Reviewer 1's Review (Heather Piwowar)

“The authors found that gene expression patterns of benign ovarian adenomas clustered separately from malignant untreated carcinomas, and that samples from neoadjuvant carcinomas collected at the time of surgery grouped with either cluster. The neoadjuvant carcinomas which clustered with the carcinomas were more apt to be associated with loss-of-function mutations in p53, and remarkably these patients trended towards enhanced long-term survival. The authors hypothesize that LOF mutations in p53 prevent p53-mediated cell-cycle arrest, thus reducing the ability of cells to repair DNA mutations caused by chemotherapy. This allows the death of these mutated cells via p53-independent pathways, eliminating their future contributions to tumor recurrence relative to being (potentially partially) repaired.

This paper would be of interest to three audiences:

* those researching the cause of chemoresistance in ovarian cancer (data, conclusions, and suggestions for future work)
* those looking for predictive genes and pathways in other domains, as the analysis is a nice model for looking beyond standard gene signature analysis for upstream underlying mutations (methods)
* those studying genome-wide patterns in other domains (raw microarray data)


The paper is easy to understand and well written for a broad audience. The science is well done, and the authors use a wide variety of techniques to investigate their hypotheses. The topic, understanding chemoresistance, is crucially important, particularly in advanced ovarian cancer. Finally, the authors reach interesting and potentially very useful conclusions.

Primary Issues

1. Please expand on a few of your techniques in the Methods section. In particular, add a description of the gene-profile overlap analysis, provide a bit more detail on the pathway analysis (such as what it means for a pathway to be "significantly altered" or any assumptions you made in determining what genes were downstream targets of p53), specify whether the survival p-value was calculated using a log-rank test or some other method.
2. The analysis does not consider other potential confounders. Do you have access to other prognostic variables, as in your previous paper (reference 26)? Particularly age and debulking status (as done in refs 6 and 10). Do these variables explain the noted difference in survival? If you are not able to do this analysis, mention as a limitation that there may be confounding.
3. It is not clear that the provided CA-125 information in Table 4 is evidence of a response to chemotherapy as stated, since the measurements also span a surgical intervention. Do you have a series of CA-125 measurements, as in ref 26? A demonstrated drop in CA-125 during chemotherapy treatment would be much more informative.
4. Please elaborate on the relationship between your results and theories and the current literature, including for example the following articles:
* Mayr D et al. Immunohistochemical analysis of drug resistance-associated proteins in ovarian carcinomas. Pathol Res Pract. 2000;196(7):469-75
* Andrews GA et al. Mutation of p53 in head and neck squamous cell carcinoma correlates with Bcl-2 expression and increased susceptibility to cisplatin-induced apoptosis. Head Neck. 2004 Oct;26(10):870-7.
* Osugi H et al. p53 null mutations detected by a p53 yeast functional assay predict a poor outcome in young esophageal carcinoma patients. Int J Oncol. 2002 Sep;21(3):637-41.
* Keshelava N et al. p53 mutations and loss of p53 function confer multidrug resistance in neuroblastoma. Med Pediatr Oncol. 2000 Dec;35(6):563-8.
5. Since the clusters are a major foundation of the paper, it is important to verify the robustness of the clustering results. Do the major clusters remain stable if the samples are reordered,resampled, or if a different weighting metric is used?
6. As the LOF findings are key to your conclusions, it would be appropriate to do a statistical test to confirm whether your observations are significantly different from chance. For example, a Fishers test on presence ofLOF mutation vs clustering group for the CA samples.
7. To aid future researchers (and address PLoS publishing guidelines http://www.plosone.org/st...), please annotate gene names and make the following raw data available: microarray CEL files, survival information, additional clinical covariates such as stage and debulking status as in table 1 in ref 26 (thank you for making some of the CA-125 available in your initial draft). It would be helpful if you also provided the results of your p53 staining and p53 LOF sequencing, indexed by sample ID. Providing the names of the 9106 probe sets in a text file would also increase the reusability of your results.
8. Please be more explicit that the survival benefit is not statistically significant.
9. Mention some specific future directions for validating your concluding hypotheses.
10. Finally, some of your results seem to contradict those of your pilot study, in which the neoadjuvant patients whose samples clustered with benign adenomas had a longer progression-free interval than the others. Please speculate into reasons behind this difference, since a lack of reproducibility plagues microarray studies.

I enjoyed reading the paper, and thank the authors for their work.”

n.b. These are the general comments made by the reviewer when reviewing the originally submitted version of this paper. The manuscript was revised before publication. Specific minor points addressed during revision of the paper are not shown.