Mg, the most abundant intracellular divalent cation in the body, plays an essential role in over 300 biological activities( Reference Flatman 1 – Reference Hans, Chaudhary and Bansal 6 ). The major food sources of Mg are whole-grain foods, nuts, legumes, green leafy vegetables and deep-ocean fish( 7 ). Epidemiological studies have found low Mg intake to be related to an elevated risk of colorectal neoplasia in some( Reference Dai, Shrubsole and Ness 8 – Reference van den Brandt, Smits and Goldbohm 11 ) but not all studies( Reference Lin, Cook and Lee 12 – Reference Wark, Lau and Norat 14 ). One potential explanation for the inconsistency is that the interaction between Mg and Ca was not considered. Several studies have suggested that Ca and Mg may directly or indirectly compete for intestinal absorption( Reference Norman, Fordtran and Brinkley 15 , Reference Hardwick, Jones and Brautbar 16 ). Over 80 % of plasma Mg is ultrafiltrated and reabsorbed in the kidneys. A high Ca intake consistently leads to significantly increased excretion of Mg via urine( Reference Domrongkitchaiporn, Ongphiphadhanakul and Stitchantrakul 17 – Reference Karkkainen, Wiersma and Lamberg-Allardt 21 ). Thus, it is possible that long-term consumption of a high Ca:Mg ratio diet may lead to Mg deficiency, even in the context of adequate Mg intake( Reference Abrams, Grusak and Stuff 22 ).
We have reported that only when individuals consume diets with low Ca:Mg ratios (i.e. below median ratios between 2·6 and 2·8) intakes of Ca and Mg may be related to a reduced risk of colorectal adenoma( Reference Dai, Shrubsole and Ness 8 ), and Ca supplementation led to a reduced risk of adenoma recurrence in one randomised clinical trial( Reference Dai, Sandler and Barry 23 ). Further, Ca:Mg intake ratios modified the associations of dietary intakes of Mg and Ca and risks of mortality due to cancer and CVD in two large-scale population-based cohort studies conducted in Chinese populations with very low (median ratio of 1·7) Ca:Mg ratios( Reference Dai, Shu and Deng 24 ). Very recently, we reported potential interactions between dietary Mg with dietary vitamin D and serum 25-hydroxyvitamin D( Reference Deng, Song and Manson 25 ).
Previous findings from our Irish population-based case–control study found no association between Ca intake and risk of oesophageal adenocarcinoma, Barrett’s oesophagus or reflux oesophagitis, whereas high vitamin D intakes were unexpectedly associated with an increased risk of oesophageal adenocarcinoma( Reference Mulholland, Murray and Anderson 26 ). Given the emerging evidence of Mg as a chemopreventive dietary agent for colorectal neoplasia described above, and its intrinsic relationship with Ca( Reference Dai, Shrubsole and Ness 8 , Reference Norman, Fordtran and Brinkley 15 – Reference Dai, Shu and Deng 24 ) and vitamin D( Reference Deng, Song and Manson 25 ), we hypothesised that the previously observed associations for Ca and vitamin D intake may differ according to Mg intake. Furthermore, to our knowledge, no study to date has investigated Mg intake in relation to risk of these oesophageal disorders. Identifying potential risk factors for oesophageal adenocarcinoma is of strong public health relevance, given its rising incidence in Western populations and associated poor survival rates( Reference Edgren, Adami and Weiderpass 27 , Reference Bosetti, Levi and Ferlay 28 ). Differences in the prevalence of oesophageal cancer and gastro-oesophageal reflux disease between Eastern and Western populations lend further support to dietary risk factors potentially contributing to their aetiology( Reference El-Serag, Sweet and Winchester 29 , Reference Ferlay, Soerjomataram and Dikshit 30 ). Gastro-oesophageal reflux is a major risk factor for oesophageal adenocarcinoma and medications commonly used to treat reflux symptoms, such as proton pump inhibitors( Reference Alexandropoulou, van Vlymen and Reid 31 ), may lead to Mg deficiency( Reference Luk, Parsons and Lee 32 , Reference Markovits, Loebstein and Halkin 33 ). Therefore, it may be even more pertinent to intervene to achieve optimal Mg intakes in this patient group.
