Inhibition of mutant EGFR in lung cancer cells triggers SOX2-FOXO6-dependent survival pathways

1) The authors do not establish whether their observations regarding SOX2 are relevant to any important clinical observation or outcomes. The paper is introduced with a discussion of resistance, but the reported study does not address resistance. Does modulation of SOX2 have effects on drug resistance?

We now show new data in Figure 5E, demonstrating that transient siRNA-mediated knockdown of SOX2, coincident with erlotinib exposure, decreases the subsequent development of stable, acquired resistance to erlotinib. While continuous culture of control PC9 cells in erlotinib for two weeks leads to the emergence of erlotinib-resistant clones, the number of resistant colonies is significantly decreased when SOX2 induction by erlotinib is prevented with siRNA.

2) To further relate the findings to drug resistance, the authors should determine what occurs following a repeat challenge with erlotinib. The authors show that SOX2 induction is transient. Does the proportion of cells that can increase SOX2 following erlotinib exposure increase in the surviving cells (i.e. does erlotinib select for cells with this phenotype)? If not, then these data may be describing some form “fractional kill” or stochastic resistance, akin to that shown in Spencer et al., Nature, 2009. This would also be interesting, but not really related to the clinically observed problem of acquired resistance to erlotinib and other similar inhibitors.

We undertook the requested retreatment experiments and we show in the new Figure 2B for HCC827 cells (and in Figure 2–figure supplement 4A for PC9 cells) that re-challenge with erlotinib after a period of recovery from two prior treatments (75-90% cell killing for each treatment), does not increase the proportion of cells with SOX2 inducible expression. Instead, SOX2 expression returns to basal levels in the absence of drug, and re-treatment re-induces SOX2 to similar levels and with similar cell distribution as with the prior treatment. These data are consistent with a stochastic model of SOX2 induction. As noted by the reviewer, a similar stochastic cell killing model was described in Spencer et al, Nature, 2009 with respect to TRAIL-induced apoptosis. However, this in no way suggests absence of clinical significance, since (as shown in Point #1 above), the transient expression of SOX2 itself enhances the subsequent emergence of stably resistant colonies. Progressively increasing fractions of SOX2 inducible cells following each erlotinib treatment are not expected in this model.

We also now show in the new Figure 2–figure supplement 4B, that in PC9 cells that have developed acquired resistance following long-term, continuous erlotinib treatment, stochastic induction of SOX2 still occurs but only at the much higher doses of erlotinib required to inhibit EGFR in these resistant cells.

Taken together, these data support the heterogeneous, transient and stochastic induction of SOX2 following erlotinib treatment as a precursor to the development of stable acquired drug resistance.

3) The data presented show that SOX2 expression is not only heterogeneous, but also very rare. In Figure 2, for example, it appears that < 20% (maybe less than 10%) of cells express SOX2 after erlotinib exposure. Thus, it is unclear why removing SOX2 from ∼10-20% of cells would have the pronounced effects on erlotinib-mediated death that are shown in Figure 4A-B. The dose of erlotinib used in Figure 2 is very low. Will the proportion of cells with SOX2 induction increase as a function of erlotinib dose?

In our experiments, the fraction of SOX2+ cells varies from 17-24% for HCC827 cells and 18-32% for PC9 cells (see Figure 2 and Figure 2–figure supplement 1). Given the threshold for immunofluorescence detection, it is likely that our immunofluorescence assay underestimates the number of cells with SOX2, and as such we propose a heterogeneous range of expression, rather than a truly rare and unique subset of the population. It is also difficult to determine precisely what level of SOX2 expression in an individual cell is required for a measurable effect on cell survival. We note that unlike the single cell data in Figure 2A, the new Figure 5B shows an immunoblot assay using total cell lysates, which obscures the contribution of individual cells to the apoptotic effect of SOX2 knockdown.

As suggested by the reviewers, we also increased the dose of erlotinib beyond that initially tested. We show in the new Figure 2–figure supplement 1, that further increasing the dose of erlotinib beyond that required for full inhibition of EGFR signaling to 1.0μM does not substantially increase the fraction of SOX2+ HCC827 cells (Increased cell death in PC9 cells at the highest concentration precludes acute SOX2 induction analysis).

4) At least 70% of patients who acquire resistance to EGFR inhibition do so on the basis of identifiable new mutations in EGFR or MET. I know the authors discuss a sort-of resistant state that SOX2. The authors should examine whether modulation of SOX2 affects selection for mutants in vivo.

