Oncogenic EML4-ALK assemblies suppress growth factor perception and modulate drug tolerance

Optogenetic profiling reveals that RTK signaling is suppressed in EML4-ALK+ cancer cells

To determine whether EML4-ALK could alter RTK signal transmission, we expressed a light-sensitive fibroblast growth factor receptor (optoFGFR³¹) in STE-1 cancer cells, which express EML4-ALK variant 1 (V1)³² (Fig. 1E, Supplementary Fig. S1A). We observed signal transmission by applying blue light stimuli and measuring pathway activation (phospho-ERK, ppERK) in the presence or absence of ALK activity. In these cells, light stimulation of optoFGFR induced only minimal ppERK signal increase above the high basal levels driven by EML4-ALK (Fig. 1F, top). Surprisingly, pretreatment with ALK inhibitor crizotinib (ALKi, 1 µM) led to exceedingly strong ERK signaling, consistently surpassing levels achieved in the absence of drug treatment (Fig. 1F, bottom). The dynamic range (fold-change) of signal induction increased as a function of light intensity and reached a maximum of ~10-fold at the highest levels of light stimulation, compared to ~1.5-fold in the absence of drug (Fig. 1G). Notably, strong ppERK increase in drug-treated cells was observed in response to even low levels of light (8 mW/cm²), a level that did not provide measurable increase in untreated cells, suggesting that ALK inhibition can sensitize cancer cells to weak RTK signals (Supplementary Fig. S1B). We note that although the magnitude of dynamic range increase was dependent on the expression levels of optoFGFR, ALKi potentiated signaling across all expression levels (Supplementary Fig. S1C). Collectively, these results indicate a strong suppressive effect of EML4-ALK activity on transmembrane RTK signaling.

EML4-ALK suppresses—and ALK inhibition restores—EGFR signaling

EGFR is found in >90% of NSCLC and modulates drug responses and resistance development in EML4-ALK+ cancers^{10,12,13,14,24}. To determine whether EML4-ALK could generally suppress endogenous transmembrane RTKs including EGFR, we measured ppERK induction upon addition of epidermal growth factor (EGF) in STE-1 and H3122 cells, two established EML4-ALK (V1)+ cell lines³² (Fig. 2A). Strong (100 ng/mL) EGF stimulation gave minimal ppERK response in the absence of drug, but pretreatment with ALKi substantially increased both the absolute magnitude as well as fold-change of ppERK in both cell lines (Fig. 2B–D, Supplementary Fig. S2A, B). As before, increased dynamic range was due to lower baseline—but also higher maximal—ppERK signaling and the magnitude of fold-change increase was dose-dependent (Fig. 2D). We found similar responses using first-line ALK inhibitor alectinib (Supplementary Fig. S2C) as well as in two additional primary patient-derived cell lines that express EML4-ALK (V1) (Fig. 2E, Supplementary Fig. S2D, E). However, EGF response was not potentiated in drug-treated cancer lines that expressed constitutively active transmembrane ALK mutants (Fig. 2F). Thus, suppression of EGFR signaling is a shared property of EML4-ALK(V1)+ cancer cells and depends on oncogene activity, but is not a general property of oncogenic ALK receptor activity.

To test the sufficiency of EML4-ALK for RTK suppression, we measured EGF response in isogenic, non-transformed lung epithelial Beas2B cells that were transfected with EML4-ALK (V1) (Fig. 2G). Exogenous EML4-ALK expression raised basal ppERK levels (t = 0, Fig. 2H), but EGF stimulation resulted in only a small increase in ppERK relative to cells transfected with a control plasmid, indicating suppression of EGFR/ERK signaling. Pre-incubation with ALKi reversed suppression and allowed stronger maximum and fold-change of ERK response (Fig. 2H, I). Collectively, our results show that EML4-ALK(V1) activity is sufficient to desensitize cellular response to EGFR stimuli and that ALK inhibition resensitizes and potentiates this response.

Mapping EGFR suppression using optogenetics

We next sought to understand the molecular mechanism by which EML4-ALK suppressed EGFR signaling. We confirmed that ALKi suppressed RAS/ERK signaling but did not alter levels of EML4-ALK (Supplementary Fig. S3A). To narrow candidate mechanisms of suppression, we first determined the duration of ALKi pre-incubation that was required to observe enhanced EGFR/ERK signaling. We preincubated both STE-1 and H3122 cells with 0–8 h of ALKi, and we analyzed ppERK levels in response to subsequent EGF stimulation (Fig. 3A). In both cell lines, an increase in ppERK amplitude was observed with as little as 5 min of ALKi pre-incubation (20 min total including stimulation) and rose to half-max with only ~15 min of pre-incubation (Fig. 3B). Such fast response suggested a primarily post-translational mechanism.