The primary aim of this investigation was to evaluate the association between Mg intake and risk of reflux oesophagitis, Barrett’s oesophagus and oesophageal adenocarcinoma. Secondary aims of this study were to evaluate the associations between intakes of the inter-related nutrients of Mg, Ca and vitamin D according to Ca:Mg intake ratios and the risk of oesophageal lesions.
Methods
Study design
Study participants were drawn from the Factors INfluencing the Barrett’s Adenocarcinoma Relationship study, an all-Ireland population-based case–control study established to investigate the aetiology of reflux oesophagitis, Barrett’s oesophagus and oesophageal adenocarcinoma( Reference Anderson, Johnston and Watson 34 – Reference Mulholland, Cantwell and Anderson 36 ). In brief, incident cases of oesophageal adenocarcinoma (n 227), long-segment Barrett’s oesophagus (n 224) and population controls (n 260) were recruited from Northern Ireland and the Republic of Ireland between March 2002 and July 2004. Reflux oesophagitis cases (n 230) were recruited from Northern Ireland only between 2004 and 2005. Barrett’s oesophagus, reflux oesophagitis and control subjects were frequency-matched within 5-year age and sex strata to oesophageal adenocarcinoma cases, upto a maximum age of 85 years. This study was conducted according to the guidelines laid down in the Declaration of Helsinki, and all procedures involving human subjects were approved by the Research Ethics Committee of the Queen’s University Belfast, Northern Ireland, Clinical Research Ethics Committee of Cork Teaching Hospitals and Research Ethics Committee Board of St. James’s Hospital, Dublin. All the subjects provided written informed consent to participate in the study.
Study participants
Incident cases of histologically confirmed oesophageal adenocarcinoma were identified from electronic pathology records in Northern Ireland and hospital clinical records in the Republic of Ireland. Non-dysplastic long-segment (≥3 cm) Barrett’s oesophagus cases were recruited if specialised intestinal metaplasia had been histologically confirmed. Reflux oesophagitis cases had erosions of the oesophageal mucosa diagnosed at endoscopy, classified as grades 2–4 or grades B, C or D using the Savary-Miller/Hetzel-Dent or Los Angeles methods( Reference Nayar and Vaezi 37 ), respectively. Population-based controls were adults with no previous history of Barrett’s oesophagus, oesophageal or other gastrointestinal cancer. Randomly selected controls were recruited via the General Practice Master Index in Northern Ireland and from four General Practices in the Republic of Ireland, representing both urban and rural areas within the Dublin and Cork regions. Response rates ranged from 42 % for controls to 69, 82 and 74 % for reflux oesophagitis, Barrett’s oesophagus and oesophageal adenocarcinoma cases, respectively.
Data collection
Trained interviewers collected data from study participants using an electronic questionnaire, which captured information on demographics, lifestyle, medication and co-morbidities. Dietary intake was assessed using a 101-item FFQ, adapted from a version of the European Prospective Investigation into Cancer and Nutrition FFQ( Reference Day, Oakes and Luben 38 ), by incorporating additional foods reported as commonly eaten in the North/South Ireland Food Consumption Survey( Reference Harrington, Robson and Kiely 39 ). Mean daily dietary intakes were calculated from the FFQ using Q-Builder (Tinuviel Software). Participants were asked to recall their dietary habits over the 12-month period 5 years before interview; BMI 5 years before the interview was calculated using self-reported weight (kg) divided by current height (m2), as measured by the interviewer. Helicobacter pylori infection status was assessed from serum samples using a Western blot assay, as previously described( Reference Anderson, Murphy and Johnston 40 ).
Statistical analysis
Participants were excluded from the analysis if they failed to complete the FFQ (n 22) or did not complete the FFQ section on dairy product intake (n 29), given our interest in studying the relationship between Mg and Ca intakes. This left 252 controls, 208 reflux oesophagitis, 212 Barrett’s oesophagus and 218 oesophageal adenocarcinoma cases for consideration in the current analysis.
Characteristics and mean nutrient intakes were compared between groups using independent t tests for continuous variables and χ 2 tests for categorical variables. Unconditional logistic regression analysis was applied to generate OR and corresponding 95 % CI for oesophageal lesions according to tertiles of intake. Reflux oesophagitis analyses were restricted to controls from Northern Ireland only, as these patients were recruited from Northern Ireland only. Tertiles of energy-adjusted nutrient intakes were defined by distribution in the appropriate controls. In order to test for trend, each person within a particular tertile was assigned the median intake value for that tertile before inclusion in the regression model.