The reviewers bring up the important point that stably acquired resistance to EGFR inhibitors can take many forms, from secondary gatekeeper mutations within EGFR, to induction of MET or other “bypass tracks”, to epithelial-mesenchymal transition (EMT) and even lineage switching from non-small cell to small cell histology. All of these complex mechanisms are noted in patients after prolonged treatment with erlotinib, but only few of these are well recapitulated using the standard cell line models, which tend to favor the gatekeeper mutation mechanism (even MET amplification is very rare in vitro). There are ample previous publications in this area. Given its complexity, we feel that defining the precise mechanism of long-term resistance in various PC9 and HCC827 clones, and the extent to which the various frequencies might be altered in the rare colonies that do ultimately emerge despite SOX2 suppression, would be beyond the scope of this manuscript.

5) BIM is identified as a key player in SOX2 signaling based on SOX2 over-expression assays (Figure 5A). However, SOX2 siRNA causes very small effects on BIM protein expression (Figure 7D). Since erlotinib causes major loss of ERK activity (Figure 7D), this would be anticipated to cause BIM stabilization. One interpretation of Figure 7D is therefore that SOX2 is not very important for BIM protein expression in this system. Figure 4C using SOX2 over-expression would appear to support this conclusion (although in that experiment, it is not clear why SOX2 is not induced by erlotinib in the control cells, but oddly is induced in the SOX2 over-expressing cells, which appear not to express SOX2 at time = 0).

The reviewers correctly point out that the activity of BIM is controlled at several levels—both transcriptional and post-translational (in the latter case, primarily by ERK-dependent phosphorylation/stabilization). Our data suggest transcriptional regulation by SOX2 as an additional mechanism that contributes to BIM regulation, but it is clearly not the only one (for example, the loss of MAPK pathway activity following erlotinib treatment likely leads to stabilization of the pool of previously transcribed BIM, that would be unaffected by SOX2 modulation of new BIM transcription). However, the effect of manipulating SOX2 levels on the degree of apoptosis clearly demonstrates that in certain cellular contexts, SOX2 can have an important, functional role.

We have replaced Figure 4C with the blot in Figure 5A (previously a supplemental figure in the original manuscript), so as to remove any confusion about how the experiment was performed. In this experiment, doxycycline was added for two hours to induce exogenous SOX2 and then removed prior to the addition of erlotinib for 24 hours, followed by preparation of lysates and immunoblot with the indicated antibodies. Here, both induction of physiologic levels of exogenous, epitope-tagged (and slower migrating) SOX2 by doxycycline, and induction of endogenous SOX2 by erlotinib, can be clearly appreciated, still resulting in decreased apoptotic PARP and Caspase-3 cleavage. We have moved the original Figure 4C to Figure 5–figure supplement 1B and clarified how the experiment was done in the accompanying legend.

6) BIM and BMF are identified as SOX2 target genes. Evidence to support this conclusion is required (e.g. ChIP assays).

As requested by the reviewers we have performed ChIP assays. In the new Figure 6B, we use CHIP seq and ChIP qPCR to demonstrate erlotinib-induced binding of endogenous SOX2 to upstream regions of both the BIM and BMF genes.

7) The conclusion that FOXO6, but not other FOXO family members, mediates effects of erlotinib on SOX2 expression is not fully established. First, redundancy of FOXO isoforms is not considered, this is especially a problem since the efficiency of FOXO1/3a/4 knockdown is not presented. Second, ChIP assays are required to test the chromatin interactions of these FOXO proteins in the cells employed in this study. The connection to FOXO6 would be strengthened if it were shown that FOXO6 function was heterogeneous and matched SOX2 expression.

The efficiency of knockdown for all FOXO siRNAs in Figure 3 is shown in the new Figure 7B using qPCR. For all FOXO family members where good quality antibodies are available, we have also added Western blot assays (new Figure 7–figure supplement 2B). We agree with the reviewer’s comment about the redundancy of FOXO isoforms, and this is shown in multiple FOXO knockdown experiments (Figure 7B). However, the triple knockdown of FOXO 1, FOXO3a and FOXO4 does not decrease SOX2 induction like single FOXO6 knockdown.

To further support the role of FOXO6 alone, we use four different siRNAs to confirm that its down regulation suppresses SOX2 induction by erlotinib in both HCC827 and PC9 cells (Figure 7D).

As requested by the reviewers, we show in the new Figure 7–figure supplement 3 that FOXO6 expression is heterogeneous, and correlated with SOX2 expression.