A Time course of EGFR sensitization was obtained by pre-incubating cancer cells with 1 µM crizotinib for a variable period before stimulation with EGF (50 ng/mL) for 15 min, followed by fixation and immunostaining for ppERK. B ppERK induction as a function of ALKi pre-incubation time. Open circles: unstimulated; closed circles: EGF-stimulated. Data points represent mean of three biological replicates, each representing 1200–2700 cells for STE-1 and 2000–3300 cells for H3122. Error bars = 95% CI. C Pinpointing the location of EML4-ALK interaction with RTK/ERK signaling. D OptoFGFR and optoSOS permit optogenetic stimulation at successive nodes of the pathway. E Quantification of ppERK levels in response to optoFGFR or optoSOS in the presence (orange) or absence (gray) of ALKi (1 µM crizotinib). X-axis represents optoFGFR or optoSOS expression quartiles. Data points represent mean ± SEM of three biological replicates, each representing 200–1100 cells. Significance assessed using one-sided T-test, n = 3 biological replicates. Open circles: unstimulated; closed circles: light stimulated. F ALK-dependent suppression is observed only with optoFGFR, suggesting that suppression happens upstream of RAS but downstream of RTK activation. G Testing the role of ERK-dependent negative feedback on RTK suppression. Light stimulation of optoSOS drives elevated levels of ppERK during ALKi treatment and sustains any potential ERK-dependent negative feedback that would otherwise be lost during ALK inhibition. H STE-1 cells were treated with either ALKi or ALKi and optoSOS stimulation, and the response to EGF was assessed. I Predicted results and implications for ERK-dependent feedback. J optoSOS did not suppress ALKi-induced potentiation of EGF response, suggesting that ERK-dependent negative feedback does not account for EGFR suppression. Data points represent means of 150–900 cells per condition. Significance assessed using one-sided T-test, n = 3 biological replicates.

To pinpoint the molecular interactions that mediated EGFR suppression, we measured ppERK response after targeted optogenetic stimulation at successive nodes of the EGFR/ERK pathway³⁰ (Fig. 3C, D). We stimulated STE-1 cell lines that stably expressed either optoFGFR or optoSOS, which allows activation of RAS/ERK signals through light-induced membrane recruitment of the SOS catalytic domain^33,34 (Fig. 3C, D, Supplementary Fig. S3B, C). As before, optoFGFR was strongly suppressed across all expression levels of the probe, and ALKi pretreatment potentiated signaling (Fig. 3E, left). By contrast, optoSOS-driven signaling was not suppressed at any expression level, and ALKi pretreatment had no effect on the maximal signal achievable (Fig. 3E, right). These results suggest that EML4-ALK suppresses signaling upstream of RAS stimulation, for example, at the receptor level. However, ALKi pre-incubation did not potentiate phosphorylation of EGFR upon EGF stimulation (Supplementary Fig. S3D, E). Collectively, our results show that EML4-ALK suppresses RTK signaling downstream of receptor phosphorylation but upstream of RAS activation. These results implicate a role for the adapter proteins that couple these two successive nodes (Fig. 3F).

Rapid post-translational negative feedback of RTKs has been described previously, for example through ERK-dependent phosphorylation of SOS1^11,35. To test the role of ERK-dependent negative feedback in our system, we decoupled ALK inhibition from the loss of ERK signaling by supplementing cells with optogenetic RAS (optoSOS) stimulation during ALK inhibition (Fig. 3G). We assessed response to EGF after 15 min of ALKi pre-incubation, a duration that was sufficiently long to resensitize EGFR (Fig. 3B) but sufficiently short to test only post-translational effects of ERK inhibition. One experimental group received oncogenic levels of optogenetic RAS/ERK signaling during the 15 min of ALKi treatment in order to maintain any ERK-dependent feedback that would have been otherwise lost through ALK inhibition (Fig. 3G–I, Supplementary Fig. S4A). Optogenetic signal supplementation did not diminish the enhanced response to EGF (Fig. 3J). Similar results were obtained after 1 and 2 h of ALKi pre-incubation (Supplementary Fig. S4B). In agreement, the levels of ERK-dependent negative regulator Spry2 did not change after acute ALKi pre-incubation (Supplementary Fig. S4C). These results show that, in EML4-ALK+ cancer cells, EGFR suppression and rapid drug-induced resensitization are not mediated by established, ERK-dependent mechanisms.