The nutrient density method was utilised to adjust for energy intake, which calculates intakes per 4184 kJ/d (1000 kcal/d) in addition to including log kJ/d (kcal/d) in the regression models( Reference Willett and Stampfer 41 ). Other confounders included in the models were age (years), sex, smoking status (current/previous/never), education (years), BMI 5 years before the interview (kg/m2), occupation (manual/non-manual), alcohol intake (g/d), regular non-steroidal anti-inflammatory drug use (weekly use for at least 6 months duration), H. pylori infection (seronegative/seropositive) and location (Northern Ireland/Republic of Ireland). We also tested for energy-adjusted intakes of carbohydrate, fat, Ca and vitamin D, as well as for antioxidant score (a summary of combined intake of vitamin C, vitamin E, total carotenoids and Se, as previously described( Reference Murphy, Anderson and Ferguson 42 , Reference Kubo, Levin and Block 43 )). Confounders were selected due to being previously known risk factors for oesophageal lesions within this study population( Reference Mulholland, Murray and Anderson 26 , Reference Anderson, Johnston and Watson 34 – Reference Mulholland, Cantwell and Anderson 36 , Reference Anderson, Murphy and Johnston 40 , Reference Murphy, Anderson and Ferguson 42 , Reference Anderson, Cantwell and Watson 44 , Reference O’Doherty, Cantwell and Murray 45 ). In separate models, we further tested for regular gastro-oesophageal reflux symptoms (ever/never) and hiatus hernia (ever/never), as it is debatable whether reflux symptoms or hiatus hernia may confound or be on the causal pathway between disease risk and the dietary variables of interest. We also sought to adjust for dietary fibre intake; however, it was too highly correlated with Mg intake (r 0·58) to be included in the statistical models. In an attempt to explore this further, we used the residuals method (regressing energy-adjusted Mg and fibre intakes) to test for fibre as a potential confounder( Reference Willett and Stampfer 41 ). Further testing for diabetes history, in line with a peer-review suggestion, was also conducted.
Stratified analyses were carried out according to categories of vitamin D intake and Ca:Mg intake ratios. The likelihood-ratio test was applied to evaluate potential interactions in stratified analyses. All the statistical analyses were carried out using Intercooled Stata version 12.0 (StataCorp LP).
Results
Descriptive characteristics for reflux oesophagitis, Barrett’s oesophagus and oesophageal adenocarcinoma cases as well as controls are shown in Table 1. Oesophageal adenocarcinoma cases were more likely to be smokers, consume more alcohol, have a higher BMI, have worked in manual occupations and completed fewer years of education compared with controls (as did Barrett’s oesophagus cases). All three case groups were more likely to have experienced gastro-oesophageal reflux symptoms compared with controls.
Table 1 Characteristics of reflux oesophagitis, Barrett’s oesophagus and oesophageal adenocarcinoma cases and controls (Numbers and percentages for categorical variables; mean values and standard deviations for continuous data)
GOR, gastro-oesophageal reflux; NSAID, non-steroidal anti-inflammatory drug.
* Cases compared with Northern Ireland controls only, calculated using the t test (continuous variables) or χ 2 test (categorical variables).
† Cases compared with all controls, calculated using the t test (continuous variables) or χ 2 test (categorical variables).
‡ Heartburn/acid reflux symptoms experienced at least once weekly or >50 times/year >5 years before the interview date.
§ Includes both type 1 and type 2 diabetes – the majority of diabetes cases were type 2 diabetics, n 4 were type 1 diabetics.
|| Ever used, defined as use at least once weekly for ≥6 months’ duration.
Daily nutrient intakes of cases and controls are outlined in Table 2. Total energy and energy-adjusted fat intakes were greater across all case groups compared with controls. Energy-adjusted carbohydrate, fibre, antioxidant, Fe and Mg intakes were lower among all case groups compared with controls. Energy-adjusted Ca intake was lower in reflux oesophagitis cases only compared with controls, although Barrett’s oesophagus and oesophageal adenocarcinoma cases were more likely than controls to have a high Ca:Mg intake ratios (Table 2). There were no significant differences in energy-adjusted vitamin D intakes from foods between case groups and controls.
