.

Having shown that xenospheres faithfully retain both biological and genetic features of xenopatients, we assessed the response of xenospheres to anti-EGFR therapy with cetuximab, and its correlation with the presence of genetic determinants of resistance. The response was assessed in a panel of 46 xenospheres, using previously described cell culture conditions (5). As observed in xenopatients (8, 11), cetuximab-resistant xenospheres mostly harbored KRAS mutations, whereas most of cetuximab-sensitive xenospheres were quadruple WT (Fig. 4A). Interestingly, among xenospheres lacking KRAS mutations (KRAS^wt), we could discriminate two subgroups (Fig. 4B): those derived from xenopatients that displayed cetuximab resistance (Ctx-R), and those derived from xenopatients that displayed cetuximab sensitivity (Ctx-S), which included tumors that underwent either shrinkage or stabilization (8, 10, 11, 21). Interestingly, KRAS^wt-Ctx-R xenospheres were enriched with cases harboring genetic alterations activating the RAS pathway (Fig. 4B). These lesions were recently described to be negative predictors of response to anti-EGFR therapy and were exploited as new targets to overcome resistance in xenopatients (8–11). Coherently, we observed similar correspondences in xenospheres (Fig. 4C): those harboring somatic mutations in the EGFR extracellular domain (CRC0104) and in ERBB2 (CRC0151) were resistant to cetuximab but sensitive to EGFR/ERBB2 small-molecule kinase inhibitors; BRAF-mutated xenospheres (CRC0480) were resistant to cetuximab but sensitive to BRAF inhibition, and were partially protected by EGF, consistently with the notion that EGFR activation protect against BRAF inhibition in colorectal cancer (Fig. 4C; ref. 25).

According to the presence of these oncogenic drivers, KRAS^wt-Ctx-R xenospheres were fully independent of exogenous EGF for their proliferation, similarly to KRAS^mut xenospheres (Supplementary Fig. S1A). Vice versa, the KRAS^wt-Ctx-S subgroup mainly included xenospheres that lacked mutations known to activate the RAS pathway (Fig. 4B) and was strongly dependent on exogenous EGF for their proliferation (Supplementary Fig. S1A). This indicates that these xenospheres lack a cell-autonomous genetic driver of proliferation. Therefore, they represent cases that, on the one hand, are susceptible to EGFR inhibition, and, on the other hand, can be affected by growth factors that can protect against different kinase inhibitors (26, 27). Remarkably, the latter could be underestimated in conventional xenopatients, as we showed for HGF (5).

Therefore, we set out to systematically investigate which signals could confer to KRAS^wt xenospheres the ability to circumvent the proliferative block imposed by EGFR inhibition with cetuximab. KRAS^wt xenospheres were kept in the presence of growth factors that activate the EGFR family (the EGFR ligands EGF, TGFα, HBEGF, EREG, AREG, and the ERBB3 ligand NRG1), and other tyrosine kinase receptors (bFGF and HGF). The response to cetuximab, measured as cell viability, was compared with that of untreated cells (Fig. 4D and E and Supplementary Fig. S1B). As expected, viability of the KRAS^wt-Ctx-R xenosphere group kept in basal medium was overall poorly affected by cetuximab, and poorly modulated in the presence of the other growth factors (Fig. 4D). In contrast, the viability of the KRAS^wt-Ctx-S xenosphere group was strongly impaired by cetuximab in basal medium, but it was at least in part preserved in the presence of other growth factors (Fig. 4E). These showed different abilities to protect from cetuximab: (i) negligible for EREG and AREG (viability < 0.4); (ii) intermediate for TGFα, HBEGF, HGF, and bFGF (0.4 < viability < 0.8); (iii) maximum for EGF and NRG1 (viability > 0.8).

Taken together, these data indicate that xenospheres reproduce the correspondence observed in xenopatients between the genetic make-up, on the one hand, and the therapeutic response to cetuximab or other targeted drugs on the other hand. Moreover, xenospheres lacking any known genetic determinants of resistance retain cetuximab sensitivity and can be exploited to qualitatively and quantitatively measure the ability of growth factors to compensate for EGFR inhibition.