We were unfortunately unable to demonstrate binding of FOXO6 to the promoter of SOX2 by chromatin immunoprecipitation, most likely due to the poor quality of available antibodies, since ChIP seq tracks for FOXO6-bound chromatin were inadequate across the genome.

8) There are a number of technical concerns with the analysis presented. First, the siRNA studies are poorly described, for example, the siRNA are not defined. Second, important controls are missing. Only one siRNA is presented for each knockdown and no rescue studies were performed. Evidence that the siRNA caused a knockdown is missing; for example in Figure 3. In other cases the extent of knockdown appears limited (e.g. FOXO6 in Figure 3C). Since the study is highly dependent on the use of siRNA approaches, these questions limit the impact of the study.

We apologize for these omissions. siRNA target sequences are now clearly defined in the Methods sections of the manuscript. The new Figure 5C demonstrates dramatic rescue of the apoptotic effect of the most potent siRNA targeting SOX2 by a siRNA resistant, exogenous SOX2.

Figure 7C also now demonstrates that four independent siRNA that effectively target FOXO6 decrease erlotinib-induced induction of SOX2 (shown above under point 7).

The degree of FOXO6 knockdown in Figure 3C at the protein level is 50% (despite 80-90% knockdown at the mRNA level), which may be due technical issues related to the antibodies used (see Discussion about ChIP for FOXO6 above), or to the relatively long half-life of the protein (we waited up to 96 hours after siRNA transfection to assay for knockdown). Consistent with some residual FOXO6 protein remaining after knockdown, we observed significant, though incomplete suppression of SOX2 induction by erlotinib (see Figure 7D).

[Editors' note: further revisions were requested prior to acceptance, as described below.]

We consider that the manuscript has been improved by the inclusion of new data and revisions to the text. However, there are a number of points in the original review that were not adequately addressed, specifically:

5) Effect of SOX2 on BIM protein expression. Figure 4C (now Figure 5–figure supplement 1B) shows that the large increase in SOX2 expression does not appear to cause major changes in BIM expression. Moreover, why is BIM expression in lane 5 slightly greater than in lane 1? These data do not make a strong case for a role of SOX2 in the regulation of BIM protein expression. This is consistent with the modest effects of SOX2 knockdown to increase BIM protein expression (Figure 5–figure supplement 4). The authors need to clarify the experimental basis for their conclusion that BIM protein contributes to the effects of SOX2.

The critical point about BIM is its relative induction compared to baseline, rather than a comparison of basal levels across two panels. As we show quantitatively in Figure 5–figure supplement 1B (together with the relevant protein bands from the referenced supplementary figure), the levels of each BIM isoform are clearly increased by 12 hours of treatment with erlotinib in the absence of ectopic SOX2 (blue lines). The erlotinib-mediated increase in BIM is clearly suppressed over the same time period by the simultaneous addition of erlotinib and ectopic SOX2 induction (red lines). The slightly increased basal BIM levels at time zero in lane 5 versus lane 1 noted by the reviewer are most likely due to experimental variation and are not relevant to the apoptotic effects observed.

This quantitation is now included with the revised Figure 5–figure supplement 1B. In addition, we in the revised Figure 5–figure supplement 1C a different experiment using tagged, ectopic SOX2 and extend the time of treatment to 24 hours. Here, the levels of ectopic SOX2 induction (lanes 5-6) are comparable to endogenous SOX2 induction by erlotinib (lane 2-3); decreased BIM levels are clearly apparent (the use of a lower percentage acrylamide gel does not permit BIM L and S to be distinguished from EL).

To address the question of large changes in SOX2 expression leading to smaller changes in BIM expression, we would not expect a linear relationship between arbitrary “levels” of SOX2 versus BIM. What is important here is the functional consequence of BIM and BMF induction (and of preventing their induction by siRNA or ectopic SOX2), which we clearly demonstrate.

As to the major data relating to SOX2 and BIM, we note the following: in our manuscript, we claim that the transient induction of SOX2, following acute suppression of mutant EGFR signaling, reduces induction of two pro-apoptotic proteins, BIM and BMF, thereby reducing cell death and setting the stage for acquired drug resistance to EGFR inhibitors. This is based on several lines of evidence:

(1) Potent suppression of BIM and BMF by ectopic SOX2 (Figure 5–figure supplement 1B, Figure 6A and its associated figure supplement 1, together with the new data above).

(2) Increased BIM following SOX2 knockdown in the presence of erlotinib (Figure 5–figure supplement 4).

(3) Direct, erlotinib-dependent binding of endogenous SOX2 to both BIM and BMF regulatory regions by ChIP assays (Figures 6A and B).