Table 2 Nutrient intakes of reflux oesophagitis, Barrett’s oesophagus and oesophageal adenocarcinoma cases and controls (Mean values and standard deviations for continuous data; numbers and percentages for categorical variables)
* Cases compared with Northern Ireland controls only, calculated using the t test (continuous variables) or χ 2 test (categorical variables).
† Cases compared with all controls, calculated using the t test (continuous variables) or χ 2 test (categorical variables).
‡ Energy-adjusted where indicated per 4184 kJ (1000 kcal).
§ Antioxidant index score reflects a summary of combined intake of vitamin C, vitamin E, total carotenoids and Se.
|| Ca:Mg intake ratio categories defined by the median value in Northern Ireland controls (3·15) for reflux oesophagitis analysis and all controls (3·05) for Barrett’s oesophagus and oesophageal adenocarcinoma analysis.
The association between Mg intake and disease risk is shown in Table 3. After adjustment for potential confounders, individuals having the highest Mg intakes from foods had significant reductions in odds of reflux oesophagitis (OR 0·31; 95 % CI 0·11, 0·87) and Barrett’s oesophagus (OR 0·29; 95 % CI 0·12, 0·71) compared with individuals having the lowest Mg intakes. Similarly strong inverse associations were detected for oesophageal adenocarcinoma in baseline models, but these became attenuated after adjustment for further confounders (tertile 3 (T3) v. T1; OR 0·77; 95 % CI 0·30, 1·99). Sensitivity analysis was conducted including further adjustment for gastro-oesophageal reflux symptoms, history of hiatus hernia, history of diabetes and daily fibre, vitamin D and Ca intakes, but these did not markedly alter the magnitude of associations observed (data not shown).
Table 3 Magnesium intake from foods and risk of reflux oesophagitis, Barrett’s oesophagus and oesophageal adenocarcinoma (Numbers; odds ratios and 95 % confidence intervals)
* Model 1: adjusted for age (years), sex, energy intake (by nutrient density method+log kJ/d (kcal/d)).
† Model 2: adjusted for model 1+smoking status (current/previous/never), BMI 5 years ago, education (years), occupation (manual/non-manual), alcohol (g/d), regular non-steroidal anti-inflammatory drug use (ever/never), Helicobacter pylori infection (seropositive/seronegative) and location (Northern Ireland/Republic of Ireland).
‡ Model 3: adjusted for model 2+antioxidant index score, energy-adjusted daily intakes of fat (g/4184 kJ per d (1000 kcal per d)) and carbohydrate (g/4184 kJ per d (1000 kcal per d)).
§ Analysis limited to Northern Ireland controls only.
Table 4 displays results from the stratified analysis of Ca and Mg intakes according to Ca:Mg intake ratios, in relation to reflux oesophagitis, Barrett’s oesophagus and oesophageal adenocarcinoma risk. The previously observed protective effect of high Mg intake was more evident in the context of a low Ca:Mg ratio intake for reflux oesophagitis (OR 0·12; 95 % CI 0·02, 0·73) and Barrett’s oesophagus (OR 0·24; 95 % CI 0·06, 0·96), although tests for interaction failed to achieve statistical significance (P=0·13 and P=0·26, respectively). Individuals having high Ca intakes also had reduced odds of reflux oesophagitis and Barrett’s oesophagus in the context of a low dietary Ca:Mg ratio. Ca:Mg ratio did not alter the null associations between either Ca or Mg and oesophageal adenocarcinoma risk (Table 4).
Table 4 Calcium and magnesium intake from foods and risk of reflux oesophagitis, Barrett’s oesophagus and oesophageal adenocarcinoma, stratified by Ca:Mg intake ratios (Odds ratios and 95 % confidence intervals)
* Ca:Mg intake ratio categories defined by the median value in Northern Ireland controls (3·15) for reflux oesophagitis analysis and all controls (3·05) for Barrett’s oesophagus and oesophageal adenocarcinoma analysis.
† Adjusted for age (years), sex, energy intake (by nutrient density method+log kJ/d (kcal/d)), smoking status (current/previous/never), BMI 5 years ago, education (years), occupation (manual/non-manual), alcohol (g/d), regular non-steroidal anti-inflammatory drug use (ever/never), Helicobacter pylori infection (seropositive/seronegative), location (Northern Ireland/Republic of Ireland), antioxidant index score and energy-adjusted daily intakes of fat (g) and carbohydrate (g).
‡ Analysis limited to Northern Ireland controls only.