In order to investigate the mechanisms of cetuximab primary resistance conferred by exogenous growth factors, we focused on the EGF family of ligands, whose expression in patients has been correlated with the cetuximab response (28–30). First, expression of EGFR family members ERBB1/2/3 (not ERBB4, undetectable in gene expression profiling) was assessed in Western blots of xenospheres kept in basal medium (Supplementary Fig. S2A). EGFR and ERBB3 proteins were expressed by most xenospheres, with the exception of one single case (CRC0729) that did not express EGFR and was completely insensitive to any of its ligands (Supplementary Fig. S2A and S2B). ERBB2 was significantly detectable in only two cases (CRC0394 and CRC0080), one of which (CRC0080) was ERBB2-amplified (Supplementary Fig. S2A).

Next, we investigated whether the different ability of EGF family ligands to sustain cetuximab resistance (as shown in Fig. 4E) was associated with ligand-specific mechanisms of EGFR family activation. EGF-like ligands such as EGF, TGFα, HBEGF, AREG, and EREG, that bind EGFR (15), can directly outcompete the antibody, whereas NRG1, which binds ERBB3 and induces formation of heterodimers with either EGFR or ERBB2, is likely to bypass EGFR inhibition by cetuximab (Supplementary Fig. S3A). KRAS^wt-Ctx-S xenospheres were kept in the presence of each ligand and biochemically analyzed. All EGF-like ligands activated ERK and AKT to a similar extent, whereas NRG1 hyperactivated both (Fig. 5A). The respective ability of EGF-like ligands to protect from cetuximab was correlated with a different ability to modulate EGFR total protein levels (Fig. 5A) and cell-surface expression (Fig. 5B): indeed, EGF (maximum protection against cetuximab) induced complete EGFR internalization and degradation, whereas TGFα and HBEGF (intermediate protection) showed only a partial effect on EGFR, and EREG and AREG (negligible protection) did not induce any EGFR internalization and degradation (Fig. 5A and B). As expected, NRG1 did not cause EGFR internalization, which mostly requires EGFR homodimerization (31, 32). Interestingly, the EGF ability to induce EGFR internalization is inversely proportional to its concentration, and becomes negligible at doses that induce cell proliferation but are fully inhibited by cetuximab (Supplementary Fig. S3B and S3C). We can thus hypothesize that EGF-like ligands (or ligand concentrations) that do not induce efficient EGFR downregulation from the cell surface leave the receptor exposed to the competing activity of cetuximab, whereas those that induce internalization provide an escape from cetuximab and a site of intracellular signaling.

The above findings highlighted that NRG1 can powerfully activate EGFR signaling—the essential proliferative pathway in colorectal cancer stem cells—in a way alternative to EGF, which bypasses cetuximab inhibition. As the effect of NRG1 on colorectal cancer stem–like cells has been only preliminarily characterized (33), we investigated its ability to sustain xenosphere propagation. Six KRAS^wt-Ctx-S xenospheres were long-term cultured either in the standard medium or in the corresponding medium where EGF was replaced by NRG1. Besides a stronger ERK and AKT activation by NRG1, as compared with EGF (Supplementary Fig. S3D and data not shown), no significant differences were observed in the long-term proliferative potential (not shown) and, interestingly, even in the global gene expression profiles (Supplementary Fig. S4), suggesting that NRG1 could be interchangeable with EGF to support xenosphere expansion.

RNAseq analysis of xenopatients unveiled that, although most of EGF-like ligands are expressed by colorectal cancer cells, thus triggering autocrine loops, NRG1 is seldom expressed, similarly to HGF (20). However, although it is known that HGF is mainly secreted by cancer-associated fibroblasts (34), the source of NRG1 is still poorly understood, although it may include bone marrow–derived mesenchymal stem cells (35) or infiltrating lymphoid cells (36). Therefore, as in the case of HGF, the immunocompromised xeno- or spheropatient can be inadequate to evaluate NRG1 contribution to cetuximab resistance.

