Author's response to reviews

Author's response to reviews Title: Associations between toenail arsenic accumulation and dietary factors in a New Hampshire populationassociations between toenail arsenic accumulation and dietary factors in a New Hampshire population Authors: Joann F Gruber (joann.gruber@unc.edu) Margaret R Karagas (margaret.karagas@dartmouth.edu) Diane Gilbert-Diamond (diane.gilbert-diamond@dartmouth.edu) Pamela J Bagley (pamela.bagley@dartmouth.edu) Vicki Sayarath (vicki.sayarath@dartmouth.edu) Tracy Punshon (tracy.punshon@dartmouth.edu) M Scot Zens (m.scot.zens@dartmouth.edu) J Steven Morris (MorrisJ@missouri.edu) Kathryn L Cottingham (kathryn.cottingham@dartmouth.edu) Version: 2 Date: 1 March 2012 Author's response to reviews: see over

29 February 2012 Dear Editor-in-Chief: Dartmouth College Department of Biological Sciences Class of 1978 Life Sciences Building, 78 N. College St. HANOVER NEW HAMPSHIRE 03755 TELEPHONE: (603) 646-0216 FAX: (603) 646-1347 Please find enclosed our revised manuscript entitled, Associations between toenail arsenic concentration and dietary factors in a New Hampshire population (MS: 8948667605919056). We were encouraged by the generally positive reviewer comments received in November 2011, and have taken nearly all of their comments into account in making our revisions. In the rest of this letter, we briefly explain how we have addressed each concern. The reviewer comments appear as bold font, and then our responses follow. In addition, to help make our changes easier to locate, we have enabled continuous line numbering in the Word document containing the manuscript and use these line numbers to direct the editor and reviewers to our changes. We have also used yellow highlights in the text to identify changes made All authors have read and approved these revisions. The manuscript has not been published previously nor is it being considered by any other peer-reviewed journal. We hope that you find this revision suitable for publication in Nutrition Journal and look forward to hearing from you. Sincerely, Kathryn L. Cottingham, Professor Department of Biological Sciences, Dartmouth College 78 N. College St., Hanover, NH 03755 Email: kathryn.cottingham@dartmouth.edu, Phone: 603-646-0216, FAX: 603-646-1347 Joann F. Gruber, Graduate Student UNC Gillings School of Global Public Health Email: joann.gruber@unc.edu, Phone: 785-288-0166.

DETAILED RESPONSE TO REVIEWERS Megan Hall Minor Essential Revisions The finding that many dietary lipids were inversely associated with toenail arsenic is unexpected and somewhat surprising. The authors speculate that: "From a physiological perspective it is conceivable that fat could form a complex with arsenic, inhibiting its absorption and resulting in decreased exposure and decreased accumulation in toenails. For example, lipid particles associated with proteins can interact with arsenic, which could provide an alternative mechanism for detoxification." There is an equally plausible explanation for these findings that deserves mention. The authors adjusted for total energy intake in multiple regression models using the "Standard Multivariate" approach, i.e., by including total energy intake as a covariate in the model. With this approach, the beta coefficient for dietary lipids represents the effect of substituting dietary fat for calories from other sources (i.e., protein and carbohydrates) - see Willett, Nutritional Epidemiology text, 1998, Chapter 11. This might suggest that protein and carbohydrate rich foods contribute more to dietary arsenic exposure in this population than do foods higher in dietary fat. We thank Dr. Hall for this suggestion, and have added the following sentence to the Results and Discussion: Another plausible explanation is that foods with higher fat content contribute relatively less dietary arsenic compared to carbohydrate- and protein-rich foods (such as rice and fish) resulting in negative correlations in our energy adjusted analysis (lines 246-248). Tejo Prakash N Nagaraja Minor Essential Revision The authors need to provide a brief outline of how NAA analysis was carried out for Arsenic estimation or how FA analysis was carried out for dietary components, although citation with references associated with methodology are given. This is important as the analysis has been carried out after sample collection way back in 1998, and the standards/protocols have varied during the time that has passed. We thank Dr. Nagaraja for pointing out the brevity in the earlier manuscript, and agree that providing a reference to the original Karagas et al. study, which in turn cites the technical references for neutron activation analysis, was not ideal. We now provide the reader with a bit more detail on the overall approach (lines 119-129), together with a specific reference for the detailed NAA procedures to make this information easier to locate for readers (line 126). However, due to the highly complex nature of the NAA analyses and the fact that the toenail arsenic data has been previously published, we have chosen not to describe the NAA procedures in detail in this manuscript. 2