We also investigated the association between disease risk and intakes of Ca, Mg and Ca:Mg intake ratios according to strata of vitamin D intakes, although no significant interactions with vitamin D intakes were detected (online Supplementary Table S1). Furthermore, the association between vitamin D intake and risk of reflux oesophagitis, Barrett’s oesophagus and oesophageal adenocarcinoma did not differ by strata of Ca:Mg intake ratios (online Supplementary Table S2).
Discussion
In this all-Ireland population-based study, high Mg intake was associated with a reduced risk of reflux oesophagitis and Barrett’s oesophagus but not oesophageal adenocarcinoma. The protective effect of Mg was particularly pronounced in the context of a low Ca:Mg ratio intake. This differential effect also applied to Ca intakes, whereby high Ca intakes were associated with reduced odds of reflux oesophagitis and Barrett’s oesophagus in the context of a low Ca:Mg intake ratio.
It is unclear why high Mg intake might particularly influence the earlier stages of cancer development, but not oesophageal adenocarcinoma. Growing evidence indicates that people with insulin resistance, the metabolic syndrome and type 2 diabetes are at high risk of Barrett’s oesophagus( Reference Rubenstein, Morgenstern and McConell 46 – Reference Iyer, Borah and Heien 50 ). Moreover, inflammation and reactive oxygen species play an important in the aetiology of Barrett’s oesophagus( Reference Nelsen, Kirihara and Takahashi 51 , Reference Poehlmann, Kuester and Malfertheiner 52 ). A number of epidemiological studies have linked low dietary intake of Mg to elevated risks of systemic inflammation( Reference Dibaba, Xun and He 53 , Reference Ziegler 54 ), insulin resistance( Reference Paolisso, Sgambato and Gambardella 55 – Reference Guerrero-Romero, Tamez-Perez and Gonzalez-Gonzalez 59 ), the metabolic syndrome( Reference Champagne 60 – Reference He, Song and Belin 62 ) and type 2 diabetes( Reference Song, Manson and Buring 58 , Reference Dong, Xun and He 63 – Reference Schulze, Schulz and Heidemann 67 ). In addition, Mg deficiency increased oxidative stress( Reference Hans, Chaudhary and Bansal 68 ), whereas Mg supplementation reduced oxidative stress in animal models( Reference Hans, Chaudhary and Bansal 6 ). Thus, it is biologically plausible that high intakes of Mg protect against Barrett’s oesophagus development.
Furthermore, one striking observation from animal studies is that Ca-adequate and Mg-deficient diets (i.e. diets with higher Ca:Mg ratios) led to increase in inflammatory responses and heart lipid peroxidation levels. Conversely, Ca-deficient and Mg-deficient diets (i.e. diets with lower Ca:Mg ratios) caused a significant reduction in heart lipid peroxidation and a normalisation of inflammatory responses( Reference Ziegler 54 , Reference Bussiere, Gueux and Rock 69 ). Thus, these findings indicate that in addition to intake of Mg, the Ca:Mg intake ratios contribute to the Mg status and, in turn, oxidative stress and inflammation status related to the development of Barrett’s oesophagus. One relevant observation is that, although Barrett’s oesophagus incidence is increasing, the prevalence rate is much lower in Asian populations( Reference Lee and Jeon 70 ) who have much lower Ca:Mg ratios compared with their Western counterparts( Reference Dai, Shu and Deng 24 ).
Interestingly, we also found that high Ca intake was related to reduced risks of reflux oesophagitis and Barrett’s oesophagus when the Ca:Mg intake ratios were below median levels. These findings are consistent with our earlier results on other gastrointestinal pre-malignant diseases, colorectal adenoma( Reference Dai, Shrubsole and Ness 8 ) and adenoma recurrence( Reference Dai, Sandler and Barry 23 ). On the basis of our previous studies conducted in US populations with high Ca:Mg ratios( Reference Dai, Shrubsole and Ness 8 , Reference Dai, Sandler and Barry 23 ) and in Chinese population with a very low Ca:Mg intake ratios( Reference Dai, Shu and Deng 24 ), it is likely that Ca:Mg ratios between 1·70 and 2·63 may be required for high intake of Mg or Ca to be protective against colorectal cancer and CVD( Reference Dai, Shu and Deng 24 ). Further studies are warranted to examine whether this is also true for reflux oesophagitis and Barrett’s oesophagus.