To circumvent this limitation, two KRAS^wt-Ctx-S xenospheres were transduced with a lentiviral construct to induce autocrine expression of human NRG1 (Supplementary Fig. S5A and S5B). In a proliferation assay, NRG1-expressing xenospheres grew in the absence of either EGF or bFGF, showing complete resistance to cetuximab, but gaining full sensitivity to lapatinib (Fig. 5C). The same result was obtained in cells cultured in conventional cell-line conditions (i.e., in the presence of serum and adhesion, Supplementary Fig. S5C). This effect was specific as no inhibition was observed with an unrelated MET inhibitor (JNJ-38877605, Fig. 5D). Interestingly, the specific EGFR small-molecule inhibitor gefitinib induced a weaker response as compared with lapatinib (Fig. 5D), suggesting that NRG1 exerts its effect through both ERBB3/EGFR and ERBB3/ERBB2 heterodimers. As in the case of stimulation with exogenous NRG1 (Supplementary Fig. S3D), NRG1-expressing xenospheres displayed constitutive ERBB3 phosphorylation, and a stronger activation of both ERK and AKT/S6 kinase pathways as compared with their parental counterparts (Fig. 5E). These pathways were inhibited by lapatinib but almost unaffected by cetuximab (Fig. 5E). EGFR was activated in both NRG1-expressing and parental xenospheres (for the presence of EGF in basal medium), but, whereas in parental xenospheres EGFR phosphorylation was completely abrogated by cetuximab, in NRG1-expressing xenospheres, it was inhibited only by lapatinib (Fig. 5E).

In order to simulate paracrine secretion, a murine fibroblast primary culture (derived from a xenopatient) was transduced with human NRG1 (Supplementary Fig. S5E). Representative KRAS^wt-Ctx-S xenospheres were stimulated with conditioned media obtained from control and NRG1-expressing fibroblasts, and treated with either cetuximab or lapatinib. Only the conditioned medium from NRG1-expressing fibroblasts could induce proliferation similarly to EGF, an effect inhibited by lapatinib but not by cetuximab (Supplementary Fig. S5E), as observed in the autocrine model (Fig. 5D).

Taken together, these experiments show that either autocrine or paracrine NRG1 powerfully stimulates proliferative EGFR signaling through ERBB3, and that this activity can be inhibited by the EGFR family kinase inhibitor lapatinib, but not by cetuximab.

In order to assess NRG1 ability to sustain cetuximab resistance in vivo, and explore ways to circumvent it, NRG1-expressing KRAS^wt-Ctx-S xenospheres (CRC0078 and CRC0059) and their parental counterparts were injected into NOD/SCID mice (Fig. 6A). Parental xenospheres were previously shown to form spheropatients highly sensitive to cetuximab treatment (Supplementary Fig. S6A; ref. 5). One month after injection, NRG1-expressing xenospheres formed tumors that reached volumes suitable for starting treatments (∼500 mm³ for CRC0078 and ∼200 mm³ for CRC0059), whereas no parental xenosphere had formed detectable tumors (Fig. 6B). These appeared later and reached volumes comparable with those expressing NRG1 only after approximately 3 months (not shown), indicating that NRG1 strongly increases the xenosphere tumorigenic potential. RNA in situ hybridization confirmed that NRG1 was highly expressed only in tumors derived from NRG1-expressing xenospheres (Fig. 6C). Spheropatients were then treated with cetuximab and/or lapatinib. CRC0078-NRG1 and CRC0059-NRG1 spheropatients were completely resistant to cetuxumab, as tumors grew with the same rate as controls (Fig. 6D). In contrast, the same tumors were arrested by lapatinib. No additive/synergistic effect was observed by lapatinib and cetuximab combination (Fig. 6D). Similar results (growth arrest) were observed by treating CRC0078-NRG1 xenopatients with two other pan-HER inhibitors, afatinib or neratinib (Fig. 6E), which cause irreversible inhibition of the catalytic site, and were shown to be more potent than lapatinib in ERBB2-amplified/mutated xenopatients (9, 37). All three inhibitors effectively and comparably downregulated ERK and AKT signaling in tumors (Fig. 6F, Supplementary Fig. S6B). These results indicate that NRG1 can powerfully stimulate tumor growth in vivo, which can be arrested only by pan-HER inhibitors.

mCRC xenopatients were fruitfully exploited by us and others as a translational platform to identify mechanisms of both primary and secondary resistance to cetuximab, and assess new targeting agents to bypass them (38). The identification of ERBB2 amplification as a mechanism of primary cetuximab resistance (8) has been recently translated into a successful clinical trial (13).

In vitro preclinical models are required as well to investigate both genetic and nongenetic mechanisms of response and resistance to conventional and targeted therapies. In colorectal cancer, different cell culture approaches have been developed, including conventional cell lines (39, 40), organoids (41–43), and spheroids (19, 44–46), each displaying inherent and complementary advantages and limitations.