Active EML4-ALK assemblies suppress EGFR through sequestration of RTK effectors

Protein condensates can act as a negative regulator in both natural and engineered systems when they sequester essential components of biochemical reactions^36,37,38. To test if EML4-ALK assemblies suppressed RTKs, we first examined EGF response in cells transfected with EML4-ALK mutants that failed to assemble condensates due to mutations in either the trimerization domain (ΔTD) or the kinase domain (K589M)¹⁹. In contrast to wt EML4-ALK (V1) (Fig. 4A, left), the condensate-deficient mutants did not stimulate ERK signaling and permitted higher levels of EGF-induced ERK activity (Fig. 4A, middle and right), suggesting that signaling-competent EML4-ALK assemblies were essential for suppression of EGFR sensitivity.

A Quantification of ppERK response to EGF (50 ng/mL) in Beas2B cells that were transfected with EML4-ALK (V1), EML4-ALK (∆TD), or kinase-dead EML4-ALK (K589M), or with an empty vector (EV). Data points represent mean ± SEM of three biological replicates, each representing 200–400 EV cells, 130-270-EML4-ALK (V1) cells, 55–160 EML4-ALK (∆TD) cells, or 260–450 EML4-ALK (K589M) cells. B Colocalization of EML4-ALK condensates with endogenously tagged GRB2 (GRB2:mNG2). See Fig. 5 for quantitation. Scale = 10 µm. C GRB2:mNG2 Beas2B cells were transiently transfected with EML4-ALK and stimulated with EGF (50 ng/mL) to visualize GRB2 translocation in the presence and absence of EML4-ALK (V1). D Impaired membrane translocation of GRB2 in the presence of EML4-ALK condensates. Time in mm:sec. See Supplementary Movie 1. E Line scan of GRB2 intensity distribution in the presence (red) or absence (gray) of EML4-ALK expression, as depicted in (C). F Quantitation of translocation of endogenous GRB2 or SOS1 in the presence (red) or absence (gray) of EML4-ALK. Boxplot indicates the median and upper/lower quartiles, and whiskers extend to 1.5*IQR. See Fig. 5 for full quantitation. G GRB2 localization and translocation were visualized upon treatment with 1 µM ALKi and subsequent stimulation with EGF (50 ng/mL). Time in hh:mm. H Quantification of kinetics of GRB2 dissociation from condensates after ALKi treatment. I ALKi restores GRB2 and SOS translocation. Plot shows median translocation of endogenous adapters in cells expressing EML4-ALK(V1) represented as a fraction of translocation in the absence of EML4-ALK (V1). Data represent medians, error bars show 1st and 3rd quartiles of 1000 bootstrapped samples (distributions found in Fig. 5F, G). Significance assessed by one-sided bootstrap test for comparison of medians. See Fig. 5F, G for underlying data and quantitation. J Immunoprecipitation of EGFR shows enhanced co-precipitation of GRB2 and SOS1 in the presence of both ALKi pretreatment and EGF in STE-1 cells. gray arrows: non-specific bands. K Densitometry quantification of three independent pulldowns. L Testing effect of GRB2 overexpression on ERK response. M Expression levels of GRB2-GFP or GFP analyzed in (N, O). N ppERK levels in the absence (open circles) or presence (closed circles) of EGF stimulation (50 ng/mL) as a function of expression levels of GFP or GRB2-GFP. Data represent mean ± SEM of three biological replicates, each representing the mean of 100–300 cells. O Absolute magnitude of ppERK increase for each expression bin from data shown in (N). Significance assessed by one-sided T-test, n = 3 biological replicates. P Fold-change of response calculated from data in (N). Q Conceptual model of how EML4-ALK suppresses transmembrane RTKs. EML4-ALK sequesters adapters like GRB2/SOS1 and prohibits their translocation to activated RTKs. ALK inhibition releases adapter sequestration and restores cellular response to RTK stimulation.

