The study uses cell lines established from multiple regions of the same tumor to describe identification of PI3Kb as a mediator of resistance in ALK+ NSCLC. Use of multiple cell lines established from different regions of the same tumors represents a novel and interesting (though insufficiently developed) aspect of the study. The study reflects a large body of work, and could be a useful resources for the community.
However, there are substantial discrepancies between the data and inferences that the authors draw, including key conclusions. Moreover, some of the choices in data presentation makes the study unnecessarily difficult to follow.
Specifically,
1. It is unclear why the authors present their findings through the lenses of ALKi resistance. The cell lines were derived from ALKi naive tumor, and the manuscript assess sensitivity of these cells without deriving cells with lower sensitivity to ALK inhibition. Restoration of sensitivity (which authors use in the title) assumes that the tumors were initially more sensitive to ALKi. Most likely, authors refer to the very quick reactivation of downstream signaling, leading to reversal from initial sensitivity observed in short term assays to loss of sensitivity in longer term assays (see point #8). But this is not self-obvious. Of note, given the short term time course of this reversal, initial responses would not be noticeable clinically given the interval between diagnostic imaging.
2. From observation that EGF reduces sensitivity of ALK+ cell lines to ALKi authors conclude that "EGFR activation was involved in establishing resistance". In addition to assuming initially higher sensitivity (without sufficient data to back this up), another issue is that resistance to ALKi can be substantially reduced by other cytokines, such as HGF and FGF, across multiple cell line models of targetable NSCLC. Following the same logic as authors used for EGFR inferences, if authors were to perform an assay looking at the impact of HGF and FGF on ALKi sensitivity most likely they would have to add cMET and FGFR activation as "is involved in establishing ALKi resistance". Further, if EGF-EGFR axis is indeed responsible for resistance, what is the source of EGF (unless authors can show evidence that FR cell lines produce EGF in meaningful quantities)?
3. I do not see how a more tightly packed growth of TR5 derived cells can "corroborate" mesenchymal phenotype. It seems to directly contradict it. Typically, mesenchymal phenotype is associated with increased invasiveness and migration and more spread out patterns of cell growth, while tightly packed colonies characterize cells with stronger luminal differentiation. Difference in vimentin staining does indicate some sort of intermediate EMT phenotype, but this clearly does not go all the way to mesenchymal morphology. Authors should either modify the claim, or clearly articulate their reasoning.
4. Scale bars need to be included in the IHC/IF images.
5. I found inferences on differences in ALKi sensitivities between different cell lines unconvincing. The differences between distinct TR cell lines are relatively minor, and the viability assays results are not that easy to interpret without information on the differences in cell proliferation, as these assays reflect a combined effects of proliferation and death. For example, H228 cell typically shows as more resistant in these growth curve assays, but it might reflect its lower proliferation compared to H3122. High degree of clumping of TR5 and associated restriction of cell proliferation might be fully sufficient to account for the minor differences reported.
6. "To identify any intra-tumor heterogeneity" - I found both phrase to be very awkward, as it conveys a degree of desperation. On the other hand, this seems to be totally unnecessary, as in the preceding paragraphs authors report a clear evidence of heterogeneity in expression of EMT related genes, EML4-ALK fusion protein, and the overall cell morphology. These differences are much more obvious and convincing compared to the slight differences in ALKi sensitivity (see previous point).
7. Figure S2G reveals a remarkable divergence between TR3 and TR5 in terms of timing of timing and extent of activation fo different downstream signaling pathways. For example, pERK seems to be much higher at the baseline in TR5 compared to TR3, but in TR3 it reaches much higher than the baseline level by 48hrs, while staying more flat at lower than the baseline levels at TR5. I found it surprising that the authors see their results under an umbrella of "rebound activation", without commenting on the differences. Also, it would be useful to have quantitation of the differences, as eyeballing is somewhat difficult. Plus, I would like to know whether the effects reported in S2G are sufficiently robust to be reproducible.
8. Results of the 15 days assay in Figure 2D are obviously in stark contrast with the results of short term viability assay shown in Figure 2C. While authors do not discuss this explicitly, the implicit interpretation is that TR cell lines quickly develop resistance due to reactivation of downstream signaling. If so, this needs to be clearly brought to light, otherwise it is not obvious why the authors discuss resistance in the first place. It would be very useful to generate a time lapse data to get an idea of the timing of the reversal. WB data suggests that the reversal might be happening as early as 48 hours post treatment initiation, but there are no functional readouts beyond the two viability snapshots in Figure 2.
9. Authors use DSS as as the main metrics to compare therapy sensitivity throughout the paper. While there might be good reasons to use DSS as the metrics, DSS is not (at least yet) widely used in the community, which makes interpreting the results more challenging and less intuitive. Moreover, it is not self-obvious that integrating sensitivity across the range of concentrations with DSS would be more informative compared to focusing on the concentrations that are relevant to those experienced by tumor cells in vivo. Thus, my suggestion is to include a more conventional metrics (such as dose response curves, or with a single dose such as in Fig5C) at least as a supplementary data to help with interpreting the results.
10. The results of the xenograft experiment appear to contradict rather than to support the main claim of the paper, as ceritinib leads to reduction in tumor volume, while tumors show modest growth under the drug combo. Authors apparently try to resolve this by claiming that "tumors did not resume growth in the combination treatment group", but this statement clearly contradicts the data shown in Figure 4F, where tumors post combination group increase in volume, with the rate of the increase catching up to post-ceritinib group after an initial delay. And the observation of the delay in tumor growth cannot does not represent a strong evidence of lower tumor burden under combination therapy. Of note, in vivo results might still be consistent with in vitro studeis. There are well-established cases (though not in H3122 model), where tumors increase in volume during therapy even though the numbers of tumor cells are going down, due to a growing volume of necrotic regions. Perhaps this is indeed one of the cases of pseudoprogression, in which case it should be possible to reveal the effect through cellularity inferences, or through use of bioliminescently labelled tumor cells. On the other hand, it appears that authors decreased a chance of successful in vivo validation by choosing to work on H3122 model, where the in vitro effects of combination therapy are much more pronounced compared to the TR cell lines (Fig. 4A).

I have the following recommendation on addressing the key issues.
1. Re-focus presentation/discussion around the issue of sensitivity rather than resistance. This would also address my issue with the claim of EGFR being the resistance mechanism.
2. Explicitly describe reactivation as a quick loss of initial ALKi sensitivity, and perform time lapse microscopy assays (or multiple time points in Crystal Violet readout) to fill in the gap in temporal dynamics.
3. Ideally, perform xenograft validation studies with FR cell lines, where the combination therapy has a clear effect in vivo, so one would expect to observe an increased tumor regression. Using luciferase expressing tumor cells might make analyses easier. Authors might need to use NSG model, which is more permissive for xenograft growth for this purpose. A weaker, but reasonable alternative might be to explicitly acknowledge the inconclusive (and ostensibly contradictory) results with the H3122 xenografts. In either case, individual tumor data needs to be shown, and data on tumor volumes prior to normalization needs to be included.

The authors have adequately addressed the concerns; I do not think additional changes are necessary.