From mCRC xenopatients, we previously derived a small cohort of “xenospheres,” showing that they retained cancer stem–like properties, and generated tumor xenografts (“spheropatients”) that reproduced both the histotype and the pharmacologic response to cetuximab of the corresponding xenopatients (5). Here, we present an expanded cohort of 58 xenospheres, encompassing the main genetic lesions predictive of cetuximab resistance. By comparing gene expression profiles and WES analysis of a group of matched xenospheres and xenopatients, we observed a remarkable correspondence between the two models. WES analysis, in particular, showed that not only the type of genetic lesions, but also their respective allelic frequency, are passed on from xenopatients to xenospheres, indicating that xenospheres do not undergo a significant genetic drift from the original tumor. Moreover, they are also genetically stable over long-term propagation, as attested by longitudinal WES analysis of secondary xenospheres and a conventional cell line derived after extensive in vitro and in vivo passages of a prototypic case. Concerning their prerogatives as a reliable in vitro model, we can conclude that xenospheres display a faithful genetic correspondence with the original tumors. This genetic fidelity is shared with tumor organoids, whose derivation has been reported to be more efficient than xenospheres (90% vs. 45%–50%) and whose growth, unlike that of xenospheres, tends to reproduce also cell pseudo-differentiation as in whole tumor tissues (41). However, with respect to organoids, xenospheres and, in general, cancer stem–like cells propagated in culture as spheroids are endowed with the advantage to be more amenable to genetic manipulation and quantitative assessment of therapeutic outcomes at cancer stem-cell level, either in vitro or after in vivo transplantation and tumor regeneration (5).

As previously shown by us in a limited number of cases (5), and recently observed also in tumor organoids (42), KRAS^mut xenospheres display EGF-independent proliferative ability and, not surprisingly, are highly resistant to cetuximab in vitro. Here, we show for the first time that also KRAS^wt xenospheres derived from cetuximab-resistant xenopatients are mainly EGF-independent, and enriched with cases harboring genetic alterations such as EGFR extracellular domain mutations, ERBB2 mutation or amplification, BRAF mutations, PTEN deletion and IGF2 overexpression, each of which can provide a mechanistic explanation for cell-autonomous proliferation, and a therapeutic target (8–11).

Conversely, we found that KRAS^wt xenospheres derived from cetuximab-sensitive xenopatients (KRAS^wt-Ctx-S) are devoid of genetic alterations able to hijack the proliferative pathway and display strong EGF-dependent proliferative ability and in vitro cetuximab sensitivity. While representing patients that likely would benefit from anti-EGFR therapy, KRAS^wt-Ctx-S xenospheres can also be susceptible to autocrine and paracrine signals that can counteract EGFR inhibition, as we previously showed for the HGF–MET axis (5). By screening the ability of a growth factors panel to induce primary resistance to cetuximab in vitro, we identified different responses that can explain preclinical and clinical observations. In particular, among the EGF family of ligands, EREG and AREG are completely unable to outcompete the antibody. Accordingly, their expression does not confer resistance, but, rather, is associated with cetuximab response both in patients (28, 29) and in xenopatients (10). Interestingly, unlike EGF, these ligands, although capable of inducing proliferation, cannot trigger EGFR internalization and degradation, suggesting that ligand-induced disappearance of EGFR from the cell surface is relevant to protect cells from cetuximab inhibition. Similarly, TGFα and HBEGF induce only partial EGFR internalization and cetuximab resistance; this correlates with their enriched expression in KRAS^wt xenopatients displaying either complete or partial resistance (disease stabilization) to cetuximab (10).

In our study, NRG1 emerged as a pivotal factor of primary resistance to EGFR inhibition. NRG1 is a ligand of the EGF family whose role in tumors has been poorly characterized. NRG1 binds the kinase-deficient ERBB3 receptor, and induces its heterodimerization with the other family members (EGFR, ERBB2, and ERBB4), which are required for ERBB3 transphosphorylation and the ensuing signaling activity (16). The mechanism of ERBB3 activation is complex and depends also on other EGF family ligands, specifically binding EGFR, as well as on dimerization with unrelated receptors such as the HGF receptor/MET (47). By this multiple interactions, ERBB3 is involved in primary and secondary resistance to EGFR or ERBB2 inhibition in several tumor types, including lung (47), head and neck (48), and ERBB2-amplified colorectal cancers (37). Concerning the specific role of NRG1, it was reported that this factor can induce in vitro resistance to various kinase inhibitors (26, 33). In colorectal cancer, high levels of circulating NRG1 correlate with weak response to cetuximab (14). However, the activity of NRG1 has been so far poorly investigated at the mechanistic level in in vitro models. The ability of NRG1 to induce cetuximab resistance has been recently reported in a single conventional colorectal cancer cell line (49), which harbored a high-copy EGFR amplification, a genetic alteration uncommon in colorectal cancer. Our study shows for the first time the protective ability of NRG1 in a large panel of mCRC stem-like cells that faithfully mirror the genetic make-up of the original tumors, thus achieving robust translational reliability. Moreover, we provide also the first evidence that NRG1 sustains long-term in vitro propagation of mCRC stem-like cells.