We hypothesized that EGFR suppression could result from cytoplasmic sequestration of RTK adapter proteins. Such adapters, including GRB2 and SOS1, are required to transmit signals from both EML4-ALK and EGFR¹⁹ and thus represent shared resources that could implement competitive inhibition. We first directly observed the association of adapters with EML4-ALK using Beas2B cells where GRB2 was fluorescently tagged at the endogenous locus¹⁹. Upon transfection with mCh-EML4-ALK, GRB2 colocalized within EML4-ALK condensates (Fig. 4B), as shown previously^19,20. We then measured translocation of GRB2 in response to EGF stimulation (Fig. 4C). In untransfected cells, GRB2 appeared diffuse in the cytoplasm but translocated to the membrane within 4D–F, Supplementary Movie 1). However, in cells with EML4-ALK, GRB2 remained sequestered in the cytoplasmic puncta and did not translocate in response to EGF (Fig. 4D–F Supplementary Movie 1). Such sequestration was also observed for SOS1 (Fig. 4F). Upon treatment of EML4-ALK-expressing cells with ALKi, adapters rapidly diffused from condensates into the cytoplasm (t_1/2 = 17 ± 2 min) (Fig. 4G, H, Supplementary Movie 2). Subsequent treatment with EGF now stimulated robust membrane translocation of both GRB2 and SOS1 despite the presence of exogenous EML4-ALK (Fig. 4G, I). Of note, ALKi treatment does not cause complete dissolution of the underlying granule-like EML4-ALK condensates in cancer cells, though their number and size diminish, consistent with previous results²⁰ (Supplementary Fig. S5A, B).

We next measured the ability of EGFR to recruit adapters in EML4-ALK+ cancer cells through co-immunoprecipitation with endogenous EGFR (Fig. 4J, K, Supplementary Figs. S5C, D, S6). In both STE-1 and H3122 cells, a short (20 min) ALKi pre-incubation yielded dramatically stronger co-precipitation of both GRB2 and SOS1 with EGFR in response to EGF stimulation compared to cells that received only ALKi or only EGF (Fig. 4J, K, Supplementary Figs. S5C, D, S6).

If adapters are sequestered in EML4-ALK assemblies, then supplementing the cell with exogenous adapters could restore sensitivity to EGFR signaling (Fig. 4L). Overexpression of GRB2-GFP exerted effects on both basal and stimulated EML4-ALK cancer cells, in a manner dependent on GRB2 expression levels (Fig. 4L–P). In STE-1 cells, exogenous GRB2-GFP both lowered basal ppERK levels and potentiated maximal ppERK response to EGF relative to GFP controls, resulting in an increase in absolute magnitude as well as fold-change of response (Fig. 4O, P). We speculate that decreased basal and modest increase in stimulated ppERK occurs because the exogenous GRB2 contributes not only to the diffuse cytoplasmic compartment but also interacts with active assemblies, altering their stoichiometry. Similar effects were observed in H3122 cells, though with smaller relative increase in absolute and fold-change ppErk response relative to GFP controls (Supplementary Fig. S5E–H). We hypothesized that, in these cells, the more modest increase could be due to continued limitation of another component, e.g., SOS1, which is expressed at ~10–100-fold lower concentration than endogenous GRB2^39,40. In agreement, transient expression of SOS1 further potentiated ppErk response in these cells (Supplementary Fig. S5I, J).

Collectively, our results show that active, multivalent EML4-ALK assemblies sequester adapters including GRB2 and SOS1 and suppress their translocation to the membrane upon EGF stimulation, providing a mechanism by which EML4-ALK can competitively suppress EGFR signaling. ALKi treatment restores the available pool of adapters and consequently restores EGFR transmission (Fig. 4Q). Because this model relies on general properties of RTK fusions (e.g., multivalency and adapter recruitment), it predicts that cancer cells driven by other RTK fusions that form condensates would similarly show repressed EGFR signaling. We thus measured EGF response in TPC-1 cells, which harbor the CCDC6-RET fusion. Like EML4-ALK, CCDC6-RET can form condensates that colocalize with many RTK adapters, including GRB2 and SOS1¹⁹. As predicted, pretreatment with RET inhibitor BLU-667 permitted stronger response to EGF compared to that observed in drug-naive cells (Supplementary Fig. S7A, B). Furthermore, we tested a panel of cancer cell lines driven by either hyperactivating mutants or amplification in EGFR or HER2 and found no similar hyperactivation of transmembrane RTK signaling upon oncogene inhibition, further linking signal suppression specifically to cytoplasmic RTK assemblies and not simply to hyperactivated RTKs (Supplementary Fig. S7C, D).

Adapter sequestration and signal suppression is a common property of EML4-ALK variants

EML4-ALK(V1) is one of ~15 EML4-ALK variants found in human cancers⁴¹. Of these, variants 1,2 and 3 account for 70–90% of cases, with a roughly 3-fold higher frequency of V1 or V3 compared to the less-common V2⁴¹. EML4-ALK variants differ by the length of the EML4 fragment that is fused to the ALK kinase domain (Fig. 5A), and the differing lengths of EML4 are associated with different molecular interactions and condensation properties²⁰. We thus asked whether EML4-ALK variants retain the ability to form multivalent assemblies, sequester GRB2 and SOS1 in the cytoplasm, and suppress RTK signal transduction, as observed in V1.