We have also clarified that the analyses were generally done contemporaneous with the toenail sampling (lines 123-124). We did not do FA analysis as part of this analysis, but instead used the FFQ and nutrient databases used by the Nurses Health Study (Willett et al., 1987). We now provide additional information to reflect this in the text (lines 113-115). Melissa Slotnick Major revisions I would recommend separating out results and discussion, and I feel that the discussion needs to be elaborated on significantly. We thank Dr. Slotnick for her many helpful suggestions. We have opted not to separate the Results and Discussion into separate sections, as it made the manuscript much longer due to the need to repeat subheadings and reorient the reader to each result. Instead, we have made the recommended changes within our joint Results/Discussion section, which we hope will prove an adequate compromise. Specifically, since there is so much speculation about why these dietary factors may influence toenail arsenic concentrations, there should be some discussion about how arsenic is incorporated into the nail (and, consequently, a bit about arsenic metabolism). It seems that this is a missing link that might tie in some of the speculation. We have expanded the paragraph in the Background section (lines 66-75) to include more details on the metabolism of arsenic, especially the affinity of inorganic arsenic for keratin in nails and hair. Additionally, the term arsenic accumulation is used in reference to toenail arsenic levels. I think one should be careful with this wording, and discuss and/or clarify if the toenail is measuring accumulation, excretion, or exposure. This is a good point (and was also raised by Dr. Steinmaus). We have changed the word accumulation to concentration throughout the manuscript, when referring to concentrations of arsenic in toenails. Additionally, it is important to discuss some of the limitations of the study. In particular, the level of error in the dietary assessment is a bit concerning and needs to be addressed. Toenails, in theory, measure arsenic exposure occurring over a longer period of time which is assumed to be relatively constant if drinking water sources haven t changed significantly. Dietary exposures can change dramatically over a period of a few months. 3

Additionally, the timing of the dietary assessment versus the timing of the toenail clipping (and the period of exposure it represents) should be addressed. We have added a new section to the Results and Discussion (lines 299-323) that mentions the limitations brought up by Dr. Slotnick and Dr. Steinmaus. We have also added a statement in the Methods that clarifies the relative timing of the food frequency questionnaire and the toenail collection (lines 121-123). Since the amount of arsenic in toenails represents approximately a one-year period (12-18 months, Fleckman 1985), the one-year FFQ should reflect the levels of dietary factors potentially influencing the arsenic concentrations during that year for the majority of the subjects. Were all of the analyses conducted on all 920 individuals? Did results change when looking at the 213 participants with water concentrations >1 mcg/l? Yes, the analyses reported in this manuscript include all individuals. Results did change somewhat when we focused on the 213 participants with arsenic concentrations >1 µg/l, in that not all of the effects detected for all subjects or the low-arsenic subjects only (<1 µg/l) were statistically significant for the higher-arsenic group. Thus, we generally attributed any observed differences to the reduced power for n=213 versus n=707 (subjects with <=1 µg/l) or n=920. To help clarify this point, we edited the sentence in the Methods referring to stratified analyses (lines 141-144) and added a short paragraph to the Results and Discussion explaining our findings for stratified data (lines 184-189). It would be interesting to see analyses examining whether or not the dietary factors assessed were modifying the association between the drinking water exposure and the toenail arsenic levels. We agree, and will consider this in future work. However, this analysis is outside the scope of our goals here regarding the relation between dietary factors and toenail arsenic. Paragraph 3 in the Background section needs to be expanded. As noted above, we have expanded paragraph three in the Background (lines 66-75) to include a brief overview of arsenic metabolism and different arsenic species. Minor revisions A visual (e.g., scatterplot) of some of the stronger associations may be interesting. We have added a scatterplot for total fat and alcohol, to help readers visualize the associations discussed in the text (lines 222 and 279). 4