(4) The functional consequences of this SOX2-mediated effect is demonstrated by the cooperative suppression of erlotinib-induced apoptosis, when both BIM and BMF induction are prevented using siRNA (Figure 6C).

7) The conclusion that FOXO6 mediates the effects of erlotinib on SOX2 expression is not well supported by the data, because it appears that the RNAi data presented is confounded by off-target effects. Thus, siFOXO6 causes decreased expression of all other FOXO mRNA tested (Figure 7B) and siFOXO1/3a/4 causes decreased FOXO6 mRNA expression (Figure 7B). These data do not allow the authors to distinguish between the roles of different FOXO proteins (and raise questions concerning what other genes might be affected). The authors do present additional data to address this point (Figure 7D) by using more than the single siRNAs employed elsewhere; but no controls for off-target effects on other FOXO proteins are presented.

Our critical finding is that only knockdown of FOXO6 significantly decreases SOX2 induction by erlotinib. Given the reviewer’s concern about possible nonspecific effects of the siRNA pool targeting FOXO6, we have now added data on different FOXO isoform levels for each of the four individual siRNAs that comprised the FOXO6 pool (see the revised Figure 7–figure supplement 2C). Two different siRNA duplexes (-01 and -02) are uniquely targeting FOXO6, without effect on other FOXO family members. Two other siRNAs within the pool (-03 and -04) do have some off-target effect on FOXOs1, 3a and 4. Note in the new figure that with 24 hours of erlotinib treatment, FOXO6 is induced by erlotinib to a much greater degree than the other FOXOs in both HCC827 and PC9 cells, hence these off-target effects are less impressive than the on-target effect on FOXO6. This new data suggests that the nonspecific effects of the FOXO6 siRNA pool on other FOXOs are likely due to duplexes -03 and -04.

In addition to two independent siRNAs (-01, -02) being highly specific in knocking down FOXO6 and reducing SOX2 expression, it is key to also note that we have tested the other FOXO family members directly for their effect on SOX2 expression (Figure 7B). siRNAs that effectively knock down either FOXO1, FOXO3a or FOXO4 have no significant effect on SOX2. Hence, the off target effects of FOXO siRNAs -03 and -04 on other FOXOs are not responsible for suppression of SOX2 expression.

In addition to addressing the reviewers’ concern regarding siRNA off-target effects, the role for FOXO6 is further supported by our demonstration of FOXO6 co-localization with SOX2 (Figure 7–figure supplement 3), as requested by the reviewers of our initial submission. With respect to other pathways, we used multiple approaches to test several other (e.g. non-FOXO) pathways implicated in SOX2 regulation, including FGF, WNT and NKX2.1 (Figure 7–figure supplement 4), and none of these affected SOX2 induction by erlotinib. Therefore, our conclusion, that FOXO6 is the primary mediator of erlotinib’s effect on SOX2 expression, is well supported by the data.

We thank the reviewer/editors for this opportunity to address the remaining concerns from our resubmitted manuscript. Together with the extensive additional data and edits provided in our revised manuscript, we provide multiple lines of evidence that SOX2 induction contributes to upfront and acquired resistance to erlotinib in EGFR-mutant lung cancer cells. These include:

(1) Enhancement of acute erlotinib-induced apoptosis and suppression of long-term acquired resistance, when SOX2 induction is prevented with siRNA (an effect that is completely rescued by a SOX2 construct that is resistant to siRNA targeting).

(2) Detailed analysis of the heterogeneous, transient and stochastic nature of SOX2 induction in multiple EGFR-mutant cell lines, xenografts and patient-derived models.

(3) A proposed mechanism for both mutant EGFR’s regulation of SOX2 through FOXO6, and SOX2’s regulation of apoptosis through BIM and BMF.

(4) A demonstration of the functional importance of high basal SOX2 for upfront erlotinib resistance, in addition to acquired drug resistance.

The critical importance of acute and transient signaling feedback pathways for the development of resistance to targeted anti-cancer therapies has been established for several oncogene-addicted cancers (primarily mutant BRAF-driven melanoma and colorectal cancer). We show that similar pathways are involved in EGFR-mutant lung cancer. These studies, by their very nature, may be different from well-characterized developmental roles of FOXO and SOX genes, but their adaptation in cancer cells and their contribution to acquired resistance to signaling inhibitors is of significant biological interest and high clinical relevance. Given the scope of our observations and the readily addressed points raised by this secondary review, we are most grateful for your consideration.