A Three common oncogenic variants of EML4-ALK (Variants 1–3) share a common ALK fragment but differ in the lengths of the EML4 domain. B Expression of mCh-EML4-ALK(V1/2/3) in GRB2:mNG2 Beas2B cells showed condensation of each variant as well as the propensity of the condensates to colocalize with GRB2. Scale = 10 µm. C Quantification of puncta per cell for each variant. D Quantification of the percent of EML4-ALK puncta that overlap with GRB2 puncta in each cell. Boxplots in (C, D) show median and upper/lower quartile, and whiskers extend to 1.5*IQR. C, D Data points represent 18 (V1), 28 (V2), and 21 (V3) cells. E Translocation was quantified by identifying the cell edge and defining a 10 pixel ring into the cytoplasm (“edge”). The remaining cell pixels beyond this ring wire designated as the cell “core”. Membrane localization was defined as the ratio of mean edge fluorescence to mean core fluorescence. Translocation was defined as the difference in adapter membrane localization after 1.5 min of EGF stimulation vs pre-stimulation. F, G Quantitation of translocation of GRB2 (F) or SOS (G) for cells transfected with one of the 3 EML4-ALK variants or for neighboring untransfected cells (wt). Due to small variations in imaging plane between acquisitions, the absolute magnitude of translocation differed between variants and drug conditions (note the differences in untransfected Beas2B responses, which are equivalent conditions between panels). However, cells with or without EML4-ALK (black vs. red in the same panel) were imaged in the same field of view and thus can be compared directly. Data points represent individual cells. For (F), n = 57(WT)/40(V1), 170(WT)/42(V2), 24(WT)/34(V3) cells. For (G), n = 46(WT)/50(V1), 50(WT)/27(V2), 67(WT)/40(V3) cells. Boxplots show median and upper/lower quartile, and whiskers extend to 1.5*IQR. H Definition of the magnitude of translocation. I, J Comparison of translocation suppression of GRB2 (F) or SOS1 (G) for each of the three variants. Data represent median translocation suppression from resampling of 1000 bootstrapped samples. Error bars show lower and upper quartiles. Significance determined by either one-sided T-test (panels F, G) or one-sided bootstrap test (panels I, J). Data for Variant 1 in (F–J) is reproduced from Fig. 4F, I. K Beas2B cells were transfected with EML4-ALK-2A-H2B-miRFP constructs for one of 3 EML4-ALK variants (V1, V2, V3), or with an H2B-iRFP control, and ppERK levels were assessed after stimulation with EGF (15 min, 50 ng/mL) in the presence or absence of ALKi (1 µM crizotinib, 2 h), through immunofluorescence. L Quantification of ppERK immunostaining after EGF stimulation of Beas2B transiently expressing EML4-ALK in the presence or absence of ALKi. Significance assessed by Hsu MCB test. n = 8 biological replicates. M ppERK response in the presence and absence of ALKi pretreatment. Data points represent the mean ppErk intensity of 20–60 cells. Significance assessed by one-sided T-test. n = 8 biological replicates. Gray bars in (M) are reproduced from (L) for direct comparison to between non-treated and ALKi-treated cells.

We expressed mCh-tagged fusions of each EML4-ALK variant in Beas2B cells where endogenous GRB2 or SOS1 was tagged with mNeonGreen2 (mNG2). All three variants could form cytoplasmic condensates (Fig. 5B, C, Supplementary Fig. S8A, C). On average, ~80–90% of EML4-ALK puncta overlapped with GRB2 puncta, and 60–90% visibly overlapped with SOS1 (Fig. 5D, Supplementary Fig. S8B, D). Treatment with ALKi (crizotinib, 1 µM) caused dissolution of GRB2 and SOS1 puncta across all variants (Supplementary Fig. S8E–H). Thus, each variant retained the ability to form multivalent assemblies of oncogene and adapters that depended on ALK kinase activity.

We next asked whether each variant could functionally sequester adapters during EGF stimulation by directly measuring cytoplasmic-to-membrane translocation of endogenous GRB2 and SOS1, as before (Fig. 5E). All variants suppressed translocation of both GRB2 and SOS1, with suppression strongest for V1 and V3 relative to V2 (Fig. 5F–J). Pre-incubation with ALKi relieved this suppression in all cases.