What was the mean toenail arsenic concentration for participants with drinking water levels greater than 1mcg/L? To facilitate comparisons among drinking water strata for those interested in the details, we have revised Table 1 to provide covariate and toenail arsenic concentration data on the entire study population, those exposed to <1 µg/l of arsenic in water, those exposed to levels between 1 and 10, and those exposed to levels >10 µg/l. Were specifics on concentrations of nutrients in dietary supplements obtained? These concentrations and consumed doses can vary widely. To what extent might this influence results? There is certainly the potential for error in the FFQ estimates. The FFQ used in this study asks participants first about multivitamins, and obtains exact brand-name information and frequency of consumption. It then asks specifically about separate supplements both length of consumption and doses for individual vitamins and minerals (vitamins A, C, B6, E; selenium, iron, zinc, calcium). Finally, it asks for whether participants take other supplements, but without dose information: folic acid, vitamin D, B-complex vitamins, cod liver oil, omega-3 fatty acids, iodine, copper, brewer s yeast, beta-carotene, and magnesium. We believe any errors due to mis-estimation are likely conservative, since errors in the estimates of the predictor variables (like the dietary factors studied here) typically bias parameter estimates in a regression model towards zero (Draper & Smith 1998). The first sentence of the conclusion needs to reflect the focus of the discussion and results. This is an interesting sentence, but the conclusion wasn t really structured in a way to support this sentence. Good point. This was not a strong topic sentence for the ensuing content. We have revised it so that it better sets up the surprising results of our study (lines 326-329). In the background and conclusion relatively well-nourished population exposure to varying levels of arsenic, varying is a relative term. I would suggest replacing it with the actual concentration/exposure values. As it is written it suggests that there is wide variability, while the analyses were really conducted on individuals with relatively low exposures. This was a good suggestion that helps clarify our message. In the Abstract, we have changed the text to read, relatively low levels instead of varying (line 32 and 46) Likewise, we have changed varying to relatively low concentrations in the Conclusion (line 327). 5

Percentiles might be nice to include for exposure variables. There s a range of drinking water arsenic concentrations, but how many individuals fall into the higher exposure range? (e.g., >10 mcg/l) As mentioned above, we have provided some stratified data in Table 1 so readers can better understand the distribution of water and toenail arsenic concentrations in the population. It may be helpful for some researchers to see descriptive stats for the dietary intake variables (e.g., range of intake, mean, median, etc). This could help to facilitate crosspopulation comparisons. We have created a new Additional File 2, which provides a table of descriptive statistics about the dietary factors related to toenail arsenic concentration in this population. Craig M. Steinmaus Minor Essential Revisions It s not clear what the authors mean by the word accumulation. As noted in response to Dr. Slotnick s review, we have changed accumulation to concentration throughout the manuscript to avoid ambiguity. The past difficulties other researchers have had with measuring arsenic in toenails, especially in low exposure studies such as this one, is not discussed (Orloff et al., J Toxicol Environ Health B, 509-24, 2009). We have added a paragraph to the new Limitations section in the Results and Discussion to address this concern (lines 299-311). The large majority of subjects were cancer cases. It seems that this should be mentioned early in the paper, maybe even in the abstract. I don t see this as a major flaw (since most are skin cancers). However, it seems at least possible that diet-toenail arsenic relationships could be different in people with cancer than those without (e.g. from the cancer process itself, from chemotherapy/medications, from changes in diet ). Thus the generalizability of these findings to non-cancer subjects should be discussed. Simply putting case-control status in the statistical model would not explain any effect modification caused by cancer status. Good point. We added a mention of the case-control study to the Abstract (line 34), Methods (line 90), and Results and Discussion (lines 311-312; 321-323). 6

As a side note, we also tried some additional analyses in response to this comment. First, we looked for interactions between our case-control status indicator variable and each of the dietary factors we explored, and no statistically significant interactions were found. Second, we looked to see whether results for Controls only were similar to those reported here, and turned up only a few differences. In both cases, though, it is not clear what role the reduced power of n=277 played in these findings. The participation rates should be given. We have added some information about participation rates in the original case-control studies to the Methods (lines 104-106) as well as response rates for FFQ collection and toenail samples (lines 120-123). It s difficult to understand the practical meaning of their results. In other words, what does a B-coefficient of -5.8 x 10(-3) mean in terms of actual toenail arsenic levels. Does it mean a doubling, a tripling, and 1% increase of arsenic toenail as we go from the upper to lower quartile of total fat? The authors should help the readers interpret their findings in practical terms. This was an important suggestion for helping make our points clearer. We have added columns to Table 2 that provide the predicted percent change in toenail arsenic when moving from the first quartile to the third quartile for each of the dietary factors, for both males and females (see lines 163-172 in Methods, and related text in Results lines 198, 223-225, and 258-259). Several of the regression coefficients change dramatically from the unadjusted to the adjusted model. For fat, the coefficient changes from -1.5 x 10(-3) to -5.8 x 10(-3), almost a 400% increase in magnitude. Several other diet variables also have very big changes with adjustment. This 400% increase seems like a huge change. Can the authors identify what factor caused such huge confounding. Our explorations with the model and data indicate that these differences are primarily due to the inclusion of the categorical variable for BMI (which is not available for all subjects) and the continuous variable Calories in the adjusted model. Fortunately, our qualitative conclusions as to which dietary factors show significant associations and the magnitudes of effect of dietary factors (% change as discussed in the previous comment) are robust to the inclusion of these factors. As such, we have opted not to draw explicit attention to this issue in the manuscript. 7