As NRG1 is produced by still elusive cells of the tumor microenvironment (20), to investigate its ability to sustain cetuximab resistance in vivo, we induced NRG1 expression in KRAS^wt-Ctx-S xenospheres, as we previously did for HGF (5). NRG1-expressing xenospheres displayed EGF-independent proliferative ability and full resistance to cetuximab. However, they were highly sensitive to lapatinib, a pan-HER inhibitor active in cancer cell lines harboring an NRG1-autocrine loop (50). Here, we show for the first time that NRG1 dramatically increases the tumorigenic potential of xenospheres injected into NOD/SCID mice. In accordance with in vitro data, NRG1-expressing spheropatients displayed complete resistance to cetuximab treatment, but high sensitivity to different pan-HER inhibitors, including lapatinib, afatinib, and neratinib.

It has been previously shown that rare genetic lesions in members of the EGFR family (EGFR and ERBB2 mutations, ERBB2 amplification) promote resistance to anti-EGFR therapy in KRAS^wt xenopatients (8, 9, 11) and patients (13), which can be bypassed by different strategies that ultimately target multiple ERBBs. Here, by using a large cohort of colorectal cancer stem–like cultures, we show that the same strategy might also be effective in counteracting nongenetic mechanisms of primary resistance driven by ligands of the EGFR family, mainly NRG1. These results provide a proof-of-concept for the clinical investigation of NRG1 as a predictive biomarker of primary resistance to selective EGFR targeting, suggesting the requirement for broad inhibition of the entire EGFR family. This should apply primarily to RAS^wt patients and, possibly, to all patients lacking other biomarkers of resistance to EGFR inhibition, such as K/NRAS or BRAF mutations, MET amplification or IGF2 overexpression. As in the patient population NRG1 expression is likely a continuous variable, its clinical development as a predictive biomarker presents with the challenge to define a reliable threshold that discriminates positive from negative cases. As the cellular origin of NRG1 remains elusive, and likely involves tumor-associated cells, evaluation of in situ expression, although feasible either with immunohistochemistry or the more sensitive RNA in situ hybridization, can be significantly jeopardized by sample-bias. Integration of complementary approaches including measurements in circulating blood (14) should be recommended.

No potential conflicts of interest were disclosed.

Conception and design: P. Luraghi, C. Boccaccio

Development of methodology: P. Luraghi, F. Sassi, F. De Bacco

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P. Luraghi, V. Bigatto, E. Cipriano, G. Reato, F. Orzan, E. Medico, A. Bertotti, L. Trusolino

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P. Luraghi, G. Reato, F. Orzan, C. Isella, S.E. Bellomo, E. Medico

Writing, review, and/or revision of the manuscript: P. Luraghi, V. Bigatto, P.M. Comoglio, A. Bertotti, L. Trusolino, C. Boccaccio

Study supervision: C. Boccaccio

This work was supported by AIRC, Italian Association for Cancer Research (“Special Program Molecular Clinical Oncology 5xMille, N. 9970”; IG 15709 to C. Boccaccio, 15572 to P.M. Comoglio, 18532 to L. Trusolino, 15571 to A. Bertotti), Transcan (L. Trusolino), TACTIC (L. Trusolino), Fondazione Piemontese per la Ricerca sul Cancro-ONLUS, 5 × 1000 Ministry of Health (L. Trusolino), and Italian Ministry of Health (Ricerca Corrente).

We thank Raffaella Albano for technical help, Eugenia Zanella and Francesco Galimi for genomic data of xenopatients, Roberta Porporato for RNA-ISH analysis, and Antonella Cignetto, Daniela Gramaglia, and Francesca Natale for assistance. P. Luraghi was recipient of a “Fondazione Umberto Veronesi” postdoctoral fellowship.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.