Adjustment for gastro-oesophageal reflux symptoms did not influence the results shown; therefore, it is unlikely that dietary Mg and Ca are mimicking properties of Ca/Mg-containing antacids in order to reduce the odds of reflux oesophagitis and Barrett’s oesophagus. However, little is known about the role of micronutrients in gastro-oesophageal reflux aetiology, and thus this potential mechanism cannot be ruled out.
We have not found a significant association between intake of Mg and risk for oesophageal adenocarcinoma or a significant interaction between intakes of vitamin D and Mg. Future larger studies are needed. Regarding the interaction with vitamin D, serum concentrations of 25-hydroxyvitamin D should also be used in future studies, as it is a more accurate biomarker of body vitamin D status than dietary intake of vitamin D.
This study has several strengths. It is a large population-based study that enabled the role of these dietary factors to be investigated throughout the oesophageal carcinogenesis pathway. Statistical analyses took into consideration a large number of potential confounders. To our knowledge, it is the first study to assess dietary Mg and Ca:Mg ratio in relation to oesophageal lesion risk.
Similar to all case–control studies, there is potential for recall bias due to the use of FFQ to enquire about habitual diet 5 years before interview. Such retrospective questioning of dietary habits is necessary to help overcome the impact of symptoms associated with prevalent disease and resulting changes to eating habits. However, it is unlikely that cases would have differentially reported food sources rich in Mg due to their diagnoses. The response rate for controls was lower than that observed for cases; however, the daily Mg intake from food in our population-based controls was similar to that estimated in national dietary surveys in Ireland, suggesting that our controls are representative of the general population( Reference Hannon, Kiely and Harrington 71 ). A further limitation of our study is that we considered only nutrient intake from foods and did not account for supplement usage. However, supplements were estimated to contribute to <5 % of total mineral intake (with the exception of Fe in females) in adults in the population at this time( Reference Hannon, Kiely and Harrington 71 ). This may lead to non-differential misclassification of Ca and Mg intakes, which usually biases associations towards the null. Future studies should take supplement usage into account and investigate their contribution to overall mineral intakes in association with risk for these diseases. Owing to the asymptomatic nature of Barrett’s oesophagus, and potentially oesophagitis, it is also possible that some of our control population may have had these conditions. Arguably, a superior control population would be individuals who have undergone endoscopy with negative findings; however, such a study design may impact on the generalisability of such controls.
Our findings indicate that high intake of Mg may protect against reflux oesophagitis and Barrett’s oesophagus. The protective effect of Mg may be particularly pronounced in the context of a low Ca:Mg ratio intake. This is also true for Ca. Future studies including cohort studies and clinical trials are necessary to confirm our findings. Our findings, if confirmed, will have important public health significance.
Acknowledgements
The Factors INfluencing the Barrett’s Adenocarcinoma Relationship (FINBAR) study group members include L. J. M. (Queen’s University Belfast), L. A. A. (Queen’s University Belfast), B. T. Johnston (Belfast Health & Social Care Trust), R. G. P. Watson (Belfast Health & Social Care Trust), J. McGuigan (Belfast Health & Social Care Trust), H. R. Ferguson (Belfast Health & Social Care Trust), S. J. Murphy (St Vincent’s Hospital Dublin), J. V. Reynolds (St James’ Hospital, Dublin) and H. Comber (National Cancer Registry of Ireland). The authors appreciate the contributions made by the study participants, their families and all of them who assisted with the study, particularly the Northern Ireland Cancer Registry and National Cancer Registry Cork.
The FINBAR study was supported by funding from Cancer Focus Northern Ireland (formerly the Ulster Cancer Foundation), the Northern Ireland R&D office and the Health Research Board. H. G. C. is currently supported by a Cancer Research UK Population Research Postdoctoral Fellowship. None of the funders had any role in the design, analysis or writing of this article.
L. J. M. and L. A. A. designed and conducted the FINBAR study research; Q. D., M. M. C., W. Z. and H. G. C. planned the nutritional and statistical analysis approach; H. G. C. analysed the data; and Q. D. and H. G. C. wrote the first draft of the paper. All the authors read and approved the final version of the manuscript.
The authors have no conflicts of interest to declare.
Supplementary material
For supplementary material/s referred to in this article, please visit http://dx.doi.org/doi:10.1017/S0007114515004444