Suppression of adapter translocation corresponded to the ability of each variant to suppress EGF-induced ERK signaling (Fig. 5K–M). Expression of each variant increased ppERK levels relative to control cells, in line with the ability of each variant to drive oncogenic RAS/ERK signaling (Fig. 5L). However, in response to EGF, cells expressing EML4-ALK showed a markedly weaker response relative to controls (Fig. 5L). Signal suppression was stronger for V1 and V3 relative to V2, consistent with weaker suppression of adapter translocation for V2 (Fig. 5F–J). Pretreatment with ALKi reversed signal suppression for all variants, again with stronger reversal for V1 and V3 relative to V2, where initial suppression was milder (Fig. 5M).

In sum, this series of experiments reveals that, despite different sizes, structures, and biophysical properties of condensation²⁰, three major EML4-ALK variants share the ability to form oncogene/adapter assemblies, sequester RTK adapters, and suppress transmembrane ligand-mediated signaling.

Single-cell analysis of RTK resensitization reveals rapid signal reactivation after ALK inhibition

RTK activity promotes drug tolerance across cancer types^15,28. We hypothesized that drug-dependent resensitization of RTKs could promote RTK signaling during ALKi therapy. We thus monitored ERK signaling in drug-treated populations of STE-1 cells using ErkKTR, a biosensor that reports on ERK activity through nuclear exclusion of a fluorescent protein⁴² (Fig. 6A). We first used ErkKTR to confirm that ALKi sensitized cells to EGF ligands, finding that ALKi-treated cells showed measurable increase in ERK activity at ~10-fold lower EGF concentrations than untreated cells (Fig. 6B, Supplementary Fig. S9A, Supplementary Movie 3). Moreover, at any concentration above this response threshold, the magnitude of single-cell responses was uniformly stronger in the presence of ALKi (Supplementary Fig. S9B, Supplementary Movie 3), paralleling our earlier results in fixed cells.

A The ErkKTR reporter indicates ERK activity through nuclear-cytoplasmic translocation of a fluorescent protein. B Sensitivity of single cells (STE-1) to 15 min EGF stimulation. Plot shows fold change of ERK activity in single cells upon stimulation with the indicated amount of EGF in the presence or absence of ALKi (crizotinib, 1 µM). Boxplot shows median and upper/lower quartiles, and whiskers extend to 1.5*IQR. Significance tests indicate increased response above 0 ng/mL EGF. ****p n = 316, 384, 421, 285, 307, 256, 294, 421 cells respectively for DMSO and 473, 727, 536, 595, 541, 425, 439, 415 cells respectively for ALKi. See Supplementary Movie 3. C Live-cell imaging of STE-1 cells expressing ErkKTR in the presence or absence of ALKi (1 µM crizotinib). See Supplementary Movie 4. D Representative single-cell traces of cytoplasmic/nuclear ErkKTR intensity ratio from conditions shown in (C). E Quantification of ErkKTR activity in the presence of ALKi or its combination with EGFRi (1 µM erlotinib) or MMPi (10 µM marimastat). F Quantification of ERK activity pulses. Boxplot shows median and upper/lower quartiles, whiskers show 1.5*IQR. Significance assessed by Hsu MCB test. n = 177 (control), 198 (ALKi), 170 (ALKi/EGFRi), and 182 (ALKi/MMPi) single cells. G Percent of cells that exhibited any pulses over 22 h of imaging. Error bars indicate 95% CI. Significance assessed by the Hsu MCB test. n = 200 cells per condition. H Apoptotic cells secrete paracrine EGFR ligands to their neighbors. Paracrine signaling can be blocked by inhibiting either EGFR or the MMPs that mediate shedding of EGFR ligands from the surface of the sender cell. I ErkKTR activity pulses are primarily observed surrounding a dying cell during ALK inhibition but not in the absence of drug or in the added presence of EGFR or MMP inhibitors. J Definition of neighbors and non-neighbors of a death event. K Quantification of pulses per cell for each death event in neighbors or in equal number of randomly chosen subset of cells not near a death event (see “Methods” for more details), n = 91, 103, 149, 20 events for ALKi, ALKi/EGFRi, ALKi/MMPi and DMSO, respectively. Boxplot shows median and upper/lower quartiles, whiskers show 1.5*IQR. L Fraction of total neighbor vs random non-neighbor cells that show any pulsing. Cell numbers as in (F). Error bars show 95% CI. Significance in (K, L) determined by independent T-tests (within treatment conditions) or by ANOVA followed by the Hsu MCB test (across treatment conditions).

We then observed ERK activity after ALKi treatment in the absence of exogenous stimulation. In untreated STE-1 cells, ERK activity was at an intermediate level, indicated by relatively equal reporter distribution between the cytoplasm and nucleus (Fig. 6C, top row, Supplementary Movie 4). Notably, this localization pattern did not change over the course of 22 h, reflecting the tonic signaling downstream of EML4-ALK. As anticipated, treatment with ALKi induced a rapid initial decrease of ERK activity (Fig. 6C, bottom, 5 m and 30 m panels). Strikingly, this initial decrease was followed by the appearance of sporadic ERK activity pulses within ~1–2 h of treatment (Fig. 6C, 2h20m panel, Supplementary Movie 4). Each ERK pulse lasted 10–20 min, and the pulse amplitude exceeded ERK activity levels observed in the absence of drug (Fig. 6D).

Activity pulses appeared sporadically but in a spatially coordinated manner, appearing either simultaneously or as a traveling wave within small clusters of neighboring cells (Fig. 6C, Supplementary Movie 4). This pattern was consistent with RTK stimulation through paracrine signaling. To test this hypothesis, we sought to block paracrine signals through inhibition of either the EGF receptor (EGFRi, 1 µM erlotinib) or matrix metalloproteases (MMPi, 10 µM marimastat), which release EGFR ligands from the cell surface to enable paracrine signaling^43,44,45. Co-treatment with ALKi and EGFRi eliminated ERK pulses after drug addition, indicating that EGFR activation causes the observed ERK pulses (Fig. 6E–G, Supplementary Movie 4). Similarly, co-treatment with ALKi and MMPi also reduced ERK reactivation pulses, though to a lesser extent than with EGFRi, potentially due to MMP-independent juxtacrine signals^46,47(Fig. 6E–G, Supplementary Movie 4). Thus, ALKi treatment decreases ERK activity but is rapidly followed by RTK reactivation mediated by paracrine signals.

Signal reactivation results from paracrine signals from dying cells

We next sought to determine the source of the paracrine signals. We observed that pulsing events appeared next to dying cells and that co-inhibition of ALKi with EGFRi or MMPi prevented this pulsing (Fig. 6H, I, Supplementary Movie 4). These observations are consistent with paracrine ligand secretion from apoptotic cells, which promotes survival of neighboring cells and homeostasis of epithelial sheets^48,49,50. To quantify this effect, we measured signal activation in cells that neighbored a dying cell within the ~hour preceding its death (Fig. 6J). We then counted ERK pulses in these neighbors (N) and compared pulse counts to those from randomly selected non-neighbors (R) over that same time interval (for more details, see “Methods”). In untreated cells, pulses were almost never observed in either the N or R populations (DMSO, Fig. 6K, L), demonstrating the need for ALK inhibition for RTK/Erk activation. However, in ALKi-treated cells, N cells pulsed significantly more than R cells (Fig. 6L). Co-treatment with either EGFRi or MMPi dramatically reduced ERK pulsing in neighbors (Fig. 6L). Ligand shedding during apoptosis is triggered by mitochondrial outer membrane permeabilization (MOMP) and thus cannot be blocked by inhibition of apoptotic caspase cleavage, which occurs downstream of MOMP⁴⁸. Accordingly, co-treatment of cells with ALKi and a caspase inhibitor suppressed apoptosis but did not suppress ERK pulsing, in line with MOMP-induced ligand shedding (Supplementary Fig. S9C–E, Supplementary Movie 5). Together, our results demonstrate that virtually all observed ERK reactivation is associated with paracrine signals associated with dying drug-treated cells. Importantly, because ERK pulses were not observed in untreated cells—even in neighbors of dying cells (Fig. 6I, K, L)—ALKi-induced RTK resensitization is an essential first step for the perception of paracrine ligands during ALKi therapy.

Signal reactivation pulses activate downstream gene expression

We next asked whether the short ALKi-dependent ERK pulses could impact cell behavior. We first asked whether ERK pulses led to downstream transcription. ERK activity controls the transcription of immediate early genes (IEGs) which begin transcription within minutes of ERK activity⁵¹. EGR1 is an IEG that has been implicated in drug resistance to ALK inhibitors¹⁶. Additionally, EGR1 expression is adaptive, such that its expression peaks by ~1 h but then decays within 1–2 h, even in the presence of constant upstream signal^29,52 (Supplementary Fig. S10A, B). Thus, the accumulation of EGR1 indicates the presence of only recent ERK activation⁵³. We examined the extent to which EGR1 accumulated in STE-1 cells upon drug treatment to understand whether drug-induced ERK pulses could drive transcription (Fig. 7A). In untreated cancer cells, EGR1 levels remained low despite high ERK signaling from EML4-ALK, consistent with EGR1 adaptation to tonic ERK signals (Fig. 7B, C). By contrast, ALK inhibition resulted in a distinct peak of EGR1-high cells that appeared 4 h after drug treatment. Co-inhibition of ALK and EGFR prevented the appearance of EGR1-positive cells, consistent with transcription resulting from paracrine signaling through EGFR (Fig. 7B, C, Supplementary Fig. S10C). Similar responses were measured in H3122 cells as well as in two additional primary patient-derived EML4-ALK(V1)+ cell lines (Supplementary Fig. S10D–I). Thus, ALKi-induced ERK activity pulses provide sufficient signal to drive gene expression changes that could regulate cell fate.

A Signaling through EGFR activates RAS/ERK and stimulates transcription, including of the immediate early gene EGR1. B Quantification of single-cell IF of EGR1 in STE-1 cells under the conditions indicated. C Overlay of EGR1 expression at the 6 h time point in (B). D Testing whether restored perception of paracrine signals can promote survival during ALKi treatment. E Quantification of pulses per cell in cells that died (D) or survived (S) through 22 h of imaging in the conditions where ERK pulsing could be observed. Boxplot shows median and upper/lower quartiles, whiskers show 1.5*IQR. Significance assessed by one-sided T-tests. n = 88 ALKi(D), 110 ALK(S), 136 ALKi/MMPi(D), and 46 ALKi/MMPi(S) cells. F Caspase-3 activation was assessed using the NucView reporter after 24 h treatment with the indicated drugs. Data points show proportion of 2300–3500 STE-1 cells and 3500–4500 H3122 cells. Significance assessed by one-sided T-test, n = 3 biological replicates. G, H DAPI imaging (left) and cell counts (right) of cell survival after 17 days of the indicated treatments in both H3122 (G) and STE-1 (H) cell lines. Significance assessed using one-sided T-test. n = 3 (H3122) and n = 4 (STE1) biological replicates. I Summary of the effects of drug-induced RTK resensitization.

Signal reactivation promotes acute drug tolerance and cell persistence during ALK inhibition

Finally, we asked whether ALKi-induced RTK resensitization and ERK pulses could counteract cell killing and promote tolerance to ALKi therapy (Fig. 7D). First, we observed that cells that survived over the first 22 h of treatment showed more ERK pulses compared with those that died (Fig. 7E, Supplementary Fig. S11A). To quantify cell death, we measured a fluorescent reporter of caspase-3 activity (NucView) after the first 24 h of drug treatment. While ALKi monotherapy led to increased caspase-3 over baseline, co-treatment with either EGFRi or MMPi significantly increased the caspase-3+ cell fraction in both STE-1 cells and H3122 cells (Fig. 7F, Supplementary Fig. S11B). Neither EGFRi nor MMPi treatment alone showed substantial killing over untreated cells. Enhanced killing was also observed when alectinib was used as the ALK inhibitor in combination with either EGFRi or MMPi (Supplementary Fig. S11C). Conversely, addition of 50 ng/mL EGF promoted survival in the presence of ALKi/MMPi in STE-1 cells (Supplementary Fig. S11D). Similarly, optogenetic pulses of optoFGFR lowered cell death in cells treated with combined ALKi/EGFRi treatment (Supplementary Fig. S11E–G). Finally, co-treatments of ALKi with either EGFRi or MMPi suppressed drug tolerance measured at 17 days in both STE-1 and H3122 cell lines (Fig. 7G, H), consistent with prior reports that demonstrated the anti-tumor efficacy of ALKi/EGFRi co-treatment in cell and xenograft models^13,14,16,54.

Our mechanistic studies also suggested that other clinically relevant drug combinations would suppress ERK reactivation and cancer cell survival. SHP2 is an intracellular phosphatase involved in transducing RTK signals^55,56 and co-therapies with SHP2 inhibitors can enhance response to ALK inhibitors^57,58. Paralleling our results with EGFRi and MMPi, SHP2i co-treatment with ALKi suppressed RTK reactivation pulses and enhanced cell killing, while SHP2i alone had little effect (Supplementary Fig. S12A–C).

Together, our results indicate that ALK inhibitors resensitize transmembrane EGFR and restore perception of paracrine ligands. Resensitized EGFR allows reactivation of survival signaling within minutes of drug treatment, which limits the cytotoxicity of ALK therapies and promotes drug tolerance, the first step towards acquired resistance (Fig. 7I). The molecular mechanisms we find to underlie these events likely contribute to previously observed synergies in multiple combination therapies and propose treatment co-targets to enhance therapy in EML4-ALK+ cancers.